Btrfs: Allocator improvements

Reduce CPU time searching for free blocks by optimizing find_first_extent_bit

Fix find_free_extent to make better use of the last_alloc hint.  Before it
was often finding blocks just before the hint.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

4 files changed, 70 insertions, 15 deletions

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 6f0825b..88e21bd 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -951,7 +951,7 @@ void btrfs_throttle(struct btrfs_root *root)
 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
 {
 	balance_dirty_pages_ratelimited_nr(
-			root->fs_info->btree_inode->i_mapping, 1);
+				   root->fs_info->btree_inode->i_mapping, 1);
 }
 
 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index af5d4a0..239e9d8 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -188,9 +188,10 @@ static u64 noinline find_search_start(struct btrfs_root *root,
 {
 	int ret;
 	struct btrfs_block_group_cache *cache = *cache_ret;
+	struct extent_io_tree *free_space_cache;
+	struct extent_state *state;
 	u64 last;
 	u64 start = 0;
-	u64 end = 0;
 	u64 cache_miss = 0;
 	u64 total_fs_bytes;
 	int wrapped = 0;
@@ -199,6 +200,8 @@ static u64 noinline find_search_start(struct btrfs_root *root,
 		goto out;
 	}
 	total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
+	free_space_cache = &root->fs_info->free_space_cache;
+
 again:
 	ret = cache_block_group(root, cache);
 	if (ret)
@@ -206,22 +209,27 @@ again:
 
 	last = max(search_start, cache->key.objectid);
 
+	spin_lock_irq(&free_space_cache->lock);
+	state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
 	while(1) {
-		ret = find_first_extent_bit(&root->fs_info->free_space_cache,
-					    last, &start, &end, EXTENT_DIRTY);
-		if (ret) {
+		if (!state) {
 			if (!cache_miss)
 				cache_miss = last;
+			spin_unlock_irq(&free_space_cache->lock);
 			goto new_group;
 		}
 
-		start = max(last, start);
-		last = end + 1;
+		start = max(last, state->start);
+		last = state->end + 1;
 		if (last - start < num) {
 			if (last == cache->key.objectid + cache->key.offset)
 				cache_miss = start;
+			do {
+				state = extent_state_next(state);
+			} while(state && !(state->state & EXTENT_DIRTY));
 			continue;
 		}
+		spin_unlock_irq(&free_space_cache->lock);
 		if (data != BTRFS_BLOCK_GROUP_MIXED &&
 		    start + num > cache->key.objectid + cache->key.offset)
 			goto new_group;
@@ -1420,6 +1428,7 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_block_group_cache *block_group;
 	int full_scan = 0;
 	int wrapped = 0;
+	int empty_cluster;
 	u64 cached_start;
 
 	WARN_ON(num_bytes < root->sectorsize);
@@ -1431,11 +1440,15 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
 		data = BTRFS_BLOCK_GROUP_MIXED;
 	}
 
-	if (!data)
+	if (!data) {
 		last_ptr = &root->fs_info->last_alloc;
+		empty_cluster = 128 * 1024;
+	}
 
-	if (data && btrfs_test_opt(root, SSD))
+	if (data && btrfs_test_opt(root, SSD)) {
 		last_ptr = &root->fs_info->last_data_alloc;
+		empty_cluster = 2 * 1024 * 1024;
+	}
 
 	if (last_ptr) {
 		if (*last_ptr)
@@ -1443,8 +1456,9 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
 		else {
 			hint_byte = hint_byte &
 				~((u64)BTRFS_BLOCK_GROUP_SIZE - 1);
-			empty_size += 2 * 1024 * 1024;
+			empty_size += empty_cluster;
 		}
+		search_start = max(search_start, hint_byte);
 	}
 
 	search_end = min(search_end,
@@ -1476,7 +1490,7 @@ check_failed:
 	if (last_ptr && *last_ptr && search_start != *last_ptr) {
 		*last_ptr = 0;
 		if (!empty_size) {
-			empty_size += 2 * 1024 * 1024;
+			empty_size += empty_cluster;
 			total_needed += empty_size;
 		}
 		search_start = find_search_start(root, &block_group,
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 9262ab3..fb64008 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -255,7 +255,7 @@ static int merge_state(struct extent_io_tree *tree,
 			state->start = other->start;
 			other->tree = NULL;
 			if (tree->last == other)
-				tree->last = NULL;
+				tree->last = state;
 			rb_erase(&other->rb_node, &tree->state);
 			free_extent_state(other);
 		}
@@ -268,7 +268,7 @@ static int merge_state(struct extent_io_tree *tree,
 			other->start = state->start;
 			state->tree = NULL;
 			if (tree->last == state)
-				tree->last = NULL;
+				tree->last = other;
 			rb_erase(&state->rb_node, &tree->state);
 			free_extent_state(state);
 		}
@@ -397,8 +397,9 @@ static int clear_state_bit(struct extent_io_tree *tree,
 	if (delete || state->state == 0) {
 		if (state->tree) {
 			clear_state_cb(tree, state, state->state);
-			if (tree->last == state)
-				tree->last = NULL;
+			if (tree->last == state) {
+				tree->last = extent_state_next(state);
+			}
 			rb_erase(&state->rb_node, &tree->state);
 			state->tree = NULL;
 			free_extent_state(state);
@@ -962,6 +963,35 @@ out:
 }
 EXPORT_SYMBOL(find_first_extent_bit);
 
+struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
+						 u64 start, int bits)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node || IS_ERR(node)) {
+		goto out;
+	}
+
+	while(1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (state->end >= start && (state->state & bits)) {
+			return state;
+		}
+		node = rb_next(node);
+		if (!node)
+			break;
+	}
+out:
+	return NULL;
+}
+EXPORT_SYMBOL(find_first_extent_bit_state);
+
 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
 			     u64 *start, u64 *end, u64 max_bytes)
 {
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 151fdad..fcc4bb0 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -81,6 +81,15 @@ struct extent_buffer {
 
 struct extent_map_tree;
 
+static inline struct extent_state *extent_state_next(struct extent_state *state)
+{
+	struct rb_node *node;
+	node = rb_next(&state->rb_node);
+	if (!node)
+		return NULL;
+	return rb_entry(node, struct extent_state, rb_node);
+}
+
 typedef struct extent_map *(get_extent_t)(struct inode *inode,
 					  struct page *page,
 					  size_t page_offset,
@@ -122,6 +131,8 @@ int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 		     gfp_t mask);
 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 			  u64 *start_ret, u64 *end_ret, int bits);
+struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
+						 u64 start, int bits);
 int extent_invalidatepage(struct extent_io_tree *tree,
 			  struct page *page, unsigned long offset);
 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,