Btrfs: use FLUSH_LIMIT for relocation in reserve_metadata_bytes

We used to allow you to set FLUSH_ALL and then just wouldn't do things like
commit transactions or wait on ordered extents if we noticed you were in a
transaction.  However now that all the flushing for FLUSH_ALL is asynchronous
we've lost the ability to tell, and we could end up deadlocking.  So instead use
FLUSH_LIMIT in reserve_metadata_bytes in relocation and then return -EAGAIN if
we error out to preserve the previous behavior.  I've also added an ASSERT() to
catch anybody else who tries to do this.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>

1 files changed, 20 insertions, 17 deletions

diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index 23e16de52e23..fc067b07e31f 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -2604,25 +2604,28 @@ static int reserve_metadata_space(struct btrfs_trans_handle *trans,
 
 	trans->block_rsv = rc->block_rsv;
 	rc->reserved_bytes += num_bytes;
+
+	/*
+	 * We are under a transaction here so we can only do limited flushing.
+	 * If we get an enospc just kick back -EAGAIN so we know to drop the
+	 * transaction and try to refill when we can flush all the things.
+	 */
 	ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
-				BTRFS_RESERVE_FLUSH_ALL);
+				BTRFS_RESERVE_FLUSH_LIMIT);
 	if (ret) {
-		if (ret == -EAGAIN) {
-			tmp = rc->extent_root->nodesize *
-				RELOCATION_RESERVED_NODES;
-			while (tmp <= rc->reserved_bytes)
-				tmp <<= 1;
-			/*
-			 * only one thread can access block_rsv at this point,
-			 * so we don't need hold lock to protect block_rsv.
-			 * we expand more reservation size here to allow enough
-			 * space for relocation and we will return earlier in
-			 * enospc case.
-			 */
-			rc->block_rsv->size = tmp + rc->extent_root->nodesize *
-					      RELOCATION_RESERVED_NODES;
-		}
-		return ret;
+		tmp = rc->extent_root->nodesize * RELOCATION_RESERVED_NODES;
+		while (tmp <= rc->reserved_bytes)
+			tmp <<= 1;
+		/*
+		 * only one thread can access block_rsv at this point,
+		 * so we don't need hold lock to protect block_rsv.
+		 * we expand more reservation size here to allow enough
+		 * space for relocation and we will return eailer in
+		 * enospc case.
+		 */
+		rc->block_rsv->size = tmp + rc->extent_root->nodesize *
+			RELOCATION_RESERVED_NODES;
+		return -EAGAIN;
 	}
 
 	return 0;