[Stgt-devel] [Patch] Segmentation fault in conn_close

[Tomas Henzl]

FUJITA Tomonori wrote:
> From: Tomas Henzl <thenzl at redhat.com>
> Subject: [Stgt-devel] [Patch] Segmentation fault in conn_close
> Date: Fri, 25 Jul 2008 14:56:22 +0200
>
>   
>> Hi,
>>
>> I noticed a segfault which is probably caused by the patch 
>> "iscsi: needs to call iscsi_free_cmd_task for commands in tx_clist"
>> b723430058dcbe3b201a2a2c38ce114217dc5273
>>
>> it looks like that for some reason iscsi_free_cmd_task(task); gets called  with
>> appropriate scmd zeroed out (not initialized ?) and then it causes a segfault
>> in list_del.
>>
>> I don't know how could it happen that this structure (scmd) is empty. The patch 
>> below solves my problem - it restores the old behaviour if the scmd is empty
>> as it was before the patch mentioned above.
>>     
>
> I think that this bug was fixed by:
>
>
> commit 73c6fab9f7f9e34aa14c359413b467be5e0ce0dc
> Author: FUJITA Tomonori <fujita.tomonori at lab.ntt.co.jp>
> Date:   Sat Jul 12 15:47:38 2008 +0900
>
>     iscsi: fix conn_close() segfaults
>
>     The problem is that conn_close() calls iscsi_free_cmd_task against
>     tasks in conn->tx_clist. But we could have non SCSI command tasks in
>     conn->tx_clist (such as NOOP). We can't call cmd_hlist_remove for such
>     tasks.
>
>     Signed-off-by: FUJITA Tomonori <fujita.tomonori at lab.ntt.co.jp>
>
>
> As the commit log says, non SCSI command tasks don't initialize scmd.
>   
Yes thanks, it looks correct now. My fault I should have downloaded the
latest
sources after vacation.