[PATCH RESEND] blk-mq: fix possible creation failure for 'debugfs_dir'

[PATCH RESEND] blk-mq: fix possible creation failure for 'debugfs_dir'

[Yu Kuai]

'q->debugfs_dir' is created in blk_register_queue(), however, it's not
removed in blk_unregister_queue() and is delayed to blk_release_queue().
Thus it's possible that del_gendisk() is done and blk_release_queue()
is not called yet, and in the mean time blk_register_queue() is called
for the new device with the same name. In this case, kernel will
compalin about creation failure for 'debugfs_dir' like following:

debugfs: Directory 'nullb1' with parent 'block' already present!

Fix the problem by moving forward the removal of 'debgfs_dir' to
blk_unregister_queue().

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
---
 block/blk-sysfs.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 88bd41d4cb59..15fc3f0786e3 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -779,11 +779,6 @@ static void blk_release_queue(struct kobject *kobj)
 	if (queue_is_mq(q))
 		blk_mq_release(q);
 
-	blk_trace_shutdown(q);
-	mutex_lock(&q->debugfs_mutex);
-	debugfs_remove_recursive(q->debugfs_dir);
-	mutex_unlock(&q->debugfs_mutex);
-
 	if (queue_is_mq(q))
 		blk_mq_debugfs_unregister(q);
 
@@ -951,5 +946,11 @@ void blk_unregister_queue(struct gendisk *disk)
 
 	mutex_unlock(&q->sysfs_dir_lock);
 
+	blk_trace_shutdown(q);
+	mutex_lock(&q->debugfs_mutex);
+	blk_mq_debugfs_unregister_hctxs(q);
+	debugfs_remove_recursive(q->debugfs_dir);
+	mutex_unlock(&q->debugfs_mutex);
+
 	kobject_put(&disk_to_dev(disk)->kobj);
 }
-- 
2.31.1

Re: [PATCH RESEND] blk-mq: fix possible creation failure for 'debugfs_dir'

[Jens Axboe]

On Fri, 15 Apr 2022 11:56:07 +0800, Yu Kuai wrote:
> 'q->debugfs_dir' is created in blk_register_queue(), however, it's not
> removed in blk_unregister_queue() and is delayed to blk_release_queue().
> Thus it's possible that del_gendisk() is done and blk_release_queue()
> is not called yet, and in the mean time blk_register_queue() is called
> for the new device with the same name. In this case, kernel will
> compalin about creation failure for 'debugfs_dir' like following:
> 
> [...]

Applied, thanks!

[1/1] blk-mq: fix possible creation failure for 'debugfs_dir'
      commit: a87c29e1a85e64b28445bb1e80505230bf2e3b4b

Best regards,
-- 
Jens Axboe

Re: [PATCH RESEND] blk-mq: fix possible creation failure for 'debugfs_dir'

[Williams, Dan J]

On Fri, 2022-04-15 at 11:56 +0800, Yu Kuai wrote:
> 'q->debugfs_dir' is created in blk_register_queue(), however, it's not
> removed in blk_unregister_queue() and is delayed to blk_release_queue().
> Thus it's possible that del_gendisk() is done and blk_release_queue()
> is not called yet, and in the mean time blk_register_queue() is called
> for the new device with the same name. In this case, kernel will
> compalin about creation failure for 'debugfs_dir' like following:
> 
> debugfs: Directory 'nullb1' with parent 'block' already present!
> 
> Fix the problem by moving forward the removal of 'debgfs_dir' to
> blk_unregister_queue().
> 
> Signed-off-by: Yu Kuai <yukuai3@huawei.com>

Looks like I'm late to the party, but for what its worth:

Tested-by: Dan Williams <dan.j.williams@intel.com>

Thanks!