1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/uuid.h>
9 #include "transaction.h"
11 #include "print-tree.h"
13 #include "space-info.h"
16 * Read a root item from the tree. In case we detect a root item smaller then
17 * sizeof(root_item), we know it's an old version of the root structure and
18 * initialize all new fields to zero. The same happens if we detect mismatching
19 * generation numbers as then we know the root was once mounted with an older
20 * kernel that was not aware of the root item structure change.
22 static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
23 struct btrfs_root_item *item)
28 len = btrfs_item_size(eb, slot);
29 read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
30 min_t(u32, len, sizeof(*item)));
31 if (len < sizeof(*item))
33 if (!need_reset && btrfs_root_generation(item)
34 != btrfs_root_generation_v2(item)) {
35 if (btrfs_root_generation_v2(item) != 0) {
36 btrfs_warn(eb->fs_info,
37 "mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
42 /* Clear all members from generation_v2 onwards. */
43 memset_startat(item, 0, generation_v2);
44 generate_random_guid(item->uuid);
49 * btrfs_find_root - lookup the root by the key.
50 * root: the root of the root tree
51 * search_key: the key to search
52 * path: the path we search
53 * root_item: the root item of the tree we look for
54 * root_key: the root key of the tree we look for
56 * If ->offset of 'search_key' is -1ULL, it means we are not sure the offset
57 * of the search key, just lookup the root with the highest offset for a
60 * If we find something return 0, otherwise > 0, < 0 on error.
62 int btrfs_find_root(struct btrfs_root *root, const struct btrfs_key *search_key,
63 struct btrfs_path *path, struct btrfs_root_item *root_item,
64 struct btrfs_key *root_key)
66 struct btrfs_key found_key;
67 struct extent_buffer *l;
71 ret = btrfs_search_slot(NULL, root, search_key, path, 0, 0);
75 if (search_key->offset != -1ULL) { /* the search key is exact */
79 BUG_ON(ret == 0); /* Logical error */
80 if (path->slots[0] == 0)
87 slot = path->slots[0];
89 btrfs_item_key_to_cpu(l, &found_key, slot);
90 if (found_key.objectid != search_key->objectid ||
91 found_key.type != BTRFS_ROOT_ITEM_KEY) {
97 btrfs_read_root_item(l, slot, root_item);
99 memcpy(root_key, &found_key, sizeof(found_key));
101 btrfs_release_path(path);
105 void btrfs_set_root_node(struct btrfs_root_item *item,
106 struct extent_buffer *node)
108 btrfs_set_root_bytenr(item, node->start);
109 btrfs_set_root_level(item, btrfs_header_level(node));
110 btrfs_set_root_generation(item, btrfs_header_generation(node));
114 * copy the data in 'item' into the btree
116 int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
117 *root, struct btrfs_key *key, struct btrfs_root_item
120 struct btrfs_fs_info *fs_info = root->fs_info;
121 struct btrfs_path *path;
122 struct extent_buffer *l;
128 path = btrfs_alloc_path();
132 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
138 "unable to find root key (%llu %u %llu) in tree %llu",
139 key->objectid, key->type, key->offset,
140 root->root_key.objectid);
142 btrfs_abort_transaction(trans, ret);
147 slot = path->slots[0];
148 ptr = btrfs_item_ptr_offset(l, slot);
149 old_len = btrfs_item_size(l, slot);
152 * If this is the first time we update the root item which originated
153 * from an older kernel, we need to enlarge the item size to make room
154 * for the added fields.
156 if (old_len < sizeof(*item)) {
157 btrfs_release_path(path);
158 ret = btrfs_search_slot(trans, root, key, path,
161 btrfs_abort_transaction(trans, ret);
165 ret = btrfs_del_item(trans, root, path);
167 btrfs_abort_transaction(trans, ret);
170 btrfs_release_path(path);
171 ret = btrfs_insert_empty_item(trans, root, path,
174 btrfs_abort_transaction(trans, ret);
178 slot = path->slots[0];
179 ptr = btrfs_item_ptr_offset(l, slot);
183 * Update generation_v2 so at the next mount we know the new root
186 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
188 write_extent_buffer(l, item, ptr, sizeof(*item));
189 btrfs_mark_buffer_dirty(path->nodes[0]);
191 btrfs_free_path(path);
195 int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
196 const struct btrfs_key *key, struct btrfs_root_item *item)
199 * Make sure generation v1 and v2 match. See update_root for details.
201 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
202 return btrfs_insert_item(trans, root, key, item, sizeof(*item));
205 int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
207 struct btrfs_root *tree_root = fs_info->tree_root;
208 struct extent_buffer *leaf;
209 struct btrfs_path *path;
210 struct btrfs_key key;
211 struct btrfs_root *root;
215 path = btrfs_alloc_path();
219 key.objectid = BTRFS_ORPHAN_OBJECTID;
220 key.type = BTRFS_ORPHAN_ITEM_KEY;
226 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
232 leaf = path->nodes[0];
233 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
234 ret = btrfs_next_leaf(tree_root, path);
239 leaf = path->nodes[0];
242 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
243 btrfs_release_path(path);
245 if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
246 key.type != BTRFS_ORPHAN_ITEM_KEY)
249 root_objectid = key.offset;
252 root = btrfs_get_fs_root(fs_info, root_objectid, false);
253 err = PTR_ERR_OR_ZERO(root);
254 if (err && err != -ENOENT) {
256 } else if (err == -ENOENT) {
257 struct btrfs_trans_handle *trans;
259 btrfs_release_path(path);
261 trans = btrfs_join_transaction(tree_root);
263 err = PTR_ERR(trans);
264 btrfs_handle_fs_error(fs_info, err,
265 "Failed to start trans to delete orphan item");
268 err = btrfs_del_orphan_item(trans, tree_root,
270 btrfs_end_transaction(trans);
272 btrfs_handle_fs_error(fs_info, err,
273 "Failed to delete root orphan item");
279 WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
280 if (btrfs_root_refs(&root->root_item) == 0) {
281 struct btrfs_key drop_key;
283 btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
285 * If we have a non-zero drop_progress then we know we
286 * made it partly through deleting this snapshot, and
287 * thus we need to make sure we block any balance from
288 * happening until this snapshot is completely dropped.
290 if (drop_key.objectid != 0 || drop_key.type != 0 ||
291 drop_key.offset != 0) {
292 set_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
293 set_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
296 set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
297 btrfs_add_dead_root(root);
299 btrfs_put_root(root);
302 btrfs_free_path(path);
306 /* drop the root item for 'key' from the tree root */
307 int btrfs_del_root(struct btrfs_trans_handle *trans,
308 const struct btrfs_key *key)
310 struct btrfs_root *root = trans->fs_info->tree_root;
311 struct btrfs_path *path;
314 path = btrfs_alloc_path();
317 ret = btrfs_search_slot(trans, root, key, path, -1, 1);
323 ret = btrfs_del_item(trans, root, path);
325 btrfs_free_path(path);
329 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
330 u64 ref_id, u64 dirid, u64 *sequence, const char *name,
334 struct btrfs_root *tree_root = trans->fs_info->tree_root;
335 struct btrfs_path *path;
336 struct btrfs_root_ref *ref;
337 struct extent_buffer *leaf;
338 struct btrfs_key key;
343 path = btrfs_alloc_path();
347 key.objectid = root_id;
348 key.type = BTRFS_ROOT_BACKREF_KEY;
351 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
355 leaf = path->nodes[0];
356 ref = btrfs_item_ptr(leaf, path->slots[0],
357 struct btrfs_root_ref);
358 ptr = (unsigned long)(ref + 1);
359 if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
360 (btrfs_root_ref_name_len(leaf, ref) != name_len) ||
361 memcmp_extent_buffer(leaf, name, ptr, name_len)) {
365 *sequence = btrfs_root_ref_sequence(leaf, ref);
367 ret = btrfs_del_item(trans, tree_root, path);
375 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
376 btrfs_release_path(path);
377 key.objectid = ref_id;
378 key.type = BTRFS_ROOT_REF_KEY;
379 key.offset = root_id;
384 btrfs_free_path(path);
389 * add a btrfs_root_ref item. type is either BTRFS_ROOT_REF_KEY
390 * or BTRFS_ROOT_BACKREF_KEY.
392 * The dirid, sequence, name and name_len refer to the directory entry
393 * that is referencing the root.
395 * For a forward ref, the root_id is the id of the tree referencing
396 * the root and ref_id is the id of the subvol or snapshot.
398 * For a back ref the root_id is the id of the subvol or snapshot and
399 * ref_id is the id of the tree referencing it.
401 * Will return 0, -ENOMEM, or anything from the CoW path
403 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
404 u64 ref_id, u64 dirid, u64 sequence, const char *name,
407 struct btrfs_root *tree_root = trans->fs_info->tree_root;
408 struct btrfs_key key;
410 struct btrfs_path *path;
411 struct btrfs_root_ref *ref;
412 struct extent_buffer *leaf;
415 path = btrfs_alloc_path();
419 key.objectid = root_id;
420 key.type = BTRFS_ROOT_BACKREF_KEY;
423 ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
424 sizeof(*ref) + name_len);
426 btrfs_abort_transaction(trans, ret);
427 btrfs_free_path(path);
431 leaf = path->nodes[0];
432 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
433 btrfs_set_root_ref_dirid(leaf, ref, dirid);
434 btrfs_set_root_ref_sequence(leaf, ref, sequence);
435 btrfs_set_root_ref_name_len(leaf, ref, name_len);
436 ptr = (unsigned long)(ref + 1);
437 write_extent_buffer(leaf, name, ptr, name_len);
438 btrfs_mark_buffer_dirty(leaf);
440 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
441 btrfs_release_path(path);
442 key.objectid = ref_id;
443 key.type = BTRFS_ROOT_REF_KEY;
444 key.offset = root_id;
448 btrfs_free_path(path);
453 * Old btrfs forgets to init root_item->flags and root_item->byte_limit
454 * for subvolumes. To work around this problem, we steal a bit from
455 * root_item->inode_item->flags, and use it to indicate if those fields
456 * have been properly initialized.
458 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
460 u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
462 if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
463 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
464 btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
465 btrfs_set_root_flags(root_item, 0);
466 btrfs_set_root_limit(root_item, 0);
470 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
471 struct btrfs_root *root)
473 struct btrfs_root_item *item = &root->root_item;
474 struct timespec64 ct;
476 ktime_get_real_ts64(&ct);
477 spin_lock(&root->root_item_lock);
478 btrfs_set_root_ctransid(item, trans->transid);
479 btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
480 btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
481 spin_unlock(&root->root_item_lock);
485 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
486 * root: the root of the parent directory
487 * rsv: block reservation
488 * items: the number of items that we need do reservation
489 * use_global_rsv: allow fallback to the global block reservation
491 * This function is used to reserve the space for snapshot/subvolume
492 * creation and deletion. Those operations are different with the
493 * common file/directory operations, they change two fs/file trees
494 * and root tree, the number of items that the qgroup reserves is
495 * different with the free space reservation. So we can not use
496 * the space reservation mechanism in start_transaction().
498 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
499 struct btrfs_block_rsv *rsv, int items,
502 u64 qgroup_num_bytes = 0;
505 struct btrfs_fs_info *fs_info = root->fs_info;
506 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
508 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
509 /* One for parent inode, two for dir entries */
510 qgroup_num_bytes = 3 * fs_info->nodesize;
511 ret = btrfs_qgroup_reserve_meta_prealloc(root,
512 qgroup_num_bytes, true,
518 num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
519 rsv->space_info = btrfs_find_space_info(fs_info,
520 BTRFS_BLOCK_GROUP_METADATA);
521 ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes,
522 BTRFS_RESERVE_FLUSH_ALL);
524 if (ret == -ENOSPC && use_global_rsv)
525 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
527 if (ret && qgroup_num_bytes)
528 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
531 spin_lock(&rsv->lock);
532 rsv->qgroup_rsv_reserved += qgroup_num_bytes;
533 spin_unlock(&rsv->lock);
538 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
539 struct btrfs_block_rsv *rsv)
541 struct btrfs_fs_info *fs_info = root->fs_info;
542 u64 qgroup_to_release;
544 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, &qgroup_to_release);
545 btrfs_qgroup_convert_reserved_meta(root, qgroup_to_release);