[PATCH] blk-mq: move cancel of requeue_work to the front of blk_exit_queue

[PATCH] blk-mq: move cancel of requeue_work to the front of blk_exit_queue

[zhengbin]

blk_exit_queue will free elevator_data, while blk_mq_requeue_work
will access it. Move cancel of requeue_work to the front of
blk_exit_queue to avoid use-after-free.

blk_exit_queue                blk_mq_requeue_work
  __elevator_exit               blk_mq_run_hw_queues
    blk_mq_exit_sched             blk_mq_run_hw_queue
      dd_exit_queue                 blk_mq_hctx_has_pending
        kfree(elevator_data)          blk_mq_sched_has_work
                                        dd_has_work

Fixes: fbc2a15e3433 ("blk-mq: move cancel of requeue_work into blk_mq_release")
Signed-off-by: zhengbin <zhengbin13@huawei.com>
---
 block/blk-mq.c    | 2 --
 block/blk-sysfs.c | 3 +++
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index f78d328..a8e6a58 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2666,8 +2666,6 @@ void blk_mq_release(struct request_queue *q)
 	struct blk_mq_hw_ctx *hctx, *next;
 	int i;

-	cancel_delayed_work_sync(&q->requeue_work);
-
 	queue_for_each_hw_ctx(q, hctx, i)
 		WARN_ON_ONCE(hctx && list_empty(&hctx->hctx_list));

diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 977c659..9bfa3ea 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -892,6 +892,9 @@ static void __blk_release_queue(struct work_struct *work)

 	blk_free_queue_stats(q->stats);

+	if (queue_is_mq(q))
+		cancel_delayed_work_sync(&q->requeue_work);
+
 	blk_exit_queue(q);

 	blk_queue_free_zone_bitmaps(q);
--
2.7.4

Re: [PATCH] blk-mq: move cancel of requeue_work to the front of blk_exit_queue

[Ming Lei]

On Mon, Aug 12, 2019 at 08:36:55PM +0800, zhengbin wrote:
> blk_exit_queue will free elevator_data, while blk_mq_requeue_work
> will access it. Move cancel of requeue_work to the front of
> blk_exit_queue to avoid use-after-free.
> 
> blk_exit_queue                blk_mq_requeue_work
>   __elevator_exit               blk_mq_run_hw_queues
>     blk_mq_exit_sched             blk_mq_run_hw_queue
>       dd_exit_queue                 blk_mq_hctx_has_pending
>         kfree(elevator_data)          blk_mq_sched_has_work
>                                         dd_has_work
> 
> Fixes: fbc2a15e3433 ("blk-mq: move cancel of requeue_work into blk_mq_release")
> Signed-off-by: zhengbin <zhengbin13@huawei.com>
> ---
>  block/blk-mq.c    | 2 --
>  block/blk-sysfs.c | 3 +++
>  2 files changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index f78d328..a8e6a58 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -2666,8 +2666,6 @@ void blk_mq_release(struct request_queue *q)
>  	struct blk_mq_hw_ctx *hctx, *next;
>  	int i;
> 
> -	cancel_delayed_work_sync(&q->requeue_work);
> -
>  	queue_for_each_hw_ctx(q, hctx, i)
>  		WARN_ON_ONCE(hctx && list_empty(&hctx->hctx_list));
> 
> diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
> index 977c659..9bfa3ea 100644
> --- a/block/blk-sysfs.c
> +++ b/block/blk-sysfs.c
> @@ -892,6 +892,9 @@ static void __blk_release_queue(struct work_struct *work)
> 
>  	blk_free_queue_stats(q->stats);
> 
> +	if (queue_is_mq(q))
> +		cancel_delayed_work_sync(&q->requeue_work);
> +
>  	blk_exit_queue(q);
> 
>  	blk_queue_free_zone_bitmaps(q);

Looks fine:

Reviewed-by: Ming Lei <ming.lei@redhat.com>

Thanks,
Ming

Re: [PATCH] blk-mq: move cancel of requeue_work to the front of blk_exit_queue

[Jens Axboe]

On 8/12/19 5:36 AM, zhengbin wrote:
> blk_exit_queue will free elevator_data, while blk_mq_requeue_work
> will access it. Move cancel of requeue_work to the front of
> blk_exit_queue to avoid use-after-free.
> 
> blk_exit_queue                blk_mq_requeue_work
>    __elevator_exit               blk_mq_run_hw_queues
>      blk_mq_exit_sched             blk_mq_run_hw_queue
>        dd_exit_queue                 blk_mq_hctx_has_pending
>          kfree(elevator_data)          blk_mq_sched_has_work
>                                          dd_has_work

Applied, thanks.

-- 
Jens Axboe