[PATCH v2] Btrfs: don't attach unnecessary extents to transaction on fsync

[Filipe Manana <fdmanana@suse.com>]

We don't need to attach ordered extents that have completed to the current
transaction. Doing so only makes us hold memory for longer than necessary
and delaying the iput of the inode until the transaction is committed (for
each created ordered extent we do an igrab and then schedule an asynchronous
iput when the ordered extent's reference count drops to 0), preventing the
inode from being evictable until the transaction commits.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
---

V2: Made the patch independent of my previous one, since that other one
    was not correct.

 fs/btrfs/ordered-data.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 72b6f0d..6c4f8e7 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -509,7 +509,21 @@ void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
 		wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
 						   &ordered->flags));
 
-		list_add_tail(&ordered->trans_list, &trans->ordered);
+		/*
+		 * If our ordered extent completed it means it updated the
+		 * fs/subvol and csum trees already, so no need to make the
+		 * current transaction's commit wait for it, as we end up
+		 * holding memory unnecessarily and delaying the inode's iput
+		 * until the transaction commit (we schedule an iput for the
+		 * inode when the ordered extent's refcount drops to 0), which
+		 * prevents it from being evictable until the transaction
+		 * commits.
+		 */
+		if (test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags))
+			btrfs_put_ordered_extent(ordered);
+		else
+			list_add_tail(&ordered->trans_list, &trans->ordered);
+
 		spin_lock_irq(&log->log_extents_lock[index]);
 	}
 	spin_unlock_irq(&log->log_extents_lock[index]);
-- 
2.1.3