linux/fs/jbd2/transaction.c

2780 lines
86 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0+
/*
* linux/fs/jbd2/transaction.c
*
* Written by Stephen C. Tweedie <sct@redhat.com>, 1998
*
* Copyright 1998 Red Hat corp --- All Rights Reserved
*
* Generic filesystem transaction handling code; part of the ext2fs
* journaling system.
*
* This file manages transactions (compound commits managed by the
* journaling code) and handles (individual atomic operations by the
* filesystem).
*/
#include <linux/time.h>
#include <linux/fs.h>
#include <linux/jbd2.h>
#include <linux/errno.h>
#include <linux/slab.h>
#include <linux/timer.h>
#include <linux/mm.h>
#include <linux/highmem.h>
#include <linux/hrtimer.h>
#include <linux/backing-dev.h>
#include <linux/bug.h>
#include <linux/module.h>
#include <linux/sched/mm.h>
#include <trace/events/jbd2.h>
static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
static struct kmem_cache *transaction_cache;
int __init jbd2_journal_init_transaction_cache(void)
{
J_ASSERT(!transaction_cache);
transaction_cache = kmem_cache_create("jbd2_transaction_s",
sizeof(transaction_t),
0,
SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
NULL);
if (!transaction_cache) {
pr_emerg("JBD2: failed to create transaction cache\n");
return -ENOMEM;
}
return 0;
}
void jbd2_journal_destroy_transaction_cache(void)
{
kmem_cache_destroy(transaction_cache);
transaction_cache = NULL;
}
void jbd2_journal_free_transaction(transaction_t *transaction)
{
if (unlikely(ZERO_OR_NULL_PTR(transaction)))
return;
kmem_cache_free(transaction_cache, transaction);
}
/*
* Base amount of descriptor blocks we reserve for each transaction.
*/
static int jbd2_descriptor_blocks_per_trans(journal_t *journal)
{
int tag_space = journal->j_blocksize - sizeof(journal_header_t);
int tags_per_block;
/* Subtract UUID */
tag_space -= 16;
if (jbd2_journal_has_csum_v2or3(journal))
tag_space -= sizeof(struct jbd2_journal_block_tail);
/* Commit code leaves a slack space of 16 bytes at the end of block */
tags_per_block = (tag_space - 16) / journal_tag_bytes(journal);
/*
* Revoke descriptors are accounted separately so we need to reserve
* space for commit block and normal transaction descriptor blocks.
*/
return 1 + DIV_ROUND_UP(journal->j_max_transaction_buffers,
tags_per_block);
}
/*
* jbd2_get_transaction: obtain a new transaction_t object.
*
* Simply initialise a new transaction. Initialize it in
* RUNNING state and add it to the current journal (which should not
* have an existing running transaction: we only make a new transaction
* once we have started to commit the old one).
*
* Preconditions:
* The journal MUST be locked. We don't perform atomic mallocs on the
* new transaction and we can't block without protecting against other
* processes trying to touch the journal while it is in transition.
*
*/
static void jbd2_get_transaction(journal_t *journal,
transaction_t *transaction)
{
transaction->t_journal = journal;
transaction->t_state = T_RUNNING;
transaction->t_start_time = ktime_get();
transaction->t_tid = journal->j_transaction_sequence++;
transaction->t_expires = jiffies + journal->j_commit_interval;
atomic_set(&transaction->t_updates, 0);
atomic_set(&transaction->t_outstanding_credits,
jbd2_descriptor_blocks_per_trans(journal) +
atomic_read(&journal->j_reserved_credits));
atomic_set(&transaction->t_outstanding_revokes, 0);
atomic_set(&transaction->t_handle_count, 0);
INIT_LIST_HEAD(&transaction->t_inode_list);
INIT_LIST_HEAD(&transaction->t_private_list);
/* Set up the commit timer for the new transaction. */
journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
add_timer(&journal->j_commit_timer);
J_ASSERT(journal->j_running_transaction == NULL);
journal->j_running_transaction = transaction;
jbd2: jbd2 stats through procfs The patch below updates the jbd stats patch to 2.6.20/jbd2. The initial patch was posted by Alex Tomas in December 2005 (http://marc.info/?l=linux-ext4&m=113538565128617&w=2). It provides statistics via procfs such as transaction lifetime and size. Sometimes, investigating performance problems, i find useful to have stats from jbd about transaction's lifetime, size, etc. here is a patch for review and inclusion probably. for example, stats after creation of 3M files in htree directory: [root@bob ~]# cat /proc/fs/jbd/sda/history R/C tid wait run lock flush log hndls block inlog ctime write drop close R 261 8260 2720 0 0 750 9892 8170 8187 C 259 750 0 4885 1 R 262 20 2200 10 0 770 9836 8170 8187 R 263 30 2200 10 0 3070 9812 8170 8187 R 264 0 5000 10 0 1340 0 0 0 C 261 8240 3212 4957 0 R 265 8260 1470 0 0 4640 9854 8170 8187 R 266 0 5000 10 0 1460 0 0 0 C 262 8210 2989 4868 0 R 267 8230 1490 10 0 4440 9875 8171 8188 R 268 0 5000 10 0 1260 0 0 0 C 263 7710 2937 4908 0 R 269 7730 1470 10 0 3330 9841 8170 8187 R 270 0 5000 10 0 830 0 0 0 C 265 8140 3234 4898 0 C 267 720 0 4849 1 R 271 8630 2740 20 0 740 9819 8170 8187 C 269 800 0 4214 1 R 272 40 2170 10 0 830 9716 8170 8187 R 273 40 2280 0 0 3530 9799 8170 8187 R 274 0 5000 10 0 990 0 0 0 where, R - line for transaction's life from T_RUNNING to T_FINISHED C - line for transaction's checkpointing tid - transaction's id wait - for how long we were waiting for new transaction to start (the longest period journal_start() took in this transaction) run - real transaction's lifetime (from T_RUNNING to T_LOCKED lock - how long we were waiting for all handles to close (time the transaction was in T_LOCKED) flush - how long it took to flush all data (data=ordered) log - how long it took to write the transaction to the log hndls - how many handles got to the transaction block - how many blocks got to the transaction inlog - how many blocks are written to the log (block + descriptors) ctime - how long it took to checkpoint the transaction write - how many blocks have been written during checkpointing drop - how many blocks have been dropped during checkpointing close - how many running transactions have been closed to checkpoint this one all times are in msec. [root@bob ~]# cat /proc/fs/jbd/sda/info 280 transaction, each upto 8192 blocks average: 1633ms waiting for transaction 3616ms running transaction 5ms transaction was being locked 1ms flushing data (in ordered mode) 1799ms logging transaction 11781 handles per transaction 5629 blocks per transaction 5641 logged blocks per transaction Signed-off-by: Johann Lombardi <johann.lombardi@bull.net> Signed-off-by: Mariusz Kozlowski <m.kozlowski@tuxland.pl> Signed-off-by: Mingming Cao <cmm@us.ibm.com> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
2008-01-29 04:58:27 +00:00
transaction->t_max_wait = 0;
transaction->t_start = jiffies;
transaction->t_requested = 0;
}
/*
* Handle management.
*
* A handle_t is an object which represents a single atomic update to a
* filesystem, and which tracks all of the modifications which form part
* of that one update.
*/
/*
* Update transaction's maximum wait time, if debugging is enabled.
*
jbd2: kill t_handle_lock transaction spinlock This patch kills t_handle_lock transaction spinlock completely from jbd2. To explain the reasoning, currently there were three sites at which this spinlock was used. 1. jbd2_journal_wait_updates() a. Based on careful code review it can be seen that, we don't need this lock here. This is since we wait for any currently ongoing updates based on a atomic variable t_updates. And we anyway don't take any t_handle_lock while in stop_this_handle(). i.e. write_lock(&journal->j_state_lock() jbd2_journal_wait_updates() stop_this_handle() while (atomic_read(txn->t_updates) { | DEFINE_WAIT(wait); | prepare_to_wait(); | if (atomic_read(txn->t_updates) if (atomic_dec_and_test(txn->t_updates)) write_unlock(&journal->j_state_lock); schedule(); wake_up() write_lock(&journal->j_state_lock); finish_wait(); } txn->t_state = T_COMMIT write_unlock(&journal->j_state_lock); b. Also note that between atomic_inc(&txn->t_updates) in start_this_handle() and jbd2_journal_wait_updates(), the synchronization happens via read_lock(journal->j_state_lock) in start_this_handle(); 2. jbd2_journal_extend() a. jbd2_journal_extend() is called with the handle of each process from task_struct. So no lock required in updating member fields of handle_t b. For member fields of h_transaction, all updates happens only via atomic APIs (which is also within read_lock()). So, no need of this transaction spinlock. 3. update_t_max_wait() Based on Jan suggestion, this can be carefully removed using atomic cmpxchg API. Note that there can be several processes which are waiting for a new transaction to be allocated and started. For doing this only one process will succeed in taking write_lock() and allocating a new txn. After that all of the process will be updating the t_max_wait (max transaction wait time). This can be done via below method w/o taking any locks using atomic cmpxchg. For more details refer [1] new = get_new_val(); old = READ_ONCE(ptr->max_val); while (old < new) old = cmpxchg(&ptr->max_val, old, new); [1]: https://lwn.net/Articles/849237/ Suggested-by: Jan Kara <jack@suse.cz> Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/d89e599658b4a1f3893a48c6feded200073037fc.1644992076.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-16 07:00:35 +00:00
* t_max_wait is carefully updated here with use of atomic compare exchange.
* Note that there could be multiplre threads trying to do this simultaneously
* hence using cmpxchg to avoid any use of locks in this case.
* With this t_max_wait can be updated w/o enabling jbd2_journal_enable_debug.
*/
static inline void update_t_max_wait(transaction_t *transaction,
unsigned long ts)
{
jbd2: kill t_handle_lock transaction spinlock This patch kills t_handle_lock transaction spinlock completely from jbd2. To explain the reasoning, currently there were three sites at which this spinlock was used. 1. jbd2_journal_wait_updates() a. Based on careful code review it can be seen that, we don't need this lock here. This is since we wait for any currently ongoing updates based on a atomic variable t_updates. And we anyway don't take any t_handle_lock while in stop_this_handle(). i.e. write_lock(&journal->j_state_lock() jbd2_journal_wait_updates() stop_this_handle() while (atomic_read(txn->t_updates) { | DEFINE_WAIT(wait); | prepare_to_wait(); | if (atomic_read(txn->t_updates) if (atomic_dec_and_test(txn->t_updates)) write_unlock(&journal->j_state_lock); schedule(); wake_up() write_lock(&journal->j_state_lock); finish_wait(); } txn->t_state = T_COMMIT write_unlock(&journal->j_state_lock); b. Also note that between atomic_inc(&txn->t_updates) in start_this_handle() and jbd2_journal_wait_updates(), the synchronization happens via read_lock(journal->j_state_lock) in start_this_handle(); 2. jbd2_journal_extend() a. jbd2_journal_extend() is called with the handle of each process from task_struct. So no lock required in updating member fields of handle_t b. For member fields of h_transaction, all updates happens only via atomic APIs (which is also within read_lock()). So, no need of this transaction spinlock. 3. update_t_max_wait() Based on Jan suggestion, this can be carefully removed using atomic cmpxchg API. Note that there can be several processes which are waiting for a new transaction to be allocated and started. For doing this only one process will succeed in taking write_lock() and allocating a new txn. After that all of the process will be updating the t_max_wait (max transaction wait time). This can be done via below method w/o taking any locks using atomic cmpxchg. For more details refer [1] new = get_new_val(); old = READ_ONCE(ptr->max_val); while (old < new) old = cmpxchg(&ptr->max_val, old, new); [1]: https://lwn.net/Articles/849237/ Suggested-by: Jan Kara <jack@suse.cz> Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/d89e599658b4a1f3893a48c6feded200073037fc.1644992076.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-16 07:00:35 +00:00
unsigned long oldts, newts;
if (time_after(transaction->t_start, ts)) {
jbd2: kill t_handle_lock transaction spinlock This patch kills t_handle_lock transaction spinlock completely from jbd2. To explain the reasoning, currently there were three sites at which this spinlock was used. 1. jbd2_journal_wait_updates() a. Based on careful code review it can be seen that, we don't need this lock here. This is since we wait for any currently ongoing updates based on a atomic variable t_updates. And we anyway don't take any t_handle_lock while in stop_this_handle(). i.e. write_lock(&journal->j_state_lock() jbd2_journal_wait_updates() stop_this_handle() while (atomic_read(txn->t_updates) { | DEFINE_WAIT(wait); | prepare_to_wait(); | if (atomic_read(txn->t_updates) if (atomic_dec_and_test(txn->t_updates)) write_unlock(&journal->j_state_lock); schedule(); wake_up() write_lock(&journal->j_state_lock); finish_wait(); } txn->t_state = T_COMMIT write_unlock(&journal->j_state_lock); b. Also note that between atomic_inc(&txn->t_updates) in start_this_handle() and jbd2_journal_wait_updates(), the synchronization happens via read_lock(journal->j_state_lock) in start_this_handle(); 2. jbd2_journal_extend() a. jbd2_journal_extend() is called with the handle of each process from task_struct. So no lock required in updating member fields of handle_t b. For member fields of h_transaction, all updates happens only via atomic APIs (which is also within read_lock()). So, no need of this transaction spinlock. 3. update_t_max_wait() Based on Jan suggestion, this can be carefully removed using atomic cmpxchg API. Note that there can be several processes which are waiting for a new transaction to be allocated and started. For doing this only one process will succeed in taking write_lock() and allocating a new txn. After that all of the process will be updating the t_max_wait (max transaction wait time). This can be done via below method w/o taking any locks using atomic cmpxchg. For more details refer [1] new = get_new_val(); old = READ_ONCE(ptr->max_val); while (old < new) old = cmpxchg(&ptr->max_val, old, new); [1]: https://lwn.net/Articles/849237/ Suggested-by: Jan Kara <jack@suse.cz> Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/d89e599658b4a1f3893a48c6feded200073037fc.1644992076.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-16 07:00:35 +00:00
newts = jbd2_time_diff(ts, transaction->t_start);
oldts = READ_ONCE(transaction->t_max_wait);
while (oldts < newts)
oldts = cmpxchg(&transaction->t_max_wait, oldts, newts);
}
}
/*
* Wait until running transaction passes to T_FLUSH state and new transaction
* can thus be started. Also starts the commit if needed. The function expects
* running transaction to exist and releases j_state_lock.
*/
static void wait_transaction_locked(journal_t *journal)
__releases(journal->j_state_lock)
{
DEFINE_WAIT(wait);
int need_to_start;
tid_t tid = journal->j_running_transaction->t_tid;
prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait,
TASK_UNINTERRUPTIBLE);
need_to_start = !tid_geq(journal->j_commit_request, tid);
read_unlock(&journal->j_state_lock);
if (need_to_start)
jbd2_log_start_commit(journal, tid);
jbd2_might_wait_for_commit(journal);
schedule();
finish_wait(&journal->j_wait_transaction_locked, &wait);
}
/*
* Wait until running transaction transitions from T_SWITCH to T_FLUSH
* state and new transaction can thus be started. The function releases
* j_state_lock.
*/
static void wait_transaction_switching(journal_t *journal)
__releases(journal->j_state_lock)
{
DEFINE_WAIT(wait);
if (WARN_ON(!journal->j_running_transaction ||
journal->j_running_transaction->t_state != T_SWITCH)) {
read_unlock(&journal->j_state_lock);
return;
}
prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait,
TASK_UNINTERRUPTIBLE);
read_unlock(&journal->j_state_lock);
/*
* We don't call jbd2_might_wait_for_commit() here as there's no
* waiting for outstanding handles happening anymore in T_SWITCH state
* and handling of reserved handles actually relies on that for
* correctness.
*/
schedule();
finish_wait(&journal->j_wait_transaction_locked, &wait);
}
static void sub_reserved_credits(journal_t *journal, int blocks)
{
atomic_sub(blocks, &journal->j_reserved_credits);
wake_up(&journal->j_wait_reserved);
}
/*
* Wait until we can add credits for handle to the running transaction. Called
* with j_state_lock held for reading. Returns 0 if handle joined the running
* transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
* caller must retry.
*
* Note: because j_state_lock may be dropped depending on the return
* value, we need to fake out sparse so ti doesn't complain about a
* locking imbalance. Callers of add_transaction_credits will need to
* make a similar accomodation.
*/
static int add_transaction_credits(journal_t *journal, int blocks,
int rsv_blocks)
__must_hold(&journal->j_state_lock)
{
transaction_t *t = journal->j_running_transaction;
int needed;
int total = blocks + rsv_blocks;
/*
* If the current transaction is locked down for commit, wait
* for the lock to be released.
*/
if (t->t_state != T_RUNNING) {
WARN_ON_ONCE(t->t_state >= T_FLUSH);
wait_transaction_locked(journal);
__acquire(&journal->j_state_lock); /* fake out sparse */
return 1;
}
/*
* If there is not enough space left in the log to write all
* potential buffers requested by this operation, we need to
* stall pending a log checkpoint to free some more log space.
*/
needed = atomic_add_return(total, &t->t_outstanding_credits);
if (needed > journal->j_max_transaction_buffers) {
/*
* If the current transaction is already too large,
* then start to commit it: we can then go back and
* attach this handle to a new transaction.
*/
atomic_sub(total, &t->t_outstanding_credits);
/*
* Is the number of reserved credits in the current transaction too
* big to fit this handle? Wait until reserved credits are freed.
*/
if (atomic_read(&journal->j_reserved_credits) + total >
journal->j_max_transaction_buffers) {
read_unlock(&journal->j_state_lock);
jbd2_might_wait_for_commit(journal);
wait_event(journal->j_wait_reserved,
atomic_read(&journal->j_reserved_credits) + total <=
journal->j_max_transaction_buffers);
__acquire(&journal->j_state_lock); /* fake out sparse */
return 1;
}
wait_transaction_locked(journal);
__acquire(&journal->j_state_lock); /* fake out sparse */
return 1;
}
/*
* The commit code assumes that it can get enough log space
* without forcing a checkpoint. This is *critical* for
* correctness: a checkpoint of a buffer which is also
* associated with a committing transaction creates a deadlock,
* so commit simply cannot force through checkpoints.
*
* We must therefore ensure the necessary space in the journal
* *before* starting to dirty potentially checkpointed buffers
* in the new transaction.
*/
if (jbd2_log_space_left(journal) < journal->j_max_transaction_buffers) {
atomic_sub(total, &t->t_outstanding_credits);
read_unlock(&journal->j_state_lock);
jbd2_might_wait_for_commit(journal);
write_lock(&journal->j_state_lock);
if (jbd2_log_space_left(journal) <
journal->j_max_transaction_buffers)
__jbd2_log_wait_for_space(journal);
write_unlock(&journal->j_state_lock);
__acquire(&journal->j_state_lock); /* fake out sparse */
return 1;
}
/* No reservation? We are done... */
if (!rsv_blocks)
return 0;
needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
/* We allow at most half of a transaction to be reserved */
if (needed > journal->j_max_transaction_buffers / 2) {
sub_reserved_credits(journal, rsv_blocks);
atomic_sub(total, &t->t_outstanding_credits);
read_unlock(&journal->j_state_lock);
jbd2_might_wait_for_commit(journal);
wait_event(journal->j_wait_reserved,
atomic_read(&journal->j_reserved_credits) + rsv_blocks
<= journal->j_max_transaction_buffers / 2);
__acquire(&journal->j_state_lock); /* fake out sparse */
return 1;
}
return 0;
}
/*
* start_this_handle: Given a handle, deal with any locking or stalling
* needed to make sure that there is enough journal space for the handle
* to begin. Attach the handle to a transaction and set up the
* transaction's buffer credits.
*/
static int start_this_handle(journal_t *journal, handle_t *handle,
gfp_t gfp_mask)
{
transaction_t *transaction, *new_transaction = NULL;
int blocks = handle->h_total_credits;
int rsv_blocks = 0;
unsigned long ts = jiffies;
if (handle->h_rsv_handle)
rsv_blocks = handle->h_rsv_handle->h_total_credits;
/*
* Limit the number of reserved credits to 1/2 of maximum transaction
* size and limit the number of total credits to not exceed maximum
* transaction size per operation.
*/
if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
(rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
printk(KERN_ERR "JBD2: %s wants too many credits "
"credits:%d rsv_credits:%d max:%d\n",
current->comm, blocks, rsv_blocks,
journal->j_max_transaction_buffers);
WARN_ON(1);
return -ENOSPC;
}
alloc_transaction:
/*
* This check is racy but it is just an optimization of allocating new
* transaction early if there are high chances we'll need it. If we
* guess wrong, we'll retry or free unused transaction.
*/
if (!data_race(journal->j_running_transaction)) {
/*
* If __GFP_FS is not present, then we may be being called from
* inside the fs writeback layer, so we MUST NOT fail.
*/
if ((gfp_mask & __GFP_FS) == 0)
gfp_mask |= __GFP_NOFAIL;
new_transaction = kmem_cache_zalloc(transaction_cache,
gfp_mask);
if (!new_transaction)
return -ENOMEM;
}
jbd2_debug(3, "New handle %p going live.\n", handle);
/*
* We need to hold j_state_lock until t_updates has been incremented,
* for proper journal barrier handling
*/
repeat:
read_lock(&journal->j_state_lock);
BUG_ON(journal->j_flags & JBD2_UNMOUNT);
if (is_journal_aborted(journal) ||
(journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
read_unlock(&journal->j_state_lock);
jbd2_journal_free_transaction(new_transaction);
return -EROFS;
}
/*
* Wait on the journal's transaction barrier if necessary. Specifically
* we allow reserved handles to proceed because otherwise commit could
* deadlock on page writeback not being able to complete.
*/
if (!handle->h_reserved && journal->j_barrier_count) {
read_unlock(&journal->j_state_lock);
wait_event(journal->j_wait_transaction_locked,
journal->j_barrier_count == 0);
goto repeat;
}
if (!journal->j_running_transaction) {
read_unlock(&journal->j_state_lock);
if (!new_transaction)
goto alloc_transaction;
write_lock(&journal->j_state_lock);
if (!journal->j_running_transaction &&
(handle->h_reserved || !journal->j_barrier_count)) {
jbd2_get_transaction(journal, new_transaction);
new_transaction = NULL;
}
write_unlock(&journal->j_state_lock);
goto repeat;
}
transaction = journal->j_running_transaction;
if (!handle->h_reserved) {
/* We may have dropped j_state_lock - restart in that case */
if (add_transaction_credits(journal, blocks, rsv_blocks)) {
/*
* add_transaction_credits releases
* j_state_lock on a non-zero return
*/
__release(&journal->j_state_lock);
goto repeat;
}
} else {
/*
* We have handle reserved so we are allowed to join T_LOCKED
* transaction and we don't have to check for transaction size
* and journal space. But we still have to wait while running
* transaction is being switched to a committing one as it
* won't wait for any handles anymore.
*/
if (transaction->t_state == T_SWITCH) {
wait_transaction_switching(journal);
goto repeat;
}
sub_reserved_credits(journal, blocks);
handle->h_reserved = 0;
}
/* OK, account for the buffers that this operation expects to
* use and add the handle to the running transaction.
*/
update_t_max_wait(transaction, ts);
handle->h_transaction = transaction;
handle->h_requested_credits = blocks;
handle->h_revoke_credits_requested = handle->h_revoke_credits;
handle->h_start_jiffies = jiffies;
atomic_inc(&transaction->t_updates);
atomic_inc(&transaction->t_handle_count);
jbd2_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
handle, blocks,
atomic_read(&transaction->t_outstanding_credits),
jbd2_log_space_left(journal));
read_unlock(&journal->j_state_lock);
current->journal_info = handle;
rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
jbd2_journal_free_transaction(new_transaction);
/*
* Ensure that no allocations done while the transaction is open are
* going to recurse back to the fs layer.
*/
handle->saved_alloc_context = memalloc_nofs_save();
return 0;
}
/* Allocate a new handle. This should probably be in a slab... */
static handle_t *new_handle(int nblocks)
{
handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
if (!handle)
return NULL;
handle->h_total_credits = nblocks;
handle->h_ref = 1;
return handle;
}
handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
int revoke_records, gfp_t gfp_mask,
unsigned int type, unsigned int line_no)
{
handle_t *handle = journal_current_handle();
int err;
if (!journal)
return ERR_PTR(-EROFS);
if (handle) {
J_ASSERT(handle->h_transaction->t_journal == journal);
handle->h_ref++;
return handle;
}
nblocks += DIV_ROUND_UP(revoke_records,
journal->j_revoke_records_per_block);
handle = new_handle(nblocks);
if (!handle)
return ERR_PTR(-ENOMEM);
if (rsv_blocks) {
handle_t *rsv_handle;
rsv_handle = new_handle(rsv_blocks);
if (!rsv_handle) {
jbd2_free_handle(handle);
return ERR_PTR(-ENOMEM);
}
rsv_handle->h_reserved = 1;
rsv_handle->h_journal = journal;
handle->h_rsv_handle = rsv_handle;
}
handle->h_revoke_credits = revoke_records;
err = start_this_handle(journal, handle, gfp_mask);
if (err < 0) {
if (handle->h_rsv_handle)
jbd2_free_handle(handle->h_rsv_handle);
jbd2_free_handle(handle);
return ERR_PTR(err);
}
handle->h_type = type;
handle->h_line_no = line_no;
trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
handle->h_transaction->t_tid, type,
line_no, nblocks);
return handle;
}
EXPORT_SYMBOL(jbd2__journal_start);
/**
* jbd2_journal_start() - Obtain a new handle.
* @journal: Journal to start transaction on.
* @nblocks: number of block buffer we might modify
*
* We make sure that the transaction can guarantee at least nblocks of
* modified buffers in the log. We block until the log can guarantee
* that much space. Additionally, if rsv_blocks > 0, we also create another
* handle with rsv_blocks reserved blocks in the journal. This handle is
* stored in h_rsv_handle. It is not attached to any particular transaction
* and thus doesn't block transaction commit. If the caller uses this reserved
* handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
* on the parent handle will dispose the reserved one. Reserved handle has to
* be converted to a normal handle using jbd2_journal_start_reserved() before
* it can be used.
*
* Return a pointer to a newly allocated handle, or an ERR_PTR() value
* on failure.
*/
handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
{
return jbd2__journal_start(journal, nblocks, 0, 0, GFP_NOFS, 0, 0);
}
EXPORT_SYMBOL(jbd2_journal_start);
static void __jbd2_journal_unreserve_handle(handle_t *handle, transaction_t *t)
{
journal_t *journal = handle->h_journal;
WARN_ON(!handle->h_reserved);
sub_reserved_credits(journal, handle->h_total_credits);
if (t)
atomic_sub(handle->h_total_credits, &t->t_outstanding_credits);
}
void jbd2_journal_free_reserved(handle_t *handle)
{
journal_t *journal = handle->h_journal;
/* Get j_state_lock to pin running transaction if it exists */
read_lock(&journal->j_state_lock);
__jbd2_journal_unreserve_handle(handle, journal->j_running_transaction);
read_unlock(&journal->j_state_lock);
jbd2_free_handle(handle);
}
EXPORT_SYMBOL(jbd2_journal_free_reserved);
/**
* jbd2_journal_start_reserved() - start reserved handle
* @handle: handle to start
* @type: for handle statistics
* @line_no: for handle statistics
*
* Start handle that has been previously reserved with jbd2_journal_reserve().
* This attaches @handle to the running transaction (or creates one if there's
* not transaction running). Unlike jbd2_journal_start() this function cannot
* block on journal commit, checkpointing, or similar stuff. It can block on
* memory allocation or frozen journal though.
*
* Return 0 on success, non-zero on error - handle is freed in that case.
*/
int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
unsigned int line_no)
{
journal_t *journal = handle->h_journal;
int ret = -EIO;
if (WARN_ON(!handle->h_reserved)) {
/* Someone passed in normal handle? Just stop it. */
jbd2_journal_stop(handle);
return ret;
}
/*
* Usefulness of mixing of reserved and unreserved handles is
* questionable. So far nobody seems to need it so just error out.
*/
if (WARN_ON(current->journal_info)) {
jbd2_journal_free_reserved(handle);
return ret;
}
handle->h_journal = NULL;
/*
* GFP_NOFS is here because callers are likely from writeback or
* similarly constrained call sites
*/
ret = start_this_handle(journal, handle, GFP_NOFS);
if (ret < 0) {
handle->h_journal = journal;
jbd2_journal_free_reserved(handle);
return ret;
}
handle->h_type = type;
handle->h_line_no = line_no;
trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
handle->h_transaction->t_tid, type,
line_no, handle->h_total_credits);
return 0;
}
EXPORT_SYMBOL(jbd2_journal_start_reserved);
/**
* jbd2_journal_extend() - extend buffer credits.
* @handle: handle to 'extend'
* @nblocks: nr blocks to try to extend by.
* @revoke_records: number of revoke records to try to extend by.
*
* Some transactions, such as large extends and truncates, can be done
* atomically all at once or in several stages. The operation requests
* a credit for a number of buffer modifications in advance, but can
* extend its credit if it needs more.
*
* jbd2_journal_extend tries to give the running handle more buffer credits.
* It does not guarantee that allocation - this is a best-effort only.
* The calling process MUST be able to deal cleanly with a failure to
* extend here.
*
* Return 0 on success, non-zero on failure.
*
* return code < 0 implies an error
* return code > 0 implies normal transaction-full status.
*/
int jbd2_journal_extend(handle_t *handle, int nblocks, int revoke_records)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal;
int result;
int wanted;
if (is_handle_aborted(handle))
return -EROFS;
journal = transaction->t_journal;
result = 1;
read_lock(&journal->j_state_lock);
/* Don't extend a locked-down transaction! */
if (transaction->t_state != T_RUNNING) {
jbd2_debug(3, "denied handle %p %d blocks: "
"transaction not running\n", handle, nblocks);
goto error_out;
}
nblocks += DIV_ROUND_UP(
handle->h_revoke_credits_requested + revoke_records,
journal->j_revoke_records_per_block) -
DIV_ROUND_UP(
handle->h_revoke_credits_requested,
journal->j_revoke_records_per_block);
wanted = atomic_add_return(nblocks,
&transaction->t_outstanding_credits);
if (wanted > journal->j_max_transaction_buffers) {
jbd2_debug(3, "denied handle %p %d blocks: "
"transaction too large\n", handle, nblocks);
atomic_sub(nblocks, &transaction->t_outstanding_credits);
jbd2: kill t_handle_lock transaction spinlock This patch kills t_handle_lock transaction spinlock completely from jbd2. To explain the reasoning, currently there were three sites at which this spinlock was used. 1. jbd2_journal_wait_updates() a. Based on careful code review it can be seen that, we don't need this lock here. This is since we wait for any currently ongoing updates based on a atomic variable t_updates. And we anyway don't take any t_handle_lock while in stop_this_handle(). i.e. write_lock(&journal->j_state_lock() jbd2_journal_wait_updates() stop_this_handle() while (atomic_read(txn->t_updates) { | DEFINE_WAIT(wait); | prepare_to_wait(); | if (atomic_read(txn->t_updates) if (atomic_dec_and_test(txn->t_updates)) write_unlock(&journal->j_state_lock); schedule(); wake_up() write_lock(&journal->j_state_lock); finish_wait(); } txn->t_state = T_COMMIT write_unlock(&journal->j_state_lock); b. Also note that between atomic_inc(&txn->t_updates) in start_this_handle() and jbd2_journal_wait_updates(), the synchronization happens via read_lock(journal->j_state_lock) in start_this_handle(); 2. jbd2_journal_extend() a. jbd2_journal_extend() is called with the handle of each process from task_struct. So no lock required in updating member fields of handle_t b. For member fields of h_transaction, all updates happens only via atomic APIs (which is also within read_lock()). So, no need of this transaction spinlock. 3. update_t_max_wait() Based on Jan suggestion, this can be carefully removed using atomic cmpxchg API. Note that there can be several processes which are waiting for a new transaction to be allocated and started. For doing this only one process will succeed in taking write_lock() and allocating a new txn. After that all of the process will be updating the t_max_wait (max transaction wait time). This can be done via below method w/o taking any locks using atomic cmpxchg. For more details refer [1] new = get_new_val(); old = READ_ONCE(ptr->max_val); while (old < new) old = cmpxchg(&ptr->max_val, old, new); [1]: https://lwn.net/Articles/849237/ Suggested-by: Jan Kara <jack@suse.cz> Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/d89e599658b4a1f3893a48c6feded200073037fc.1644992076.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-16 07:00:35 +00:00
goto error_out;
}
trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
transaction->t_tid,
handle->h_type, handle->h_line_no,
handle->h_total_credits,
nblocks);
handle->h_total_credits += nblocks;
handle->h_requested_credits += nblocks;
handle->h_revoke_credits += revoke_records;
handle->h_revoke_credits_requested += revoke_records;
result = 0;
jbd2_debug(3, "extended handle %p by %d\n", handle, nblocks);
error_out:
read_unlock(&journal->j_state_lock);
return result;
}
static void stop_this_handle(handle_t *handle)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
int revokes;
J_ASSERT(journal_current_handle() == handle);
J_ASSERT(atomic_read(&transaction->t_updates) > 0);
current->journal_info = NULL;
/*
* Subtract necessary revoke descriptor blocks from handle credits. We
* take care to account only for revoke descriptor blocks the
* transaction will really need as large sequences of transactions with
* small numbers of revokes are relatively common.
*/
revokes = handle->h_revoke_credits_requested - handle->h_revoke_credits;
if (revokes) {
int t_revokes, revoke_descriptors;
int rr_per_blk = journal->j_revoke_records_per_block;
WARN_ON_ONCE(DIV_ROUND_UP(revokes, rr_per_blk)
> handle->h_total_credits);
t_revokes = atomic_add_return(revokes,
&transaction->t_outstanding_revokes);
revoke_descriptors =
DIV_ROUND_UP(t_revokes, rr_per_blk) -
DIV_ROUND_UP(t_revokes - revokes, rr_per_blk);
handle->h_total_credits -= revoke_descriptors;
}
atomic_sub(handle->h_total_credits,
&transaction->t_outstanding_credits);
if (handle->h_rsv_handle)
__jbd2_journal_unreserve_handle(handle->h_rsv_handle,
transaction);
if (atomic_dec_and_test(&transaction->t_updates))
wake_up(&journal->j_wait_updates);
This merge window saw the the following new featuers added to ext4: * Direct I/O via iomap (required the iomap-for-next branch from Darrick as a prereq). * Support for using dioread-nolock where the block size < page size. * Support for encryption for file systems where the block size < page size. * Rework of journal credits handling so a revoke-heavy workload will not cause the journal to run out of space. * Replace bit-spinlocks with spinlocks in jbd2 Also included were some bug fixes and cleanups, mostly to clean up corner cases from fuzzed file systems and error path handling. -----BEGIN PGP SIGNATURE----- iQEzBAABCAAdFiEEK2m5VNv+CHkogTfJ8vlZVpUNgaMFAl3dHxoACgkQ8vlZVpUN gaMZswf5AbtQhTEJDXO7Pc1ull38GIGFgAv7uAth0TymLC3h1/FEYWW0crEPFsDr 1Eei55UUVOYrMMUKQ4P7wlLX0cIh3XDPMWnRFuqBoV5/ZOsH/ZSbkY//TG2Xze/v 9wXIH/RKQnzbRtXffJ1+DnvmXJk+HFm1R1gjl0nfyUXGrnlSfqJxhLSczyd6bJJq ehi/tso5UC/4EQsAIdWp7VWsAdaHcZ7ogHqDoy8dXpM1equ408iml7VlKr8R+Nr7 5ANpCISXChSlLLYm0NYN5vhO8upF5uDxWLdCtxVPL5kFdM2m/ELjXw9h9C+78l7C EWJGlGlxvx07Px+e+bfStEsoixpWBg== =0eko -----END PGP SIGNATURE----- Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4 Pull ext4 updates from Ted Ts'o: "This merge window saw the the following new featuers added to ext4: - Direct I/O via iomap (required the iomap-for-next branch from Darrick as a prereq). - Support for using dioread-nolock where the block size < page size. - Support for encryption for file systems where the block size < page size. - Rework of journal credits handling so a revoke-heavy workload will not cause the journal to run out of space. - Replace bit-spinlocks with spinlocks in jbd2 Also included were some bug fixes and cleanups, mostly to clean up corner cases from fuzzed file systems and error path handling" * tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (59 commits) ext4: work around deleting a file with i_nlink == 0 safely ext4: add more paranoia checking in ext4_expand_extra_isize handling jbd2: make jbd2_handle_buffer_credits() handle reserved handles ext4: fix a bug in ext4_wait_for_tail_page_commit ext4: bio_alloc with __GFP_DIRECT_RECLAIM never fails ext4: code cleanup for get_next_id ext4: fix leak of quota reservations ext4: remove unused variable warning in parse_options() ext4: Enable encryption for subpage-sized blocks fs/buffer.c: support fscrypt in block_read_full_page() ext4: Add error handling for io_end_vec struct allocation jbd2: Fine tune estimate of necessary descriptor blocks jbd2: Provide trace event for handle restarts ext4: Reserve revoke credits for freed blocks jbd2: Make credit checking more strict jbd2: Rename h_buffer_credits to h_total_credits jbd2: Reserve space for revoke descriptor blocks jbd2: Drop jbd2_space_needed() jbd2: Account descriptor blocks into t_outstanding_credits jbd2: Factor out common parts of stopping and restarting a handle ...
2019-11-30 18:53:02 +00:00
rwsem_release(&journal->j_trans_commit_map, _THIS_IP_);
/*
* Scope of the GFP_NOFS context is over here and so we can restore the
* original alloc context.
*/
memalloc_nofs_restore(handle->saved_alloc_context);
}
/**
* jbd2__journal_restart() - restart a handle .
* @handle: handle to restart
* @nblocks: nr credits requested
* @revoke_records: number of revoke record credits requested
* @gfp_mask: memory allocation flags (for start_this_handle)
*
* Restart a handle for a multi-transaction filesystem
* operation.
*
* If the jbd2_journal_extend() call above fails to grant new buffer credits
* to a running handle, a call to jbd2_journal_restart will commit the
* handle's transaction so far and reattach the handle to a new
* transaction capable of guaranteeing the requested number of
* credits. We preserve reserved handle if there's any attached to the
* passed in handle.
*/
int jbd2__journal_restart(handle_t *handle, int nblocks, int revoke_records,
gfp_t gfp_mask)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal;
tid_t tid;
int need_to_start;
int ret;
/* If we've had an abort of any type, don't even think about
* actually doing the restart! */
if (is_handle_aborted(handle))
return 0;
journal = transaction->t_journal;
tid = transaction->t_tid;
/*
* First unlink the handle from its current transaction, and start the
* commit on that.
*/
jbd2_debug(2, "restarting handle %p\n", handle);
stop_this_handle(handle);
handle->h_transaction = NULL;
/*
* TODO: If we use READ_ONCE / WRITE_ONCE for j_commit_request we can
* get rid of pointless j_state_lock traffic like this.
*/
read_lock(&journal->j_state_lock);
need_to_start = !tid_geq(journal->j_commit_request, tid);
read_unlock(&journal->j_state_lock);
if (need_to_start)
jbd2_log_start_commit(journal, tid);
handle->h_total_credits = nblocks +
DIV_ROUND_UP(revoke_records,
journal->j_revoke_records_per_block);
handle->h_revoke_credits = revoke_records;
ret = start_this_handle(journal, handle, gfp_mask);
trace_jbd2_handle_restart(journal->j_fs_dev->bd_dev,
ret ? 0 : handle->h_transaction->t_tid,
handle->h_type, handle->h_line_no,
handle->h_total_credits);
return ret;
}
EXPORT_SYMBOL(jbd2__journal_restart);
int jbd2_journal_restart(handle_t *handle, int nblocks)
{
return jbd2__journal_restart(handle, nblocks, 0, GFP_NOFS);
}
EXPORT_SYMBOL(jbd2_journal_restart);
/*
* Waits for any outstanding t_updates to finish.
* This is called with write j_state_lock held.
*/
void jbd2_journal_wait_updates(journal_t *journal)
{
DEFINE_WAIT(wait);
while (1) {
/*
* Note that the running transaction can get freed under us if
* this transaction is getting committed in
* jbd2_journal_commit_transaction() ->
* jbd2_journal_free_transaction(). This can only happen when we
* release j_state_lock -> schedule() -> acquire j_state_lock.
* Hence we should everytime retrieve new j_running_transaction
* value (after j_state_lock release acquire cycle), else it may
* lead to use-after-free of old freed transaction.
*/
transaction_t *transaction = journal->j_running_transaction;
if (!transaction)
break;
prepare_to_wait(&journal->j_wait_updates, &wait,
TASK_UNINTERRUPTIBLE);
if (!atomic_read(&transaction->t_updates)) {
finish_wait(&journal->j_wait_updates, &wait);
break;
}
write_unlock(&journal->j_state_lock);
schedule();
finish_wait(&journal->j_wait_updates, &wait);
write_lock(&journal->j_state_lock);
}
}
/**
* jbd2_journal_lock_updates () - establish a transaction barrier.
* @journal: Journal to establish a barrier on.
*
* This locks out any further updates from being started, and blocks
* until all existing updates have completed, returning only once the
* journal is in a quiescent state with no updates running.
*
* The journal lock should not be held on entry.
*/
void jbd2_journal_lock_updates(journal_t *journal)
{
jbd2_might_wait_for_commit(journal);
write_lock(&journal->j_state_lock);
++journal->j_barrier_count;
/* Wait until there are no reserved handles */
if (atomic_read(&journal->j_reserved_credits)) {
write_unlock(&journal->j_state_lock);
wait_event(journal->j_wait_reserved,
atomic_read(&journal->j_reserved_credits) == 0);
write_lock(&journal->j_state_lock);
}
/* Wait until there are no running t_updates */
jbd2_journal_wait_updates(journal);
write_unlock(&journal->j_state_lock);
/*
* We have now established a barrier against other normal updates, but
* we also need to barrier against other jbd2_journal_lock_updates() calls
* to make sure that we serialise special journal-locked operations
* too.
*/
mutex_lock(&journal->j_barrier);
}
/**
* jbd2_journal_unlock_updates () - release barrier
* @journal: Journal to release the barrier on.
*
* Release a transaction barrier obtained with jbd2_journal_lock_updates().
*
* Should be called without the journal lock held.
*/
void jbd2_journal_unlock_updates (journal_t *journal)
{
J_ASSERT(journal->j_barrier_count != 0);
mutex_unlock(&journal->j_barrier);
write_lock(&journal->j_state_lock);
--journal->j_barrier_count;
write_unlock(&journal->j_state_lock);
wake_up_all(&journal->j_wait_transaction_locked);
}
static void warn_dirty_buffer(struct buffer_head *bh)
{
printk(KERN_WARNING
"JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
"There's a risk of filesystem corruption in case of system "
"crash.\n",
bh->b_bdev, (unsigned long long)bh->b_blocknr);
}
/* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
static void jbd2_freeze_jh_data(struct journal_head *jh)
{
char *source;
struct buffer_head *bh = jh2bh(jh);
J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
source = kmap_local_folio(bh->b_folio, bh_offset(bh));
/* Fire data frozen trigger just before we copy the data */
jbd2_buffer_frozen_trigger(jh, source, jh->b_triggers);
memcpy(jh->b_frozen_data, source, bh->b_size);
kunmap_local(source);
/*
* Now that the frozen data is saved off, we need to store any matching
* triggers.
*/
jh->b_frozen_triggers = jh->b_triggers;
}
/*
* If the buffer is already part of the current transaction, then there
* is nothing we need to do. If it is already part of a prior
* transaction which we are still committing to disk, then we need to
* make sure that we do not overwrite the old copy: we do copy-out to
* preserve the copy going to disk. We also account the buffer against
* the handle's metadata buffer credits (unless the buffer is already
* part of the transaction, that is).
*
*/
static int
do_get_write_access(handle_t *handle, struct journal_head *jh,
int force_copy)
{
struct buffer_head *bh;
transaction_t *transaction = handle->h_transaction;
journal_t *journal;
int error;
char *frozen_buffer = NULL;
unsigned long start_lock, time_lock;
journal = transaction->t_journal;
jbd2_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
JBUFFER_TRACE(jh, "entry");
repeat:
bh = jh2bh(jh);
/* @@@ Need to check for errors here at some point. */
start_lock = jiffies;
lock_buffer(bh);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
/* If it takes too long to lock the buffer, trace it */
time_lock = jbd2_time_diff(start_lock, jiffies);
if (time_lock > HZ/10)
trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
jiffies_to_msecs(time_lock));
/* We now hold the buffer lock so it is safe to query the buffer
* state. Is the buffer dirty?
*
* If so, there are two possibilities. The buffer may be
* non-journaled, and undergoing a quite legitimate writeback.
* Otherwise, it is journaled, and we don't expect dirty buffers
* in that state (the buffers should be marked JBD_Dirty
* instead.) So either the IO is being done under our own
* control and this is a bug, or it's a third party IO such as
* dump(8) (which may leave the buffer scheduled for read ---
* ie. locked but not dirty) or tune2fs (which may actually have
* the buffer dirtied, ugh.) */
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
if (buffer_dirty(bh) && jh->b_transaction) {
warn_dirty_buffer(bh);
/*
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
* We need to clean the dirty flag and we must do it under the
* buffer lock to be sure we don't race with running write-out.
*/
JBUFFER_TRACE(jh, "Journalling dirty buffer");
clear_buffer_dirty(bh);
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
/*
* The buffer is going to be added to BJ_Reserved list now and
* nothing guarantees jbd2_journal_dirty_metadata() will be
* ever called for it. So we need to set jbddirty bit here to
* make sure the buffer is dirtied and written out when the
* journaling machinery is done with it.
*/
set_buffer_jbddirty(bh);
}
error = -EROFS;
if (is_handle_aborted(handle)) {
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
unlock_buffer(bh);
goto out;
}
error = 0;
/*
* The buffer is already part of this transaction if b_transaction or
* b_next_transaction points to it
*/
if (jh->b_transaction == transaction ||
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
jh->b_next_transaction == transaction) {
unlock_buffer(bh);
goto done;
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
}
/*
* this is the first time this transaction is touching this buffer,
* reset the modified flag
*/
jh->b_modified = 0;
/*
* If the buffer is not journaled right now, we need to make sure it
* doesn't get written to disk before the caller actually commits the
* new data
*/
if (!jh->b_transaction) {
JBUFFER_TRACE(jh, "no transaction");
J_ASSERT_JH(jh, !jh->b_next_transaction);
JBUFFER_TRACE(jh, "file as BJ_Reserved");
/*
* Make sure all stores to jh (b_modified, b_frozen_data) are
* visible before attaching it to the running transaction.
* Paired with barrier in jbd2_write_access_granted()
*/
smp_wmb();
spin_lock(&journal->j_list_lock);
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
if (test_clear_buffer_dirty(bh)) {
/*
* Execute buffer dirty clearing and jh->b_transaction
* assignment under journal->j_list_lock locked to
* prevent bh being removed from checkpoint list if
* the buffer is in an intermediate state (not dirty
* and jh->b_transaction is NULL).
*/
JBUFFER_TRACE(jh, "Journalling dirty buffer");
set_buffer_jbddirty(bh);
}
__jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
spin_unlock(&journal->j_list_lock);
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
unlock_buffer(bh);
goto done;
}
jbd2: fix data missing when reusing bh which is ready to be checkpointed Following process will make data lost and could lead to a filesystem corrupted problem: 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and jh->b_transaction = NULL 2. T1 is added into journal->j_checkpoint_transactions. 3. Get bh prepare to write while doing checkpoing: PA PB do_get_write_access jbd2_log_do_checkpoint spin_lock(&jh->b_state_lock) if (buffer_dirty(bh)) clear_buffer_dirty(bh) // clear buffer dirty set_buffer_jbddirty(bh) transaction = journal->j_checkpoint_transactions jh = transaction->t_checkpoint_list if (!buffer_dirty(bh)) __jbd2_journal_remove_checkpoint(jh) // bh won't be flushed jbd2_cleanup_journal_tail __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved) 4. Aborting journal/Power-cut before writing latest bh on journal area. In this way we get a corrupted filesystem with bh's data lost. Fix it by moving the clearing of buffer_dirty bit just before the call to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction assignment are under journal->j_list_lock locked, so that jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed even bh is currently not dirty. And journal_shrink_one_cp_list() won't remove jh from checkpoint list if the buffer head is reused in do_get_write_access(). Fetch a reproducer in [Link]. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898 Cc: <stable@kernel.org> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com> Suggested-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20230110015327.1181863-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2023-01-10 01:53:27 +00:00
unlock_buffer(bh);
/*
* If there is already a copy-out version of this buffer, then we don't
* need to make another one
*/
if (jh->b_frozen_data) {
JBUFFER_TRACE(jh, "has frozen data");
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
goto attach_next;
}
JBUFFER_TRACE(jh, "owned by older transaction");
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
/*
* There is one case we have to be very careful about. If the
* committing transaction is currently writing this buffer out to disk
* and has NOT made a copy-out, then we cannot modify the buffer
* contents at all right now. The essence of copy-out is that it is
* the extra copy, not the primary copy, which gets journaled. If the
* primary copy is already going to disk then we cannot do copy-out
* here.
*/
if (buffer_shadow(bh)) {
JBUFFER_TRACE(jh, "on shadow: sleep");
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
goto repeat;
}
/*
* Only do the copy if the currently-owning transaction still needs it.
* If buffer isn't on BJ_Metadata list, the committing transaction is
* past that stage (here we use the fact that BH_Shadow is set under
* bh_state lock together with refiling to BJ_Shadow list and at this
* point we know the buffer doesn't have BH_Shadow set).
*
* Subtle point, though: if this is a get_undo_access, then we will be
* relying on the frozen_data to contain the new value of the
* committed_data record after the transaction, so we HAVE to force the
* frozen_data copy in that case.
*/
if (jh->b_jlist == BJ_Metadata || force_copy) {
JBUFFER_TRACE(jh, "generate frozen data");
if (!frozen_buffer) {
JBUFFER_TRACE(jh, "allocate memory for buffer");
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
GFP_NOFS | __GFP_NOFAIL);
goto repeat;
}
jh->b_frozen_data = frozen_buffer;
frozen_buffer = NULL;
jbd2_freeze_jh_data(jh);
}
attach_next:
/*
* Make sure all stores to jh (b_modified, b_frozen_data) are visible
* before attaching it to the running transaction. Paired with barrier
* in jbd2_write_access_granted()
*/
smp_wmb();
jh->b_next_transaction = transaction;
done:
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
/*
* If we are about to journal a buffer, then any revoke pending on it is
* no longer valid
*/
jbd2_journal_cancel_revoke(handle, jh);
out:
if (unlikely(frozen_buffer)) /* It's usually NULL */
jbd2_free(frozen_buffer, bh->b_size);
JBUFFER_TRACE(jh, "exit");
return error;
}
/* Fast check whether buffer is already attached to the required transaction */
jbd2: fix null committed data return in undo_access introduced jbd2_write_access_granted() to improve write|undo_access speed, but missed to check the status of b_committed_data which caused a kernel panic on ocfs2. [ 6538.405938] ------------[ cut here ]------------ [ 6538.406686] kernel BUG at fs/ocfs2/suballoc.c:2400! [ 6538.406686] invalid opcode: 0000 [#1] SMP [ 6538.406686] Modules linked in: ocfs2 nfsd lockd grace nfs_acl auth_rpcgss sunrpc autofs4 ocfs2_dlmfs ocfs2_stack_o2cb ocfs2_dlm ocfs2_nodemanager ocfs2_stackglue configfs sd_mod sg ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables be2iscsi iscsi_boot_sysfs bnx2i cnic uio cxgb4i cxgb4 cxgb3i libcxgbi cxgb3 mdio ib_iser rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr ipv6 iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi ppdev xen_kbdfront xen_netfront xen_fbfront parport_pc parport pcspkr i2c_piix4 acpi_cpufreq ext4 jbd2 mbcache xen_blkfront floppy pata_acpi ata_generic ata_piix cirrus ttm drm_kms_helper drm fb_sys_fops sysimgblt sysfillrect i2c_core syscopyarea dm_mirror dm_region_hash dm_log dm_mod [ 6538.406686] CPU: 1 PID: 16265 Comm: mmap_truncate Not tainted 4.3.0 #1 [ 6538.406686] Hardware name: Xen HVM domU, BIOS 4.3.1OVM 05/14/2014 [ 6538.406686] task: ffff88007c2bab00 ti: ffff880075b78000 task.ti: ffff880075b78000 [ 6538.406686] RIP: 0010:[<ffffffffa06a286b>] [<ffffffffa06a286b>] ocfs2_block_group_clear_bits+0x23b/0x250 [ocfs2] [ 6538.406686] RSP: 0018:ffff880075b7b7f8 EFLAGS: 00010246 [ 6538.406686] RAX: ffff8800760c5b40 RBX: ffff88006c06a000 RCX: ffffffffa06e6df0 [ 6538.406686] RDX: 0000000000000000 RSI: ffff88007a6f6ea0 RDI: ffff88007a760430 [ 6538.406686] RBP: ffff880075b7b878 R08: 0000000000000002 R09: 0000000000000001 [ 6538.406686] R10: ffffffffa06769be R11: 0000000000000000 R12: 0000000000000001 [ 6538.406686] R13: ffffffffa06a1750 R14: 0000000000000001 R15: ffff88007a6f6ea0 [ 6538.406686] FS: 00007f17fde30720(0000) GS:ffff88007f040000(0000) knlGS:0000000000000000 [ 6538.406686] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 6538.406686] CR2: 0000000000601730 CR3: 000000007aea0000 CR4: 00000000000406e0 [ 6538.406686] Stack: [ 6538.406686] ffff88007c2bb5b0 ffff880075b7b8e0 ffff88007a7604b0 ffff88006c640800 [ 6538.406686] ffff88007a7604b0 ffff880075d77390 0000000075b7b878 ffffffffa06a309d [ 6538.406686] ffff880075d752d8 ffff880075b7b990 ffff880075b7b898 0000000000000000 [ 6538.406686] Call Trace: [ 6538.406686] [<ffffffffa06a309d>] ? ocfs2_read_group_descriptor+0x6d/0xa0 [ocfs2] [ 6538.406686] [<ffffffffa06a3654>] _ocfs2_free_suballoc_bits+0xe4/0x320 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa06a397e>] _ocfs2_free_clusters+0xee/0x210 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa0682d50>] ? ocfs2_extend_trans+0x50/0x1a0 [ocfs2] [ 6538.406686] [<ffffffffa06a3ad5>] ocfs2_free_clusters+0x15/0x20 [ocfs2] [ 6538.406686] [<ffffffffa065072c>] ocfs2_replay_truncate_records+0xfc/0x290 [ocfs2] [ 6538.406686] [<ffffffffa06843ac>] ? ocfs2_start_trans+0xec/0x1d0 [ocfs2] [ 6538.406686] [<ffffffffa0654600>] __ocfs2_flush_truncate_log+0x140/0x2d0 [ocfs2] [ 6538.406686] [<ffffffffa0654394>] ? ocfs2_reserve_blocks_for_rec_trunc.clone.0+0x44/0x170 [ocfs2] [ 6538.406686] [<ffffffffa065acd4>] ocfs2_remove_btree_range+0x374/0x630 [ocfs2] [ 6538.406686] [<ffffffffa017486b>] ? jbd2_journal_stop+0x25b/0x470 [jbd2] [ 6538.406686] [<ffffffffa065d5b5>] ocfs2_commit_truncate+0x305/0x670 [ocfs2] [ 6538.406686] [<ffffffffa0683430>] ? ocfs2_journal_access_eb+0x20/0x20 [ocfs2] [ 6538.406686] [<ffffffffa067adb7>] ocfs2_truncate_file+0x297/0x380 [ocfs2] [ 6538.406686] [<ffffffffa01759e4>] ? jbd2_journal_begin_ordered_truncate+0x64/0xc0 [jbd2] [ 6538.406686] [<ffffffffa067c7a2>] ocfs2_setattr+0x572/0x860 [ocfs2] [ 6538.406686] [<ffffffff810e4a3f>] ? current_fs_time+0x3f/0x50 [ 6538.406686] [<ffffffff812124b7>] notify_change+0x1d7/0x340 [ 6538.406686] [<ffffffff8121abf9>] ? generic_getxattr+0x79/0x80 [ 6538.406686] [<ffffffff811f5876>] do_truncate+0x66/0x90 [ 6538.406686] [<ffffffff81120e30>] ? __audit_syscall_entry+0xb0/0x110 [ 6538.406686] [<ffffffff811f5bb3>] do_sys_ftruncate.clone.0+0xf3/0x120 [ 6538.406686] [<ffffffff811f5bee>] SyS_ftruncate+0xe/0x10 [ 6538.406686] [<ffffffff816aa2ae>] entry_SYSCALL_64_fastpath+0x12/0x71 [ 6538.406686] Code: 28 48 81 ee b0 04 00 00 48 8b 92 50 fb ff ff 48 8b 80 b0 03 00 00 48 39 90 88 00 00 00 0f 84 30 fe ff ff 0f 0b eb fe 0f 0b eb fe <0f> 0b 0f 1f 00 eb fb 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00 [ 6538.406686] RIP [<ffffffffa06a286b>] ocfs2_block_group_clear_bits+0x23b/0x250 [ocfs2] [ 6538.406686] RSP <ffff880075b7b7f8> [ 6538.691128] ---[ end trace 31cd7011d6770d7e ]--- [ 6538.694492] Kernel panic - not syncing: Fatal exception [ 6538.695484] Kernel Offset: disabled Fixes: de92c8caf16c("jbd2: speedup jbd2_journal_get_[write|undo]_access()") Cc: <stable@vger.kernel.org> Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2015-12-04 17:29:28 +00:00
static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
bool undo)
{
struct journal_head *jh;
bool ret = false;
/* Dirty buffers require special handling... */
if (buffer_dirty(bh))
return false;
/*
* RCU protects us from dereferencing freed pages. So the checks we do
* are guaranteed not to oops. However the jh slab object can get freed
* & reallocated while we work with it. So we have to be careful. When
* we see jh attached to the running transaction, we know it must stay
* so until the transaction is committed. Thus jh won't be freed and
* will be attached to the same bh while we run. However it can
* happen jh gets freed, reallocated, and attached to the transaction
* just after we get pointer to it from bh. So we have to be careful
* and recheck jh still belongs to our bh before we return success.
*/
rcu_read_lock();
if (!buffer_jbd(bh))
goto out;
/* This should be bh2jh() but that doesn't work with inline functions */
jh = READ_ONCE(bh->b_private);
if (!jh)
goto out;
jbd2: fix null committed data return in undo_access introduced jbd2_write_access_granted() to improve write|undo_access speed, but missed to check the status of b_committed_data which caused a kernel panic on ocfs2. [ 6538.405938] ------------[ cut here ]------------ [ 6538.406686] kernel BUG at fs/ocfs2/suballoc.c:2400! [ 6538.406686] invalid opcode: 0000 [#1] SMP [ 6538.406686] Modules linked in: ocfs2 nfsd lockd grace nfs_acl auth_rpcgss sunrpc autofs4 ocfs2_dlmfs ocfs2_stack_o2cb ocfs2_dlm ocfs2_nodemanager ocfs2_stackglue configfs sd_mod sg ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables be2iscsi iscsi_boot_sysfs bnx2i cnic uio cxgb4i cxgb4 cxgb3i libcxgbi cxgb3 mdio ib_iser rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr ipv6 iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi ppdev xen_kbdfront xen_netfront xen_fbfront parport_pc parport pcspkr i2c_piix4 acpi_cpufreq ext4 jbd2 mbcache xen_blkfront floppy pata_acpi ata_generic ata_piix cirrus ttm drm_kms_helper drm fb_sys_fops sysimgblt sysfillrect i2c_core syscopyarea dm_mirror dm_region_hash dm_log dm_mod [ 6538.406686] CPU: 1 PID: 16265 Comm: mmap_truncate Not tainted 4.3.0 #1 [ 6538.406686] Hardware name: Xen HVM domU, BIOS 4.3.1OVM 05/14/2014 [ 6538.406686] task: ffff88007c2bab00 ti: ffff880075b78000 task.ti: ffff880075b78000 [ 6538.406686] RIP: 0010:[<ffffffffa06a286b>] [<ffffffffa06a286b>] ocfs2_block_group_clear_bits+0x23b/0x250 [ocfs2] [ 6538.406686] RSP: 0018:ffff880075b7b7f8 EFLAGS: 00010246 [ 6538.406686] RAX: ffff8800760c5b40 RBX: ffff88006c06a000 RCX: ffffffffa06e6df0 [ 6538.406686] RDX: 0000000000000000 RSI: ffff88007a6f6ea0 RDI: ffff88007a760430 [ 6538.406686] RBP: ffff880075b7b878 R08: 0000000000000002 R09: 0000000000000001 [ 6538.406686] R10: ffffffffa06769be R11: 0000000000000000 R12: 0000000000000001 [ 6538.406686] R13: ffffffffa06a1750 R14: 0000000000000001 R15: ffff88007a6f6ea0 [ 6538.406686] FS: 00007f17fde30720(0000) GS:ffff88007f040000(0000) knlGS:0000000000000000 [ 6538.406686] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 6538.406686] CR2: 0000000000601730 CR3: 000000007aea0000 CR4: 00000000000406e0 [ 6538.406686] Stack: [ 6538.406686] ffff88007c2bb5b0 ffff880075b7b8e0 ffff88007a7604b0 ffff88006c640800 [ 6538.406686] ffff88007a7604b0 ffff880075d77390 0000000075b7b878 ffffffffa06a309d [ 6538.406686] ffff880075d752d8 ffff880075b7b990 ffff880075b7b898 0000000000000000 [ 6538.406686] Call Trace: [ 6538.406686] [<ffffffffa06a309d>] ? ocfs2_read_group_descriptor+0x6d/0xa0 [ocfs2] [ 6538.406686] [<ffffffffa06a3654>] _ocfs2_free_suballoc_bits+0xe4/0x320 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa06a397e>] _ocfs2_free_clusters+0xee/0x210 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa0682d50>] ? ocfs2_extend_trans+0x50/0x1a0 [ocfs2] [ 6538.406686] [<ffffffffa06a3ad5>] ocfs2_free_clusters+0x15/0x20 [ocfs2] [ 6538.406686] [<ffffffffa065072c>] ocfs2_replay_truncate_records+0xfc/0x290 [ocfs2] [ 6538.406686] [<ffffffffa06843ac>] ? ocfs2_start_trans+0xec/0x1d0 [ocfs2] [ 6538.406686] [<ffffffffa0654600>] __ocfs2_flush_truncate_log+0x140/0x2d0 [ocfs2] [ 6538.406686] [<ffffffffa0654394>] ? ocfs2_reserve_blocks_for_rec_trunc.clone.0+0x44/0x170 [ocfs2] [ 6538.406686] [<ffffffffa065acd4>] ocfs2_remove_btree_range+0x374/0x630 [ocfs2] [ 6538.406686] [<ffffffffa017486b>] ? jbd2_journal_stop+0x25b/0x470 [jbd2] [ 6538.406686] [<ffffffffa065d5b5>] ocfs2_commit_truncate+0x305/0x670 [ocfs2] [ 6538.406686] [<ffffffffa0683430>] ? ocfs2_journal_access_eb+0x20/0x20 [ocfs2] [ 6538.406686] [<ffffffffa067adb7>] ocfs2_truncate_file+0x297/0x380 [ocfs2] [ 6538.406686] [<ffffffffa01759e4>] ? jbd2_journal_begin_ordered_truncate+0x64/0xc0 [jbd2] [ 6538.406686] [<ffffffffa067c7a2>] ocfs2_setattr+0x572/0x860 [ocfs2] [ 6538.406686] [<ffffffff810e4a3f>] ? current_fs_time+0x3f/0x50 [ 6538.406686] [<ffffffff812124b7>] notify_change+0x1d7/0x340 [ 6538.406686] [<ffffffff8121abf9>] ? generic_getxattr+0x79/0x80 [ 6538.406686] [<ffffffff811f5876>] do_truncate+0x66/0x90 [ 6538.406686] [<ffffffff81120e30>] ? __audit_syscall_entry+0xb0/0x110 [ 6538.406686] [<ffffffff811f5bb3>] do_sys_ftruncate.clone.0+0xf3/0x120 [ 6538.406686] [<ffffffff811f5bee>] SyS_ftruncate+0xe/0x10 [ 6538.406686] [<ffffffff816aa2ae>] entry_SYSCALL_64_fastpath+0x12/0x71 [ 6538.406686] Code: 28 48 81 ee b0 04 00 00 48 8b 92 50 fb ff ff 48 8b 80 b0 03 00 00 48 39 90 88 00 00 00 0f 84 30 fe ff ff 0f 0b eb fe 0f 0b eb fe <0f> 0b 0f 1f 00 eb fb 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00 [ 6538.406686] RIP [<ffffffffa06a286b>] ocfs2_block_group_clear_bits+0x23b/0x250 [ocfs2] [ 6538.406686] RSP <ffff880075b7b7f8> [ 6538.691128] ---[ end trace 31cd7011d6770d7e ]--- [ 6538.694492] Kernel panic - not syncing: Fatal exception [ 6538.695484] Kernel Offset: disabled Fixes: de92c8caf16c("jbd2: speedup jbd2_journal_get_[write|undo]_access()") Cc: <stable@vger.kernel.org> Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2015-12-04 17:29:28 +00:00
/* For undo access buffer must have data copied */
if (undo && !jh->b_committed_data)
goto out;
jbd2: fix data races at struct journal_head journal_head::b_transaction and journal_head::b_next_transaction could be accessed concurrently as noticed by KCSAN, LTP: starting fsync04 /dev/zero: Can't open blockdev EXT4-fs (loop0): mounting ext3 file system using the ext4 subsystem EXT4-fs (loop0): mounted filesystem with ordered data mode. Opts: (null) ================================================================== BUG: KCSAN: data-race in __jbd2_journal_refile_buffer [jbd2] / jbd2_write_access_granted [jbd2] write to 0xffff99f9b1bd0e30 of 8 bytes by task 25721 on cpu 70: __jbd2_journal_refile_buffer+0xdd/0x210 [jbd2] __jbd2_journal_refile_buffer at fs/jbd2/transaction.c:2569 jbd2_journal_commit_transaction+0x2d15/0x3f20 [jbd2] (inlined by) jbd2_journal_commit_transaction at fs/jbd2/commit.c:1034 kjournald2+0x13b/0x450 [jbd2] kthread+0x1cd/0x1f0 ret_from_fork+0x27/0x50 read to 0xffff99f9b1bd0e30 of 8 bytes by task 25724 on cpu 68: jbd2_write_access_granted+0x1b2/0x250 [jbd2] jbd2_write_access_granted at fs/jbd2/transaction.c:1155 jbd2_journal_get_write_access+0x2c/0x60 [jbd2] __ext4_journal_get_write_access+0x50/0x90 [ext4] ext4_mb_mark_diskspace_used+0x158/0x620 [ext4] ext4_mb_new_blocks+0x54f/0xca0 [ext4] ext4_ind_map_blocks+0xc79/0x1b40 [ext4] ext4_map_blocks+0x3b4/0x950 [ext4] _ext4_get_block+0xfc/0x270 [ext4] ext4_get_block+0x3b/0x50 [ext4] __block_write_begin_int+0x22e/0xae0 __block_write_begin+0x39/0x50 ext4_write_begin+0x388/0xb50 [ext4] generic_perform_write+0x15d/0x290 ext4_buffered_write_iter+0x11f/0x210 [ext4] ext4_file_write_iter+0xce/0x9e0 [ext4] new_sync_write+0x29c/0x3b0 __vfs_write+0x92/0xa0 vfs_write+0x103/0x260 ksys_write+0x9d/0x130 __x64_sys_write+0x4c/0x60 do_syscall_64+0x91/0xb05 entry_SYSCALL_64_after_hwframe+0x49/0xbe 5 locks held by fsync04/25724: #0: ffff99f9911093f8 (sb_writers#13){.+.+}, at: vfs_write+0x21c/0x260 #1: ffff99f9db4c0348 (&sb->s_type->i_mutex_key#15){+.+.}, at: ext4_buffered_write_iter+0x65/0x210 [ext4] #2: ffff99f5e7dfcf58 (jbd2_handle){++++}, at: start_this_handle+0x1c1/0x9d0 [jbd2] #3: ffff99f9db4c0168 (&ei->i_data_sem){++++}, at: ext4_map_blocks+0x176/0x950 [ext4] #4: ffffffff99086b40 (rcu_read_lock){....}, at: jbd2_write_access_granted+0x4e/0x250 [jbd2] irq event stamp: 1407125 hardirqs last enabled at (1407125): [<ffffffff980da9b7>] __find_get_block+0x107/0x790 hardirqs last disabled at (1407124): [<ffffffff980da8f9>] __find_get_block+0x49/0x790 softirqs last enabled at (1405528): [<ffffffff98a0034c>] __do_softirq+0x34c/0x57c softirqs last disabled at (1405521): [<ffffffff97cc67a2>] irq_exit+0xa2/0xc0 Reported by Kernel Concurrency Sanitizer on: CPU: 68 PID: 25724 Comm: fsync04 Tainted: G L 5.6.0-rc2-next-20200221+ #7 Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 07/10/2019 The plain reads are outside of jh->b_state_lock critical section which result in data races. Fix them by adding pairs of READ|WRITE_ONCE(). Reviewed-by: Jan Kara <jack@suse.cz> Signed-off-by: Qian Cai <cai@lca.pw> Link: https://lore.kernel.org/r/20200222043111.2227-1-cai@lca.pw Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2020-02-22 04:31:11 +00:00
if (READ_ONCE(jh->b_transaction) != handle->h_transaction &&
READ_ONCE(jh->b_next_transaction) != handle->h_transaction)
goto out;
/*
* There are two reasons for the barrier here:
* 1) Make sure to fetch b_bh after we did previous checks so that we
* detect when jh went through free, realloc, attach to transaction
* while we were checking. Paired with implicit barrier in that path.
* 2) So that access to bh done after jbd2_write_access_granted()
* doesn't get reordered and see inconsistent state of concurrent
* do_get_write_access().
*/
smp_mb();
if (unlikely(jh->b_bh != bh))
goto out;
ret = true;
out:
rcu_read_unlock();
return ret;
}
/**
* jbd2_journal_get_write_access() - notify intent to modify a buffer
* for metadata (not data) update.
* @handle: transaction to add buffer modifications to
* @bh: bh to be used for metadata writes
*
* Returns: error code or 0 on success.
*
* In full data journalling mode the buffer may be of type BJ_AsyncData,
* because we're ``write()ing`` a buffer which is also part of a shared mapping.
*/
int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
{
struct journal_head *jh;
journal_t *journal;
int rc;
jbd2: fix ocfs2 corrupt when clearing block group bits I found a NULL pointer dereference in ocfs2_block_group_clear_bits(). The running environment: kernel version: 4.19 A cluster with two nodes, 5 luns mounted on two nodes, and do some file operations like dd/fallocate/truncate/rm on every lun with storage network disconnection. The fallocate operation on dm-23-45 caused an null pointer dereference. The information of NULL pointer dereference as follows: [577992.878282] JBD2: Error -5 detected when updating journal superblock for dm-23-45. [577992.878290] Aborting journal on device dm-23-45. ... [577992.890778] JBD2: Error -5 detected when updating journal superblock for dm-24-46. [577992.890908] __journal_remove_journal_head: freeing b_committed_data [577992.890916] (fallocate,88392,52):ocfs2_extend_trans:474 ERROR: status = -30 [577992.890918] __journal_remove_journal_head: freeing b_committed_data [577992.890920] (fallocate,88392,52):ocfs2_rotate_tree_right:2500 ERROR: status = -30 [577992.890922] __journal_remove_journal_head: freeing b_committed_data [577992.890924] (fallocate,88392,52):ocfs2_do_insert_extent:4382 ERROR: status = -30 [577992.890928] (fallocate,88392,52):ocfs2_insert_extent:4842 ERROR: status = -30 [577992.890928] __journal_remove_journal_head: freeing b_committed_data [577992.890930] (fallocate,88392,52):ocfs2_add_clusters_in_btree:4947 ERROR: status = -30 [577992.890933] __journal_remove_journal_head: freeing b_committed_data [577992.890939] __journal_remove_journal_head: freeing b_committed_data [577992.890949] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000020 [577992.890950] Mem abort info: [577992.890951] ESR = 0x96000004 [577992.890952] Exception class = DABT (current EL), IL = 32 bits [577992.890952] SET = 0, FnV = 0 [577992.890953] EA = 0, S1PTW = 0 [577992.890954] Data abort info: [577992.890955] ISV = 0, ISS = 0x00000004 [577992.890956] CM = 0, WnR = 0 [577992.890958] user pgtable: 4k pages, 48-bit VAs, pgdp = 00000000f8da07a9 [577992.890960] [0000000000000020] pgd=0000000000000000 [577992.890964] Internal error: Oops: 96000004 [#1] SMP [577992.890965] Process fallocate (pid: 88392, stack limit = 0x00000000013db2fd) [577992.890968] CPU: 52 PID: 88392 Comm: fallocate Kdump: loaded Tainted: G W OE 4.19.36 #1 [577992.890969] Hardware name: Huawei TaiShan 2280 V2/BC82AMDD, BIOS 0.98 08/25/2019 [577992.890971] pstate: 60400009 (nZCv daif +PAN -UAO) [577992.891054] pc : _ocfs2_free_suballoc_bits+0x63c/0x968 [ocfs2] [577992.891082] lr : _ocfs2_free_suballoc_bits+0x618/0x968 [ocfs2] [577992.891084] sp : ffff0000c8e2b810 [577992.891085] x29: ffff0000c8e2b820 x28: 0000000000000000 [577992.891087] x27: 00000000000006f3 x26: ffffa07957b02e70 [577992.891089] x25: ffff807c59d50000 x24: 00000000000006f2 [577992.891091] x23: 0000000000000001 x22: ffff807bd39abc30 [577992.891093] x21: ffff0000811d9000 x20: ffffa07535d6a000 [577992.891097] x19: ffff000001681638 x18: ffffffffffffffff [577992.891098] x17: 0000000000000000 x16: ffff000080a03df0 [577992.891100] x15: ffff0000811d9708 x14: 203d207375746174 [577992.891101] x13: 73203a524f525245 x12: 20373439343a6565 [577992.891103] x11: 0000000000000038 x10: 0101010101010101 [577992.891106] x9 : ffffa07c68a85d70 x8 : 7f7f7f7f7f7f7f7f [577992.891109] x7 : 0000000000000000 x6 : 0000000000000080 [577992.891110] x5 : 0000000000000000 x4 : 0000000000000002 [577992.891112] x3 : ffff000001713390 x2 : 2ff90f88b1c22f00 [577992.891114] x1 : ffff807bd39abc30 x0 : 0000000000000000 [577992.891116] Call trace: [577992.891139] _ocfs2_free_suballoc_bits+0x63c/0x968 [ocfs2] [577992.891162] _ocfs2_free_clusters+0x100/0x290 [ocfs2] [577992.891185] ocfs2_free_clusters+0x50/0x68 [ocfs2] [577992.891206] ocfs2_add_clusters_in_btree+0x198/0x5e0 [ocfs2] [577992.891227] ocfs2_add_inode_data+0x94/0xc8 [ocfs2] [577992.891248] ocfs2_extend_allocation+0x1bc/0x7a8 [ocfs2] [577992.891269] ocfs2_allocate_extents+0x14c/0x338 [ocfs2] [577992.891290] __ocfs2_change_file_space+0x3f8/0x610 [ocfs2] [577992.891309] ocfs2_fallocate+0xe4/0x128 [ocfs2] [577992.891316] vfs_fallocate+0x11c/0x250 [577992.891317] ksys_fallocate+0x54/0x88 [577992.891319] __arm64_sys_fallocate+0x28/0x38 [577992.891323] el0_svc_common+0x78/0x130 [577992.891325] el0_svc_handler+0x38/0x78 [577992.891327] el0_svc+0x8/0xc My analysis process as follows: ocfs2_fallocate __ocfs2_change_file_space ocfs2_allocate_extents ocfs2_extend_allocation ocfs2_add_inode_data ocfs2_add_clusters_in_btree ocfs2_insert_extent ocfs2_do_insert_extent ocfs2_rotate_tree_right ocfs2_extend_rotate_transaction ocfs2_extend_trans jbd2_journal_restart jbd2__journal_restart /* handle->h_transaction is NULL, * is_handle_aborted(handle) is true */ handle->h_transaction = NULL; start_this_handle return -EROFS; ocfs2_free_clusters _ocfs2_free_clusters _ocfs2_free_suballoc_bits ocfs2_block_group_clear_bits ocfs2_journal_access_gd __ocfs2_journal_access jbd2_journal_get_undo_access /* I think jbd2_write_access_granted() will * return true, because do_get_write_access() * will return -EROFS. */ if (jbd2_write_access_granted(...)) return 0; do_get_write_access /* handle->h_transaction is NULL, it will * return -EROFS here, so do_get_write_access() * was not called. */ if (is_handle_aborted(handle)) return -EROFS; /* bh2jh(group_bh) is NULL, caused NULL pointer dereference */ undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data; If handle->h_transaction == NULL, then jbd2_write_access_granted() does not really guarantee that journal_head will stay around, not even speaking of its b_committed_data. The bh2jh(group_bh) can be removed after ocfs2_journal_access_gd() and before call "bh2jh(group_bh)->b_committed_data". So, we should move is_handle_aborted() check from do_get_write_access() into jbd2_journal_get_undo_access() and jbd2_journal_get_write_access() before the call to jbd2_write_access_granted(). Link: https://lore.kernel.org/r/f72a623f-b3f1-381a-d91d-d22a1c83a336@huawei.com Signed-off-by: Yan Wang <wangyan122@huawei.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jun Piao <piaojun@huawei.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: stable@kernel.org
2020-02-20 13:46:14 +00:00
if (is_handle_aborted(handle))
return -EROFS;
journal = handle->h_transaction->t_journal;
if (jbd2_check_fs_dev_write_error(journal)) {
/*
* If the fs dev has writeback errors, it may have failed
* to async write out metadata buffers in the background.
* In this case, we could read old data from disk and write
* it out again, which may lead to on-disk filesystem
* inconsistency. Aborting journal can avoid it happen.
*/
jbd2_journal_abort(journal, -EIO);
return -EIO;
}
jbd2: fix null committed data return in undo_access introduced jbd2_write_access_granted() to improve write|undo_access speed, but missed to check the status of b_committed_data which caused a kernel panic on ocfs2. [ 6538.405938] ------------[ cut here ]------------ [ 6538.406686] kernel BUG at fs/ocfs2/suballoc.c:2400! [ 6538.406686] invalid opcode: 0000 [#1] SMP [ 6538.406686] Modules linked in: ocfs2 nfsd lockd grace nfs_acl auth_rpcgss sunrpc autofs4 ocfs2_dlmfs ocfs2_stack_o2cb ocfs2_dlm ocfs2_nodemanager ocfs2_stackglue configfs sd_mod sg ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables be2iscsi iscsi_boot_sysfs bnx2i cnic uio cxgb4i cxgb4 cxgb3i libcxgbi cxgb3 mdio ib_iser rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr ipv6 iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi ppdev xen_kbdfront xen_netfront xen_fbfront parport_pc parport pcspkr i2c_piix4 acpi_cpufreq ext4 jbd2 mbcache xen_blkfront floppy pata_acpi ata_generic ata_piix cirrus ttm drm_kms_helper drm fb_sys_fops sysimgblt sysfillrect i2c_core syscopyarea dm_mirror dm_region_hash dm_log dm_mod [ 6538.406686] CPU: 1 PID: 16265 Comm: mmap_truncate Not tainted 4.3.0 #1 [ 6538.406686] Hardware name: Xen HVM domU, BIOS 4.3.1OVM 05/14/2014 [ 6538.406686] task: ffff88007c2bab00 ti: ffff880075b78000 task.ti: ffff880075b78000 [ 6538.406686] RIP: 0010:[<ffffffffa06a286b>] [<ffffffffa06a286b>] ocfs2_block_group_clear_bits+0x23b/0x250 [ocfs2] [ 6538.406686] RSP: 0018:ffff880075b7b7f8 EFLAGS: 00010246 [ 6538.406686] RAX: ffff8800760c5b40 RBX: ffff88006c06a000 RCX: ffffffffa06e6df0 [ 6538.406686] RDX: 0000000000000000 RSI: ffff88007a6f6ea0 RDI: ffff88007a760430 [ 6538.406686] RBP: ffff880075b7b878 R08: 0000000000000002 R09: 0000000000000001 [ 6538.406686] R10: ffffffffa06769be R11: 0000000000000000 R12: 0000000000000001 [ 6538.406686] R13: ffffffffa06a1750 R14: 0000000000000001 R15: ffff88007a6f6ea0 [ 6538.406686] FS: 00007f17fde30720(0000) GS:ffff88007f040000(0000) knlGS:0000000000000000 [ 6538.406686] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 6538.406686] CR2: 0000000000601730 CR3: 000000007aea0000 CR4: 00000000000406e0 [ 6538.406686] Stack: [ 6538.406686] ffff88007c2bb5b0 ffff880075b7b8e0 ffff88007a7604b0 ffff88006c640800 [ 6538.406686] ffff88007a7604b0 ffff880075d77390 0000000075b7b878 ffffffffa06a309d [ 6538.406686] ffff880075d752d8 ffff880075b7b990 ffff880075b7b898 0000000000000000 [ 6538.406686] Call Trace: [ 6538.406686] [<ffffffffa06a309d>] ? ocfs2_read_group_descriptor+0x6d/0xa0 [ocfs2] [ 6538.406686] [<ffffffffa06a3654>] _ocfs2_free_suballoc_bits+0xe4/0x320 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa06a397e>] _ocfs2_free_clusters+0xee/0x210 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa0682d50>] ? ocfs2_extend_trans+0x50/0x1a0 [ocfs2] [ 6538.406686] [<ffffffffa06a3ad5>] ocfs2_free_clusters+0x15/0x20 [ocfs2] [ 6538.406686] [<ffffffffa065072c>] ocfs2_replay_truncate_records+0xfc/0x290 [ocfs2] [ 6538.406686] [<ffffffffa06843ac>] ? ocfs2_start_trans+0xec/0x1d0 [ocfs2] [ 6538.406686] [<ffffffffa0654600>] __ocfs2_flush_truncate_log+0x140/0x2d0 [ocfs2] [ 6538.406686] [<ffffffffa0654394>] ? ocfs2_reserve_blocks_for_rec_trunc.clone.0+0x44/0x170 [ocfs2] [ 6538.406686] [<ffffffffa065acd4>] ocfs2_remove_btree_range+0x374/0x630 [ocfs2] [ 6538.406686] [<ffffffffa017486b>] ? jbd2_journal_stop+0x25b/0x470 [jbd2] [ 6538.406686] [<ffffffffa065d5b5>] ocfs2_commit_truncate+0x305/0x670 [ocfs2] [ 6538.406686] [<ffffffffa0683430>] ? ocfs2_journal_access_eb+0x20/0x20 [ocfs2] [ 6538.406686] [<ffffffffa067adb7>] ocfs2_truncate_file+0x297/0x380 [ocfs2] [ 6538.406686] [<ffffffffa01759e4>] ? jbd2_journal_begin_ordered_truncate+0x64/0xc0 [jbd2] [ 6538.406686] [<ffffffffa067c7a2>] ocfs2_setattr+0x572/0x860 [ocfs2] [ 6538.406686] [<ffffffff810e4a3f>] ? current_fs_time+0x3f/0x50 [ 6538.406686] [<ffffffff812124b7>] notify_change+0x1d7/0x340 [ 6538.406686] [<ffffffff8121abf9>] ? generic_getxattr+0x79/0x80 [ 6538.406686] [<ffffffff811f5876>] do_truncate+0x66/0x90 [ 6538.406686] [<ffffffff81120e30>] ? __audit_syscall_entry+0xb0/0x110 [ 6538.406686] [<ffffffff811f5bb3>] do_sys_ftruncate.clone.0+0xf3/0x120 [ 6538.406686] [<ffffffff811f5bee>] SyS_ftruncate+0xe/0x10 [ 6538.406686] [<ffffffff816aa2ae>] entry_SYSCALL_64_fastpath+0x12/0x71 [ 6538.406686] Code: 28 48 81 ee b0 04 00 00 48 8b 92 50 fb ff ff 48 8b 80 b0 03 00 00 48 39 90 88 00 00 00 0f 84 30 fe ff ff 0f 0b eb fe 0f 0b eb fe <0f> 0b 0f 1f 00 eb fb 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00 [ 6538.406686] RIP [<ffffffffa06a286b>] ocfs2_block_group_clear_bits+0x23b/0x250 [ocfs2] [ 6538.406686] RSP <ffff880075b7b7f8> [ 6538.691128] ---[ end trace 31cd7011d6770d7e ]--- [ 6538.694492] Kernel panic - not syncing: Fatal exception [ 6538.695484] Kernel Offset: disabled Fixes: de92c8caf16c("jbd2: speedup jbd2_journal_get_[write|undo]_access()") Cc: <stable@vger.kernel.org> Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2015-12-04 17:29:28 +00:00
if (jbd2_write_access_granted(handle, bh, false))
return 0;
jh = jbd2_journal_add_journal_head(bh);
/* We do not want to get caught playing with fields which the
* log thread also manipulates. Make sure that the buffer
* completes any outstanding IO before proceeding. */
rc = do_get_write_access(handle, jh, 0);
jbd2_journal_put_journal_head(jh);
return rc;
}
/*
* When the user wants to journal a newly created buffer_head
* (ie. getblk() returned a new buffer and we are going to populate it
* manually rather than reading off disk), then we need to keep the
* buffer_head locked until it has been completely filled with new
* data. In this case, we should be able to make the assertion that
* the bh is not already part of an existing transaction.
*
* The buffer should already be locked by the caller by this point.
* There is no lock ranking violation: it was a newly created,
* unlocked buffer beforehand. */
/**
* jbd2_journal_get_create_access () - notify intent to use newly created bh
* @handle: transaction to new buffer to
* @bh: new buffer.
*
* Call this if you create a new bh.
*/
int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal;
struct journal_head *jh = jbd2_journal_add_journal_head(bh);
int err;
jbd2_debug(5, "journal_head %p\n", jh);
err = -EROFS;
if (is_handle_aborted(handle))
goto out;
journal = transaction->t_journal;
err = 0;
JBUFFER_TRACE(jh, "entry");
/*
* The buffer may already belong to this transaction due to pre-zeroing
* in the filesystem's new_block code. It may also be on the previous,
* committing transaction's lists, but it HAS to be in Forget state in
* that case: the transaction must have deleted the buffer for it to be
* reused here.
*/
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
jh->b_transaction == NULL ||
(jh->b_transaction == journal->j_committing_transaction &&
jh->b_jlist == BJ_Forget)));
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
if (jh->b_transaction == NULL) {
/*
* Previous jbd2_journal_forget() could have left the buffer
* with jbddirty bit set because it was being committed. When
* the commit finished, we've filed the buffer for
* checkpointing and marked it dirty. Now we are reallocating
* the buffer so the transaction freeing it must have
* committed and so it's safe to clear the dirty bit.
*/
clear_buffer_dirty(jh2bh(jh));
/* first access by this transaction */
jh->b_modified = 0;
JBUFFER_TRACE(jh, "file as BJ_Reserved");
spin_lock(&journal->j_list_lock);
__jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
spin_unlock(&journal->j_list_lock);
} else if (jh->b_transaction == journal->j_committing_transaction) {
/* first access by this transaction */
jh->b_modified = 0;
JBUFFER_TRACE(jh, "set next transaction");
spin_lock(&journal->j_list_lock);
jh->b_next_transaction = transaction;
spin_unlock(&journal->j_list_lock);
}
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
/*
* akpm: I added this. ext3_alloc_branch can pick up new indirect
* blocks which contain freed but then revoked metadata. We need
* to cancel the revoke in case we end up freeing it yet again
* and the reallocating as data - this would cause a second revoke,
* which hits an assertion error.
*/
JBUFFER_TRACE(jh, "cancelling revoke");
jbd2_journal_cancel_revoke(handle, jh);
out:
jbd2_journal_put_journal_head(jh);
return err;
}
/**
* jbd2_journal_get_undo_access() - Notify intent to modify metadata with
* non-rewindable consequences
* @handle: transaction
* @bh: buffer to undo
*
* Sometimes there is a need to distinguish between metadata which has
* been committed to disk and that which has not. The ext3fs code uses
* this for freeing and allocating space, we have to make sure that we
* do not reuse freed space until the deallocation has been committed,
* since if we overwrote that space we would make the delete
* un-rewindable in case of a crash.
*
* To deal with that, jbd2_journal_get_undo_access requests write access to a
* buffer for parts of non-rewindable operations such as delete
* operations on the bitmaps. The journaling code must keep a copy of
* the buffer's contents prior to the undo_access call until such time
* as we know that the buffer has definitely been committed to disk.
*
* We never need to know which transaction the committed data is part
* of, buffers touched here are guaranteed to be dirtied later and so
* will be committed to a new transaction in due course, at which point
* we can discard the old committed data pointer.
*
* Returns error number or 0 on success.
*/
int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
{
int err;
struct journal_head *jh;
char *committed_data = NULL;
jbd2: fix ocfs2 corrupt when clearing block group bits I found a NULL pointer dereference in ocfs2_block_group_clear_bits(). The running environment: kernel version: 4.19 A cluster with two nodes, 5 luns mounted on two nodes, and do some file operations like dd/fallocate/truncate/rm on every lun with storage network disconnection. The fallocate operation on dm-23-45 caused an null pointer dereference. The information of NULL pointer dereference as follows: [577992.878282] JBD2: Error -5 detected when updating journal superblock for dm-23-45. [577992.878290] Aborting journal on device dm-23-45. ... [577992.890778] JBD2: Error -5 detected when updating journal superblock for dm-24-46. [577992.890908] __journal_remove_journal_head: freeing b_committed_data [577992.890916] (fallocate,88392,52):ocfs2_extend_trans:474 ERROR: status = -30 [577992.890918] __journal_remove_journal_head: freeing b_committed_data [577992.890920] (fallocate,88392,52):ocfs2_rotate_tree_right:2500 ERROR: status = -30 [577992.890922] __journal_remove_journal_head: freeing b_committed_data [577992.890924] (fallocate,88392,52):ocfs2_do_insert_extent:4382 ERROR: status = -30 [577992.890928] (fallocate,88392,52):ocfs2_insert_extent:4842 ERROR: status = -30 [577992.890928] __journal_remove_journal_head: freeing b_committed_data [577992.890930] (fallocate,88392,52):ocfs2_add_clusters_in_btree:4947 ERROR: status = -30 [577992.890933] __journal_remove_journal_head: freeing b_committed_data [577992.890939] __journal_remove_journal_head: freeing b_committed_data [577992.890949] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000020 [577992.890950] Mem abort info: [577992.890951] ESR = 0x96000004 [577992.890952] Exception class = DABT (current EL), IL = 32 bits [577992.890952] SET = 0, FnV = 0 [577992.890953] EA = 0, S1PTW = 0 [577992.890954] Data abort info: [577992.890955] ISV = 0, ISS = 0x00000004 [577992.890956] CM = 0, WnR = 0 [577992.890958] user pgtable: 4k pages, 48-bit VAs, pgdp = 00000000f8da07a9 [577992.890960] [0000000000000020] pgd=0000000000000000 [577992.890964] Internal error: Oops: 96000004 [#1] SMP [577992.890965] Process fallocate (pid: 88392, stack limit = 0x00000000013db2fd) [577992.890968] CPU: 52 PID: 88392 Comm: fallocate Kdump: loaded Tainted: G W OE 4.19.36 #1 [577992.890969] Hardware name: Huawei TaiShan 2280 V2/BC82AMDD, BIOS 0.98 08/25/2019 [577992.890971] pstate: 60400009 (nZCv daif +PAN -UAO) [577992.891054] pc : _ocfs2_free_suballoc_bits+0x63c/0x968 [ocfs2] [577992.891082] lr : _ocfs2_free_suballoc_bits+0x618/0x968 [ocfs2] [577992.891084] sp : ffff0000c8e2b810 [577992.891085] x29: ffff0000c8e2b820 x28: 0000000000000000 [577992.891087] x27: 00000000000006f3 x26: ffffa07957b02e70 [577992.891089] x25: ffff807c59d50000 x24: 00000000000006f2 [577992.891091] x23: 0000000000000001 x22: ffff807bd39abc30 [577992.891093] x21: ffff0000811d9000 x20: ffffa07535d6a000 [577992.891097] x19: ffff000001681638 x18: ffffffffffffffff [577992.891098] x17: 0000000000000000 x16: ffff000080a03df0 [577992.891100] x15: ffff0000811d9708 x14: 203d207375746174 [577992.891101] x13: 73203a524f525245 x12: 20373439343a6565 [577992.891103] x11: 0000000000000038 x10: 0101010101010101 [577992.891106] x9 : ffffa07c68a85d70 x8 : 7f7f7f7f7f7f7f7f [577992.891109] x7 : 0000000000000000 x6 : 0000000000000080 [577992.891110] x5 : 0000000000000000 x4 : 0000000000000002 [577992.891112] x3 : ffff000001713390 x2 : 2ff90f88b1c22f00 [577992.891114] x1 : ffff807bd39abc30 x0 : 0000000000000000 [577992.891116] Call trace: [577992.891139] _ocfs2_free_suballoc_bits+0x63c/0x968 [ocfs2] [577992.891162] _ocfs2_free_clusters+0x100/0x290 [ocfs2] [577992.891185] ocfs2_free_clusters+0x50/0x68 [ocfs2] [577992.891206] ocfs2_add_clusters_in_btree+0x198/0x5e0 [ocfs2] [577992.891227] ocfs2_add_inode_data+0x94/0xc8 [ocfs2] [577992.891248] ocfs2_extend_allocation+0x1bc/0x7a8 [ocfs2] [577992.891269] ocfs2_allocate_extents+0x14c/0x338 [ocfs2] [577992.891290] __ocfs2_change_file_space+0x3f8/0x610 [ocfs2] [577992.891309] ocfs2_fallocate+0xe4/0x128 [ocfs2] [577992.891316] vfs_fallocate+0x11c/0x250 [577992.891317] ksys_fallocate+0x54/0x88 [577992.891319] __arm64_sys_fallocate+0x28/0x38 [577992.891323] el0_svc_common+0x78/0x130 [577992.891325] el0_svc_handler+0x38/0x78 [577992.891327] el0_svc+0x8/0xc My analysis process as follows: ocfs2_fallocate __ocfs2_change_file_space ocfs2_allocate_extents ocfs2_extend_allocation ocfs2_add_inode_data ocfs2_add_clusters_in_btree ocfs2_insert_extent ocfs2_do_insert_extent ocfs2_rotate_tree_right ocfs2_extend_rotate_transaction ocfs2_extend_trans jbd2_journal_restart jbd2__journal_restart /* handle->h_transaction is NULL, * is_handle_aborted(handle) is true */ handle->h_transaction = NULL; start_this_handle return -EROFS; ocfs2_free_clusters _ocfs2_free_clusters _ocfs2_free_suballoc_bits ocfs2_block_group_clear_bits ocfs2_journal_access_gd __ocfs2_journal_access jbd2_journal_get_undo_access /* I think jbd2_write_access_granted() will * return true, because do_get_write_access() * will return -EROFS. */ if (jbd2_write_access_granted(...)) return 0; do_get_write_access /* handle->h_transaction is NULL, it will * return -EROFS here, so do_get_write_access() * was not called. */ if (is_handle_aborted(handle)) return -EROFS; /* bh2jh(group_bh) is NULL, caused NULL pointer dereference */ undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data; If handle->h_transaction == NULL, then jbd2_write_access_granted() does not really guarantee that journal_head will stay around, not even speaking of its b_committed_data. The bh2jh(group_bh) can be removed after ocfs2_journal_access_gd() and before call "bh2jh(group_bh)->b_committed_data". So, we should move is_handle_aborted() check from do_get_write_access() into jbd2_journal_get_undo_access() and jbd2_journal_get_write_access() before the call to jbd2_write_access_granted(). Link: https://lore.kernel.org/r/f72a623f-b3f1-381a-d91d-d22a1c83a336@huawei.com Signed-off-by: Yan Wang <wangyan122@huawei.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jun Piao <piaojun@huawei.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: stable@kernel.org
2020-02-20 13:46:14 +00:00
if (is_handle_aborted(handle))
return -EROFS;
jbd2: fix null committed data return in undo_access introduced jbd2_write_access_granted() to improve write|undo_access speed, but missed to check the status of b_committed_data which caused a kernel panic on ocfs2. [ 6538.405938] ------------[ cut here ]------------ [ 6538.406686] kernel BUG at fs/ocfs2/suballoc.c:2400! [ 6538.406686] invalid opcode: 0000 [#1] SMP [ 6538.406686] Modules linked in: ocfs2 nfsd lockd grace nfs_acl auth_rpcgss sunrpc autofs4 ocfs2_dlmfs ocfs2_stack_o2cb ocfs2_dlm ocfs2_nodemanager ocfs2_stackglue configfs sd_mod sg ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables be2iscsi iscsi_boot_sysfs bnx2i cnic uio cxgb4i cxgb4 cxgb3i libcxgbi cxgb3 mdio ib_iser rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr ipv6 iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi ppdev xen_kbdfront xen_netfront xen_fbfront parport_pc parport pcspkr i2c_piix4 acpi_cpufreq ext4 jbd2 mbcache xen_blkfront floppy pata_acpi ata_generic ata_piix cirrus ttm drm_kms_helper drm fb_sys_fops sysimgblt sysfillrect i2c_core syscopyarea dm_mirror dm_region_hash dm_log dm_mod [ 6538.406686] CPU: 1 PID: 16265 Comm: mmap_truncate Not tainted 4.3.0 #1 [ 6538.406686] Hardware name: Xen HVM domU, BIOS 4.3.1OVM 05/14/2014 [ 6538.406686] task: ffff88007c2bab00 ti: ffff880075b78000 task.ti: ffff880075b78000 [ 6538.406686] RIP: 0010:[<ffffffffa06a286b>] [<ffffffffa06a286b>] ocfs2_block_group_clear_bits+0x23b/0x250 [ocfs2] [ 6538.406686] RSP: 0018:ffff880075b7b7f8 EFLAGS: 00010246 [ 6538.406686] RAX: ffff8800760c5b40 RBX: ffff88006c06a000 RCX: ffffffffa06e6df0 [ 6538.406686] RDX: 0000000000000000 RSI: ffff88007a6f6ea0 RDI: ffff88007a760430 [ 6538.406686] RBP: ffff880075b7b878 R08: 0000000000000002 R09: 0000000000000001 [ 6538.406686] R10: ffffffffa06769be R11: 0000000000000000 R12: 0000000000000001 [ 6538.406686] R13: ffffffffa06a1750 R14: 0000000000000001 R15: ffff88007a6f6ea0 [ 6538.406686] FS: 00007f17fde30720(0000) GS:ffff88007f040000(0000) knlGS:0000000000000000 [ 6538.406686] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 6538.406686] CR2: 0000000000601730 CR3: 000000007aea0000 CR4: 00000000000406e0 [ 6538.406686] Stack: [ 6538.406686] ffff88007c2bb5b0 ffff880075b7b8e0 ffff88007a7604b0 ffff88006c640800 [ 6538.406686] ffff88007a7604b0 ffff880075d77390 0000000075b7b878 ffffffffa06a309d [ 6538.406686] ffff880075d752d8 ffff880075b7b990 ffff880075b7b898 0000000000000000 [ 6538.406686] Call Trace: [ 6538.406686] [<ffffffffa06a309d>] ? ocfs2_read_group_descriptor+0x6d/0xa0 [ocfs2] [ 6538.406686] [<ffffffffa06a3654>] _ocfs2_free_suballoc_bits+0xe4/0x320 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa06a397e>] _ocfs2_free_clusters+0xee/0x210 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa06a1750>] ? ocfs2_put_slot+0xf0/0xf0 [ocfs2] [ 6538.406686] [<ffffffffa0682d50>] ? ocfs2_extend_trans+0x50/0x1a0 [ocfs2] [ 6538.406686] [<ffffffffa06a3ad5>] ocfs2_free_clusters+0x15/0x20 [ocfs2] [ 6538.406686] [<ffffffffa065072c>] ocfs2_replay_truncate_records+0xfc/0x290 [ocfs2] [ 6538.406686] [<ffffffffa06843ac>] ? ocfs2_start_trans+0xec/0x1d0 [ocfs2] [ 6538.406686] [<ffffffffa0654600>] __ocfs2_flush_truncate_log+0x140/0x2d0 [ocfs2] [ 6538.406686] [<ffffffffa0654394>] ? ocfs2_reserve_blocks_for_rec_trunc.clone.0+0x44/0x170 [ocfs2] [ 6538.406686] [<ffffffffa065acd4>] ocfs2_remove_btree_range+0x374/0x630 [ocfs2] [ 6538.406686] [<ffffffffa017486b>] ? jbd2_journal_stop+0x25b/0x470 [jbd2] [ 6538.406686] [<ffffffffa065d5b5>] ocfs2_commit_truncate+0x305/0x670 [ocfs2] [ 6538.406686] [<ffffffffa0683430>] ? ocfs2_journal_access_eb+0x20/0x20 [ocfs2] [ 6538.406686] [<ffffffffa067adb7>] ocfs2_truncate_file+0x297/0x380 [ocfs2] [ 6538.406686] [<ffffffffa01759e4>] ? jbd2_journal_begin_ordered_truncate+0x64/0xc0 [jbd2] [ 6538.406686] [<ffffffffa067c7a2>] ocfs2_setattr+0x572/0x860 [ocfs2] [ 6538.406686] [<ffffffff810e4a3f>] ? current_fs_time+0x3f/0x50 [ 6538.406686] [<ffffffff812124b7>] notify_change+0x1d7/0x340 [ 6538.406686] [<ffffffff8121abf9>] ? generic_getxattr+0x79/0x80 [ 6538.406686] [<ffffffff811f5876>] do_truncate+0x66/0x90 [ 6538.406686] [<ffffffff81120e30>] ? __audit_syscall_entry+0xb0/0x110 [ 6538.406686] [<ffffffff811f5bb3>] do_sys_ftruncate.clone.0+0xf3/0x120 [ 6538.406686] [<ffffffff811f5bee>] SyS_ftruncate+0xe/0x10 [ 6538.406686] [<ffffffff816aa2ae>] entry_SYSCALL_64_fastpath+0x12/0x71 [ 6538.406686] Code: 28 48 81 ee b0 04 00 00 48 8b 92 50 fb ff ff 48 8b 80 b0 03 00 00 48 39 90 88 00 00 00 0f 84 30 fe ff ff 0f 0b eb fe 0f 0b eb fe <0f> 0b 0f 1f 00 eb fb 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00 [ 6538.406686] RIP [<ffffffffa06a286b>] ocfs2_block_group_clear_bits+0x23b/0x250 [ocfs2] [ 6538.406686] RSP <ffff880075b7b7f8> [ 6538.691128] ---[ end trace 31cd7011d6770d7e ]--- [ 6538.694492] Kernel panic - not syncing: Fatal exception [ 6538.695484] Kernel Offset: disabled Fixes: de92c8caf16c("jbd2: speedup jbd2_journal_get_[write|undo]_access()") Cc: <stable@vger.kernel.org> Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2015-12-04 17:29:28 +00:00
if (jbd2_write_access_granted(handle, bh, true))
return 0;
jh = jbd2_journal_add_journal_head(bh);
JBUFFER_TRACE(jh, "entry");
/*
* Do this first --- it can drop the journal lock, so we want to
* make sure that obtaining the committed_data is done
* atomically wrt. completion of any outstanding commits.
*/
err = do_get_write_access(handle, jh, 1);
if (err)
goto out;
repeat:
if (!jh->b_committed_data)
committed_data = jbd2_alloc(jh2bh(jh)->b_size,
GFP_NOFS|__GFP_NOFAIL);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
if (!jh->b_committed_data) {
/* Copy out the current buffer contents into the
* preserved, committed copy. */
JBUFFER_TRACE(jh, "generate b_committed data");
if (!committed_data) {
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
goto repeat;
}
jh->b_committed_data = committed_data;
committed_data = NULL;
memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
}
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
out:
jbd2_journal_put_journal_head(jh);
if (unlikely(committed_data))
jbd2_free(committed_data, bh->b_size);
return err;
}
jbd2: Add buffer triggers Filesystems often to do compute intensive operation on some metadata. If this operation is repeated many times, it can be very expensive. It would be much nicer if the operation could be performed once before a buffer goes to disk. This adds triggers to jbd2 buffer heads. Just before writing a metadata buffer to the journal, jbd2 will optionally call a commit trigger associated with the buffer. If the journal is aborted, an abort trigger will be called on any dirty buffers as they are dropped from pending transactions. ocfs2 will use this feature. Initially I tried to come up with a more generic trigger that could be used for non-buffer-related events like transaction completion. It doesn't tie nicely, because the information a buffer trigger needs (specific to a journal_head) isn't the same as what a transaction trigger needs (specific to a tranaction_t or perhaps journal_t). So I implemented a buffer set, with the understanding that journal/transaction wide triggers should be implemented separately. There is only one trigger set allowed per buffer. I can't think of any reason to attach more than one set. Contrast this with a journal or transaction in which multiple places may want to watch the entire transaction separately. The trigger sets are considered static allocation from the jbd2 perspective. ocfs2 will just have one trigger set per block type, setting the same set on every bh of the same type. Signed-off-by: Joel Becker <joel.becker@oracle.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: <linux-ext4@vger.kernel.org> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2008-09-11 22:35:47 +00:00
/**
* jbd2_journal_set_triggers() - Add triggers for commit writeout
jbd2: Add buffer triggers Filesystems often to do compute intensive operation on some metadata. If this operation is repeated many times, it can be very expensive. It would be much nicer if the operation could be performed once before a buffer goes to disk. This adds triggers to jbd2 buffer heads. Just before writing a metadata buffer to the journal, jbd2 will optionally call a commit trigger associated with the buffer. If the journal is aborted, an abort trigger will be called on any dirty buffers as they are dropped from pending transactions. ocfs2 will use this feature. Initially I tried to come up with a more generic trigger that could be used for non-buffer-related events like transaction completion. It doesn't tie nicely, because the information a buffer trigger needs (specific to a journal_head) isn't the same as what a transaction trigger needs (specific to a tranaction_t or perhaps journal_t). So I implemented a buffer set, with the understanding that journal/transaction wide triggers should be implemented separately. There is only one trigger set allowed per buffer. I can't think of any reason to attach more than one set. Contrast this with a journal or transaction in which multiple places may want to watch the entire transaction separately. The trigger sets are considered static allocation from the jbd2 perspective. ocfs2 will just have one trigger set per block type, setting the same set on every bh of the same type. Signed-off-by: Joel Becker <joel.becker@oracle.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: <linux-ext4@vger.kernel.org> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2008-09-11 22:35:47 +00:00
* @bh: buffer to trigger on
* @type: struct jbd2_buffer_trigger_type containing the trigger(s).
*
* Set any triggers on this journal_head. This is always safe, because
* triggers for a committing buffer will be saved off, and triggers for
* a running transaction will match the buffer in that transaction.
*
* Call with NULL to clear the triggers.
*/
void jbd2_journal_set_triggers(struct buffer_head *bh,
struct jbd2_buffer_trigger_type *type)
{
struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
jbd2: Add buffer triggers Filesystems often to do compute intensive operation on some metadata. If this operation is repeated many times, it can be very expensive. It would be much nicer if the operation could be performed once before a buffer goes to disk. This adds triggers to jbd2 buffer heads. Just before writing a metadata buffer to the journal, jbd2 will optionally call a commit trigger associated with the buffer. If the journal is aborted, an abort trigger will be called on any dirty buffers as they are dropped from pending transactions. ocfs2 will use this feature. Initially I tried to come up with a more generic trigger that could be used for non-buffer-related events like transaction completion. It doesn't tie nicely, because the information a buffer trigger needs (specific to a journal_head) isn't the same as what a transaction trigger needs (specific to a tranaction_t or perhaps journal_t). So I implemented a buffer set, with the understanding that journal/transaction wide triggers should be implemented separately. There is only one trigger set allowed per buffer. I can't think of any reason to attach more than one set. Contrast this with a journal or transaction in which multiple places may want to watch the entire transaction separately. The trigger sets are considered static allocation from the jbd2 perspective. ocfs2 will just have one trigger set per block type, setting the same set on every bh of the same type. Signed-off-by: Joel Becker <joel.becker@oracle.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: <linux-ext4@vger.kernel.org> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2008-09-11 22:35:47 +00:00
if (WARN_ON_ONCE(!jh))
return;
jbd2: Add buffer triggers Filesystems often to do compute intensive operation on some metadata. If this operation is repeated many times, it can be very expensive. It would be much nicer if the operation could be performed once before a buffer goes to disk. This adds triggers to jbd2 buffer heads. Just before writing a metadata buffer to the journal, jbd2 will optionally call a commit trigger associated with the buffer. If the journal is aborted, an abort trigger will be called on any dirty buffers as they are dropped from pending transactions. ocfs2 will use this feature. Initially I tried to come up with a more generic trigger that could be used for non-buffer-related events like transaction completion. It doesn't tie nicely, because the information a buffer trigger needs (specific to a journal_head) isn't the same as what a transaction trigger needs (specific to a tranaction_t or perhaps journal_t). So I implemented a buffer set, with the understanding that journal/transaction wide triggers should be implemented separately. There is only one trigger set allowed per buffer. I can't think of any reason to attach more than one set. Contrast this with a journal or transaction in which multiple places may want to watch the entire transaction separately. The trigger sets are considered static allocation from the jbd2 perspective. ocfs2 will just have one trigger set per block type, setting the same set on every bh of the same type. Signed-off-by: Joel Becker <joel.becker@oracle.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: <linux-ext4@vger.kernel.org> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2008-09-11 22:35:47 +00:00
jh->b_triggers = type;
jbd2_journal_put_journal_head(jh);
jbd2: Add buffer triggers Filesystems often to do compute intensive operation on some metadata. If this operation is repeated many times, it can be very expensive. It would be much nicer if the operation could be performed once before a buffer goes to disk. This adds triggers to jbd2 buffer heads. Just before writing a metadata buffer to the journal, jbd2 will optionally call a commit trigger associated with the buffer. If the journal is aborted, an abort trigger will be called on any dirty buffers as they are dropped from pending transactions. ocfs2 will use this feature. Initially I tried to come up with a more generic trigger that could be used for non-buffer-related events like transaction completion. It doesn't tie nicely, because the information a buffer trigger needs (specific to a journal_head) isn't the same as what a transaction trigger needs (specific to a tranaction_t or perhaps journal_t). So I implemented a buffer set, with the understanding that journal/transaction wide triggers should be implemented separately. There is only one trigger set allowed per buffer. I can't think of any reason to attach more than one set. Contrast this with a journal or transaction in which multiple places may want to watch the entire transaction separately. The trigger sets are considered static allocation from the jbd2 perspective. ocfs2 will just have one trigger set per block type, setting the same set on every bh of the same type. Signed-off-by: Joel Becker <joel.becker@oracle.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: <linux-ext4@vger.kernel.org> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2008-09-11 22:35:47 +00:00
}
jbd2/ocfs2: Fix block checksumming when a buffer is used in several transactions OCFS2 uses t_commit trigger to compute and store checksum of the just committed blocks. When a buffer has b_frozen_data, checksum is computed for it instead of b_data but this can result in an old checksum being written to the filesystem in the following scenario: 1) transaction1 is opened 2) handle1 is opened 3) journal_access(handle1, bh) - This sets jh->b_transaction to transaction1 4) modify(bh) 5) journal_dirty(handle1, bh) 6) handle1 is closed 7) start committing transaction1, opening transaction2 8) handle2 is opened 9) journal_access(handle2, bh) - This copies off b_frozen_data to make it safe for transaction1 to commit. jh->b_next_transaction is set to transaction2. 10) jbd2_journal_write_metadata() checksums b_frozen_data 11) the journal correctly writes b_frozen_data to the disk journal 12) handle2 is closed - There was no dirty call for the bh on handle2, so it is never queued for any more journal operation 13) Checkpointing finally happens, and it just spools the bh via normal buffer writeback. This will write b_data, which was never triggered on and thus contains a wrong (old) checksum. This patch fixes the problem by calling the trigger at the moment data is frozen for journal commit - i.e., either when b_frozen_data is created by do_get_write_access or just before we write a buffer to the log if b_frozen_data does not exist. We also rename the trigger to t_frozen as that better describes when it is called. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Mark Fasheh <mfasheh@suse.com> Signed-off-by: Joel Becker <joel.becker@oracle.com>
2010-07-14 05:56:33 +00:00
void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
jbd2: Add buffer triggers Filesystems often to do compute intensive operation on some metadata. If this operation is repeated many times, it can be very expensive. It would be much nicer if the operation could be performed once before a buffer goes to disk. This adds triggers to jbd2 buffer heads. Just before writing a metadata buffer to the journal, jbd2 will optionally call a commit trigger associated with the buffer. If the journal is aborted, an abort trigger will be called on any dirty buffers as they are dropped from pending transactions. ocfs2 will use this feature. Initially I tried to come up with a more generic trigger that could be used for non-buffer-related events like transaction completion. It doesn't tie nicely, because the information a buffer trigger needs (specific to a journal_head) isn't the same as what a transaction trigger needs (specific to a tranaction_t or perhaps journal_t). So I implemented a buffer set, with the understanding that journal/transaction wide triggers should be implemented separately. There is only one trigger set allowed per buffer. I can't think of any reason to attach more than one set. Contrast this with a journal or transaction in which multiple places may want to watch the entire transaction separately. The trigger sets are considered static allocation from the jbd2 perspective. ocfs2 will just have one trigger set per block type, setting the same set on every bh of the same type. Signed-off-by: Joel Becker <joel.becker@oracle.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: <linux-ext4@vger.kernel.org> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2008-09-11 22:35:47 +00:00
struct jbd2_buffer_trigger_type *triggers)
{
struct buffer_head *bh = jh2bh(jh);
jbd2/ocfs2: Fix block checksumming when a buffer is used in several transactions OCFS2 uses t_commit trigger to compute and store checksum of the just committed blocks. When a buffer has b_frozen_data, checksum is computed for it instead of b_data but this can result in an old checksum being written to the filesystem in the following scenario: 1) transaction1 is opened 2) handle1 is opened 3) journal_access(handle1, bh) - This sets jh->b_transaction to transaction1 4) modify(bh) 5) journal_dirty(handle1, bh) 6) handle1 is closed 7) start committing transaction1, opening transaction2 8) handle2 is opened 9) journal_access(handle2, bh) - This copies off b_frozen_data to make it safe for transaction1 to commit. jh->b_next_transaction is set to transaction2. 10) jbd2_journal_write_metadata() checksums b_frozen_data 11) the journal correctly writes b_frozen_data to the disk journal 12) handle2 is closed - There was no dirty call for the bh on handle2, so it is never queued for any more journal operation 13) Checkpointing finally happens, and it just spools the bh via normal buffer writeback. This will write b_data, which was never triggered on and thus contains a wrong (old) checksum. This patch fixes the problem by calling the trigger at the moment data is frozen for journal commit - i.e., either when b_frozen_data is created by do_get_write_access or just before we write a buffer to the log if b_frozen_data does not exist. We also rename the trigger to t_frozen as that better describes when it is called. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Mark Fasheh <mfasheh@suse.com> Signed-off-by: Joel Becker <joel.becker@oracle.com>
2010-07-14 05:56:33 +00:00
if (!triggers || !triggers->t_frozen)
jbd2: Add buffer triggers Filesystems often to do compute intensive operation on some metadata. If this operation is repeated many times, it can be very expensive. It would be much nicer if the operation could be performed once before a buffer goes to disk. This adds triggers to jbd2 buffer heads. Just before writing a metadata buffer to the journal, jbd2 will optionally call a commit trigger associated with the buffer. If the journal is aborted, an abort trigger will be called on any dirty buffers as they are dropped from pending transactions. ocfs2 will use this feature. Initially I tried to come up with a more generic trigger that could be used for non-buffer-related events like transaction completion. It doesn't tie nicely, because the information a buffer trigger needs (specific to a journal_head) isn't the same as what a transaction trigger needs (specific to a tranaction_t or perhaps journal_t). So I implemented a buffer set, with the understanding that journal/transaction wide triggers should be implemented separately. There is only one trigger set allowed per buffer. I can't think of any reason to attach more than one set. Contrast this with a journal or transaction in which multiple places may want to watch the entire transaction separately. The trigger sets are considered static allocation from the jbd2 perspective. ocfs2 will just have one trigger set per block type, setting the same set on every bh of the same type. Signed-off-by: Joel Becker <joel.becker@oracle.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: <linux-ext4@vger.kernel.org> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2008-09-11 22:35:47 +00:00
return;
jbd2/ocfs2: Fix block checksumming when a buffer is used in several transactions OCFS2 uses t_commit trigger to compute and store checksum of the just committed blocks. When a buffer has b_frozen_data, checksum is computed for it instead of b_data but this can result in an old checksum being written to the filesystem in the following scenario: 1) transaction1 is opened 2) handle1 is opened 3) journal_access(handle1, bh) - This sets jh->b_transaction to transaction1 4) modify(bh) 5) journal_dirty(handle1, bh) 6) handle1 is closed 7) start committing transaction1, opening transaction2 8) handle2 is opened 9) journal_access(handle2, bh) - This copies off b_frozen_data to make it safe for transaction1 to commit. jh->b_next_transaction is set to transaction2. 10) jbd2_journal_write_metadata() checksums b_frozen_data 11) the journal correctly writes b_frozen_data to the disk journal 12) handle2 is closed - There was no dirty call for the bh on handle2, so it is never queued for any more journal operation 13) Checkpointing finally happens, and it just spools the bh via normal buffer writeback. This will write b_data, which was never triggered on and thus contains a wrong (old) checksum. This patch fixes the problem by calling the trigger at the moment data is frozen for journal commit - i.e., either when b_frozen_data is created by do_get_write_access or just before we write a buffer to the log if b_frozen_data does not exist. We also rename the trigger to t_frozen as that better describes when it is called. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Mark Fasheh <mfasheh@suse.com> Signed-off-by: Joel Becker <joel.becker@oracle.com>
2010-07-14 05:56:33 +00:00
triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
jbd2: Add buffer triggers Filesystems often to do compute intensive operation on some metadata. If this operation is repeated many times, it can be very expensive. It would be much nicer if the operation could be performed once before a buffer goes to disk. This adds triggers to jbd2 buffer heads. Just before writing a metadata buffer to the journal, jbd2 will optionally call a commit trigger associated with the buffer. If the journal is aborted, an abort trigger will be called on any dirty buffers as they are dropped from pending transactions. ocfs2 will use this feature. Initially I tried to come up with a more generic trigger that could be used for non-buffer-related events like transaction completion. It doesn't tie nicely, because the information a buffer trigger needs (specific to a journal_head) isn't the same as what a transaction trigger needs (specific to a tranaction_t or perhaps journal_t). So I implemented a buffer set, with the understanding that journal/transaction wide triggers should be implemented separately. There is only one trigger set allowed per buffer. I can't think of any reason to attach more than one set. Contrast this with a journal or transaction in which multiple places may want to watch the entire transaction separately. The trigger sets are considered static allocation from the jbd2 perspective. ocfs2 will just have one trigger set per block type, setting the same set on every bh of the same type. Signed-off-by: Joel Becker <joel.becker@oracle.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: <linux-ext4@vger.kernel.org> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2008-09-11 22:35:47 +00:00
}
void jbd2_buffer_abort_trigger(struct journal_head *jh,
struct jbd2_buffer_trigger_type *triggers)
{
if (!triggers || !triggers->t_abort)
return;
triggers->t_abort(triggers, jh2bh(jh));
}
/**
* jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
* @handle: transaction to add buffer to.
* @bh: buffer to mark
*
* mark dirty metadata which needs to be journaled as part of the current
* transaction.
*
* The buffer must have previously had jbd2_journal_get_write_access()
* called so that it has a valid journal_head attached to the buffer
* head.
*
* The buffer is placed on the transaction's metadata list and is marked
* as belonging to the transaction.
*
* Returns error number or 0 on success.
*
* Special care needs to be taken if the buffer already belongs to the
* current committing transaction (in which case we should have frozen
* data present for that commit). In that case, we don't relink the
* buffer: that only gets done when the old transaction finally
* completes its commit.
*/
int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal;
struct journal_head *jh;
int ret = 0;
if (!buffer_jbd(bh))
return -EUCLEAN;
/*
* We don't grab jh reference here since the buffer must be part
* of the running transaction.
*/
jh = bh2jh(bh);
jbd2_debug(5, "journal_head %p\n", jh);
JBUFFER_TRACE(jh, "entry");
/*
* This and the following assertions are unreliable since we may see jh
* in inconsistent state unless we grab bh_state lock. But this is
* crucial to catch bugs so let's do a reliable check until the
* lockless handling is fully proven.
*/
if (data_race(jh->b_transaction != transaction &&
jh->b_next_transaction != transaction)) {
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
J_ASSERT_JH(jh, jh->b_transaction == transaction ||
jh->b_next_transaction == transaction);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
}
if (jh->b_modified == 1) {
/* If it's in our transaction it must be in BJ_Metadata list. */
if (data_race(jh->b_transaction == transaction &&
jh->b_jlist != BJ_Metadata)) {
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
if (jh->b_transaction == transaction &&
jh->b_jlist != BJ_Metadata)
pr_err("JBD2: assertion failure: h_type=%u "
"h_line_no=%u block_no=%llu jlist=%u\n",
handle->h_type, handle->h_line_no,
(unsigned long long) bh->b_blocknr,
jh->b_jlist);
J_ASSERT_JH(jh, jh->b_transaction != transaction ||
jh->b_jlist == BJ_Metadata);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
}
goto out;
}
journal = transaction->t_journal;
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
jbd2: fix assertion 'jh->b_frozen_data == NULL' failure when journal aborted Following process will fail assertion 'jh->b_frozen_data == NULL' in jbd2_journal_dirty_metadata(): jbd2_journal_commit_transaction unlink(dir/a) jh->b_transaction = trans1 jh->b_jlist = BJ_Metadata journal->j_running_transaction = NULL trans1->t_state = T_COMMIT unlink(dir/b) handle->h_trans = trans2 do_get_write_access jh->b_modified = 0 jh->b_frozen_data = frozen_buffer jh->b_next_transaction = trans2 jbd2_journal_dirty_metadata is_handle_aborted is_journal_aborted // return false --> jbd2 abort <-- while (commit_transaction->t_buffers) if (is_journal_aborted) jbd2_journal_refile_buffer __jbd2_journal_refile_buffer WRITE_ONCE(jh->b_transaction, jh->b_next_transaction) WRITE_ONCE(jh->b_next_transaction, NULL) __jbd2_journal_file_buffer(jh, BJ_Reserved) J_ASSERT_JH(jh, jh->b_frozen_data == NULL) // assertion failure ! The reproducer (See detail in [Link]) reports: ------------[ cut here ]------------ kernel BUG at fs/jbd2/transaction.c:1629! invalid opcode: 0000 [#1] PREEMPT SMP CPU: 2 PID: 584 Comm: unlink Tainted: G W 5.19.0-rc6-00115-g4a57a8400075-dirty #697 RIP: 0010:jbd2_journal_dirty_metadata+0x3c5/0x470 RSP: 0018:ffffc90000be7ce0 EFLAGS: 00010202 Call Trace: <TASK> __ext4_handle_dirty_metadata+0xa0/0x290 ext4_handle_dirty_dirblock+0x10c/0x1d0 ext4_delete_entry+0x104/0x200 __ext4_unlink+0x22b/0x360 ext4_unlink+0x275/0x390 vfs_unlink+0x20b/0x4c0 do_unlinkat+0x42f/0x4c0 __x64_sys_unlink+0x37/0x50 do_syscall_64+0x35/0x80 After journal aborting, __jbd2_journal_refile_buffer() is executed with holding @jh->b_state_lock, we can fix it by moving 'is_handle_aborted()' into the area protected by @jh->b_state_lock. Link: https://bugzilla.kernel.org/show_bug.cgi?id=216251 Fixes: 470decc613ab20 ("[PATCH] jbd2: initial copy of files from jbd") Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Link: https://lore.kernel.org/r/20220715125152.4022726-1-chengzhihao1@huawei.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-07-15 12:51:52 +00:00
if (is_handle_aborted(handle)) {
/*
* Check journal aborting with @jh->b_state_lock locked,
* since 'jh->b_transaction' could be replaced with
* 'jh->b_next_transaction' during old transaction
* committing if journal aborted, which may fail
* assertion on 'jh->b_frozen_data == NULL'.
*/
ret = -EROFS;
goto out_unlock_bh;
}
if (jh->b_modified == 0) {
/*
* This buffer's got modified and becoming part
* of the transaction. This needs to be done
* once a transaction -bzzz
*/
if (WARN_ON_ONCE(jbd2_handle_buffer_credits(handle) <= 0)) {
ret = -ENOSPC;
goto out_unlock_bh;
}
jh->b_modified = 1;
handle->h_total_credits--;
}
/*
* fastpath, to avoid expensive locking. If this buffer is already
* on the running transaction's metadata list there is nothing to do.
* Nobody can take it off again because there is a handle open.
* I _think_ we're OK here with SMP barriers - a mistaken decision will
* result in this test being false, so we go in and take the locks.
*/
if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
JBUFFER_TRACE(jh, "fastpath");
if (unlikely(jh->b_transaction !=
journal->j_running_transaction)) {
printk(KERN_ERR "JBD2: %s: "
"jh->b_transaction (%llu, %p, %u) != "
"journal->j_running_transaction (%p, %u)\n",
journal->j_devname,
(unsigned long long) bh->b_blocknr,
jh->b_transaction,
jh->b_transaction ? jh->b_transaction->t_tid : 0,
journal->j_running_transaction,
journal->j_running_transaction ?
journal->j_running_transaction->t_tid : 0);
ret = -EINVAL;
}
goto out_unlock_bh;
}
set_buffer_jbddirty(bh);
/*
* Metadata already on the current transaction list doesn't
* need to be filed. Metadata on another transaction's list must
* be committing, and will be refiled once the commit completes:
* leave it alone for now.
*/
if (jh->b_transaction != transaction) {
JBUFFER_TRACE(jh, "already on other transaction");
if (unlikely(((jh->b_transaction !=
journal->j_committing_transaction)) ||
(jh->b_next_transaction != transaction))) {
printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
"bad jh for block %llu: "
"transaction (%p, %u), "
"jh->b_transaction (%p, %u), "
"jh->b_next_transaction (%p, %u), jlist %u\n",
journal->j_devname,
(unsigned long long) bh->b_blocknr,
transaction, transaction->t_tid,
jh->b_transaction,
jh->b_transaction ?
jh->b_transaction->t_tid : 0,
jh->b_next_transaction,
jh->b_next_transaction ?
jh->b_next_transaction->t_tid : 0,
jh->b_jlist);
WARN_ON(1);
ret = -EINVAL;
}
/* And this case is illegal: we can't reuse another
* transaction's data buffer, ever. */
goto out_unlock_bh;
}
/* That test should have eliminated the following case: */
J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
JBUFFER_TRACE(jh, "file as BJ_Metadata");
spin_lock(&journal->j_list_lock);
__jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
spin_unlock(&journal->j_list_lock);
out_unlock_bh:
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
out:
JBUFFER_TRACE(jh, "exit");
return ret;
}
/**
* jbd2_journal_forget() - bforget() for potentially-journaled buffers.
* @handle: transaction handle
* @bh: bh to 'forget'
*
* We can only do the bforget if there are no commits pending against the
* buffer. If the buffer is dirty in the current running transaction we
* can safely unlink it.
*
* bh may not be a journalled buffer at all - it may be a non-JBD
* buffer which came off the hashtable. Check for this.
*
* Decrements bh->b_count by one.
*
* Allow this call even if the handle has aborted --- it may be part of
* the caller's cleanup after an abort.
*/
int jbd2_journal_forget(handle_t *handle, struct buffer_head *bh)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal;
struct journal_head *jh;
int drop_reserve = 0;
int err = 0;
jbd2: fix possible journal overflow issues There are several cases where the running transaction can get buffers added to its BJ_Metadata list which it never dirtied, which makes its t_nr_buffers counter end up larger than its t_outstanding_credits counter. This will cause issues when starting new transactions as while we are logging buffers we decrement t_outstanding_buffers, so when t_outstanding_buffers goes negative, we will report that we need less space in the journal than we actually need, so transactions will be started even though there may not be enough room for them. In the worst case scenario (which admittedly is almost impossible to reproduce) this will result in the journal running out of space. The fix is to only refile buffers from the committing transaction to the running transactions BJ_Modified list when b_modified is set on that journal, which is the only way to be sure if the running transaction has modified that buffer. This patch also fixes an accounting error in journal_forget, it is possible that we can call journal_forget on a buffer without having modified it, only gotten write access to it, so instead of freeing a credit, we only do so if the buffer was modified. The assert will help catch if this problem occurs. Without these two patches I could hit this assert within minutes of running postmark, with them this issue no longer arises. Cc: <linux-ext4@vger.kernel.org> Cc: Jan Kara <jack@ucw.cz> Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2008-04-17 14:38:59 +00:00
int was_modified = 0;
if (is_handle_aborted(handle))
return -EROFS;
journal = transaction->t_journal;
BUFFER_TRACE(bh, "entry");
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
jh = jbd2_journal_grab_journal_head(bh);
if (!jh) {
__bforget(bh);
return 0;
}
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
/* Critical error: attempting to delete a bitmap buffer, maybe?
* Don't do any jbd operations, and return an error. */
if (!J_EXPECT_JH(jh, !jh->b_committed_data,
"inconsistent data on disk")) {
err = -EIO;
goto drop;
}
/* keep track of whether or not this transaction modified us */
jbd2: fix possible journal overflow issues There are several cases where the running transaction can get buffers added to its BJ_Metadata list which it never dirtied, which makes its t_nr_buffers counter end up larger than its t_outstanding_credits counter. This will cause issues when starting new transactions as while we are logging buffers we decrement t_outstanding_buffers, so when t_outstanding_buffers goes negative, we will report that we need less space in the journal than we actually need, so transactions will be started even though there may not be enough room for them. In the worst case scenario (which admittedly is almost impossible to reproduce) this will result in the journal running out of space. The fix is to only refile buffers from the committing transaction to the running transactions BJ_Modified list when b_modified is set on that journal, which is the only way to be sure if the running transaction has modified that buffer. This patch also fixes an accounting error in journal_forget, it is possible that we can call journal_forget on a buffer without having modified it, only gotten write access to it, so instead of freeing a credit, we only do so if the buffer was modified. The assert will help catch if this problem occurs. Without these two patches I could hit this assert within minutes of running postmark, with them this issue no longer arises. Cc: <linux-ext4@vger.kernel.org> Cc: Jan Kara <jack@ucw.cz> Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2008-04-17 14:38:59 +00:00
was_modified = jh->b_modified;
/*
* The buffer's going from the transaction, we must drop
* all references -bzzz
*/
jh->b_modified = 0;
if (jh->b_transaction == transaction) {
J_ASSERT_JH(jh, !jh->b_frozen_data);
/* If we are forgetting a buffer which is already part
* of this transaction, then we can just drop it from
* the transaction immediately. */
clear_buffer_dirty(bh);
clear_buffer_jbddirty(bh);
JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
jbd2: fix possible journal overflow issues There are several cases where the running transaction can get buffers added to its BJ_Metadata list which it never dirtied, which makes its t_nr_buffers counter end up larger than its t_outstanding_credits counter. This will cause issues when starting new transactions as while we are logging buffers we decrement t_outstanding_buffers, so when t_outstanding_buffers goes negative, we will report that we need less space in the journal than we actually need, so transactions will be started even though there may not be enough room for them. In the worst case scenario (which admittedly is almost impossible to reproduce) this will result in the journal running out of space. The fix is to only refile buffers from the committing transaction to the running transactions BJ_Modified list when b_modified is set on that journal, which is the only way to be sure if the running transaction has modified that buffer. This patch also fixes an accounting error in journal_forget, it is possible that we can call journal_forget on a buffer without having modified it, only gotten write access to it, so instead of freeing a credit, we only do so if the buffer was modified. The assert will help catch if this problem occurs. Without these two patches I could hit this assert within minutes of running postmark, with them this issue no longer arises. Cc: <linux-ext4@vger.kernel.org> Cc: Jan Kara <jack@ucw.cz> Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2008-04-17 14:38:59 +00:00
/*
* we only want to drop a reference if this transaction
* modified the buffer
*/
if (was_modified)
drop_reserve = 1;
/*
* We are no longer going to journal this buffer.
* However, the commit of this transaction is still
* important to the buffer: the delete that we are now
* processing might obsolete an old log entry, so by
* committing, we can satisfy the buffer's checkpoint.
*
* So, if we have a checkpoint on the buffer, we should
* now refile the buffer on our BJ_Forget list so that
* we know to remove the checkpoint after we commit.
*/
spin_lock(&journal->j_list_lock);
if (jh->b_cp_transaction) {
__jbd2_journal_temp_unlink_buffer(jh);
__jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
} else {
__jbd2_journal_unfile_buffer(jh);
jbd2_journal_put_journal_head(jh);
}
spin_unlock(&journal->j_list_lock);
} else if (jh->b_transaction) {
J_ASSERT_JH(jh, (jh->b_transaction ==
journal->j_committing_transaction));
/* However, if the buffer is still owned by a prior
* (committing) transaction, we can't drop it yet... */
JBUFFER_TRACE(jh, "belongs to older transaction");
jbd2: clear dirty flag when revoking a buffer from an older transaction Now, we capture a data corruption problem on ext4 while we're truncating an extent index block. Imaging that if we are revoking a buffer which has been journaled by the committing transaction, the buffer's jbddirty flag will not be cleared in jbd2_journal_forget(), so the commit code will set the buffer dirty flag again after refile the buffer. fsx kjournald2 jbd2_journal_commit_transaction jbd2_journal_revoke commit phase 1~5... jbd2_journal_forget belongs to older transaction commit phase 6 jbddirty not clear __jbd2_journal_refile_buffer __jbd2_journal_unfile_buffer test_clear_buffer_jbddirty mark_buffer_dirty Finally, if the freed extent index block was allocated again as data block by some other files, it may corrupt the file data after writing cached pages later, such as during unmount time. (In general, clean_bdev_aliases() related helpers should be invoked after re-allocation to prevent the above corruption, but unfortunately we missed it when zeroout the head of extra extent blocks in ext4_ext_handle_unwritten_extents()). This patch mark buffer as freed and set j_next_transaction to the new transaction when it already belongs to the committing transaction in jbd2_journal_forget(), so that commit code knows it should clear dirty bits when it is done with the buffer. This problem can be reproduced by xfstests generic/455 easily with seeds (3246 3247 3248 3249). Signed-off-by: zhangyi (F) <yi.zhang@huawei.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jan Kara <jack@suse.cz> Cc: stable@vger.kernel.org
2019-02-11 04:23:04 +00:00
/* ... but we CAN drop it from the new transaction through
* marking the buffer as freed and set j_next_transaction to
* the new transaction, so that not only the commit code
* knows it should clear dirty bits when it is done with the
* buffer, but also the buffer can be checkpointed only
* after the new transaction commits. */
jbd2: clear dirty flag when revoking a buffer from an older transaction Now, we capture a data corruption problem on ext4 while we're truncating an extent index block. Imaging that if we are revoking a buffer which has been journaled by the committing transaction, the buffer's jbddirty flag will not be cleared in jbd2_journal_forget(), so the commit code will set the buffer dirty flag again after refile the buffer. fsx kjournald2 jbd2_journal_commit_transaction jbd2_journal_revoke commit phase 1~5... jbd2_journal_forget belongs to older transaction commit phase 6 jbddirty not clear __jbd2_journal_refile_buffer __jbd2_journal_unfile_buffer test_clear_buffer_jbddirty mark_buffer_dirty Finally, if the freed extent index block was allocated again as data block by some other files, it may corrupt the file data after writing cached pages later, such as during unmount time. (In general, clean_bdev_aliases() related helpers should be invoked after re-allocation to prevent the above corruption, but unfortunately we missed it when zeroout the head of extra extent blocks in ext4_ext_handle_unwritten_extents()). This patch mark buffer as freed and set j_next_transaction to the new transaction when it already belongs to the committing transaction in jbd2_journal_forget(), so that commit code knows it should clear dirty bits when it is done with the buffer. This problem can be reproduced by xfstests generic/455 easily with seeds (3246 3247 3248 3249). Signed-off-by: zhangyi (F) <yi.zhang@huawei.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jan Kara <jack@suse.cz> Cc: stable@vger.kernel.org
2019-02-11 04:23:04 +00:00
set_buffer_freed(bh);
if (!jh->b_next_transaction) {
spin_lock(&journal->j_list_lock);
jbd2: clear dirty flag when revoking a buffer from an older transaction Now, we capture a data corruption problem on ext4 while we're truncating an extent index block. Imaging that if we are revoking a buffer which has been journaled by the committing transaction, the buffer's jbddirty flag will not be cleared in jbd2_journal_forget(), so the commit code will set the buffer dirty flag again after refile the buffer. fsx kjournald2 jbd2_journal_commit_transaction jbd2_journal_revoke commit phase 1~5... jbd2_journal_forget belongs to older transaction commit phase 6 jbddirty not clear __jbd2_journal_refile_buffer __jbd2_journal_unfile_buffer test_clear_buffer_jbddirty mark_buffer_dirty Finally, if the freed extent index block was allocated again as data block by some other files, it may corrupt the file data after writing cached pages later, such as during unmount time. (In general, clean_bdev_aliases() related helpers should be invoked after re-allocation to prevent the above corruption, but unfortunately we missed it when zeroout the head of extra extent blocks in ext4_ext_handle_unwritten_extents()). This patch mark buffer as freed and set j_next_transaction to the new transaction when it already belongs to the committing transaction in jbd2_journal_forget(), so that commit code knows it should clear dirty bits when it is done with the buffer. This problem can be reproduced by xfstests generic/455 easily with seeds (3246 3247 3248 3249). Signed-off-by: zhangyi (F) <yi.zhang@huawei.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jan Kara <jack@suse.cz> Cc: stable@vger.kernel.org
2019-02-11 04:23:04 +00:00
jh->b_next_transaction = transaction;
spin_unlock(&journal->j_list_lock);
jbd2: clear dirty flag when revoking a buffer from an older transaction Now, we capture a data corruption problem on ext4 while we're truncating an extent index block. Imaging that if we are revoking a buffer which has been journaled by the committing transaction, the buffer's jbddirty flag will not be cleared in jbd2_journal_forget(), so the commit code will set the buffer dirty flag again after refile the buffer. fsx kjournald2 jbd2_journal_commit_transaction jbd2_journal_revoke commit phase 1~5... jbd2_journal_forget belongs to older transaction commit phase 6 jbddirty not clear __jbd2_journal_refile_buffer __jbd2_journal_unfile_buffer test_clear_buffer_jbddirty mark_buffer_dirty Finally, if the freed extent index block was allocated again as data block by some other files, it may corrupt the file data after writing cached pages later, such as during unmount time. (In general, clean_bdev_aliases() related helpers should be invoked after re-allocation to prevent the above corruption, but unfortunately we missed it when zeroout the head of extra extent blocks in ext4_ext_handle_unwritten_extents()). This patch mark buffer as freed and set j_next_transaction to the new transaction when it already belongs to the committing transaction in jbd2_journal_forget(), so that commit code knows it should clear dirty bits when it is done with the buffer. This problem can be reproduced by xfstests generic/455 easily with seeds (3246 3247 3248 3249). Signed-off-by: zhangyi (F) <yi.zhang@huawei.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jan Kara <jack@suse.cz> Cc: stable@vger.kernel.org
2019-02-11 04:23:04 +00:00
} else {
J_ASSERT(jh->b_next_transaction == transaction);
jbd2: fix possible journal overflow issues There are several cases where the running transaction can get buffers added to its BJ_Metadata list which it never dirtied, which makes its t_nr_buffers counter end up larger than its t_outstanding_credits counter. This will cause issues when starting new transactions as while we are logging buffers we decrement t_outstanding_buffers, so when t_outstanding_buffers goes negative, we will report that we need less space in the journal than we actually need, so transactions will be started even though there may not be enough room for them. In the worst case scenario (which admittedly is almost impossible to reproduce) this will result in the journal running out of space. The fix is to only refile buffers from the committing transaction to the running transactions BJ_Modified list when b_modified is set on that journal, which is the only way to be sure if the running transaction has modified that buffer. This patch also fixes an accounting error in journal_forget, it is possible that we can call journal_forget on a buffer without having modified it, only gotten write access to it, so instead of freeing a credit, we only do so if the buffer was modified. The assert will help catch if this problem occurs. Without these two patches I could hit this assert within minutes of running postmark, with them this issue no longer arises. Cc: <linux-ext4@vger.kernel.org> Cc: Jan Kara <jack@ucw.cz> Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2008-04-17 14:38:59 +00:00
/*
* only drop a reference if this transaction modified
* the buffer
*/
if (was_modified)
drop_reserve = 1;
}
} else {
/*
* Finally, if the buffer is not belongs to any
* transaction, we can just drop it now if it has no
* checkpoint.
*/
spin_lock(&journal->j_list_lock);
if (!jh->b_cp_transaction) {
JBUFFER_TRACE(jh, "belongs to none transaction");
spin_unlock(&journal->j_list_lock);
goto drop;
}
/*
* Otherwise, if the buffer has been written to disk,
* it is safe to remove the checkpoint and drop it.
*/
if (jbd2_journal_try_remove_checkpoint(jh) >= 0) {
spin_unlock(&journal->j_list_lock);
goto drop;
}
/*
* The buffer is still not written to disk, we should
* attach this buffer to current transaction so that the
* buffer can be checkpointed only after the current
* transaction commits.
*/
clear_buffer_dirty(bh);
__jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
spin_unlock(&journal->j_list_lock);
}
drop:
__brelse(bh);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
jbd2_journal_put_journal_head(jh);
if (drop_reserve) {
/* no need to reserve log space for this block -bzzz */
handle->h_total_credits++;
}
return err;
}
/**
* jbd2_journal_stop() - complete a transaction
* @handle: transaction to complete.
*
* All done for a particular handle.
*
* There is not much action needed here. We just return any remaining
* buffer credits to the transaction and remove the handle. The only
* complication is that we need to start a commit operation if the
* filesystem is marked for synchronous update.
*
* jbd2_journal_stop itself will not usually return an error, but it may
* do so in unusual circumstances. In particular, expect it to
* return -EIO if a jbd2_journal_abort has been executed since the
* transaction began.
*/
int jbd2_journal_stop(handle_t *handle)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal;
int err = 0, wait_for_commit = 0;
tid_t tid;
pid_t pid;
if (--handle->h_ref > 0) {
jbd2_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
handle->h_ref);
if (is_handle_aborted(handle))
return -EIO;
return 0;
}
ext4: fix NULL pointer dereference when journal restart fails Currently when journal restart fails, we'll have the h_transaction of the handle set to NULL to indicate that the handle has been effectively aborted. We handle this situation quietly in the jbd2_journal_stop() and just free the handle and exit because everything else has been done before we attempted (and failed) to restart the journal. Unfortunately there are a number of problems with that approach introduced with commit 41a5b913197c "jbd2: invalidate handle if jbd2_journal_restart() fails" First of all in ext4 jbd2_journal_stop() will be called through __ext4_journal_stop() where we would try to get a hold of the superblock by dereferencing h_transaction which in this case would lead to NULL pointer dereference and crash. In addition we're going to free the handle regardless of the refcount which is bad as well, because others up the call chain will still reference the handle so we might potentially reference already freed memory. Moreover it's expected that we'll get aborted handle as well as detached handle in some of the journalling function as the error propagates up the stack, so it's unnecessary to call WARN_ON every time we get detached handle. And finally we might leak some memory by forgetting to free reserved handle in jbd2_journal_stop() in the case where handle was detached from the transaction (h_transaction is NULL). Fix the NULL pointer dereference in __ext4_journal_stop() by just calling jbd2_journal_stop() quietly as suggested by Jan Kara. Also fix the potential memory leak in jbd2_journal_stop() and use proper handle refcounting before we attempt to free it to avoid use-after-free issues. And finally remove all WARN_ON(!transaction) from the code so that we do not get random traces when something goes wrong because when journal restart fails we will get to some of those functions. Cc: stable@vger.kernel.org Signed-off-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jan Kara <jack@suse.cz>
2015-05-14 22:55:18 +00:00
if (!transaction) {
/*
* Handle is already detached from the transaction so there is
* nothing to do other than free the handle.
ext4: fix NULL pointer dereference when journal restart fails Currently when journal restart fails, we'll have the h_transaction of the handle set to NULL to indicate that the handle has been effectively aborted. We handle this situation quietly in the jbd2_journal_stop() and just free the handle and exit because everything else has been done before we attempted (and failed) to restart the journal. Unfortunately there are a number of problems with that approach introduced with commit 41a5b913197c "jbd2: invalidate handle if jbd2_journal_restart() fails" First of all in ext4 jbd2_journal_stop() will be called through __ext4_journal_stop() where we would try to get a hold of the superblock by dereferencing h_transaction which in this case would lead to NULL pointer dereference and crash. In addition we're going to free the handle regardless of the refcount which is bad as well, because others up the call chain will still reference the handle so we might potentially reference already freed memory. Moreover it's expected that we'll get aborted handle as well as detached handle in some of the journalling function as the error propagates up the stack, so it's unnecessary to call WARN_ON every time we get detached handle. And finally we might leak some memory by forgetting to free reserved handle in jbd2_journal_stop() in the case where handle was detached from the transaction (h_transaction is NULL). Fix the NULL pointer dereference in __ext4_journal_stop() by just calling jbd2_journal_stop() quietly as suggested by Jan Kara. Also fix the potential memory leak in jbd2_journal_stop() and use proper handle refcounting before we attempt to free it to avoid use-after-free issues. And finally remove all WARN_ON(!transaction) from the code so that we do not get random traces when something goes wrong because when journal restart fails we will get to some of those functions. Cc: stable@vger.kernel.org Signed-off-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jan Kara <jack@suse.cz>
2015-05-14 22:55:18 +00:00
*/
memalloc_nofs_restore(handle->saved_alloc_context);
goto free_and_exit;
ext4: fix NULL pointer dereference when journal restart fails Currently when journal restart fails, we'll have the h_transaction of the handle set to NULL to indicate that the handle has been effectively aborted. We handle this situation quietly in the jbd2_journal_stop() and just free the handle and exit because everything else has been done before we attempted (and failed) to restart the journal. Unfortunately there are a number of problems with that approach introduced with commit 41a5b913197c "jbd2: invalidate handle if jbd2_journal_restart() fails" First of all in ext4 jbd2_journal_stop() will be called through __ext4_journal_stop() where we would try to get a hold of the superblock by dereferencing h_transaction which in this case would lead to NULL pointer dereference and crash. In addition we're going to free the handle regardless of the refcount which is bad as well, because others up the call chain will still reference the handle so we might potentially reference already freed memory. Moreover it's expected that we'll get aborted handle as well as detached handle in some of the journalling function as the error propagates up the stack, so it's unnecessary to call WARN_ON every time we get detached handle. And finally we might leak some memory by forgetting to free reserved handle in jbd2_journal_stop() in the case where handle was detached from the transaction (h_transaction is NULL). Fix the NULL pointer dereference in __ext4_journal_stop() by just calling jbd2_journal_stop() quietly as suggested by Jan Kara. Also fix the potential memory leak in jbd2_journal_stop() and use proper handle refcounting before we attempt to free it to avoid use-after-free issues. And finally remove all WARN_ON(!transaction) from the code so that we do not get random traces when something goes wrong because when journal restart fails we will get to some of those functions. Cc: stable@vger.kernel.org Signed-off-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Jan Kara <jack@suse.cz>
2015-05-14 22:55:18 +00:00
}
journal = transaction->t_journal;
tid = transaction->t_tid;
if (is_handle_aborted(handle))
err = -EIO;
jbd2_debug(4, "Handle %p going down\n", handle);
trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
tid, handle->h_type, handle->h_line_no,
jiffies - handle->h_start_jiffies,
handle->h_sync, handle->h_requested_credits,
(handle->h_requested_credits -
handle->h_total_credits));
/*
* Implement synchronous transaction batching. If the handle
* was synchronous, don't force a commit immediately. Let's
* yield and let another thread piggyback onto this
* transaction. Keep doing that while new threads continue to
* arrive. It doesn't cost much - we're about to run a commit
* and sleep on IO anyway. Speeds up many-threaded, many-dir
* operations by 30x or more...
*
* We try and optimize the sleep time against what the
* underlying disk can do, instead of having a static sleep
* time. This is useful for the case where our storage is so
* fast that it is more optimal to go ahead and force a flush
* and wait for the transaction to be committed than it is to
* wait for an arbitrary amount of time for new writers to
* join the transaction. We achieve this by measuring how
* long it takes to commit a transaction, and compare it with
* how long this transaction has been running, and if run time
* < commit time then we sleep for the delta and commit. This
* greatly helps super fast disks that would see slowdowns as
* more threads started doing fsyncs.
*
* But don't do this if this process was the most recent one
* to perform a synchronous write. We do this to detect the
* case where a single process is doing a stream of sync
* writes. No point in waiting for joiners in that case.
*
* Setting max_batch_time to 0 disables this completely.
*/
pid = current->pid;
if (handle->h_sync && journal->j_last_sync_writer != pid &&
journal->j_max_batch_time) {
u64 commit_time, trans_time;
journal->j_last_sync_writer = pid;
read_lock(&journal->j_state_lock);
commit_time = journal->j_average_commit_time;
read_unlock(&journal->j_state_lock);
trans_time = ktime_to_ns(ktime_sub(ktime_get(),
transaction->t_start_time));
commit_time = max_t(u64, commit_time,
1000*journal->j_min_batch_time);
commit_time = min_t(u64, commit_time,
1000*journal->j_max_batch_time);
if (trans_time < commit_time) {
ktime_t expires = ktime_add_ns(ktime_get(),
commit_time);
set_current_state(TASK_UNINTERRUPTIBLE);
schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
}
}
if (handle->h_sync)
transaction->t_synchronous_commit = 1;
/*
* If the handle is marked SYNC, we need to set another commit
* going! We also want to force a commit if the transaction is too
* old now.
*/
if (handle->h_sync ||
time_after_eq(jiffies, transaction->t_expires)) {
/* Do this even for aborted journals: an abort still
* completes the commit thread, it just doesn't write
* anything to disk. */
jbd2_debug(2, "transaction too old, requesting commit for "
"handle %p\n", handle);
/* This is non-blocking */
jbd2_log_start_commit(journal, tid);
/*
* Special case: JBD2_SYNC synchronous updates require us
* to wait for the commit to complete.
*/
if (handle->h_sync && !(current->flags & PF_MEMALLOC))
wait_for_commit = 1;
}
/*
* Once stop_this_handle() drops t_updates, the transaction could start
* committing on us and eventually disappear. So we must not
* dereference transaction pointer again after calling
* stop_this_handle().
*/
stop_this_handle(handle);
if (wait_for_commit)
err = jbd2_log_wait_commit(journal, tid);
free_and_exit:
if (handle->h_rsv_handle)
jbd2_free_handle(handle->h_rsv_handle);
jbd2_free_handle(handle);
return err;
}
/*
*
* List management code snippets: various functions for manipulating the
* transaction buffer lists.
*
*/
/*
* Append a buffer to a transaction list, given the transaction's list head
* pointer.
*
* j_list_lock is held.
*
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
* jh->b_state_lock is held.
*/
static inline void
__blist_add_buffer(struct journal_head **list, struct journal_head *jh)
{
if (!*list) {
jh->b_tnext = jh->b_tprev = jh;
*list = jh;
} else {
/* Insert at the tail of the list to preserve order */
struct journal_head *first = *list, *last = first->b_tprev;
jh->b_tprev = last;
jh->b_tnext = first;
last->b_tnext = first->b_tprev = jh;
}
}
/*
* Remove a buffer from a transaction list, given the transaction's list
* head pointer.
*
* Called with j_list_lock held, and the journal may not be locked.
*
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
* jh->b_state_lock is held.
*/
static inline void
__blist_del_buffer(struct journal_head **list, struct journal_head *jh)
{
if (*list == jh) {
*list = jh->b_tnext;
if (*list == jh)
*list = NULL;
}
jh->b_tprev->b_tnext = jh->b_tnext;
jh->b_tnext->b_tprev = jh->b_tprev;
}
/*
* Remove a buffer from the appropriate transaction list.
*
* Note that this function can *change* the value of
* bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
* t_reserved_list. If the caller is holding onto a copy of one of these
* pointers, it could go bad. Generally the caller needs to re-read the
* pointer from the transaction_t.
*
* Called under j_list_lock.
*/
static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
{
struct journal_head **list = NULL;
transaction_t *transaction;
struct buffer_head *bh = jh2bh(jh);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
lockdep_assert_held(&jh->b_state_lock);
transaction = jh->b_transaction;
if (transaction)
assert_spin_locked(&transaction->t_journal->j_list_lock);
J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
if (jh->b_jlist != BJ_None)
J_ASSERT_JH(jh, transaction != NULL);
switch (jh->b_jlist) {
case BJ_None:
return;
case BJ_Metadata:
transaction->t_nr_buffers--;
J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
list = &transaction->t_buffers;
break;
case BJ_Forget:
list = &transaction->t_forget;
break;
case BJ_Shadow:
list = &transaction->t_shadow_list;
break;
case BJ_Reserved:
list = &transaction->t_reserved_list;
break;
}
__blist_del_buffer(list, jh);
jh->b_jlist = BJ_None;
if (transaction && is_journal_aborted(transaction->t_journal))
clear_buffer_jbddirty(bh);
else if (test_clear_buffer_jbddirty(bh))
mark_buffer_dirty(bh); /* Expose it to the VM */
}
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
/*
* Remove buffer from all transactions. The caller is responsible for dropping
* the jh reference that belonged to the transaction.
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
*
* Called with bh_state lock and j_list_lock
*/
static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
{
J_ASSERT_JH(jh, jh->b_transaction != NULL);
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
__jbd2_journal_temp_unlink_buffer(jh);
jh->b_transaction = NULL;
}
void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
{
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
struct buffer_head *bh = jh2bh(jh);
/* Get reference so that buffer cannot be freed before we unlock it */
get_bh(bh);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
spin_lock(&journal->j_list_lock);
__jbd2_journal_unfile_buffer(jh);
spin_unlock(&journal->j_list_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
jbd2_journal_put_journal_head(jh);
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
__brelse(bh);
}
/**
* jbd2_journal_try_to_free_buffers() - try to free page buffers.
* @journal: journal for operation
* @folio: Folio to detach data from.
*
* For all the buffers on this page,
* if they are fully written out ordered data, move them onto BUF_CLEAN
* so try_to_free_buffers() can reap them.
*
* This function returns non-zero if we wish try_to_free_buffers()
* to be called. We do this if the page is releasable by try_to_free_buffers().
* We also do it if the page has locked or dirty buffers and the caller wants
* us to perform sync or async writeout.
*
* This complicates JBD locking somewhat. We aren't protected by the
* BKL here. We wish to remove the buffer from its committing or
* running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
*
* This may *change* the value of transaction_t->t_datalist, so anyone
* who looks at t_datalist needs to lock against this function.
*
* Even worse, someone may be doing a jbd2_journal_dirty_data on this
* buffer. So we need to lock against that. jbd2_journal_dirty_data()
* will come out of the lock with the buffer dirty, which makes it
* ineligible for release here.
*
* Who else is affected by this? hmm... Really the only contender
* is do_get_write_access() - it could be looking at the buffer while
* journal_try_to_free_buffer() is changing its state. But that
* cannot happen because we never reallocate freed data as metadata
* while the data is part of a transaction. Yes?
*
* Return false on failure, true on success
*/
bool jbd2_journal_try_to_free_buffers(journal_t *journal, struct folio *folio)
{
struct buffer_head *head;
struct buffer_head *bh;
bool ret = false;
J_ASSERT(folio_test_locked(folio));
head = folio_buffers(folio);
bh = head;
do {
struct journal_head *jh;
/*
* We take our own ref against the journal_head here to avoid
* having to add tons of locking around each instance of
* jbd2_journal_put_journal_head().
*/
jh = jbd2_journal_grab_journal_head(bh);
if (!jh)
continue;
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
if (!jh->b_transaction && !jh->b_next_transaction) {
spin_lock(&journal->j_list_lock);
/* Remove written-back checkpointed metadata buffer */
if (jh->b_cp_transaction != NULL)
jbd2_journal_try_remove_checkpoint(jh);
spin_unlock(&journal->j_list_lock);
}
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
jbd2_journal_put_journal_head(jh);
if (buffer_jbd(bh))
goto busy;
} while ((bh = bh->b_this_page) != head);
ret = try_to_free_buffers(folio);
busy:
return ret;
}
/*
* This buffer is no longer needed. If it is on an older transaction's
* checkpoint list we need to record it on this transaction's forget list
* to pin this buffer (and hence its checkpointing transaction) down until
* this transaction commits. If the buffer isn't on a checkpoint list, we
* release it.
* Returns non-zero if JBD no longer has an interest in the buffer.
*
* Called under j_list_lock.
*
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
* Called under jh->b_state_lock.
*/
static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
{
int may_free = 1;
struct buffer_head *bh = jh2bh(jh);
if (jh->b_cp_transaction) {
JBUFFER_TRACE(jh, "on running+cp transaction");
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
__jbd2_journal_temp_unlink_buffer(jh);
/*
* We don't want to write the buffer anymore, clear the
* bit so that we don't confuse checks in
* __journal_file_buffer
*/
clear_buffer_dirty(bh);
__jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
may_free = 0;
} else {
JBUFFER_TRACE(jh, "on running transaction");
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
__jbd2_journal_unfile_buffer(jh);
jbd2_journal_put_journal_head(jh);
}
return may_free;
}
/*
* jbd2_journal_invalidate_folio
*
* This code is tricky. It has a number of cases to deal with.
*
* There are two invariants which this code relies on:
*
* i_size must be updated on disk before we start calling invalidate_folio
* on the data.
*
* This is done in ext3 by defining an ext3_setattr method which
* updates i_size before truncate gets going. By maintaining this
* invariant, we can be sure that it is safe to throw away any buffers
* attached to the current transaction: once the transaction commits,
* we know that the data will not be needed.
*
* Note however that we can *not* throw away data belonging to the
* previous, committing transaction!
*
* Any disk blocks which *are* part of the previous, committing
* transaction (and which therefore cannot be discarded immediately) are
* not going to be reused in the new running transaction
*
* The bitmap committed_data images guarantee this: any block which is
* allocated in one transaction and removed in the next will be marked
* as in-use in the committed_data bitmap, so cannot be reused until
* the next transaction to delete the block commits. This means that
* leaving committing buffers dirty is quite safe: the disk blocks
* cannot be reallocated to a different file and so buffer aliasing is
* not possible.
*
*
* The above applies mainly to ordered data mode. In writeback mode we
* don't make guarantees about the order in which data hits disk --- in
* particular we don't guarantee that new dirty data is flushed before
* transaction commit --- so it is always safe just to discard data
* immediately in that mode. --sct
*/
/*
* The journal_unmap_buffer helper function returns zero if the buffer
* concerned remains pinned as an anonymous buffer belonging to an older
* transaction.
*
* We're outside-transaction here. Either or both of j_running_transaction
* and j_committing_transaction may be NULL.
*/
static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
int partial_page)
{
transaction_t *transaction;
struct journal_head *jh;
int may_free = 1;
BUFFER_TRACE(bh, "entry");
/*
* It is safe to proceed here without the j_list_lock because the
* buffers cannot be stolen by try_to_free_buffers as long as we are
* holding the page lock. --sct
*/
jh = jbd2_journal_grab_journal_head(bh);
if (!jh)
goto zap_buffer_unlocked;
/* OK, we have data buffer in journaled mode */
write_lock(&journal->j_state_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
spin_lock(&journal->j_list_lock);
/*
* We cannot remove the buffer from checkpoint lists until the
* transaction adding inode to orphan list (let's call it T)
* is committed. Otherwise if the transaction changing the
* buffer would be cleaned from the journal before T is
* committed, a crash will cause that the correct contents of
* the buffer will be lost. On the other hand we have to
* clear the buffer dirty bit at latest at the moment when the
* transaction marking the buffer as freed in the filesystem
* structures is committed because from that moment on the
* block can be reallocated and used by a different page.
* Since the block hasn't been freed yet but the inode has
* already been added to orphan list, it is safe for us to add
* the buffer to BJ_Forget list of the newest transaction.
*
* Also we have to clear buffer_mapped flag of a truncated buffer
* because the buffer_head may be attached to the page straddling
* i_size (can happen only when blocksize < pagesize) and thus the
* buffer_head can be reused when the file is extended again. So we end
* up keeping around invalidated buffers attached to transactions'
* BJ_Forget list just to stop checkpointing code from cleaning up
* the transaction this buffer was modified in.
*/
transaction = jh->b_transaction;
if (transaction == NULL) {
/* First case: not on any transaction. If it
* has no checkpoint link, then we can zap it:
* it's a writeback-mode buffer so we don't care
* if it hits disk safely. */
if (!jh->b_cp_transaction) {
JBUFFER_TRACE(jh, "not on any transaction: zap");
goto zap_buffer;
}
if (!buffer_dirty(bh)) {
/* bdflush has written it. We can drop it now */
__jbd2_journal_remove_checkpoint(jh);
goto zap_buffer;
}
/* OK, it must be in the journal but still not
* written fully to disk: it's metadata or
* journaled data... */
if (journal->j_running_transaction) {
/* ... and once the current transaction has
* committed, the buffer won't be needed any
* longer. */
JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
may_free = __dispose_buffer(jh,
journal->j_running_transaction);
goto zap_buffer;
} else {
/* There is no currently-running transaction. So the
* orphan record which we wrote for this file must have
* passed into commit. We must attach this buffer to
* the committing transaction, if it exists. */
if (journal->j_committing_transaction) {
JBUFFER_TRACE(jh, "give to committing trans");
may_free = __dispose_buffer(jh,
journal->j_committing_transaction);
goto zap_buffer;
} else {
/* The orphan record's transaction has
* committed. We can cleanse this buffer */
clear_buffer_jbddirty(bh);
__jbd2_journal_remove_checkpoint(jh);
goto zap_buffer;
}
}
} else if (transaction == journal->j_committing_transaction) {
JBUFFER_TRACE(jh, "on committing transaction");
/*
* The buffer is committing, we simply cannot touch
* it. If the page is straddling i_size we have to wait
* for commit and try again.
*/
if (partial_page) {
spin_unlock(&journal->j_list_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
write_unlock(&journal->j_state_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
jbd2_journal_put_journal_head(jh);
/* Already zapped buffer? Nothing to do... */
if (!bh->b_bdev)
return 0;
return -EBUSY;
}
/*
* OK, buffer won't be reachable after truncate. We just clear
* b_modified to not confuse transaction credit accounting, and
* set j_next_transaction to the running transaction (if there
* is one) and mark buffer as freed so that commit code knows
* it should clear dirty bits when it is done with the buffer.
*/
set_buffer_freed(bh);
if (journal->j_running_transaction && buffer_jbddirty(bh))
jh->b_next_transaction = journal->j_running_transaction;
jh->b_modified = 0;
spin_unlock(&journal->j_list_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
write_unlock(&journal->j_state_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
jbd2_journal_put_journal_head(jh);
return 0;
} else {
/* Good, the buffer belongs to the running transaction.
* We are writing our own transaction's data, not any
* previous one's, so it is safe to throw it away
* (remember that we expect the filesystem to have set
* i_size already for this truncate so recovery will not
* expose the disk blocks we are discarding here.) */
J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
JBUFFER_TRACE(jh, "on running transaction");
may_free = __dispose_buffer(jh, transaction);
}
zap_buffer:
/*
* This is tricky. Although the buffer is truncated, it may be reused
* if blocksize < pagesize and it is attached to the page straddling
* EOF. Since the buffer might have been added to BJ_Forget list of the
* running transaction, journal_get_write_access() won't clear
* b_modified and credit accounting gets confused. So clear b_modified
* here.
*/
jh->b_modified = 0;
spin_unlock(&journal->j_list_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
write_unlock(&journal->j_state_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
jbd2_journal_put_journal_head(jh);
zap_buffer_unlocked:
clear_buffer_dirty(bh);
J_ASSERT_BH(bh, !buffer_jbddirty(bh));
clear_buffer_mapped(bh);
clear_buffer_req(bh);
clear_buffer_new(bh);
clear_buffer_delay(bh);
clear_buffer_unwritten(bh);
bh->b_bdev = NULL;
return may_free;
}
/**
* jbd2_journal_invalidate_folio()
* @journal: journal to use for flush...
* @folio: folio to flush
* @offset: start of the range to invalidate
* @length: length of the range to invalidate
*
* Reap page buffers containing data after in the specified range in page.
* Can return -EBUSY if buffers are part of the committing transaction and
* the page is straddling i_size. Caller then has to wait for current commit
* and try again.
*/
int jbd2_journal_invalidate_folio(journal_t *journal, struct folio *folio,
size_t offset, size_t length)
{
struct buffer_head *head, *bh, *next;
unsigned int stop = offset + length;
unsigned int curr_off = 0;
int partial_page = (offset || length < folio_size(folio));
int may_free = 1;
int ret = 0;
if (!folio_test_locked(folio))
BUG();
head = folio_buffers(folio);
if (!head)
return 0;
BUG_ON(stop > folio_size(folio) || stop < length);
/* We will potentially be playing with lists other than just the
* data lists (especially for journaled data mode), so be
* cautious in our locking. */
bh = head;
do {
unsigned int next_off = curr_off + bh->b_size;
next = bh->b_this_page;
if (next_off > stop)
return 0;
if (offset <= curr_off) {
/* This block is wholly outside the truncation point */
lock_buffer(bh);
ret = journal_unmap_buffer(journal, bh, partial_page);
unlock_buffer(bh);
if (ret < 0)
return ret;
may_free &= ret;
}
curr_off = next_off;
bh = next;
} while (bh != head);
if (!partial_page) {
if (may_free && try_to_free_buffers(folio))
J_ASSERT(!folio_buffers(folio));
}
return 0;
}
/*
* File a buffer on the given transaction list.
*/
void __jbd2_journal_file_buffer(struct journal_head *jh,
transaction_t *transaction, int jlist)
{
struct journal_head **list = NULL;
int was_dirty = 0;
struct buffer_head *bh = jh2bh(jh);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
lockdep_assert_held(&jh->b_state_lock);
assert_spin_locked(&transaction->t_journal->j_list_lock);
J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
J_ASSERT_JH(jh, jh->b_transaction == transaction ||
jh->b_transaction == NULL);
if (jh->b_transaction && jh->b_jlist == jlist)
return;
if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
jlist == BJ_Shadow || jlist == BJ_Forget) {
/*
* For metadata buffers, we track dirty bit in buffer_jbddirty
* instead of buffer_dirty. We should not see a dirty bit set
* here because we clear it in do_get_write_access but e.g.
* tune2fs can modify the sb and set the dirty bit at any time
* so we try to gracefully handle that.
*/
if (buffer_dirty(bh))
warn_dirty_buffer(bh);
if (test_clear_buffer_dirty(bh) ||
test_clear_buffer_jbddirty(bh))
was_dirty = 1;
}
if (jh->b_transaction)
__jbd2_journal_temp_unlink_buffer(jh);
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
else
jbd2_journal_grab_journal_head(bh);
jh->b_transaction = transaction;
switch (jlist) {
case BJ_None:
J_ASSERT_JH(jh, !jh->b_committed_data);
J_ASSERT_JH(jh, !jh->b_frozen_data);
return;
case BJ_Metadata:
transaction->t_nr_buffers++;
list = &transaction->t_buffers;
break;
case BJ_Forget:
list = &transaction->t_forget;
break;
case BJ_Shadow:
list = &transaction->t_shadow_list;
break;
case BJ_Reserved:
list = &transaction->t_reserved_list;
break;
}
__blist_add_buffer(list, jh);
jh->b_jlist = jlist;
if (was_dirty)
set_buffer_jbddirty(bh);
}
void jbd2_journal_file_buffer(struct journal_head *jh,
transaction_t *transaction, int jlist)
{
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
spin_lock(&transaction->t_journal->j_list_lock);
__jbd2_journal_file_buffer(jh, transaction, jlist);
spin_unlock(&transaction->t_journal->j_list_lock);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
}
/*
* Remove a buffer from its current buffer list in preparation for
* dropping it from its current transaction entirely. If the buffer has
* already started to be used by a subsequent transaction, refile the
* buffer on that transaction's metadata list.
*
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
* Called under j_list_lock
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
* Called under jh->b_state_lock
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
*
* When this function returns true, there's no next transaction to refile to
* and the caller has to drop jh reference through
* jbd2_journal_put_journal_head().
*/
bool __jbd2_journal_refile_buffer(struct journal_head *jh)
{
int was_dirty, jlist;
struct buffer_head *bh = jh2bh(jh);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
lockdep_assert_held(&jh->b_state_lock);
if (jh->b_transaction)
assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
/* If the buffer is now unused, just drop it. */
if (jh->b_next_transaction == NULL) {
__jbd2_journal_unfile_buffer(jh);
return true;
}
/*
* It has been modified by a later transaction: add it to the new
* transaction's metadata list.
*/
was_dirty = test_clear_buffer_jbddirty(bh);
__jbd2_journal_temp_unlink_buffer(jh);
/*
* b_transaction must be set, otherwise the new b_transaction won't
* be holding jh reference
*/
J_ASSERT_JH(jh, jh->b_transaction != NULL);
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
/*
* We set b_transaction here because b_next_transaction will inherit
* our jh reference and thus __jbd2_journal_file_buffer() must not
* take a new one.
*/
jbd2: fix data races at struct journal_head journal_head::b_transaction and journal_head::b_next_transaction could be accessed concurrently as noticed by KCSAN, LTP: starting fsync04 /dev/zero: Can't open blockdev EXT4-fs (loop0): mounting ext3 file system using the ext4 subsystem EXT4-fs (loop0): mounted filesystem with ordered data mode. Opts: (null) ================================================================== BUG: KCSAN: data-race in __jbd2_journal_refile_buffer [jbd2] / jbd2_write_access_granted [jbd2] write to 0xffff99f9b1bd0e30 of 8 bytes by task 25721 on cpu 70: __jbd2_journal_refile_buffer+0xdd/0x210 [jbd2] __jbd2_journal_refile_buffer at fs/jbd2/transaction.c:2569 jbd2_journal_commit_transaction+0x2d15/0x3f20 [jbd2] (inlined by) jbd2_journal_commit_transaction at fs/jbd2/commit.c:1034 kjournald2+0x13b/0x450 [jbd2] kthread+0x1cd/0x1f0 ret_from_fork+0x27/0x50 read to 0xffff99f9b1bd0e30 of 8 bytes by task 25724 on cpu 68: jbd2_write_access_granted+0x1b2/0x250 [jbd2] jbd2_write_access_granted at fs/jbd2/transaction.c:1155 jbd2_journal_get_write_access+0x2c/0x60 [jbd2] __ext4_journal_get_write_access+0x50/0x90 [ext4] ext4_mb_mark_diskspace_used+0x158/0x620 [ext4] ext4_mb_new_blocks+0x54f/0xca0 [ext4] ext4_ind_map_blocks+0xc79/0x1b40 [ext4] ext4_map_blocks+0x3b4/0x950 [ext4] _ext4_get_block+0xfc/0x270 [ext4] ext4_get_block+0x3b/0x50 [ext4] __block_write_begin_int+0x22e/0xae0 __block_write_begin+0x39/0x50 ext4_write_begin+0x388/0xb50 [ext4] generic_perform_write+0x15d/0x290 ext4_buffered_write_iter+0x11f/0x210 [ext4] ext4_file_write_iter+0xce/0x9e0 [ext4] new_sync_write+0x29c/0x3b0 __vfs_write+0x92/0xa0 vfs_write+0x103/0x260 ksys_write+0x9d/0x130 __x64_sys_write+0x4c/0x60 do_syscall_64+0x91/0xb05 entry_SYSCALL_64_after_hwframe+0x49/0xbe 5 locks held by fsync04/25724: #0: ffff99f9911093f8 (sb_writers#13){.+.+}, at: vfs_write+0x21c/0x260 #1: ffff99f9db4c0348 (&sb->s_type->i_mutex_key#15){+.+.}, at: ext4_buffered_write_iter+0x65/0x210 [ext4] #2: ffff99f5e7dfcf58 (jbd2_handle){++++}, at: start_this_handle+0x1c1/0x9d0 [jbd2] #3: ffff99f9db4c0168 (&ei->i_data_sem){++++}, at: ext4_map_blocks+0x176/0x950 [ext4] #4: ffffffff99086b40 (rcu_read_lock){....}, at: jbd2_write_access_granted+0x4e/0x250 [jbd2] irq event stamp: 1407125 hardirqs last enabled at (1407125): [<ffffffff980da9b7>] __find_get_block+0x107/0x790 hardirqs last disabled at (1407124): [<ffffffff980da8f9>] __find_get_block+0x49/0x790 softirqs last enabled at (1405528): [<ffffffff98a0034c>] __do_softirq+0x34c/0x57c softirqs last disabled at (1405521): [<ffffffff97cc67a2>] irq_exit+0xa2/0xc0 Reported by Kernel Concurrency Sanitizer on: CPU: 68 PID: 25724 Comm: fsync04 Tainted: G L 5.6.0-rc2-next-20200221+ #7 Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 07/10/2019 The plain reads are outside of jh->b_state_lock critical section which result in data races. Fix them by adding pairs of READ|WRITE_ONCE(). Reviewed-by: Jan Kara <jack@suse.cz> Signed-off-by: Qian Cai <cai@lca.pw> Link: https://lore.kernel.org/r/20200222043111.2227-1-cai@lca.pw Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2020-02-22 04:31:11 +00:00
WRITE_ONCE(jh->b_transaction, jh->b_next_transaction);
WRITE_ONCE(jh->b_next_transaction, NULL);
if (buffer_freed(bh))
jlist = BJ_Forget;
else if (jh->b_modified)
jlist = BJ_Metadata;
else
jlist = BJ_Reserved;
__jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
if (was_dirty)
set_buffer_jbddirty(bh);
return false;
}
/*
jbd2: Fix oops in jbd2_journal_remove_journal_head() jbd2_journal_remove_journal_head() can oops when trying to access journal_head returned by bh2jh(). This is caused for example by the following race: TASK1 TASK2 jbd2_journal_commit_transaction() ... processing t_forget list __jbd2_journal_refile_buffer(jh); if (!jh->b_transaction) { jbd_unlock_bh_state(bh); jbd2_journal_try_to_free_buffers() jbd2_journal_grab_journal_head(bh) jbd_lock_bh_state(bh) __journal_try_to_free_buffer() jbd2_journal_put_journal_head(jh) jbd2_journal_remove_journal_head(bh); jbd2_journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not part of any transaction and thus frees journal_head before TASK1 gets to doing so. Note that even buffer_head can be released by try_to_free_buffers() after jbd2_journal_put_journal_head() which adds even larger opportunity for oops (but I didn't see this happen in reality). Fix the problem by making transactions hold their own journal_head reference (in b_jcount). That way we don't have to remove journal_head explicitely via jbd2_journal_remove_journal_head() and instead just remove journal_head when b_jcount drops to zero. The result of this is that [__]jbd2_journal_refile_buffer(), [__]jbd2_journal_unfile_buffer(), and __jdb2_journal_remove_checkpoint() can free journal_head which needs modification of a few callers. Also we have to be careful because once journal_head is removed, buffer_head might be freed as well. So we have to get our own buffer_head reference where it matters. Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
2011-06-13 19:38:22 +00:00
* __jbd2_journal_refile_buffer() with necessary locking added. We take our
* bh reference so that we can safely unlock bh.
*
* The jh and bh may be freed by this call.
*/
void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
{
bool drop;
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_lock(&jh->b_state_lock);
spin_lock(&journal->j_list_lock);
drop = __jbd2_journal_refile_buffer(jh);
jbd2: Make state lock a spinlock Bit-spinlocks are problematic on PREEMPT_RT if functions which might sleep on RT, e.g. spin_lock(), alloc/free(), are invoked inside the lock held region because bit spinlocks disable preemption even on RT. A first attempt was to replace state lock with a spinlock placed in struct buffer_head and make the locking conditional on PREEMPT_RT and DEBUG_BIT_SPINLOCKS. Jan pointed out that there is a 4 byte hole in struct journal_head where a regular spinlock fits in and he would not object to convert the state lock to a spinlock unconditionally. Aside of solving the RT problem, this also gains lockdep coverage for the journal head state lock (bit-spinlocks are not covered by lockdep as it's hard to fit a lockdep map into a single bit). The trivial change would have been to convert the jbd_*lock_bh_state() inlines, but that comes with the downside that these functions take a buffer head pointer which needs to be converted to a journal head pointer which adds another level of indirection. As almost all functions which use this lock have a journal head pointer readily available, it makes more sense to remove the lock helper inlines and write out spin_*lock() at all call sites. Fixup all locking comments as well. Suggested-by: Jan Kara <jack@suse.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Jan Kara <jack@suse.cz> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Mark Fasheh <mark@fasheh.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Jan Kara <jack@suse.com> Cc: linux-ext4@vger.kernel.org Link: https://lore.kernel.org/r/20190809124233.13277-7-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2019-08-09 12:42:32 +00:00
spin_unlock(&jh->b_state_lock);
spin_unlock(&journal->j_list_lock);
if (drop)
jbd2_journal_put_journal_head(jh);
}
/*
* File inode in the inode list of the handle's transaction
*/
static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
unsigned long flags, loff_t start_byte, loff_t end_byte)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal;
if (is_handle_aborted(handle))
return -EROFS;
journal = transaction->t_journal;
jbd2_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
transaction->t_tid);
spin_lock(&journal->j_list_lock);
jinode->i_flags |= flags;
if (jinode->i_dirty_end) {
jinode->i_dirty_start = min(jinode->i_dirty_start, start_byte);
jinode->i_dirty_end = max(jinode->i_dirty_end, end_byte);
} else {
jinode->i_dirty_start = start_byte;
jinode->i_dirty_end = end_byte;
}
/* Is inode already attached where we need it? */
if (jinode->i_transaction == transaction ||
jinode->i_next_transaction == transaction)
goto done;
/*
* We only ever set this variable to 1 so the test is safe. Since
* t_need_data_flush is likely to be set, we do the test to save some
* cacheline bouncing
*/
if (!transaction->t_need_data_flush)
transaction->t_need_data_flush = 1;
/* On some different transaction's list - should be
* the committing one */
if (jinode->i_transaction) {
J_ASSERT(jinode->i_next_transaction == NULL);
J_ASSERT(jinode->i_transaction ==
journal->j_committing_transaction);
jinode->i_next_transaction = transaction;
goto done;
}
/* Not on any transaction list... */
J_ASSERT(!jinode->i_next_transaction);
jinode->i_transaction = transaction;
list_add(&jinode->i_list, &transaction->t_inode_list);
done:
spin_unlock(&journal->j_list_lock);
return 0;
}
int jbd2_journal_inode_ranged_write(handle_t *handle,
struct jbd2_inode *jinode, loff_t start_byte, loff_t length)
{
return jbd2_journal_file_inode(handle, jinode,
JI_WRITE_DATA | JI_WAIT_DATA, start_byte,
start_byte + length - 1);
}
int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode,
loff_t start_byte, loff_t length)
{
return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA,
start_byte, start_byte + length - 1);
}
/*
* File truncate and transaction commit interact with each other in a
* non-trivial way. If a transaction writing data block A is
* committing, we cannot discard the data by truncate until we have
* written them. Otherwise if we crashed after the transaction with
* write has committed but before the transaction with truncate has
* committed, we could see stale data in block A. This function is a
* helper to solve this problem. It starts writeout of the truncated
* part in case it is in the committing transaction.
*
* Filesystem code must call this function when inode is journaled in
* ordered mode before truncation happens and after the inode has been
* placed on orphan list with the new inode size. The second condition
* avoids the race that someone writes new data and we start
* committing the transaction after this function has been called but
* before a transaction for truncate is started (and furthermore it
* allows us to optimize the case where the addition to orphan list
* happens in the same transaction as write --- we don't have to write
* any data in such case).
*/
int jbd2_journal_begin_ordered_truncate(journal_t *journal,
struct jbd2_inode *jinode,
loff_t new_size)
{
transaction_t *inode_trans, *commit_trans;
int ret = 0;
/* This is a quick check to avoid locking if not necessary */
if (!jinode->i_transaction)
goto out;
/* Locks are here just to force reading of recent values, it is
* enough that the transaction was not committing before we started
* a transaction adding the inode to orphan list */
read_lock(&journal->j_state_lock);
commit_trans = journal->j_committing_transaction;
read_unlock(&journal->j_state_lock);
spin_lock(&journal->j_list_lock);
inode_trans = jinode->i_transaction;
spin_unlock(&journal->j_list_lock);
if (inode_trans == commit_trans) {
ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
new_size, LLONG_MAX);
if (ret)
jbd2_journal_abort(journal, ret);
}
out:
return ret;
}