Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <d4efe597-94b0-5878-fb54-1c6deb915659@acm.org>
Date:   Sun, 17 Jul 2022 07:49:12 -0700
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.11.0
Subject: Re: [RFC] block: Add a NULL check in blk_mq_free_rqs()
Content-Language: en-US
To:     Gautam Menghani <gautammenghani201@gmail.com>, axboe@kernel.dk
Cc:     linux-block@vger.kernel.org, linux-kernel@vger.kernel.org,
        skhan@linuxfoundation.org
References: <20220717102242.219424-1-gautammenghani201@gmail.com>
From:   Bart Van Assche <bvanassche@acm.org>
In-Reply-To: <20220717102242.219424-1-gautammenghani201@gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 7/17/22 03:22, Gautam Menghani wrote:
> Syzbot reported a general protection fault in the function
> blk_mq_clear_rq_mapping() in the file block/blk-mq.c.
> The issue is that the variable drv_tags is NULL, and this
> originates from the struct blk_mq_tag_set. The dashboard link for this
> issue is :
> syzkaller.appspot.com/bug?id=c3ce4caa4fc58c156d4903984131cdfa38eee354
> 
> This patch fixes the above bug, but there is another syzbot bug which is
> related to this and getting triggered after the call to
> blk_mq_clear_rq_mapping(). As a result, I cannot determine if the issue
> is really solved. The link to other issue:
> syzkaller.appspot.com/bug?id=7643cea70f1d0ce15f5f4bc39488918837ad4233
> 
> Please provide feedback/suggestions on the same.
> 
> Signed-off-by: Gautam Menghani <gautammenghani201@gmail.com>
> ---
>   block/blk-mq.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 93d9d60980fb..c1dd1b78b95c 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -3092,7 +3092,8 @@ void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,
>   		}
>   	}
>   
> -	blk_mq_clear_rq_mapping(drv_tags, tags);
> +	if (drv_tags)
> +		blk_mq_clear_rq_mapping(drv_tags, tags);
>   
>   	while (!list_empty(&tags->page_list)) {
>   		page = list_first_entry(&tags->page_list, struct page, lru);

I don't see how drv_tags could be NULL without triggering a race 
condition. Please take a look at the nbd driver to see whether the root 
cause is perhaps in that driver instead of in the block layer core.

Thanks,

Bart.