1 files changed, 92 insertions, 16 deletions

diff --git a/fs/btrfs/delayed-inode.c b/fs/btrfs/delayed-inode.c
index b52c672f4c1..c7ddf8a01c5 100644
--- a/fs/btrfs/delayed-inode.c
+++ b/fs/btrfs/delayed-inode.c
@@ -591,7 +591,7 @@ static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 		return 0;
 
 	src_rsv = trans->block_rsv;
-	dst_rsv = &root->fs_info->global_block_rsv;
+	dst_rsv = &root->fs_info->delayed_block_rsv;
 
 	num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
@@ -609,7 +609,7 @@ static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 	if (!item->bytes_reserved)
 		return;
 
-	rsv = &root->fs_info->global_block_rsv;
+	rsv = &root->fs_info->delayed_block_rsv;
 	btrfs_block_rsv_release(root, rsv,
 				item->bytes_reserved);
 }
@@ -617,24 +617,102 @@ static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 static int btrfs_delayed_inode_reserve_metadata(
 					struct btrfs_trans_handle *trans,
 					struct btrfs_root *root,
+					struct inode *inode,
 					struct btrfs_delayed_node *node)
 {
 	struct btrfs_block_rsv *src_rsv;
 	struct btrfs_block_rsv *dst_rsv;
 	u64 num_bytes;
 	int ret;
-
-	if (!trans->bytes_reserved)
-		return 0;
+	int release = false;
 
 	src_rsv = trans->block_rsv;
-	dst_rsv = &root->fs_info->global_block_rsv;
+	dst_rsv = &root->fs_info->delayed_block_rsv;
 
 	num_bytes = btrfs_calc_trans_metadata_size(root, 1);
+
+	/*
+	 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
+	 * which doesn't reserve space for speed.  This is a problem since we
+	 * still need to reserve space for this update, so try to reserve the
+	 * space.
+	 *
+	 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
+	 * we're accounted for.
+	 */
+	if (!src_rsv || (!trans->bytes_reserved &&
+	    src_rsv != &root->fs_info->delalloc_block_rsv)) {
+		ret = btrfs_block_rsv_add_noflush(root, dst_rsv, num_bytes);
+		/*
+		 * Since we're under a transaction reserve_metadata_bytes could
+		 * try to commit the transaction which will make it return
+		 * EAGAIN to make us stop the transaction we have, so return
+		 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
+		 */
+		if (ret == -EAGAIN)
+			ret = -ENOSPC;
+		if (!ret)
+			node->bytes_reserved = num_bytes;
+		return ret;
+	} else if (src_rsv == &root->fs_info->delalloc_block_rsv) {
+		spin_lock(&BTRFS_I(inode)->lock);
+		if (BTRFS_I(inode)->delalloc_meta_reserved) {
+			BTRFS_I(inode)->delalloc_meta_reserved = 0;
+			spin_unlock(&BTRFS_I(inode)->lock);
+			release = true;
+			goto migrate;
+		}
+		spin_unlock(&BTRFS_I(inode)->lock);
+
+		/* Ok we didn't have space pre-reserved.  This shouldn't happen
+		 * too often but it can happen if we do delalloc to an existing
+		 * inode which gets dirtied because of the time update, and then
+		 * isn't touched again until after the transaction commits and
+		 * then we try to write out the data.  First try to be nice and
+		 * reserve something strictly for us.  If not be a pain and try
+		 * to steal from the delalloc block rsv.
+		 */
+		ret = btrfs_block_rsv_add_noflush(root, dst_rsv, num_bytes);
+		if (!ret)
+			goto out;
+
+		ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
+		if (!ret)
+			goto out;
+
+		/*
+		 * Ok this is a problem, let's just steal from the global rsv
+		 * since this really shouldn't happen that often.
+		 */
+		WARN_ON(1);
+		ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
+					      dst_rsv, num_bytes);
+		goto out;
+	}
+
+migrate:
 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
+
+out:
+	/*
+	 * Migrate only takes a reservation, it doesn't touch the size of the
+	 * block_rsv.  This is to simplify people who don't normally have things
+	 * migrated from their block rsv.  If they go to release their
+	 * reservation, that will decrease the size as well, so if migrate
+	 * reduced size we'd end up with a negative size.  But for the
+	 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
+	 * but we could in fact do this reserve/migrate dance several times
+	 * between the time we did the original reservation and we'd clean it
+	 * up.  So to take care of this, release the space for the meta
+	 * reservation here.  I think it may be time for a documentation page on
+	 * how block rsvs. work.
+	 */
 	if (!ret)
 		node->bytes_reserved = num_bytes;
 
+	if (release)
+		btrfs_block_rsv_release(root, src_rsv, num_bytes);
+
 	return ret;
 }
 
@@ -646,7 +724,7 @@ static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
 	if (!node->bytes_reserved)
 		return;
 
-	rsv = &root->fs_info->global_block_rsv;
+	rsv = &root->fs_info->delayed_block_rsv;
 	btrfs_block_rsv_release(root, rsv,
 				node->bytes_reserved);
 	node->bytes_reserved = 0;
@@ -1026,7 +1104,7 @@ int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
 	path->leave_spinning = 1;
 
 	block_rsv = trans->block_rsv;
-	trans->block_rsv = &root->fs_info->global_block_rsv;
+	trans->block_rsv = &root->fs_info->delayed_block_rsv;
 
 	delayed_root = btrfs_get_delayed_root(root);
 
@@ -1069,7 +1147,7 @@ static int __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
 	path->leave_spinning = 1;
 
 	block_rsv = trans->block_rsv;
-	trans->block_rsv = &node->root->fs_info->global_block_rsv;
+	trans->block_rsv = &node->root->fs_info->delayed_block_rsv;
 
 	ret = btrfs_insert_delayed_items(trans, path, node->root, node);
 	if (!ret)
@@ -1149,7 +1227,7 @@ static void btrfs_async_run_delayed_node_done(struct btrfs_work *work)
 		goto free_path;
 
 	block_rsv = trans->block_rsv;
-	trans->block_rsv = &root->fs_info->global_block_rsv;
+	trans->block_rsv = &root->fs_info->delayed_block_rsv;
 
 	ret = btrfs_insert_delayed_items(trans, path, root, delayed_node);
 	if (!ret)
@@ -1685,12 +1763,10 @@ int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
 		goto release_node;
 	}
 
-	ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
-	/*
-	 * we must reserve enough space when we start a new transaction,
-	 * so reserving metadata failure is impossible
-	 */
-	BUG_ON(ret);
+	ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
+						   delayed_node);
+	if (ret)
+		goto release_node;
 
 	fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
 	delayed_node->inode_dirty = 1;