Btrfs: faster/more efficient insertion of file extent items

This is an extension to my previous commit titled:

  "Btrfs: faster file extent item replace operations"
  (hash 1acae57b161ef1282f565ef907f72aeed0eb71d9)

Instead of inserting the new file extent item if we deleted existing
file extent items covering our target file range, also allow to insert
the new file extent item if we didn't find any existing items to delete
and replace_extent != 0, since in this case our caller would do another
tree search to insert the new file extent item anyway, therefore just
combine the two tree searches into a single one, saving cpu time, reducing
lock contention and reducing btree node/leaf COW operations.

This covers the case where applications keep doing tail append writes to
files, which for example is the case of Apache CouchDB (its database and
view index files are always open with O_APPEND).

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

1 files changed, 30 insertions, 22 deletions

diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 0165b8672f09..006af2f4dd98 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -720,7 +720,7 @@ int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 	if (drop_cache)
 		btrfs_drop_extent_cache(inode, start, end - 1, 0);
 
-	if (start >= BTRFS_I(inode)->disk_i_size)
+	if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
 		modify_tree = 0;
 
 	while (1) {
@@ -938,34 +938,42 @@ next_slot:
 		 * Set path->slots[0] to first slot, so that after the delete
 		 * if items are move off from our leaf to its immediate left or
 		 * right neighbor leafs, we end up with a correct and adjusted
-		 * path->slots[0] for our insertion.
+		 * path->slots[0] for our insertion (if replace_extent != 0).
 		 */
 		path->slots[0] = del_slot;
 		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 		if (ret)
 			btrfs_abort_transaction(trans, root, ret);
+	}
 
-		leaf = path->nodes[0];
-		/*
-		 * leaf eb has flag EXTENT_BUFFER_STALE if it was deleted (that
-		 * is, its contents got pushed to its neighbors), in which case
-		 * it means path->locks[0] == 0
-		 */
-		if (!ret && replace_extent && leafs_visited == 1 &&
-		    path->locks[0] &&
-		    btrfs_leaf_free_space(root, leaf) >=
-		    sizeof(struct btrfs_item) + extent_item_size) {
-
-			key.objectid = ino;
-			key.type = BTRFS_EXTENT_DATA_KEY;
-			key.offset = start;
-			setup_items_for_insert(root, path, &key,
-					       &extent_item_size,
-					       extent_item_size,
-					       sizeof(struct btrfs_item) +
-					       extent_item_size, 1);
-			*key_inserted = 1;
+	leaf = path->nodes[0];
+	/*
+	 * If btrfs_del_items() was called, it might have deleted a leaf, in
+	 * which case it unlocked our path, so check path->locks[0] matches a
+	 * write lock.
+	 */
+	if (!ret && replace_extent && leafs_visited == 1 &&
+	    (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
+	     path->locks[0] == BTRFS_WRITE_LOCK) &&
+	    btrfs_leaf_free_space(root, leaf) >=
+	    sizeof(struct btrfs_item) + extent_item_size) {
+
+		key.objectid = ino;
+		key.type = BTRFS_EXTENT_DATA_KEY;
+		key.offset = start;
+		if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
+			struct btrfs_key slot_key;
+
+			btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
+			if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
+				path->slots[0]++;
 		}
+		setup_items_for_insert(root, path, &key,
+				       &extent_item_size,
+				       extent_item_size,
+				       sizeof(struct btrfs_item) +
+				       extent_item_size, 1);
+		*key_inserted = 1;
 	}
 
 	if (!replace_extent || !(*key_inserted))