When a target sends Check Condition, whilst initiator is busy xmiting
re-queued data, could lead to race between iscsi_complete_task() and
iscsi_xmit_task() and eventually crashing with the following kernel
backtrace.
Commit 6f8830f5bbab ("scsi: libiscsi: add lock around task lists to fix
list corruption regression") introduced "taskqueuelock" to fix list
corruption during the race, but this wasn't enough.
Re-setting of conn->task to NULL, could race with iscsi_xmit_task().
iscsi_complete_task()
{
....
if (conn->task == task)
conn->task = NULL;
}
conn->task in iscsi_xmit_task() could be NULL and so will be task.
__iscsi_get_task(task) will crash (NullPtr de-ref), trying to access
refcount.
This commit will take extra conn->session->back_lock in iscsi_xmit_task()
to ensure iscsi_xmit_task() waits for iscsi_complete_task(), if
iscsi_complete_task() wins the race. If iscsi_xmit_task() wins the race,
iscsi_xmit_task() increments task->refcount
(__iscsi_get_task) ensuring iscsi_complete_task() will not iscsi_free_task().
Signed-off-by: Anoob Soman <anoob.soman@citrix.com> Signed-off-by: Bob Liu <bob.liu@oracle.com> Acked-by: Lee Duncan <lduncan@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>