linux/net/rds/send.c

1456 lines
38 KiB
C
Raw Normal View History

/*
* Copyright (c) 2006, 2018 Oracle and/or its affiliates. All rights reserved.
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
* General Public License (GPL) Version 2, available from the file
* COPYING in the main directory of this source tree, or the
* OpenIB.org BSD license below:
*
* Redistribution and use in source and binary forms, with or
* without modification, are permitted provided that the following
* conditions are met:
*
* - Redistributions of source code must retain the above
* copyright notice, this list of conditions and the following
* disclaimer.
*
* - Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials
* provided with the distribution.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
*/
#include <linux/kernel.h>
#include <linux/moduleparam.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/gfp.h>
#include <net/sock.h>
#include <linux/in.h>
#include <linux/list.h>
#include <linux/ratelimit.h>
#include <linux/export.h>
#include <linux/sizes.h>
#include "rds.h"
/* When transmitting messages in rds_send_xmit, we need to emerge from
* time to time and briefly release the CPU. Otherwise the softlock watchdog
* will kick our shin.
* Also, it seems fairer to not let one busy connection stall all the
* others.
*
* send_batch_count is the number of times we'll loop in send_xmit. Setting
* it to 0 will restore the old behavior (where we looped until we had
* drained the queue).
*/
static int send_batch_count = SZ_1K;
module_param(send_batch_count, int, 0444);
MODULE_PARM_DESC(send_batch_count, " batch factor when working the send queue");
static void rds_send_remove_from_sock(struct list_head *messages, int status);
/*
* Reset the send state. Callers must ensure that this doesn't race with
* rds_send_xmit().
*/
void rds_send_path_reset(struct rds_conn_path *cp)
{
struct rds_message *rm, *tmp;
unsigned long flags;
if (cp->cp_xmit_rm) {
rm = cp->cp_xmit_rm;
cp->cp_xmit_rm = NULL;
/* Tell the user the RDMA op is no longer mapped by the
* transport. This isn't entirely true (it's flushed out
* independently) but as the connection is down, there's
* no ongoing RDMA to/from that memory */
rds_message_unmapped(rm);
rds_message_put(rm);
}
cp->cp_xmit_sg = 0;
cp->cp_xmit_hdr_off = 0;
cp->cp_xmit_data_off = 0;
cp->cp_xmit_atomic_sent = 0;
cp->cp_xmit_rdma_sent = 0;
cp->cp_xmit_data_sent = 0;
cp->cp_conn->c_map_queued = 0;
cp->cp_unacked_packets = rds_sysctl_max_unacked_packets;
cp->cp_unacked_bytes = rds_sysctl_max_unacked_bytes;
/* Mark messages as retransmissions, and move them to the send q */
spin_lock_irqsave(&cp->cp_lock, flags);
list_for_each_entry_safe(rm, tmp, &cp->cp_retrans, m_conn_item) {
set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags);
set_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags);
}
list_splice_init(&cp->cp_retrans, &cp->cp_send_queue);
spin_unlock_irqrestore(&cp->cp_lock, flags);
}
EXPORT_SYMBOL_GPL(rds_send_path_reset);
static int acquire_in_xmit(struct rds_conn_path *cp)
{
return test_and_set_bit(RDS_IN_XMIT, &cp->cp_flags) == 0;
}
static void release_in_xmit(struct rds_conn_path *cp)
{
clear_bit(RDS_IN_XMIT, &cp->cp_flags);
smp_mb__after_atomic();
/*
* We don't use wait_on_bit()/wake_up_bit() because our waking is in a
* hot path and finding waiters is very rare. We don't want to walk
* the system-wide hashed waitqueue buckets in the fast path only to
* almost never find waiters.
*/
if (waitqueue_active(&cp->cp_waitq))
wake_up_all(&cp->cp_waitq);
}
/*
* We're making the conscious trade-off here to only send one message
* down the connection at a time.
* Pro:
* - tx queueing is a simple fifo list
* - reassembly is optional and easily done by transports per conn
* - no per flow rx lookup at all, straight to the socket
* - less per-frag memory and wire overhead
* Con:
* - queued acks can be delayed behind large messages
* Depends:
* - small message latency is higher behind queued large messages
* - large message latency isn't starved by intervening small sends
*/
int rds_send_xmit(struct rds_conn_path *cp)
{
struct rds_connection *conn = cp->cp_conn;
struct rds_message *rm;
unsigned long flags;
unsigned int tmp;
struct scatterlist *sg;
int ret = 0;
LIST_HEAD(to_be_dropped);
int batch_count;
unsigned long send_gen = 0;
restart:
batch_count = 0;
/*
* sendmsg calls here after having queued its message on the send
* queue. We only have one task feeding the connection at a time. If
* another thread is already feeding the queue then we back off. This
* avoids blocking the caller and trading per-connection data between
* caches per message.
*/
if (!acquire_in_xmit(cp)) {
rds_stats_inc(s_send_lock_contention);
ret = -ENOMEM;
goto out;
}
rds: tcp: use rds_destroy_pending() to synchronize netns/module teardown and rds connection/workq management An rds_connection can get added during netns deletion between lines 528 and 529 of 506 static void rds_tcp_kill_sock(struct net *net) : /* code to pull out all the rds_connections that should be destroyed */ : 528 spin_unlock_irq(&rds_tcp_conn_lock); 529 list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) 530 rds_conn_destroy(tc->t_cpath->cp_conn); Such an rds_connection would miss out the rds_conn_destroy() loop (that cancels all pending work) and (if it was scheduled after netns deletion) could trigger the use-after-free. A similar race-window exists for the module unload path in rds_tcp_exit -> rds_tcp_destroy_conns Concurrency with netns deletion (rds_tcp_kill_sock()) must be handled by checking check_net() before enqueuing new work or adding new connections. Concurrency with module-unload is handled by maintaining a module specific flag that is set at the start of the module exit function, and must be checked before enqueuing new work or adding new connections. This commit refactors existing RDS_DESTROY_PENDING checks added by commit 3db6e0d172c9 ("rds: use RCU to synchronize work-enqueue with connection teardown") and consolidates all the concurrency checks listed above into the function rds_destroy_pending(). Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-02-03 12:26:51 +00:00
if (rds_destroy_pending(cp->cp_conn)) {
release_in_xmit(cp);
ret = -ENETUNREACH; /* dont requeue send work */
goto out;
}
/*
* we record the send generation after doing the xmit acquire.
* if someone else manages to jump in and do some work, we'll use
* this to avoid a goto restart farther down.
*
* The acquire_in_xmit() check above ensures that only one
* caller can increment c_send_gen at any time.
*/
send_gen = READ_ONCE(cp->cp_send_gen) + 1;
WRITE_ONCE(cp->cp_send_gen, send_gen);
/*
* rds_conn_shutdown() sets the conn state and then tests RDS_IN_XMIT,
* we do the opposite to avoid races.
*/
if (!rds_conn_path_up(cp)) {
release_in_xmit(cp);
ret = 0;
goto out;
}
if (conn->c_trans->xmit_path_prepare)
conn->c_trans->xmit_path_prepare(cp);
/*
* spin trying to push headers and data down the connection until
* the connection doesn't make forward progress.
*/
while (1) {
rm = cp->cp_xmit_rm;
/*
* If between sending messages, we can send a pending congestion
* map update.
*/
if (!rm && test_and_clear_bit(0, &conn->c_map_queued)) {
rm = rds_cong_update_alloc(conn);
if (IS_ERR(rm)) {
ret = PTR_ERR(rm);
break;
}
rm->data.op_active = 1;
rm->m_inc.i_conn_path = cp;
rm->m_inc.i_conn = cp->cp_conn;
cp->cp_xmit_rm = rm;
}
/*
* If not already working on one, grab the next message.
*
* cp_xmit_rm holds a ref while we're sending this message down
* the connction. We can use this ref while holding the
* send_sem.. rds_send_reset() is serialized with it.
*/
if (!rm) {
unsigned int len;
batch_count++;
/* we want to process as big a batch as we can, but
* we also want to avoid softlockups. If we've been
* through a lot of messages, lets back off and see
* if anyone else jumps in
*/
if (batch_count >= send_batch_count)
goto over_batch;
spin_lock_irqsave(&cp->cp_lock, flags);
if (!list_empty(&cp->cp_send_queue)) {
rm = list_entry(cp->cp_send_queue.next,
struct rds_message,
m_conn_item);
rds_message_addref(rm);
/*
* Move the message from the send queue to the retransmit
* list right away.
*/
list_move_tail(&rm->m_conn_item,
&cp->cp_retrans);
}
spin_unlock_irqrestore(&cp->cp_lock, flags);
if (!rm)
break;
/* Unfortunately, the way Infiniband deals with
* RDMA to a bad MR key is by moving the entire
* queue pair to error state. We cold possibly
* recover from that, but right now we drop the
* connection.
* Therefore, we never retransmit messages with RDMA ops.
*/
2016-11-16 21:29:49 +00:00
if (test_bit(RDS_MSG_FLUSH, &rm->m_flags) ||
(rm->rdma.op_active &&
test_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags))) {
spin_lock_irqsave(&cp->cp_lock, flags);
if (test_and_clear_bit(RDS_MSG_ON_CONN, &rm->m_flags))
list_move(&rm->m_conn_item, &to_be_dropped);
spin_unlock_irqrestore(&cp->cp_lock, flags);
continue;
}
/* Require an ACK every once in a while */
len = ntohl(rm->m_inc.i_hdr.h_len);
if (cp->cp_unacked_packets == 0 ||
cp->cp_unacked_bytes < len) {
set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags);
cp->cp_unacked_packets =
rds_sysctl_max_unacked_packets;
cp->cp_unacked_bytes =
rds_sysctl_max_unacked_bytes;
rds_stats_inc(s_send_ack_required);
} else {
cp->cp_unacked_bytes -= len;
cp->cp_unacked_packets--;
}
cp->cp_xmit_rm = rm;
}
/* The transport either sends the whole rdma or none of it */
if (rm->rdma.op_active && !cp->cp_xmit_rdma_sent) {
rm->m_final_op = &rm->rdma;
/* The transport owns the mapped memory for now.
* You can't unmap it while it's on the send queue
*/
set_bit(RDS_MSG_MAPPED, &rm->m_flags);
ret = conn->c_trans->xmit_rdma(conn, &rm->rdma);
if (ret) {
clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
wake_up_interruptible(&rm->m_flush_wait);
break;
}
cp->cp_xmit_rdma_sent = 1;
}
if (rm->atomic.op_active && !cp->cp_xmit_atomic_sent) {
rm->m_final_op = &rm->atomic;
/* The transport owns the mapped memory for now.
* You can't unmap it while it's on the send queue
*/
set_bit(RDS_MSG_MAPPED, &rm->m_flags);
ret = conn->c_trans->xmit_atomic(conn, &rm->atomic);
if (ret) {
clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
wake_up_interruptible(&rm->m_flush_wait);
break;
}
cp->cp_xmit_atomic_sent = 1;
}
/*
* A number of cases require an RDS header to be sent
* even if there is no data.
* We permit 0-byte sends; rds-ping depends on this.
* However, if there are exclusively attached silent ops,
* we skip the hdr/data send, to enable silent operation.
*/
if (rm->data.op_nents == 0) {
int ops_present;
int all_ops_are_silent = 1;
ops_present = (rm->atomic.op_active || rm->rdma.op_active);
if (rm->atomic.op_active && !rm->atomic.op_silent)
all_ops_are_silent = 0;
if (rm->rdma.op_active && !rm->rdma.op_silent)
all_ops_are_silent = 0;
if (ops_present && all_ops_are_silent
&& !rm->m_rdma_cookie)
rm->data.op_active = 0;
}
if (rm->data.op_active && !cp->cp_xmit_data_sent) {
rm->m_final_op = &rm->data;
ret = conn->c_trans->xmit(conn, rm,
cp->cp_xmit_hdr_off,
cp->cp_xmit_sg,
cp->cp_xmit_data_off);
if (ret <= 0)
break;
if (cp->cp_xmit_hdr_off < sizeof(struct rds_header)) {
tmp = min_t(int, ret,
sizeof(struct rds_header) -
cp->cp_xmit_hdr_off);
cp->cp_xmit_hdr_off += tmp;
ret -= tmp;
}
sg = &rm->data.op_sg[cp->cp_xmit_sg];
while (ret) {
tmp = min_t(int, ret, sg->length -
cp->cp_xmit_data_off);
cp->cp_xmit_data_off += tmp;
ret -= tmp;
if (cp->cp_xmit_data_off == sg->length) {
cp->cp_xmit_data_off = 0;
sg++;
cp->cp_xmit_sg++;
BUG_ON(ret != 0 && cp->cp_xmit_sg ==
rm->data.op_nents);
}
}
if (cp->cp_xmit_hdr_off == sizeof(struct rds_header) &&
(cp->cp_xmit_sg == rm->data.op_nents))
cp->cp_xmit_data_sent = 1;
}
/*
* A rm will only take multiple times through this loop
* if there is a data op. Thus, if the data is sent (or there was
* none), then we're done with the rm.
*/
if (!rm->data.op_active || cp->cp_xmit_data_sent) {
cp->cp_xmit_rm = NULL;
cp->cp_xmit_sg = 0;
cp->cp_xmit_hdr_off = 0;
cp->cp_xmit_data_off = 0;
cp->cp_xmit_rdma_sent = 0;
cp->cp_xmit_atomic_sent = 0;
cp->cp_xmit_data_sent = 0;
rds_message_put(rm);
}
}
over_batch:
if (conn->c_trans->xmit_path_complete)
conn->c_trans->xmit_path_complete(cp);
release_in_xmit(cp);
/* Nuke any messages we decided not to retransmit. */
if (!list_empty(&to_be_dropped)) {
/* irqs on here, so we can put(), unlike above */
list_for_each_entry(rm, &to_be_dropped, m_conn_item)
rds_message_put(rm);
rds_send_remove_from_sock(&to_be_dropped, RDS_RDMA_DROPPED);
}
/*
* Other senders can queue a message after we last test the send queue
* but before we clear RDS_IN_XMIT. In that case they'd back off and
* not try and send their newly queued message. We need to check the
* send queue after having cleared RDS_IN_XMIT so that their message
* doesn't get stuck on the send queue.
*
* If the transport cannot continue (i.e ret != 0), then it must
* call us when more room is available, such as from the tx
* completion handler.
*
* We have an extra generation check here so that if someone manages
* to jump in after our release_in_xmit, we'll see that they have done
* some work and we will skip our goto
*/
if (ret == 0) {
bool raced;
smp_mb();
raced = send_gen != READ_ONCE(cp->cp_send_gen);
if ((test_bit(0, &conn->c_map_queued) ||
!list_empty(&cp->cp_send_queue)) && !raced) {
if (batch_count < send_batch_count)
goto restart;
rcu_read_lock();
rds: tcp: use rds_destroy_pending() to synchronize netns/module teardown and rds connection/workq management An rds_connection can get added during netns deletion between lines 528 and 529 of 506 static void rds_tcp_kill_sock(struct net *net) : /* code to pull out all the rds_connections that should be destroyed */ : 528 spin_unlock_irq(&rds_tcp_conn_lock); 529 list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) 530 rds_conn_destroy(tc->t_cpath->cp_conn); Such an rds_connection would miss out the rds_conn_destroy() loop (that cancels all pending work) and (if it was scheduled after netns deletion) could trigger the use-after-free. A similar race-window exists for the module unload path in rds_tcp_exit -> rds_tcp_destroy_conns Concurrency with netns deletion (rds_tcp_kill_sock()) must be handled by checking check_net() before enqueuing new work or adding new connections. Concurrency with module-unload is handled by maintaining a module specific flag that is set at the start of the module exit function, and must be checked before enqueuing new work or adding new connections. This commit refactors existing RDS_DESTROY_PENDING checks added by commit 3db6e0d172c9 ("rds: use RCU to synchronize work-enqueue with connection teardown") and consolidates all the concurrency checks listed above into the function rds_destroy_pending(). Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-02-03 12:26:51 +00:00
if (rds_destroy_pending(cp->cp_conn))
ret = -ENETUNREACH;
else
queue_delayed_work(rds_wq, &cp->cp_send_w, 1);
rcu_read_unlock();
} else if (raced) {
rds_stats_inc(s_send_lock_queue_raced);
}
}
out:
return ret;
}
EXPORT_SYMBOL_GPL(rds_send_xmit);
static void rds_send_sndbuf_remove(struct rds_sock *rs, struct rds_message *rm)
{
u32 len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
assert_spin_locked(&rs->rs_lock);
BUG_ON(rs->rs_snd_bytes < len);
rs->rs_snd_bytes -= len;
if (rs->rs_snd_bytes == 0)
rds_stats_inc(s_send_queue_empty);
}
static inline int rds_send_is_acked(struct rds_message *rm, u64 ack,
is_acked_func is_acked)
{
if (is_acked)
return is_acked(rm, ack);
return be64_to_cpu(rm->m_inc.i_hdr.h_sequence) <= ack;
}
/*
* This is pretty similar to what happens below in the ACK
* handling code - except that we call here as soon as we get
* the IB send completion on the RDMA op and the accompanying
* message.
*/
void rds_rdma_send_complete(struct rds_message *rm, int status)
{
struct rds_sock *rs = NULL;
struct rm_rdma_op *ro;
struct rds_notifier *notifier;
unsigned long flags;
unsigned int notify = 0;
spin_lock_irqsave(&rm->m_rs_lock, flags);
notify = rm->rdma.op_notify | rm->data.op_notify;
ro = &rm->rdma;
if (test_bit(RDS_MSG_ON_SOCK, &rm->m_flags) &&
ro->op_active && notify && ro->op_notifier) {
notifier = ro->op_notifier;
rs = rm->m_rs;
sock_hold(rds_rs_to_sk(rs));
notifier->n_status = status;
spin_lock(&rs->rs_lock);
list_add_tail(&notifier->n_list, &rs->rs_notify_queue);
spin_unlock(&rs->rs_lock);
ro->op_notifier = NULL;
}
spin_unlock_irqrestore(&rm->m_rs_lock, flags);
if (rs) {
rds_wake_sk_sleep(rs);
sock_put(rds_rs_to_sk(rs));
}
}
EXPORT_SYMBOL_GPL(rds_rdma_send_complete);
/*
* Just like above, except looks at atomic op
*/
void rds_atomic_send_complete(struct rds_message *rm, int status)
{
struct rds_sock *rs = NULL;
struct rm_atomic_op *ao;
struct rds_notifier *notifier;
unsigned long flags;
spin_lock_irqsave(&rm->m_rs_lock, flags);
ao = &rm->atomic;
if (test_bit(RDS_MSG_ON_SOCK, &rm->m_flags)
&& ao->op_active && ao->op_notify && ao->op_notifier) {
notifier = ao->op_notifier;
rs = rm->m_rs;
sock_hold(rds_rs_to_sk(rs));
notifier->n_status = status;
spin_lock(&rs->rs_lock);
list_add_tail(&notifier->n_list, &rs->rs_notify_queue);
spin_unlock(&rs->rs_lock);
ao->op_notifier = NULL;
}
spin_unlock_irqrestore(&rm->m_rs_lock, flags);
if (rs) {
rds_wake_sk_sleep(rs);
sock_put(rds_rs_to_sk(rs));
}
}
EXPORT_SYMBOL_GPL(rds_atomic_send_complete);
/*
* This is the same as rds_rdma_send_complete except we
* don't do any locking - we have all the ingredients (message,
* socket, socket lock) and can just move the notifier.
*/
static inline void
__rds_send_complete(struct rds_sock *rs, struct rds_message *rm, int status)
{
struct rm_rdma_op *ro;
struct rm_atomic_op *ao;
ro = &rm->rdma;
if (ro->op_active && ro->op_notify && ro->op_notifier) {
ro->op_notifier->n_status = status;
list_add_tail(&ro->op_notifier->n_list, &rs->rs_notify_queue);
ro->op_notifier = NULL;
}
ao = &rm->atomic;
if (ao->op_active && ao->op_notify && ao->op_notifier) {
ao->op_notifier->n_status = status;
list_add_tail(&ao->op_notifier->n_list, &rs->rs_notify_queue);
ao->op_notifier = NULL;
}
/* No need to wake the app - caller does this */
}
/*
* This removes messages from the socket's list if they're on it. The list
* argument must be private to the caller, we must be able to modify it
* without locks. The messages must have a reference held for their
* position on the list. This function will drop that reference after
* removing the messages from the 'messages' list regardless of if it found
* the messages on the socket list or not.
*/
static void rds_send_remove_from_sock(struct list_head *messages, int status)
{
unsigned long flags;
struct rds_sock *rs = NULL;
struct rds_message *rm;
while (!list_empty(messages)) {
int was_on_sock = 0;
rm = list_entry(messages->next, struct rds_message,
m_conn_item);
list_del_init(&rm->m_conn_item);
/*
* If we see this flag cleared then we're *sure* that someone
* else beat us to removing it from the sock. If we race
* with their flag update we'll get the lock and then really
* see that the flag has been cleared.
*
* The message spinlock makes sure nobody clears rm->m_rs
* while we're messing with it. It does not prevent the
* message from being removed from the socket, though.
*/
spin_lock_irqsave(&rm->m_rs_lock, flags);
if (!test_bit(RDS_MSG_ON_SOCK, &rm->m_flags))
goto unlock_and_drop;
if (rs != rm->m_rs) {
if (rs) {
rds_wake_sk_sleep(rs);
sock_put(rds_rs_to_sk(rs));
}
rs = rm->m_rs;
net/rds: fix possible double free on sock tear down I got a report of a double free happening at RDS slab cache. One suspicion was that may be somewhere we were doing a sock_hold/sock_put on an already freed sock. Thus after providing a kernel with the following change: static inline void sock_hold(struct sock *sk) { - atomic_inc(&sk->sk_refcnt); + if (!atomic_inc_not_zero(&sk->sk_refcnt)) + WARN(1, "Trying to hold sock already gone: %p (family: %hd)\n", + sk, sk->sk_family); } The warning successfuly triggered: Trying to hold sock already gone: ffff81f6dda61280 (family: 21) WARNING: at include/net/sock.h:350 sock_hold() Call Trace: <IRQ> [<ffffffff8adac135>] :rds:rds_send_remove_from_sock+0xf0/0x21b [<ffffffff8adad35c>] :rds:rds_send_drop_acked+0xbf/0xcf [<ffffffff8addf546>] :rds_rdma:rds_ib_recv_tasklet_fn+0x256/0x2dc [<ffffffff8009899a>] tasklet_action+0x8f/0x12b [<ffffffff800125a2>] __do_softirq+0x89/0x133 [<ffffffff8005f30c>] call_softirq+0x1c/0x28 [<ffffffff8006e644>] do_softirq+0x2c/0x7d [<ffffffff8006e4d4>] do_IRQ+0xee/0xf7 [<ffffffff8005e625>] ret_from_intr+0x0/0xa <EOI> Looking at the call chain above, the only way I think this would be possible is if somewhere we already released the same socket->sock which is assigned to the rds_message at rds_send_remove_from_sock. Which seems only possible to happen after the tear down done on rds_release. rds_release properly calls rds_send_drop_to to drop the socket from any rds_message, and some proper synchronization is in place to avoid race with rds_send_drop_acked/rds_send_remove_from_sock. However, I still see a very narrow window where it may be possible we touch a sock already released: when rds_release races with rds_send_drop_acked, we check RDS_MSG_ON_CONN to avoid cleanup on the same rds_message, but in this specific case we don't clear rm->m_rs. In this case, it seems we could then go on at rds_send_drop_to and after it returns, the sock is freed by last sock_put on rds_release, with concurrently we being at rds_send_remove_from_sock; then at some point in the loop at rds_send_remove_from_sock we process an rds_message which didn't have rm->m_rs unset for a freed sock, and a possible sock_hold on an sock already gone at rds_release happens. This hopefully address the described condition above and avoids a double free on "second last" sock_put. In addition, I removed the comment about socket destruction on top of rds_send_drop_acked: we call rds_send_drop_to in rds_release and we should have things properly serialized there, thus I can't see the comment being accurate there. Signed-off-by: Herton R. Krzesinski <herton@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-10-01 21:49:54 +00:00
if (rs)
sock_hold(rds_rs_to_sk(rs));
}
net/rds: fix possible double free on sock tear down I got a report of a double free happening at RDS slab cache. One suspicion was that may be somewhere we were doing a sock_hold/sock_put on an already freed sock. Thus after providing a kernel with the following change: static inline void sock_hold(struct sock *sk) { - atomic_inc(&sk->sk_refcnt); + if (!atomic_inc_not_zero(&sk->sk_refcnt)) + WARN(1, "Trying to hold sock already gone: %p (family: %hd)\n", + sk, sk->sk_family); } The warning successfuly triggered: Trying to hold sock already gone: ffff81f6dda61280 (family: 21) WARNING: at include/net/sock.h:350 sock_hold() Call Trace: <IRQ> [<ffffffff8adac135>] :rds:rds_send_remove_from_sock+0xf0/0x21b [<ffffffff8adad35c>] :rds:rds_send_drop_acked+0xbf/0xcf [<ffffffff8addf546>] :rds_rdma:rds_ib_recv_tasklet_fn+0x256/0x2dc [<ffffffff8009899a>] tasklet_action+0x8f/0x12b [<ffffffff800125a2>] __do_softirq+0x89/0x133 [<ffffffff8005f30c>] call_softirq+0x1c/0x28 [<ffffffff8006e644>] do_softirq+0x2c/0x7d [<ffffffff8006e4d4>] do_IRQ+0xee/0xf7 [<ffffffff8005e625>] ret_from_intr+0x0/0xa <EOI> Looking at the call chain above, the only way I think this would be possible is if somewhere we already released the same socket->sock which is assigned to the rds_message at rds_send_remove_from_sock. Which seems only possible to happen after the tear down done on rds_release. rds_release properly calls rds_send_drop_to to drop the socket from any rds_message, and some proper synchronization is in place to avoid race with rds_send_drop_acked/rds_send_remove_from_sock. However, I still see a very narrow window where it may be possible we touch a sock already released: when rds_release races with rds_send_drop_acked, we check RDS_MSG_ON_CONN to avoid cleanup on the same rds_message, but in this specific case we don't clear rm->m_rs. In this case, it seems we could then go on at rds_send_drop_to and after it returns, the sock is freed by last sock_put on rds_release, with concurrently we being at rds_send_remove_from_sock; then at some point in the loop at rds_send_remove_from_sock we process an rds_message which didn't have rm->m_rs unset for a freed sock, and a possible sock_hold on an sock already gone at rds_release happens. This hopefully address the described condition above and avoids a double free on "second last" sock_put. In addition, I removed the comment about socket destruction on top of rds_send_drop_acked: we call rds_send_drop_to in rds_release and we should have things properly serialized there, thus I can't see the comment being accurate there. Signed-off-by: Herton R. Krzesinski <herton@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-10-01 21:49:54 +00:00
if (!rs)
goto unlock_and_drop;
spin_lock(&rs->rs_lock);
if (test_and_clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags)) {
struct rm_rdma_op *ro = &rm->rdma;
struct rds_notifier *notifier;
list_del_init(&rm->m_sock_item);
rds_send_sndbuf_remove(rs, rm);
if (ro->op_active && ro->op_notifier &&
(ro->op_notify || (ro->op_recverr && status))) {
notifier = ro->op_notifier;
list_add_tail(&notifier->n_list,
&rs->rs_notify_queue);
if (!notifier->n_status)
notifier->n_status = status;
rm->rdma.op_notifier = NULL;
}
was_on_sock = 1;
}
spin_unlock(&rs->rs_lock);
unlock_and_drop:
spin_unlock_irqrestore(&rm->m_rs_lock, flags);
rds_message_put(rm);
if (was_on_sock)
rds_message_put(rm);
}
if (rs) {
rds_wake_sk_sleep(rs);
sock_put(rds_rs_to_sk(rs));
}
}
/*
* Transports call here when they've determined that the receiver queued
* messages up to, and including, the given sequence number. Messages are
* moved to the retrans queue when rds_send_xmit picks them off the send
* queue. This means that in the TCP case, the message may not have been
* assigned the m_ack_seq yet - but that's fine as long as tcp_is_acked
* checks the RDS_MSG_HAS_ACK_SEQ bit.
*/
void rds_send_path_drop_acked(struct rds_conn_path *cp, u64 ack,
is_acked_func is_acked)
{
struct rds_message *rm, *tmp;
unsigned long flags;
LIST_HEAD(list);
spin_lock_irqsave(&cp->cp_lock, flags);
list_for_each_entry_safe(rm, tmp, &cp->cp_retrans, m_conn_item) {
if (!rds_send_is_acked(rm, ack, is_acked))
break;
list_move(&rm->m_conn_item, &list);
clear_bit(RDS_MSG_ON_CONN, &rm->m_flags);
}
/* order flag updates with spin locks */
if (!list_empty(&list))
smp_mb__after_atomic();
spin_unlock_irqrestore(&cp->cp_lock, flags);
/* now remove the messages from the sock list as needed */
rds_send_remove_from_sock(&list, RDS_RDMA_SUCCESS);
}
EXPORT_SYMBOL_GPL(rds_send_path_drop_acked);
void rds_send_drop_acked(struct rds_connection *conn, u64 ack,
is_acked_func is_acked)
{
WARN_ON(conn->c_trans->t_mp_capable);
rds_send_path_drop_acked(&conn->c_path[0], ack, is_acked);
}
EXPORT_SYMBOL_GPL(rds_send_drop_acked);
void rds_send_drop_to(struct rds_sock *rs, struct sockaddr_in6 *dest)
{
struct rds_message *rm, *tmp;
struct rds_connection *conn;
struct rds_conn_path *cp;
unsigned long flags;
LIST_HEAD(list);
/* get all the messages we're dropping under the rs lock */
spin_lock_irqsave(&rs->rs_lock, flags);
list_for_each_entry_safe(rm, tmp, &rs->rs_send_queue, m_sock_item) {
if (dest &&
(!ipv6_addr_equal(&dest->sin6_addr, &rm->m_daddr) ||
dest->sin6_port != rm->m_inc.i_hdr.h_dport))
continue;
list_move(&rm->m_sock_item, &list);
rds_send_sndbuf_remove(rs, rm);
clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags);
}
/* order flag updates with the rs lock */
smp_mb__after_atomic();
spin_unlock_irqrestore(&rs->rs_lock, flags);
if (list_empty(&list))
return;
/* Remove the messages from the conn */
list_for_each_entry(rm, &list, m_sock_item) {
conn = rm->m_inc.i_conn;
if (conn->c_trans->t_mp_capable)
cp = rm->m_inc.i_conn_path;
else
cp = &conn->c_path[0];
spin_lock_irqsave(&cp->cp_lock, flags);
/*
* Maybe someone else beat us to removing rm from the conn.
* If we race with their flag update we'll get the lock and
* then really see that the flag has been cleared.
*/
if (!test_and_clear_bit(RDS_MSG_ON_CONN, &rm->m_flags)) {
spin_unlock_irqrestore(&cp->cp_lock, flags);
continue;
}
list_del_init(&rm->m_conn_item);
spin_unlock_irqrestore(&cp->cp_lock, flags);
/*
* Couldn't grab m_rs_lock in top loop (lock ordering),
* but we can now.
*/
spin_lock_irqsave(&rm->m_rs_lock, flags);
spin_lock(&rs->rs_lock);
__rds_send_complete(rs, rm, RDS_RDMA_CANCELED);
spin_unlock(&rs->rs_lock);
spin_unlock_irqrestore(&rm->m_rs_lock, flags);
rds_message_put(rm);
}
rds_wake_sk_sleep(rs);
while (!list_empty(&list)) {
rm = list_entry(list.next, struct rds_message, m_sock_item);
list_del_init(&rm->m_sock_item);
rds_message_wait(rm);
/* just in case the code above skipped this message
* because RDS_MSG_ON_CONN wasn't set, run it again here
* taking m_rs_lock is the only thing that keeps us
* from racing with ack processing.
*/
spin_lock_irqsave(&rm->m_rs_lock, flags);
spin_lock(&rs->rs_lock);
__rds_send_complete(rs, rm, RDS_RDMA_CANCELED);
spin_unlock(&rs->rs_lock);
spin_unlock_irqrestore(&rm->m_rs_lock, flags);
rds_message_put(rm);
}
}
/*
* we only want this to fire once so we use the callers 'queued'. It's
* possible that another thread can race with us and remove the
* message from the flow with RDS_CANCEL_SENT_TO.
*/
static int rds_send_queue_rm(struct rds_sock *rs, struct rds_connection *conn,
struct rds_conn_path *cp,
struct rds_message *rm, __be16 sport,
__be16 dport, int *queued)
{
unsigned long flags;
u32 len;
if (*queued)
goto out;
len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
/* this is the only place which holds both the socket's rs_lock
* and the connection's c_lock */
spin_lock_irqsave(&rs->rs_lock, flags);
/*
* If there is a little space in sndbuf, we don't queue anything,
* and userspace gets -EAGAIN. But poll() indicates there's send
* room. This can lead to bad behavior (spinning) if snd_bytes isn't
* freed up by incoming acks. So we check the *old* value of
* rs_snd_bytes here to allow the last msg to exceed the buffer,
* and poll() now knows no more data can be sent.
*/
if (rs->rs_snd_bytes < rds_sk_sndbuf(rs)) {
rs->rs_snd_bytes += len;
/* let recv side know we are close to send space exhaustion.
* This is probably not the optimal way to do it, as this
* means we set the flag on *all* messages as soon as our
* throughput hits a certain threshold.
*/
if (rs->rs_snd_bytes >= rds_sk_sndbuf(rs) / 2)
set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags);
list_add_tail(&rm->m_sock_item, &rs->rs_send_queue);
set_bit(RDS_MSG_ON_SOCK, &rm->m_flags);
rds_message_addref(rm);
sock_hold(rds_rs_to_sk(rs));
rm->m_rs = rs;
/* The code ordering is a little weird, but we're
trying to minimize the time we hold c_lock */
rds_message_populate_header(&rm->m_inc.i_hdr, sport, dport, 0);
rm->m_inc.i_conn = conn;
rm->m_inc.i_conn_path = cp;
rds_message_addref(rm);
spin_lock(&cp->cp_lock);
rm->m_inc.i_hdr.h_sequence = cpu_to_be64(cp->cp_next_tx_seq++);
list_add_tail(&rm->m_conn_item, &cp->cp_send_queue);
set_bit(RDS_MSG_ON_CONN, &rm->m_flags);
spin_unlock(&cp->cp_lock);
rdsdebug("queued msg %p len %d, rs %p bytes %d seq %llu\n",
rm, len, rs, rs->rs_snd_bytes,
(unsigned long long)be64_to_cpu(rm->m_inc.i_hdr.h_sequence));
*queued = 1;
}
spin_unlock_irqrestore(&rs->rs_lock, flags);
out:
return *queued;
}
/*
* rds_message is getting to be quite complicated, and we'd like to allocate
* it all in one go. This figures out how big it needs to be up front.
*/
static int rds_rm_size(struct msghdr *msg, int num_sgs)
{
struct cmsghdr *cmsg;
int size = 0;
int cmsg_groups = 0;
int retval;
bool zcopy_cookie = false;
for_each_cmsghdr(cmsg, msg) {
if (!CMSG_OK(msg, cmsg))
return -EINVAL;
if (cmsg->cmsg_level != SOL_RDS)
continue;
switch (cmsg->cmsg_type) {
case RDS_CMSG_RDMA_ARGS:
cmsg_groups |= 1;
retval = rds_rdma_extra_size(CMSG_DATA(cmsg));
if (retval < 0)
return retval;
size += retval;
break;
case RDS_CMSG_ZCOPY_COOKIE:
zcopy_cookie = true;
/* fall through */
case RDS_CMSG_RDMA_DEST:
case RDS_CMSG_RDMA_MAP:
cmsg_groups |= 2;
/* these are valid but do no add any size */
break;
case RDS_CMSG_ATOMIC_CSWP:
case RDS_CMSG_ATOMIC_FADD:
case RDS_CMSG_MASKED_ATOMIC_CSWP:
case RDS_CMSG_MASKED_ATOMIC_FADD:
cmsg_groups |= 1;
size += sizeof(struct scatterlist);
break;
default:
return -EINVAL;
}
}
if ((msg->msg_flags & MSG_ZEROCOPY) && !zcopy_cookie)
return -EINVAL;
size += num_sgs * sizeof(struct scatterlist);
/* Ensure (DEST, MAP) are never used with (ARGS, ATOMIC) */
if (cmsg_groups == 3)
return -EINVAL;
return size;
}
static int rds_cmsg_zcopy(struct rds_sock *rs, struct rds_message *rm,
struct cmsghdr *cmsg)
{
u32 *cookie;
if (cmsg->cmsg_len < CMSG_LEN(sizeof(*cookie)) ||
!rm->data.op_mmp_znotifier)
return -EINVAL;
cookie = CMSG_DATA(cmsg);
rm->data.op_mmp_znotifier->z_cookie = *cookie;
return 0;
}
static int rds_cmsg_send(struct rds_sock *rs, struct rds_message *rm,
struct msghdr *msg, int *allocated_mr)
{
struct cmsghdr *cmsg;
int ret = 0;
for_each_cmsghdr(cmsg, msg) {
if (!CMSG_OK(msg, cmsg))
return -EINVAL;
if (cmsg->cmsg_level != SOL_RDS)
continue;
/* As a side effect, RDMA_DEST and RDMA_MAP will set
* rm->rdma.m_rdma_cookie and rm->rdma.m_rdma_mr.
*/
switch (cmsg->cmsg_type) {
case RDS_CMSG_RDMA_ARGS:
ret = rds_cmsg_rdma_args(rs, rm, cmsg);
break;
case RDS_CMSG_RDMA_DEST:
ret = rds_cmsg_rdma_dest(rs, rm, cmsg);
break;
case RDS_CMSG_RDMA_MAP:
ret = rds_cmsg_rdma_map(rs, rm, cmsg);
if (!ret)
*allocated_mr = 1;
else if (ret == -ENODEV)
/* Accommodate the get_mr() case which can fail
* if connection isn't established yet.
*/
ret = -EAGAIN;
break;
case RDS_CMSG_ATOMIC_CSWP:
case RDS_CMSG_ATOMIC_FADD:
case RDS_CMSG_MASKED_ATOMIC_CSWP:
case RDS_CMSG_MASKED_ATOMIC_FADD:
ret = rds_cmsg_atomic(rs, rm, cmsg);
break;
case RDS_CMSG_ZCOPY_COOKIE:
ret = rds_cmsg_zcopy(rs, rm, cmsg);
break;
default:
return -EINVAL;
}
if (ret)
break;
}
return ret;
}
static int rds_send_mprds_hash(struct rds_sock *rs, struct rds_connection *conn)
{
int hash;
if (conn->c_npaths == 0)
hash = RDS_MPATH_HASH(rs, RDS_MPATH_WORKERS);
else
hash = RDS_MPATH_HASH(rs, conn->c_npaths);
if (conn->c_npaths == 0 && hash != 0) {
rds_send_ping(conn, 0);
/* The underlying connection is not up yet. Need to wait
* until it is up to be sure that the non-zero c_path can be
* used. But if we are interrupted, we have to use the zero
* c_path in case the connection ends up being non-MP capable.
*/
if (conn->c_npaths == 0)
if (wait_event_interruptible(conn->c_hs_waitq,
conn->c_npaths != 0))
hash = 0;
if (conn->c_npaths == 1)
hash = 0;
}
return hash;
}
static int rds_rdma_bytes(struct msghdr *msg, size_t *rdma_bytes)
{
struct rds_rdma_args *args;
struct cmsghdr *cmsg;
for_each_cmsghdr(cmsg, msg) {
if (!CMSG_OK(msg, cmsg))
return -EINVAL;
if (cmsg->cmsg_level != SOL_RDS)
continue;
if (cmsg->cmsg_type == RDS_CMSG_RDMA_ARGS) {
RDS: Check cmsg_len before dereferencing CMSG_DATA RDS currently doesn't check if the length of the control message is large enough to hold the required data, before dereferencing the control message data. This results in following crash: BUG: KASAN: stack-out-of-bounds in rds_rdma_bytes net/rds/send.c:1013 [inline] BUG: KASAN: stack-out-of-bounds in rds_sendmsg+0x1f02/0x1f90 net/rds/send.c:1066 Read of size 8 at addr ffff8801c928fb70 by task syzkaller455006/3157 CPU: 0 PID: 3157 Comm: syzkaller455006 Not tainted 4.15.0-rc3+ #161 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 print_address_description+0x73/0x250 mm/kasan/report.c:252 kasan_report_error mm/kasan/report.c:351 [inline] kasan_report+0x25b/0x340 mm/kasan/report.c:409 __asan_report_load8_noabort+0x14/0x20 mm/kasan/report.c:430 rds_rdma_bytes net/rds/send.c:1013 [inline] rds_sendmsg+0x1f02/0x1f90 net/rds/send.c:1066 sock_sendmsg_nosec net/socket.c:628 [inline] sock_sendmsg+0xca/0x110 net/socket.c:638 ___sys_sendmsg+0x320/0x8b0 net/socket.c:2018 __sys_sendmmsg+0x1ee/0x620 net/socket.c:2108 SYSC_sendmmsg net/socket.c:2139 [inline] SyS_sendmmsg+0x35/0x60 net/socket.c:2134 entry_SYSCALL_64_fastpath+0x1f/0x96 RIP: 0033:0x43fe49 RSP: 002b:00007fffbe244ad8 EFLAGS: 00000217 ORIG_RAX: 0000000000000133 RAX: ffffffffffffffda RBX: 00000000004002c8 RCX: 000000000043fe49 RDX: 0000000000000001 RSI: 000000002020c000 RDI: 0000000000000003 RBP: 00000000006ca018 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000217 R12: 00000000004017b0 R13: 0000000000401840 R14: 0000000000000000 R15: 0000000000000000 To fix this, we verify that the cmsg_len is large enough to hold the data to be read, before proceeding further. Reported-by: syzbot <syzkaller-bugs@googlegroups.com> Signed-off-by: Avinash Repaka <avinash.repaka@oracle.com> Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-12-22 04:17:04 +00:00
if (cmsg->cmsg_len <
CMSG_LEN(sizeof(struct rds_rdma_args)))
return -EINVAL;
args = CMSG_DATA(cmsg);
*rdma_bytes += args->remote_vec.bytes;
}
}
return 0;
}
int rds_sendmsg(struct socket *sock, struct msghdr *msg, size_t payload_len)
{
struct sock *sk = sock->sk;
struct rds_sock *rs = rds_sk_to_rs(sk);
DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
__be16 dport;
struct rds_message *rm = NULL;
struct rds_connection *conn;
int ret = 0;
int queued = 0, allocated_mr = 0;
int nonblock = msg->msg_flags & MSG_DONTWAIT;
long timeo = sock_sndtimeo(sk, nonblock);
struct rds_conn_path *cpath;
struct in6_addr daddr;
__u32 scope_id = 0;
size_t total_payload_len = payload_len, rdma_payload_len = 0;
bool zcopy = ((msg->msg_flags & MSG_ZEROCOPY) &&
sock_flag(rds_rs_to_sk(rs), SOCK_ZEROCOPY));
int num_sgs = ceil(payload_len, PAGE_SIZE);
int namelen;
/* Mirror Linux UDP mirror of BSD error message compatibility */
/* XXX: Perhaps MSG_MORE someday */
if (msg->msg_flags & ~(MSG_DONTWAIT | MSG_CMSG_COMPAT | MSG_ZEROCOPY)) {
ret = -EOPNOTSUPP;
goto out;
}
namelen = msg->msg_namelen;
if (namelen != 0) {
if (namelen < sizeof(*usin)) {
ret = -EINVAL;
goto out;
}
switch (usin->sin_family) {
case AF_INET:
if (usin->sin_addr.s_addr == htonl(INADDR_ANY) ||
usin->sin_addr.s_addr == htonl(INADDR_BROADCAST) ||
IN_MULTICAST(ntohl(usin->sin_addr.s_addr))) {
ret = -EINVAL;
goto out;
}
ipv6_addr_set_v4mapped(usin->sin_addr.s_addr, &daddr);
dport = usin->sin_port;
break;
#if IS_ENABLED(CONFIG_IPV6)
case AF_INET6: {
int addr_type;
if (namelen < sizeof(*sin6)) {
ret = -EINVAL;
goto out;
}
addr_type = ipv6_addr_type(&sin6->sin6_addr);
if (!(addr_type & IPV6_ADDR_UNICAST)) {
__be32 addr4;
if (!(addr_type & IPV6_ADDR_MAPPED)) {
ret = -EINVAL;
goto out;
}
/* It is a mapped address. Need to do some
* sanity checks.
*/
addr4 = sin6->sin6_addr.s6_addr32[3];
if (addr4 == htonl(INADDR_ANY) ||
addr4 == htonl(INADDR_BROADCAST) ||
IN_MULTICAST(ntohl(addr4))) {
ret = -EINVAL;
goto out;
}
}
if (addr_type & IPV6_ADDR_LINKLOCAL) {
if (sin6->sin6_scope_id == 0) {
ret = -EINVAL;
goto out;
}
scope_id = sin6->sin6_scope_id;
}
daddr = sin6->sin6_addr;
dport = sin6->sin6_port;
break;
}
#endif
default:
ret = -EINVAL;
goto out;
}
} else {
/* We only care about consistency with ->connect() */
lock_sock(sk);
daddr = rs->rs_conn_addr;
dport = rs->rs_conn_port;
scope_id = rs->rs_bound_scope_id;
release_sock(sk);
}
lock_sock(sk);
if (ipv6_addr_any(&rs->rs_bound_addr) || ipv6_addr_any(&daddr)) {
release_sock(sk);
ret = -ENOTCONN;
goto out;
} else if (namelen != 0) {
/* Cannot send to an IPv4 address using an IPv6 source
* address and cannot send to an IPv6 address using an
* IPv4 source address.
*/
if (ipv6_addr_v4mapped(&daddr) ^
ipv6_addr_v4mapped(&rs->rs_bound_addr)) {
release_sock(sk);
ret = -EOPNOTSUPP;
goto out;
}
/* If the socket is already bound to a link local address,
* it can only send to peers on the same link. But allow
* communicating beween link local and non-link local address.
*/
if (scope_id != rs->rs_bound_scope_id) {
if (!scope_id) {
scope_id = rs->rs_bound_scope_id;
} else if (rs->rs_bound_scope_id) {
release_sock(sk);
ret = -EINVAL;
goto out;
}
}
}
release_sock(sk);
ret = rds_rdma_bytes(msg, &rdma_payload_len);
if (ret)
goto out;
total_payload_len += rdma_payload_len;
if (max_t(size_t, payload_len, rdma_payload_len) > RDS_MAX_MSG_SIZE) {
ret = -EMSGSIZE;
goto out;
}
RDS: return EMSGSIZE for oversize requests before processing/queueing rds_send_queue_rm() allows for the "current datagram" being queued to exceed SO_SNDBUF thresholds by checking bytes queued without counting in length of current datagram. (Since sk_sndbuf is set to twice requested SO_SNDBUF value as a kernel heuristic this is usually fine!) If this "current datagram" squeezing past the threshold is itself many times the size of the sk_sndbuf threshold itself then even twice the SO_SNDBUF does not save us and it gets queued but cannot be transmitted. Threads block and deadlock and device becomes unusable. The check for this datagram not exceeding SNDBUF thresholds (EMSGSIZE) is not done on this datagram as that check is only done if queueing attempt fails. (Datagrams that follow this datagram fail queueing attempts, go through the check and eventually trip EMSGSIZE error but zero length datagrams silently fail!) This fix moves the check for datagrams exceeding SNDBUF limits before any processing or queueing is attempted and returns EMSGSIZE early in the rds_sndmsg() code. This change also ensures that all datagrams get checked for exceeding SNDBUF/sk_sndbuf size limits and the large datagrams that exceed those limits do not get to rds_send_queue_rm() code for processing. Signed-off-by: Mukesh Kacker <mukesh.kacker@oracle.com> Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org> Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-08-22 22:45:34 +00:00
if (payload_len > rds_sk_sndbuf(rs)) {
ret = -EMSGSIZE;
goto out;
}
if (zcopy) {
if (rs->rs_transport->t_type != RDS_TRANS_TCP) {
ret = -EOPNOTSUPP;
goto out;
}
num_sgs = iov_iter_npages(&msg->msg_iter, INT_MAX);
}
/* size of rm including all sgs */
ret = rds_rm_size(msg, num_sgs);
if (ret < 0)
goto out;
rm = rds_message_alloc(ret, GFP_KERNEL);
if (!rm) {
ret = -ENOMEM;
goto out;
}
/* Attach data to the rm */
if (payload_len) {
rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
if (!rm->data.op_sg) {
ret = -ENOMEM;
goto out;
}
ret = rds_message_copy_from_user(rm, &msg->msg_iter, zcopy);
if (ret)
goto out;
}
rm->data.op_active = 1;
rm->m_daddr = daddr;
/* rds_conn_create has a spinlock that runs with IRQ off.
* Caching the conn in the socket helps a lot. */
if (rs->rs_conn && ipv6_addr_equal(&rs->rs_conn->c_faddr, &daddr))
conn = rs->rs_conn;
else {
conn = rds_conn_create_outgoing(sock_net(sock->sk),
&rs->rs_bound_addr, &daddr,
rs->rs_transport,
sock->sk->sk_allocation,
scope_id);
if (IS_ERR(conn)) {
ret = PTR_ERR(conn);
goto out;
}
rs->rs_conn = conn;
}
RDS: RDMA: Fix the NULL-ptr deref in rds_ib_get_mr Registration of a memory region(MR) through FRMR/fastreg(unlike FMR) needs a connection/qp. With a proxy qp, this dependency on connection will be removed, but that needs more infrastructure patches, which is a work in progress. As an intermediate fix, the get_mr returns EOPNOTSUPP when connection details are not populated. The MR registration through sendmsg() will continue to work even with fast registration, since connection in this case is formed upfront. This patch fixes the following crash: kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN Modules linked in: CPU: 1 PID: 4244 Comm: syzkaller468044 Not tainted 4.16.0-rc6+ #361 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:rds_ib_get_mr+0x5c/0x230 net/rds/ib_rdma.c:544 RSP: 0018:ffff8801b059f890 EFLAGS: 00010202 RAX: dffffc0000000000 RBX: ffff8801b07e1300 RCX: ffffffff8562d96e RDX: 000000000000000d RSI: 0000000000000001 RDI: 0000000000000068 RBP: ffff8801b059f8b8 R08: ffffed0036274244 R09: ffff8801b13a1200 R10: 0000000000000004 R11: ffffed0036274243 R12: ffff8801b13a1200 R13: 0000000000000001 R14: ffff8801ca09fa9c R15: 0000000000000000 FS: 00007f4d050af700(0000) GS:ffff8801db300000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007f4d050aee78 CR3: 00000001b0d9b006 CR4: 00000000001606e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: __rds_rdma_map+0x710/0x1050 net/rds/rdma.c:271 rds_get_mr_for_dest+0x1d4/0x2c0 net/rds/rdma.c:357 rds_setsockopt+0x6cc/0x980 net/rds/af_rds.c:347 SYSC_setsockopt net/socket.c:1849 [inline] SyS_setsockopt+0x189/0x360 net/socket.c:1828 do_syscall_64+0x281/0x940 arch/x86/entry/common.c:287 entry_SYSCALL_64_after_hwframe+0x42/0xb7 RIP: 0033:0x4456d9 RSP: 002b:00007f4d050aedb8 EFLAGS: 00000246 ORIG_RAX: 0000000000000036 RAX: ffffffffffffffda RBX: 00000000006dac3c RCX: 00000000004456d9 RDX: 0000000000000007 RSI: 0000000000000114 RDI: 0000000000000004 RBP: 00000000006dac38 R08: 00000000000000a0 R09: 0000000000000000 R10: 0000000020000380 R11: 0000000000000246 R12: 0000000000000000 R13: 00007fffbfb36d6f R14: 00007f4d050af9c0 R15: 0000000000000005 Code: fa 48 c1 ea 03 80 3c 02 00 0f 85 cc 01 00 00 4c 8b bb 80 04 00 00 48 b8 00 00 00 00 00 fc ff df 49 8d 7f 68 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 9c 01 00 00 4d 8b 7f 68 48 b8 00 00 00 00 00 RIP: rds_ib_get_mr+0x5c/0x230 net/rds/ib_rdma.c:544 RSP: ffff8801b059f890 ---[ end trace 7e1cea13b85473b0 ]--- Reported-by: syzbot+b51c77ef956678a65834@syzkaller.appspotmail.com Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by: Avinash Repaka <avinash.repaka@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-25 03:31:58 +00:00
if (conn->c_trans->t_mp_capable)
cpath = &conn->c_path[rds_send_mprds_hash(rs, conn)];
else
cpath = &conn->c_path[0];
rm->m_conn_path = cpath;
/* Parse any control messages the user may have included. */
ret = rds_cmsg_send(rs, rm, msg, &allocated_mr);
if (ret) {
/* Trigger connection so that its ready for the next retry */
if (ret == -EAGAIN)
rds_conn_connect_if_down(conn);
goto out;
}
if (rm->rdma.op_active && !conn->c_trans->xmit_rdma) {
printk_ratelimited(KERN_NOTICE "rdma_op %p conn xmit_rdma %p\n",
&rm->rdma, conn->c_trans->xmit_rdma);
ret = -EOPNOTSUPP;
goto out;
}
if (rm->atomic.op_active && !conn->c_trans->xmit_atomic) {
printk_ratelimited(KERN_NOTICE "atomic_op %p conn xmit_atomic %p\n",
&rm->atomic, conn->c_trans->xmit_atomic);
ret = -EOPNOTSUPP;
goto out;
}
rds: tcp: use rds_destroy_pending() to synchronize netns/module teardown and rds connection/workq management An rds_connection can get added during netns deletion between lines 528 and 529 of 506 static void rds_tcp_kill_sock(struct net *net) : /* code to pull out all the rds_connections that should be destroyed */ : 528 spin_unlock_irq(&rds_tcp_conn_lock); 529 list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) 530 rds_conn_destroy(tc->t_cpath->cp_conn); Such an rds_connection would miss out the rds_conn_destroy() loop (that cancels all pending work) and (if it was scheduled after netns deletion) could trigger the use-after-free. A similar race-window exists for the module unload path in rds_tcp_exit -> rds_tcp_destroy_conns Concurrency with netns deletion (rds_tcp_kill_sock()) must be handled by checking check_net() before enqueuing new work or adding new connections. Concurrency with module-unload is handled by maintaining a module specific flag that is set at the start of the module exit function, and must be checked before enqueuing new work or adding new connections. This commit refactors existing RDS_DESTROY_PENDING checks added by commit 3db6e0d172c9 ("rds: use RCU to synchronize work-enqueue with connection teardown") and consolidates all the concurrency checks listed above into the function rds_destroy_pending(). Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-02-03 12:26:51 +00:00
if (rds_destroy_pending(conn)) {
ret = -EAGAIN;
goto out;
}
rds_conn_path_connect_if_down(cpath);
ret = rds_cong_wait(conn->c_fcong, dport, nonblock, rs);
if (ret) {
rs->rs_seen_congestion = 1;
goto out;
}
while (!rds_send_queue_rm(rs, conn, cpath, rm, rs->rs_bound_port,
dport, &queued)) {
rds_stats_inc(s_send_queue_full);
RDS: return EMSGSIZE for oversize requests before processing/queueing rds_send_queue_rm() allows for the "current datagram" being queued to exceed SO_SNDBUF thresholds by checking bytes queued without counting in length of current datagram. (Since sk_sndbuf is set to twice requested SO_SNDBUF value as a kernel heuristic this is usually fine!) If this "current datagram" squeezing past the threshold is itself many times the size of the sk_sndbuf threshold itself then even twice the SO_SNDBUF does not save us and it gets queued but cannot be transmitted. Threads block and deadlock and device becomes unusable. The check for this datagram not exceeding SNDBUF thresholds (EMSGSIZE) is not done on this datagram as that check is only done if queueing attempt fails. (Datagrams that follow this datagram fail queueing attempts, go through the check and eventually trip EMSGSIZE error but zero length datagrams silently fail!) This fix moves the check for datagrams exceeding SNDBUF limits before any processing or queueing is attempted and returns EMSGSIZE early in the rds_sndmsg() code. This change also ensures that all datagrams get checked for exceeding SNDBUF/sk_sndbuf size limits and the large datagrams that exceed those limits do not get to rds_send_queue_rm() code for processing. Signed-off-by: Mukesh Kacker <mukesh.kacker@oracle.com> Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org> Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-08-22 22:45:34 +00:00
if (nonblock) {
ret = -EAGAIN;
goto out;
}
timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
rds_send_queue_rm(rs, conn, cpath, rm,
rs->rs_bound_port,
dport,
&queued),
timeo);
rdsdebug("sendmsg woke queued %d timeo %ld\n", queued, timeo);
if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
continue;
ret = timeo;
if (ret == 0)
ret = -ETIMEDOUT;
goto out;
}
/*
* By now we've committed to the send. We reuse rds_send_worker()
* to retry sends in the rds thread if the transport asks us to.
*/
rds_stats_inc(s_send_queued);
ret = rds_send_xmit(cpath);
if (ret == -ENOMEM || ret == -EAGAIN) {
ret = 0;
rcu_read_lock();
rds: tcp: use rds_destroy_pending() to synchronize netns/module teardown and rds connection/workq management An rds_connection can get added during netns deletion between lines 528 and 529 of 506 static void rds_tcp_kill_sock(struct net *net) : /* code to pull out all the rds_connections that should be destroyed */ : 528 spin_unlock_irq(&rds_tcp_conn_lock); 529 list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) 530 rds_conn_destroy(tc->t_cpath->cp_conn); Such an rds_connection would miss out the rds_conn_destroy() loop (that cancels all pending work) and (if it was scheduled after netns deletion) could trigger the use-after-free. A similar race-window exists for the module unload path in rds_tcp_exit -> rds_tcp_destroy_conns Concurrency with netns deletion (rds_tcp_kill_sock()) must be handled by checking check_net() before enqueuing new work or adding new connections. Concurrency with module-unload is handled by maintaining a module specific flag that is set at the start of the module exit function, and must be checked before enqueuing new work or adding new connections. This commit refactors existing RDS_DESTROY_PENDING checks added by commit 3db6e0d172c9 ("rds: use RCU to synchronize work-enqueue with connection teardown") and consolidates all the concurrency checks listed above into the function rds_destroy_pending(). Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-02-03 12:26:51 +00:00
if (rds_destroy_pending(cpath->cp_conn))
ret = -ENETUNREACH;
else
queue_delayed_work(rds_wq, &cpath->cp_send_w, 1);
rcu_read_unlock();
}
if (ret)
goto out;
rds_message_put(rm);
return payload_len;
out:
/* If the user included a RDMA_MAP cmsg, we allocated a MR on the fly.
* If the sendmsg goes through, we keep the MR. If it fails with EAGAIN
* or in any other way, we need to destroy the MR again */
if (allocated_mr)
rds_rdma_unuse(rs, rds_rdma_cookie_key(rm->m_rdma_cookie), 1);
if (rm)
rds_message_put(rm);
return ret;
}
/*
* send out a probe. Can be shared by rds_send_ping,
* rds_send_pong, rds_send_hb.
* rds_send_hb should use h_flags
* RDS_FLAG_HB_PING|RDS_FLAG_ACK_REQUIRED
* or
* RDS_FLAG_HB_PONG|RDS_FLAG_ACK_REQUIRED
*/
static int
rds_send_probe(struct rds_conn_path *cp, __be16 sport,
__be16 dport, u8 h_flags)
{
struct rds_message *rm;
unsigned long flags;
int ret = 0;
rm = rds_message_alloc(0, GFP_ATOMIC);
if (!rm) {
ret = -ENOMEM;
goto out;
}
rm->m_daddr = cp->cp_conn->c_faddr;
rm->data.op_active = 1;
rds_conn_path_connect_if_down(cp);
ret = rds_cong_wait(cp->cp_conn->c_fcong, dport, 1, NULL);
if (ret)
goto out;
spin_lock_irqsave(&cp->cp_lock, flags);
list_add_tail(&rm->m_conn_item, &cp->cp_send_queue);
set_bit(RDS_MSG_ON_CONN, &rm->m_flags);
rds_message_addref(rm);
rm->m_inc.i_conn = cp->cp_conn;
rm->m_inc.i_conn_path = cp;
rds_message_populate_header(&rm->m_inc.i_hdr, sport, dport,
cp->cp_next_tx_seq);
rm->m_inc.i_hdr.h_flags |= h_flags;
cp->cp_next_tx_seq++;
if (RDS_HS_PROBE(be16_to_cpu(sport), be16_to_cpu(dport)) &&
cp->cp_conn->c_trans->t_mp_capable) {
u16 npaths = cpu_to_be16(RDS_MPATH_WORKERS);
u32 my_gen_num = cpu_to_be32(cp->cp_conn->c_my_gen_num);
rds_message_add_extension(&rm->m_inc.i_hdr,
RDS_EXTHDR_NPATHS, &npaths,
sizeof(npaths));
2016-11-16 21:29:49 +00:00
rds_message_add_extension(&rm->m_inc.i_hdr,
RDS_EXTHDR_GEN_NUM,
&my_gen_num,
2016-11-16 21:29:49 +00:00
sizeof(u32));
}
spin_unlock_irqrestore(&cp->cp_lock, flags);
rds_stats_inc(s_send_queued);
rds_stats_inc(s_send_pong);
RDS: fix rds-ping deadlock over TCP transport Sowmini found hang with rds-ping while testing RDS over TCP. Its a corner case and doesn't happen always. The issue is not reproducible with IB transport. Its clear from below dump why we see it with RDS TCP. [<ffffffff8153b7e5>] do_tcp_setsockopt+0xb5/0x740 [<ffffffff8153bec4>] tcp_setsockopt+0x24/0x30 [<ffffffff814d57d4>] sock_common_setsockopt+0x14/0x20 [<ffffffffa096071d>] rds_tcp_xmit_prepare+0x5d/0x70 [rds_tcp] [<ffffffffa093b5f7>] rds_send_xmit+0xd7/0x740 [rds] [<ffffffffa093bda2>] rds_send_pong+0x142/0x180 [rds] [<ffffffffa0939d34>] rds_recv_incoming+0x274/0x330 [rds] [<ffffffff810815ae>] ? ttwu_queue+0x11e/0x130 [<ffffffff814dcacd>] ? skb_copy_bits+0x6d/0x2c0 [<ffffffffa0960350>] rds_tcp_data_recv+0x2f0/0x3d0 [rds_tcp] [<ffffffff8153d836>] tcp_read_sock+0x96/0x1c0 [<ffffffffa0960060>] ? rds_tcp_recv_init+0x40/0x40 [rds_tcp] [<ffffffff814d6a90>] ? sock_def_write_space+0xa0/0xa0 [<ffffffffa09604d1>] rds_tcp_data_ready+0xa1/0xf0 [rds_tcp] [<ffffffff81545249>] tcp_data_queue+0x379/0x5b0 [<ffffffffa0960cdb>] ? rds_tcp_write_space+0xbb/0x110 [rds_tcp] [<ffffffff81547fd2>] tcp_rcv_established+0x2e2/0x6e0 [<ffffffff81552602>] tcp_v4_do_rcv+0x122/0x220 [<ffffffff81553627>] tcp_v4_rcv+0x867/0x880 [<ffffffff8152e0b3>] ip_local_deliver_finish+0xa3/0x220 This happens because rds_send_xmit() chain wants to take sock_lock which is already taken by tcp_v4_rcv() on its way to rds_tcp_data_ready(). Commit db6526dcb51b ("RDS: use rds_send_xmit() state instead of RDS_LL_SEND_FULL") which was trying to opportunistically finish the send request in same thread context. But because of above recursive lock hang with RDS TCP, the send work from rds_send_pong() needs to deferred to worker to avoid lock up. Given RDS ping is more of connectivity test than performance critical path, its should be ok even for transport like IB. Reported-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Acked-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org> Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Acked-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-10-17 02:13:21 +00:00
/* schedule the send work on rds_wq */
rcu_read_lock();
rds: tcp: use rds_destroy_pending() to synchronize netns/module teardown and rds connection/workq management An rds_connection can get added during netns deletion between lines 528 and 529 of 506 static void rds_tcp_kill_sock(struct net *net) : /* code to pull out all the rds_connections that should be destroyed */ : 528 spin_unlock_irq(&rds_tcp_conn_lock); 529 list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) 530 rds_conn_destroy(tc->t_cpath->cp_conn); Such an rds_connection would miss out the rds_conn_destroy() loop (that cancels all pending work) and (if it was scheduled after netns deletion) could trigger the use-after-free. A similar race-window exists for the module unload path in rds_tcp_exit -> rds_tcp_destroy_conns Concurrency with netns deletion (rds_tcp_kill_sock()) must be handled by checking check_net() before enqueuing new work or adding new connections. Concurrency with module-unload is handled by maintaining a module specific flag that is set at the start of the module exit function, and must be checked before enqueuing new work or adding new connections. This commit refactors existing RDS_DESTROY_PENDING checks added by commit 3db6e0d172c9 ("rds: use RCU to synchronize work-enqueue with connection teardown") and consolidates all the concurrency checks listed above into the function rds_destroy_pending(). Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-02-03 12:26:51 +00:00
if (!rds_destroy_pending(cp->cp_conn))
queue_delayed_work(rds_wq, &cp->cp_send_w, 1);
rcu_read_unlock();
rds_message_put(rm);
return 0;
out:
if (rm)
rds_message_put(rm);
return ret;
}
int
rds_send_pong(struct rds_conn_path *cp, __be16 dport)
{
return rds_send_probe(cp, 0, dport, 0);
}
void
rds_send_ping(struct rds_connection *conn, int cp_index)
{
unsigned long flags;
struct rds_conn_path *cp = &conn->c_path[cp_index];
spin_lock_irqsave(&cp->cp_lock, flags);
if (conn->c_ping_triggered) {
spin_unlock_irqrestore(&cp->cp_lock, flags);
return;
}
conn->c_ping_triggered = 1;
spin_unlock_irqrestore(&cp->cp_lock, flags);
rds_send_probe(cp, cpu_to_be16(RDS_FLAG_PROBE_PORT), 0, 0);
}
EXPORT_SYMBOL_GPL(rds_send_ping);