btrfs: qgroup: avoid start/commit empty transaction when flushing reservations

When flushing reservations we are using btrfs_join_transaction() to get a
handle for the current transaction and then commit it to try to release
space. However btrfs_join_transaction() has some undesirable consequences:

1) If there's no running transaction, it will create one, and we will
   commit it right after. This is unnecessary because it will not release
   any space, and it will result in unnecessary IO and rotation of backup
   roots in the superblock;

2) If there's a current transaction and that transaction is committing
   (its state is >= TRANS_STATE_COMMIT_DOING), it will wait for that
   transaction to almost finish its commit (for its state to be >=
   TRANS_STATE_UNBLOCKED) and then start and return a new transaction.

   We will then commit that new transaction, which is pointless because
   all we wanted was to wait for the current (previous) transaction to
   fully finish its commit (state == TRANS_STATE_COMPLETED), and by
   starting and committing a new transaction we are wasting IO too and
   causing unnecessary rotation of backup roots in the superblock.

So improve this by using btrfs_attach_transaction_barrier() instead, which
does not create a new transaction if there's none running, and if there's
a current transaction that is committing, it will wait for it to fully
commit and not create a new transaction.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c
index b9c5dff960de4e0f46ea06099eb27513e7f45ebe..63fdbbf61f15294d8a53ed148578673bc217c14f 100644 (file)
--- a/fs/btrfs/qgroup.c
+++ b/fs/btrfs/qgroup.c
@@ -1341,12 +1341,14 @@ static int flush_reservations(struct btrfs_fs_info *fs_info)
        if (ret)
                return ret;
        btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
-       trans = btrfs_join_transaction(fs_info->tree_root);
-       if (IS_ERR(trans))
-               return PTR_ERR(trans);
-       ret = btrfs_commit_transaction(trans);
 
-       return ret;
+       trans = btrfs_attach_transaction_barrier(fs_info->tree_root);
+       if (IS_ERR(trans)) {
+               ret = PTR_ERR(trans);
+               return (ret == -ENOENT) ? 0 : ret;
+       }
+
+       return btrfs_commit_transaction(trans);
 }
 
 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)