scsi: libiscsi: add lock around task lists to fix list corruption regression
authorChris Leech <cleech@redhat.com>
Tue, 28 Feb 2017 00:58:36 +0000 (16:58 -0800)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 1 Mar 2017 03:05:22 +0000 (22:05 -0500)
commit6f8830f5bbab16e54f261de187f3df4644a5b977
tree323e3313764b7dcc5ec5e5c12779e278e4a73742
parentd1a9ccc4b1374a5a7762031fd8e4e398c68549e6
scsi: libiscsi: add lock around task lists to fix list corruption regression

There's a rather long standing regression from the commit "libiscsi:
Reduce locking contention in fast path"

Depending on iSCSI target behavior, it's possible to hit the case in
iscsi_complete_task where the task is still on a pending list
(!list_empty(&task->running)).  When that happens the task is removed
from the list while holding the session back_lock, but other task list
modification occur under the frwd_lock.  That leads to linked list
corruption and eventually a panicked system.

Rather than back out the session lock split entirely, in order to try
and keep some of the performance gains this patch adds another lock to
maintain the task lists integrity.

Major enterprise supported kernels have been backing out the lock split
for while now, thanks to the efforts at IBM where a lab setup has the
most reliable reproducer I've seen on this issue.  This patch has been
tested there successfully.

Signed-off-by: Chris Leech <cleech@redhat.com>
Fixes: 659743b02c41 ("[SCSI] libiscsi: Reduce locking contention in fast path")
Reported-by: Prashantha Subbarao <psubbara@us.ibm.com>
Reviewed-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com>
Cc: <stable@vger.kernel.org> # v3.15+
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/libiscsi.c
include/scsi/libiscsi.h