[Stgt-devel] close completes before synchronize cache response

[Pete Wyckoff]

Commit ae2f80cbc432fe8cc4da94bdf289a7d856f11ac2:

    Date:   Sun Aug 12 08:18:45 2007 +0900

    close a connection after sending a logout response
    
    For now we need to just close a connection after sending a logout
    response.
    
causes some initiator complaints with 2.6.22 on logout:

sd 15:0:0:1: [sdb] Synchronizing SCSI cache
iscsi: cmd 0x35 is not queued (6)
iscsi: cmd 0x35 is not queued (6)
iscsi: cmd 0x35 is not queued (6)
sd 15:0:0:1: [sdb] Result: hostbyte=DID_NO_CONNECT driverbyte=DRIVER_OK,SUGGEST_OK

What happens on the target is:

tgtd: [12:00:25.295827] iscsi_scsi_cmd_rx_start(1387) 1 35 0 0 0 1 5b
tgtd: [12:00:25.295836] iscsi_task_queue(1332) 4752 4752 1
tgtd: [12:00:25.295842] target_cmd_queue(532) 0x630240 35 1
tgtd: [12:00:25.295848] target_cmd_queue(549) 0x630240 35 1 1
tgtd: [12:00:25.295853] bs_sync_cmd_submit(296) 6 0 0 0 0 0x630240
tgtd: [12:00:25.295861] target_cmd_queue(558) 5b 35 0 0 0 0 1
tgtd: [12:00:25.295877] bs_sync_worker_fn(140) got 0x630240
tgtd: [12:00:25.307434] iscsi_task_queue(1332) 4753 4753 46
tgtd: [12:00:25.307471] iscsi_task_tx_start(1680) found a task 0 0 0 0
tgtd: [12:00:25.307524] conn_close(97) connection closed
tgtd: [12:00:25.307532] iscsi_event_handler(2045) connection closed
tgtd: [12:00:25.608082] bs_sync_worker_fn(156) io done 0x630240 35 0 0
tgtd: [12:00:25.608136] bs_sync_ack_fn(94) found 0x630240
tgtd: [12:00:25.608167] bs_sync_handler(193) back to tgtd, 0x630240
tgtd: [12:00:25.608200] __cmd_done(618) 0 0 0 0
tgtd: [12:00:25.608210] conn_free(79) freeing connection
tgtd: [12:00:25.608217] it_nexus_destroy(124) 1 1

(Mostly unmodified git, just timestamps and little stuff.)

The SYNCHRONIZE_CACHE command (0x35) gets started in bs_sync by a
separate thread.  Then the immediate LOGOUT command (0x46) is
received and sets conn->state = STATE_CLOSE, as added in this patch.
After 300 ms, bs_sync completes, and returns its result, but iscsi
does not send it because the connection is closed.  Linux
complains.

If I remove bs_sync and instead use a non-threaded BS that processes
commands synchronously, close processing waits until after the cache
has been synchronized, and the initiator does not complain.

Things do work, but are a bit messy.  I don't have a good idea
for a workaround just now.  What was the problem that this patch
fixed?  Can we delay the close until all command responses have been
sent somehow?

		-- Pete