btrfs-progs: add the ability to fix shifted item offsets
[platform/upstream/btrfs-progs.git] / cmds-check.c
index c15d8d0..75b984f 100644 (file)
@@ -38,6 +38,8 @@
 #include "commands.h"
 #include "free-space-cache.h"
 #include "btrfsck.h"
+#include "qgroup-verify.h"
+#include "rbtree-utils.h"
 
 static u64 bytes_used = 0;
 static u64 total_csum_bytes = 0;
@@ -49,7 +51,11 @@ static u64 data_bytes_allocated = 0;
 static u64 data_bytes_referenced = 0;
 static int found_old_backref = 0;
 static LIST_HEAD(duplicate_extents);
+static LIST_HEAD(delete_items);
 static int repair = 0;
+static int no_holes = 0;
+static int init_extent_tree = 0;
+static int check_data_csum = 0;
 
 struct extent_backref {
        struct list_head list;
@@ -89,16 +95,17 @@ struct extent_record {
        struct list_head list;
        struct cache_extent cache;
        struct btrfs_disk_key parent_key;
-       unsigned int found_rec;
        u64 start;
        u64 max_size;
        u64 nr;
        u64 refs;
        u64 extent_item_refs;
        u64 generation;
+       u64 parent_generation;
        u64 info_objectid;
-       u64 num_duplicates;
+       u32 num_duplicates;
        u8 info_level;
+       unsigned int found_rec:1;
        unsigned int content_checked:1;
        unsigned int owner_ref_checked:1;
        unsigned int is_root:1;
@@ -229,8 +236,25 @@ struct walk_control {
        int root_level;
 };
 
+struct bad_item {
+       struct btrfs_key key;
+       u64 root_id;
+       struct list_head list;
+};
+
 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
 
+static void record_root_in_trans(struct btrfs_trans_handle *trans,
+                                struct btrfs_root *root)
+{
+       if (root->last_trans != trans->transid) {
+               root->track_dirty = 1;
+               root->last_trans = trans->transid;
+               root->commit_root = root->node;
+               extent_buffer_get(root->node);
+       }
+}
+
 static u8 imode_to_type(u32 imode)
 {
 #define S_SHIFT 12
@@ -284,8 +308,22 @@ static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
        return rec;
 }
 
-static void print_inode_error(int errors)
+static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
 {
+       u64 root_objectid = root->root_key.objectid;
+       int errors = rec->errors;
+
+       if (!errors)
+               return;
+       /* reloc root errors, we print its corresponding fs root objectid*/
+       if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
+               root_objectid = root->root_key.offset;
+               fprintf(stderr, "reloc");
+       }
+       fprintf(stderr, "root %llu inode %llu errors %x",
+               (unsigned long long) root_objectid,
+               (unsigned long long) rec->ino, rec->errors);
+
        if (errors & I_ERR_NO_INODE_ITEM)
                fprintf(stderr, ", no inode item");
        if (errors & I_ERR_NO_ORPHAN_ITEM)
@@ -449,8 +487,9 @@ static void maybe_free_inode_rec(struct cache_tree *inode_cache,
                        rec->errors |= I_ERR_FILE_NBYTES_WRONG;
                if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
                        rec->first_extent_gap = 0;
-               if (rec->nlink > 0 && (rec->extent_end < rec->isize ||
-                   rec->first_extent_gap < rec->isize))
+               if (rec->nlink > 0 && !no_holes &&
+                   (rec->extent_end < rec->isize ||
+                    rec->first_extent_gap < rec->isize))
                        rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
        }
 
@@ -524,6 +563,8 @@ static struct inode_backref *get_inode_backref(struct inode_record *rec,
        struct inode_backref *backref;
 
        list_for_each_entry(backref, &rec->backrefs, list) {
+               if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
+                       break;
                if (backref->dir != dir || backref->namelen != namelen)
                        continue;
                if (memcmp(name, backref->name, namelen))
@@ -579,9 +620,10 @@ static int add_inode_backref(struct cache_tree *inode_cache,
                        backref->errors |= REF_ERR_DUP_INODE_REF;
                if (backref->found_dir_index && backref->index != index)
                        backref->errors |= REF_ERR_INDEX_UNMATCH;
+               else
+                       backref->index = index;
 
                backref->ref_type = itemtype;
-               backref->index = index;
                backref->found_inode_ref = 1;
        } else {
                BUG_ON(1);
@@ -867,7 +909,8 @@ static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
        key.offset = child_root_id;
        ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
                                0, 0);
-       BUG_ON(ret < 0);
+       if (ret < 0)
+               return ret;
        btrfs_release_path(&path);
        if (!ret)
                return 1;
@@ -877,15 +920,14 @@ static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
        key.offset = 0;
        ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
                                0, 0);
-       BUG_ON(ret <= 0);
+       if (ret < 0)
+               goto out;
 
        while (1) {
                leaf = path.nodes[0];
                if (path.slots[0] >= btrfs_header_nritems(leaf)) {
                        ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
-                       BUG_ON(ret < 0);
-
-                       if (ret > 0)
+                       if (ret)
                                break;
                        leaf = path.nodes[0];
                }
@@ -904,8 +946,10 @@ static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
 
                path.slots[0]++;
        }
-
+out:
        btrfs_release_path(&path);
+       if (ret < 0)
+               return ret;
        return has_parent? 0 : -1;
 }
 
@@ -963,7 +1007,11 @@ static int process_dir_item(struct btrfs_root *root,
                                          namebuf, len, filetype,
                                          key->type, error);
                } else {
-                       fprintf(stderr, "warning line %d\n", __LINE__);
+                       fprintf(stderr, "invalid location in dir item %u\n",
+                               location.type);
+                       add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
+                                         key->objectid, key->offset, namebuf,
+                                         len, filetype, key->type, error);
                }
 
                len = sizeof(*di) + name_len + data_len;
@@ -1058,14 +1106,15 @@ static int process_inode_extref(struct extent_buffer *eb,
 
 }
 
-static u64 count_csum_range(struct btrfs_root *root, u64 start, u64 len)
+static int count_csum_range(struct btrfs_root *root, u64 start,
+                           u64 len, u64 *found)
 {
        struct btrfs_key key;
        struct btrfs_path path;
        struct extent_buffer *leaf;
-       int ret ;
+       int ret;
        size_t size;
-       u64 found = 0;
+       *found = 0;
        u64 csum_end;
        u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 
@@ -1077,7 +1126,8 @@ static u64 count_csum_range(struct btrfs_root *root, u64 start, u64 len)
 
        ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
                                &key, &path, 0, 0);
-       BUG_ON(ret < 0);
+       if (ret < 0)
+               goto out;
        if (ret > 0 && path.slots[0] > 0) {
                leaf = path.nodes[0];
                btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
@@ -1090,9 +1140,10 @@ static u64 count_csum_range(struct btrfs_root *root, u64 start, u64 len)
                leaf = path.nodes[0];
                if (path.slots[0] >= btrfs_header_nritems(leaf)) {
                        ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
-                       BUG_ON(ret < 0);
                        if (ret > 0)
                                break;
+                       else if (ret < 0)
+                               goto out;
                        leaf = path.nodes[0];
                }
 
@@ -1114,13 +1165,16 @@ static u64 count_csum_range(struct btrfs_root *root, u64 start, u64 len)
                        size = min(csum_end - start, len);
                        len -= size;
                        start += size;
-                       found += size;
+                       *found += size;
                }
 
                path.slots[0]++;
        }
+out:
+       if (ret < 0)
+               return ret;
        btrfs_release_path(&path);
-       return found;
+       return 0;
 }
 
 static int process_file_extent(struct btrfs_root *root,
@@ -1135,6 +1189,7 @@ static int process_file_extent(struct btrfs_root *root,
        u64 extent_offset = 0;
        u64 mask = root->sectorsize - 1;
        int extent_type;
+       int ret;
 
        rec = active_node->current;
        BUG_ON(rec->ino != key->objectid || rec->refs > 1);
@@ -1155,7 +1210,7 @@ static int process_file_extent(struct btrfs_root *root,
        extent_type = btrfs_file_extent_type(eb, fi);
 
        if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
-               num_bytes = btrfs_file_extent_inline_len(eb, fi);
+               num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
                if (num_bytes == 0)
                        rec->errors |= I_ERR_BAD_FILE_EXTENT;
                rec->found_size += num_bytes;
@@ -1189,7 +1244,9 @@ static int process_file_extent(struct btrfs_root *root,
                else
                        disk_bytenr += extent_offset;
 
-               found = count_csum_range(root, disk_bytenr, num_bytes);
+               ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
+               if (ret < 0)
+                       return ret;
                if (extent_type == BTRFS_FILE_EXTENT_REG) {
                        if (found > 0)
                                rec->found_csum_item = 1;
@@ -1225,6 +1282,8 @@ static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
 
                if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
                        continue;
+               if (key.type == BTRFS_ORPHAN_ITEM_KEY)
+                       continue;
 
                if (active_node->current == NULL ||
                    active_node->current->ino < key.objectid) {
@@ -1269,7 +1328,6 @@ static void reada_walk_down(struct btrfs_root *root,
        u32 nritems;
        u32 blocksize;
        int i;
-       int ret;
        int level;
 
        level = btrfs_header_level(node);
@@ -1281,10 +1339,58 @@ static void reada_walk_down(struct btrfs_root *root,
        for (i = slot; i < nritems; i++) {
                bytenr = btrfs_node_blockptr(node, i);
                ptr_gen = btrfs_node_ptr_generation(node, i);
-               ret = readahead_tree_block(root, bytenr, blocksize, ptr_gen);
-               if (ret)
-                       break;
+               readahead_tree_block(root, bytenr, blocksize, ptr_gen);
+       }
+}
+
+/*
+ * Check the child node/leaf by the following condition:
+ * 1. the first item key of the node/leaf should be the same with the one
+ *    in parent.
+ * 2. block in parent node should match the child node/leaf.
+ * 3. generation of parent node and child's header should be consistent.
+ *
+ * Or the child node/leaf pointed by the key in parent is not valid.
+ *
+ * We hope to check leaf owner too, but since subvol may share leaves,
+ * which makes leaf owner check not so strong, key check should be
+ * sufficient enough for that case.
+ */
+static int check_child_node(struct btrfs_root *root,
+                           struct extent_buffer *parent, int slot,
+                           struct extent_buffer *child)
+{
+       struct btrfs_key parent_key;
+       struct btrfs_key child_key;
+       int ret = 0;
+
+       btrfs_node_key_to_cpu(parent, &parent_key, slot);
+       if (btrfs_header_level(child) == 0)
+               btrfs_item_key_to_cpu(child, &child_key, 0);
+       else
+               btrfs_node_key_to_cpu(child, &child_key, 0);
+
+       if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
+               ret = -EINVAL;
+               fprintf(stderr,
+                       "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
+                       parent_key.objectid, parent_key.type, parent_key.offset,
+                       child_key.objectid, child_key.type, child_key.offset);
        }
+       if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
+               ret = -EINVAL;
+               fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
+                       btrfs_node_blockptr(parent, slot),
+                       btrfs_header_bytenr(child));
+       }
+       if (btrfs_node_ptr_generation(parent, slot) !=
+           btrfs_header_generation(child)) {
+               ret = -EINVAL;
+               fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
+                       btrfs_header_generation(child),
+                       btrfs_node_ptr_generation(parent, slot));
+       }
+       return ret;
 }
 
 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
@@ -1329,6 +1435,8 @@ static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
                        break;
                if (*level == 0) {
                        ret = process_one_leaf(root, cur, wc);
+                       if (ret < 0)
+                               err = ret;
                        break;
                }
                bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
@@ -1360,6 +1468,11 @@ static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
                        }
                }
 
+               ret = check_child_node(root, cur, path->slots[*level], next);
+               if (ret) {
+                       err = ret;
+                       goto out;
+               }
                *level = *level - 1;
                free_extent_buffer(path->nodes[*level]);
                path->nodes[*level] = next;
@@ -1418,28 +1531,14 @@ out:
        return ret;
 }
 
-static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
+static int repair_inode_isize(struct btrfs_trans_handle *trans,
+                             struct btrfs_root *root, struct btrfs_path *path,
+                             struct inode_record *rec)
 {
-       struct btrfs_trans_handle *trans;
-       struct btrfs_path *path;
        struct btrfs_inode_item *ei;
        struct btrfs_key key;
        int ret;
 
-       /* So far we just fix dir isize wrong */
-       if (!(rec->errors & I_ERR_DIR_ISIZE_WRONG))
-               return 1;
-
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
-
-       trans = btrfs_start_transaction(root, 1);
-       if (IS_ERR(trans)) {
-               btrfs_free_path(path);
-               return PTR_ERR(trans);
-       }
-
        key.objectid = rec->ino;
        key.type = BTRFS_INODE_ITEM_KEY;
        key.offset = (u64)-1;
@@ -1469,6 +1568,215 @@ static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
        printf("reset isize for dir %Lu root %Lu\n", rec->ino,
               root->root_key.objectid);
 out:
+       btrfs_release_path(path);
+       return ret;
+}
+
+static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
+                                   struct btrfs_root *root,
+                                   struct btrfs_path *path,
+                                   struct inode_record *rec)
+{
+       struct btrfs_key key;
+       int ret;
+
+       key.objectid = BTRFS_ORPHAN_OBJECTID;
+       key.type = BTRFS_ORPHAN_ITEM_KEY;
+       key.offset = rec->ino;
+
+       ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+       btrfs_release_path(path);
+       if (!ret)
+               rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
+       return ret;
+}
+
+static int add_missing_dir_index(struct btrfs_root *root,
+                                struct cache_tree *inode_cache,
+                                struct inode_record *rec,
+                                struct inode_backref *backref)
+{
+       struct btrfs_path *path;
+       struct btrfs_trans_handle *trans;
+       struct btrfs_dir_item *dir_item;
+       struct extent_buffer *leaf;
+       struct btrfs_key key;
+       struct btrfs_disk_key disk_key;
+       struct inode_record *dir_rec;
+       unsigned long name_ptr;
+       u32 data_size = sizeof(*dir_item) + backref->namelen;
+       int ret;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       trans = btrfs_start_transaction(root, 1);
+       if (IS_ERR(trans)) {
+               btrfs_free_path(path);
+               return PTR_ERR(trans);
+       }
+
+       fprintf(stderr, "repairing missing dir index item for inode %llu\n",
+               (unsigned long long)rec->ino);
+       key.objectid = backref->dir;
+       key.type = BTRFS_DIR_INDEX_KEY;
+       key.offset = backref->index;
+
+       ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
+       BUG_ON(ret);
+
+       leaf = path->nodes[0];
+       dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
+
+       disk_key.objectid = cpu_to_le64(rec->ino);
+       disk_key.type = BTRFS_INODE_ITEM_KEY;
+       disk_key.offset = 0;
+
+       btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
+       btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
+       btrfs_set_dir_data_len(leaf, dir_item, 0);
+       btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
+       name_ptr = (unsigned long)(dir_item + 1);
+       write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
+       btrfs_mark_buffer_dirty(leaf);
+       btrfs_free_path(path);
+       btrfs_commit_transaction(trans, root);
+
+       backref->found_dir_index = 1;
+       dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
+       if (!dir_rec)
+               return 0;
+       dir_rec->found_size += backref->namelen;
+       if (dir_rec->found_size == dir_rec->isize &&
+           (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
+               dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
+       if (dir_rec->found_size != dir_rec->isize)
+               dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
+
+       return 0;
+}
+
+static int delete_dir_index(struct btrfs_root *root,
+                           struct cache_tree *inode_cache,
+                           struct inode_record *rec,
+                           struct inode_backref *backref)
+{
+       struct btrfs_trans_handle *trans;
+       struct btrfs_dir_item *di;
+       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);
+               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,
+                                   backref->name, backref->namelen,
+                                   backref->index, -1);
+       if (IS_ERR(di)) {
+               ret = PTR_ERR(di);
+               btrfs_free_path(path);
+               btrfs_commit_transaction(trans, root);
+               if (ret == -ENOENT)
+                       return 0;
+               return ret;
+       }
+
+       if (!di)
+               ret = btrfs_del_item(trans, root, path);
+       else
+               ret = btrfs_delete_one_dir_name(trans, root, path, di);
+       BUG_ON(ret);
+       btrfs_free_path(path);
+       btrfs_commit_transaction(trans, root);
+       return ret;
+}
+
+static int repair_inode_backrefs(struct btrfs_root *root,
+                                struct inode_record *rec,
+                                struct cache_tree *inode_cache,
+                                int delete)
+{
+       struct inode_backref *tmp, *backref;
+       u64 root_dirid = btrfs_root_dirid(&root->root_item);
+       int ret = 0;
+       int repaired = 0;
+
+       list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
+               /* Index 0 for root dir's are special, don't mess with it */
+               if (rec->ino == root_dirid && backref->index == 0)
+                       continue;
+
+               if (delete &&
+                   ((backref->found_dir_index && !backref->found_inode_ref) ||
+                    (backref->found_dir_index && backref->found_inode_ref &&
+                     (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
+                       ret = delete_dir_index(root, inode_cache, rec, backref);
+                       if (ret)
+                               break;
+                       repaired++;
+                       list_del(&backref->list);
+                       free(backref);
+               }
+
+               if (!delete && !backref->found_dir_index &&
+                   backref->found_dir_item && backref->found_inode_ref) {
+                       ret = add_missing_dir_index(root, inode_cache, rec,
+                                                   backref);
+                       if (ret)
+                               break;
+                       repaired++;
+                       if (backref->found_dir_item &&
+                           backref->found_dir_index &&
+                           backref->found_dir_index) {
+                               if (!backref->errors &&
+                                   backref->found_inode_ref) {
+                                       list_del(&backref->list);
+                                       free(backref);
+                               }
+                       }
+               }
+
+       }
+       return ret ? ret : repaired;
+}
+
+static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
+{
+       struct btrfs_trans_handle *trans;
+       struct btrfs_path *path;
+       int ret = 0;
+
+       if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG | I_ERR_NO_ORPHAN_ITEM)))
+               return rec->errors;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       trans = btrfs_start_transaction(root, 1);
+       if (IS_ERR(trans)) {
+               btrfs_free_path(path);
+               return PTR_ERR(trans);
+       }
+
+       if (rec->errors & I_ERR_DIR_ISIZE_WRONG)
+               ret = repair_inode_isize(trans, root, path, rec);
+       if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
+               ret = repair_inode_orphan_item(trans, root, path, rec);
        btrfs_commit_transaction(trans, root);
        btrfs_free_path(path);
        return ret;
@@ -1481,7 +1789,9 @@ static int check_inode_recs(struct btrfs_root *root,
        struct ptr_node *node;
        struct inode_record *rec;
        struct inode_backref *backref;
+       int stage = 0;
        int ret;
+       int err = 0;
        u64 error = 0;
        u64 root_dirid = btrfs_root_dirid(&root->root_item);
 
@@ -1491,6 +1801,57 @@ static int check_inode_recs(struct btrfs_root *root,
                return 0;
        }
 
+       /*
+        * We need to repair backrefs first because we could change some of the
+        * errors in the inode recs.
+        *
+        * We also need to go through and delete invalid backrefs first and then
+        * add the correct ones second.  We do this because we may get EEXIST
+        * when adding back the correct index because we hadn't yet deleted the
+        * invalid index.
+        *
+        * For example, if we were missing a dir index then the directories
+        * isize would be wrong, so if we fixed the isize to what we thought it
+        * would be and then fixed the backref we'd still have a invalid fs, so
+        * we need to add back the dir index and then check to see if the isize
+        * is still wrong.
+        */
+       while (stage < 3) {
+               stage++;
+               if (stage == 3 && !err)
+                       break;
+
+               cache = search_cache_extent(inode_cache, 0);
+               while (repair && cache) {
+                       node = container_of(cache, struct ptr_node, cache);
+                       rec = node->data;
+                       cache = next_cache_extent(cache);
+
+                       /* Need to free everything up and rescan */
+                       if (stage == 3) {
+                               remove_cache_extent(inode_cache, &node->cache);
+                               free(node);
+                               free_inode_rec(rec);
+                               continue;
+                       }
+
+                       if (list_empty(&rec->backrefs))
+                               continue;
+
+                       ret = repair_inode_backrefs(root, rec, inode_cache,
+                                                   stage == 1);
+                       if (ret < 0) {
+                               err = ret;
+                               stage = 2;
+                               break;
+                       } if (ret > 0) {
+                               err = -EAGAIN;
+                       }
+               }
+       }
+       if (err)
+               return err;
+
        rec = get_inode_rec(inode_cache, root_dirid, 0);
        if (rec) {
                ret = check_root_dir(rec);
@@ -1544,10 +1905,7 @@ static int check_inode_recs(struct btrfs_root *root,
                        rec->errors |= I_ERR_NO_INODE_ITEM;
                if (rec->found_link != rec->nlink)
                        rec->errors |= I_ERR_LINK_COUNT_WRONG;
-               fprintf(stderr, "root %llu inode %llu errors %x",
-                       (unsigned long long) root->root_key.objectid,
-                       (unsigned long long) rec->ino, rec->errors);
-               print_inode_error(rec->errors);
+               print_inode_error(root, rec);
                list_for_each_entry(backref, &rec->backrefs, list) {
                        if (!backref->found_dir_item)
                                backref->errors |= REF_ERR_NO_DIR_ITEM;
@@ -1556,7 +1914,7 @@ static int check_inode_recs(struct btrfs_root *root,
                        if (!backref->found_inode_ref)
                                backref->errors |= REF_ERR_NO_INODE_REF;
                        fprintf(stderr, "\tunresolved ref dir %llu index %llu"
-                               " namelen %u name %s filetype %d error %x",
+                               " namelen %u name %s filetype %d errors %x",
                                (unsigned long long)backref->dir,
                                (unsigned long long)backref->index,
                                backref->namelen, backref->name,
@@ -1692,6 +2050,7 @@ static int merge_root_recs(struct btrfs_root *root,
        struct ptr_node *node;
        struct inode_record *rec;
        struct inode_backref *backref;
+       int ret = 0;
 
        if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
                free_inode_recs_tree(src_cache);
@@ -1707,7 +2066,10 @@ static int merge_root_recs(struct btrfs_root *root,
                remove_cache_extent(src_cache, &node->cache);
                free(node);
 
-               if (!is_child_root(root, root->objectid, rec->ino))
+               ret = is_child_root(root, root->objectid, rec->ino);
+               if (ret < 0)
+                       break;
+               else if (ret == 0)
                        goto skip;
 
                list_for_each_entry(backref, &rec->backrefs, list) {
@@ -1728,6 +2090,8 @@ static int merge_root_recs(struct btrfs_root *root,
 skip:
                free_inode_rec(rec);
        }
+       if (ret < 0)
+               return ret;
        return 0;
 }
 
@@ -1832,12 +2196,13 @@ static int check_root_refs(struct btrfs_root *root,
                        if (!backref->errors && rec->found_root_item)
                                continue;
                        fprintf(stderr, "\tunresolved ref root %llu dir %llu"
-                               " index %llu namelen %u name %s error %x\n",
+                               " index %llu namelen %u name %s errors %x\n",
                                (unsigned long long)backref->ref_root,
                                (unsigned long long)backref->dir,
                                (unsigned long long)backref->index,
                                backref->namelen, backref->name,
                                backref->errors);
+                       print_ref_error(backref->errors);
                }
        }
        return errors > 0 ? 1 : 0;
@@ -1885,6 +2250,7 @@ static int check_fs_root(struct btrfs_root *root,
                         struct walk_control *wc)
 {
        int ret = 0;
+       int err = 0;
        int wret;
        int level;
        struct btrfs_path path;
@@ -1922,7 +2288,8 @@ static int check_fs_root(struct btrfs_root *root,
                level = root_item->drop_level;
                path.lowest_level = level;
                wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
-               BUG_ON(wret < 0);
+               if (wret < 0)
+                       goto skip_walking;
                btrfs_node_key(path.nodes[level], &found_key,
                                path.slots[level]);
                WARN_ON(memcmp(&found_key, &root_item->drop_progress,
@@ -1942,9 +2309,12 @@ static int check_fs_root(struct btrfs_root *root,
                if (wret != 0)
                        break;
        }
+skip_walking:
        btrfs_release_path(&path);
 
-       merge_root_recs(root, &root_node.root_cache, root_cache);
+       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;
@@ -1952,19 +2322,18 @@ static int check_fs_root(struct btrfs_root *root,
                                root_node.current);
        }
 
-       ret = check_inode_recs(root, &root_node.inode_cache);
+       err = check_inode_recs(root, &root_node.inode_cache);
+       if (!ret)
+               ret = err;
        return ret;
 }
 
 static int fs_root_objectid(u64 objectid)
 {
-       if (objectid == BTRFS_FS_TREE_OBJECTID ||
-           objectid == BTRFS_TREE_RELOC_OBJECTID ||
-           objectid == BTRFS_DATA_RELOC_TREE_OBJECTID ||
-           (objectid >= BTRFS_FIRST_FREE_OBJECTID &&
-            objectid <= BTRFS_LAST_FREE_OBJECTID))
+       if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
+           objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
                return 1;
-       return 0;
+       return is_fstree(objectid);
 }
 
 static int check_fs_roots(struct btrfs_root *root,
@@ -1973,7 +2342,7 @@ static int check_fs_roots(struct btrfs_root *root,
        struct btrfs_path path;
        struct btrfs_key key;
        struct walk_control wc;
-       struct extent_buffer *leaf;
+       struct extent_buffer *leaf, *tree_node;
        struct btrfs_root *tmp_root;
        struct btrfs_root *tree_root = root->fs_info->tree_root;
        int ret;
@@ -1989,32 +2358,57 @@ static int check_fs_roots(struct btrfs_root *root,
        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);
-       BUG_ON(ret < 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 != 0)
+                       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)) {
-                       tmp_root = btrfs_read_fs_root_no_cache(root->fs_info,
-                                                              &key);
+                       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;
-                       btrfs_free_fs_root(tmp_root);
+                       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,
@@ -2023,8 +2417,10 @@ static int check_fs_roots(struct btrfs_root *root,
 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__);
 
@@ -2289,14 +2685,249 @@ static int record_bad_block_io(struct btrfs_fs_info *info,
        return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
 }
 
-static int check_block(struct btrfs_root *root,
-                      struct cache_tree *extent_cache,
+static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
+                      struct extent_buffer *buf, int slot)
+{
+       if (btrfs_header_level(buf)) {
+               struct btrfs_key_ptr ptr1, ptr2;
+
+               read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
+                                  sizeof(struct btrfs_key_ptr));
+               read_extent_buffer(buf, &ptr2,
+                                  btrfs_node_key_ptr_offset(slot + 1),
+                                  sizeof(struct btrfs_key_ptr));
+               write_extent_buffer(buf, &ptr1,
+                                   btrfs_node_key_ptr_offset(slot + 1),
+                                   sizeof(struct btrfs_key_ptr));
+               write_extent_buffer(buf, &ptr2,
+                                   btrfs_node_key_ptr_offset(slot),
+                                   sizeof(struct btrfs_key_ptr));
+               if (slot == 0) {
+                       struct btrfs_disk_key key;
+                       btrfs_node_key(buf, &key, 0);
+                       btrfs_fixup_low_keys(root, path, &key,
+                                            btrfs_header_level(buf) + 1);
+               }
+       } else {
+               struct btrfs_item *item1, *item2;
+               struct btrfs_key k1, k2;
+               char *item1_data, *item2_data;
+               u32 item1_offset, item2_offset, item1_size, item2_size;
+
+               item1 = btrfs_item_nr(slot);
+               item2 = btrfs_item_nr(slot + 1);
+               btrfs_item_key_to_cpu(buf, &k1, slot);
+               btrfs_item_key_to_cpu(buf, &k2, slot + 1);
+               item1_offset = btrfs_item_offset(buf, item1);
+               item2_offset = btrfs_item_offset(buf, item2);
+               item1_size = btrfs_item_size(buf, item1);
+               item2_size = btrfs_item_size(buf, item2);
+
+               item1_data = malloc(item1_size);
+               if (!item1_data)
+                       return -ENOMEM;
+               item2_data = malloc(item2_size);
+               if (!item2_data) {
+                       free(item1_data);
+                       return -ENOMEM;
+               }
+
+               read_extent_buffer(buf, item1_data, item1_offset, item1_size);
+               read_extent_buffer(buf, item2_data, item2_offset, item2_size);
+
+               write_extent_buffer(buf, item1_data, item2_offset, item2_size);
+               write_extent_buffer(buf, item2_data, item1_offset, item1_size);
+               free(item1_data);
+               free(item2_data);
+
+               btrfs_set_item_offset(buf, item1, item2_offset);
+               btrfs_set_item_offset(buf, item2, item1_offset);
+               btrfs_set_item_size(buf, item1, item2_size);
+               btrfs_set_item_size(buf, item2, item1_size);
+
+               path->slots[0] = slot;
+               btrfs_set_item_key_unsafe(root, path, &k2);
+               path->slots[0] = slot + 1;
+               btrfs_set_item_key_unsafe(root, path, &k1);
+       }
+       return 0;
+}
+
+static int fix_key_order(struct btrfs_trans_handle *trans,
+                        struct btrfs_root *root,
+                        struct extent_buffer *buf)
+{
+       struct btrfs_path *path;
+       struct btrfs_key k1, k2;
+       int i;
+       int level;
+       int ret;
+
+       k1.objectid = btrfs_header_owner(buf);
+       k1.type = BTRFS_ROOT_ITEM_KEY;
+       k1.offset = (u64)-1;
+
+       root = btrfs_read_fs_root(root->fs_info, &k1);
+       if (IS_ERR(root))
+               return -EIO;
+
+       record_root_in_trans(trans, root);
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -EIO;
+
+       level = btrfs_header_level(buf);
+       path->lowest_level = level;
+       path->skip_check_block = 1;
+       if (level)
+               btrfs_node_key_to_cpu(buf, &k1, 0);
+       else
+               btrfs_item_key_to_cpu(buf, &k1, 0);
+
+       ret = btrfs_search_slot(trans, root, &k1, path, 0, 1);
+       if (ret) {
+               btrfs_free_path(path);
+               return -EIO;
+       }
+
+       buf = path->nodes[level];
+       for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
+               if (level) {
+                       btrfs_node_key_to_cpu(buf, &k1, i);
+                       btrfs_node_key_to_cpu(buf, &k2, i + 1);
+               } else {
+                       btrfs_item_key_to_cpu(buf, &k1, i);
+                       btrfs_item_key_to_cpu(buf, &k2, i + 1);
+               }
+               if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
+                       continue;
+               ret = swap_values(root, path, buf, i);
+               if (ret)
+                       break;
+               btrfs_mark_buffer_dirty(buf);
+               i = 0;
+       }
+
+       btrfs_free_path(path);
+       return ret;
+}
+
+static int fix_item_offset(struct btrfs_trans_handle *trans,
+                          struct btrfs_root *root,
+                          struct extent_buffer *buf)
+{
+       struct btrfs_path *path;
+       struct btrfs_key k1;
+       int i;
+       int level;
+       int ret;
+
+       k1.objectid = btrfs_header_owner(buf);
+       k1.type = BTRFS_ROOT_ITEM_KEY;
+       k1.offset = (u64)-1;
+
+       root = btrfs_read_fs_root(root->fs_info, &k1);
+       if (IS_ERR(root))
+               return -EIO;
+
+       record_root_in_trans(trans, root);
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -EIO;
+
+       level = btrfs_header_level(buf);
+       path->lowest_level = level;
+       path->skip_check_block = 1;
+       if (level)
+               btrfs_node_key_to_cpu(buf, &k1, 0);
+       else
+               btrfs_item_key_to_cpu(buf, &k1, 0);
+
+       ret = btrfs_search_slot(trans, root, &k1, path, 0, 1);
+       if (ret) {
+               btrfs_free_path(path);
+               return -EIO;
+       }
+
+       buf = path->nodes[level];
+       for (i = 0; i < btrfs_header_nritems(buf); i++) {
+               unsigned int shift = 0, offset;
+
+               if (i == 0 && btrfs_item_end_nr(buf, i) !=
+                   BTRFS_LEAF_DATA_SIZE(root)) {
+                       if (btrfs_item_end_nr(buf, i) >
+                           BTRFS_LEAF_DATA_SIZE(root)) {
+                               fprintf(stderr, "item is off the end of the "
+                                       "leaf, can't fix\n");
+                               ret = -EIO;
+                               break;
+                       }
+                       shift = BTRFS_LEAF_DATA_SIZE(root) -
+                               btrfs_item_end_nr(buf, i);
+               } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
+                          btrfs_item_offset_nr(buf, i - 1)) {
+                       if (btrfs_item_end_nr(buf, i) >
+                           btrfs_item_offset_nr(buf, i - 1)) {
+                               fprintf(stderr, "items overlap, can't fix\n");
+                               ret = -EIO;
+                               break;
+                       }
+                       shift = btrfs_item_offset_nr(buf, i - 1) -
+                               btrfs_item_end_nr(buf, i);
+               }
+               if (!shift)
+                       continue;
+
+               printf("Shifting item nr %d by %u bytes in block %llu\n",
+                      i, shift, (unsigned long long)buf->start);
+               offset = btrfs_item_offset_nr(buf, i);
+               memmove_extent_buffer(buf,
+                                     btrfs_leaf_data(buf) + offset + shift,
+                                     btrfs_leaf_data(buf) + offset,
+                                     btrfs_item_size_nr(buf, i));
+               btrfs_set_item_offset(buf, btrfs_item_nr(i),
+                                     offset + shift);
+               btrfs_mark_buffer_dirty(buf);
+       }
+
+       /*
+        * We may have moved things, in which case we want to exit so we don't
+        * write those changes out.  Once we have proper abort functionality in
+        * progs this can be changed to something nicer.
+        */
+       BUG_ON(ret);
+       btrfs_free_path(path);
+       return ret;
+}
+
+/*
+ * Attempt to fix basic block failures.  If we can't fix it for whatever reason
+ * then just return -EIO.
+ */
+static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
+                               struct btrfs_root *root,
+                               struct extent_buffer *buf,
+                               enum btrfs_tree_block_status status)
+{
+       if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
+               return fix_key_order(trans, root, buf);
+       if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
+               return fix_item_offset(trans, root, buf);
+       return -EIO;
+}
+
+static int check_block(struct btrfs_trans_handle *trans,
+                      struct btrfs_root *root,
+                      struct cache_tree *extent_cache,
                       struct extent_buffer *buf, u64 flags)
 {
        struct extent_record *rec;
        struct cache_extent *cache;
        struct btrfs_key key;
-       int ret = 1;
+       enum btrfs_tree_block_status status;
+       int ret = 0;
        int level;
 
        cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
@@ -2318,13 +2949,25 @@ static int check_block(struct btrfs_root *root,
        rec->info_level = level;
 
        if (btrfs_is_leaf(buf))
-               ret = btrfs_check_leaf(root, &rec->parent_key, buf);
+               status = btrfs_check_leaf(root, &rec->parent_key, buf);
        else
-               ret = btrfs_check_node(root, &rec->parent_key, buf);
-
-       if (ret) {
-               fprintf(stderr, "bad block %llu\n",
-                       (unsigned long long)buf->start);
+               status = btrfs_check_node(root, &rec->parent_key, buf);
+
+       if (status != BTRFS_TREE_BLOCK_CLEAN) {
+               if (repair)
+                       status = try_to_fix_bad_block(trans, root, buf,
+                                                     status);
+               if (status != BTRFS_TREE_BLOCK_CLEAN) {
+                       ret = -EIO;
+                       fprintf(stderr, "bad block %llu\n",
+                               (unsigned long long)buf->start);
+               } else {
+                       /*
+                        * Signal to callers we need to start the scan over
+                        * again since we'll have cow'ed blocks.
+                        */
+                       ret = -EAGAIN;
+               }
        } else {
                rec->content_checked = 1;
                if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
@@ -2452,7 +3095,7 @@ static struct data_backref *alloc_data_backref(struct extent_record *rec,
 }
 
 static int add_extent_rec(struct cache_tree *extent_cache,
-                         struct btrfs_key *parent_key,
+                         struct btrfs_key *parent_key, u64 parent_gen,
                          u64 start, u64 nr, u64 extent_item_refs,
                          int is_root, int inc_ref, int set_checked,
                          int metadata, int extent_rec, u64 max_size)
@@ -2528,6 +3171,8 @@ static int add_extent_rec(struct cache_tree *extent_cache,
 
                if (parent_key)
                        btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
+               if (parent_gen)
+                       rec->parent_generation = parent_gen;
 
                if (rec->max_size < max_size)
                        rec->max_size = max_size;
@@ -2539,7 +3184,7 @@ static int add_extent_rec(struct cache_tree *extent_cache,
        rec->start = start;
        rec->max_size = max_size;
        rec->nr = max(nr, max_size);
-       rec->found_rec = extent_rec;
+       rec->found_rec = !!extent_rec;
        rec->content_checked = 0;
        rec->owner_ref_checked = 0;
        rec->num_duplicates = 0;
@@ -2568,6 +3213,11 @@ static int add_extent_rec(struct cache_tree *extent_cache,
        else
                memset(&rec->parent_key, 0, sizeof(*parent_key));
 
+       if (parent_gen)
+               rec->parent_generation = parent_gen;
+       else
+               rec->parent_generation = 0;
+
        rec->cache.start = start;
        rec->cache.size = nr;
        ret = insert_cache_extent(extent_cache, &rec->cache);
@@ -2589,7 +3239,7 @@ static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
 
        cache = lookup_cache_extent(extent_cache, bytenr, 1);
        if (!cache) {
-               add_extent_rec(extent_cache, NULL, bytenr,
+               add_extent_rec(extent_cache, NULL, 0, bytenr,
                               1, 0, 0, 0, 0, 1, 0, 0);
                cache = lookup_cache_extent(extent_cache, bytenr, 1);
                if (!cache)
@@ -2624,6 +3274,7 @@ static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
                }
                back->node.found_extent_tree = 1;
        }
+       maybe_free_extent_rec(extent_cache, rec);
        return 0;
 }
 
@@ -2637,7 +3288,7 @@ static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
 
        cache = lookup_cache_extent(extent_cache, bytenr, 1);
        if (!cache) {
-               add_extent_rec(extent_cache, NULL, bytenr, 1, 0, 0, 0, 0,
+               add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
                               0, 0, max_size);
                cache = lookup_cache_extent(extent_cache, bytenr, 1);
                if (!cache)
@@ -2678,7 +3329,7 @@ static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
        } else {
                if (back->node.found_extent_tree) {
                        fprintf(stderr, "Extent back ref already exists "
-                               "for %llu parent %llu root %llu"
+                               "for %llu parent %llu root %llu "
                                "owner %llu offset %llu num_refs %lu\n",
                                (unsigned long long)bytenr,
                                (unsigned long long)parent,
@@ -2690,6 +3341,7 @@ static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
                back->num_refs = num_refs;
                back->node.found_extent_tree = 1;
        }
+       maybe_free_extent_rec(extent_cache, rec);
        return 0;
 }
 
@@ -2717,7 +3369,7 @@ static int pick_next_pending(struct cache_tree *pending,
        cache = search_cache_extent(reada, 0);
        if (cache) {
                bits[0].start = cache->start;
-               bits[1].size = cache->size;
+               bits[0].size = cache->size;
                *reada_bits = 1;
                return 1;
        }
@@ -2777,6 +3429,8 @@ static void free_chunk_record(struct cache_extent *cache)
        struct chunk_record *rec;
 
        rec = container_of(cache, struct chunk_record, cache);
+       list_del_init(&rec->list);
+       list_del_init(&rec->dextents);
        free(rec);
 }
 
@@ -2813,6 +3467,7 @@ static void free_block_group_record(struct cache_extent *cache)
        struct block_group_record *rec;
 
        rec = container_of(cache, struct block_group_record, cache);
+       list_del_init(&rec->list);
        free(rec);
 }
 
@@ -2846,6 +3501,10 @@ static void free_device_extent_record(struct cache_extent *cache)
        struct device_extent_record *rec;
 
        rec = container_of(cache, struct device_extent_record, cache);
+       if (!list_empty(&rec->chunk_list))
+               list_del_init(&rec->chunk_list);
+       if (!list_empty(&rec->device_list))
+               list_del_init(&rec->device_list);
        free(rec);
 }
 
@@ -3124,7 +3783,7 @@ static int process_extent_item(struct btrfs_root *root,
 #else
                BUG();
 #endif
-               return add_extent_rec(extent_cache, NULL, key.objectid,
+               return add_extent_rec(extent_cache, NULL, 0, key.objectid,
                                      num_bytes, refs, 0, 0, 0, metadata, 1,
                                      num_bytes);
        }
@@ -3132,7 +3791,7 @@ static int process_extent_item(struct btrfs_root *root,
        ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
        refs = btrfs_extent_refs(eb, ei);
 
-       add_extent_rec(extent_cache, NULL, key.objectid, num_bytes,
+       add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
                       refs, 0, 0, 0, metadata, 1, num_bytes);
 
        ptr = (unsigned long)(ei + 1);
@@ -3406,6 +4065,109 @@ static int check_space_cache(struct btrfs_root *root)
        return error ? -EINVAL : 0;
 }
 
+static int read_extent_data(struct btrfs_root *root, char *data,
+                       u64 logical, u64 *len, int mirror)
+{
+       u64 offset = 0;
+       struct btrfs_multi_bio *multi = NULL;
+       struct btrfs_fs_info *info = root->fs_info;
+       struct btrfs_device *device;
+       int ret = 0;
+       u64 max_len = *len;
+
+       ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
+                             &multi, mirror, NULL);
+       if (ret) {
+               fprintf(stderr, "Couldn't map the block %llu\n",
+                               logical + offset);
+               goto err;
+       }
+       device = multi->stripes[0].dev;
+
+       if (device->fd == 0)
+               goto err;
+       if (*len > max_len)
+               *len = max_len;
+
+       ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
+       if (ret != *len)
+               ret = -EIO;
+       else
+               ret = 0;
+err:
+       kfree(multi);
+       return ret;
+}
+
+static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
+                       u64 num_bytes, unsigned long leaf_offset,
+                       struct extent_buffer *eb) {
+
+       u64 offset = 0;
+       u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
+       char *data;
+       unsigned long csum_offset;
+       u32 csum;
+       u32 csum_expected;
+       u64 read_len;
+       u64 data_checked = 0;
+       u64 tmp;
+       int ret = 0;
+       int mirror;
+       int num_copies;
+
+       if (num_bytes % root->sectorsize)
+               return -EINVAL;
+
+       data = malloc(num_bytes);
+       if (!data)
+               return -ENOMEM;
+
+       while (offset < num_bytes) {
+               mirror = 0;
+again:
+               read_len = num_bytes - offset;
+               /* read as much space once a time */
+               ret = read_extent_data(root, data + offset,
+                               bytenr + offset, &read_len, mirror);
+               if (ret)
+                       goto out;
+               data_checked = 0;
+               /* verify every 4k data's checksum */
+               while (data_checked < read_len) {
+                       csum = ~(u32)0;
+                       tmp = offset + data_checked;
+
+                       csum = btrfs_csum_data(NULL, (char *)data + tmp,
+                                              csum, root->sectorsize);
+                       btrfs_csum_final(csum, (char *)&csum);
+
+                       csum_offset = leaf_offset +
+                                tmp / root->sectorsize * csum_size;
+                       read_extent_buffer(eb, (char *)&csum_expected,
+                                          csum_offset, csum_size);
+                       /* try another mirror */
+                       if (csum != csum_expected) {
+                               fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
+                                               mirror, bytenr + tmp,
+                                               csum, csum_expected);
+                               num_copies = btrfs_num_copies(
+                                               &root->fs_info->mapping_tree,
+                                               bytenr, num_bytes);
+                               if (mirror < num_copies - 1) {
+                                       mirror += 1;
+                                       goto again;
+                               }
+                       }
+                       data_checked += root->sectorsize;
+               }
+               offset += read_len;
+       }
+out:
+       free(data);
+       return ret;
+}
+
 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
                               u64 num_bytes)
 {
@@ -3422,8 +4184,7 @@ static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
 
        key.objectid = bytenr;
        key.type = BTRFS_EXTENT_ITEM_KEY;
-       key.offset = 0;
-
+       key.offset = (u64)-1;
 
 again:
        ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
@@ -3433,10 +4194,17 @@ again:
                btrfs_free_path(path);
                return ret;
        } else if (ret) {
-               if (path->slots[0])
+               if (path->slots[0] > 0) {
                        path->slots[0]--;
-               else
-                       btrfs_prev_leaf(root, path);
+               } else {
+                       ret = btrfs_prev_leaf(root, path);
+                       if (ret < 0) {
+                               goto out;
+                       } else if (ret > 0) {
+                               ret = 0;
+                               goto out;
+                       }
+               }
        }
 
        btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
@@ -3446,13 +4214,22 @@ again:
         * bytenr, so walk back one more just in case.  Dear future traveler,
         * first congrats on mastering time travel.  Now if it's not too much
         * trouble could you go back to 2006 and tell Chris to make the
-        * BLOCK_GROUP_ITEM_KEY lower than the EXTENT_ITEM_KEY please?
+        * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
+        * EXTENT_ITEM_KEY please?
         */
-       if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
-               if (path->slots[0])
+       while (key.type > BTRFS_EXTENT_ITEM_KEY) {
+               if (path->slots[0] > 0) {
                        path->slots[0]--;
-               else
-                       btrfs_prev_leaf(root, path);
+               } else {
+                       ret = btrfs_prev_leaf(root, path);
+                       if (ret < 0) {
+                               goto out;
+                       } else if (ret > 0) {
+                               ret = 0;
+                               goto out;
+                       }
+               }
+               btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
        }
 
        while (num_bytes) {
@@ -3524,7 +4301,8 @@ again:
        }
        ret = 0;
 
-       if (num_bytes) {
+out:
+       if (num_bytes && !ret) {
                fprintf(stderr, "There are no extents for csum range "
                        "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
                ret = 1;
@@ -3543,8 +4321,14 @@ static int check_csums(struct btrfs_root *root)
        u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
        int errors = 0;
        int ret;
+       u64 data_len;
+       unsigned long leaf_offset;
 
        root = root->fs_info->csum_root;
+       if (!extent_buffer_uptodate(root->node)) {
+               fprintf(stderr, "No valid csum tree found\n");
+               return -ENOENT;
+       }
 
        key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
        key.type = BTRFS_EXTENT_CSUM_KEY;
@@ -3584,6 +4368,16 @@ static int check_csums(struct btrfs_root *root)
                        continue;
                }
 
+               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]);
+               ret = check_extent_csums(root, key.offset, data_len,
+                                        leaf_offset, leaf);
+               if (ret)
+                       break;
+skip_csum_check:
                if (!num_bytes) {
                        offset = key.offset;
                } else if (key.offset != offset + num_bytes) {
@@ -3597,9 +4391,7 @@ static int check_csums(struct btrfs_root *root)
                        offset = key.offset;
                        num_bytes = 0;
                }
-
-               num_bytes += (btrfs_item_size_nr(leaf, path->slots[0]) /
-                             csum_size) * root->sectorsize;
+               num_bytes += data_len;
                path->slots[0]++;
        }
 
@@ -3622,7 +4414,8 @@ static int is_dropped_key(struct btrfs_key *key,
        return 0;
 }
 
-static int run_next_block(struct btrfs_root *root,
+static int run_next_block(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *root,
                          struct block_info *bits,
                          int bits_nr,
                          u64 *last,
@@ -3644,7 +4437,8 @@ static int run_next_block(struct btrfs_root *root,
        u64 owner;
        u64 flags;
        u64 ptr;
-       int ret;
+       u64 gen = 0;
+       int ret = 0;
        int i;
        int nritems;
        struct btrfs_key key;
@@ -3687,14 +4481,16 @@ static int run_next_block(struct btrfs_root *root,
                remove_cache_extent(nodes, cache);
                free(cache);
        }
-       cache = lookup_cache_extent(seen, bytenr, size);
+       cache = lookup_cache_extent(extent_cache, bytenr, size);
        if (cache) {
-               remove_cache_extent(seen, cache);
-               free(cache);
+               struct extent_record *rec;
+
+               rec = container_of(cache, struct extent_record, cache);
+               gen = rec->parent_generation;
        }
 
        /* fixme, get the real parent transid */
-       buf = read_tree_block(root, bytenr, size, 0);
+       buf = read_tree_block(root, bytenr, size, gen);
        if (!extent_buffer_uptodate(buf)) {
                record_bad_block_io(root->fs_info,
                                    extent_cache, bytenr, size);
@@ -3703,11 +4499,19 @@ static int run_next_block(struct btrfs_root *root,
 
        nritems = btrfs_header_nritems(buf);
 
-       ret = btrfs_lookup_extent_info(NULL, root, bytenr,
+       /*
+        * FIXME, this only works only if we don't have any full
+        * backref mode.
+        */
+       if (!init_extent_tree) {
+               ret = btrfs_lookup_extent_info(NULL, root, bytenr,
                                       btrfs_header_level(buf), 1, NULL,
                                       &flags);
-       if (ret < 0)
-               flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
+               if (ret < 0)
+                       flags = 0;
+       } else {
+               flags = 0;
+       }
 
        if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
                parent = bytenr;
@@ -3717,7 +4521,7 @@ static int run_next_block(struct btrfs_root *root,
                owner = btrfs_header_owner(buf);
        }
 
-       ret = check_block(root, extent_cache, buf, flags);
+       ret = check_block(trans, root, extent_cache, buf, flags);
        if (ret)
                goto out;
 
@@ -3803,6 +4607,23 @@ static int run_next_block(struct btrfs_root *root,
                                        0, root->sectorsize);
                                continue;
                        }
+                       if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
+                               struct bad_item *bad;
+
+                               if (key.objectid == BTRFS_ORPHAN_OBJECTID)
+                                       continue;
+                               if (!owner)
+                                       continue;
+                               bad = malloc(sizeof(struct bad_item));
+                               if (!bad)
+                                       continue;
+                               INIT_LIST_HEAD(&bad->list);
+                               memcpy(&bad->key, &key,
+                                      sizeof(struct btrfs_key));
+                               bad->root_id = owner;
+                               list_add_tail(&bad->list, &delete_items);
+                               continue;
+                       }
                        if (key.type != BTRFS_EXTENT_DATA_KEY)
                                continue;
                        fi = btrfs_item_ptr(buf, i,
@@ -3825,7 +4646,6 @@ static int run_next_block(struct btrfs_root *root,
                                parent, owner, key.objectid, key.offset -
                                btrfs_file_extent_offset(buf, fi), 1, 1,
                                btrfs_file_extent_disk_num_bytes(buf, fi));
-                       BUG_ON(ret);
                }
        } else {
                int level;
@@ -3850,6 +4670,7 @@ static int run_next_block(struct btrfs_root *root,
                                }
                        }
                        ret = add_extent_rec(extent_cache, &key,
+                                            btrfs_node_ptr_generation(buf, i),
                                             ptr, size, 0, 0, 1, 0, 1, 0,
                                             size);
                        BUG_ON(ret);
@@ -3877,7 +4698,7 @@ static int run_next_block(struct btrfs_root *root,
                found_old_backref = 1;
 out:
        free_extent_buffer(buf);
-       return 0;
+       return ret;
 }
 
 static int add_root_to_pending(struct extent_buffer *buf,
@@ -3891,7 +4712,7 @@ static int add_root_to_pending(struct extent_buffer *buf,
                add_pending(nodes, seen, buf->start, buf->len);
        else
                add_pending(pending, seen, buf->start, buf->len);
-       add_extent_rec(extent_cache, NULL, buf->start, buf->len,
+       add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
                       0, 1, 1, 0, 1, 0, buf->len);
 
        if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
@@ -4321,12 +5142,7 @@ static int repair_ref(struct btrfs_trans_handle *trans,
         * Have to make sure that this root gets updated when we commit the
         * transaction
         */
-       root->track_dirty = 1;
-       if (root->last_trans != trans->transid) {
-               root->last_trans = trans->transid;
-               root->commit_root = root->node;
-               extent_buffer_get(root->node);
-       }
+       record_root_in_trans(trans, root);
 
        /*
         * Ok we have the key of the file extent we want to fix, now we can cow
@@ -4432,15 +5248,17 @@ static int verify_backrefs(struct btrfs_trans_handle *trans,
                return 0;
 
        list_for_each_entry(back, &rec->backrefs, list) {
+               if (back->full_backref || !back->is_data)
+                       continue;
+
                dback = (struct data_backref *)back;
+
                /*
                 * We only pay attention to backrefs that we found a real
                 * backref for.
                 */
                if (dback->found_ref == 0)
                        continue;
-               if (back->full_backref)
-                       continue;
 
                /*
                 * For now we only catch when the bytes don't match, not the
@@ -4502,7 +5320,7 @@ static int verify_backrefs(struct btrfs_trans_handle *trans,
        if (!best) {
                entry = find_entry(&entries, rec->start, rec->nr);
                if (!entry && (!broken_entries || !rec->found_rec)) {
-                       fprintf(stderr, "Backrefs don't agree with eachother "
+                       fprintf(stderr, "Backrefs don't agree with each other "
                                "and extent record doesn't agree with anybody,"
                                " so we can't fix bytenr %Lu bytes %Lu\n",
                                rec->start, rec->nr);
@@ -4556,6 +5374,9 @@ static int verify_backrefs(struct btrfs_trans_handle *trans,
         * references and fix up the ones that don't match.
         */
        list_for_each_entry(back, &rec->backrefs, list) {
+               if (back->full_backref || !back->is_data)
+                       continue;
+
                dback = (struct data_backref *)back;
 
                /*
@@ -4564,8 +5385,6 @@ static int verify_backrefs(struct btrfs_trans_handle *trans,
                 */
                if (dback->found_ref == 0)
                        continue;
-               if (back->full_backref)
-                       continue;
 
                if (dback->bytes == best->bytes &&
                    dback->disk_bytenr == best->bytenr)
@@ -4785,14 +5604,16 @@ static int find_possible_backrefs(struct btrfs_trans_handle *trans,
        int ret;
 
        list_for_each_entry(back, &rec->backrefs, list) {
+               /* Don't care about full backrefs (poor unloved backrefs) */
+               if (back->full_backref || !back->is_data)
+                       continue;
+
                dback = (struct data_backref *)back;
 
                /* We found this one, we don't need to do a lookup */
                if (dback->found_ref)
                        continue;
-               /* Don't care about full backrefs (poor unloved backrefs) */
-               if (back->full_backref)
-                       continue;
+
                key.objectid = dback->root;
                key.type = BTRFS_ROOT_ITEM_KEY;
                key.offset = (u64)-1;
@@ -4873,12 +5694,20 @@ static int fixup_extent_refs(struct btrfs_trans_handle *trans,
        int allocated = 0;
        u64 flags = 0;
 
-       /* remember our flags for recreating the extent */
-       ret = btrfs_lookup_extent_info(NULL, info->extent_root, rec->start,
-                                      rec->max_size, rec->metadata, NULL,
-                                      &flags);
-       if (ret < 0)
-               flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
+       /*
+        * remember our flags for recreating the extent.
+        * FIXME, if we have cleared extent tree, we can not
+        * lookup extent info in extent tree.
+        */
+       if (!init_extent_tree) {
+               ret = btrfs_lookup_extent_info(NULL, info->extent_root,
+                                       rec->start, rec->max_size,
+                                       rec->metadata, NULL, &flags);
+               if (ret < 0)
+                       flags = 0;
+       } else {
+               flags = 0;
+       }
 
        path = btrfs_alloc_path();
        if (!path)
@@ -5409,7 +6238,7 @@ static int check_device_used(struct device_record *dev_rec,
                if (dev_extent_rec->objectid != dev_rec->devid)
                        break;
 
-               list_del(&dev_extent_rec->device_list);
+               list_del_init(&dev_extent_rec->device_list);
                total_byte += dev_extent_rec->length;
                cache = next_cache_extent(cache);
        }
@@ -5526,7 +6355,8 @@ again:
        btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
        ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
                                        &key, &path, 0, 0);
-       BUG_ON(ret < 0);
+       if (ret < 0)
+               goto out;
        while(1) {
                leaf = path.nodes[0];
                slot = path.slots[0];
@@ -5550,6 +6380,10 @@ again:
                                                      btrfs_level_size(root,
                                                      btrfs_root_level(&ri)),
                                                      0);
+                               if (!buf) {
+                                       ret = -EIO;
+                                       goto out;
+                               }
                                add_root_to_pending(buf, &extent_cache,
                                                    &pending, &seen, &nodes,
                                                    &found_key);
@@ -5571,9 +6405,9 @@ again:
        }
        btrfs_release_path(&path);
        while (1) {
-               ret = run_next_block(root, bits, bits_nr, &last, &pending,
-                                    &seen, &reada, &nodes, &extent_cache,
-                                    &chunk_cache, &dev_cache,
+               ret = run_next_block(trans, root, bits, bits_nr, &last,
+                                    &pending, &seen, &reada, &nodes,
+                                    &extent_cache, &chunk_cache, &dev_cache,
                                     &block_group_cache, &dev_extent_cache,
                                     NULL);
                if (ret != 0)
@@ -5594,10 +6428,14 @@ again:
                                      btrfs_root_bytenr(&rec->ri),
                                      btrfs_level_size(root,
                                      btrfs_root_level(&rec->ri)), 0);
+               if (!buf) {
+                       ret = -EIO;
+                       goto out;
+               }
                add_root_to_pending(buf, &extent_cache, &pending,
                                    &seen, &nodes, &rec->found_key);
                while (1) {
-                       ret = run_next_block(root, bits, bits_nr, &last,
+                       ret = run_next_block(trans, root, bits, bits_nr, &last,
                                             &pending, &seen, &reada,
                                             &nodes, &extent_cache,
                                             &chunk_cache, &dev_cache,
@@ -5612,7 +6450,8 @@ again:
                free(rec);
        }
 
-       ret = check_extent_refs(trans, root, &extent_cache);
+       if (ret >= 0)
+               ret = check_extent_refs(trans, root, &extent_cache);
        if (ret == -EAGAIN) {
                ret = btrfs_commit_transaction(trans, root);
                if (ret)
@@ -5629,6 +6468,10 @@ again:
                free_extent_cache_tree(&pending);
                free_extent_cache_tree(&reada);
                free_extent_cache_tree(&nodes);
+               free_chunk_cache_tree(&chunk_cache);
+               free_block_group_tree(&block_group_cache);
+               free_device_cache_tree(&dev_cache);
+               free_device_extent_tree(&dev_extent_cache);
                free_extent_record_cache(root->fs_info, &extent_cache);
                goto again;
        }
@@ -5642,12 +6485,12 @@ again:
        if (err && !ret)
                ret = err;
 
+out:
        if (trans) {
                err = btrfs_commit_transaction(trans, root);
                if (!ret)
                        ret = err;
        }
-out:
        if (repair) {
                free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
                root->fs_info->fsck_extent_cache = NULL;
@@ -5659,6 +6502,10 @@ out:
        free_device_cache_tree(&dev_cache);
        free_block_group_tree(&block_group_cache);
        free_device_extent_tree(&dev_extent_cache);
+       free_extent_cache_tree(&seen);
+       free_extent_cache_tree(&pending);
+       free_extent_cache_tree(&reada);
+       free_extent_cache_tree(&nodes);
        return ret;
 }
 
@@ -5668,6 +6515,7 @@ static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
        struct extent_buffer *c;
        struct extent_buffer *old = root->node;
        int level;
+       int ret;
        struct btrfs_disk_key disk_key = {0,0,0};
 
        level = 0;
@@ -5684,6 +6532,7 @@ static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
        if (IS_ERR(c)) {
                c = old;
                extent_buffer_get(c);
+               overwrite = 1;
        }
 init:
        memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
@@ -5701,7 +6550,26 @@ init:
                            BTRFS_UUID_SIZE);
 
        btrfs_mark_buffer_dirty(c);
-
+       /*
+        * this case can happen in the following case:
+        *
+        * 1.overwrite previous root.
+        *
+        * 2.reinit reloc data root, this is because we skip pin
+        * down reloc data tree before which means we can allocate
+        * same block bytenr here.
+        */
+       if (old->start == c->start) {
+               btrfs_set_root_generation(&root->root_item,
+                                         trans->transid);
+               root->root_item.level = btrfs_header_level(root->node);
+               ret = btrfs_update_root(trans, root->fs_info->tree_root,
+                                       &root->root_key, &root->root_item);
+               if (ret) {
+                       free_extent_buffer(c);
+                       return ret;
+               }
+       }
        free_extent_buffer(old);
        root->node = c;
        add_root_to_dirty_list(root);
@@ -5721,6 +6589,15 @@ static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
        int ret;
        int i;
 
+       /*
+        * If we have pinned this block before, don't pin it again.
+        * This can not only avoid forever loop with broken filesystem
+        * but also give us some speedups.
+        */
+       if (test_range_bit(&fs_info->pinned_extents, eb->start,
+                          eb->start + eb->len - 1, EXTENT_DIRTY, 0))
+               return 0;
+
        btrfs_pin_extent(fs_info, eb->start, eb->len);
 
        leafsize = btrfs_super_leafsize(fs_info->super_copy);
@@ -5792,11 +6669,13 @@ 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 extent_buffer *leaf;
        struct btrfs_chunk *chunk;
        struct btrfs_key key;
        int ret;
+       u64 start;
 
        path = btrfs_alloc_path();
        if (!path)
@@ -5847,8 +6726,19 @@ static int reset_block_groups(struct btrfs_fs_info *fs_info)
                                      btrfs_chunk_type(leaf, chunk),
                                      key.objectid, key.offset,
                                      btrfs_chunk_length(leaf, chunk));
+               set_extent_dirty(&fs_info->free_space_cache, key.offset,
+                                key.offset + btrfs_chunk_length(leaf, chunk),
+                                GFP_NOFS);
                path->slots[0]++;
        }
+       start = 0;
+       while (1) {
+               cache = btrfs_lookup_first_block_group(fs_info, start);
+               if (!cache)
+                       break;
+               cache->cached = 1;
+               start = cache->key.objectid + cache->key.offset;
+       }
 
        btrfs_free_path(path);
        return 0;
@@ -5877,7 +6767,10 @@ static int reset_balance(struct btrfs_trans_handle *trans,
        if (ret) {
                if (ret > 0)
                        ret = 0;
-               goto out;
+               if (!ret)
+                       goto reinit_data_reloc;
+               else
+                       goto out;
        }
 
        ret = btrfs_del_item(trans, root, path);
@@ -5939,6 +6832,7 @@ static int reset_balance(struct btrfs_trans_handle *trans,
        }
        btrfs_release_path(path);
 
+reinit_data_reloc:
        key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
        key.type = BTRFS_ROOT_ITEM_KEY;
        key.offset = (u64)-1;
@@ -5947,21 +6841,19 @@ static int reset_balance(struct btrfs_trans_handle *trans,
                fprintf(stderr, "Error reading data reloc tree\n");
                return PTR_ERR(root);
        }
-       root->track_dirty = 1;
-       if (root->last_trans != trans->transid) {
-               root->last_trans = trans->transid;
-               root->commit_root = root->node;
-               extent_buffer_get(root->node);
-       }
+       record_root_in_trans(trans, root);
        ret = btrfs_fsck_reinit_root(trans, root, 0);
+       if (ret)
+               goto out;
+       ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 out:
        btrfs_free_path(path);
        return ret;
 }
 
-static int reinit_extent_tree(struct btrfs_fs_info *fs_info)
+static int reinit_extent_tree(struct btrfs_trans_handle *trans,
+                             struct btrfs_fs_info *fs_info)
 {
-       struct btrfs_trans_handle *trans;
        u64 start = 0;
        int ret;
 
@@ -5980,12 +6872,6 @@ static int reinit_extent_tree(struct btrfs_fs_info *fs_info)
                return -EINVAL;
        }
 
-       trans = btrfs_start_transaction(fs_info->extent_root, 1);
-       if (IS_ERR(trans)) {
-               fprintf(stderr, "Error starting transaction\n");
-               return PTR_ERR(trans);
-       }
-
        /*
         * first we need to walk all of the trees except the extent tree and pin
         * down the bytes that are in use so we don't overwrite any existing
@@ -6020,12 +6906,6 @@ static int reinit_extent_tree(struct btrfs_fs_info *fs_info)
                return ret;
        }
 
-       ret = reset_balance(trans, fs_info);
-       if (ret) {
-               fprintf(stderr, "error reseting the pending balance\n");
-               return ret;
-       }
-
        /*
         * Now we have all the in-memory block groups setup so we can make
         * allocations properly, and the metadata we care about is safe since we
@@ -6048,11 +6928,11 @@ static int reinit_extent_tree(struct btrfs_fs_info *fs_info)
                btrfs_extent_post_op(trans, fs_info->extent_root);
        }
 
-       /*
-        * Ok now we commit and run the normal fsck, which will add extent
-        * entries for all of the items it finds.
-        */
-       return btrfs_commit_transaction(trans, fs_info->extent_root);
+       ret = reset_balance(trans, fs_info);
+       if (ret)
+               fprintf(stderr, "error reseting the pending balance\n");
+
+       return ret;
 }
 
 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
@@ -6096,12 +6976,166 @@ static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb
        return ret;
 }
 
+static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
+{
+       struct btrfs_path *path;
+       struct btrfs_trans_handle *trans;
+       struct btrfs_key key;
+       int ret;
+
+       printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
+              bad->key.type, bad->key.offset);
+       key.objectid = bad->root_id;
+       key.type = BTRFS_ROOT_ITEM_KEY;
+       key.offset = (u64)-1;
+
+       root = btrfs_read_fs_root(root->fs_info, &key);
+       if (IS_ERR(root)) {
+               fprintf(stderr, "Couldn't find owner root %llu\n",
+                       key.objectid);
+               return PTR_ERR(root);
+       }
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       trans = btrfs_start_transaction(root, 1);
+       if (IS_ERR(trans)) {
+               btrfs_free_path(path);
+               return PTR_ERR(trans);
+       }
+
+       ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
+       if (ret) {
+               if (ret > 0)
+                       ret = 0;
+               goto out;
+       }
+       ret = btrfs_del_item(trans, root, path);
+out:
+       btrfs_commit_transaction(trans, root);
+       btrfs_free_path(path);
+       return ret;
+}
+
+static int zero_log_tree(struct btrfs_root *root)
+{
+       struct btrfs_trans_handle *trans;
+       int ret;
+
+       trans = btrfs_start_transaction(root, 1);
+       if (IS_ERR(trans)) {
+               ret = PTR_ERR(trans);
+               return ret;
+       }
+       btrfs_set_super_log_root(root->fs_info->super_copy, 0);
+       btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
+       ret = btrfs_commit_transaction(trans, root);
+       return ret;
+}
+
+static int populate_csum(struct btrfs_trans_handle *trans,
+                        struct btrfs_root *csum_root, char *buf, u64 start,
+                        u64 len)
+{
+       u64 offset = 0;
+       u64 sectorsize;
+       int ret = 0;
+
+       while (offset < len) {
+               sectorsize = csum_root->sectorsize;
+               ret = read_extent_data(csum_root, buf, start + offset,
+                                      &sectorsize, 0);
+               if (ret)
+                       break;
+               ret = btrfs_csum_file_block(trans, csum_root, start + len,
+                                           start + offset, buf, sectorsize);
+               if (ret)
+                       break;
+               offset += sectorsize;
+       }
+       return ret;
+}
+
+static int fill_csum_tree(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_extent_item *ei;
+       struct extent_buffer *leaf;
+       char *buf;
+       struct btrfs_key key;
+       int ret;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       key.objectid = 0;
+       key.type = BTRFS_EXTENT_ITEM_KEY;
+       key.offset = 0;
+
+       ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
+       if (ret < 0) {
+               btrfs_free_path(path);
+               return ret;
+       }
+
+       buf = malloc(csum_root->sectorsize);
+       if (!buf) {
+               btrfs_free_path(path);
+               return -ENOMEM;
+       }
+
+       while (1) {
+               if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
+                       ret = btrfs_next_leaf(extent_root, path);
+                       if (ret < 0)
+                               break;
+                       if (ret) {
+                               ret = 0;
+                               break;
+                       }
+               }
+               leaf = path->nodes[0];
+
+               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+               if (key.type != BTRFS_EXTENT_ITEM_KEY) {
+                       path->slots[0]++;
+                       continue;
+               }
+
+               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]++;
+                       continue;
+               }
+
+               ret = populate_csum(trans, csum_root, buf, key.objectid,
+                                   key.offset);
+               if (ret)
+                       break;
+               path->slots[0]++;
+       }
+
+       btrfs_free_path(path);
+       free(buf);
+       return ret;
+}
+
 static struct option long_options[] = {
        { "super", 1, NULL, 's' },
        { "repair", 0, NULL, 0 },
        { "init-csum-tree", 0, NULL, 0 },
        { "init-extent-tree", 0, NULL, 0 },
+       { "check-data-csum", 0, NULL, 0 },
        { "backup", 0, NULL, 0 },
+       { "subvol-extents", no_argument, NULL, 'E' },
+       { "qgroup-report", 0, NULL, 'Q' },
        { NULL, 0, NULL, 0}
 };
 
@@ -6114,6 +7148,9 @@ const char * const cmd_check_usage[] = {
        "--repair                    try to repair the filesystem",
        "--init-csum-tree            create a new CRC tree",
        "--init-extent-tree          create a new extent tree",
+       "--check-data-csum           verify checkums of data blocks",
+       "--qgroup-report             print a report on qgroup consistency",
+       "--subvol-extents            print subvolume extents and sharing state",
        NULL
 };
 
@@ -6123,13 +7160,14 @@ int cmd_check(int argc, char **argv)
        struct btrfs_root *root;
        struct btrfs_fs_info *info;
        u64 bytenr = 0;
+       u64 subvolid = 0;
        char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
        int ret;
-       int num;
+       u64 num;
        int option_index = 0;
        int init_csum_tree = 0;
-       int init_extent_tree = 0;
-       enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_PARTIAL;
+       int qgroup_report = 0;
+       enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_EXCLUSIVE;
 
        while(1) {
                int c;
@@ -6143,11 +7181,23 @@ int cmd_check(int argc, char **argv)
                                ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
                                break;
                        case 's':
-                               num = atol(optarg);
-                               bytenr = btrfs_sb_offset(num);
-                               printf("using SB copy %d, bytenr %llu\n", num,
+                               num = arg_strtou64(optarg);
+                               if (num >= BTRFS_SUPER_MIRROR_MAX) {
+                                       fprintf(stderr,
+                                               "ERROR: super mirror should be less than: %d\n",
+                                               BTRFS_SUPER_MIRROR_MAX);
+                                       exit(1);
+                               }
+                               bytenr = btrfs_sb_offset(((int)num));
+                               printf("using SB copy %llu, bytenr %llu\n", num,
                                       (unsigned long long)bytenr);
                                break;
+                       case 'Q':
+                               qgroup_report = 1;
+                               break;
+                       case 'E':
+                               subvolid = arg_strtou64(optarg);
+                               break;
                        case '?':
                        case 'h':
                                usage(cmd_check_usage);
@@ -6159,18 +7209,20 @@ int cmd_check(int argc, char **argv)
                } else if (option_index == 2) {
                        printf("Creating a new CRC tree\n");
                        init_csum_tree = 1;
+                       repair = 1;
                        ctree_flags |= OPEN_CTREE_WRITES;
                } else if (option_index == 3) {
                        init_extent_tree = 1;
                        ctree_flags |= (OPEN_CTREE_WRITES |
                                        OPEN_CTREE_NO_BLOCK_GROUPS);
                        repair = 1;
+               } else if (option_index == 4) {
+                       check_data_csum = 1;
                }
-
        }
        argc = argc - optind;
 
-       if (argc != 1)
+       if (check_argc_exact(argc, 1))
                usage(cmd_check_usage);
 
        radix_tree_init();
@@ -6178,62 +7230,119 @@ int cmd_check(int argc, char **argv)
 
        if((ret = check_mounted(argv[optind])) < 0) {
                fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
-               return ret;
+               goto err_out;
        } else if(ret) {
                fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
-               return -EBUSY;
+               ret = -EBUSY;
+               goto err_out;
        }
 
+       /* only allow partial opening under repair mode */
+       if (repair)
+               ctree_flags |= OPEN_CTREE_PARTIAL;
+
        info = open_ctree_fs_info(argv[optind], bytenr, 0, ctree_flags);
        if (!info) {
                fprintf(stderr, "Couldn't open file system\n");
-               return -EIO;
+               ret = -EIO;
+               goto err_out;
+       }
+
+       root = info->fs_root;
+       /*
+        * repair mode will force us to commit transaction which
+        * will make us fail to load log tree when mounting.
+        */
+       if (repair && btrfs_super_log_root(info->super_copy)) {
+               ret = ask_user("repair mode will force to clear out log tree, Are you sure?");
+               if (!ret) {
+                       ret = 1;
+                       goto close_out;
+               }
+               ret = zero_log_tree(root);
+               if (ret) {
+                       fprintf(stderr, "fail to zero log tree\n");
+                       goto close_out;
+               }
        }
 
        uuid_unparse(info->super_copy->fsid, uuidbuf);
+       if (qgroup_report) {
+               printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
+                      uuidbuf);
+               ret = qgroup_verify_all(info);
+               if (ret == 0)
+                       print_qgroup_report(1);
+               goto close_out;
+       }
+       if (subvolid) {
+               printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
+                      subvolid, argv[optind], uuidbuf);
+               ret = print_extent_state(info, subvolid);
+               goto close_out;
+       }
        printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
 
        if (!extent_buffer_uptodate(info->tree_root->node) ||
            !extent_buffer_uptodate(info->dev_root->node) ||
            !extent_buffer_uptodate(info->chunk_root->node)) {
                fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
-               return -EIO;
-       }
-
-       root = info->fs_root;
-       if (init_extent_tree) {
-               printf("Creating a new extent tree\n");
-               ret = reinit_extent_tree(info);
-               if (ret)
-                       return ret;
-       }
-       if (!extent_buffer_uptodate(info->extent_root->node)) {
-               fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
-               return -EIO;
+               ret = -EIO;
+               goto close_out;
        }
 
-       fprintf(stderr, "checking extents\n");
-       if (init_csum_tree) {
+       if (init_extent_tree || init_csum_tree) {
                struct btrfs_trans_handle *trans;
 
-               fprintf(stderr, "Reinit crc root\n");
-               trans = btrfs_start_transaction(info->csum_root, 1);
+               trans = btrfs_start_transaction(info->extent_root, 0);
                if (IS_ERR(trans)) {
                        fprintf(stderr, "Error starting transaction\n");
-                       return PTR_ERR(trans);
+                       ret = PTR_ERR(trans);
+                       goto close_out;
                }
 
-               ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
-               if (ret) {
-                       fprintf(stderr, "crc root initialization failed\n");
-                       return -EIO;
+               if (init_extent_tree) {
+                       printf("Creating a new extent tree\n");
+                       ret = reinit_extent_tree(trans, info);
+                       if (ret)
+                               goto close_out;
                }
 
-               ret = btrfs_commit_transaction(trans, info->csum_root);
+               if (init_csum_tree) {
+                       fprintf(stderr, "Reinit crc root\n");
+                       ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
+                       if (ret) {
+                               fprintf(stderr, "crc root initialization failed\n");
+                               ret = -EIO;
+                               goto close_out;
+                       }
+
+                       ret = fill_csum_tree(trans, info->csum_root);
+                       if (ret) {
+                               fprintf(stderr, "crc refilling failed\n");
+                               return -EIO;
+                       }
+               }
+               /*
+                * Ok now we commit and run the normal fsck, which will add
+                * extent entries for all of the items it finds.
+                */
+               ret = btrfs_commit_transaction(trans, info->extent_root);
                if (ret)
-                       exit(1);
-               goto out;
+                       goto close_out;
        }
+       if (!extent_buffer_uptodate(info->extent_root->node)) {
+               fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
+               ret = -EIO;
+               goto close_out;
+       }
+       if (!extent_buffer_uptodate(info->csum_root->node)) {
+               fprintf(stderr, "Checksum root corrupted, rerun with --init-csum-tree option\n");
+               ret = -EIO;
+               goto close_out;
+       }
+
+       fprintf(stderr, "checking extents\n");
        ret = check_chunks_and_extents(root);
        if (ret)
                fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
@@ -6243,6 +7352,14 @@ int cmd_check(int argc, char **argv)
        if (ret)
                goto out;
 
+       /*
+        * We used to have to have these hole extents in between our real
+        * extents so if we don't have this flag set we need to make sure there
+        * are no gaps in the file extents for inodes, otherwise we can just
+        * ignore it when this happens.
+        */
+       no_holes = btrfs_fs_incompat(root->fs_info,
+                                    BTRFS_FEATURE_INCOMPAT_NO_HOLES);
        fprintf(stderr, "checking fs roots\n");
        ret = check_fs_roots(root, &root_cache);
        if (ret)
@@ -6263,19 +7380,36 @@ int cmd_check(int argc, char **argv)
 
                eb = list_first_entry(&root->fs_info->recow_ebs,
                                      struct extent_buffer, recow);
+               list_del_init(&eb->recow);
                ret = recow_extent_buffer(root, eb);
                if (ret)
                        break;
        }
 
+       while (!list_empty(&delete_items)) {
+               struct bad_item *bad;
+
+               bad = list_first_entry(&delete_items, struct bad_item, list);
+               list_del_init(&bad->list);
+               if (repair)
+                       ret = delete_bad_item(root, bad);
+               free(bad);
+       }
+
+       if (info->quota_enabled) {
+               int err;
+               fprintf(stderr, "checking quota groups\n");
+               err = qgroup_verify_all(info);
+               if (err)
+                       goto out;
+       }
+
        if (!list_empty(&root->fs_info->recow_ebs)) {
                fprintf(stderr, "Transid errors in file system\n");
                ret = 1;
        }
 out:
-       free_root_recs_tree(&root_cache);
-       close_ctree(root);
-
+       print_qgroup_report(0);
        if (found_old_backref) { /*
                 * there was a disk format change when mixed
                 * backref was in testing tree. The old format
@@ -6302,5 +7436,10 @@ out:
                (unsigned long long)data_bytes_allocated,
                (unsigned long long)data_bytes_referenced);
        printf("%s\n", BTRFS_BUILD_VERSION);
+
+       free_root_recs_tree(&root_cache);
+close_out:
+       close_ctree(root);
+err_out:
        return ret;
 }