nbd: fix possible sysfs duplicate warning
authorXiubo Li <xiubli@redhat.com>
Thu, 19 Sep 2019 06:14:27 +0000 (11:44 +0530)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 6 Nov 2019 12:06:07 +0000 (13:06 +0100)
[ Upstream commit 862488105b84ca744b3d8ff131e0fcfe10644be1 ]

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.

Reported-by: Mike Christie <mchristi@redhat.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/block/nbd.c

index bc2fa4e..d445195 100644 (file)
@@ -228,8 +228,8 @@ 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);
        }
 }