1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/uuid.h>
11 #include "transaction.h"
13 #include "print-tree.h"
15 #include "space-info.h"
16 #include "accessors.h"
17 #include "root-tree.h"
21 * Read a root item from the tree. In case we detect a root item smaller then
22 * sizeof(root_item), we know it's an old version of the root structure and
23 * initialize all new fields to zero. The same happens if we detect mismatching
24 * generation numbers as then we know the root was once mounted with an older
25 * kernel that was not aware of the root item structure change.
27 static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
28 struct btrfs_root_item *item)
33 len = btrfs_item_size(eb, slot);
34 read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
35 min_t(u32, len, sizeof(*item)));
36 if (len < sizeof(*item))
38 if (!need_reset && btrfs_root_generation(item)
39 != btrfs_root_generation_v2(item)) {
40 if (btrfs_root_generation_v2(item) != 0) {
41 btrfs_warn(eb->fs_info,
42 "mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
47 /* Clear all members from generation_v2 onwards. */
48 memset_startat(item, 0, generation_v2);
49 generate_random_guid(item->uuid);
54 * Lookup the root by the key.
56 * root: the root of the root tree
57 * search_key: the key to search
58 * path: the path we search
59 * root_item: the root item of the tree we look for
60 * root_key: the root key of the tree we look for
62 * If ->offset of 'search_key' is -1ULL, it means we are not sure the offset
63 * of the search key, just lookup the root with the highest offset for a
66 * If we find something return 0, otherwise > 0, < 0 on error.
68 int btrfs_find_root(struct btrfs_root *root, const struct btrfs_key *search_key,
69 struct btrfs_path *path, struct btrfs_root_item *root_item,
70 struct btrfs_key *root_key)
72 struct btrfs_key found_key;
73 struct extent_buffer *l;
77 ret = btrfs_search_slot(NULL, root, search_key, path, 0, 0);
81 if (search_key->offset != -1ULL) { /* the search key is exact */
85 BUG_ON(ret == 0); /* Logical error */
86 if (path->slots[0] == 0)
93 slot = path->slots[0];
95 btrfs_item_key_to_cpu(l, &found_key, slot);
96 if (found_key.objectid != search_key->objectid ||
97 found_key.type != BTRFS_ROOT_ITEM_KEY) {
103 btrfs_read_root_item(l, slot, root_item);
105 memcpy(root_key, &found_key, sizeof(found_key));
107 btrfs_release_path(path);
111 void btrfs_set_root_node(struct btrfs_root_item *item,
112 struct extent_buffer *node)
114 btrfs_set_root_bytenr(item, node->start);
115 btrfs_set_root_level(item, btrfs_header_level(node));
116 btrfs_set_root_generation(item, btrfs_header_generation(node));
120 * copy the data in 'item' into the btree
122 int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
123 *root, struct btrfs_key *key, struct btrfs_root_item
126 struct btrfs_fs_info *fs_info = root->fs_info;
127 struct btrfs_path *path;
128 struct extent_buffer *l;
134 path = btrfs_alloc_path();
138 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
144 "unable to find root key (%llu %u %llu) in tree %llu",
145 key->objectid, key->type, key->offset,
146 root->root_key.objectid);
148 btrfs_abort_transaction(trans, ret);
153 slot = path->slots[0];
154 ptr = btrfs_item_ptr_offset(l, slot);
155 old_len = btrfs_item_size(l, slot);
158 * If this is the first time we update the root item which originated
159 * from an older kernel, we need to enlarge the item size to make room
160 * for the added fields.
162 if (old_len < sizeof(*item)) {
163 btrfs_release_path(path);
164 ret = btrfs_search_slot(trans, root, key, path,
167 btrfs_abort_transaction(trans, ret);
171 ret = btrfs_del_item(trans, root, path);
173 btrfs_abort_transaction(trans, ret);
176 btrfs_release_path(path);
177 ret = btrfs_insert_empty_item(trans, root, path,
180 btrfs_abort_transaction(trans, ret);
184 slot = path->slots[0];
185 ptr = btrfs_item_ptr_offset(l, slot);
189 * Update generation_v2 so at the next mount we know the new root
192 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
194 write_extent_buffer(l, item, ptr, sizeof(*item));
195 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
197 btrfs_free_path(path);
201 int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
202 const struct btrfs_key *key, struct btrfs_root_item *item)
205 * Make sure generation v1 and v2 match. See update_root for details.
207 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
208 return btrfs_insert_item(trans, root, key, item, sizeof(*item));
211 int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
213 struct btrfs_root *tree_root = fs_info->tree_root;
214 struct extent_buffer *leaf;
215 struct btrfs_path *path;
216 struct btrfs_key key;
217 struct btrfs_root *root;
221 path = btrfs_alloc_path();
225 key.objectid = BTRFS_ORPHAN_OBJECTID;
226 key.type = BTRFS_ORPHAN_ITEM_KEY;
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_objectid = key.offset;
258 root = btrfs_get_fs_root(fs_info, root_objectid, false);
259 err = PTR_ERR_OR_ZERO(root);
260 if (err && err != -ENOENT) {
262 } else if (err == -ENOENT) {
263 struct btrfs_trans_handle *trans;
265 btrfs_release_path(path);
267 trans = btrfs_join_transaction(tree_root);
269 err = PTR_ERR(trans);
270 btrfs_handle_fs_error(fs_info, err,
271 "Failed to start trans to delete orphan item");
274 err = btrfs_del_orphan_item(trans, tree_root,
276 btrfs_end_transaction(trans);
278 btrfs_handle_fs_error(fs_info, err,
279 "Failed to delete root orphan item");
285 WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
286 if (btrfs_root_refs(&root->root_item) == 0) {
287 struct btrfs_key drop_key;
289 btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
291 * If we have a non-zero drop_progress then we know we
292 * made it partly through deleting this snapshot, and
293 * thus we need to make sure we block any balance from
294 * happening until this snapshot is completely dropped.
296 if (drop_key.objectid != 0 || drop_key.type != 0 ||
297 drop_key.offset != 0) {
298 set_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
299 set_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
302 set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
303 btrfs_add_dead_root(root);
305 btrfs_put_root(root);
308 btrfs_free_path(path);
312 /* drop the root item for 'key' from the tree root */
313 int btrfs_del_root(struct btrfs_trans_handle *trans,
314 const struct btrfs_key *key)
316 struct btrfs_root *root = trans->fs_info->tree_root;
317 struct btrfs_path *path;
320 path = btrfs_alloc_path();
323 ret = btrfs_search_slot(trans, root, key, path, -1, 1);
329 ret = btrfs_del_item(trans, root, path);
331 btrfs_free_path(path);
335 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
336 u64 ref_id, u64 dirid, u64 *sequence,
337 const struct fscrypt_str *name)
339 struct btrfs_root *tree_root = trans->fs_info->tree_root;
340 struct btrfs_path *path;
341 struct btrfs_root_ref *ref;
342 struct extent_buffer *leaf;
343 struct btrfs_key key;
347 path = btrfs_alloc_path();
351 key.objectid = root_id;
352 key.type = BTRFS_ROOT_BACKREF_KEY;
355 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
358 } else if (ret == 0) {
359 leaf = path->nodes[0];
360 ref = btrfs_item_ptr(leaf, path->slots[0],
361 struct btrfs_root_ref);
362 ptr = (unsigned long)(ref + 1);
363 if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
364 (btrfs_root_ref_name_len(leaf, ref) != name->len) ||
365 memcmp_extent_buffer(leaf, name->name, ptr, name->len)) {
369 *sequence = btrfs_root_ref_sequence(leaf, ref);
371 ret = btrfs_del_item(trans, tree_root, path);
379 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
380 btrfs_release_path(path);
381 key.objectid = ref_id;
382 key.type = BTRFS_ROOT_REF_KEY;
383 key.offset = root_id;
388 btrfs_free_path(path);
393 * add a btrfs_root_ref item. type is either BTRFS_ROOT_REF_KEY
394 * or BTRFS_ROOT_BACKREF_KEY.
396 * The dirid, sequence, name and name_len refer to the directory entry
397 * that is referencing the root.
399 * For a forward ref, the root_id is the id of the tree referencing
400 * the root and ref_id is the id of the subvol or snapshot.
402 * For a back ref the root_id is the id of the subvol or snapshot and
403 * ref_id is the id of the tree referencing it.
405 * Will return 0, -ENOMEM, or anything from the CoW path
407 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
408 u64 ref_id, u64 dirid, u64 sequence,
409 const struct fscrypt_str *name)
411 struct btrfs_root *tree_root = trans->fs_info->tree_root;
412 struct btrfs_key key;
414 struct btrfs_path *path;
415 struct btrfs_root_ref *ref;
416 struct extent_buffer *leaf;
419 path = btrfs_alloc_path();
423 key.objectid = root_id;
424 key.type = BTRFS_ROOT_BACKREF_KEY;
427 ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
428 sizeof(*ref) + name->len);
430 btrfs_abort_transaction(trans, ret);
431 btrfs_free_path(path);
435 leaf = path->nodes[0];
436 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
437 btrfs_set_root_ref_dirid(leaf, ref, dirid);
438 btrfs_set_root_ref_sequence(leaf, ref, sequence);
439 btrfs_set_root_ref_name_len(leaf, ref, name->len);
440 ptr = (unsigned long)(ref + 1);
441 write_extent_buffer(leaf, name->name, ptr, name->len);
442 btrfs_mark_buffer_dirty(trans, leaf);
444 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
445 btrfs_release_path(path);
446 key.objectid = ref_id;
447 key.type = BTRFS_ROOT_REF_KEY;
448 key.offset = root_id;
452 btrfs_free_path(path);
457 * Old btrfs forgets to init root_item->flags and root_item->byte_limit
458 * for subvolumes. To work around this problem, we steal a bit from
459 * root_item->inode_item->flags, and use it to indicate if those fields
460 * have been properly initialized.
462 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
464 u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
466 if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
467 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
468 btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
469 btrfs_set_root_flags(root_item, 0);
470 btrfs_set_root_limit(root_item, 0);
474 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
475 struct btrfs_root *root)
477 struct btrfs_root_item *item = &root->root_item;
478 struct timespec64 ct;
480 ktime_get_real_ts64(&ct);
481 spin_lock(&root->root_item_lock);
482 btrfs_set_root_ctransid(item, trans->transid);
483 btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
484 btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
485 spin_unlock(&root->root_item_lock);
489 * Reserve space for subvolume operation.
491 * root: the root of the parent directory
492 * rsv: block reservation
493 * items: the number of items that we need do reservation
494 * use_global_rsv: allow fallback to the global block reservation
496 * This function is used to reserve the space for snapshot/subvolume
497 * creation and deletion. Those operations are different with the
498 * common file/directory operations, they change two fs/file trees
499 * and root tree, the number of items that the qgroup reserves is
500 * different with the free space reservation. So we can not use
501 * the space reservation mechanism in start_transaction().
503 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
504 struct btrfs_block_rsv *rsv, int items,
507 u64 qgroup_num_bytes = 0;
510 struct btrfs_fs_info *fs_info = root->fs_info;
511 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
513 if (btrfs_qgroup_enabled(fs_info)) {
514 /* One for parent inode, two for dir entries */
515 qgroup_num_bytes = 3 * fs_info->nodesize;
516 ret = btrfs_qgroup_reserve_meta_prealloc(root,
517 qgroup_num_bytes, true,
523 num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
524 rsv->space_info = btrfs_find_space_info(fs_info,
525 BTRFS_BLOCK_GROUP_METADATA);
526 ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes,
527 BTRFS_RESERVE_FLUSH_ALL);
529 if (ret == -ENOSPC && use_global_rsv)
530 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
532 if (ret && qgroup_num_bytes)
533 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
536 spin_lock(&rsv->lock);
537 rsv->qgroup_rsv_reserved += qgroup_num_bytes;
538 spin_unlock(&rsv->lock);
543 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
544 struct btrfs_block_rsv *rsv)
546 struct btrfs_fs_info *fs_info = root->fs_info;
547 u64 qgroup_to_release;
549 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, &qgroup_to_release);
550 btrfs_qgroup_convert_reserved_meta(root, qgroup_to_release);