5 files changed, 510 insertions, 101 deletions

diff --git a/btrfs-corrupt-block.c b/btrfs-corrupt-block.c
index f0c14a9d..10cae00a 100644
--- a/btrfs-corrupt-block.c
+++ b/btrfs-corrupt-block.c
@@ -105,6 +105,8 @@ static void print_usage(void)
 		"specify -i for the inode and -f for the field to corrupt)\n");
 	fprintf(stderr, "\t-m The metadata block to corrupt (must also "
 		"specify -f for the field to corrupt)\n");
+	fprintf(stderr, "\t-K The key to corrupt in the format "
+		"<num>,<num>,<num> (must also specify -f for the field)\n");
 	fprintf(stderr, "\t-f The field in the item to corrupt\n");
 	exit(1);
 }
@@ -306,6 +308,13 @@ enum btrfs_metadata_block_field {
 	BTRFS_METADATA_BLOCK_BAD,
 };
 
+enum btrfs_key_field {
+	BTRFS_KEY_OBJECTID,
+	BTRFS_KEY_TYPE,
+	BTRFS_KEY_OFFSET,
+	BTRFS_KEY_BAD,
+};
+
 static enum btrfs_inode_field convert_inode_field(char *field)
 {
 	if (!strncmp(field, "isize", FIELD_BUF_LEN))
@@ -328,6 +337,17 @@ convert_metadata_block_field(char *field)
 	return BTRFS_METADATA_BLOCK_BAD;
 }
 
+static enum btrfs_key_field convert_key_field(char *field)
+{
+	if (!strncmp(field, "objectid", FIELD_BUF_LEN))
+		return BTRFS_KEY_OBJECTID;
+	if (!strncmp(field, "type", FIELD_BUF_LEN))
+		return BTRFS_KEY_TYPE;
+	if (!strncmp(field, "offset", FIELD_BUF_LEN))
+		return BTRFS_KEY_OFFSET;
+	return BTRFS_KEY_BAD;
+}
+
 static u64 generate_u64(u64 orig)
 {
 	u64 ret;
@@ -337,6 +357,73 @@ static u64 generate_u64(u64 orig)
 	return ret;
 }
 
+static u8 generate_u8(u8 orig)
+{
+	u8 ret;
+	do {
+		ret = rand();
+	} while (ret == orig);
+	return ret;
+}
+
+static int corrupt_key(struct btrfs_root *root, struct btrfs_key *key,
+		       char *field)
+{
+	enum btrfs_key_field corrupt_field = convert_key_field(field);
+	struct btrfs_path *path;
+	struct btrfs_trans_handle *trans;
+	int ret;
+
+	root = root->fs_info->fs_root;
+	if (corrupt_field == BTRFS_KEY_BAD) {
+		fprintf(stderr, "Invalid field %s\n", field);
+		return -EINVAL;
+	}
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	trans = btrfs_start_transaction(root, 1);
+	if (IS_ERR(trans)) {
+		btrfs_free_path(path);
+		return PTR_ERR(trans);
+	}
+
+	ret = btrfs_search_slot(trans, root, key, path, 0, 1);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		fprintf(stderr, "Couldn't find the key to corrupt\n");
+		ret = -ENOENT;
+		goto out;
+	}
+
+	switch (corrupt_field) {
+	case BTRFS_KEY_OBJECTID:
+		key->objectid = generate_u64(key->objectid);
+		break;
+	case BTRFS_KEY_TYPE:
+		key->type = generate_u8(key->type);
+		break;
+	case BTRFS_KEY_OFFSET:
+		key->offset = generate_u64(key->objectid);
+		break;
+	default:
+		fprintf(stderr, "Invalid field %s, %d\n", field,
+			corrupt_field);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	btrfs_set_item_key_unsafe(root, path, key);
+out:
+	btrfs_free_path(path);
+	btrfs_commit_transaction(trans, root);
+	return ret;
+}
+
+
 static int corrupt_inode(struct btrfs_trans_handle *trans,
 			 struct btrfs_root *root, u64 inode, char *field)
 {
@@ -548,6 +635,7 @@ static struct option long_options[] = {
 	{ "file-extent", 1, NULL, 'x'},
 	{ "metadata-block", 1, NULL, 'm'},
 	{ "field", 1, NULL, 'f'},
+	{ "key", 1, NULL, 'K'},
 	{ 0, 0, 0, 0}
 };
 
@@ -696,6 +784,7 @@ out:
 int main(int ac, char **av)
 {
 	struct cache_tree root_cache;
+	struct btrfs_key key;
 	struct btrfs_root *root;
 	struct extent_buffer *eb;
 	char *dev;
@@ -717,10 +806,11 @@ int main(int ac, char **av)
 
 	field[0] = '\0';
 	srand(128);
+	memset(&key, 0, sizeof(key));
 
 	while(1) {
 		int c;
-		c = getopt_long(ac, av, "l:c:b:eEkuUi:f:x:m:", long_options,
+		c = getopt_long(ac, av, "l:c:b:eEkuUi:f:x:m:K:", long_options,
 				&option_index);
 		if (c < 0)
 			break;
@@ -787,6 +877,17 @@ int main(int ac, char **av)
 					print_usage();
 				}
 				break;
+			case 'K':
+				ret = sscanf(optarg, "%llu,%u,%llu",
+					     &key.objectid,
+					     (unsigned int *)&key.type,
+					     &key.offset);
+				if (ret != 3) {
+					fprintf(stderr, "error reading key "
+						"%d\n", errno);
+					print_usage();
+				}
+				break;
 			default:
 				print_usage();
 		}
@@ -882,6 +983,12 @@ int main(int ac, char **av)
 		ret = corrupt_metadata_block(root, metadata_block, field);
 		goto out_close;
 	}
+	if (key.objectid || key.offset || key.type) {
+		if (!strlen(field))
+			print_usage();
+		ret = corrupt_key(root, &key, field);
+		goto out_close;
+	}
 	/*
 	 * If we made it here and we have extent set then we didn't specify
 	 * inode and we're screwed.
diff --git a/cmds-check.c b/cmds-check.c
index cb5ed3e8..c658879c 100644
--- a/cmds-check.c
+++ b/cmds-check.c
@@ -49,6 +49,7 @@ static u64 data_bytes_allocated = 0;
 static u64 data_bytes_referenced = 0;
 static int found_old_backref = 0;
 static LIST_HEAD(duplicate_extents);
+static LIST_HEAD(delete_items);
 static int repair = 0;
 
 struct extent_backref {
@@ -229,6 +230,12 @@ struct walk_control {
 	int root_level;
 };
 
+struct bad_item {
+	struct btrfs_key key;
+	u64 root_id;
+	struct list_head list;
+};
+
 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
 
 static u8 imode_to_type(u32 imode)
@@ -1225,6 +1232,8 @@ static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
 
 		if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
 			continue;
+		if (key.type == BTRFS_ORPHAN_ITEM_KEY)
+			continue;
 
 		if (active_node->current == NULL ||
 		    active_node->current->ino < key.objectid) {
@@ -1418,28 +1427,14 @@ out:
 	return ret;
 }
 
-static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
+static int repair_inode_isize(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root, struct btrfs_path *path,
+			      struct inode_record *rec)
 {
-	struct btrfs_trans_handle *trans;
-	struct btrfs_path *path;
 	struct btrfs_inode_item *ei;
 	struct btrfs_key key;
 	int ret;
 
-	/* So far we just fix dir isize wrong */
-	if (!(rec->errors & I_ERR_DIR_ISIZE_WRONG))
-		return 1;
-
-	path = btrfs_alloc_path();
-	if (!path)
-		return -ENOMEM;
-
-	trans = btrfs_start_transaction(root, 1);
-	if (IS_ERR(trans)) {
-		btrfs_free_path(path);
-		return PTR_ERR(trans);
-	}
-
 	key.objectid = rec->ino;
 	key.type = BTRFS_INODE_ITEM_KEY;
 	key.offset = (u64)-1;
@@ -1469,6 +1464,53 @@ static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
 	printf("reset isize for dir %Lu root %Lu\n", rec->ino,
 	       root->root_key.objectid);
 out:
+	btrfs_release_path(path);
+	return ret;
+}
+
+static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *root,
+				    struct btrfs_path *path,
+				    struct inode_record *rec)
+{
+	struct btrfs_key key;
+	int ret;
+
+	key.objectid = BTRFS_ORPHAN_OBJECTID;
+	key.type = BTRFS_ORPHAN_ITEM_KEY;
+	key.offset = rec->ino;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+	btrfs_release_path(path);
+	if (!ret)
+		rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
+	return ret;
+}
+
+static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_path *path;
+	int ret = 0;
+
+	/* So far we just fix dir isize wrong */
+	if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG | I_ERR_NO_ORPHAN_ITEM)))
+		return 1;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	trans = btrfs_start_transaction(root, 1);
+	if (IS_ERR(trans)) {
+		btrfs_free_path(path);
+		return PTR_ERR(trans);
+	}
+
+	if (rec->errors & I_ERR_DIR_ISIZE_WRONG)
+		ret = repair_inode_isize(trans, root, path, rec);
+	if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
+		ret = repair_inode_orphan_item(trans, root, path, rec);
 	btrfs_commit_transaction(trans, root);
 	btrfs_free_path(path);
 	return ret;
@@ -2005,8 +2047,8 @@ static int check_fs_roots(struct btrfs_root *root,
 		btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
 		if (key.type == BTRFS_ROOT_ITEM_KEY &&
 		    fs_root_objectid(key.objectid)) {
-			tmp_root = btrfs_read_fs_root_no_cache(root->fs_info,
-							       &key);
+			key.offset = (u64)-1;
+			tmp_root = btrfs_read_fs_root(root->fs_info, &key);
 			if (IS_ERR(tmp_root)) {
 				err = 1;
 				goto next;
@@ -2014,7 +2056,6 @@ static int check_fs_roots(struct btrfs_root *root,
 			ret = check_fs_root(tmp_root, root_cache, &wc);
 			if (ret)
 				err = 1;
-			btrfs_free_fs_root(tmp_root);
 		} else if (key.type == BTRFS_ROOT_REF_KEY ||
 			   key.type == BTRFS_ROOT_BACKREF_KEY) {
 			process_root_ref(leaf, path.slots[0], &key,
@@ -2289,13 +2330,148 @@ static int record_bad_block_io(struct btrfs_fs_info *info,
 	return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
 }
 
-static int check_block(struct btrfs_root *root,
+static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
+		       struct extent_buffer *buf, int slot)
+{
+	if (btrfs_header_level(buf)) {
+		struct btrfs_key_ptr ptr1, ptr2;
+
+		read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
+				   sizeof(struct btrfs_key_ptr));
+		read_extent_buffer(buf, &ptr2,
+				   btrfs_node_key_ptr_offset(slot + 1),
+				   sizeof(struct btrfs_key_ptr));
+		write_extent_buffer(buf, &ptr1,
+				    btrfs_node_key_ptr_offset(slot + 1),
+				    sizeof(struct btrfs_key_ptr));
+		write_extent_buffer(buf, &ptr2,
+				    btrfs_node_key_ptr_offset(slot),
+				    sizeof(struct btrfs_key_ptr));
+		if (slot == 0) {
+			struct btrfs_disk_key key;
+			btrfs_node_key(buf, &key, 0);
+			btrfs_fixup_low_keys(root, path, &key,
+					     btrfs_header_level(buf) + 1);
+		}
+	} else {
+		struct btrfs_item *item1, *item2;
+		struct btrfs_key k1, k2;
+		char *item1_data, *item2_data;
+		u32 item1_offset, item2_offset, item1_size, item2_size;
+
+		item1 = btrfs_item_nr(slot);
+		item2 = btrfs_item_nr(slot + 1);
+		btrfs_item_key_to_cpu(buf, &k1, slot);
+		btrfs_item_key_to_cpu(buf, &k2, slot + 1);
+		item1_offset = btrfs_item_offset(buf, item1);
+		item2_offset = btrfs_item_offset(buf, item2);
+		item1_size = btrfs_item_size(buf, item1);
+		item2_size = btrfs_item_size(buf, item2);
+
+		item1_data = malloc(item1_size);
+		if (!item1_data)
+			return -ENOMEM;
+		item2_data = malloc(item2_size);
+		if (!item2_data) {
+			free(item2_data);
+			return -ENOMEM;
+		}
+
+		read_extent_buffer(buf, item1_data, item1_offset, item1_size);
+		read_extent_buffer(buf, item2_data, item2_offset, item2_size);
+
+		write_extent_buffer(buf, item1_data, item2_offset, item2_size);
+		write_extent_buffer(buf, item2_data, item1_offset, item1_size);
+		free(item1_data);
+		free(item2_data);
+
+		btrfs_set_item_offset(buf, item1, item2_offset);
+		btrfs_set_item_offset(buf, item2, item1_offset);
+		btrfs_set_item_size(buf, item1, item2_size);
+		btrfs_set_item_size(buf, item2, item1_size);
+
+		path->slots[0] = slot;
+		btrfs_set_item_key_unsafe(root, path, &k2);
+		path->slots[0] = slot + 1;
+		btrfs_set_item_key_unsafe(root, path, &k1);
+	}
+	return 0;
+}
+
+/*
+ * Attempt to fix basic block failures.  Currently we only handle bad key
+ * orders, we will cycle through the keys and swap them if necessary.
+ */
+static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct extent_buffer *buf,
+				struct btrfs_disk_key *parent_key,
+				enum btrfs_tree_block_status status)
+{
+	struct btrfs_path *path;
+	struct btrfs_key k1, k2;
+	int i;
+	int ret;
+
+	if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
+		return -EIO;
+
+	k1.objectid = btrfs_header_owner(buf);
+	k1.type = BTRFS_ROOT_ITEM_KEY;
+	k1.offset = (u64)-1;
+
+	root = btrfs_read_fs_root(root->fs_info, &k1);
+	if (IS_ERR(root))
+		return -EIO;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -EIO;
+
+	path->lowest_level = btrfs_header_level(buf);
+	path->skip_check_block = 1;
+	if (btrfs_header_level(buf))
+		btrfs_node_key_to_cpu(buf, &k1, 0);
+	else
+		btrfs_item_key_to_cpu(buf, &k1, 0);
+
+	ret = btrfs_search_slot(trans, root, &k1, path, 0, 1);
+	if (ret) {
+		btrfs_free_path(path);
+		return -EIO;
+	}
+
+	buf = path->nodes[0];
+	for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
+		if (btrfs_header_level(buf)) {
+			btrfs_node_key_to_cpu(buf, &k1, i);
+			btrfs_node_key_to_cpu(buf, &k2, i + 1);
+		} else {
+			btrfs_item_key_to_cpu(buf, &k1, i);
+			btrfs_item_key_to_cpu(buf, &k2, i + 1);
+		}
+		if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
+			continue;
+		ret = swap_values(root, path, buf, i);
+		if (ret)
+			break;
+		btrfs_mark_buffer_dirty(buf);
+		i = 0;
+	}
+
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int check_block(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root,
 		       struct cache_tree *extent_cache,
 		       struct extent_buffer *buf, u64 flags)
 {
 	struct extent_record *rec;
 	struct cache_extent *cache;
 	struct btrfs_key key;
+	enum btrfs_tree_block_status status;
 	int ret = 1;
 	int level;
 
@@ -2318,13 +2494,26 @@ static int check_block(struct btrfs_root *root,
 	rec->info_level = level;
 
 	if (btrfs_is_leaf(buf))
-		ret = btrfs_check_leaf(root, &rec->parent_key, buf);
+		status = btrfs_check_leaf(root, &rec->parent_key, buf);
 	else
-		ret = btrfs_check_node(root, &rec->parent_key, buf);
-
-	if (ret) {
-		fprintf(stderr, "bad block %llu\n",
-			(unsigned long long)buf->start);
+		status = btrfs_check_node(root, &rec->parent_key, buf);
+
+	if (status != BTRFS_TREE_BLOCK_CLEAN) {
+		if (repair)
+			status = try_to_fix_bad_block(trans, root, buf,
+						      &rec->parent_key,
+						      status);
+		if (status != BTRFS_TREE_BLOCK_CLEAN) {
+			ret = -EIO;
+			fprintf(stderr, "bad block %llu\n",
+				(unsigned long long)buf->start);
+		} else {
+			/*
+			 * Signal to callers we need to start the scan over
+			 * again since we'll have cow'ed blocks.
+			 */
+			ret = -EAGAIN;
+		}
 	} else {
 		rec->content_checked = 1;
 		if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
@@ -3622,7 +3811,8 @@ static int is_dropped_key(struct btrfs_key *key,
 	return 0;
 }
 
-static int run_next_block(struct btrfs_root *root,
+static int run_next_block(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
 			  struct block_info *bits,
 			  int bits_nr,
 			  u64 *last,
@@ -3644,7 +3834,7 @@ static int run_next_block(struct btrfs_root *root,
 	u64 owner;
 	u64 flags;
 	u64 ptr;
-	int ret;
+	int ret = 0;
 	int i;
 	int nritems;
 	struct btrfs_key key;
@@ -3717,7 +3907,7 @@ static int run_next_block(struct btrfs_root *root,
 		owner = btrfs_header_owner(buf);
 	}
 
-	ret = check_block(root, extent_cache, buf, flags);
+	ret = check_block(trans, root, extent_cache, buf, flags);
 	if (ret)
 		goto out;
 
@@ -3803,6 +3993,23 @@ static int run_next_block(struct btrfs_root *root,
 					0, root->sectorsize);
 				continue;
 			}
+			if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
+				struct bad_item *bad;
+
+				if (key.objectid == BTRFS_ORPHAN_OBJECTID)
+					continue;
+				if (!owner)
+					continue;
+				bad = malloc(sizeof(struct bad_item));
+				if (!bad)
+					continue;
+				INIT_LIST_HEAD(&bad->list);
+				memcpy(&bad->key, &key,
+				       sizeof(struct btrfs_key));
+				bad->root_id = owner;
+				list_add_tail(&bad->list, &delete_items);
+				continue;
+			}
 			if (key.type != BTRFS_EXTENT_DATA_KEY)
 				continue;
 			fi = btrfs_item_ptr(buf, i,
@@ -3877,7 +4084,7 @@ static int run_next_block(struct btrfs_root *root,
 		found_old_backref = 1;
 out:
 	free_extent_buffer(buf);
-	return 0;
+	return ret;
 }
 
 static int add_root_to_pending(struct extent_buffer *buf,
@@ -5571,9 +5778,9 @@ again:
 	}
 	btrfs_release_path(&path);
 	while (1) {
-		ret = run_next_block(root, bits, bits_nr, &last, &pending,
-				     &seen, &reada, &nodes, &extent_cache,
-				     &chunk_cache, &dev_cache,
+		ret = run_next_block(trans, root, bits, bits_nr, &last,
+				     &pending, &seen, &reada, &nodes,
+				     &extent_cache, &chunk_cache, &dev_cache,
 				     &block_group_cache, &dev_extent_cache,
 				     NULL);
 		if (ret != 0)
@@ -5597,7 +5804,7 @@ again:
 		add_root_to_pending(buf, &extent_cache, &pending,
 				    &seen, &nodes, &rec->found_key);
 		while (1) {
-			ret = run_next_block(root, bits, bits_nr, &last,
+			ret = run_next_block(trans, root, bits, bits_nr, &last,
 					     &pending, &seen, &reada,
 					     &nodes, &extent_cache,
 					     &chunk_cache, &dev_cache,
@@ -5612,7 +5819,8 @@ again:
 		free(rec);
 	}
 
-	ret = check_extent_refs(trans, root, &extent_cache);
+	if (ret >= 0)
+		ret = check_extent_refs(trans, root, &extent_cache);
 	if (ret == -EAGAIN) {
 		ret = btrfs_commit_transaction(trans, root);
 		if (ret)
@@ -6096,6 +6304,49 @@ static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb
 	return ret;
 }
 
+static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
+{
+	struct btrfs_path *path;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_key key;
+	int ret;
+
+	printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
+	       bad->key.type, bad->key.offset);
+	key.objectid = bad->root_id;
+	key.type = BTRFS_ROOT_ITEM_KEY;
+	key.offset = (u64)-1;
+
+	root = btrfs_read_fs_root(root->fs_info, &key);
+	if (IS_ERR(root)) {
+		fprintf(stderr, "Couldn't find owner root %llu\n",
+			key.objectid);
+		return PTR_ERR(root);
+	}
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	trans = btrfs_start_transaction(root, 1);
+	if (IS_ERR(trans)) {
+		btrfs_free_path(path);
+		return PTR_ERR(trans);
+	}
+
+	ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
+	if (ret) {
+		if (ret > 0)
+			ret = 0;
+		goto out;
+	}
+	ret = btrfs_del_item(trans, root, path);
+out:
+	btrfs_commit_transaction(trans, root);
+	btrfs_free_path(path);
+	return ret;
+}
+
 static struct option long_options[] = {
 	{ "super", 1, NULL, 's' },
 	{ "repair", 0, NULL, 0 },
@@ -6274,6 +6525,16 @@ int cmd_check(int argc, char **argv)
 			break;
 	}
 
+	while (!list_empty(&delete_items)) {
+		struct bad_item *bad;
+
+		bad = list_first_entry(&delete_items, struct bad_item, list);
+		list_del_init(&bad->list);
+		if (repair)
+			ret = delete_bad_item(root, bad);
+		free(bad);
+	}
+
 	if (!list_empty(&root->fs_info->recow_ebs)) {
 		fprintf(stderr, "Transid errors in file system\n");
 		ret = 1;
diff --git a/ctree.c b/ctree.c
index 2d4f773b..9e5b30f3 100644
--- a/ctree.c
+++ b/ctree.c
@@ -371,31 +371,35 @@ int btrfs_cow_block(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
-/*
- * compare two keys in a memcmp fashion
- */
-static int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
+int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
 {
-	struct btrfs_key k1;
-
-	btrfs_disk_key_to_cpu(&k1, disk);
-
-	if (k1.objectid > k2->objectid)
+	if (k1->objectid > k2->objectid)
 		return 1;
-	if (k1.objectid < k2->objectid)
+	if (k1->objectid < k2->objectid)
 		return -1;
-	if (k1.type > k2->type)
+	if (k1->type > k2->type)
 		return 1;
-	if (k1.type < k2->type)
+	if (k1->type < k2->type)
 		return -1;
-	if (k1.offset > k2->offset)
+	if (k1->offset > k2->offset)
 		return 1;
-	if (k1.offset < k2->offset)
+	if (k1->offset < k2->offset)
 		return -1;
 	return 0;
 }
 
 /*
+ * compare two keys in a memcmp fashion
+ */
+static int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
+{
+	struct btrfs_key k1;
+
+	btrfs_disk_key_to_cpu(&k1, disk);
+	return btrfs_comp_cpu_keys(&k1, k2);
+}
+
+/*
  * The leaf data grows from end-to-front in the node.
  * this returns the address of the start of the last item,
  * which is the stop of the leaf data stack
@@ -409,30 +413,33 @@ static inline unsigned int leaf_data_end(struct btrfs_root *root,
 	return btrfs_item_offset_nr(leaf, nr - 1);
 }
 
-int btrfs_check_node(struct btrfs_root *root,
-		      struct btrfs_disk_key *parent_key,
-		      struct extent_buffer *buf)
+enum btrfs_tree_block_status
+btrfs_check_node(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
+		 struct extent_buffer *buf)
 {
 	int i;
 	struct btrfs_key cpukey;
 	struct btrfs_disk_key key;
 	u32 nritems = btrfs_header_nritems(buf);
+	enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
 
 	if (nritems == 0 || nritems > BTRFS_NODEPTRS_PER_BLOCK(root))
 		goto fail;
 
+	ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
 	if (parent_key && parent_key->type) {
 		btrfs_node_key(buf, &key, 0);
 		if (memcmp(parent_key, &key, sizeof(key)))
 			goto fail;
 	}
+	ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
 	for (i = 0; nritems > 1 && i < nritems - 2; i++) {
 		btrfs_node_key(buf, &key, i);
 		btrfs_node_key_to_cpu(buf, &cpukey, i + 1);
 		if (btrfs_comp_keys(&key, &cpukey) >= 0)
 			goto fail;
 	}
-	return 0;
+	return BTRFS_TREE_BLOCK_CLEAN;
 fail:
 	if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
 		if (parent_key)
@@ -443,17 +450,18 @@ fail:
 						buf->start, buf->len,
 						btrfs_header_level(buf));
 	}
-	return -EIO;
+	return ret;
 }
 
-int btrfs_check_leaf(struct btrfs_root *root,
-		      struct btrfs_disk_key *parent_key,
-		      struct extent_buffer *buf)
+enum btrfs_tree_block_status
+btrfs_check_leaf(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
+		 struct extent_buffer *buf)
 {
 	int i;
 	struct btrfs_key cpukey;
 	struct btrfs_disk_key key;
 	u32 nritems = btrfs_header_nritems(buf);
+	enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
 
 	if (nritems * sizeof(struct btrfs_item) > buf->len)  {
 		fprintf(stderr, "invalid number of items %llu\n",
@@ -462,11 +470,13 @@ int btrfs_check_leaf(struct btrfs_root *root,
 	}
 
 	if (btrfs_header_level(buf) != 0) {
+		ret = BTRFS_TREE_BLOCK_INVALID_LEVEL;
 		fprintf(stderr, "leaf is not a leaf %llu\n",
 		       (unsigned long long)btrfs_header_bytenr(buf));
 		goto fail;
 	}
 	if (btrfs_leaf_free_space(root, buf) < 0) {
+		ret = BTRFS_TREE_BLOCK_INVALID_FREE_SPACE;
 		fprintf(stderr, "leaf free space incorrect %llu %d\n",
 			(unsigned long long)btrfs_header_bytenr(buf),
 			btrfs_leaf_free_space(root, buf));
@@ -474,11 +484,12 @@ int btrfs_check_leaf(struct btrfs_root *root,
 	}
 
 	if (nritems == 0)
-		return 0;
+		return BTRFS_TREE_BLOCK_CLEAN;
 
 	btrfs_item_key(buf, &key, 0);
 	if (parent_key && parent_key->type &&
 	    memcmp(parent_key, &key, sizeof(key))) {
+		ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
 		fprintf(stderr, "leaf parent key incorrect %llu\n",
 		       (unsigned long long)btrfs_header_bytenr(buf));
 		goto fail;
@@ -487,11 +498,13 @@ int btrfs_check_leaf(struct btrfs_root *root,
 		btrfs_item_key(buf, &key, i);
 		btrfs_item_key_to_cpu(buf, &cpukey, i + 1);
 		if (btrfs_comp_keys(&key, &cpukey) >= 0) {
+			ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
 			fprintf(stderr, "bad key ordering %d %d\n", i, i+1);
 			goto fail;
 		}
 		if (btrfs_item_offset_nr(buf, i) !=
 			btrfs_item_end_nr(buf, i + 1)) {
+			ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
 			fprintf(stderr, "incorrect offsets %u %u\n",
 				btrfs_item_offset_nr(buf, i),
 				btrfs_item_end_nr(buf, i + 1));
@@ -499,13 +512,14 @@ int btrfs_check_leaf(struct btrfs_root *root,
 		}
 		if (i == 0 && btrfs_item_end_nr(buf, i) !=
 		    BTRFS_LEAF_DATA_SIZE(root)) {
+			ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
 			fprintf(stderr, "bad item end %u wanted %u\n",
 				btrfs_item_end_nr(buf, i),
 				(unsigned)BTRFS_LEAF_DATA_SIZE(root));
 			goto fail;
 		}
 	}
-	return 0;
+	return BTRFS_TREE_BLOCK_CLEAN;
 fail:
 	if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
 		if (parent_key)
@@ -516,7 +530,7 @@ fail:
 		btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
 						buf->start, buf->len, 0);
 	}
-	return -EIO;
+	return ret;
 }
 
 static int noinline check_block(struct btrfs_root *root,
@@ -525,15 +539,22 @@ static int noinline check_block(struct btrfs_root *root,
 	struct btrfs_disk_key key;
 	struct btrfs_disk_key *key_ptr = NULL;
 	struct extent_buffer *parent;
+	enum btrfs_tree_block_status ret;
 
+	if (path->skip_check_block)
+		return 0;
 	if (path->nodes[level + 1]) {
 		parent = path->nodes[level + 1];
 		btrfs_node_key(parent, &key, path->slots[level + 1]);
 		key_ptr = &key;
 	}
 	if (level == 0)
-		return btrfs_check_leaf(root, key_ptr, path->nodes[0]);
-	return btrfs_check_node(root, key_ptr, path->nodes[level]);
+		ret =  btrfs_check_leaf(root, key_ptr, path->nodes[0]);
+	else
+		ret = btrfs_check_node(root, key_ptr, path->nodes[level]);
+	if (ret == BTRFS_TREE_BLOCK_CLEAN)
+		return 0;
+	return -EIO;
 }
 
 /*
@@ -1114,16 +1135,11 @@ again:
  * This is used after shifting pointers to the left, so it stops
  * fixing up pointers when a given leaf/node is not in slot 0 of the
  * higher levels
- *
- * If this fails to write a tree block, it returns -1, but continues
- * fixing up the blocks in ram so the tree is consistent.
  */
-static int fixup_low_keys(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root, struct btrfs_path *path,
+void btrfs_fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
 			  struct btrfs_disk_key *key, int level)
 {
 	int i;
-	int ret = 0;
 	struct extent_buffer *t;
 
 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
@@ -1136,7 +1152,6 @@ static int fixup_low_keys(struct btrfs_trans_handle *trans,
 		if (tslot != 0)
 			break;
 	}
-	return ret;
 }
 
 /*
@@ -1145,8 +1160,7 @@ static int fixup_low_keys(struct btrfs_trans_handle *trans,
  * This function isn't completely safe. It's the caller's responsibility
  * that the new key won't break the order
  */
-int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root, struct btrfs_path *path,
+int btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
 			    struct btrfs_key *new_key)
 {
 	struct btrfs_disk_key disk_key;
@@ -1170,11 +1184,33 @@ int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
 	btrfs_set_item_key(eb, &disk_key, slot);
 	btrfs_mark_buffer_dirty(eb);
 	if (slot == 0)
-		fixup_low_keys(trans, root, path, &disk_key, 1);
+		btrfs_fixup_low_keys(root, path, &disk_key, 1);
 	return 0;
 }
 
 /*
+ * update an item key without the safety checks.  This is meant to be called by
+ * fsck only.
+ */
+void btrfs_set_item_key_unsafe(struct btrfs_root *root,
+			       struct btrfs_path *path,
+			       struct btrfs_key *new_key)
+{
+	struct btrfs_disk_key disk_key;
+	struct extent_buffer *eb;
+	int slot;
+
+	eb = path->nodes[0];
+	slot = path->slots[0];
+
+	btrfs_cpu_key_to_disk(&disk_key, new_key);
+	btrfs_set_item_key(eb, &disk_key, slot);
+	btrfs_mark_buffer_dirty(eb);
+	if (slot == 0)
+		btrfs_fixup_low_keys(root, path, &disk_key, 1);
+}
+
+/*
  * try to push data from one node into the next node left in the
  * tree.
  *
@@ -1706,7 +1742,6 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
 	u32 right_nritems;
 	u32 nr;
 	int ret = 0;
-	int wret;
 	u32 this_item_size;
 	u32 old_left_item_size;
 
@@ -1830,9 +1865,7 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
 		btrfs_mark_buffer_dirty(right);
 
 	btrfs_item_key(right, &disk_key, 0);
-	wret = fixup_low_keys(trans, root, path, &disk_key, 1);
-	if (wret)
-		ret = wret;
+	btrfs_fixup_low_keys(root, path, &disk_key, 1);
 
 	/* then fixup the leaf pointer in the path */
 	if (path->slots[0] < push_items) {
@@ -2052,10 +2085,8 @@ again:
 			path->nodes[0] = right;
 			path->slots[0] = 0;
 			if (path->slots[1] == 0) {
-				wret = fixup_low_keys(trans, root,
-						path, &disk_key, 1);
-				if (wret)
-					ret = wret;
+				btrfs_fixup_low_keys(root, path,
+						     &disk_key, 1);
 			}
 		}
 		btrfs_mark_buffer_dirty(right);
@@ -2270,7 +2301,7 @@ int btrfs_truncate_item(struct btrfs_trans_handle *trans,
 		btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
 		btrfs_set_item_key(leaf, &disk_key, slot);
 		if (slot == 0)
-			fixup_low_keys(trans, root, path, &disk_key, 1);
+			btrfs_fixup_low_keys(root, path, &disk_key, 1);
 	}
 
 	item = btrfs_item_nr(slot);
@@ -2448,7 +2479,7 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
 	ret = 0;
 	if (slot == 0) {
 		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
-		ret = fixup_low_keys(trans, root, path, &disk_key, 1);
+		btrfs_fixup_low_keys(root, path, &disk_key, 1);
 	}
 
 	if (btrfs_leaf_free_space(root, leaf) < 0) {
@@ -2499,7 +2530,6 @@ int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 	struct extent_buffer *parent = path->nodes[level];
 	u32 nritems;
 	int ret = 0;
-	int wret;
 
 	nritems = btrfs_header_nritems(parent);
 	if (slot != nritems -1) {
@@ -2519,9 +2549,7 @@ int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		struct btrfs_disk_key disk_key;
 
 		btrfs_node_key(parent, &disk_key, 0);
-		wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
-		if (wret)
-			ret = wret;
+		btrfs_fixup_low_keys(root, path, &disk_key, level + 1);
 	}
 	btrfs_mark_buffer_dirty(parent);
 	return ret;
@@ -2621,10 +2649,7 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 			struct btrfs_disk_key disk_key;
 
 			btrfs_item_key(leaf, &disk_key, 0);
-			wret = fixup_low_keys(trans, root, path,
-					      &disk_key, 1);
-			if (wret)
-				ret = wret;
+			btrfs_fixup_low_keys(root, path, &disk_key, 1);
 		}
 
 		/* delete the leaf if it is mostly empty */
diff --git a/ctree.h b/ctree.h
index 6d101848..22a5c6ae 100644
--- a/ctree.h
+++ b/ctree.h
@@ -547,6 +547,7 @@ struct btrfs_path {
 	unsigned int keep_locks:1;
 	unsigned int skip_locking:1;
 	unsigned int leave_spinning:1;
+	unsigned int skip_check_block:1;
 };
 
 /*
@@ -647,6 +648,16 @@ typedef enum {
 	BTRFS_ENCRYPTION_LAST = 1,
 } btrfs_encryption_type;
 
+enum btrfs_tree_block_status {
+	BTRFS_TREE_BLOCK_CLEAN,
+	BTRFS_TREE_BLOCK_INVALID_NRITEMS,
+	BTRFS_TREE_BLOCK_INVALID_PARENT_KEY,
+	BTRFS_TREE_BLOCK_BAD_KEY_ORDER,
+	BTRFS_TREE_BLOCK_INVALID_LEVEL,
+	BTRFS_TREE_BLOCK_INVALID_FREE_SPACE,
+	BTRFS_TREE_BLOCK_INVALID_OFFSETS,
+};
+
 struct btrfs_inode_item {
 	/* nfs style generation number */
 	__le64 generation;
@@ -2202,14 +2213,15 @@ int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
 			      u64 file_pos, u64 disk_bytenr,
 			      u64 num_bytes);
 /* ctree.c */
+int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2);
 int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		   struct btrfs_path *path, int level, int slot);
-int btrfs_check_node(struct btrfs_root *root,
-		      struct btrfs_disk_key *parent_key,
-		      struct extent_buffer *buf);
-int btrfs_check_leaf(struct btrfs_root *root,
-		      struct btrfs_disk_key *parent_key,
-		      struct extent_buffer *buf);
+enum btrfs_tree_block_status
+btrfs_check_node(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
+		 struct extent_buffer *buf);
+enum btrfs_tree_block_status
+btrfs_check_leaf(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
+		 struct extent_buffer *buf);
 void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
 			     int level, int slot, u64 objectid);
 struct extent_buffer *read_node_slot(struct btrfs_root *root,
@@ -2279,9 +2291,13 @@ static inline int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path);
 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf);
-int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root, struct btrfs_path *path,
+void btrfs_fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
+			  struct btrfs_disk_key *key, int level);
+int btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
 			    struct btrfs_key *new_key);
+void btrfs_set_item_key_unsafe(struct btrfs_root *root,
+			       struct btrfs_path *path,
+			       struct btrfs_key *new_key);
 
 /* root-item.c */
 int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
diff --git a/file-item.c b/file-item.c
index f53a5cab..6f3708b8 100644
--- a/file-item.c
+++ b/file-item.c
@@ -367,7 +367,7 @@ static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
 		BUG_ON(ret);
 
 		key->offset = end_byte;
-		ret = btrfs_set_item_key_safe(trans, root, path, key);
+		ret = btrfs_set_item_key_safe(root, path, key);
 		BUG_ON(ret);
 	} else {
 		BUG();