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)
29 len = btrfs_item_size_nr(eb, slot);
30 read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
31 min_t(u32, len, sizeof(*item)));
32 if (len < sizeof(*item))
34 if (!need_reset && btrfs_root_generation(item)
35 != btrfs_root_generation_v2(item)) {
36 if (btrfs_root_generation_v2(item) != 0) {
37 btrfs_warn(eb->fs_info,
38 "mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
43 memset(&item->generation_v2, 0,
44 sizeof(*item) - offsetof(struct btrfs_root_item,
48 memcpy(item->uuid, uuid.b, BTRFS_UUID_SIZE);
53 * btrfs_find_root - lookup the root by the key.
54 * root: the root of the root tree
55 * search_key: the key to search
56 * path: the path we search
57 * root_item: the root item of the tree we look for
58 * root_key: the root key of the tree we look for
60 * If ->offset of 'search_key' is -1ULL, it means we are not sure the offset
61 * of the search key, just lookup the root with the highest offset for a
64 * If we find something return 0, otherwise > 0, < 0 on error.
66 int btrfs_find_root(struct btrfs_root *root, const struct btrfs_key *search_key,
67 struct btrfs_path *path, struct btrfs_root_item *root_item,
68 struct btrfs_key *root_key)
70 struct btrfs_key found_key;
71 struct extent_buffer *l;
75 ret = btrfs_search_slot(NULL, root, search_key, path, 0, 0);
79 if (search_key->offset != -1ULL) { /* the search key is exact */
83 BUG_ON(ret == 0); /* Logical error */
84 if (path->slots[0] == 0)
91 slot = path->slots[0];
93 btrfs_item_key_to_cpu(l, &found_key, slot);
94 if (found_key.objectid != search_key->objectid ||
95 found_key.type != BTRFS_ROOT_ITEM_KEY) {
101 btrfs_read_root_item(l, slot, root_item);
103 memcpy(root_key, &found_key, sizeof(found_key));
105 btrfs_release_path(path);
109 void btrfs_set_root_node(struct btrfs_root_item *item,
110 struct extent_buffer *node)
112 btrfs_set_root_bytenr(item, node->start);
113 btrfs_set_root_level(item, btrfs_header_level(node));
114 btrfs_set_root_generation(item, btrfs_header_generation(node));
118 * copy the data in 'item' into the btree
120 int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
121 *root, struct btrfs_key *key, struct btrfs_root_item
124 struct btrfs_fs_info *fs_info = root->fs_info;
125 struct btrfs_path *path;
126 struct extent_buffer *l;
132 path = btrfs_alloc_path();
136 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
142 "unable to find root key (%llu %u %llu) in tree %llu",
143 key->objectid, key->type, key->offset,
144 root->root_key.objectid);
146 btrfs_abort_transaction(trans, ret);
151 slot = path->slots[0];
152 ptr = btrfs_item_ptr_offset(l, slot);
153 old_len = btrfs_item_size_nr(l, slot);
156 * If this is the first time we update the root item which originated
157 * from an older kernel, we need to enlarge the item size to make room
158 * for the added fields.
160 if (old_len < sizeof(*item)) {
161 btrfs_release_path(path);
162 ret = btrfs_search_slot(trans, root, key, path,
165 btrfs_abort_transaction(trans, ret);
169 ret = btrfs_del_item(trans, root, path);
171 btrfs_abort_transaction(trans, ret);
174 btrfs_release_path(path);
175 ret = btrfs_insert_empty_item(trans, root, path,
178 btrfs_abort_transaction(trans, ret);
182 slot = path->slots[0];
183 ptr = btrfs_item_ptr_offset(l, slot);
187 * Update generation_v2 so at the next mount we know the new root
190 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
192 write_extent_buffer(l, item, ptr, sizeof(*item));
193 btrfs_mark_buffer_dirty(path->nodes[0]);
195 btrfs_free_path(path);
199 int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
200 const struct btrfs_key *key, struct btrfs_root_item *item)
203 * Make sure generation v1 and v2 match. See update_root for details.
205 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
206 return btrfs_insert_item(trans, root, key, item, sizeof(*item));
209 int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
211 struct btrfs_root *tree_root = fs_info->tree_root;
212 struct extent_buffer *leaf;
213 struct btrfs_path *path;
214 struct btrfs_key key;
215 struct btrfs_key root_key;
216 struct btrfs_root *root;
220 path = btrfs_alloc_path();
224 key.objectid = BTRFS_ORPHAN_OBJECTID;
225 key.type = BTRFS_ORPHAN_ITEM_KEY;
228 root_key.type = BTRFS_ROOT_ITEM_KEY;
229 root_key.offset = (u64)-1;
232 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
238 leaf = path->nodes[0];
239 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
240 ret = btrfs_next_leaf(tree_root, path);
245 leaf = path->nodes[0];
248 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
249 btrfs_release_path(path);
251 if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
252 key.type != BTRFS_ORPHAN_ITEM_KEY)
255 root_key.objectid = key.offset;
259 * The root might have been inserted already, as before we look
260 * for orphan roots, log replay might have happened, which
261 * triggers a transaction commit and qgroup accounting, which
262 * in turn reads and inserts fs roots while doing backref
265 root = btrfs_lookup_fs_root(fs_info, root_key.objectid);
267 WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
269 if (btrfs_root_refs(&root->root_item) == 0) {
270 set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
271 btrfs_add_dead_root(root);
276 root = btrfs_read_fs_root(tree_root, &root_key);
277 err = PTR_ERR_OR_ZERO(root);
278 if (err && err != -ENOENT) {
280 } else if (err == -ENOENT) {
281 struct btrfs_trans_handle *trans;
283 btrfs_release_path(path);
285 trans = btrfs_join_transaction(tree_root);
287 err = PTR_ERR(trans);
288 btrfs_handle_fs_error(fs_info, err,
289 "Failed to start trans to delete orphan item");
292 err = btrfs_del_orphan_item(trans, tree_root,
294 btrfs_end_transaction(trans);
296 btrfs_handle_fs_error(fs_info, err,
297 "Failed to delete root orphan item");
303 err = btrfs_init_fs_root(root);
305 btrfs_free_fs_root(root);
309 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
311 err = btrfs_insert_fs_root(fs_info, root);
313 BUG_ON(err == -EEXIST);
314 btrfs_free_fs_root(root);
318 if (btrfs_root_refs(&root->root_item) == 0) {
319 set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
320 btrfs_add_dead_root(root);
324 btrfs_free_path(path);
328 /* drop the root item for 'key' from the tree root */
329 int btrfs_del_root(struct btrfs_trans_handle *trans,
330 const struct btrfs_key *key)
332 struct btrfs_root *root = trans->fs_info->tree_root;
333 struct btrfs_path *path;
336 path = btrfs_alloc_path();
339 ret = btrfs_search_slot(trans, root, key, path, -1, 1);
345 ret = btrfs_del_item(trans, root, path);
347 btrfs_free_path(path);
351 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
352 u64 ref_id, u64 dirid, u64 *sequence, const char *name,
356 struct btrfs_root *tree_root = trans->fs_info->tree_root;
357 struct btrfs_path *path;
358 struct btrfs_root_ref *ref;
359 struct extent_buffer *leaf;
360 struct btrfs_key key;
365 path = btrfs_alloc_path();
369 key.objectid = root_id;
370 key.type = BTRFS_ROOT_BACKREF_KEY;
373 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
376 leaf = path->nodes[0];
377 ref = btrfs_item_ptr(leaf, path->slots[0],
378 struct btrfs_root_ref);
380 WARN_ON(btrfs_root_ref_dirid(leaf, ref) != dirid);
381 WARN_ON(btrfs_root_ref_name_len(leaf, ref) != name_len);
382 ptr = (unsigned long)(ref + 1);
383 WARN_ON(memcmp_extent_buffer(leaf, name, ptr, name_len));
384 *sequence = btrfs_root_ref_sequence(leaf, ref);
386 ret = btrfs_del_item(trans, tree_root, path);
394 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
395 btrfs_release_path(path);
396 key.objectid = ref_id;
397 key.type = BTRFS_ROOT_REF_KEY;
398 key.offset = root_id;
403 btrfs_free_path(path);
408 * add a btrfs_root_ref item. type is either BTRFS_ROOT_REF_KEY
409 * or BTRFS_ROOT_BACKREF_KEY.
411 * The dirid, sequence, name and name_len refer to the directory entry
412 * that is referencing the root.
414 * For a forward ref, the root_id is the id of the tree referencing
415 * the root and ref_id is the id of the subvol or snapshot.
417 * For a back ref the root_id is the id of the subvol or snapshot and
418 * ref_id is the id of the tree referencing it.
420 * Will return 0, -ENOMEM, or anything from the CoW path
422 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
423 u64 ref_id, u64 dirid, u64 sequence, const char *name,
426 struct btrfs_root *tree_root = trans->fs_info->tree_root;
427 struct btrfs_key key;
429 struct btrfs_path *path;
430 struct btrfs_root_ref *ref;
431 struct extent_buffer *leaf;
434 path = btrfs_alloc_path();
438 key.objectid = root_id;
439 key.type = BTRFS_ROOT_BACKREF_KEY;
442 ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
443 sizeof(*ref) + name_len);
445 btrfs_abort_transaction(trans, ret);
446 btrfs_free_path(path);
450 leaf = path->nodes[0];
451 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
452 btrfs_set_root_ref_dirid(leaf, ref, dirid);
453 btrfs_set_root_ref_sequence(leaf, ref, sequence);
454 btrfs_set_root_ref_name_len(leaf, ref, name_len);
455 ptr = (unsigned long)(ref + 1);
456 write_extent_buffer(leaf, name, ptr, name_len);
457 btrfs_mark_buffer_dirty(leaf);
459 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
460 btrfs_release_path(path);
461 key.objectid = ref_id;
462 key.type = BTRFS_ROOT_REF_KEY;
463 key.offset = root_id;
467 btrfs_free_path(path);
472 * Old btrfs forgets to init root_item->flags and root_item->byte_limit
473 * for subvolumes. To work around this problem, we steal a bit from
474 * root_item->inode_item->flags, and use it to indicate if those fields
475 * have been properly initialized.
477 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
479 u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
481 if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
482 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
483 btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
484 btrfs_set_root_flags(root_item, 0);
485 btrfs_set_root_limit(root_item, 0);
489 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
490 struct btrfs_root *root)
492 struct btrfs_root_item *item = &root->root_item;
493 struct timespec64 ct;
495 ktime_get_real_ts64(&ct);
496 spin_lock(&root->root_item_lock);
497 btrfs_set_root_ctransid(item, trans->transid);
498 btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
499 btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
500 spin_unlock(&root->root_item_lock);
504 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
505 * root: the root of the parent directory
506 * rsv: block reservation
507 * items: the number of items that we need do reservation
508 * use_global_rsv: allow fallback to the global block reservation
510 * This function is used to reserve the space for snapshot/subvolume
511 * creation and deletion. Those operations are different with the
512 * common file/directory operations, they change two fs/file trees
513 * and root tree, the number of items that the qgroup reserves is
514 * different with the free space reservation. So we can not use
515 * the space reservation mechanism in start_transaction().
517 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
518 struct btrfs_block_rsv *rsv, int items,
521 u64 qgroup_num_bytes = 0;
524 struct btrfs_fs_info *fs_info = root->fs_info;
525 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
527 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
528 /* One for parent inode, two for dir entries */
529 qgroup_num_bytes = 3 * fs_info->nodesize;
530 ret = btrfs_qgroup_reserve_meta_prealloc(root,
531 qgroup_num_bytes, true);
536 num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
537 rsv->space_info = btrfs_find_space_info(fs_info,
538 BTRFS_BLOCK_GROUP_METADATA);
539 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
540 BTRFS_RESERVE_FLUSH_ALL);
542 if (ret == -ENOSPC && use_global_rsv)
543 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
545 if (ret && qgroup_num_bytes)
546 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
551 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
552 struct btrfs_block_rsv *rsv)
554 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);