Patchwork [139/241] block: Don't access request after it might be freed

login
register
mail settings
Submitter Herton Ronaldo Krzesinski
Date Dec. 13, 2012, 1:58 p.m.
Message ID <1355407206-17100-140-git-send-email-herton.krzesinski@canonical.com>
Download mbox | patch
Permalink /patch/205989/
State New
Headers show

Comments

Herton Ronaldo Krzesinski - Dec. 13, 2012, 1:58 p.m.
3.5.7.2 -stable review patch.  If anyone has any objections, please let me know.

------------------

From: Roland Dreier <roland@purestorage.com>

commit 893d290f1d7496db97c9471bc352ad4a11dc8a25 upstream.

After we've done __elv_add_request() and __blk_run_queue() in
blk_execute_rq_nowait(), the request might finish and be freed
immediately.  Therefore checking if the type is REQ_TYPE_PM_RESUME
isn't safe afterwards, because if it isn't, rq might be gone.
Instead, check beforehand and stash the result in a temporary.

This fixes crashes in blk_execute_rq_nowait() I get occasionally when
running with lots of memory debugging options enabled -- I think this
race is usually harmless because the window for rq to be reallocated
is so small.

Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
[ herton: adjust context ]
Signed-off-by: Herton Ronaldo Krzesinski <herton.krzesinski@canonical.com>
---
 block/blk-exec.c |   10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
Ben Hutchings - Dec. 15, 2012, 7:40 p.m.
On Thu, 2012-12-13 at 11:58 -0200, Herton Ronaldo Krzesinski wrote:
> 3.5.7.2 -stable review patch.  If anyone has any objections, please let me know.
> 
> ------------------
> 
> From: Roland Dreier <roland@purestorage.com>
> 
> commit 893d290f1d7496db97c9471bc352ad4a11dc8a25 upstream.
> 
> After we've done __elv_add_request() and __blk_run_queue() in
> blk_execute_rq_nowait(), the request might finish and be freed
> immediately.  Therefore checking if the type is REQ_TYPE_PM_RESUME
> isn't safe afterwards, because if it isn't, rq might be gone.
> Instead, check beforehand and stash the result in a temporary.
> 
> This fixes crashes in blk_execute_rq_nowait() I get occasionally when
> running with lots of memory debugging options enabled -- I think this
> race is usually harmless because the window for rq to be reallocated
> is so small.
> 
> Signed-off-by: Roland Dreier <roland@purestorage.com>
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> [ herton: adjust context ]
> Signed-off-by: Herton Ronaldo Krzesinski <herton.krzesinski@canonical.com>

This is missing from 3.{0,4} but I did apply it to 3.2, again with the
need to adjust context.

Perhaps the intermediate fixes to blk_execute_rq_nowait() that resulted
in the changed context should also be applied to stable updates.  The
fixes in question are:

8ba61435d73f block: add missing blk_queue_dead() checks
- applied in 3.3; missing from 3.{0,2}.y
e81ca6fe85b7 [SCSI] block: Fix blk_execute_rq_nowait() dead queue handling
- applied in 3.6; missing from 3.{0,2,4,5}.y
893d290f1d74 block: Don't access request after it might be freed
- applied in 3.7, then to 3.{2,6}.y; missing from 3.{0,4}.y

Jens?

Ben.
> ---
>  block/blk-exec.c |   10 +++++++++-
>  1 file changed, 9 insertions(+), 1 deletion(-)
> 
> diff --git a/block/blk-exec.c b/block/blk-exec.c
> index fb2cbd5..9925fbe 100644
> --- a/block/blk-exec.c
> +++ b/block/blk-exec.c
> @@ -49,8 +49,16 @@ void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
>  			   rq_end_io_fn *done)
>  {
>  	int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
> +	bool is_pm_resume;
>  
>  	WARN_ON(irqs_disabled());
> +
> +	/*
> +	 * need to check this before __blk_run_queue(), because rq can
> +	 * be freed before that returns.
> +	 */
> +	is_pm_resume = rq->cmd_type == REQ_TYPE_PM_RESUME;
> +
>  	spin_lock_irq(q->queue_lock);
>  
>  	if (unlikely(blk_queue_dead(q))) {
> @@ -66,7 +74,7 @@ void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
>  	__elv_add_request(q, rq, where);
>  	__blk_run_queue(q);
>  	/* the queue is stopped so it won't be run */
> -	if (rq->cmd_type == REQ_TYPE_PM_RESUME)
> +	if (is_pm_resume)
>  		q->request_fn(q);
>  	spin_unlock_irq(q->queue_lock);
>  }

Patch

diff --git a/block/blk-exec.c b/block/blk-exec.c
index fb2cbd5..9925fbe 100644
--- a/block/blk-exec.c
+++ b/block/blk-exec.c
@@ -49,8 +49,16 @@  void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
 			   rq_end_io_fn *done)
 {
 	int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
+	bool is_pm_resume;
 
 	WARN_ON(irqs_disabled());
+
+	/*
+	 * need to check this before __blk_run_queue(), because rq can
+	 * be freed before that returns.
+	 */
+	is_pm_resume = rq->cmd_type == REQ_TYPE_PM_RESUME;
+
 	spin_lock_irq(q->queue_lock);
 
 	if (unlikely(blk_queue_dead(q))) {
@@ -66,7 +74,7 @@  void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
 	__elv_add_request(q, rq, where);
 	__blk_run_queue(q);
 	/* the queue is stopped so it won't be run */
-	if (rq->cmd_type == REQ_TYPE_PM_RESUME)
+	if (is_pm_resume)
 		q->request_fn(q);
 	spin_unlock_irq(q->queue_lock);
 }