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