[PATCH 5.15] io-wq: max_worker fixes

[PATCH 5.15] io-wq: max_worker fixes

[Pavel Begunkov]

First, fix nr_workers checks against max_workers, with max_worker
registration, it may pretty easily happen that nr_workers > max_workers.

Also, synchronise writing to acct->max_worker with wqe->lock. It's not
an actual problem, but as we don't care about io_wqe_create_worker(),
it's better than WRITE_ONCE()/READ_ONCE().

Fixes: 2e480058ddc2 ("io-wq: provide a way to limit max number of workers")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 fs/io-wq.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/fs/io-wq.c b/fs/io-wq.c
index 811299ac9684..cdf1719f6be6 100644
--- a/fs/io-wq.c
+++ b/fs/io-wq.c
@@ -253,7 +253,7 @@ static bool io_wqe_create_worker(struct io_wqe *wqe, struct io_wqe_acct *acct)
 		pr_warn_once("io-wq is not configured for unbound workers");
 
 	raw_spin_lock(&wqe->lock);
-	if (acct->nr_workers == acct->max_workers) {
+	if (acct->nr_workers >= acct->max_workers) {
 		raw_spin_unlock(&wqe->lock);
 		return true;
 	}
@@ -1290,15 +1290,18 @@ int io_wq_max_workers(struct io_wq *wq, int *new_count)
 
 	rcu_read_lock();
 	for_each_node(node) {
+		struct io_wqe *wqe = wq->wqes[node];
 		struct io_wqe_acct *acct;
 
+		raw_spin_lock(&wqe->lock);
 		for (i = 0; i < IO_WQ_ACCT_NR; i++) {
-			acct = &wq->wqes[node]->acct[i];
+			acct = &wqe->acct[i];
 			prev = max_t(int, acct->max_workers, prev);
 			if (new_count[i])
 				acct->max_workers = new_count[i];
 			new_count[i] = prev;
 		}
+		raw_spin_unlock(&wqe->lock);
 	}
 	rcu_read_unlock();
 	return 0;
-- 
2.33.0

Re: [PATCH 5.15] io-wq: max_worker fixes

[Jens Axboe]

On Tue, 19 Oct 2021 20:31:26 +0100, Pavel Begunkov wrote:
> First, fix nr_workers checks against max_workers, with max_worker
> registration, it may pretty easily happen that nr_workers > max_workers.
> 
> Also, synchronise writing to acct->max_worker with wqe->lock. It's not
> an actual problem, but as we don't care about io_wqe_create_worker(),
> it's better than WRITE_ONCE()/READ_ONCE().
> 
> [...]

Applied, thanks!

[1/1] io-wq: max_worker fixes
      commit: 9f7f7ea493a1d5c202760cd6145d6dde744f71a2

Best regards,
-- 
Jens Axboe

Re: [PATCH 5.15] io-wq: max_worker fixes

[Pavel Begunkov]

On 10/19/21 20:31, Pavel Begunkov wrote:
> First, fix nr_workers checks against max_workers, with max_worker
> registration, it may pretty easily happen that nr_workers > max_workers.
> 
> Also, synchronise writing to acct->max_worker with wqe->lock. It's not
> an actual problem, but as we don't care about io_wqe_create_worker(),
> it's better than WRITE_ONCE()/READ_ONCE().

Jens, can you add

Reported-by: Beld Zhang <beldzhang@gmail.com>

> 
> Fixes: 2e480058ddc2 ("io-wq: provide a way to limit max number of workers")
> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
> ---
>   fs/io-wq.c | 7 +++++--
>   1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/io-wq.c b/fs/io-wq.c
> index 811299ac9684..cdf1719f6be6 100644
> --- a/fs/io-wq.c
> +++ b/fs/io-wq.c
> @@ -253,7 +253,7 @@ static bool io_wqe_create_worker(struct io_wqe *wqe, struct io_wqe_acct *acct)
>   		pr_warn_once("io-wq is not configured for unbound workers");
>   
>   	raw_spin_lock(&wqe->lock);
> -	if (acct->nr_workers == acct->max_workers) {
> +	if (acct->nr_workers >= acct->max_workers) {
>   		raw_spin_unlock(&wqe->lock);
>   		return true;
>   	}
> @@ -1290,15 +1290,18 @@ int io_wq_max_workers(struct io_wq *wq, int *new_count)
>   
>   	rcu_read_lock();
>   	for_each_node(node) {
> +		struct io_wqe *wqe = wq->wqes[node];
>   		struct io_wqe_acct *acct;
>   
> +		raw_spin_lock(&wqe->lock);
>   		for (i = 0; i < IO_WQ_ACCT_NR; i++) {
> -			acct = &wq->wqes[node]->acct[i];
> +			acct = &wqe->acct[i];
>   			prev = max_t(int, acct->max_workers, prev);
>   			if (new_count[i])
>   				acct->max_workers = new_count[i];
>   			new_count[i] = prev;
>   		}
> +		raw_spin_unlock(&wqe->lock);
>   	}
>   	rcu_read_unlock();
>   	return 0;
> 

-- 
Pavel Begunkov

Re: [PATCH 5.15] io-wq: max_worker fixes

[Jens Axboe]

On 10/19/21 4:54 PM, Pavel Begunkov wrote:
> On 10/19/21 20:31, Pavel Begunkov wrote:
>> First, fix nr_workers checks against max_workers, with max_worker
>> registration, it may pretty easily happen that nr_workers > max_workers.
>>
>> Also, synchronise writing to acct->max_worker with wqe->lock. It's not
>> an actual problem, but as we don't care about io_wqe_create_worker(),
>> it's better than WRITE_ONCE()/READ_ONCE().
> 
> Jens, can you add
> 
> Reported-by: Beld Zhang <beldzhang@gmail.com>

Yep, done.

-- 
Jens Axboe