projects/linux·blob master · refs

527 lines · 14.1 KiB

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 bool need_reset = false;
31
32 len = btrfs_item_size(eb, slot);
33 read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
34 min_t(u32, len, sizeof(*item)));
35 if (len < sizeof(*item))
36 need_reset = true;
37 if (!need_reset && btrfs_root_generation(item)
38 != btrfs_root_generation_v2(item)) {
39 if (btrfs_root_generation_v2(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 = true;
44 }
45 if (need_reset) {
46 /* Clear all members from generation_v2 onwards. */
47 memset_startat(item, 0, generation_v2);
48 generate_random_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, 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 /*
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 (unlikely(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(l, &found_key, 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(l, slot, root_item);
110 if (root_key)
111 memcpy(root_key, &found_key, sizeof(found_key));
112out:
113 btrfs_release_path(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(item, node->start);
121 btrfs_set_root_level(item, btrfs_header_level(node));
122 btrfs_set_root_generation(item, btrfs_header_generation(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 BTRFS_PATH_AUTO_FREE(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, path, 0, 1);
145 if (ret < 0)
146 return ret;
147
148 if (unlikely(ret > 0)) {
149 btrfs_crit(fs_info,
150 "unable to find root key " BTRFS_KEY_FMT " in tree %llu",
151 BTRFS_KEY_FMT_VALUE(key), btrfs_root_id(root));
152 ret = -EUCLEAN;
153 btrfs_abort_transaction(trans, ret);
154 return ret;
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(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(path);
169 ret = btrfs_search_slot(trans, root, key, path,
170 -1, 1);
171 if (unlikely(ret < 0)) {
172 btrfs_abort_transaction(trans, ret);
173 return ret;
174 }
175
176 ret = btrfs_del_item(trans, root, path);
177 if (unlikely(ret < 0)) {
178 btrfs_abort_transaction(trans, ret);
179 return ret;
180 }
181 btrfs_release_path(path);
182 ret = btrfs_insert_empty_item(trans, root, path,
183 key, sizeof(*item));
184 if (unlikely(ret < 0)) {
185 btrfs_abort_transaction(trans, ret);
186 return ret;
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(item, btrfs_root_generation(item));
198
199 write_extent_buffer(l, item, ptr, sizeof(*item));
200 return ret;
201}
202
203int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
204 const struct btrfs_key *key, struct btrfs_root_item *item)
205{
206 /*
207 * Make sure generation v1 and v2 match. See update_root for details.
208 */
209 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
210 return btrfs_insert_item(trans, root, key, item, sizeof(*item));
211}
212
213int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
214{
215 struct btrfs_root *tree_root = fs_info->tree_root;
216 struct extent_buffer *leaf;
217 BTRFS_PATH_AUTO_FREE(path);
218 struct btrfs_key key;
219 struct btrfs_root *root;
220
221 path = btrfs_alloc_path();
222 if (!path)
223 return -ENOMEM;
224
225 key.objectid = BTRFS_ORPHAN_OBJECTID;
226 key.type = BTRFS_ORPHAN_ITEM_KEY;
227 key.offset = 0;
228
229 while (1) {
230 u64 root_objectid;
231 int ret;
232
233 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
234 if (ret < 0)
235 return ret;
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 return ret;
242 else if (ret > 0)
243 return 0;
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 return 0;
253
254 root_objectid = key.offset;
255 key.offset++;
256
257 root = btrfs_get_fs_root(fs_info, root_objectid, false);
258 ret = PTR_ERR_OR_ZERO(root);
259 if (ret && ret != -ENOENT) {
260 return ret;
261 } else if (ret == -ENOENT) {
262 struct btrfs_trans_handle *trans;
263
264 trans = btrfs_join_transaction(tree_root);
265 if (IS_ERR(trans)) {
266 ret = PTR_ERR(trans);
267 btrfs_err(fs_info,
268 "failed to join transaction to delete orphan item: %pe",
269 ERR_PTR(ret));
270 return ret;
271 }
272 ret = btrfs_del_orphan_item(trans, tree_root, root_objectid);
273 btrfs_end_transaction(trans);
274 if (ret) {
275 btrfs_err(fs_info,
276 "failed to delete root orphan item: %pe", ERR_PTR(ret));
277 return ret;
278 }
279 continue;
280 }
281
282 WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
283 if (btrfs_root_refs(&root->root_item) == 0) {
284 struct btrfs_key drop_key;
285
286 btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
287 /*
288 * If we have a non-zero drop_progress then we know we
289 * made it partly through deleting this snapshot, and
290 * thus we need to make sure we block any balance from
291 * happening until this snapshot is completely dropped.
292 */
293 if (drop_key.objectid != 0 || drop_key.type != 0 ||
294 drop_key.offset != 0) {
295 set_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
296 set_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
297 }
298
299 set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
300 btrfs_add_dead_root(root);
301 }
302 btrfs_put_root(root);
303 }
304
305 return 0;
306}
307
308/* drop the root item for 'key' from the tree root */
309int btrfs_del_root(struct btrfs_trans_handle *trans,
310 const struct btrfs_key *key)
311{
312 struct btrfs_root *root = trans->fs_info->tree_root;
313 BTRFS_PATH_AUTO_FREE(path);
314 int ret;
315
316 path = btrfs_alloc_path();
317 if (!path)
318 return -ENOMEM;
319 ret = btrfs_search_slot(trans, root, key, path, -1, 1);
320 if (ret < 0)
321 return ret;
322 if (unlikely(ret > 0))
323 /* The root must exist but we did not find it by the key. */
324 return -EUCLEAN;
325
326 return btrfs_del_item(trans, root, path);
327}
328
329int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
330 u64 ref_id, u64 dirid, u64 *sequence,
331 const struct fscrypt_str *name)
332{
333 struct btrfs_root *tree_root = trans->fs_info->tree_root;
334 BTRFS_PATH_AUTO_FREE(path);
335 struct btrfs_root_ref *ref;
336 struct extent_buffer *leaf;
337 struct btrfs_key key;
338 unsigned long ptr;
339 int ret;
340
341 path = btrfs_alloc_path();
342 if (!path)
343 return -ENOMEM;
344
345 key.objectid = root_id;
346 key.type = BTRFS_ROOT_BACKREF_KEY;
347 key.offset = ref_id;
348again:
349 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
350 if (ret < 0) {
351 return ret;
352 } else if (ret == 0) {
353 leaf = path->nodes[0];
354 ref = btrfs_item_ptr(leaf, path->slots[0],
355 struct btrfs_root_ref);
356 ptr = (unsigned long)(ref + 1);
357 if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
358 (btrfs_root_ref_name_len(leaf, ref) != name->len) ||
359 memcmp_extent_buffer(leaf, name->name, ptr, name->len))
360 return -ENOENT;
361
362 *sequence = btrfs_root_ref_sequence(leaf, ref);
363
364 ret = btrfs_del_item(trans, tree_root, path);
365 if (ret)
366 return ret;
367 } else {
368 return -ENOENT;
369 }
370
371 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
372 btrfs_release_path(path);
373 key.objectid = ref_id;
374 key.type = BTRFS_ROOT_REF_KEY;
375 key.offset = root_id;
376 goto again;
377 }
378
379 return ret;
380}
381
382/*
383 * add a btrfs_root_ref item. type is either BTRFS_ROOT_REF_KEY
384 * or BTRFS_ROOT_BACKREF_KEY.
385 *
386 * The dirid, sequence, name and name_len refer to the directory entry
387 * that is referencing the root.
388 *
389 * For a forward ref, the root_id is the id of the tree referencing
390 * the root and ref_id is the id of the subvol or snapshot.
391 *
392 * For a back ref the root_id is the id of the subvol or snapshot and
393 * ref_id is the id of the tree referencing it.
394 *
395 * Will return 0, -ENOMEM, or anything from the CoW path
396 */
397int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
398 u64 ref_id, u64 dirid, u64 sequence,
399 const struct fscrypt_str *name)
400{
401 struct btrfs_root *tree_root = trans->fs_info->tree_root;
402 struct btrfs_key key;
403 int ret;
404 BTRFS_PATH_AUTO_FREE(path);
405 struct btrfs_root_ref *ref;
406 struct extent_buffer *leaf;
407 unsigned long ptr;
408
409 path = btrfs_alloc_path();
410 if (!path)
411 return -ENOMEM;
412
413 key.objectid = root_id;
414 key.type = BTRFS_ROOT_BACKREF_KEY;
415 key.offset = ref_id;
416again:
417 ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
418 sizeof(*ref) + name->len);
419 if (unlikely(ret)) {
420 btrfs_abort_transaction(trans, ret);
421 return ret;
422 }
423
424 leaf = path->nodes[0];
425 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
426 btrfs_set_root_ref_dirid(leaf, ref, dirid);
427 btrfs_set_root_ref_sequence(leaf, ref, sequence);
428 btrfs_set_root_ref_name_len(leaf, ref, name->len);
429 ptr = (unsigned long)(ref + 1);
430 write_extent_buffer(leaf, name->name, ptr, name->len);
431
432 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
433 btrfs_release_path(path);
434 key.objectid = ref_id;
435 key.type = BTRFS_ROOT_REF_KEY;
436 key.offset = root_id;
437 goto again;
438 }
439
440 return 0;
441}
442
443/*
444 * Old btrfs forgets to init root_item->flags and root_item->byte_limit
445 * for subvolumes. To work around this problem, we steal a bit from
446 * root_item->inode_item->flags, and use it to indicate if those fields
447 * have been properly initialized.
448 */
449void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
450{
451 u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
452
453 if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
454 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
455 btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
456 btrfs_set_root_flags(root_item, 0);
457 btrfs_set_root_limit(root_item, 0);
458 }
459}
460
461void btrfs_update_root_times(struct btrfs_trans_handle *trans,
462 struct btrfs_root *root)
463{
464 struct btrfs_root_item *item = &root->root_item;
465 struct timespec64 ct;
466
467 ktime_get_real_ts64(&ct);
468 spin_lock(&root->root_item_lock);
469 btrfs_set_root_ctransid(item, trans->transid);
470 btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
471 btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
472 spin_unlock(&root->root_item_lock);
473}
474
475/*
476 * Reserve space for subvolume operation.
477 *
478 * root: the root of the parent directory
479 * rsv: block reservation
480 * items: the number of items that we need do reservation
481 * use_global_rsv: allow fallback to the global block reservation
482 *
483 * This function is used to reserve the space for snapshot/subvolume
484 * creation and deletion. Those operations are different with the
485 * common file/directory operations, they change two fs/file trees
486 * and root tree, the number of items that the qgroup reserves is
487 * different with the free space reservation. So we can not use
488 * the space reservation mechanism in start_transaction().
489 */
490int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
491 struct btrfs_block_rsv *rsv, int items,
492 bool use_global_rsv)
493{
494 u64 qgroup_num_bytes = 0;
495 u64 num_bytes;
496 int ret;
497 struct btrfs_fs_info *fs_info = root->fs_info;
498 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
499
500 if (btrfs_qgroup_enabled(fs_info)) {
501 /* One for parent inode, two for dir entries */
502 qgroup_num_bytes = 3 * fs_info->nodesize;
503 ret = btrfs_qgroup_reserve_meta_prealloc(root,
504 qgroup_num_bytes, true,
505 false);
506 if (ret)
507 return ret;
508 }
509
510 num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
511 rsv->space_info = btrfs_find_space_info(fs_info,
512 BTRFS_BLOCK_GROUP_METADATA);
513 ret = btrfs_block_rsv_add(rsv, num_bytes, BTRFS_RESERVE_FLUSH_ALL);
514
515 if (ret == -ENOSPC && use_global_rsv)
516 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
517
518 if (ret && qgroup_num_bytes)
519 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
520
521 if (!ret) {
522 spin_lock(&rsv->lock);
523 rsv->qgroup_rsv_reserved += qgroup_num_bytes;
524 spin_unlock(&rsv->lock);
525 }
526 return ret;
527}