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 <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}
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 "check-integrity.h"
33#include "rcu-string.h"
34#include "dev-replace.h"
35#include "raid56.h"
36#include "sysfs.h"
37#include "qgroup.h"
38#include "compression.h"
39#include "tree-checker.h"
40#include "ref-verify.h"
41#include "block-group.h"
42#include "discard.h"
43#include "space-info.h"
44#include "zoned.h"
45#include "subpage.h"
46#include "fs.h"
47#include "accessors.h"
48#include "extent-tree.h"
49#include "root-tree.h"
50#include "defrag.h"
51#include "uuid-tree.h"
52#include "relocation.h"
53#include "scrub.h"
54#include "super.h"
55
56#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
57 BTRFS_HEADER_FLAG_RELOC |\
58 BTRFS_SUPER_FLAG_ERROR |\
59 BTRFS_SUPER_FLAG_SEEDING |\
60 BTRFS_SUPER_FLAG_METADUMP |\
61 BTRFS_SUPER_FLAG_METADUMP_V2)
62
63static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_fs_info *fs_info);
66static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
67static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
68 struct extent_io_tree *dirty_pages,
69 int mark);
70static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
71 struct extent_io_tree *pinned_extents);
72static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
73static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
74
75static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
76{
77 if (fs_info->csum_shash)
78 crypto_free_shash(fs_info->csum_shash);
79}
80
81/*
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
85 */
86struct async_submit_bio {
87 struct btrfs_inode *inode;
88 struct bio *bio;
89 enum btrfs_wq_submit_cmd submit_cmd;
90 int mirror_num;
91
92 /* Optional parameter for used by direct io */
93 u64 dio_file_offset;
94 struct btrfs_work work;
95 blk_status_t status;
96};
97
98/*
99 * Compute the csum of a btree block and store the result to provided buffer.
100 */
101static void csum_tree_block(struct extent_buffer *buf, u8 *result)
102{
103 struct btrfs_fs_info *fs_info = buf->fs_info;
104 const int num_pages = num_extent_pages(buf);
105 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
106 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
107 char *kaddr;
108 int i;
109
110 shash->tfm = fs_info->csum_shash;
111 crypto_shash_init(shash);
112 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
113 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
114 first_page_part - BTRFS_CSUM_SIZE);
115
116 for (i = 1; i < num_pages; i++) {
117 kaddr = page_address(buf->pages[i]);
118 crypto_shash_update(shash, kaddr, PAGE_SIZE);
119 }
120 memset(result, 0, BTRFS_CSUM_SIZE);
121 crypto_shash_final(shash, result);
122}
123
124/*
125 * we can't consider a given block up to date unless the transid of the
126 * block matches the transid in the parent node's pointer. This is how we
127 * detect blocks that either didn't get written at all or got written
128 * in the wrong place.
129 */
130static int verify_parent_transid(struct extent_io_tree *io_tree,
131 struct extent_buffer *eb, u64 parent_transid,
132 int atomic)
133{
134 struct extent_state *cached_state = NULL;
135 int ret;
136
137 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
138 return 0;
139
140 if (atomic)
141 return -EAGAIN;
142
143 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state);
144 if (extent_buffer_uptodate(eb) &&
145 btrfs_header_generation(eb) == parent_transid) {
146 ret = 0;
147 goto out;
148 }
149 btrfs_err_rl(eb->fs_info,
150"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
151 eb->start, eb->read_mirror,
152 parent_transid, btrfs_header_generation(eb));
153 ret = 1;
154 clear_extent_buffer_uptodate(eb);
155out:
156 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
157 &cached_state);
158 return ret;
159}
160
161static bool btrfs_supported_super_csum(u16 csum_type)
162{
163 switch (csum_type) {
164 case BTRFS_CSUM_TYPE_CRC32:
165 case BTRFS_CSUM_TYPE_XXHASH:
166 case BTRFS_CSUM_TYPE_SHA256:
167 case BTRFS_CSUM_TYPE_BLAKE2:
168 return true;
169 default:
170 return false;
171 }
172}
173
174/*
175 * Return 0 if the superblock checksum type matches the checksum value of that
176 * algorithm. Pass the raw disk superblock data.
177 */
178int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
179 const struct btrfs_super_block *disk_sb)
180{
181 char result[BTRFS_CSUM_SIZE];
182 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
183
184 shash->tfm = fs_info->csum_shash;
185
186 /*
187 * The super_block structure does not span the whole
188 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
189 * filled with zeros and is included in the checksum.
190 */
191 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
192 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
193
194 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
195 return 1;
196
197 return 0;
198}
199
200int btrfs_verify_level_key(struct extent_buffer *eb, int level,
201 struct btrfs_key *first_key, u64 parent_transid)
202{
203 struct btrfs_fs_info *fs_info = eb->fs_info;
204 int found_level;
205 struct btrfs_key found_key;
206 int ret;
207
208 found_level = btrfs_header_level(eb);
209 if (found_level != level) {
210 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
211 KERN_ERR "BTRFS: tree level check failed\n");
212 btrfs_err(fs_info,
213"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
214 eb->start, level, found_level);
215 return -EIO;
216 }
217
218 if (!first_key)
219 return 0;
220
221 /*
222 * For live tree block (new tree blocks in current transaction),
223 * we need proper lock context to avoid race, which is impossible here.
224 * So we only checks tree blocks which is read from disk, whose
225 * generation <= fs_info->last_trans_committed.
226 */
227 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
228 return 0;
229
230 /* We have @first_key, so this @eb must have at least one item */
231 if (btrfs_header_nritems(eb) == 0) {
232 btrfs_err(fs_info,
233 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
234 eb->start);
235 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
236 return -EUCLEAN;
237 }
238
239 if (found_level)
240 btrfs_node_key_to_cpu(eb, &found_key, 0);
241 else
242 btrfs_item_key_to_cpu(eb, &found_key, 0);
243 ret = btrfs_comp_cpu_keys(first_key, &found_key);
244
245 if (ret) {
246 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
247 KERN_ERR "BTRFS: tree first key check failed\n");
248 btrfs_err(fs_info,
249"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
250 eb->start, parent_transid, first_key->objectid,
251 first_key->type, first_key->offset,
252 found_key.objectid, found_key.type,
253 found_key.offset);
254 }
255 return ret;
256}
257
258static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
259 int mirror_num)
260{
261 struct btrfs_fs_info *fs_info = eb->fs_info;
262 u64 start = eb->start;
263 int i, num_pages = num_extent_pages(eb);
264 int ret = 0;
265
266 if (sb_rdonly(fs_info->sb))
267 return -EROFS;
268
269 for (i = 0; i < num_pages; i++) {
270 struct page *p = eb->pages[i];
271
272 ret = btrfs_repair_io_failure(fs_info, 0, start, PAGE_SIZE,
273 start, p, start - page_offset(p), mirror_num);
274 if (ret)
275 break;
276 start += PAGE_SIZE;
277 }
278
279 return ret;
280}
281
282/*
283 * helper to read a given tree block, doing retries as required when
284 * the checksums don't match and we have alternate mirrors to try.
285 *
286 * @check: expected tree parentness check, see the comments of the
287 * structure for details.
288 */
289int btrfs_read_extent_buffer(struct extent_buffer *eb,
290 struct btrfs_tree_parent_check *check)
291{
292 struct btrfs_fs_info *fs_info = eb->fs_info;
293 int failed = 0;
294 int ret;
295 int num_copies = 0;
296 int mirror_num = 0;
297 int failed_mirror = 0;
298
299 ASSERT(check);
300
301 while (1) {
302 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
303 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
304 if (!ret)
305 break;
306
307 num_copies = btrfs_num_copies(fs_info,
308 eb->start, eb->len);
309 if (num_copies == 1)
310 break;
311
312 if (!failed_mirror) {
313 failed = 1;
314 failed_mirror = eb->read_mirror;
315 }
316
317 mirror_num++;
318 if (mirror_num == failed_mirror)
319 mirror_num++;
320
321 if (mirror_num > num_copies)
322 break;
323 }
324
325 if (failed && !ret && failed_mirror)
326 btrfs_repair_eb_io_failure(eb, failed_mirror);
327
328 return ret;
329}
330
331static int csum_one_extent_buffer(struct extent_buffer *eb)
332{
333 struct btrfs_fs_info *fs_info = eb->fs_info;
334 u8 result[BTRFS_CSUM_SIZE];
335 int ret;
336
337 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
338 offsetof(struct btrfs_header, fsid),
339 BTRFS_FSID_SIZE) == 0);
340 csum_tree_block(eb, result);
341
342 if (btrfs_header_level(eb))
343 ret = btrfs_check_node(eb);
344 else
345 ret = btrfs_check_leaf_full(eb);
346
347 if (ret < 0)
348 goto error;
349
350 /*
351 * Also check the generation, the eb reached here must be newer than
352 * last committed. Or something seriously wrong happened.
353 */
354 if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
355 ret = -EUCLEAN;
356 btrfs_err(fs_info,
357 "block=%llu bad generation, have %llu expect > %llu",
358 eb->start, btrfs_header_generation(eb),
359 fs_info->last_trans_committed);
360 goto error;
361 }
362 write_extent_buffer(eb, result, 0, fs_info->csum_size);
363
364 return 0;
365
366error:
367 btrfs_print_tree(eb, 0);
368 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
369 eb->start);
370 /*
371 * Be noisy if this is an extent buffer from a log tree. We don't abort
372 * a transaction in case there's a bad log tree extent buffer, we just
373 * fallback to a transaction commit. Still we want to know when there is
374 * a bad log tree extent buffer, as that may signal a bug somewhere.
375 */
376 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
377 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
378 return ret;
379}
380
381/* Checksum all dirty extent buffers in one bio_vec */
382static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
383 struct bio_vec *bvec)
384{
385 struct page *page = bvec->bv_page;
386 u64 bvec_start = page_offset(page) + bvec->bv_offset;
387 u64 cur;
388 int ret = 0;
389
390 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
391 cur += fs_info->nodesize) {
392 struct extent_buffer *eb;
393 bool uptodate;
394
395 eb = find_extent_buffer(fs_info, cur);
396 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
397 fs_info->nodesize);
398
399 /* A dirty eb shouldn't disappear from buffer_radix */
400 if (WARN_ON(!eb))
401 return -EUCLEAN;
402
403 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
404 free_extent_buffer(eb);
405 return -EUCLEAN;
406 }
407 if (WARN_ON(!uptodate)) {
408 free_extent_buffer(eb);
409 return -EUCLEAN;
410 }
411
412 ret = csum_one_extent_buffer(eb);
413 free_extent_buffer(eb);
414 if (ret < 0)
415 return ret;
416 }
417 return ret;
418}
419
420/*
421 * Checksum a dirty tree block before IO. This has extra checks to make sure
422 * we only fill in the checksum field in the first page of a multi-page block.
423 * For subpage extent buffers we need bvec to also read the offset in the page.
424 */
425static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
426{
427 struct page *page = bvec->bv_page;
428 u64 start = page_offset(page);
429 u64 found_start;
430 struct extent_buffer *eb;
431
432 if (fs_info->nodesize < PAGE_SIZE)
433 return csum_dirty_subpage_buffers(fs_info, bvec);
434
435 eb = (struct extent_buffer *)page->private;
436 if (page != eb->pages[0])
437 return 0;
438
439 found_start = btrfs_header_bytenr(eb);
440
441 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
442 WARN_ON(found_start != 0);
443 return 0;
444 }
445
446 /*
447 * Please do not consolidate these warnings into a single if.
448 * It is useful to know what went wrong.
449 */
450 if (WARN_ON(found_start != start))
451 return -EUCLEAN;
452 if (WARN_ON(!PageUptodate(page)))
453 return -EUCLEAN;
454
455 return csum_one_extent_buffer(eb);
456}
457
458static int check_tree_block_fsid(struct extent_buffer *eb)
459{
460 struct btrfs_fs_info *fs_info = eb->fs_info;
461 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
462 u8 fsid[BTRFS_FSID_SIZE];
463 u8 *metadata_uuid;
464
465 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
466 BTRFS_FSID_SIZE);
467 /*
468 * Checking the incompat flag is only valid for the current fs. For
469 * seed devices it's forbidden to have their uuid changed so reading
470 * ->fsid in this case is fine
471 */
472 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
473 metadata_uuid = fs_devices->metadata_uuid;
474 else
475 metadata_uuid = fs_devices->fsid;
476
477 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
478 return 0;
479
480 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
481 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
482 return 0;
483
484 return 1;
485}
486
487/* Do basic extent buffer checks at read time */
488static int validate_extent_buffer(struct extent_buffer *eb,
489 struct btrfs_tree_parent_check *check)
490{
491 struct btrfs_fs_info *fs_info = eb->fs_info;
492 u64 found_start;
493 const u32 csum_size = fs_info->csum_size;
494 u8 found_level;
495 u8 result[BTRFS_CSUM_SIZE];
496 const u8 *header_csum;
497 int ret = 0;
498
499 ASSERT(check);
500
501 found_start = btrfs_header_bytenr(eb);
502 if (found_start != eb->start) {
503 btrfs_err_rl(fs_info,
504 "bad tree block start, mirror %u want %llu have %llu",
505 eb->read_mirror, eb->start, found_start);
506 ret = -EIO;
507 goto out;
508 }
509 if (check_tree_block_fsid(eb)) {
510 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
511 eb->start, eb->read_mirror);
512 ret = -EIO;
513 goto out;
514 }
515 found_level = btrfs_header_level(eb);
516 if (found_level >= BTRFS_MAX_LEVEL) {
517 btrfs_err(fs_info,
518 "bad tree block level, mirror %u level %d on logical %llu",
519 eb->read_mirror, btrfs_header_level(eb), eb->start);
520 ret = -EIO;
521 goto out;
522 }
523
524 csum_tree_block(eb, result);
525 header_csum = page_address(eb->pages[0]) +
526 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
527
528 if (memcmp(result, header_csum, csum_size) != 0) {
529 btrfs_warn_rl(fs_info,
530"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
531 eb->start, eb->read_mirror,
532 CSUM_FMT_VALUE(csum_size, header_csum),
533 CSUM_FMT_VALUE(csum_size, result),
534 btrfs_header_level(eb));
535 ret = -EUCLEAN;
536 goto out;
537 }
538
539 if (found_level != check->level) {
540 btrfs_err(fs_info,
541 "level verify failed on logical %llu mirror %u wanted %u found %u",
542 eb->start, eb->read_mirror, check->level, found_level);
543 ret = -EIO;
544 goto out;
545 }
546 if (unlikely(check->transid &&
547 btrfs_header_generation(eb) != check->transid)) {
548 btrfs_err_rl(eb->fs_info,
549"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
550 eb->start, eb->read_mirror, check->transid,
551 btrfs_header_generation(eb));
552 ret = -EIO;
553 goto out;
554 }
555 if (check->has_first_key) {
556 struct btrfs_key *expect_key = &check->first_key;
557 struct btrfs_key found_key;
558
559 if (found_level)
560 btrfs_node_key_to_cpu(eb, &found_key, 0);
561 else
562 btrfs_item_key_to_cpu(eb, &found_key, 0);
563 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
564 btrfs_err(fs_info,
565"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
566 eb->start, check->transid,
567 expect_key->objectid,
568 expect_key->type, expect_key->offset,
569 found_key.objectid, found_key.type,
570 found_key.offset);
571 ret = -EUCLEAN;
572 goto out;
573 }
574 }
575 if (check->owner_root) {
576 ret = btrfs_check_eb_owner(eb, check->owner_root);
577 if (ret < 0)
578 goto out;
579 }
580
581 /*
582 * If this is a leaf block and it is corrupt, set the corrupt bit so
583 * that we don't try and read the other copies of this block, just
584 * return -EIO.
585 */
586 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
587 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
588 ret = -EIO;
589 }
590
591 if (found_level > 0 && btrfs_check_node(eb))
592 ret = -EIO;
593
594 if (!ret)
595 set_extent_buffer_uptodate(eb);
596 else
597 btrfs_err(fs_info,
598 "read time tree block corruption detected on logical %llu mirror %u",
599 eb->start, eb->read_mirror);
600out:
601 return ret;
602}
603
604static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
605 int mirror, struct btrfs_tree_parent_check *check)
606{
607 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
608 struct extent_buffer *eb;
609 bool reads_done;
610 int ret = 0;
611
612 ASSERT(check);
613
614 /*
615 * We don't allow bio merge for subpage metadata read, so we should
616 * only get one eb for each endio hook.
617 */
618 ASSERT(end == start + fs_info->nodesize - 1);
619 ASSERT(PagePrivate(page));
620
621 eb = find_extent_buffer(fs_info, start);
622 /*
623 * When we are reading one tree block, eb must have been inserted into
624 * the radix tree. If not, something is wrong.
625 */
626 ASSERT(eb);
627
628 reads_done = atomic_dec_and_test(&eb->io_pages);
629 /* Subpage read must finish in page read */
630 ASSERT(reads_done);
631
632 eb->read_mirror = mirror;
633 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
634 ret = -EIO;
635 goto err;
636 }
637 ret = validate_extent_buffer(eb, check);
638 if (ret < 0)
639 goto err;
640
641 set_extent_buffer_uptodate(eb);
642
643 free_extent_buffer(eb);
644 return ret;
645err:
646 /*
647 * end_bio_extent_readpage decrements io_pages in case of error,
648 * make sure it has something to decrement.
649 */
650 atomic_inc(&eb->io_pages);
651 clear_extent_buffer_uptodate(eb);
652 free_extent_buffer(eb);
653 return ret;
654}
655
656int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
657 struct page *page, u64 start, u64 end,
658 int mirror)
659{
660 struct extent_buffer *eb;
661 int ret = 0;
662 int reads_done;
663
664 ASSERT(page->private);
665
666 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
667 return validate_subpage_buffer(page, start, end, mirror,
668 &bbio->parent_check);
669
670 eb = (struct extent_buffer *)page->private;
671
672 /*
673 * The pending IO might have been the only thing that kept this buffer
674 * in memory. Make sure we have a ref for all this other checks
675 */
676 atomic_inc(&eb->refs);
677
678 reads_done = atomic_dec_and_test(&eb->io_pages);
679 if (!reads_done)
680 goto err;
681
682 eb->read_mirror = mirror;
683 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
684 ret = -EIO;
685 goto err;
686 }
687 ret = validate_extent_buffer(eb, &bbio->parent_check);
688err:
689 if (ret) {
690 /*
691 * our io error hook is going to dec the io pages
692 * again, we have to make sure it has something
693 * to decrement
694 */
695 atomic_inc(&eb->io_pages);
696 clear_extent_buffer_uptodate(eb);
697 }
698 free_extent_buffer(eb);
699
700 return ret;
701}
702
703static void run_one_async_start(struct btrfs_work *work)
704{
705 struct async_submit_bio *async;
706 blk_status_t ret;
707
708 async = container_of(work, struct async_submit_bio, work);
709 switch (async->submit_cmd) {
710 case WQ_SUBMIT_METADATA:
711 ret = btree_submit_bio_start(async->bio);
712 break;
713 case WQ_SUBMIT_DATA:
714 ret = btrfs_submit_bio_start(async->inode, async->bio);
715 break;
716 case WQ_SUBMIT_DATA_DIO:
717 ret = btrfs_submit_bio_start_direct_io(async->inode,
718 async->bio, async->dio_file_offset);
719 break;
720 }
721 if (ret)
722 async->status = ret;
723}
724
725/*
726 * In order to insert checksums into the metadata in large chunks, we wait
727 * until bio submission time. All the pages in the bio are checksummed and
728 * sums are attached onto the ordered extent record.
729 *
730 * At IO completion time the csums attached on the ordered extent record are
731 * inserted into the tree.
732 */
733static void run_one_async_done(struct btrfs_work *work)
734{
735 struct async_submit_bio *async =
736 container_of(work, struct async_submit_bio, work);
737 struct btrfs_inode *inode = async->inode;
738 struct btrfs_bio *bbio = btrfs_bio(async->bio);
739
740 /* If an error occurred we just want to clean up the bio and move on */
741 if (async->status) {
742 btrfs_bio_end_io(bbio, async->status);
743 return;
744 }
745
746 /*
747 * All of the bios that pass through here are from async helpers.
748 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
749 * This changes nothing when cgroups aren't in use.
750 */
751 async->bio->bi_opf |= REQ_CGROUP_PUNT;
752 btrfs_submit_bio(inode->root->fs_info, async->bio, async->mirror_num);
753}
754
755static void run_one_async_free(struct btrfs_work *work)
756{
757 struct async_submit_bio *async;
758
759 async = container_of(work, struct async_submit_bio, work);
760 kfree(async);
761}
762
763/*
764 * Submit bio to an async queue.
765 *
766 * Retrun:
767 * - true if the work has been succesfuly submitted
768 * - false in case of error
769 */
770bool btrfs_wq_submit_bio(struct btrfs_inode *inode, struct bio *bio, int mirror_num,
771 u64 dio_file_offset, enum btrfs_wq_submit_cmd cmd)
772{
773 struct btrfs_fs_info *fs_info = inode->root->fs_info;
774 struct async_submit_bio *async;
775
776 async = kmalloc(sizeof(*async), GFP_NOFS);
777 if (!async)
778 return false;
779
780 async->inode = inode;
781 async->bio = bio;
782 async->mirror_num = mirror_num;
783 async->submit_cmd = cmd;
784
785 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
786 run_one_async_free);
787
788 async->dio_file_offset = dio_file_offset;
789
790 async->status = 0;
791
792 if (op_is_sync(bio->bi_opf))
793 btrfs_queue_work(fs_info->hipri_workers, &async->work);
794 else
795 btrfs_queue_work(fs_info->workers, &async->work);
796 return true;
797}
798
799static blk_status_t btree_csum_one_bio(struct bio *bio)
800{
801 struct bio_vec *bvec;
802 struct btrfs_root *root;
803 int ret = 0;
804 struct bvec_iter_all iter_all;
805
806 ASSERT(!bio_flagged(bio, BIO_CLONED));
807 bio_for_each_segment_all(bvec, bio, iter_all) {
808 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
809 ret = csum_dirty_buffer(root->fs_info, bvec);
810 if (ret)
811 break;
812 }
813
814 return errno_to_blk_status(ret);
815}
816
817blk_status_t btree_submit_bio_start(struct bio *bio)
818{
819 /*
820 * when we're called for a write, we're already in the async
821 * submission context. Just jump into btrfs_submit_bio.
822 */
823 return btree_csum_one_bio(bio);
824}
825
826static bool should_async_write(struct btrfs_fs_info *fs_info,
827 struct btrfs_inode *bi)
828{
829 if (btrfs_is_zoned(fs_info))
830 return false;
831 if (atomic_read(&bi->sync_writers))
832 return false;
833 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
834 return false;
835 return true;
836}
837
838void btrfs_submit_metadata_bio(struct btrfs_inode *inode, struct bio *bio, int mirror_num)
839{
840 struct btrfs_fs_info *fs_info = inode->root->fs_info;
841 struct btrfs_bio *bbio = btrfs_bio(bio);
842 blk_status_t ret;
843
844 bio->bi_opf |= REQ_META;
845 bbio->is_metadata = 1;
846
847 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
848 btrfs_submit_bio(fs_info, bio, mirror_num);
849 return;
850 }
851
852 /*
853 * Kthread helpers are used to submit writes so that checksumming can
854 * happen in parallel across all CPUs.
855 */
856 if (should_async_write(fs_info, inode) &&
857 btrfs_wq_submit_bio(inode, bio, mirror_num, 0, WQ_SUBMIT_METADATA))
858 return;
859
860 ret = btree_csum_one_bio(bio);
861 if (ret) {
862 btrfs_bio_end_io(bbio, ret);
863 return;
864 }
865
866 btrfs_submit_bio(fs_info, bio, mirror_num);
867}
868
869#ifdef CONFIG_MIGRATION
870static int btree_migrate_folio(struct address_space *mapping,
871 struct folio *dst, struct folio *src, enum migrate_mode mode)
872{
873 /*
874 * we can't safely write a btree page from here,
875 * we haven't done the locking hook
876 */
877 if (folio_test_dirty(src))
878 return -EAGAIN;
879 /*
880 * Buffers may be managed in a filesystem specific way.
881 * We must have no buffers or drop them.
882 */
883 if (folio_get_private(src) &&
884 !filemap_release_folio(src, GFP_KERNEL))
885 return -EAGAIN;
886 return migrate_folio(mapping, dst, src, mode);
887}
888#else
889#define btree_migrate_folio NULL
890#endif
891
892static int btree_writepages(struct address_space *mapping,
893 struct writeback_control *wbc)
894{
895 struct btrfs_fs_info *fs_info;
896 int ret;
897
898 if (wbc->sync_mode == WB_SYNC_NONE) {
899
900 if (wbc->for_kupdate)
901 return 0;
902
903 fs_info = BTRFS_I(mapping->host)->root->fs_info;
904 /* this is a bit racy, but that's ok */
905 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
906 BTRFS_DIRTY_METADATA_THRESH,
907 fs_info->dirty_metadata_batch);
908 if (ret < 0)
909 return 0;
910 }
911 return btree_write_cache_pages(mapping, wbc);
912}
913
914static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
915{
916 if (folio_test_writeback(folio) || folio_test_dirty(folio))
917 return false;
918
919 return try_release_extent_buffer(&folio->page);
920}
921
922static void btree_invalidate_folio(struct folio *folio, size_t offset,
923 size_t length)
924{
925 struct extent_io_tree *tree;
926 tree = &BTRFS_I(folio->mapping->host)->io_tree;
927 extent_invalidate_folio(tree, folio, offset);
928 btree_release_folio(folio, GFP_NOFS);
929 if (folio_get_private(folio)) {
930 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
931 "folio private not zero on folio %llu",
932 (unsigned long long)folio_pos(folio));
933 folio_detach_private(folio);
934 }
935}
936
937#ifdef DEBUG
938static bool btree_dirty_folio(struct address_space *mapping,
939 struct folio *folio)
940{
941 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
942 struct btrfs_subpage *subpage;
943 struct extent_buffer *eb;
944 int cur_bit = 0;
945 u64 page_start = folio_pos(folio);
946
947 if (fs_info->sectorsize == PAGE_SIZE) {
948 eb = folio_get_private(folio);
949 BUG_ON(!eb);
950 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
951 BUG_ON(!atomic_read(&eb->refs));
952 btrfs_assert_tree_write_locked(eb);
953 return filemap_dirty_folio(mapping, folio);
954 }
955 subpage = folio_get_private(folio);
956
957 ASSERT(subpage->dirty_bitmap);
958 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
959 unsigned long flags;
960 u64 cur;
961 u16 tmp = (1 << cur_bit);
962
963 spin_lock_irqsave(&subpage->lock, flags);
964 if (!(tmp & subpage->dirty_bitmap)) {
965 spin_unlock_irqrestore(&subpage->lock, flags);
966 cur_bit++;
967 continue;
968 }
969 spin_unlock_irqrestore(&subpage->lock, flags);
970 cur = page_start + cur_bit * fs_info->sectorsize;
971
972 eb = find_extent_buffer(fs_info, cur);
973 ASSERT(eb);
974 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
975 ASSERT(atomic_read(&eb->refs));
976 btrfs_assert_tree_write_locked(eb);
977 free_extent_buffer(eb);
978
979 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
980 }
981 return filemap_dirty_folio(mapping, folio);
982}
983#else
984#define btree_dirty_folio filemap_dirty_folio
985#endif
986
987static const struct address_space_operations btree_aops = {
988 .writepages = btree_writepages,
989 .release_folio = btree_release_folio,
990 .invalidate_folio = btree_invalidate_folio,
991 .migrate_folio = btree_migrate_folio,
992 .dirty_folio = btree_dirty_folio,
993};
994
995struct extent_buffer *btrfs_find_create_tree_block(
996 struct btrfs_fs_info *fs_info,
997 u64 bytenr, u64 owner_root,
998 int level)
999{
1000 if (btrfs_is_testing(fs_info))
1001 return alloc_test_extent_buffer(fs_info, bytenr);
1002 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1003}
1004
1005/*
1006 * Read tree block at logical address @bytenr and do variant basic but critical
1007 * verification.
1008 *
1009 * @check: expected tree parentness check, see comments of the
1010 * structure for details.
1011 */
1012struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1013 struct btrfs_tree_parent_check *check)
1014{
1015 struct extent_buffer *buf = NULL;
1016 int ret;
1017
1018 ASSERT(check);
1019
1020 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
1021 check->level);
1022 if (IS_ERR(buf))
1023 return buf;
1024
1025 ret = btrfs_read_extent_buffer(buf, check);
1026 if (ret) {
1027 free_extent_buffer_stale(buf);
1028 return ERR_PTR(ret);
1029 }
1030 if (btrfs_check_eb_owner(buf, check->owner_root)) {
1031 free_extent_buffer_stale(buf);
1032 return ERR_PTR(-EUCLEAN);
1033 }
1034 return buf;
1035
1036}
1037
1038void btrfs_clean_tree_block(struct extent_buffer *buf)
1039{
1040 struct btrfs_fs_info *fs_info = buf->fs_info;
1041 if (btrfs_header_generation(buf) ==
1042 fs_info->running_transaction->transid) {
1043 btrfs_assert_tree_write_locked(buf);
1044
1045 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1046 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1047 -buf->len,
1048 fs_info->dirty_metadata_batch);
1049 clear_extent_buffer_dirty(buf);
1050 }
1051 }
1052}
1053
1054static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1055 u64 objectid)
1056{
1057 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1058
1059 memset(&root->root_key, 0, sizeof(root->root_key));
1060 memset(&root->root_item, 0, sizeof(root->root_item));
1061 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1062 root->fs_info = fs_info;
1063 root->root_key.objectid = objectid;
1064 root->node = NULL;
1065 root->commit_root = NULL;
1066 root->state = 0;
1067 RB_CLEAR_NODE(&root->rb_node);
1068
1069 root->last_trans = 0;
1070 root->free_objectid = 0;
1071 root->nr_delalloc_inodes = 0;
1072 root->nr_ordered_extents = 0;
1073 root->inode_tree = RB_ROOT;
1074 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1075
1076 btrfs_init_root_block_rsv(root);
1077
1078 INIT_LIST_HEAD(&root->dirty_list);
1079 INIT_LIST_HEAD(&root->root_list);
1080 INIT_LIST_HEAD(&root->delalloc_inodes);
1081 INIT_LIST_HEAD(&root->delalloc_root);
1082 INIT_LIST_HEAD(&root->ordered_extents);
1083 INIT_LIST_HEAD(&root->ordered_root);
1084 INIT_LIST_HEAD(&root->reloc_dirty_list);
1085 INIT_LIST_HEAD(&root->logged_list[0]);
1086 INIT_LIST_HEAD(&root->logged_list[1]);
1087 spin_lock_init(&root->inode_lock);
1088 spin_lock_init(&root->delalloc_lock);
1089 spin_lock_init(&root->ordered_extent_lock);
1090 spin_lock_init(&root->accounting_lock);
1091 spin_lock_init(&root->log_extents_lock[0]);
1092 spin_lock_init(&root->log_extents_lock[1]);
1093 spin_lock_init(&root->qgroup_meta_rsv_lock);
1094 mutex_init(&root->objectid_mutex);
1095 mutex_init(&root->log_mutex);
1096 mutex_init(&root->ordered_extent_mutex);
1097 mutex_init(&root->delalloc_mutex);
1098 init_waitqueue_head(&root->qgroup_flush_wait);
1099 init_waitqueue_head(&root->log_writer_wait);
1100 init_waitqueue_head(&root->log_commit_wait[0]);
1101 init_waitqueue_head(&root->log_commit_wait[1]);
1102 INIT_LIST_HEAD(&root->log_ctxs[0]);
1103 INIT_LIST_HEAD(&root->log_ctxs[1]);
1104 atomic_set(&root->log_commit[0], 0);
1105 atomic_set(&root->log_commit[1], 0);
1106 atomic_set(&root->log_writers, 0);
1107 atomic_set(&root->log_batch, 0);
1108 refcount_set(&root->refs, 1);
1109 atomic_set(&root->snapshot_force_cow, 0);
1110 atomic_set(&root->nr_swapfiles, 0);
1111 root->log_transid = 0;
1112 root->log_transid_committed = -1;
1113 root->last_log_commit = 0;
1114 root->anon_dev = 0;
1115 if (!dummy) {
1116 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1117 IO_TREE_ROOT_DIRTY_LOG_PAGES);
1118 extent_io_tree_init(fs_info, &root->log_csum_range,
1119 IO_TREE_LOG_CSUM_RANGE);
1120 }
1121
1122 spin_lock_init(&root->root_item_lock);
1123 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1124#ifdef CONFIG_BTRFS_DEBUG
1125 INIT_LIST_HEAD(&root->leak_list);
1126 spin_lock(&fs_info->fs_roots_radix_lock);
1127 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1128 spin_unlock(&fs_info->fs_roots_radix_lock);
1129#endif
1130}
1131
1132static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1133 u64 objectid, gfp_t flags)
1134{
1135 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1136 if (root)
1137 __setup_root(root, fs_info, objectid);
1138 return root;
1139}
1140
1141#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1142/* Should only be used by the testing infrastructure */
1143struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1144{
1145 struct btrfs_root *root;
1146
1147 if (!fs_info)
1148 return ERR_PTR(-EINVAL);
1149
1150 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1151 if (!root)
1152 return ERR_PTR(-ENOMEM);
1153
1154 /* We don't use the stripesize in selftest, set it as sectorsize */
1155 root->alloc_bytenr = 0;
1156
1157 return root;
1158}
1159#endif
1160
1161static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1162{
1163 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1164 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1165
1166 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1167}
1168
1169static int global_root_key_cmp(const void *k, const struct rb_node *node)
1170{
1171 const struct btrfs_key *key = k;
1172 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1173
1174 return btrfs_comp_cpu_keys(key, &root->root_key);
1175}
1176
1177int btrfs_global_root_insert(struct btrfs_root *root)
1178{
1179 struct btrfs_fs_info *fs_info = root->fs_info;
1180 struct rb_node *tmp;
1181
1182 write_lock(&fs_info->global_root_lock);
1183 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1184 write_unlock(&fs_info->global_root_lock);
1185 ASSERT(!tmp);
1186
1187 return tmp ? -EEXIST : 0;
1188}
1189
1190void btrfs_global_root_delete(struct btrfs_root *root)
1191{
1192 struct btrfs_fs_info *fs_info = root->fs_info;
1193
1194 write_lock(&fs_info->global_root_lock);
1195 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1196 write_unlock(&fs_info->global_root_lock);
1197}
1198
1199struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1200 struct btrfs_key *key)
1201{
1202 struct rb_node *node;
1203 struct btrfs_root *root = NULL;
1204
1205 read_lock(&fs_info->global_root_lock);
1206 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1207 if (node)
1208 root = container_of(node, struct btrfs_root, rb_node);
1209 read_unlock(&fs_info->global_root_lock);
1210
1211 return root;
1212}
1213
1214static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1215{
1216 struct btrfs_block_group *block_group;
1217 u64 ret;
1218
1219 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1220 return 0;
1221
1222 if (bytenr)
1223 block_group = btrfs_lookup_block_group(fs_info, bytenr);
1224 else
1225 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1226 ASSERT(block_group);
1227 if (!block_group)
1228 return 0;
1229 ret = block_group->global_root_id;
1230 btrfs_put_block_group(block_group);
1231
1232 return ret;
1233}
1234
1235struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1236{
1237 struct btrfs_key key = {
1238 .objectid = BTRFS_CSUM_TREE_OBJECTID,
1239 .type = BTRFS_ROOT_ITEM_KEY,
1240 .offset = btrfs_global_root_id(fs_info, bytenr),
1241 };
1242
1243 return btrfs_global_root(fs_info, &key);
1244}
1245
1246struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1247{
1248 struct btrfs_key key = {
1249 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
1250 .type = BTRFS_ROOT_ITEM_KEY,
1251 .offset = btrfs_global_root_id(fs_info, bytenr),
1252 };
1253
1254 return btrfs_global_root(fs_info, &key);
1255}
1256
1257struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
1258{
1259 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
1260 return fs_info->block_group_root;
1261 return btrfs_extent_root(fs_info, 0);
1262}
1263
1264struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1265 u64 objectid)
1266{
1267 struct btrfs_fs_info *fs_info = trans->fs_info;
1268 struct extent_buffer *leaf;
1269 struct btrfs_root *tree_root = fs_info->tree_root;
1270 struct btrfs_root *root;
1271 struct btrfs_key key;
1272 unsigned int nofs_flag;
1273 int ret = 0;
1274
1275 /*
1276 * We're holding a transaction handle, so use a NOFS memory allocation
1277 * context to avoid deadlock if reclaim happens.
1278 */
1279 nofs_flag = memalloc_nofs_save();
1280 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1281 memalloc_nofs_restore(nofs_flag);
1282 if (!root)
1283 return ERR_PTR(-ENOMEM);
1284
1285 root->root_key.objectid = objectid;
1286 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1287 root->root_key.offset = 0;
1288
1289 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1290 BTRFS_NESTING_NORMAL);
1291 if (IS_ERR(leaf)) {
1292 ret = PTR_ERR(leaf);
1293 leaf = NULL;
1294 goto fail;
1295 }
1296
1297 root->node = leaf;
1298 btrfs_mark_buffer_dirty(leaf);
1299
1300 root->commit_root = btrfs_root_node(root);
1301 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1302
1303 btrfs_set_root_flags(&root->root_item, 0);
1304 btrfs_set_root_limit(&root->root_item, 0);
1305 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1306 btrfs_set_root_generation(&root->root_item, trans->transid);
1307 btrfs_set_root_level(&root->root_item, 0);
1308 btrfs_set_root_refs(&root->root_item, 1);
1309 btrfs_set_root_used(&root->root_item, leaf->len);
1310 btrfs_set_root_last_snapshot(&root->root_item, 0);
1311 btrfs_set_root_dirid(&root->root_item, 0);
1312 if (is_fstree(objectid))
1313 generate_random_guid(root->root_item.uuid);
1314 else
1315 export_guid(root->root_item.uuid, &guid_null);
1316 btrfs_set_root_drop_level(&root->root_item, 0);
1317
1318 btrfs_tree_unlock(leaf);
1319
1320 key.objectid = objectid;
1321 key.type = BTRFS_ROOT_ITEM_KEY;
1322 key.offset = 0;
1323 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1324 if (ret)
1325 goto fail;
1326
1327 return root;
1328
1329fail:
1330 btrfs_put_root(root);
1331
1332 return ERR_PTR(ret);
1333}
1334
1335static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1336 struct btrfs_fs_info *fs_info)
1337{
1338 struct btrfs_root *root;
1339
1340 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1341 if (!root)
1342 return ERR_PTR(-ENOMEM);
1343
1344 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1345 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1346 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1347
1348 return root;
1349}
1350
1351int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1352 struct btrfs_root *root)
1353{
1354 struct extent_buffer *leaf;
1355
1356 /*
1357 * DON'T set SHAREABLE bit for log trees.
1358 *
1359 * Log trees are not exposed to user space thus can't be snapshotted,
1360 * and they go away before a real commit is actually done.
1361 *
1362 * They do store pointers to file data extents, and those reference
1363 * counts still get updated (along with back refs to the log tree).
1364 */
1365
1366 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1367 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1368 if (IS_ERR(leaf))
1369 return PTR_ERR(leaf);
1370
1371 root->node = leaf;
1372
1373 btrfs_mark_buffer_dirty(root->node);
1374 btrfs_tree_unlock(root->node);
1375
1376 return 0;
1377}
1378
1379int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1380 struct btrfs_fs_info *fs_info)
1381{
1382 struct btrfs_root *log_root;
1383
1384 log_root = alloc_log_tree(trans, fs_info);
1385 if (IS_ERR(log_root))
1386 return PTR_ERR(log_root);
1387
1388 if (!btrfs_is_zoned(fs_info)) {
1389 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1390
1391 if (ret) {
1392 btrfs_put_root(log_root);
1393 return ret;
1394 }
1395 }
1396
1397 WARN_ON(fs_info->log_root_tree);
1398 fs_info->log_root_tree = log_root;
1399 return 0;
1400}
1401
1402int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1403 struct btrfs_root *root)
1404{
1405 struct btrfs_fs_info *fs_info = root->fs_info;
1406 struct btrfs_root *log_root;
1407 struct btrfs_inode_item *inode_item;
1408 int ret;
1409
1410 log_root = alloc_log_tree(trans, fs_info);
1411 if (IS_ERR(log_root))
1412 return PTR_ERR(log_root);
1413
1414 ret = btrfs_alloc_log_tree_node(trans, log_root);
1415 if (ret) {
1416 btrfs_put_root(log_root);
1417 return ret;
1418 }
1419
1420 log_root->last_trans = trans->transid;
1421 log_root->root_key.offset = root->root_key.objectid;
1422
1423 inode_item = &log_root->root_item.inode;
1424 btrfs_set_stack_inode_generation(inode_item, 1);
1425 btrfs_set_stack_inode_size(inode_item, 3);
1426 btrfs_set_stack_inode_nlink(inode_item, 1);
1427 btrfs_set_stack_inode_nbytes(inode_item,
1428 fs_info->nodesize);
1429 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1430
1431 btrfs_set_root_node(&log_root->root_item, log_root->node);
1432
1433 WARN_ON(root->log_root);
1434 root->log_root = log_root;
1435 root->log_transid = 0;
1436 root->log_transid_committed = -1;
1437 root->last_log_commit = 0;
1438 return 0;
1439}
1440
1441static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1442 struct btrfs_path *path,
1443 struct btrfs_key *key)
1444{
1445 struct btrfs_root *root;
1446 struct btrfs_tree_parent_check check = { 0 };
1447 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1448 u64 generation;
1449 int ret;
1450 int level;
1451
1452 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1453 if (!root)
1454 return ERR_PTR(-ENOMEM);
1455
1456 ret = btrfs_find_root(tree_root, key, path,
1457 &root->root_item, &root->root_key);
1458 if (ret) {
1459 if (ret > 0)
1460 ret = -ENOENT;
1461 goto fail;
1462 }
1463
1464 generation = btrfs_root_generation(&root->root_item);
1465 level = btrfs_root_level(&root->root_item);
1466 check.level = level;
1467 check.transid = generation;
1468 check.owner_root = key->objectid;
1469 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1470 &check);
1471 if (IS_ERR(root->node)) {
1472 ret = PTR_ERR(root->node);
1473 root->node = NULL;
1474 goto fail;
1475 }
1476 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1477 ret = -EIO;
1478 goto fail;
1479 }
1480
1481 /*
1482 * For real fs, and not log/reloc trees, root owner must
1483 * match its root node owner
1484 */
1485 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1486 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1487 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1488 root->root_key.objectid != btrfs_header_owner(root->node)) {
1489 btrfs_crit(fs_info,
1490"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1491 root->root_key.objectid, root->node->start,
1492 btrfs_header_owner(root->node),
1493 root->root_key.objectid);
1494 ret = -EUCLEAN;
1495 goto fail;
1496 }
1497 root->commit_root = btrfs_root_node(root);
1498 return root;
1499fail:
1500 btrfs_put_root(root);
1501 return ERR_PTR(ret);
1502}
1503
1504struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1505 struct btrfs_key *key)
1506{
1507 struct btrfs_root *root;
1508 struct btrfs_path *path;
1509
1510 path = btrfs_alloc_path();
1511 if (!path)
1512 return ERR_PTR(-ENOMEM);
1513 root = read_tree_root_path(tree_root, path, key);
1514 btrfs_free_path(path);
1515
1516 return root;
1517}
1518
1519/*
1520 * Initialize subvolume root in-memory structure
1521 *
1522 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1523 */
1524static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1525{
1526 int ret;
1527 unsigned int nofs_flag;
1528
1529 /*
1530 * We might be called under a transaction (e.g. indirect backref
1531 * resolution) which could deadlock if it triggers memory reclaim
1532 */
1533 nofs_flag = memalloc_nofs_save();
1534 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1535 memalloc_nofs_restore(nofs_flag);
1536 if (ret)
1537 goto fail;
1538
1539 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1540 !btrfs_is_data_reloc_root(root)) {
1541 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1542 btrfs_check_and_init_root_item(&root->root_item);
1543 }
1544
1545 /*
1546 * Don't assign anonymous block device to roots that are not exposed to
1547 * userspace, the id pool is limited to 1M
1548 */
1549 if (is_fstree(root->root_key.objectid) &&
1550 btrfs_root_refs(&root->root_item) > 0) {
1551 if (!anon_dev) {
1552 ret = get_anon_bdev(&root->anon_dev);
1553 if (ret)
1554 goto fail;
1555 } else {
1556 root->anon_dev = anon_dev;
1557 }
1558 }
1559
1560 mutex_lock(&root->objectid_mutex);
1561 ret = btrfs_init_root_free_objectid(root);
1562 if (ret) {
1563 mutex_unlock(&root->objectid_mutex);
1564 goto fail;
1565 }
1566
1567 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1568
1569 mutex_unlock(&root->objectid_mutex);
1570
1571 return 0;
1572fail:
1573 /* The caller is responsible to call btrfs_free_fs_root */
1574 return ret;
1575}
1576
1577static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1578 u64 root_id)
1579{
1580 struct btrfs_root *root;
1581
1582 spin_lock(&fs_info->fs_roots_radix_lock);
1583 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1584 (unsigned long)root_id);
1585 if (root)
1586 root = btrfs_grab_root(root);
1587 spin_unlock(&fs_info->fs_roots_radix_lock);
1588 return root;
1589}
1590
1591static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1592 u64 objectid)
1593{
1594 struct btrfs_key key = {
1595 .objectid = objectid,
1596 .type = BTRFS_ROOT_ITEM_KEY,
1597 .offset = 0,
1598 };
1599
1600 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1601 return btrfs_grab_root(fs_info->tree_root);
1602 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1603 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1604 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1605 return btrfs_grab_root(fs_info->chunk_root);
1606 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1607 return btrfs_grab_root(fs_info->dev_root);
1608 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1609 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1610 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1611 return btrfs_grab_root(fs_info->quota_root) ?
1612 fs_info->quota_root : ERR_PTR(-ENOENT);
1613 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1614 return btrfs_grab_root(fs_info->uuid_root) ?
1615 fs_info->uuid_root : ERR_PTR(-ENOENT);
1616 if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID)
1617 return btrfs_grab_root(fs_info->block_group_root) ?
1618 fs_info->block_group_root : ERR_PTR(-ENOENT);
1619 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1620 struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1621
1622 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1623 }
1624 return NULL;
1625}
1626
1627int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1628 struct btrfs_root *root)
1629{
1630 int ret;
1631
1632 ret = radix_tree_preload(GFP_NOFS);
1633 if (ret)
1634 return ret;
1635
1636 spin_lock(&fs_info->fs_roots_radix_lock);
1637 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1638 (unsigned long)root->root_key.objectid,
1639 root);
1640 if (ret == 0) {
1641 btrfs_grab_root(root);
1642 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1643 }
1644 spin_unlock(&fs_info->fs_roots_radix_lock);
1645 radix_tree_preload_end();
1646
1647 return ret;
1648}
1649
1650void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1651{
1652#ifdef CONFIG_BTRFS_DEBUG
1653 struct btrfs_root *root;
1654
1655 while (!list_empty(&fs_info->allocated_roots)) {
1656 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1657
1658 root = list_first_entry(&fs_info->allocated_roots,
1659 struct btrfs_root, leak_list);
1660 btrfs_err(fs_info, "leaked root %s refcount %d",
1661 btrfs_root_name(&root->root_key, buf),
1662 refcount_read(&root->refs));
1663 while (refcount_read(&root->refs) > 1)
1664 btrfs_put_root(root);
1665 btrfs_put_root(root);
1666 }
1667#endif
1668}
1669
1670static void free_global_roots(struct btrfs_fs_info *fs_info)
1671{
1672 struct btrfs_root *root;
1673 struct rb_node *node;
1674
1675 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1676 root = rb_entry(node, struct btrfs_root, rb_node);
1677 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1678 btrfs_put_root(root);
1679 }
1680}
1681
1682void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1683{
1684 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1685 percpu_counter_destroy(&fs_info->delalloc_bytes);
1686 percpu_counter_destroy(&fs_info->ordered_bytes);
1687 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1688 btrfs_free_csum_hash(fs_info);
1689 btrfs_free_stripe_hash_table(fs_info);
1690 btrfs_free_ref_cache(fs_info);
1691 kfree(fs_info->balance_ctl);
1692 kfree(fs_info->delayed_root);
1693 free_global_roots(fs_info);
1694 btrfs_put_root(fs_info->tree_root);
1695 btrfs_put_root(fs_info->chunk_root);
1696 btrfs_put_root(fs_info->dev_root);
1697 btrfs_put_root(fs_info->quota_root);
1698 btrfs_put_root(fs_info->uuid_root);
1699 btrfs_put_root(fs_info->fs_root);
1700 btrfs_put_root(fs_info->data_reloc_root);
1701 btrfs_put_root(fs_info->block_group_root);
1702 btrfs_check_leaked_roots(fs_info);
1703 btrfs_extent_buffer_leak_debug_check(fs_info);
1704 kfree(fs_info->super_copy);
1705 kfree(fs_info->super_for_commit);
1706 kfree(fs_info->subpage_info);
1707 kvfree(fs_info);
1708}
1709
1710
1711/*
1712 * Get an in-memory reference of a root structure.
1713 *
1714 * For essential trees like root/extent tree, we grab it from fs_info directly.
1715 * For subvolume trees, we check the cached filesystem roots first. If not
1716 * found, then read it from disk and add it to cached fs roots.
1717 *
1718 * Caller should release the root by calling btrfs_put_root() after the usage.
1719 *
1720 * NOTE: Reloc and log trees can't be read by this function as they share the
1721 * same root objectid.
1722 *
1723 * @objectid: root id
1724 * @anon_dev: preallocated anonymous block device number for new roots,
1725 * pass 0 for new allocation.
1726 * @check_ref: whether to check root item references, If true, return -ENOENT
1727 * for orphan roots
1728 */
1729static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1730 u64 objectid, dev_t anon_dev,
1731 bool check_ref)
1732{
1733 struct btrfs_root *root;
1734 struct btrfs_path *path;
1735 struct btrfs_key key;
1736 int ret;
1737
1738 root = btrfs_get_global_root(fs_info, objectid);
1739 if (root)
1740 return root;
1741again:
1742 root = btrfs_lookup_fs_root(fs_info, objectid);
1743 if (root) {
1744 /* Shouldn't get preallocated anon_dev for cached roots */
1745 ASSERT(!anon_dev);
1746 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1747 btrfs_put_root(root);
1748 return ERR_PTR(-ENOENT);
1749 }
1750 return root;
1751 }
1752
1753 key.objectid = objectid;
1754 key.type = BTRFS_ROOT_ITEM_KEY;
1755 key.offset = (u64)-1;
1756 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1757 if (IS_ERR(root))
1758 return root;
1759
1760 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1761 ret = -ENOENT;
1762 goto fail;
1763 }
1764
1765 ret = btrfs_init_fs_root(root, anon_dev);
1766 if (ret)
1767 goto fail;
1768
1769 path = btrfs_alloc_path();
1770 if (!path) {
1771 ret = -ENOMEM;
1772 goto fail;
1773 }
1774 key.objectid = BTRFS_ORPHAN_OBJECTID;
1775 key.type = BTRFS_ORPHAN_ITEM_KEY;
1776 key.offset = objectid;
1777
1778 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1779 btrfs_free_path(path);
1780 if (ret < 0)
1781 goto fail;
1782 if (ret == 0)
1783 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1784
1785 ret = btrfs_insert_fs_root(fs_info, root);
1786 if (ret) {
1787 if (ret == -EEXIST) {
1788 btrfs_put_root(root);
1789 goto again;
1790 }
1791 goto fail;
1792 }
1793 return root;
1794fail:
1795 /*
1796 * If our caller provided us an anonymous device, then it's his
1797 * responsibility to free it in case we fail. So we have to set our
1798 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1799 * and once again by our caller.
1800 */
1801 if (anon_dev)
1802 root->anon_dev = 0;
1803 btrfs_put_root(root);
1804 return ERR_PTR(ret);
1805}
1806
1807/*
1808 * Get in-memory reference of a root structure
1809 *
1810 * @objectid: tree objectid
1811 * @check_ref: if set, verify that the tree exists and the item has at least
1812 * one reference
1813 */
1814struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1815 u64 objectid, bool check_ref)
1816{
1817 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1818}
1819
1820/*
1821 * Get in-memory reference of a root structure, created as new, optionally pass
1822 * the anonymous block device id
1823 *
1824 * @objectid: tree objectid
1825 * @anon_dev: if zero, allocate a new anonymous block device or use the
1826 * parameter value
1827 */
1828struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1829 u64 objectid, dev_t anon_dev)
1830{
1831 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1832}
1833
1834/*
1835 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1836 * @fs_info: the fs_info
1837 * @objectid: the objectid we need to lookup
1838 *
1839 * This is exclusively used for backref walking, and exists specifically because
1840 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1841 * creation time, which means we may have to read the tree_root in order to look
1842 * up a fs root that is not in memory. If the root is not in memory we will
1843 * read the tree root commit root and look up the fs root from there. This is a
1844 * temporary root, it will not be inserted into the radix tree as it doesn't
1845 * have the most uptodate information, it'll simply be discarded once the
1846 * backref code is finished using the root.
1847 */
1848struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1849 struct btrfs_path *path,
1850 u64 objectid)
1851{
1852 struct btrfs_root *root;
1853 struct btrfs_key key;
1854
1855 ASSERT(path->search_commit_root && path->skip_locking);
1856
1857 /*
1858 * This can return -ENOENT if we ask for a root that doesn't exist, but
1859 * since this is called via the backref walking code we won't be looking
1860 * up a root that doesn't exist, unless there's corruption. So if root
1861 * != NULL just return it.
1862 */
1863 root = btrfs_get_global_root(fs_info, objectid);
1864 if (root)
1865 return root;
1866
1867 root = btrfs_lookup_fs_root(fs_info, objectid);
1868 if (root)
1869 return root;
1870
1871 key.objectid = objectid;
1872 key.type = BTRFS_ROOT_ITEM_KEY;
1873 key.offset = (u64)-1;
1874 root = read_tree_root_path(fs_info->tree_root, path, &key);
1875 btrfs_release_path(path);
1876
1877 return root;
1878}
1879
1880static int cleaner_kthread(void *arg)
1881{
1882 struct btrfs_fs_info *fs_info = arg;
1883 int again;
1884
1885 while (1) {
1886 again = 0;
1887
1888 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1889
1890 /* Make the cleaner go to sleep early. */
1891 if (btrfs_need_cleaner_sleep(fs_info))
1892 goto sleep;
1893
1894 /*
1895 * Do not do anything if we might cause open_ctree() to block
1896 * before we have finished mounting the filesystem.
1897 */
1898 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1899 goto sleep;
1900
1901 if (!mutex_trylock(&fs_info->cleaner_mutex))
1902 goto sleep;
1903
1904 /*
1905 * Avoid the problem that we change the status of the fs
1906 * during the above check and trylock.
1907 */
1908 if (btrfs_need_cleaner_sleep(fs_info)) {
1909 mutex_unlock(&fs_info->cleaner_mutex);
1910 goto sleep;
1911 }
1912
1913 btrfs_run_delayed_iputs(fs_info);
1914
1915 again = btrfs_clean_one_deleted_snapshot(fs_info);
1916 mutex_unlock(&fs_info->cleaner_mutex);
1917
1918 /*
1919 * The defragger has dealt with the R/O remount and umount,
1920 * needn't do anything special here.
1921 */
1922 btrfs_run_defrag_inodes(fs_info);
1923
1924 /*
1925 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1926 * with relocation (btrfs_relocate_chunk) and relocation
1927 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1928 * after acquiring fs_info->reclaim_bgs_lock. So we
1929 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1930 * unused block groups.
1931 */
1932 btrfs_delete_unused_bgs(fs_info);
1933
1934 /*
1935 * Reclaim block groups in the reclaim_bgs list after we deleted
1936 * all unused block_groups. This possibly gives us some more free
1937 * space.
1938 */
1939 btrfs_reclaim_bgs(fs_info);
1940sleep:
1941 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1942 if (kthread_should_park())
1943 kthread_parkme();
1944 if (kthread_should_stop())
1945 return 0;
1946 if (!again) {
1947 set_current_state(TASK_INTERRUPTIBLE);
1948 schedule();
1949 __set_current_state(TASK_RUNNING);
1950 }
1951 }
1952}
1953
1954static int transaction_kthread(void *arg)
1955{
1956 struct btrfs_root *root = arg;
1957 struct btrfs_fs_info *fs_info = root->fs_info;
1958 struct btrfs_trans_handle *trans;
1959 struct btrfs_transaction *cur;
1960 u64 transid;
1961 time64_t delta;
1962 unsigned long delay;
1963 bool cannot_commit;
1964
1965 do {
1966 cannot_commit = false;
1967 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1968 mutex_lock(&fs_info->transaction_kthread_mutex);
1969
1970 spin_lock(&fs_info->trans_lock);
1971 cur = fs_info->running_transaction;
1972 if (!cur) {
1973 spin_unlock(&fs_info->trans_lock);
1974 goto sleep;
1975 }
1976
1977 delta = ktime_get_seconds() - cur->start_time;
1978 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1979 cur->state < TRANS_STATE_COMMIT_START &&
1980 delta < fs_info->commit_interval) {
1981 spin_unlock(&fs_info->trans_lock);
1982 delay -= msecs_to_jiffies((delta - 1) * 1000);
1983 delay = min(delay,
1984 msecs_to_jiffies(fs_info->commit_interval * 1000));
1985 goto sleep;
1986 }
1987 transid = cur->transid;
1988 spin_unlock(&fs_info->trans_lock);
1989
1990 /* If the file system is aborted, this will always fail. */
1991 trans = btrfs_attach_transaction(root);
1992 if (IS_ERR(trans)) {
1993 if (PTR_ERR(trans) != -ENOENT)
1994 cannot_commit = true;
1995 goto sleep;
1996 }
1997 if (transid == trans->transid) {
1998 btrfs_commit_transaction(trans);
1999 } else {
2000 btrfs_end_transaction(trans);
2001 }
2002sleep:
2003 wake_up_process(fs_info->cleaner_kthread);
2004 mutex_unlock(&fs_info->transaction_kthread_mutex);
2005
2006 if (BTRFS_FS_ERROR(fs_info))
2007 btrfs_cleanup_transaction(fs_info);
2008 if (!kthread_should_stop() &&
2009 (!btrfs_transaction_blocked(fs_info) ||
2010 cannot_commit))
2011 schedule_timeout_interruptible(delay);
2012 } while (!kthread_should_stop());
2013 return 0;
2014}
2015
2016/*
2017 * This will find the highest generation in the array of root backups. The
2018 * index of the highest array is returned, or -EINVAL if we can't find
2019 * anything.
2020 *
2021 * We check to make sure the array is valid by comparing the
2022 * generation of the latest root in the array with the generation
2023 * in the super block. If they don't match we pitch it.
2024 */
2025static int find_newest_super_backup(struct btrfs_fs_info *info)
2026{
2027 const u64 newest_gen = btrfs_super_generation(info->super_copy);
2028 u64 cur;
2029 struct btrfs_root_backup *root_backup;
2030 int i;
2031
2032 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2033 root_backup = info->super_copy->super_roots + i;
2034 cur = btrfs_backup_tree_root_gen(root_backup);
2035 if (cur == newest_gen)
2036 return i;
2037 }
2038
2039 return -EINVAL;
2040}
2041
2042/*
2043 * copy all the root pointers into the super backup array.
2044 * this will bump the backup pointer by one when it is
2045 * done
2046 */
2047static void backup_super_roots(struct btrfs_fs_info *info)
2048{
2049 const int next_backup = info->backup_root_index;
2050 struct btrfs_root_backup *root_backup;
2051
2052 root_backup = info->super_for_commit->super_roots + next_backup;
2053
2054 /*
2055 * make sure all of our padding and empty slots get zero filled
2056 * regardless of which ones we use today
2057 */
2058 memset(root_backup, 0, sizeof(*root_backup));
2059
2060 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2061
2062 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2063 btrfs_set_backup_tree_root_gen(root_backup,
2064 btrfs_header_generation(info->tree_root->node));
2065
2066 btrfs_set_backup_tree_root_level(root_backup,
2067 btrfs_header_level(info->tree_root->node));
2068
2069 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2070 btrfs_set_backup_chunk_root_gen(root_backup,
2071 btrfs_header_generation(info->chunk_root->node));
2072 btrfs_set_backup_chunk_root_level(root_backup,
2073 btrfs_header_level(info->chunk_root->node));
2074
2075 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
2076 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2077 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2078
2079 btrfs_set_backup_extent_root(root_backup,
2080 extent_root->node->start);
2081 btrfs_set_backup_extent_root_gen(root_backup,
2082 btrfs_header_generation(extent_root->node));
2083 btrfs_set_backup_extent_root_level(root_backup,
2084 btrfs_header_level(extent_root->node));
2085
2086 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2087 btrfs_set_backup_csum_root_gen(root_backup,
2088 btrfs_header_generation(csum_root->node));
2089 btrfs_set_backup_csum_root_level(root_backup,
2090 btrfs_header_level(csum_root->node));
2091 }
2092
2093 /*
2094 * we might commit during log recovery, which happens before we set
2095 * the fs_root. Make sure it is valid before we fill it in.
2096 */
2097 if (info->fs_root && info->fs_root->node) {
2098 btrfs_set_backup_fs_root(root_backup,
2099 info->fs_root->node->start);
2100 btrfs_set_backup_fs_root_gen(root_backup,
2101 btrfs_header_generation(info->fs_root->node));
2102 btrfs_set_backup_fs_root_level(root_backup,
2103 btrfs_header_level(info->fs_root->node));
2104 }
2105
2106 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2107 btrfs_set_backup_dev_root_gen(root_backup,
2108 btrfs_header_generation(info->dev_root->node));
2109 btrfs_set_backup_dev_root_level(root_backup,
2110 btrfs_header_level(info->dev_root->node));
2111
2112 btrfs_set_backup_total_bytes(root_backup,
2113 btrfs_super_total_bytes(info->super_copy));
2114 btrfs_set_backup_bytes_used(root_backup,
2115 btrfs_super_bytes_used(info->super_copy));
2116 btrfs_set_backup_num_devices(root_backup,
2117 btrfs_super_num_devices(info->super_copy));
2118
2119 /*
2120 * if we don't copy this out to the super_copy, it won't get remembered
2121 * for the next commit
2122 */
2123 memcpy(&info->super_copy->super_roots,
2124 &info->super_for_commit->super_roots,
2125 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2126}
2127
2128/*
2129 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2130 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2131 *
2132 * fs_info - filesystem whose backup roots need to be read
2133 * priority - priority of backup root required
2134 *
2135 * Returns backup root index on success and -EINVAL otherwise.
2136 */
2137static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2138{
2139 int backup_index = find_newest_super_backup(fs_info);
2140 struct btrfs_super_block *super = fs_info->super_copy;
2141 struct btrfs_root_backup *root_backup;
2142
2143 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2144 if (priority == 0)
2145 return backup_index;
2146
2147 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2148 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2149 } else {
2150 return -EINVAL;
2151 }
2152
2153 root_backup = super->super_roots + backup_index;
2154
2155 btrfs_set_super_generation(super,
2156 btrfs_backup_tree_root_gen(root_backup));
2157 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2158 btrfs_set_super_root_level(super,
2159 btrfs_backup_tree_root_level(root_backup));
2160 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2161
2162 /*
2163 * Fixme: the total bytes and num_devices need to match or we should
2164 * need a fsck
2165 */
2166 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2167 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2168
2169 return backup_index;
2170}
2171
2172/* helper to cleanup workers */
2173static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2174{
2175 btrfs_destroy_workqueue(fs_info->fixup_workers);
2176 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2177 btrfs_destroy_workqueue(fs_info->hipri_workers);
2178 btrfs_destroy_workqueue(fs_info->workers);
2179 if (fs_info->endio_workers)
2180 destroy_workqueue(fs_info->endio_workers);
2181 if (fs_info->rmw_workers)
2182 destroy_workqueue(fs_info->rmw_workers);
2183 if (fs_info->compressed_write_workers)
2184 destroy_workqueue(fs_info->compressed_write_workers);
2185 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2186 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2187 btrfs_destroy_workqueue(fs_info->delayed_workers);
2188 btrfs_destroy_workqueue(fs_info->caching_workers);
2189 btrfs_destroy_workqueue(fs_info->flush_workers);
2190 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2191 if (fs_info->discard_ctl.discard_workers)
2192 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2193 /*
2194 * Now that all other work queues are destroyed, we can safely destroy
2195 * the queues used for metadata I/O, since tasks from those other work
2196 * queues can do metadata I/O operations.
2197 */
2198 if (fs_info->endio_meta_workers)
2199 destroy_workqueue(fs_info->endio_meta_workers);
2200}
2201
2202static void free_root_extent_buffers(struct btrfs_root *root)
2203{
2204 if (root) {
2205 free_extent_buffer(root->node);
2206 free_extent_buffer(root->commit_root);
2207 root->node = NULL;
2208 root->commit_root = NULL;
2209 }
2210}
2211
2212static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2213{
2214 struct btrfs_root *root, *tmp;
2215
2216 rbtree_postorder_for_each_entry_safe(root, tmp,
2217 &fs_info->global_root_tree,
2218 rb_node)
2219 free_root_extent_buffers(root);
2220}
2221
2222/* helper to cleanup tree roots */
2223static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2224{
2225 free_root_extent_buffers(info->tree_root);
2226
2227 free_global_root_pointers(info);
2228 free_root_extent_buffers(info->dev_root);
2229 free_root_extent_buffers(info->quota_root);
2230 free_root_extent_buffers(info->uuid_root);
2231 free_root_extent_buffers(info->fs_root);
2232 free_root_extent_buffers(info->data_reloc_root);
2233 free_root_extent_buffers(info->block_group_root);
2234 if (free_chunk_root)
2235 free_root_extent_buffers(info->chunk_root);
2236}
2237
2238void btrfs_put_root(struct btrfs_root *root)
2239{
2240 if (!root)
2241 return;
2242
2243 if (refcount_dec_and_test(&root->refs)) {
2244 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2245 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2246 if (root->anon_dev)
2247 free_anon_bdev(root->anon_dev);
2248 btrfs_drew_lock_destroy(&root->snapshot_lock);
2249 free_root_extent_buffers(root);
2250#ifdef CONFIG_BTRFS_DEBUG
2251 spin_lock(&root->fs_info->fs_roots_radix_lock);
2252 list_del_init(&root->leak_list);
2253 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2254#endif
2255 kfree(root);
2256 }
2257}
2258
2259void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2260{
2261 int ret;
2262 struct btrfs_root *gang[8];
2263 int i;
2264
2265 while (!list_empty(&fs_info->dead_roots)) {
2266 gang[0] = list_entry(fs_info->dead_roots.next,
2267 struct btrfs_root, root_list);
2268 list_del(&gang[0]->root_list);
2269
2270 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2271 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2272 btrfs_put_root(gang[0]);
2273 }
2274
2275 while (1) {
2276 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2277 (void **)gang, 0,
2278 ARRAY_SIZE(gang));
2279 if (!ret)
2280 break;
2281 for (i = 0; i < ret; i++)
2282 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2283 }
2284}
2285
2286static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2287{
2288 mutex_init(&fs_info->scrub_lock);
2289 atomic_set(&fs_info->scrubs_running, 0);
2290 atomic_set(&fs_info->scrub_pause_req, 0);
2291 atomic_set(&fs_info->scrubs_paused, 0);
2292 atomic_set(&fs_info->scrub_cancel_req, 0);
2293 init_waitqueue_head(&fs_info->scrub_pause_wait);
2294 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2295}
2296
2297static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2298{
2299 spin_lock_init(&fs_info->balance_lock);
2300 mutex_init(&fs_info->balance_mutex);
2301 atomic_set(&fs_info->balance_pause_req, 0);
2302 atomic_set(&fs_info->balance_cancel_req, 0);
2303 fs_info->balance_ctl = NULL;
2304 init_waitqueue_head(&fs_info->balance_wait_q);
2305 atomic_set(&fs_info->reloc_cancel_req, 0);
2306}
2307
2308static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2309{
2310 struct inode *inode = fs_info->btree_inode;
2311 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
2312 fs_info->tree_root);
2313
2314 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2315 set_nlink(inode, 1);
2316 /*
2317 * we set the i_size on the btree inode to the max possible int.
2318 * the real end of the address space is determined by all of
2319 * the devices in the system
2320 */
2321 inode->i_size = OFFSET_MAX;
2322 inode->i_mapping->a_ops = &btree_aops;
2323
2324 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2325 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2326 IO_TREE_BTREE_INODE_IO);
2327 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2328
2329 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2330 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2331 BTRFS_I(inode)->location.type = 0;
2332 BTRFS_I(inode)->location.offset = 0;
2333 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2334 __insert_inode_hash(inode, hash);
2335}
2336
2337static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2338{
2339 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2340 init_rwsem(&fs_info->dev_replace.rwsem);
2341 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2342}
2343
2344static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2345{
2346 spin_lock_init(&fs_info->qgroup_lock);
2347 mutex_init(&fs_info->qgroup_ioctl_lock);
2348 fs_info->qgroup_tree = RB_ROOT;
2349 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2350 fs_info->qgroup_seq = 1;
2351 fs_info->qgroup_ulist = NULL;
2352 fs_info->qgroup_rescan_running = false;
2353 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
2354 mutex_init(&fs_info->qgroup_rescan_lock);
2355}
2356
2357static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2358{
2359 u32 max_active = fs_info->thread_pool_size;
2360 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2361
2362 fs_info->workers =
2363 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2364 fs_info->hipri_workers =
2365 btrfs_alloc_workqueue(fs_info, "worker-high",
2366 flags | WQ_HIGHPRI, max_active, 16);
2367
2368 fs_info->delalloc_workers =
2369 btrfs_alloc_workqueue(fs_info, "delalloc",
2370 flags, max_active, 2);
2371
2372 fs_info->flush_workers =
2373 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2374 flags, max_active, 0);
2375
2376 fs_info->caching_workers =
2377 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2378
2379 fs_info->fixup_workers =
2380 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2381
2382 fs_info->endio_workers =
2383 alloc_workqueue("btrfs-endio", flags, max_active);
2384 fs_info->endio_meta_workers =
2385 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2386 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2387 fs_info->endio_write_workers =
2388 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2389 max_active, 2);
2390 fs_info->compressed_write_workers =
2391 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2392 fs_info->endio_freespace_worker =
2393 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2394 max_active, 0);
2395 fs_info->delayed_workers =
2396 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2397 max_active, 0);
2398 fs_info->qgroup_rescan_workers =
2399 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2400 fs_info->discard_ctl.discard_workers =
2401 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2402
2403 if (!(fs_info->workers && fs_info->hipri_workers &&
2404 fs_info->delalloc_workers && fs_info->flush_workers &&
2405 fs_info->endio_workers && fs_info->endio_meta_workers &&
2406 fs_info->compressed_write_workers &&
2407 fs_info->endio_write_workers &&
2408 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2409 fs_info->caching_workers && fs_info->fixup_workers &&
2410 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2411 fs_info->discard_ctl.discard_workers)) {
2412 return -ENOMEM;
2413 }
2414
2415 return 0;
2416}
2417
2418static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2419{
2420 struct crypto_shash *csum_shash;
2421 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2422
2423 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2424
2425 if (IS_ERR(csum_shash)) {
2426 btrfs_err(fs_info, "error allocating %s hash for checksum",
2427 csum_driver);
2428 return PTR_ERR(csum_shash);
2429 }
2430
2431 fs_info->csum_shash = csum_shash;
2432
2433 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2434 btrfs_super_csum_name(csum_type),
2435 crypto_shash_driver_name(csum_shash));
2436 return 0;
2437}
2438
2439static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2440 struct btrfs_fs_devices *fs_devices)
2441{
2442 int ret;
2443 struct btrfs_tree_parent_check check = { 0 };
2444 struct btrfs_root *log_tree_root;
2445 struct btrfs_super_block *disk_super = fs_info->super_copy;
2446 u64 bytenr = btrfs_super_log_root(disk_super);
2447 int level = btrfs_super_log_root_level(disk_super);
2448
2449 if (fs_devices->rw_devices == 0) {
2450 btrfs_warn(fs_info, "log replay required on RO media");
2451 return -EIO;
2452 }
2453
2454 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2455 GFP_KERNEL);
2456 if (!log_tree_root)
2457 return -ENOMEM;
2458
2459 check.level = level;
2460 check.transid = fs_info->generation + 1;
2461 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2462 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2463 if (IS_ERR(log_tree_root->node)) {
2464 btrfs_warn(fs_info, "failed to read log tree");
2465 ret = PTR_ERR(log_tree_root->node);
2466 log_tree_root->node = NULL;
2467 btrfs_put_root(log_tree_root);
2468 return ret;
2469 }
2470 if (!extent_buffer_uptodate(log_tree_root->node)) {
2471 btrfs_err(fs_info, "failed to read log tree");
2472 btrfs_put_root(log_tree_root);
2473 return -EIO;
2474 }
2475
2476 /* returns with log_tree_root freed on success */
2477 ret = btrfs_recover_log_trees(log_tree_root);
2478 if (ret) {
2479 btrfs_handle_fs_error(fs_info, ret,
2480 "Failed to recover log tree");
2481 btrfs_put_root(log_tree_root);
2482 return ret;
2483 }
2484
2485 if (sb_rdonly(fs_info->sb)) {
2486 ret = btrfs_commit_super(fs_info);
2487 if (ret)
2488 return ret;
2489 }
2490
2491 return 0;
2492}
2493
2494static int load_global_roots_objectid(struct btrfs_root *tree_root,
2495 struct btrfs_path *path, u64 objectid,
2496 const char *name)
2497{
2498 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2499 struct btrfs_root *root;
2500 u64 max_global_id = 0;
2501 int ret;
2502 struct btrfs_key key = {
2503 .objectid = objectid,
2504 .type = BTRFS_ROOT_ITEM_KEY,
2505 .offset = 0,
2506 };
2507 bool found = false;
2508
2509 /* If we have IGNOREDATACSUMS skip loading these roots. */
2510 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2511 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2512 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2513 return 0;
2514 }
2515
2516 while (1) {
2517 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2518 if (ret < 0)
2519 break;
2520
2521 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2522 ret = btrfs_next_leaf(tree_root, path);
2523 if (ret) {
2524 if (ret > 0)
2525 ret = 0;
2526 break;
2527 }
2528 }
2529 ret = 0;
2530
2531 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2532 if (key.objectid != objectid)
2533 break;
2534 btrfs_release_path(path);
2535
2536 /*
2537 * Just worry about this for extent tree, it'll be the same for
2538 * everybody.
2539 */
2540 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2541 max_global_id = max(max_global_id, key.offset);
2542
2543 found = true;
2544 root = read_tree_root_path(tree_root, path, &key);
2545 if (IS_ERR(root)) {
2546 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2547 ret = PTR_ERR(root);
2548 break;
2549 }
2550 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2551 ret = btrfs_global_root_insert(root);
2552 if (ret) {
2553 btrfs_put_root(root);
2554 break;
2555 }
2556 key.offset++;
2557 }
2558 btrfs_release_path(path);
2559
2560 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2561 fs_info->nr_global_roots = max_global_id + 1;
2562
2563 if (!found || ret) {
2564 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2565 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2566
2567 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2568 ret = ret ? ret : -ENOENT;
2569 else
2570 ret = 0;
2571 btrfs_err(fs_info, "failed to load root %s", name);
2572 }
2573 return ret;
2574}
2575
2576static int load_global_roots(struct btrfs_root *tree_root)
2577{
2578 struct btrfs_path *path;
2579 int ret = 0;
2580
2581 path = btrfs_alloc_path();
2582 if (!path)
2583 return -ENOMEM;
2584
2585 ret = load_global_roots_objectid(tree_root, path,
2586 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2587 if (ret)
2588 goto out;
2589 ret = load_global_roots_objectid(tree_root, path,
2590 BTRFS_CSUM_TREE_OBJECTID, "csum");
2591 if (ret)
2592 goto out;
2593 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2594 goto out;
2595 ret = load_global_roots_objectid(tree_root, path,
2596 BTRFS_FREE_SPACE_TREE_OBJECTID,
2597 "free space");
2598out:
2599 btrfs_free_path(path);
2600 return ret;
2601}
2602
2603static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2604{
2605 struct btrfs_root *tree_root = fs_info->tree_root;
2606 struct btrfs_root *root;
2607 struct btrfs_key location;
2608 int ret;
2609
2610 BUG_ON(!fs_info->tree_root);
2611
2612 ret = load_global_roots(tree_root);
2613 if (ret)
2614 return ret;
2615
2616 location.type = BTRFS_ROOT_ITEM_KEY;
2617 location.offset = 0;
2618
2619 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2620 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2621 root = btrfs_read_tree_root(tree_root, &location);
2622 if (IS_ERR(root)) {
2623 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2624 ret = PTR_ERR(root);
2625 goto out;
2626 }
2627 } else {
2628 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2629 fs_info->block_group_root = root;
2630 }
2631 }
2632
2633 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2634 root = btrfs_read_tree_root(tree_root, &location);
2635 if (IS_ERR(root)) {
2636 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2637 ret = PTR_ERR(root);
2638 goto out;
2639 }
2640 } else {
2641 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2642 fs_info->dev_root = root;
2643 }
2644 /* Initialize fs_info for all devices in any case */
2645 ret = btrfs_init_devices_late(fs_info);
2646 if (ret)
2647 goto out;
2648
2649 /*
2650 * This tree can share blocks with some other fs tree during relocation
2651 * and we need a proper setup by btrfs_get_fs_root
2652 */
2653 root = btrfs_get_fs_root(tree_root->fs_info,
2654 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2655 if (IS_ERR(root)) {
2656 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2657 ret = PTR_ERR(root);
2658 goto out;
2659 }
2660 } else {
2661 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2662 fs_info->data_reloc_root = root;
2663 }
2664
2665 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2666 root = btrfs_read_tree_root(tree_root, &location);
2667 if (!IS_ERR(root)) {
2668 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2669 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2670 fs_info->quota_root = root;
2671 }
2672
2673 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2674 root = btrfs_read_tree_root(tree_root, &location);
2675 if (IS_ERR(root)) {
2676 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2677 ret = PTR_ERR(root);
2678 if (ret != -ENOENT)
2679 goto out;
2680 }
2681 } else {
2682 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2683 fs_info->uuid_root = root;
2684 }
2685
2686 return 0;
2687out:
2688 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2689 location.objectid, ret);
2690 return ret;
2691}
2692
2693/*
2694 * Real super block validation
2695 * NOTE: super csum type and incompat features will not be checked here.
2696 *
2697 * @sb: super block to check
2698 * @mirror_num: the super block number to check its bytenr:
2699 * 0 the primary (1st) sb
2700 * 1, 2 2nd and 3rd backup copy
2701 * -1 skip bytenr check
2702 */
2703int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2704 struct btrfs_super_block *sb, int mirror_num)
2705{
2706 u64 nodesize = btrfs_super_nodesize(sb);
2707 u64 sectorsize = btrfs_super_sectorsize(sb);
2708 int ret = 0;
2709
2710 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2711 btrfs_err(fs_info, "no valid FS found");
2712 ret = -EINVAL;
2713 }
2714 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2715 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2716 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2717 ret = -EINVAL;
2718 }
2719 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2720 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2721 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2722 ret = -EINVAL;
2723 }
2724 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2725 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2726 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2727 ret = -EINVAL;
2728 }
2729 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2730 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2731 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2732 ret = -EINVAL;
2733 }
2734
2735 /*
2736 * Check sectorsize and nodesize first, other check will need it.
2737 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2738 */
2739 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2740 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2741 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2742 ret = -EINVAL;
2743 }
2744
2745 /*
2746 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2747 *
2748 * We can support 16K sectorsize with 64K page size without problem,
2749 * but such sectorsize/pagesize combination doesn't make much sense.
2750 * 4K will be our future standard, PAGE_SIZE is supported from the very
2751 * beginning.
2752 */
2753 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2754 btrfs_err(fs_info,
2755 "sectorsize %llu not yet supported for page size %lu",
2756 sectorsize, PAGE_SIZE);
2757 ret = -EINVAL;
2758 }
2759
2760 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2761 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2762 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2763 ret = -EINVAL;
2764 }
2765 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2766 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2767 le32_to_cpu(sb->__unused_leafsize), nodesize);
2768 ret = -EINVAL;
2769 }
2770
2771 /* Root alignment check */
2772 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2773 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2774 btrfs_super_root(sb));
2775 ret = -EINVAL;
2776 }
2777 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2778 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2779 btrfs_super_chunk_root(sb));
2780 ret = -EINVAL;
2781 }
2782 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2783 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2784 btrfs_super_log_root(sb));
2785 ret = -EINVAL;
2786 }
2787
2788 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2789 BTRFS_FSID_SIZE)) {
2790 btrfs_err(fs_info,
2791 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2792 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2793 ret = -EINVAL;
2794 }
2795
2796 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2797 memcmp(fs_info->fs_devices->metadata_uuid,
2798 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2799 btrfs_err(fs_info,
2800"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2801 fs_info->super_copy->metadata_uuid,
2802 fs_info->fs_devices->metadata_uuid);
2803 ret = -EINVAL;
2804 }
2805
2806 /*
2807 * Artificial requirement for block-group-tree to force newer features
2808 * (free-space-tree, no-holes) so the test matrix is smaller.
2809 */
2810 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2811 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2812 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2813 btrfs_err(fs_info,
2814 "block-group-tree feature requires fres-space-tree and no-holes");
2815 ret = -EINVAL;
2816 }
2817
2818 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2819 BTRFS_FSID_SIZE) != 0) {
2820 btrfs_err(fs_info,
2821 "dev_item UUID does not match metadata fsid: %pU != %pU",
2822 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2823 ret = -EINVAL;
2824 }
2825
2826 /*
2827 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2828 * done later
2829 */
2830 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2831 btrfs_err(fs_info, "bytes_used is too small %llu",
2832 btrfs_super_bytes_used(sb));
2833 ret = -EINVAL;
2834 }
2835 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2836 btrfs_err(fs_info, "invalid stripesize %u",
2837 btrfs_super_stripesize(sb));
2838 ret = -EINVAL;
2839 }
2840 if (btrfs_super_num_devices(sb) > (1UL << 31))
2841 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2842 btrfs_super_num_devices(sb));
2843 if (btrfs_super_num_devices(sb) == 0) {
2844 btrfs_err(fs_info, "number of devices is 0");
2845 ret = -EINVAL;
2846 }
2847
2848 if (mirror_num >= 0 &&
2849 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2850 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2851 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2852 ret = -EINVAL;
2853 }
2854
2855 /*
2856 * Obvious sys_chunk_array corruptions, it must hold at least one key
2857 * and one chunk
2858 */
2859 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2860 btrfs_err(fs_info, "system chunk array too big %u > %u",
2861 btrfs_super_sys_array_size(sb),
2862 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2863 ret = -EINVAL;
2864 }
2865 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2866 + sizeof(struct btrfs_chunk)) {
2867 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2868 btrfs_super_sys_array_size(sb),
2869 sizeof(struct btrfs_disk_key)
2870 + sizeof(struct btrfs_chunk));
2871 ret = -EINVAL;
2872 }
2873
2874 /*
2875 * The generation is a global counter, we'll trust it more than the others
2876 * but it's still possible that it's the one that's wrong.
2877 */
2878 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2879 btrfs_warn(fs_info,
2880 "suspicious: generation < chunk_root_generation: %llu < %llu",
2881 btrfs_super_generation(sb),
2882 btrfs_super_chunk_root_generation(sb));
2883 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2884 && btrfs_super_cache_generation(sb) != (u64)-1)
2885 btrfs_warn(fs_info,
2886 "suspicious: generation < cache_generation: %llu < %llu",
2887 btrfs_super_generation(sb),
2888 btrfs_super_cache_generation(sb));
2889
2890 return ret;
2891}
2892
2893/*
2894 * Validation of super block at mount time.
2895 * Some checks already done early at mount time, like csum type and incompat
2896 * flags will be skipped.
2897 */
2898static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2899{
2900 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2901}
2902
2903/*
2904 * Validation of super block at write time.
2905 * Some checks like bytenr check will be skipped as their values will be
2906 * overwritten soon.
2907 * Extra checks like csum type and incompat flags will be done here.
2908 */
2909static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2910 struct btrfs_super_block *sb)
2911{
2912 int ret;
2913
2914 ret = btrfs_validate_super(fs_info, sb, -1);
2915 if (ret < 0)
2916 goto out;
2917 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2918 ret = -EUCLEAN;
2919 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2920 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2921 goto out;
2922 }
2923 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2924 ret = -EUCLEAN;
2925 btrfs_err(fs_info,
2926 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2927 btrfs_super_incompat_flags(sb),
2928 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2929 goto out;
2930 }
2931out:
2932 if (ret < 0)
2933 btrfs_err(fs_info,
2934 "super block corruption detected before writing it to disk");
2935 return ret;
2936}
2937
2938static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2939{
2940 struct btrfs_tree_parent_check check = {
2941 .level = level,
2942 .transid = gen,
2943 .owner_root = root->root_key.objectid
2944 };
2945 int ret = 0;
2946
2947 root->node = read_tree_block(root->fs_info, bytenr, &check);
2948 if (IS_ERR(root->node)) {
2949 ret = PTR_ERR(root->node);
2950 root->node = NULL;
2951 return ret;
2952 }
2953 if (!extent_buffer_uptodate(root->node)) {
2954 free_extent_buffer(root->node);
2955 root->node = NULL;
2956 return -EIO;
2957 }
2958
2959 btrfs_set_root_node(&root->root_item, root->node);
2960 root->commit_root = btrfs_root_node(root);
2961 btrfs_set_root_refs(&root->root_item, 1);
2962 return ret;
2963}
2964
2965static int load_important_roots(struct btrfs_fs_info *fs_info)
2966{
2967 struct btrfs_super_block *sb = fs_info->super_copy;
2968 u64 gen, bytenr;
2969 int level, ret;
2970
2971 bytenr = btrfs_super_root(sb);
2972 gen = btrfs_super_generation(sb);
2973 level = btrfs_super_root_level(sb);
2974 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2975 if (ret) {
2976 btrfs_warn(fs_info, "couldn't read tree root");
2977 return ret;
2978 }
2979 return 0;
2980}
2981
2982static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2983{
2984 int backup_index = find_newest_super_backup(fs_info);
2985 struct btrfs_super_block *sb = fs_info->super_copy;
2986 struct btrfs_root *tree_root = fs_info->tree_root;
2987 bool handle_error = false;
2988 int ret = 0;
2989 int i;
2990
2991 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2992 if (handle_error) {
2993 if (!IS_ERR(tree_root->node))
2994 free_extent_buffer(tree_root->node);
2995 tree_root->node = NULL;
2996
2997 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2998 break;
2999
3000 free_root_pointers(fs_info, 0);
3001
3002 /*
3003 * Don't use the log in recovery mode, it won't be
3004 * valid
3005 */
3006 btrfs_set_super_log_root(sb, 0);
3007
3008 /* We can't trust the free space cache either */
3009 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3010
3011 ret = read_backup_root(fs_info, i);
3012 backup_index = ret;
3013 if (ret < 0)
3014 return ret;
3015 }
3016
3017 ret = load_important_roots(fs_info);
3018 if (ret) {
3019 handle_error = true;
3020 continue;
3021 }
3022
3023 /*
3024 * No need to hold btrfs_root::objectid_mutex since the fs
3025 * hasn't been fully initialised and we are the only user
3026 */
3027 ret = btrfs_init_root_free_objectid(tree_root);
3028 if (ret < 0) {
3029 handle_error = true;
3030 continue;
3031 }
3032
3033 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
3034
3035 ret = btrfs_read_roots(fs_info);
3036 if (ret < 0) {
3037 handle_error = true;
3038 continue;
3039 }
3040
3041 /* All successful */
3042 fs_info->generation = btrfs_header_generation(tree_root->node);
3043 fs_info->last_trans_committed = fs_info->generation;
3044 fs_info->last_reloc_trans = 0;
3045
3046 /* Always begin writing backup roots after the one being used */
3047 if (backup_index < 0) {
3048 fs_info->backup_root_index = 0;
3049 } else {
3050 fs_info->backup_root_index = backup_index + 1;
3051 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3052 }
3053 break;
3054 }
3055
3056 return ret;
3057}
3058
3059void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3060{
3061 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
3062 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
3063 INIT_LIST_HEAD(&fs_info->trans_list);
3064 INIT_LIST_HEAD(&fs_info->dead_roots);
3065 INIT_LIST_HEAD(&fs_info->delayed_iputs);
3066 INIT_LIST_HEAD(&fs_info->delalloc_roots);
3067 INIT_LIST_HEAD(&fs_info->caching_block_groups);
3068 spin_lock_init(&fs_info->delalloc_root_lock);
3069 spin_lock_init(&fs_info->trans_lock);
3070 spin_lock_init(&fs_info->fs_roots_radix_lock);
3071 spin_lock_init(&fs_info->delayed_iput_lock);
3072 spin_lock_init(&fs_info->defrag_inodes_lock);
3073 spin_lock_init(&fs_info->super_lock);
3074 spin_lock_init(&fs_info->buffer_lock);
3075 spin_lock_init(&fs_info->unused_bgs_lock);
3076 spin_lock_init(&fs_info->treelog_bg_lock);
3077 spin_lock_init(&fs_info->zone_active_bgs_lock);
3078 spin_lock_init(&fs_info->relocation_bg_lock);
3079 rwlock_init(&fs_info->tree_mod_log_lock);
3080 rwlock_init(&fs_info->global_root_lock);
3081 mutex_init(&fs_info->unused_bg_unpin_mutex);
3082 mutex_init(&fs_info->reclaim_bgs_lock);
3083 mutex_init(&fs_info->reloc_mutex);
3084 mutex_init(&fs_info->delalloc_root_mutex);
3085 mutex_init(&fs_info->zoned_meta_io_lock);
3086 mutex_init(&fs_info->zoned_data_reloc_io_lock);
3087 seqlock_init(&fs_info->profiles_lock);
3088
3089 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
3090 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
3091 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
3092 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
3093 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
3094 BTRFS_LOCKDEP_TRANS_COMMIT_START);
3095 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
3096 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
3097 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
3098 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
3099 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
3100 BTRFS_LOCKDEP_TRANS_COMPLETED);
3101
3102 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3103 INIT_LIST_HEAD(&fs_info->space_info);
3104 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3105 INIT_LIST_HEAD(&fs_info->unused_bgs);
3106 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3107 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3108#ifdef CONFIG_BTRFS_DEBUG
3109 INIT_LIST_HEAD(&fs_info->allocated_roots);
3110 INIT_LIST_HEAD(&fs_info->allocated_ebs);
3111 spin_lock_init(&fs_info->eb_leak_lock);
3112#endif
3113 extent_map_tree_init(&fs_info->mapping_tree);
3114 btrfs_init_block_rsv(&fs_info->global_block_rsv,
3115 BTRFS_BLOCK_RSV_GLOBAL);
3116 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3117 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3118 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3119 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3120 BTRFS_BLOCK_RSV_DELOPS);
3121 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3122 BTRFS_BLOCK_RSV_DELREFS);
3123
3124 atomic_set(&fs_info->async_delalloc_pages, 0);
3125 atomic_set(&fs_info->defrag_running, 0);
3126 atomic_set(&fs_info->nr_delayed_iputs, 0);
3127 atomic64_set(&fs_info->tree_mod_seq, 0);
3128 fs_info->global_root_tree = RB_ROOT;
3129 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3130 fs_info->metadata_ratio = 0;
3131 fs_info->defrag_inodes = RB_ROOT;
3132 atomic64_set(&fs_info->free_chunk_space, 0);
3133 fs_info->tree_mod_log = RB_ROOT;
3134 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3135 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3136 btrfs_init_ref_verify(fs_info);
3137
3138 fs_info->thread_pool_size = min_t(unsigned long,
3139 num_online_cpus() + 2, 8);
3140
3141 INIT_LIST_HEAD(&fs_info->ordered_roots);
3142 spin_lock_init(&fs_info->ordered_root_lock);
3143
3144 btrfs_init_scrub(fs_info);
3145#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3146 fs_info->check_integrity_print_mask = 0;
3147#endif
3148 btrfs_init_balance(fs_info);
3149 btrfs_init_async_reclaim_work(fs_info);
3150
3151 rwlock_init(&fs_info->block_group_cache_lock);
3152 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3153
3154 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3155 IO_TREE_FS_EXCLUDED_EXTENTS);
3156
3157 mutex_init(&fs_info->ordered_operations_mutex);
3158 mutex_init(&fs_info->tree_log_mutex);
3159 mutex_init(&fs_info->chunk_mutex);
3160 mutex_init(&fs_info->transaction_kthread_mutex);
3161 mutex_init(&fs_info->cleaner_mutex);
3162 mutex_init(&fs_info->ro_block_group_mutex);
3163 init_rwsem(&fs_info->commit_root_sem);
3164 init_rwsem(&fs_info->cleanup_work_sem);
3165 init_rwsem(&fs_info->subvol_sem);
3166 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3167
3168 btrfs_init_dev_replace_locks(fs_info);
3169 btrfs_init_qgroup(fs_info);
3170 btrfs_discard_init(fs_info);
3171
3172 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3173 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3174
3175 init_waitqueue_head(&fs_info->transaction_throttle);
3176 init_waitqueue_head(&fs_info->transaction_wait);
3177 init_waitqueue_head(&fs_info->transaction_blocked_wait);
3178 init_waitqueue_head(&fs_info->async_submit_wait);
3179 init_waitqueue_head(&fs_info->delayed_iputs_wait);
3180
3181 /* Usable values until the real ones are cached from the superblock */
3182 fs_info->nodesize = 4096;
3183 fs_info->sectorsize = 4096;
3184 fs_info->sectorsize_bits = ilog2(4096);
3185 fs_info->stripesize = 4096;
3186
3187 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3188
3189 spin_lock_init(&fs_info->swapfile_pins_lock);
3190 fs_info->swapfile_pins = RB_ROOT;
3191
3192 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3193 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3194}
3195
3196static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3197{
3198 int ret;
3199
3200 fs_info->sb = sb;
3201 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3202 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3203
3204 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3205 if (ret)
3206 return ret;
3207
3208 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3209 if (ret)
3210 return ret;
3211
3212 fs_info->dirty_metadata_batch = PAGE_SIZE *
3213 (1 + ilog2(nr_cpu_ids));
3214
3215 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3216 if (ret)
3217 return ret;
3218
3219 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3220 GFP_KERNEL);
3221 if (ret)
3222 return ret;
3223
3224 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3225 GFP_KERNEL);
3226 if (!fs_info->delayed_root)
3227 return -ENOMEM;
3228 btrfs_init_delayed_root(fs_info->delayed_root);
3229
3230 if (sb_rdonly(sb))
3231 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3232
3233 return btrfs_alloc_stripe_hash_table(fs_info);
3234}
3235
3236static int btrfs_uuid_rescan_kthread(void *data)
3237{
3238 struct btrfs_fs_info *fs_info = data;
3239 int ret;
3240
3241 /*
3242 * 1st step is to iterate through the existing UUID tree and
3243 * to delete all entries that contain outdated data.
3244 * 2nd step is to add all missing entries to the UUID tree.
3245 */
3246 ret = btrfs_uuid_tree_iterate(fs_info);
3247 if (ret < 0) {
3248 if (ret != -EINTR)
3249 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3250 ret);
3251 up(&fs_info->uuid_tree_rescan_sem);
3252 return ret;
3253 }
3254 return btrfs_uuid_scan_kthread(data);
3255}
3256
3257static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3258{
3259 struct task_struct *task;
3260
3261 down(&fs_info->uuid_tree_rescan_sem);
3262 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3263 if (IS_ERR(task)) {
3264 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3265 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3266 up(&fs_info->uuid_tree_rescan_sem);
3267 return PTR_ERR(task);
3268 }
3269
3270 return 0;
3271}
3272
3273/*
3274 * Some options only have meaning at mount time and shouldn't persist across
3275 * remounts, or be displayed. Clear these at the end of mount and remount
3276 * code paths.
3277 */
3278void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3279{
3280 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3281 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3282}
3283
3284/*
3285 * Mounting logic specific to read-write file systems. Shared by open_ctree
3286 * and btrfs_remount when remounting from read-only to read-write.
3287 */
3288int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3289{
3290 int ret;
3291 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3292 bool clear_free_space_tree = false;
3293
3294 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3295 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3296 clear_free_space_tree = true;
3297 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3298 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3299 btrfs_warn(fs_info, "free space tree is invalid");
3300 clear_free_space_tree = true;
3301 }
3302
3303 if (clear_free_space_tree) {
3304 btrfs_info(fs_info, "clearing free space tree");
3305 ret = btrfs_clear_free_space_tree(fs_info);
3306 if (ret) {
3307 btrfs_warn(fs_info,
3308 "failed to clear free space tree: %d", ret);
3309 goto out;
3310 }
3311 }
3312
3313 /*
3314 * btrfs_find_orphan_roots() is responsible for finding all the dead
3315 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3316 * them into the fs_info->fs_roots_radix tree. This must be done before
3317 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3318 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3319 * item before the root's tree is deleted - this means that if we unmount
3320 * or crash before the deletion completes, on the next mount we will not
3321 * delete what remains of the tree because the orphan item does not
3322 * exists anymore, which is what tells us we have a pending deletion.
3323 */
3324 ret = btrfs_find_orphan_roots(fs_info);
3325 if (ret)
3326 goto out;
3327
3328 ret = btrfs_cleanup_fs_roots(fs_info);
3329 if (ret)
3330 goto out;
3331
3332 down_read(&fs_info->cleanup_work_sem);
3333 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3334 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3335 up_read(&fs_info->cleanup_work_sem);
3336 goto out;
3337 }
3338 up_read(&fs_info->cleanup_work_sem);
3339
3340 mutex_lock(&fs_info->cleaner_mutex);
3341 ret = btrfs_recover_relocation(fs_info);
3342 mutex_unlock(&fs_info->cleaner_mutex);
3343 if (ret < 0) {
3344 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3345 goto out;
3346 }
3347
3348 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3349 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3350 btrfs_info(fs_info, "creating free space tree");
3351 ret = btrfs_create_free_space_tree(fs_info);
3352 if (ret) {
3353 btrfs_warn(fs_info,
3354 "failed to create free space tree: %d", ret);
3355 goto out;
3356 }
3357 }
3358
3359 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3360 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3361 if (ret)
3362 goto out;
3363 }
3364
3365 ret = btrfs_resume_balance_async(fs_info);
3366 if (ret)
3367 goto out;
3368
3369 ret = btrfs_resume_dev_replace_async(fs_info);
3370 if (ret) {
3371 btrfs_warn(fs_info, "failed to resume dev_replace");
3372 goto out;
3373 }
3374
3375 btrfs_qgroup_rescan_resume(fs_info);
3376
3377 if (!fs_info->uuid_root) {
3378 btrfs_info(fs_info, "creating UUID tree");
3379 ret = btrfs_create_uuid_tree(fs_info);
3380 if (ret) {
3381 btrfs_warn(fs_info,
3382 "failed to create the UUID tree %d", ret);
3383 goto out;
3384 }
3385 }
3386
3387out:
3388 return ret;
3389}
3390
3391/*
3392 * Do various sanity and dependency checks of different features.
3393 *
3394 * @is_rw_mount: If the mount is read-write.
3395 *
3396 * This is the place for less strict checks (like for subpage or artificial
3397 * feature dependencies).
3398 *
3399 * For strict checks or possible corruption detection, see
3400 * btrfs_validate_super().
3401 *
3402 * This should be called after btrfs_parse_options(), as some mount options
3403 * (space cache related) can modify on-disk format like free space tree and
3404 * screw up certain feature dependencies.
3405 */
3406int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3407{
3408 struct btrfs_super_block *disk_super = fs_info->super_copy;
3409 u64 incompat = btrfs_super_incompat_flags(disk_super);
3410 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3411 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3412
3413 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3414 btrfs_err(fs_info,
3415 "cannot mount because of unknown incompat features (0x%llx)",
3416 incompat);
3417 return -EINVAL;
3418 }
3419
3420 /* Runtime limitation for mixed block groups. */
3421 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3422 (fs_info->sectorsize != fs_info->nodesize)) {
3423 btrfs_err(fs_info,
3424"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3425 fs_info->nodesize, fs_info->sectorsize);
3426 return -EINVAL;
3427 }
3428
3429 /* Mixed backref is an always-enabled feature. */
3430 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3431
3432 /* Set compression related flags just in case. */
3433 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3434 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3435 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3436 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3437
3438 /*
3439 * An ancient flag, which should really be marked deprecated.
3440 * Such runtime limitation doesn't really need a incompat flag.
3441 */
3442 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3443 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3444
3445 if (compat_ro_unsupp && is_rw_mount) {
3446 btrfs_err(fs_info,
3447 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3448 compat_ro);
3449 return -EINVAL;
3450 }
3451
3452 /*
3453 * We have unsupported RO compat features, although RO mounted, we
3454 * should not cause any metadata writes, including log replay.
3455 * Or we could screw up whatever the new feature requires.
3456 */
3457 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3458 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3459 btrfs_err(fs_info,
3460"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3461 compat_ro);
3462 return -EINVAL;
3463 }
3464
3465 /*
3466 * Artificial limitations for block group tree, to force
3467 * block-group-tree to rely on no-holes and free-space-tree.
3468 */
3469 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3470 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3471 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3472 btrfs_err(fs_info,
3473"block-group-tree feature requires no-holes and free-space-tree features");
3474 return -EINVAL;
3475 }
3476
3477 /*
3478 * Subpage runtime limitation on v1 cache.
3479 *
3480 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3481 * we're already defaulting to v2 cache, no need to bother v1 as it's
3482 * going to be deprecated anyway.
3483 */
3484 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3485 btrfs_warn(fs_info,
3486 "v1 space cache is not supported for page size %lu with sectorsize %u",
3487 PAGE_SIZE, fs_info->sectorsize);
3488 return -EINVAL;
3489 }
3490
3491 /* This can be called by remount, we need to protect the super block. */
3492 spin_lock(&fs_info->super_lock);
3493 btrfs_set_super_incompat_flags(disk_super, incompat);
3494 spin_unlock(&fs_info->super_lock);
3495
3496 return 0;
3497}
3498
3499int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3500 char *options)
3501{
3502 u32 sectorsize;
3503 u32 nodesize;
3504 u32 stripesize;
3505 u64 generation;
3506 u64 features;
3507 u16 csum_type;
3508 struct btrfs_super_block *disk_super;
3509 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3510 struct btrfs_root *tree_root;
3511 struct btrfs_root *chunk_root;
3512 int ret;
3513 int err = -EINVAL;
3514 int level;
3515
3516 ret = init_mount_fs_info(fs_info, sb);
3517 if (ret) {
3518 err = ret;
3519 goto fail;
3520 }
3521
3522 /* These need to be init'ed before we start creating inodes and such. */
3523 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3524 GFP_KERNEL);
3525 fs_info->tree_root = tree_root;
3526 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3527 GFP_KERNEL);
3528 fs_info->chunk_root = chunk_root;
3529 if (!tree_root || !chunk_root) {
3530 err = -ENOMEM;
3531 goto fail;
3532 }
3533
3534 fs_info->btree_inode = new_inode(sb);
3535 if (!fs_info->btree_inode) {
3536 err = -ENOMEM;
3537 goto fail;
3538 }
3539 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3540 btrfs_init_btree_inode(fs_info);
3541
3542 invalidate_bdev(fs_devices->latest_dev->bdev);
3543
3544 /*
3545 * Read super block and check the signature bytes only
3546 */
3547 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3548 if (IS_ERR(disk_super)) {
3549 err = PTR_ERR(disk_super);
3550 goto fail_alloc;
3551 }
3552
3553 /*
3554 * Verify the type first, if that or the checksum value are
3555 * corrupted, we'll find out
3556 */
3557 csum_type = btrfs_super_csum_type(disk_super);
3558 if (!btrfs_supported_super_csum(csum_type)) {
3559 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3560 csum_type);
3561 err = -EINVAL;
3562 btrfs_release_disk_super(disk_super);
3563 goto fail_alloc;
3564 }
3565
3566 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3567
3568 ret = btrfs_init_csum_hash(fs_info, csum_type);
3569 if (ret) {
3570 err = ret;
3571 btrfs_release_disk_super(disk_super);
3572 goto fail_alloc;
3573 }
3574
3575 /*
3576 * We want to check superblock checksum, the type is stored inside.
3577 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3578 */
3579 if (btrfs_check_super_csum(fs_info, disk_super)) {
3580 btrfs_err(fs_info, "superblock checksum mismatch");
3581 err = -EINVAL;
3582 btrfs_release_disk_super(disk_super);
3583 goto fail_alloc;
3584 }
3585
3586 /*
3587 * super_copy is zeroed at allocation time and we never touch the
3588 * following bytes up to INFO_SIZE, the checksum is calculated from
3589 * the whole block of INFO_SIZE
3590 */
3591 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3592 btrfs_release_disk_super(disk_super);
3593
3594 disk_super = fs_info->super_copy;
3595
3596
3597 features = btrfs_super_flags(disk_super);
3598 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3599 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3600 btrfs_set_super_flags(disk_super, features);
3601 btrfs_info(fs_info,
3602 "found metadata UUID change in progress flag, clearing");
3603 }
3604
3605 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3606 sizeof(*fs_info->super_for_commit));
3607
3608 ret = btrfs_validate_mount_super(fs_info);
3609 if (ret) {
3610 btrfs_err(fs_info, "superblock contains fatal errors");
3611 err = -EINVAL;
3612 goto fail_alloc;
3613 }
3614
3615 if (!btrfs_super_root(disk_super))
3616 goto fail_alloc;
3617
3618 /* check FS state, whether FS is broken. */
3619 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3620 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3621
3622 /*
3623 * In the long term, we'll store the compression type in the super
3624 * block, and it'll be used for per file compression control.
3625 */
3626 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3627
3628
3629 /* Set up fs_info before parsing mount options */
3630 nodesize = btrfs_super_nodesize(disk_super);
3631 sectorsize = btrfs_super_sectorsize(disk_super);
3632 stripesize = sectorsize;
3633 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3634 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3635
3636 fs_info->nodesize = nodesize;
3637 fs_info->sectorsize = sectorsize;
3638 fs_info->sectorsize_bits = ilog2(sectorsize);
3639 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3640 fs_info->stripesize = stripesize;
3641
3642 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3643 if (ret) {
3644 err = ret;
3645 goto fail_alloc;
3646 }
3647
3648 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3649 if (ret < 0) {
3650 err = ret;
3651 goto fail_alloc;
3652 }
3653
3654 if (sectorsize < PAGE_SIZE) {
3655 struct btrfs_subpage_info *subpage_info;
3656
3657 /*
3658 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3659 * going to be deprecated.
3660 *
3661 * Force to use v2 cache for subpage case.
3662 */
3663 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3664 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3665 "forcing free space tree for sector size %u with page size %lu",
3666 sectorsize, PAGE_SIZE);
3667
3668 btrfs_warn(fs_info,
3669 "read-write for sector size %u with page size %lu is experimental",
3670 sectorsize, PAGE_SIZE);
3671 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3672 if (!subpage_info)
3673 goto fail_alloc;
3674 btrfs_init_subpage_info(subpage_info, sectorsize);
3675 fs_info->subpage_info = subpage_info;
3676 }
3677
3678 ret = btrfs_init_workqueues(fs_info);
3679 if (ret) {
3680 err = ret;
3681 goto fail_sb_buffer;
3682 }
3683
3684 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3685 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3686
3687 sb->s_blocksize = sectorsize;
3688 sb->s_blocksize_bits = blksize_bits(sectorsize);
3689 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3690
3691 mutex_lock(&fs_info->chunk_mutex);
3692 ret = btrfs_read_sys_array(fs_info);
3693 mutex_unlock(&fs_info->chunk_mutex);
3694 if (ret) {
3695 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3696 goto fail_sb_buffer;
3697 }
3698
3699 generation = btrfs_super_chunk_root_generation(disk_super);
3700 level = btrfs_super_chunk_root_level(disk_super);
3701 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3702 generation, level);
3703 if (ret) {
3704 btrfs_err(fs_info, "failed to read chunk root");
3705 goto fail_tree_roots;
3706 }
3707
3708 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3709 offsetof(struct btrfs_header, chunk_tree_uuid),
3710 BTRFS_UUID_SIZE);
3711
3712 ret = btrfs_read_chunk_tree(fs_info);
3713 if (ret) {
3714 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3715 goto fail_tree_roots;
3716 }
3717
3718 /*
3719 * At this point we know all the devices that make this filesystem,
3720 * including the seed devices but we don't know yet if the replace
3721 * target is required. So free devices that are not part of this
3722 * filesystem but skip the replace target device which is checked
3723 * below in btrfs_init_dev_replace().
3724 */
3725 btrfs_free_extra_devids(fs_devices);
3726 if (!fs_devices->latest_dev->bdev) {
3727 btrfs_err(fs_info, "failed to read devices");
3728 goto fail_tree_roots;
3729 }
3730
3731 ret = init_tree_roots(fs_info);
3732 if (ret)
3733 goto fail_tree_roots;
3734
3735 /*
3736 * Get zone type information of zoned block devices. This will also
3737 * handle emulation of a zoned filesystem if a regular device has the
3738 * zoned incompat feature flag set.
3739 */
3740 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3741 if (ret) {
3742 btrfs_err(fs_info,
3743 "zoned: failed to read device zone info: %d",
3744 ret);
3745 goto fail_block_groups;
3746 }
3747
3748 /*
3749 * If we have a uuid root and we're not being told to rescan we need to
3750 * check the generation here so we can set the
3751 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3752 * transaction during a balance or the log replay without updating the
3753 * uuid generation, and then if we crash we would rescan the uuid tree,
3754 * even though it was perfectly fine.
3755 */
3756 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3757 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3758 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3759
3760 ret = btrfs_verify_dev_extents(fs_info);
3761 if (ret) {
3762 btrfs_err(fs_info,
3763 "failed to verify dev extents against chunks: %d",
3764 ret);
3765 goto fail_block_groups;
3766 }
3767 ret = btrfs_recover_balance(fs_info);
3768 if (ret) {
3769 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3770 goto fail_block_groups;
3771 }
3772
3773 ret = btrfs_init_dev_stats(fs_info);
3774 if (ret) {
3775 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3776 goto fail_block_groups;
3777 }
3778
3779 ret = btrfs_init_dev_replace(fs_info);
3780 if (ret) {
3781 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3782 goto fail_block_groups;
3783 }
3784
3785 ret = btrfs_check_zoned_mode(fs_info);
3786 if (ret) {
3787 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3788 ret);
3789 goto fail_block_groups;
3790 }
3791
3792 ret = btrfs_sysfs_add_fsid(fs_devices);
3793 if (ret) {
3794 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3795 ret);
3796 goto fail_block_groups;
3797 }
3798
3799 ret = btrfs_sysfs_add_mounted(fs_info);
3800 if (ret) {
3801 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3802 goto fail_fsdev_sysfs;
3803 }
3804
3805 ret = btrfs_init_space_info(fs_info);
3806 if (ret) {
3807 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3808 goto fail_sysfs;
3809 }
3810
3811 ret = btrfs_read_block_groups(fs_info);
3812 if (ret) {
3813 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3814 goto fail_sysfs;
3815 }
3816
3817 btrfs_free_zone_cache(fs_info);
3818
3819 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3820 !btrfs_check_rw_degradable(fs_info, NULL)) {
3821 btrfs_warn(fs_info,
3822 "writable mount is not allowed due to too many missing devices");
3823 goto fail_sysfs;
3824 }
3825
3826 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3827 "btrfs-cleaner");
3828 if (IS_ERR(fs_info->cleaner_kthread))
3829 goto fail_sysfs;
3830
3831 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3832 tree_root,
3833 "btrfs-transaction");
3834 if (IS_ERR(fs_info->transaction_kthread))
3835 goto fail_cleaner;
3836
3837 if (!btrfs_test_opt(fs_info, NOSSD) &&
3838 !fs_info->fs_devices->rotating) {
3839 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3840 }
3841
3842 /*
3843 * For devices supporting discard turn on discard=async automatically,
3844 * unless it's already set or disabled. This could be turned off by
3845 * nodiscard for the same mount.
3846 */
3847 if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3848 btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3849 btrfs_test_opt(fs_info, NODISCARD)) &&
3850 fs_info->fs_devices->discardable) {
3851 btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3852 "auto enabling async discard");
3853 btrfs_clear_opt(fs_info->mount_opt, NODISCARD);
3854 }
3855
3856#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3857 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3858 ret = btrfsic_mount(fs_info, fs_devices,
3859 btrfs_test_opt(fs_info,
3860 CHECK_INTEGRITY_DATA) ? 1 : 0,
3861 fs_info->check_integrity_print_mask);
3862 if (ret)
3863 btrfs_warn(fs_info,
3864 "failed to initialize integrity check module: %d",
3865 ret);
3866 }
3867#endif
3868 ret = btrfs_read_qgroup_config(fs_info);
3869 if (ret)
3870 goto fail_trans_kthread;
3871
3872 if (btrfs_build_ref_tree(fs_info))
3873 btrfs_err(fs_info, "couldn't build ref tree");
3874
3875 /* do not make disk changes in broken FS or nologreplay is given */
3876 if (btrfs_super_log_root(disk_super) != 0 &&
3877 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3878 btrfs_info(fs_info, "start tree-log replay");
3879 ret = btrfs_replay_log(fs_info, fs_devices);
3880 if (ret) {
3881 err = ret;
3882 goto fail_qgroup;
3883 }
3884 }
3885
3886 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3887 if (IS_ERR(fs_info->fs_root)) {
3888 err = PTR_ERR(fs_info->fs_root);
3889 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3890 fs_info->fs_root = NULL;
3891 goto fail_qgroup;
3892 }
3893
3894 if (sb_rdonly(sb))
3895 goto clear_oneshot;
3896
3897 ret = btrfs_start_pre_rw_mount(fs_info);
3898 if (ret) {
3899 close_ctree(fs_info);
3900 return ret;
3901 }
3902 btrfs_discard_resume(fs_info);
3903
3904 if (fs_info->uuid_root &&
3905 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3906 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3907 btrfs_info(fs_info, "checking UUID tree");
3908 ret = btrfs_check_uuid_tree(fs_info);
3909 if (ret) {
3910 btrfs_warn(fs_info,
3911 "failed to check the UUID tree: %d", ret);
3912 close_ctree(fs_info);
3913 return ret;
3914 }
3915 }
3916
3917 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3918
3919 /* Kick the cleaner thread so it'll start deleting snapshots. */
3920 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3921 wake_up_process(fs_info->cleaner_kthread);
3922
3923clear_oneshot:
3924 btrfs_clear_oneshot_options(fs_info);
3925 return 0;
3926
3927fail_qgroup:
3928 btrfs_free_qgroup_config(fs_info);
3929fail_trans_kthread:
3930 kthread_stop(fs_info->transaction_kthread);
3931 btrfs_cleanup_transaction(fs_info);
3932 btrfs_free_fs_roots(fs_info);
3933fail_cleaner:
3934 kthread_stop(fs_info->cleaner_kthread);
3935
3936 /*
3937 * make sure we're done with the btree inode before we stop our
3938 * kthreads
3939 */
3940 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3941
3942fail_sysfs:
3943 btrfs_sysfs_remove_mounted(fs_info);
3944
3945fail_fsdev_sysfs:
3946 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3947
3948fail_block_groups:
3949 btrfs_put_block_group_cache(fs_info);
3950
3951fail_tree_roots:
3952 if (fs_info->data_reloc_root)
3953 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3954 free_root_pointers(fs_info, true);
3955 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3956
3957fail_sb_buffer:
3958 btrfs_stop_all_workers(fs_info);
3959 btrfs_free_block_groups(fs_info);
3960fail_alloc:
3961 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3962
3963 iput(fs_info->btree_inode);
3964fail:
3965 btrfs_close_devices(fs_info->fs_devices);
3966 return err;
3967}
3968ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3969
3970static void btrfs_end_super_write(struct bio *bio)
3971{
3972 struct btrfs_device *device = bio->bi_private;
3973 struct bio_vec *bvec;
3974 struct bvec_iter_all iter_all;
3975 struct page *page;
3976
3977 bio_for_each_segment_all(bvec, bio, iter_all) {
3978 page = bvec->bv_page;
3979
3980 if (bio->bi_status) {
3981 btrfs_warn_rl_in_rcu(device->fs_info,
3982 "lost page write due to IO error on %s (%d)",
3983 btrfs_dev_name(device),
3984 blk_status_to_errno(bio->bi_status));
3985 ClearPageUptodate(page);
3986 SetPageError(page);
3987 btrfs_dev_stat_inc_and_print(device,
3988 BTRFS_DEV_STAT_WRITE_ERRS);
3989 } else {
3990 SetPageUptodate(page);
3991 }
3992
3993 put_page(page);
3994 unlock_page(page);
3995 }
3996
3997 bio_put(bio);
3998}
3999
4000struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
4001 int copy_num, bool drop_cache)
4002{
4003 struct btrfs_super_block *super;
4004 struct page *page;
4005 u64 bytenr, bytenr_orig;
4006 struct address_space *mapping = bdev->bd_inode->i_mapping;
4007 int ret;
4008
4009 bytenr_orig = btrfs_sb_offset(copy_num);
4010 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
4011 if (ret == -ENOENT)
4012 return ERR_PTR(-EINVAL);
4013 else if (ret)
4014 return ERR_PTR(ret);
4015
4016 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
4017 return ERR_PTR(-EINVAL);
4018
4019 if (drop_cache) {
4020 /* This should only be called with the primary sb. */
4021 ASSERT(copy_num == 0);
4022
4023 /*
4024 * Drop the page of the primary superblock, so later read will
4025 * always read from the device.
4026 */
4027 invalidate_inode_pages2_range(mapping,
4028 bytenr >> PAGE_SHIFT,
4029 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
4030 }
4031
4032 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
4033 if (IS_ERR(page))
4034 return ERR_CAST(page);
4035
4036 super = page_address(page);
4037 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
4038 btrfs_release_disk_super(super);
4039 return ERR_PTR(-ENODATA);
4040 }
4041
4042 if (btrfs_super_bytenr(super) != bytenr_orig) {
4043 btrfs_release_disk_super(super);
4044 return ERR_PTR(-EINVAL);
4045 }
4046
4047 return super;
4048}
4049
4050
4051struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
4052{
4053 struct btrfs_super_block *super, *latest = NULL;
4054 int i;
4055 u64 transid = 0;
4056
4057 /* we would like to check all the supers, but that would make
4058 * a btrfs mount succeed after a mkfs from a different FS.
4059 * So, we need to add a special mount option to scan for
4060 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
4061 */
4062 for (i = 0; i < 1; i++) {
4063 super = btrfs_read_dev_one_super(bdev, i, false);
4064 if (IS_ERR(super))
4065 continue;
4066
4067 if (!latest || btrfs_super_generation(super) > transid) {
4068 if (latest)
4069 btrfs_release_disk_super(super);
4070
4071 latest = super;
4072 transid = btrfs_super_generation(super);
4073 }
4074 }
4075
4076 return super;
4077}
4078
4079/*
4080 * Write superblock @sb to the @device. Do not wait for completion, all the
4081 * pages we use for writing are locked.
4082 *
4083 * Write @max_mirrors copies of the superblock, where 0 means default that fit
4084 * the expected device size at commit time. Note that max_mirrors must be
4085 * same for write and wait phases.
4086 *
4087 * Return number of errors when page is not found or submission fails.
4088 */
4089static int write_dev_supers(struct btrfs_device *device,
4090 struct btrfs_super_block *sb, int max_mirrors)
4091{
4092 struct btrfs_fs_info *fs_info = device->fs_info;
4093 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
4094 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4095 int i;
4096 int errors = 0;
4097 int ret;
4098 u64 bytenr, bytenr_orig;
4099
4100 if (max_mirrors == 0)
4101 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4102
4103 shash->tfm = fs_info->csum_shash;
4104
4105 for (i = 0; i < max_mirrors; i++) {
4106 struct page *page;
4107 struct bio *bio;
4108 struct btrfs_super_block *disk_super;
4109
4110 bytenr_orig = btrfs_sb_offset(i);
4111 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4112 if (ret == -ENOENT) {
4113 continue;
4114 } else if (ret < 0) {
4115 btrfs_err(device->fs_info,
4116 "couldn't get super block location for mirror %d",
4117 i);
4118 errors++;
4119 continue;
4120 }
4121 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4122 device->commit_total_bytes)
4123 break;
4124
4125 btrfs_set_super_bytenr(sb, bytenr_orig);
4126
4127 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4128 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4129 sb->csum);
4130
4131 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4132 GFP_NOFS);
4133 if (!page) {
4134 btrfs_err(device->fs_info,
4135 "couldn't get super block page for bytenr %llu",
4136 bytenr);
4137 errors++;
4138 continue;
4139 }
4140
4141 /* Bump the refcount for wait_dev_supers() */
4142 get_page(page);
4143
4144 disk_super = page_address(page);
4145 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4146
4147 /*
4148 * Directly use bios here instead of relying on the page cache
4149 * to do I/O, so we don't lose the ability to do integrity
4150 * checking.
4151 */
4152 bio = bio_alloc(device->bdev, 1,
4153 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4154 GFP_NOFS);
4155 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4156 bio->bi_private = device;
4157 bio->bi_end_io = btrfs_end_super_write;
4158 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4159 offset_in_page(bytenr));
4160
4161 /*
4162 * We FUA only the first super block. The others we allow to
4163 * go down lazy and there's a short window where the on-disk
4164 * copies might still contain the older version.
4165 */
4166 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4167 bio->bi_opf |= REQ_FUA;
4168
4169 btrfsic_check_bio(bio);
4170 submit_bio(bio);
4171
4172 if (btrfs_advance_sb_log(device, i))
4173 errors++;
4174 }
4175 return errors < i ? 0 : -1;
4176}
4177
4178/*
4179 * Wait for write completion of superblocks done by write_dev_supers,
4180 * @max_mirrors same for write and wait phases.
4181 *
4182 * Return number of errors when page is not found or not marked up to
4183 * date.
4184 */
4185static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4186{
4187 int i;
4188 int errors = 0;
4189 bool primary_failed = false;
4190 int ret;
4191 u64 bytenr;
4192
4193 if (max_mirrors == 0)
4194 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4195
4196 for (i = 0; i < max_mirrors; i++) {
4197 struct page *page;
4198
4199 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4200 if (ret == -ENOENT) {
4201 break;
4202 } else if (ret < 0) {
4203 errors++;
4204 if (i == 0)
4205 primary_failed = true;
4206 continue;
4207 }
4208 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4209 device->commit_total_bytes)
4210 break;
4211
4212 page = find_get_page(device->bdev->bd_inode->i_mapping,
4213 bytenr >> PAGE_SHIFT);
4214 if (!page) {
4215 errors++;
4216 if (i == 0)
4217 primary_failed = true;
4218 continue;
4219 }
4220 /* Page is submitted locked and unlocked once the IO completes */
4221 wait_on_page_locked(page);
4222 if (PageError(page)) {
4223 errors++;
4224 if (i == 0)
4225 primary_failed = true;
4226 }
4227
4228 /* Drop our reference */
4229 put_page(page);
4230
4231 /* Drop the reference from the writing run */
4232 put_page(page);
4233 }
4234
4235 /* log error, force error return */
4236 if (primary_failed) {
4237 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4238 device->devid);
4239 return -1;
4240 }
4241
4242 return errors < i ? 0 : -1;
4243}
4244
4245/*
4246 * endio for the write_dev_flush, this will wake anyone waiting
4247 * for the barrier when it is done
4248 */
4249static void btrfs_end_empty_barrier(struct bio *bio)
4250{
4251 bio_uninit(bio);
4252 complete(bio->bi_private);
4253}
4254
4255/*
4256 * Submit a flush request to the device if it supports it. Error handling is
4257 * done in the waiting counterpart.
4258 */
4259static void write_dev_flush(struct btrfs_device *device)
4260{
4261 struct bio *bio = &device->flush_bio;
4262
4263#ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4264 /*
4265 * When a disk has write caching disabled, we skip submission of a bio
4266 * with flush and sync requests before writing the superblock, since
4267 * it's not needed. However when the integrity checker is enabled, this
4268 * results in reports that there are metadata blocks referred by a
4269 * superblock that were not properly flushed. So don't skip the bio
4270 * submission only when the integrity checker is enabled for the sake
4271 * of simplicity, since this is a debug tool and not meant for use in
4272 * non-debug builds.
4273 */
4274 if (!bdev_write_cache(device->bdev))
4275 return;
4276#endif
4277
4278 bio_init(bio, device->bdev, NULL, 0,
4279 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4280 bio->bi_end_io = btrfs_end_empty_barrier;
4281 init_completion(&device->flush_wait);
4282 bio->bi_private = &device->flush_wait;
4283
4284 btrfsic_check_bio(bio);
4285 submit_bio(bio);
4286 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4287}
4288
4289/*
4290 * If the flush bio has been submitted by write_dev_flush, wait for it.
4291 */
4292static blk_status_t wait_dev_flush(struct btrfs_device *device)
4293{
4294 struct bio *bio = &device->flush_bio;
4295
4296 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4297 return BLK_STS_OK;
4298
4299 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4300 wait_for_completion_io(&device->flush_wait);
4301
4302 return bio->bi_status;
4303}
4304
4305static int check_barrier_error(struct btrfs_fs_info *fs_info)
4306{
4307 if (!btrfs_check_rw_degradable(fs_info, NULL))
4308 return -EIO;
4309 return 0;
4310}
4311
4312/*
4313 * send an empty flush down to each device in parallel,
4314 * then wait for them
4315 */
4316static int barrier_all_devices(struct btrfs_fs_info *info)
4317{
4318 struct list_head *head;
4319 struct btrfs_device *dev;
4320 int errors_wait = 0;
4321 blk_status_t ret;
4322
4323 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4324 /* send down all the barriers */
4325 head = &info->fs_devices->devices;
4326 list_for_each_entry(dev, head, dev_list) {
4327 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4328 continue;
4329 if (!dev->bdev)
4330 continue;
4331 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4332 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4333 continue;
4334
4335 write_dev_flush(dev);
4336 dev->last_flush_error = BLK_STS_OK;
4337 }
4338
4339 /* wait for all the barriers */
4340 list_for_each_entry(dev, head, dev_list) {
4341 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4342 continue;
4343 if (!dev->bdev) {
4344 errors_wait++;
4345 continue;
4346 }
4347 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4348 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4349 continue;
4350
4351 ret = wait_dev_flush(dev);
4352 if (ret) {
4353 dev->last_flush_error = ret;
4354 btrfs_dev_stat_inc_and_print(dev,
4355 BTRFS_DEV_STAT_FLUSH_ERRS);
4356 errors_wait++;
4357 }
4358 }
4359
4360 if (errors_wait) {
4361 /*
4362 * At some point we need the status of all disks
4363 * to arrive at the volume status. So error checking
4364 * is being pushed to a separate loop.
4365 */
4366 return check_barrier_error(info);
4367 }
4368 return 0;
4369}
4370
4371int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4372{
4373 int raid_type;
4374 int min_tolerated = INT_MAX;
4375
4376 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4377 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4378 min_tolerated = min_t(int, min_tolerated,
4379 btrfs_raid_array[BTRFS_RAID_SINGLE].
4380 tolerated_failures);
4381
4382 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4383 if (raid_type == BTRFS_RAID_SINGLE)
4384 continue;
4385 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4386 continue;
4387 min_tolerated = min_t(int, min_tolerated,
4388 btrfs_raid_array[raid_type].
4389 tolerated_failures);
4390 }
4391
4392 if (min_tolerated == INT_MAX) {
4393 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4394 min_tolerated = 0;
4395 }
4396
4397 return min_tolerated;
4398}
4399
4400int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4401{
4402 struct list_head *head;
4403 struct btrfs_device *dev;
4404 struct btrfs_super_block *sb;
4405 struct btrfs_dev_item *dev_item;
4406 int ret;
4407 int do_barriers;
4408 int max_errors;
4409 int total_errors = 0;
4410 u64 flags;
4411
4412 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4413
4414 /*
4415 * max_mirrors == 0 indicates we're from commit_transaction,
4416 * not from fsync where the tree roots in fs_info have not
4417 * been consistent on disk.
4418 */
4419 if (max_mirrors == 0)
4420 backup_super_roots(fs_info);
4421
4422 sb = fs_info->super_for_commit;
4423 dev_item = &sb->dev_item;
4424
4425 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4426 head = &fs_info->fs_devices->devices;
4427 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4428
4429 if (do_barriers) {
4430 ret = barrier_all_devices(fs_info);
4431 if (ret) {
4432 mutex_unlock(
4433 &fs_info->fs_devices->device_list_mutex);
4434 btrfs_handle_fs_error(fs_info, ret,
4435 "errors while submitting device barriers.");
4436 return ret;
4437 }
4438 }
4439
4440 list_for_each_entry(dev, head, dev_list) {
4441 if (!dev->bdev) {
4442 total_errors++;
4443 continue;
4444 }
4445 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4446 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4447 continue;
4448
4449 btrfs_set_stack_device_generation(dev_item, 0);
4450 btrfs_set_stack_device_type(dev_item, dev->type);
4451 btrfs_set_stack_device_id(dev_item, dev->devid);
4452 btrfs_set_stack_device_total_bytes(dev_item,
4453 dev->commit_total_bytes);
4454 btrfs_set_stack_device_bytes_used(dev_item,
4455 dev->commit_bytes_used);
4456 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4457 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4458 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4459 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4460 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4461 BTRFS_FSID_SIZE);
4462
4463 flags = btrfs_super_flags(sb);
4464 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4465
4466 ret = btrfs_validate_write_super(fs_info, sb);
4467 if (ret < 0) {
4468 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4469 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4470 "unexpected superblock corruption detected");
4471 return -EUCLEAN;
4472 }
4473
4474 ret = write_dev_supers(dev, sb, max_mirrors);
4475 if (ret)
4476 total_errors++;
4477 }
4478 if (total_errors > max_errors) {
4479 btrfs_err(fs_info, "%d errors while writing supers",
4480 total_errors);
4481 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4482
4483 /* FUA is masked off if unsupported and can't be the reason */
4484 btrfs_handle_fs_error(fs_info, -EIO,
4485 "%d errors while writing supers",
4486 total_errors);
4487 return -EIO;
4488 }
4489
4490 total_errors = 0;
4491 list_for_each_entry(dev, head, dev_list) {
4492 if (!dev->bdev)
4493 continue;
4494 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4495 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4496 continue;
4497
4498 ret = wait_dev_supers(dev, max_mirrors);
4499 if (ret)
4500 total_errors++;
4501 }
4502 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4503 if (total_errors > max_errors) {
4504 btrfs_handle_fs_error(fs_info, -EIO,
4505 "%d errors while writing supers",
4506 total_errors);
4507 return -EIO;
4508 }
4509 return 0;
4510}
4511
4512/* Drop a fs root from the radix tree and free it. */
4513void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4514 struct btrfs_root *root)
4515{
4516 bool drop_ref = false;
4517
4518 spin_lock(&fs_info->fs_roots_radix_lock);
4519 radix_tree_delete(&fs_info->fs_roots_radix,
4520 (unsigned long)root->root_key.objectid);
4521 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4522 drop_ref = true;
4523 spin_unlock(&fs_info->fs_roots_radix_lock);
4524
4525 if (BTRFS_FS_ERROR(fs_info)) {
4526 ASSERT(root->log_root == NULL);
4527 if (root->reloc_root) {
4528 btrfs_put_root(root->reloc_root);
4529 root->reloc_root = NULL;
4530 }
4531 }
4532
4533 if (drop_ref)
4534 btrfs_put_root(root);
4535}
4536
4537int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4538{
4539 u64 root_objectid = 0;
4540 struct btrfs_root *gang[8];
4541 int i = 0;
4542 int err = 0;
4543 unsigned int ret = 0;
4544
4545 while (1) {
4546 spin_lock(&fs_info->fs_roots_radix_lock);
4547 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4548 (void **)gang, root_objectid,
4549 ARRAY_SIZE(gang));
4550 if (!ret) {
4551 spin_unlock(&fs_info->fs_roots_radix_lock);
4552 break;
4553 }
4554 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4555
4556 for (i = 0; i < ret; i++) {
4557 /* Avoid to grab roots in dead_roots */
4558 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4559 gang[i] = NULL;
4560 continue;
4561 }
4562 /* grab all the search result for later use */
4563 gang[i] = btrfs_grab_root(gang[i]);
4564 }
4565 spin_unlock(&fs_info->fs_roots_radix_lock);
4566
4567 for (i = 0; i < ret; i++) {
4568 if (!gang[i])
4569 continue;
4570 root_objectid = gang[i]->root_key.objectid;
4571 err = btrfs_orphan_cleanup(gang[i]);
4572 if (err)
4573 break;
4574 btrfs_put_root(gang[i]);
4575 }
4576 root_objectid++;
4577 }
4578
4579 /* release the uncleaned roots due to error */
4580 for (; i < ret; i++) {
4581 if (gang[i])
4582 btrfs_put_root(gang[i]);
4583 }
4584 return err;
4585}
4586
4587int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4588{
4589 struct btrfs_root *root = fs_info->tree_root;
4590 struct btrfs_trans_handle *trans;
4591
4592 mutex_lock(&fs_info->cleaner_mutex);
4593 btrfs_run_delayed_iputs(fs_info);
4594 mutex_unlock(&fs_info->cleaner_mutex);
4595 wake_up_process(fs_info->cleaner_kthread);
4596
4597 /* wait until ongoing cleanup work done */
4598 down_write(&fs_info->cleanup_work_sem);
4599 up_write(&fs_info->cleanup_work_sem);
4600
4601 trans = btrfs_join_transaction(root);
4602 if (IS_ERR(trans))
4603 return PTR_ERR(trans);
4604 return btrfs_commit_transaction(trans);
4605}
4606
4607static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4608{
4609 struct btrfs_transaction *trans;
4610 struct btrfs_transaction *tmp;
4611 bool found = false;
4612
4613 if (list_empty(&fs_info->trans_list))
4614 return;
4615
4616 /*
4617 * This function is only called at the very end of close_ctree(),
4618 * thus no other running transaction, no need to take trans_lock.
4619 */
4620 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4621 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4622 struct extent_state *cached = NULL;
4623 u64 dirty_bytes = 0;
4624 u64 cur = 0;
4625 u64 found_start;
4626 u64 found_end;
4627
4628 found = true;
4629 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4630 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4631 dirty_bytes += found_end + 1 - found_start;
4632 cur = found_end + 1;
4633 }
4634 btrfs_warn(fs_info,
4635 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4636 trans->transid, dirty_bytes);
4637 btrfs_cleanup_one_transaction(trans, fs_info);
4638
4639 if (trans == fs_info->running_transaction)
4640 fs_info->running_transaction = NULL;
4641 list_del_init(&trans->list);
4642
4643 btrfs_put_transaction(trans);
4644 trace_btrfs_transaction_commit(fs_info);
4645 }
4646 ASSERT(!found);
4647}
4648
4649void __cold close_ctree(struct btrfs_fs_info *fs_info)
4650{
4651 int ret;
4652
4653 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4654
4655 /*
4656 * If we had UNFINISHED_DROPS we could still be processing them, so
4657 * clear that bit and wake up relocation so it can stop.
4658 * We must do this before stopping the block group reclaim task, because
4659 * at btrfs_relocate_block_group() we wait for this bit, and after the
4660 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4661 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4662 * return 1.
4663 */
4664 btrfs_wake_unfinished_drop(fs_info);
4665
4666 /*
4667 * We may have the reclaim task running and relocating a data block group,
4668 * in which case it may create delayed iputs. So stop it before we park
4669 * the cleaner kthread otherwise we can get new delayed iputs after
4670 * parking the cleaner, and that can make the async reclaim task to hang
4671 * if it's waiting for delayed iputs to complete, since the cleaner is
4672 * parked and can not run delayed iputs - this will make us hang when
4673 * trying to stop the async reclaim task.
4674 */
4675 cancel_work_sync(&fs_info->reclaim_bgs_work);
4676 /*
4677 * We don't want the cleaner to start new transactions, add more delayed
4678 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4679 * because that frees the task_struct, and the transaction kthread might
4680 * still try to wake up the cleaner.
4681 */
4682 kthread_park(fs_info->cleaner_kthread);
4683
4684 /* wait for the qgroup rescan worker to stop */
4685 btrfs_qgroup_wait_for_completion(fs_info, false);
4686
4687 /* wait for the uuid_scan task to finish */
4688 down(&fs_info->uuid_tree_rescan_sem);
4689 /* avoid complains from lockdep et al., set sem back to initial state */
4690 up(&fs_info->uuid_tree_rescan_sem);
4691
4692 /* pause restriper - we want to resume on mount */
4693 btrfs_pause_balance(fs_info);
4694
4695 btrfs_dev_replace_suspend_for_unmount(fs_info);
4696
4697 btrfs_scrub_cancel(fs_info);
4698
4699 /* wait for any defraggers to finish */
4700 wait_event(fs_info->transaction_wait,
4701 (atomic_read(&fs_info->defrag_running) == 0));
4702
4703 /* clear out the rbtree of defraggable inodes */
4704 btrfs_cleanup_defrag_inodes(fs_info);
4705
4706 /*
4707 * After we parked the cleaner kthread, ordered extents may have
4708 * completed and created new delayed iputs. If one of the async reclaim
4709 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4710 * can hang forever trying to stop it, because if a delayed iput is
4711 * added after it ran btrfs_run_delayed_iputs() and before it called
4712 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4713 * no one else to run iputs.
4714 *
4715 * So wait for all ongoing ordered extents to complete and then run
4716 * delayed iputs. This works because once we reach this point no one
4717 * can either create new ordered extents nor create delayed iputs
4718 * through some other means.
4719 *
4720 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4721 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4722 * but the delayed iput for the respective inode is made only when doing
4723 * the final btrfs_put_ordered_extent() (which must happen at
4724 * btrfs_finish_ordered_io() when we are unmounting).
4725 */
4726 btrfs_flush_workqueue(fs_info->endio_write_workers);
4727 /* Ordered extents for free space inodes. */
4728 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4729 btrfs_run_delayed_iputs(fs_info);
4730
4731 cancel_work_sync(&fs_info->async_reclaim_work);
4732 cancel_work_sync(&fs_info->async_data_reclaim_work);
4733 cancel_work_sync(&fs_info->preempt_reclaim_work);
4734
4735 /* Cancel or finish ongoing discard work */
4736 btrfs_discard_cleanup(fs_info);
4737
4738 if (!sb_rdonly(fs_info->sb)) {
4739 /*
4740 * The cleaner kthread is stopped, so do one final pass over
4741 * unused block groups.
4742 */
4743 btrfs_delete_unused_bgs(fs_info);
4744
4745 /*
4746 * There might be existing delayed inode workers still running
4747 * and holding an empty delayed inode item. We must wait for
4748 * them to complete first because they can create a transaction.
4749 * This happens when someone calls btrfs_balance_delayed_items()
4750 * and then a transaction commit runs the same delayed nodes
4751 * before any delayed worker has done something with the nodes.
4752 * We must wait for any worker here and not at transaction
4753 * commit time since that could cause a deadlock.
4754 * This is a very rare case.
4755 */
4756 btrfs_flush_workqueue(fs_info->delayed_workers);
4757
4758 ret = btrfs_commit_super(fs_info);
4759 if (ret)
4760 btrfs_err(fs_info, "commit super ret %d", ret);
4761 }
4762
4763 if (BTRFS_FS_ERROR(fs_info))
4764 btrfs_error_commit_super(fs_info);
4765
4766 kthread_stop(fs_info->transaction_kthread);
4767 kthread_stop(fs_info->cleaner_kthread);
4768
4769 ASSERT(list_empty(&fs_info->delayed_iputs));
4770 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4771
4772 if (btrfs_check_quota_leak(fs_info)) {
4773 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4774 btrfs_err(fs_info, "qgroup reserved space leaked");
4775 }
4776
4777 btrfs_free_qgroup_config(fs_info);
4778 ASSERT(list_empty(&fs_info->delalloc_roots));
4779
4780 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4781 btrfs_info(fs_info, "at unmount delalloc count %lld",
4782 percpu_counter_sum(&fs_info->delalloc_bytes));
4783 }
4784
4785 if (percpu_counter_sum(&fs_info->ordered_bytes))
4786 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4787 percpu_counter_sum(&fs_info->ordered_bytes));
4788
4789 btrfs_sysfs_remove_mounted(fs_info);
4790 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4791
4792 btrfs_put_block_group_cache(fs_info);
4793
4794 /*
4795 * we must make sure there is not any read request to
4796 * submit after we stopping all workers.
4797 */
4798 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4799 btrfs_stop_all_workers(fs_info);
4800
4801 /* We shouldn't have any transaction open at this point */
4802 warn_about_uncommitted_trans(fs_info);
4803
4804 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4805 free_root_pointers(fs_info, true);
4806 btrfs_free_fs_roots(fs_info);
4807
4808 /*
4809 * We must free the block groups after dropping the fs_roots as we could
4810 * have had an IO error and have left over tree log blocks that aren't
4811 * cleaned up until the fs roots are freed. This makes the block group
4812 * accounting appear to be wrong because there's pending reserved bytes,
4813 * so make sure we do the block group cleanup afterwards.
4814 */
4815 btrfs_free_block_groups(fs_info);
4816
4817 iput(fs_info->btree_inode);
4818
4819#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4820 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4821 btrfsic_unmount(fs_info->fs_devices);
4822#endif
4823
4824 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4825 btrfs_close_devices(fs_info->fs_devices);
4826}
4827
4828int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4829 int atomic)
4830{
4831 int ret;
4832 struct inode *btree_inode = buf->pages[0]->mapping->host;
4833
4834 ret = extent_buffer_uptodate(buf);
4835 if (!ret)
4836 return ret;
4837
4838 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4839 parent_transid, atomic);
4840 if (ret == -EAGAIN)
4841 return ret;
4842 return !ret;
4843}
4844
4845void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4846{
4847 struct btrfs_fs_info *fs_info = buf->fs_info;
4848 u64 transid = btrfs_header_generation(buf);
4849 int was_dirty;
4850
4851#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4852 /*
4853 * This is a fast path so only do this check if we have sanity tests
4854 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4855 * outside of the sanity tests.
4856 */
4857 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4858 return;
4859#endif
4860 btrfs_assert_tree_write_locked(buf);
4861 if (transid != fs_info->generation)
4862 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4863 buf->start, transid, fs_info->generation);
4864 was_dirty = set_extent_buffer_dirty(buf);
4865 if (!was_dirty)
4866 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4867 buf->len,
4868 fs_info->dirty_metadata_batch);
4869#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4870 /*
4871 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4872 * but item data not updated.
4873 * So here we should only check item pointers, not item data.
4874 */
4875 if (btrfs_header_level(buf) == 0 &&
4876 btrfs_check_leaf_relaxed(buf)) {
4877 btrfs_print_leaf(buf);
4878 ASSERT(0);
4879 }
4880#endif
4881}
4882
4883static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4884 int flush_delayed)
4885{
4886 /*
4887 * looks as though older kernels can get into trouble with
4888 * this code, they end up stuck in balance_dirty_pages forever
4889 */
4890 int ret;
4891
4892 if (current->flags & PF_MEMALLOC)
4893 return;
4894
4895 if (flush_delayed)
4896 btrfs_balance_delayed_items(fs_info);
4897
4898 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4899 BTRFS_DIRTY_METADATA_THRESH,
4900 fs_info->dirty_metadata_batch);
4901 if (ret > 0) {
4902 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4903 }
4904}
4905
4906void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4907{
4908 __btrfs_btree_balance_dirty(fs_info, 1);
4909}
4910
4911void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4912{
4913 __btrfs_btree_balance_dirty(fs_info, 0);
4914}
4915
4916static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4917{
4918 /* cleanup FS via transaction */
4919 btrfs_cleanup_transaction(fs_info);
4920
4921 mutex_lock(&fs_info->cleaner_mutex);
4922 btrfs_run_delayed_iputs(fs_info);
4923 mutex_unlock(&fs_info->cleaner_mutex);
4924
4925 down_write(&fs_info->cleanup_work_sem);
4926 up_write(&fs_info->cleanup_work_sem);
4927}
4928
4929static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4930{
4931 struct btrfs_root *gang[8];
4932 u64 root_objectid = 0;
4933 int ret;
4934
4935 spin_lock(&fs_info->fs_roots_radix_lock);
4936 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4937 (void **)gang, root_objectid,
4938 ARRAY_SIZE(gang))) != 0) {
4939 int i;
4940
4941 for (i = 0; i < ret; i++)
4942 gang[i] = btrfs_grab_root(gang[i]);
4943 spin_unlock(&fs_info->fs_roots_radix_lock);
4944
4945 for (i = 0; i < ret; i++) {
4946 if (!gang[i])
4947 continue;
4948 root_objectid = gang[i]->root_key.objectid;
4949 btrfs_free_log(NULL, gang[i]);
4950 btrfs_put_root(gang[i]);
4951 }
4952 root_objectid++;
4953 spin_lock(&fs_info->fs_roots_radix_lock);
4954 }
4955 spin_unlock(&fs_info->fs_roots_radix_lock);
4956 btrfs_free_log_root_tree(NULL, fs_info);
4957}
4958
4959static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4960{
4961 struct btrfs_ordered_extent *ordered;
4962
4963 spin_lock(&root->ordered_extent_lock);
4964 /*
4965 * This will just short circuit the ordered completion stuff which will
4966 * make sure the ordered extent gets properly cleaned up.
4967 */
4968 list_for_each_entry(ordered, &root->ordered_extents,
4969 root_extent_list)
4970 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4971 spin_unlock(&root->ordered_extent_lock);
4972}
4973
4974static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4975{
4976 struct btrfs_root *root;
4977 struct list_head splice;
4978
4979 INIT_LIST_HEAD(&splice);
4980
4981 spin_lock(&fs_info->ordered_root_lock);
4982 list_splice_init(&fs_info->ordered_roots, &splice);
4983 while (!list_empty(&splice)) {
4984 root = list_first_entry(&splice, struct btrfs_root,
4985 ordered_root);
4986 list_move_tail(&root->ordered_root,
4987 &fs_info->ordered_roots);
4988
4989 spin_unlock(&fs_info->ordered_root_lock);
4990 btrfs_destroy_ordered_extents(root);
4991
4992 cond_resched();
4993 spin_lock(&fs_info->ordered_root_lock);
4994 }
4995 spin_unlock(&fs_info->ordered_root_lock);
4996
4997 /*
4998 * We need this here because if we've been flipped read-only we won't
4999 * get sync() from the umount, so we need to make sure any ordered
5000 * extents that haven't had their dirty pages IO start writeout yet
5001 * actually get run and error out properly.
5002 */
5003 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
5004}
5005
5006static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
5007 struct btrfs_fs_info *fs_info)
5008{
5009 struct rb_node *node;
5010 struct btrfs_delayed_ref_root *delayed_refs;
5011 struct btrfs_delayed_ref_node *ref;
5012 int ret = 0;
5013
5014 delayed_refs = &trans->delayed_refs;
5015
5016 spin_lock(&delayed_refs->lock);
5017 if (atomic_read(&delayed_refs->num_entries) == 0) {
5018 spin_unlock(&delayed_refs->lock);
5019 btrfs_debug(fs_info, "delayed_refs has NO entry");
5020 return ret;
5021 }
5022
5023 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
5024 struct btrfs_delayed_ref_head *head;
5025 struct rb_node *n;
5026 bool pin_bytes = false;
5027
5028 head = rb_entry(node, struct btrfs_delayed_ref_head,
5029 href_node);
5030 if (btrfs_delayed_ref_lock(delayed_refs, head))
5031 continue;
5032
5033 spin_lock(&head->lock);
5034 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
5035 ref = rb_entry(n, struct btrfs_delayed_ref_node,
5036 ref_node);
5037 ref->in_tree = 0;
5038 rb_erase_cached(&ref->ref_node, &head->ref_tree);
5039 RB_CLEAR_NODE(&ref->ref_node);
5040 if (!list_empty(&ref->add_list))
5041 list_del(&ref->add_list);
5042 atomic_dec(&delayed_refs->num_entries);
5043 btrfs_put_delayed_ref(ref);
5044 }
5045 if (head->must_insert_reserved)
5046 pin_bytes = true;
5047 btrfs_free_delayed_extent_op(head->extent_op);
5048 btrfs_delete_ref_head(delayed_refs, head);
5049 spin_unlock(&head->lock);
5050 spin_unlock(&delayed_refs->lock);
5051 mutex_unlock(&head->mutex);
5052
5053 if (pin_bytes) {
5054 struct btrfs_block_group *cache;
5055
5056 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
5057 BUG_ON(!cache);
5058
5059 spin_lock(&cache->space_info->lock);
5060 spin_lock(&cache->lock);
5061 cache->pinned += head->num_bytes;
5062 btrfs_space_info_update_bytes_pinned(fs_info,
5063 cache->space_info, head->num_bytes);
5064 cache->reserved -= head->num_bytes;
5065 cache->space_info->bytes_reserved -= head->num_bytes;
5066 spin_unlock(&cache->lock);
5067 spin_unlock(&cache->space_info->lock);
5068
5069 btrfs_put_block_group(cache);
5070
5071 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
5072 head->bytenr + head->num_bytes - 1);
5073 }
5074 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
5075 btrfs_put_delayed_ref_head(head);
5076 cond_resched();
5077 spin_lock(&delayed_refs->lock);
5078 }
5079 btrfs_qgroup_destroy_extent_records(trans);
5080
5081 spin_unlock(&delayed_refs->lock);
5082
5083 return ret;
5084}
5085
5086static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
5087{
5088 struct btrfs_inode *btrfs_inode;
5089 struct list_head splice;
5090
5091 INIT_LIST_HEAD(&splice);
5092
5093 spin_lock(&root->delalloc_lock);
5094 list_splice_init(&root->delalloc_inodes, &splice);
5095
5096 while (!list_empty(&splice)) {
5097 struct inode *inode = NULL;
5098 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5099 delalloc_inodes);
5100 __btrfs_del_delalloc_inode(root, btrfs_inode);
5101 spin_unlock(&root->delalloc_lock);
5102
5103 /*
5104 * Make sure we get a live inode and that it'll not disappear
5105 * meanwhile.
5106 */
5107 inode = igrab(&btrfs_inode->vfs_inode);
5108 if (inode) {
5109 invalidate_inode_pages2(inode->i_mapping);
5110 iput(inode);
5111 }
5112 spin_lock(&root->delalloc_lock);
5113 }
5114 spin_unlock(&root->delalloc_lock);
5115}
5116
5117static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5118{
5119 struct btrfs_root *root;
5120 struct list_head splice;
5121
5122 INIT_LIST_HEAD(&splice);
5123
5124 spin_lock(&fs_info->delalloc_root_lock);
5125 list_splice_init(&fs_info->delalloc_roots, &splice);
5126 while (!list_empty(&splice)) {
5127 root = list_first_entry(&splice, struct btrfs_root,
5128 delalloc_root);
5129 root = btrfs_grab_root(root);
5130 BUG_ON(!root);
5131 spin_unlock(&fs_info->delalloc_root_lock);
5132
5133 btrfs_destroy_delalloc_inodes(root);
5134 btrfs_put_root(root);
5135
5136 spin_lock(&fs_info->delalloc_root_lock);
5137 }
5138 spin_unlock(&fs_info->delalloc_root_lock);
5139}
5140
5141static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5142 struct extent_io_tree *dirty_pages,
5143 int mark)
5144{
5145 int ret;
5146 struct extent_buffer *eb;
5147 u64 start = 0;
5148 u64 end;
5149
5150 while (1) {
5151 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5152 mark, NULL);
5153 if (ret)
5154 break;
5155
5156 clear_extent_bits(dirty_pages, start, end, mark);
5157 while (start <= end) {
5158 eb = find_extent_buffer(fs_info, start);
5159 start += fs_info->nodesize;
5160 if (!eb)
5161 continue;
5162 wait_on_extent_buffer_writeback(eb);
5163
5164 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5165 &eb->bflags))
5166 clear_extent_buffer_dirty(eb);
5167 free_extent_buffer_stale(eb);
5168 }
5169 }
5170
5171 return ret;
5172}
5173
5174static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5175 struct extent_io_tree *unpin)
5176{
5177 u64 start;
5178 u64 end;
5179 int ret;
5180
5181 while (1) {
5182 struct extent_state *cached_state = NULL;
5183
5184 /*
5185 * The btrfs_finish_extent_commit() may get the same range as
5186 * ours between find_first_extent_bit and clear_extent_dirty.
5187 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5188 * the same extent range.
5189 */
5190 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5191 ret = find_first_extent_bit(unpin, 0, &start, &end,
5192 EXTENT_DIRTY, &cached_state);
5193 if (ret) {
5194 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5195 break;
5196 }
5197
5198 clear_extent_dirty(unpin, start, end, &cached_state);
5199 free_extent_state(cached_state);
5200 btrfs_error_unpin_extent_range(fs_info, start, end);
5201 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5202 cond_resched();
5203 }
5204
5205 return 0;
5206}
5207
5208static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5209{
5210 struct inode *inode;
5211
5212 inode = cache->io_ctl.inode;
5213 if (inode) {
5214 invalidate_inode_pages2(inode->i_mapping);
5215 BTRFS_I(inode)->generation = 0;
5216 cache->io_ctl.inode = NULL;
5217 iput(inode);
5218 }
5219 ASSERT(cache->io_ctl.pages == NULL);
5220 btrfs_put_block_group(cache);
5221}
5222
5223void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5224 struct btrfs_fs_info *fs_info)
5225{
5226 struct btrfs_block_group *cache;
5227
5228 spin_lock(&cur_trans->dirty_bgs_lock);
5229 while (!list_empty(&cur_trans->dirty_bgs)) {
5230 cache = list_first_entry(&cur_trans->dirty_bgs,
5231 struct btrfs_block_group,
5232 dirty_list);
5233
5234 if (!list_empty(&cache->io_list)) {
5235 spin_unlock(&cur_trans->dirty_bgs_lock);
5236 list_del_init(&cache->io_list);
5237 btrfs_cleanup_bg_io(cache);
5238 spin_lock(&cur_trans->dirty_bgs_lock);
5239 }
5240
5241 list_del_init(&cache->dirty_list);
5242 spin_lock(&cache->lock);
5243 cache->disk_cache_state = BTRFS_DC_ERROR;
5244 spin_unlock(&cache->lock);
5245
5246 spin_unlock(&cur_trans->dirty_bgs_lock);
5247 btrfs_put_block_group(cache);
5248 btrfs_delayed_refs_rsv_release(fs_info, 1);
5249 spin_lock(&cur_trans->dirty_bgs_lock);
5250 }
5251 spin_unlock(&cur_trans->dirty_bgs_lock);
5252
5253 /*
5254 * Refer to the definition of io_bgs member for details why it's safe
5255 * to use it without any locking
5256 */
5257 while (!list_empty(&cur_trans->io_bgs)) {
5258 cache = list_first_entry(&cur_trans->io_bgs,
5259 struct btrfs_block_group,
5260 io_list);
5261
5262 list_del_init(&cache->io_list);
5263 spin_lock(&cache->lock);
5264 cache->disk_cache_state = BTRFS_DC_ERROR;
5265 spin_unlock(&cache->lock);
5266 btrfs_cleanup_bg_io(cache);
5267 }
5268}
5269
5270void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5271 struct btrfs_fs_info *fs_info)
5272{
5273 struct btrfs_device *dev, *tmp;
5274
5275 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5276 ASSERT(list_empty(&cur_trans->dirty_bgs));
5277 ASSERT(list_empty(&cur_trans->io_bgs));
5278
5279 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5280 post_commit_list) {
5281 list_del_init(&dev->post_commit_list);
5282 }
5283
5284 btrfs_destroy_delayed_refs(cur_trans, fs_info);
5285
5286 cur_trans->state = TRANS_STATE_COMMIT_START;
5287 wake_up(&fs_info->transaction_blocked_wait);
5288
5289 cur_trans->state = TRANS_STATE_UNBLOCKED;
5290 wake_up(&fs_info->transaction_wait);
5291
5292 btrfs_destroy_delayed_inodes(fs_info);
5293
5294 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5295 EXTENT_DIRTY);
5296 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5297
5298 btrfs_free_redirty_list(cur_trans);
5299
5300 cur_trans->state =TRANS_STATE_COMPLETED;
5301 wake_up(&cur_trans->commit_wait);
5302}
5303
5304static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5305{
5306 struct btrfs_transaction *t;
5307
5308 mutex_lock(&fs_info->transaction_kthread_mutex);
5309
5310 spin_lock(&fs_info->trans_lock);
5311 while (!list_empty(&fs_info->trans_list)) {
5312 t = list_first_entry(&fs_info->trans_list,
5313 struct btrfs_transaction, list);
5314 if (t->state >= TRANS_STATE_COMMIT_START) {
5315 refcount_inc(&t->use_count);
5316 spin_unlock(&fs_info->trans_lock);
5317 btrfs_wait_for_commit(fs_info, t->transid);
5318 btrfs_put_transaction(t);
5319 spin_lock(&fs_info->trans_lock);
5320 continue;
5321 }
5322 if (t == fs_info->running_transaction) {
5323 t->state = TRANS_STATE_COMMIT_DOING;
5324 spin_unlock(&fs_info->trans_lock);
5325 /*
5326 * We wait for 0 num_writers since we don't hold a trans
5327 * handle open currently for this transaction.
5328 */
5329 wait_event(t->writer_wait,
5330 atomic_read(&t->num_writers) == 0);
5331 } else {
5332 spin_unlock(&fs_info->trans_lock);
5333 }
5334 btrfs_cleanup_one_transaction(t, fs_info);
5335
5336 spin_lock(&fs_info->trans_lock);
5337 if (t == fs_info->running_transaction)
5338 fs_info->running_transaction = NULL;
5339 list_del_init(&t->list);
5340 spin_unlock(&fs_info->trans_lock);
5341
5342 btrfs_put_transaction(t);
5343 trace_btrfs_transaction_commit(fs_info);
5344 spin_lock(&fs_info->trans_lock);
5345 }
5346 spin_unlock(&fs_info->trans_lock);
5347 btrfs_destroy_all_ordered_extents(fs_info);
5348 btrfs_destroy_delayed_inodes(fs_info);
5349 btrfs_assert_delayed_root_empty(fs_info);
5350 btrfs_destroy_all_delalloc_inodes(fs_info);
5351 btrfs_drop_all_logs(fs_info);
5352 mutex_unlock(&fs_info->transaction_kthread_mutex);
5353
5354 return 0;
5355}
5356
5357int btrfs_init_root_free_objectid(struct btrfs_root *root)
5358{
5359 struct btrfs_path *path;
5360 int ret;
5361 struct extent_buffer *l;
5362 struct btrfs_key search_key;
5363 struct btrfs_key found_key;
5364 int slot;
5365
5366 path = btrfs_alloc_path();
5367 if (!path)
5368 return -ENOMEM;
5369
5370 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5371 search_key.type = -1;
5372 search_key.offset = (u64)-1;
5373 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5374 if (ret < 0)
5375 goto error;
5376 BUG_ON(ret == 0); /* Corruption */
5377 if (path->slots[0] > 0) {
5378 slot = path->slots[0] - 1;
5379 l = path->nodes[0];
5380 btrfs_item_key_to_cpu(l, &found_key, slot);
5381 root->free_objectid = max_t(u64, found_key.objectid + 1,
5382 BTRFS_FIRST_FREE_OBJECTID);
5383 } else {
5384 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5385 }
5386 ret = 0;
5387error:
5388 btrfs_free_path(path);
5389 return ret;
5390}
5391
5392int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5393{
5394 int ret;
5395 mutex_lock(&root->objectid_mutex);
5396
5397 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5398 btrfs_warn(root->fs_info,
5399 "the objectid of root %llu reaches its highest value",
5400 root->root_key.objectid);
5401 ret = -ENOSPC;
5402 goto out;
5403 }
5404
5405 *objectid = root->free_objectid++;
5406 ret = 0;
5407out:
5408 mutex_unlock(&root->objectid_mutex);
5409 return ret;
5410}