1 files changed, 1592 insertions, 1102 deletions

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index edc7d208c5c..359a754c782 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -21,6 +21,7 @@
 #include <linux/blkdev.h>
 #include <linux/sort.h>
 #include <linux/rcupdate.h>
+#include <linux/kthread.h>
 #include "compat.h"
 #include "hash.h"
 #include "ctree.h"
@@ -31,12 +32,12 @@
 #include "locking.h"
 #include "free-space-cache.h"
 
-static int update_reserved_extents(struct btrfs_root *root,
-				   u64 bytenr, u64 num, int reserve);
 static int update_block_group(struct btrfs_trans_handle *trans,
 			      struct btrfs_root *root,
 			      u64 bytenr, u64 num_bytes, int alloc,
 			      int mark_free);
+static int update_reserved_extents(struct btrfs_block_group_cache *cache,
+				   u64 num_bytes, int reserve);
 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 				struct btrfs_root *root,
 				u64 bytenr, u64 num_bytes, u64 parent,
@@ -56,10 +57,26 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 				     u64 parent, u64 root_objectid,
 				     u64 flags, struct btrfs_disk_key *key,
 				     int level, struct btrfs_key *ins);
-
 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *extent_root, u64 alloc_bytes,
 			  u64 flags, int force);
+static int pin_down_bytes(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  struct btrfs_path *path,
+			  u64 bytenr, u64 num_bytes,
+			  int is_data, int reserved,
+			  struct extent_buffer **must_clean);
+static int find_next_key(struct btrfs_path *path, int level,
+			 struct btrfs_key *key);
+static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
+			    int dump_block_groups);
+
+static noinline int
+block_group_cache_done(struct btrfs_block_group_cache *cache)
+{
+	smp_mb();
+	return cache->cached == BTRFS_CACHE_FINISHED;
+}
 
 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 {
@@ -145,21 +162,96 @@ block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 	return ret;
 }
 
+static int add_excluded_extent(struct btrfs_root *root,
+			       u64 start, u64 num_bytes)
+{
+	u64 end = start + num_bytes - 1;
+	set_extent_bits(&root->fs_info->freed_extents[0],
+			start, end, EXTENT_UPTODATE, GFP_NOFS);
+	set_extent_bits(&root->fs_info->freed_extents[1],
+			start, end, EXTENT_UPTODATE, GFP_NOFS);
+	return 0;
+}
+
+static void free_excluded_extents(struct btrfs_root *root,
+				  struct btrfs_block_group_cache *cache)
+{
+	u64 start, end;
+
+	start = cache->key.objectid;
+	end = start + cache->key.offset - 1;
+
+	clear_extent_bits(&root->fs_info->freed_extents[0],
+			  start, end, EXTENT_UPTODATE, GFP_NOFS);
+	clear_extent_bits(&root->fs_info->freed_extents[1],
+			  start, end, EXTENT_UPTODATE, GFP_NOFS);
+}
+
+static int exclude_super_stripes(struct btrfs_root *root,
+				 struct btrfs_block_group_cache *cache)
+{
+	u64 bytenr;
+	u64 *logical;
+	int stripe_len;
+	int i, nr, ret;
+
+	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
+		bytenr = btrfs_sb_offset(i);
+		ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
+				       cache->key.objectid, bytenr,
+				       0, &logical, &nr, &stripe_len);
+		BUG_ON(ret);
+
+		while (nr--) {
+			cache->bytes_super += stripe_len;
+			ret = add_excluded_extent(root, logical[nr],
+						  stripe_len);
+			BUG_ON(ret);
+		}
+
+		kfree(logical);
+	}
+	return 0;
+}
+
+static struct btrfs_caching_control *
+get_caching_control(struct btrfs_block_group_cache *cache)
+{
+	struct btrfs_caching_control *ctl;
+
+	spin_lock(&cache->lock);
+	if (cache->cached != BTRFS_CACHE_STARTED) {
+		spin_unlock(&cache->lock);
+		return NULL;
+	}
+
+	ctl = cache->caching_ctl;
+	atomic_inc(&ctl->count);
+	spin_unlock(&cache->lock);
+	return ctl;
+}
+
+static void put_caching_control(struct btrfs_caching_control *ctl)
+{
+	if (atomic_dec_and_test(&ctl->count))
+		kfree(ctl);
+}
+
 /*
  * this is only called by cache_block_group, since we could have freed extents
  * we need to check the pinned_extents for any extents that can't be used yet
  * since their free space will be released as soon as the transaction commits.
  */
-static int add_new_free_space(struct btrfs_block_group_cache *block_group,
+static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 			      struct btrfs_fs_info *info, u64 start, u64 end)
 {
-	u64 extent_start, extent_end, size;
+	u64 extent_start, extent_end, size, total_added = 0;
 	int ret;
 
 	while (start < end) {
-		ret = find_first_extent_bit(&info->pinned_extents, start,
+		ret = find_first_extent_bit(info->pinned_extents, start,
 					    &extent_start, &extent_end,
-					    EXTENT_DIRTY);
+					    EXTENT_DIRTY | EXTENT_UPTODATE);
 		if (ret)
 			break;
 
@@ -167,6 +259,7 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 			start = extent_end + 1;
 		} else if (extent_start > start && extent_start < end) {
 			size = extent_start - start;
+			total_added += size;
 			ret = btrfs_add_free_space(block_group, start,
 						   size);
 			BUG_ON(ret);
@@ -178,112 +271,183 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 
 	if (start < end) {
 		size = end - start;
+		total_added += size;
 		ret = btrfs_add_free_space(block_group, start, size);
 		BUG_ON(ret);
 	}
 
-	return 0;
-}
-
-static int remove_sb_from_cache(struct btrfs_root *root,
-				struct btrfs_block_group_cache *cache)
-{
-	u64 bytenr;
-	u64 *logical;
-	int stripe_len;
-	int i, nr, ret;
-
-	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
-		bytenr = btrfs_sb_offset(i);
-		ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
-				       cache->key.objectid, bytenr, 0,
-				       &logical, &nr, &stripe_len);
-		BUG_ON(ret);
-		while (nr--) {
-			btrfs_remove_free_space(cache, logical[nr],
-						stripe_len);
-		}
-		kfree(logical);
-	}
-	return 0;
+	return total_added;
 }
 
-static int cache_block_group(struct btrfs_root *root,
-			     struct btrfs_block_group_cache *block_group)
+static int caching_kthread(void *data)
 {
+	struct btrfs_block_group_cache *block_group = data;
+	struct btrfs_fs_info *fs_info = block_group->fs_info;
+	struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
+	struct btrfs_root *extent_root = fs_info->extent_root;
 	struct btrfs_path *path;
-	int ret = 0;
-	struct btrfs_key key;
 	struct extent_buffer *leaf;
-	int slot;
-	u64 last;
-
-	if (!block_group)
-		return 0;
-
-	root = root->fs_info->extent_root;
-
-	if (block_group->cached)
-		return 0;
+	struct btrfs_key key;
+	u64 total_found = 0;
+	u64 last = 0;
+	u32 nritems;
+	int ret = 0;
 
 	path = btrfs_alloc_path();
 	if (!path)
 		return -ENOMEM;
 
-	path->reada = 2;
+	exclude_super_stripes(extent_root, block_group);
+	spin_lock(&block_group->space_info->lock);
+	block_group->space_info->bytes_super += block_group->bytes_super;
+	spin_unlock(&block_group->space_info->lock);
+
+	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
+
 	/*
-	 * we get into deadlocks with paths held by callers of this function.
-	 * since the alloc_mutex is protecting things right now, just
-	 * skip the locking here
+	 * We don't want to deadlock with somebody trying to allocate a new
+	 * extent for the extent root while also trying to search the extent
+	 * root to add free space.  So we skip locking and search the commit
+	 * root, since its read-only
 	 */
 	path->skip_locking = 1;
-	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
+	path->search_commit_root = 1;
+	path->reada = 2;
+
 	key.objectid = last;
 	key.offset = 0;
-	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
-	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+again:
+	mutex_lock(&caching_ctl->mutex);
+	/* need to make sure the commit_root doesn't disappear */
+	down_read(&fs_info->extent_commit_sem);
+
+	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 	if (ret < 0)
 		goto err;
 
+	leaf = path->nodes[0];
+	nritems = btrfs_header_nritems(leaf);
+
 	while (1) {
-		leaf = path->nodes[0];
-		slot = path->slots[0];
-		if (slot >= btrfs_header_nritems(leaf)) {
-			ret = btrfs_next_leaf(root, path);
-			if (ret < 0)
-				goto err;
-			if (ret == 0)
-				continue;
-			else
+		smp_mb();
+		if (fs_info->closing > 1) {
+			last = (u64)-1;
+			break;
+		}
+
+		if (path->slots[0] < nritems) {
+			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		} else {
+			ret = find_next_key(path, 0, &key);
+			if (ret)
 				break;
+
+			caching_ctl->progress = last;
+			btrfs_release_path(extent_root, path);
+			up_read(&fs_info->extent_commit_sem);
+			mutex_unlock(&caching_ctl->mutex);
+			if (btrfs_transaction_in_commit(fs_info))
+				schedule_timeout(1);
+			else
+				cond_resched();
+			goto again;
+		}
+
+		if (key.objectid < block_group->key.objectid) {
+			path->slots[0]++;
+			continue;
 		}
-		btrfs_item_key_to_cpu(leaf, &key, slot);
-		if (key.objectid < block_group->key.objectid)
-			goto next;
 
 		if (key.objectid >= block_group->key.objectid +
 		    block_group->key.offset)
 			break;
 
-		if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
-			add_new_free_space(block_group, root->fs_info, last,
-					   key.objectid);
-
+		if (key.type == BTRFS_EXTENT_ITEM_KEY) {
+			total_found += add_new_free_space(block_group,
+							  fs_info, last,
+							  key.objectid);
 			last = key.objectid + key.offset;
+
+			if (total_found > (1024 * 1024 * 2)) {
+				total_found = 0;
+				wake_up(&caching_ctl->wait);
+			}
 		}
-next:
 		path->slots[0]++;
 	}
+	ret = 0;
 
-	add_new_free_space(block_group, root->fs_info, last,
-			   block_group->key.objectid +
-			   block_group->key.offset);
+	total_found += add_new_free_space(block_group, fs_info, last,
+					  block_group->key.objectid +
+					  block_group->key.offset);
+	caching_ctl->progress = (u64)-1;
+
+	spin_lock(&block_group->lock);
+	block_group->caching_ctl = NULL;
+	block_group->cached = BTRFS_CACHE_FINISHED;
+	spin_unlock(&block_group->lock);
 
-	block_group->cached = 1;
-	remove_sb_from_cache(root, block_group);
-	ret = 0;
 err:
 	btrfs_free_path(path);
+	up_read(&fs_info->extent_commit_sem);
+
+	free_excluded_extents(extent_root, block_group);
+
+	mutex_unlock(&caching_ctl->mutex);
+	wake_up(&caching_ctl->wait);
+
+	put_caching_control(caching_ctl);
+	atomic_dec(&block_group->space_info->caching_threads);
+	return 0;
+}
+
+static int cache_block_group(struct btrfs_block_group_cache *cache)
+{
+	struct btrfs_fs_info *fs_info = cache->fs_info;
+	struct btrfs_caching_control *caching_ctl;
+	struct task_struct *tsk;
+	int ret = 0;
+
+	smp_mb();
+	if (cache->cached != BTRFS_CACHE_NO)
+		return 0;
+
+	caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
+	BUG_ON(!caching_ctl);
+
+	INIT_LIST_HEAD(&caching_ctl->list);
+	mutex_init(&caching_ctl->mutex);
+	init_waitqueue_head(&caching_ctl->wait);
+	caching_ctl->block_group = cache;
+	caching_ctl->progress = cache->key.objectid;
+	/* one for caching kthread, one for caching block group list */
+	atomic_set(&caching_ctl->count, 2);
+
+	spin_lock(&cache->lock);
+	if (cache->cached != BTRFS_CACHE_NO) {
+		spin_unlock(&cache->lock);
+		kfree(caching_ctl);
+		return 0;
+	}
+	cache->caching_ctl = caching_ctl;
+	cache->cached = BTRFS_CACHE_STARTED;
+	spin_unlock(&cache->lock);
+
+	down_write(&fs_info->extent_commit_sem);
+	list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
+	up_write(&fs_info->extent_commit_sem);
+
+	atomic_inc(&cache->space_info->caching_threads);
+
+	tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
+			  cache->key.objectid);
+	if (IS_ERR(tsk)) {
+		ret = PTR_ERR(tsk);
+		printk(KERN_ERR "error running thread %d\n", ret);
+		BUG();
+	}
+
 	return ret;
 }
 
@@ -990,15 +1154,13 @@ static inline int extent_ref_type(u64 parent, u64 owner)
 	return type;
 }
 
-static int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
+static int find_next_key(struct btrfs_path *path, int level,
+			 struct btrfs_key *key)
 
 {
-	int level;
-	BUG_ON(!path->keep_locks);
-	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
+	for (; level < BTRFS_MAX_LEVEL; level++) {
 		if (!path->nodes[level])
 			break;
-		btrfs_assert_tree_locked(path->nodes[level]);
 		if (path->slots[level] + 1 >=
 		    btrfs_header_nritems(path->nodes[level]))
 			continue;
@@ -1158,7 +1320,8 @@ int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
 		 * For simplicity, we just do not add new inline back
 		 * ref if there is any kind of item for this block
 		 */
-		if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
+		if (find_next_key(path, 0, &key) == 0 &&
+		    key.objectid == bytenr &&
 		    key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
 			err = -EAGAIN;
 			goto out;
@@ -1409,7 +1572,8 @@ static int remove_extent_backref(struct btrfs_trans_handle *trans,
 static void btrfs_issue_discard(struct block_device *bdev,
 				u64 start, u64 len)
 {
-	blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
+	blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
+			     DISCARD_FL_BARRIER);
 }
 #endif
 
@@ -1554,7 +1718,6 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 						 parent, ref_root, flags,
 						 ref->objectid, ref->offset,
 						 &ins, node->ref_mod);
-		update_reserved_extents(root, ins.objectid, ins.offset, 0);
 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
 		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
 					     node->num_bytes, parent,
@@ -1680,7 +1843,6 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 						extent_op->flags_to_set,
 						&extent_op->key,
 						ref->level, &ins);
-		update_reserved_extents(root, ins.objectid, ins.offset, 0);
 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
 		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
 					     node->num_bytes, parent, ref_root,
@@ -1715,16 +1877,32 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 		BUG_ON(extent_op);
 		head = btrfs_delayed_node_to_head(node);
 		if (insert_reserved) {
+			int mark_free = 0;
+			struct extent_buffer *must_clean = NULL;
+
+			ret = pin_down_bytes(trans, root, NULL,
+					     node->bytenr, node->num_bytes,
+					     head->is_data, 1, &must_clean);
+			if (ret > 0)
+				mark_free = 1;
+
+			if (must_clean) {
+				clean_tree_block(NULL, root, must_clean);
+				btrfs_tree_unlock(must_clean);
+				free_extent_buffer(must_clean);
+			}
 			if (head->is_data) {
 				ret = btrfs_del_csums(trans, root,
 						      node->bytenr,
 						      node->num_bytes);
 				BUG_ON(ret);
 			}
-			btrfs_update_pinned_extents(root, node->bytenr,
-						    node->num_bytes, 1);
-			update_reserved_extents(root, node->bytenr,
-						node->num_bytes, 0);
+			if (mark_free) {
+				ret = btrfs_free_reserved_extent(root,
+							node->bytenr,
+							node->num_bytes);
+				BUG_ON(ret);
+			}
 		}
 		mutex_unlock(&head->mutex);
 		return 0;
@@ -2388,13 +2566,29 @@ fail:
 
 }
 
+static struct btrfs_block_group_cache *
+next_block_group(struct btrfs_root *root,
+		 struct btrfs_block_group_cache *cache)
+{
+	struct rb_node *node;
+	spin_lock(&root->fs_info->block_group_cache_lock);
+	node = rb_next(&cache->cache_node);
+	btrfs_put_block_group(cache);
+	if (node) {
+		cache = rb_entry(node, struct btrfs_block_group_cache,
+				 cache_node);
+		atomic_inc(&cache->count);
+	} else
+		cache = NULL;
+	spin_unlock(&root->fs_info->block_group_cache_lock);
+	return cache;
+}
+
 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 				   struct btrfs_root *root)
 {
-	struct btrfs_block_group_cache *cache, *entry;
-	struct rb_node *n;
+	struct btrfs_block_group_cache *cache;
 	int err = 0;
-	int werr = 0;
 	struct btrfs_path *path;
 	u64 last = 0;
 
@@ -2403,39 +2597,35 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 
 	while (1) {
-		cache = NULL;
-		spin_lock(&root->fs_info->block_group_cache_lock);
-		for (n = rb_first(&root->fs_info->block_group_cache_tree);
-		     n; n = rb_next(n)) {
-			entry = rb_entry(n, struct btrfs_block_group_cache,
-					 cache_node);
-			if (entry->dirty) {
-				cache = entry;
-				break;
-			}
+		if (last == 0) {
+			err = btrfs_run_delayed_refs(trans, root,
+						     (unsigned long)-1);
+			BUG_ON(err);
 		}
-		spin_unlock(&root->fs_info->block_group_cache_lock);
 
-		if (!cache)
-			break;
+		cache = btrfs_lookup_first_block_group(root->fs_info, last);
+		while (cache) {
+			if (cache->dirty)
+				break;
+			cache = next_block_group(root, cache);
+		}
+		if (!cache) {
+			if (last == 0)
+				break;
+			last = 0;
+			continue;
+		}
 
 		cache->dirty = 0;
-		last += cache->key.offset;
+		last = cache->key.objectid + cache->key.offset;
 
-		err = write_one_cache_group(trans, root,
-					    path, cache);
-		/*
-		 * if we fail to write the cache group, we want
-		 * to keep it marked dirty in hopes that a later
-		 * write will work
-		 */
-		if (err) {
-			werr = err;
-			continue;
-		}
+		err = write_one_cache_group(trans, root, path, cache);
+		BUG_ON(err);
+		btrfs_put_block_group(cache);
 	}
+
 	btrfs_free_path(path);
-	return werr;
+	return 0;
 }
 
 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
@@ -2485,6 +2675,7 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
 	found->force_alloc = 0;
 	*space_info = found;
 	list_add_rcu(&found->list, &info->space_info);
+	atomic_set(&found->caching_threads, 0);
 	return 0;
 }
 
@@ -2576,60 +2767,346 @@ void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
 						       alloc_target);
 }
 
+static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
+{
+	u64 num_bytes;
+	int level;
+
+	level = BTRFS_MAX_LEVEL - 2;
+	/*
+	 * NOTE: these calculations are absolutely the worst possible case.
+	 * This assumes that _every_ item we insert will require a new leaf, and
+	 * that the tree has grown to its maximum level size.
+	 */
+
+	/*
+	 * for every item we insert we could insert both an extent item and a
+	 * extent ref item.  Then for ever item we insert, we will need to cow
+	 * both the original leaf, plus the leaf to the left and right of it.
+	 *
+	 * Unless we are talking about the extent root, then we just want the
+	 * number of items * 2, since we just need the extent item plus its ref.
+	 */
+	if (root == root->fs_info->extent_root)
+		num_bytes = num_items * 2;
+	else
+		num_bytes = (num_items + (2 * num_items)) * 3;
+
+	/*
+	 * num_bytes is total number of leaves we could need times the leaf
+	 * size, and then for every leaf we could end up cow'ing 2 nodes per
+	 * level, down to the leaf level.
+	 */
+	num_bytes = (num_bytes * root->leafsize) +
+		(num_bytes * (level * 2)) * root->nodesize;
+
+	return num_bytes;
+}
+
 /*
- * for now this just makes sure we have at least 5% of our metadata space free
- * for use.
+ * Unreserve metadata space for delalloc.  If we have less reserved credits than
+ * we have extents, this function does nothing.
  */
-int btrfs_check_metadata_free_space(struct btrfs_root *root)
+int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
+					  struct inode *inode, int num_items)
 {
 	struct btrfs_fs_info *info = root->fs_info;
 	struct btrfs_space_info *meta_sinfo;
-	u64 alloc_target, thresh;
-	int committed = 0, ret;
+	u64 num_bytes;
+	u64 alloc_target;
+	bool bug = false;
 
 	/* get the space info for where the metadata will live */
 	alloc_target = btrfs_get_alloc_profile(root, 0);
 	meta_sinfo = __find_space_info(info, alloc_target);
 
-again:
+	num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
+					   num_items);
+
 	spin_lock(&meta_sinfo->lock);
-	if (!meta_sinfo->full)
-		thresh = meta_sinfo->total_bytes * 80;
-	else
-		thresh = meta_sinfo->total_bytes * 95;
+	if (BTRFS_I(inode)->delalloc_reserved_extents <=
+	    BTRFS_I(inode)->delalloc_extents) {
+		spin_unlock(&meta_sinfo->lock);
+		return 0;
+	}
+
+	BTRFS_I(inode)->delalloc_reserved_extents--;
+	BUG_ON(BTRFS_I(inode)->delalloc_reserved_extents < 0);
+
+	if (meta_sinfo->bytes_delalloc < num_bytes) {
+		bug = true;
+		meta_sinfo->bytes_delalloc = 0;
+	} else {
+		meta_sinfo->bytes_delalloc -= num_bytes;
+	}
+	spin_unlock(&meta_sinfo->lock);
+
+	BUG_ON(bug);
+
+	return 0;
+}
+
+static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
+{
+	u64 thresh;
+
+	thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+		meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+		meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+		meta_sinfo->bytes_may_use;
 
+	thresh = meta_sinfo->total_bytes - thresh;
+	thresh *= 80;
 	do_div(thresh, 100);
+	if (thresh <= meta_sinfo->bytes_delalloc)
+		meta_sinfo->force_delalloc = 1;
+	else
+		meta_sinfo->force_delalloc = 0;
+}
 
-	if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
-	    meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
-		struct btrfs_trans_handle *trans;
-		if (!meta_sinfo->full) {
-			meta_sinfo->force_alloc = 1;
-			spin_unlock(&meta_sinfo->lock);
+static int maybe_allocate_chunk(struct btrfs_root *root,
+				 struct btrfs_space_info *info)
+{
+	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
+	struct btrfs_trans_handle *trans;
+	bool wait = false;
+	int ret = 0;
+	u64 min_metadata;
+	u64 free_space;
 
-			trans = btrfs_start_transaction(root, 1);
-			if (!trans)
-				return -ENOMEM;
+	free_space = btrfs_super_total_bytes(disk_super);
+	/*
+	 * we allow the metadata to grow to a max of either 5gb or 5% of the
+	 * space in the volume.
+	 */
+	min_metadata = min((u64)5 * 1024 * 1024 * 1024,
+			     div64_u64(free_space * 5, 100));
+	if (info->total_bytes >= min_metadata) {
+		spin_unlock(&info->lock);
+		return 0;
+	}
 
-			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
-					     2 * 1024 * 1024, alloc_target, 0);
-			btrfs_end_transaction(trans, root);
+	if (info->full) {
+		spin_unlock(&info->lock);
+		return 0;
+	}
+
+	if (!info->allocating_chunk) {
+		info->force_alloc = 1;
+		info->allocating_chunk = 1;
+		init_waitqueue_head(&info->wait);
+	} else {
+		wait = true;
+	}
+
+	spin_unlock(&info->lock);
+
+	if (wait) {
+		wait_event(info->wait,
+			   !info->allocating_chunk);
+		return 1;
+	}
+
+	trans = btrfs_start_transaction(root, 1);
+	if (!trans) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+			     4096 + 2 * 1024 * 1024,
+			     info->flags, 0);
+	btrfs_end_transaction(trans, root);
+	if (ret)
+		goto out;
+out:
+	spin_lock(&info->lock);
+	info->allocating_chunk = 0;
+	spin_unlock(&info->lock);
+	wake_up(&info->wait);
+
+	if (ret)
+		return 0;
+	return 1;
+}
+
+/*
+ * Reserve metadata space for delalloc.
+ */
+int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
+					struct inode *inode, int num_items)
+{
+	struct btrfs_fs_info *info = root->fs_info;
+	struct btrfs_space_info *meta_sinfo;
+	u64 num_bytes;
+	u64 used;
+	u64 alloc_target;
+	int flushed = 0;
+	int force_delalloc;
+
+	/* get the space info for where the metadata will live */
+	alloc_target = btrfs_get_alloc_profile(root, 0);
+	meta_sinfo = __find_space_info(info, alloc_target);
+
+	num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
+					   num_items);
+again:
+	spin_lock(&meta_sinfo->lock);
+
+	force_delalloc = meta_sinfo->force_delalloc;
+
+	if (unlikely(!meta_sinfo->bytes_root))
+		meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
+
+	if (!flushed)
+		meta_sinfo->bytes_delalloc += num_bytes;
+
+	used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+		meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+		meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+		meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
+
+	if (used > meta_sinfo->total_bytes) {
+		flushed++;
+
+		if (flushed == 1) {
+			if (maybe_allocate_chunk(root, meta_sinfo))
+				goto again;
+			flushed++;
+		} else {
+			spin_unlock(&meta_sinfo->lock);
+		}
+
+		if (flushed == 2) {
+			filemap_flush(inode->i_mapping);
+			goto again;
+		} else if (flushed == 3) {
+			btrfs_start_delalloc_inodes(root);
+			btrfs_wait_ordered_extents(root, 0);
 			goto again;
 		}
+		spin_lock(&meta_sinfo->lock);
+		meta_sinfo->bytes_delalloc -= num_bytes;
 		spin_unlock(&meta_sinfo->lock);
+		printk(KERN_ERR "enospc, has %d, reserved %d\n",
+		       BTRFS_I(inode)->delalloc_extents,
+		       BTRFS_I(inode)->delalloc_reserved_extents);
+		dump_space_info(meta_sinfo, 0, 0);
+		return -ENOSPC;
+	}
 
-		if (!committed) {
-			committed = 1;
-			trans = btrfs_join_transaction(root, 1);
-			if (!trans)
-				return -ENOMEM;
-			ret = btrfs_commit_transaction(trans, root);
-			if (ret)
-				return ret;
+	BTRFS_I(inode)->delalloc_reserved_extents++;
+	check_force_delalloc(meta_sinfo);
+	spin_unlock(&meta_sinfo->lock);
+
+	if (!flushed && force_delalloc)
+		filemap_flush(inode->i_mapping);
+
+	return 0;
+}
+
+/*
+ * unreserve num_items number of items worth of metadata space.  This needs to
+ * be paired with btrfs_reserve_metadata_space.
+ *
+ * NOTE: if you have the option, run this _AFTER_ you do a
+ * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
+ * oprations which will result in more used metadata, so we want to make sure we
+ * can do that without issue.
+ */
+int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
+{
+	struct btrfs_fs_info *info = root->fs_info;
+	struct btrfs_space_info *meta_sinfo;
+	u64 num_bytes;
+	u64 alloc_target;
+	bool bug = false;
+
+	/* get the space info for where the metadata will live */
+	alloc_target = btrfs_get_alloc_profile(root, 0);
+	meta_sinfo = __find_space_info(info, alloc_target);
+
+	num_bytes = calculate_bytes_needed(root, num_items);
+
+	spin_lock(&meta_sinfo->lock);
+	if (meta_sinfo->bytes_may_use < num_bytes) {
+		bug = true;
+		meta_sinfo->bytes_may_use = 0;
+	} else {
+		meta_sinfo->bytes_may_use -= num_bytes;
+	}
+	spin_unlock(&meta_sinfo->lock);
+
+	BUG_ON(bug);
+
+	return 0;
+}
+
+/*
+ * Reserve some metadata space for use.  We'll calculate the worste case number
+ * of bytes that would be needed to modify num_items number of items.  If we
+ * have space, fantastic, if not, you get -ENOSPC.  Please call
+ * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
+ * items you reserved, since whatever metadata you needed should have already
+ * been allocated.
+ *
+ * This will commit the transaction to make more space if we don't have enough
+ * metadata space.  THe only time we don't do this is if we're reserving space
+ * inside of a transaction, then we will just return -ENOSPC and it is the
+ * callers responsibility to handle it properly.
+ */
+int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
+{
+	struct btrfs_fs_info *info = root->fs_info;
+	struct btrfs_space_info *meta_sinfo;
+	u64 num_bytes;
+	u64 used;
+	u64 alloc_target;
+	int retries = 0;
+
+	/* get the space info for where the metadata will live */
+	alloc_target = btrfs_get_alloc_profile(root, 0);
+	meta_sinfo = __find_space_info(info, alloc_target);
+
+	num_bytes = calculate_bytes_needed(root, num_items);
+again:
+	spin_lock(&meta_sinfo->lock);
+
+	if (unlikely(!meta_sinfo->bytes_root))
+		meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
+
+	if (!retries)
+		meta_sinfo->bytes_may_use += num_bytes;
+
+	used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+		meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+		meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+		meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
+
+	if (used > meta_sinfo->total_bytes) {
+		retries++;
+		if (retries == 1) {
+			if (maybe_allocate_chunk(root, meta_sinfo))
+				goto again;
+			retries++;
+		} else {
+			spin_unlock(&meta_sinfo->lock);
+		}
+
+		if (retries == 2) {
+			btrfs_start_delalloc_inodes(root);
+			btrfs_wait_ordered_extents(root, 0);
 			goto again;
 		}
+		spin_lock(&meta_sinfo->lock);
+		meta_sinfo->bytes_may_use -= num_bytes;
+		spin_unlock(&meta_sinfo->lock);
+
+		dump_space_info(meta_sinfo, 0, 0);
 		return -ENOSPC;
 	}
+
+	check_force_delalloc(meta_sinfo);
 	spin_unlock(&meta_sinfo->lock);
 
 	return 0;
@@ -2649,13 +3126,16 @@ int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
 	bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 
 	data_sinfo = BTRFS_I(inode)->space_info;
+	if (!data_sinfo)
+		goto alloc;
+
 again:
 	/* make sure we have enough space to handle the data first */
 	spin_lock(&data_sinfo->lock);
 	if (data_sinfo->total_bytes - data_sinfo->bytes_used -
 	    data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
 	    data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
-	    data_sinfo->bytes_may_use < bytes) {
+	    data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
 		struct btrfs_trans_handle *trans;
 
 		/*
@@ -2667,7 +3147,7 @@ again:
 
 			data_sinfo->force_alloc = 1;
 			spin_unlock(&data_sinfo->lock);
-
+alloc:
 			alloc_target = btrfs_get_alloc_profile(root, 1);
 			trans = btrfs_start_transaction(root, 1);
 			if (!trans)
@@ -2679,12 +3159,17 @@ again:
 			btrfs_end_transaction(trans, root);
 			if (ret)
 				return ret;
+
+			if (!data_sinfo) {
+				btrfs_set_inode_space_info(root, inode);
+				data_sinfo = BTRFS_I(inode)->space_info;
+			}
 			goto again;
 		}
 		spin_unlock(&data_sinfo->lock);
 
 		/* commit the current transaction and try again */
-		if (!committed) {
+		if (!committed && !root->fs_info->open_ioctl_trans) {
 			committed = 1;
 			trans = btrfs_join_transaction(root, 1);
 			if (!trans)
@@ -2697,7 +3182,7 @@ again:
 
 		printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
 		       ", %llu bytes_used, %llu bytes_reserved, "
-		       "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
+		       "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
 		       "%llu total\n", (unsigned long long)bytes,
 		       (unsigned long long)data_sinfo->bytes_delalloc,
 		       (unsigned long long)data_sinfo->bytes_used,
@@ -2712,7 +3197,7 @@ again:
 	BTRFS_I(inode)->reserved_bytes += bytes;
 	spin_unlock(&data_sinfo->lock);
 
-	return btrfs_check_metadata_free_space(root);
+	return 0;
 }
 
 /*
@@ -2811,17 +3296,15 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 	BUG_ON(!space_info);
 
 	spin_lock(&space_info->lock);
-	if (space_info->force_alloc) {
+	if (space_info->force_alloc)
 		force = 1;
-		space_info->force_alloc = 0;
-	}
 	if (space_info->full) {
 		spin_unlock(&space_info->lock);
 		goto out;
 	}
 
 	thresh = space_info->total_bytes - space_info->bytes_readonly;
-	thresh = div_factor(thresh, 6);
+	thresh = div_factor(thresh, 8);
 	if (!force &&
 	   (space_info->bytes_used + space_info->bytes_pinned +
 	    space_info->bytes_reserved + alloc_bytes) < thresh) {
@@ -2835,7 +3318,7 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 	 * we keep a reasonable number of metadata chunks allocated in the
 	 * FS as well.
 	 */
-	if (flags & BTRFS_BLOCK_GROUP_DATA) {
+	if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
 		fs_info->data_chunk_allocations++;
 		if (!(fs_info->data_chunk_allocations %
 		      fs_info->metadata_ratio))
@@ -2843,8 +3326,11 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 	}
 
 	ret = btrfs_alloc_chunk(trans, extent_root, flags);
+	spin_lock(&space_info->lock);
 	if (ret)
 		space_info->full = 1;
+	space_info->force_alloc = 0;
+	spin_unlock(&space_info->lock);
 out:
 	mutex_unlock(&extent_root->fs_info->chunk_mutex);
 	return ret;
@@ -2893,10 +3379,12 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 		num_bytes = min(total, cache->key.offset - byte_in_group);
 		if (alloc) {
 			old_val += num_bytes;
+			btrfs_set_block_group_used(&cache->item, old_val);
+			cache->reserved -= num_bytes;
 			cache->space_info->bytes_used += num_bytes;
+			cache->space_info->bytes_reserved -= num_bytes;
 			if (cache->ro)
 				cache->space_info->bytes_readonly -= num_bytes;
-			btrfs_set_block_group_used(&cache->item, old_val);
 			spin_unlock(&cache->lock);
 			spin_unlock(&cache->space_info->lock);
 		} else {
@@ -2941,110 +3429,136 @@ static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
 	return bytenr;
 }
 
-int btrfs_update_pinned_extents(struct btrfs_root *root,
-				u64 bytenr, u64 num, int pin)
+/*
+ * this function must be called within transaction
+ */
+int btrfs_pin_extent(struct btrfs_root *root,
+		     u64 bytenr, u64 num_bytes, int reserved)
 {
-	u64 len;
-	struct btrfs_block_group_cache *cache;
 	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_block_group_cache *cache;
+
+	cache = btrfs_lookup_block_group(fs_info, bytenr);
+	BUG_ON(!cache);
+
+	spin_lock(&cache->space_info->lock);
+	spin_lock(&cache->lock);
+	cache->pinned += num_bytes;
+	cache->space_info->bytes_pinned += num_bytes;
+	if (reserved) {
+		cache->reserved -= num_bytes;
+		cache->space_info->bytes_reserved -= num_bytes;
+	}
+	spin_unlock(&cache->lock);
+	spin_unlock(&cache->space_info->lock);
+
+	btrfs_put_block_group(cache);
+
+	set_extent_dirty(fs_info->pinned_extents,
+			 bytenr, bytenr + num_bytes - 1, GFP_NOFS);