1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6 #include <linux/fs.h>
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
22 #include "ctree.h"
23 #include "disk-io.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "bio.h"
27 #include "print-tree.h"
28 #include "locking.h"
29 #include "tree-log.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "dev-replace.h"
33 #include "raid56.h"
34 #include "sysfs.h"
35 #include "qgroup.h"
36 #include "compression.h"
37 #include "tree-checker.h"
38 #include "ref-verify.h"
39 #include "block-group.h"
40 #include "discard.h"
41 #include "space-info.h"
42 #include "zoned.h"
43 #include "subpage.h"
44 #include "fs.h"
45 #include "accessors.h"
46 #include "extent-tree.h"
47 #include "root-tree.h"
48 #include "defrag.h"
49 #include "uuid-tree.h"
50 #include "relocation.h"
51 #include "scrub.h"
52 #include "super.h"
53
54 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
55 BTRFS_HEADER_FLAG_RELOC |\
56 BTRFS_SUPER_FLAG_ERROR |\
57 BTRFS_SUPER_FLAG_SEEDING |\
58 BTRFS_SUPER_FLAG_METADUMP |\
59 BTRFS_SUPER_FLAG_METADUMP_V2)
60
61 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
62 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
63
64 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
65 {
66 if (fs_info->csum_shash)
67 crypto_free_shash(fs_info->csum_shash);
68 }
69
70 /*
71 * Compute the csum of a btree block and store the result to provided buffer.
72 */
73 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
74 {
75 struct btrfs_fs_info *fs_info = buf->fs_info;
76 int num_pages;
77 u32 first_page_part;
78 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
79 char *kaddr;
80 int i;
81
82 shash->tfm = fs_info->csum_shash;
83 crypto_shash_init(shash);
84
85 if (buf->addr) {
86 /* Pages are contiguous, handle them as a big one. */
87 kaddr = buf->addr;
88 first_page_part = fs_info->nodesize;
89 num_pages = 1;
90 } else {
91 kaddr = folio_address(buf->folios[0]);
92 first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
93 num_pages = num_extent_pages(buf);
94 }
95
96 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
97 first_page_part - BTRFS_CSUM_SIZE);
98
99 /*
100 * Multiple single-page folios case would reach here.
101 *
102 * nodesize <= PAGE_SIZE and large folio all handled by above
103 * crypto_shash_update() already.
104 */
105 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
106 kaddr = folio_address(buf->folios[i]);
107 crypto_shash_update(shash, kaddr, PAGE_SIZE);
108 }
109 memset(result, 0, BTRFS_CSUM_SIZE);
110 crypto_shash_final(shash, result);
111 }
112
113 /*
114 * we can't consider a given block up to date unless the transid of the
115 * block matches the transid in the parent node's pointer. This is how we
116 * detect blocks that either didn't get written at all or got written
117 * in the wrong place.
118 */
119 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
120 {
121 if (!extent_buffer_uptodate(eb))
122 return 0;
123
124 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
125 return 1;
126
127 if (atomic)
128 return -EAGAIN;
129
130 if (!extent_buffer_uptodate(eb) ||
131 btrfs_header_generation(eb) != parent_transid) {
132 btrfs_err_rl(eb->fs_info,
133 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
134 eb->start, eb->read_mirror,
135 parent_transid, btrfs_header_generation(eb));
136 clear_extent_buffer_uptodate(eb);
137 return 0;
138 }
139 return 1;
140 }
141
142 static bool btrfs_supported_super_csum(u16 csum_type)
143 {
144 switch (csum_type) {
145 case BTRFS_CSUM_TYPE_CRC32:
146 case BTRFS_CSUM_TYPE_XXHASH:
147 case BTRFS_CSUM_TYPE_SHA256:
148 case BTRFS_CSUM_TYPE_BLAKE2:
149 return true;
150 default:
151 return false;
152 }
153 }
154
155 /*
156 * Return 0 if the superblock checksum type matches the checksum value of that
157 * algorithm. Pass the raw disk superblock data.
158 */
159 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
160 const struct btrfs_super_block *disk_sb)
161 {
162 char result[BTRFS_CSUM_SIZE];
163 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
164
165 shash->tfm = fs_info->csum_shash;
166
167 /*
168 * The super_block structure does not span the whole
169 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
170 * filled with zeros and is included in the checksum.
171 */
172 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
173 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
174
175 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
176 return 1;
177
178 return 0;
179 }
180
181 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
182 int mirror_num)
183 {
184 struct btrfs_fs_info *fs_info = eb->fs_info;
185 int num_folios = num_extent_folios(eb);
186 int ret = 0;
187
188 if (sb_rdonly(fs_info->sb))
189 return -EROFS;
190
191 for (int i = 0; i < num_folios; i++) {
192 struct folio *folio = eb->folios[i];
193 u64 start = max_t(u64, eb->start, folio_pos(folio));
194 u64 end = min_t(u64, eb->start + eb->len,
195 folio_pos(folio) + eb->folio_size);
196 u32 len = end - start;
197
198 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
199 start, folio, offset_in_folio(folio, start),
200 mirror_num);
201 if (ret)
202 break;
203 }
204
205 return ret;
206 }
207
208 /*
209 * helper to read a given tree block, doing retries as required when
210 * the checksums don't match and we have alternate mirrors to try.
211 *
212 * @check: expected tree parentness check, see the comments of the
213 * structure for details.
214 */
215 int btrfs_read_extent_buffer(struct extent_buffer *eb,
216 const struct btrfs_tree_parent_check *check)
217 {
218 struct btrfs_fs_info *fs_info = eb->fs_info;
219 int failed = 0;
220 int ret;
221 int num_copies = 0;
222 int mirror_num = 0;
223 int failed_mirror = 0;
224
225 ASSERT(check);
226
227 while (1) {
228 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
229 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
230 if (!ret)
231 break;
232
233 num_copies = btrfs_num_copies(fs_info,
234 eb->start, eb->len);
235 if (num_copies == 1)
236 break;
237
238 if (!failed_mirror) {
239 failed = 1;
240 failed_mirror = eb->read_mirror;
241 }
242
243 mirror_num++;
244 if (mirror_num == failed_mirror)
245 mirror_num++;
246
247 if (mirror_num > num_copies)
248 break;
249 }
250
251 if (failed && !ret && failed_mirror)
252 btrfs_repair_eb_io_failure(eb, failed_mirror);
253
254 return ret;
255 }
256
257 /*
258 * Checksum a dirty tree block before IO.
259 */
260 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
261 {
262 struct extent_buffer *eb = bbio->private;
263 struct btrfs_fs_info *fs_info = eb->fs_info;
264 u64 found_start = btrfs_header_bytenr(eb);
265 u64 last_trans;
266 u8 result[BTRFS_CSUM_SIZE];
267 int ret;
268
269 /* Btree blocks are always contiguous on disk. */
270 if (WARN_ON_ONCE(bbio->file_offset != eb->start))
271 return BLK_STS_IOERR;
272 if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
273 return BLK_STS_IOERR;
274
275 /*
276 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
277 * checksum it but zero-out its content. This is done to preserve
278 * ordering of I/O without unnecessarily writing out data.
279 */
280 if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
281 memzero_extent_buffer(eb, 0, eb->len);
282 return BLK_STS_OK;
283 }
284
285 if (WARN_ON_ONCE(found_start != eb->start))
286 return BLK_STS_IOERR;
287 if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
288 eb->start, eb->len)))
289 return BLK_STS_IOERR;
290
291 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
292 offsetof(struct btrfs_header, fsid),
293 BTRFS_FSID_SIZE) == 0);
294 csum_tree_block(eb, result);
295
296 if (btrfs_header_level(eb))
297 ret = btrfs_check_node(eb);
298 else
299 ret = btrfs_check_leaf(eb);
300
301 if (ret < 0)
302 goto error;
303
304 /*
305 * Also check the generation, the eb reached here must be newer than
306 * last committed. Or something seriously wrong happened.
307 */
308 last_trans = btrfs_get_last_trans_committed(fs_info);
309 if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
310 ret = -EUCLEAN;
311 btrfs_err(fs_info,
312 "block=%llu bad generation, have %llu expect > %llu",
313 eb->start, btrfs_header_generation(eb), last_trans);
314 goto error;
315 }
316 write_extent_buffer(eb, result, 0, fs_info->csum_size);
317 return BLK_STS_OK;
318
319 error:
320 btrfs_print_tree(eb, 0);
321 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
322 eb->start);
323 /*
324 * Be noisy if this is an extent buffer from a log tree. We don't abort
325 * a transaction in case there's a bad log tree extent buffer, we just
326 * fallback to a transaction commit. Still we want to know when there is
327 * a bad log tree extent buffer, as that may signal a bug somewhere.
328 */
329 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
330 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
331 return errno_to_blk_status(ret);
332 }
333
334 static bool check_tree_block_fsid(struct extent_buffer *eb)
335 {
336 struct btrfs_fs_info *fs_info = eb->fs_info;
337 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
338 u8 fsid[BTRFS_FSID_SIZE];
339
340 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
341 BTRFS_FSID_SIZE);
342
343 /*
344 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
345 * This is then overwritten by metadata_uuid if it is present in the
346 * device_list_add(). The same true for a seed device as well. So use of
347 * fs_devices::metadata_uuid is appropriate here.
348 */
349 if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
350 return false;
351
352 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
353 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
354 return false;
355
356 return true;
357 }
358
359 /* Do basic extent buffer checks at read time */
360 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
361 const struct btrfs_tree_parent_check *check)
362 {
363 struct btrfs_fs_info *fs_info = eb->fs_info;
364 u64 found_start;
365 const u32 csum_size = fs_info->csum_size;
366 u8 found_level;
367 u8 result[BTRFS_CSUM_SIZE];
368 const u8 *header_csum;
369 int ret = 0;
370 const bool ignore_csum = btrfs_test_opt(fs_info, IGNOREMETACSUMS);
371
372 ASSERT(check);
373
374 found_start = btrfs_header_bytenr(eb);
375 if (found_start != eb->start) {
376 btrfs_err_rl(fs_info,
377 "bad tree block start, mirror %u want %llu have %llu",
378 eb->read_mirror, eb->start, found_start);
379 ret = -EIO;
380 goto out;
381 }
382 if (check_tree_block_fsid(eb)) {
383 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
384 eb->start, eb->read_mirror);
385 ret = -EIO;
386 goto out;
387 }
388 found_level = btrfs_header_level(eb);
389 if (found_level >= BTRFS_MAX_LEVEL) {
390 btrfs_err(fs_info,
391 "bad tree block level, mirror %u level %d on logical %llu",
392 eb->read_mirror, btrfs_header_level(eb), eb->start);
393 ret = -EIO;
394 goto out;
395 }
396
397 csum_tree_block(eb, result);
398 header_csum = folio_address(eb->folios[0]) +
399 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
400
401 if (memcmp(result, header_csum, csum_size) != 0) {
402 btrfs_warn_rl(fs_info,
403 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d%s",
404 eb->start, eb->read_mirror,
405 CSUM_FMT_VALUE(csum_size, header_csum),
406 CSUM_FMT_VALUE(csum_size, result),
407 btrfs_header_level(eb),
408 ignore_csum ? ", ignored" : "");
409 if (!ignore_csum) {
410 ret = -EUCLEAN;
411 goto out;
412 }
413 }
414
415 if (found_level != check->level) {
416 btrfs_err(fs_info,
417 "level verify failed on logical %llu mirror %u wanted %u found %u",
418 eb->start, eb->read_mirror, check->level, found_level);
419 ret = -EIO;
420 goto out;
421 }
422 if (unlikely(check->transid &&
423 btrfs_header_generation(eb) != check->transid)) {
424 btrfs_err_rl(eb->fs_info,
425 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
426 eb->start, eb->read_mirror, check->transid,
427 btrfs_header_generation(eb));
428 ret = -EIO;
429 goto out;
430 }
431 if (check->has_first_key) {
432 const struct btrfs_key *expect_key = &check->first_key;
433 struct btrfs_key found_key;
434
435 if (found_level)
436 btrfs_node_key_to_cpu(eb, &found_key, 0);
437 else
438 btrfs_item_key_to_cpu(eb, &found_key, 0);
439 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
440 btrfs_err(fs_info,
441 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
442 eb->start, check->transid,
443 expect_key->objectid,
444 expect_key->type, expect_key->offset,
445 found_key.objectid, found_key.type,
446 found_key.offset);
447 ret = -EUCLEAN;
448 goto out;
449 }
450 }
451 if (check->owner_root) {
452 ret = btrfs_check_eb_owner(eb, check->owner_root);
453 if (ret < 0)
454 goto out;
455 }
456
457 /*
458 * If this is a leaf block and it is corrupt, set the corrupt bit so
459 * that we don't try and read the other copies of this block, just
460 * return -EIO.
461 */
462 if (found_level == 0 && btrfs_check_leaf(eb)) {
463 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
464 ret = -EIO;
465 }
466
467 if (found_level > 0 && btrfs_check_node(eb))
468 ret = -EIO;
469
470 if (ret)
471 btrfs_err(fs_info,
472 "read time tree block corruption detected on logical %llu mirror %u",
473 eb->start, eb->read_mirror);
474 out:
475 return ret;
476 }
477
478 #ifdef CONFIG_MIGRATION
479 static int btree_migrate_folio(struct address_space *mapping,
480 struct folio *dst, struct folio *src, enum migrate_mode mode)
481 {
482 /*
483 * we can't safely write a btree page from here,
484 * we haven't done the locking hook
485 */
486 if (folio_test_dirty(src))
487 return -EAGAIN;
488 /*
489 * Buffers may be managed in a filesystem specific way.
490 * We must have no buffers or drop them.
491 */
492 if (folio_get_private(src) &&
493 !filemap_release_folio(src, GFP_KERNEL))
494 return -EAGAIN;
495 return migrate_folio(mapping, dst, src, mode);
496 }
497 #else
498 #define btree_migrate_folio NULL
499 #endif
500
501 static int btree_writepages(struct address_space *mapping,
502 struct writeback_control *wbc)
503 {
504 int ret;
505
506 if (wbc->sync_mode == WB_SYNC_NONE) {
507 struct btrfs_fs_info *fs_info;
508
509 if (wbc->for_kupdate)
510 return 0;
511
512 fs_info = inode_to_fs_info(mapping->host);
513 /* this is a bit racy, but that's ok */
514 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
515 BTRFS_DIRTY_METADATA_THRESH,
516 fs_info->dirty_metadata_batch);
517 if (ret < 0)
518 return 0;
519 }
520 return btree_write_cache_pages(mapping, wbc);
521 }
522
523 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
524 {
525 if (folio_test_writeback(folio) || folio_test_dirty(folio))
526 return false;
527
528 return try_release_extent_buffer(&folio->page);
529 }
530
531 static void btree_invalidate_folio(struct folio *folio, size_t offset,
532 size_t length)
533 {
534 struct extent_io_tree *tree;
535
536 tree = &folio_to_inode(folio)->io_tree;
537 extent_invalidate_folio(tree, folio, offset);
538 btree_release_folio(folio, GFP_NOFS);
539 if (folio_get_private(folio)) {
540 btrfs_warn(folio_to_fs_info(folio),
541 "folio private not zero on folio %llu",
542 (unsigned long long)folio_pos(folio));
543 folio_detach_private(folio);
544 }
545 }
546
547 #ifdef DEBUG
548 static bool btree_dirty_folio(struct address_space *mapping,
549 struct folio *folio)
550 {
551 struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
552 struct btrfs_subpage_info *spi = fs_info->subpage_info;
553 struct btrfs_subpage *subpage;
554 struct extent_buffer *eb;
555 int cur_bit = 0;
556 u64 page_start = folio_pos(folio);
557
558 if (fs_info->sectorsize == PAGE_SIZE) {
559 eb = folio_get_private(folio);
560 BUG_ON(!eb);
561 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
562 BUG_ON(!atomic_read(&eb->refs));
563 btrfs_assert_tree_write_locked(eb);
564 return filemap_dirty_folio(mapping, folio);
565 }
566
567 ASSERT(spi);
568 subpage = folio_get_private(folio);
569
570 for (cur_bit = spi->dirty_offset;
571 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
572 cur_bit++) {
573 unsigned long flags;
574 u64 cur;
575
576 spin_lock_irqsave(&subpage->lock, flags);
577 if (!test_bit(cur_bit, subpage->bitmaps)) {
578 spin_unlock_irqrestore(&subpage->lock, flags);
579 continue;
580 }
581 spin_unlock_irqrestore(&subpage->lock, flags);
582 cur = page_start + cur_bit * fs_info->sectorsize;
583
584 eb = find_extent_buffer(fs_info, cur);
585 ASSERT(eb);
586 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
587 ASSERT(atomic_read(&eb->refs));
588 btrfs_assert_tree_write_locked(eb);
589 free_extent_buffer(eb);
590
591 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
592 }
593 return filemap_dirty_folio(mapping, folio);
594 }
595 #else
596 #define btree_dirty_folio filemap_dirty_folio
597 #endif
598
599 static const struct address_space_operations btree_aops = {
600 .writepages = btree_writepages,
601 .release_folio = btree_release_folio,
602 .invalidate_folio = btree_invalidate_folio,
603 .migrate_folio = btree_migrate_folio,
604 .dirty_folio = btree_dirty_folio,
605 };
606
607 struct extent_buffer *btrfs_find_create_tree_block(
608 struct btrfs_fs_info *fs_info,
609 u64 bytenr, u64 owner_root,
610 int level)
611 {
612 if (btrfs_is_testing(fs_info))
613 return alloc_test_extent_buffer(fs_info, bytenr);
614 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
615 }
616
617 /*
618 * Read tree block at logical address @bytenr and do variant basic but critical
619 * verification.
620 *
621 * @check: expected tree parentness check, see comments of the
622 * structure for details.
623 */
624 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
625 struct btrfs_tree_parent_check *check)
626 {
627 struct extent_buffer *buf = NULL;
628 int ret;
629
630 ASSERT(check);
631
632 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
633 check->level);
634 if (IS_ERR(buf))
635 return buf;
636
637 ret = btrfs_read_extent_buffer(buf, check);
638 if (ret) {
639 free_extent_buffer_stale(buf);
640 return ERR_PTR(ret);
641 }
642 return buf;
643
644 }
645
646 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
647 u64 objectid)
648 {
649 bool dummy = btrfs_is_testing(fs_info);
650
651 memset(&root->root_key, 0, sizeof(root->root_key));
652 memset(&root->root_item, 0, sizeof(root->root_item));
653 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
654 root->fs_info = fs_info;
655 root->root_key.objectid = objectid;
656 root->node = NULL;
657 root->commit_root = NULL;
658 root->state = 0;
659 RB_CLEAR_NODE(&root->rb_node);
660
661 btrfs_set_root_last_trans(root, 0);
662 root->free_objectid = 0;
663 root->nr_delalloc_inodes = 0;
664 root->nr_ordered_extents = 0;
665 xa_init(&root->inodes);
666 xa_init(&root->delayed_nodes);
667
668 btrfs_init_root_block_rsv(root);
669
670 INIT_LIST_HEAD(&root->dirty_list);
671 INIT_LIST_HEAD(&root->root_list);
672 INIT_LIST_HEAD(&root->delalloc_inodes);
673 INIT_LIST_HEAD(&root->delalloc_root);
674 INIT_LIST_HEAD(&root->ordered_extents);
675 INIT_LIST_HEAD(&root->ordered_root);
676 INIT_LIST_HEAD(&root->reloc_dirty_list);
677 spin_lock_init(&root->delalloc_lock);
678 spin_lock_init(&root->ordered_extent_lock);
679 spin_lock_init(&root->accounting_lock);
680 spin_lock_init(&root->qgroup_meta_rsv_lock);
681 mutex_init(&root->objectid_mutex);
682 mutex_init(&root->log_mutex);
683 mutex_init(&root->ordered_extent_mutex);
684 mutex_init(&root->delalloc_mutex);
685 init_waitqueue_head(&root->qgroup_flush_wait);
686 init_waitqueue_head(&root->log_writer_wait);
687 init_waitqueue_head(&root->log_commit_wait[0]);
688 init_waitqueue_head(&root->log_commit_wait[1]);
689 INIT_LIST_HEAD(&root->log_ctxs[0]);
690 INIT_LIST_HEAD(&root->log_ctxs[1]);
691 atomic_set(&root->log_commit[0], 0);
692 atomic_set(&root->log_commit[1], 0);
693 atomic_set(&root->log_writers, 0);
694 atomic_set(&root->log_batch, 0);
695 refcount_set(&root->refs, 1);
696 atomic_set(&root->snapshot_force_cow, 0);
697 atomic_set(&root->nr_swapfiles, 0);
698 btrfs_set_root_log_transid(root, 0);
699 root->log_transid_committed = -1;
700 btrfs_set_root_last_log_commit(root, 0);
701 root->anon_dev = 0;
702 if (!dummy) {
703 extent_io_tree_init(fs_info, &root->dirty_log_pages,
704 IO_TREE_ROOT_DIRTY_LOG_PAGES);
705 extent_io_tree_init(fs_info, &root->log_csum_range,
706 IO_TREE_LOG_CSUM_RANGE);
707 }
708
709 spin_lock_init(&root->root_item_lock);
710 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
711 #ifdef CONFIG_BTRFS_DEBUG
712 INIT_LIST_HEAD(&root->leak_list);
713 spin_lock(&fs_info->fs_roots_radix_lock);
714 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
715 spin_unlock(&fs_info->fs_roots_radix_lock);
716 #endif
717 }
718
719 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
720 u64 objectid, gfp_t flags)
721 {
722 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
723 if (root)
724 __setup_root(root, fs_info, objectid);
725 return root;
726 }
727
728 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
729 /* Should only be used by the testing infrastructure */
730 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
731 {
732 struct btrfs_root *root;
733
734 if (!fs_info)
735 return ERR_PTR(-EINVAL);
736
737 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
738 if (!root)
739 return ERR_PTR(-ENOMEM);
740
741 /* We don't use the stripesize in selftest, set it as sectorsize */
742 root->alloc_bytenr = 0;
743
744 return root;
745 }
746 #endif
747
748 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
749 {
750 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
751 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
752
753 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
754 }
755
756 static int global_root_key_cmp(const void *k, const struct rb_node *node)
757 {
758 const struct btrfs_key *key = k;
759 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
760
761 return btrfs_comp_cpu_keys(key, &root->root_key);
762 }
763
764 int btrfs_global_root_insert(struct btrfs_root *root)
765 {
766 struct btrfs_fs_info *fs_info = root->fs_info;
767 struct rb_node *tmp;
768 int ret = 0;
769
770 write_lock(&fs_info->global_root_lock);
771 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
772 write_unlock(&fs_info->global_root_lock);
773
774 if (tmp) {
775 ret = -EEXIST;
776 btrfs_warn(fs_info, "global root %llu %llu already exists",
777 btrfs_root_id(root), root->root_key.offset);
778 }
779 return ret;
780 }
781
782 void btrfs_global_root_delete(struct btrfs_root *root)
783 {
784 struct btrfs_fs_info *fs_info = root->fs_info;
785
786 write_lock(&fs_info->global_root_lock);
787 rb_erase(&root->rb_node, &fs_info->global_root_tree);
788 write_unlock(&fs_info->global_root_lock);
789 }
790
791 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
792 struct btrfs_key *key)
793 {
794 struct rb_node *node;
795 struct btrfs_root *root = NULL;
796
797 read_lock(&fs_info->global_root_lock);
798 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
799 if (node)
800 root = container_of(node, struct btrfs_root, rb_node);
801 read_unlock(&fs_info->global_root_lock);
802
803 return root;
804 }
805
806 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
807 {
808 struct btrfs_block_group *block_group;
809 u64 ret;
810
811 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
812 return 0;
813
814 if (bytenr)
815 block_group = btrfs_lookup_block_group(fs_info, bytenr);
816 else
817 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
818 ASSERT(block_group);
819 if (!block_group)
820 return 0;
821 ret = block_group->global_root_id;
822 btrfs_put_block_group(block_group);
823
824 return ret;
825 }
826
827 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
828 {
829 struct btrfs_key key = {
830 .objectid = BTRFS_CSUM_TREE_OBJECTID,
831 .type = BTRFS_ROOT_ITEM_KEY,
832 .offset = btrfs_global_root_id(fs_info, bytenr),
833 };
834
835 return btrfs_global_root(fs_info, &key);
836 }
837
838 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
839 {
840 struct btrfs_key key = {
841 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
842 .type = BTRFS_ROOT_ITEM_KEY,
843 .offset = btrfs_global_root_id(fs_info, bytenr),
844 };
845
846 return btrfs_global_root(fs_info, &key);
847 }
848
849 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
850 u64 objectid)
851 {
852 struct btrfs_fs_info *fs_info = trans->fs_info;
853 struct extent_buffer *leaf;
854 struct btrfs_root *tree_root = fs_info->tree_root;
855 struct btrfs_root *root;
856 struct btrfs_key key;
857 unsigned int nofs_flag;
858 int ret = 0;
859
860 /*
861 * We're holding a transaction handle, so use a NOFS memory allocation
862 * context to avoid deadlock if reclaim happens.
863 */
864 nofs_flag = memalloc_nofs_save();
865 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
866 memalloc_nofs_restore(nofs_flag);
867 if (!root)
868 return ERR_PTR(-ENOMEM);
869
870 root->root_key.objectid = objectid;
871 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
872 root->root_key.offset = 0;
873
874 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
875 0, BTRFS_NESTING_NORMAL);
876 if (IS_ERR(leaf)) {
877 ret = PTR_ERR(leaf);
878 leaf = NULL;
879 goto fail;
880 }
881
882 root->node = leaf;
883 btrfs_mark_buffer_dirty(trans, leaf);
884
885 root->commit_root = btrfs_root_node(root);
886 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
887
888 btrfs_set_root_flags(&root->root_item, 0);
889 btrfs_set_root_limit(&root->root_item, 0);
890 btrfs_set_root_bytenr(&root->root_item, leaf->start);
891 btrfs_set_root_generation(&root->root_item, trans->transid);
892 btrfs_set_root_level(&root->root_item, 0);
893 btrfs_set_root_refs(&root->root_item, 1);
894 btrfs_set_root_used(&root->root_item, leaf->len);
895 btrfs_set_root_last_snapshot(&root->root_item, 0);
896 btrfs_set_root_dirid(&root->root_item, 0);
897 if (is_fstree(objectid))
898 generate_random_guid(root->root_item.uuid);
899 else
900 export_guid(root->root_item.uuid, &guid_null);
901 btrfs_set_root_drop_level(&root->root_item, 0);
902
903 btrfs_tree_unlock(leaf);
904
905 key.objectid = objectid;
906 key.type = BTRFS_ROOT_ITEM_KEY;
907 key.offset = 0;
908 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
909 if (ret)
910 goto fail;
911
912 return root;
913
914 fail:
915 btrfs_put_root(root);
916
917 return ERR_PTR(ret);
918 }
919
920 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
921 struct btrfs_fs_info *fs_info)
922 {
923 struct btrfs_root *root;
924
925 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
926 if (!root)
927 return ERR_PTR(-ENOMEM);
928
929 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
930 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
931 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
932
933 return root;
934 }
935
936 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
937 struct btrfs_root *root)
938 {
939 struct extent_buffer *leaf;
940
941 /*
942 * DON'T set SHAREABLE bit for log trees.
943 *
944 * Log trees are not exposed to user space thus can't be snapshotted,
945 * and they go away before a real commit is actually done.
946 *
947 * They do store pointers to file data extents, and those reference
948 * counts still get updated (along with back refs to the log tree).
949 */
950
951 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
952 NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
953 if (IS_ERR(leaf))
954 return PTR_ERR(leaf);
955
956 root->node = leaf;
957
958 btrfs_mark_buffer_dirty(trans, root->node);
959 btrfs_tree_unlock(root->node);
960
961 return 0;
962 }
963
964 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
965 struct btrfs_fs_info *fs_info)
966 {
967 struct btrfs_root *log_root;
968
969 log_root = alloc_log_tree(trans, fs_info);
970 if (IS_ERR(log_root))
971 return PTR_ERR(log_root);
972
973 if (!btrfs_is_zoned(fs_info)) {
974 int ret = btrfs_alloc_log_tree_node(trans, log_root);
975
976 if (ret) {
977 btrfs_put_root(log_root);
978 return ret;
979 }
980 }
981
982 WARN_ON(fs_info->log_root_tree);
983 fs_info->log_root_tree = log_root;
984 return 0;
985 }
986
987 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
988 struct btrfs_root *root)
989 {
990 struct btrfs_fs_info *fs_info = root->fs_info;
991 struct btrfs_root *log_root;
992 struct btrfs_inode_item *inode_item;
993 int ret;
994
995 log_root = alloc_log_tree(trans, fs_info);
996 if (IS_ERR(log_root))
997 return PTR_ERR(log_root);
998
999 ret = btrfs_alloc_log_tree_node(trans, log_root);
1000 if (ret) {
1001 btrfs_put_root(log_root);
1002 return ret;
1003 }
1004
1005 btrfs_set_root_last_trans(log_root, trans->transid);
1006 log_root->root_key.offset = btrfs_root_id(root);
1007
1008 inode_item = &log_root->root_item.inode;
1009 btrfs_set_stack_inode_generation(inode_item, 1);
1010 btrfs_set_stack_inode_size(inode_item, 3);
1011 btrfs_set_stack_inode_nlink(inode_item, 1);
1012 btrfs_set_stack_inode_nbytes(inode_item,
1013 fs_info->nodesize);
1014 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1015
1016 btrfs_set_root_node(&log_root->root_item, log_root->node);
1017
1018 WARN_ON(root->log_root);
1019 root->log_root = log_root;
1020 btrfs_set_root_log_transid(root, 0);
1021 root->log_transid_committed = -1;
1022 btrfs_set_root_last_log_commit(root, 0);
1023 return 0;
1024 }
1025
1026 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1027 struct btrfs_path *path,
1028 const struct btrfs_key *key)
1029 {
1030 struct btrfs_root *root;
1031 struct btrfs_tree_parent_check check = { 0 };
1032 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1033 u64 generation;
1034 int ret;
1035 int level;
1036
1037 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1038 if (!root)
1039 return ERR_PTR(-ENOMEM);
1040
1041 ret = btrfs_find_root(tree_root, key, path,
1042 &root->root_item, &root->root_key);
1043 if (ret) {
1044 if (ret > 0)
1045 ret = -ENOENT;
1046 goto fail;
1047 }
1048
1049 generation = btrfs_root_generation(&root->root_item);
1050 level = btrfs_root_level(&root->root_item);
1051 check.level = level;
1052 check.transid = generation;
1053 check.owner_root = key->objectid;
1054 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1055 &check);
1056 if (IS_ERR(root->node)) {
1057 ret = PTR_ERR(root->node);
1058 root->node = NULL;
1059 goto fail;
1060 }
1061 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1062 ret = -EIO;
1063 goto fail;
1064 }
1065
1066 /*
1067 * For real fs, and not log/reloc trees, root owner must
1068 * match its root node owner
1069 */
1070 if (!btrfs_is_testing(fs_info) &&
1071 btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1072 btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID &&
1073 btrfs_root_id(root) != btrfs_header_owner(root->node)) {
1074 btrfs_crit(fs_info,
1075 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1076 btrfs_root_id(root), root->node->start,
1077 btrfs_header_owner(root->node),
1078 btrfs_root_id(root));
1079 ret = -EUCLEAN;
1080 goto fail;
1081 }
1082 root->commit_root = btrfs_root_node(root);
1083 return root;
1084 fail:
1085 btrfs_put_root(root);
1086 return ERR_PTR(ret);
1087 }
1088
1089 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1090 const struct btrfs_key *key)
1091 {
1092 struct btrfs_root *root;
1093 struct btrfs_path *path;
1094
1095 path = btrfs_alloc_path();
1096 if (!path)
1097 return ERR_PTR(-ENOMEM);
1098 root = read_tree_root_path(tree_root, path, key);
1099 btrfs_free_path(path);
1100
1101 return root;
1102 }
1103
1104 /*
1105 * Initialize subvolume root in-memory structure
1106 *
1107 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1108 */
1109 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1110 {
1111 int ret;
1112
1113 btrfs_drew_lock_init(&root->snapshot_lock);
1114
1115 if (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1116 !btrfs_is_data_reloc_root(root) &&
1117 is_fstree(btrfs_root_id(root))) {
1118 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1119 btrfs_check_and_init_root_item(&root->root_item);
1120 }
1121
1122 /*
1123 * Don't assign anonymous block device to roots that are not exposed to
1124 * userspace, the id pool is limited to 1M
1125 */
1126 if (is_fstree(btrfs_root_id(root)) &&
1127 btrfs_root_refs(&root->root_item) > 0) {
1128 if (!anon_dev) {
1129 ret = get_anon_bdev(&root->anon_dev);
1130 if (ret)
1131 goto fail;
1132 } else {
1133 root->anon_dev = anon_dev;
1134 }
1135 }
1136
1137 mutex_lock(&root->objectid_mutex);
1138 ret = btrfs_init_root_free_objectid(root);
1139 if (ret) {
1140 mutex_unlock(&root->objectid_mutex);
1141 goto fail;
1142 }
1143
1144 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1145
1146 mutex_unlock(&root->objectid_mutex);
1147
1148 return 0;
1149 fail:
1150 /* The caller is responsible to call btrfs_free_fs_root */
1151 return ret;
1152 }
1153
1154 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1155 u64 root_id)
1156 {
1157 struct btrfs_root *root;
1158
1159 spin_lock(&fs_info->fs_roots_radix_lock);
1160 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1161 (unsigned long)root_id);
1162 root = btrfs_grab_root(root);
1163 spin_unlock(&fs_info->fs_roots_radix_lock);
1164 return root;
1165 }
1166
1167 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1168 u64 objectid)
1169 {
1170 struct btrfs_key key = {
1171 .objectid = objectid,
1172 .type = BTRFS_ROOT_ITEM_KEY,
1173 .offset = 0,
1174 };
1175
1176 switch (objectid) {
1177 case BTRFS_ROOT_TREE_OBJECTID:
1178 return btrfs_grab_root(fs_info->tree_root);
1179 case BTRFS_EXTENT_TREE_OBJECTID:
1180 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1181 case BTRFS_CHUNK_TREE_OBJECTID:
1182 return btrfs_grab_root(fs_info->chunk_root);
1183 case BTRFS_DEV_TREE_OBJECTID:
1184 return btrfs_grab_root(fs_info->dev_root);
1185 case BTRFS_CSUM_TREE_OBJECTID:
1186 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1187 case BTRFS_QUOTA_TREE_OBJECTID:
1188 return btrfs_grab_root(fs_info->quota_root);
1189 case BTRFS_UUID_TREE_OBJECTID:
1190 return btrfs_grab_root(fs_info->uuid_root);
1191 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1192 return btrfs_grab_root(fs_info->block_group_root);
1193 case BTRFS_FREE_SPACE_TREE_OBJECTID:
1194 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1195 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1196 return btrfs_grab_root(fs_info->stripe_root);
1197 default:
1198 return NULL;
1199 }
1200 }
1201
1202 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1203 struct btrfs_root *root)
1204 {
1205 int ret;
1206
1207 ret = radix_tree_preload(GFP_NOFS);
1208 if (ret)
1209 return ret;
1210
1211 spin_lock(&fs_info->fs_roots_radix_lock);
1212 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1213 (unsigned long)btrfs_root_id(root),
1214 root);
1215 if (ret == 0) {
1216 btrfs_grab_root(root);
1217 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1218 }
1219 spin_unlock(&fs_info->fs_roots_radix_lock);
1220 radix_tree_preload_end();
1221
1222 return ret;
1223 }
1224
1225 void btrfs_check_leaked_roots(const struct btrfs_fs_info *fs_info)
1226 {
1227 #ifdef CONFIG_BTRFS_DEBUG
1228 struct btrfs_root *root;
1229
1230 while (!list_empty(&fs_info->allocated_roots)) {
1231 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1232
1233 root = list_first_entry(&fs_info->allocated_roots,
1234 struct btrfs_root, leak_list);
1235 btrfs_err(fs_info, "leaked root %s refcount %d",
1236 btrfs_root_name(&root->root_key, buf),
1237 refcount_read(&root->refs));
1238 WARN_ON_ONCE(1);
1239 while (refcount_read(&root->refs) > 1)
1240 btrfs_put_root(root);
1241 btrfs_put_root(root);
1242 }
1243 #endif
1244 }
1245
1246 static void free_global_roots(struct btrfs_fs_info *fs_info)
1247 {
1248 struct btrfs_root *root;
1249 struct rb_node *node;
1250
1251 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1252 root = rb_entry(node, struct btrfs_root, rb_node);
1253 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1254 btrfs_put_root(root);
1255 }
1256 }
1257
1258 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1259 {
1260 struct percpu_counter *em_counter = &fs_info->evictable_extent_maps;
1261
1262 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1263 percpu_counter_destroy(&fs_info->delalloc_bytes);
1264 percpu_counter_destroy(&fs_info->ordered_bytes);
1265 if (percpu_counter_initialized(em_counter))
1266 ASSERT(percpu_counter_sum_positive(em_counter) == 0);
1267 percpu_counter_destroy(em_counter);
1268 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1269 btrfs_free_csum_hash(fs_info);
1270 btrfs_free_stripe_hash_table(fs_info);
1271 btrfs_free_ref_cache(fs_info);
1272 kfree(fs_info->balance_ctl);
1273 kfree(fs_info->delayed_root);
1274 free_global_roots(fs_info);
1275 btrfs_put_root(fs_info->tree_root);
1276 btrfs_put_root(fs_info->chunk_root);
1277 btrfs_put_root(fs_info->dev_root);
1278 btrfs_put_root(fs_info->quota_root);
1279 btrfs_put_root(fs_info->uuid_root);
1280 btrfs_put_root(fs_info->fs_root);
1281 btrfs_put_root(fs_info->data_reloc_root);
1282 btrfs_put_root(fs_info->block_group_root);
1283 btrfs_put_root(fs_info->stripe_root);
1284 btrfs_check_leaked_roots(fs_info);
1285 btrfs_extent_buffer_leak_debug_check(fs_info);
1286 kfree(fs_info->super_copy);
1287 kfree(fs_info->super_for_commit);
1288 kfree(fs_info->subpage_info);
1289 kvfree(fs_info);
1290 }
1291
1292
1293 /*
1294 * Get an in-memory reference of a root structure.
1295 *
1296 * For essential trees like root/extent tree, we grab it from fs_info directly.
1297 * For subvolume trees, we check the cached filesystem roots first. If not
1298 * found, then read it from disk and add it to cached fs roots.
1299 *
1300 * Caller should release the root by calling btrfs_put_root() after the usage.
1301 *
1302 * NOTE: Reloc and log trees can't be read by this function as they share the
1303 * same root objectid.
1304 *
1305 * @objectid: root id
1306 * @anon_dev: preallocated anonymous block device number for new roots,
1307 * pass NULL for a new allocation.
1308 * @check_ref: whether to check root item references, If true, return -ENOENT
1309 * for orphan roots
1310 */
1311 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1312 u64 objectid, dev_t *anon_dev,
1313 bool check_ref)
1314 {
1315 struct btrfs_root *root;
1316 struct btrfs_path *path;
1317 struct btrfs_key key;
1318 int ret;
1319
1320 root = btrfs_get_global_root(fs_info, objectid);
1321 if (root)
1322 return root;
1323
1324 /*
1325 * If we're called for non-subvolume trees, and above function didn't
1326 * find one, do not try to read it from disk.
1327 *
1328 * This is namely for free-space-tree and quota tree, which can change
1329 * at runtime and should only be grabbed from fs_info.
1330 */
1331 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1332 return ERR_PTR(-ENOENT);
1333 again:
1334 root = btrfs_lookup_fs_root(fs_info, objectid);
1335 if (root) {
1336 /*
1337 * Some other caller may have read out the newly inserted
1338 * subvolume already (for things like backref walk etc). Not
1339 * that common but still possible. In that case, we just need
1340 * to free the anon_dev.
1341 */
1342 if (unlikely(anon_dev && *anon_dev)) {
1343 free_anon_bdev(*anon_dev);
1344 *anon_dev = 0;
1345 }
1346
1347 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1348 btrfs_put_root(root);
1349 return ERR_PTR(-ENOENT);
1350 }
1351 return root;
1352 }
1353
1354 key.objectid = objectid;
1355 key.type = BTRFS_ROOT_ITEM_KEY;
1356 key.offset = (u64)-1;
1357 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1358 if (IS_ERR(root))
1359 return root;
1360
1361 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1362 ret = -ENOENT;
1363 goto fail;
1364 }
1365
1366 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1367 if (ret)
1368 goto fail;
1369
1370 path = btrfs_alloc_path();
1371 if (!path) {
1372 ret = -ENOMEM;
1373 goto fail;
1374 }
1375 key.objectid = BTRFS_ORPHAN_OBJECTID;
1376 key.type = BTRFS_ORPHAN_ITEM_KEY;
1377 key.offset = objectid;
1378
1379 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1380 btrfs_free_path(path);
1381 if (ret < 0)
1382 goto fail;
1383 if (ret == 0)
1384 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1385
1386 ret = btrfs_insert_fs_root(fs_info, root);
1387 if (ret) {
1388 if (ret == -EEXIST) {
1389 btrfs_put_root(root);
1390 goto again;
1391 }
1392 goto fail;
1393 }
1394 return root;
1395 fail:
1396 /*
1397 * If our caller provided us an anonymous device, then it's his
1398 * responsibility to free it in case we fail. So we have to set our
1399 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1400 * and once again by our caller.
1401 */
1402 if (anon_dev && *anon_dev)
1403 root->anon_dev = 0;
1404 btrfs_put_root(root);
1405 return ERR_PTR(ret);
1406 }
1407
1408 /*
1409 * Get in-memory reference of a root structure
1410 *
1411 * @objectid: tree objectid
1412 * @check_ref: if set, verify that the tree exists and the item has at least
1413 * one reference
1414 */
1415 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1416 u64 objectid, bool check_ref)
1417 {
1418 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1419 }
1420
1421 /*
1422 * Get in-memory reference of a root structure, created as new, optionally pass
1423 * the anonymous block device id
1424 *
1425 * @objectid: tree objectid
1426 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1427 * parameter value if not NULL
1428 */
1429 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1430 u64 objectid, dev_t *anon_dev)
1431 {
1432 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1433 }
1434
1435 /*
1436 * Return a root for the given objectid.
1437 *
1438 * @fs_info: the fs_info
1439 * @objectid: the objectid we need to lookup
1440 *
1441 * This is exclusively used for backref walking, and exists specifically because
1442 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1443 * creation time, which means we may have to read the tree_root in order to look
1444 * up a fs root that is not in memory. If the root is not in memory we will
1445 * read the tree root commit root and look up the fs root from there. This is a
1446 * temporary root, it will not be inserted into the radix tree as it doesn't
1447 * have the most uptodate information, it'll simply be discarded once the
1448 * backref code is finished using the root.
1449 */
1450 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1451 struct btrfs_path *path,
1452 u64 objectid)
1453 {
1454 struct btrfs_root *root;
1455 struct btrfs_key key;
1456
1457 ASSERT(path->search_commit_root && path->skip_locking);
1458
1459 /*
1460 * This can return -ENOENT if we ask for a root that doesn't exist, but
1461 * since this is called via the backref walking code we won't be looking
1462 * up a root that doesn't exist, unless there's corruption. So if root
1463 * != NULL just return it.
1464 */
1465 root = btrfs_get_global_root(fs_info, objectid);
1466 if (root)
1467 return root;
1468
1469 root = btrfs_lookup_fs_root(fs_info, objectid);
1470 if (root)
1471 return root;
1472
1473 key.objectid = objectid;
1474 key.type = BTRFS_ROOT_ITEM_KEY;
1475 key.offset = (u64)-1;
1476 root = read_tree_root_path(fs_info->tree_root, path, &key);
1477 btrfs_release_path(path);
1478
1479 return root;
1480 }
1481
1482 static int cleaner_kthread(void *arg)
1483 {
1484 struct btrfs_fs_info *fs_info = arg;
1485 int again;
1486
1487 while (1) {
1488 again = 0;
1489
1490 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1491
1492 /* Make the cleaner go to sleep early. */
1493 if (btrfs_need_cleaner_sleep(fs_info))
1494 goto sleep;
1495
1496 /*
1497 * Do not do anything if we might cause open_ctree() to block
1498 * before we have finished mounting the filesystem.
1499 */
1500 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1501 goto sleep;
1502
1503 if (!mutex_trylock(&fs_info->cleaner_mutex))
1504 goto sleep;
1505
1506 /*
1507 * Avoid the problem that we change the status of the fs
1508 * during the above check and trylock.
1509 */
1510 if (btrfs_need_cleaner_sleep(fs_info)) {
1511 mutex_unlock(&fs_info->cleaner_mutex);
1512 goto sleep;
1513 }
1514
1515 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1516 btrfs_sysfs_feature_update(fs_info);
1517
1518 btrfs_run_delayed_iputs(fs_info);
1519
1520 again = btrfs_clean_one_deleted_snapshot(fs_info);
1521 mutex_unlock(&fs_info->cleaner_mutex);
1522
1523 /*
1524 * The defragger has dealt with the R/O remount and umount,
1525 * needn't do anything special here.
1526 */
1527 btrfs_run_defrag_inodes(fs_info);
1528
1529 /*
1530 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1531 * with relocation (btrfs_relocate_chunk) and relocation
1532 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1533 * after acquiring fs_info->reclaim_bgs_lock. So we
1534 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1535 * unused block groups.
1536 */
1537 btrfs_delete_unused_bgs(fs_info);
1538
1539 /*
1540 * Reclaim block groups in the reclaim_bgs list after we deleted
1541 * all unused block_groups. This possibly gives us some more free
1542 * space.
1543 */
1544 btrfs_reclaim_bgs(fs_info);
1545 sleep:
1546 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1547 if (kthread_should_park())
1548 kthread_parkme();
1549 if (kthread_should_stop())
1550 return 0;
1551 if (!again) {
1552 set_current_state(TASK_INTERRUPTIBLE);
1553 schedule();
1554 __set_current_state(TASK_RUNNING);
1555 }
1556 }
1557 }
1558
1559 static int transaction_kthread(void *arg)
1560 {
1561 struct btrfs_root *root = arg;
1562 struct btrfs_fs_info *fs_info = root->fs_info;
1563 struct btrfs_trans_handle *trans;
1564 struct btrfs_transaction *cur;
1565 u64 transid;
1566 time64_t delta;
1567 unsigned long delay;
1568 bool cannot_commit;
1569
1570 do {
1571 cannot_commit = false;
1572 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1573 mutex_lock(&fs_info->transaction_kthread_mutex);
1574
1575 spin_lock(&fs_info->trans_lock);
1576 cur = fs_info->running_transaction;
1577 if (!cur) {
1578 spin_unlock(&fs_info->trans_lock);
1579 goto sleep;
1580 }
1581
1582 delta = ktime_get_seconds() - cur->start_time;
1583 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1584 cur->state < TRANS_STATE_COMMIT_PREP &&
1585 delta < fs_info->commit_interval) {
1586 spin_unlock(&fs_info->trans_lock);
1587 delay -= msecs_to_jiffies((delta - 1) * 1000);
1588 delay = min(delay,
1589 msecs_to_jiffies(fs_info->commit_interval * 1000));
1590 goto sleep;
1591 }
1592 transid = cur->transid;
1593 spin_unlock(&fs_info->trans_lock);
1594
1595 /* If the file system is aborted, this will always fail. */
1596 trans = btrfs_attach_transaction(root);
1597 if (IS_ERR(trans)) {
1598 if (PTR_ERR(trans) != -ENOENT)
1599 cannot_commit = true;
1600 goto sleep;
1601 }
1602 if (transid == trans->transid) {
1603 btrfs_commit_transaction(trans);
1604 } else {
1605 btrfs_end_transaction(trans);
1606 }
1607 sleep:
1608 wake_up_process(fs_info->cleaner_kthread);
1609 mutex_unlock(&fs_info->transaction_kthread_mutex);
1610
1611 if (BTRFS_FS_ERROR(fs_info))
1612 btrfs_cleanup_transaction(fs_info);
1613 if (!kthread_should_stop() &&
1614 (!btrfs_transaction_blocked(fs_info) ||
1615 cannot_commit))
1616 schedule_timeout_interruptible(delay);
1617 } while (!kthread_should_stop());
1618 return 0;
1619 }
1620
1621 /*
1622 * This will find the highest generation in the array of root backups. The
1623 * index of the highest array is returned, or -EINVAL if we can't find
1624 * anything.
1625 *
1626 * We check to make sure the array is valid by comparing the
1627 * generation of the latest root in the array with the generation
1628 * in the super block. If they don't match we pitch it.
1629 */
1630 static int find_newest_super_backup(struct btrfs_fs_info *info)
1631 {
1632 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1633 u64 cur;
1634 struct btrfs_root_backup *root_backup;
1635 int i;
1636
1637 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1638 root_backup = info->super_copy->super_roots + i;
1639 cur = btrfs_backup_tree_root_gen(root_backup);
1640 if (cur == newest_gen)
1641 return i;
1642 }
1643
1644 return -EINVAL;
1645 }
1646
1647 /*
1648 * copy all the root pointers into the super backup array.
1649 * this will bump the backup pointer by one when it is
1650 * done
1651 */
1652 static void backup_super_roots(struct btrfs_fs_info *info)
1653 {
1654 const int next_backup = info->backup_root_index;
1655 struct btrfs_root_backup *root_backup;
1656
1657 root_backup = info->super_for_commit->super_roots + next_backup;
1658
1659 /*
1660 * make sure all of our padding and empty slots get zero filled
1661 * regardless of which ones we use today
1662 */
1663 memset(root_backup, 0, sizeof(*root_backup));
1664
1665 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1666
1667 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1668 btrfs_set_backup_tree_root_gen(root_backup,
1669 btrfs_header_generation(info->tree_root->node));
1670
1671 btrfs_set_backup_tree_root_level(root_backup,
1672 btrfs_header_level(info->tree_root->node));
1673
1674 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1675 btrfs_set_backup_chunk_root_gen(root_backup,
1676 btrfs_header_generation(info->chunk_root->node));
1677 btrfs_set_backup_chunk_root_level(root_backup,
1678 btrfs_header_level(info->chunk_root->node));
1679
1680 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1681 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1682 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1683
1684 btrfs_set_backup_extent_root(root_backup,
1685 extent_root->node->start);
1686 btrfs_set_backup_extent_root_gen(root_backup,
1687 btrfs_header_generation(extent_root->node));
1688 btrfs_set_backup_extent_root_level(root_backup,
1689 btrfs_header_level(extent_root->node));
1690
1691 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1692 btrfs_set_backup_csum_root_gen(root_backup,
1693 btrfs_header_generation(csum_root->node));
1694 btrfs_set_backup_csum_root_level(root_backup,
1695 btrfs_header_level(csum_root->node));
1696 }
1697
1698 /*
1699 * we might commit during log recovery, which happens before we set
1700 * the fs_root. Make sure it is valid before we fill it in.
1701 */
1702 if (info->fs_root && info->fs_root->node) {
1703 btrfs_set_backup_fs_root(root_backup,
1704 info->fs_root->node->start);
1705 btrfs_set_backup_fs_root_gen(root_backup,
1706 btrfs_header_generation(info->fs_root->node));
1707 btrfs_set_backup_fs_root_level(root_backup,
1708 btrfs_header_level(info->fs_root->node));
1709 }
1710
1711 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1712 btrfs_set_backup_dev_root_gen(root_backup,
1713 btrfs_header_generation(info->dev_root->node));
1714 btrfs_set_backup_dev_root_level(root_backup,
1715 btrfs_header_level(info->dev_root->node));
1716
1717 btrfs_set_backup_total_bytes(root_backup,
1718 btrfs_super_total_bytes(info->super_copy));
1719 btrfs_set_backup_bytes_used(root_backup,
1720 btrfs_super_bytes_used(info->super_copy));
1721 btrfs_set_backup_num_devices(root_backup,
1722 btrfs_super_num_devices(info->super_copy));
1723
1724 /*
1725 * if we don't copy this out to the super_copy, it won't get remembered
1726 * for the next commit
1727 */
1728 memcpy(&info->super_copy->super_roots,
1729 &info->super_for_commit->super_roots,
1730 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1731 }
1732
1733 /*
1734 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1735 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1736 *
1737 * @fs_info: filesystem whose backup roots need to be read
1738 * @priority: priority of backup root required
1739 *
1740 * Returns backup root index on success and -EINVAL otherwise.
1741 */
1742 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1743 {
1744 int backup_index = find_newest_super_backup(fs_info);
1745 struct btrfs_super_block *super = fs_info->super_copy;
1746 struct btrfs_root_backup *root_backup;
1747
1748 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1749 if (priority == 0)
1750 return backup_index;
1751
1752 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1753 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1754 } else {
1755 return -EINVAL;
1756 }
1757
1758 root_backup = super->super_roots + backup_index;
1759
1760 btrfs_set_super_generation(super,
1761 btrfs_backup_tree_root_gen(root_backup));
1762 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1763 btrfs_set_super_root_level(super,
1764 btrfs_backup_tree_root_level(root_backup));
1765 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1766
1767 /*
1768 * Fixme: the total bytes and num_devices need to match or we should
1769 * need a fsck
1770 */
1771 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1772 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1773
1774 return backup_index;
1775 }
1776
1777 /* helper to cleanup workers */
1778 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1779 {
1780 btrfs_destroy_workqueue(fs_info->fixup_workers);
1781 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1782 btrfs_destroy_workqueue(fs_info->workers);
1783 if (fs_info->endio_workers)
1784 destroy_workqueue(fs_info->endio_workers);
1785 if (fs_info->rmw_workers)
1786 destroy_workqueue(fs_info->rmw_workers);
1787 if (fs_info->compressed_write_workers)
1788 destroy_workqueue(fs_info->compressed_write_workers);
1789 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1790 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1791 btrfs_destroy_workqueue(fs_info->delayed_workers);
1792 btrfs_destroy_workqueue(fs_info->caching_workers);
1793 btrfs_destroy_workqueue(fs_info->flush_workers);
1794 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1795 if (fs_info->discard_ctl.discard_workers)
1796 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1797 /*
1798 * Now that all other work queues are destroyed, we can safely destroy
1799 * the queues used for metadata I/O, since tasks from those other work
1800 * queues can do metadata I/O operations.
1801 */
1802 if (fs_info->endio_meta_workers)
1803 destroy_workqueue(fs_info->endio_meta_workers);
1804 }
1805
1806 static void free_root_extent_buffers(struct btrfs_root *root)
1807 {
1808 if (root) {
1809 free_extent_buffer(root->node);
1810 free_extent_buffer(root->commit_root);
1811 root->node = NULL;
1812 root->commit_root = NULL;
1813 }
1814 }
1815
1816 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1817 {
1818 struct btrfs_root *root, *tmp;
1819
1820 rbtree_postorder_for_each_entry_safe(root, tmp,
1821 &fs_info->global_root_tree,
1822 rb_node)
1823 free_root_extent_buffers(root);
1824 }
1825
1826 /* helper to cleanup tree roots */
1827 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1828 {
1829 free_root_extent_buffers(info->tree_root);
1830
1831 free_global_root_pointers(info);
1832 free_root_extent_buffers(info->dev_root);
1833 free_root_extent_buffers(info->quota_root);
1834 free_root_extent_buffers(info->uuid_root);
1835 free_root_extent_buffers(info->fs_root);
1836 free_root_extent_buffers(info->data_reloc_root);
1837 free_root_extent_buffers(info->block_group_root);
1838 free_root_extent_buffers(info->stripe_root);
1839 if (free_chunk_root)
1840 free_root_extent_buffers(info->chunk_root);
1841 }
1842
1843 void btrfs_put_root(struct btrfs_root *root)
1844 {
1845 if (!root)
1846 return;
1847
1848 if (refcount_dec_and_test(&root->refs)) {
1849 if (WARN_ON(!xa_empty(&root->inodes)))
1850 xa_destroy(&root->inodes);
1851 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1852 if (root->anon_dev)
1853 free_anon_bdev(root->anon_dev);
1854 free_root_extent_buffers(root);
1855 #ifdef CONFIG_BTRFS_DEBUG
1856 spin_lock(&root->fs_info->fs_roots_radix_lock);
1857 list_del_init(&root->leak_list);
1858 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1859 #endif
1860 kfree(root);
1861 }
1862 }
1863
1864 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1865 {
1866 int ret;
1867 struct btrfs_root *gang[8];
1868 int i;
1869
1870 while (!list_empty(&fs_info->dead_roots)) {
1871 gang[0] = list_entry(fs_info->dead_roots.next,
1872 struct btrfs_root, root_list);
1873 list_del(&gang[0]->root_list);
1874
1875 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1876 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1877 btrfs_put_root(gang[0]);
1878 }
1879
1880 while (1) {
1881 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1882 (void **)gang, 0,
1883 ARRAY_SIZE(gang));
1884 if (!ret)
1885 break;
1886 for (i = 0; i < ret; i++)
1887 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1888 }
1889 }
1890
1891 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1892 {
1893 mutex_init(&fs_info->scrub_lock);
1894 atomic_set(&fs_info->scrubs_running, 0);
1895 atomic_set(&fs_info->scrub_pause_req, 0);
1896 atomic_set(&fs_info->scrubs_paused, 0);
1897 atomic_set(&fs_info->scrub_cancel_req, 0);
1898 init_waitqueue_head(&fs_info->scrub_pause_wait);
1899 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1900 }
1901
1902 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1903 {
1904 spin_lock_init(&fs_info->balance_lock);
1905 mutex_init(&fs_info->balance_mutex);
1906 atomic_set(&fs_info->balance_pause_req, 0);
1907 atomic_set(&fs_info->balance_cancel_req, 0);
1908 fs_info->balance_ctl = NULL;
1909 init_waitqueue_head(&fs_info->balance_wait_q);
1910 atomic_set(&fs_info->reloc_cancel_req, 0);
1911 }
1912
1913 static int btrfs_init_btree_inode(struct super_block *sb)
1914 {
1915 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1916 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1917 fs_info->tree_root);
1918 struct inode *inode;
1919
1920 inode = new_inode(sb);
1921 if (!inode)
1922 return -ENOMEM;
1923
1924 btrfs_set_inode_number(BTRFS_I(inode), BTRFS_BTREE_INODE_OBJECTID);
1925 set_nlink(inode, 1);
1926 /*
1927 * we set the i_size on the btree inode to the max possible int.
1928 * the real end of the address space is determined by all of
1929 * the devices in the system
1930 */
1931 inode->i_size = OFFSET_MAX;
1932 inode->i_mapping->a_ops = &btree_aops;
1933 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1934
1935 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1936 IO_TREE_BTREE_INODE_IO);
1937 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1938
1939 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1940 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1941 __insert_inode_hash(inode, hash);
1942 fs_info->btree_inode = inode;
1943
1944 return 0;
1945 }
1946
1947 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1948 {
1949 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1950 init_rwsem(&fs_info->dev_replace.rwsem);
1951 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1952 }
1953
1954 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1955 {
1956 spin_lock_init(&fs_info->qgroup_lock);
1957 mutex_init(&fs_info->qgroup_ioctl_lock);
1958 fs_info->qgroup_tree = RB_ROOT;
1959 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1960 fs_info->qgroup_seq = 1;
1961 fs_info->qgroup_ulist = NULL;
1962 fs_info->qgroup_rescan_running = false;
1963 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1964 mutex_init(&fs_info->qgroup_rescan_lock);
1965 }
1966
1967 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1968 {
1969 u32 max_active = fs_info->thread_pool_size;
1970 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1971 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1972
1973 fs_info->workers =
1974 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1975
1976 fs_info->delalloc_workers =
1977 btrfs_alloc_workqueue(fs_info, "delalloc",
1978 flags, max_active, 2);
1979
1980 fs_info->flush_workers =
1981 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1982 flags, max_active, 0);
1983
1984 fs_info->caching_workers =
1985 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1986
1987 fs_info->fixup_workers =
1988 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1989
1990 fs_info->endio_workers =
1991 alloc_workqueue("btrfs-endio", flags, max_active);
1992 fs_info->endio_meta_workers =
1993 alloc_workqueue("btrfs-endio-meta", flags, max_active);
1994 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1995 fs_info->endio_write_workers =
1996 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1997 max_active, 2);
1998 fs_info->compressed_write_workers =
1999 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2000 fs_info->endio_freespace_worker =
2001 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2002 max_active, 0);
2003 fs_info->delayed_workers =
2004 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2005 max_active, 0);
2006 fs_info->qgroup_rescan_workers =
2007 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2008 ordered_flags);
2009 fs_info->discard_ctl.discard_workers =
2010 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2011
2012 if (!(fs_info->workers &&
2013 fs_info->delalloc_workers && fs_info->flush_workers &&
2014 fs_info->endio_workers && fs_info->endio_meta_workers &&
2015 fs_info->compressed_write_workers &&
2016 fs_info->endio_write_workers &&
2017 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2018 fs_info->caching_workers && fs_info->fixup_workers &&
2019 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2020 fs_info->discard_ctl.discard_workers)) {
2021 return -ENOMEM;
2022 }
2023
2024 return 0;
2025 }
2026
2027 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2028 {
2029 struct crypto_shash *csum_shash;
2030 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2031
2032 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2033
2034 if (IS_ERR(csum_shash)) {
2035 btrfs_err(fs_info, "error allocating %s hash for checksum",
2036 csum_driver);
2037 return PTR_ERR(csum_shash);
2038 }
2039
2040 fs_info->csum_shash = csum_shash;
2041
2042 /*
2043 * Check if the checksum implementation is a fast accelerated one.
2044 * As-is this is a bit of a hack and should be replaced once the csum
2045 * implementations provide that information themselves.
2046 */
2047 switch (csum_type) {
2048 case BTRFS_CSUM_TYPE_CRC32:
2049 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2050 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2051 break;
2052 case BTRFS_CSUM_TYPE_XXHASH:
2053 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2054 break;
2055 default:
2056 break;
2057 }
2058
2059 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2060 btrfs_super_csum_name(csum_type),
2061 crypto_shash_driver_name(csum_shash));
2062 return 0;
2063 }
2064
2065 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2066 struct btrfs_fs_devices *fs_devices)
2067 {
2068 int ret;
2069 struct btrfs_tree_parent_check check = { 0 };
2070 struct btrfs_root *log_tree_root;
2071 struct btrfs_super_block *disk_super = fs_info->super_copy;
2072 u64 bytenr = btrfs_super_log_root(disk_super);
2073 int level = btrfs_super_log_root_level(disk_super);
2074
2075 if (fs_devices->rw_devices == 0) {
2076 btrfs_warn(fs_info, "log replay required on RO media");
2077 return -EIO;
2078 }
2079
2080 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2081 GFP_KERNEL);
2082 if (!log_tree_root)
2083 return -ENOMEM;
2084
2085 check.level = level;
2086 check.transid = fs_info->generation + 1;
2087 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2088 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2089 if (IS_ERR(log_tree_root->node)) {
2090 btrfs_warn(fs_info, "failed to read log tree");
2091 ret = PTR_ERR(log_tree_root->node);
2092 log_tree_root->node = NULL;
2093 btrfs_put_root(log_tree_root);
2094 return ret;
2095 }
2096 if (!extent_buffer_uptodate(log_tree_root->node)) {
2097 btrfs_err(fs_info, "failed to read log tree");
2098 btrfs_put_root(log_tree_root);
2099 return -EIO;
2100 }
2101
2102 /* returns with log_tree_root freed on success */
2103 ret = btrfs_recover_log_trees(log_tree_root);
2104 if (ret) {
2105 btrfs_handle_fs_error(fs_info, ret,
2106 "Failed to recover log tree");
2107 btrfs_put_root(log_tree_root);
2108 return ret;
2109 }
2110
2111 if (sb_rdonly(fs_info->sb)) {
2112 ret = btrfs_commit_super(fs_info);
2113 if (ret)
2114 return ret;
2115 }
2116
2117 return 0;
2118 }
2119
2120 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2121 struct btrfs_path *path, u64 objectid,
2122 const char *name)
2123 {
2124 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2125 struct btrfs_root *root;
2126 u64 max_global_id = 0;
2127 int ret;
2128 struct btrfs_key key = {
2129 .objectid = objectid,
2130 .type = BTRFS_ROOT_ITEM_KEY,
2131 .offset = 0,
2132 };
2133 bool found = false;
2134
2135 /* If we have IGNOREDATACSUMS skip loading these roots. */
2136 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2137 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2138 set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2139 return 0;
2140 }
2141
2142 while (1) {
2143 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2144 if (ret < 0)
2145 break;
2146
2147 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2148 ret = btrfs_next_leaf(tree_root, path);
2149 if (ret) {
2150 if (ret > 0)
2151 ret = 0;
2152 break;
2153 }
2154 }
2155 ret = 0;
2156
2157 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2158 if (key.objectid != objectid)
2159 break;
2160 btrfs_release_path(path);
2161
2162 /*
2163 * Just worry about this for extent tree, it'll be the same for
2164 * everybody.
2165 */
2166 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2167 max_global_id = max(max_global_id, key.offset);
2168
2169 found = true;
2170 root = read_tree_root_path(tree_root, path, &key);
2171 if (IS_ERR(root)) {
2172 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2173 ret = PTR_ERR(root);
2174 break;
2175 }
2176 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2177 ret = btrfs_global_root_insert(root);
2178 if (ret) {
2179 btrfs_put_root(root);
2180 break;
2181 }
2182 key.offset++;
2183 }
2184 btrfs_release_path(path);
2185
2186 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2187 fs_info->nr_global_roots = max_global_id + 1;
2188
2189 if (!found || ret) {
2190 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2191 set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2192
2193 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2194 ret = ret ? ret : -ENOENT;
2195 else
2196 ret = 0;
2197 btrfs_err(fs_info, "failed to load root %s", name);
2198 }
2199 return ret;
2200 }
2201
2202 static int load_global_roots(struct btrfs_root *tree_root)
2203 {
2204 struct btrfs_path *path;
2205 int ret = 0;
2206
2207 path = btrfs_alloc_path();
2208 if (!path)
2209 return -ENOMEM;
2210
2211 ret = load_global_roots_objectid(tree_root, path,
2212 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2213 if (ret)
2214 goto out;
2215 ret = load_global_roots_objectid(tree_root, path,
2216 BTRFS_CSUM_TREE_OBJECTID, "csum");
2217 if (ret)
2218 goto out;
2219 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2220 goto out;
2221 ret = load_global_roots_objectid(tree_root, path,
2222 BTRFS_FREE_SPACE_TREE_OBJECTID,
2223 "free space");
2224 out:
2225 btrfs_free_path(path);
2226 return ret;
2227 }
2228
2229 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2230 {
2231 struct btrfs_root *tree_root = fs_info->tree_root;
2232 struct btrfs_root *root;
2233 struct btrfs_key location;
2234 int ret;
2235
2236 ASSERT(fs_info->tree_root);
2237
2238 ret = load_global_roots(tree_root);
2239 if (ret)
2240 return ret;
2241
2242 location.type = BTRFS_ROOT_ITEM_KEY;
2243 location.offset = 0;
2244
2245 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2246 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2247 root = btrfs_read_tree_root(tree_root, &location);
2248 if (IS_ERR(root)) {
2249 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2250 ret = PTR_ERR(root);
2251 goto out;
2252 }
2253 } else {
2254 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2255 fs_info->block_group_root = root;
2256 }
2257 }
2258
2259 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2260 root = btrfs_read_tree_root(tree_root, &location);
2261 if (IS_ERR(root)) {
2262 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2263 ret = PTR_ERR(root);
2264 goto out;
2265 }
2266 } else {
2267 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2268 fs_info->dev_root = root;
2269 }
2270 /* Initialize fs_info for all devices in any case */
2271 ret = btrfs_init_devices_late(fs_info);
2272 if (ret)
2273 goto out;
2274
2275 /*
2276 * This tree can share blocks with some other fs tree during relocation
2277 * and we need a proper setup by btrfs_get_fs_root
2278 */
2279 root = btrfs_get_fs_root(tree_root->fs_info,
2280 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2281 if (IS_ERR(root)) {
2282 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2283 ret = PTR_ERR(root);
2284 goto out;
2285 }
2286 } else {
2287 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2288 fs_info->data_reloc_root = root;
2289 }
2290
2291 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2292 root = btrfs_read_tree_root(tree_root, &location);
2293 if (!IS_ERR(root)) {
2294 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2295 fs_info->quota_root = root;
2296 }
2297
2298 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2299 root = btrfs_read_tree_root(tree_root, &location);
2300 if (IS_ERR(root)) {
2301 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2302 ret = PTR_ERR(root);
2303 if (ret != -ENOENT)
2304 goto out;
2305 }
2306 } else {
2307 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2308 fs_info->uuid_root = root;
2309 }
2310
2311 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2312 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2313 root = btrfs_read_tree_root(tree_root, &location);
2314 if (IS_ERR(root)) {
2315 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2316 ret = PTR_ERR(root);
2317 goto out;
2318 }
2319 } else {
2320 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2321 fs_info->stripe_root = root;
2322 }
2323 }
2324
2325 return 0;
2326 out:
2327 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2328 location.objectid, ret);
2329 return ret;
2330 }
2331
2332 /*
2333 * Real super block validation
2334 * NOTE: super csum type and incompat features will not be checked here.
2335 *
2336 * @sb: super block to check
2337 * @mirror_num: the super block number to check its bytenr:
2338 * 0 the primary (1st) sb
2339 * 1, 2 2nd and 3rd backup copy
2340 * -1 skip bytenr check
2341 */
2342 int btrfs_validate_super(const struct btrfs_fs_info *fs_info,
2343 const struct btrfs_super_block *sb, int mirror_num)
2344 {
2345 u64 nodesize = btrfs_super_nodesize(sb);
2346 u64 sectorsize = btrfs_super_sectorsize(sb);
2347 int ret = 0;
2348 const bool ignore_flags = btrfs_test_opt(fs_info, IGNORESUPERFLAGS);
2349
2350 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2351 btrfs_err(fs_info, "no valid FS found");
2352 ret = -EINVAL;
2353 }
2354 if ((btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)) {
2355 if (!ignore_flags) {
2356 btrfs_err(fs_info,
2357 "unrecognized or unsupported super flag 0x%llx",
2358 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2359 ret = -EINVAL;
2360 } else {
2361 btrfs_info(fs_info,
2362 "unrecognized or unsupported super flags: 0x%llx, ignored",
2363 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2364 }
2365 }
2366 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2367 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2368 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2369 ret = -EINVAL;
2370 }
2371 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2372 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2373 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2374 ret = -EINVAL;
2375 }
2376 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2377 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2378 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2379 ret = -EINVAL;
2380 }
2381
2382 /*
2383 * Check sectorsize and nodesize first, other check will need it.
2384 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2385 */
2386 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2387 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2388 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2389 ret = -EINVAL;
2390 }
2391
2392 /*
2393 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2394 *
2395 * We can support 16K sectorsize with 64K page size without problem,
2396 * but such sectorsize/pagesize combination doesn't make much sense.
2397 * 4K will be our future standard, PAGE_SIZE is supported from the very
2398 * beginning.
2399 */
2400 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2401 btrfs_err(fs_info,
2402 "sectorsize %llu not yet supported for page size %lu",
2403 sectorsize, PAGE_SIZE);
2404 ret = -EINVAL;
2405 }
2406
2407 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2408 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2409 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2410 ret = -EINVAL;
2411 }
2412 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2413 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2414 le32_to_cpu(sb->__unused_leafsize), nodesize);
2415 ret = -EINVAL;
2416 }
2417
2418 /* Root alignment check */
2419 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2420 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2421 btrfs_super_root(sb));
2422 ret = -EINVAL;
2423 }
2424 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2425 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2426 btrfs_super_chunk_root(sb));
2427 ret = -EINVAL;
2428 }
2429 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2430 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2431 btrfs_super_log_root(sb));
2432 ret = -EINVAL;
2433 }
2434
2435 if (!fs_info->fs_devices->temp_fsid &&
2436 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2437 btrfs_err(fs_info,
2438 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2439 sb->fsid, fs_info->fs_devices->fsid);
2440 ret = -EINVAL;
2441 }
2442
2443 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2444 BTRFS_FSID_SIZE) != 0) {
2445 btrfs_err(fs_info,
2446 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2447 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2448 ret = -EINVAL;
2449 }
2450
2451 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2452 BTRFS_FSID_SIZE) != 0) {
2453 btrfs_err(fs_info,
2454 "dev_item UUID does not match metadata fsid: %pU != %pU",
2455 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2456 ret = -EINVAL;
2457 }
2458
2459 /*
2460 * Artificial requirement for block-group-tree to force newer features
2461 * (free-space-tree, no-holes) so the test matrix is smaller.
2462 */
2463 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2464 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2465 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2466 btrfs_err(fs_info,
2467 "block-group-tree feature requires free-space-tree and no-holes");
2468 ret = -EINVAL;
2469 }
2470
2471 /*
2472 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2473 * done later
2474 */
2475 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2476 btrfs_err(fs_info, "bytes_used is too small %llu",
2477 btrfs_super_bytes_used(sb));
2478 ret = -EINVAL;
2479 }
2480 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2481 btrfs_err(fs_info, "invalid stripesize %u",
2482 btrfs_super_stripesize(sb));
2483 ret = -EINVAL;
2484 }
2485 if (btrfs_super_num_devices(sb) > (1UL << 31))
2486 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2487 btrfs_super_num_devices(sb));
2488 if (btrfs_super_num_devices(sb) == 0) {
2489 btrfs_err(fs_info, "number of devices is 0");
2490 ret = -EINVAL;
2491 }
2492
2493 if (mirror_num >= 0 &&
2494 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2495 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2496 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2497 ret = -EINVAL;
2498 }
2499
2500 /*
2501 * Obvious sys_chunk_array corruptions, it must hold at least one key
2502 * and one chunk
2503 */
2504 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2505 btrfs_err(fs_info, "system chunk array too big %u > %u",
2506 btrfs_super_sys_array_size(sb),
2507 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2508 ret = -EINVAL;
2509 }
2510 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2511 + sizeof(struct btrfs_chunk)) {
2512 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2513 btrfs_super_sys_array_size(sb),
2514 sizeof(struct btrfs_disk_key)
2515 + sizeof(struct btrfs_chunk));
2516 ret = -EINVAL;
2517 }
2518
2519 /*
2520 * The generation is a global counter, we'll trust it more than the others
2521 * but it's still possible that it's the one that's wrong.
2522 */
2523 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2524 btrfs_warn(fs_info,
2525 "suspicious: generation < chunk_root_generation: %llu < %llu",
2526 btrfs_super_generation(sb),
2527 btrfs_super_chunk_root_generation(sb));
2528 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2529 && btrfs_super_cache_generation(sb) != (u64)-1)
2530 btrfs_warn(fs_info,
2531 "suspicious: generation < cache_generation: %llu < %llu",
2532 btrfs_super_generation(sb),
2533 btrfs_super_cache_generation(sb));
2534
2535 return ret;
2536 }
2537
2538 /*
2539 * Validation of super block at mount time.
2540 * Some checks already done early at mount time, like csum type and incompat
2541 * flags will be skipped.
2542 */
2543 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2544 {
2545 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2546 }
2547
2548 /*
2549 * Validation of super block at write time.
2550 * Some checks like bytenr check will be skipped as their values will be
2551 * overwritten soon.
2552 * Extra checks like csum type and incompat flags will be done here.
2553 */
2554 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2555 struct btrfs_super_block *sb)
2556 {
2557 int ret;
2558
2559 ret = btrfs_validate_super(fs_info, sb, -1);
2560 if (ret < 0)
2561 goto out;
2562 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2563 ret = -EUCLEAN;
2564 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2565 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2566 goto out;
2567 }
2568 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2569 ret = -EUCLEAN;
2570 btrfs_err(fs_info,
2571 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2572 btrfs_super_incompat_flags(sb),
2573 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2574 goto out;
2575 }
2576 out:
2577 if (ret < 0)
2578 btrfs_err(fs_info,
2579 "super block corruption detected before writing it to disk");
2580 return ret;
2581 }
2582
2583 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2584 {
2585 struct btrfs_tree_parent_check check = {
2586 .level = level,
2587 .transid = gen,
2588 .owner_root = btrfs_root_id(root)
2589 };
2590 int ret = 0;
2591
2592 root->node = read_tree_block(root->fs_info, bytenr, &check);
2593 if (IS_ERR(root->node)) {
2594 ret = PTR_ERR(root->node);
2595 root->node = NULL;
2596 return ret;
2597 }
2598 if (!extent_buffer_uptodate(root->node)) {
2599 free_extent_buffer(root->node);
2600 root->node = NULL;
2601 return -EIO;
2602 }
2603
2604 btrfs_set_root_node(&root->root_item, root->node);
2605 root->commit_root = btrfs_root_node(root);
2606 btrfs_set_root_refs(&root->root_item, 1);
2607 return ret;
2608 }
2609
2610 static int load_important_roots(struct btrfs_fs_info *fs_info)
2611 {
2612 struct btrfs_super_block *sb = fs_info->super_copy;
2613 u64 gen, bytenr;
2614 int level, ret;
2615
2616 bytenr = btrfs_super_root(sb);
2617 gen = btrfs_super_generation(sb);
2618 level = btrfs_super_root_level(sb);
2619 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2620 if (ret) {
2621 btrfs_warn(fs_info, "couldn't read tree root");
2622 return ret;
2623 }
2624 return 0;
2625 }
2626
2627 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2628 {
2629 int backup_index = find_newest_super_backup(fs_info);
2630 struct btrfs_super_block *sb = fs_info->super_copy;
2631 struct btrfs_root *tree_root = fs_info->tree_root;
2632 bool handle_error = false;
2633 int ret = 0;
2634 int i;
2635
2636 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2637 if (handle_error) {
2638 if (!IS_ERR(tree_root->node))
2639 free_extent_buffer(tree_root->node);
2640 tree_root->node = NULL;
2641
2642 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2643 break;
2644
2645 free_root_pointers(fs_info, 0);
2646
2647 /*
2648 * Don't use the log in recovery mode, it won't be
2649 * valid
2650 */
2651 btrfs_set_super_log_root(sb, 0);
2652
2653 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2654 ret = read_backup_root(fs_info, i);
2655 backup_index = ret;
2656 if (ret < 0)
2657 return ret;
2658 }
2659
2660 ret = load_important_roots(fs_info);
2661 if (ret) {
2662 handle_error = true;
2663 continue;
2664 }
2665
2666 /*
2667 * No need to hold btrfs_root::objectid_mutex since the fs
2668 * hasn't been fully initialised and we are the only user
2669 */
2670 ret = btrfs_init_root_free_objectid(tree_root);
2671 if (ret < 0) {
2672 handle_error = true;
2673 continue;
2674 }
2675
2676 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2677
2678 ret = btrfs_read_roots(fs_info);
2679 if (ret < 0) {
2680 handle_error = true;
2681 continue;
2682 }
2683
2684 /* All successful */
2685 fs_info->generation = btrfs_header_generation(tree_root->node);
2686 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2687 fs_info->last_reloc_trans = 0;
2688
2689 /* Always begin writing backup roots after the one being used */
2690 if (backup_index < 0) {
2691 fs_info->backup_root_index = 0;
2692 } else {
2693 fs_info->backup_root_index = backup_index + 1;
2694 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2695 }
2696 break;
2697 }
2698
2699 return ret;
2700 }
2701
2702 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2703 {
2704 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2705 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2706 INIT_LIST_HEAD(&fs_info->trans_list);
2707 INIT_LIST_HEAD(&fs_info->dead_roots);
2708 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2709 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2710 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2711 spin_lock_init(&fs_info->delalloc_root_lock);
2712 spin_lock_init(&fs_info->trans_lock);
2713 spin_lock_init(&fs_info->fs_roots_radix_lock);
2714 spin_lock_init(&fs_info->delayed_iput_lock);
2715 spin_lock_init(&fs_info->defrag_inodes_lock);
2716 spin_lock_init(&fs_info->super_lock);
2717 spin_lock_init(&fs_info->buffer_lock);
2718 spin_lock_init(&fs_info->unused_bgs_lock);
2719 spin_lock_init(&fs_info->treelog_bg_lock);
2720 spin_lock_init(&fs_info->zone_active_bgs_lock);
2721 spin_lock_init(&fs_info->relocation_bg_lock);
2722 rwlock_init(&fs_info->tree_mod_log_lock);
2723 rwlock_init(&fs_info->global_root_lock);
2724 mutex_init(&fs_info->unused_bg_unpin_mutex);
2725 mutex_init(&fs_info->reclaim_bgs_lock);
2726 mutex_init(&fs_info->reloc_mutex);
2727 mutex_init(&fs_info->delalloc_root_mutex);
2728 mutex_init(&fs_info->zoned_meta_io_lock);
2729 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2730 seqlock_init(&fs_info->profiles_lock);
2731
2732 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2733 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2734 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2735 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2736 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2737 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2738 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2739 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2740 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2741 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2742 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2743 BTRFS_LOCKDEP_TRANS_COMPLETED);
2744
2745 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2746 INIT_LIST_HEAD(&fs_info->space_info);
2747 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2748 INIT_LIST_HEAD(&fs_info->unused_bgs);
2749 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2750 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2751 #ifdef CONFIG_BTRFS_DEBUG
2752 INIT_LIST_HEAD(&fs_info->allocated_roots);
2753 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2754 spin_lock_init(&fs_info->eb_leak_lock);
2755 #endif
2756 fs_info->mapping_tree = RB_ROOT_CACHED;
2757 rwlock_init(&fs_info->mapping_tree_lock);
2758 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2759 BTRFS_BLOCK_RSV_GLOBAL);
2760 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2761 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2762 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2763 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2764 BTRFS_BLOCK_RSV_DELOPS);
2765 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2766 BTRFS_BLOCK_RSV_DELREFS);
2767
2768 atomic_set(&fs_info->async_delalloc_pages, 0);
2769 atomic_set(&fs_info->defrag_running, 0);
2770 atomic_set(&fs_info->nr_delayed_iputs, 0);
2771 atomic64_set(&fs_info->tree_mod_seq, 0);
2772 fs_info->global_root_tree = RB_ROOT;
2773 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2774 fs_info->metadata_ratio = 0;
2775 fs_info->defrag_inodes = RB_ROOT;
2776 atomic64_set(&fs_info->free_chunk_space, 0);
2777 fs_info->tree_mod_log = RB_ROOT;
2778 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2779 btrfs_init_ref_verify(fs_info);
2780
2781 fs_info->thread_pool_size = min_t(unsigned long,
2782 num_online_cpus() + 2, 8);
2783
2784 INIT_LIST_HEAD(&fs_info->ordered_roots);
2785 spin_lock_init(&fs_info->ordered_root_lock);
2786
2787 btrfs_init_scrub(fs_info);
2788 btrfs_init_balance(fs_info);
2789 btrfs_init_async_reclaim_work(fs_info);
2790
2791 rwlock_init(&fs_info->block_group_cache_lock);
2792 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2793
2794 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2795 IO_TREE_FS_EXCLUDED_EXTENTS);
2796
2797 mutex_init(&fs_info->ordered_operations_mutex);
2798 mutex_init(&fs_info->tree_log_mutex);
2799 mutex_init(&fs_info->chunk_mutex);
2800 mutex_init(&fs_info->transaction_kthread_mutex);
2801 mutex_init(&fs_info->cleaner_mutex);
2802 mutex_init(&fs_info->ro_block_group_mutex);
2803 init_rwsem(&fs_info->commit_root_sem);
2804 init_rwsem(&fs_info->cleanup_work_sem);
2805 init_rwsem(&fs_info->subvol_sem);
2806 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2807
2808 btrfs_init_dev_replace_locks(fs_info);
2809 btrfs_init_qgroup(fs_info);
2810 btrfs_discard_init(fs_info);
2811
2812 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2813 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2814
2815 init_waitqueue_head(&fs_info->transaction_throttle);
2816 init_waitqueue_head(&fs_info->transaction_wait);
2817 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2818 init_waitqueue_head(&fs_info->async_submit_wait);
2819 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2820
2821 /* Usable values until the real ones are cached from the superblock */
2822 fs_info->nodesize = 4096;
2823 fs_info->sectorsize = 4096;
2824 fs_info->sectorsize_bits = ilog2(4096);
2825 fs_info->stripesize = 4096;
2826
2827 /* Default compress algorithm when user does -o compress */
2828 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2829
2830 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2831
2832 spin_lock_init(&fs_info->swapfile_pins_lock);
2833 fs_info->swapfile_pins = RB_ROOT;
2834
2835 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2836 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2837 }
2838
2839 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2840 {
2841 int ret;
2842
2843 fs_info->sb = sb;
2844 /* Temporary fixed values for block size until we read the superblock. */
2845 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2846 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2847
2848 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2849 if (ret)
2850 return ret;
2851
2852 ret = percpu_counter_init(&fs_info->evictable_extent_maps, 0, GFP_KERNEL);
2853 if (ret)
2854 return ret;
2855
2856 spin_lock_init(&fs_info->extent_map_shrinker_lock);
2857
2858 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2859 if (ret)
2860 return ret;
2861
2862 fs_info->dirty_metadata_batch = PAGE_SIZE *
2863 (1 + ilog2(nr_cpu_ids));
2864
2865 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2866 if (ret)
2867 return ret;
2868
2869 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2870 GFP_KERNEL);
2871 if (ret)
2872 return ret;
2873
2874 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2875 GFP_KERNEL);
2876 if (!fs_info->delayed_root)
2877 return -ENOMEM;
2878 btrfs_init_delayed_root(fs_info->delayed_root);
2879
2880 if (sb_rdonly(sb))
2881 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2882 if (btrfs_test_opt(fs_info, IGNOREMETACSUMS))
2883 set_bit(BTRFS_FS_STATE_SKIP_META_CSUMS, &fs_info->fs_state);
2884
2885 return btrfs_alloc_stripe_hash_table(fs_info);
2886 }
2887
2888 static int btrfs_uuid_rescan_kthread(void *data)
2889 {
2890 struct btrfs_fs_info *fs_info = data;
2891 int ret;
2892
2893 /*
2894 * 1st step is to iterate through the existing UUID tree and
2895 * to delete all entries that contain outdated data.
2896 * 2nd step is to add all missing entries to the UUID tree.
2897 */
2898 ret = btrfs_uuid_tree_iterate(fs_info);
2899 if (ret < 0) {
2900 if (ret != -EINTR)
2901 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2902 ret);
2903 up(&fs_info->uuid_tree_rescan_sem);
2904 return ret;
2905 }
2906 return btrfs_uuid_scan_kthread(data);
2907 }
2908
2909 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2910 {
2911 struct task_struct *task;
2912
2913 down(&fs_info->uuid_tree_rescan_sem);
2914 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2915 if (IS_ERR(task)) {
2916 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2917 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2918 up(&fs_info->uuid_tree_rescan_sem);
2919 return PTR_ERR(task);
2920 }
2921
2922 return 0;
2923 }
2924
2925 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2926 {
2927 u64 root_objectid = 0;
2928 struct btrfs_root *gang[8];
2929 int ret = 0;
2930
2931 while (1) {
2932 unsigned int found;
2933
2934 spin_lock(&fs_info->fs_roots_radix_lock);
2935 found = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2936 (void **)gang, root_objectid,
2937 ARRAY_SIZE(gang));
2938 if (!found) {
2939 spin_unlock(&fs_info->fs_roots_radix_lock);
2940 break;
2941 }
2942 root_objectid = btrfs_root_id(gang[found - 1]) + 1;
2943
2944 for (int i = 0; i < found; i++) {
2945 /* Avoid to grab roots in dead_roots. */
2946 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2947 gang[i] = NULL;
2948 continue;
2949 }
2950 /* Grab all the search result for later use. */
2951 gang[i] = btrfs_grab_root(gang[i]);
2952 }
2953 spin_unlock(&fs_info->fs_roots_radix_lock);
2954
2955 for (int i = 0; i < found; i++) {
2956 if (!gang[i])
2957 continue;
2958 root_objectid = btrfs_root_id(gang[i]);
2959 /*
2960 * Continue to release the remaining roots after the first
2961 * error without cleanup and preserve the first error
2962 * for the return.
2963 */
2964 if (!ret)
2965 ret = btrfs_orphan_cleanup(gang[i]);
2966 btrfs_put_root(gang[i]);
2967 }
2968 if (ret)
2969 break;
2970
2971 root_objectid++;
2972 }
2973 return ret;
2974 }
2975
2976 /*
2977 * Mounting logic specific to read-write file systems. Shared by open_ctree
2978 * and btrfs_remount when remounting from read-only to read-write.
2979 */
2980 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2981 {
2982 int ret;
2983 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2984 bool rebuild_free_space_tree = false;
2985
2986 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2987 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2988 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2989 btrfs_warn(fs_info,
2990 "'clear_cache' option is ignored with extent tree v2");
2991 else
2992 rebuild_free_space_tree = true;
2993 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2994 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2995 btrfs_warn(fs_info, "free space tree is invalid");
2996 rebuild_free_space_tree = true;
2997 }
2998
2999 if (rebuild_free_space_tree) {
3000 btrfs_info(fs_info, "rebuilding free space tree");
3001 ret = btrfs_rebuild_free_space_tree(fs_info);
3002 if (ret) {
3003 btrfs_warn(fs_info,
3004 "failed to rebuild free space tree: %d", ret);
3005 goto out;
3006 }
3007 }
3008
3009 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3010 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3011 btrfs_info(fs_info, "disabling free space tree");
3012 ret = btrfs_delete_free_space_tree(fs_info);
3013 if (ret) {
3014 btrfs_warn(fs_info,
3015 "failed to disable free space tree: %d", ret);
3016 goto out;
3017 }
3018 }
3019
3020 /*
3021 * btrfs_find_orphan_roots() is responsible for finding all the dead
3022 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3023 * them into the fs_info->fs_roots_radix tree. This must be done before
3024 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3025 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3026 * item before the root's tree is deleted - this means that if we unmount
3027 * or crash before the deletion completes, on the next mount we will not
3028 * delete what remains of the tree because the orphan item does not
3029 * exists anymore, which is what tells us we have a pending deletion.
3030 */
3031 ret = btrfs_find_orphan_roots(fs_info);
3032 if (ret)
3033 goto out;
3034
3035 ret = btrfs_cleanup_fs_roots(fs_info);
3036 if (ret)
3037 goto out;
3038
3039 down_read(&fs_info->cleanup_work_sem);
3040 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3041 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3042 up_read(&fs_info->cleanup_work_sem);
3043 goto out;
3044 }
3045 up_read(&fs_info->cleanup_work_sem);
3046
3047 mutex_lock(&fs_info->cleaner_mutex);
3048 ret = btrfs_recover_relocation(fs_info);
3049 mutex_unlock(&fs_info->cleaner_mutex);
3050 if (ret < 0) {
3051 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3052 goto out;
3053 }
3054
3055 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3056 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3057 btrfs_info(fs_info, "creating free space tree");
3058 ret = btrfs_create_free_space_tree(fs_info);
3059 if (ret) {
3060 btrfs_warn(fs_info,
3061 "failed to create free space tree: %d", ret);
3062 goto out;
3063 }
3064 }
3065
3066 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3067 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3068 if (ret)
3069 goto out;
3070 }
3071
3072 ret = btrfs_resume_balance_async(fs_info);
3073 if (ret)
3074 goto out;
3075
3076 ret = btrfs_resume_dev_replace_async(fs_info);
3077 if (ret) {
3078 btrfs_warn(fs_info, "failed to resume dev_replace");
3079 goto out;
3080 }
3081
3082 btrfs_qgroup_rescan_resume(fs_info);
3083
3084 if (!fs_info->uuid_root) {
3085 btrfs_info(fs_info, "creating UUID tree");
3086 ret = btrfs_create_uuid_tree(fs_info);
3087 if (ret) {
3088 btrfs_warn(fs_info,
3089 "failed to create the UUID tree %d", ret);
3090 goto out;
3091 }
3092 }
3093
3094 out:
3095 return ret;
3096 }
3097
3098 /*
3099 * Do various sanity and dependency checks of different features.
3100 *
3101 * @is_rw_mount: If the mount is read-write.
3102 *
3103 * This is the place for less strict checks (like for subpage or artificial
3104 * feature dependencies).
3105 *
3106 * For strict checks or possible corruption detection, see
3107 * btrfs_validate_super().
3108 *
3109 * This should be called after btrfs_parse_options(), as some mount options
3110 * (space cache related) can modify on-disk format like free space tree and
3111 * screw up certain feature dependencies.
3112 */
3113 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3114 {
3115 struct btrfs_super_block *disk_super = fs_info->super_copy;
3116 u64 incompat = btrfs_super_incompat_flags(disk_super);
3117 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3118 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3119
3120 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3121 btrfs_err(fs_info,
3122 "cannot mount because of unknown incompat features (0x%llx)",
3123 incompat);
3124 return -EINVAL;
3125 }
3126
3127 /* Runtime limitation for mixed block groups. */
3128 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3129 (fs_info->sectorsize != fs_info->nodesize)) {
3130 btrfs_err(fs_info,
3131 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3132 fs_info->nodesize, fs_info->sectorsize);
3133 return -EINVAL;
3134 }
3135
3136 /* Mixed backref is an always-enabled feature. */
3137 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3138
3139 /* Set compression related flags just in case. */
3140 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3141 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3142 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3143 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3144
3145 /*
3146 * An ancient flag, which should really be marked deprecated.
3147 * Such runtime limitation doesn't really need a incompat flag.
3148 */
3149 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3150 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3151
3152 if (compat_ro_unsupp && is_rw_mount) {
3153 btrfs_err(fs_info,
3154 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3155 compat_ro);
3156 return -EINVAL;
3157 }
3158
3159 /*
3160 * We have unsupported RO compat features, although RO mounted, we
3161 * should not cause any metadata writes, including log replay.
3162 * Or we could screw up whatever the new feature requires.
3163 */
3164 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3165 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3166 btrfs_err(fs_info,
3167 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3168 compat_ro);
3169 return -EINVAL;
3170 }
3171
3172 /*
3173 * Artificial limitations for block group tree, to force
3174 * block-group-tree to rely on no-holes and free-space-tree.
3175 */
3176 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3177 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3178 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3179 btrfs_err(fs_info,
3180 "block-group-tree feature requires no-holes and free-space-tree features");
3181 return -EINVAL;
3182 }
3183
3184 /*
3185 * Subpage runtime limitation on v1 cache.
3186 *
3187 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3188 * we're already defaulting to v2 cache, no need to bother v1 as it's
3189 * going to be deprecated anyway.
3190 */
3191 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3192 btrfs_warn(fs_info,
3193 "v1 space cache is not supported for page size %lu with sectorsize %u",
3194 PAGE_SIZE, fs_info->sectorsize);
3195 return -EINVAL;
3196 }
3197
3198 /* This can be called by remount, we need to protect the super block. */
3199 spin_lock(&fs_info->super_lock);
3200 btrfs_set_super_incompat_flags(disk_super, incompat);
3201 spin_unlock(&fs_info->super_lock);
3202
3203 return 0;
3204 }
3205
3206 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3207 const char *options)
3208 {
3209 u32 sectorsize;
3210 u32 nodesize;
3211 u32 stripesize;
3212 u64 generation;
3213 u16 csum_type;
3214 struct btrfs_super_block *disk_super;
3215 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3216 struct btrfs_root *tree_root;
3217 struct btrfs_root *chunk_root;
3218 int ret;
3219 int level;
3220
3221 ret = init_mount_fs_info(fs_info, sb);
3222 if (ret)
3223 goto fail;
3224
3225 /* These need to be init'ed before we start creating inodes and such. */
3226 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3227 GFP_KERNEL);
3228 fs_info->tree_root = tree_root;
3229 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3230 GFP_KERNEL);
3231 fs_info->chunk_root = chunk_root;
3232 if (!tree_root || !chunk_root) {
3233 ret = -ENOMEM;
3234 goto fail;
3235 }
3236
3237 ret = btrfs_init_btree_inode(sb);
3238 if (ret)
3239 goto fail;
3240
3241 invalidate_bdev(fs_devices->latest_dev->bdev);
3242
3243 /*
3244 * Read super block and check the signature bytes only
3245 */
3246 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3247 if (IS_ERR(disk_super)) {
3248 ret = PTR_ERR(disk_super);
3249 goto fail_alloc;
3250 }
3251
3252 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3253 /*
3254 * Verify the type first, if that or the checksum value are
3255 * corrupted, we'll find out
3256 */
3257 csum_type = btrfs_super_csum_type(disk_super);
3258 if (!btrfs_supported_super_csum(csum_type)) {
3259 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3260 csum_type);
3261 ret = -EINVAL;
3262 btrfs_release_disk_super(disk_super);
3263 goto fail_alloc;
3264 }
3265
3266 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3267
3268 ret = btrfs_init_csum_hash(fs_info, csum_type);
3269 if (ret) {
3270 btrfs_release_disk_super(disk_super);
3271 goto fail_alloc;
3272 }
3273
3274 /*
3275 * We want to check superblock checksum, the type is stored inside.
3276 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3277 */
3278 if (btrfs_check_super_csum(fs_info, disk_super)) {
3279 btrfs_err(fs_info, "superblock checksum mismatch");
3280 ret = -EINVAL;
3281 btrfs_release_disk_super(disk_super);
3282 goto fail_alloc;
3283 }
3284
3285 /*
3286 * super_copy is zeroed at allocation time and we never touch the
3287 * following bytes up to INFO_SIZE, the checksum is calculated from
3288 * the whole block of INFO_SIZE
3289 */
3290 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3291 btrfs_release_disk_super(disk_super);
3292
3293 disk_super = fs_info->super_copy;
3294
3295 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3296 sizeof(*fs_info->super_for_commit));
3297
3298 ret = btrfs_validate_mount_super(fs_info);
3299 if (ret) {
3300 btrfs_err(fs_info, "superblock contains fatal errors");
3301 ret = -EINVAL;
3302 goto fail_alloc;
3303 }
3304
3305 if (!btrfs_super_root(disk_super)) {
3306 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3307 ret = -EINVAL;
3308 goto fail_alloc;
3309 }
3310
3311 /* check FS state, whether FS is broken. */
3312 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3313 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3314
3315 /* Set up fs_info before parsing mount options */
3316 nodesize = btrfs_super_nodesize(disk_super);
3317 sectorsize = btrfs_super_sectorsize(disk_super);
3318 stripesize = sectorsize;
3319 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3320 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3321
3322 fs_info->nodesize = nodesize;
3323 fs_info->sectorsize = sectorsize;
3324 fs_info->sectorsize_bits = ilog2(sectorsize);
3325 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3326 fs_info->stripesize = stripesize;
3327
3328 /*
3329 * Handle the space caching options appropriately now that we have the
3330 * super block loaded and validated.
3331 */
3332 btrfs_set_free_space_cache_settings(fs_info);
3333
3334 if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3335 ret = -EINVAL;
3336 goto fail_alloc;
3337 }
3338
3339 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3340 if (ret < 0)
3341 goto fail_alloc;
3342
3343 /*
3344 * At this point our mount options are validated, if we set ->max_inline
3345 * to something non-standard make sure we truncate it to sectorsize.
3346 */
3347 fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3348
3349 if (sectorsize < PAGE_SIZE) {
3350 struct btrfs_subpage_info *subpage_info;
3351
3352 btrfs_warn(fs_info,
3353 "read-write for sector size %u with page size %lu is experimental",
3354 sectorsize, PAGE_SIZE);
3355 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3356 if (!subpage_info) {
3357 ret = -ENOMEM;
3358 goto fail_alloc;
3359 }
3360 btrfs_init_subpage_info(subpage_info, sectorsize);
3361 fs_info->subpage_info = subpage_info;
3362 }
3363
3364 ret = btrfs_init_workqueues(fs_info);
3365 if (ret)
3366 goto fail_sb_buffer;
3367
3368 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3369 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3370
3371 /* Update the values for the current filesystem. */
3372 sb->s_blocksize = sectorsize;
3373 sb->s_blocksize_bits = blksize_bits(sectorsize);
3374 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3375
3376 mutex_lock(&fs_info->chunk_mutex);
3377 ret = btrfs_read_sys_array(fs_info);
3378 mutex_unlock(&fs_info->chunk_mutex);
3379 if (ret) {
3380 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3381 goto fail_sb_buffer;
3382 }
3383
3384 generation = btrfs_super_chunk_root_generation(disk_super);
3385 level = btrfs_super_chunk_root_level(disk_super);
3386 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3387 generation, level);
3388 if (ret) {
3389 btrfs_err(fs_info, "failed to read chunk root");
3390 goto fail_tree_roots;
3391 }
3392
3393 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3394 offsetof(struct btrfs_header, chunk_tree_uuid),
3395 BTRFS_UUID_SIZE);
3396
3397 ret = btrfs_read_chunk_tree(fs_info);
3398 if (ret) {
3399 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3400 goto fail_tree_roots;
3401 }
3402
3403 /*
3404 * At this point we know all the devices that make this filesystem,
3405 * including the seed devices but we don't know yet if the replace
3406 * target is required. So free devices that are not part of this
3407 * filesystem but skip the replace target device which is checked
3408 * below in btrfs_init_dev_replace().
3409 */
3410 btrfs_free_extra_devids(fs_devices);
3411 if (!fs_devices->latest_dev->bdev) {
3412 btrfs_err(fs_info, "failed to read devices");
3413 ret = -EIO;
3414 goto fail_tree_roots;
3415 }
3416
3417 ret = init_tree_roots(fs_info);
3418 if (ret)
3419 goto fail_tree_roots;
3420
3421 /*
3422 * Get zone type information of zoned block devices. This will also
3423 * handle emulation of a zoned filesystem if a regular device has the
3424 * zoned incompat feature flag set.
3425 */
3426 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3427 if (ret) {
3428 btrfs_err(fs_info,
3429 "zoned: failed to read device zone info: %d", ret);
3430 goto fail_block_groups;
3431 }
3432
3433 /*
3434 * If we have a uuid root and we're not being told to rescan we need to
3435 * check the generation here so we can set the
3436 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3437 * transaction during a balance or the log replay without updating the
3438 * uuid generation, and then if we crash we would rescan the uuid tree,
3439 * even though it was perfectly fine.
3440 */
3441 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3442 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3443 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3444
3445 ret = btrfs_verify_dev_extents(fs_info);
3446 if (ret) {
3447 btrfs_err(fs_info,
3448 "failed to verify dev extents against chunks: %d",
3449 ret);
3450 goto fail_block_groups;
3451 }
3452 ret = btrfs_recover_balance(fs_info);
3453 if (ret) {
3454 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3455 goto fail_block_groups;
3456 }
3457
3458 ret = btrfs_init_dev_stats(fs_info);
3459 if (ret) {
3460 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3461 goto fail_block_groups;
3462 }
3463
3464 ret = btrfs_init_dev_replace(fs_info);
3465 if (ret) {
3466 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3467 goto fail_block_groups;
3468 }
3469
3470 ret = btrfs_check_zoned_mode(fs_info);
3471 if (ret) {
3472 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3473 ret);
3474 goto fail_block_groups;
3475 }
3476
3477 ret = btrfs_sysfs_add_fsid(fs_devices);
3478 if (ret) {
3479 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3480 ret);
3481 goto fail_block_groups;
3482 }
3483
3484 ret = btrfs_sysfs_add_mounted(fs_info);
3485 if (ret) {
3486 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3487 goto fail_fsdev_sysfs;
3488 }
3489
3490 ret = btrfs_init_space_info(fs_info);
3491 if (ret) {
3492 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3493 goto fail_sysfs;
3494 }
3495
3496 ret = btrfs_read_block_groups(fs_info);
3497 if (ret) {
3498 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3499 goto fail_sysfs;
3500 }
3501
3502 btrfs_free_zone_cache(fs_info);
3503
3504 btrfs_check_active_zone_reservation(fs_info);
3505
3506 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3507 !btrfs_check_rw_degradable(fs_info, NULL)) {
3508 btrfs_warn(fs_info,
3509 "writable mount is not allowed due to too many missing devices");
3510 ret = -EINVAL;
3511 goto fail_sysfs;
3512 }
3513
3514 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3515 "btrfs-cleaner");
3516 if (IS_ERR(fs_info->cleaner_kthread)) {
3517 ret = PTR_ERR(fs_info->cleaner_kthread);
3518 goto fail_sysfs;
3519 }
3520
3521 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3522 tree_root,
3523 "btrfs-transaction");
3524 if (IS_ERR(fs_info->transaction_kthread)) {
3525 ret = PTR_ERR(fs_info->transaction_kthread);
3526 goto fail_cleaner;
3527 }
3528
3529 ret = btrfs_read_qgroup_config(fs_info);
3530 if (ret)
3531 goto fail_trans_kthread;
3532
3533 if (btrfs_build_ref_tree(fs_info))
3534 btrfs_err(fs_info, "couldn't build ref tree");
3535
3536 /* do not make disk changes in broken FS or nologreplay is given */
3537 if (btrfs_super_log_root(disk_super) != 0 &&
3538 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3539 btrfs_info(fs_info, "start tree-log replay");
3540 ret = btrfs_replay_log(fs_info, fs_devices);
3541 if (ret)
3542 goto fail_qgroup;
3543 }
3544
3545 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3546 if (IS_ERR(fs_info->fs_root)) {
3547 ret = PTR_ERR(fs_info->fs_root);
3548 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3549 fs_info->fs_root = NULL;
3550 goto fail_qgroup;
3551 }
3552
3553 if (sb_rdonly(sb))
3554 return 0;
3555
3556 ret = btrfs_start_pre_rw_mount(fs_info);
3557 if (ret) {
3558 close_ctree(fs_info);
3559 return ret;
3560 }
3561 btrfs_discard_resume(fs_info);
3562
3563 if (fs_info->uuid_root &&
3564 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3565 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3566 btrfs_info(fs_info, "checking UUID tree");
3567 ret = btrfs_check_uuid_tree(fs_info);
3568 if (ret) {
3569 btrfs_warn(fs_info,
3570 "failed to check the UUID tree: %d", ret);
3571 close_ctree(fs_info);
3572 return ret;
3573 }
3574 }
3575
3576 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3577
3578 /* Kick the cleaner thread so it'll start deleting snapshots. */
3579 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3580 wake_up_process(fs_info->cleaner_kthread);
3581
3582 return 0;
3583
3584 fail_qgroup:
3585 btrfs_free_qgroup_config(fs_info);
3586 fail_trans_kthread:
3587 kthread_stop(fs_info->transaction_kthread);
3588 btrfs_cleanup_transaction(fs_info);
3589 btrfs_free_fs_roots(fs_info);
3590 fail_cleaner:
3591 kthread_stop(fs_info->cleaner_kthread);
3592
3593 /*
3594 * make sure we're done with the btree inode before we stop our
3595 * kthreads
3596 */
3597 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3598
3599 fail_sysfs:
3600 btrfs_sysfs_remove_mounted(fs_info);
3601
3602 fail_fsdev_sysfs:
3603 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3604
3605 fail_block_groups:
3606 btrfs_put_block_group_cache(fs_info);
3607
3608 fail_tree_roots:
3609 if (fs_info->data_reloc_root)
3610 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3611 free_root_pointers(fs_info, true);
3612 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3613
3614 fail_sb_buffer:
3615 btrfs_stop_all_workers(fs_info);
3616 btrfs_free_block_groups(fs_info);
3617 fail_alloc:
3618 btrfs_mapping_tree_free(fs_info);
3619
3620 iput(fs_info->btree_inode);
3621 fail:
3622 btrfs_close_devices(fs_info->fs_devices);
3623 ASSERT(ret < 0);
3624 return ret;
3625 }
3626 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3627
3628 static void btrfs_end_super_write(struct bio *bio)
3629 {
3630 struct btrfs_device *device = bio->bi_private;
3631 struct folio_iter fi;
3632
3633 bio_for_each_folio_all(fi, bio) {
3634 if (bio->bi_status) {
3635 btrfs_warn_rl_in_rcu(device->fs_info,
3636 "lost super block write due to IO error on %s (%d)",
3637 btrfs_dev_name(device),
3638 blk_status_to_errno(bio->bi_status));
3639 btrfs_dev_stat_inc_and_print(device,
3640 BTRFS_DEV_STAT_WRITE_ERRS);
3641 /* Ensure failure if the primary sb fails. */
3642 if (bio->bi_opf & REQ_FUA)
3643 atomic_add(BTRFS_SUPER_PRIMARY_WRITE_ERROR,
3644 &device->sb_write_errors);
3645 else
3646 atomic_inc(&device->sb_write_errors);
3647 }
3648 folio_unlock(fi.folio);
3649 folio_put(fi.folio);
3650 }
3651
3652 bio_put(bio);
3653 }
3654
3655 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3656 int copy_num, bool drop_cache)
3657 {
3658 struct btrfs_super_block *super;
3659 struct page *page;
3660 u64 bytenr, bytenr_orig;
3661 struct address_space *mapping = bdev->bd_mapping;
3662 int ret;
3663
3664 bytenr_orig = btrfs_sb_offset(copy_num);
3665 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3666 if (ret == -ENOENT)
3667 return ERR_PTR(-EINVAL);
3668 else if (ret)
3669 return ERR_PTR(ret);
3670
3671 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3672 return ERR_PTR(-EINVAL);
3673
3674 if (drop_cache) {
3675 /* This should only be called with the primary sb. */
3676 ASSERT(copy_num == 0);
3677
3678 /*
3679 * Drop the page of the primary superblock, so later read will
3680 * always read from the device.
3681 */
3682 invalidate_inode_pages2_range(mapping,
3683 bytenr >> PAGE_SHIFT,
3684 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3685 }
3686
3687 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3688 if (IS_ERR(page))
3689 return ERR_CAST(page);
3690
3691 super = page_address(page);
3692 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3693 btrfs_release_disk_super(super);
3694 return ERR_PTR(-ENODATA);
3695 }
3696
3697 if (btrfs_super_bytenr(super) != bytenr_orig) {
3698 btrfs_release_disk_super(super);
3699 return ERR_PTR(-EINVAL);
3700 }
3701
3702 return super;
3703 }
3704
3705
3706 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3707 {
3708 struct btrfs_super_block *super, *latest = NULL;
3709 int i;
3710 u64 transid = 0;
3711
3712 /* we would like to check all the supers, but that would make
3713 * a btrfs mount succeed after a mkfs from a different FS.
3714 * So, we need to add a special mount option to scan for
3715 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3716 */
3717 for (i = 0; i < 1; i++) {
3718 super = btrfs_read_dev_one_super(bdev, i, false);
3719 if (IS_ERR(super))
3720 continue;
3721
3722 if (!latest || btrfs_super_generation(super) > transid) {
3723 if (latest)
3724 btrfs_release_disk_super(super);
3725
3726 latest = super;
3727 transid = btrfs_super_generation(super);
3728 }
3729 }
3730
3731 return super;
3732 }
3733
3734 /*
3735 * Write superblock @sb to the @device. Do not wait for completion, all the
3736 * folios we use for writing are locked.
3737 *
3738 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3739 * the expected device size at commit time. Note that max_mirrors must be
3740 * same for write and wait phases.
3741 *
3742 * Return number of errors when folio is not found or submission fails.
3743 */
3744 static int write_dev_supers(struct btrfs_device *device,
3745 struct btrfs_super_block *sb, int max_mirrors)
3746 {
3747 struct btrfs_fs_info *fs_info = device->fs_info;
3748 struct address_space *mapping = device->bdev->bd_mapping;
3749 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3750 int i;
3751 int ret;
3752 u64 bytenr, bytenr_orig;
3753
3754 atomic_set(&device->sb_write_errors, 0);
3755
3756 if (max_mirrors == 0)
3757 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3758
3759 shash->tfm = fs_info->csum_shash;
3760
3761 for (i = 0; i < max_mirrors; i++) {
3762 struct folio *folio;
3763 struct bio *bio;
3764 struct btrfs_super_block *disk_super;
3765 size_t offset;
3766
3767 bytenr_orig = btrfs_sb_offset(i);
3768 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3769 if (ret == -ENOENT) {
3770 continue;
3771 } else if (ret < 0) {
3772 btrfs_err(device->fs_info,
3773 "couldn't get super block location for mirror %d",
3774 i);
3775 atomic_inc(&device->sb_write_errors);
3776 continue;
3777 }
3778 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3779 device->commit_total_bytes)
3780 break;
3781
3782 btrfs_set_super_bytenr(sb, bytenr_orig);
3783
3784 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3785 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3786 sb->csum);
3787
3788 folio = __filemap_get_folio(mapping, bytenr >> PAGE_SHIFT,
3789 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3790 GFP_NOFS);
3791 if (IS_ERR(folio)) {
3792 btrfs_err(device->fs_info,
3793 "couldn't get super block page for bytenr %llu",
3794 bytenr);
3795 atomic_inc(&device->sb_write_errors);
3796 continue;
3797 }
3798 ASSERT(folio_order(folio) == 0);
3799
3800 offset = offset_in_folio(folio, bytenr);
3801 disk_super = folio_address(folio) + offset;
3802 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3803
3804 /*
3805 * Directly use bios here instead of relying on the page cache
3806 * to do I/O, so we don't lose the ability to do integrity
3807 * checking.
3808 */
3809 bio = bio_alloc(device->bdev, 1,
3810 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3811 GFP_NOFS);
3812 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3813 bio->bi_private = device;
3814 bio->bi_end_io = btrfs_end_super_write;
3815 bio_add_folio_nofail(bio, folio, BTRFS_SUPER_INFO_SIZE, offset);
3816
3817 /*
3818 * We FUA only the first super block. The others we allow to
3819 * go down lazy and there's a short window where the on-disk
3820 * copies might still contain the older version.
3821 */
3822 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3823 bio->bi_opf |= REQ_FUA;
3824 submit_bio(bio);
3825
3826 if (btrfs_advance_sb_log(device, i))
3827 atomic_inc(&device->sb_write_errors);
3828 }
3829 return atomic_read(&device->sb_write_errors) < i ? 0 : -1;
3830 }
3831
3832 /*
3833 * Wait for write completion of superblocks done by write_dev_supers,
3834 * @max_mirrors same for write and wait phases.
3835 *
3836 * Return -1 if primary super block write failed or when there were no super block
3837 * copies written. Otherwise 0.
3838 */
3839 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3840 {
3841 int i;
3842 int errors = 0;
3843 bool primary_failed = false;
3844 int ret;
3845 u64 bytenr;
3846
3847 if (max_mirrors == 0)
3848 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3849
3850 for (i = 0; i < max_mirrors; i++) {
3851 struct folio *folio;
3852
3853 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3854 if (ret == -ENOENT) {
3855 break;
3856 } else if (ret < 0) {
3857 errors++;
3858 if (i == 0)
3859 primary_failed = true;
3860 continue;
3861 }
3862 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3863 device->commit_total_bytes)
3864 break;
3865
3866 folio = filemap_get_folio(device->bdev->bd_mapping,
3867 bytenr >> PAGE_SHIFT);
3868 /* If the folio has been removed, then we know it completed. */
3869 if (IS_ERR(folio))
3870 continue;
3871 ASSERT(folio_order(folio) == 0);
3872
3873 /* Folio will be unlocked once the write completes. */
3874 folio_wait_locked(folio);
3875 folio_put(folio);
3876 }
3877
3878 errors += atomic_read(&device->sb_write_errors);
3879 if (errors >= BTRFS_SUPER_PRIMARY_WRITE_ERROR)
3880 primary_failed = true;
3881 if (primary_failed) {
3882 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3883 device->devid);
3884 return -1;
3885 }
3886
3887 return errors < i ? 0 : -1;
3888 }
3889
3890 /*
3891 * endio for the write_dev_flush, this will wake anyone waiting
3892 * for the barrier when it is done
3893 */
3894 static void btrfs_end_empty_barrier(struct bio *bio)
3895 {
3896 bio_uninit(bio);
3897 complete(bio->bi_private);
3898 }
3899
3900 /*
3901 * Submit a flush request to the device if it supports it. Error handling is
3902 * done in the waiting counterpart.
3903 */
3904 static void write_dev_flush(struct btrfs_device *device)
3905 {
3906 struct bio *bio = &device->flush_bio;
3907
3908 device->last_flush_error = BLK_STS_OK;
3909
3910 bio_init(bio, device->bdev, NULL, 0,
3911 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3912 bio->bi_end_io = btrfs_end_empty_barrier;
3913 init_completion(&device->flush_wait);
3914 bio->bi_private = &device->flush_wait;
3915 submit_bio(bio);
3916 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3917 }
3918
3919 /*
3920 * If the flush bio has been submitted by write_dev_flush, wait for it.
3921 * Return true for any error, and false otherwise.
3922 */
3923 static bool wait_dev_flush(struct btrfs_device *device)
3924 {
3925 struct bio *bio = &device->flush_bio;
3926
3927 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3928 return false;
3929
3930 wait_for_completion_io(&device->flush_wait);
3931
3932 if (bio->bi_status) {
3933 device->last_flush_error = bio->bi_status;
3934 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3935 return true;
3936 }
3937
3938 return false;
3939 }
3940
3941 /*
3942 * send an empty flush down to each device in parallel,
3943 * then wait for them
3944 */
3945 static int barrier_all_devices(struct btrfs_fs_info *info)
3946 {
3947 struct list_head *head;
3948 struct btrfs_device *dev;
3949 int errors_wait = 0;
3950
3951 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3952 /* send down all the barriers */
3953 head = &info->fs_devices->devices;
3954 list_for_each_entry(dev, head, dev_list) {
3955 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3956 continue;
3957 if (!dev->bdev)
3958 continue;
3959 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3960 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3961 continue;
3962
3963 write_dev_flush(dev);
3964 }
3965
3966 /* wait for all the barriers */
3967 list_for_each_entry(dev, head, dev_list) {
3968 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3969 continue;
3970 if (!dev->bdev) {
3971 errors_wait++;
3972 continue;
3973 }
3974 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3975 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3976 continue;
3977
3978 if (wait_dev_flush(dev))
3979 errors_wait++;
3980 }
3981
3982 /*
3983 * Checks last_flush_error of disks in order to determine the device
3984 * state.
3985 */
3986 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3987 return -EIO;
3988
3989 return 0;
3990 }
3991
3992 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3993 {
3994 int raid_type;
3995 int min_tolerated = INT_MAX;
3996
3997 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3998 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3999 min_tolerated = min_t(int, min_tolerated,
4000 btrfs_raid_array[BTRFS_RAID_SINGLE].
4001 tolerated_failures);
4002
4003 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4004 if (raid_type == BTRFS_RAID_SINGLE)
4005 continue;
4006 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4007 continue;
4008 min_tolerated = min_t(int, min_tolerated,
4009 btrfs_raid_array[raid_type].
4010 tolerated_failures);
4011 }
4012
4013 if (min_tolerated == INT_MAX) {
4014 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4015 min_tolerated = 0;
4016 }
4017
4018 return min_tolerated;
4019 }
4020
4021 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4022 {
4023 struct list_head *head;
4024 struct btrfs_device *dev;
4025 struct btrfs_super_block *sb;
4026 struct btrfs_dev_item *dev_item;
4027 int ret;
4028 int do_barriers;
4029 int max_errors;
4030 int total_errors = 0;
4031 u64 flags;
4032
4033 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4034
4035 /*
4036 * max_mirrors == 0 indicates we're from commit_transaction,
4037 * not from fsync where the tree roots in fs_info have not
4038 * been consistent on disk.
4039 */
4040 if (max_mirrors == 0)
4041 backup_super_roots(fs_info);
4042
4043 sb = fs_info->super_for_commit;
4044 dev_item = &sb->dev_item;
4045
4046 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4047 head = &fs_info->fs_devices->devices;
4048 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4049
4050 if (do_barriers) {
4051 ret = barrier_all_devices(fs_info);
4052 if (ret) {
4053 mutex_unlock(
4054 &fs_info->fs_devices->device_list_mutex);
4055 btrfs_handle_fs_error(fs_info, ret,
4056 "errors while submitting device barriers.");
4057 return ret;
4058 }
4059 }
4060
4061 list_for_each_entry(dev, head, dev_list) {
4062 if (!dev->bdev) {
4063 total_errors++;
4064 continue;
4065 }
4066 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4067 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4068 continue;
4069
4070 btrfs_set_stack_device_generation(dev_item, 0);
4071 btrfs_set_stack_device_type(dev_item, dev->type);
4072 btrfs_set_stack_device_id(dev_item, dev->devid);
4073 btrfs_set_stack_device_total_bytes(dev_item,
4074 dev->commit_total_bytes);
4075 btrfs_set_stack_device_bytes_used(dev_item,
4076 dev->commit_bytes_used);
4077 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4078 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4079 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4080 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4081 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4082 BTRFS_FSID_SIZE);
4083
4084 flags = btrfs_super_flags(sb);
4085 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4086
4087 ret = btrfs_validate_write_super(fs_info, sb);
4088 if (ret < 0) {
4089 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4090 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4091 "unexpected superblock corruption detected");
4092 return -EUCLEAN;
4093 }
4094
4095 ret = write_dev_supers(dev, sb, max_mirrors);
4096 if (ret)
4097 total_errors++;
4098 }
4099 if (total_errors > max_errors) {
4100 btrfs_err(fs_info, "%d errors while writing supers",
4101 total_errors);
4102 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4103
4104 /* FUA is masked off if unsupported and can't be the reason */
4105 btrfs_handle_fs_error(fs_info, -EIO,
4106 "%d errors while writing supers",
4107 total_errors);
4108 return -EIO;
4109 }
4110
4111 total_errors = 0;
4112 list_for_each_entry(dev, head, dev_list) {
4113 if (!dev->bdev)
4114 continue;
4115 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4116 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4117 continue;
4118
4119 ret = wait_dev_supers(dev, max_mirrors);
4120 if (ret)
4121 total_errors++;
4122 }
4123 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4124 if (total_errors > max_errors) {
4125 btrfs_handle_fs_error(fs_info, -EIO,
4126 "%d errors while writing supers",
4127 total_errors);
4128 return -EIO;
4129 }
4130 return 0;
4131 }
4132
4133 /* Drop a fs root from the radix tree and free it. */
4134 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4135 struct btrfs_root *root)
4136 {
4137 bool drop_ref = false;
4138
4139 spin_lock(&fs_info->fs_roots_radix_lock);
4140 radix_tree_delete(&fs_info->fs_roots_radix,
4141 (unsigned long)btrfs_root_id(root));
4142 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4143 drop_ref = true;
4144 spin_unlock(&fs_info->fs_roots_radix_lock);
4145
4146 if (BTRFS_FS_ERROR(fs_info)) {
4147 ASSERT(root->log_root == NULL);
4148 if (root->reloc_root) {
4149 btrfs_put_root(root->reloc_root);
4150 root->reloc_root = NULL;
4151 }
4152 }
4153
4154 if (drop_ref)
4155 btrfs_put_root(root);
4156 }
4157
4158 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4159 {
4160 mutex_lock(&fs_info->cleaner_mutex);
4161 btrfs_run_delayed_iputs(fs_info);
4162 mutex_unlock(&fs_info->cleaner_mutex);
4163 wake_up_process(fs_info->cleaner_kthread);
4164
4165 /* wait until ongoing cleanup work done */
4166 down_write(&fs_info->cleanup_work_sem);
4167 up_write(&fs_info->cleanup_work_sem);
4168
4169 return btrfs_commit_current_transaction(fs_info->tree_root);
4170 }
4171
4172 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4173 {
4174 struct btrfs_transaction *trans;
4175 struct btrfs_transaction *tmp;
4176 bool found = false;
4177
4178 /*
4179 * This function is only called at the very end of close_ctree(),
4180 * thus no other running transaction, no need to take trans_lock.
4181 */
4182 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4183 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4184 struct extent_state *cached = NULL;
4185 u64 dirty_bytes = 0;
4186 u64 cur = 0;
4187 u64 found_start;
4188 u64 found_end;
4189
4190 found = true;
4191 while (find_first_extent_bit(&trans->dirty_pages, cur,
4192 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4193 dirty_bytes += found_end + 1 - found_start;
4194 cur = found_end + 1;
4195 }
4196 btrfs_warn(fs_info,
4197 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4198 trans->transid, dirty_bytes);
4199 btrfs_cleanup_one_transaction(trans, fs_info);
4200
4201 if (trans == fs_info->running_transaction)
4202 fs_info->running_transaction = NULL;
4203 list_del_init(&trans->list);
4204
4205 btrfs_put_transaction(trans);
4206 trace_btrfs_transaction_commit(fs_info);
4207 }
4208 ASSERT(!found);
4209 }
4210
4211 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4212 {
4213 int ret;
4214
4215 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4216
4217 /*
4218 * If we had UNFINISHED_DROPS we could still be processing them, so
4219 * clear that bit and wake up relocation so it can stop.
4220 * We must do this before stopping the block group reclaim task, because
4221 * at btrfs_relocate_block_group() we wait for this bit, and after the
4222 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4223 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4224 * return 1.
4225 */
4226 btrfs_wake_unfinished_drop(fs_info);
4227
4228 /*
4229 * We may have the reclaim task running and relocating a data block group,
4230 * in which case it may create delayed iputs. So stop it before we park
4231 * the cleaner kthread otherwise we can get new delayed iputs after
4232 * parking the cleaner, and that can make the async reclaim task to hang
4233 * if it's waiting for delayed iputs to complete, since the cleaner is
4234 * parked and can not run delayed iputs - this will make us hang when
4235 * trying to stop the async reclaim task.
4236 */
4237 cancel_work_sync(&fs_info->reclaim_bgs_work);
4238 /*
4239 * We don't want the cleaner to start new transactions, add more delayed
4240 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4241 * because that frees the task_struct, and the transaction kthread might
4242 * still try to wake up the cleaner.
4243 */
4244 kthread_park(fs_info->cleaner_kthread);
4245
4246 /* wait for the qgroup rescan worker to stop */
4247 btrfs_qgroup_wait_for_completion(fs_info, false);
4248
4249 /* wait for the uuid_scan task to finish */
4250 down(&fs_info->uuid_tree_rescan_sem);
4251 /* avoid complains from lockdep et al., set sem back to initial state */
4252 up(&fs_info->uuid_tree_rescan_sem);
4253
4254 /* pause restriper - we want to resume on mount */
4255 btrfs_pause_balance(fs_info);
4256
4257 btrfs_dev_replace_suspend_for_unmount(fs_info);
4258
4259 btrfs_scrub_cancel(fs_info);
4260
4261 /* wait for any defraggers to finish */
4262 wait_event(fs_info->transaction_wait,
4263 (atomic_read(&fs_info->defrag_running) == 0));
4264
4265 /* clear out the rbtree of defraggable inodes */
4266 btrfs_cleanup_defrag_inodes(fs_info);
4267
4268 /*
4269 * Wait for any fixup workers to complete.
4270 * If we don't wait for them here and they are still running by the time
4271 * we call kthread_stop() against the cleaner kthread further below, we
4272 * get an use-after-free on the cleaner because the fixup worker adds an
4273 * inode to the list of delayed iputs and then attempts to wakeup the
4274 * cleaner kthread, which was already stopped and destroyed. We parked
4275 * already the cleaner, but below we run all pending delayed iputs.
4276 */
4277 btrfs_flush_workqueue(fs_info->fixup_workers);
4278
4279 /*
4280 * After we parked the cleaner kthread, ordered extents may have
4281 * completed and created new delayed iputs. If one of the async reclaim
4282 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4283 * can hang forever trying to stop it, because if a delayed iput is
4284 * added after it ran btrfs_run_delayed_iputs() and before it called
4285 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4286 * no one else to run iputs.
4287 *
4288 * So wait for all ongoing ordered extents to complete and then run
4289 * delayed iputs. This works because once we reach this point no one
4290 * can either create new ordered extents nor create delayed iputs
4291 * through some other means.
4292 *
4293 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4294 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4295 * but the delayed iput for the respective inode is made only when doing
4296 * the final btrfs_put_ordered_extent() (which must happen at
4297 * btrfs_finish_ordered_io() when we are unmounting).
4298 */
4299 btrfs_flush_workqueue(fs_info->endio_write_workers);
4300 /* Ordered extents for free space inodes. */
4301 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4302 btrfs_run_delayed_iputs(fs_info);
4303
4304 cancel_work_sync(&fs_info->async_reclaim_work);
4305 cancel_work_sync(&fs_info->async_data_reclaim_work);
4306 cancel_work_sync(&fs_info->preempt_reclaim_work);
4307
4308 /* Cancel or finish ongoing discard work */
4309 btrfs_discard_cleanup(fs_info);
4310
4311 if (!sb_rdonly(fs_info->sb)) {
4312 /*
4313 * The cleaner kthread is stopped, so do one final pass over
4314 * unused block groups.
4315 */
4316 btrfs_delete_unused_bgs(fs_info);
4317
4318 /*
4319 * There might be existing delayed inode workers still running
4320 * and holding an empty delayed inode item. We must wait for
4321 * them to complete first because they can create a transaction.
4322 * This happens when someone calls btrfs_balance_delayed_items()
4323 * and then a transaction commit runs the same delayed nodes
4324 * before any delayed worker has done something with the nodes.
4325 * We must wait for any worker here and not at transaction
4326 * commit time since that could cause a deadlock.
4327 * This is a very rare case.
4328 */
4329 btrfs_flush_workqueue(fs_info->delayed_workers);
4330
4331 ret = btrfs_commit_super(fs_info);
4332 if (ret)
4333 btrfs_err(fs_info, "commit super ret %d", ret);
4334 }
4335
4336 if (BTRFS_FS_ERROR(fs_info))
4337 btrfs_error_commit_super(fs_info);
4338
4339 kthread_stop(fs_info->transaction_kthread);
4340 kthread_stop(fs_info->cleaner_kthread);
4341
4342 ASSERT(list_empty(&fs_info->delayed_iputs));
4343 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4344
4345 if (btrfs_check_quota_leak(fs_info)) {
4346 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4347 btrfs_err(fs_info, "qgroup reserved space leaked");
4348 }
4349
4350 btrfs_free_qgroup_config(fs_info);
4351 ASSERT(list_empty(&fs_info->delalloc_roots));
4352
4353 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4354 btrfs_info(fs_info, "at unmount delalloc count %lld",
4355 percpu_counter_sum(&fs_info->delalloc_bytes));
4356 }
4357
4358 if (percpu_counter_sum(&fs_info->ordered_bytes))
4359 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4360 percpu_counter_sum(&fs_info->ordered_bytes));
4361
4362 btrfs_sysfs_remove_mounted(fs_info);
4363 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4364
4365 btrfs_put_block_group_cache(fs_info);
4366
4367 /*
4368 * we must make sure there is not any read request to
4369 * submit after we stopping all workers.
4370 */
4371 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4372 btrfs_stop_all_workers(fs_info);
4373
4374 /* We shouldn't have any transaction open at this point */
4375 warn_about_uncommitted_trans(fs_info);
4376
4377 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4378 free_root_pointers(fs_info, true);
4379 btrfs_free_fs_roots(fs_info);
4380
4381 /*
4382 * We must free the block groups after dropping the fs_roots as we could
4383 * have had an IO error and have left over tree log blocks that aren't
4384 * cleaned up until the fs roots are freed. This makes the block group
4385 * accounting appear to be wrong because there's pending reserved bytes,
4386 * so make sure we do the block group cleanup afterwards.
4387 */
4388 btrfs_free_block_groups(fs_info);
4389
4390 iput(fs_info->btree_inode);
4391
4392 btrfs_mapping_tree_free(fs_info);
4393 btrfs_close_devices(fs_info->fs_devices);
4394 }
4395
4396 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4397 struct extent_buffer *buf)
4398 {
4399 struct btrfs_fs_info *fs_info = buf->fs_info;
4400 u64 transid = btrfs_header_generation(buf);
4401
4402 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4403 /*
4404 * This is a fast path so only do this check if we have sanity tests
4405 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4406 * outside of the sanity tests.
4407 */
4408 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4409 return;
4410 #endif
4411 /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4412 ASSERT(trans->transid == fs_info->generation);
4413 btrfs_assert_tree_write_locked(buf);
4414 if (unlikely(transid != fs_info->generation)) {
4415 btrfs_abort_transaction(trans, -EUCLEAN);
4416 btrfs_crit(fs_info,
4417 "dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4418 buf->start, transid, fs_info->generation);
4419 }
4420 set_extent_buffer_dirty(buf);
4421 }
4422
4423 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4424 int flush_delayed)
4425 {
4426 /*
4427 * looks as though older kernels can get into trouble with
4428 * this code, they end up stuck in balance_dirty_pages forever
4429 */
4430 int ret;
4431
4432 if (current->flags & PF_MEMALLOC)
4433 return;
4434
4435 if (flush_delayed)
4436 btrfs_balance_delayed_items(fs_info);
4437
4438 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4439 BTRFS_DIRTY_METADATA_THRESH,
4440 fs_info->dirty_metadata_batch);
4441 if (ret > 0) {
4442 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4443 }
4444 }
4445
4446 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4447 {
4448 __btrfs_btree_balance_dirty(fs_info, 1);
4449 }
4450
4451 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4452 {
4453 __btrfs_btree_balance_dirty(fs_info, 0);
4454 }
4455
4456 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4457 {
4458 /* cleanup FS via transaction */
4459 btrfs_cleanup_transaction(fs_info);
4460
4461 mutex_lock(&fs_info->cleaner_mutex);
4462 btrfs_run_delayed_iputs(fs_info);
4463 mutex_unlock(&fs_info->cleaner_mutex);
4464
4465 down_write(&fs_info->cleanup_work_sem);
4466 up_write(&fs_info->cleanup_work_sem);
4467 }
4468
4469 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4470 {
4471 struct btrfs_root *gang[8];
4472 u64 root_objectid = 0;
4473 int ret;
4474
4475 spin_lock(&fs_info->fs_roots_radix_lock);
4476 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4477 (void **)gang, root_objectid,
4478 ARRAY_SIZE(gang))) != 0) {
4479 int i;
4480
4481 for (i = 0; i < ret; i++)
4482 gang[i] = btrfs_grab_root(gang[i]);
4483 spin_unlock(&fs_info->fs_roots_radix_lock);
4484
4485 for (i = 0; i < ret; i++) {
4486 if (!gang[i])
4487 continue;
4488 root_objectid = btrfs_root_id(gang[i]);
4489 btrfs_free_log(NULL, gang[i]);
4490 btrfs_put_root(gang[i]);
4491 }
4492 root_objectid++;
4493 spin_lock(&fs_info->fs_roots_radix_lock);
4494 }
4495 spin_unlock(&fs_info->fs_roots_radix_lock);
4496 btrfs_free_log_root_tree(NULL, fs_info);
4497 }
4498
4499 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4500 {
4501 struct btrfs_ordered_extent *ordered;
4502
4503 spin_lock(&root->ordered_extent_lock);
4504 /*
4505 * This will just short circuit the ordered completion stuff which will
4506 * make sure the ordered extent gets properly cleaned up.
4507 */
4508 list_for_each_entry(ordered, &root->ordered_extents,
4509 root_extent_list)
4510 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4511 spin_unlock(&root->ordered_extent_lock);
4512 }
4513
4514 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4515 {
4516 struct btrfs_root *root;
4517 LIST_HEAD(splice);
4518
4519 spin_lock(&fs_info->ordered_root_lock);
4520 list_splice_init(&fs_info->ordered_roots, &splice);
4521 while (!list_empty(&splice)) {
4522 root = list_first_entry(&splice, struct btrfs_root,
4523 ordered_root);
4524 list_move_tail(&root->ordered_root,
4525 &fs_info->ordered_roots);
4526
4527 spin_unlock(&fs_info->ordered_root_lock);
4528 btrfs_destroy_ordered_extents(root);
4529
4530 cond_resched();
4531 spin_lock(&fs_info->ordered_root_lock);
4532 }
4533 spin_unlock(&fs_info->ordered_root_lock);
4534
4535 /*
4536 * We need this here because if we've been flipped read-only we won't
4537 * get sync() from the umount, so we need to make sure any ordered
4538 * extents that haven't had their dirty pages IO start writeout yet
4539 * actually get run and error out properly.
4540 */
4541 btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
4542 }
4543
4544 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4545 struct btrfs_fs_info *fs_info)
4546 {
4547 struct rb_node *node;
4548 struct btrfs_delayed_ref_root *delayed_refs = &trans->delayed_refs;
4549 struct btrfs_delayed_ref_node *ref;
4550
4551 spin_lock(&delayed_refs->lock);
4552 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4553 struct btrfs_delayed_ref_head *head;
4554 struct rb_node *n;
4555 bool pin_bytes = false;
4556
4557 head = rb_entry(node, struct btrfs_delayed_ref_head,
4558 href_node);
4559 if (btrfs_delayed_ref_lock(delayed_refs, head))
4560 continue;
4561
4562 spin_lock(&head->lock);
4563 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4564 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4565 ref_node);
4566 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4567 RB_CLEAR_NODE(&ref->ref_node);
4568 if (!list_empty(&ref->add_list))
4569 list_del(&ref->add_list);
4570 atomic_dec(&delayed_refs->num_entries);
4571 btrfs_put_delayed_ref(ref);
4572 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4573 }
4574 if (head->must_insert_reserved)
4575 pin_bytes = true;
4576 btrfs_free_delayed_extent_op(head->extent_op);
4577 btrfs_delete_ref_head(delayed_refs, head);
4578 spin_unlock(&head->lock);
4579 spin_unlock(&delayed_refs->lock);
4580 mutex_unlock(&head->mutex);
4581
4582 if (pin_bytes) {
4583 struct btrfs_block_group *cache;
4584
4585 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4586 BUG_ON(!cache);
4587
4588 spin_lock(&cache->space_info->lock);
4589 spin_lock(&cache->lock);
4590 cache->pinned += head->num_bytes;
4591 btrfs_space_info_update_bytes_pinned(fs_info,
4592 cache->space_info, head->num_bytes);
4593 cache->reserved -= head->num_bytes;
4594 cache->space_info->bytes_reserved -= head->num_bytes;
4595 spin_unlock(&cache->lock);
4596 spin_unlock(&cache->space_info->lock);
4597
4598 btrfs_put_block_group(cache);
4599
4600 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4601 head->bytenr + head->num_bytes - 1);
4602 }
4603 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4604 btrfs_put_delayed_ref_head(head);
4605 cond_resched();
4606 spin_lock(&delayed_refs->lock);
4607 }
4608 btrfs_qgroup_destroy_extent_records(trans);
4609
4610 spin_unlock(&delayed_refs->lock);
4611 }
4612
4613 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4614 {
4615 struct btrfs_inode *btrfs_inode;
4616 LIST_HEAD(splice);
4617
4618 spin_lock(&root->delalloc_lock);
4619 list_splice_init(&root->delalloc_inodes, &splice);
4620
4621 while (!list_empty(&splice)) {
4622 struct inode *inode = NULL;
4623 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4624 delalloc_inodes);
4625 btrfs_del_delalloc_inode(btrfs_inode);
4626 spin_unlock(&root->delalloc_lock);
4627
4628 /*
4629 * Make sure we get a live inode and that it'll not disappear
4630 * meanwhile.
4631 */
4632 inode = igrab(&btrfs_inode->vfs_inode);
4633 if (inode) {
4634 unsigned int nofs_flag;
4635
4636 nofs_flag = memalloc_nofs_save();
4637 invalidate_inode_pages2(inode->i_mapping);
4638 memalloc_nofs_restore(nofs_flag);
4639 iput(inode);
4640 }
4641 spin_lock(&root->delalloc_lock);
4642 }
4643 spin_unlock(&root->delalloc_lock);
4644 }
4645
4646 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4647 {
4648 struct btrfs_root *root;
4649 LIST_HEAD(splice);
4650
4651 spin_lock(&fs_info->delalloc_root_lock);
4652 list_splice_init(&fs_info->delalloc_roots, &splice);
4653 while (!list_empty(&splice)) {
4654 root = list_first_entry(&splice, struct btrfs_root,
4655 delalloc_root);
4656 root = btrfs_grab_root(root);
4657 BUG_ON(!root);
4658 spin_unlock(&fs_info->delalloc_root_lock);
4659
4660 btrfs_destroy_delalloc_inodes(root);
4661 btrfs_put_root(root);
4662
4663 spin_lock(&fs_info->delalloc_root_lock);
4664 }
4665 spin_unlock(&fs_info->delalloc_root_lock);
4666 }
4667
4668 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4669 struct extent_io_tree *dirty_pages,
4670 int mark)
4671 {
4672 struct extent_buffer *eb;
4673 u64 start = 0;
4674 u64 end;
4675
4676 while (find_first_extent_bit(dirty_pages, start, &start, &end,
4677 mark, NULL)) {
4678 clear_extent_bits(dirty_pages, start, end, mark);
4679 while (start <= end) {
4680 eb = find_extent_buffer(fs_info, start);
4681 start += fs_info->nodesize;
4682 if (!eb)
4683 continue;
4684
4685 btrfs_tree_lock(eb);
4686 wait_on_extent_buffer_writeback(eb);
4687 btrfs_clear_buffer_dirty(NULL, eb);
4688 btrfs_tree_unlock(eb);
4689
4690 free_extent_buffer_stale(eb);
4691 }
4692 }
4693 }
4694
4695 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4696 struct extent_io_tree *unpin)
4697 {
4698 u64 start;
4699 u64 end;
4700
4701 while (1) {
4702 struct extent_state *cached_state = NULL;
4703
4704 /*
4705 * The btrfs_finish_extent_commit() may get the same range as
4706 * ours between find_first_extent_bit and clear_extent_dirty.
4707 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4708 * the same extent range.
4709 */
4710 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4711 if (!find_first_extent_bit(unpin, 0, &start, &end,
4712 EXTENT_DIRTY, &cached_state)) {
4713 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4714 break;
4715 }
4716
4717 clear_extent_dirty(unpin, start, end, &cached_state);
4718 free_extent_state(cached_state);
4719 btrfs_error_unpin_extent_range(fs_info, start, end);
4720 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4721 cond_resched();
4722 }
4723 }
4724
4725 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4726 {
4727 struct inode *inode;
4728
4729 inode = cache->io_ctl.inode;
4730 if (inode) {
4731 unsigned int nofs_flag;
4732
4733 nofs_flag = memalloc_nofs_save();
4734 invalidate_inode_pages2(inode->i_mapping);
4735 memalloc_nofs_restore(nofs_flag);
4736
4737 BTRFS_I(inode)->generation = 0;
4738 cache->io_ctl.inode = NULL;
4739 iput(inode);
4740 }
4741 ASSERT(cache->io_ctl.pages == NULL);
4742 btrfs_put_block_group(cache);
4743 }
4744
4745 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4746 struct btrfs_fs_info *fs_info)
4747 {
4748 struct btrfs_block_group *cache;
4749
4750 spin_lock(&cur_trans->dirty_bgs_lock);
4751 while (!list_empty(&cur_trans->dirty_bgs)) {
4752 cache = list_first_entry(&cur_trans->dirty_bgs,
4753 struct btrfs_block_group,
4754 dirty_list);
4755
4756 if (!list_empty(&cache->io_list)) {
4757 spin_unlock(&cur_trans->dirty_bgs_lock);
4758 list_del_init(&cache->io_list);
4759 btrfs_cleanup_bg_io(cache);
4760 spin_lock(&cur_trans->dirty_bgs_lock);
4761 }
4762
4763 list_del_init(&cache->dirty_list);
4764 spin_lock(&cache->lock);
4765 cache->disk_cache_state = BTRFS_DC_ERROR;
4766 spin_unlock(&cache->lock);
4767
4768 spin_unlock(&cur_trans->dirty_bgs_lock);
4769 btrfs_put_block_group(cache);
4770 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4771 spin_lock(&cur_trans->dirty_bgs_lock);
4772 }
4773 spin_unlock(&cur_trans->dirty_bgs_lock);
4774
4775 /*
4776 * Refer to the definition of io_bgs member for details why it's safe
4777 * to use it without any locking
4778 */
4779 while (!list_empty(&cur_trans->io_bgs)) {
4780 cache = list_first_entry(&cur_trans->io_bgs,
4781 struct btrfs_block_group,
4782 io_list);
4783
4784 list_del_init(&cache->io_list);
4785 spin_lock(&cache->lock);
4786 cache->disk_cache_state = BTRFS_DC_ERROR;
4787 spin_unlock(&cache->lock);
4788 btrfs_cleanup_bg_io(cache);
4789 }
4790 }
4791
4792 static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4793 {
4794 struct btrfs_root *gang[8];
4795 int i;
4796 int ret;
4797
4798 spin_lock(&fs_info->fs_roots_radix_lock);
4799 while (1) {
4800 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4801 (void **)gang, 0,
4802 ARRAY_SIZE(gang),
4803 BTRFS_ROOT_TRANS_TAG);
4804 if (ret == 0)
4805 break;
4806 for (i = 0; i < ret; i++) {
4807 struct btrfs_root *root = gang[i];
4808
4809 btrfs_qgroup_free_meta_all_pertrans(root);
4810 radix_tree_tag_clear(&fs_info->fs_roots_radix,
4811 (unsigned long)btrfs_root_id(root),
4812 BTRFS_ROOT_TRANS_TAG);
4813 }
4814 }
4815 spin_unlock(&fs_info->fs_roots_radix_lock);
4816 }
4817
4818 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4819 struct btrfs_fs_info *fs_info)
4820 {
4821 struct btrfs_device *dev, *tmp;
4822
4823 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4824 ASSERT(list_empty(&cur_trans->dirty_bgs));
4825 ASSERT(list_empty(&cur_trans->io_bgs));
4826
4827 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4828 post_commit_list) {
4829 list_del_init(&dev->post_commit_list);
4830 }
4831
4832 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4833
4834 cur_trans->state = TRANS_STATE_COMMIT_START;
4835 wake_up(&fs_info->transaction_blocked_wait);
4836
4837 cur_trans->state = TRANS_STATE_UNBLOCKED;
4838 wake_up(&fs_info->transaction_wait);
4839
4840 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4841 EXTENT_DIRTY);
4842 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4843
4844 cur_trans->state =TRANS_STATE_COMPLETED;
4845 wake_up(&cur_trans->commit_wait);
4846 }
4847
4848 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4849 {
4850 struct btrfs_transaction *t;
4851
4852 mutex_lock(&fs_info->transaction_kthread_mutex);
4853
4854 spin_lock(&fs_info->trans_lock);
4855 while (!list_empty(&fs_info->trans_list)) {
4856 t = list_first_entry(&fs_info->trans_list,
4857 struct btrfs_transaction, list);
4858 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4859 refcount_inc(&t->use_count);
4860 spin_unlock(&fs_info->trans_lock);
4861 btrfs_wait_for_commit(fs_info, t->transid);
4862 btrfs_put_transaction(t);
4863 spin_lock(&fs_info->trans_lock);
4864 continue;
4865 }
4866 if (t == fs_info->running_transaction) {
4867 t->state = TRANS_STATE_COMMIT_DOING;
4868 spin_unlock(&fs_info->trans_lock);
4869 /*
4870 * We wait for 0 num_writers since we don't hold a trans
4871 * handle open currently for this transaction.
4872 */
4873 wait_event(t->writer_wait,
4874 atomic_read(&t->num_writers) == 0);
4875 } else {
4876 spin_unlock(&fs_info->trans_lock);
4877 }
4878 btrfs_cleanup_one_transaction(t, fs_info);
4879
4880 spin_lock(&fs_info->trans_lock);
4881 if (t == fs_info->running_transaction)
4882 fs_info->running_transaction = NULL;
4883 list_del_init(&t->list);
4884 spin_unlock(&fs_info->trans_lock);
4885
4886 btrfs_put_transaction(t);
4887 trace_btrfs_transaction_commit(fs_info);
4888 spin_lock(&fs_info->trans_lock);
4889 }
4890 spin_unlock(&fs_info->trans_lock);
4891 btrfs_destroy_all_ordered_extents(fs_info);
4892 btrfs_destroy_delayed_inodes(fs_info);
4893 btrfs_assert_delayed_root_empty(fs_info);
4894 btrfs_destroy_all_delalloc_inodes(fs_info);
4895 btrfs_drop_all_logs(fs_info);
4896 btrfs_free_all_qgroup_pertrans(fs_info);
4897 mutex_unlock(&fs_info->transaction_kthread_mutex);
4898
4899 return 0;
4900 }
4901
4902 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4903 {
4904 struct btrfs_path *path;
4905 int ret;
4906 struct extent_buffer *l;
4907 struct btrfs_key search_key;
4908 struct btrfs_key found_key;
4909 int slot;
4910
4911 path = btrfs_alloc_path();
4912 if (!path)
4913 return -ENOMEM;
4914
4915 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4916 search_key.type = -1;
4917 search_key.offset = (u64)-1;
4918 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4919 if (ret < 0)
4920 goto error;
4921 if (ret == 0) {
4922 /*
4923 * Key with offset -1 found, there would have to exist a root
4924 * with such id, but this is out of valid range.
4925 */
4926 ret = -EUCLEAN;
4927 goto error;
4928 }
4929 if (path->slots[0] > 0) {
4930 slot = path->slots[0] - 1;
4931 l = path->nodes[0];
4932 btrfs_item_key_to_cpu(l, &found_key, slot);
4933 root->free_objectid = max_t(u64, found_key.objectid + 1,
4934 BTRFS_FIRST_FREE_OBJECTID);
4935 } else {
4936 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4937 }
4938 ret = 0;
4939 error:
4940 btrfs_free_path(path);
4941 return ret;
4942 }
4943
4944 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4945 {
4946 int ret;
4947 mutex_lock(&root->objectid_mutex);
4948
4949 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4950 btrfs_warn(root->fs_info,
4951 "the objectid of root %llu reaches its highest value",
4952 btrfs_root_id(root));
4953 ret = -ENOSPC;
4954 goto out;
4955 }
4956
4957 *objectid = root->free_objectid++;
4958 ret = 0;
4959 out:
4960 mutex_unlock(&root->objectid_mutex);
4961 return ret;
4962 }
4963
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.