btrfs-progs: add the ability to fix shifted item offsets
[platform/upstream/btrfs-progs.git] / cmds-check.c
index fede71e..75b984f 100644 (file)
@@ -563,6 +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))
@@ -618,9 +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);
@@ -1004,7 +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;
@@ -1650,35 +1657,101 @@ static int add_missing_dir_index(struct btrfs_root *root,
        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)
+                                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 (!backref->found_dir_index && backref->found_inode_ref) {
-                       ret = add_missing_dir_index(root, inode_cache, rec,
-                                                   backref);
+               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 (backref->found_dir_item && backref->found_dir_index) {
-                       if (!backref->errors && backref->found_inode_ref) {
-                               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;
+       }
+       return ret ? ret : repaired;
 }
 
 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
@@ -1716,7 +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);
 
@@ -1730,22 +1805,52 @@ static int check_inode_recs(struct btrfs_root *root,
         * 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.
         */
-       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);
+       while (stage < 3) {
+               stage++;
+               if (stage == 3 && !err)
+                       break;
 
-               if (list_empty(&rec->backrefs))
-                       continue;
-               repair_inode_backrefs(root, rec, inode_cache);
+               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) {
@@ -2237,7 +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;
@@ -2253,6 +2358,7 @@ 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;
@@ -2261,7 +2367,13 @@ static int check_fs_roots(struct btrfs_root *root,
                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);
@@ -2288,6 +2400,11 @@ static int check_fs_roots(struct btrfs_root *root,
                                goto next;
                        }
                        ret = check_fs_root(tmp_root, root_cache, &wc);
+                       if (ret == -EAGAIN) {
+                               free_root_recs_tree(root_cache);
+                               btrfs_release_path(&path);
+                               goto again;
+                       }
                        if (ret)
                                err = 1;
                        if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
@@ -2636,15 +2753,9 @@ static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
        return 0;
 }
 
-/*
- * Attempt to fix basic block failures.  Currently we only handle bad key
- * orders, we will cycle through the keys and swap them if necessary.
- */
-static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
-                               struct btrfs_root *root,
-                               struct extent_buffer *buf,
-                               struct btrfs_disk_key *parent_key,
-                               enum btrfs_tree_block_status status)
+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;
@@ -2652,9 +2763,6 @@ static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
        int level;
        int ret;
 
-       if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
-               return -EIO;
-
        k1.objectid = btrfs_header_owner(buf);
        k1.type = BTRFS_ROOT_ITEM_KEY;
        k1.offset = (u64)-1;
@@ -2705,6 +2813,111 @@ static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
        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,
@@ -2743,7 +2956,6 @@ static int check_block(struct btrfs_trans_handle *trans,
        if (status != BTRFS_TREE_BLOCK_CLEAN) {
                if (repair)
                        status = try_to_fix_bad_block(trans, root, buf,
-                                                     &rec->parent_key,
                                                      status);
                if (status != BTRFS_TREE_BLOCK_CLEAN) {
                        ret = -EIO;
@@ -3217,6 +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);
 }
 
@@ -3253,6 +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);
 }
 
@@ -3286,6 +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);
 }
 
@@ -6019,7 +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);
        }
@@ -6249,6 +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;
        }