Re: [PATCH 12/20] target/tmr: LUN reset cause cmd premature free.

From: Quinn Tran <quinn.tran@qlogic.com>
To: Christoph Hellwig <hch@infradead.org>, Hannes Reinecke <hare@suse.de>
Cc: "target-devel@vger.kernel.org" <target-devel@vger.kernel.org>,
	"nab@linux-iscsi.org" <nab@linux-iscsi.org>,
	Giridhar Malavali <giridhar.malavali@qlogic.com>,
	linux-scsi <linux-scsi@vger.kernel.org>,
	Himanshu Madhani <himanshu.madhani@qlogic.com>
Subject: Re: [PATCH 12/20] target/tmr: LUN reset cause cmd premature free.
Date: Wed, 9 Dec 2015 20:11:06 +0000	[thread overview]
Message-ID: <D28DC7E3.1B099%quinn.tran@qlogic.com> (raw)
In-Reply-To: <20151208024859.GB32255@infradead.org>

[-- Attachment #1: Type: text/plain, Size: 1907 bytes --]

Christoph, 

Understood.  I was not sure which way the community is swinging.
For what it¹s worth, this fix was required to stabilize one of our
customer environment in the older kernel.

Regards,
Quinn Tran

On 12/7/15, 6:48 PM, "target-devel-owner@vger.kernel.org on behalf of
Christoph Hellwig" <target-devel-owner@vger.kernel.org on behalf of
hch@infradead.org> wrote:

>On Mon, Dec 07, 2015 at 07:48:59PM -0500, Himanshu Madhani wrote:
>> From: Quinn Tran <quinn.tran@qlogic.com>
>> 
>> During LUN/Target reset, the TMR code attempt to intercept
>> cmds and try to aborted them.  Current code assume cmds are
>> always intercepted at the back end device.  The cleanup code
>> would issue a "queue_status() & check_stop_free()" to terminate
>> the command.  However, when a cmd is intercepted at the front
>> end/Fabric layer, current code introduce premature free or
>> cause Fabric to double free.
>> 
>> When command is intercepted at Fabric layer, it means a
>> check_stop_free(cmd_kref--) has been called.  The extra
>> check_stop_free in the Lun Reset cleanup code causes early
>> free.  When a cmd in the Fabric layer is completed, the normal
>> free code adds another another free which introduce a double free.
>> 
>> To fix the issue:
>> - add a new flag/CMD_T_SENT_STATUS to track command that have
>>  made it down to fabric layer after back end good/bad completion.
>> - if cmd reach Fabric Layer at Lun Reset time, add an extra
>>  cmd_kref count to prevent premature free.
>
>This seems lke something solved by Bart's abort rewrite in a much nicer
>way.  All this special casing based on magic flags isn't maintainable
>in the long run.
>--
>To unsubscribe from this list: send the line "unsubscribe target-devel" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: winmail.dat --]
[-- Type: application/ms-tnef, Size: 4974 bytes --]