New: Repeated unit attentions trigger initiator lockup

https://bugzilla.kernel.org/show_bug.cgi?id=97641

            Bug ID: 97641
           Summary: Repeated unit attentions trigger initiator lockup
           Product: IO/Storage
           Version: 2.5
    Kernel Version: 4.1-rc1
          Hardware: All
                OS: Linux
              Tree: Mainline
            Status: NEW
          Severity: normal
          Priority: P1
         Component: SCSI
          Assignee: [email protected]
          Reporter: [email protected]
        Regression: No

Created attachment 175651
--> https://bugzilla.kernel.org/attachment.cgi?id=175651&action=edit
initiator-deadlock.txt

Two virtual machines were configured such that one had the role of iSCSI target system and the other of iSCSI initiator system. After iSCSI login a fio job was started on the initiator system against LUN 0 of the target system. When I start
run the following test at the target side:

# cd /sys/kernel/scst_tgt/targets/iscsi/iqn.2012-06.be.bva\:tgt1/luns/
# while true; do echo del 1 > mgmt; echo add diskf0 1 > mgmt; done

that results in the following error messages at the initiator side:

iscsid: Can not allocate memory for receive context.
kernel: [ 231.226002] connection1:0: detected conn error (1011)

and exactly 30 seconds later the initiator system locks up reporting the following (see also the initiator-deadlock.txt attachment):

[ INFO: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected ] 4.1.0-rc1-debug+ #2 Not tainted
------------------------------------------------------
swapper/0/0 [HC0[0]:SC1[1]:HE0:SE0] is trying to acquire:
(&(&session->frwd_lock)->rlock){+.-...}, at: [<ffffffffc0185f42>]
iscsi_eh_cmd_timed_out+0x42/0x330 [libiscsi]

and this task is already holding:
(&(&q->__queue_lock)->rlock){-.-...}, at: [<ffffffff81328037>]
blk_rq_timed_out_timer+0x27/0x130
which would create a new lock dependency:
(&(&q->__queue_lock)->rlock){-.-...} ->
(&(&session->frwd_lock)->rlock){+.-...}

You are receiving this mail because:
You are the assignee for the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=97641

--- Comment #1 from Bart Van Assche <[email protected]> --- The fio job at the initiator side was started as follows:

mkfs.ext4 -O ^has_journal ${LUN0_DEV_NODE}
umount /mnt
mount ${LUN0_DEV_NODE} /mnt

fio --verify=md5 -rw=randwrite --size=10m --bs=4K --loops=10000000 \
    --iodepth=64 --group_reporting --sync=1 --direct=1 --ioengine=libaio \
    --directory=/mnt --name=test --thread --numjobs=1

You are receiving this mail because:
You are the assignee for the bug.