[PATCH] block: mark GFP_NOIO around sysfs ->store()

[PATCH] block: mark GFP_NOIO around sysfs ->store()

[Ming Lei]

sysfs ->store is called with queue freezed, meantime we have several
->store() callbacks(update_nr_requests, wbt, scheduler) to allocate
memory with GFP_KERNEL which may run into direct reclaim code path,
then potential deadlock can be caused.

Fix the issue by marking NOIO around sysfs ->store()

Reported-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 block/blk-sysfs.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index e828be777206..e09b455874bf 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -681,6 +681,7 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
 	struct queue_sysfs_entry *entry = to_queue(attr);
 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
 	struct request_queue *q = disk->queue;
+	unsigned int noio_flag;
 	ssize_t res;
 
 	if (!entry->store_limit && !entry->store)
@@ -711,7 +712,9 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
 
 	mutex_lock(&q->sysfs_lock);
 	blk_mq_freeze_queue(q);
+	noio_flag = memalloc_noio_save();
 	res = entry->store(disk, page, length);
+	memalloc_noio_restore(noio_flag);
 	blk_mq_unfreeze_queue(q);
 	mutex_unlock(&q->sysfs_lock);
 	return res;
-- 
2.44.0

Re: [PATCH] block: mark GFP_NOIO around sysfs ->store()

[Christoph Hellwig]

On Mon, Jan 13, 2025 at 09:58:33AM +0800, Ming Lei wrote:
> sysfs ->store is called with queue freezed, meantime we have several
> ->store() callbacks(update_nr_requests, wbt, scheduler) to allocate
> memory with GFP_KERNEL which may run into direct reclaim code path,
> then potential deadlock can be caused.
> 
> Fix the issue by marking NOIO around sysfs ->store()

Yes, that's a good thing, and we should aim for more of that for
block layer code that requires NOIO:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH] block: mark GFP_NOIO around sysfs ->store()

[John Garry]

On 13/01/2025 01:58, Ming Lei wrote:
> sysfs ->store is called with queue freezed, meantime we have several
> ->store() callbacks(update_nr_requests, wbt, scheduler) to allocate
> memory with GFP_KERNEL which may run into direct reclaim code path,
> then potential deadlock can be caused.
> 
> Fix the issue by marking NOIO around sysfs ->store()
> 
> Reported-by: Thomas Hellström<thomas.hellstrom@linux.intel.com>
> Cc:stable@vger.kernel.org
> Signed-off-by: Ming Lei<ming.lei@redhat.com>

I guess that you should be including a link to 
https://lore.kernel.org/linux-block/Z4RkemI9f6N5zoEF@fedora/T/#mc774c65eeca5c024d29695f9ac6152b87763f305

Regardless, FWIW:
Reviewed-by: John Garry <john.g.garry@oracle.com>

Re: [PATCH] block: mark GFP_NOIO around sysfs ->store()

[Ming Lei]

On Mon, Jan 13, 2025 at 08:23:19AM +0000, John Garry wrote:
> On 13/01/2025 01:58, Ming Lei wrote:
> > sysfs ->store is called with queue freezed, meantime we have several
> > ->store() callbacks(update_nr_requests, wbt, scheduler) to allocate
> > memory with GFP_KERNEL which may run into direct reclaim code path,
> > then potential deadlock can be caused.
> > 
> > Fix the issue by marking NOIO around sysfs ->store()
> > 
> > Reported-by: Thomas Hellström<thomas.hellstrom@linux.intel.com>
> > Cc:stable@vger.kernel.org
> > Signed-off-by: Ming Lei<ming.lei@redhat.com>
> 
> I guess that you should be including a link to https://lore.kernel.org/linux-block/Z4RkemI9f6N5zoEF@fedora/T/#mc774c65eeca5c024d29695f9ac6152b87763f305
> 

Indeed, I will add a Closes tag in V2.


> Regardless, FWIW:
> Reviewed-by: John Garry <john.g.garry@oracle.com>

Thanks,


-- 
Ming

Re: [PATCH] block: mark GFP_NOIO around sysfs ->store()

[Jens Axboe]

On Mon, 13 Jan 2025 09:58:33 +0800, Ming Lei wrote:
> sysfs ->store is called with queue freezed, meantime we have several
> ->store() callbacks(update_nr_requests, wbt, scheduler) to allocate
> memory with GFP_KERNEL which may run into direct reclaim code path,
> then potential deadlock can be caused.
> 
> Fix the issue by marking NOIO around sysfs ->store()
> 
> [...]

Applied, thanks!

[1/1] block: mark GFP_NOIO around sysfs ->store()
      commit: 7c0be4ead1f8f5f8be0803f347de0de81e3b8e1c

Best regards,
-- 
Jens Axboe

Re: [PATCH] block: mark GFP_NOIO around sysfs ->store()

[xu.xin16]

> sysfs ->store is called with queue freezed, meantime we have several
> ->store() callbacks(update_nr_requests, wbt, scheduler) to allocate
> memory with GFP_KERNEL which may run into direct reclaim code path,
> then potential deadlock can be caused.
> 
> Fix the issue by marking NOIO around sysfs ->store()
> 
> Reported-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: stable@vger.kernel.org
> Signed-off-by: Ming Lei <ming.lei@redhat.com>
> ---
>  block/blk-sysfs.c | 3 +++
>  1 file changed, 3 insertions(+)

Excuse me, does the issue to fix comes from f1be1788a32e ("block: model freeze &
enter queue as lock for supporting lockdep") ?

Thanks.
> 
> diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
> index e828be777206..e09b455874bf 100644
> --- a/block/blk-sysfs.c
> +++ b/block/blk-sysfs.c
> @@ -681,6 +681,7 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
>  	struct queue_sysfs_entry *entry = to_queue(attr);
>  	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
>  	struct request_queue *q = disk->queue;
> +	unsigned int noio_flag;
>  	ssize_t res;
>  
>  	if (!entry->store_limit && !entry->store)
> @@ -711,7 +712,9 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
>  
>  	mutex_lock(&q->sysfs_lock);
>  	blk_mq_freeze_queue(q);
> +	noio_flag = memalloc_noio_save();
>  	res = entry->store(disk, page, length);
> +	memalloc_noio_restore(noio_flag);
>  	blk_mq_unfreeze_queue(q);
>  	mutex_unlock(&q->sysfs_lock);
>  	return res;
> -- 
> 2.44.0
>

Re: [PATCH] block: mark GFP_NOIO around sysfs ->store()

[Ming Lei]

On Wed, Aug 06, 2025 at 10:50:49AM +0800, xu.xin16@zte.com.cn wrote:
> > sysfs ->store is called with queue freezed, meantime we have several
> > ->store() callbacks(update_nr_requests, wbt, scheduler) to allocate
> > memory with GFP_KERNEL which may run into direct reclaim code path,
> > then potential deadlock can be caused.
> > 
> > Fix the issue by marking NOIO around sysfs ->store()
> > 
> > Reported-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Ming Lei <ming.lei@redhat.com>
> > ---
> >  block/blk-sysfs.c | 3 +++
> >  1 file changed, 3 insertions(+)
> 
> Excuse me, does the issue to fix comes from f1be1788a32e ("block: model freeze &
> enter queue as lock for supporting lockdep") ?

The above commit just starts to show the potential deadlock risk, which exists for
long time.

And now it becomes not necessary because blk_mq_freeze_queue() includes
memalloc_noio_save().

 
Thanks,
Ming