nbd: fix possible sysfs duplicate warning

Commit Message

Xiubo Li Sept. 19, 2019, 6:14 a.m. UTC

From: Xiubo Li <[e-mail]>

1. nbd_put takes the mutex and drops nbd->ref to 0. It then does
idr_remove and drops the mutex.

2. nbd_genl_connect takes the mutex. idr_find/idr_for_each fails
to find an existing device, so it does nbd_dev_add.

3. just before the nbd_put could call nbd_dev_remove or not finished
totally, but if nbd_dev_add try to add_disk, we can hit:

debugfs: Directory 'nbd1' with parent 'block' already present!

This patch will make sure all the disk add/remove stuff are done
by holding the nbd_index_mutex lock.

Signed-off-by: Xiubo Li <[e-mail]>
Reported-by: Mike Christie <[e-mail]>
---
 drivers/block/nbd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Josef Bacik Oct. 10, 2019, 1:56 p.m. UTC | #1

On Thu, Sep 19, 2019 at 11:44:27AM +0530, [e-mail] wrote:
> From: Xiubo Li <[e-mail]>
> 
> 1. nbd_put takes the mutex and drops nbd->ref to 0. It then does
> idr_remove and drops the mutex.
> 
> 2. nbd_genl_connect takes the mutex. idr_find/idr_for_each fails
> to find an existing device, so it does nbd_dev_add.
> 
> 3. just before the nbd_put could call nbd_dev_remove or not finished
> totally, but if nbd_dev_add try to add_disk, we can hit:
> 
> debugfs: Directory 'nbd1' with parent 'block' already present!
> 
> This patch will make sure all the disk add/remove stuff are done
> by holding the nbd_index_mutex lock.
> 
> Signed-off-by: Xiubo Li <[e-mail]>
> Reported-by: Mike Christie <[e-mail]>

Sorry, don't know how I missed this.

Reviewed-by: Josef Bacik <[e-mail]>

Thanks,

Josef

Jens Axboe Oct. 10, 2019, 3:45 p.m. UTC | #2

On 9/19/19 12:14 AM, [e-mail] wrote:
> From: Xiubo Li <[e-mail]>
> 
> 1. nbd_put takes the mutex and drops nbd->ref to 0. It then does
> idr_remove and drops the mutex.
> 
> 2. nbd_genl_connect takes the mutex. idr_find/idr_for_each fails
> to find an existing device, so it does nbd_dev_add.
> 
> 3. just before the nbd_put could call nbd_dev_remove or not finished
> totally, but if nbd_dev_add try to add_disk, we can hit:
> 
> debugfs: Directory 'nbd1' with parent 'block' already present!
> 
> This patch will make sure all the disk add/remove stuff are done
> by holding the nbd_index_mutex lock.

Applied, thanks.

Xiubo Li Oct. 11, 2019, 12:39 a.m. UTC | #3

On 2019/10/10 21:56, Josef Bacik wrote:
> On Thu, Sep 19, 2019 at 11:44:27AM +0530, [e-mail] wrote:
>> From: Xiubo Li <[e-mail]>
>>
>> 1. nbd_put takes the mutex and drops nbd->ref to 0. It then does
>> idr_remove and drops the mutex.
>>
>> 2. nbd_genl_connect takes the mutex. idr_find/idr_for_each fails
>> to find an existing device, so it does nbd_dev_add.
>>
>> 3. just before the nbd_put could call nbd_dev_remove or not finished
>> totally, but if nbd_dev_add try to add_disk, we can hit:
>>
>> debugfs: Directory 'nbd1' with parent 'block' already present!
>>
>> This patch will make sure all the disk add/remove stuff are done
>> by holding the nbd_index_mutex lock.
>>
>> Signed-off-by: Xiubo Li <[e-mail]>
>> Reported-by: Mike Christie <[e-mail]>
> Sorry, don't know how I missed this.

Many thanks.

Xiubo

> Reviewed-by: Josef Bacik <[e-mail]>
>
> Thanks,
>
> Josef

Patch

diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
index ac07e8c94c79..478aa86fc1f2 100644
--- a/drivers/block/nbd.c
+++ b/drivers/block/nbd.c
@@ -248,8 +248,8 @@  static void nbd_put(struct nbd_device *nbd)
 	if (refcount_dec_and_mutex_lock(&nbd->refs,
 					&nbd_index_mutex)) {
 		idr_remove(&nbd_index_idr, nbd->index);
-		mutex_unlock(&nbd_index_mutex);
 		nbd_dev_remove(nbd);
+		mutex_unlock(&nbd_index_mutex);
 	}
 }