linux/fs/xfs/xfs_log_cil.c

1898 lines
58 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* Copyright (c) 2010 Red Hat, Inc. All Rights Reserved.
*/
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_shared.h"
#include "xfs_trans_resv.h"
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
#include "xfs_mount.h"
#include "xfs_extent_busy.h"
#include "xfs_trans.h"
#include "xfs_trans_priv.h"
#include "xfs_log.h"
#include "xfs_log_priv.h"
#include "xfs_trace.h"
struct workqueue_struct *xfs_discard_wq;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* Allocate a new ticket. Failing to get a new ticket makes it really hard to
* recover, so we don't allow failure here. Also, we allocate in a context that
* we don't want to be issuing transactions from, so we need to tell the
* allocation code this as well.
*
* We don't reserve any space for the ticket - we are going to steal whatever
* space we require from transactions as they commit. To ensure we reserve all
* the space required, we need to set the current reservation of the ticket to
* zero so that we know to steal the initial transaction overhead from the
* first transaction commit.
*/
static struct xlog_ticket *
xlog_cil_ticket_alloc(
struct xlog *log)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
{
struct xlog_ticket *tic;
tic = xlog_ticket_alloc(log, 0, 1, 0);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* set the current reservation to zero so we know to steal the basic
* transaction overhead reservation from the first transaction commit.
*/
tic->t_curr_res = 0;
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
tic->t_iclog_hdrs = 0;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
return tic;
}
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
static inline void
xlog_cil_set_iclog_hdr_count(struct xfs_cil *cil)
{
struct xlog *log = cil->xc_log;
atomic_set(&cil->xc_iclog_hdrs,
(XLOG_CIL_BLOCKING_SPACE_LIMIT(log) /
(log->l_iclog_size - log->l_iclog_hsize)));
}
/*
* Check if the current log item was first committed in this sequence.
* We can't rely on just the log item being in the CIL, we have to check
* the recorded commit sequence number.
*
* Note: for this to be used in a non-racy manner, it has to be called with
* CIL flushing locked out. As a result, it should only be used during the
* transaction commit process when deciding what to format into the item.
*/
static bool
xlog_item_in_current_chkpt(
struct xfs_cil *cil,
struct xfs_log_item *lip)
{
if (test_bit(XLOG_CIL_EMPTY, &cil->xc_flags))
return false;
/*
* li_seq is written on the first commit of a log item to record the
* first checkpoint it is written to. Hence if it is different to the
* current sequence, we're in a new checkpoint.
*/
return lip->li_seq == READ_ONCE(cil->xc_current_sequence);
}
bool
xfs_log_item_in_current_chkpt(
struct xfs_log_item *lip)
{
return xlog_item_in_current_chkpt(lip->li_log->l_cilp, lip);
}
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
/*
* Unavoidable forward declaration - xlog_cil_push_work() calls
* xlog_cil_ctx_alloc() itself.
*/
static void xlog_cil_push_work(struct work_struct *work);
static struct xfs_cil_ctx *
xlog_cil_ctx_alloc(void)
{
struct xfs_cil_ctx *ctx;
ctx = kmem_zalloc(sizeof(*ctx), KM_NOFS);
INIT_LIST_HEAD(&ctx->committing);
INIT_LIST_HEAD(&ctx->busy_extents);
INIT_LIST_HEAD(&ctx->log_items);
INIT_LIST_HEAD(&ctx->lv_chain);
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
INIT_WORK(&ctx->push_work, xlog_cil_push_work);
return ctx;
}
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
/*
* Aggregate the CIL per cpu structures into global counts, lists, etc and
* clear the percpu state ready for the next context to use. This is called
* from the push code with the context lock held exclusively, hence nothing else
* will be accessing or modifying the per-cpu counters.
*/
static void
xlog_cil_push_pcp_aggregate(
struct xfs_cil *cil,
struct xfs_cil_ctx *ctx)
{
struct xlog_cil_pcp *cilpcp;
int cpu;
for_each_online_cpu(cpu) {
cilpcp = per_cpu_ptr(cil->xc_pcp, cpu);
ctx->ticket->t_curr_res += cilpcp->space_reserved;
cilpcp->space_reserved = 0;
if (!list_empty(&cilpcp->busy_extents)) {
list_splice_init(&cilpcp->busy_extents,
&ctx->busy_extents);
}
if (!list_empty(&cilpcp->log_items))
list_splice_init(&cilpcp->log_items, &ctx->log_items);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
/*
* We're in the middle of switching cil contexts. Reset the
* counter we use to detect when the current context is nearing
* full.
*/
cilpcp->space_used = 0;
}
}
/*
* Aggregate the CIL per-cpu space used counters into the global atomic value.
* This is called when the per-cpu counter aggregation will first pass the soft
* limit threshold so we can switch to atomic counter aggregation for accurate
* detection of hard limit traversal.
*/
static void
xlog_cil_insert_pcp_aggregate(
struct xfs_cil *cil,
struct xfs_cil_ctx *ctx)
{
struct xlog_cil_pcp *cilpcp;
int cpu;
int count = 0;
/* Trigger atomic updates then aggregate only for the first caller */
if (!test_and_clear_bit(XLOG_CIL_PCP_SPACE, &cil->xc_flags))
return;
for_each_online_cpu(cpu) {
int old, prev;
cilpcp = per_cpu_ptr(cil->xc_pcp, cpu);
do {
old = cilpcp->space_used;
prev = cmpxchg(&cilpcp->space_used, old, 0);
} while (old != prev);
count += old;
}
atomic_add(count, &ctx->space_used);
}
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
static void
xlog_cil_ctx_switch(
struct xfs_cil *cil,
struct xfs_cil_ctx *ctx)
{
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
xlog_cil_set_iclog_hdr_count(cil);
set_bit(XLOG_CIL_EMPTY, &cil->xc_flags);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
set_bit(XLOG_CIL_PCP_SPACE, &cil->xc_flags);
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
ctx->sequence = ++cil->xc_current_sequence;
ctx->cil = cil;
cil->xc_ctx = ctx;
}
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* After the first stage of log recovery is done, we know where the head and
* tail of the log are. We need this log initialisation done before we can
* initialise the first CIL checkpoint context.
*
* Here we allocate a log ticket to track space usage during a CIL push. This
* ticket is passed to xlog_write() directly so that we don't slowly leak log
* space by failing to account for space used by log headers and additional
* region headers for split regions.
*/
void
xlog_cil_init_post_recovery(
struct xlog *log)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
{
log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
log->l_cilp->xc_ctx->sequence = 1;
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
xlog_cil_set_iclog_hdr_count(log->l_cilp);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
}
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
static inline int
xlog_cil_iovec_space(
uint niovecs)
{
return round_up((sizeof(struct xfs_log_vec) +
niovecs * sizeof(struct xfs_log_iovec)),
sizeof(uint64_t));
}
/*
* Allocate or pin log vector buffers for CIL insertion.
*
* The CIL currently uses disposable buffers for copying a snapshot of the
* modified items into the log during a push. The biggest problem with this is
* the requirement to allocate the disposable buffer during the commit if:
* a) does not exist; or
* b) it is too small
*
* If we do this allocation within xlog_cil_insert_format_items(), it is done
* under the xc_ctx_lock, which means that a CIL push cannot occur during
* the memory allocation. This means that we have a potential deadlock situation
* under low memory conditions when we have lots of dirty metadata pinned in
* the CIL and we need a CIL commit to occur to free memory.
*
* To avoid this, we need to move the memory allocation outside the
* xc_ctx_lock, but because the log vector buffers are disposable, that opens
* up a TOCTOU race condition w.r.t. the CIL committing and removing the log
* vector buffers between the check and the formatting of the item into the
* log vector buffer within the xc_ctx_lock.
*
* Because the log vector buffer needs to be unchanged during the CIL push
* process, we cannot share the buffer between the transaction commit (which
* modifies the buffer) and the CIL push context that is writing the changes
* into the log. This means skipping preallocation of buffer space is
* unreliable, but we most definitely do not want to be allocating and freeing
* buffers unnecessarily during commits when overwrites can be done safely.
*
* The simplest solution to this problem is to allocate a shadow buffer when a
* log item is committed for the second time, and then to only use this buffer
* if necessary. The buffer can remain attached to the log item until such time
* it is needed, and this is the buffer that is reallocated to match the size of
* the incoming modification. Then during the formatting of the item we can swap
* the active buffer with the new one if we can't reuse the existing buffer. We
* don't free the old buffer as it may be reused on the next modification if
* it's size is right, otherwise we'll free and reallocate it at that point.
*
* This function builds a vector for the changes in each log item in the
* transaction. It then works out the length of the buffer needed for each log
* item, allocates them and attaches the vector to the log item in preparation
* for the formatting step which occurs under the xc_ctx_lock.
*
* While this means the memory footprint goes up, it avoids the repeated
* alloc/free pattern that repeated modifications of an item would otherwise
* cause, and hence minimises the CPU overhead of such behaviour.
*/
static void
xlog_cil_alloc_shadow_bufs(
struct xlog *log,
struct xfs_trans *tp)
{
struct xfs_log_item *lip;
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
list_for_each_entry(lip, &tp->t_items, li_trans) {
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
struct xfs_log_vec *lv;
int niovecs = 0;
int nbytes = 0;
int buf_size;
bool ordered = false;
/* Skip items which aren't dirty in this transaction. */
if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
continue;
/* get number of vecs and size of data to be stored */
lip->li_ops->iop_size(lip, &niovecs, &nbytes);
/*
* Ordered items need to be tracked but we do not wish to write
* them. We need a logvec to track the object, but we do not
* need an iovec or buffer to be allocated for copying data.
*/
if (niovecs == XFS_LOG_VEC_ORDERED) {
ordered = true;
niovecs = 0;
nbytes = 0;
}
/*
xfs: reserve space and initialise xlog_op_header in item formatting Current xlog_write() adds op headers to the log manually for every log item region that is in the vector passed to it. While xlog_write() needs to stamp the transaction ID into the ophdr, we already know it's length, flags, clientid, etc at CIL commit time. This means the only time that xlog write really needs to format and reserve space for a new ophdr is when a region is split across two iclogs. Adding the opheader and accounting for it as part of the normal formatted item region means we simplify the accounting of space used by a transaction and we don't have to special case reserving of space in for the ophdrs in xlog_write(). It also means we can largely initialise the ophdr in transaction commit instead of xlog_write, making the xlog_write formatting inner loop much tighter. xlog_prepare_iovec() is now too large to stay as an inline function, so we move it out of line and into xfs_log.c. Object sizes: text data bss dec hex filename 1125934 305951 484 1432369 15db31 fs/xfs/built-in.a.before 1123360 305951 484 1429795 15d123 fs/xfs/built-in.a.after So the code is a roughly 2.5kB smaller with xlog_prepare_iovec() now out of line, even though it grew in size itself. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-04-21 00:34:59 +00:00
* We 64-bit align the length of each iovec so that the start of
* the next one is naturally aligned. We'll need to account for
* that slack space here.
*
* We also add the xlog_op_header to each region when
* formatting, but that's not accounted to the size of the item
* at this point. Hence we'll need an addition number of bytes
* for each vector to hold an opheader.
*
* Then round nbytes up to 64-bit alignment so that the initial
* buffer alignment is easy to calculate and verify.
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
*/
xfs: reserve space and initialise xlog_op_header in item formatting Current xlog_write() adds op headers to the log manually for every log item region that is in the vector passed to it. While xlog_write() needs to stamp the transaction ID into the ophdr, we already know it's length, flags, clientid, etc at CIL commit time. This means the only time that xlog write really needs to format and reserve space for a new ophdr is when a region is split across two iclogs. Adding the opheader and accounting for it as part of the normal formatted item region means we simplify the accounting of space used by a transaction and we don't have to special case reserving of space in for the ophdrs in xlog_write(). It also means we can largely initialise the ophdr in transaction commit instead of xlog_write, making the xlog_write formatting inner loop much tighter. xlog_prepare_iovec() is now too large to stay as an inline function, so we move it out of line and into xfs_log.c. Object sizes: text data bss dec hex filename 1125934 305951 484 1432369 15db31 fs/xfs/built-in.a.before 1123360 305951 484 1429795 15d123 fs/xfs/built-in.a.after So the code is a roughly 2.5kB smaller with xlog_prepare_iovec() now out of line, even though it grew in size itself. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-04-21 00:34:59 +00:00
nbytes += niovecs *
(sizeof(uint64_t) + sizeof(struct xlog_op_header));
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
nbytes = round_up(nbytes, sizeof(uint64_t));
/*
* The data buffer needs to start 64-bit aligned, so round up
* that space to ensure we can align it appropriately and not
* overrun the buffer.
*/
buf_size = nbytes + xlog_cil_iovec_space(niovecs);
/*
* if we have no shadow buffer, or it is too small, we need to
* reallocate it.
*/
if (!lip->li_lv_shadow ||
buf_size > lip->li_lv_shadow->lv_size) {
/*
* We free and allocate here as a realloc would copy
xfs: reduce kvmalloc overhead for CIL shadow buffers Oh, let me count the ways that the kvmalloc API sucks dog eggs. The problem is when we are logging lots of large objects, we hit kvmalloc really damn hard with costly order allocations, and behaviour utterly sucks: - 49.73% xlog_cil_commit - 31.62% kvmalloc_node - 29.96% __kmalloc_node - 29.38% kmalloc_large_node - 29.33% __alloc_pages - 24.33% __alloc_pages_slowpath.constprop.0 - 18.35% __alloc_pages_direct_compact - 17.39% try_to_compact_pages - compact_zone_order - 15.26% compact_zone 5.29% __pageblock_pfn_to_page 3.71% PageHuge - 1.44% isolate_migratepages_block 0.71% set_pfnblock_flags_mask 1.11% get_pfnblock_flags_mask - 0.81% get_page_from_freelist - 0.59% _raw_spin_lock_irqsave - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 3.24% try_to_free_pages - 3.14% shrink_node - 2.94% shrink_slab.constprop.0 - 0.89% super_cache_count - 0.66% xfs_fs_nr_cached_objects - 0.65% xfs_reclaim_inodes_count 0.55% xfs_perag_get_tag 0.58% kfree_rcu_shrink_count - 2.09% get_page_from_freelist - 1.03% _raw_spin_lock_irqsave - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 4.88% get_page_from_freelist - 3.66% _raw_spin_lock_irqsave - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 1.63% __vmalloc_node - __vmalloc_node_range - 1.10% __alloc_pages_bulk - 0.93% __alloc_pages - 0.92% get_page_from_freelist - 0.89% rmqueue_bulk - 0.69% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 13.73% memcpy_erms - 2.22% kvfree On this workload, that's almost a dozen CPUs all trying to compact and reclaim memory inside kvmalloc_node at the same time. Yet it is regularly falling back to vmalloc despite all that compaction, page and shrinker reclaim that direct reclaim is doing. Copying all the metadata is taking far less CPU time than allocating the storage! Direct reclaim should be considered extremely harmful. This is a high frequency, high throughput, CPU usage and latency sensitive allocation. We've got memory there, and we're using kvmalloc to allow memory allocation to avoid doing lots of work to try to do contiguous allocations. Except it still does *lots of costly work* that is unnecessary. Worse: the only way to avoid the slowpath page allocation trying to do compaction on costly allocations is to turn off direct reclaim (i.e. remove __GFP_RECLAIM_DIRECT from the gfp flags). Unfortunately, the stupid kvmalloc API then says "oh, this isn't a GFP_KERNEL allocation context, so you only get kmalloc!". This cuts off the vmalloc fallback, and this leads to almost instant OOM problems which ends up in filesystems deadlocks, shutdowns and/or kernel crashes. I want some basic kvmalloc behaviour: - kmalloc for a contiguous range with fail fast semantics - no compaction direct reclaim if the allocation enters the slow path. - run normal vmalloc (i.e. GFP_KERNEL) if kmalloc fails The really, really stupid part about this is these kvmalloc() calls are run under memalloc_nofs task context, so all the allocations are always reduced to GFP_NOFS regardless of the fact that kvmalloc requires GFP_KERNEL to be passed in. IOWs, we're already telling kvmalloc to behave differently to the gfp flags we pass in, but it still won't allow vmalloc to be run with anything other than GFP_KERNEL. So, this patch open codes the kvmalloc() in the commit path to have the above described behaviour. The result is we more than halve the CPU time spend doing kvmalloc() in this path and transaction commits with 64kB objects in them more than doubles. i.e. we get ~5x reduction in CPU usage per costly-sized kvmalloc() invocation and the profile looks like this: - 37.60% xlog_cil_commit 16.01% memcpy_erms - 8.45% __kmalloc - 8.04% kmalloc_order_trace - 8.03% kmalloc_order - 7.93% alloc_pages - 7.90% __alloc_pages - 4.05% __alloc_pages_slowpath.constprop.0 - 2.18% get_page_from_freelist - 1.77% wake_all_kswapds .... - __wake_up_common_lock - 0.94% _raw_spin_lock_irqsave - 3.72% get_page_from_freelist - 2.43% _raw_spin_lock_irqsave - 5.72% vmalloc - 5.72% __vmalloc_node_range - 4.81% __get_vm_area_node.constprop.0 - 3.26% alloc_vmap_area - 2.52% _raw_spin_lock - 1.46% _raw_spin_lock 0.56% __alloc_pages_bulk - 4.66% kvfree - 3.25% vfree - __vfree - 3.23% __vunmap - 1.95% remove_vm_area - 1.06% free_vmap_area_noflush - 0.82% _raw_spin_lock - 0.68% _raw_spin_lock - 0.92% _raw_spin_lock - 1.40% kfree - 1.36% __free_pages - 1.35% __free_pages_ok - 1.02% _raw_spin_lock_irqsave It's worth noting that over 50% of the CPU time spent allocating these shadow buffers is now spent on spinlocks. So the shadow buffer allocation overhead is greatly reduced by getting rid of direct reclaim from kmalloc, and could probably be made even less costly if vmalloc() didn't use global spinlocks to protect it's structures. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-01-05 01:22:18 +00:00
* unnecessary data. We don't use kvzalloc() for the
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
* same reason - we don't need to zero the data area in
* the buffer, only the log vector header and the iovec
* storage.
*/
kmem_free(lip->li_lv_shadow);
xfs: can't use kmem_zalloc() for attribute buffers Because heap allocation of 64kB buffers will fail: .... XFS: fs_mark(8414) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8417) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8409) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8428) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8430) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8437) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8433) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8406) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8412) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8432) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) XFS: fs_mark(8424) possible memory allocation deadlock size 65768 in kmem_alloc (mode:0x2d40) .... I'd use kvmalloc() instead, but.... - 48.19% xfs_attr_create_intent - 46.89% xfs_attri_init - kvmalloc_node - 46.04% __kmalloc_node - kmalloc_large_node - 45.99% __alloc_pages - 39.39% __alloc_pages_slowpath.constprop.0 - 38.89% __alloc_pages_direct_compact - 38.71% try_to_compact_pages - compact_zone_order - compact_zone - 21.09% isolate_migratepages_block 10.31% PageHuge 5.82% set_pfnblock_flags_mask 0.86% get_pfnblock_flags_mask - 4.48% __reset_isolation_suitable 4.44% __reset_isolation_pfn - 3.56% __pageblock_pfn_to_page 1.33% pfn_to_online_page 2.83% get_pfnblock_flags_mask - 0.87% migrate_pages 0.86% compaction_alloc 0.84% find_suitable_fallback - 6.60% get_page_from_freelist 4.99% clear_page_erms - 1.19% _raw_spin_lock_irqsave - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.86% __vmalloc_node_range 0.65% __alloc_pages_bulk .... this is just yet another reminder of how much kvmalloc() sucks. So lift xlog_cil_kvmalloc(), rename it to xlog_kvmalloc() and use that instead.... We also clean up the attribute name and value lengths as they no longer need to be rounded out to sizes compatible with log vectors. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-12 05:12:57 +00:00
lv = xlog_kvmalloc(buf_size);
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
memset(lv, 0, xlog_cil_iovec_space(niovecs));
INIT_LIST_HEAD(&lv->lv_list);
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
lv->lv_item = lip;
lv->lv_size = buf_size;
if (ordered)
lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
else
lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
lip->li_lv_shadow = lv;
} else {
/* same or smaller, optimise common overwrite case */
lv = lip->li_lv_shadow;
if (ordered)
lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
else
lv->lv_buf_len = 0;
lv->lv_bytes = 0;
}
/* Ensure the lv is set up according to ->iop_size */
lv->lv_niovecs = niovecs;
/* The allocated data region lies beyond the iovec region */
lv->lv_buf = (char *)lv + xlog_cil_iovec_space(niovecs);
}
}
/*
* Prepare the log item for insertion into the CIL. Calculate the difference in
* log space it will consume, and if it is a new item pin it as well.
*/
STATIC void
xfs_cil_prepare_item(
struct xlog *log,
struct xfs_log_vec *lv,
struct xfs_log_vec *old_lv,
int *diff_len)
{
/* Account for the new LV being passed in */
if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED)
xfs: log vector rounding leaks log space The addition of direct formatting of log items into the CIL linear buffer added alignment restrictions that the start of each vector needed to be 64 bit aligned. Hence padding was added in xlog_finish_iovec() to round up the vector length to ensure the next vector started with the correct alignment. This adds a small number of bytes to the size of the linear buffer that is otherwise unused. The issue is that we then use the linear buffer size to determine the log space used by the log item, and this includes the unused space. Hence when we account for space used by the log item, it's more than is actually written into the iclogs, and hence we slowly leak this space. This results on log hangs when reserving space, with threads getting stuck with these stack traces: Call Trace: [<ffffffff81d15989>] schedule+0x29/0x70 [<ffffffff8150d3a2>] xlog_grant_head_wait+0xa2/0x1a0 [<ffffffff8150d55d>] xlog_grant_head_check+0xbd/0x140 [<ffffffff8150ee33>] xfs_log_reserve+0x103/0x220 [<ffffffff814b7f05>] xfs_trans_reserve+0x2f5/0x310 ..... The 4 bytes is significant. Brain Foster did all the hard work in tracking down a reproducable leak to inode chunk allocation (it went away with the ikeep mount option). His rough numbers were that creating 50,000 inodes leaked 11 log blocks. This turns out to be roughly 800 inode chunks or 1600 inode cluster buffers. That works out at roughly 4 bytes per cluster buffer logged, and at that I started looking for a 4 byte leak in the buffer logging code. What I found was that a struct xfs_buf_log_format structure for an inode cluster buffer is 28 bytes in length. This gets rounded up to 32 bytes, but the vector length remains 28 bytes. Hence the CIL ticket reservation is decremented by 32 bytes (via lv->lv_buf_len) for that vector rather than 28 bytes which are written into the log. The fix for this problem is to separately track the bytes used by the log vectors in the item and use that instead of the buffer length when accounting for the log space that will be used by the formatted log item. Again, thanks to Brian Foster for doing all the hard work and long hours to isolate this leak and make finding the bug relatively simple. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-19 22:18:09 +00:00
*diff_len += lv->lv_bytes;
/*
* If there is no old LV, this is the first time we've seen the item in
* this CIL context and so we need to pin it. If we are replacing the
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
* old_lv, then remove the space it accounts for and make it the shadow
* buffer for later freeing. In both cases we are now switching to the
* shadow buffer, so update the pointer to it appropriately.
*/
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
if (!old_lv) {
if (lv->lv_item->li_ops->iop_pin)
lv->lv_item->li_ops->iop_pin(lv->lv_item);
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
lv->lv_item->li_lv_shadow = NULL;
} else if (old_lv != lv) {
ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
xfs: log vector rounding leaks log space The addition of direct formatting of log items into the CIL linear buffer added alignment restrictions that the start of each vector needed to be 64 bit aligned. Hence padding was added in xlog_finish_iovec() to round up the vector length to ensure the next vector started with the correct alignment. This adds a small number of bytes to the size of the linear buffer that is otherwise unused. The issue is that we then use the linear buffer size to determine the log space used by the log item, and this includes the unused space. Hence when we account for space used by the log item, it's more than is actually written into the iclogs, and hence we slowly leak this space. This results on log hangs when reserving space, with threads getting stuck with these stack traces: Call Trace: [<ffffffff81d15989>] schedule+0x29/0x70 [<ffffffff8150d3a2>] xlog_grant_head_wait+0xa2/0x1a0 [<ffffffff8150d55d>] xlog_grant_head_check+0xbd/0x140 [<ffffffff8150ee33>] xfs_log_reserve+0x103/0x220 [<ffffffff814b7f05>] xfs_trans_reserve+0x2f5/0x310 ..... The 4 bytes is significant. Brain Foster did all the hard work in tracking down a reproducable leak to inode chunk allocation (it went away with the ikeep mount option). His rough numbers were that creating 50,000 inodes leaked 11 log blocks. This turns out to be roughly 800 inode chunks or 1600 inode cluster buffers. That works out at roughly 4 bytes per cluster buffer logged, and at that I started looking for a 4 byte leak in the buffer logging code. What I found was that a struct xfs_buf_log_format structure for an inode cluster buffer is 28 bytes in length. This gets rounded up to 32 bytes, but the vector length remains 28 bytes. Hence the CIL ticket reservation is decremented by 32 bytes (via lv->lv_buf_len) for that vector rather than 28 bytes which are written into the log. The fix for this problem is to separately track the bytes used by the log vectors in the item and use that instead of the buffer length when accounting for the log space that will be used by the formatted log item. Again, thanks to Brian Foster for doing all the hard work and long hours to isolate this leak and make finding the bug relatively simple. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-19 22:18:09 +00:00
*diff_len -= old_lv->lv_bytes;
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
lv->lv_item->li_lv_shadow = old_lv;
}
/* attach new log vector to log item */
lv->lv_item->li_lv = lv;
/*
* If this is the first time the item is being committed to the
* CIL, store the sequence number on the log item so we can
* tell in future commits whether this is the first checkpoint
* the item is being committed into.
*/
if (!lv->lv_item->li_seq)
lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
}
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* Format log item into a flat buffers
*
* For delayed logging, we need to hold a formatted buffer containing all the
* changes on the log item. This enables us to relog the item in memory and
* write it out asynchronously without needing to relock the object that was
* modified at the time it gets written into the iclog.
*
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
* This function takes the prepared log vectors attached to each log item, and
* formats the changes into the log vector buffer. The buffer it uses is
* dependent on the current state of the vector in the CIL - the shadow lv is
* guaranteed to be large enough for the current modification, but we will only
* use that if we can't reuse the existing lv. If we can't reuse the existing
* lv, then simple swap it out for the shadow lv. We don't free it - that is
* done lazily either by th enext modification or the freeing of the log item.
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
*
* We don't set up region headers during this process; we simply copy the
* regions into the flat buffer. We can do this because we still have to do a
* formatting step to write the regions into the iclog buffer. Writing the
* ophdrs during the iclog write means that we can support splitting large
* regions across iclog boundares without needing a change in the format of the
* item/region encapsulation.
*
* Hence what we need to do now is change the rewrite the vector array to point
* to the copied region inside the buffer we just allocated. This allows us to
* format the regions into the iclog as though they are being formatted
* directly out of the objects themselves.
*/
static void
xlog_cil_insert_format_items(
struct xlog *log,
struct xfs_trans *tp,
int *diff_len)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
{
struct xfs_log_item *lip;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/* Bail out if we didn't find a log item. */
if (list_empty(&tp->t_items)) {
ASSERT(0);
return;
}
list_for_each_entry(lip, &tp->t_items, li_trans) {
struct xfs_log_vec *lv;
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
struct xfs_log_vec *old_lv = NULL;
struct xfs_log_vec *shadow;
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 06:04:51 +00:00
bool ordered = false;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/* Skip items which aren't dirty in this transaction. */
if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
continue;
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 06:04:51 +00:00
/*
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
* The formatting size information is already attached to
* the shadow lv on the log item.
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 06:04:51 +00:00
*/
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
shadow = lip->li_lv_shadow;
if (shadow->lv_buf_len == XFS_LOG_VEC_ORDERED)
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 06:04:51 +00:00
ordered = true;
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
/* Skip items that do not have any vectors for writing */
if (!shadow->lv_niovecs && !ordered)
continue;
/* compare to existing item size */
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
old_lv = lip->li_lv;
if (lip->li_lv && shadow->lv_size <= lip->li_lv->lv_size) {
/* same or smaller, optimise common overwrite case */
lv = lip->li_lv;
if (ordered)
goto insert;
/*
* set the item up as though it is a new insertion so
* that the space reservation accounting is correct.
*/
xfs: log vector rounding leaks log space The addition of direct formatting of log items into the CIL linear buffer added alignment restrictions that the start of each vector needed to be 64 bit aligned. Hence padding was added in xlog_finish_iovec() to round up the vector length to ensure the next vector started with the correct alignment. This adds a small number of bytes to the size of the linear buffer that is otherwise unused. The issue is that we then use the linear buffer size to determine the log space used by the log item, and this includes the unused space. Hence when we account for space used by the log item, it's more than is actually written into the iclogs, and hence we slowly leak this space. This results on log hangs when reserving space, with threads getting stuck with these stack traces: Call Trace: [<ffffffff81d15989>] schedule+0x29/0x70 [<ffffffff8150d3a2>] xlog_grant_head_wait+0xa2/0x1a0 [<ffffffff8150d55d>] xlog_grant_head_check+0xbd/0x140 [<ffffffff8150ee33>] xfs_log_reserve+0x103/0x220 [<ffffffff814b7f05>] xfs_trans_reserve+0x2f5/0x310 ..... The 4 bytes is significant. Brain Foster did all the hard work in tracking down a reproducable leak to inode chunk allocation (it went away with the ikeep mount option). His rough numbers were that creating 50,000 inodes leaked 11 log blocks. This turns out to be roughly 800 inode chunks or 1600 inode cluster buffers. That works out at roughly 4 bytes per cluster buffer logged, and at that I started looking for a 4 byte leak in the buffer logging code. What I found was that a struct xfs_buf_log_format structure for an inode cluster buffer is 28 bytes in length. This gets rounded up to 32 bytes, but the vector length remains 28 bytes. Hence the CIL ticket reservation is decremented by 32 bytes (via lv->lv_buf_len) for that vector rather than 28 bytes which are written into the log. The fix for this problem is to separately track the bytes used by the log vectors in the item and use that instead of the buffer length when accounting for the log space that will be used by the formatted log item. Again, thanks to Brian Foster for doing all the hard work and long hours to isolate this leak and make finding the bug relatively simple. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-19 22:18:09 +00:00
*diff_len -= lv->lv_bytes;
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
/* Ensure the lv is set up according to ->iop_size */
lv->lv_niovecs = shadow->lv_niovecs;
/* reset the lv buffer information for new formatting */
lv->lv_buf_len = 0;
lv->lv_bytes = 0;
lv->lv_buf = (char *)lv +
xlog_cil_iovec_space(lv->lv_niovecs);
} else {
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
/* switch to shadow buffer! */
lv = shadow;
lv->lv_item = lip;
if (ordered) {
/* track as an ordered logvec */
ASSERT(lip->li_lv == NULL);
goto insert;
}
}
ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
lip->li_ops->iop_format(lip, lv);
insert:
xfs_cil_prepare_item(log, lv, old_lv, diff_len);
}
}
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
/*
* The use of lockless waitqueue_active() requires that the caller has
* serialised itself against the wakeup call in xlog_cil_push_work(). That
* can be done by either holding the push lock or the context lock.
*/
static inline bool
xlog_cil_over_hard_limit(
struct xlog *log,
int32_t space_used)
{
if (waitqueue_active(&log->l_cilp->xc_push_wait))
return true;
if (space_used >= XLOG_CIL_BLOCKING_SPACE_LIMIT(log))
return true;
return false;
}
/*
* Insert the log items into the CIL and calculate the difference in space
* consumed by the item. Add the space to the checkpoint ticket and calculate
* if the change requires additional log metadata. If it does, take that space
* as well. Remove the amount of space we added to the checkpoint ticket from
* the current transaction ticket so that the accounting works out correctly.
*/
static void
xlog_cil_insert_items(
struct xlog *log,
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
struct xfs_trans *tp,
uint32_t released_space)
{
struct xfs_cil *cil = log->l_cilp;
struct xfs_cil_ctx *ctx = cil->xc_ctx;
struct xfs_log_item *lip;
int len = 0;
int iovhdr_res = 0, split_res = 0, ctx_res = 0;
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
int space_used;
int order;
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
struct xlog_cil_pcp *cilpcp;
ASSERT(tp);
/*
* We can do this safely because the context can't checkpoint until we
* are done so it doesn't matter exactly how we update the CIL.
*/
xlog_cil_insert_format_items(log, tp, &len);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
/*
* Subtract the space released by intent cancelation from the space we
* consumed so that we remove it from the CIL space and add it back to
* the current transaction reservation context.
*/
len -= released_space;
/*
* Grab the per-cpu pointer for the CIL before we start any accounting.
* That ensures that we are running with pre-emption disabled and so we
* can't be scheduled away between split sample/update operations that
* are done without outside locking to serialise them.
*/
cilpcp = get_cpu_ptr(cil->xc_pcp);
/*
* We need to take the CIL checkpoint unit reservation on the first
* commit into the CIL. Test the XLOG_CIL_EMPTY bit first so we don't
* unnecessarily do an atomic op in the fast path here. We can clear the
* XLOG_CIL_EMPTY bit as we are under the xc_ctx_lock here and that
* needs to be held exclusively to reset the XLOG_CIL_EMPTY bit.
*/
if (test_bit(XLOG_CIL_EMPTY, &cil->xc_flags) &&
test_and_clear_bit(XLOG_CIL_EMPTY, &cil->xc_flags))
ctx_res = ctx->ticket->t_unit_res;
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
/*
* Check if we need to steal iclog headers. atomic_read() is not a
* locked atomic operation, so we can check the value before we do any
* real atomic ops in the fast path. If we've already taken the CIL unit
* reservation from this commit, we've already got one iclog header
* space reserved so we have to account for that otherwise we risk
* overrunning the reservation on this ticket.
*
* If the CIL is already at the hard limit, we might need more header
* space that originally reserved. So steal more header space from every
* commit that occurs once we are over the hard limit to ensure the CIL
* push won't run out of reservation space.
*
* This can steal more than we need, but that's OK.
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
*
* The cil->xc_ctx_lock provides the serialisation necessary for safely
* calling xlog_cil_over_hard_limit() in this context.
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
*/
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
space_used = atomic_read(&ctx->space_used) + cilpcp->space_used + len;
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
if (atomic_read(&cil->xc_iclog_hdrs) > 0 ||
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
xlog_cil_over_hard_limit(log, space_used)) {
split_res = log->l_iclog_hsize +
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
sizeof(struct xlog_op_header);
if (ctx_res)
ctx_res += split_res * (tp->t_ticket->t_iclog_hdrs - 1);
else
ctx_res = split_res * tp->t_ticket->t_iclog_hdrs;
atomic_sub(tp->t_ticket->t_iclog_hdrs, &cil->xc_iclog_hdrs);
}
cilpcp->space_reserved += ctx_res;
xfs: rework per-iclog header CIL reservation For every iclog that a CIL push will use up, we need to ensure we have space reserved for the iclog header in each iclog. It is extremely difficult to do this accurately with a per-cpu counter without expensive summing of the counter in every commit. However, we know what the maximum CIL size is going to be because of the hard space limit we have, and hence we know exactly how many iclogs we are going to need to write out the CIL. We are constrained by the requirement that small transactions only have reservation space for a single iclog header built into them. At commit time we don't know how much of the current transaction reservation is made up of iclog header reservations as calculated by xfs_log_calc_unit_res() when the ticket was reserved. As larger reservations have multiple header spaces reserved, we can steal more than one iclog header reservation at a time, but we only steal the exact number needed for the given log vector size delta. As a result, we don't know exactly when we are going to steal iclog header reservations, nor do we know exactly how many we are going to need for a given CIL. To make things simple, start by calculating the worst case number of iclog headers a full CIL push will require. Record this into an atomic variable in the CIL. Then add a byte counter to the log ticket that records exactly how much iclog header space has been reserved in this ticket by xfs_log_calc_unit_res(). This tells us exactly how much space we can steal from the ticket at transaction commit time. Now, at transaction commit time, we can check if the CIL has a full iclog header reservation and, if not, steal the entire reservation the current ticket holds for iclog headers. This minimises the number of times we need to do atomic operations in the fast path, but still guarantees we get all the reservations we need. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-01 16:12:52 +00:00
/*
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
* Accurately account when over the soft limit, otherwise fold the
* percpu count into the global count if over the per-cpu threshold.
*/
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
if (!test_bit(XLOG_CIL_PCP_SPACE, &cil->xc_flags)) {
atomic_add(len, &ctx->space_used);
} else if (cilpcp->space_used + len >
(XLOG_CIL_SPACE_LIMIT(log) / num_online_cpus())) {
space_used = atomic_add_return(cilpcp->space_used + len,
&ctx->space_used);
cilpcp->space_used = 0;
/*
* If we just transitioned over the soft limit, we need to
* transition to the global atomic counter.
*/
if (space_used >= XLOG_CIL_SPACE_LIMIT(log))
xlog_cil_insert_pcp_aggregate(cil, ctx);
} else {
cilpcp->space_used += len;
}
/* attach the transaction to the CIL if it has any busy extents */
if (!list_empty(&tp->t_busy))
list_splice_init(&tp->t_busy, &cilpcp->busy_extents);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
/*
* Now update the order of everything modified in the transaction
* and insert items into the CIL if they aren't already there.
* We do this here so we only need to take the CIL lock once during
* the transaction commit.
*/
order = atomic_inc_return(&ctx->order_id);
list_for_each_entry(lip, &tp->t_items, li_trans) {
/* Skip items which aren't dirty in this transaction. */
if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
continue;
lip->li_order_id = order;
if (!list_empty(&lip->li_cil))
continue;
list_add_tail(&lip->li_cil, &cilpcp->log_items);
}
put_cpu_ptr(cilpcp);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
/*
* If we've overrun the reservation, dump the tx details before we move
* the log items. Shutdown is imminent...
*/
tp->t_ticket->t_curr_res -= ctx_res + len;
if (WARN_ON(tp->t_ticket->t_curr_res < 0)) {
xfs_warn(log->l_mp, "Transaction log reservation overrun:");
xfs_warn(log->l_mp,
" log items: %d bytes (iov hdrs: %d bytes)",
len, iovhdr_res);
xfs_warn(log->l_mp, " split region headers: %d bytes",
split_res);
xfs_warn(log->l_mp, " ctx ticket: %d bytes", ctx_res);
xlog_print_trans(tp);
xfs: log shutdown triggers should only shut down the log We've got a mess on our hands. 1. xfs_trans_commit() cannot cancel transactions because the mount is shut down - that causes dirty, aborted, unlogged log items to sit unpinned in memory and potentially get written to disk before the log is shut down. Hence xfs_trans_commit() can only abort transactions when xlog_is_shutdown() is true. 2. xfs_force_shutdown() is used in places to cause the current modification to be aborted via xfs_trans_commit() because it may be impractical or impossible to cancel the transaction directly, and hence xfs_trans_commit() must cancel transactions when xfs_is_shutdown() is true in this situation. But we can't do that because of #1. 3. Log IO errors cause log shutdowns by calling xfs_force_shutdown() to shut down the mount and then the log from log IO completion. 4. xfs_force_shutdown() can result in a log force being issued, which has to wait for log IO completion before it will mark the log as shut down. If #3 races with some other shutdown trigger that runs a log force, we rely on xfs_force_shutdown() silently ignoring #3 and avoiding shutting down the log until the failed log force completes. 5. To ensure #2 always works, we have to ensure that xfs_force_shutdown() does not return until the the log is shut down. But in the case of #4, this will result in a deadlock because the log Io completion will block waiting for a log force to complete which is blocked waiting for log IO to complete.... So the very first thing we have to do here to untangle this mess is dissociate log shutdown triggers from mount shutdowns. We already have xlog_forced_shutdown, which will atomically transistion to the log a shutdown state. Due to internal asserts it cannot be called multiple times, but was done simply because the only place that could call it was xfs_do_force_shutdown() (i.e. the mount shutdown!) and that could only call it once and once only. So the first thing we do is remove the asserts. We then convert all the internal log shutdown triggers to call xlog_force_shutdown() directly instead of xfs_force_shutdown(). This allows the log shutdown triggers to shut down the log without needing to care about mount based shutdown constraints. This means we shut down the log independently of the mount and the mount may not notice this until it's next attempt to read or modify metadata. At that point (e.g. xfs_trans_commit()) it will see that the log is shutdown, error out and shutdown the mount. To ensure that all the unmount behaviours and asserts track correctly as a result of a log shutdown, propagate the shutdown up to the mount if it is not already set. This keeps the mount and log state in sync, and saves a huge amount of hassle where code fails because of a log shutdown but only checks for mount shutdowns and hence ends up doing the wrong thing. Cleaning up that mess is an exercise for another day. This enables us to address the other problems noted above in followup patches. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 01:22:01 +00:00
xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
}
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
}
static void
xlog_cil_free_logvec(
struct list_head *lv_chain)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
{
struct xfs_log_vec *lv;
while (!list_empty(lv_chain)) {
lv = list_first_entry(lv_chain, struct xfs_log_vec, lv_list);
list_del_init(&lv->lv_list);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
kmem_free(lv);
}
}
static void
xlog_discard_endio_work(
struct work_struct *work)
{
struct xfs_cil_ctx *ctx =
container_of(work, struct xfs_cil_ctx, discard_endio_work);
struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
kmem_free(ctx);
}
/*
* Queue up the actual completion to a thread to avoid IRQ-safe locking for
* pagb_lock. Note that we need a unbounded workqueue, otherwise we might
* get the execution delayed up to 30 seconds for weird reasons.
*/
static void
xlog_discard_endio(
struct bio *bio)
{
struct xfs_cil_ctx *ctx = bio->bi_private;
INIT_WORK(&ctx->discard_endio_work, xlog_discard_endio_work);
queue_work(xfs_discard_wq, &ctx->discard_endio_work);
bio_put(bio);
}
static void
xlog_discard_busy_extents(
struct xfs_mount *mp,
struct xfs_cil_ctx *ctx)
{
struct list_head *list = &ctx->busy_extents;
struct xfs_extent_busy *busyp;
struct bio *bio = NULL;
struct blk_plug plug;
int error = 0;
ASSERT(xfs_has_discard(mp));
blk_start_plug(&plug);
list_for_each_entry(busyp, list, list) {
trace_xfs_discard_extent(mp, busyp->agno, busyp->bno,
busyp->length);
error = __blkdev_issue_discard(mp->m_ddev_targp->bt_bdev,
XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno),
XFS_FSB_TO_BB(mp, busyp->length),
GFP_NOFS, &bio);
if (error && error != -EOPNOTSUPP) {
xfs_info(mp,
"discard failed for extent [0x%llx,%u], error %d",
(unsigned long long)busyp->bno,
busyp->length,
error);
break;
}
}
if (bio) {
bio->bi_private = ctx;
bio->bi_end_io = xlog_discard_endio;
submit_bio(bio);
} else {
xlog_discard_endio_work(&ctx->discard_endio_work);
}
blk_finish_plug(&plug);
}
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* Mark all items committed and clear busy extents. We free the log vector
* chains in a separate pass so that we unpin the log items as quickly as
* possible.
*/
static void
xlog_cil_committed(
struct xfs_cil_ctx *ctx)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
{
struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
bool abort = xlog_is_shutdown(ctx->cil->xc_log);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
xfs: wake commit waiters on CIL abort before log item abort XFS shutdown deadlocks have been reproduced by fstest generic/475. The deadlock signature involves log I/O completion running error handling to abort logged items and waiting for an inode cluster buffer lock in the buffer item unpin handler. The buffer lock is held by xfsaild attempting to flush an inode. The buffer happens to be pinned and so xfs_iflush() triggers an async log force to begin work required to get it unpinned. The log force is blocked waiting on the commit completion, which never occurs and thus leaves the filesystem deadlocked. The root problem is that aborted log I/O completion pots commit completion behind callback completion, which is unexpected for async log forces. Under normal running conditions, an async log force returns to the caller once the CIL ctx has been formatted/submitted and the commit completion event triggered at the tail end of xlog_cil_push(). If the filesystem has shutdown, however, we rely on xlog_cil_committed() to trigger the completion event and it happens to do so after running log item unpin callbacks. This makes it unsafe to invoke an async log force from contexts that hold locks that might also be required in log completion processing. To address this problem, wake commit completion waiters before aborting log items in the log I/O completion handler. This ensures that an async log force will not deadlock on held locks if the filesystem happens to shutdown. Note that it is still unsafe to issue a sync log force while holding such locks because a sync log force explicitly waits on the force completion, which occurs after log I/O completion processing. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-04-12 14:39:20 +00:00
/*
* If the I/O failed, we're aborting the commit and already shutdown.
* Wake any commit waiters before aborting the log items so we don't
* block async log pushers on callbacks. Async log pushers explicitly do
* not wait on log force completion because they may be holding locks
* required to unpin items.
*/
if (abort) {
spin_lock(&ctx->cil->xc_push_lock);
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
wake_up_all(&ctx->cil->xc_start_wait);
xfs: wake commit waiters on CIL abort before log item abort XFS shutdown deadlocks have been reproduced by fstest generic/475. The deadlock signature involves log I/O completion running error handling to abort logged items and waiting for an inode cluster buffer lock in the buffer item unpin handler. The buffer lock is held by xfsaild attempting to flush an inode. The buffer happens to be pinned and so xfs_iflush() triggers an async log force to begin work required to get it unpinned. The log force is blocked waiting on the commit completion, which never occurs and thus leaves the filesystem deadlocked. The root problem is that aborted log I/O completion pots commit completion behind callback completion, which is unexpected for async log forces. Under normal running conditions, an async log force returns to the caller once the CIL ctx has been formatted/submitted and the commit completion event triggered at the tail end of xlog_cil_push(). If the filesystem has shutdown, however, we rely on xlog_cil_committed() to trigger the completion event and it happens to do so after running log item unpin callbacks. This makes it unsafe to invoke an async log force from contexts that hold locks that might also be required in log completion processing. To address this problem, wake commit completion waiters before aborting log items in the log I/O completion handler. This ensures that an async log force will not deadlock on held locks if the filesystem happens to shutdown. Note that it is still unsafe to issue a sync log force while holding such locks because a sync log force explicitly waits on the force completion, which occurs after log I/O completion processing. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-04-12 14:39:20 +00:00
wake_up_all(&ctx->cil->xc_commit_wait);
spin_unlock(&ctx->cil->xc_push_lock);
}
xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, &ctx->lv_chain,
xfs: bulk AIL insertion during transaction commit When inserting items into the AIL from the transaction committed callbacks, we take the AIL lock for every single item that is to be inserted. For a CIL checkpoint commit, this can be tens of thousands of individual inserts, yet almost all of the items will be inserted at the same point in the AIL because they have the same index. To reduce the overhead and contention on the AIL lock for such operations, introduce a "bulk insert" operation which allows a list of log items with the same LSN to be inserted in a single operation via a list splice. To do this, we need to pre-sort the log items being committed into a temporary list for insertion. The complexity is that not every log item will end up with the same LSN, and not every item is actually inserted into the AIL. Items that don't match the commit LSN will be inserted and unpinned as per the current one-at-a-time method (relatively rare), while items that are not to be inserted will be unpinned and freed immediately. Items that are to be inserted at the given commit lsn are placed in a temporary array and inserted into the AIL in bulk each time the array fills up. As a result of this, we trade off AIL hold time for a significant reduction in traffic. lock_stat output shows that the worst case hold time is unchanged, but contention from AIL inserts drops by an order of magnitude and the number of lock traversal decreases significantly. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-12-20 01:02:19 +00:00
ctx->start_lsn, abort);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
xfs_extent_busy_sort(&ctx->busy_extents);
xfs_extent_busy_clear(mp, &ctx->busy_extents,
xfs_has_discard(mp) && !abort);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
spin_lock(&ctx->cil->xc_push_lock);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
list_del(&ctx->committing);
spin_unlock(&ctx->cil->xc_push_lock);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
xlog_cil_free_logvec(&ctx->lv_chain);
if (!list_empty(&ctx->busy_extents))
xlog_discard_busy_extents(mp, ctx);
else
kmem_free(ctx);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
}
void
xlog_cil_process_committed(
struct list_head *list)
{
struct xfs_cil_ctx *ctx;
while ((ctx = list_first_entry_or_null(list,
struct xfs_cil_ctx, iclog_entry))) {
list_del(&ctx->iclog_entry);
xlog_cil_committed(ctx);
}
}
/*
* Record the LSN of the iclog we were just granted space to start writing into.
* If the context doesn't have a start_lsn recorded, then this iclog will
* contain the start record for the checkpoint. Otherwise this write contains
* the commit record for the checkpoint.
*/
void
xlog_cil_set_ctx_write_state(
struct xfs_cil_ctx *ctx,
struct xlog_in_core *iclog)
{
struct xfs_cil *cil = ctx->cil;
xfs_lsn_t lsn = be64_to_cpu(iclog->ic_header.h_lsn);
ASSERT(!ctx->commit_lsn);
if (!ctx->start_lsn) {
spin_lock(&cil->xc_push_lock);
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
/*
* The LSN we need to pass to the log items on transaction
* commit is the LSN reported by the first log vector write, not
* the commit lsn. If we use the commit record lsn then we can
xfs: drop async cache flushes from CIL commits. Jan Kara reported a performance regression in dbench that he bisected down to commit bad77c375e8d ("xfs: CIL checkpoint flushes caches unconditionally"). Whilst developing the journal flush/fua optimisations this cache was part of, it appeared to made a significant difference to performance. However, now that this patchset has settled and all the correctness issues fixed, there does not appear to be any significant performance benefit to asynchronous cache flushes. In fact, the opposite is true on some storage types and workloads, where additional cache flushes that can occur from fsync heavy workloads have measurable and significant impact on overall throughput. Local dbench testing shows little difference on dbench runs with sync vs async cache flushes on either fast or slow SSD storage, and no difference in streaming concurrent async transaction workloads like fs-mark. Fast NVME storage. From `dbench -t 30`, CIL scale: clients async sync BW Latency BW Latency 1 935.18 0.855 915.64 0.903 8 2404.51 6.873 2341.77 6.511 16 3003.42 6.460 2931.57 6.529 32 3697.23 7.939 3596.28 7.894 128 7237.43 15.495 7217.74 11.588 512 5079.24 90.587 5167.08 95.822 fsmark, 32 threads, create w/ 64 byte xattr w/32k logbsize create chown unlink async 1m41s 1m16s 2m03s sync 1m40s 1m19s 1m54s Slower SATA SSD storage: From `dbench -t 30`, CIL scale: clients async sync BW Latency BW Latency 1 78.59 15.792 83.78 10.729 8 367.88 92.067 404.63 59.943 16 564.51 72.524 602.71 76.089 32 831.66 105.984 870.26 110.482 128 1659.76 102.969 1624.73 91.356 512 2135.91 223.054 2603.07 161.160 fsmark, 16 threads, create w/32k logbsize create unlink async 5m06s 4m15s sync 5m00s 4m22s And on Jan's test machine: 5.18-rc8-vanilla 5.18-rc8-patched Amean 1 71.22 ( 0.00%) 64.94 * 8.81%* Amean 2 93.03 ( 0.00%) 84.80 * 8.85%* Amean 4 150.54 ( 0.00%) 137.51 * 8.66%* Amean 8 252.53 ( 0.00%) 242.24 * 4.08%* Amean 16 454.13 ( 0.00%) 439.08 * 3.31%* Amean 32 835.24 ( 0.00%) 829.74 * 0.66%* Amean 64 1740.59 ( 0.00%) 1686.73 * 3.09%* Performance and cache flush behaviour is restored to pre-regression levels. As such, we can now consider the async cache flush mechanism an unnecessary exercise in premature optimisation and hence we can now remove it and the infrastructure it requires completely. Fixes: bad77c375e8d ("xfs: CIL checkpoint flushes caches unconditionally") Reported-and-tested-by: Jan Kara <jack@suse.cz> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 01:22:02 +00:00
* move the grant write head beyond the tail LSN and overwrite
* it.
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
*/
ctx->start_lsn = lsn;
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
wake_up_all(&cil->xc_start_wait);
spin_unlock(&cil->xc_push_lock);
xfs: drop async cache flushes from CIL commits. Jan Kara reported a performance regression in dbench that he bisected down to commit bad77c375e8d ("xfs: CIL checkpoint flushes caches unconditionally"). Whilst developing the journal flush/fua optimisations this cache was part of, it appeared to made a significant difference to performance. However, now that this patchset has settled and all the correctness issues fixed, there does not appear to be any significant performance benefit to asynchronous cache flushes. In fact, the opposite is true on some storage types and workloads, where additional cache flushes that can occur from fsync heavy workloads have measurable and significant impact on overall throughput. Local dbench testing shows little difference on dbench runs with sync vs async cache flushes on either fast or slow SSD storage, and no difference in streaming concurrent async transaction workloads like fs-mark. Fast NVME storage. From `dbench -t 30`, CIL scale: clients async sync BW Latency BW Latency 1 935.18 0.855 915.64 0.903 8 2404.51 6.873 2341.77 6.511 16 3003.42 6.460 2931.57 6.529 32 3697.23 7.939 3596.28 7.894 128 7237.43 15.495 7217.74 11.588 512 5079.24 90.587 5167.08 95.822 fsmark, 32 threads, create w/ 64 byte xattr w/32k logbsize create chown unlink async 1m41s 1m16s 2m03s sync 1m40s 1m19s 1m54s Slower SATA SSD storage: From `dbench -t 30`, CIL scale: clients async sync BW Latency BW Latency 1 78.59 15.792 83.78 10.729 8 367.88 92.067 404.63 59.943 16 564.51 72.524 602.71 76.089 32 831.66 105.984 870.26 110.482 128 1659.76 102.969 1624.73 91.356 512 2135.91 223.054 2603.07 161.160 fsmark, 16 threads, create w/32k logbsize create unlink async 5m06s 4m15s sync 5m00s 4m22s And on Jan's test machine: 5.18-rc8-vanilla 5.18-rc8-patched Amean 1 71.22 ( 0.00%) 64.94 * 8.81%* Amean 2 93.03 ( 0.00%) 84.80 * 8.85%* Amean 4 150.54 ( 0.00%) 137.51 * 8.66%* Amean 8 252.53 ( 0.00%) 242.24 * 4.08%* Amean 16 454.13 ( 0.00%) 439.08 * 3.31%* Amean 32 835.24 ( 0.00%) 829.74 * 0.66%* Amean 64 1740.59 ( 0.00%) 1686.73 * 3.09%* Performance and cache flush behaviour is restored to pre-regression levels. As such, we can now consider the async cache flush mechanism an unnecessary exercise in premature optimisation and hence we can now remove it and the infrastructure it requires completely. Fixes: bad77c375e8d ("xfs: CIL checkpoint flushes caches unconditionally") Reported-and-tested-by: Jan Kara <jack@suse.cz> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 01:22:02 +00:00
/*
* Make sure the metadata we are about to overwrite in the log
* has been flushed to stable storage before this iclog is
* issued.
*/
spin_lock(&cil->xc_log->l_icloglock);
iclog->ic_flags |= XLOG_ICL_NEED_FLUSH;
spin_unlock(&cil->xc_log->l_icloglock);
return;
}
/*
* Take a reference to the iclog for the context so that we still hold
* it when xlog_write is done and has released it. This means the
* context controls when the iclog is released for IO.
*/
atomic_inc(&iclog->ic_refcnt);
/*
* xlog_state_get_iclog_space() guarantees there is enough space in the
* iclog for an entire commit record, so we can attach the context
* callbacks now. This needs to be done before we make the commit_lsn
* visible to waiters so that checkpoints with commit records in the
* same iclog order their IO completion callbacks in the same order that
* the commit records appear in the iclog.
*/
spin_lock(&cil->xc_log->l_icloglock);
list_add_tail(&ctx->iclog_entry, &iclog->ic_callbacks);
spin_unlock(&cil->xc_log->l_icloglock);
/*
* Now we can record the commit LSN and wake anyone waiting for this
* sequence to have the ordered commit record assigned to a physical
* location in the log.
*/
spin_lock(&cil->xc_push_lock);
ctx->commit_iclog = iclog;
ctx->commit_lsn = lsn;
wake_up_all(&cil->xc_commit_wait);
spin_unlock(&cil->xc_push_lock);
}
/*
* Ensure that the order of log writes follows checkpoint sequence order. This
* relies on the context LSN being zero until the log write has guaranteed the
* LSN that the log write will start at via xlog_state_get_iclog_space().
*/
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
enum _record_type {
_START_RECORD,
_COMMIT_RECORD,
};
static int
xlog_cil_order_write(
struct xfs_cil *cil,
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
xfs_csn_t sequence,
enum _record_type record)
{
struct xfs_cil_ctx *ctx;
restart:
spin_lock(&cil->xc_push_lock);
list_for_each_entry(ctx, &cil->xc_committing, committing) {
/*
* Avoid getting stuck in this loop because we were woken by the
* shutdown, but then went back to sleep once already in the
* shutdown state.
*/
if (xlog_is_shutdown(cil->xc_log)) {
spin_unlock(&cil->xc_push_lock);
return -EIO;
}
/*
* Higher sequences will wait for this one so skip them.
* Don't wait for our own sequence, either.
*/
if (ctx->sequence >= sequence)
continue;
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
/* Wait until the LSN for the record has been recorded. */
switch (record) {
case _START_RECORD:
if (!ctx->start_lsn) {
xlog_wait(&cil->xc_start_wait, &cil->xc_push_lock);
goto restart;
}
break;
case _COMMIT_RECORD:
if (!ctx->commit_lsn) {
xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
goto restart;
}
break;
}
}
spin_unlock(&cil->xc_push_lock);
return 0;
}
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
/*
* Write out the log vector change now attached to the CIL context. This will
* write a start record that needs to be strictly ordered in ascending CIL
* sequence order so that log recovery will always use in-order start LSNs when
* replaying checkpoints.
*/
static int
xlog_cil_write_chain(
struct xfs_cil_ctx *ctx,
uint32_t chain_len)
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
{
struct xlog *log = ctx->cil->xc_log;
int error;
error = xlog_cil_order_write(ctx->cil, ctx->sequence, _START_RECORD);
if (error)
return error;
return xlog_write(log, ctx, &ctx->lv_chain, ctx->ticket, chain_len);
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
}
/*
* Write out the commit record of a checkpoint transaction to close off a
* running log write. These commit records are strictly ordered in ascending CIL
* sequence order so that log recovery will always replay the checkpoints in the
* correct order.
*/
static int
xlog_cil_write_commit_record(
struct xfs_cil_ctx *ctx)
{
struct xlog *log = ctx->cil->xc_log;
struct xlog_op_header ophdr = {
.oh_clientid = XFS_TRANSACTION,
.oh_tid = cpu_to_be32(ctx->ticket->t_tid),
.oh_flags = XLOG_COMMIT_TRANS,
};
struct xfs_log_iovec reg = {
.i_addr = &ophdr,
.i_len = sizeof(struct xlog_op_header),
.i_type = XLOG_REG_TYPE_COMMIT,
};
struct xfs_log_vec vec = {
.lv_niovecs = 1,
.lv_iovecp = &reg,
};
int error;
LIST_HEAD(lv_chain);
list_add(&vec.lv_list, &lv_chain);
if (xlog_is_shutdown(log))
return -EIO;
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
error = xlog_cil_order_write(ctx->cil, ctx->sequence, _COMMIT_RECORD);
if (error)
return error;
/* account for space used by record data */
ctx->ticket->t_curr_res -= reg.i_len;
error = xlog_write(log, ctx, &lv_chain, ctx->ticket, reg.i_len);
if (error)
xfs: log shutdown triggers should only shut down the log We've got a mess on our hands. 1. xfs_trans_commit() cannot cancel transactions because the mount is shut down - that causes dirty, aborted, unlogged log items to sit unpinned in memory and potentially get written to disk before the log is shut down. Hence xfs_trans_commit() can only abort transactions when xlog_is_shutdown() is true. 2. xfs_force_shutdown() is used in places to cause the current modification to be aborted via xfs_trans_commit() because it may be impractical or impossible to cancel the transaction directly, and hence xfs_trans_commit() must cancel transactions when xfs_is_shutdown() is true in this situation. But we can't do that because of #1. 3. Log IO errors cause log shutdowns by calling xfs_force_shutdown() to shut down the mount and then the log from log IO completion. 4. xfs_force_shutdown() can result in a log force being issued, which has to wait for log IO completion before it will mark the log as shut down. If #3 races with some other shutdown trigger that runs a log force, we rely on xfs_force_shutdown() silently ignoring #3 and avoiding shutting down the log until the failed log force completes. 5. To ensure #2 always works, we have to ensure that xfs_force_shutdown() does not return until the the log is shut down. But in the case of #4, this will result in a deadlock because the log Io completion will block waiting for a log force to complete which is blocked waiting for log IO to complete.... So the very first thing we have to do here to untangle this mess is dissociate log shutdown triggers from mount shutdowns. We already have xlog_forced_shutdown, which will atomically transistion to the log a shutdown state. Due to internal asserts it cannot be called multiple times, but was done simply because the only place that could call it was xfs_do_force_shutdown() (i.e. the mount shutdown!) and that could only call it once and once only. So the first thing we do is remove the asserts. We then convert all the internal log shutdown triggers to call xlog_force_shutdown() directly instead of xfs_force_shutdown(). This allows the log shutdown triggers to shut down the log without needing to care about mount based shutdown constraints. This means we shut down the log independently of the mount and the mount may not notice this until it's next attempt to read or modify metadata. At that point (e.g. xfs_trans_commit()) it will see that the log is shutdown, error out and shutdown the mount. To ensure that all the unmount behaviours and asserts track correctly as a result of a log shutdown, propagate the shutdown up to the mount if it is not already set. This keeps the mount and log state in sync, and saves a huge amount of hassle where code fails because of a log shutdown but only checks for mount shutdowns and hence ends up doing the wrong thing. Cleaning up that mess is an exercise for another day. This enables us to address the other problems noted above in followup patches. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 01:22:01 +00:00
xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
return error;
}
struct xlog_cil_trans_hdr {
struct xlog_op_header oph[2];
struct xfs_trans_header thdr;
struct xfs_log_iovec lhdr[2];
};
/*
* Build a checkpoint transaction header to begin the journal transaction. We
* need to account for the space used by the transaction header here as it is
* not accounted for in xlog_write().
*
* This is the only place we write a transaction header, so we also build the
* log opheaders that indicate the start of a log transaction and wrap the
* transaction header. We keep the start record in it's own log vector rather
* than compacting them into a single region as this ends up making the logic
* in xlog_write() for handling empty opheaders for start, commit and unmount
* records much simpler.
*/
static void
xlog_cil_build_trans_hdr(
struct xfs_cil_ctx *ctx,
struct xlog_cil_trans_hdr *hdr,
struct xfs_log_vec *lvhdr,
int num_iovecs)
{
struct xlog_ticket *tic = ctx->ticket;
__be32 tid = cpu_to_be32(tic->t_tid);
memset(hdr, 0, sizeof(*hdr));
/* Log start record */
hdr->oph[0].oh_tid = tid;
hdr->oph[0].oh_clientid = XFS_TRANSACTION;
hdr->oph[0].oh_flags = XLOG_START_TRANS;
/* log iovec region pointer */
hdr->lhdr[0].i_addr = &hdr->oph[0];
hdr->lhdr[0].i_len = sizeof(struct xlog_op_header);
hdr->lhdr[0].i_type = XLOG_REG_TYPE_LRHEADER;
/* log opheader */
hdr->oph[1].oh_tid = tid;
hdr->oph[1].oh_clientid = XFS_TRANSACTION;
hdr->oph[1].oh_len = cpu_to_be32(sizeof(struct xfs_trans_header));
/* transaction header in host byte order format */
hdr->thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
hdr->thdr.th_type = XFS_TRANS_CHECKPOINT;
hdr->thdr.th_tid = tic->t_tid;
hdr->thdr.th_num_items = num_iovecs;
/* log iovec region pointer */
hdr->lhdr[1].i_addr = &hdr->oph[1];
hdr->lhdr[1].i_len = sizeof(struct xlog_op_header) +
sizeof(struct xfs_trans_header);
hdr->lhdr[1].i_type = XLOG_REG_TYPE_TRANSHDR;
lvhdr->lv_niovecs = 2;
lvhdr->lv_iovecp = &hdr->lhdr[0];
lvhdr->lv_bytes = hdr->lhdr[0].i_len + hdr->lhdr[1].i_len;
tic->t_curr_res -= lvhdr->lv_bytes;
}
/*
* CIL item reordering compare function. We want to order in ascending ID order,
* but we want to leave items with the same ID in the order they were added to
* the list. This is important for operations like reflink where we log 4 order
* dependent intents in a single transaction when we overwrite an existing
* shared extent with a new shared extent. i.e. BUI(unmap), CUI(drop),
* CUI (inc), BUI(remap)...
*/
static int
xlog_cil_order_cmp(
void *priv,
const struct list_head *a,
const struct list_head *b)
{
struct xfs_log_vec *l1 = container_of(a, struct xfs_log_vec, lv_list);
struct xfs_log_vec *l2 = container_of(b, struct xfs_log_vec, lv_list);
return l1->lv_order_id > l2->lv_order_id;
}
/*
* Pull all the log vectors off the items in the CIL, and remove the items from
* the CIL. We don't need the CIL lock here because it's only needed on the
* transaction commit side which is currently locked out by the flush lock.
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
*
* If a log item is marked with a whiteout, we do not need to write it to the
* journal and so we just move them to the whiteout list for the caller to
* dispose of appropriately.
*/
static void
xlog_cil_build_lv_chain(
struct xfs_cil_ctx *ctx,
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
struct list_head *whiteouts,
uint32_t *num_iovecs,
uint32_t *num_bytes)
{
while (!list_empty(&ctx->log_items)) {
struct xfs_log_item *item;
struct xfs_log_vec *lv;
item = list_first_entry(&ctx->log_items,
struct xfs_log_item, li_cil);
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
if (test_bit(XFS_LI_WHITEOUT, &item->li_flags)) {
list_move(&item->li_cil, whiteouts);
trace_xfs_cil_whiteout_skip(item);
continue;
}
lv = item->li_lv;
lv->lv_order_id = item->li_order_id;
/* we don't write ordered log vectors */
if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED)
*num_bytes += lv->lv_bytes;
*num_iovecs += lv->lv_niovecs;
list_add_tail(&lv->lv_list, &ctx->lv_chain);
list_del_init(&item->li_cil);
item->li_order_id = 0;
item->li_lv = NULL;
}
}
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
static void
xlog_cil_cleanup_whiteouts(
struct list_head *whiteouts)
{
while (!list_empty(whiteouts)) {
struct xfs_log_item *item = list_first_entry(whiteouts,
struct xfs_log_item, li_cil);
list_del_init(&item->li_cil);
trace_xfs_cil_whiteout_unpin(item);
item->li_ops->iop_unpin(item, 1);
}
}
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* Push the Committed Item List to the log.
*
* If the current sequence is the same as xc_push_seq we need to do a flush. If
* xc_push_seq is less than the current sequence, then it has already been
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
* flushed and we don't need to do anything - the caller will wait for it to
* complete if necessary.
*
* xc_push_seq is checked unlocked against the sequence number for a match.
* Hence we can allow log forces to run racily and not issue pushes for the
* same sequence twice. If we get a race between multiple pushes for the same
* sequence they will block on the first one and then abort, hence avoiding
* needless pushes.
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
*/
static void
xlog_cil_push_work(
struct work_struct *work)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
{
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
struct xfs_cil_ctx *ctx =
container_of(work, struct xfs_cil_ctx, push_work);
struct xfs_cil *cil = ctx->cil;
struct xlog *log = cil->xc_log;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
struct xfs_cil_ctx *new_ctx;
int num_iovecs = 0;
int num_bytes = 0;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
int error = 0;
struct xlog_cil_trans_hdr thdr;
struct xfs_log_vec lvhdr = {};
xfs: fix ordering violation between cache flushes and tail updates There is a race between the new CIL async data device metadata IO completion cache flush and the log tail in the iclog the flush covers being updated. This can be seen by repeating generic/482 in a loop and eventually log recovery fails with a failures such as this: XFS (dm-3): Starting recovery (logdev: internal) XFS (dm-3): bad inode magic/vsn daddr 228352 #0 (magic=0) XFS (dm-3): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x37c00 xfs_inode_buf_verify XFS (dm-3): Unmount and run xfs_repair XFS (dm-3): First 128 bytes of corrupted metadata buffer: 00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000030: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000040: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000050: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000070: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ XFS (dm-3): metadata I/O error in "xlog_recover_items_pass2+0x55/0xc0" at daddr 0x37c00 len 32 error 117 Analysis of the logwrite replay shows that there were no writes to the data device between the FUA @ write 124 and the FUA at write @ 125, but log recovery @ 125 failed. The difference was the one log write @ 125 moved the tail of the log forwards from (1,8) to (1,32) and so the inode create intent in (1,8) was not replayed and so the inode cluster was zero on disk when replay of the first inode item in (1,32) was attempted. What this meant was that the journal write that occurred at @ 125 did not ensure that metadata completed before the iclog was written was correctly on stable storage. The tail of the log moved forward, so IO must have been completed between the two iclog writes. This means that there is a race condition between the unconditional async cache flush in the CIL push work and the tail LSN that is written to the iclog. This happens like so: CIL push work AIL push work ------------- ------------- Add to committing list start async data dev cache flush ..... <flush completes> <all writes to old tail lsn are stable> xlog_write .... push inode create buffer <start IO> ..... xlog_write(commit record) .... <IO completes> log tail moves xlog_assign_tail_lsn() start_lsn == commit_lsn <no iclog preflush!> xlog_state_release_iclog __xlog_state_release_iclog() <writes *new* tail_lsn into iclog> xlog_sync() .... submit_bio() <tail in log moves forward without flushing written metadata> Essentially, this can only occur if the commit iclog is issued without a cache flush. If the iclog bio is submitted with REQ_PREFLUSH, then it will guarantee that all the completed IO is one stable storage before the iclog bio with the new tail LSN in it is written to the log. IOWs, the tail lsn that is written to the iclog needs to be sampled *before* we issue the cache flush that guarantees all IO up to that LSN has been completed. To fix this without giving up the performance advantage of the flush/FUA optimisations (e.g. g/482 runtime halves with 5.14-rc1 compared to 5.13), we need to ensure that we always issue a cache flush if the tail LSN changes between the initial async flush and the commit record being written. THis requires sampling the tail_lsn before we start the flush, and then passing the sampled tail LSN to xlog_state_release_iclog() so it can determine if the the tail LSN has changed while writing the checkpoint. If the tail LSN has changed, then it needs to set the NEED_FLUSH flag on the iclog and we'll issue another cache flush before writing the iclog. Fixes: eef983ffeae7 ("xfs: journal IO cache flush reductions") Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-07-27 23:23:48 +00:00
xfs_csn_t push_seq;
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
bool push_commit_stable;
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
LIST_HEAD (whiteouts);
xfs: xlog_sync() manually adjusts grant head space When xlog_sync() rounds off the tail the iclog that is being flushed, it manually subtracts that space from the grant heads. This space is actually reserved by the transaction ticket that covers the xlog_sync() call from xlog_write(), but we don't plumb the ticket down far enough for it to account for the space consumed in the current log ticket. The grant heads are hot, so we really should be accounting this to the ticket is we can, rather than adding thousands of extra grant head updates every CIL commit. Interestingly, this actually indicates a potential log space overrun can occur when we force the log. By the time that xfs_log_force() pushes out an active iclog and consumes the roundoff space, the reservation for that roundoff space has been returned to the grant heads and is no longer covered by a reservation. In theory the roundoff added to log force on an already full log could push the write head past the tail. In practice, the CIL commit that writes to the log and needs the iclog pushed will have reserved space for roundoff, so when it releases the ticket there will still be physical space for the roundoff to be committed to the log, even though it is no longer reserved. This roundoff won't be enough space to allow a transaction to be woken if the log is full, so overruns should not actually occur in practice. That said, it indicates that we should not release the CIL context log ticket until after we've released the commit iclog. It also means that xlog_sync() still needs the direct grant head manipulation if we don't provide it with a ticket. Log forces are rare when we are in fast paths running 1.5 million transactions/s that make the grant heads hot, so let's optimise the hot case and pass CIL log tickets down to the xlog_sync() code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:56:09 +00:00
struct xlog_ticket *ticket;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
new_ctx = xlog_cil_ctx_alloc();
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
new_ctx->ticket = xlog_cil_ticket_alloc(log);
down_write(&cil->xc_ctx_lock);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
spin_lock(&cil->xc_push_lock);
push_seq = cil->xc_push_seq;
ASSERT(push_seq <= ctx->sequence);
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
push_commit_stable = cil->xc_push_commit_stable;
cil->xc_push_commit_stable = false;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
xfs: Fix CIL throttle hang when CIL space used going backwards A hang with tasks stuck on the CIL hard throttle was reported and largely diagnosed by Donald Buczek, who discovered that it was a result of the CIL context space usage decrementing in committed transactions once the hard throttle limit had been hit and processes were already blocked. This resulted in the CIL push not waking up those waiters because the CIL context was no longer over the hard throttle limit. The surprising aspect of this was the CIL space usage going backwards regularly enough to trigger this situation. Assumptions had been made in design that the relogging process would only increase the size of the objects in the CIL, and so that space would only increase. This change and commit message fixes the issue and documents the result of an audit of the triggers that can cause the CIL space to go backwards, how large the backwards steps tend to be, the frequency in which they occur, and what the impact on the CIL accounting code is. Even though the CIL ctx->space_used can go backwards, it will only do so if the log item is already logged to the CIL and contains a space reservation for it's entire logged state. This is tracked by the shadow buffer state on the log item. If the item is not previously logged in the CIL it has no shadow buffer nor log vector, and hence the entire size of the logged item copied to the log vector is accounted to the CIL space usage. i.e. it will always go up in this case. If the item has a log vector (i.e. already in the CIL) and the size decreases, then the existing log vector will be overwritten and the space usage will go down. This is the only condition where the space usage reduces, and it can only occur when an item is already tracked in the CIL. Hence we are safe from CIL space usage underruns as a result of log items decreasing in size when they are relogged. Typically this reduction in CIL usage occurs from metadata blocks being free, such as when a btree block merge occurs or a directory enter/xattr entry is removed and the da-tree is reduced in size. This generally results in a reduction in size of around a single block in the CIL, but also tends to increase the number of log vectors because the parent and sibling nodes in the tree needs to be updated when a btree block is removed. If a multi-level merge occurs, then we see reduction in size of 2+ blocks, but again the log vector count goes up. The other vector is inode fork size changes, which only log the current size of the fork and ignore the previously logged size when the fork is relogged. Hence if we are removing items from the inode fork (dir/xattr removal in shortform, extent record removal in extent form, etc) the relogged size of the inode for can decrease. No other log items can decrease in size either because they are a fixed size (e.g. dquots) or they cannot be relogged (e.g. relogging an intent actually creates a new intent log item and doesn't relog the old item at all.) Hence the only two vectors for CIL context size reduction are relogging inode forks and marking buffers active in the CIL as stale. Long story short: the majority of the code does the right thing and handles the reduction in log item size correctly, and only the CIL hard throttle implementation is problematic and needs fixing. This patch makes that fix, as well as adds comments in the log item code that result in items shrinking in size when they are relogged as a clear reminder that this can and does happen frequently. The throttle fix is based upon the change Donald proposed, though it goes further to ensure that once the throttle is activated, it captures all tasks until the CIL push issues a wakeup, regardless of whether the CIL space used has gone back under the throttle threshold. This ensures that we prevent tasks reducing the CIL slightly under the throttle threshold and then making more changes that push it well over the throttle limit. This is acheived by checking if the throttle wait queue is already active as a condition of throttling. Hence once we start throttling, we continue to apply the throttle until the CIL context push wakes everything on the wait queue. We can use waitqueue_active() for the waitqueue manipulations and checks as they are all done under the ctx->xc_push_lock. Hence the waitqueue has external serialisation and we can safely peek inside the wait queue without holding the internal waitqueue locks. Many thanks to Donald for his diagnostic and analysis work to isolate the cause of this hang. Reported-and-tested-by: Donald Buczek <buczek@molgen.mpg.de> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:51 +00:00
* As we are about to switch to a new, empty CIL context, we no longer
* need to throttle tasks on CIL space overruns. Wake any waiters that
* the hard push throttle may have caught so they can start committing
* to the new context. The ctx->xc_push_lock provides the serialisation
* necessary for safely using the lockless waitqueue_active() check in
* this context.
*/
xfs: Fix CIL throttle hang when CIL space used going backwards A hang with tasks stuck on the CIL hard throttle was reported and largely diagnosed by Donald Buczek, who discovered that it was a result of the CIL context space usage decrementing in committed transactions once the hard throttle limit had been hit and processes were already blocked. This resulted in the CIL push not waking up those waiters because the CIL context was no longer over the hard throttle limit. The surprising aspect of this was the CIL space usage going backwards regularly enough to trigger this situation. Assumptions had been made in design that the relogging process would only increase the size of the objects in the CIL, and so that space would only increase. This change and commit message fixes the issue and documents the result of an audit of the triggers that can cause the CIL space to go backwards, how large the backwards steps tend to be, the frequency in which they occur, and what the impact on the CIL accounting code is. Even though the CIL ctx->space_used can go backwards, it will only do so if the log item is already logged to the CIL and contains a space reservation for it's entire logged state. This is tracked by the shadow buffer state on the log item. If the item is not previously logged in the CIL it has no shadow buffer nor log vector, and hence the entire size of the logged item copied to the log vector is accounted to the CIL space usage. i.e. it will always go up in this case. If the item has a log vector (i.e. already in the CIL) and the size decreases, then the existing log vector will be overwritten and the space usage will go down. This is the only condition where the space usage reduces, and it can only occur when an item is already tracked in the CIL. Hence we are safe from CIL space usage underruns as a result of log items decreasing in size when they are relogged. Typically this reduction in CIL usage occurs from metadata blocks being free, such as when a btree block merge occurs or a directory enter/xattr entry is removed and the da-tree is reduced in size. This generally results in a reduction in size of around a single block in the CIL, but also tends to increase the number of log vectors because the parent and sibling nodes in the tree needs to be updated when a btree block is removed. If a multi-level merge occurs, then we see reduction in size of 2+ blocks, but again the log vector count goes up. The other vector is inode fork size changes, which only log the current size of the fork and ignore the previously logged size when the fork is relogged. Hence if we are removing items from the inode fork (dir/xattr removal in shortform, extent record removal in extent form, etc) the relogged size of the inode for can decrease. No other log items can decrease in size either because they are a fixed size (e.g. dquots) or they cannot be relogged (e.g. relogging an intent actually creates a new intent log item and doesn't relog the old item at all.) Hence the only two vectors for CIL context size reduction are relogging inode forks and marking buffers active in the CIL as stale. Long story short: the majority of the code does the right thing and handles the reduction in log item size correctly, and only the CIL hard throttle implementation is problematic and needs fixing. This patch makes that fix, as well as adds comments in the log item code that result in items shrinking in size when they are relogged as a clear reminder that this can and does happen frequently. The throttle fix is based upon the change Donald proposed, though it goes further to ensure that once the throttle is activated, it captures all tasks until the CIL push issues a wakeup, regardless of whether the CIL space used has gone back under the throttle threshold. This ensures that we prevent tasks reducing the CIL slightly under the throttle threshold and then making more changes that push it well over the throttle limit. This is acheived by checking if the throttle wait queue is already active as a condition of throttling. Hence once we start throttling, we continue to apply the throttle until the CIL context push wakes everything on the wait queue. We can use waitqueue_active() for the waitqueue manipulations and checks as they are all done under the ctx->xc_push_lock. Hence the waitqueue has external serialisation and we can safely peek inside the wait queue without holding the internal waitqueue locks. Many thanks to Donald for his diagnostic and analysis work to isolate the cause of this hang. Reported-and-tested-by: Donald Buczek <buczek@molgen.mpg.de> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:51 +00:00
if (waitqueue_active(&cil->xc_push_wait))
wake_up_all(&cil->xc_push_wait);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
xlog_cil_push_pcp_aggregate(cil, ctx);
/*
* Check if we've anything to push. If there is nothing, then we don't
* move on to a new sequence number and so we have to be able to push
* this sequence again later.
*/
if (test_bit(XLOG_CIL_EMPTY, &cil->xc_flags)) {
cil->xc_push_seq = 0;
spin_unlock(&cil->xc_push_lock);
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
goto out_skip;
}
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
/* check for a previously pushed sequence */
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
if (push_seq < ctx->sequence) {
xfs: xlog_cil_force_lsn doesn't always wait correctly When running a tight mount/unmount loop on an older kernel, RedHat QE found that unmount would occasionally hang in xfs_buf_unpin_wait() on the superblock buffer. Tracing and other debug work by Eric Sandeen indicated that it was hanging on the writing of the superblock during unmount immediately after logging the superblock counters in a synchronous transaction. Further debug indicated that the synchronous transaction was not waiting for completion correctly, and we narrowed it down to xlog_cil_force_lsn() returning NULLCOMMITLSN and hence not pushing the transaction in the iclog buffer to disk correctly. While this unmount superblock write code is now very different in mainline kernels, the xlog_cil_force_lsn() code is identical, and it was bisected to the backport of commit f876e44 ("xfs: always do log forces via the workqueue"). This commit made the CIL push asynchronous for log forces and hence exposed a race condition that couldn't occur on a synchronous push. Essentially, the xlog_cil_force_lsn() relied implicitly on the fact that the sequence push would be complete by the time xlog_cil_push_now() returned, resulting in the context being pushed being in the committing list. When it was made asynchronous, it was recognised that there was a race condition in detecting whether an asynchronous push has started or not and code was added to handle it. Unfortunately, the fix was not quite right and left a race condition where it it would detect an empty CIL while a push was in progress before the context had been added to the committing list. This was incorrectly seen as a "nothing to do" condition and so would tell xfs_log_force_lsn() that there is nothing to wait for, and hence it would push the iclogbufs in memory. The fix is simple, but explaining the logic and the race condition is a lot more complex. The fix is to add the context to the committing list before we start emptying the CIL. This allows us to detect the difference between an empty "do nothing" push and a push that has not started by adding a discrete "emptying the CIL" state to avoid the transient, incorrect "empty" condition that the (unchanged) waiting code was seeing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-09-23 05:57:59 +00:00
spin_unlock(&cil->xc_push_lock);
goto out_skip;
xfs: xlog_cil_force_lsn doesn't always wait correctly When running a tight mount/unmount loop on an older kernel, RedHat QE found that unmount would occasionally hang in xfs_buf_unpin_wait() on the superblock buffer. Tracing and other debug work by Eric Sandeen indicated that it was hanging on the writing of the superblock during unmount immediately after logging the superblock counters in a synchronous transaction. Further debug indicated that the synchronous transaction was not waiting for completion correctly, and we narrowed it down to xlog_cil_force_lsn() returning NULLCOMMITLSN and hence not pushing the transaction in the iclog buffer to disk correctly. While this unmount superblock write code is now very different in mainline kernels, the xlog_cil_force_lsn() code is identical, and it was bisected to the backport of commit f876e44 ("xfs: always do log forces via the workqueue"). This commit made the CIL push asynchronous for log forces and hence exposed a race condition that couldn't occur on a synchronous push. Essentially, the xlog_cil_force_lsn() relied implicitly on the fact that the sequence push would be complete by the time xlog_cil_push_now() returned, resulting in the context being pushed being in the committing list. When it was made asynchronous, it was recognised that there was a race condition in detecting whether an asynchronous push has started or not and code was added to handle it. Unfortunately, the fix was not quite right and left a race condition where it it would detect an empty CIL while a push was in progress before the context had been added to the committing list. This was incorrectly seen as a "nothing to do" condition and so would tell xfs_log_force_lsn() that there is nothing to wait for, and hence it would push the iclogbufs in memory. The fix is simple, but explaining the logic and the race condition is a lot more complex. The fix is to add the context to the committing list before we start emptying the CIL. This allows us to detect the difference between an empty "do nothing" push and a push that has not started by adding a discrete "emptying the CIL" state to avoid the transient, incorrect "empty" condition that the (unchanged) waiting code was seeing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-09-23 05:57:59 +00:00
}
/*
* We are now going to push this context, so add it to the committing
* list before we do anything else. This ensures that anyone waiting on
* this push can easily detect the difference between a "push in
* progress" and "CIL is empty, nothing to do".
*
* IOWs, a wait loop can now check for:
* the current sequence not being found on the committing list;
* an empty CIL; and
* an unchanged sequence number
* to detect a push that had nothing to do and therefore does not need
* waiting on. If the CIL is not empty, we get put on the committing
* list before emptying the CIL and bumping the sequence number. Hence
* an empty CIL and an unchanged sequence number means we jumped out
* above after doing nothing.
*
* Hence the waiter will either find the commit sequence on the
* committing list or the sequence number will be unchanged and the CIL
* still dirty. In that latter case, the push has not yet started, and
* so the waiter will have to continue trying to check the CIL
* committing list until it is found. In extreme cases of delay, the
* sequence may fully commit between the attempts the wait makes to wait
* on the commit sequence.
*/
list_add(&ctx->committing, &cil->xc_committing);
spin_unlock(&cil->xc_push_lock);
xlog_cil_build_lv_chain(ctx, &whiteouts, &num_iovecs, &num_bytes);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
* Switch the contexts so we can drop the context lock and move out
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
* of a shared context. We can't just go straight to the commit record,
* though - we need to synchronise with previous and future commits so
* that the commit records are correctly ordered in the log to ensure
* that we process items during log IO completion in the correct order.
*
* For example, if we get an EFI in one checkpoint and the EFD in the
* next (e.g. due to log forces), we do not want the checkpoint with
* the EFD to be committed before the checkpoint with the EFI. Hence
* we must strictly order the commit records of the checkpoints so
* that: a) the checkpoint callbacks are attached to the iclogs in the
* correct order; and b) the checkpoints are replayed in correct order
* in log recovery.
*
* Hence we need to add this context to the committing context list so
* that higher sequences will wait for us to write out a commit record
* before they do.
*
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
* xfs_log_force_seq requires us to mirror the new sequence into the cil
* structure atomically with the addition of this sequence to the
* committing list. This also ensures that we can do unlocked checks
* against the current sequence in log forces without risking
* deferencing a freed context pointer.
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
*/
spin_lock(&cil->xc_push_lock);
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
xlog_cil_ctx_switch(cil, new_ctx);
spin_unlock(&cil->xc_push_lock);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
up_write(&cil->xc_ctx_lock);
/*
* Sort the log vector chain before we add the transaction headers.
* This ensures we always have the transaction headers at the start
* of the chain.
*/
list_sort(NULL, &ctx->lv_chain, xlog_cil_order_cmp);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* Build a checkpoint transaction header and write it to the log to
* begin the transaction. We need to account for the space used by the
* transaction header here as it is not accounted for in xlog_write().
* Add the lvhdr to the head of the lv chain we pass to xlog_write() so
* it gets written into the iclog first.
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
*/
xlog_cil_build_trans_hdr(ctx, &thdr, &lvhdr, num_iovecs);
num_bytes += lvhdr.lv_bytes;
list_add(&lvhdr.lv_list, &ctx->lv_chain);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* Take the lvhdr back off the lv_chain immediately after calling
* xlog_cil_write_chain() as it should not be passed to log IO
* completion.
*/
error = xlog_cil_write_chain(ctx, num_bytes);
list_del(&lvhdr.lv_list);
if (error)
goto out_abort_free_ticket;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
error = xlog_cil_write_commit_record(ctx);
if (error)
goto out_abort_free_ticket;
xfs: xlog_sync() manually adjusts grant head space When xlog_sync() rounds off the tail the iclog that is being flushed, it manually subtracts that space from the grant heads. This space is actually reserved by the transaction ticket that covers the xlog_sync() call from xlog_write(), but we don't plumb the ticket down far enough for it to account for the space consumed in the current log ticket. The grant heads are hot, so we really should be accounting this to the ticket is we can, rather than adding thousands of extra grant head updates every CIL commit. Interestingly, this actually indicates a potential log space overrun can occur when we force the log. By the time that xfs_log_force() pushes out an active iclog and consumes the roundoff space, the reservation for that roundoff space has been returned to the grant heads and is no longer covered by a reservation. In theory the roundoff added to log force on an already full log could push the write head past the tail. In practice, the CIL commit that writes to the log and needs the iclog pushed will have reserved space for roundoff, so when it releases the ticket there will still be physical space for the roundoff to be committed to the log, even though it is no longer reserved. This roundoff won't be enough space to allow a transaction to be woken if the log is full, so overruns should not actually occur in practice. That said, it indicates that we should not release the CIL context log ticket until after we've released the commit iclog. It also means that xlog_sync() still needs the direct grant head manipulation if we don't provide it with a ticket. Log forces are rare when we are in fast paths running 1.5 million transactions/s that make the grant heads hot, so let's optimise the hot case and pass CIL log tickets down to the xlog_sync() code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:56:09 +00:00
/*
* Grab the ticket from the ctx so we can ungrant it after releasing the
* commit_iclog. The ctx may be freed by the time we return from
* releasing the commit_iclog (i.e. checkpoint has been completed and
* callback run) so we can't reference the ctx after the call to
* xlog_state_release_iclog().
*/
ticket = ctx->ticket;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
xfs: separate CIL commit record IO To allow for iclog IO device cache flush behaviour to be optimised, we first need to separate out the commit record iclog IO from the rest of the checkpoint so we can wait for the checkpoint IO to complete before we issue the commit record. This separation is only necessary if the commit record is being written into a different iclog to the start of the checkpoint as the upcoming cache flushing changes requires completion ordering against the other iclogs submitted by the checkpoint. If the entire checkpoint and commit is in the one iclog, then they are both covered by the one set of cache flush primitives on the iclog and hence there is no need to separate them for ordering. Otherwise, we need to wait for all the previous iclogs to complete so they are ordered correctly and made stable by the REQ_PREFLUSH that the commit record iclog IO issues. This guarantees that if a reader sees the commit record in the journal, they will also see the entire checkpoint that commit record closes off. This also provides the guarantee that when the commit record IO completes, we can safely unpin all the log items in the checkpoint so they can be written back because the entire checkpoint is stable in the journal. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:48 +00:00
/*
xfs: don't wait on future iclogs when pushing the CIL The iclogbuf ring attached to the struct xlog is circular, hence the first and last iclogs in the ring can only be determined by comparing them against the log->l_iclog pointer. In xfs_cil_push_work(), we want to wait on previous iclogs that were issued so that we can flush them to stable storage with the commit record write, and it simply waits on the previous iclog in the ring. This, however, leads to CIL push hangs in generic/019 like so: task:kworker/u33:0 state:D stack:12680 pid: 7 ppid: 2 flags:0x00004000 Workqueue: xfs-cil/pmem1 xlog_cil_push_work Call Trace: __schedule+0x30b/0x9f0 schedule+0x68/0xe0 xlog_wait_on_iclog+0x121/0x190 ? wake_up_q+0xa0/0xa0 xlog_cil_push_work+0x994/0xa10 ? _raw_spin_lock+0x15/0x20 ? xfs_swap_extents+0x920/0x920 process_one_work+0x1ab/0x390 worker_thread+0x56/0x3d0 ? rescuer_thread+0x3c0/0x3c0 kthread+0x14d/0x170 ? __kthread_bind_mask+0x70/0x70 ret_from_fork+0x1f/0x30 With other threads blocking in either xlog_state_get_iclog_space() waiting for iclog space or xlog_grant_head_wait() waiting for log reservation space. The problem here is that the previous iclog on the ring might actually be a future iclog. That is, if log->l_iclog points at commit_iclog, commit_iclog is the first (oldest) iclog in the ring and there are no previous iclogs pending as they have all completed their IO and been activated again. IOWs, commit_iclog->ic_prev points to an iclog that will be written in the future, not one that has been written in the past. Hence, in this case, waiting on the ->ic_prev iclog is incorrect behaviour, and depending on the state of the future iclog, we can end up with a circular ABA wait cycle and we hang. The fix is made more complex by the fact that many iclogs states cannot be used to determine if the iclog is a past or future iclog. Hence we have to determine past iclogs by checking the LSN of the iclog rather than their state. A past ACTIVE iclog will have a LSN of zero, while a future ACTIVE iclog will have a LSN greater than the current iclog. We don't wait on either of these cases. Similarly, a future iclog that hasn't completed IO will have an LSN greater than the current iclog and so we don't wait on them. A past iclog that is still undergoing IO completion will have a LSN less than the current iclog and those are the only iclogs that we need to wait on. Hence we can use the iclog LSN to determine what iclogs we need to wait on here. Fixes: 5fd9256ce156 ("xfs: separate CIL commit record IO") Reported-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-25 18:21:02 +00:00
* If the checkpoint spans multiple iclogs, wait for all previous iclogs
* to complete before we submit the commit_iclog. We can't use state
* checks for this - ACTIVE can be either a past completed iclog or a
* future iclog being filled, while WANT_SYNC through SYNC_DONE can be a
* past or future iclog awaiting IO or ordered IO completion to be run.
* In the latter case, if it's a future iclog and we wait on it, the we
* will hang because it won't get processed through to ic_force_wait
* wakeup until this commit_iclog is written to disk. Hence we use the
* iclog header lsn and compare it to the commit lsn to determine if we
* need to wait on iclogs or not.
xfs: separate CIL commit record IO To allow for iclog IO device cache flush behaviour to be optimised, we first need to separate out the commit record iclog IO from the rest of the checkpoint so we can wait for the checkpoint IO to complete before we issue the commit record. This separation is only necessary if the commit record is being written into a different iclog to the start of the checkpoint as the upcoming cache flushing changes requires completion ordering against the other iclogs submitted by the checkpoint. If the entire checkpoint and commit is in the one iclog, then they are both covered by the one set of cache flush primitives on the iclog and hence there is no need to separate them for ordering. Otherwise, we need to wait for all the previous iclogs to complete so they are ordered correctly and made stable by the REQ_PREFLUSH that the commit record iclog IO issues. This guarantees that if a reader sees the commit record in the journal, they will also see the entire checkpoint that commit record closes off. This also provides the guarantee that when the commit record IO completes, we can safely unpin all the log items in the checkpoint so they can be written back because the entire checkpoint is stable in the journal. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:48 +00:00
*/
spin_lock(&log->l_icloglock);
if (ctx->start_lsn != ctx->commit_lsn) {
xfs: don't wait on future iclogs when pushing the CIL The iclogbuf ring attached to the struct xlog is circular, hence the first and last iclogs in the ring can only be determined by comparing them against the log->l_iclog pointer. In xfs_cil_push_work(), we want to wait on previous iclogs that were issued so that we can flush them to stable storage with the commit record write, and it simply waits on the previous iclog in the ring. This, however, leads to CIL push hangs in generic/019 like so: task:kworker/u33:0 state:D stack:12680 pid: 7 ppid: 2 flags:0x00004000 Workqueue: xfs-cil/pmem1 xlog_cil_push_work Call Trace: __schedule+0x30b/0x9f0 schedule+0x68/0xe0 xlog_wait_on_iclog+0x121/0x190 ? wake_up_q+0xa0/0xa0 xlog_cil_push_work+0x994/0xa10 ? _raw_spin_lock+0x15/0x20 ? xfs_swap_extents+0x920/0x920 process_one_work+0x1ab/0x390 worker_thread+0x56/0x3d0 ? rescuer_thread+0x3c0/0x3c0 kthread+0x14d/0x170 ? __kthread_bind_mask+0x70/0x70 ret_from_fork+0x1f/0x30 With other threads blocking in either xlog_state_get_iclog_space() waiting for iclog space or xlog_grant_head_wait() waiting for log reservation space. The problem here is that the previous iclog on the ring might actually be a future iclog. That is, if log->l_iclog points at commit_iclog, commit_iclog is the first (oldest) iclog in the ring and there are no previous iclogs pending as they have all completed their IO and been activated again. IOWs, commit_iclog->ic_prev points to an iclog that will be written in the future, not one that has been written in the past. Hence, in this case, waiting on the ->ic_prev iclog is incorrect behaviour, and depending on the state of the future iclog, we can end up with a circular ABA wait cycle and we hang. The fix is made more complex by the fact that many iclogs states cannot be used to determine if the iclog is a past or future iclog. Hence we have to determine past iclogs by checking the LSN of the iclog rather than their state. A past ACTIVE iclog will have a LSN of zero, while a future ACTIVE iclog will have a LSN greater than the current iclog. We don't wait on either of these cases. Similarly, a future iclog that hasn't completed IO will have an LSN greater than the current iclog and so we don't wait on them. A past iclog that is still undergoing IO completion will have a LSN less than the current iclog and those are the only iclogs that we need to wait on. Hence we can use the iclog LSN to determine what iclogs we need to wait on here. Fixes: 5fd9256ce156 ("xfs: separate CIL commit record IO") Reported-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-25 18:21:02 +00:00
xfs_lsn_t plsn;
plsn = be64_to_cpu(ctx->commit_iclog->ic_prev->ic_header.h_lsn);
if (plsn && XFS_LSN_CMP(plsn, ctx->commit_lsn) < 0) {
xfs: don't wait on future iclogs when pushing the CIL The iclogbuf ring attached to the struct xlog is circular, hence the first and last iclogs in the ring can only be determined by comparing them against the log->l_iclog pointer. In xfs_cil_push_work(), we want to wait on previous iclogs that were issued so that we can flush them to stable storage with the commit record write, and it simply waits on the previous iclog in the ring. This, however, leads to CIL push hangs in generic/019 like so: task:kworker/u33:0 state:D stack:12680 pid: 7 ppid: 2 flags:0x00004000 Workqueue: xfs-cil/pmem1 xlog_cil_push_work Call Trace: __schedule+0x30b/0x9f0 schedule+0x68/0xe0 xlog_wait_on_iclog+0x121/0x190 ? wake_up_q+0xa0/0xa0 xlog_cil_push_work+0x994/0xa10 ? _raw_spin_lock+0x15/0x20 ? xfs_swap_extents+0x920/0x920 process_one_work+0x1ab/0x390 worker_thread+0x56/0x3d0 ? rescuer_thread+0x3c0/0x3c0 kthread+0x14d/0x170 ? __kthread_bind_mask+0x70/0x70 ret_from_fork+0x1f/0x30 With other threads blocking in either xlog_state_get_iclog_space() waiting for iclog space or xlog_grant_head_wait() waiting for log reservation space. The problem here is that the previous iclog on the ring might actually be a future iclog. That is, if log->l_iclog points at commit_iclog, commit_iclog is the first (oldest) iclog in the ring and there are no previous iclogs pending as they have all completed their IO and been activated again. IOWs, commit_iclog->ic_prev points to an iclog that will be written in the future, not one that has been written in the past. Hence, in this case, waiting on the ->ic_prev iclog is incorrect behaviour, and depending on the state of the future iclog, we can end up with a circular ABA wait cycle and we hang. The fix is made more complex by the fact that many iclogs states cannot be used to determine if the iclog is a past or future iclog. Hence we have to determine past iclogs by checking the LSN of the iclog rather than their state. A past ACTIVE iclog will have a LSN of zero, while a future ACTIVE iclog will have a LSN greater than the current iclog. We don't wait on either of these cases. Similarly, a future iclog that hasn't completed IO will have an LSN greater than the current iclog and so we don't wait on them. A past iclog that is still undergoing IO completion will have a LSN less than the current iclog and those are the only iclogs that we need to wait on. Hence we can use the iclog LSN to determine what iclogs we need to wait on here. Fixes: 5fd9256ce156 ("xfs: separate CIL commit record IO") Reported-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-25 18:21:02 +00:00
/*
* Waiting on ic_force_wait orders the completion of
* iclogs older than ic_prev. Hence we only need to wait
* on the most recent older iclog here.
*/
xlog_wait_on_iclog(ctx->commit_iclog->ic_prev);
xfs: don't wait on future iclogs when pushing the CIL The iclogbuf ring attached to the struct xlog is circular, hence the first and last iclogs in the ring can only be determined by comparing them against the log->l_iclog pointer. In xfs_cil_push_work(), we want to wait on previous iclogs that were issued so that we can flush them to stable storage with the commit record write, and it simply waits on the previous iclog in the ring. This, however, leads to CIL push hangs in generic/019 like so: task:kworker/u33:0 state:D stack:12680 pid: 7 ppid: 2 flags:0x00004000 Workqueue: xfs-cil/pmem1 xlog_cil_push_work Call Trace: __schedule+0x30b/0x9f0 schedule+0x68/0xe0 xlog_wait_on_iclog+0x121/0x190 ? wake_up_q+0xa0/0xa0 xlog_cil_push_work+0x994/0xa10 ? _raw_spin_lock+0x15/0x20 ? xfs_swap_extents+0x920/0x920 process_one_work+0x1ab/0x390 worker_thread+0x56/0x3d0 ? rescuer_thread+0x3c0/0x3c0 kthread+0x14d/0x170 ? __kthread_bind_mask+0x70/0x70 ret_from_fork+0x1f/0x30 With other threads blocking in either xlog_state_get_iclog_space() waiting for iclog space or xlog_grant_head_wait() waiting for log reservation space. The problem here is that the previous iclog on the ring might actually be a future iclog. That is, if log->l_iclog points at commit_iclog, commit_iclog is the first (oldest) iclog in the ring and there are no previous iclogs pending as they have all completed their IO and been activated again. IOWs, commit_iclog->ic_prev points to an iclog that will be written in the future, not one that has been written in the past. Hence, in this case, waiting on the ->ic_prev iclog is incorrect behaviour, and depending on the state of the future iclog, we can end up with a circular ABA wait cycle and we hang. The fix is made more complex by the fact that many iclogs states cannot be used to determine if the iclog is a past or future iclog. Hence we have to determine past iclogs by checking the LSN of the iclog rather than their state. A past ACTIVE iclog will have a LSN of zero, while a future ACTIVE iclog will have a LSN greater than the current iclog. We don't wait on either of these cases. Similarly, a future iclog that hasn't completed IO will have an LSN greater than the current iclog and so we don't wait on them. A past iclog that is still undergoing IO completion will have a LSN less than the current iclog and those are the only iclogs that we need to wait on. Hence we can use the iclog LSN to determine what iclogs we need to wait on here. Fixes: 5fd9256ce156 ("xfs: separate CIL commit record IO") Reported-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-25 18:21:02 +00:00
spin_lock(&log->l_icloglock);
}
/*
* We need to issue a pre-flush so that the ordering for this
* checkpoint is correctly preserved down to stable storage.
*/
ctx->commit_iclog->ic_flags |= XLOG_ICL_NEED_FLUSH;
xfs: separate CIL commit record IO To allow for iclog IO device cache flush behaviour to be optimised, we first need to separate out the commit record iclog IO from the rest of the checkpoint so we can wait for the checkpoint IO to complete before we issue the commit record. This separation is only necessary if the commit record is being written into a different iclog to the start of the checkpoint as the upcoming cache flushing changes requires completion ordering against the other iclogs submitted by the checkpoint. If the entire checkpoint and commit is in the one iclog, then they are both covered by the one set of cache flush primitives on the iclog and hence there is no need to separate them for ordering. Otherwise, we need to wait for all the previous iclogs to complete so they are ordered correctly and made stable by the REQ_PREFLUSH that the commit record iclog IO issues. This guarantees that if a reader sees the commit record in the journal, they will also see the entire checkpoint that commit record closes off. This also provides the guarantee that when the commit record IO completes, we can safely unpin all the log items in the checkpoint so they can be written back because the entire checkpoint is stable in the journal. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:48 +00:00
}
xfs: journal IO cache flush reductions Currently every journal IO is issued as REQ_PREFLUSH | REQ_FUA to guarantee the ordering requirements the journal has w.r.t. metadata writeback. THe two ordering constraints are: 1. we cannot overwrite metadata in the journal until we guarantee that the dirty metadata has been written back in place and is stable. 2. we cannot write back dirty metadata until it has been written to the journal and guaranteed to be stable (and hence recoverable) in the journal. The ordering guarantees of #1 are provided by REQ_PREFLUSH. This causes the journal IO to issue a cache flush and wait for it to complete before issuing the write IO to the journal. Hence all completed metadata IO is guaranteed to be stable before the journal overwrites the old metadata. The ordering guarantees of #2 are provided by the REQ_FUA, which ensures the journal writes do not complete until they are on stable storage. Hence by the time the last journal IO in a checkpoint completes, we know that the entire checkpoint is on stable storage and we can unpin the dirty metadata and allow it to be written back. This is the mechanism by which ordering was first implemented in XFS way back in 2002 by commit 95d97c36e5155075ba2eb22b17562cfcc53fcf96 ("Add support for drive write cache flushing") in the xfs-archive tree. A lot has changed since then, most notably we now use delayed logging to checkpoint the filesystem to the journal rather than write each individual transaction to the journal. Cache flushes on journal IO are necessary when individual transactions are wholly contained within a single iclog. However, CIL checkpoints are single transactions that typically span hundreds to thousands of individual journal writes, and so the requirements for device cache flushing have changed. That is, the ordering rules I state above apply to ordering of atomic transactions recorded in the journal, not to the journal IO itself. Hence we need to ensure metadata is stable before we start writing a new transaction to the journal (guarantee #1), and we need to ensure the entire transaction is stable in the journal before we start metadata writeback (guarantee #2). Hence we only need a REQ_PREFLUSH on the journal IO that starts a new journal transaction to provide #1, and it is not on any other journal IO done within the context of that journal transaction. The CIL checkpoint already issues a cache flush before it starts writing to the log, so we no longer need the iclog IO to issue a REQ_REFLUSH for us. Hence if XLOG_START_TRANS is passed to xlog_write(), we no longer need to mark the first iclog in the log write with REQ_PREFLUSH for this case. As an added bonus, this ordering mechanism works for both internal and external logs, meaning we can remove the explicit data device cache flushes from the iclog write code when using external logs. Given the new ordering semantics of commit records for the CIL, we need iclogs containing commit records to issue a REQ_PREFLUSH. We also require unmount records to do this. Hence for both XLOG_COMMIT_TRANS and XLOG_UNMOUNT_TRANS xlog_write() calls we need to mark the first iclog being written with REQ_PREFLUSH. For both commit records and unmount records, we also want them immediately on stable storage, so we want to also mark the iclogs that contain these records to be marked REQ_FUA. That means if a record is split across multiple iclogs, they are all marked REQ_FUA and not just the last one so that when the transaction is completed all the parts of the record are on stable storage. And for external logs, unmount records need a pre-write data device cache flush similar to the CIL checkpoint cache pre-flush as the internal iclog write code does not do this implicitly anymore. As an optimisation, when the commit record lands in the same iclog as the journal transaction starts, we don't need to wait for anything and can simply use REQ_FUA to provide guarantee #2. This means that for fsync() heavy workloads, the cache flush behaviour is completely unchanged and there is no degradation in performance as a result of optimise the multi-IO transaction case. The most notable sign that there is less IO latency on my test machine (nvme SSDs) is that the "noiclogs" rate has dropped substantially. This metric indicates that the CIL push is blocking in xlog_get_iclog_space() waiting for iclog IO completion to occur. With 8 iclogs of 256kB, the rate is appoximately 1 noiclog event to every 4 iclog writes. IOWs, every 4th call to xlog_get_iclog_space() is blocking waiting for log IO. With the changes in this patch, this drops to 1 noiclog event for every 100 iclog writes. Hence it is clear that log IO is completing much faster than it was previously, but it is also clear that for large iclog sizes, this isn't the performance limiting factor on this hardware. With smaller iclogs (32kB), however, there is a substantial difference. With the cache flush modifications, the journal is now running at over 4000 write IOPS, and the journal throughput is largely identical to the 256kB iclogs and the noiclog event rate stays low at about 1:50 iclog writes. The existing code tops out at about 2500 IOPS as the number of cache flushes dominate performance and latency. The noiclog event rate is about 1:4, and the performance variance is quite large as the journal throughput can fall to less than half the peak sustained rate when the cache flush rate prevents metadata writeback from keeping up and the log runs out of space and throttles reservations. As a result: logbsize fsmark create rate rm -rf before 32kb 152851+/-5.3e+04 5m28s patched 32kb 221533+/-1.1e+04 5m24s before 256kb 220239+/-6.2e+03 4m58s patched 256kb 228286+/-9.2e+03 5m06s The rm -rf times are included because I ran them, but the differences are largely noise. This workload is largely metadata read IO latency bound and the changes to the journal cache flushing doesn't really make any noticable difference to behaviour apart from a reduction in noiclog events from background CIL pushing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:51 +00:00
/*
* The commit iclog must be written to stable storage to guarantee
* journal IO vs metadata writeback IO is correctly ordered on stable
* storage.
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
*
* If the push caller needs the commit to be immediately stable and the
* commit_iclog is not yet marked as XLOG_STATE_WANT_SYNC to indicate it
* will be written when released, switch it's state to WANT_SYNC right
* now.
xfs: journal IO cache flush reductions Currently every journal IO is issued as REQ_PREFLUSH | REQ_FUA to guarantee the ordering requirements the journal has w.r.t. metadata writeback. THe two ordering constraints are: 1. we cannot overwrite metadata in the journal until we guarantee that the dirty metadata has been written back in place and is stable. 2. we cannot write back dirty metadata until it has been written to the journal and guaranteed to be stable (and hence recoverable) in the journal. The ordering guarantees of #1 are provided by REQ_PREFLUSH. This causes the journal IO to issue a cache flush and wait for it to complete before issuing the write IO to the journal. Hence all completed metadata IO is guaranteed to be stable before the journal overwrites the old metadata. The ordering guarantees of #2 are provided by the REQ_FUA, which ensures the journal writes do not complete until they are on stable storage. Hence by the time the last journal IO in a checkpoint completes, we know that the entire checkpoint is on stable storage and we can unpin the dirty metadata and allow it to be written back. This is the mechanism by which ordering was first implemented in XFS way back in 2002 by commit 95d97c36e5155075ba2eb22b17562cfcc53fcf96 ("Add support for drive write cache flushing") in the xfs-archive tree. A lot has changed since then, most notably we now use delayed logging to checkpoint the filesystem to the journal rather than write each individual transaction to the journal. Cache flushes on journal IO are necessary when individual transactions are wholly contained within a single iclog. However, CIL checkpoints are single transactions that typically span hundreds to thousands of individual journal writes, and so the requirements for device cache flushing have changed. That is, the ordering rules I state above apply to ordering of atomic transactions recorded in the journal, not to the journal IO itself. Hence we need to ensure metadata is stable before we start writing a new transaction to the journal (guarantee #1), and we need to ensure the entire transaction is stable in the journal before we start metadata writeback (guarantee #2). Hence we only need a REQ_PREFLUSH on the journal IO that starts a new journal transaction to provide #1, and it is not on any other journal IO done within the context of that journal transaction. The CIL checkpoint already issues a cache flush before it starts writing to the log, so we no longer need the iclog IO to issue a REQ_REFLUSH for us. Hence if XLOG_START_TRANS is passed to xlog_write(), we no longer need to mark the first iclog in the log write with REQ_PREFLUSH for this case. As an added bonus, this ordering mechanism works for both internal and external logs, meaning we can remove the explicit data device cache flushes from the iclog write code when using external logs. Given the new ordering semantics of commit records for the CIL, we need iclogs containing commit records to issue a REQ_PREFLUSH. We also require unmount records to do this. Hence for both XLOG_COMMIT_TRANS and XLOG_UNMOUNT_TRANS xlog_write() calls we need to mark the first iclog being written with REQ_PREFLUSH. For both commit records and unmount records, we also want them immediately on stable storage, so we want to also mark the iclogs that contain these records to be marked REQ_FUA. That means if a record is split across multiple iclogs, they are all marked REQ_FUA and not just the last one so that when the transaction is completed all the parts of the record are on stable storage. And for external logs, unmount records need a pre-write data device cache flush similar to the CIL checkpoint cache pre-flush as the internal iclog write code does not do this implicitly anymore. As an optimisation, when the commit record lands in the same iclog as the journal transaction starts, we don't need to wait for anything and can simply use REQ_FUA to provide guarantee #2. This means that for fsync() heavy workloads, the cache flush behaviour is completely unchanged and there is no degradation in performance as a result of optimise the multi-IO transaction case. The most notable sign that there is less IO latency on my test machine (nvme SSDs) is that the "noiclogs" rate has dropped substantially. This metric indicates that the CIL push is blocking in xlog_get_iclog_space() waiting for iclog IO completion to occur. With 8 iclogs of 256kB, the rate is appoximately 1 noiclog event to every 4 iclog writes. IOWs, every 4th call to xlog_get_iclog_space() is blocking waiting for log IO. With the changes in this patch, this drops to 1 noiclog event for every 100 iclog writes. Hence it is clear that log IO is completing much faster than it was previously, but it is also clear that for large iclog sizes, this isn't the performance limiting factor on this hardware. With smaller iclogs (32kB), however, there is a substantial difference. With the cache flush modifications, the journal is now running at over 4000 write IOPS, and the journal throughput is largely identical to the 256kB iclogs and the noiclog event rate stays low at about 1:50 iclog writes. The existing code tops out at about 2500 IOPS as the number of cache flushes dominate performance and latency. The noiclog event rate is about 1:4, and the performance variance is quite large as the journal throughput can fall to less than half the peak sustained rate when the cache flush rate prevents metadata writeback from keeping up and the log runs out of space and throttles reservations. As a result: logbsize fsmark create rate rm -rf before 32kb 152851+/-5.3e+04 5m28s patched 32kb 221533+/-1.1e+04 5m24s before 256kb 220239+/-6.2e+03 4m58s patched 256kb 228286+/-9.2e+03 5m06s The rm -rf times are included because I ran them, but the differences are largely noise. This workload is largely metadata read IO latency bound and the changes to the journal cache flushing doesn't really make any noticable difference to behaviour apart from a reduction in noiclog events from background CIL pushing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:51 +00:00
*/
ctx->commit_iclog->ic_flags |= XLOG_ICL_NEED_FUA;
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
if (push_commit_stable &&
ctx->commit_iclog->ic_state == XLOG_STATE_ACTIVE)
xlog_state_switch_iclogs(log, ctx->commit_iclog, 0);
xfs: xlog_sync() manually adjusts grant head space When xlog_sync() rounds off the tail the iclog that is being flushed, it manually subtracts that space from the grant heads. This space is actually reserved by the transaction ticket that covers the xlog_sync() call from xlog_write(), but we don't plumb the ticket down far enough for it to account for the space consumed in the current log ticket. The grant heads are hot, so we really should be accounting this to the ticket is we can, rather than adding thousands of extra grant head updates every CIL commit. Interestingly, this actually indicates a potential log space overrun can occur when we force the log. By the time that xfs_log_force() pushes out an active iclog and consumes the roundoff space, the reservation for that roundoff space has been returned to the grant heads and is no longer covered by a reservation. In theory the roundoff added to log force on an already full log could push the write head past the tail. In practice, the CIL commit that writes to the log and needs the iclog pushed will have reserved space for roundoff, so when it releases the ticket there will still be physical space for the roundoff to be committed to the log, even though it is no longer reserved. This roundoff won't be enough space to allow a transaction to be woken if the log is full, so overruns should not actually occur in practice. That said, it indicates that we should not release the CIL context log ticket until after we've released the commit iclog. It also means that xlog_sync() still needs the direct grant head manipulation if we don't provide it with a ticket. Log forces are rare when we are in fast paths running 1.5 million transactions/s that make the grant heads hot, so let's optimise the hot case and pass CIL log tickets down to the xlog_sync() code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:56:09 +00:00
ticket = ctx->ticket;
xlog_state_release_iclog(log, ctx->commit_iclog, ticket);
xfs: don't run shutdown callbacks on active iclogs When the log is shutdown, it currently walks all the iclogs and runs callbacks that are attached to the iclogs, regardless of whether the iclog is queued for IO completion or not. This creates a problem for contexts attaching callbacks to iclogs in that a racing shutdown can run the callbacks even before the attaching context has finished processing the iclog and releasing it for IO submission. If the callback processing of the iclog frees the structure that is attached to the iclog, then this leads to an UAF scenario that can only be protected against by holding the icloglock from the point callbacks are attached through to the release of the iclog. While we currently do this, it is not practical or sustainable. Hence we need to make shutdown processing the responsibility of the context that holds active references to the iclog. We know that the contexts attaching callbacks to the iclog must have active references to the iclog, and that means they must be in either ACTIVE or WANT_SYNC states. xlog_state_do_callback() will skip over iclogs in these states -except- when the log is shut down. xlog_state_do_callback() checks the state of the iclogs while holding the icloglock, therefore the reference count/state change that occurs in xlog_state_release_iclog() after the callbacks are atomic w.r.t. shutdown processing. We can't push the responsibility of callback cleanup onto the CIL context because we can have ACTIVE iclogs that have callbacks attached that have already been released. Hence we really need to internalise the cleanup of callbacks into xlog_state_release_iclog() processing. Indeed, we already have that internalisation via: xlog_state_release_iclog drop last reference ->SYNCING xlog_sync xlog_write_iclog if (log_is_shutdown) xlog_state_done_syncing() xlog_state_do_callback() <process shutdown on iclog that is now in SYNCING state> The problem is that xlog_state_release_iclog() aborts before doing anything if the log is already shut down. It assumes that the callbacks have already been cleaned up, and it doesn't need to do any cleanup. Hence the fix is to remove the xlog_is_shutdown() check from xlog_state_release_iclog() so that reference counts are correctly released from the iclogs, and when the reference count is zero we always transition to SYNCING if the log is shut down. Hence we'll always enter the xlog_sync() path in a shutdown and eventually end up erroring out the iclog IO and running xlog_state_do_callback() to process the callbacks attached to the iclog. This allows us to stop processing referenced ACTIVE/WANT_SYNC iclogs directly in the shutdown code, and in doing so gets rid of the UAF vector that currently exists. This then decouples the adding of callbacks to the iclogs from xlog_state_release_iclog() as we guarantee that xlog_state_release_iclog() will process the callbacks if the log has been shut down before xlog_state_release_iclog() has been called. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:41 +00:00
/* Not safe to reference ctx now! */
xfs: journal IO cache flush reductions Currently every journal IO is issued as REQ_PREFLUSH | REQ_FUA to guarantee the ordering requirements the journal has w.r.t. metadata writeback. THe two ordering constraints are: 1. we cannot overwrite metadata in the journal until we guarantee that the dirty metadata has been written back in place and is stable. 2. we cannot write back dirty metadata until it has been written to the journal and guaranteed to be stable (and hence recoverable) in the journal. The ordering guarantees of #1 are provided by REQ_PREFLUSH. This causes the journal IO to issue a cache flush and wait for it to complete before issuing the write IO to the journal. Hence all completed metadata IO is guaranteed to be stable before the journal overwrites the old metadata. The ordering guarantees of #2 are provided by the REQ_FUA, which ensures the journal writes do not complete until they are on stable storage. Hence by the time the last journal IO in a checkpoint completes, we know that the entire checkpoint is on stable storage and we can unpin the dirty metadata and allow it to be written back. This is the mechanism by which ordering was first implemented in XFS way back in 2002 by commit 95d97c36e5155075ba2eb22b17562cfcc53fcf96 ("Add support for drive write cache flushing") in the xfs-archive tree. A lot has changed since then, most notably we now use delayed logging to checkpoint the filesystem to the journal rather than write each individual transaction to the journal. Cache flushes on journal IO are necessary when individual transactions are wholly contained within a single iclog. However, CIL checkpoints are single transactions that typically span hundreds to thousands of individual journal writes, and so the requirements for device cache flushing have changed. That is, the ordering rules I state above apply to ordering of atomic transactions recorded in the journal, not to the journal IO itself. Hence we need to ensure metadata is stable before we start writing a new transaction to the journal (guarantee #1), and we need to ensure the entire transaction is stable in the journal before we start metadata writeback (guarantee #2). Hence we only need a REQ_PREFLUSH on the journal IO that starts a new journal transaction to provide #1, and it is not on any other journal IO done within the context of that journal transaction. The CIL checkpoint already issues a cache flush before it starts writing to the log, so we no longer need the iclog IO to issue a REQ_REFLUSH for us. Hence if XLOG_START_TRANS is passed to xlog_write(), we no longer need to mark the first iclog in the log write with REQ_PREFLUSH for this case. As an added bonus, this ordering mechanism works for both internal and external logs, meaning we can remove the explicit data device cache flushes from the iclog write code when using external logs. Given the new ordering semantics of commit records for the CIL, we need iclogs containing commit records to issue a REQ_PREFLUSH. We also require unmount records to do this. Hence for both XLOG_COMMIT_TRANS and XLOG_UNMOUNT_TRANS xlog_write() calls we need to mark the first iclog being written with REQ_PREFLUSH. For both commit records and unmount records, we also want them immediately on stable storage, so we want to also mark the iclogs that contain these records to be marked REQ_FUA. That means if a record is split across multiple iclogs, they are all marked REQ_FUA and not just the last one so that when the transaction is completed all the parts of the record are on stable storage. And for external logs, unmount records need a pre-write data device cache flush similar to the CIL checkpoint cache pre-flush as the internal iclog write code does not do this implicitly anymore. As an optimisation, when the commit record lands in the same iclog as the journal transaction starts, we don't need to wait for anything and can simply use REQ_FUA to provide guarantee #2. This means that for fsync() heavy workloads, the cache flush behaviour is completely unchanged and there is no degradation in performance as a result of optimise the multi-IO transaction case. The most notable sign that there is less IO latency on my test machine (nvme SSDs) is that the "noiclogs" rate has dropped substantially. This metric indicates that the CIL push is blocking in xlog_get_iclog_space() waiting for iclog IO completion to occur. With 8 iclogs of 256kB, the rate is appoximately 1 noiclog event to every 4 iclog writes. IOWs, every 4th call to xlog_get_iclog_space() is blocking waiting for log IO. With the changes in this patch, this drops to 1 noiclog event for every 100 iclog writes. Hence it is clear that log IO is completing much faster than it was previously, but it is also clear that for large iclog sizes, this isn't the performance limiting factor on this hardware. With smaller iclogs (32kB), however, there is a substantial difference. With the cache flush modifications, the journal is now running at over 4000 write IOPS, and the journal throughput is largely identical to the 256kB iclogs and the noiclog event rate stays low at about 1:50 iclog writes. The existing code tops out at about 2500 IOPS as the number of cache flushes dominate performance and latency. The noiclog event rate is about 1:4, and the performance variance is quite large as the journal throughput can fall to less than half the peak sustained rate when the cache flush rate prevents metadata writeback from keeping up and the log runs out of space and throttles reservations. As a result: logbsize fsmark create rate rm -rf before 32kb 152851+/-5.3e+04 5m28s patched 32kb 221533+/-1.1e+04 5m24s before 256kb 220239+/-6.2e+03 4m58s patched 256kb 228286+/-9.2e+03 5m06s The rm -rf times are included because I ran them, but the differences are largely noise. This workload is largely metadata read IO latency bound and the changes to the journal cache flushing doesn't really make any noticable difference to behaviour apart from a reduction in noiclog events from background CIL pushing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:51 +00:00
spin_unlock(&log->l_icloglock);
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
xlog_cil_cleanup_whiteouts(&whiteouts);
xfs: xlog_sync() manually adjusts grant head space When xlog_sync() rounds off the tail the iclog that is being flushed, it manually subtracts that space from the grant heads. This space is actually reserved by the transaction ticket that covers the xlog_sync() call from xlog_write(), but we don't plumb the ticket down far enough for it to account for the space consumed in the current log ticket. The grant heads are hot, so we really should be accounting this to the ticket is we can, rather than adding thousands of extra grant head updates every CIL commit. Interestingly, this actually indicates a potential log space overrun can occur when we force the log. By the time that xfs_log_force() pushes out an active iclog and consumes the roundoff space, the reservation for that roundoff space has been returned to the grant heads and is no longer covered by a reservation. In theory the roundoff added to log force on an already full log could push the write head past the tail. In practice, the CIL commit that writes to the log and needs the iclog pushed will have reserved space for roundoff, so when it releases the ticket there will still be physical space for the roundoff to be committed to the log, even though it is no longer reserved. This roundoff won't be enough space to allow a transaction to be woken if the log is full, so overruns should not actually occur in practice. That said, it indicates that we should not release the CIL context log ticket until after we've released the commit iclog. It also means that xlog_sync() still needs the direct grant head manipulation if we don't provide it with a ticket. Log forces are rare when we are in fast paths running 1.5 million transactions/s that make the grant heads hot, so let's optimise the hot case and pass CIL log tickets down to the xlog_sync() code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:56:09 +00:00
xfs_log_ticket_ungrant(log, ticket);
return;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
out_skip:
up_write(&cil->xc_ctx_lock);
xfs_log_ticket_put(new_ctx->ticket);
kmem_free(new_ctx);
return;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
out_abort_free_ticket:
ASSERT(xlog_is_shutdown(log));
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
xlog_cil_cleanup_whiteouts(&whiteouts);
if (!ctx->commit_iclog) {
xfs: xlog_sync() manually adjusts grant head space When xlog_sync() rounds off the tail the iclog that is being flushed, it manually subtracts that space from the grant heads. This space is actually reserved by the transaction ticket that covers the xlog_sync() call from xlog_write(), but we don't plumb the ticket down far enough for it to account for the space consumed in the current log ticket. The grant heads are hot, so we really should be accounting this to the ticket is we can, rather than adding thousands of extra grant head updates every CIL commit. Interestingly, this actually indicates a potential log space overrun can occur when we force the log. By the time that xfs_log_force() pushes out an active iclog and consumes the roundoff space, the reservation for that roundoff space has been returned to the grant heads and is no longer covered by a reservation. In theory the roundoff added to log force on an already full log could push the write head past the tail. In practice, the CIL commit that writes to the log and needs the iclog pushed will have reserved space for roundoff, so when it releases the ticket there will still be physical space for the roundoff to be committed to the log, even though it is no longer reserved. This roundoff won't be enough space to allow a transaction to be woken if the log is full, so overruns should not actually occur in practice. That said, it indicates that we should not release the CIL context log ticket until after we've released the commit iclog. It also means that xlog_sync() still needs the direct grant head manipulation if we don't provide it with a ticket. Log forces are rare when we are in fast paths running 1.5 million transactions/s that make the grant heads hot, so let's optimise the hot case and pass CIL log tickets down to the xlog_sync() code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:56:09 +00:00
xfs_log_ticket_ungrant(log, ctx->ticket);
xlog_cil_committed(ctx);
return;
}
spin_lock(&log->l_icloglock);
xfs: xlog_sync() manually adjusts grant head space When xlog_sync() rounds off the tail the iclog that is being flushed, it manually subtracts that space from the grant heads. This space is actually reserved by the transaction ticket that covers the xlog_sync() call from xlog_write(), but we don't plumb the ticket down far enough for it to account for the space consumed in the current log ticket. The grant heads are hot, so we really should be accounting this to the ticket is we can, rather than adding thousands of extra grant head updates every CIL commit. Interestingly, this actually indicates a potential log space overrun can occur when we force the log. By the time that xfs_log_force() pushes out an active iclog and consumes the roundoff space, the reservation for that roundoff space has been returned to the grant heads and is no longer covered by a reservation. In theory the roundoff added to log force on an already full log could push the write head past the tail. In practice, the CIL commit that writes to the log and needs the iclog pushed will have reserved space for roundoff, so when it releases the ticket there will still be physical space for the roundoff to be committed to the log, even though it is no longer reserved. This roundoff won't be enough space to allow a transaction to be woken if the log is full, so overruns should not actually occur in practice. That said, it indicates that we should not release the CIL context log ticket until after we've released the commit iclog. It also means that xlog_sync() still needs the direct grant head manipulation if we don't provide it with a ticket. Log forces are rare when we are in fast paths running 1.5 million transactions/s that make the grant heads hot, so let's optimise the hot case and pass CIL log tickets down to the xlog_sync() code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:56:09 +00:00
ticket = ctx->ticket;
xlog_state_release_iclog(log, ctx->commit_iclog, ticket);
/* Not safe to reference ctx now! */
spin_unlock(&log->l_icloglock);
xfs: xlog_sync() manually adjusts grant head space When xlog_sync() rounds off the tail the iclog that is being flushed, it manually subtracts that space from the grant heads. This space is actually reserved by the transaction ticket that covers the xlog_sync() call from xlog_write(), but we don't plumb the ticket down far enough for it to account for the space consumed in the current log ticket. The grant heads are hot, so we really should be accounting this to the ticket is we can, rather than adding thousands of extra grant head updates every CIL commit. Interestingly, this actually indicates a potential log space overrun can occur when we force the log. By the time that xfs_log_force() pushes out an active iclog and consumes the roundoff space, the reservation for that roundoff space has been returned to the grant heads and is no longer covered by a reservation. In theory the roundoff added to log force on an already full log could push the write head past the tail. In practice, the CIL commit that writes to the log and needs the iclog pushed will have reserved space for roundoff, so when it releases the ticket there will still be physical space for the roundoff to be committed to the log, even though it is no longer reserved. This roundoff won't be enough space to allow a transaction to be woken if the log is full, so overruns should not actually occur in practice. That said, it indicates that we should not release the CIL context log ticket until after we've released the commit iclog. It also means that xlog_sync() still needs the direct grant head manipulation if we don't provide it with a ticket. Log forces are rare when we are in fast paths running 1.5 million transactions/s that make the grant heads hot, so let's optimise the hot case and pass CIL log tickets down to the xlog_sync() code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:56:09 +00:00
xfs_log_ticket_ungrant(log, ticket);
}
/*
* We need to push CIL every so often so we don't cache more than we can fit in
* the log. The limit really is that a checkpoint can't be more than half the
* log (the current checkpoint is not allowed to overwrite the previous
* checkpoint), but commit latency and memory usage limit this to a smaller
* size.
*/
static void
xlog_cil_push_background(
struct xlog *log) __releases(cil->xc_ctx_lock)
{
struct xfs_cil *cil = log->l_cilp;
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
int space_used = atomic_read(&cil->xc_ctx->space_used);
/*
* The cil won't be empty because we are called while holding the
* context lock so whatever we added to the CIL will still be there.
*/
ASSERT(!test_bit(XLOG_CIL_EMPTY, &cil->xc_flags));
/*
* We are done if:
* - we haven't used up all the space available yet; or
* - we've already queued up a push; and
* - we're not over the hard limit; and
* - nothing has been over the hard limit.
*
* If so, we don't need to take the push lock as there's nothing to do.
*/
if (space_used < XLOG_CIL_SPACE_LIMIT(log) ||
(cil->xc_push_seq == cil->xc_current_sequence &&
space_used < XLOG_CIL_BLOCKING_SPACE_LIMIT(log) &&
!waitqueue_active(&cil->xc_push_wait))) {
up_read(&cil->xc_ctx_lock);
return;
}
spin_lock(&cil->xc_push_lock);
if (cil->xc_push_seq < cil->xc_current_sequence) {
cil->xc_push_seq = cil->xc_current_sequence;
queue_work(cil->xc_push_wq, &cil->xc_ctx->push_work);
}
/*
* Drop the context lock now, we can't hold that if we need to sleep
* because we are over the blocking threshold. The push_lock is still
* held, so blocking threshold sleep/wakeup is still correctly
* serialised here.
*/
up_read(&cil->xc_ctx_lock);
/*
* If we are well over the space limit, throttle the work that is being
xfs: Fix CIL throttle hang when CIL space used going backwards A hang with tasks stuck on the CIL hard throttle was reported and largely diagnosed by Donald Buczek, who discovered that it was a result of the CIL context space usage decrementing in committed transactions once the hard throttle limit had been hit and processes were already blocked. This resulted in the CIL push not waking up those waiters because the CIL context was no longer over the hard throttle limit. The surprising aspect of this was the CIL space usage going backwards regularly enough to trigger this situation. Assumptions had been made in design that the relogging process would only increase the size of the objects in the CIL, and so that space would only increase. This change and commit message fixes the issue and documents the result of an audit of the triggers that can cause the CIL space to go backwards, how large the backwards steps tend to be, the frequency in which they occur, and what the impact on the CIL accounting code is. Even though the CIL ctx->space_used can go backwards, it will only do so if the log item is already logged to the CIL and contains a space reservation for it's entire logged state. This is tracked by the shadow buffer state on the log item. If the item is not previously logged in the CIL it has no shadow buffer nor log vector, and hence the entire size of the logged item copied to the log vector is accounted to the CIL space usage. i.e. it will always go up in this case. If the item has a log vector (i.e. already in the CIL) and the size decreases, then the existing log vector will be overwritten and the space usage will go down. This is the only condition where the space usage reduces, and it can only occur when an item is already tracked in the CIL. Hence we are safe from CIL space usage underruns as a result of log items decreasing in size when they are relogged. Typically this reduction in CIL usage occurs from metadata blocks being free, such as when a btree block merge occurs or a directory enter/xattr entry is removed and the da-tree is reduced in size. This generally results in a reduction in size of around a single block in the CIL, but also tends to increase the number of log vectors because the parent and sibling nodes in the tree needs to be updated when a btree block is removed. If a multi-level merge occurs, then we see reduction in size of 2+ blocks, but again the log vector count goes up. The other vector is inode fork size changes, which only log the current size of the fork and ignore the previously logged size when the fork is relogged. Hence if we are removing items from the inode fork (dir/xattr removal in shortform, extent record removal in extent form, etc) the relogged size of the inode for can decrease. No other log items can decrease in size either because they are a fixed size (e.g. dquots) or they cannot be relogged (e.g. relogging an intent actually creates a new intent log item and doesn't relog the old item at all.) Hence the only two vectors for CIL context size reduction are relogging inode forks and marking buffers active in the CIL as stale. Long story short: the majority of the code does the right thing and handles the reduction in log item size correctly, and only the CIL hard throttle implementation is problematic and needs fixing. This patch makes that fix, as well as adds comments in the log item code that result in items shrinking in size when they are relogged as a clear reminder that this can and does happen frequently. The throttle fix is based upon the change Donald proposed, though it goes further to ensure that once the throttle is activated, it captures all tasks until the CIL push issues a wakeup, regardless of whether the CIL space used has gone back under the throttle threshold. This ensures that we prevent tasks reducing the CIL slightly under the throttle threshold and then making more changes that push it well over the throttle limit. This is acheived by checking if the throttle wait queue is already active as a condition of throttling. Hence once we start throttling, we continue to apply the throttle until the CIL context push wakes everything on the wait queue. We can use waitqueue_active() for the waitqueue manipulations and checks as they are all done under the ctx->xc_push_lock. Hence the waitqueue has external serialisation and we can safely peek inside the wait queue without holding the internal waitqueue locks. Many thanks to Donald for his diagnostic and analysis work to isolate the cause of this hang. Reported-and-tested-by: Donald Buczek <buczek@molgen.mpg.de> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:51 +00:00
* done until the push work on this context has begun. Enforce the hard
* throttle on all transaction commits once it has been activated, even
* if the committing transactions have resulted in the space usage
* dipping back down under the hard limit.
*
* The ctx->xc_push_lock provides the serialisation necessary for safely
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
* calling xlog_cil_over_hard_limit() in this context.
*/
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
if (xlog_cil_over_hard_limit(log, space_used)) {
trace_xfs_log_cil_wait(log, cil->xc_ctx->ticket);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
ASSERT(space_used < log->l_logsize);
xlog_wait(&cil->xc_push_wait, &cil->xc_push_lock);
return;
}
spin_unlock(&cil->xc_push_lock);
}
/*
* xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
* number that is passed. When it returns, the work will be queued for
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
* @push_seq, but it won't be completed.
*
* If the caller is performing a synchronous force, we will flush the workqueue
* to get previously queued work moving to minimise the wait time they will
* undergo waiting for all outstanding pushes to complete. The caller is
* expected to do the required waiting for push_seq to complete.
*
* If the caller is performing an async push, we need to ensure that the
* checkpoint is fully flushed out of the iclogs when we finish the push. If we
* don't do this, then the commit record may remain sitting in memory in an
* ACTIVE iclog. This then requires another full log force to push to disk,
* which defeats the purpose of having an async, non-blocking CIL force
* mechanism. Hence in this case we need to pass a flag to the push work to
* indicate it needs to flush the commit record itself.
*/
static void
xlog_cil_push_now(
struct xlog *log,
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
xfs_lsn_t push_seq,
bool async)
{
struct xfs_cil *cil = log->l_cilp;
if (!cil)
return;
ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
/* start on any pending background push to minimise wait time on it */
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
if (!async)
flush_workqueue(cil->xc_push_wq);
xfs: async CIL flushes need pending pushes to be made stable When the AIL tries to flush the CIL, it relies on the CIL push ending up on stable storage without having to wait for and manipulate iclog state directly. However, if there is already a pending CIL push when the AIL tries to flush the CIL, it won't set the cil->xc_push_commit_stable flag and so the CIL push will not actively flush the commit record iclog. generic/530 when run on a single CPU test VM can trigger this fairly reliably. This test exercises unlinked inode recovery, and can result in inodes being pinned in memory by ongoing modifications to the inode cluster buffer to record unlinked list modifications. As a result, the first inode unlinked in a buffer can pin the tail of the log whilst the inode cluster buffer is pinned by the current checkpoint that has been pushed but isn't on stable storage because because the cil->xc_push_commit_stable was not set. This results in the log/AIL effectively deadlocking until something triggers the commit record iclog to be pushed to stable storage (i.e. the periodic log worker calling xfs_log_force()). The fix is two-fold - first we should always set the cil->xc_push_commit_stable when xlog_cil_flush() is called, regardless of whether there is already a pending push or not. Second, if the CIL is empty, we should trigger an iclog flush to ensure that the iclogs of the last checkpoint have actually been submitted to disk as that checkpoint may not have been run under stable completion constraints. Reported-and-tested-by: Matthew Wilcox <willy@infradead.org> Fixes: 0020a190cf3e ("xfs: AIL needs asynchronous CIL forcing") Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-17 16:09:11 +00:00
spin_lock(&cil->xc_push_lock);
/*
* If this is an async flush request, we always need to set the
* xc_push_commit_stable flag even if something else has already queued
* a push. The flush caller is asking for the CIL to be on stable
* storage when the next push completes, so regardless of who has queued
* the push, the flush requires stable semantics from it.
*/
cil->xc_push_commit_stable = async;
/*
* If the CIL is empty or we've already pushed the sequence then
xfs: async CIL flushes need pending pushes to be made stable When the AIL tries to flush the CIL, it relies on the CIL push ending up on stable storage without having to wait for and manipulate iclog state directly. However, if there is already a pending CIL push when the AIL tries to flush the CIL, it won't set the cil->xc_push_commit_stable flag and so the CIL push will not actively flush the commit record iclog. generic/530 when run on a single CPU test VM can trigger this fairly reliably. This test exercises unlinked inode recovery, and can result in inodes being pinned in memory by ongoing modifications to the inode cluster buffer to record unlinked list modifications. As a result, the first inode unlinked in a buffer can pin the tail of the log whilst the inode cluster buffer is pinned by the current checkpoint that has been pushed but isn't on stable storage because because the cil->xc_push_commit_stable was not set. This results in the log/AIL effectively deadlocking until something triggers the commit record iclog to be pushed to stable storage (i.e. the periodic log worker calling xfs_log_force()). The fix is two-fold - first we should always set the cil->xc_push_commit_stable when xlog_cil_flush() is called, regardless of whether there is already a pending push or not. Second, if the CIL is empty, we should trigger an iclog flush to ensure that the iclogs of the last checkpoint have actually been submitted to disk as that checkpoint may not have been run under stable completion constraints. Reported-and-tested-by: Matthew Wilcox <willy@infradead.org> Fixes: 0020a190cf3e ("xfs: AIL needs asynchronous CIL forcing") Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-17 16:09:11 +00:00
* there's no more work that we need to do.
*/
if (test_bit(XLOG_CIL_EMPTY, &cil->xc_flags) ||
push_seq <= cil->xc_push_seq) {
spin_unlock(&cil->xc_push_lock);
return;
}
cil->xc_push_seq = push_seq;
queue_work(cil->xc_push_wq, &cil->xc_ctx->push_work);
spin_unlock(&cil->xc_push_lock);
}
xfs: prevent deadlock trying to cover an active log Recent analysis of a deadlocked XFS filesystem from a kernel crash dump indicated that the filesystem was stuck waiting for log space. The short story of the hang on the RHEL6 kernel is this: - the tail of the log is pinned by an inode - the inode has been pushed by the xfsaild - the inode has been flushed to it's backing buffer and is currently flush locked and hence waiting for backing buffer IO to complete and remove it from the AIL - the backing buffer is marked for write - it is on the delayed write queue - the inode buffer has been modified directly and logged recently due to unlinked inode list modification - the backing buffer is pinned in memory as it is in the active CIL context. - the xfsbufd won't start buffer writeback because it is pinned - xfssyncd won't force the log because it sees the log as needing to be covered and hence wants to issue a dummy transaction to move the log covering state machine along. Hence there is no trigger to force the CIL to the log and hence unpin the inode buffer and therefore complete the inode IO, remove it from the AIL and hence move the tail of the log along, allowing transactions to start again. Mainline kernels also have the same deadlock, though the signature is slightly different - the inode buffer never reaches the delayed write lists because xfs_buf_item_push() sees that it is pinned and hence never adds it to the delayed write list that the xfsaild flushes. There are two possible solutions here. The first is to simply force the log before trying to cover the log and so ensure that the CIL is emptied before we try to reserve space for the dummy transaction in the xfs_log_worker(). While this might work most of the time, it is still racy and is no guarantee that we don't get stuck in xfs_trans_reserve waiting for log space to come free. Hence it's not the best way to solve the problem. The second solution is to modify xfs_log_need_covered() to be aware of the CIL. We only should be attempting to cover the log if there is no current activity in the log - covering the log is the process of ensuring that the head and tail in the log on disk are identical (i.e. the log is clean and at idle). Hence, by definition, if there are items in the CIL then the log is not at idle and so we don't need to attempt to cover it. When we don't need to cover the log because it is active or idle, we issue a log force from xfs_log_worker() - if the log is idle, then this does nothing. However, if the log is active due to there being items in the CIL, it will force the items in the CIL to the log and unpin them. In the case of the above deadlock scenario, instead of xfs_log_worker() getting stuck in xfs_trans_reserve() attempting to cover the log, it will instead force the log, thereby unpinning the inode buffer, allowing IO to be issued and complete and hence removing the inode that was pinning the tail of the log from the AIL. At that point, everything will start moving along again. i.e. the xfs_log_worker turns back into a watchdog that can alleviate deadlocks based around pinned items that prevent the tail of the log from being moved... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-14 22:17:49 +00:00
bool
xlog_cil_empty(
struct xlog *log)
{
struct xfs_cil *cil = log->l_cilp;
bool empty = false;
spin_lock(&cil->xc_push_lock);
if (test_bit(XLOG_CIL_EMPTY, &cil->xc_flags))
xfs: prevent deadlock trying to cover an active log Recent analysis of a deadlocked XFS filesystem from a kernel crash dump indicated that the filesystem was stuck waiting for log space. The short story of the hang on the RHEL6 kernel is this: - the tail of the log is pinned by an inode - the inode has been pushed by the xfsaild - the inode has been flushed to it's backing buffer and is currently flush locked and hence waiting for backing buffer IO to complete and remove it from the AIL - the backing buffer is marked for write - it is on the delayed write queue - the inode buffer has been modified directly and logged recently due to unlinked inode list modification - the backing buffer is pinned in memory as it is in the active CIL context. - the xfsbufd won't start buffer writeback because it is pinned - xfssyncd won't force the log because it sees the log as needing to be covered and hence wants to issue a dummy transaction to move the log covering state machine along. Hence there is no trigger to force the CIL to the log and hence unpin the inode buffer and therefore complete the inode IO, remove it from the AIL and hence move the tail of the log along, allowing transactions to start again. Mainline kernels also have the same deadlock, though the signature is slightly different - the inode buffer never reaches the delayed write lists because xfs_buf_item_push() sees that it is pinned and hence never adds it to the delayed write list that the xfsaild flushes. There are two possible solutions here. The first is to simply force the log before trying to cover the log and so ensure that the CIL is emptied before we try to reserve space for the dummy transaction in the xfs_log_worker(). While this might work most of the time, it is still racy and is no guarantee that we don't get stuck in xfs_trans_reserve waiting for log space to come free. Hence it's not the best way to solve the problem. The second solution is to modify xfs_log_need_covered() to be aware of the CIL. We only should be attempting to cover the log if there is no current activity in the log - covering the log is the process of ensuring that the head and tail in the log on disk are identical (i.e. the log is clean and at idle). Hence, by definition, if there are items in the CIL then the log is not at idle and so we don't need to attempt to cover it. When we don't need to cover the log because it is active or idle, we issue a log force from xfs_log_worker() - if the log is idle, then this does nothing. However, if the log is active due to there being items in the CIL, it will force the items in the CIL to the log and unpin them. In the case of the above deadlock scenario, instead of xfs_log_worker() getting stuck in xfs_trans_reserve() attempting to cover the log, it will instead force the log, thereby unpinning the inode buffer, allowing IO to be issued and complete and hence removing the inode that was pinning the tail of the log from the AIL. At that point, everything will start moving along again. i.e. the xfs_log_worker turns back into a watchdog that can alleviate deadlocks based around pinned items that prevent the tail of the log from being moved... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-14 22:17:49 +00:00
empty = true;
spin_unlock(&cil->xc_push_lock);
return empty;
}
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
/*
* If there are intent done items in this transaction and the related intent was
* committed in the current (same) CIL checkpoint, we don't need to write either
* the intent or intent done item to the journal as the change will be
* journalled atomically within this checkpoint. As we cannot remove items from
* the CIL here, mark the related intent with a whiteout so that the CIL push
* can remove it rather than writing it to the journal. Then remove the intent
* done item from the current transaction and release it so it doesn't get put
* into the CIL at all.
*/
static uint32_t
xlog_cil_process_intents(
struct xfs_cil *cil,
struct xfs_trans *tp)
{
struct xfs_log_item *lip, *ilip, *next;
uint32_t len = 0;
list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
if (!(lip->li_ops->flags & XFS_ITEM_INTENT_DONE))
continue;
ilip = lip->li_ops->iop_intent(lip);
if (!ilip || !xlog_item_in_current_chkpt(cil, ilip))
continue;
set_bit(XFS_LI_WHITEOUT, &ilip->li_flags);
trace_xfs_cil_whiteout_mark(ilip);
len += ilip->li_lv->lv_bytes;
kmem_free(ilip->li_lv);
ilip->li_lv = NULL;
xfs_trans_del_item(lip);
lip->li_ops->iop_release(lip);
}
return len;
}
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
/*
* Commit a transaction with the given vector to the Committed Item List.
*
* To do this, we need to format the item, pin it in memory if required and
* account for the space used by the transaction. Once we have done that we
* need to release the unused reservation for the transaction, attach the
* transaction to the checkpoint context so we carry the busy extents through
* to checkpoint completion, and then unlock all the items in the transaction.
*
* Called with the context lock already held in read mode to lock out
* background commit, returns without it held once background commits are
* allowed again.
*/
void
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
xlog_cil_commit(
struct xlog *log,
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
struct xfs_trans *tp,
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
xfs_csn_t *commit_seq,
bool regrant)
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
{
struct xfs_cil *cil = log->l_cilp;
struct xfs_log_item *lip, *next;
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
uint32_t released_space = 0;
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
xfs: allocate log vector buffers outside CIL context lock One of the problems we currently have with delayed logging is that under serious memory pressure we can deadlock memory reclaim. THis occurs when memory reclaim (such as run by kswapd) is reclaiming XFS inodes and issues a log force to unpin inodes that are dirty in the CIL. The CIL is pushed, but this will only occur once it gets the CIL context lock to ensure that all committing transactions are complete and no new transactions start being committed to the CIL while the push switches to a new context. The deadlock occurs when the CIL context lock is held by a committing process that is doing memory allocation for log vector buffers, and that allocation is then blocked on memory reclaim making progress. Memory reclaim, however, is blocked waiting for a log force to make progress, and so we effectively deadlock at this point. To solve this problem, we have to move the CIL log vector buffer allocation outside of the context lock so that memory reclaim can always make progress when it needs to force the log. The problem with doing this is that a CIL push can take place while we are determining if we need to allocate a new log vector buffer for an item and hence the current log vector may go away without warning. That means we canot rely on the existing log vector being present when we finally grab the context lock and so we must have a replacement buffer ready to go at all times. To ensure this, introduce a "shadow log vector" buffer that is always guaranteed to be present when we gain the CIL context lock and format the item. This shadow buffer may or may not be used during the formatting, but if the log item does not have an existing log vector buffer or that buffer is too small for the new modifications, we swap it for the new shadow buffer and format the modifications into that new log vector buffer. The result of this is that for any object we modify more than once in a given CIL checkpoint, we double the memory required to track dirty regions in the log. For single modifications then we consume the shadow log vectorwe allocate on commit, and that gets consumed by the checkpoint. However, if we make multiple modifications, then the second transaction commit will allocate a shadow log vector and hence we will end up with double the memory usage as only one of the log vectors is consumed by the CIL checkpoint. The remaining shadow vector will be freed when th elog item is freed. This can probably be optimised in future - access to the shadow log vector is serialised by the object lock (as opposited to the active log vector, which is controlled by the CIL context lock) and so we can probably free shadow log vector from some objects when the log item is marked clean on removal from the AIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-07-21 23:52:35 +00:00
/*
* Do all necessary memory allocation before we lock the CIL.
* This ensures the allocation does not deadlock with a CIL
* push in memory reclaim (e.g. from kswapd).
*/
xlog_cil_alloc_shadow_bufs(log, tp);
/* lock out background commit */
down_read(&cil->xc_ctx_lock);
xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 01:50:29 +00:00
if (tp->t_flags & XFS_TRANS_HAS_INTENT_DONE)
released_space = xlog_cil_process_intents(cil, tp);
xlog_cil_insert_items(log, tp, released_space);
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
if (regrant && !xlog_is_shutdown(log))
xfs_log_ticket_regrant(log, tp->t_ticket);
else
xfs_log_ticket_ungrant(log, tp->t_ticket);
tp->t_ticket = NULL;
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
xfs_trans_unreserve_and_mod_sb(tp);
/*
* Once all the items of the transaction have been copied to the CIL,
* the items can be unlocked and possibly freed.
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
*
* This needs to be done before we drop the CIL context lock because we
* have to update state in the log items and unlock them before they go
* to disk. If we don't, then the CIL checkpoint can race with us and
* we can run checkpoint completion before we've updated and unlocked
* the log items. This affects (at least) processing of stale buffers,
* inodes and EFIs.
*/
trace_xfs_trans_commit_items(tp, _RET_IP_);
list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
xfs_trans_del_item(lip);
if (lip->li_ops->iop_committing)
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
lip->li_ops->iop_committing(lip, cil->xc_ctx->sequence);
}
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
if (commit_seq)
*commit_seq = cil->xc_ctx->sequence;
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
/* xlog_cil_push_background() releases cil->xc_ctx_lock */
xlog_cil_push_background(log);
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
}
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
/*
* Flush the CIL to stable storage but don't wait for it to complete. This
* requires the CIL push to ensure the commit record for the push hits the disk,
* but otherwise is no different to a push done from a log force.
*/
void
xlog_cil_flush(
struct xlog *log)
{
xfs_csn_t seq = log->l_cilp->xc_current_sequence;
trace_xfs_log_force(log->l_mp, seq, _RET_IP_);
xlog_cil_push_now(log, seq, true);
xfs: async CIL flushes need pending pushes to be made stable When the AIL tries to flush the CIL, it relies on the CIL push ending up on stable storage without having to wait for and manipulate iclog state directly. However, if there is already a pending CIL push when the AIL tries to flush the CIL, it won't set the cil->xc_push_commit_stable flag and so the CIL push will not actively flush the commit record iclog. generic/530 when run on a single CPU test VM can trigger this fairly reliably. This test exercises unlinked inode recovery, and can result in inodes being pinned in memory by ongoing modifications to the inode cluster buffer to record unlinked list modifications. As a result, the first inode unlinked in a buffer can pin the tail of the log whilst the inode cluster buffer is pinned by the current checkpoint that has been pushed but isn't on stable storage because because the cil->xc_push_commit_stable was not set. This results in the log/AIL effectively deadlocking until something triggers the commit record iclog to be pushed to stable storage (i.e. the periodic log worker calling xfs_log_force()). The fix is two-fold - first we should always set the cil->xc_push_commit_stable when xlog_cil_flush() is called, regardless of whether there is already a pending push or not. Second, if the CIL is empty, we should trigger an iclog flush to ensure that the iclogs of the last checkpoint have actually been submitted to disk as that checkpoint may not have been run under stable completion constraints. Reported-and-tested-by: Matthew Wilcox <willy@infradead.org> Fixes: 0020a190cf3e ("xfs: AIL needs asynchronous CIL forcing") Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-17 16:09:11 +00:00
/*
* If the CIL is empty, make sure that any previous checkpoint that may
* still be in an active iclog is pushed to stable storage.
*/
if (test_bit(XLOG_CIL_EMPTY, &log->l_cilp->xc_flags))
xfs: async CIL flushes need pending pushes to be made stable When the AIL tries to flush the CIL, it relies on the CIL push ending up on stable storage without having to wait for and manipulate iclog state directly. However, if there is already a pending CIL push when the AIL tries to flush the CIL, it won't set the cil->xc_push_commit_stable flag and so the CIL push will not actively flush the commit record iclog. generic/530 when run on a single CPU test VM can trigger this fairly reliably. This test exercises unlinked inode recovery, and can result in inodes being pinned in memory by ongoing modifications to the inode cluster buffer to record unlinked list modifications. As a result, the first inode unlinked in a buffer can pin the tail of the log whilst the inode cluster buffer is pinned by the current checkpoint that has been pushed but isn't on stable storage because because the cil->xc_push_commit_stable was not set. This results in the log/AIL effectively deadlocking until something triggers the commit record iclog to be pushed to stable storage (i.e. the periodic log worker calling xfs_log_force()). The fix is two-fold - first we should always set the cil->xc_push_commit_stable when xlog_cil_flush() is called, regardless of whether there is already a pending push or not. Second, if the CIL is empty, we should trigger an iclog flush to ensure that the iclogs of the last checkpoint have actually been submitted to disk as that checkpoint may not have been run under stable completion constraints. Reported-and-tested-by: Matthew Wilcox <willy@infradead.org> Fixes: 0020a190cf3e ("xfs: AIL needs asynchronous CIL forcing") Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-17 16:09:11 +00:00
xfs_log_force(log->l_mp, 0);
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
}
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* Conditionally push the CIL based on the sequence passed in.
*
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
* We only need to push if we haven't already pushed the sequence number given.
* Hence the only time we will trigger a push here is if the push sequence is
* the same as the current context.
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
*
* We return the current commit lsn to allow the callers to determine if a
* iclog flush is necessary following this call.
*/
xfs_lsn_t
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
xlog_cil_force_seq(
struct xlog *log,
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
xfs_csn_t sequence)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
{
struct xfs_cil *cil = log->l_cilp;
struct xfs_cil_ctx *ctx;
xfs_lsn_t commit_lsn = NULLCOMMITLSN;
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
ASSERT(sequence <= cil->xc_current_sequence);
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
if (!sequence)
sequence = cil->xc_current_sequence;
trace_xfs_log_force(log->l_mp, sequence, _RET_IP_);
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
/*
* check to see if we need to force out the current context.
* xlog_cil_push() handles racing pushes for the same sequence,
* so no need to deal with it here.
*/
restart:
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
xlog_cil_push_now(log, sequence, false);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
/*
* See if we can find a previous sequence still committing.
* We need to wait for all previous sequence commits to complete
* before allowing the force of push_seq to go ahead. Hence block
* on commits for those as well.
*/
spin_lock(&cil->xc_push_lock);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
list_for_each_entry(ctx, &cil->xc_committing, committing) {
/*
* Avoid getting stuck in this loop because we were woken by the
* shutdown, but then went back to sleep once already in the
* shutdown state.
*/
if (xlog_is_shutdown(log))
goto out_shutdown;
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
if (ctx->sequence > sequence)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
continue;
if (!ctx->commit_lsn) {
/*
* It is still being pushed! Wait for the push to
* complete, then start again from the beginning.
*/
xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
XFS_STATS_INC(log->l_mp, xs_log_force_sleep);
xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
goto restart;
}
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
if (ctx->sequence != sequence)
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
continue;
/* found it! */
commit_lsn = ctx->commit_lsn;
}
/*
* The call to xlog_cil_push_now() executes the push in the background.
* Hence by the time we have got here it our sequence may not have been
* pushed yet. This is true if the current sequence still matches the
* push sequence after the above wait loop and the CIL still contains
xfs: xlog_cil_force_lsn doesn't always wait correctly When running a tight mount/unmount loop on an older kernel, RedHat QE found that unmount would occasionally hang in xfs_buf_unpin_wait() on the superblock buffer. Tracing and other debug work by Eric Sandeen indicated that it was hanging on the writing of the superblock during unmount immediately after logging the superblock counters in a synchronous transaction. Further debug indicated that the synchronous transaction was not waiting for completion correctly, and we narrowed it down to xlog_cil_force_lsn() returning NULLCOMMITLSN and hence not pushing the transaction in the iclog buffer to disk correctly. While this unmount superblock write code is now very different in mainline kernels, the xlog_cil_force_lsn() code is identical, and it was bisected to the backport of commit f876e44 ("xfs: always do log forces via the workqueue"). This commit made the CIL push asynchronous for log forces and hence exposed a race condition that couldn't occur on a synchronous push. Essentially, the xlog_cil_force_lsn() relied implicitly on the fact that the sequence push would be complete by the time xlog_cil_push_now() returned, resulting in the context being pushed being in the committing list. When it was made asynchronous, it was recognised that there was a race condition in detecting whether an asynchronous push has started or not and code was added to handle it. Unfortunately, the fix was not quite right and left a race condition where it it would detect an empty CIL while a push was in progress before the context had been added to the committing list. This was incorrectly seen as a "nothing to do" condition and so would tell xfs_log_force_lsn() that there is nothing to wait for, and hence it would push the iclogbufs in memory. The fix is simple, but explaining the logic and the race condition is a lot more complex. The fix is to add the context to the committing list before we start emptying the CIL. This allows us to detect the difference between an empty "do nothing" push and a push that has not started by adding a discrete "emptying the CIL" state to avoid the transient, incorrect "empty" condition that the (unchanged) waiting code was seeing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-09-23 05:57:59 +00:00
* dirty objects. This is guaranteed by the push code first adding the
* context to the committing list before emptying the CIL.
*
xfs: xlog_cil_force_lsn doesn't always wait correctly When running a tight mount/unmount loop on an older kernel, RedHat QE found that unmount would occasionally hang in xfs_buf_unpin_wait() on the superblock buffer. Tracing and other debug work by Eric Sandeen indicated that it was hanging on the writing of the superblock during unmount immediately after logging the superblock counters in a synchronous transaction. Further debug indicated that the synchronous transaction was not waiting for completion correctly, and we narrowed it down to xlog_cil_force_lsn() returning NULLCOMMITLSN and hence not pushing the transaction in the iclog buffer to disk correctly. While this unmount superblock write code is now very different in mainline kernels, the xlog_cil_force_lsn() code is identical, and it was bisected to the backport of commit f876e44 ("xfs: always do log forces via the workqueue"). This commit made the CIL push asynchronous for log forces and hence exposed a race condition that couldn't occur on a synchronous push. Essentially, the xlog_cil_force_lsn() relied implicitly on the fact that the sequence push would be complete by the time xlog_cil_push_now() returned, resulting in the context being pushed being in the committing list. When it was made asynchronous, it was recognised that there was a race condition in detecting whether an asynchronous push has started or not and code was added to handle it. Unfortunately, the fix was not quite right and left a race condition where it it would detect an empty CIL while a push was in progress before the context had been added to the committing list. This was incorrectly seen as a "nothing to do" condition and so would tell xfs_log_force_lsn() that there is nothing to wait for, and hence it would push the iclogbufs in memory. The fix is simple, but explaining the logic and the race condition is a lot more complex. The fix is to add the context to the committing list before we start emptying the CIL. This allows us to detect the difference between an empty "do nothing" push and a push that has not started by adding a discrete "emptying the CIL" state to avoid the transient, incorrect "empty" condition that the (unchanged) waiting code was seeing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-09-23 05:57:59 +00:00
* Hence if we don't find the context in the committing list and the
* current sequence number is unchanged then the CIL contents are
* significant. If the CIL is empty, if means there was nothing to push
* and that means there is nothing to wait for. If the CIL is not empty,
* it means we haven't yet started the push, because if it had started
* we would have found the context on the committing list.
*/
if (sequence == cil->xc_current_sequence &&
!test_bit(XLOG_CIL_EMPTY, &cil->xc_flags)) {
spin_unlock(&cil->xc_push_lock);
goto restart;
}
spin_unlock(&cil->xc_push_lock);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
return commit_lsn;
/*
* We detected a shutdown in progress. We need to trigger the log force
* to pass through it's iclog state machine error handling, even though
* we are already in a shutdown state. Hence we can't return
* NULLCOMMITLSN here as that has special meaning to log forces (i.e.
* LSN is already stable), so we return a zero LSN instead.
*/
out_shutdown:
spin_unlock(&cil->xc_push_lock);
return 0;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 04:37:18 +00:00
}
xfs: Ensure inode allocation buffers are fully replayed With delayed logging, we can get inode allocation buffers in the same transaction inode unlink buffers. We don't currently mark inode allocation buffers in the log, so inode unlink buffers take precedence over allocation buffers. The result is that when they are combined into the same checkpoint, only the unlinked inode chain fields are replayed, resulting in uninitialised inode buffers being detected when the next inode modification is replayed. To fix this, we need to ensure that we do not set the inode buffer flag in the buffer log item format flags if the inode allocation has not already hit the log. To avoid requiring a change to log recovery, we really need to make this a modification that relies only on in-memory sate. We can do this by checking during buffer log formatting (while the CIL cannot be flushed) if we are still in the same sequence when we commit the unlink transaction as the inode allocation transaction. If we are, then we do not add the inode buffer flag to the buffer log format item flags. This means the entire buffer will be replayed, not just the unlinked fields. We do this while CIL flusheѕ are locked out to ensure that we don't race with the sequence numbers changing and hence fail to put the inode buffer flag in the buffer format flags when we really need to. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-20 13:19:42 +00:00
/*
* Move dead percpu state to the relevant CIL context structures.
*
* We have to lock the CIL context here to ensure that nothing is modifying
* the percpu state, either addition or removal. Both of these are done under
* the CIL context lock, so grabbing that exclusively here will ensure we can
* safely drain the cilpcp for the CPU that is dying.
*/
void
xlog_cil_pcp_dead(
struct xlog *log,
unsigned int cpu)
{
struct xfs_cil *cil = log->l_cilp;
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
struct xlog_cil_pcp *cilpcp = per_cpu_ptr(cil->xc_pcp, cpu);
struct xfs_cil_ctx *ctx;
down_write(&cil->xc_ctx_lock);
ctx = cil->xc_ctx;
if (ctx->ticket)
ctx->ticket->t_curr_res += cilpcp->space_reserved;
cilpcp->space_reserved = 0;
if (!list_empty(&cilpcp->log_items))
list_splice_init(&cilpcp->log_items, &ctx->log_items);
if (!list_empty(&cilpcp->busy_extents))
list_splice_init(&cilpcp->busy_extents, &ctx->busy_extents);
atomic_add(cilpcp->space_used, &ctx->space_used);
xfs: implement percpu cil space used calculation Now that we have the CIL percpu structures in place, implement the space used counter as a per-cpu counter. We have to be really careful now about ensuring that the checks and updates run without arbitrary delays, which means they need to run with pre-emption disabled. We do this by careful placement of the get_cpu_ptr/put_cpu_ptr calls to access the per-cpu structures for that CPU. We need to be able to reliably detect that the CIL has reached the hard limit threshold so we can take extra reservations for the iclog headers when the space used overruns the original reservation. hence we factor out xlog_cil_over_hard_limit() from xlog_cil_push_background(). The global CIL space used is an atomic variable that is backed by per-cpu aggregation to minimise the number of atomic updates we do to the global state in the fast path. While we are under the soft limit, we aggregate only when the per-cpu aggregation is over the proportion of the soft limit assigned to that CPU. This means that all CPUs can use all but one byte of their aggregation threshold and we will not go over the soft limit. Hence once we detect that we've gone over both a per-cpu aggregation threshold and the soft limit, we know that we have only exceeded the soft limit by one per-cpu aggregation threshold. Even if all CPUs hit this at the same time, we can't be over the hard limit, so we can run an aggregation back into the atomic counter at this point and still be under the hard limit. At this point, we will be over the soft limit and hence we'll aggregate into the global atomic used space directly rather than the per-cpu counters, hence providing accurate detection of hard limit excursion for accounting and reservation purposes. Hence we get the best of both worlds - lockless, scalable per-cpu fast path plus accurate, atomic detection of hard limit excursion. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 08:50:59 +00:00
cilpcp->space_used = 0;
up_write(&cil->xc_ctx_lock);
}
/*
* Perform initial CIL structure initialisation.
*/
int
xlog_cil_init(
struct xlog *log)
{
struct xfs_cil *cil;
struct xfs_cil_ctx *ctx;
struct xlog_cil_pcp *cilpcp;
int cpu;
cil = kmem_zalloc(sizeof(*cil), KM_MAYFAIL);
if (!cil)
return -ENOMEM;
/*
* Limit the CIL pipeline depth to 4 concurrent works to bound the
* concurrency the log spinlocks will be exposed to.
*/
cil->xc_push_wq = alloc_workqueue("xfs-cil/%s",
XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_UNBOUND),
4, log->l_mp->m_super->s_id);
if (!cil->xc_push_wq)
goto out_destroy_cil;
cil->xc_log = log;
cil->xc_pcp = alloc_percpu(struct xlog_cil_pcp);
if (!cil->xc_pcp)
goto out_destroy_wq;
for_each_possible_cpu(cpu) {
cilpcp = per_cpu_ptr(cil->xc_pcp, cpu);
INIT_LIST_HEAD(&cilpcp->busy_extents);
INIT_LIST_HEAD(&cilpcp->log_items);
}
INIT_LIST_HEAD(&cil->xc_committing);
spin_lock_init(&cil->xc_push_lock);
init_waitqueue_head(&cil->xc_push_wait);
init_rwsem(&cil->xc_ctx_lock);
xfs: order CIL checkpoint start records Because log recovery depends on strictly ordered start records as well as strictly ordered commit records. This is a zero day bug in the way XFS writes pipelined transactions to the journal which is exposed by fixing the zero day bug that prevents the CIL from pipelining checkpoints. This re-introduces explicit concurrent commits back into the on-disk journal and hence out of order start records. The XFS journal commit code has never ordered start records and we have relied on strict commit record ordering for correct recovery ordering of concurrently written transactions. Unfortunately, root cause analysis uncovered the fact that log recovery uses the LSN of the start record for transaction commit processing. Hence, whilst the commits are processed in strict order by recovery, the LSNs associated with the commits can be out of order and so recovery may stamp incorrect LSNs into objects and/or misorder intents in the AIL for later processing. This can result in log recovery failures and/or on disk corruption, sometimes silent. Because this is a long standing log recovery issue, we can't just fix log recovery and call it good. This still leaves older kernels susceptible to recovery failures and corruption when replaying a log from a kernel that pipelines checkpoints. There is also the issue that in-memory ordering for AIL pushing and data integrity operations are based on checkpoint start LSNs, and if the start LSN is incorrect in the journal, it is also incorrect in memory. Hence there's really only one choice for fixing this zero-day bug: we need to strictly order checkpoint start records in ascending sequence order in the log, the same way we already strictly order commit records. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:44 +00:00
init_waitqueue_head(&cil->xc_start_wait);
init_waitqueue_head(&cil->xc_commit_wait);
log->l_cilp = cil;
xfs: CIL work is serialised, not pipelined Because we use a single work structure attached to the CIL rather than the CIL context, we can only queue a single work item at a time. This results in the CIL being single threaded and limits performance when it becomes CPU bound. The design of the CIL is that it is pipelined and multiple commits can be running concurrently, but the way the work is currently implemented means that it is not pipelining as it was intended. The critical work to switch the CIL context can take a few milliseconds to run, but the rest of the CIL context flush can take hundreds of milliseconds to complete. The context switching is the serialisation point of the CIL, once the context has been switched the rest of the context push can run asynchrnously with all other context pushes. Hence we can move the work to the CIL context so that we can run multiple CIL pushes at the same time and spread the majority of the work out over multiple CPUs. We can keep the per-cpu CIL commit state on the CIL rather than the context, because the context is pinned to the CIL until the switch is done and we aggregate and drain the per-cpu state held on the CIL during the context switch. However, because we no longer serialise the CIL work, we can have effectively unlimited CIL pushes in progress. We don't want to do this - not only does it create contention on the iclogs and the state machine locks, we can run the log right out of space with outstanding pushes. Instead, limit the work concurrency to 4 concurrent works being processed at a time. This is enough concurrency to remove the CIL from being a CPU bound bottleneck but not enough to create new contention points or unbound concurrency issues. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-11 01:00:45 +00:00
ctx = xlog_cil_ctx_alloc();
xlog_cil_ctx_switch(cil, ctx);
return 0;
out_destroy_wq:
destroy_workqueue(cil->xc_push_wq);
out_destroy_cil:
kmem_free(cil);
return -ENOMEM;
}
void
xlog_cil_destroy(
struct xlog *log)
{
struct xfs_cil *cil = log->l_cilp;
if (cil->xc_ctx) {
if (cil->xc_ctx->ticket)
xfs_log_ticket_put(cil->xc_ctx->ticket);
kmem_free(cil->xc_ctx);
}
ASSERT(test_bit(XLOG_CIL_EMPTY, &cil->xc_flags));
free_percpu(cil->xc_pcp);
destroy_workqueue(cil->xc_push_wq);
kmem_free(cil);
}