1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/err.h>
7#include <linux/uuid.h>
8#include "ctree.h"
9#include "fs.h"
10#include "messages.h"
11#include "transaction.h"
12#include "disk-io.h"
13#include "qgroup.h"
14#include "space-info.h"
15#include "accessors.h"
16#include "root-tree.h"
17#include "orphan.h"
18
19/*
20 * Read a root item from the tree. In case we detect a root item smaller then
21 * sizeof(root_item), we know it's an old version of the root structure and
22 * initialize all new fields to zero. The same happens if we detect mismatching
23 * generation numbers as then we know the root was once mounted with an older
24 * kernel that was not aware of the root item structure change.
25 */
26static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
27 struct btrfs_root_item *item)
28{
29 u32 len;
30 int need_reset = 0;
31
32 len = btrfs_item_size(eb, slot);
33 read_extent_buffer(eb, dst: item, btrfs_item_ptr_offset(eb, slot),
34 min_t(u32, len, sizeof(*item)));
35 if (len < sizeof(*item))
36 need_reset = 1;
37 if (!need_reset && btrfs_root_generation(s: item)
38 != btrfs_root_generation_v2(s: item)) {
39 if (btrfs_root_generation_v2(s: item) != 0) {
40 btrfs_warn(eb->fs_info,
41 "mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
42 }
43 need_reset = 1;
44 }
45 if (need_reset) {
46 /* Clear all members from generation_v2 onwards. */
47 memset_startat(item, 0, generation_v2);
48 generate_random_guid(guid: item->uuid);
49 }
50}
51
52/*
53 * Lookup the root by the key.
54 *
55 * root: the root of the root tree
56 * search_key: the key to search
57 * path: the path we search
58 * root_item: the root item of the tree we look for
59 * root_key: the root key of the tree we look for
60 *
61 * If ->offset of 'search_key' is -1ULL, it means we are not sure the offset
62 * of the search key, just lookup the root with the highest offset for a
63 * given objectid.
64 *
65 * If we find something return 0, otherwise > 0, < 0 on error.
66 */
67int btrfs_find_root(struct btrfs_root *root, const struct btrfs_key *search_key,
68 struct btrfs_path *path, struct btrfs_root_item *root_item,
69 struct btrfs_key *root_key)
70{
71 struct btrfs_key found_key;
72 struct extent_buffer *l;
73 int ret;
74 int slot;
75
76 ret = btrfs_search_slot(NULL, root, key: search_key, p: path, ins_len: 0, cow: 0);
77 if (ret < 0)
78 return ret;
79
80 if (search_key->offset != -1ULL) { /* the search key is exact */
81 if (ret > 0)
82 goto out;
83 } else {
84 /*
85 * Key with offset -1 found, there would have to exist a root
86 * with such id, but this is out of the valid range.
87 */
88 if (ret == 0) {
89 ret = -EUCLEAN;
90 goto out;
91 }
92 if (path->slots[0] == 0)
93 goto out;
94 path->slots[0]--;
95 ret = 0;
96 }
97
98 l = path->nodes[0];
99 slot = path->slots[0];
100
101 btrfs_item_key_to_cpu(eb: l, cpu_key: &found_key, nr: slot);
102 if (found_key.objectid != search_key->objectid ||
103 found_key.type != BTRFS_ROOT_ITEM_KEY) {
104 ret = 1;
105 goto out;
106 }
107
108 if (root_item)
109 btrfs_read_root_item(eb: l, slot, item: root_item);
110 if (root_key)
111 memcpy(root_key, &found_key, sizeof(found_key));
112out:
113 btrfs_release_path(p: path);
114 return ret;
115}
116
117void btrfs_set_root_node(struct btrfs_root_item *item,
118 struct extent_buffer *node)
119{
120 btrfs_set_root_bytenr(s: item, val: node->start);
121 btrfs_set_root_level(s: item, val: btrfs_header_level(eb: node));
122 btrfs_set_root_generation(s: item, val: btrfs_header_generation(eb: node));
123}
124
125/*
126 * copy the data in 'item' into the btree
127 */
128int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
129 *root, struct btrfs_key *key, struct btrfs_root_item
130 *item)
131{
132 struct btrfs_fs_info *fs_info = root->fs_info;
133 struct btrfs_path *path;
134 struct extent_buffer *l;
135 int ret;
136 int slot;
137 unsigned long ptr;
138 u32 old_len;
139
140 path = btrfs_alloc_path();
141 if (!path)
142 return -ENOMEM;
143
144 ret = btrfs_search_slot(trans, root, key, p: path, ins_len: 0, cow: 1);
145 if (ret < 0)
146 goto out;
147
148 if (ret > 0) {
149 btrfs_crit(fs_info,
150 "unable to find root key (%llu %u %llu) in tree %llu",
151 key->objectid, key->type, key->offset, btrfs_root_id(root));
152 ret = -EUCLEAN;
153 btrfs_abort_transaction(trans, ret);
154 goto out;
155 }
156
157 l = path->nodes[0];
158 slot = path->slots[0];
159 ptr = btrfs_item_ptr_offset(l, slot);
160 old_len = btrfs_item_size(eb: l, slot);
161
162 /*
163 * If this is the first time we update the root item which originated
164 * from an older kernel, we need to enlarge the item size to make room
165 * for the added fields.
166 */
167 if (old_len < sizeof(*item)) {
168 btrfs_release_path(p: path);
169 ret = btrfs_search_slot(trans, root, key, p: path,
170 ins_len: -1, cow: 1);
171 if (ret < 0) {
172 btrfs_abort_transaction(trans, ret);
173 goto out;
174 }
175
176 ret = btrfs_del_item(trans, root, path);
177 if (ret < 0) {
178 btrfs_abort_transaction(trans, ret);
179 goto out;
180 }
181 btrfs_release_path(p: path);
182 ret = btrfs_insert_empty_item(trans, root, path,
183 key, data_size: sizeof(*item));
184 if (ret < 0) {
185 btrfs_abort_transaction(trans, ret);
186 goto out;
187 }
188 l = path->nodes[0];
189 slot = path->slots[0];
190 ptr = btrfs_item_ptr_offset(l, slot);
191 }
192
193 /*
194 * Update generation_v2 so at the next mount we know the new root
195 * fields are valid.
196 */
197 btrfs_set_root_generation_v2(s: item, val: btrfs_root_generation(s: item));
198
199 write_extent_buffer(eb: l, src: item, start: ptr, len: sizeof(*item));
200out:
201 btrfs_free_path(p: path);
202 return ret;
203}
204
205int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
206 const struct btrfs_key *key, struct btrfs_root_item *item)
207{
208 /*
209 * Make sure generation v1 and v2 match. See update_root for details.
210 */
211 btrfs_set_root_generation_v2(s: item, val: btrfs_root_generation(s: item));
212 return btrfs_insert_item(trans, root, key, data: item, data_size: sizeof(*item));
213}
214
215int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
216{
217 struct btrfs_root *tree_root = fs_info->tree_root;
218 struct extent_buffer *leaf;
219 struct btrfs_path *path;
220 struct btrfs_key key;
221 struct btrfs_root *root;
222 int err = 0;
223 int ret;
224
225 path = btrfs_alloc_path();
226 if (!path)
227 return -ENOMEM;
228
229 key.objectid = BTRFS_ORPHAN_OBJECTID;
230 key.type = BTRFS_ORPHAN_ITEM_KEY;
231 key.offset = 0;
232
233 while (1) {
234 u64 root_objectid;
235
236 ret = btrfs_search_slot(NULL, root: tree_root, key: &key, p: path, ins_len: 0, cow: 0);
237 if (ret < 0) {
238 err = ret;
239 break;
240 }
241
242 leaf = path->nodes[0];
243 if (path->slots[0] >= btrfs_header_nritems(eb: leaf)) {
244 ret = btrfs_next_leaf(root: tree_root, path);
245 if (ret < 0)
246 err = ret;
247 if (ret != 0)
248 break;
249 leaf = path->nodes[0];
250 }
251
252 btrfs_item_key_to_cpu(eb: leaf, cpu_key: &key, nr: path->slots[0]);
253 btrfs_release_path(p: path);
254
255 if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
256 key.type != BTRFS_ORPHAN_ITEM_KEY)
257 break;
258
259 root_objectid = key.offset;
260 key.offset++;
261
262 root = btrfs_get_fs_root(fs_info, objectid: root_objectid, check_ref: false);
263 err = PTR_ERR_OR_ZERO(ptr: root);
264 if (err && err != -ENOENT) {
265 break;
266 } else if (err == -ENOENT) {
267 struct btrfs_trans_handle *trans;
268
269 btrfs_release_path(p: path);
270
271 trans = btrfs_join_transaction(root: tree_root);
272 if (IS_ERR(ptr: trans)) {
273 err = PTR_ERR(ptr: trans);
274 btrfs_handle_fs_error(fs_info, err,
275 "Failed to start trans to delete orphan item");
276 break;
277 }
278 err = btrfs_del_orphan_item(trans, root: tree_root,
279 offset: root_objectid);
280 btrfs_end_transaction(trans);
281 if (err) {
282 btrfs_handle_fs_error(fs_info, err,
283 "Failed to delete root orphan item");
284 break;
285 }
286 continue;
287 }
288
289 WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
290 if (btrfs_root_refs(s: &root->root_item) == 0) {
291 struct btrfs_key drop_key;
292
293 btrfs_disk_key_to_cpu(cpu_key: &drop_key, disk_key: &root->root_item.drop_progress);
294 /*
295 * If we have a non-zero drop_progress then we know we
296 * made it partly through deleting this snapshot, and
297 * thus we need to make sure we block any balance from
298 * happening until this snapshot is completely dropped.
299 */
300 if (drop_key.objectid != 0 || drop_key.type != 0 ||
301 drop_key.offset != 0) {
302 set_bit(nr: BTRFS_FS_UNFINISHED_DROPS, addr: &fs_info->flags);
303 set_bit(nr: BTRFS_ROOT_UNFINISHED_DROP, addr: &root->state);
304 }
305
306 set_bit(nr: BTRFS_ROOT_DEAD_TREE, addr: &root->state);
307 btrfs_add_dead_root(root);
308 }
309 btrfs_put_root(root);
310 }
311
312 btrfs_free_path(p: path);
313 return err;
314}
315
316/* drop the root item for 'key' from the tree root */
317int btrfs_del_root(struct btrfs_trans_handle *trans,
318 const struct btrfs_key *key)
319{
320 struct btrfs_root *root = trans->fs_info->tree_root;
321 struct btrfs_path *path;
322 int ret;
323
324 path = btrfs_alloc_path();
325 if (!path)
326 return -ENOMEM;
327 ret = btrfs_search_slot(trans, root, key, p: path, ins_len: -1, cow: 1);
328 if (ret < 0)
329 goto out;
330 if (ret != 0) {
331 /* The root must exist but we did not find it by the key. */
332 ret = -EUCLEAN;
333 goto out;
334 }
335
336 ret = btrfs_del_item(trans, root, path);
337out:
338 btrfs_free_path(p: path);
339 return ret;
340}
341
342int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
343 u64 ref_id, u64 dirid, u64 *sequence,
344 const struct fscrypt_str *name)
345{
346 struct btrfs_root *tree_root = trans->fs_info->tree_root;
347 struct btrfs_path *path;
348 struct btrfs_root_ref *ref;
349 struct extent_buffer *leaf;
350 struct btrfs_key key;
351 unsigned long ptr;
352 int ret;
353
354 path = btrfs_alloc_path();
355 if (!path)
356 return -ENOMEM;
357
358 key.objectid = root_id;
359 key.type = BTRFS_ROOT_BACKREF_KEY;
360 key.offset = ref_id;
361again:
362 ret = btrfs_search_slot(trans, root: tree_root, key: &key, p: path, ins_len: -1, cow: 1);
363 if (ret < 0) {
364 goto out;
365 } else if (ret == 0) {
366 leaf = path->nodes[0];
367 ref = btrfs_item_ptr(leaf, path->slots[0],
368 struct btrfs_root_ref);
369 ptr = (unsigned long)(ref + 1);
370 if ((btrfs_root_ref_dirid(eb: leaf, s: ref) != dirid) ||
371 (btrfs_root_ref_name_len(eb: leaf, s: ref) != name->len) ||
372 memcmp_extent_buffer(eb: leaf, ptrv: name->name, start: ptr, len: name->len)) {
373 ret = -ENOENT;
374 goto out;
375 }
376 *sequence = btrfs_root_ref_sequence(eb: leaf, s: ref);
377
378 ret = btrfs_del_item(trans, root: tree_root, path);
379 if (ret)
380 goto out;
381 } else {
382 ret = -ENOENT;
383 goto out;
384 }
385
386 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
387 btrfs_release_path(p: path);
388 key.objectid = ref_id;
389 key.type = BTRFS_ROOT_REF_KEY;
390 key.offset = root_id;
391 goto again;
392 }
393
394out:
395 btrfs_free_path(p: path);
396 return ret;
397}
398
399/*
400 * add a btrfs_root_ref item. type is either BTRFS_ROOT_REF_KEY
401 * or BTRFS_ROOT_BACKREF_KEY.
402 *
403 * The dirid, sequence, name and name_len refer to the directory entry
404 * that is referencing the root.
405 *
406 * For a forward ref, the root_id is the id of the tree referencing
407 * the root and ref_id is the id of the subvol or snapshot.
408 *
409 * For a back ref the root_id is the id of the subvol or snapshot and
410 * ref_id is the id of the tree referencing it.
411 *
412 * Will return 0, -ENOMEM, or anything from the CoW path
413 */
414int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
415 u64 ref_id, u64 dirid, u64 sequence,
416 const struct fscrypt_str *name)
417{
418 struct btrfs_root *tree_root = trans->fs_info->tree_root;
419 struct btrfs_key key;
420 int ret;
421 struct btrfs_path *path;
422 struct btrfs_root_ref *ref;
423 struct extent_buffer *leaf;
424 unsigned long ptr;
425
426 path = btrfs_alloc_path();
427 if (!path)
428 return -ENOMEM;
429
430 key.objectid = root_id;
431 key.type = BTRFS_ROOT_BACKREF_KEY;
432 key.offset = ref_id;
433again:
434 ret = btrfs_insert_empty_item(trans, root: tree_root, path, key: &key,
435 data_size: sizeof(*ref) + name->len);
436 if (ret) {
437 btrfs_abort_transaction(trans, ret);
438 btrfs_free_path(p: path);
439 return ret;
440 }
441
442 leaf = path->nodes[0];
443 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
444 btrfs_set_root_ref_dirid(eb: leaf, s: ref, val: dirid);
445 btrfs_set_root_ref_sequence(eb: leaf, s: ref, val: sequence);
446 btrfs_set_root_ref_name_len(eb: leaf, s: ref, val: name->len);
447 ptr = (unsigned long)(ref + 1);
448 write_extent_buffer(eb: leaf, src: name->name, start: ptr, len: name->len);
449
450 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
451 btrfs_release_path(p: path);
452 key.objectid = ref_id;
453 key.type = BTRFS_ROOT_REF_KEY;
454 key.offset = root_id;
455 goto again;
456 }
457
458 btrfs_free_path(p: path);
459 return 0;
460}
461
462/*
463 * Old btrfs forgets to init root_item->flags and root_item->byte_limit
464 * for subvolumes. To work around this problem, we steal a bit from
465 * root_item->inode_item->flags, and use it to indicate if those fields
466 * have been properly initialized.
467 */
468void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
469{
470 u64 inode_flags = btrfs_stack_inode_flags(s: &root_item->inode);
471
472 if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
473 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
474 btrfs_set_stack_inode_flags(s: &root_item->inode, val: inode_flags);
475 btrfs_set_root_flags(s: root_item, val: 0);
476 btrfs_set_root_limit(s: root_item, val: 0);
477 }
478}
479
480void btrfs_update_root_times(struct btrfs_trans_handle *trans,
481 struct btrfs_root *root)
482{
483 struct btrfs_root_item *item = &root->root_item;
484 struct timespec64 ct;
485
486 ktime_get_real_ts64(tv: &ct);
487 spin_lock(lock: &root->root_item_lock);
488 btrfs_set_root_ctransid(s: item, val: trans->transid);
489 btrfs_set_stack_timespec_sec(s: &item->ctime, val: ct.tv_sec);
490 btrfs_set_stack_timespec_nsec(s: &item->ctime, val: ct.tv_nsec);
491 spin_unlock(lock: &root->root_item_lock);
492}
493
494/*
495 * Reserve space for subvolume operation.
496 *
497 * root: the root of the parent directory
498 * rsv: block reservation
499 * items: the number of items that we need do reservation
500 * use_global_rsv: allow fallback to the global block reservation
501 *
502 * This function is used to reserve the space for snapshot/subvolume
503 * creation and deletion. Those operations are different with the
504 * common file/directory operations, they change two fs/file trees
505 * and root tree, the number of items that the qgroup reserves is
506 * different with the free space reservation. So we can not use
507 * the space reservation mechanism in start_transaction().
508 */
509int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
510 struct btrfs_block_rsv *rsv, int items,
511 bool use_global_rsv)
512{
513 u64 qgroup_num_bytes = 0;
514 u64 num_bytes;
515 int ret;
516 struct btrfs_fs_info *fs_info = root->fs_info;
517 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
518
519 if (btrfs_qgroup_enabled(fs_info)) {
520 /* One for parent inode, two for dir entries */
521 qgroup_num_bytes = 3 * fs_info->nodesize;
522 ret = btrfs_qgroup_reserve_meta_prealloc(root,
523 num_bytes: qgroup_num_bytes, enforce: true,
524 noflush: false);
525 if (ret)
526 return ret;
527 }
528
529 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items: items);
530 rsv->space_info = btrfs_find_space_info(info: fs_info,
531 BTRFS_BLOCK_GROUP_METADATA);
532 ret = btrfs_block_rsv_add(fs_info, block_rsv: rsv, num_bytes,
533 flush: BTRFS_RESERVE_FLUSH_ALL);
534
535 if (ret == -ENOSPC && use_global_rsv)
536 ret = btrfs_block_rsv_migrate(src_rsv: global_rsv, dst_rsv: rsv, num_bytes, update_size: true);
537
538 if (ret && qgroup_num_bytes)
539 btrfs_qgroup_free_meta_prealloc(root, num_bytes: qgroup_num_bytes);
540
541 if (!ret) {
542 spin_lock(lock: &rsv->lock);
543 rsv->qgroup_rsv_reserved += qgroup_num_bytes;
544 spin_unlock(lock: &rsv->lock);
545 }
546 return ret;
547}
548

source code of linux/fs/btrfs/root-tree.c