btrfs-progs: check: fix the return value bug of cmd_check()
[platform/upstream/btrfs-progs.git] / cmds-check.c
index 24c729b..14816bd 100644 (file)
@@ -41,6 +41,7 @@
 #include "rbtree-utils.h"
 #include "backref.h"
 #include "ulist.h"
+#include "hash.h"
 
 enum task_position {
        TASK_EXTENTS,
@@ -2250,31 +2251,25 @@ static int delete_dir_index(struct btrfs_root *root,
 {
        struct btrfs_trans_handle *trans;
        struct btrfs_dir_item *di;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        int ret = 0;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
        trans = btrfs_start_transaction(root, 1);
-       if (IS_ERR(trans)) {
-               btrfs_free_path(path);
+       if (IS_ERR(trans))
                return PTR_ERR(trans);
-       }
-
 
        fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
                (unsigned long long)backref->dir,
                BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
                (unsigned long long)root->objectid);
 
-       di = btrfs_lookup_dir_index(trans, root, path, backref->dir,
+       btrfs_init_path(&path);
+       di = btrfs_lookup_dir_index(trans, root, &path, backref->dir,
                                    backref->name, backref->namelen,
                                    backref->index, -1);
        if (IS_ERR(di)) {
                ret = PTR_ERR(di);
-               btrfs_free_path(path);
+               btrfs_release_path(&path);
                btrfs_commit_transaction(trans, root);
                if (ret == -ENOENT)
                        return 0;
@@ -2282,11 +2277,11 @@ static int delete_dir_index(struct btrfs_root *root,
        }
 
        if (!di)
-               ret = btrfs_del_item(trans, root, path);
+               ret = btrfs_del_item(trans, root, &path);
        else
-               ret = btrfs_delete_one_dir_name(trans, root, path, di);
+               ret = btrfs_delete_one_dir_name(trans, root, &path, di);
        BUG_ON(ret);
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        btrfs_commit_transaction(trans, root);
        return ret;
 }
@@ -2689,48 +2684,46 @@ out:
  */
 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
 {
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct btrfs_key key;
        struct btrfs_key found_key;
        struct btrfs_file_extent_item *fi;
        u8 type;
        int ret = 0;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               goto out;
+       btrfs_init_path(&path);
        key.objectid = ino;
        key.type = BTRFS_EXTENT_DATA_KEY;
        key.offset = 0;
 
-       ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+       ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
        if (ret < 0) {
                ret = 0;
                goto out;
        }
-       if (ret && path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
-               ret = btrfs_next_leaf(root, path);
+       if (ret && path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
+               ret = btrfs_next_leaf(root, &path);
                if (ret) {
                        ret = 0;
                        goto out;
                }
        }
        while (1) {
-               btrfs_item_key_to_cpu(path->nodes[0], &found_key,
-                                     path->slots[0]);
+               btrfs_item_key_to_cpu(path.nodes[0], &found_key,
+                                     path.slots[0]);
                if (found_key.objectid != ino ||
                    found_key.type != BTRFS_EXTENT_DATA_KEY)
                        break;
-               fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
+               fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
                                    struct btrfs_file_extent_item);
-               type = btrfs_file_extent_type(path->nodes[0], fi);
+               type = btrfs_file_extent_type(path.nodes[0], fi);
                if (type != BTRFS_FILE_EXTENT_INLINE) {
                        ret = 1;
                        goto out;
                }
        }
 out:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -2921,7 +2914,7 @@ out:
 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
 {
        struct btrfs_trans_handle *trans;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        int ret = 0;
 
        if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
@@ -2933,10 +2926,6 @@ static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
                             I_ERR_FILE_NBYTES_WRONG)))
                return rec->errors;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
        /*
         * For nlink repair, it may create a dir and add link, so
         * 2 for parent(256)'s dir_index and dir_item
@@ -2945,27 +2934,26 @@ static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
         * 2 for lost+found dir's dir_index and dir_item for the file
         */
        trans = btrfs_start_transaction(root, 7);
-       if (IS_ERR(trans)) {
-               btrfs_free_path(path);
+       if (IS_ERR(trans))
                return PTR_ERR(trans);
-       }
 
+       btrfs_init_path(&path);
        if (rec->errors & I_ERR_NO_INODE_ITEM)
-               ret = repair_inode_no_item(trans, root, path, rec);
+               ret = repair_inode_no_item(trans, root, &path, rec);
        if (!ret && rec->errors & I_ERR_FILE_EXTENT_ORPHAN)
-               ret = repair_inode_orphan_extent(trans, root, path, rec);
+               ret = repair_inode_orphan_extent(trans, root, &path, rec);
        if (!ret && rec->errors & I_ERR_FILE_EXTENT_DISCOUNT)
-               ret = repair_inode_discount_extent(trans, root, path, rec);
+               ret = repair_inode_discount_extent(trans, root, &path, rec);
        if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
-               ret = repair_inode_isize(trans, root, path, rec);
+               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);
+               ret = repair_inode_orphan_item(trans, root, &path, rec);
        if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
-               ret = repair_inode_nlinks(trans, root, path, rec);
+               ret = repair_inode_nlinks(trans, root, &path, rec);
        if (!ret && rec->errors & I_ERR_FILE_NBYTES_WRONG)
-               ret = repair_inode_nbytes(trans, root, path, rec);
+               ret = repair_inode_nbytes(trans, root, &path, rec);
        btrfs_commit_transaction(trans, root);
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -3498,7 +3486,7 @@ static int repair_btree(struct btrfs_root *root,
                        struct cache_tree *corrupt_blocks)
 {
        struct btrfs_trans_handle *trans;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct btrfs_corrupt_block *corrupt;
        struct cache_extent *cache;
        struct btrfs_key key;
@@ -3509,23 +3497,20 @@ static int repair_btree(struct btrfs_root *root,
        if (cache_tree_empty(corrupt_blocks))
                return 0;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
        trans = btrfs_start_transaction(root, 1);
        if (IS_ERR(trans)) {
                ret = PTR_ERR(trans);
                fprintf(stderr, "Error starting transaction: %s\n",
                        strerror(-ret));
-               goto out_free_path;
+               return ret;
        }
+       btrfs_init_path(&path);
        cache = first_cache_extent(corrupt_blocks);
        while (cache) {
                corrupt = container_of(cache, struct btrfs_corrupt_block,
                                       cache);
                level = corrupt->level;
-               path->lowest_level = level;
+               path.lowest_level = level;
                key.objectid = corrupt->key.objectid;
                key.type = corrupt->key.type;
                key.offset = corrupt->key.offset;
@@ -3536,22 +3521,22 @@ static int repair_btree(struct btrfs_root *root,
                 * so ins_len set to 0 here.
                 * Balance will be done after all corrupt node/leaf is deleted.
                 */
-               ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+               ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
                if (ret < 0)
                        goto out;
-               offset = btrfs_node_blockptr(path->nodes[level],
-                                            path->slots[level]);
+               offset = btrfs_node_blockptr(path.nodes[level],
+                                            path.slots[level]);
 
                /* Remove the ptr */
-               ret = btrfs_del_ptr(trans, root, path, level,
-                                   path->slots[level]);
+               ret = btrfs_del_ptr(trans, root, &path, level,
+                                   path.slots[level]);
                if (ret < 0)
                        goto out;
                /*
                 * Remove the corresponding extent
                 * return value is not concerned.
                 */
-               btrfs_release_path(path);
+               btrfs_release_path(&path);
                ret = btrfs_free_extent(trans, root, offset, root->nodesize,
                                        0, root->root_key.objectid,
                                        level - 1, 0);
@@ -3564,18 +3549,17 @@ static int repair_btree(struct btrfs_root *root,
                corrupt = container_of(cache, struct btrfs_corrupt_block,
                                       cache);
                memcpy(&key, &corrupt->key, sizeof(key));
-               ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+               ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
                if (ret < 0)
                        goto out;
                /* return will always >0 since it won't find the item */
                ret = 0;
-               btrfs_release_path(path);
+               btrfs_release_path(&path);
                cache = next_cache_extent(cache);
        }
 out:
        btrfs_commit_transaction(trans, root);
-out_free_path:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -3715,131 +3699,1219 @@ skip_walking:
                                printf("Btree for root %llu is fixed\n",
                                       root->root_key.objectid);
                }
-       }
+       }
+
+       err = merge_root_recs(root, &root_node.root_cache, root_cache);
+       if (err < 0)
+               ret = err;
+
+       if (root_node.current) {
+               root_node.current->checked = 1;
+               maybe_free_inode_rec(&root_node.inode_cache,
+                               root_node.current);
+       }
+
+       err = check_inode_recs(root, &root_node.inode_cache);
+       if (!ret)
+               ret = err;
+
+       free_corrupt_blocks_tree(&corrupt_blocks);
+       root->fs_info->corrupt_blocks = NULL;
+       free_orphan_data_extents(&root->orphan_data_extents);
+       return ret;
+}
+
+static int fs_root_objectid(u64 objectid)
+{
+       if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
+           objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
+               return 1;
+       return is_fstree(objectid);
+}
+
+static int check_fs_roots(struct btrfs_root *root,
+                         struct cache_tree *root_cache)
+{
+       struct btrfs_path path;
+       struct btrfs_key key;
+       struct walk_control wc;
+       struct extent_buffer *leaf, *tree_node;
+       struct btrfs_root *tmp_root;
+       struct btrfs_root *tree_root = root->fs_info->tree_root;
+       int ret;
+       int err = 0;
+
+       if (ctx.progress_enabled) {
+               ctx.tp = TASK_FS_ROOTS;
+               task_start(ctx.info);
+       }
+
+       /*
+        * Just in case we made any changes to the extent tree that weren't
+        * reflected into the free space cache yet.
+        */
+       if (repair)
+               reset_cached_block_groups(root->fs_info);
+       memset(&wc, 0, sizeof(wc));
+       cache_tree_init(&wc.shared);
+       btrfs_init_path(&path);
+
+again:
+       key.offset = 0;
+       key.objectid = 0;
+       key.type = BTRFS_ROOT_ITEM_KEY;
+       ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
+       if (ret < 0) {
+               err = 1;
+               goto out;
+       }
+       tree_node = tree_root->node;
+       while (1) {
+               if (tree_node != tree_root->node) {
+                       free_root_recs_tree(root_cache);
+                       btrfs_release_path(&path);
+                       goto again;
+               }
+               leaf = path.nodes[0];
+               if (path.slots[0] >= btrfs_header_nritems(leaf)) {
+                       ret = btrfs_next_leaf(tree_root, &path);
+                       if (ret) {
+                               if (ret < 0)
+                                       err = 1;
+                               break;
+                       }
+                       leaf = path.nodes[0];
+               }
+               btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
+               if (key.type == BTRFS_ROOT_ITEM_KEY &&
+                   fs_root_objectid(key.objectid)) {
+                       if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
+                               tmp_root = btrfs_read_fs_root_no_cache(
+                                               root->fs_info, &key);
+                       } else {
+                               key.offset = (u64)-1;
+                               tmp_root = btrfs_read_fs_root(
+                                               root->fs_info, &key);
+                       }
+                       if (IS_ERR(tmp_root)) {
+                               err = 1;
+                               goto next;
+                       }
+                       ret = check_fs_root(tmp_root, root_cache, &wc);
+                       if (ret == -EAGAIN) {
+                               free_root_recs_tree(root_cache);
+                               btrfs_release_path(&path);
+                               goto again;
+                       }
+                       if (ret)
+                               err = 1;
+                       if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
+                               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,
+                                        root_cache);
+               }
+next:
+               path.slots[0]++;
+       }
+out:
+       btrfs_release_path(&path);
+       if (err)
+               free_extent_cache_tree(&wc.shared);
+       if (!cache_tree_empty(&wc.shared))
+               fprintf(stderr, "warning line %d\n", __LINE__);
+
+       task_stop(ctx.info);
+
+       return err;
+}
+
+#define ROOT_DIR_ERROR         (1<<1)  /* bad ROOT_DIR */
+#define DIR_ITEM_MISSING       (1<<2)  /* DIR_ITEM not found */
+#define DIR_ITEM_MISMATCH      (1<<3)  /* DIR_ITEM found but not match */
+#define INODE_REF_MISSING      (1<<4)  /* INODE_REF/INODE_EXTREF not found */
+#define INODE_ITEM_MISSING     (1<<5)  /* INODE_ITEM not found */
+#define INODE_ITEM_MISMATCH    (1<<6)  /* INODE_ITEM found but not match */
+#define FILE_EXTENT_ERROR      (1<<7)  /* bad FILE_EXTENT */
+#define ODD_CSUM_ITEM          (1<<8)  /* CSUM_ITEM error */
+#define CSUM_ITEM_MISSING      (1<<9)  /* CSUM_ITEM not found */
+#define LINK_COUNT_ERROR       (1<<10) /* INODE_ITEM nlink count error */
+#define NBYTES_ERROR           (1<<11) /* INODE_ITEM nbytes count error */
+#define ISIZE_ERROR            (1<<12) /* INODE_ITEM size count error */
+#define ORPHAN_ITEM            (1<<13) /* INODE_ITEM no reference */
+#define NO_INODE_ITEM          (1<<14) /* no inode_item */
+#define LAST_ITEM              (1<<15) /* Complete this tree traversal */
+#define ROOT_REF_MISSING       (1<<16) /* ROOT_REF not found */
+#define ROOT_REF_MISMATCH      (1<<17) /* ROOT_REF found but not match */
+
+/*
+ * Find DIR_ITEM/DIR_INDEX for the given key and check it with the specified
+ * INODE_REF/INODE_EXTREF match.
+ *
+ * @root:      the root of the fs/file tree
+ * @ref_key:   the key of the INODE_REF/INODE_EXTREF
+ * @key:       the key of the DIR_ITEM/DIR_INDEX
+ * @index:     the index in the INODE_REF/INODE_EXTREF, be used to
+ *             distinguish root_dir between normal dir/file
+ * @name:      the name in the INODE_REF/INODE_EXTREF
+ * @namelen:   the length of name in the INODE_REF/INODE_EXTREF
+ * @mode:      the st_mode of INODE_ITEM
+ *
+ * Return 0 if no error occurred.
+ * Return ROOT_DIR_ERROR if found DIR_ITEM/DIR_INDEX for root_dir.
+ * Return DIR_ITEM_MISSING if couldn't find DIR_ITEM/DIR_INDEX for normal
+ * dir/file.
+ * Return DIR_ITEM_MISMATCH if INODE_REF/INODE_EXTREF and DIR_ITEM/DIR_INDEX
+ * not match for normal dir/file.
+ */
+static int find_dir_item(struct btrfs_root *root, struct btrfs_key *ref_key,
+                        struct btrfs_key *key, u64 index, char *name,
+                        u32 namelen, u32 mode)
+{
+       struct btrfs_path path;
+       struct extent_buffer *node;
+       struct btrfs_dir_item *di;
+       struct btrfs_key location;
+       char namebuf[BTRFS_NAME_LEN] = {0};
+       u32 total;
+       u32 cur = 0;
+       u32 len;
+       u32 name_len;
+       u32 data_len;
+       u8 filetype;
+       int slot;
+       int ret;
+
+       btrfs_init_path(&path);
+       ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
+       if (ret < 0) {
+               ret = DIR_ITEM_MISSING;
+               goto out;
+       }
+
+       /* Process root dir and goto out*/
+       if (index == 0) {
+               if (ret == 0) {
+                       ret = ROOT_DIR_ERROR;
+                       error(
+                       "root %llu INODE %s[%llu %llu] ROOT_DIR shouldn't have %s",
+                               root->objectid,
+                               ref_key->type == BTRFS_INODE_REF_KEY ?
+                                       "REF" : "EXTREF",
+                               ref_key->objectid, ref_key->offset,
+                               key->type == BTRFS_DIR_ITEM_KEY ?
+                                       "DIR_ITEM" : "DIR_INDEX");
+               } else {
+                       ret = 0;
+               }
+
+               goto out;
+       }
+
+       /* Process normal file/dir */
+       if (ret > 0) {
+               ret = DIR_ITEM_MISSING;
+               error(
+               "root %llu INODE %s[%llu %llu] doesn't have related %s[%llu %llu] namelen %u filename %s filetype %d",
+                       root->objectid,
+                       ref_key->type == BTRFS_INODE_REF_KEY ? "REF" : "EXTREF",
+                       ref_key->objectid, ref_key->offset,
+                       key->type == BTRFS_DIR_ITEM_KEY ?
+                               "DIR_ITEM" : "DIR_INDEX",
+                       key->objectid, key->offset, namelen, name,
+                       imode_to_type(mode));
+               goto out;
+       }
+
+       /* Check whether inode_id/filetype/name match */
+       node = path.nodes[0];
+       slot = path.slots[0];
+       di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
+       total = btrfs_item_size_nr(node, slot);
+       while (cur < total) {
+               ret = DIR_ITEM_MISMATCH;
+               name_len = btrfs_dir_name_len(node, di);
+               data_len = btrfs_dir_data_len(node, di);
+
+               btrfs_dir_item_key_to_cpu(node, di, &location);
+               if (location.objectid != ref_key->objectid ||
+                   location.type !=  BTRFS_INODE_ITEM_KEY ||
+                   location.offset != 0)
+                       goto next;
+
+               filetype = btrfs_dir_type(node, di);
+               if (imode_to_type(mode) != filetype)
+                       goto next;
+
+               if (name_len <= BTRFS_NAME_LEN) {
+                       len = name_len;
+               } else {
+                       len = BTRFS_NAME_LEN;
+                       warning("root %llu %s[%llu %llu] name too long %u, trimmed",
+                       root->objectid,
+                       key->type == BTRFS_DIR_ITEM_KEY ?
+                       "DIR_ITEM" : "DIR_INDEX",
+                       key->objectid, key->offset, name_len);
+               }
+               read_extent_buffer(node, namebuf, (unsigned long)(di + 1), len);
+               if (len != namelen || strncmp(namebuf, name, len))
+                       goto next;
+
+               ret = 0;
+               goto out;
+next:
+               len = sizeof(*di) + name_len + data_len;
+               di = (struct btrfs_dir_item *)((char *)di + len);
+               cur += len;
+       }
+       if (ret == DIR_ITEM_MISMATCH)
+               error(
+               "root %llu INODE %s[%llu %llu] and %s[%llu %llu] mismatch namelen %u filename %s filetype %d",
+                       root->objectid,
+                       ref_key->type == BTRFS_INODE_REF_KEY ? "REF" : "EXTREF",
+                       ref_key->objectid, ref_key->offset,
+                       key->type == BTRFS_DIR_ITEM_KEY ?
+                               "DIR_ITEM" : "DIR_INDEX",
+                       key->objectid, key->offset, namelen, name,
+                       imode_to_type(mode));
+out:
+       btrfs_release_path(&path);
+       return ret;
+}
+
+/*
+ * Traverse the given INODE_REF and call find_dir_item() to find related
+ * DIR_ITEM/DIR_INDEX.
+ *
+ * @root:      the root of the fs/file tree
+ * @ref_key:   the key of the INODE_REF
+ * @refs:      the count of INODE_REF
+ * @mode:      the st_mode of INODE_ITEM
+ *
+ * Return 0 if no error occurred.
+ */
+static int check_inode_ref(struct btrfs_root *root, struct btrfs_key *ref_key,
+                          struct extent_buffer *node, int slot, u64 *refs,
+                          int mode)
+{
+       struct btrfs_key key;
+       struct btrfs_inode_ref *ref;
+       char namebuf[BTRFS_NAME_LEN] = {0};
+       u32 total;
+       u32 cur = 0;
+       u32 len;
+       u32 name_len;
+       u64 index;
+       int ret, err = 0;
+
+       ref = btrfs_item_ptr(node, slot, struct btrfs_inode_ref);
+       total = btrfs_item_size_nr(node, slot);
+
+next:
+       /* Update inode ref count */
+       (*refs)++;
+
+       index = btrfs_inode_ref_index(node, ref);
+       name_len = btrfs_inode_ref_name_len(node, ref);
+       if (name_len <= BTRFS_NAME_LEN) {
+               len = name_len;
+       } else {
+               len = BTRFS_NAME_LEN;
+               warning("root %llu INODE_REF[%llu %llu] name too long",
+                       root->objectid, ref_key->objectid, ref_key->offset);
+       }
+
+       read_extent_buffer(node, namebuf, (unsigned long)(ref + 1), len);
+
+       /* Check root dir ref name */
+       if (index == 0 && strncmp(namebuf, "..", name_len)) {
+               error("root %llu INODE_REF[%llu %llu] ROOT_DIR name shouldn't be %s",
+                     root->objectid, ref_key->objectid, ref_key->offset,
+                     namebuf);
+               err |= ROOT_DIR_ERROR;
+       }
+
+       /* Find related DIR_INDEX */
+       key.objectid = ref_key->offset;
+       key.type = BTRFS_DIR_INDEX_KEY;
+       key.offset = index;
+       ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
+       err |= ret;
+
+       /* Find related dir_item */
+       key.objectid = ref_key->offset;
+       key.type = BTRFS_DIR_ITEM_KEY;
+       key.offset = btrfs_name_hash(namebuf, len);
+       ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
+       err |= ret;
+
+       len = sizeof(*ref) + name_len;
+       ref = (struct btrfs_inode_ref *)((char *)ref + len);
+       cur += len;
+       if (cur < total)
+               goto next;
+
+       return err;
+}
+
+/*
+ * Traverse the given INODE_EXTREF and call find_dir_item() to find related
+ * DIR_ITEM/DIR_INDEX.
+ *
+ * @root:      the root of the fs/file tree
+ * @ref_key:   the key of the INODE_EXTREF
+ * @refs:      the count of INODE_EXTREF
+ * @mode:      the st_mode of INODE_ITEM
+ *
+ * Return 0 if no error occurred.
+ */
+static int check_inode_extref(struct btrfs_root *root,
+                             struct btrfs_key *ref_key,
+                             struct extent_buffer *node, int slot, u64 *refs,
+                             int mode)
+{
+       struct btrfs_key key;
+       struct btrfs_inode_extref *extref;
+       char namebuf[BTRFS_NAME_LEN] = {0};
+       u32 total;
+       u32 cur = 0;
+       u32 len;
+       u32 name_len;
+       u64 index;
+       u64 parent;
+       int ret;
+       int err = 0;
+
+       extref = btrfs_item_ptr(node, slot, struct btrfs_inode_extref);
+       total = btrfs_item_size_nr(node, slot);
+
+next:
+       /* update inode ref count */
+       (*refs)++;
+       name_len = btrfs_inode_extref_name_len(node, extref);
+       index = btrfs_inode_extref_index(node, extref);
+       parent = btrfs_inode_extref_parent(node, extref);
+       if (name_len <= BTRFS_NAME_LEN) {
+               len = name_len;
+       } else {
+               len = BTRFS_NAME_LEN;
+               warning("root %llu INODE_EXTREF[%llu %llu] name too long",
+                       root->objectid, ref_key->objectid, ref_key->offset);
+       }
+       read_extent_buffer(node, namebuf, (unsigned long)(extref + 1), len);
+
+       /* Check root dir ref name */
+       if (index == 0 && strncmp(namebuf, "..", name_len)) {
+               error("root %llu INODE_EXTREF[%llu %llu] ROOT_DIR name shouldn't be %s",
+                     root->objectid, ref_key->objectid, ref_key->offset,
+                     namebuf);
+               err |= ROOT_DIR_ERROR;
+       }
+
+       /* find related dir_index */
+       key.objectid = parent;
+       key.type = BTRFS_DIR_INDEX_KEY;
+       key.offset = index;
+       ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
+       err |= ret;
+
+       /* find related dir_item */
+       key.objectid = parent;
+       key.type = BTRFS_DIR_ITEM_KEY;
+       key.offset = btrfs_name_hash(namebuf, len);
+       ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
+       err |= ret;
+
+       len = sizeof(*extref) + name_len;
+       extref = (struct btrfs_inode_extref *)((char *)extref + len);
+       cur += len;
+
+       if (cur < total)
+               goto next;
+
+       return err;
+}
+
+/*
+ * Find INODE_REF/INODE_EXTREF for the given key and check it with the specified
+ * DIR_ITEM/DIR_INDEX match.
+ *
+ * @root:      the root of the fs/file tree
+ * @key:       the key of the INODE_REF/INODE_EXTREF
+ * @name:      the name in the INODE_REF/INODE_EXTREF
+ * @namelen:   the length of name in the INODE_REF/INODE_EXTREF
+ * @index:     the index in the INODE_REF/INODE_EXTREF, for DIR_ITEM set index
+ * to (u64)-1
+ * @ext_ref:   the EXTENDED_IREF feature
+ *
+ * Return 0 if no error occurred.
+ * Return >0 for error bitmap
+ */
+static int find_inode_ref(struct btrfs_root *root, struct btrfs_key *key,
+                         char *name, int namelen, u64 index,
+                         unsigned int ext_ref)
+{
+       struct btrfs_path path;
+       struct btrfs_inode_ref *ref;
+       struct btrfs_inode_extref *extref;
+       struct extent_buffer *node;
+       char ref_namebuf[BTRFS_NAME_LEN] = {0};
+       u32 total;
+       u32 cur = 0;
+       u32 len;
+       u32 ref_namelen;
+       u64 ref_index;
+       u64 parent;
+       u64 dir_id;
+       int slot;
+       int ret;
+
+       btrfs_init_path(&path);
+       ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
+       if (ret) {
+               ret = INODE_REF_MISSING;
+               goto extref;
+       }
+
+       node = path.nodes[0];
+       slot = path.slots[0];
+
+       ref = btrfs_item_ptr(node, slot, struct btrfs_inode_ref);
+       total = btrfs_item_size_nr(node, slot);
+
+       /* Iterate all entry of INODE_REF */
+       while (cur < total) {
+               ret = INODE_REF_MISSING;
+
+               ref_namelen = btrfs_inode_ref_name_len(node, ref);
+               ref_index = btrfs_inode_ref_index(node, ref);
+               if (index != (u64)-1 && index != ref_index)
+                       goto next_ref;
+
+               if (ref_namelen <= BTRFS_NAME_LEN) {
+                       len = ref_namelen;
+               } else {
+                       len = BTRFS_NAME_LEN;
+                       warning("root %llu INODE %s[%llu %llu] name too long",
+                               root->objectid,
+                               key->type == BTRFS_INODE_REF_KEY ?
+                                       "REF" : "EXTREF",
+                               key->objectid, key->offset);
+               }
+               read_extent_buffer(node, ref_namebuf, (unsigned long)(ref + 1),
+                                  len);
+
+               if (len != namelen || strncmp(ref_namebuf, name, len))
+                       goto next_ref;
+
+               ret = 0;
+               goto out;
+next_ref:
+               len = sizeof(*ref) + ref_namelen;
+               ref = (struct btrfs_inode_ref *)((char *)ref + len);
+               cur += len;
+       }
+
+extref:
+       /* Skip if not support EXTENDED_IREF feature */
+       if (!ext_ref)
+               goto out;
+
+       btrfs_release_path(&path);
+       btrfs_init_path(&path);
+
+       dir_id = key->offset;
+       key->type = BTRFS_INODE_EXTREF_KEY;
+       key->offset = btrfs_extref_hash(dir_id, name, namelen);
+
+       ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
+       if (ret) {
+               ret = INODE_REF_MISSING;
+               goto out;
+       }
+
+       node = path.nodes[0];
+       slot = path.slots[0];
+
+       extref = btrfs_item_ptr(node, slot, struct btrfs_inode_extref);
+       cur = 0;
+       total = btrfs_item_size_nr(node, slot);
+
+       /* Iterate all entry of INODE_EXTREF */
+       while (cur < total) {
+               ret = INODE_REF_MISSING;
+
+               ref_namelen = btrfs_inode_extref_name_len(node, extref);
+               ref_index = btrfs_inode_extref_index(node, extref);
+               parent = btrfs_inode_extref_parent(node, extref);
+               if (index != (u64)-1 && index != ref_index)
+                       goto next_extref;
+
+               if (parent != dir_id)
+                       goto next_extref;
+
+               if (ref_namelen <= BTRFS_NAME_LEN) {
+                       len = ref_namelen;
+               } else {
+                       len = BTRFS_NAME_LEN;
+                       warning("Warning: root %llu INODE %s[%llu %llu] name too long\n",
+                               root->objectid,
+                               key->type == BTRFS_INODE_REF_KEY ?
+                                       "REF" : "EXTREF",
+                               key->objectid, key->offset);
+               }
+               read_extent_buffer(node, ref_namebuf,
+                                  (unsigned long)(extref + 1), len);
+
+               if (len != namelen || strncmp(ref_namebuf, name, len))
+                       goto next_extref;
+
+               ret = 0;
+               goto out;
+
+next_extref:
+               len = sizeof(*extref) + ref_namelen;
+               extref = (struct btrfs_inode_extref *)((char *)extref + len);
+               cur += len;
+
+       }
+out:
+       btrfs_release_path(&path);
+       return ret;
+}
+
+/*
+ * Traverse the given DIR_ITEM/DIR_INDEX and check related INODE_ITEM and
+ * call find_inode_ref() to check related INODE_REF/INODE_EXTREF.
+ *
+ * @root:      the root of the fs/file tree
+ * @key:       the key of the INODE_REF/INODE_EXTREF
+ * @size:      the st_size of the INODE_ITEM
+ * @ext_ref:   the EXTENDED_IREF feature
+ *
+ * Return 0 if no error occurred.
+ */
+static int check_dir_item(struct btrfs_root *root, struct btrfs_key *key,
+                         struct extent_buffer *node, int slot, u64 *size,
+                         unsigned int ext_ref)
+{
+       struct btrfs_dir_item *di;
+       struct btrfs_inode_item *ii;
+       struct btrfs_path path;
+       struct btrfs_key location;
+       char namebuf[BTRFS_NAME_LEN] = {0};
+       u32 total;
+       u32 cur = 0;
+       u32 len;
+       u32 name_len;
+       u32 data_len;
+       u8 filetype;
+       u32 mode;
+       u64 index;
+       int ret;
+       int err = 0;
+
+       /*
+        * For DIR_ITEM set index to (u64)-1, so that find_inode_ref
+        * ignore index check.
+        */
+       index = (key->type == BTRFS_DIR_INDEX_KEY) ? key->offset : (u64)-1;
+
+       di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
+       total = btrfs_item_size_nr(node, slot);
+
+       while (cur < total) {
+               data_len = btrfs_dir_data_len(node, di);
+               if (data_len)
+                       error("root %llu %s[%llu %llu] data_len shouldn't be %u",
+                             root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
+                             "DIR_ITEM" : "DIR_INDEX",
+                             key->objectid, key->offset, data_len);
+
+               name_len = btrfs_dir_name_len(node, di);
+               if (name_len <= BTRFS_NAME_LEN) {
+                       len = name_len;
+               } else {
+                       len = BTRFS_NAME_LEN;
+                       warning("root %llu %s[%llu %llu] name too long",
+                               root->objectid,
+                               key->type == BTRFS_DIR_ITEM_KEY ?
+                               "DIR_ITEM" : "DIR_INDEX",
+                               key->objectid, key->offset);
+               }
+               (*size) += name_len;
+
+               read_extent_buffer(node, namebuf, (unsigned long)(di + 1), len);
+               filetype = btrfs_dir_type(node, di);
+
+               btrfs_init_path(&path);
+               btrfs_dir_item_key_to_cpu(node, di, &location);
+
+               /* Ignore related ROOT_ITEM check */
+               if (location.type == BTRFS_ROOT_ITEM_KEY)
+                       goto next;
+
+               /* Check relative INODE_ITEM(existence/filetype) */
+               ret = btrfs_search_slot(NULL, root, &location, &path, 0, 0);
+               if (ret) {
+                       err |= INODE_ITEM_MISSING;
+                       error("root %llu %s[%llu %llu] couldn't find relative INODE_ITEM[%llu] namelen %u filename %s filetype %x",
+                             root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
+                             "DIR_ITEM" : "DIR_INDEX", key->objectid,
+                             key->offset, location.objectid, name_len,
+                             namebuf, filetype);
+                       goto next;
+               }
+
+               ii = btrfs_item_ptr(path.nodes[0], path.slots[0],
+                                   struct btrfs_inode_item);
+               mode = btrfs_inode_mode(path.nodes[0], ii);
+
+               if (imode_to_type(mode) != filetype) {
+                       err |= INODE_ITEM_MISMATCH;
+                       error("root %llu %s[%llu %llu] relative INODE_ITEM filetype mismatch namelen %u filename %s filetype %d",
+                             root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
+                             "DIR_ITEM" : "DIR_INDEX", key->objectid,
+                             key->offset, name_len, namebuf, filetype);
+               }
+
+               /* Check relative INODE_REF/INODE_EXTREF */
+               location.type = BTRFS_INODE_REF_KEY;
+               location.offset = key->objectid;
+               ret = find_inode_ref(root, &location, namebuf, len,
+                                      index, ext_ref);
+               err |= ret;
+               if (ret & INODE_REF_MISSING)
+                       error("root %llu %s[%llu %llu] relative INODE_REF missing namelen %u filename %s filetype %d",
+                             root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
+                             "DIR_ITEM" : "DIR_INDEX", key->objectid,
+                             key->offset, name_len, namebuf, filetype);
+
+next:
+               btrfs_release_path(&path);
+               len = sizeof(*di) + name_len + data_len;
+               di = (struct btrfs_dir_item *)((char *)di + len);
+               cur += len;
+
+               if (key->type == BTRFS_DIR_INDEX_KEY && cur < total) {
+                       error("root %llu DIR_INDEX[%llu %llu] should contain only one entry",
+                             root->objectid, key->objectid, key->offset);
+                       break;
+               }
+       }
+
+       return err;
+}
+
+/*
+ * Check file extent datasum/hole, update the size of the file extents,
+ * check and update the last offset of the file extent.
+ *
+ * @root:      the root of fs/file tree.
+ * @fkey:      the key of the file extent.
+ * @nodatasum: INODE_NODATASUM feature.
+ * @size:      the sum of all EXTENT_DATA items size for this inode.
+ * @end:       the offset of the last extent.
+ *
+ * Return 0 if no error occurred.
+ */
+static int check_file_extent(struct btrfs_root *root, struct btrfs_key *fkey,
+                            struct extent_buffer *node, int slot,
+                            unsigned int nodatasum, u64 *size, u64 *end)
+{
+       struct btrfs_file_extent_item *fi;
+       u64 disk_bytenr;
+       u64 disk_num_bytes;
+       u64 extent_num_bytes;
+       u64 found;
+       unsigned int extent_type;
+       unsigned int is_hole;
+       int ret;
+       int err = 0;
+
+       fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
+
+       extent_type = btrfs_file_extent_type(node, fi);
+       /* Skip if file extent is inline */
+       if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+               struct btrfs_item *e = btrfs_item_nr(slot);
+               u32 item_inline_len;
+
+               item_inline_len = btrfs_file_extent_inline_item_len(node, e);
+               extent_num_bytes = btrfs_file_extent_inline_len(node, slot, fi);
+               if (extent_num_bytes == 0 ||
+                   extent_num_bytes != item_inline_len)
+                       err |= FILE_EXTENT_ERROR;
+               *size += extent_num_bytes;
+               return err;
+       }
+
+       /* Check extent type */
+       if (extent_type != BTRFS_FILE_EXTENT_REG &&
+                       extent_type != BTRFS_FILE_EXTENT_PREALLOC) {
+               err |= FILE_EXTENT_ERROR;
+               error("root %llu EXTENT_DATA[%llu %llu] type bad",
+                     root->objectid, fkey->objectid, fkey->offset);
+               return err;
+       }
+
+       /* Check REG_EXTENT/PREALLOC_EXTENT */
+       disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
+       disk_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
+       extent_num_bytes = btrfs_file_extent_num_bytes(node, fi);
+       is_hole = (disk_bytenr == 0) && (disk_num_bytes == 0);
+
+       /* Check EXTENT_DATA datasum */
+       ret = count_csum_range(root, disk_bytenr, disk_num_bytes, &found);
+       if (found > 0 && nodatasum) {
+               err |= ODD_CSUM_ITEM;
+               error("root %llu EXTENT_DATA[%llu %llu] nodatasum shouldn't have datasum",
+                     root->objectid, fkey->objectid, fkey->offset);
+       } else if (extent_type == BTRFS_FILE_EXTENT_REG && !nodatasum &&
+                  !is_hole &&
+                  (ret < 0 || found == 0 || found < disk_num_bytes)) {
+               err |= CSUM_ITEM_MISSING;
+               error("root %llu EXTENT_DATA[%llu %llu] datasum missing",
+                     root->objectid, fkey->objectid, fkey->offset);
+       } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC && found > 0) {
+               err |= ODD_CSUM_ITEM;
+               error("root %llu EXTENT_DATA[%llu %llu] prealloc shouldn't have datasum",
+                     root->objectid, fkey->objectid, fkey->offset);
+       }
+
+       /* Check EXTENT_DATA hole */
+       if (no_holes && is_hole) {
+               err |= FILE_EXTENT_ERROR;
+               error("root %llu EXTENT_DATA[%llu %llu] shouldn't be hole",
+                     root->objectid, fkey->objectid, fkey->offset);
+       } else if (!no_holes && *end != fkey->offset) {
+               err |= FILE_EXTENT_ERROR;
+               error("root %llu EXTENT_DATA[%llu %llu] interrupt",
+                     root->objectid, fkey->objectid, fkey->offset);
+       }
+
+       *end += extent_num_bytes;
+       if (!is_hole)
+               *size += extent_num_bytes;
+
+       return err;
+}
+
+/*
+ * Check INODE_ITEM and related ITEMs (the same inode number)
+ * 1. check link count
+ * 2. check inode ref/extref
+ * 3. check dir item/index
+ *
+ * @ext_ref:   the EXTENDED_IREF feature
+ *
+ * Return 0 if no error occurred.
+ * Return >0 for error or hit the traversal is done(by error bitmap)
+ */
+static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
+                           unsigned int ext_ref)
+{
+       struct extent_buffer *node;
+       struct btrfs_inode_item *ii;
+       struct btrfs_key key;
+       u64 inode_id;
+       u32 mode;
+       u64 nlink;
+       u64 nbytes;
+       u64 isize;
+       u64 size = 0;
+       u64 refs = 0;
+       u64 extent_end = 0;
+       u64 extent_size = 0;
+       unsigned int dir;
+       unsigned int nodatasum;
+       int slot;
+       int ret;
+       int err = 0;
+
+       node = path->nodes[0];
+       slot = path->slots[0];
+
+       btrfs_item_key_to_cpu(node, &key, slot);
+       inode_id = key.objectid;
+
+       if (inode_id == BTRFS_ORPHAN_OBJECTID) {
+               ret = btrfs_next_item(root, path);
+               if (ret > 0)
+                       err |= LAST_ITEM;
+               return err;
+       }
+
+       ii = btrfs_item_ptr(node, slot, struct btrfs_inode_item);
+       isize = btrfs_inode_size(node, ii);
+       nbytes = btrfs_inode_nbytes(node, ii);
+       mode = btrfs_inode_mode(node, ii);
+       dir = imode_to_type(mode) == BTRFS_FT_DIR;
+       nlink = btrfs_inode_nlink(node, ii);
+       nodatasum = btrfs_inode_flags(node, ii) & BTRFS_INODE_NODATASUM;
+
+       while (1) {
+               ret = btrfs_next_item(root, path);
+               if (ret < 0) {
+                       /* out will fill 'err' rusing current statistics */
+                       goto out;
+               } else if (ret > 0) {
+                       err |= LAST_ITEM;
+                       goto out;
+               }
+
+               node = path->nodes[0];
+               slot = path->slots[0];
+               btrfs_item_key_to_cpu(node, &key, slot);
+               if (key.objectid != inode_id)
+                       goto out;
+
+               switch (key.type) {
+               case BTRFS_INODE_REF_KEY:
+                       ret = check_inode_ref(root, &key, node, slot, &refs,
+                                             mode);
+                       err |= ret;
+                       break;
+               case BTRFS_INODE_EXTREF_KEY:
+                       if (key.type == BTRFS_INODE_EXTREF_KEY && !ext_ref)
+                               warning("root %llu EXTREF[%llu %llu] isn't supported",
+                                       root->objectid, key.objectid,
+                                       key.offset);
+                       ret = check_inode_extref(root, &key, node, slot, &refs,
+                                                mode);
+                       err |= ret;
+                       break;
+               case BTRFS_DIR_ITEM_KEY:
+               case BTRFS_DIR_INDEX_KEY:
+                       if (!dir) {
+                               warning("root %llu INODE[%llu] mode %u shouldn't have DIR_INDEX[%llu %llu]",
+                                       root->objectid, inode_id,
+                                       imode_to_type(mode), key.objectid,
+                                       key.offset);
+                       }
+                       ret = check_dir_item(root, &key, node, slot, &size,
+                                            ext_ref);
+                       err |= ret;
+                       break;
+               case BTRFS_EXTENT_DATA_KEY:
+                       if (dir) {
+                               warning("root %llu DIR INODE[%llu] shouldn't EXTENT_DATA[%llu %llu]",
+                                       root->objectid, inode_id, key.objectid,
+                                       key.offset);
+                       }
+                       ret = check_file_extent(root, &key, node, slot,
+                                               nodatasum, &extent_size,
+                                               &extent_end);
+                       err |= ret;
+                       break;
+               case BTRFS_XATTR_ITEM_KEY:
+                       break;
+               default:
+                       error("ITEM[%llu %u %llu] UNKNOWN TYPE",
+                             key.objectid, key.type, key.offset);
+               }
+       }
+
+out:
+       /* verify INODE_ITEM nlink/isize/nbytes */
+       if (dir) {
+               if (nlink != 1) {
+                       err |= LINK_COUNT_ERROR;
+                       error("root %llu DIR INODE[%llu] shouldn't have more than one link(%llu)",
+                             root->objectid, inode_id, nlink);
+               }
+
+               /*
+                * Just a warning, as dir inode nbytes is just an
+                * instructive value.
+                */
+               if (!IS_ALIGNED(nbytes, root->nodesize)) {
+                       warning("root %llu DIR INODE[%llu] nbytes should be aligned to %u",
+                               root->objectid, inode_id, root->nodesize);
+               }
+
+               if (isize != size) {
+                       err |= ISIZE_ERROR;
+                       error("root %llu DIR INODE [%llu] size(%llu) not equal to %llu",
+                             root->objectid, inode_id, isize, size);
+               }
+       } else {
+               if (nlink != refs) {
+                       err |= LINK_COUNT_ERROR;
+                       error("root %llu INODE[%llu] nlink(%llu) not equal to inode_refs(%llu)",
+                             root->objectid, inode_id, nlink, refs);
+               } else if (!nlink) {
+                       err |= ORPHAN_ITEM;
+               }
+
+               if (!nbytes && !no_holes && extent_end < isize) {
+                       err |= NBYTES_ERROR;
+                       error("root %llu INODE[%llu] size (%llu) should have a file extent hole",
+                             root->objectid, inode_id, isize);
+               }
+
+               if (nbytes != extent_size) {
+                       err |= NBYTES_ERROR;
+                       error("root %llu INODE[%llu] nbytes(%llu) not equal to extent_size(%llu)",
+                             root->objectid, inode_id, nbytes, extent_size);
+               }
+       }
+
+       return err;
+}
+
+/*
+ * Iterate all item on the tree and call check_inode_item() to check.
+ *
+ * @root:      the root of the tree to be checked.
+ * @ext_ref:   the EXTENDED_IREF feature
+ *
+ * Return 0 if no error found.
+ * Return <0 for error.
+ * All internal error bitmap will be converted to -EIO, to avoid
+ * mixing negative and postive return value.
+ */
+static int check_fs_root_v2(struct btrfs_root *root, unsigned int ext_ref)
+{
+       struct btrfs_path *path;
+       struct btrfs_key key;
+       u64 inode_id;
+       int ret, err = 0;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       key.objectid = 0;
+       key.type = 0;
+       key.offset = 0;
+
+       ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+       if (ret < 0)
+               goto out;
+
+       while (1) {
+               btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+               /*
+                * All check must start with inode item, skip if not
+                */
+               if (key.type == BTRFS_INODE_ITEM_KEY) {
+                       ret = check_inode_item(root, path, ext_ref);
+                       err |= ret;
+                       if (err & LAST_ITEM)
+                               goto out;
+                       continue;
+               }
+               error("root %llu ITEM[%llu %u %llu] isn't INODE_ITEM, skip to next inode",
+                     root->objectid, key.objectid, key.type,
+                     key.offset);
 
-       err = merge_root_recs(root, &root_node.root_cache, root_cache);
-       if (err < 0)
-               ret = err;
+               err |= NO_INODE_ITEM;
+               inode_id = key.objectid;
 
-       if (root_node.current) {
-               root_node.current->checked = 1;
-               maybe_free_inode_rec(&root_node.inode_cache,
-                               root_node.current);
+               /*
+                * skip to next inode
+                * TODO: Maybe search_slot() will be faster?
+                */
+               do {
+                       ret = btrfs_next_item(root, path);
+                       if (ret > 0) {
+                               goto out;
+                       } else if (ret < 0) {
+                               err = ret;
+                               goto out;
+                       }
+                       btrfs_item_key_to_cpu(path->nodes[0], &key,
+                                             path->slots[0]);
+               } while (inode_id == key.objectid);
        }
 
-       err = check_inode_recs(root, &root_node.inode_cache);
-       if (!ret)
-               ret = err;
-
-       free_corrupt_blocks_tree(&corrupt_blocks);
-       root->fs_info->corrupt_blocks = NULL;
-       free_orphan_data_extents(&root->orphan_data_extents);
+out:
+       err &= ~LAST_ITEM;
+       if (err && !ret)
+               ret = -EIO;
+       btrfs_free_path(path);
        return ret;
 }
 
-static int fs_root_objectid(u64 objectid)
+/*
+ * Find the relative ref for root_ref and root_backref.
+ *
+ * @root:      the root of the root tree.
+ * @ref_key:   the key of the root ref.
+ *
+ * Return 0 if no error occurred.
+ */
+static int check_root_ref(struct btrfs_root *root, struct btrfs_key *ref_key,
+                         struct extent_buffer *node, int slot)
 {
-       if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
-           objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
-               return 1;
-       return is_fstree(objectid);
+       struct btrfs_path *path;
+       struct btrfs_key key;
+       struct btrfs_root_ref *ref;
+       struct btrfs_root_ref *backref;
+       char ref_name[BTRFS_NAME_LEN] = {0};
+       char backref_name[BTRFS_NAME_LEN] = {0};
+       u64 ref_dirid;
+       u64 ref_seq;
+       u32 ref_namelen;
+       u64 backref_dirid;
+       u64 backref_seq;
+       u32 backref_namelen;
+       u32 len;
+       int ret;
+       int err = 0;
+
+       ref = btrfs_item_ptr(node, slot, struct btrfs_root_ref);
+       ref_dirid = btrfs_root_ref_dirid(node, ref);
+       ref_seq = btrfs_root_ref_sequence(node, ref);
+       ref_namelen = btrfs_root_ref_name_len(node, ref);
+
+       if (ref_namelen <= BTRFS_NAME_LEN) {
+               len = ref_namelen;
+       } else {
+               len = BTRFS_NAME_LEN;
+               warning("%s[%llu %llu] ref_name too long",
+                       ref_key->type == BTRFS_ROOT_REF_KEY ?
+                       "ROOT_REF" : "ROOT_BACKREF", ref_key->objectid,
+                       ref_key->offset);
+       }
+       read_extent_buffer(node, ref_name, (unsigned long)(ref + 1), len);
+
+       /* Find relative root_ref */
+       key.objectid = ref_key->offset;
+       key.type = BTRFS_ROOT_BACKREF_KEY + BTRFS_ROOT_REF_KEY - ref_key->type;
+       key.offset = ref_key->objectid;
+
+       path = btrfs_alloc_path();
+       ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+       if (ret) {
+               err |= ROOT_REF_MISSING;
+               error("%s[%llu %llu] couldn't find relative ref",
+                     ref_key->type == BTRFS_ROOT_REF_KEY ?
+                     "ROOT_REF" : "ROOT_BACKREF",
+                     ref_key->objectid, ref_key->offset);
+               goto out;
+       }
+
+       backref = btrfs_item_ptr(path->nodes[0], path->slots[0],
+                                struct btrfs_root_ref);
+       backref_dirid = btrfs_root_ref_dirid(path->nodes[0], backref);
+       backref_seq = btrfs_root_ref_sequence(path->nodes[0], backref);
+       backref_namelen = btrfs_root_ref_name_len(path->nodes[0], backref);
+
+       if (backref_namelen <= BTRFS_NAME_LEN) {
+               len = backref_namelen;
+       } else {
+               len = BTRFS_NAME_LEN;
+               warning("%s[%llu %llu] ref_name too long",
+                       key.type == BTRFS_ROOT_REF_KEY ?
+                       "ROOT_REF" : "ROOT_BACKREF",
+                       key.objectid, key.offset);
+       }
+       read_extent_buffer(path->nodes[0], backref_name,
+                          (unsigned long)(backref + 1), len);
+
+       if (ref_dirid != backref_dirid || ref_seq != backref_seq ||
+           ref_namelen != backref_namelen ||
+           strncmp(ref_name, backref_name, len)) {
+               err |= ROOT_REF_MISMATCH;
+               error("%s[%llu %llu] mismatch relative ref",
+                     ref_key->type == BTRFS_ROOT_REF_KEY ?
+                     "ROOT_REF" : "ROOT_BACKREF",
+                     ref_key->objectid, ref_key->offset);
+       }
+out:
+       btrfs_free_path(path);
+       return err;
 }
 
-static int check_fs_roots(struct btrfs_root *root,
-                         struct cache_tree *root_cache)
+/*
+ * Check all fs/file tree in low_memory mode.
+ *
+ * 1. for fs tree root item, call check_fs_root_v2()
+ * 2. for fs tree root ref/backref, call check_root_ref()
+ *
+ * Return 0 if no error occurred.
+ */
+static int check_fs_roots_v2(struct btrfs_fs_info *fs_info)
 {
-       struct btrfs_path path;
+       struct btrfs_root *tree_root = fs_info->tree_root;
+       struct btrfs_root *cur_root = NULL;
+       struct btrfs_path *path;
        struct btrfs_key key;
-       struct walk_control wc;
-       struct extent_buffer *leaf, *tree_node;
-       struct btrfs_root *tmp_root;
-       struct btrfs_root *tree_root = root->fs_info->tree_root;
+       struct extent_buffer *node;
+       unsigned int ext_ref;
+       int slot;
        int ret;
        int err = 0;
 
-       if (ctx.progress_enabled) {
-               ctx.tp = TASK_FS_ROOTS;
-               task_start(ctx.info);
-       }
+       ext_ref = btrfs_fs_incompat(fs_info,
+                                   BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF);
 
-       /*
-        * Just in case we made any changes to the extent tree that weren't
-        * reflected into the free space cache yet.
-        */
-       if (repair)
-               reset_cached_block_groups(root->fs_info);
-       memset(&wc, 0, sizeof(wc));
-       cache_tree_init(&wc.shared);
-       btrfs_init_path(&path);
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
 
-again:
+       key.objectid = BTRFS_FS_TREE_OBJECTID;
        key.offset = 0;
-       key.objectid = 0;
        key.type = BTRFS_ROOT_ITEM_KEY;
-       ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
+
+       ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
        if (ret < 0) {
-               err = 1;
+               err = ret;
+               goto out;
+       } else if (ret > 0) {
+               err = -ENOENT;
                goto out;
        }
-       tree_node = tree_root->node;
+
        while (1) {
-               if (tree_node != tree_root->node) {
-                       free_root_recs_tree(root_cache);
-                       btrfs_release_path(&path);
-                       goto again;
-               }
-               leaf = path.nodes[0];
-               if (path.slots[0] >= btrfs_header_nritems(leaf)) {
-                       ret = btrfs_next_leaf(tree_root, &path);
-                       if (ret) {
-                               if (ret < 0)
-                                       err = 1;
-                               break;
-                       }
-                       leaf = path.nodes[0];
-               }
-               btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
+               node = path->nodes[0];
+               slot = path->slots[0];
+               btrfs_item_key_to_cpu(node, &key, slot);
+               if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
+                       goto out;
                if (key.type == BTRFS_ROOT_ITEM_KEY &&
                    fs_root_objectid(key.objectid)) {
                        if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
-                               tmp_root = btrfs_read_fs_root_no_cache(
-                                               root->fs_info, &key);
+                               cur_root = btrfs_read_fs_root_no_cache(fs_info,
+                                                                      &key);
                        } else {
                                key.offset = (u64)-1;
-                               tmp_root = btrfs_read_fs_root(
-                                               root->fs_info, &key);
+                               cur_root = btrfs_read_fs_root(fs_info, &key);
                        }
-                       if (IS_ERR(tmp_root)) {
-                               err = 1;
+
+                       if (IS_ERR(cur_root)) {
+                               error("Fail to read fs/subvol tree: %lld",
+                                     key.objectid);
+                               err = -EIO;
                                goto next;
                        }
-                       ret = check_fs_root(tmp_root, root_cache, &wc);
-                       if (ret == -EAGAIN) {
-                               free_root_recs_tree(root_cache);
-                               btrfs_release_path(&path);
-                               goto again;
-                       }
-                       if (ret)
-                               err = 1;
+
+                       ret = check_fs_root_v2(cur_root, ext_ref);
+                       err |= ret;
+
                        if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
-                               btrfs_free_fs_root(tmp_root);
+                               btrfs_free_fs_root(cur_root);
                } else if (key.type == BTRFS_ROOT_REF_KEY ||
-                          key.type == BTRFS_ROOT_BACKREF_KEY) {
-                       process_root_ref(leaf, path.slots[0], &key,
-                                        root_cache);
+                               key.type == BTRFS_ROOT_BACKREF_KEY) {
+                       ret = check_root_ref(tree_root, &key, node, slot);
+                       err |= ret;
                }
 next:
-               path.slots[0]++;
+               ret = btrfs_next_item(tree_root, path);
+               if (ret > 0)
+                       goto out;
+               if (ret < 0) {
+                       err = ret;
+                       goto out;
+               }
        }
-out:
-       btrfs_release_path(&path);
-       if (err)
-               free_extent_cache_tree(&wc.shared);
-       if (!cache_tree_empty(&wc.shared))
-               fprintf(stderr, "warning line %d\n", __LINE__);
-
-       task_stop(ctx.info);
 
+out:
+       btrfs_free_path(path);
        return err;
 }
 
@@ -4317,7 +5389,7 @@ static int try_to_fix_bad_block(struct btrfs_root *root,
        struct ulist *roots;
        struct ulist_node *node;
        struct btrfs_root *search_root;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct ulist_iterator iter;
        struct btrfs_key root_key, key;
        int ret;
@@ -4326,17 +5398,11 @@ static int try_to_fix_bad_block(struct btrfs_root *root,
            status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
                return -EIO;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -EIO;
-
-       ret = btrfs_find_all_roots(NULL, root->fs_info, buf->start,
-                                  0, &roots);
-       if (ret) {
-               btrfs_free_path(path);
+       ret = btrfs_find_all_roots(NULL, root->fs_info, buf->start, 0, &roots);
+       if (ret)
                return -EIO;
-       }
 
+       btrfs_init_path(&path);
        ULIST_ITER_INIT(&iter);
        while ((node = ulist_next(roots, &iter))) {
                root_key.objectid = node->val;
@@ -4356,31 +5422,31 @@ static int try_to_fix_bad_block(struct btrfs_root *root,
                        break;
                }
 
-               path->lowest_level = btrfs_header_level(buf);
-               path->skip_check_block = 1;
-               if (path->lowest_level)
+               path.lowest_level = btrfs_header_level(buf);
+               path.skip_check_block = 1;
+               if (path.lowest_level)
                        btrfs_node_key_to_cpu(buf, &key, 0);
                else
                        btrfs_item_key_to_cpu(buf, &key, 0);
-               ret = btrfs_search_slot(trans, search_root, &key, path, 0, 1);
+               ret = btrfs_search_slot(trans, search_root, &key, &path, 0, 1);
                if (ret) {
                        ret = -EIO;
                        btrfs_commit_transaction(trans, search_root);
                        break;
                }
                if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
-                       ret = fix_key_order(trans, search_root, path);
+                       ret = fix_key_order(trans, search_root, &path);
                else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
-                       ret = fix_item_offset(trans, search_root, path);
+                       ret = fix_item_offset(trans, search_root, &path);
                if (ret) {
                        btrfs_commit_transaction(trans, search_root);
                        break;
                }
-               btrfs_release_path(path);
+               btrfs_release_path(&path);
                btrfs_commit_transaction(trans, search_root);
        }
        ulist_free(roots);
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -5563,31 +6629,27 @@ static int check_cache_range(struct btrfs_root *root,
 static int verify_space_cache(struct btrfs_root *root,
                              struct btrfs_block_group_cache *cache)
 {
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct extent_buffer *leaf;
        struct btrfs_key key;
        u64 last;
        int ret = 0;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
        root = root->fs_info->extent_root;
 
        last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 
+       btrfs_init_path(&path);
        key.objectid = last;
        key.offset = 0;
        key.type = BTRFS_EXTENT_ITEM_KEY;
-
-       ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+       ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
        if (ret < 0)
                goto out;
        ret = 0;
        while (1) {
-               if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
-                       ret = btrfs_next_leaf(root, path);
+               if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
+                       ret = btrfs_next_leaf(root, &path);
                        if (ret < 0)
                                goto out;
                        if (ret > 0) {
@@ -5595,13 +6657,13 @@ static int verify_space_cache(struct btrfs_root *root,
                                break;
                        }
                }
-               leaf = path->nodes[0];
-               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+               leaf = path.nodes[0];
+               btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
                if (key.objectid >= cache->key.offset + cache->key.objectid)
                        break;
                if (key.type != BTRFS_EXTENT_ITEM_KEY &&
                    key.type != BTRFS_METADATA_ITEM_KEY) {
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
 
@@ -5610,7 +6672,7 @@ static int verify_space_cache(struct btrfs_root *root,
                                last = key.objectid + key.offset;
                        else
                                last = key.objectid + root->nodesize;
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
 
@@ -5622,7 +6684,7 @@ static int verify_space_cache(struct btrfs_root *root,
                        last = key.objectid + key.offset;
                else
                        last = key.objectid + root->nodesize;
-               path->slots[0]++;
+               path.slots[0]++;
        }
 
        if (last < cache->key.objectid + cache->key.offset)
@@ -5631,7 +6693,7 @@ static int verify_space_cache(struct btrfs_root *root,
                                        cache->key.offset - last);
 
 out:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
 
        if (!ret &&
            !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
@@ -5788,33 +6850,28 @@ out:
 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
                               u64 num_bytes)
 {
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct extent_buffer *leaf;
        struct btrfs_key key;
        int ret;
 
-       path = btrfs_alloc_path();
-       if (!path) {
-               fprintf(stderr, "Error allocating path\n");
-               return -ENOMEM;
-       }
-
+       btrfs_init_path(&path);
        key.objectid = bytenr;
        key.type = BTRFS_EXTENT_ITEM_KEY;
        key.offset = (u64)-1;
 
 again:
-       ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
+       ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, &path,
                                0, 0);
        if (ret < 0) {
                fprintf(stderr, "Error looking up extent record %d\n", ret);
-               btrfs_free_path(path);
+               btrfs_release_path(&path);
                return ret;
        } else if (ret) {
-               if (path->slots[0] > 0) {
-                       path->slots[0]--;
+               if (path.slots[0] > 0) {
+                       path.slots[0]--;
                } else {
-                       ret = btrfs_prev_leaf(root, path);
+                       ret = btrfs_prev_leaf(root, &path);
                        if (ret < 0) {
                                goto out;
                        } else if (ret > 0) {
@@ -5824,7 +6881,7 @@ again:
                }
        }
 
-       btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+       btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
 
        /*
         * Block group items come before extent items if they have the same
@@ -5835,10 +6892,10 @@ again:
         * EXTENT_ITEM_KEY please?
         */
        while (key.type > BTRFS_EXTENT_ITEM_KEY) {
-               if (path->slots[0] > 0) {
-                       path->slots[0]--;
+               if (path.slots[0] > 0) {
+                       path.slots[0]--;
                } else {
-                       ret = btrfs_prev_leaf(root, path);
+                       ret = btrfs_prev_leaf(root, &path);
                        if (ret < 0) {
                                goto out;
                        } else if (ret > 0) {
@@ -5846,29 +6903,29 @@ again:
                                goto out;
                        }
                }
-               btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+               btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
        }
 
        while (num_bytes) {
-               if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
-                       ret = btrfs_next_leaf(root, path);
+               if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
+                       ret = btrfs_next_leaf(root, &path);
                        if (ret < 0) {
                                fprintf(stderr, "Error going to next leaf "
                                        "%d\n", ret);
-                               btrfs_free_path(path);
+                               btrfs_release_path(&path);
                                return ret;
                        } else if (ret) {
                                break;
                        }
                }
-               leaf = path->nodes[0];
-               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+               leaf = path.nodes[0];
+               btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
                if (key.type != BTRFS_EXTENT_ITEM_KEY) {
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
                if (key.objectid + key.offset < bytenr) {
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
                if (key.objectid > bytenr + num_bytes)
@@ -5901,7 +6958,7 @@ again:
                                 * in real life, but no harm in coding it up
                                 * anyway just in case.
                                 */
-                               btrfs_release_path(path);
+                               btrfs_release_path(&path);
                                ret = check_extent_exists(root, new_start,
                                                          new_bytes);
                                if (ret) {
@@ -5914,7 +6971,7 @@ again:
                        }
                        num_bytes = key.objectid - bytenr;
                }
-               path->slots[0]++;
+               path.slots[0]++;
        }
        ret = 0;
 
@@ -5925,13 +6982,13 @@ out:
                ret = 1;
        }
 
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
 static int check_csums(struct btrfs_root *root)
 {
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct extent_buffer *leaf;
        struct btrfs_key key;
        u64 offset = 0, num_bytes = 0;
@@ -5947,28 +7004,24 @@ static int check_csums(struct btrfs_root *root)
                return -ENOENT;
        }
 
+       btrfs_init_path(&path);
        key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
        key.type = BTRFS_EXTENT_CSUM_KEY;
        key.offset = 0;
-
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
-       ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+       ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
        if (ret < 0) {
                fprintf(stderr, "Error searching csum tree %d\n", ret);
-               btrfs_free_path(path);
+               btrfs_release_path(&path);
                return ret;
        }
 
-       if (ret > 0 && path->slots[0])
-               path->slots[0]--;
+       if (ret > 0 && path.slots[0])
+               path.slots[0]--;
        ret = 0;
 
        while (1) {
-               if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
-                       ret = btrfs_next_leaf(root, path);
+               if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
+                       ret = btrfs_next_leaf(root, &path);
                        if (ret < 0) {
                                fprintf(stderr, "Error going to next leaf "
                                        "%d\n", ret);
@@ -5977,19 +7030,19 @@ static int check_csums(struct btrfs_root *root)
                        if (ret)
                                break;
                }
-               leaf = path->nodes[0];
+               leaf = path.nodes[0];
 
-               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+               btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
                if (key.type != BTRFS_EXTENT_CSUM_KEY) {
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
 
-               data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
+               data_len = (btrfs_item_size_nr(leaf, path.slots[0]) /
                              csum_size) * root->sectorsize;
                if (!check_data_csum)
                        goto skip_csum_check;
-               leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
+               leaf_offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
                ret = check_extent_csums(root, key.offset, data_len,
                                         leaf_offset, leaf);
                if (ret)
@@ -6009,10 +7062,10 @@ skip_csum_check:
                        num_bytes = 0;
                }
                num_bytes += data_len;
-               path->slots[0]++;
+               path.slots[0]++;
        }
 
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return errors;
 }
 
@@ -6706,7 +7759,6 @@ static int record_extent(struct btrfs_trans_handle *trans,
        struct extent_buffer *leaf;
        struct btrfs_key ins_key;
        struct btrfs_extent_item *ei;
-       struct tree_backref *tback;
        struct data_backref *dback;
        struct btrfs_tree_block_info *bi;
 
@@ -6742,7 +7794,6 @@ static int record_extent(struct btrfs_trans_handle *trans,
                } else {
                        struct btrfs_disk_key copy_key;;
 
-                       tback = to_tree_backref(back);
                        bi = (struct btrfs_tree_block_info *)(ei + 1);
                        memset_extent_buffer(leaf, 0, (unsigned long)bi,
                                             sizeof(*bi));
@@ -6808,6 +7859,7 @@ static int record_extent(struct btrfs_trans_handle *trans,
                                dback->found_ref);
        } else {
                u64 parent;
+               struct tree_backref *tback;
 
                tback = to_tree_backref(back);
                if (back->full_backref)
@@ -6845,11 +7897,6 @@ static struct extent_entry *find_most_right_entry(struct list_head *entries)
        struct extent_entry *entry, *best = NULL, *prev = NULL;
 
        list_for_each_entry(entry, entries, list) {
-               if (!prev) {
-                       prev = entry;
-                       continue;
-               }
-
                /*
                 * If there are as many broken entries as entries then we know
                 * not to trust this particular entry.
@@ -6858,6 +7905,16 @@ static struct extent_entry *find_most_right_entry(struct list_head *entries)
                        continue;
 
                /*
+                * Special case, when there are only two entries and 'best' is
+                * the first one
+                */
+               if (!prev) {
+                       best = entry;
+                       prev = entry;
+                       continue;
+               }
+
+               /*
                 * If our current entry == best then we can't be sure our best
                 * is really the best, so we need to keep searching.
                 */
@@ -7303,17 +8360,13 @@ static int delete_duplicate_records(struct btrfs_root *root,
 {
        struct btrfs_trans_handle *trans;
        LIST_HEAD(delete_list);
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct extent_record *tmp, *good, *n;
        int nr_del = 0;
        int ret = 0, err;
        struct btrfs_key key;
 
-       path = btrfs_alloc_path();
-       if (!path) {
-               ret = -ENOMEM;
-               goto out;
-       }
+       btrfs_init_path(&path);
 
        good = rec;
        /* Find the record that covers all of the duplicates. */
@@ -7365,16 +8418,16 @@ static int delete_duplicate_records(struct btrfs_root *root,
                        abort();
                }
 
-               ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+               ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
                if (ret) {
                        if (ret > 0)
                                ret = -EINVAL;
                        break;
                }
-               ret = btrfs_del_item(trans, root, path);
+               ret = btrfs_del_item(trans, root, &path);
                if (ret)
                        break;
-               btrfs_release_path(path);
+               btrfs_release_path(&path);
                nr_del++;
        }
        err = btrfs_commit_transaction(trans, root);
@@ -7395,7 +8448,7 @@ out:
                free(tmp);
        }
 
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
 
        if (!ret && !nr_del)
                rec->num_duplicates = 0;
@@ -7508,15 +8561,13 @@ static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
        struct extent_backref *back;
        struct data_backref *dback;
        struct orphan_data_extent *orphan;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        int recorded_data_ref = 0;
        int ret = 0;
 
        if (rec->metadata)
                return 1;
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
+       btrfs_init_path(&path);
        list_for_each_entry(back, &rec->backrefs, list) {
                if (back->full_backref || !back->is_data ||
                    !back->found_extent_tree)
@@ -7538,8 +8589,8 @@ static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
                key.type = BTRFS_EXTENT_DATA_KEY;
                key.offset = dback->offset;
 
-               ret = btrfs_search_slot(NULL, dest_root, &key, path, 0, 0);
-               btrfs_release_path(path);
+               ret = btrfs_search_slot(NULL, dest_root, &key, &path, 0, 0);
+               btrfs_release_path(&path);
                /*
                 * For ret < 0, it's OK since the fs-tree may be corrupted,
                 * we need to record it for inode/file extent rebuild.
@@ -7566,7 +8617,7 @@ static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
                recorded_data_ref = 1;
        }
 out:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        if (!ret)
                return !recorded_data_ref;
        else
@@ -7584,7 +8635,7 @@ static int fixup_extent_refs(struct btrfs_fs_info *info,
 {
        struct btrfs_trans_handle *trans = NULL;
        int ret;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct list_head *cur = rec->backrefs.next;
        struct cache_extent *cache;
        struct extent_backref *back;
@@ -7594,10 +8645,7 @@ static int fixup_extent_refs(struct btrfs_fs_info *info,
        if (rec->flag_block_full_backref)
                flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
+       btrfs_init_path(&path);
        if (rec->refs != rec->extent_item_refs && !rec->metadata) {
                /*
                 * Sometimes the backrefs themselves are so broken they don't
@@ -7606,13 +8654,13 @@ static int fixup_extent_refs(struct btrfs_fs_info *info,
                 * them into the list if we find the backref so that
                 * verify_backrefs can figure out what to do.
                 */
-               ret = find_possible_backrefs(info, path, extent_cache, rec);
+               ret = find_possible_backrefs(info, &path, extent_cache, rec);
                if (ret < 0)
                        goto out;
        }
 
        /* step one, make sure all of the backrefs agree */
-       ret = verify_backrefs(info, path, rec);
+       ret = verify_backrefs(info, &path, rec);
        if (ret < 0)
                goto out;
 
@@ -7623,7 +8671,7 @@ static int fixup_extent_refs(struct btrfs_fs_info *info,
        }
 
        /* step two, delete all the existing records */
-       ret = delete_extent_records(trans, info->extent_root, path,
+       ret = delete_extent_records(trans, info->extent_root, &path,
                                    rec->start, rec->max_size);
 
        if (ret < 0)
@@ -7650,7 +8698,7 @@ static int fixup_extent_refs(struct btrfs_fs_info *info,
                        continue;
 
                rec->bad_full_backref = 0;
-               ret = record_extent(trans, info, path, rec, back, allocated, flags);
+               ret = record_extent(trans, info, &path, rec, back, allocated, flags);
                allocated = 1;
 
                if (ret)
@@ -7663,7 +8711,7 @@ out:
                        ret = err;
        }
 
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -7672,7 +8720,7 @@ static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
 {
        struct btrfs_trans_handle *trans;
        struct btrfs_root *root = fs_info->extent_root;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct btrfs_extent_item *ei;
        struct btrfs_key key;
        u64 flags;
@@ -7687,32 +8735,27 @@ static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
                key.offset = rec->max_size;
        }
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
        trans = btrfs_start_transaction(root, 0);
-       if (IS_ERR(trans)) {
-               btrfs_free_path(path);
+       if (IS_ERR(trans))
                return PTR_ERR(trans);
-       }
 
-       ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+       btrfs_init_path(&path);
+       ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
        if (ret < 0) {
-               btrfs_free_path(path);
+               btrfs_release_path(&path);
                btrfs_commit_transaction(trans, root);
                return ret;
        } else if (ret) {
                fprintf(stderr, "Didn't find extent for %llu\n",
                        (unsigned long long)rec->start);
-               btrfs_free_path(path);
+               btrfs_release_path(&path);
                btrfs_commit_transaction(trans, root);
                return -ENOENT;
        }
 
-       ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
+       ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
                            struct btrfs_extent_item);
-       flags = btrfs_extent_flags(path->nodes[0], ei);
+       flags = btrfs_extent_flags(path.nodes[0], ei);
        if (rec->flag_block_full_backref) {
                fprintf(stderr, "setting full backref on %llu\n",
                        (unsigned long long)key.objectid);
@@ -7722,9 +8765,9 @@ static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
                        (unsigned long long)key.objectid);
                flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
        }
-       btrfs_set_extent_flags(path->nodes[0], ei, flags);
-       btrfs_mark_buffer_dirty(path->nodes[0]);
-       btrfs_free_path(path);
+       btrfs_set_extent_flags(path.nodes[0], ei, flags);
+       btrfs_mark_buffer_dirty(path.nodes[0]);
+       btrfs_release_path(&path);
        return btrfs_commit_transaction(trans, root);
 }
 
@@ -10263,24 +11306,20 @@ static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
 static int reset_block_groups(struct btrfs_fs_info *fs_info)
 {
        struct btrfs_block_group_cache *cache;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct extent_buffer *leaf;
        struct btrfs_chunk *chunk;
        struct btrfs_key key;
        int ret;
        u64 start;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
+       btrfs_init_path(&path);
        key.objectid = 0;
        key.type = BTRFS_CHUNK_ITEM_KEY;
        key.offset = 0;
-
-       ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
+       ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, &path, 0, 0);
        if (ret < 0) {
-               btrfs_free_path(path);
+               btrfs_release_path(&path);
                return ret;
        }
 
@@ -10295,10 +11334,10 @@ static int reset_block_groups(struct btrfs_fs_info *fs_info)
 
        /* First we need to create the in-memory block groups */
        while (1) {
-               if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
-                       ret = btrfs_next_leaf(fs_info->chunk_root, path);
+               if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
+                       ret = btrfs_next_leaf(fs_info->chunk_root, &path);
                        if (ret < 0) {
-                               btrfs_free_path(path);
+                               btrfs_release_path(&path);
                                return ret;
                        }
                        if (ret) {
@@ -10306,15 +11345,14 @@ static int reset_block_groups(struct btrfs_fs_info *fs_info)
                                break;
                        }
                }
-               leaf = path->nodes[0];
-               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+               leaf = path.nodes[0];
+               btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
                if (key.type != BTRFS_CHUNK_ITEM_KEY) {
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
 
-               chunk = btrfs_item_ptr(leaf, path->slots[0],
-                                      struct btrfs_chunk);
+               chunk = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_chunk);
                btrfs_add_block_group(fs_info, 0,
                                      btrfs_chunk_type(leaf, chunk),
                                      key.objectid, key.offset,
@@ -10322,7 +11360,7 @@ static int reset_block_groups(struct btrfs_fs_info *fs_info)
                set_extent_dirty(&fs_info->free_space_cache, key.offset,
                                 key.offset + btrfs_chunk_length(leaf, chunk),
                                 GFP_NOFS);
-               path->slots[0]++;
+               path.slots[0]++;
        }
        start = 0;
        while (1) {
@@ -10333,7 +11371,7 @@ static int reset_block_groups(struct btrfs_fs_info *fs_info)
                start = cache->key.objectid + cache->key.offset;
        }
 
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return 0;
 }
 
@@ -10341,22 +11379,18 @@ static int reset_balance(struct btrfs_trans_handle *trans,
                         struct btrfs_fs_info *fs_info)
 {
        struct btrfs_root *root = fs_info->tree_root;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct extent_buffer *leaf;
        struct btrfs_key key;
        int del_slot, del_nr = 0;
        int ret;
        int found = 0;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
+       btrfs_init_path(&path);
        key.objectid = BTRFS_BALANCE_OBJECTID;
        key.type = BTRFS_BALANCE_ITEM_KEY;
        key.offset = 0;
-
-       ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+       ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
        if (ret) {
                if (ret > 0)
                        ret = 0;
@@ -10366,64 +11400,63 @@ static int reset_balance(struct btrfs_trans_handle *trans,
                        goto out;
        }
 
-       ret = btrfs_del_item(trans, root, path);
+       ret = btrfs_del_item(trans, root, &path);
        if (ret)
                goto out;
-       btrfs_release_path(path);
+       btrfs_release_path(&path);
 
        key.objectid = BTRFS_TREE_RELOC_OBJECTID;
        key.type = BTRFS_ROOT_ITEM_KEY;
        key.offset = 0;
-
-       ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+       ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
        if (ret < 0)
                goto out;
        while (1) {
-               if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
+               if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
                        if (!found)
                                break;
 
                        if (del_nr) {
-                               ret = btrfs_del_items(trans, root, path,
+                               ret = btrfs_del_items(trans, root, &path,
                                                      del_slot, del_nr);
                                del_nr = 0;
                                if (ret)
                                        goto out;
                        }
                        key.offset++;
-                       btrfs_release_path(path);
+                       btrfs_release_path(&path);
 
                        found = 0;
-                       ret = btrfs_search_slot(trans, root, &key, path,
+                       ret = btrfs_search_slot(trans, root, &key, &path,
                                                -1, 1);
                        if (ret < 0)
                                goto out;
                        continue;
                }
                found = 1;
-               leaf = path->nodes[0];
-               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+               leaf = path.nodes[0];
+               btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
                if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
                        break;
                if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
                if (!del_nr) {
-                       del_slot = path->slots[0];
+                       del_slot = path.slots[0];
                        del_nr = 1;
                } else {
                        del_nr++;
                }
-               path->slots[0]++;
+               path.slots[0]++;
        }
 
        if (del_nr) {
-               ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
+               ret = btrfs_del_items(trans, root, &path, del_slot, del_nr);
                if (ret)
                        goto out;
        }
-       btrfs_release_path(path);
+       btrfs_release_path(&path);
 
 reinit_data_reloc:
        key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
@@ -10441,7 +11474,7 @@ reinit_data_reloc:
                goto out;
        ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 out:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -10531,7 +11564,7 @@ static int reinit_extent_tree(struct btrfs_trans_handle *trans,
 
 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
 {
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct btrfs_trans_handle *trans;
        struct btrfs_key key;
        int ret;
@@ -10548,31 +11581,26 @@ static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb
                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);
+       if (IS_ERR(trans))
                return PTR_ERR(trans);
-       }
 
-       path->lowest_level = btrfs_header_level(eb);
-       if (path->lowest_level)
+       btrfs_init_path(&path);
+       path.lowest_level = btrfs_header_level(eb);
+       if (path.lowest_level)
                btrfs_node_key_to_cpu(eb, &key, 0);
        else
                btrfs_item_key_to_cpu(eb, &key, 0);
 
-       ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+       ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
        btrfs_commit_transaction(trans, root);
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
 {
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct btrfs_trans_handle *trans;
        struct btrfs_key key;
        int ret;
@@ -10590,26 +11618,21 @@ static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
                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);
+       if (IS_ERR(trans))
                return PTR_ERR(trans);
-       }
 
-       ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
+       btrfs_init_path(&path);
+       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);
+       ret = btrfs_del_item(trans, root, &path);
 out:
        btrfs_commit_transaction(trans, root);
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -10656,7 +11679,7 @@ static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
                                      struct btrfs_root *csum_root,
                                      struct btrfs_root *cur_root)
 {
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct btrfs_key key;
        struct extent_buffer *node;
        struct btrfs_file_extent_item *fi;
@@ -10666,30 +11689,25 @@ static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
        int slot = 0;
        int ret = 0;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
        buf = malloc(cur_root->fs_info->csum_root->sectorsize);
-       if (!buf) {
-               ret = -ENOMEM;
-               goto out;
-       }
+       if (!buf)
+               return -ENOMEM;
 
+       btrfs_init_path(&path);
        key.objectid = 0;
        key.offset = 0;
        key.type = 0;
-
-       ret = btrfs_search_slot(NULL, cur_root, &key, path, 0, 0);
+       ret = btrfs_search_slot(NULL, cur_root, &key, &path, 0, 0);
        if (ret < 0)
                goto out;
        /* Iterate all regular file extents and fill its csum */
        while (1) {
-               btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+               btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
 
                if (key.type != BTRFS_EXTENT_DATA_KEY)
                        goto next;
-               node = path->nodes[0];
-               slot = path->slots[0];
+               node = path.nodes[0];
+               slot = path.slots[0];
                fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
                if (btrfs_file_extent_type(node, fi) != BTRFS_FILE_EXTENT_REG)
                        goto next;
@@ -10706,7 +11724,7 @@ next:
                 * TODO: if next leaf is corrupted, jump to nearest next valid
                 * leaf.
                 */
-               ret = btrfs_next_item(cur_root, path);
+               ret = btrfs_next_item(cur_root, &path);
                if (ret < 0)
                        goto out;
                if (ret > 0) {
@@ -10716,7 +11734,7 @@ next:
        }
 
 out:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        free(buf);
        return ret;
 }
@@ -10725,7 +11743,7 @@ static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
                                  struct btrfs_root *csum_root)
 {
        struct btrfs_fs_info *fs_info = csum_root->fs_info;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct btrfs_root *tree_root = fs_info->tree_root;
        struct btrfs_root *cur_root;
        struct extent_buffer *node;
@@ -10733,15 +11751,11 @@ static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
        int slot = 0;
        int ret = 0;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
+       btrfs_init_path(&path);
        key.objectid = BTRFS_FS_TREE_OBJECTID;
        key.offset = 0;
        key.type = BTRFS_ROOT_ITEM_KEY;
-
-       ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
+       ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
        if (ret < 0)
                goto out;
        if (ret > 0) {
@@ -10750,8 +11764,8 @@ static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
        }
 
        while (1) {
-               node = path->nodes[0];
-               slot = path->slots[0];
+               node = path.nodes[0];
+               slot = path.slots[0];
                btrfs_item_key_to_cpu(node, &key, slot);
                if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
                        goto out;
@@ -10772,7 +11786,7 @@ static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
                if (ret < 0)
                        goto out;
 next:
-               ret = btrfs_next_item(tree_root, path);
+               ret = btrfs_next_item(tree_root, &path);
                if (ret > 0) {
                        ret = 0;
                        goto out;
@@ -10782,7 +11796,7 @@ next:
        }
 
 out:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -10790,36 +11804,32 @@ static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
                                      struct btrfs_root *csum_root)
 {
        struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        struct btrfs_extent_item *ei;
        struct extent_buffer *leaf;
        char *buf;
        struct btrfs_key key;
        int ret;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
+       btrfs_init_path(&path);
        key.objectid = 0;
        key.type = BTRFS_EXTENT_ITEM_KEY;
        key.offset = 0;
-
-       ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
+       ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
        if (ret < 0) {
-               btrfs_free_path(path);
+               btrfs_release_path(&path);
                return ret;
        }
 
        buf = malloc(csum_root->sectorsize);
        if (!buf) {
-               btrfs_free_path(path);
+               btrfs_release_path(&path);
                return -ENOMEM;
        }
 
        while (1) {
-               if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
-                       ret = btrfs_next_leaf(extent_root, path);
+               if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
+                       ret = btrfs_next_leaf(extent_root, &path);
                        if (ret < 0)
                                break;
                        if (ret) {
@@ -10827,19 +11837,19 @@ static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
                                break;
                        }
                }
-               leaf = path->nodes[0];
+               leaf = path.nodes[0];
 
-               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+               btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
                if (key.type != BTRFS_EXTENT_ITEM_KEY) {
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
 
-               ei = btrfs_item_ptr(leaf, path->slots[0],
+               ei = btrfs_item_ptr(leaf, path.slots[0],
                                    struct btrfs_extent_item);
                if (!(btrfs_extent_flags(leaf, ei) &
                      BTRFS_EXTENT_FLAG_DATA)) {
-                       path->slots[0]++;
+                       path.slots[0]++;
                        continue;
                }
 
@@ -10847,10 +11857,10 @@ static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
                                    key.offset);
                if (ret)
                        break;
-               path->slots[0]++;
+               path.slots[0]++;
        }
 
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        free(buf);
        return ret;
 }
@@ -10898,7 +11908,7 @@ static int build_roots_info_cache(struct btrfs_fs_info *info)
        int ret = 0;
        struct btrfs_key key;
        struct extent_buffer *leaf;
-       struct btrfs_path *path;
+       struct btrfs_path path;
 
        if (!roots_info_cache) {
                roots_info_cache = malloc(sizeof(*roots_info_cache));
@@ -10907,24 +11917,20 @@ static int build_roots_info_cache(struct btrfs_fs_info *info)
                cache_tree_init(roots_info_cache);
        }
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
+       btrfs_init_path(&path);
        key.objectid = 0;
        key.type = BTRFS_EXTENT_ITEM_KEY;
        key.offset = 0;
-
-       ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
+       ret = btrfs_search_slot(NULL, info->extent_root, &key, &path, 0, 0);
        if (ret < 0)
                goto out;
-       leaf = path->nodes[0];
+       leaf = path.nodes[0];
 
        while (1) {
                struct btrfs_key found_key;
                struct btrfs_extent_item *ei;
                struct btrfs_extent_inline_ref *iref;
-               int slot = path->slots[0];
+               int slot = path.slots[0];
                int type;
                u64 flags;
                u64 root_id;
@@ -10933,18 +11939,18 @@ static int build_roots_info_cache(struct btrfs_fs_info *info)
                struct root_item_info *rii;
 
                if (slot >= btrfs_header_nritems(leaf)) {
-                       ret = btrfs_next_leaf(info->extent_root, path);
+                       ret = btrfs_next_leaf(info->extent_root, &path);
                        if (ret < 0) {
                                break;
                        } else if (ret) {
                                ret = 0;
                                break;
                        }
-                       leaf = path->nodes[0];
-                       slot = path->slots[0];
+                       leaf = path.nodes[0];
+                       slot = path.slots[0];
                }
 
-               btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+               btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
 
                if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
                    found_key.type != BTRFS_METADATA_ITEM_KEY)
@@ -11007,11 +12013,11 @@ static int build_roots_info_cache(struct btrfs_fs_info *info)
                        rii->node_count++;
                }
 next:
-               path->slots[0]++;
+               path.slots[0]++;
        }
 
 out:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
 
        return ret;
 }
@@ -11107,7 +12113,7 @@ static int maybe_repair_root_item(struct btrfs_fs_info *info,
  */
 static int repair_root_items(struct btrfs_fs_info *info)
 {
-       struct btrfs_path *path = NULL;
+       struct btrfs_path path;
        struct btrfs_key key;
        struct extent_buffer *leaf;
        struct btrfs_trans_handle *trans = NULL;
@@ -11115,16 +12121,12 @@ static int repair_root_items(struct btrfs_fs_info *info)
        int bad_roots = 0;
        int need_trans = 0;
 
+       btrfs_init_path(&path);
+
        ret = build_roots_info_cache(info);
        if (ret)
                goto out;
 
-       path = btrfs_alloc_path();
-       if (!path) {
-               ret = -ENOMEM;
-               goto out;
-       }
-
        key.objectid = BTRFS_FIRST_FREE_OBJECTID;
        key.type = BTRFS_ROOT_ITEM_KEY;
        key.offset = 0;
@@ -11143,19 +12145,19 @@ again:
                }
        }
 
-       ret = btrfs_search_slot(trans, info->tree_root, &key, path,
+       ret = btrfs_search_slot(trans, info->tree_root, &key, &path,
                                0, trans ? 1 : 0);
        if (ret < 0)
                goto out;
-       leaf = path->nodes[0];
+       leaf = path.nodes[0];
 
        while (1) {
                struct btrfs_key found_key;
 
-               if (path->slots[0] >= btrfs_header_nritems(leaf)) {
-                       int no_more_keys = find_next_key(path, &key);
+               if (path.slots[0] >= btrfs_header_nritems(leaf)) {
+                       int no_more_keys = find_next_key(&path, &key);
 
-                       btrfs_release_path(path);
+                       btrfs_release_path(&path);
                        if (trans) {
                                ret = btrfs_commit_transaction(trans,
                                                               info->tree_root);
@@ -11169,14 +12171,14 @@ again:
                        goto again;
                }
 
-               btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+               btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
 
                if (found_key.type != BTRFS_ROOT_ITEM_KEY)
                        goto next;
                if (found_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
                        goto next;
 
-               ret = maybe_repair_root_item(info, path, &found_key,
+               ret = maybe_repair_root_item(info, &path, &found_key,
                                             trans ? 0 : 1);
                if (ret < 0)
                        goto out;
@@ -11184,18 +12186,18 @@ again:
                        if (!trans && repair) {
                                need_trans = 1;
                                key = found_key;
-                               btrfs_release_path(path);
+                               btrfs_release_path(&path);
                                goto again;
                        }
                        bad_roots++;
                }
 next:
-               path->slots[0]++;
+               path.slots[0]++;
        }
        ret = 0;
 out:
        free_roots_info_cache();
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        if (trans)
                btrfs_commit_transaction(trans, info->tree_root);
        if (ret < 0)
@@ -11262,7 +12264,6 @@ const char * const cmd_check_usage[] = {
        "--chunk-root <bytenr>       use the given bytenr for the chunk tree root",
        "-p|--progress               indicate progress",
        "--clear-space-cache v1|v2   clear space cache for v1 or v2",
-       "                            NOTE: v1 support implemented",
        NULL
 };
 
@@ -11277,6 +12278,7 @@ int cmd_check(int argc, char **argv)
        u64 chunk_root_bytenr = 0;
        char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
        int ret;
+       int err = 0;
        u64 num;
        int init_csum_tree = 0;
        int readonly = 0;
@@ -11384,13 +12386,16 @@ int cmd_check(int argc, char **argv)
                                }
                                break;
                        case GETOPT_VAL_CLEAR_SPACE_CACHE:
-                               if (strcmp(optarg, "v1") != 0) {
+                               if (strcmp(optarg, "v1") == 0) {
+                                       clear_space_cache = 1;
+                               } else if (strcmp(optarg, "v2") == 0) {
+                                       clear_space_cache = 2;
+                                       ctree_flags |= OPEN_CTREE_INVALIDATE_FST;
+                               } else {
                                        error(
-                       "only v1 support implmented, unrecognized value %s",
-                       optarg);
+               "invalid argument to --clear-space-cache, must be v1 or v2");
                                        exit(1);
                                }
-                               clear_space_cache = 1;
                                ctree_flags |= OPEN_CTREE_WRITES;
                                break;
                }
@@ -11423,10 +12428,12 @@ int cmd_check(int argc, char **argv)
 
        if((ret = check_mounted(argv[optind])) < 0) {
                error("could not check mount status: %s", strerror(-ret));
+               err |= !!ret;
                goto err_out;
        } else if(ret) {
                error("%s is currently mounted, aborting", argv[optind]);
                ret = -EBUSY;
+               err |= !!ret;
                goto err_out;
        }
 
@@ -11439,16 +12446,17 @@ int cmd_check(int argc, char **argv)
        if (!info) {
                error("cannot open file system");
                ret = -EIO;
+               err |= !!ret;
                goto err_out;
        }
 
        global_info = info;
        root = info->fs_root;
-       if (clear_space_cache) {
+       if (clear_space_cache == 1) {
                if (btrfs_fs_compat_ro(info,
                                BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
                        error(
-                       "free space cache v2 detected, clearing not implemented");
+               "free space cache v2 detected, use --clear-space-cache v2");
                        ret = 1;
                        goto close_out;
                }
@@ -11461,6 +12469,22 @@ int cmd_check(int argc, char **argv)
                        printf("Free space cache cleared\n");
                }
                goto close_out;
+       } else if (clear_space_cache == 2) {
+               if (!btrfs_fs_compat_ro(info,
+                                       BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
+                       printf("no free space cache v2 to clear\n");
+                       ret = 0;
+                       goto close_out;
+               }
+               printf("Clear free space cache v2\n");
+               ret = btrfs_clear_free_space_tree(info);
+               if (ret) {
+                       error("failed to clear free space cache v2: %d", ret);
+                       ret = 1;
+               } else {
+                       printf("free space cache v2 cleared\n");
+               }
+               goto close_out;
        }
 
        /*
@@ -11471,9 +12495,11 @@ int cmd_check(int argc, char **argv)
                ret = ask_user("repair mode will force to clear out log tree, are you sure?");
                if (!ret) {
                        ret = 1;
+                       err |= !!ret;
                        goto close_out;
                }
                ret = zero_log_tree(root);
+               err |= !!ret;
                if (ret) {
                        error("failed to zero log tree: %d", ret);
                        goto close_out;
@@ -11485,6 +12511,7 @@ int cmd_check(int argc, char **argv)
                printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
                       uuidbuf);
                ret = qgroup_verify_all(info);
+               err |= !!ret;
                if (ret == 0)
                        report_qgroups(1);
                goto close_out;
@@ -11493,6 +12520,7 @@ int cmd_check(int argc, char **argv)
                printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
                       subvolid, argv[optind], uuidbuf);
                ret = print_extent_state(info, subvolid);
+               err |= !!ret;
                goto close_out;
        }
        printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
@@ -11501,6 +12529,7 @@ int cmd_check(int argc, char **argv)
            !extent_buffer_uptodate(info->dev_root->node) ||
            !extent_buffer_uptodate(info->chunk_root->node)) {
                error("critical roots corrupted, unable to check the filesystem");
+               err |= !!ret;
                ret = -EIO;
                goto close_out;
        }
@@ -11512,12 +12541,14 @@ int cmd_check(int argc, char **argv)
                if (IS_ERR(trans)) {
                        error("error starting transaction");
                        ret = PTR_ERR(trans);
+                       err |= !!ret;
                        goto close_out;
                }
 
                if (init_extent_tree) {
                        printf("Creating a new extent tree\n");
                        ret = reinit_extent_tree(trans, info);
+                       err |= !!ret;
                        if (ret)
                                goto close_out;
                }
@@ -11529,11 +12560,13 @@ int cmd_check(int argc, char **argv)
                                error("checksum tree initialization failed: %d",
                                                ret);
                                ret = -EIO;
+                               err |= !!ret;
                                goto close_out;
                        }
 
                        ret = fill_csum_tree(trans, info->csum_root,
                                             init_extent_tree);
+                       err |= !!ret;
                        if (ret) {
                                error("checksum tree refilling failed: %d", ret);
                                return -EIO;
@@ -11544,30 +12577,36 @@ int cmd_check(int argc, char **argv)
                 * extent entries for all of the items it finds.
                 */
                ret = btrfs_commit_transaction(trans, info->extent_root);
+               err |= !!ret;
                if (ret)
                        goto close_out;
        }
        if (!extent_buffer_uptodate(info->extent_root->node)) {
                error("critical: extent_root, unable to check the filesystem");
                ret = -EIO;
+               err |= !!ret;
                goto close_out;
        }
        if (!extent_buffer_uptodate(info->csum_root->node)) {
                error("critical: csum_root, unable to check the filesystem");
                ret = -EIO;
+               err |= !!ret;
                goto close_out;
        }
 
        if (!ctx.progress_enabled)
-               printf("checking extents");
+               fprintf(stderr, "checking extents\n");
        if (check_mode == CHECK_MODE_LOWMEM)
                ret = check_chunks_and_extents_v2(root);
        else
                ret = check_chunks_and_extents(root);
+       err |= !!ret;
        if (ret)
-               printf("Errors found in extent allocation tree or chunk allocation");
+               error(
+               "errors found in extent allocation tree or chunk allocation");
 
        ret = repair_root_items(info);
+       err |= !!ret;
        if (ret < 0)
                goto close_out;
        if (repair) {
@@ -11580,6 +12619,7 @@ int cmd_check(int argc, char **argv)
                fprintf(stderr,
                        "Please run a filesystem check with the option --repair to fix them.\n");
                ret = 1;
+               err |= !!ret;
                goto close_out;
        }
 
@@ -11590,6 +12630,7 @@ int cmd_check(int argc, char **argv)
                        fprintf(stderr, "checking free space cache\n");
        }
        ret = check_space_cache(root);
+       err |= !!ret;
        if (ret)
                goto out;
 
@@ -11603,19 +12644,28 @@ int cmd_check(int argc, char **argv)
                                     BTRFS_FEATURE_INCOMPAT_NO_HOLES);
        if (!ctx.progress_enabled)
                fprintf(stderr, "checking fs roots\n");
-       ret = check_fs_roots(root, &root_cache);
+       if (check_mode == CHECK_MODE_LOWMEM)
+               ret = check_fs_roots_v2(root->fs_info);
+       else
+               ret = check_fs_roots(root, &root_cache);
+       err |= !!ret;
        if (ret)
                goto out;
 
        fprintf(stderr, "checking csums\n");
        ret = check_csums(root);
+       err |= !!ret;
        if (ret)
                goto out;
 
        fprintf(stderr, "checking root refs\n");
-       ret = check_root_refs(root, &root_cache);
-       if (ret)
-               goto out;
+       /* For low memory mode, check_fs_roots_v2 handles root refs */
+       if (check_mode != CHECK_MODE_LOWMEM) {
+               ret = check_root_refs(root, &root_cache);
+               err |= !!ret;
+               if (ret)
+                       goto out;
+       }
 
        while (repair && !list_empty(&root->fs_info->recow_ebs)) {
                struct extent_buffer *eb;
@@ -11624,6 +12674,7 @@ int cmd_check(int argc, char **argv)
                                      struct extent_buffer, recow);
                list_del_init(&eb->recow);
                ret = recow_extent_buffer(root, eb);
+               err |= !!ret;
                if (ret)
                        break;
        }
@@ -11633,32 +12684,33 @@ int cmd_check(int argc, char **argv)
 
                bad = list_first_entry(&delete_items, struct bad_item, list);
                list_del_init(&bad->list);
-               if (repair)
+               if (repair) {
                        ret = delete_bad_item(root, bad);
+                       err |= !!ret;
+               }
                free(bad);
        }
 
        if (info->quota_enabled) {
-               int err;
                fprintf(stderr, "checking quota groups\n");
-               err = qgroup_verify_all(info);
-               if (err)
+               ret = qgroup_verify_all(info);
+               err |= !!ret;
+               if (ret)
                        goto out;
                report_qgroups(0);
-               err = repair_qgroups(info, &qgroups_repaired);
+               ret = repair_qgroups(info, &qgroups_repaired);
+               err |= !!ret;
                if (err)
                        goto out;
+               ret = 0;
        }
 
        if (!list_empty(&root->fs_info->recow_ebs)) {
                error("transid errors in file system");
                ret = 1;
+               err |= !!ret;
        }
 out:
-       /* Don't override original ret */
-       if (!ret && qgroups_repaired)
-               ret = qgroups_repaired;
-
        if (found_old_backref) { /*
                 * there was a disk format change when mixed
                 * backref was in testing tree. The old format
@@ -11668,7 +12720,7 @@ out:
                       "The old format is not supported! *"
                       "\n * Please mount the FS in readonly mode, "
                       "backup data and re-format the FS. *\n\n");
-               ret = 1;
+               err |= 1;
        }
        printf("found %llu bytes used err is %d\n",
               (unsigned long long)bytes_used, ret);
@@ -11693,5 +12745,5 @@ err_out:
        if (ctx.progress_enabled)
                task_deinit(ctx.info);
 
-       return ret;
+       return err;
 }