Re: [PATCH 6.0 517/862] sbitmap: Avoid leaving waitqueue in invalid state in __sbq_wake_up()

[Hugh Dickins]

On Wed, 19 Oct 2022, Greg Kroah-Hartman wrote:

> From: Jan Kara <jack@xxxxxxx>
> 
> [ Upstream commit 48c033314f372478548203c583529f53080fd078 ]
> 
> When __sbq_wake_up() decrements wait_cnt to 0 but races with someone
> else waking the waiter on the waitqueue (so the waitqueue becomes
> empty), it exits without reseting wait_cnt to wake_batch number. Once
> wait_cnt is 0, nobody will ever reset the wait_cnt or wake the new
> waiters resulting in possible deadlocks or busyloops. Fix the problem by
> making sure we reset wait_cnt even if we didn't wake up anybody in the
> end.
> 
> Fixes: 040b83fcecfb ("sbitmap: fix possible io hung due to lost wakeup")
> Reported-by: Keith Busch <kbusch@xxxxxxxxxx>
> Signed-off-by: Jan Kara <jack@xxxxxxx>
> Link: https://lore.kernel.org/r/20220908130937.2795-1-jack@xxxxxxx
> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
> Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>

I have no authority on linux-block, but I'll say NAK to this one
(and 479/862), and let Jens and Jan overrule me if they disagree.

This was another of several 6.1-rc1 commits which had given me lost
wakeups never suffered before; was not tagged Cc stable; and (unless I've
missed it on lore) never had AUTOSEL posted to linux-block or linux-kernel.

Hugh

> ---
>  lib/sbitmap.c | 18 +++++++++++++++---
>  1 file changed, 15 insertions(+), 3 deletions(-)
> 
> diff --git a/lib/sbitmap.c b/lib/sbitmap.c
> index 1f31147872e6..bb1970ad4875 100644
> --- a/lib/sbitmap.c
> +++ b/lib/sbitmap.c
> @@ -605,6 +605,7 @@ static bool __sbq_wake_up(struct sbitmap_queue *sbq)
>  	struct sbq_wait_state *ws;
>  	unsigned int wake_batch;
>  	int wait_cnt;
> +	bool ret;
>  
>  	ws = sbq_wake_ptr(sbq);
>  	if (!ws)
> @@ -615,12 +616,23 @@ static bool __sbq_wake_up(struct sbitmap_queue *sbq)
>  	 * For concurrent callers of this, callers should call this function
>  	 * again to wakeup a new batch on a different 'ws'.
>  	 */
> -	if (wait_cnt < 0 || !waitqueue_active(&ws->wait))
> +	if (wait_cnt < 0)
>  		return true;
>  
> +	/*
> +	 * If we decremented queue without waiters, retry to avoid lost
> +	 * wakeups.
> +	 */
>  	if (wait_cnt > 0)
> -		return false;
> +		return !waitqueue_active(&ws->wait);
>  
> +	/*
> +	 * When wait_cnt == 0, we have to be particularly careful as we are
> +	 * responsible to reset wait_cnt regardless whether we've actually
> +	 * woken up anybody. But in case we didn't wakeup anybody, we still
> +	 * need to retry.
> +	 */
> +	ret = !waitqueue_active(&ws->wait);
>  	wake_batch = READ_ONCE(sbq->wake_batch);
>  
>  	/*
> @@ -649,7 +661,7 @@ static bool __sbq_wake_up(struct sbitmap_queue *sbq)
>  	sbq_index_atomic_inc(&sbq->wake_index);
>  	atomic_set(&ws->wait_cnt, wake_batch);
>  
> -	return false;
> +	return ret;
>  }
>  
>  void sbitmap_queue_wake_up(struct sbitmap_queue *sbq)
> -- 
> 2.35.1
> 
> 
> 
>