[PATCH] block: don't release queue's sysfs lock during switching elevator

[PATCH] block: don't release queue's sysfs lock during switching elevator

[Ming Lei]

cecf5d87ff20 ("block: split .sysfs_lock into two locks") starts to
release & acquire sysfs_lock before registering/un-registering elevator
queue during switching elevator for avoiding potential deadlock from
showing & storing 'queue/iosched' attributes and removing elevator's
kobject.

Turns out there isn't such deadlock because 'q->sysfs_lock' isn't
required in .show & .store of queue/iosched's attributes, and just
elevator's sysfs lock is acquired in elv_iosched_store() and
elv_iosched_show(). So it is safe to hold queue's sysfs lock when
registering/un-registering elevator queue.

The biggest issue is that commit cecf5d87ff20 assumes that concurrent
write on 'queue/scheduler' can't happen. However, this assumption isn't
true, because kernfs_fop_write() only guarantees that concurrent write
aren't called on the same open file, but the write could be from
different open on the file. So we can't release & re-acquire queue's
sysfs lock during switching elevator, otherwise use-after-free on
elevator could be triggered.

Fixes the issue by not releasing queue's sysfs lock during switching
elevator.

Fixes: cecf5d87ff20 ("block: split .sysfs_lock into two locks")
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: Mike Snitzer <snitzer@redhat.com>
Cc: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 block/blk-sysfs.c | 13 ++++---------
 block/elevator.c  | 31 +------------------------------
 2 files changed, 5 insertions(+), 39 deletions(-)

diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index b82736c781c5..962fc0c44381 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -989,13 +989,11 @@ int blk_register_queue(struct gendisk *disk)
 		blk_mq_debugfs_register(q);
 	}
 
-	/*
-	 * The flag of QUEUE_FLAG_REGISTERED isn't set yet, so elevator
-	 * switch won't happen at all.
-	 */
+	mutex_lock(&q->sysfs_lock);
 	if (q->elevator) {
 		ret = elv_register_queue(q, false);
 		if (ret) {
+			mutex_unlock(&q->sysfs_lock);
 			mutex_unlock(&q->sysfs_dir_lock);
 			kobject_del(&q->kobj);
 			blk_trace_remove_sysfs(dev);
@@ -1005,7 +1003,6 @@ int blk_register_queue(struct gendisk *disk)
 		has_elevator = true;
 	}
 
-	mutex_lock(&q->sysfs_lock);
 	blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
 	wbt_enable_default(q);
 	blk_throtl_register_queue(q);
@@ -1062,12 +1059,10 @@ void blk_unregister_queue(struct gendisk *disk)
 	kobject_del(&q->kobj);
 	blk_trace_remove_sysfs(disk_to_dev(disk));
 
-	/*
-	 * q->kobj has been removed, so it is safe to check if elevator
-	 * exists without holding q->sysfs_lock.
-	 */
+	mutex_lock(&q->sysfs_lock);
 	if (q->elevator)
 		elv_unregister_queue(q);
+	mutex_unlock(&q->sysfs_lock);
 	mutex_unlock(&q->sysfs_dir_lock);
 
 	kobject_put(&disk_to_dev(disk)->kobj);
diff --git a/block/elevator.c b/block/elevator.c
index bba10e83478a..5437059c9261 100644
--- a/block/elevator.c
+++ b/block/elevator.c
@@ -503,9 +503,7 @@ int elv_register_queue(struct request_queue *q, bool uevent)
 		if (uevent)
 			kobject_uevent(&e->kobj, KOBJ_ADD);
 
-		mutex_lock(&q->sysfs_lock);
 		e->registered = 1;
-		mutex_unlock(&q->sysfs_lock);
 	}
 	return error;
 }
@@ -523,11 +521,9 @@ void elv_unregister_queue(struct request_queue *q)
 		kobject_uevent(&e->kobj, KOBJ_REMOVE);
 		kobject_del(&e->kobj);
 
-		mutex_lock(&q->sysfs_lock);
 		e->registered = 0;
 		/* Re-enable throttling in case elevator disabled it */
 		wbt_enable_default(q);
-		mutex_unlock(&q->sysfs_lock);
 	}
 }
 
@@ -590,32 +586,11 @@ int elevator_switch_mq(struct request_queue *q,
 	lockdep_assert_held(&q->sysfs_lock);
 
 	if (q->elevator) {
-		if (q->elevator->registered) {
-			mutex_unlock(&q->sysfs_lock);
-
-			/*
-			 * Concurrent elevator switch can't happen becasue
-			 * sysfs write is always exclusively on same file.
-			 *
-			 * Also the elevator queue won't be freed after
-			 * sysfs_lock is released becasue kobject_del() in
-			 * blk_unregister_queue() waits for completion of
-			 * .store & .show on its attributes.
-			 */
+		if (q->elevator->registered)
 			elv_unregister_queue(q);
 
-			mutex_lock(&q->sysfs_lock);
-		}
 		ioc_clear_queue(q);
 		elevator_exit(q, q->elevator);
-
-		/*
-		 * sysfs_lock may be dropped, so re-check if queue is
-		 * unregistered. If yes, don't switch to new elevator
-		 * any more
-		 */
-		if (!blk_queue_registered(q))
-			return 0;
 	}
 
 	ret = blk_mq_init_sched(q, new_e);
@@ -623,11 +598,7 @@ int elevator_switch_mq(struct request_queue *q,
 		goto out;
 
 	if (new_e) {
-		mutex_unlock(&q->sysfs_lock);
-
 		ret = elv_register_queue(q, true);
-
-		mutex_lock(&q->sysfs_lock);
 		if (ret) {
 			elevator_exit(q, q->elevator);
 			goto out;
-- 
2.20.1

Re: [PATCH] block: don't release queue's sysfs lock during switching elevator

[Bart Van Assche]

On 9/23/19 8:12 AM, Ming Lei wrote:
> @@ -523,11 +521,9 @@ void elv_unregister_queue(struct request_queue *q)
>   		kobject_uevent(&e->kobj, KOBJ_REMOVE);
>   		kobject_del(&e->kobj);
>   
> -		mutex_lock(&q->sysfs_lock);
>   		e->registered = 0;
>   		/* Re-enable throttling in case elevator disabled it */
>   		wbt_enable_default(q);
> -		mutex_unlock(&q->sysfs_lock);
>   	}
>   }

Does this patch cause sysfs_lock to be held around 
kobject_del(&e->kobj)? Since sysfs_lock is locked from inside 
elv_attr_show() and elv_attr_store(), does this mean that this patch 
reintroduces the lock inversion problem that was fixed recently?

Thanks,

Bart.

Re: [PATCH] block: don't release queue's sysfs lock during switching elevator

[Ming Lei]

On Tue, Sep 24, 2019 at 11:37:09AM -0700, Bart Van Assche wrote:
> On 9/23/19 8:12 AM, Ming Lei wrote:
> > @@ -523,11 +521,9 @@ void elv_unregister_queue(struct request_queue *q)
> >   		kobject_uevent(&e->kobj, KOBJ_REMOVE);
> >   		kobject_del(&e->kobj);
> > -		mutex_lock(&q->sysfs_lock);
> >   		e->registered = 0;
> >   		/* Re-enable throttling in case elevator disabled it */
> >   		wbt_enable_default(q);
> > -		mutex_unlock(&q->sysfs_lock);
> >   	}
> >   }
> 
> Does this patch cause sysfs_lock to be held around kobject_del(&e->kobj)?

Yes.

> Since sysfs_lock is locked from inside elv_attr_show() and elv_attr_store(),

The request queue's sysfs_lock isn't required in elv_attr_show() and
elv_attr_store(), and only elevator's sysfs_lock is needed in the two
functions.

> does this mean that this patch reintroduces the lock inversion problem that
> was fixed recently?

No.

The lock inversion issue only existed on kobjects of q->kobj & q->mq_obj,
which was fixed already given the queue's sysfs_lock is required in
.show/.store callback of these two kobjects' attributes.


thanks,
Ming

Re: [PATCH] block: don't release queue's sysfs lock during switching elevator

[Bart Van Assche]

On 9/24/19 5:13 PM, Ming Lei wrote:
> On Tue, Sep 24, 2019 at 11:37:09AM -0700, Bart Van Assche wrote:
>> On 9/23/19 8:12 AM, Ming Lei wrote:
>>> @@ -523,11 +521,9 @@ void elv_unregister_queue(struct request_queue *q)
>>>    		kobject_uevent(&e->kobj, KOBJ_REMOVE);
>>>    		kobject_del(&e->kobj);
>>> -		mutex_lock(&q->sysfs_lock);
>>>    		e->registered = 0;
>>>    		/* Re-enable throttling in case elevator disabled it */
>>>    		wbt_enable_default(q);
>>> -		mutex_unlock(&q->sysfs_lock);
>>>    	}
>>>    }
>>
>> Does this patch cause sysfs_lock to be held around kobject_del(&e->kobj)?
> 
> Yes.
> 
>> Since sysfs_lock is locked from inside elv_attr_show() and elv_attr_store(),
> 
> The request queue's sysfs_lock isn't required in elv_attr_show() and
> elv_attr_store(), and only elevator's sysfs_lock is needed in the two
> functions.
> 
>> does this mean that this patch reintroduces the lock inversion problem that
>> was fixed recently?
> 
> No.
> 
> The lock inversion issue only existed on kobjects of q->kobj & q->mq_obj,
> which was fixed already given the queue's sysfs_lock is required in
> .show/.store callback of these two kobjects' attributes.

I had confused e->kobj and eq->kobj, hence my comments. After having 
taken another look at your patch:

Reviewed-by: Bart Van Assche <bvanassche@acm.org>

Re: [PATCH] block: don't release queue's sysfs lock during switching elevator

[Jens Axboe]

On 9/23/19 5:12 PM, Ming Lei wrote:
> cecf5d87ff20 ("block: split .sysfs_lock into two locks") starts to
> release & acquire sysfs_lock before registering/un-registering elevator
> queue during switching elevator for avoiding potential deadlock from
> showing & storing 'queue/iosched' attributes and removing elevator's
> kobject.
> 
> Turns out there isn't such deadlock because 'q->sysfs_lock' isn't
> required in .show & .store of queue/iosched's attributes, and just
> elevator's sysfs lock is acquired in elv_iosched_store() and
> elv_iosched_show(). So it is safe to hold queue's sysfs lock when
> registering/un-registering elevator queue.
> 
> The biggest issue is that commit cecf5d87ff20 assumes that concurrent
> write on 'queue/scheduler' can't happen. However, this assumption isn't
> true, because kernfs_fop_write() only guarantees that concurrent write
> aren't called on the same open file, but the write could be from
> different open on the file. So we can't release & re-acquire queue's
> sysfs lock during switching elevator, otherwise use-after-free on
> elevator could be triggered.
> 
> Fixes the issue by not releasing queue's sysfs lock during switching
> elevator.

Applied, thanks Ming.

-- 
Jens Axboe