Btrfs: join tree mod log code with the code holding back delayed refs

We've got two mechanisms both required for reliable backref resolving (tree
mod log and holding back delayed refs). You cannot make use of one without
the other. So instead of requiring the user of this mechanism to setup both
correctly, we join them into a single interface.

Additionally, we stop inserting non-blockers into fs_info->tree_mod_seq_list
as we did before, which was of no value.

Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>

1 files changed, 171 insertions, 104 deletions

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 67fe46f..bef68ab 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -321,7 +321,7 @@ struct tree_mod_root {
 struct tree_mod_elem {
 	struct rb_node node;
 	u64 index;		/* shifted logical */
-	struct seq_list elem;
+	u64 seq;
 	enum mod_log_op op;
 
 	/* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
@@ -341,20 +341,50 @@ struct tree_mod_elem {
 	struct tree_mod_root old_root;
 };
 
-static inline void
-__get_tree_mod_seq(struct btrfs_fs_info *fs_info, struct seq_list *elem)
+static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
 {
-	elem->seq = atomic_inc_return(&fs_info->tree_mod_seq);
-	list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
+	read_lock(&fs_info->tree_mod_log_lock);
 }
 
-void btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
-			    struct seq_list *elem)
+static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
+{
+	read_unlock(&fs_info->tree_mod_log_lock);
+}
+
+static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
+{
+	write_lock(&fs_info->tree_mod_log_lock);
+}
+
+static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
 {
-	elem->flags = 1;
+	write_unlock(&fs_info->tree_mod_log_lock);
+}
+
+/*
+ * This adds a new blocker to the tree mod log's blocker list if the @elem
+ * passed does not already have a sequence number set. So when a caller expects
+ * to record tree modifications, it should ensure to set elem->seq to zero
+ * before calling btrfs_get_tree_mod_seq.
+ * Returns a fresh, unused tree log modification sequence number, even if no new
+ * blocker was added.
+ */
+u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
+			   struct seq_list *elem)
+{
+	u64 seq;
+
+	tree_mod_log_write_lock(fs_info);
 	spin_lock(&fs_info->tree_mod_seq_lock);
-	__get_tree_mod_seq(fs_info, elem);
+	if (!elem->seq) {
+		elem->seq = btrfs_inc_tree_mod_seq(fs_info);
+		list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
+	}
+	seq = btrfs_inc_tree_mod_seq(fs_info);
 	spin_unlock(&fs_info->tree_mod_seq_lock);
+	tree_mod_log_write_unlock(fs_info);
+
+	return seq;
 }
 
 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
@@ -371,41 +401,46 @@ void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
 	if (!seq_putting)
 		return;
 
-	BUG_ON(!(elem->flags & 1));
 	spin_lock(&fs_info->tree_mod_seq_lock);
 	list_del(&elem->list);
+	elem->seq = 0;
 
 	list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
-		if ((cur_elem->flags & 1) && cur_elem->seq < min_seq) {
+		if (cur_elem->seq < min_seq) {
 			if (seq_putting > cur_elem->seq) {
 				/*
 				 * blocker with lower sequence number exists, we
 				 * cannot remove anything from the log
 				 */
-				goto out;
+				spin_unlock(&fs_info->tree_mod_seq_lock);
+				return;
 			}
 			min_seq = cur_elem->seq;
 		}
 	}
+	spin_unlock(&fs_info->tree_mod_seq_lock);
+
+	/*
+	 * we removed the lowest blocker from the blocker list, so there may be
+	 * more processible delayed refs.
+	 */
+	wake_up(&fs_info->tree_mod_seq_wait);
 
 	/*
 	 * anything that's lower than the lowest existing (read: blocked)
 	 * sequence number can be removed from the tree.
 	 */
-	write_lock(&fs_info->tree_mod_log_lock);
+	tree_mod_log_write_lock(fs_info);
 	tm_root = &fs_info->tree_mod_log;
 	for (node = rb_first(tm_root); node; node = next) {
 		next = rb_next(node);
 		tm = container_of(node, struct tree_mod_elem, node);
-		if (tm->elem.seq > min_seq)
+		if (tm->seq > min_seq)
 			continue;
 		rb_erase(node, tm_root);
-		list_del(&tm->elem.list);
 		kfree(tm);
 	}
-	write_unlock(&fs_info->tree_mod_log_lock);
-out:
-	spin_unlock(&fs_info->tree_mod_seq_lock);
+	tree_mod_log_write_unlock(fs_info);
 }
 
 /*
@@ -423,11 +458,9 @@ __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
 	struct rb_node **new;
 	struct rb_node *parent = NULL;
 	struct tree_mod_elem *cur;
-	int ret = 0;
 
-	BUG_ON(!tm || !tm->elem.seq);
+	BUG_ON(!tm || !tm->seq);
 
-	write_lock(&fs_info->tree_mod_log_lock);
 	tm_root = &fs_info->tree_mod_log;
 	new = &tm_root->rb_node;
 	while (*new) {
@@ -437,88 +470,81 @@ __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
 			new = &((*new)->rb_left);
 		else if (cur->index > tm->index)
 			new = &((*new)->rb_right);
-		else if (cur->elem.seq < tm->elem.seq)
+		else if (cur->seq < tm->seq)
 			new = &((*new)->rb_left);
-		else if (cur->elem.seq > tm->elem.seq)
+		else if (cur->seq > tm->seq)
 			new = &((*new)->rb_right);
 		else {
 			kfree(tm);
-			ret = -EEXIST;
-			goto unlock;
+			return -EEXIST;
 		}
 	}
 
 	rb_link_node(&tm->node, parent, new);
 	rb_insert_color(&tm->node, tm_root);
-unlock:
-	write_unlock(&fs_info->tree_mod_log_lock);
-	return ret;
+	return 0;
 }
 
+/*
+ * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
+ * returns zero with the tree_mod_log_lock acquired. The caller must hold
+ * this until all tree mod log insertions are recorded in the rb tree and then
+ * call tree_mod_log_write_unlock() to release.
+ */
 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
 				    struct extent_buffer *eb) {
 	smp_mb();
 	if (list_empty(&(fs_info)->tree_mod_seq_list))
 		return 1;
-	if (!eb)
-		return 0;
-	if (btrfs_header_level(eb) == 0)
+	if (eb && btrfs_header_level(eb) == 0)
 		return 1;
+
+	tree_mod_log_write_lock(fs_info);
+	if (list_empty(&fs_info->tree_mod_seq_list)) {
+		/*
+		 * someone emptied the list while we were waiting for the lock.
+		 * we must not add to the list when no blocker exists.
+		 */
+		tree_mod_log_write_unlock(fs_info);
+		return 1;
+	}
+
 	return 0;
 }
 
 /*
- * This allocates memory and gets a tree modification sequence number when
- * needed.
+ * This allocates memory and gets a tree modification sequence number.
  *
- * Returns 0 when no sequence number is needed, < 0 on error.
- * Returns 1 when a sequence number was added. In this case,
- * fs_info->tree_mod_seq_lock was acquired and must be released by the caller
- * after inserting into the rb tree.
+ * Returns <0 on error.
+ * Returns >0 (the added sequence number) on success.
  */
 static inline int tree_mod_alloc(struct btrfs_fs_info *fs_info, gfp_t flags,
 				 struct tree_mod_elem **tm_ret)
 {
 	struct tree_mod_elem *tm;
-	int seq;
 
-	if (tree_mod_dont_log(fs_info, NULL))
-		return 0;
-
-	tm = *tm_ret = kzalloc(sizeof(*tm), flags);
+	/*
+	 * once we switch from spin locks to something different, we should
+	 * honor the flags parameter here.
+	 */
+	tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
 	if (!tm)
 		return -ENOMEM;
 
-	tm->elem.flags = 0;
-	spin_lock(&fs_info->tree_mod_seq_lock);
-	if (list_empty(&fs_info->tree_mod_seq_list)) {
-		/*
-		 * someone emptied the list while we were waiting for the lock.
-		 * we must not add to the list, because no blocker exists. items
-		 * are removed from the list only when the existing blocker is
-		 * removed from the list.
-		 */
-		kfree(tm);
-		seq = 0;
-		spin_unlock(&fs_info->tree_mod_seq_lock);
-	} else {
-		__get_tree_mod_seq(fs_info, &tm->elem);
-		seq = tm->elem.seq;
-	}
-
-	return seq;
+	tm->seq = btrfs_inc_tree_mod_seq(fs_info);
+	return tm->seq;
 }
 
-static noinline int
-tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
-			     struct extent_buffer *eb, int slot,
-			     enum mod_log_op op, gfp_t flags)
+static inline int
+__tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
+			  struct extent_buffer *eb, int slot,
+			  enum mod_log_op op, gfp_t flags)
 {
-	struct tree_mod_elem *tm;
 	int ret;
+	struct tree_mod_elem *tm;
 
 	ret = tree_mod_alloc(fs_info, flags, &tm);
-	if (ret <= 0)
+	if (ret < 0)
 		return ret;
 
 	tm->index = eb->start >> PAGE_CACHE_SHIFT;
@@ -530,8 +556,22 @@ tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
 	tm->slot = slot;
 	tm->generation = btrfs_node_ptr_generation(eb, slot);
 
-	ret = __tree_mod_log_insert(fs_info, tm);
-	spin_unlock(&fs_info->tree_mod_seq_lock);
+	return __tree_mod_log_insert(fs_info, tm);
+}
+
+static noinline int
+tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
+			     struct extent_buffer *eb, int slot,
+			     enum mod_log_op op, gfp_t flags)
+{
+	int ret;
+
+	if (tree_mod_dont_log(fs_info, eb))
+		return 0;
+
+	ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags);
+
+	tree_mod_log_write_unlock(fs_info);
 	return ret;
 }
 
@@ -543,6 +583,14 @@ tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
 }
 
 static noinline int
+tree_mod_log_insert_key_locked(struct btrfs_fs_info *fs_info,
+			     struct extent_buffer *eb, int slot,
+			     enum mod_log_op op)
+{
+	return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS);
+}
+
+static noinline int
 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
 			 struct extent_buffer *eb, int dst_slot, int src_slot,
 			 int nr_items, gfp_t flags)
@@ -555,14 +603,14 @@ tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
 		return 0;
 
 	for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
-		ret = tree_mod_log_insert_key(fs_info, eb, i + dst_slot,
+		ret = tree_mod_log_insert_key_locked(fs_info, eb, i + dst_slot,
 					      MOD_LOG_KEY_REMOVE_WHILE_MOVING);
 		BUG_ON(ret < 0);
 	}
 
 	ret = tree_mod_alloc(fs_info, flags, &tm);
-	if (ret <= 0)
-		return ret;
+	if (ret < 0)
+		goto out;
 
 	tm->index = eb->start >> PAGE_CACHE_SHIFT;
 	tm->slot = src_slot;
@@ -571,10 +619,26 @@ tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
 	tm->op = MOD_LOG_MOVE_KEYS;
 
 	ret = __tree_mod_log_insert(fs_info, tm);
-	spin_unlock(&fs_info->tree_mod_seq_lock);
+out:
+	tree_mod_log_write_unlock(fs_info);
 	return ret;
 }
 
+static inline void
+__tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
+{
+	int i;
+	u32 nritems;
+	int ret;
+
+	nritems = btrfs_header_nritems(eb);
+	for (i = nritems - 1; i >= 0; i--) {
+		ret = tree_mod_log_insert_key_locked(fs_info, eb, i,
+					      MOD_LOG_KEY_REMOVE_WHILE_FREEING);
+		BUG_ON(ret < 0);
+	}
+}
+
 static noinline int
 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
 			 struct extent_buffer *old_root,
@@ -583,9 +647,14 @@ tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
 	struct tree_mod_elem *tm;
 	int ret;
 
+	if (tree_mod_dont_log(fs_info, NULL))
+		return 0;
+
+	__tree_mod_log_free_eb(fs_info, old_root);
+
 	ret = tree_mod_alloc(fs_info, flags, &tm);
-	if (ret <= 0)
-		return ret;
+	if (ret < 0)
+		goto out;
 
 	tm->index = new_root->start >> PAGE_CACHE_SHIFT;
 	tm->old_root.logical = old_root->start;
@@ -594,7 +663,8 @@ tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
 	tm->op = MOD_LOG_ROOT_REPLACE;
 
 	ret = __tree_mod_log_insert(fs_info, tm);
-	spin_unlock(&fs_info->tree_mod_seq_lock);
+out:
+	tree_mod_log_write_unlock(fs_info);
 	return ret;
 }
 
@@ -608,7 +678,7 @@ __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
 	struct tree_mod_elem *found = NULL;
 	u64 index = start >> PAGE_CACHE_SHIFT;
 
-	read_lock(&fs_info->tree_mod_log_lock);
+	tree_mod_log_read_lock(fs_info);
 	tm_root = &fs_info->tree_mod_log;
 	node = tm_root->rb_node;
 	while (node) {
@@ -617,18 +687,18 @@ __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
 			node = node->rb_left;
 		} else if (cur->index > index) {
 			node = node->rb_right;
-		} else if (cur->elem.seq < min_seq) {
+		} else if (cur->seq < min_seq) {
 			node = node->rb_left;
 		} else if (!smallest) {
 			/* we want the node with the highest seq */
 			if (found)
-				BUG_ON(found->elem.seq > cur->elem.seq);
+				BUG_ON(found->seq > cur->seq);
 			found = cur;
 			node = node->rb_left;
-		} else if (cur->elem.seq > min_seq) {
+		} else if (cur->seq > min_seq) {
 			/* we want the node with the smallest seq */
 			if (found)
-				BUG_ON(found->elem.seq < cur->elem.seq);
+				BUG_ON(found->seq < cur->seq);
 			found = cur;
 			node = node->rb_right;
 		} else {
@@ -636,7 +706,7 @@ __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
 			break;
 		}
 	}
-	read_unlock(&fs_info->tree_mod_log_lock);
+	tree_mod_log_read_unlock(fs_info);
 
 	return found;
 }
@@ -664,7 +734,7 @@ tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
 	return __tree_mod_log_search(fs_info, start, min_seq, 0);
 }
 
-static inline void
+static noinline void
 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
 		     struct extent_buffer *src, unsigned long dst_offset,
 		     unsigned long src_offset, int nr_items)
@@ -675,18 +745,23 @@ tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
 	if (tree_mod_dont_log(fs_info, NULL))
 		return;
 
-	if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
+	if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) {
+		tree_mod_log_write_unlock(fs_info);
 		return;
+	}
 
-	/* speed this up by single seq for all operations? */
 	for (i = 0; i < nr_items; i++) {
-		ret = tree_mod_log_insert_key(fs_info, src, i + src_offset,
-					      MOD_LOG_KEY_REMOVE);
+		ret = tree_mod_log_insert_key_locked(fs_info, src,
+						     i + src_offset,
+						     MOD_LOG_KEY_REMOVE);
 		BUG_ON(ret < 0);
-		ret = tree_mod_log_insert_key(fs_info, dst, i + dst_offset,
-					      MOD_LOG_KEY_ADD);
+		ret = tree_mod_log_insert_key_locked(fs_info, dst,
+						     i + dst_offset,
+						     MOD_LOG_KEY_ADD);
 		BUG_ON(ret < 0);
 	}
+
+	tree_mod_log_write_unlock(fs_info);
 }
 
 static inline void
@@ -699,7 +774,7 @@ tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
 	BUG_ON(ret < 0);
 }
 
-static inline void
+static noinline void
 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
 			  struct extent_buffer *eb,
 			  struct btrfs_disk_key *disk_key, int slot, int atomic)
@@ -712,30 +787,22 @@ tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
 	BUG_ON(ret < 0);
 }
 
-static void tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
-				 struct extent_buffer *eb)
+static noinline void
+tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
 {
-	int i;
-	int ret;
-	u32 nritems;
-
 	if (tree_mod_dont_log(fs_info, eb))
 		return;
 
-	nritems = btrfs_header_nritems(eb);
-	for (i = nritems - 1; i >= 0; i--) {
-		ret = tree_mod_log_insert_key(fs_info, eb, i,
-					      MOD_LOG_KEY_REMOVE_WHILE_FREEING);
-		BUG_ON(ret < 0);
-	}
+	__tree_mod_log_free_eb(fs_info, eb);
+
+	tree_mod_log_write_unlock(fs_info);
 }
 
-static inline void
+static noinline void
 tree_mod_log_set_root_pointer(struct btrfs_root *root,
 			      struct extent_buffer *new_root_node)
 {
 	int ret;
-	tree_mod_log_free_eb(root->fs_info, root->node);
 	ret = tree_mod_log_insert_root(root->fs_info, root->node,
 				       new_root_node, GFP_NOFS);
 	BUG_ON(ret < 0);
@@ -1069,7 +1136,7 @@ __tree_mod_log_rewind(struct extent_buffer *eb, u64 time_seq,
 	unsigned long p_size = sizeof(struct btrfs_key_ptr);
 
 	n = btrfs_header_nritems(eb);
-	while (tm && tm->elem.seq >= time_seq) {
+	while (tm && tm->seq >= time_seq) {
 		/*
 		 * all the operations are recorded with the operator used for
 		 * the modification. as we're going backwards, we do the