Drop locks in btrfs_search_slot when reading a tree block.

One lock per btree block can make for significant congestion if everyone
has to wait for IO at the high levels of the btree.  This drops
locks held by a path when doing reads during a tree search.

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

4 files changed, 38 insertions, 39 deletions

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index dff4da082d0..1b756fae279 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -63,7 +63,6 @@ void btrfs_free_path(struct btrfs_path *p)
 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
 {
 	int i;
-	int skip = p->skip_locking;
 	int keep = p->keep_locks;
 
 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
@@ -76,7 +75,6 @@ void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
 		free_extent_buffer(p->nodes[i]);
 	}
 	memset(p, 0, sizeof(*p));
-	p->skip_locking = skip;
 	p->keep_locks = keep;
 }
 
@@ -1137,7 +1135,6 @@ static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
 		return;
 
 	node = path->nodes[level];
-	WARN_ON(!path->skip_locking && !btrfs_tree_locked(node));
 
 	search = btrfs_node_blockptr(node, slot);
 	blocksize = btrfs_level_size(root, level - 1);
@@ -1192,6 +1189,7 @@ static void unlock_up(struct btrfs_path *path, int level, int lowest_unlock)
 {
 	int i;
 	int skip_level = level;
+	int no_skips = 0;
 	struct extent_buffer *t;
 
 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
@@ -1199,27 +1197,24 @@ static void unlock_up(struct btrfs_path *path, int level, int lowest_unlock)
 			break;
 		if (!path->locks[i])
 			break;
-		if (path->slots[i] == 0) {
+		if (!no_skips && path->slots[i] == 0) {
 			skip_level = i + 1;
 			continue;
 		}
-		if (path->keep_locks) {
+		if (!no_skips && path->keep_locks) {
 			u32 nritems;
 			t = path->nodes[i];
 			nritems = btrfs_header_nritems(t);
-			if (nritems < 2 || path->slots[i] >= nritems - 2) {
-if (path->keep_locks) {
-//printk("path %p skip level now %d\n", path, skip_level);
-}
+			if (nritems < 1 || path->slots[i] >= nritems - 1) {
 				skip_level = i + 1;
 				continue;
 			}
 		}
+		if (skip_level < i && i >= lowest_unlock)
+			no_skips = 1;
+
 		t = path->nodes[i];
 		if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
-if (path->keep_locks) {
-//printk("path %p unlocking level %d slot %d nritems %d skip_level %d\n", path, i, path->slots[i], btrfs_header_nritems(t), skip_level);
-}
 			btrfs_tree_unlock(t);
 			path->locks[i] = 0;
 		}
@@ -1244,6 +1239,7 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 		      ins_len, int cow)
 {
 	struct extent_buffer *b;
+	struct extent_buffer *tmp;
 	int slot;
 	int ret;
 	int level;
@@ -1263,10 +1259,7 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 	if (ins_len < 0)
 		lowest_unlock = 2;
 again:
-	if (!p->skip_locking)
-		b = btrfs_lock_root_node(root);
-	else
-		b = btrfs_root_node(root);
+	b = btrfs_lock_root_node(root);
 
 	while (b) {
 		level = btrfs_header_level(b);
@@ -1286,8 +1279,7 @@ again:
 			WARN_ON(1);
 		level = btrfs_header_level(b);
 		p->nodes[level] = b;
-		if (!p->skip_locking)
-			p->locks[level] = 1;
+		p->locks[level] = 1;
 		ret = check_block(root, p, level);
 		if (ret)
 			return -1;
@@ -1328,10 +1320,29 @@ again:
 				reada_for_search(root, p, level, slot,
 						 key->objectid);
 
-			b = read_node_slot(root, b, slot);
-			if (!p->skip_locking)
-				btrfs_tree_lock(b);
-			unlock_up(p, level + 1, lowest_unlock);
+			tmp = btrfs_find_tree_block(root,
+					  btrfs_node_blockptr(b, slot),
+					  btrfs_level_size(root, level - 1));
+			if (tmp && btrfs_buffer_uptodate(tmp,
+				   btrfs_node_ptr_generation(b, slot))) {
+				b = tmp;
+			} else {
+				/*
+				 * reduce lock contention at high levels
+				 * of the btree by dropping locks before
+				 * we read.
+				 */
+				if (level > 1) {
+					btrfs_release_path(NULL, p);
+					if (tmp)
+						free_extent_buffer(tmp);
+					goto again;
+				} else {
+					b = read_node_slot(root, b, slot);
+				}
+			}
+			btrfs_tree_lock(b);
+			unlock_up(p, level, lowest_unlock);
 		} else {
 			p->slots[level] = slot;
 			if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
@@ -3007,17 +3018,8 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
 			reada_for_search(root, path, level, slot, 0);
 
 		next = read_node_slot(root, c, slot);
-		if (!path->skip_locking) {
-			if (!btrfs_tree_locked(c)) {
-				int i;
-				WARN_ON(1);
-printk("path %p no lock on level %d\n", path, level);
-for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
-printk("path %p level %d slot %d nritems %d\n", path, i, path->slots[i], btrfs_header_nritems(path->nodes[i]));
-}
-			}
-			btrfs_tree_lock(next);
-		}
+		WARN_ON(!btrfs_tree_locked(c));
+		btrfs_tree_lock(next);
 		break;
 	}
 	path->slots[level] = slot;
@@ -3035,10 +3037,8 @@ printk("path %p level %d slot %d nritems %d\n", path, i, path->slots[i], btrfs_h
 		if (level == 1 && path->locks[1] && path->reada)
 			reada_for_search(root, path, level, slot, 0);
 		next = read_node_slot(root, next, 0);
-		if (!path->skip_locking) {
-			WARN_ON(!btrfs_tree_locked(path->nodes[level]));
-			btrfs_tree_lock(next);
-		}
+		WARN_ON(!btrfs_tree_locked(path->nodes[level]));
+		btrfs_tree_lock(next);
 	}
 done:
 	unlock_up(path, 0, 1);
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 692b8ea42de..9ea12d42741 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -336,7 +336,6 @@ struct btrfs_path {
 	/* keep some upper locks as we walk down */
 	int keep_locks;
 	int lowest_level;
-	int skip_locking;
 };
 
 /*
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index f638803549e..ffc363d2fb2 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1684,6 +1684,7 @@ void btrfs_throttle(struct btrfs_root *root)
 #else
 		blk_congestion_wait(WRITE, HZ/20);
 #endif
+
 	}
 }
 
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 890b9e9d8e2..0905653dd3f 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -88,7 +88,6 @@ static int cache_block_group(struct btrfs_root *root,
 		return -ENOMEM;
 
 	path->reada = 2;
-	path->skip_locking = 1;
 	first_free = block_group->key.objectid;
 	key.objectid = block_group->key.objectid;
 	key.offset = 0;