Comment 91 for bug 1569925

Commit has been merged upstream:

commit d754941225a7dbc61f6dd2173fa9498049f9a7ee
Author: Rafael David Tinoco <email address hidden>
Date: Thu Dec 7 19:59:13 2017 -0200

    scsi: libiscsi: Allow sd_shutdown on bad transport

    If, for any reason, userland shuts down iscsi transport interfaces
    before proper logouts - like when logging in to LUNs manually, without
    logging out on server shutdown, or when automated scripts can't
    umount/logout from logged LUNs - kernel will hang forever on its
    sd_sync_cache() logic, after issuing the SYNCHRONIZE_CACHE cmd to all
    still existent paths.

    PID: 1 TASK: ffff8801a69b8000 CPU: 1 COMMAND: "systemd-shutdow"
     #0 [ffff8801a69c3a30] __schedule at ffffffff8183e9ee
     #1 [ffff8801a69c3a80] schedule at ffffffff8183f0d5
     #2 [ffff8801a69c3a98] schedule_timeout at ffffffff81842199
     #3 [ffff8801a69c3b40] io_schedule_timeout at ffffffff8183e604
     #4 [ffff8801a69c3b70] wait_for_completion_io_timeout at ffffffff8183fc6c
     #5 [ffff8801a69c3bd0] blk_execute_rq at ffffffff813cfe10
     #6 [ffff8801a69c3c88] scsi_execute at ffffffff815c3fc7
     #7 [ffff8801a69c3cc8] scsi_execute_req_flags at ffffffff815c60fe
     #8 [ffff8801a69c3d30] sd_sync_cache at ffffffff815d37d7
     #9 [ffff8801a69c3da8] sd_shutdown at ffffffff815d3c3c

    This happens because iscsi_eh_cmd_timed_out(), the transport layer
    timeout helper, would tell the queue timeout function (scsi_times_out)
    to reset the request timer over and over, until the session state is
    back to logged in state. Unfortunately, during server shutdown, this
    might never happen again.

    Other option would be "not to handle" the issue in the transport
    layer. That would trigger the error handler logic, which would also need
    the session state to be logged in again.

    Best option, for such case, is to tell upper layers that the command was
    handled during the transport layer error handler helper, marking it as
    DID_NO_CONNECT, which will allow completion and inform about the
    problem.

    After the session was marked as ISCSI_STATE_FAILED, due to the first
    timeout during the server shutdown phase, all subsequent cmds will fail
    to be queued, allowing upper logic to fail faster.

    Signed-off-by: Rafael David Tinoco <email address hidden>
    Reviewed-by: Lee Duncan <email address hidden>
    Signed-off-by: Martin K. Petersen <email address hidden>

Applied in Bionic:

https://lists.ubuntu.com/archives/kernel-team/2018-January/089509.html

And Fix Committed.

Ack-ed for Trusty, Xenial, Artful:

https://lists.ubuntu.com/archives/kernel-team/2018-January/089774.html
https://lists.ubuntu.com/archives/kernel-team/2018-February/089809.html

Waiting to be Fix Committed there. A soon as it is released to -proposed I'll verify it and provide feedback so it can be Fix Released.