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 "print-tree.h"
14 #include "qgroup.h"
15 #include "space-info.h"
16 #include "accessors.h"
17 #include "root-tree.h"
18 #include "orphan.h"
19
20 /*
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.
26 */
btrfs_read_root_item(struct extent_buffer * eb,int slot,struct btrfs_root_item * item)27 static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
28 struct btrfs_root_item *item)
29 {
30 u32 len;
31 int need_reset = 0;
32
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))
37 need_reset = 1;
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.");
43 }
44 need_reset = 1;
45 }
46 if (need_reset) {
47 /* Clear all members from generation_v2 onwards. */
48 memset_startat(item, 0, generation_v2);
49 generate_random_guid(item->uuid);
50 }
51 }
52
53 /*
54 * btrfs_find_root - lookup the root by the key.
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 */
btrfs_find_root(struct btrfs_root * root,const struct btrfs_key * search_key,struct btrfs_path * path,struct btrfs_root_item * root_item,struct btrfs_key * root_key)67 int 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, search_key, path, 0, 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 BUG_ON(ret == 0); /* Logical error */
85 if (path->slots[0] == 0)
86 goto out;
87 path->slots[0]--;
88 ret = 0;
89 }
90
91 l = path->nodes[0];
92 slot = path->slots[0];
93
94 btrfs_item_key_to_cpu(l, &found_key, slot);
95 if (found_key.objectid != search_key->objectid ||
96 found_key.type != BTRFS_ROOT_ITEM_KEY) {
97 ret = 1;
98 goto out;
99 }
100
101 if (root_item)
102 btrfs_read_root_item(l, slot, root_item);
103 if (root_key)
104 memcpy(root_key, &found_key, sizeof(found_key));
105 out:
106 btrfs_release_path(path);
107 return ret;
108 }
109
btrfs_set_root_node(struct btrfs_root_item * item,struct extent_buffer * node)110 void btrfs_set_root_node(struct btrfs_root_item *item,
111 struct extent_buffer *node)
112 {
113 btrfs_set_root_bytenr(item, node->start);
114 btrfs_set_root_level(item, btrfs_header_level(node));
115 btrfs_set_root_generation(item, btrfs_header_generation(node));
116 }
117
118 /*
119 * copy the data in 'item' into the btree
120 */
btrfs_update_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_key * key,struct btrfs_root_item * item)121 int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
122 *root, struct btrfs_key *key, struct btrfs_root_item
123 *item)
124 {
125 struct btrfs_fs_info *fs_info = root->fs_info;
126 struct btrfs_path *path;
127 struct extent_buffer *l;
128 int ret;
129 int slot;
130 unsigned long ptr;
131 u32 old_len;
132
133 path = btrfs_alloc_path();
134 if (!path)
135 return -ENOMEM;
136
137 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
138 if (ret < 0)
139 goto out;
140
141 if (ret > 0) {
142 btrfs_crit(fs_info,
143 "unable to find root key (%llu %u %llu) in tree %llu",
144 key->objectid, key->type, key->offset,
145 root->root_key.objectid);
146 ret = -EUCLEAN;
147 btrfs_abort_transaction(trans, ret);
148 goto out;
149 }
150
151 l = path->nodes[0];
152 slot = path->slots[0];
153 ptr = btrfs_item_ptr_offset(l, slot);
154 old_len = btrfs_item_size(l, slot);
155
156 /*
157 * If this is the first time we update the root item which originated
158 * from an older kernel, we need to enlarge the item size to make room
159 * for the added fields.
160 */
161 if (old_len < sizeof(*item)) {
162 btrfs_release_path(path);
163 ret = btrfs_search_slot(trans, root, key, path,
164 -1, 1);
165 if (ret < 0) {
166 btrfs_abort_transaction(trans, ret);
167 goto out;
168 }
169
170 ret = btrfs_del_item(trans, root, path);
171 if (ret < 0) {
172 btrfs_abort_transaction(trans, ret);
173 goto out;
174 }
175 btrfs_release_path(path);
176 ret = btrfs_insert_empty_item(trans, root, path,
177 key, sizeof(*item));
178 if (ret < 0) {
179 btrfs_abort_transaction(trans, ret);
180 goto out;
181 }
182 l = path->nodes[0];
183 slot = path->slots[0];
184 ptr = btrfs_item_ptr_offset(l, slot);
185 }
186
187 /*
188 * Update generation_v2 so at the next mount we know the new root
189 * fields are valid.
190 */
191 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
192
193 write_extent_buffer(l, item, ptr, sizeof(*item));
194 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
195 out:
196 btrfs_free_path(path);
197 return ret;
198 }
199
btrfs_insert_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct btrfs_key * key,struct btrfs_root_item * item)200 int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
201 const struct btrfs_key *key, struct btrfs_root_item *item)
202 {
203 /*
204 * Make sure generation v1 and v2 match. See update_root for details.
205 */
206 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
207 return btrfs_insert_item(trans, root, key, item, sizeof(*item));
208 }
209
btrfs_find_orphan_roots(struct btrfs_fs_info * fs_info)210 int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
211 {
212 struct btrfs_root *tree_root = fs_info->tree_root;
213 struct extent_buffer *leaf;
214 struct btrfs_path *path;
215 struct btrfs_key key;
216 struct btrfs_root *root;
217 int err = 0;
218 int ret;
219
220 path = btrfs_alloc_path();
221 if (!path)
222 return -ENOMEM;
223
224 key.objectid = BTRFS_ORPHAN_OBJECTID;
225 key.type = BTRFS_ORPHAN_ITEM_KEY;
226 key.offset = 0;
227
228 while (1) {
229 u64 root_objectid;
230
231 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
232 if (ret < 0) {
233 err = ret;
234 break;
235 }
236
237 leaf = path->nodes[0];
238 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
239 ret = btrfs_next_leaf(tree_root, path);
240 if (ret < 0)
241 err = ret;
242 if (ret != 0)
243 break;
244 leaf = path->nodes[0];
245 }
246
247 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
248 btrfs_release_path(path);
249
250 if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
251 key.type != BTRFS_ORPHAN_ITEM_KEY)
252 break;
253
254 root_objectid = key.offset;
255 key.offset++;
256
257 root = btrfs_get_fs_root(fs_info, root_objectid, false);
258 err = PTR_ERR_OR_ZERO(root);
259 if (err && err != -ENOENT) {
260 break;
261 } else if (err == -ENOENT) {
262 struct btrfs_trans_handle *trans;
263
264 btrfs_release_path(path);
265
266 trans = btrfs_join_transaction(tree_root);
267 if (IS_ERR(trans)) {
268 err = PTR_ERR(trans);
269 btrfs_handle_fs_error(fs_info, err,
270 "Failed to start trans to delete orphan item");
271 break;
272 }
273 err = btrfs_del_orphan_item(trans, tree_root,
274 root_objectid);
275 btrfs_end_transaction(trans);
276 if (err) {
277 btrfs_handle_fs_error(fs_info, err,
278 "Failed to delete root orphan item");
279 break;
280 }
281 continue;
282 }
283
284 WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
285 if (btrfs_root_refs(&root->root_item) == 0) {
286 struct btrfs_key drop_key;
287
288 btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
289 /*
290 * If we have a non-zero drop_progress then we know we
291 * made it partly through deleting this snapshot, and
292 * thus we need to make sure we block any balance from
293 * happening until this snapshot is completely dropped.
294 */
295 if (drop_key.objectid != 0 || drop_key.type != 0 ||
296 drop_key.offset != 0) {
297 set_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
298 set_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
299 }
300
301 set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
302 btrfs_add_dead_root(root);
303 }
304 btrfs_put_root(root);
305 }
306
307 btrfs_free_path(path);
308 return err;
309 }
310
311 /* drop the root item for 'key' from the tree root */
btrfs_del_root(struct btrfs_trans_handle * trans,const struct btrfs_key * key)312 int btrfs_del_root(struct btrfs_trans_handle *trans,
313 const struct btrfs_key *key)
314 {
315 struct btrfs_root *root = trans->fs_info->tree_root;
316 struct btrfs_path *path;
317 int ret;
318
319 path = btrfs_alloc_path();
320 if (!path)
321 return -ENOMEM;
322 ret = btrfs_search_slot(trans, root, key, path, -1, 1);
323 if (ret < 0)
324 goto out;
325
326 BUG_ON(ret != 0);
327
328 ret = btrfs_del_item(trans, root, path);
329 out:
330 btrfs_free_path(path);
331 return ret;
332 }
333
btrfs_del_root_ref(struct btrfs_trans_handle * trans,u64 root_id,u64 ref_id,u64 dirid,u64 * sequence,const struct fscrypt_str * name)334 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
335 u64 ref_id, u64 dirid, u64 *sequence,
336 const struct fscrypt_str *name)
337 {
338 struct btrfs_root *tree_root = trans->fs_info->tree_root;
339 struct btrfs_path *path;
340 struct btrfs_root_ref *ref;
341 struct extent_buffer *leaf;
342 struct btrfs_key key;
343 unsigned long ptr;
344 int ret;
345
346 path = btrfs_alloc_path();
347 if (!path)
348 return -ENOMEM;
349
350 key.objectid = root_id;
351 key.type = BTRFS_ROOT_BACKREF_KEY;
352 key.offset = ref_id;
353 again:
354 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
355 if (ret < 0) {
356 goto out;
357 } else if (ret == 0) {
358 leaf = path->nodes[0];
359 ref = btrfs_item_ptr(leaf, path->slots[0],
360 struct btrfs_root_ref);
361 ptr = (unsigned long)(ref + 1);
362 if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
363 (btrfs_root_ref_name_len(leaf, ref) != name->len) ||
364 memcmp_extent_buffer(leaf, name->name, ptr, name->len)) {
365 ret = -ENOENT;
366 goto out;
367 }
368 *sequence = btrfs_root_ref_sequence(leaf, ref);
369
370 ret = btrfs_del_item(trans, tree_root, path);
371 if (ret)
372 goto out;
373 } else {
374 ret = -ENOENT;
375 goto out;
376 }
377
378 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
379 btrfs_release_path(path);
380 key.objectid = ref_id;
381 key.type = BTRFS_ROOT_REF_KEY;
382 key.offset = root_id;
383 goto again;
384 }
385
386 out:
387 btrfs_free_path(path);
388 return ret;
389 }
390
391 /*
392 * add a btrfs_root_ref item. type is either BTRFS_ROOT_REF_KEY
393 * or BTRFS_ROOT_BACKREF_KEY.
394 *
395 * The dirid, sequence, name and name_len refer to the directory entry
396 * that is referencing the root.
397 *
398 * For a forward ref, the root_id is the id of the tree referencing
399 * the root and ref_id is the id of the subvol or snapshot.
400 *
401 * For a back ref the root_id is the id of the subvol or snapshot and
402 * ref_id is the id of the tree referencing it.
403 *
404 * Will return 0, -ENOMEM, or anything from the CoW path
405 */
btrfs_add_root_ref(struct btrfs_trans_handle * trans,u64 root_id,u64 ref_id,u64 dirid,u64 sequence,const struct fscrypt_str * name)406 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
407 u64 ref_id, u64 dirid, u64 sequence,
408 const struct fscrypt_str *name)
409 {
410 struct btrfs_root *tree_root = trans->fs_info->tree_root;
411 struct btrfs_key key;
412 int ret;
413 struct btrfs_path *path;
414 struct btrfs_root_ref *ref;
415 struct extent_buffer *leaf;
416 unsigned long ptr;
417
418 path = btrfs_alloc_path();
419 if (!path)
420 return -ENOMEM;
421
422 key.objectid = root_id;
423 key.type = BTRFS_ROOT_BACKREF_KEY;
424 key.offset = ref_id;
425 again:
426 ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
427 sizeof(*ref) + name->len);
428 if (ret) {
429 btrfs_abort_transaction(trans, ret);
430 btrfs_free_path(path);
431 return ret;
432 }
433
434 leaf = path->nodes[0];
435 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
436 btrfs_set_root_ref_dirid(leaf, ref, dirid);
437 btrfs_set_root_ref_sequence(leaf, ref, sequence);
438 btrfs_set_root_ref_name_len(leaf, ref, name->len);
439 ptr = (unsigned long)(ref + 1);
440 write_extent_buffer(leaf, name->name, ptr, name->len);
441 btrfs_mark_buffer_dirty(trans, leaf);
442
443 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
444 btrfs_release_path(path);
445 key.objectid = ref_id;
446 key.type = BTRFS_ROOT_REF_KEY;
447 key.offset = root_id;
448 goto again;
449 }
450
451 btrfs_free_path(path);
452 return 0;
453 }
454
455 /*
456 * Old btrfs forgets to init root_item->flags and root_item->byte_limit
457 * for subvolumes. To work around this problem, we steal a bit from
458 * root_item->inode_item->flags, and use it to indicate if those fields
459 * have been properly initialized.
460 */
btrfs_check_and_init_root_item(struct btrfs_root_item * root_item)461 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
462 {
463 u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
464
465 if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
466 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
467 btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
468 btrfs_set_root_flags(root_item, 0);
469 btrfs_set_root_limit(root_item, 0);
470 }
471 }
472
btrfs_update_root_times(struct btrfs_trans_handle * trans,struct btrfs_root * root)473 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
474 struct btrfs_root *root)
475 {
476 struct btrfs_root_item *item = &root->root_item;
477 struct timespec64 ct;
478
479 ktime_get_real_ts64(&ct);
480 spin_lock(&root->root_item_lock);
481 btrfs_set_root_ctransid(item, trans->transid);
482 btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
483 btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
484 spin_unlock(&root->root_item_lock);
485 }
486
487 /*
488 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
489 * root: the root of the parent directory
490 * rsv: block reservation
491 * items: the number of items that we need do reservation
492 * use_global_rsv: allow fallback to the global block reservation
493 *
494 * This function is used to reserve the space for snapshot/subvolume
495 * creation and deletion. Those operations are different with the
496 * common file/directory operations, they change two fs/file trees
497 * and root tree, the number of items that the qgroup reserves is
498 * different with the free space reservation. So we can not use
499 * the space reservation mechanism in start_transaction().
500 */
btrfs_subvolume_reserve_metadata(struct btrfs_root * root,struct btrfs_block_rsv * rsv,int items,bool use_global_rsv)501 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
502 struct btrfs_block_rsv *rsv, int items,
503 bool use_global_rsv)
504 {
505 u64 qgroup_num_bytes = 0;
506 u64 num_bytes;
507 int ret;
508 struct btrfs_fs_info *fs_info = root->fs_info;
509 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
510
511 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
512 /* One for parent inode, two for dir entries */
513 qgroup_num_bytes = 3 * fs_info->nodesize;
514 ret = btrfs_qgroup_reserve_meta_prealloc(root,
515 qgroup_num_bytes, true,
516 false);
517 if (ret)
518 return ret;
519 }
520
521 num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
522 rsv->space_info = btrfs_find_space_info(fs_info,
523 BTRFS_BLOCK_GROUP_METADATA);
524 ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes,
525 BTRFS_RESERVE_FLUSH_ALL);
526
527 if (ret == -ENOSPC && use_global_rsv)
528 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
529
530 if (ret && qgroup_num_bytes)
531 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
532
533 if (!ret) {
534 spin_lock(&rsv->lock);
535 rsv->qgroup_rsv_reserved += qgroup_num_bytes;
536 spin_unlock(&rsv->lock);
537 }
538 return ret;
539 }
540
btrfs_subvolume_release_metadata(struct btrfs_root * root,struct btrfs_block_rsv * rsv)541 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
542 struct btrfs_block_rsv *rsv)
543 {
544 struct btrfs_fs_info *fs_info = root->fs_info;
545 u64 qgroup_to_release;
546
547 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, &qgroup_to_release);
548 btrfs_qgroup_convert_reserved_meta(root, qgroup_to_release);
549 }
550