[v2] Btrfs: free btrfs_device in place

Commit Message

Liu Bo Oct. 11, 2017, 6:13 p.m. UTC

It's pointless to defer it to a kthread helper as we're not under any
special context.

A bit more about device's lifetime, filesystems use an exclusive way
to access devices and hold a reference count on the devices in use.
Now that we've run blkdev_put() in btrfs_close_bdev(), device->bdev's
lifetime ends at btrfs_close_bdev(), not free_device(), and %bdev is
the only thing that others who need to access it really care about.

Since free_device() only frees the resources of 'struct btrfs_device',
this change won't result in the problem, ie. others like mkfs and md
are unable to access the device immediately after we do umount.

Signed-off-by: Liu Bo <[e-mail]>
Reviewed-by: Anand Jain <[e-mail]>
---

v2: Clarify the lifetime of device and device->bdev respectively and
    clear the concern about raising the 'device is in use' problem.

 fs/btrfs/volumes.c | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)

Patch

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index d983cea..4a72c45 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -836,26 +836,16 @@  void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 	mutex_unlock(&uuid_mutex);
 }
 
-static void __free_device(struct work_struct *work)
+static void free_device(struct rcu_head *head)
 {
 	struct btrfs_device *device;
 
-	device = container_of(work, struct btrfs_device, rcu_work);
+	device = container_of(head, struct btrfs_device, rcu);
 	rcu_string_free(device->name);
 	bio_put(device->flush_bio);
 	kfree(device);
 }
 
-static void free_device(struct rcu_head *head)
-{
-	struct btrfs_device *device;
-
-	device = container_of(head, struct btrfs_device, rcu);
-
-	INIT_WORK(&device->rcu_work, __free_device);
-	schedule_work(&device->rcu_work);
-}
-
 static void btrfs_close_bdev(struct btrfs_device *device)
 {
 	if (device->bdev && device->writeable) {