Btrfs: faster file extent item replace operations

When writing to a file we drop existing file extent items that cover the
write range and then add a new file extent item that represents that write
range.

Before this change we were doing a tree lookup to remove the file extent
items, and then after we did another tree lookup to insert the new file
extent item.
Most of the time all the file extent items we need to drop are located
within a single leaf - this is the leaf where our new file extent item ends
up at. Therefore, in this common case just combine these 2 operations into
a single one.

By avoiding the second btree navigation for insertion of the new file extent
item, we reduce btree node/leaf lock acquisitions/releases, btree block/leaf
COW operations, CPU time on btree node/leaf key binary searches, etc.

Besides for file writes, this is an operation that happens for file fsync's
as well. However log btrees are much less likely to big as big as regular
fs btrees, therefore the impact of this change is smaller.

The following benchmark was performed against an SSD drive and a
HDD drive, both for random and sequential writes:

  sysbench --test=fileio --file-num=4096 --file-total-size=8G \
     --file-test-mode=[rndwr|seqwr] --num-threads=512 \
     --file-block-size=8192 \ --max-requests=1000000 \
     --file-fsync-freq=0 --file-io-mode=sync [prepare|run]

All results below are averages of 10 runs of the respective test.

** SSD sequential writes

Before this change: 225.88 Mb/sec
After this change:  277.26 Mb/sec

** SSD random writes

Before this change: 49.91 Mb/sec
After this change:  56.39 Mb/sec

** HDD sequential writes

Before this change: 68.53 Mb/sec
After this change:  69.87 Mb/sec

** HDD random writes

Before this change: 13.04 Mb/sec
After this change:  14.39 Mb/sec

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

fs/btrfs/ctree.h

@@ -3772,7 +3772,10 @@ extern const struct file_operations btrfs_file_operations;
 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 			 struct btrfs_root *root, struct inode *inode,
 			 struct btrfs_path *path, u64 start, u64 end,
-			 u64 *drop_end, int drop_cache);
+			 u64 *drop_end, int drop_cache,
+			 int replace_extent,
+			 u32 extent_item_size,
+			 int *key_inserted);
 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root, struct inode *inode, u64 start,
 		       u64 end, int drop_cache);

fs/btrfs/file.c

@@ -692,7 +692,10 @@ next:
 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 			 struct btrfs_root *root, struct inode *inode,
 			 struct btrfs_path *path, u64 start, u64 end,
-			 u64 *drop_end, int drop_cache)
+			 u64 *drop_end, int drop_cache,
+			 int replace_extent,
+			 u32 extent_item_size,
+			 int *key_inserted)
 {
 	struct extent_buffer *leaf;
 	struct btrfs_file_extent_item *fi;
@@ -712,6 +715,7 @@ int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 	int modify_tree = -1;
 	int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
 	int found = 0;
+	int leafs_visited = 0;
 
 	if (drop_cache)
 		btrfs_drop_extent_cache(inode, start, end - 1, 0);
@@ -733,6 +737,7 @@ int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 				path->slots[0]--;
 		}
 		ret = 0;
+		leafs_visited++;
 next_slot:
 		leaf = path->nodes[0];
 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
@@ -744,6 +749,7 @@ next_slot:
 				ret = 0;
 				break;
 			}
+			leafs_visited++;
 			leaf = path->nodes[0];
 			recow = 1;
 		}
@@ -927,14 +933,44 @@ next_slot:
 	}
 
 	if (!ret && del_nr > 0) {
+		/*
+		 * 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] = 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;
+		}
 	}
 
+	if (!replace_extent || !(*key_inserted))
+		btrfs_release_path(path);
 	if (drop_end)
 		*drop_end = found ? min(end, extent_end) : end;
-	btrfs_release_path(path);
 	return ret;
 }
 
@@ -949,7 +985,7 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 	if (!path)
 		return -ENOMEM;
 	ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
-				   drop_cache);
+				   drop_cache, 0, 0, NULL);
 	btrfs_free_path(path);
 	return ret;
 }
@@ -2219,7 +2255,7 @@ static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
 	while (cur_offset < lockend) {
 		ret = __btrfs_drop_extents(trans, root, inode, path,
 					   cur_offset, lockend + 1,
-					   &drop_end, 1);
+					   &drop_end, 1, 0, 0, NULL);
 		if (ret != -ENOSPC)
 			break;
 

fs/btrfs/inode.c

@@ -125,13 +125,12 @@ static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  * no overlapping inline items exist in the btree
  */
 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
+				struct btrfs_path *path, int extent_inserted,
 				struct btrfs_root *root, struct inode *inode,
 				u64 start, size_t size, size_t compressed_size,
 				int compress_type,
 				struct page **compressed_pages)
 {
-	struct btrfs_key key;
-	struct btrfs_path *path;
 	struct extent_buffer *leaf;
 	struct page *page = NULL;
 	char *kaddr;
@@ -140,29 +139,29 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 	int err = 0;
 	int ret;
 	size_t cur_size = size;
-	size_t datasize;
 	unsigned long offset;
 
 	if (compressed_size && compressed_pages)
 		cur_size = compressed_size;
 
-	path = btrfs_alloc_path();
-	if (!path)
-		return -ENOMEM;
+	inode_add_bytes(inode, size);
 
-	path->leave_spinning = 1;
+	if (!extent_inserted) {
+		struct btrfs_key key;
+		size_t datasize;
 
-	key.objectid = btrfs_ino(inode);
-	key.offset = start;
-	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
-	datasize = btrfs_file_extent_calc_inline_size(cur_size);
+		key.objectid = btrfs_ino(inode);
+		key.offset = start;
+		btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 
-	inode_add_bytes(inode, size);
-	ret = btrfs_insert_empty_item(trans, root, path, &key,
-				      datasize);
-	if (ret) {
-		err = ret;
-		goto fail;
+		datasize = btrfs_file_extent_calc_inline_size(cur_size);
+		path->leave_spinning = 1;
+		ret = btrfs_insert_empty_item(trans, root, path, &key,
+					      datasize);
+		if (ret) {
+			err = ret;
+			goto fail;
+		}
 	}
 	leaf = path->nodes[0];
 	ei = btrfs_item_ptr(leaf, path->slots[0],
@@ -203,7 +202,7 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 		page_cache_release(page);
 	}
 	btrfs_mark_buffer_dirty(leaf);
-	btrfs_free_path(path);
+	btrfs_release_path(path);
 
 	/*
 	 * we're an inline extent, so nobody can
@@ -219,7 +218,6 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 
 	return ret;
 fail:
-	btrfs_free_path(path);
 	return err;
 }
 
@@ -242,6 +240,9 @@ static noinline int cow_file_range_inline(struct btrfs_root *root,
 	u64 aligned_end = ALIGN(end, root->sectorsize);
 	u64 data_len = inline_len;
 	int ret;
+	struct btrfs_path *path;
+	int extent_inserted = 0;
+	u32 extent_item_size;
 
 	if (compressed_size)
 		data_len = compressed_size;
@@ -256,12 +257,27 @@ static noinline int cow_file_range_inline(struct btrfs_root *root,
 		return 1;
 	}
 
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
 	trans = btrfs_join_transaction(root);
-	if (IS_ERR(trans))
+	if (IS_ERR(trans)) {
+		btrfs_free_path(path);
 		return PTR_ERR(trans);
+	}
 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 
-	ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
+	if (compressed_size && compressed_pages)
+		extent_item_size = btrfs_file_extent_calc_inline_size(
+		   compressed_size);
+	else
+		extent_item_size = btrfs_file_extent_calc_inline_size(
+		    inline_len);
+
+	ret = __btrfs_drop_extents(trans, root, inode, path,
+				   start, aligned_end, NULL,
+				   1, 1, extent_item_size, &extent_inserted);
 	if (ret) {
 		btrfs_abort_transaction(trans, root, ret);
 		goto out;
@@ -269,7 +285,8 @@ static noinline int cow_file_range_inline(struct btrfs_root *root,
 
 	if (isize > actual_end)
 		inline_len = min_t(u64, isize, actual_end);
-	ret = insert_inline_extent(trans, root, inode, start,
+	ret = insert_inline_extent(trans, path, extent_inserted,
+				   root, inode, start,
 				   inline_len, compressed_size,
 				   compress_type, compressed_pages);
 	if (ret && ret != -ENOSPC) {
@@ -284,6 +301,7 @@ static noinline int cow_file_range_inline(struct btrfs_root *root,
 	btrfs_delalloc_release_metadata(inode, end + 1 - start);
 	btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 out:
+	btrfs_free_path(path);
 	btrfs_end_transaction(trans, root);
 	return ret;
 }
@@ -1841,14 +1859,13 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_path *path;
 	struct extent_buffer *leaf;
 	struct btrfs_key ins;
+	int extent_inserted = 0;
 	int ret;
 
 	path = btrfs_alloc_path();
 	if (!path)
 		return -ENOMEM;
 
-	path->leave_spinning = 1;
-
 	/*
 	 * we may be replacing one extent in the tree with another.
 	 * The new extent is pinned in the extent map, and we don't want
@@ -1858,17 +1875,23 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	 * the caller is expected to unpin it and allow it to be merged
 	 * with the others.
 	 */
-	ret = btrfs_drop_extents(trans, root, inode, file_pos,
-				 file_pos + num_bytes, 0);
+	ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
+				   file_pos + num_bytes, NULL, 0,
+				   1, sizeof(*fi), &extent_inserted);
 	if (ret)
 		goto out;
 
-	ins.objectid = btrfs_ino(inode);
-	ins.offset = file_pos;
-	ins.type = BTRFS_EXTENT_DATA_KEY;
-	ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
-	if (ret)
-		goto out;
+	if (!extent_inserted) {
+		ins.objectid = btrfs_ino(inode);
+		ins.offset = file_pos;
+		ins.type = BTRFS_EXTENT_DATA_KEY;
+
+		path->leave_spinning = 1;
+		ret = btrfs_insert_empty_item(trans, root, path, &ins,
+					      sizeof(*fi));
+		if (ret)
+			goto out;
+	}
 	leaf = path->nodes[0];
 	fi = btrfs_item_ptr(leaf, path->slots[0],
 			    struct btrfs_file_extent_item);

fs/btrfs/tree-log.c

@@ -3495,21 +3495,27 @@ static int log_one_extent(struct btrfs_trans_handle *trans,
 	int ret;
 	int index = log->log_transid % 2;
 	bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
-
-	ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
-				   em->start + em->len, NULL, 0);
-	if (ret)
-		return ret;
+	int extent_inserted = 0;
 
 	INIT_LIST_HEAD(&ordered_sums);
 	btrfs_init_map_token(&token);
-	key.objectid = btrfs_ino(inode);
-	key.type = BTRFS_EXTENT_DATA_KEY;
-	key.offset = em->start;
 
-	ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*fi));
+	ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
+				   em->start + em->len, NULL, 0, 1,
+				   sizeof(*fi), &extent_inserted);
 	if (ret)
 		return ret;
+
+	if (!extent_inserted) {
+		key.objectid = btrfs_ino(inode);
+		key.type = BTRFS_EXTENT_DATA_KEY;
+		key.offset = em->start;
+
+		ret = btrfs_insert_empty_item(trans, log, path, &key,
+					      sizeof(*fi));
+		if (ret)
+			return ret;
+	}
 	leaf = path->nodes[0];
 	fi = btrfs_item_ptr(leaf, path->slots[0],
 			    struct btrfs_file_extent_item);