Re: [PATCH] scsi_prep_fn() check for empty queue

[Maxim Uvarov <maxim.uvarov@oracle.com>]

On 06/26/2013 01:20 PM, Bart Van Assche wrote:
> On 06/26/13 11:02, Maxim Uvarov wrote:
>> This fix:
>> end_request: I/O error, dev sdc, sector 976576
>>   rport-0:0-3: blocked FC remote port time out: removing target and 
>> saving
>> binding
>> BUG: unable to handle kernel NULL pointer dereference at 
>> 0000000000000400
>> IP: [<ffffffff812f0cc2>] scsi_prep_state_check+0xe/0x99
>>   [<ffffffff812f1f9d>] scsi_setup_blk_pc_cmnd+0x1b/0x115
>>   [<ffffffff812f20c0>] scsi_prep_fn+0x29/0x3b
>>   [<ffffffff8121cfb9>] blk_peek_request+0xe1/0x1b3
>>   [<ffffffff812f1400>] scsi_request_fn+0x3a/0x4d2
>>   [<ffffffff8121d916>] __generic_unplug_device+0x32/0x36
>>   [<ffffffff81220f4b>] blk_execute_rq_nowait+0x77/0x9e
>>   [<ffffffff81221018>] blk_execute_rq+0xa6/0xde
>>   [<ffffffff8144f24b>] ? printk+0x41/0x46
>>   [<ffffffffa00a21c5>] ? get_rdac_req+0x81/0xe8 [scsi_dh_rdac]
>>   [<ffffffffa00a273a>] send_mode_select+0x29f/0x489 [scsi_dh_rdac]
>>   [<ffffffff810c5d9b>] ? probe_workqueue_execution+0xb1/0xce
>>   [<ffffffff81071e38>] worker_thread+0x1a9/0x237
>>   [<ffffffffa00a249b>] ? send_mode_select+0x0/0x489 [scsi_dh_rdac]
>>   [<ffffffff8107651b>] ? autoremove_wake_function+0x0/0x39
>>   [<ffffffff81071c8f>] ? worker_thread+0x0/0x237
>>   [<ffffffff81076222>] kthread+0x7f/0x87
>>   [<ffffffff81012d2a>] child_rip+0xa/0x20
>>   [<ffffffff810761a3>] ? kthread+0x0/0x87
>>   [<ffffffff81012d20>] ? child_rip+0x0/0x20
>> Signed-off-by: Maxim Uvarov <maxim.uvarov@oracle.com>
>> ---
>>   drivers/scsi/scsi_lib.c |    3 +++
>>   1 file changed, 3 insertions(+)
>>
>> diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
>> index 86d5220..8e89ed9 100644
>> --- a/drivers/scsi/scsi_lib.c
>> +++ b/drivers/scsi/scsi_lib.c
>> @@ -1295,6 +1295,9 @@ int scsi_prep_fn(struct request_queue *q, 
>> struct request *req)
>>       struct scsi_device *sdev = q->queuedata;
>>       int ret = BLKPREP_KILL;
>>
>> +    if (!sdev)
>> +        return ret;
>> +
>>       if (req->cmd_type == REQ_TYPE_BLOCK_PC)
>>           ret = scsi_setup_blk_pc_cmnd(sdev, req);
>>       return scsi_prep_return(q, req, ret);
>
> Sorry but this patch does not look like a proper fix to me. What you 
> probably need is a scsi_device_get() call in scsi_dh_rdac.c somewhere 
> before the queue_work(kmpath_rdacd, &ctlr->ms_work) call and a 
> scsi_device_put() call once send_mode_select() has finished using the 
> sdev.
>
> Bart.
Thanks Bart. It looks reasonable. I will do some testing for your solution.

Maxim.