From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <stable-owner@vger.kernel.org>
Received: from mail-pl1-f193.google.com ([209.85.214.193]:42835 "EHLO
        mail-pl1-f193.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
        with ESMTP id S1727846AbeKOBGG (ORCPT
        <rfc822;stable@vger.kernel.org>); Wed, 14 Nov 2018 20:06:06 -0500
Message-ID: <1542207748.192870.14.camel@acm.org>
Subject: Re: [PATCH V2] SCSI: fix queue cleanup race before queue
 initialization is done
From: Bart Van Assche <bvanassche@acm.org>
To: Ming Lei <ming.lei@redhat.com>, Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org, Andrew Jones <drjones@redhat.com>,
        Bart Van Assche <bart.vanassche@wdc.com>,
        linux-scsi@vger.kernel.org,
        "Martin K . Petersen" <martin.petersen@oracle.com>,
        Christoph Hellwig <hch@lst.de>,
        "James E . J . Bottomley" <jejb@linux.vnet.ibm.com>,
        stable <stable@vger.kernel.org>,
        "jianchao . wang" <jianchao.w.wang@oracle.com>
Date: Wed, 14 Nov 2018 07:02:28 -0800
In-Reply-To: <20181114082551.12141-1-ming.lei@redhat.com>
References: <20181114082551.12141-1-ming.lei@redhat.com>
Content-Type: text/plain; charset="UTF-7"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: stable-owner@vger.kernel.org
List-ID: <stable.vger.kernel.org>

On Wed, 2018-11-14 at 16:25 +-0800, Ming Lei wrote:
+AD4 --- a/drivers/scsi/scsi+AF8-lib.c
+AD4 +-+-+- b/drivers/scsi/scsi+AF8-lib.c
+AD4 +AEAAQA -697,6 +-697,12 +AEAAQA static bool scsi+AF8-end+AF8-request(struct request +ACo-req,
+AD4 blk+AF8-status+AF8-t error,
+AD4  		 +ACo-/
+AD4  		scsi+AF8-mq+AF8-uninit+AF8-cmd(cmd)+ADs
+AD4  
+AD4 +-		/+ACo
+AD4 +-		 +ACo queue is still alive, so grab the ref for preventing it
+AD4 +-		 +ACo from being cleaned up during running queue.
+AD4 +-		 +ACo-/
+AD4 +-		percpu+AF8-ref+AF8-get(+ACY-q-+AD4-q+AF8-usage+AF8-counter)+ADs
+AD4 +-

I think the above comment is misleading. In the block layer a queue is called
alive if the +ACI-dying+ACI flag has not been set. When the above call to
percpu+AF8-ref+AF8-get() occurs it is not guaranteed that that flag has not yet been
set. But it is guaranteed that q-+AD4-q+AF8-usage+AF8-counter is not zero. I would prefer
if the comment would be modified.

What's not clear to me is why this patch only protects the blk-mq path but
not the legacy path. Does the legacy path need similar protection? It also
triggers a queue run after having finished a request.

Otherwise this patch looks fine to me.

Bart.