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 "volumes.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
47#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
48 BTRFS_HEADER_FLAG_RELOC |\
49 BTRFS_SUPER_FLAG_ERROR |\
50 BTRFS_SUPER_FLAG_SEEDING |\
51 BTRFS_SUPER_FLAG_METADUMP |\
52 BTRFS_SUPER_FLAG_METADUMP_V2)
53
54static void end_workqueue_fn(struct btrfs_work *work);
55static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_fs_info *fs_info);
58static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 struct extent_io_tree *dirty_pages,
61 int mark);
62static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *pinned_extents);
64static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66
67/*
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
71 */
72struct btrfs_end_io_wq {
73 struct bio *bio;
74 bio_end_io_t *end_io;
75 void *private;
76 struct btrfs_fs_info *info;
77 blk_status_t status;
78 enum btrfs_wq_endio_type metadata;
79 struct btrfs_work work;
80};
81
82static struct kmem_cache *btrfs_end_io_wq_cache;
83
84int __init btrfs_end_io_wq_init(void)
85{
86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq),
88 0,
89 SLAB_MEM_SPREAD,
90 NULL);
91 if (!btrfs_end_io_wq_cache)
92 return -ENOMEM;
93 return 0;
94}
95
96void __cold btrfs_end_io_wq_exit(void)
97{
98 kmem_cache_destroy(btrfs_end_io_wq_cache);
99}
100
101static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102{
103 if (fs_info->csum_shash)
104 crypto_free_shash(fs_info->csum_shash);
105}
106
107/*
108 * async submit bios are used to offload expensive checksumming
109 * onto the worker threads. They checksum file and metadata bios
110 * just before they are sent down the IO stack.
111 */
112struct async_submit_bio {
113 struct inode *inode;
114 struct bio *bio;
115 extent_submit_bio_start_t *submit_bio_start;
116 int mirror_num;
117
118 /* Optional parameter for submit_bio_start used by direct io */
119 u64 dio_file_offset;
120 struct btrfs_work work;
121 blk_status_t status;
122};
123
124/*
125 * Lockdep class keys for extent_buffer->lock's in this root. For a given
126 * eb, the lockdep key is determined by the btrfs_root it belongs to and
127 * the level the eb occupies in the tree.
128 *
129 * Different roots are used for different purposes and may nest inside each
130 * other and they require separate keysets. As lockdep keys should be
131 * static, assign keysets according to the purpose of the root as indicated
132 * by btrfs_root->root_key.objectid. This ensures that all special purpose
133 * roots have separate keysets.
134 *
135 * Lock-nesting across peer nodes is always done with the immediate parent
136 * node locked thus preventing deadlock. As lockdep doesn't know this, use
137 * subclass to avoid triggering lockdep warning in such cases.
138 *
139 * The key is set by the readpage_end_io_hook after the buffer has passed
140 * csum validation but before the pages are unlocked. It is also set by
141 * btrfs_init_new_buffer on freshly allocated blocks.
142 *
143 * We also add a check to make sure the highest level of the tree is the
144 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
145 * needs update as well.
146 */
147#ifdef CONFIG_DEBUG_LOCK_ALLOC
148# if BTRFS_MAX_LEVEL != 8
149# error
150# endif
151
152#define DEFINE_LEVEL(stem, level) \
153 .names[level] = "btrfs-" stem "-0" #level,
154
155#define DEFINE_NAME(stem) \
156 DEFINE_LEVEL(stem, 0) \
157 DEFINE_LEVEL(stem, 1) \
158 DEFINE_LEVEL(stem, 2) \
159 DEFINE_LEVEL(stem, 3) \
160 DEFINE_LEVEL(stem, 4) \
161 DEFINE_LEVEL(stem, 5) \
162 DEFINE_LEVEL(stem, 6) \
163 DEFINE_LEVEL(stem, 7)
164
165static struct btrfs_lockdep_keyset {
166 u64 id; /* root objectid */
167 /* Longest entry: btrfs-free-space-00 */
168 char names[BTRFS_MAX_LEVEL][20];
169 struct lock_class_key keys[BTRFS_MAX_LEVEL];
170} btrfs_lockdep_keysets[] = {
171 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") },
172 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") },
173 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") },
174 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") },
175 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
176 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
177 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
178 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
179 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
180 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
181 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
182 { .id = 0, DEFINE_NAME("tree") },
183};
184
185#undef DEFINE_LEVEL
186#undef DEFINE_NAME
187
188void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
189 int level)
190{
191 struct btrfs_lockdep_keyset *ks;
192
193 BUG_ON(level >= ARRAY_SIZE(ks->keys));
194
195 /* find the matching keyset, id 0 is the default entry */
196 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
197 if (ks->id == objectid)
198 break;
199
200 lockdep_set_class_and_name(&eb->lock,
201 &ks->keys[level], ks->names[level]);
202}
203
204#endif
205
206/*
207 * Compute the csum of a btree block and store the result to provided buffer.
208 */
209static void csum_tree_block(struct extent_buffer *buf, u8 *result)
210{
211 struct btrfs_fs_info *fs_info = buf->fs_info;
212 const int num_pages = num_extent_pages(buf);
213 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
214 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
215 char *kaddr;
216 int i;
217
218 shash->tfm = fs_info->csum_shash;
219 crypto_shash_init(shash);
220 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
221 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
222 first_page_part - BTRFS_CSUM_SIZE);
223
224 for (i = 1; i < num_pages; i++) {
225 kaddr = page_address(buf->pages[i]);
226 crypto_shash_update(shash, kaddr, PAGE_SIZE);
227 }
228 memset(result, 0, BTRFS_CSUM_SIZE);
229 crypto_shash_final(shash, result);
230}
231
232/*
233 * we can't consider a given block up to date unless the transid of the
234 * block matches the transid in the parent node's pointer. This is how we
235 * detect blocks that either didn't get written at all or got written
236 * in the wrong place.
237 */
238static int verify_parent_transid(struct extent_io_tree *io_tree,
239 struct extent_buffer *eb, u64 parent_transid,
240 int atomic)
241{
242 struct extent_state *cached_state = NULL;
243 int ret;
244
245 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
246 return 0;
247
248 if (atomic)
249 return -EAGAIN;
250
251 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
252 &cached_state);
253 if (extent_buffer_uptodate(eb) &&
254 btrfs_header_generation(eb) == parent_transid) {
255 ret = 0;
256 goto out;
257 }
258 btrfs_err_rl(eb->fs_info,
259 "parent transid verify failed on %llu wanted %llu found %llu",
260 eb->start,
261 parent_transid, btrfs_header_generation(eb));
262 ret = 1;
263 clear_extent_buffer_uptodate(eb);
264out:
265 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
266 &cached_state);
267 return ret;
268}
269
270static bool btrfs_supported_super_csum(u16 csum_type)
271{
272 switch (csum_type) {
273 case BTRFS_CSUM_TYPE_CRC32:
274 case BTRFS_CSUM_TYPE_XXHASH:
275 case BTRFS_CSUM_TYPE_SHA256:
276 case BTRFS_CSUM_TYPE_BLAKE2:
277 return true;
278 default:
279 return false;
280 }
281}
282
283/*
284 * Return 0 if the superblock checksum type matches the checksum value of that
285 * algorithm. Pass the raw disk superblock data.
286 */
287static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
288 char *raw_disk_sb)
289{
290 struct btrfs_super_block *disk_sb =
291 (struct btrfs_super_block *)raw_disk_sb;
292 char result[BTRFS_CSUM_SIZE];
293 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
294
295 shash->tfm = fs_info->csum_shash;
296
297 /*
298 * The super_block structure does not span the whole
299 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
300 * filled with zeros and is included in the checksum.
301 */
302 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
303 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
304
305 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
306 return 1;
307
308 return 0;
309}
310
311int btrfs_verify_level_key(struct extent_buffer *eb, int level,
312 struct btrfs_key *first_key, u64 parent_transid)
313{
314 struct btrfs_fs_info *fs_info = eb->fs_info;
315 int found_level;
316 struct btrfs_key found_key;
317 int ret;
318
319 found_level = btrfs_header_level(eb);
320 if (found_level != level) {
321 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
322 KERN_ERR "BTRFS: tree level check failed\n");
323 btrfs_err(fs_info,
324"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
325 eb->start, level, found_level);
326 return -EIO;
327 }
328
329 if (!first_key)
330 return 0;
331
332 /*
333 * For live tree block (new tree blocks in current transaction),
334 * we need proper lock context to avoid race, which is impossible here.
335 * So we only checks tree blocks which is read from disk, whose
336 * generation <= fs_info->last_trans_committed.
337 */
338 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
339 return 0;
340
341 /* We have @first_key, so this @eb must have at least one item */
342 if (btrfs_header_nritems(eb) == 0) {
343 btrfs_err(fs_info,
344 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
345 eb->start);
346 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
347 return -EUCLEAN;
348 }
349
350 if (found_level)
351 btrfs_node_key_to_cpu(eb, &found_key, 0);
352 else
353 btrfs_item_key_to_cpu(eb, &found_key, 0);
354 ret = btrfs_comp_cpu_keys(first_key, &found_key);
355
356 if (ret) {
357 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
358 KERN_ERR "BTRFS: tree first key check failed\n");
359 btrfs_err(fs_info,
360"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
361 eb->start, parent_transid, first_key->objectid,
362 first_key->type, first_key->offset,
363 found_key.objectid, found_key.type,
364 found_key.offset);
365 }
366 return ret;
367}
368
369/*
370 * helper to read a given tree block, doing retries as required when
371 * the checksums don't match and we have alternate mirrors to try.
372 *
373 * @parent_transid: expected transid, skip check if 0
374 * @level: expected level, mandatory check
375 * @first_key: expected key of first slot, skip check if NULL
376 */
377static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
378 u64 parent_transid, int level,
379 struct btrfs_key *first_key)
380{
381 struct btrfs_fs_info *fs_info = eb->fs_info;
382 struct extent_io_tree *io_tree;
383 int failed = 0;
384 int ret;
385 int num_copies = 0;
386 int mirror_num = 0;
387 int failed_mirror = 0;
388
389 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
390 while (1) {
391 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
392 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
393 if (!ret) {
394 if (verify_parent_transid(io_tree, eb,
395 parent_transid, 0))
396 ret = -EIO;
397 else if (btrfs_verify_level_key(eb, level,
398 first_key, parent_transid))
399 ret = -EUCLEAN;
400 else
401 break;
402 }
403
404 num_copies = btrfs_num_copies(fs_info,
405 eb->start, eb->len);
406 if (num_copies == 1)
407 break;
408
409 if (!failed_mirror) {
410 failed = 1;
411 failed_mirror = eb->read_mirror;
412 }
413
414 mirror_num++;
415 if (mirror_num == failed_mirror)
416 mirror_num++;
417
418 if (mirror_num > num_copies)
419 break;
420 }
421
422 if (failed && !ret && failed_mirror)
423 btrfs_repair_eb_io_failure(eb, failed_mirror);
424
425 return ret;
426}
427
428static int csum_one_extent_buffer(struct extent_buffer *eb)
429{
430 struct btrfs_fs_info *fs_info = eb->fs_info;
431 u8 result[BTRFS_CSUM_SIZE];
432 int ret;
433
434 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
435 offsetof(struct btrfs_header, fsid),
436 BTRFS_FSID_SIZE) == 0);
437 csum_tree_block(eb, result);
438
439 if (btrfs_header_level(eb))
440 ret = btrfs_check_node(eb);
441 else
442 ret = btrfs_check_leaf_full(eb);
443
444 if (ret < 0) {
445 btrfs_print_tree(eb, 0);
446 btrfs_err(fs_info,
447 "block=%llu write time tree block corruption detected",
448 eb->start);
449 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
450 return ret;
451 }
452 write_extent_buffer(eb, result, 0, fs_info->csum_size);
453
454 return 0;
455}
456
457/* Checksum all dirty extent buffers in one bio_vec */
458static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
459 struct bio_vec *bvec)
460{
461 struct page *page = bvec->bv_page;
462 u64 bvec_start = page_offset(page) + bvec->bv_offset;
463 u64 cur;
464 int ret = 0;
465
466 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
467 cur += fs_info->nodesize) {
468 struct extent_buffer *eb;
469 bool uptodate;
470
471 eb = find_extent_buffer(fs_info, cur);
472 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
473 fs_info->nodesize);
474
475 /* A dirty eb shouldn't disappear from buffer_radix */
476 if (WARN_ON(!eb))
477 return -EUCLEAN;
478
479 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
480 free_extent_buffer(eb);
481 return -EUCLEAN;
482 }
483 if (WARN_ON(!uptodate)) {
484 free_extent_buffer(eb);
485 return -EUCLEAN;
486 }
487
488 ret = csum_one_extent_buffer(eb);
489 free_extent_buffer(eb);
490 if (ret < 0)
491 return ret;
492 }
493 return ret;
494}
495
496/*
497 * Checksum a dirty tree block before IO. This has extra checks to make sure
498 * we only fill in the checksum field in the first page of a multi-page block.
499 * For subpage extent buffers we need bvec to also read the offset in the page.
500 */
501static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
502{
503 struct page *page = bvec->bv_page;
504 u64 start = page_offset(page);
505 u64 found_start;
506 struct extent_buffer *eb;
507
508 if (fs_info->sectorsize < PAGE_SIZE)
509 return csum_dirty_subpage_buffers(fs_info, bvec);
510
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
513 return 0;
514
515 found_start = btrfs_header_bytenr(eb);
516
517 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
518 WARN_ON(found_start != 0);
519 return 0;
520 }
521
522 /*
523 * Please do not consolidate these warnings into a single if.
524 * It is useful to know what went wrong.
525 */
526 if (WARN_ON(found_start != start))
527 return -EUCLEAN;
528 if (WARN_ON(!PageUptodate(page)))
529 return -EUCLEAN;
530
531 return csum_one_extent_buffer(eb);
532}
533
534static int check_tree_block_fsid(struct extent_buffer *eb)
535{
536 struct btrfs_fs_info *fs_info = eb->fs_info;
537 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
538 u8 fsid[BTRFS_FSID_SIZE];
539 u8 *metadata_uuid;
540
541 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
542 BTRFS_FSID_SIZE);
543 /*
544 * Checking the incompat flag is only valid for the current fs. For
545 * seed devices it's forbidden to have their uuid changed so reading
546 * ->fsid in this case is fine
547 */
548 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
549 metadata_uuid = fs_devices->metadata_uuid;
550 else
551 metadata_uuid = fs_devices->fsid;
552
553 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
554 return 0;
555
556 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
557 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
558 return 0;
559
560 return 1;
561}
562
563/* Do basic extent buffer checks at read time */
564static int validate_extent_buffer(struct extent_buffer *eb)
565{
566 struct btrfs_fs_info *fs_info = eb->fs_info;
567 u64 found_start;
568 const u32 csum_size = fs_info->csum_size;
569 u8 found_level;
570 u8 result[BTRFS_CSUM_SIZE];
571 const u8 *header_csum;
572 int ret = 0;
573
574 found_start = btrfs_header_bytenr(eb);
575 if (found_start != eb->start) {
576 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
577 eb->start, found_start);
578 ret = -EIO;
579 goto out;
580 }
581 if (check_tree_block_fsid(eb)) {
582 btrfs_err_rl(fs_info, "bad fsid on block %llu",
583 eb->start);
584 ret = -EIO;
585 goto out;
586 }
587 found_level = btrfs_header_level(eb);
588 if (found_level >= BTRFS_MAX_LEVEL) {
589 btrfs_err(fs_info, "bad tree block level %d on %llu",
590 (int)btrfs_header_level(eb), eb->start);
591 ret = -EIO;
592 goto out;
593 }
594
595 csum_tree_block(eb, result);
596 header_csum = page_address(eb->pages[0]) +
597 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
598
599 if (memcmp(result, header_csum, csum_size) != 0) {
600 btrfs_warn_rl(fs_info,
601 "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
602 eb->start,
603 CSUM_FMT_VALUE(csum_size, header_csum),
604 CSUM_FMT_VALUE(csum_size, result),
605 btrfs_header_level(eb));
606 ret = -EUCLEAN;
607 goto out;
608 }
609
610 /*
611 * If this is a leaf block and it is corrupt, set the corrupt bit so
612 * that we don't try and read the other copies of this block, just
613 * return -EIO.
614 */
615 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
616 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
617 ret = -EIO;
618 }
619
620 if (found_level > 0 && btrfs_check_node(eb))
621 ret = -EIO;
622
623 if (!ret)
624 set_extent_buffer_uptodate(eb);
625 else
626 btrfs_err(fs_info,
627 "block=%llu read time tree block corruption detected",
628 eb->start);
629out:
630 return ret;
631}
632
633static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
634 int mirror)
635{
636 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
637 struct extent_buffer *eb;
638 bool reads_done;
639 int ret = 0;
640
641 /*
642 * We don't allow bio merge for subpage metadata read, so we should
643 * only get one eb for each endio hook.
644 */
645 ASSERT(end == start + fs_info->nodesize - 1);
646 ASSERT(PagePrivate(page));
647
648 eb = find_extent_buffer(fs_info, start);
649 /*
650 * When we are reading one tree block, eb must have been inserted into
651 * the radix tree. If not, something is wrong.
652 */
653 ASSERT(eb);
654
655 reads_done = atomic_dec_and_test(&eb->io_pages);
656 /* Subpage read must finish in page read */
657 ASSERT(reads_done);
658
659 eb->read_mirror = mirror;
660 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
661 ret = -EIO;
662 goto err;
663 }
664 ret = validate_extent_buffer(eb);
665 if (ret < 0)
666 goto err;
667
668 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
669 btree_readahead_hook(eb, ret);
670
671 set_extent_buffer_uptodate(eb);
672
673 free_extent_buffer(eb);
674 return ret;
675err:
676 /*
677 * end_bio_extent_readpage decrements io_pages in case of error,
678 * make sure it has something to decrement.
679 */
680 atomic_inc(&eb->io_pages);
681 clear_extent_buffer_uptodate(eb);
682 free_extent_buffer(eb);
683 return ret;
684}
685
686int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
687 struct page *page, u64 start, u64 end,
688 int mirror)
689{
690 struct extent_buffer *eb;
691 int ret = 0;
692 int reads_done;
693
694 ASSERT(page->private);
695
696 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
697 return validate_subpage_buffer(page, start, end, mirror);
698
699 eb = (struct extent_buffer *)page->private;
700
701 /*
702 * The pending IO might have been the only thing that kept this buffer
703 * in memory. Make sure we have a ref for all this other checks
704 */
705 atomic_inc(&eb->refs);
706
707 reads_done = atomic_dec_and_test(&eb->io_pages);
708 if (!reads_done)
709 goto err;
710
711 eb->read_mirror = mirror;
712 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
713 ret = -EIO;
714 goto err;
715 }
716 ret = validate_extent_buffer(eb);
717err:
718 if (reads_done &&
719 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
720 btree_readahead_hook(eb, ret);
721
722 if (ret) {
723 /*
724 * our io error hook is going to dec the io pages
725 * again, we have to make sure it has something
726 * to decrement
727 */
728 atomic_inc(&eb->io_pages);
729 clear_extent_buffer_uptodate(eb);
730 }
731 free_extent_buffer(eb);
732
733 return ret;
734}
735
736static void end_workqueue_bio(struct bio *bio)
737{
738 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
739 struct btrfs_fs_info *fs_info;
740 struct btrfs_workqueue *wq;
741
742 fs_info = end_io_wq->info;
743 end_io_wq->status = bio->bi_status;
744
745 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
746 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
747 wq = fs_info->endio_meta_write_workers;
748 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
749 wq = fs_info->endio_freespace_worker;
750 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
751 wq = fs_info->endio_raid56_workers;
752 else
753 wq = fs_info->endio_write_workers;
754 } else {
755 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
756 wq = fs_info->endio_raid56_workers;
757 else if (end_io_wq->metadata)
758 wq = fs_info->endio_meta_workers;
759 else
760 wq = fs_info->endio_workers;
761 }
762
763 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
764 btrfs_queue_work(wq, &end_io_wq->work);
765}
766
767blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
768 enum btrfs_wq_endio_type metadata)
769{
770 struct btrfs_end_io_wq *end_io_wq;
771
772 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
773 if (!end_io_wq)
774 return BLK_STS_RESOURCE;
775
776 end_io_wq->private = bio->bi_private;
777 end_io_wq->end_io = bio->bi_end_io;
778 end_io_wq->info = info;
779 end_io_wq->status = 0;
780 end_io_wq->bio = bio;
781 end_io_wq->metadata = metadata;
782
783 bio->bi_private = end_io_wq;
784 bio->bi_end_io = end_workqueue_bio;
785 return 0;
786}
787
788static void run_one_async_start(struct btrfs_work *work)
789{
790 struct async_submit_bio *async;
791 blk_status_t ret;
792
793 async = container_of(work, struct async_submit_bio, work);
794 ret = async->submit_bio_start(async->inode, async->bio,
795 async->dio_file_offset);
796 if (ret)
797 async->status = ret;
798}
799
800/*
801 * In order to insert checksums into the metadata in large chunks, we wait
802 * until bio submission time. All the pages in the bio are checksummed and
803 * sums are attached onto the ordered extent record.
804 *
805 * At IO completion time the csums attached on the ordered extent record are
806 * inserted into the tree.
807 */
808static void run_one_async_done(struct btrfs_work *work)
809{
810 struct async_submit_bio *async;
811 struct inode *inode;
812 blk_status_t ret;
813
814 async = container_of(work, struct async_submit_bio, work);
815 inode = async->inode;
816
817 /* If an error occurred we just want to clean up the bio and move on */
818 if (async->status) {
819 async->bio->bi_status = async->status;
820 bio_endio(async->bio);
821 return;
822 }
823
824 /*
825 * All of the bios that pass through here are from async helpers.
826 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
827 * This changes nothing when cgroups aren't in use.
828 */
829 async->bio->bi_opf |= REQ_CGROUP_PUNT;
830 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
831 if (ret) {
832 async->bio->bi_status = ret;
833 bio_endio(async->bio);
834 }
835}
836
837static void run_one_async_free(struct btrfs_work *work)
838{
839 struct async_submit_bio *async;
840
841 async = container_of(work, struct async_submit_bio, work);
842 kfree(async);
843}
844
845blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
846 int mirror_num, unsigned long bio_flags,
847 u64 dio_file_offset,
848 extent_submit_bio_start_t *submit_bio_start)
849{
850 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
851 struct async_submit_bio *async;
852
853 async = kmalloc(sizeof(*async), GFP_NOFS);
854 if (!async)
855 return BLK_STS_RESOURCE;
856
857 async->inode = inode;
858 async->bio = bio;
859 async->mirror_num = mirror_num;
860 async->submit_bio_start = submit_bio_start;
861
862 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
863 run_one_async_free);
864
865 async->dio_file_offset = dio_file_offset;
866
867 async->status = 0;
868
869 if (op_is_sync(bio->bi_opf))
870 btrfs_set_work_high_priority(&async->work);
871
872 btrfs_queue_work(fs_info->workers, &async->work);
873 return 0;
874}
875
876static blk_status_t btree_csum_one_bio(struct bio *bio)
877{
878 struct bio_vec *bvec;
879 struct btrfs_root *root;
880 int ret = 0;
881 struct bvec_iter_all iter_all;
882
883 ASSERT(!bio_flagged(bio, BIO_CLONED));
884 bio_for_each_segment_all(bvec, bio, iter_all) {
885 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
886 ret = csum_dirty_buffer(root->fs_info, bvec);
887 if (ret)
888 break;
889 }
890
891 return errno_to_blk_status(ret);
892}
893
894static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
895 u64 dio_file_offset)
896{
897 /*
898 * when we're called for a write, we're already in the async
899 * submission context. Just jump into btrfs_map_bio
900 */
901 return btree_csum_one_bio(bio);
902}
903
904static bool should_async_write(struct btrfs_fs_info *fs_info,
905 struct btrfs_inode *bi)
906{
907 if (btrfs_is_zoned(fs_info))
908 return false;
909 if (atomic_read(&bi->sync_writers))
910 return false;
911 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
912 return false;
913 return true;
914}
915
916blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
917 int mirror_num, unsigned long bio_flags)
918{
919 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
920 blk_status_t ret;
921
922 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
923 /*
924 * called for a read, do the setup so that checksum validation
925 * can happen in the async kernel threads
926 */
927 ret = btrfs_bio_wq_end_io(fs_info, bio,
928 BTRFS_WQ_ENDIO_METADATA);
929 if (ret)
930 goto out_w_error;
931 ret = btrfs_map_bio(fs_info, bio, mirror_num);
932 } else if (!should_async_write(fs_info, BTRFS_I(inode))) {
933 ret = btree_csum_one_bio(bio);
934 if (ret)
935 goto out_w_error;
936 ret = btrfs_map_bio(fs_info, bio, mirror_num);
937 } else {
938 /*
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
941 */
942 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
943 0, btree_submit_bio_start);
944 }
945
946 if (ret)
947 goto out_w_error;
948 return 0;
949
950out_w_error:
951 bio->bi_status = ret;
952 bio_endio(bio);
953 return ret;
954}
955
956#ifdef CONFIG_MIGRATION
957static int btree_migratepage(struct address_space *mapping,
958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
960{
961 /*
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
964 */
965 if (PageDirty(page))
966 return -EAGAIN;
967 /*
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
970 */
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
973 return -EAGAIN;
974 return migrate_page(mapping, newpage, page, mode);
975}
976#endif
977
978
979static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
981{
982 struct btrfs_fs_info *fs_info;
983 int ret;
984
985 if (wbc->sync_mode == WB_SYNC_NONE) {
986
987 if (wbc->for_kupdate)
988 return 0;
989
990 fs_info = BTRFS_I(mapping->host)->root->fs_info;
991 /* this is a bit racy, but that's ok */
992 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
993 BTRFS_DIRTY_METADATA_THRESH,
994 fs_info->dirty_metadata_batch);
995 if (ret < 0)
996 return 0;
997 }
998 return btree_write_cache_pages(mapping, wbc);
999}
1000
1001static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1002{
1003 if (PageWriteback(page) || PageDirty(page))
1004 return 0;
1005
1006 return try_release_extent_buffer(page);
1007}
1008
1009static void btree_invalidatepage(struct page *page, unsigned int offset,
1010 unsigned int length)
1011{
1012 struct extent_io_tree *tree;
1013 tree = &BTRFS_I(page->mapping->host)->io_tree;
1014 extent_invalidatepage(tree, page, offset);
1015 btree_releasepage(page, GFP_NOFS);
1016 if (PagePrivate(page)) {
1017 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1018 "page private not zero on page %llu",
1019 (unsigned long long)page_offset(page));
1020 detach_page_private(page);
1021 }
1022}
1023
1024static int btree_set_page_dirty(struct page *page)
1025{
1026#ifdef DEBUG
1027 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1028 struct btrfs_subpage *subpage;
1029 struct extent_buffer *eb;
1030 int cur_bit = 0;
1031 u64 page_start = page_offset(page);
1032
1033 if (fs_info->sectorsize == PAGE_SIZE) {
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1040 return __set_page_dirty_nobuffers(page);
1041 }
1042 ASSERT(PagePrivate(page) && page->private);
1043 subpage = (struct btrfs_subpage *)page->private;
1044
1045 ASSERT(subpage->dirty_bitmap);
1046 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1047 unsigned long flags;
1048 u64 cur;
1049 u16 tmp = (1 << cur_bit);
1050
1051 spin_lock_irqsave(&subpage->lock, flags);
1052 if (!(tmp & subpage->dirty_bitmap)) {
1053 spin_unlock_irqrestore(&subpage->lock, flags);
1054 cur_bit++;
1055 continue;
1056 }
1057 spin_unlock_irqrestore(&subpage->lock, flags);
1058 cur = page_start + cur_bit * fs_info->sectorsize;
1059
1060 eb = find_extent_buffer(fs_info, cur);
1061 ASSERT(eb);
1062 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063 ASSERT(atomic_read(&eb->refs));
1064 btrfs_assert_tree_locked(eb);
1065 free_extent_buffer(eb);
1066
1067 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1068 }
1069#endif
1070 return __set_page_dirty_nobuffers(page);
1071}
1072
1073static const struct address_space_operations btree_aops = {
1074 .writepages = btree_writepages,
1075 .releasepage = btree_releasepage,
1076 .invalidatepage = btree_invalidatepage,
1077#ifdef CONFIG_MIGRATION
1078 .migratepage = btree_migratepage,
1079#endif
1080 .set_page_dirty = btree_set_page_dirty,
1081};
1082
1083struct extent_buffer *btrfs_find_create_tree_block(
1084 struct btrfs_fs_info *fs_info,
1085 u64 bytenr, u64 owner_root,
1086 int level)
1087{
1088 if (btrfs_is_testing(fs_info))
1089 return alloc_test_extent_buffer(fs_info, bytenr);
1090 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1091}
1092
1093/*
1094 * Read tree block at logical address @bytenr and do variant basic but critical
1095 * verification.
1096 *
1097 * @owner_root: the objectid of the root owner for this block.
1098 * @parent_transid: expected transid of this tree block, skip check if 0
1099 * @level: expected level, mandatory check
1100 * @first_key: expected key in slot 0, skip check if NULL
1101 */
1102struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1103 u64 owner_root, u64 parent_transid,
1104 int level, struct btrfs_key *first_key)
1105{
1106 struct extent_buffer *buf = NULL;
1107 int ret;
1108
1109 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1110 if (IS_ERR(buf))
1111 return buf;
1112
1113 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1114 level, first_key);
1115 if (ret) {
1116 free_extent_buffer_stale(buf);
1117 return ERR_PTR(ret);
1118 }
1119 return buf;
1120
1121}
1122
1123void btrfs_clean_tree_block(struct extent_buffer *buf)
1124{
1125 struct btrfs_fs_info *fs_info = buf->fs_info;
1126 if (btrfs_header_generation(buf) ==
1127 fs_info->running_transaction->transid) {
1128 btrfs_assert_tree_locked(buf);
1129
1130 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1131 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1132 -buf->len,
1133 fs_info->dirty_metadata_batch);
1134 clear_extent_buffer_dirty(buf);
1135 }
1136 }
1137}
1138
1139static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1140 u64 objectid)
1141{
1142 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1143 root->fs_info = fs_info;
1144 root->node = NULL;
1145 root->commit_root = NULL;
1146 root->state = 0;
1147 root->orphan_cleanup_state = 0;
1148
1149 root->last_trans = 0;
1150 root->free_objectid = 0;
1151 root->nr_delalloc_inodes = 0;
1152 root->nr_ordered_extents = 0;
1153 root->inode_tree = RB_ROOT;
1154 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1155 root->block_rsv = NULL;
1156
1157 INIT_LIST_HEAD(&root->dirty_list);
1158 INIT_LIST_HEAD(&root->root_list);
1159 INIT_LIST_HEAD(&root->delalloc_inodes);
1160 INIT_LIST_HEAD(&root->delalloc_root);
1161 INIT_LIST_HEAD(&root->ordered_extents);
1162 INIT_LIST_HEAD(&root->ordered_root);
1163 INIT_LIST_HEAD(&root->reloc_dirty_list);
1164 INIT_LIST_HEAD(&root->logged_list[0]);
1165 INIT_LIST_HEAD(&root->logged_list[1]);
1166 spin_lock_init(&root->inode_lock);
1167 spin_lock_init(&root->delalloc_lock);
1168 spin_lock_init(&root->ordered_extent_lock);
1169 spin_lock_init(&root->accounting_lock);
1170 spin_lock_init(&root->log_extents_lock[0]);
1171 spin_lock_init(&root->log_extents_lock[1]);
1172 spin_lock_init(&root->qgroup_meta_rsv_lock);
1173 mutex_init(&root->objectid_mutex);
1174 mutex_init(&root->log_mutex);
1175 mutex_init(&root->ordered_extent_mutex);
1176 mutex_init(&root->delalloc_mutex);
1177 init_waitqueue_head(&root->qgroup_flush_wait);
1178 init_waitqueue_head(&root->log_writer_wait);
1179 init_waitqueue_head(&root->log_commit_wait[0]);
1180 init_waitqueue_head(&root->log_commit_wait[1]);
1181 INIT_LIST_HEAD(&root->log_ctxs[0]);
1182 INIT_LIST_HEAD(&root->log_ctxs[1]);
1183 atomic_set(&root->log_commit[0], 0);
1184 atomic_set(&root->log_commit[1], 0);
1185 atomic_set(&root->log_writers, 0);
1186 atomic_set(&root->log_batch, 0);
1187 refcount_set(&root->refs, 1);
1188 atomic_set(&root->snapshot_force_cow, 0);
1189 atomic_set(&root->nr_swapfiles, 0);
1190 root->log_transid = 0;
1191 root->log_transid_committed = -1;
1192 root->last_log_commit = 0;
1193 if (!dummy) {
1194 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1195 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1196 extent_io_tree_init(fs_info, &root->log_csum_range,
1197 IO_TREE_LOG_CSUM_RANGE, NULL);
1198 }
1199
1200 memset(&root->root_key, 0, sizeof(root->root_key));
1201 memset(&root->root_item, 0, sizeof(root->root_item));
1202 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1203 root->root_key.objectid = objectid;
1204 root->anon_dev = 0;
1205
1206 spin_lock_init(&root->root_item_lock);
1207 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1208#ifdef CONFIG_BTRFS_DEBUG
1209 INIT_LIST_HEAD(&root->leak_list);
1210 spin_lock(&fs_info->fs_roots_radix_lock);
1211 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1212 spin_unlock(&fs_info->fs_roots_radix_lock);
1213#endif
1214}
1215
1216static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1217 u64 objectid, gfp_t flags)
1218{
1219 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1220 if (root)
1221 __setup_root(root, fs_info, objectid);
1222 return root;
1223}
1224
1225#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1226/* Should only be used by the testing infrastructure */
1227struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1228{
1229 struct btrfs_root *root;
1230
1231 if (!fs_info)
1232 return ERR_PTR(-EINVAL);
1233
1234 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1235 if (!root)
1236 return ERR_PTR(-ENOMEM);
1237
1238 /* We don't use the stripesize in selftest, set it as sectorsize */
1239 root->alloc_bytenr = 0;
1240
1241 return root;
1242}
1243#endif
1244
1245struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1246 u64 objectid)
1247{
1248 struct btrfs_fs_info *fs_info = trans->fs_info;
1249 struct extent_buffer *leaf;
1250 struct btrfs_root *tree_root = fs_info->tree_root;
1251 struct btrfs_root *root;
1252 struct btrfs_key key;
1253 unsigned int nofs_flag;
1254 int ret = 0;
1255
1256 /*
1257 * We're holding a transaction handle, so use a NOFS memory allocation
1258 * context to avoid deadlock if reclaim happens.
1259 */
1260 nofs_flag = memalloc_nofs_save();
1261 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1262 memalloc_nofs_restore(nofs_flag);
1263 if (!root)
1264 return ERR_PTR(-ENOMEM);
1265
1266 root->root_key.objectid = objectid;
1267 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1268 root->root_key.offset = 0;
1269
1270 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1271 BTRFS_NESTING_NORMAL);
1272 if (IS_ERR(leaf)) {
1273 ret = PTR_ERR(leaf);
1274 leaf = NULL;
1275 goto fail_unlock;
1276 }
1277
1278 root->node = leaf;
1279 btrfs_mark_buffer_dirty(leaf);
1280
1281 root->commit_root = btrfs_root_node(root);
1282 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1283
1284 btrfs_set_root_flags(&root->root_item, 0);
1285 btrfs_set_root_limit(&root->root_item, 0);
1286 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1287 btrfs_set_root_generation(&root->root_item, trans->transid);
1288 btrfs_set_root_level(&root->root_item, 0);
1289 btrfs_set_root_refs(&root->root_item, 1);
1290 btrfs_set_root_used(&root->root_item, leaf->len);
1291 btrfs_set_root_last_snapshot(&root->root_item, 0);
1292 btrfs_set_root_dirid(&root->root_item, 0);
1293 if (is_fstree(objectid))
1294 generate_random_guid(root->root_item.uuid);
1295 else
1296 export_guid(root->root_item.uuid, &guid_null);
1297 btrfs_set_root_drop_level(&root->root_item, 0);
1298
1299 btrfs_tree_unlock(leaf);
1300
1301 key.objectid = objectid;
1302 key.type = BTRFS_ROOT_ITEM_KEY;
1303 key.offset = 0;
1304 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1305 if (ret)
1306 goto fail;
1307
1308 return root;
1309
1310fail_unlock:
1311 if (leaf)
1312 btrfs_tree_unlock(leaf);
1313fail:
1314 btrfs_put_root(root);
1315
1316 return ERR_PTR(ret);
1317}
1318
1319static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1320 struct btrfs_fs_info *fs_info)
1321{
1322 struct btrfs_root *root;
1323
1324 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1325 if (!root)
1326 return ERR_PTR(-ENOMEM);
1327
1328 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1329 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1330 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1331
1332 return root;
1333}
1334
1335int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1336 struct btrfs_root *root)
1337{
1338 struct extent_buffer *leaf;
1339
1340 /*
1341 * DON'T set SHAREABLE bit for log trees.
1342 *
1343 * Log trees are not exposed to user space thus can't be snapshotted,
1344 * and they go away before a real commit is actually done.
1345 *
1346 * They do store pointers to file data extents, and those reference
1347 * counts still get updated (along with back refs to the log tree).
1348 */
1349
1350 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1351 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1352 if (IS_ERR(leaf))
1353 return PTR_ERR(leaf);
1354
1355 root->node = leaf;
1356
1357 btrfs_mark_buffer_dirty(root->node);
1358 btrfs_tree_unlock(root->node);
1359
1360 return 0;
1361}
1362
1363int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1364 struct btrfs_fs_info *fs_info)
1365{
1366 struct btrfs_root *log_root;
1367
1368 log_root = alloc_log_tree(trans, fs_info);
1369 if (IS_ERR(log_root))
1370 return PTR_ERR(log_root);
1371
1372 if (!btrfs_is_zoned(fs_info)) {
1373 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1374
1375 if (ret) {
1376 btrfs_put_root(log_root);
1377 return ret;
1378 }
1379 }
1380
1381 WARN_ON(fs_info->log_root_tree);
1382 fs_info->log_root_tree = log_root;
1383 return 0;
1384}
1385
1386int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root)
1388{
1389 struct btrfs_fs_info *fs_info = root->fs_info;
1390 struct btrfs_root *log_root;
1391 struct btrfs_inode_item *inode_item;
1392 int ret;
1393
1394 log_root = alloc_log_tree(trans, fs_info);
1395 if (IS_ERR(log_root))
1396 return PTR_ERR(log_root);
1397
1398 ret = btrfs_alloc_log_tree_node(trans, log_root);
1399 if (ret) {
1400 btrfs_put_root(log_root);
1401 return ret;
1402 }
1403
1404 log_root->last_trans = trans->transid;
1405 log_root->root_key.offset = root->root_key.objectid;
1406
1407 inode_item = &log_root->root_item.inode;
1408 btrfs_set_stack_inode_generation(inode_item, 1);
1409 btrfs_set_stack_inode_size(inode_item, 3);
1410 btrfs_set_stack_inode_nlink(inode_item, 1);
1411 btrfs_set_stack_inode_nbytes(inode_item,
1412 fs_info->nodesize);
1413 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1414
1415 btrfs_set_root_node(&log_root->root_item, log_root->node);
1416
1417 WARN_ON(root->log_root);
1418 root->log_root = log_root;
1419 root->log_transid = 0;
1420 root->log_transid_committed = -1;
1421 root->last_log_commit = 0;
1422 return 0;
1423}
1424
1425static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1426 struct btrfs_path *path,
1427 struct btrfs_key *key)
1428{
1429 struct btrfs_root *root;
1430 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1431 u64 generation;
1432 int ret;
1433 int level;
1434
1435 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1436 if (!root)
1437 return ERR_PTR(-ENOMEM);
1438
1439 ret = btrfs_find_root(tree_root, key, path,
1440 &root->root_item, &root->root_key);
1441 if (ret) {
1442 if (ret > 0)
1443 ret = -ENOENT;
1444 goto fail;
1445 }
1446
1447 generation = btrfs_root_generation(&root->root_item);
1448 level = btrfs_root_level(&root->root_item);
1449 root->node = read_tree_block(fs_info,
1450 btrfs_root_bytenr(&root->root_item),
1451 key->objectid, generation, level, NULL);
1452 if (IS_ERR(root->node)) {
1453 ret = PTR_ERR(root->node);
1454 root->node = NULL;
1455 goto fail;
1456 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1457 ret = -EIO;
1458 goto fail;
1459 }
1460 root->commit_root = btrfs_root_node(root);
1461 return root;
1462fail:
1463 btrfs_put_root(root);
1464 return ERR_PTR(ret);
1465}
1466
1467struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1468 struct btrfs_key *key)
1469{
1470 struct btrfs_root *root;
1471 struct btrfs_path *path;
1472
1473 path = btrfs_alloc_path();
1474 if (!path)
1475 return ERR_PTR(-ENOMEM);
1476 root = read_tree_root_path(tree_root, path, key);
1477 btrfs_free_path(path);
1478
1479 return root;
1480}
1481
1482/*
1483 * Initialize subvolume root in-memory structure
1484 *
1485 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1486 */
1487static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1488{
1489 int ret;
1490 unsigned int nofs_flag;
1491
1492 /*
1493 * We might be called under a transaction (e.g. indirect backref
1494 * resolution) which could deadlock if it triggers memory reclaim
1495 */
1496 nofs_flag = memalloc_nofs_save();
1497 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1498 memalloc_nofs_restore(nofs_flag);
1499 if (ret)
1500 goto fail;
1501
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1503 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1504 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1505 btrfs_check_and_init_root_item(&root->root_item);
1506 }
1507
1508 /*
1509 * Don't assign anonymous block device to roots that are not exposed to
1510 * userspace, the id pool is limited to 1M
1511 */
1512 if (is_fstree(root->root_key.objectid) &&
1513 btrfs_root_refs(&root->root_item) > 0) {
1514 if (!anon_dev) {
1515 ret = get_anon_bdev(&root->anon_dev);
1516 if (ret)
1517 goto fail;
1518 } else {
1519 root->anon_dev = anon_dev;
1520 }
1521 }
1522
1523 mutex_lock(&root->objectid_mutex);
1524 ret = btrfs_init_root_free_objectid(root);
1525 if (ret) {
1526 mutex_unlock(&root->objectid_mutex);
1527 goto fail;
1528 }
1529
1530 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1531
1532 mutex_unlock(&root->objectid_mutex);
1533
1534 return 0;
1535fail:
1536 /* The caller is responsible to call btrfs_free_fs_root */
1537 return ret;
1538}
1539
1540static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1541 u64 root_id)
1542{
1543 struct btrfs_root *root;
1544
1545 spin_lock(&fs_info->fs_roots_radix_lock);
1546 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1547 (unsigned long)root_id);
1548 if (root)
1549 root = btrfs_grab_root(root);
1550 spin_unlock(&fs_info->fs_roots_radix_lock);
1551 return root;
1552}
1553
1554static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1555 u64 objectid)
1556{
1557 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1558 return btrfs_grab_root(fs_info->tree_root);
1559 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1560 return btrfs_grab_root(fs_info->extent_root);
1561 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1562 return btrfs_grab_root(fs_info->chunk_root);
1563 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1564 return btrfs_grab_root(fs_info->dev_root);
1565 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1566 return btrfs_grab_root(fs_info->csum_root);
1567 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1568 return btrfs_grab_root(fs_info->quota_root) ?
1569 fs_info->quota_root : ERR_PTR(-ENOENT);
1570 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1571 return btrfs_grab_root(fs_info->uuid_root) ?
1572 fs_info->uuid_root : ERR_PTR(-ENOENT);
1573 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1574 return btrfs_grab_root(fs_info->free_space_root) ?
1575 fs_info->free_space_root : ERR_PTR(-ENOENT);
1576 return NULL;
1577}
1578
1579int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1580 struct btrfs_root *root)
1581{
1582 int ret;
1583
1584 ret = radix_tree_preload(GFP_NOFS);
1585 if (ret)
1586 return ret;
1587
1588 spin_lock(&fs_info->fs_roots_radix_lock);
1589 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1590 (unsigned long)root->root_key.objectid,
1591 root);
1592 if (ret == 0) {
1593 btrfs_grab_root(root);
1594 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1595 }
1596 spin_unlock(&fs_info->fs_roots_radix_lock);
1597 radix_tree_preload_end();
1598
1599 return ret;
1600}
1601
1602void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1603{
1604#ifdef CONFIG_BTRFS_DEBUG
1605 struct btrfs_root *root;
1606
1607 while (!list_empty(&fs_info->allocated_roots)) {
1608 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1609
1610 root = list_first_entry(&fs_info->allocated_roots,
1611 struct btrfs_root, leak_list);
1612 btrfs_err(fs_info, "leaked root %s refcount %d",
1613 btrfs_root_name(&root->root_key, buf),
1614 refcount_read(&root->refs));
1615 while (refcount_read(&root->refs) > 1)
1616 btrfs_put_root(root);
1617 btrfs_put_root(root);
1618 }
1619#endif
1620}
1621
1622void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1623{
1624 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1625 percpu_counter_destroy(&fs_info->delalloc_bytes);
1626 percpu_counter_destroy(&fs_info->ordered_bytes);
1627 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1628 btrfs_free_csum_hash(fs_info);
1629 btrfs_free_stripe_hash_table(fs_info);
1630 btrfs_free_ref_cache(fs_info);
1631 kfree(fs_info->balance_ctl);
1632 kfree(fs_info->delayed_root);
1633 btrfs_put_root(fs_info->extent_root);
1634 btrfs_put_root(fs_info->tree_root);
1635 btrfs_put_root(fs_info->chunk_root);
1636 btrfs_put_root(fs_info->dev_root);
1637 btrfs_put_root(fs_info->csum_root);
1638 btrfs_put_root(fs_info->quota_root);
1639 btrfs_put_root(fs_info->uuid_root);
1640 btrfs_put_root(fs_info->free_space_root);
1641 btrfs_put_root(fs_info->fs_root);
1642 btrfs_put_root(fs_info->data_reloc_root);
1643 btrfs_check_leaked_roots(fs_info);
1644 btrfs_extent_buffer_leak_debug_check(fs_info);
1645 kfree(fs_info->super_copy);
1646 kfree(fs_info->super_for_commit);
1647 kvfree(fs_info);
1648}
1649
1650
1651/*
1652 * Get an in-memory reference of a root structure.
1653 *
1654 * For essential trees like root/extent tree, we grab it from fs_info directly.
1655 * For subvolume trees, we check the cached filesystem roots first. If not
1656 * found, then read it from disk and add it to cached fs roots.
1657 *
1658 * Caller should release the root by calling btrfs_put_root() after the usage.
1659 *
1660 * NOTE: Reloc and log trees can't be read by this function as they share the
1661 * same root objectid.
1662 *
1663 * @objectid: root id
1664 * @anon_dev: preallocated anonymous block device number for new roots,
1665 * pass 0 for new allocation.
1666 * @check_ref: whether to check root item references, If true, return -ENOENT
1667 * for orphan roots
1668 */
1669static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1670 u64 objectid, dev_t anon_dev,
1671 bool check_ref)
1672{
1673 struct btrfs_root *root;
1674 struct btrfs_path *path;
1675 struct btrfs_key key;
1676 int ret;
1677
1678 root = btrfs_get_global_root(fs_info, objectid);
1679 if (root)
1680 return root;
1681again:
1682 root = btrfs_lookup_fs_root(fs_info, objectid);
1683 if (root) {
1684 /* Shouldn't get preallocated anon_dev for cached roots */
1685 ASSERT(!anon_dev);
1686 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1687 btrfs_put_root(root);
1688 return ERR_PTR(-ENOENT);
1689 }
1690 return root;
1691 }
1692
1693 key.objectid = objectid;
1694 key.type = BTRFS_ROOT_ITEM_KEY;
1695 key.offset = (u64)-1;
1696 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1697 if (IS_ERR(root))
1698 return root;
1699
1700 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1701 ret = -ENOENT;
1702 goto fail;
1703 }
1704
1705 ret = btrfs_init_fs_root(root, anon_dev);
1706 if (ret)
1707 goto fail;
1708
1709 path = btrfs_alloc_path();
1710 if (!path) {
1711 ret = -ENOMEM;
1712 goto fail;
1713 }
1714 key.objectid = BTRFS_ORPHAN_OBJECTID;
1715 key.type = BTRFS_ORPHAN_ITEM_KEY;
1716 key.offset = objectid;
1717
1718 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1719 btrfs_free_path(path);
1720 if (ret < 0)
1721 goto fail;
1722 if (ret == 0)
1723 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1724
1725 ret = btrfs_insert_fs_root(fs_info, root);
1726 if (ret) {
1727 btrfs_put_root(root);
1728 if (ret == -EEXIST)
1729 goto again;
1730 goto fail;
1731 }
1732 return root;
1733fail:
1734 btrfs_put_root(root);
1735 return ERR_PTR(ret);
1736}
1737
1738/*
1739 * Get in-memory reference of a root structure
1740 *
1741 * @objectid: tree objectid
1742 * @check_ref: if set, verify that the tree exists and the item has at least
1743 * one reference
1744 */
1745struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1746 u64 objectid, bool check_ref)
1747{
1748 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1749}
1750
1751/*
1752 * Get in-memory reference of a root structure, created as new, optionally pass
1753 * the anonymous block device id
1754 *
1755 * @objectid: tree objectid
1756 * @anon_dev: if zero, allocate a new anonymous block device or use the
1757 * parameter value
1758 */
1759struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1760 u64 objectid, dev_t anon_dev)
1761{
1762 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1763}
1764
1765/*
1766 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1767 * @fs_info: the fs_info
1768 * @objectid: the objectid we need to lookup
1769 *
1770 * This is exclusively used for backref walking, and exists specifically because
1771 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1772 * creation time, which means we may have to read the tree_root in order to look
1773 * up a fs root that is not in memory. If the root is not in memory we will
1774 * read the tree root commit root and look up the fs root from there. This is a
1775 * temporary root, it will not be inserted into the radix tree as it doesn't
1776 * have the most uptodate information, it'll simply be discarded once the
1777 * backref code is finished using the root.
1778 */
1779struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1780 struct btrfs_path *path,
1781 u64 objectid)
1782{
1783 struct btrfs_root *root;
1784 struct btrfs_key key;
1785
1786 ASSERT(path->search_commit_root && path->skip_locking);
1787
1788 /*
1789 * This can return -ENOENT if we ask for a root that doesn't exist, but
1790 * since this is called via the backref walking code we won't be looking
1791 * up a root that doesn't exist, unless there's corruption. So if root
1792 * != NULL just return it.
1793 */
1794 root = btrfs_get_global_root(fs_info, objectid);
1795 if (root)
1796 return root;
1797
1798 root = btrfs_lookup_fs_root(fs_info, objectid);
1799 if (root)
1800 return root;
1801
1802 key.objectid = objectid;
1803 key.type = BTRFS_ROOT_ITEM_KEY;
1804 key.offset = (u64)-1;
1805 root = read_tree_root_path(fs_info->tree_root, path, &key);
1806 btrfs_release_path(path);
1807
1808 return root;
1809}
1810
1811/*
1812 * called by the kthread helper functions to finally call the bio end_io
1813 * functions. This is where read checksum verification actually happens
1814 */
1815static void end_workqueue_fn(struct btrfs_work *work)
1816{
1817 struct bio *bio;
1818 struct btrfs_end_io_wq *end_io_wq;
1819
1820 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1821 bio = end_io_wq->bio;
1822
1823 bio->bi_status = end_io_wq->status;
1824 bio->bi_private = end_io_wq->private;
1825 bio->bi_end_io = end_io_wq->end_io;
1826 bio_endio(bio);
1827 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1828}
1829
1830static int cleaner_kthread(void *arg)
1831{
1832 struct btrfs_root *root = arg;
1833 struct btrfs_fs_info *fs_info = root->fs_info;
1834 int again;
1835
1836 while (1) {
1837 again = 0;
1838
1839 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1840
1841 /* Make the cleaner go to sleep early. */
1842 if (btrfs_need_cleaner_sleep(fs_info))
1843 goto sleep;
1844
1845 /*
1846 * Do not do anything if we might cause open_ctree() to block
1847 * before we have finished mounting the filesystem.
1848 */
1849 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1850 goto sleep;
1851
1852 if (!mutex_trylock(&fs_info->cleaner_mutex))
1853 goto sleep;
1854
1855 /*
1856 * Avoid the problem that we change the status of the fs
1857 * during the above check and trylock.
1858 */
1859 if (btrfs_need_cleaner_sleep(fs_info)) {
1860 mutex_unlock(&fs_info->cleaner_mutex);
1861 goto sleep;
1862 }
1863
1864 btrfs_run_delayed_iputs(fs_info);
1865
1866 again = btrfs_clean_one_deleted_snapshot(root);
1867 mutex_unlock(&fs_info->cleaner_mutex);
1868
1869 /*
1870 * The defragger has dealt with the R/O remount and umount,
1871 * needn't do anything special here.
1872 */
1873 btrfs_run_defrag_inodes(fs_info);
1874
1875 /*
1876 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1877 * with relocation (btrfs_relocate_chunk) and relocation
1878 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1879 * after acquiring fs_info->reclaim_bgs_lock. So we
1880 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1881 * unused block groups.
1882 */
1883 btrfs_delete_unused_bgs(fs_info);
1884
1885 /*
1886 * Reclaim block groups in the reclaim_bgs list after we deleted
1887 * all unused block_groups. This possibly gives us some more free
1888 * space.
1889 */
1890 btrfs_reclaim_bgs(fs_info);
1891sleep:
1892 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1893 if (kthread_should_park())
1894 kthread_parkme();
1895 if (kthread_should_stop())
1896 return 0;
1897 if (!again) {
1898 set_current_state(TASK_INTERRUPTIBLE);
1899 schedule();
1900 __set_current_state(TASK_RUNNING);
1901 }
1902 }
1903}
1904
1905static int transaction_kthread(void *arg)
1906{
1907 struct btrfs_root *root = arg;
1908 struct btrfs_fs_info *fs_info = root->fs_info;
1909 struct btrfs_trans_handle *trans;
1910 struct btrfs_transaction *cur;
1911 u64 transid;
1912 time64_t delta;
1913 unsigned long delay;
1914 bool cannot_commit;
1915
1916 do {
1917 cannot_commit = false;
1918 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1919 mutex_lock(&fs_info->transaction_kthread_mutex);
1920
1921 spin_lock(&fs_info->trans_lock);
1922 cur = fs_info->running_transaction;
1923 if (!cur) {
1924 spin_unlock(&fs_info->trans_lock);
1925 goto sleep;
1926 }
1927
1928 delta = ktime_get_seconds() - cur->start_time;
1929 if (cur->state < TRANS_STATE_COMMIT_START &&
1930 delta < fs_info->commit_interval) {
1931 spin_unlock(&fs_info->trans_lock);
1932 delay -= msecs_to_jiffies((delta - 1) * 1000);
1933 delay = min(delay,
1934 msecs_to_jiffies(fs_info->commit_interval * 1000));
1935 goto sleep;
1936 }
1937 transid = cur->transid;
1938 spin_unlock(&fs_info->trans_lock);
1939
1940 /* If the file system is aborted, this will always fail. */
1941 trans = btrfs_attach_transaction(root);
1942 if (IS_ERR(trans)) {
1943 if (PTR_ERR(trans) != -ENOENT)
1944 cannot_commit = true;
1945 goto sleep;
1946 }
1947 if (transid == trans->transid) {
1948 btrfs_commit_transaction(trans);
1949 } else {
1950 btrfs_end_transaction(trans);
1951 }
1952sleep:
1953 wake_up_process(fs_info->cleaner_kthread);
1954 mutex_unlock(&fs_info->transaction_kthread_mutex);
1955
1956 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1957 &fs_info->fs_state)))
1958 btrfs_cleanup_transaction(fs_info);
1959 if (!kthread_should_stop() &&
1960 (!btrfs_transaction_blocked(fs_info) ||
1961 cannot_commit))
1962 schedule_timeout_interruptible(delay);
1963 } while (!kthread_should_stop());
1964 return 0;
1965}
1966
1967/*
1968 * This will find the highest generation in the array of root backups. The
1969 * index of the highest array is returned, or -EINVAL if we can't find
1970 * anything.
1971 *
1972 * We check to make sure the array is valid by comparing the
1973 * generation of the latest root in the array with the generation
1974 * in the super block. If they don't match we pitch it.
1975 */
1976static int find_newest_super_backup(struct btrfs_fs_info *info)
1977{
1978 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1979 u64 cur;
1980 struct btrfs_root_backup *root_backup;
1981 int i;
1982
1983 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1984 root_backup = info->super_copy->super_roots + i;
1985 cur = btrfs_backup_tree_root_gen(root_backup);
1986 if (cur == newest_gen)
1987 return i;
1988 }
1989
1990 return -EINVAL;
1991}
1992
1993/*
1994 * copy all the root pointers into the super backup array.
1995 * this will bump the backup pointer by one when it is
1996 * done
1997 */
1998static void backup_super_roots(struct btrfs_fs_info *info)
1999{
2000 const int next_backup = info->backup_root_index;
2001 struct btrfs_root_backup *root_backup;
2002
2003 root_backup = info->super_for_commit->super_roots + next_backup;
2004
2005 /*
2006 * make sure all of our padding and empty slots get zero filled
2007 * regardless of which ones we use today
2008 */
2009 memset(root_backup, 0, sizeof(*root_backup));
2010
2011 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2012
2013 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2014 btrfs_set_backup_tree_root_gen(root_backup,
2015 btrfs_header_generation(info->tree_root->node));
2016
2017 btrfs_set_backup_tree_root_level(root_backup,
2018 btrfs_header_level(info->tree_root->node));
2019
2020 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2021 btrfs_set_backup_chunk_root_gen(root_backup,
2022 btrfs_header_generation(info->chunk_root->node));
2023 btrfs_set_backup_chunk_root_level(root_backup,
2024 btrfs_header_level(info->chunk_root->node));
2025
2026 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2027 btrfs_set_backup_extent_root_gen(root_backup,
2028 btrfs_header_generation(info->extent_root->node));
2029 btrfs_set_backup_extent_root_level(root_backup,
2030 btrfs_header_level(info->extent_root->node));
2031
2032 /*
2033 * we might commit during log recovery, which happens before we set
2034 * the fs_root. Make sure it is valid before we fill it in.
2035 */
2036 if (info->fs_root && info->fs_root->node) {
2037 btrfs_set_backup_fs_root(root_backup,
2038 info->fs_root->node->start);
2039 btrfs_set_backup_fs_root_gen(root_backup,
2040 btrfs_header_generation(info->fs_root->node));
2041 btrfs_set_backup_fs_root_level(root_backup,
2042 btrfs_header_level(info->fs_root->node));
2043 }
2044
2045 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2046 btrfs_set_backup_dev_root_gen(root_backup,
2047 btrfs_header_generation(info->dev_root->node));
2048 btrfs_set_backup_dev_root_level(root_backup,
2049 btrfs_header_level(info->dev_root->node));
2050
2051 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2052 btrfs_set_backup_csum_root_gen(root_backup,
2053 btrfs_header_generation(info->csum_root->node));
2054 btrfs_set_backup_csum_root_level(root_backup,
2055 btrfs_header_level(info->csum_root->node));
2056
2057 btrfs_set_backup_total_bytes(root_backup,
2058 btrfs_super_total_bytes(info->super_copy));
2059 btrfs_set_backup_bytes_used(root_backup,
2060 btrfs_super_bytes_used(info->super_copy));
2061 btrfs_set_backup_num_devices(root_backup,
2062 btrfs_super_num_devices(info->super_copy));
2063
2064 /*
2065 * if we don't copy this out to the super_copy, it won't get remembered
2066 * for the next commit
2067 */
2068 memcpy(&info->super_copy->super_roots,
2069 &info->super_for_commit->super_roots,
2070 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2071}
2072
2073/*
2074 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2075 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2076 *
2077 * fs_info - filesystem whose backup roots need to be read
2078 * priority - priority of backup root required
2079 *
2080 * Returns backup root index on success and -EINVAL otherwise.
2081 */
2082static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2083{
2084 int backup_index = find_newest_super_backup(fs_info);
2085 struct btrfs_super_block *super = fs_info->super_copy;
2086 struct btrfs_root_backup *root_backup;
2087
2088 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2089 if (priority == 0)
2090 return backup_index;
2091
2092 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2093 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2094 } else {
2095 return -EINVAL;
2096 }
2097
2098 root_backup = super->super_roots + backup_index;
2099
2100 btrfs_set_super_generation(super,
2101 btrfs_backup_tree_root_gen(root_backup));
2102 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2103 btrfs_set_super_root_level(super,
2104 btrfs_backup_tree_root_level(root_backup));
2105 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2106
2107 /*
2108 * Fixme: the total bytes and num_devices need to match or we should
2109 * need a fsck
2110 */
2111 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2112 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2113
2114 return backup_index;
2115}
2116
2117/* helper to cleanup workers */
2118static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2119{
2120 btrfs_destroy_workqueue(fs_info->fixup_workers);
2121 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2122 btrfs_destroy_workqueue(fs_info->workers);
2123 btrfs_destroy_workqueue(fs_info->endio_workers);
2124 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2125 btrfs_destroy_workqueue(fs_info->rmw_workers);
2126 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2127 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2128 btrfs_destroy_workqueue(fs_info->delayed_workers);
2129 btrfs_destroy_workqueue(fs_info->caching_workers);
2130 btrfs_destroy_workqueue(fs_info->readahead_workers);
2131 btrfs_destroy_workqueue(fs_info->flush_workers);
2132 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2133 if (fs_info->discard_ctl.discard_workers)
2134 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2135 /*
2136 * Now that all other work queues are destroyed, we can safely destroy
2137 * the queues used for metadata I/O, since tasks from those other work
2138 * queues can do metadata I/O operations.
2139 */
2140 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2141 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2142}
2143
2144static void free_root_extent_buffers(struct btrfs_root *root)
2145{
2146 if (root) {
2147 free_extent_buffer(root->node);
2148 free_extent_buffer(root->commit_root);
2149 root->node = NULL;
2150 root->commit_root = NULL;
2151 }
2152}
2153
2154/* helper to cleanup tree roots */
2155static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2156{
2157 free_root_extent_buffers(info->tree_root);
2158
2159 free_root_extent_buffers(info->dev_root);
2160 free_root_extent_buffers(info->extent_root);
2161 free_root_extent_buffers(info->csum_root);
2162 free_root_extent_buffers(info->quota_root);
2163 free_root_extent_buffers(info->uuid_root);
2164 free_root_extent_buffers(info->fs_root);
2165 free_root_extent_buffers(info->data_reloc_root);
2166 if (free_chunk_root)
2167 free_root_extent_buffers(info->chunk_root);
2168 free_root_extent_buffers(info->free_space_root);
2169}
2170
2171void btrfs_put_root(struct btrfs_root *root)
2172{
2173 if (!root)
2174 return;
2175
2176 if (refcount_dec_and_test(&root->refs)) {
2177 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2178 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2179 if (root->anon_dev)
2180 free_anon_bdev(root->anon_dev);
2181 btrfs_drew_lock_destroy(&root->snapshot_lock);
2182 free_root_extent_buffers(root);
2183#ifdef CONFIG_BTRFS_DEBUG
2184 spin_lock(&root->fs_info->fs_roots_radix_lock);
2185 list_del_init(&root->leak_list);
2186 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2187#endif
2188 kfree(root);
2189 }
2190}
2191
2192void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2193{
2194 int ret;
2195 struct btrfs_root *gang[8];
2196 int i;
2197
2198 while (!list_empty(&fs_info->dead_roots)) {
2199 gang[0] = list_entry(fs_info->dead_roots.next,
2200 struct btrfs_root, root_list);
2201 list_del(&gang[0]->root_list);
2202
2203 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2204 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2205 btrfs_put_root(gang[0]);
2206 }
2207
2208 while (1) {
2209 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2210 (void **)gang, 0,
2211 ARRAY_SIZE(gang));
2212 if (!ret)
2213 break;
2214 for (i = 0; i < ret; i++)
2215 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2216 }
2217}
2218
2219static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2220{
2221 mutex_init(&fs_info->scrub_lock);
2222 atomic_set(&fs_info->scrubs_running, 0);
2223 atomic_set(&fs_info->scrub_pause_req, 0);
2224 atomic_set(&fs_info->scrubs_paused, 0);
2225 atomic_set(&fs_info->scrub_cancel_req, 0);
2226 init_waitqueue_head(&fs_info->scrub_pause_wait);
2227 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2228}
2229
2230static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2231{
2232 spin_lock_init(&fs_info->balance_lock);
2233 mutex_init(&fs_info->balance_mutex);
2234 atomic_set(&fs_info->balance_pause_req, 0);
2235 atomic_set(&fs_info->balance_cancel_req, 0);
2236 fs_info->balance_ctl = NULL;
2237 init_waitqueue_head(&fs_info->balance_wait_q);
2238 atomic_set(&fs_info->reloc_cancel_req, 0);
2239}
2240
2241static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2242{
2243 struct inode *inode = fs_info->btree_inode;
2244
2245 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2246 set_nlink(inode, 1);
2247 /*
2248 * we set the i_size on the btree inode to the max possible int.
2249 * the real end of the address space is determined by all of
2250 * the devices in the system
2251 */
2252 inode->i_size = OFFSET_MAX;
2253 inode->i_mapping->a_ops = &btree_aops;
2254
2255 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2256 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2257 IO_TREE_BTREE_INODE_IO, inode);
2258 BTRFS_I(inode)->io_tree.track_uptodate = false;
2259 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2260
2261 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2262 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2263 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2264 btrfs_insert_inode_hash(inode);
2265}
2266
2267static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2268{
2269 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2270 init_rwsem(&fs_info->dev_replace.rwsem);
2271 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2272}
2273
2274static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2275{
2276 spin_lock_init(&fs_info->qgroup_lock);
2277 mutex_init(&fs_info->qgroup_ioctl_lock);
2278 fs_info->qgroup_tree = RB_ROOT;
2279 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2280 fs_info->qgroup_seq = 1;
2281 fs_info->qgroup_ulist = NULL;
2282 fs_info->qgroup_rescan_running = false;
2283 mutex_init(&fs_info->qgroup_rescan_lock);
2284}
2285
2286static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2287 struct btrfs_fs_devices *fs_devices)
2288{
2289 u32 max_active = fs_info->thread_pool_size;
2290 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2291
2292 fs_info->workers =
2293 btrfs_alloc_workqueue(fs_info, "worker",
2294 flags | WQ_HIGHPRI, max_active, 16);
2295
2296 fs_info->delalloc_workers =
2297 btrfs_alloc_workqueue(fs_info, "delalloc",
2298 flags, max_active, 2);
2299
2300 fs_info->flush_workers =
2301 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2302 flags, max_active, 0);
2303
2304 fs_info->caching_workers =
2305 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2306
2307 fs_info->fixup_workers =
2308 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2309
2310 /*
2311 * endios are largely parallel and should have a very
2312 * low idle thresh
2313 */
2314 fs_info->endio_workers =
2315 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2316 fs_info->endio_meta_workers =
2317 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2318 max_active, 4);
2319 fs_info->endio_meta_write_workers =
2320 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2321 max_active, 2);
2322 fs_info->endio_raid56_workers =
2323 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2324 max_active, 4);
2325 fs_info->rmw_workers =
2326 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2327 fs_info->endio_write_workers =
2328 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2329 max_active, 2);
2330 fs_info->endio_freespace_worker =
2331 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2332 max_active, 0);
2333 fs_info->delayed_workers =
2334 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2335 max_active, 0);
2336 fs_info->readahead_workers =
2337 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2338 max_active, 2);
2339 fs_info->qgroup_rescan_workers =
2340 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2341 fs_info->discard_ctl.discard_workers =
2342 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2343
2344 if (!(fs_info->workers && fs_info->delalloc_workers &&
2345 fs_info->flush_workers &&
2346 fs_info->endio_workers && fs_info->endio_meta_workers &&
2347 fs_info->endio_meta_write_workers &&
2348 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2349 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2350 fs_info->caching_workers && fs_info->readahead_workers &&
2351 fs_info->fixup_workers && fs_info->delayed_workers &&
2352 fs_info->qgroup_rescan_workers &&
2353 fs_info->discard_ctl.discard_workers)) {
2354 return -ENOMEM;
2355 }
2356
2357 return 0;
2358}
2359
2360static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2361{
2362 struct crypto_shash *csum_shash;
2363 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2364
2365 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2366
2367 if (IS_ERR(csum_shash)) {
2368 btrfs_err(fs_info, "error allocating %s hash for checksum",
2369 csum_driver);
2370 return PTR_ERR(csum_shash);
2371 }
2372
2373 fs_info->csum_shash = csum_shash;
2374
2375 return 0;
2376}
2377
2378static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2379 struct btrfs_fs_devices *fs_devices)
2380{
2381 int ret;
2382 struct btrfs_root *log_tree_root;
2383 struct btrfs_super_block *disk_super = fs_info->super_copy;
2384 u64 bytenr = btrfs_super_log_root(disk_super);
2385 int level = btrfs_super_log_root_level(disk_super);
2386
2387 if (fs_devices->rw_devices == 0) {
2388 btrfs_warn(fs_info, "log replay required on RO media");
2389 return -EIO;
2390 }
2391
2392 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2393 GFP_KERNEL);
2394 if (!log_tree_root)
2395 return -ENOMEM;
2396
2397 log_tree_root->node = read_tree_block(fs_info, bytenr,
2398 BTRFS_TREE_LOG_OBJECTID,
2399 fs_info->generation + 1, level,
2400 NULL);
2401 if (IS_ERR(log_tree_root->node)) {
2402 btrfs_warn(fs_info, "failed to read log tree");
2403 ret = PTR_ERR(log_tree_root->node);
2404 log_tree_root->node = NULL;
2405 btrfs_put_root(log_tree_root);
2406 return ret;
2407 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2408 btrfs_err(fs_info, "failed to read log tree");
2409 btrfs_put_root(log_tree_root);
2410 return -EIO;
2411 }
2412 /* returns with log_tree_root freed on success */
2413 ret = btrfs_recover_log_trees(log_tree_root);
2414 if (ret) {
2415 btrfs_handle_fs_error(fs_info, ret,
2416 "Failed to recover log tree");
2417 btrfs_put_root(log_tree_root);
2418 return ret;
2419 }
2420
2421 if (sb_rdonly(fs_info->sb)) {
2422 ret = btrfs_commit_super(fs_info);
2423 if (ret)
2424 return ret;
2425 }
2426
2427 return 0;
2428}
2429
2430static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2431{
2432 struct btrfs_root *tree_root = fs_info->tree_root;
2433 struct btrfs_root *root;
2434 struct btrfs_key location;
2435 int ret;
2436
2437 BUG_ON(!fs_info->tree_root);
2438
2439 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2440 location.type = BTRFS_ROOT_ITEM_KEY;
2441 location.offset = 0;
2442
2443 root = btrfs_read_tree_root(tree_root, &location);
2444 if (IS_ERR(root)) {
2445 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2446 ret = PTR_ERR(root);
2447 goto out;
2448 }
2449 } else {
2450 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2451 fs_info->extent_root = root;
2452 }
2453
2454 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2455 root = btrfs_read_tree_root(tree_root, &location);
2456 if (IS_ERR(root)) {
2457 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2458 ret = PTR_ERR(root);
2459 goto out;
2460 }
2461 } else {
2462 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2463 fs_info->dev_root = root;
2464 }
2465 /* Initialize fs_info for all devices in any case */
2466 btrfs_init_devices_late(fs_info);
2467
2468 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2469 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2470 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2471 root = btrfs_read_tree_root(tree_root, &location);
2472 if (IS_ERR(root)) {
2473 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2474 ret = PTR_ERR(root);
2475 goto out;
2476 }
2477 } else {
2478 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2479 fs_info->csum_root = root;
2480 }
2481 }
2482
2483 /*
2484 * This tree can share blocks with some other fs tree during relocation
2485 * and we need a proper setup by btrfs_get_fs_root
2486 */
2487 root = btrfs_get_fs_root(tree_root->fs_info,
2488 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2489 if (IS_ERR(root)) {
2490 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2491 ret = PTR_ERR(root);
2492 goto out;
2493 }
2494 } else {
2495 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2496 fs_info->data_reloc_root = root;
2497 }
2498
2499 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2500 root = btrfs_read_tree_root(tree_root, &location);
2501 if (!IS_ERR(root)) {
2502 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2503 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2504 fs_info->quota_root = root;
2505 }
2506
2507 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2508 root = btrfs_read_tree_root(tree_root, &location);
2509 if (IS_ERR(root)) {
2510 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2511 ret = PTR_ERR(root);
2512 if (ret != -ENOENT)
2513 goto out;
2514 }
2515 } else {
2516 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2517 fs_info->uuid_root = root;
2518 }
2519
2520 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2521 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2522 root = btrfs_read_tree_root(tree_root, &location);
2523 if (IS_ERR(root)) {
2524 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2525 ret = PTR_ERR(root);
2526 goto out;
2527 }
2528 } else {
2529 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2530 fs_info->free_space_root = root;
2531 }
2532 }
2533
2534 return 0;
2535out:
2536 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2537 location.objectid, ret);
2538 return ret;
2539}
2540
2541/*
2542 * Real super block validation
2543 * NOTE: super csum type and incompat features will not be checked here.
2544 *
2545 * @sb: super block to check
2546 * @mirror_num: the super block number to check its bytenr:
2547 * 0 the primary (1st) sb
2548 * 1, 2 2nd and 3rd backup copy
2549 * -1 skip bytenr check
2550 */
2551static int validate_super(struct btrfs_fs_info *fs_info,
2552 struct btrfs_super_block *sb, int mirror_num)
2553{
2554 u64 nodesize = btrfs_super_nodesize(sb);
2555 u64 sectorsize = btrfs_super_sectorsize(sb);
2556 int ret = 0;
2557
2558 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2559 btrfs_err(fs_info, "no valid FS found");
2560 ret = -EINVAL;
2561 }
2562 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2563 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2564 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2565 ret = -EINVAL;
2566 }
2567 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2568 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2569 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2570 ret = -EINVAL;
2571 }
2572 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2573 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2574 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2575 ret = -EINVAL;
2576 }
2577 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2578 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2579 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2580 ret = -EINVAL;
2581 }
2582
2583 /*
2584 * Check sectorsize and nodesize first, other check will need it.
2585 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2586 */
2587 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2588 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2589 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2590 ret = -EINVAL;
2591 }
2592
2593 /*
2594 * For 4K page size, we only support 4K sector size.
2595 * For 64K page size, we support read-write for 64K sector size, and
2596 * read-only for 4K sector size.
2597 */
2598 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2599 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2600 sectorsize != SZ_64K))) {
2601 btrfs_err(fs_info,
2602 "sectorsize %llu not yet supported for page size %lu",
2603 sectorsize, PAGE_SIZE);
2604 ret = -EINVAL;
2605 }
2606
2607 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2608 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2609 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2610 ret = -EINVAL;
2611 }
2612 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2613 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2614 le32_to_cpu(sb->__unused_leafsize), nodesize);
2615 ret = -EINVAL;
2616 }
2617
2618 /* Root alignment check */
2619 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2620 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2621 btrfs_super_root(sb));
2622 ret = -EINVAL;
2623 }
2624 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2625 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2626 btrfs_super_chunk_root(sb));
2627 ret = -EINVAL;
2628 }
2629 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2630 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2631 btrfs_super_log_root(sb));
2632 ret = -EINVAL;
2633 }
2634
2635 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2636 BTRFS_FSID_SIZE)) {
2637 btrfs_err(fs_info,
2638 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2639 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2640 ret = -EINVAL;
2641 }
2642
2643 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2644 memcmp(fs_info->fs_devices->metadata_uuid,
2645 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2646 btrfs_err(fs_info,
2647"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2648 fs_info->super_copy->metadata_uuid,
2649 fs_info->fs_devices->metadata_uuid);
2650 ret = -EINVAL;
2651 }
2652
2653 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2654 BTRFS_FSID_SIZE) != 0) {
2655 btrfs_err(fs_info,
2656 "dev_item UUID does not match metadata fsid: %pU != %pU",
2657 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2658 ret = -EINVAL;
2659 }
2660
2661 /*
2662 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2663 * done later
2664 */
2665 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2666 btrfs_err(fs_info, "bytes_used is too small %llu",
2667 btrfs_super_bytes_used(sb));
2668 ret = -EINVAL;
2669 }
2670 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2671 btrfs_err(fs_info, "invalid stripesize %u",
2672 btrfs_super_stripesize(sb));
2673 ret = -EINVAL;
2674 }
2675 if (btrfs_super_num_devices(sb) > (1UL << 31))
2676 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2677 btrfs_super_num_devices(sb));
2678 if (btrfs_super_num_devices(sb) == 0) {
2679 btrfs_err(fs_info, "number of devices is 0");
2680 ret = -EINVAL;
2681 }
2682
2683 if (mirror_num >= 0 &&
2684 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2685 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2686 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2687 ret = -EINVAL;
2688 }
2689
2690 /*
2691 * Obvious sys_chunk_array corruptions, it must hold at least one key
2692 * and one chunk
2693 */
2694 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2695 btrfs_err(fs_info, "system chunk array too big %u > %u",
2696 btrfs_super_sys_array_size(sb),
2697 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2698 ret = -EINVAL;
2699 }
2700 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2701 + sizeof(struct btrfs_chunk)) {
2702 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2703 btrfs_super_sys_array_size(sb),
2704 sizeof(struct btrfs_disk_key)
2705 + sizeof(struct btrfs_chunk));
2706 ret = -EINVAL;
2707 }
2708
2709 /*
2710 * The generation is a global counter, we'll trust it more than the others
2711 * but it's still possible that it's the one that's wrong.
2712 */
2713 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2714 btrfs_warn(fs_info,
2715 "suspicious: generation < chunk_root_generation: %llu < %llu",
2716 btrfs_super_generation(sb),
2717 btrfs_super_chunk_root_generation(sb));
2718 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2719 && btrfs_super_cache_generation(sb) != (u64)-1)
2720 btrfs_warn(fs_info,
2721 "suspicious: generation < cache_generation: %llu < %llu",
2722 btrfs_super_generation(sb),
2723 btrfs_super_cache_generation(sb));
2724
2725 return ret;
2726}
2727
2728/*
2729 * Validation of super block at mount time.
2730 * Some checks already done early at mount time, like csum type and incompat
2731 * flags will be skipped.
2732 */
2733static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2734{
2735 return validate_super(fs_info, fs_info->super_copy, 0);
2736}
2737
2738/*
2739 * Validation of super block at write time.
2740 * Some checks like bytenr check will be skipped as their values will be
2741 * overwritten soon.
2742 * Extra checks like csum type and incompat flags will be done here.
2743 */
2744static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2745 struct btrfs_super_block *sb)
2746{
2747 int ret;
2748
2749 ret = validate_super(fs_info, sb, -1);
2750 if (ret < 0)
2751 goto out;
2752 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2753 ret = -EUCLEAN;
2754 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2755 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2756 goto out;
2757 }
2758 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2759 ret = -EUCLEAN;
2760 btrfs_err(fs_info,
2761 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2762 btrfs_super_incompat_flags(sb),
2763 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2764 goto out;
2765 }
2766out:
2767 if (ret < 0)
2768 btrfs_err(fs_info,
2769 "super block corruption detected before writing it to disk");
2770 return ret;
2771}
2772
2773static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2774{
2775 int backup_index = find_newest_super_backup(fs_info);
2776 struct btrfs_super_block *sb = fs_info->super_copy;
2777 struct btrfs_root *tree_root = fs_info->tree_root;
2778 bool handle_error = false;
2779 int ret = 0;
2780 int i;
2781
2782 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2783 u64 generation;
2784 int level;
2785
2786 if (handle_error) {
2787 if (!IS_ERR(tree_root->node))
2788 free_extent_buffer(tree_root->node);
2789 tree_root->node = NULL;
2790
2791 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2792 break;
2793
2794 free_root_pointers(fs_info, 0);
2795
2796 /*
2797 * Don't use the log in recovery mode, it won't be
2798 * valid
2799 */
2800 btrfs_set_super_log_root(sb, 0);
2801
2802 /* We can't trust the free space cache either */
2803 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2804
2805 ret = read_backup_root(fs_info, i);
2806 backup_index = ret;
2807 if (ret < 0)
2808 return ret;
2809 }
2810 generation = btrfs_super_generation(sb);
2811 level = btrfs_super_root_level(sb);
2812 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2813 BTRFS_ROOT_TREE_OBJECTID,
2814 generation, level, NULL);
2815 if (IS_ERR(tree_root->node)) {
2816 handle_error = true;
2817 ret = PTR_ERR(tree_root->node);
2818 tree_root->node = NULL;
2819 btrfs_warn(fs_info, "couldn't read tree root");
2820 continue;
2821
2822 } else if (!extent_buffer_uptodate(tree_root->node)) {
2823 handle_error = true;
2824 ret = -EIO;
2825 btrfs_warn(fs_info, "error while reading tree root");
2826 continue;
2827 }
2828
2829 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2830 tree_root->commit_root = btrfs_root_node(tree_root);
2831 btrfs_set_root_refs(&tree_root->root_item, 1);
2832
2833 /*
2834 * No need to hold btrfs_root::objectid_mutex since the fs
2835 * hasn't been fully initialised and we are the only user
2836 */
2837 ret = btrfs_init_root_free_objectid(tree_root);
2838 if (ret < 0) {
2839 handle_error = true;
2840 continue;
2841 }
2842
2843 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2844
2845 ret = btrfs_read_roots(fs_info);
2846 if (ret < 0) {
2847 handle_error = true;
2848 continue;
2849 }
2850
2851 /* All successful */
2852 fs_info->generation = generation;
2853 fs_info->last_trans_committed = generation;
2854
2855 /* Always begin writing backup roots after the one being used */
2856 if (backup_index < 0) {
2857 fs_info->backup_root_index = 0;
2858 } else {
2859 fs_info->backup_root_index = backup_index + 1;
2860 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2861 }
2862 break;
2863 }
2864
2865 return ret;
2866}
2867
2868void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2869{
2870 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2871 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2872 INIT_LIST_HEAD(&fs_info->trans_list);
2873 INIT_LIST_HEAD(&fs_info->dead_roots);
2874 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2875 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2876 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2877 spin_lock_init(&fs_info->delalloc_root_lock);
2878 spin_lock_init(&fs_info->trans_lock);
2879 spin_lock_init(&fs_info->fs_roots_radix_lock);
2880 spin_lock_init(&fs_info->delayed_iput_lock);
2881 spin_lock_init(&fs_info->defrag_inodes_lock);
2882 spin_lock_init(&fs_info->super_lock);
2883 spin_lock_init(&fs_info->buffer_lock);
2884 spin_lock_init(&fs_info->unused_bgs_lock);
2885 spin_lock_init(&fs_info->treelog_bg_lock);
2886 rwlock_init(&fs_info->tree_mod_log_lock);
2887 mutex_init(&fs_info->unused_bg_unpin_mutex);
2888 mutex_init(&fs_info->reclaim_bgs_lock);
2889 mutex_init(&fs_info->reloc_mutex);
2890 mutex_init(&fs_info->delalloc_root_mutex);
2891 mutex_init(&fs_info->zoned_meta_io_lock);
2892 seqlock_init(&fs_info->profiles_lock);
2893
2894 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2895 INIT_LIST_HEAD(&fs_info->space_info);
2896 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2897 INIT_LIST_HEAD(&fs_info->unused_bgs);
2898 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2899#ifdef CONFIG_BTRFS_DEBUG
2900 INIT_LIST_HEAD(&fs_info->allocated_roots);
2901 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2902 spin_lock_init(&fs_info->eb_leak_lock);
2903#endif
2904 extent_map_tree_init(&fs_info->mapping_tree);
2905 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2906 BTRFS_BLOCK_RSV_GLOBAL);
2907 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2908 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2909 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2910 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2911 BTRFS_BLOCK_RSV_DELOPS);
2912 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2913 BTRFS_BLOCK_RSV_DELREFS);
2914
2915 atomic_set(&fs_info->async_delalloc_pages, 0);
2916 atomic_set(&fs_info->defrag_running, 0);
2917 atomic_set(&fs_info->reada_works_cnt, 0);
2918 atomic_set(&fs_info->nr_delayed_iputs, 0);
2919 atomic64_set(&fs_info->tree_mod_seq, 0);
2920 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2921 fs_info->metadata_ratio = 0;
2922 fs_info->defrag_inodes = RB_ROOT;
2923 atomic64_set(&fs_info->free_chunk_space, 0);
2924 fs_info->tree_mod_log = RB_ROOT;
2925 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2926 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2927 /* readahead state */
2928 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2929 spin_lock_init(&fs_info->reada_lock);
2930 btrfs_init_ref_verify(fs_info);
2931
2932 fs_info->thread_pool_size = min_t(unsigned long,
2933 num_online_cpus() + 2, 8);
2934
2935 INIT_LIST_HEAD(&fs_info->ordered_roots);
2936 spin_lock_init(&fs_info->ordered_root_lock);
2937
2938 btrfs_init_scrub(fs_info);
2939#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2940 fs_info->check_integrity_print_mask = 0;
2941#endif
2942 btrfs_init_balance(fs_info);
2943 btrfs_init_async_reclaim_work(fs_info);
2944
2945 spin_lock_init(&fs_info->block_group_cache_lock);
2946 fs_info->block_group_cache_tree = RB_ROOT;
2947 fs_info->first_logical_byte = (u64)-1;
2948
2949 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2950 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2951 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2952
2953 mutex_init(&fs_info->ordered_operations_mutex);
2954 mutex_init(&fs_info->tree_log_mutex);
2955 mutex_init(&fs_info->chunk_mutex);
2956 mutex_init(&fs_info->transaction_kthread_mutex);
2957 mutex_init(&fs_info->cleaner_mutex);
2958 mutex_init(&fs_info->ro_block_group_mutex);
2959 init_rwsem(&fs_info->commit_root_sem);
2960 init_rwsem(&fs_info->cleanup_work_sem);
2961 init_rwsem(&fs_info->subvol_sem);
2962 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2963
2964 btrfs_init_dev_replace_locks(fs_info);
2965 btrfs_init_qgroup(fs_info);
2966 btrfs_discard_init(fs_info);
2967
2968 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2969 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2970
2971 init_waitqueue_head(&fs_info->transaction_throttle);
2972 init_waitqueue_head(&fs_info->transaction_wait);
2973 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2974 init_waitqueue_head(&fs_info->async_submit_wait);
2975 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2976
2977 /* Usable values until the real ones are cached from the superblock */
2978 fs_info->nodesize = 4096;
2979 fs_info->sectorsize = 4096;
2980 fs_info->sectorsize_bits = ilog2(4096);
2981 fs_info->stripesize = 4096;
2982
2983 spin_lock_init(&fs_info->swapfile_pins_lock);
2984 fs_info->swapfile_pins = RB_ROOT;
2985
2986 spin_lock_init(&fs_info->send_reloc_lock);
2987 fs_info->send_in_progress = 0;
2988
2989 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2990 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2991}
2992
2993static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2994{
2995 int ret;
2996
2997 fs_info->sb = sb;
2998 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2999 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3000
3001 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3002 if (ret)
3003 return ret;
3004
3005 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3006 if (ret)
3007 return ret;
3008
3009 fs_info->dirty_metadata_batch = PAGE_SIZE *
3010 (1 + ilog2(nr_cpu_ids));
3011
3012 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3013 if (ret)
3014 return ret;
3015
3016 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3017 GFP_KERNEL);
3018 if (ret)
3019 return ret;
3020
3021 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3022 GFP_KERNEL);
3023 if (!fs_info->delayed_root)
3024 return -ENOMEM;
3025 btrfs_init_delayed_root(fs_info->delayed_root);
3026
3027 if (sb_rdonly(sb))
3028 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3029
3030 return btrfs_alloc_stripe_hash_table(fs_info);
3031}
3032
3033static int btrfs_uuid_rescan_kthread(void *data)
3034{
3035 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3036 int ret;
3037
3038 /*
3039 * 1st step is to iterate through the existing UUID tree and
3040 * to delete all entries that contain outdated data.
3041 * 2nd step is to add all missing entries to the UUID tree.
3042 */
3043 ret = btrfs_uuid_tree_iterate(fs_info);
3044 if (ret < 0) {
3045 if (ret != -EINTR)
3046 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3047 ret);
3048 up(&fs_info->uuid_tree_rescan_sem);
3049 return ret;
3050 }
3051 return btrfs_uuid_scan_kthread(data);
3052}
3053
3054static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3055{
3056 struct task_struct *task;
3057
3058 down(&fs_info->uuid_tree_rescan_sem);
3059 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3060 if (IS_ERR(task)) {
3061 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3062 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3063 up(&fs_info->uuid_tree_rescan_sem);
3064 return PTR_ERR(task);
3065 }
3066
3067 return 0;
3068}
3069
3070/*
3071 * Some options only have meaning at mount time and shouldn't persist across
3072 * remounts, or be displayed. Clear these at the end of mount and remount
3073 * code paths.
3074 */
3075void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3076{
3077 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3078 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3079}
3080
3081/*
3082 * Mounting logic specific to read-write file systems. Shared by open_ctree
3083 * and btrfs_remount when remounting from read-only to read-write.
3084 */
3085int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3086{
3087 int ret;
3088 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3089 bool clear_free_space_tree = false;
3090
3091 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3092 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3093 clear_free_space_tree = true;
3094 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3095 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3096 btrfs_warn(fs_info, "free space tree is invalid");
3097 clear_free_space_tree = true;
3098 }
3099
3100 if (clear_free_space_tree) {
3101 btrfs_info(fs_info, "clearing free space tree");
3102 ret = btrfs_clear_free_space_tree(fs_info);
3103 if (ret) {
3104 btrfs_warn(fs_info,
3105 "failed to clear free space tree: %d", ret);
3106 goto out;
3107 }
3108 }
3109
3110 /*
3111 * btrfs_find_orphan_roots() is responsible for finding all the dead
3112 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3113 * them into the fs_info->fs_roots_radix tree. This must be done before
3114 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3115 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3116 * item before the root's tree is deleted - this means that if we unmount
3117 * or crash before the deletion completes, on the next mount we will not
3118 * delete what remains of the tree because the orphan item does not
3119 * exists anymore, which is what tells us we have a pending deletion.
3120 */
3121 ret = btrfs_find_orphan_roots(fs_info);
3122 if (ret)
3123 goto out;
3124
3125 ret = btrfs_cleanup_fs_roots(fs_info);
3126 if (ret)
3127 goto out;
3128
3129 down_read(&fs_info->cleanup_work_sem);
3130 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3131 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3132 up_read(&fs_info->cleanup_work_sem);
3133 goto out;
3134 }
3135 up_read(&fs_info->cleanup_work_sem);
3136
3137 mutex_lock(&fs_info->cleaner_mutex);
3138 ret = btrfs_recover_relocation(fs_info->tree_root);
3139 mutex_unlock(&fs_info->cleaner_mutex);
3140 if (ret < 0) {
3141 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3142 goto out;
3143 }
3144
3145 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3146 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3147 btrfs_info(fs_info, "creating free space tree");
3148 ret = btrfs_create_free_space_tree(fs_info);
3149 if (ret) {
3150 btrfs_warn(fs_info,
3151 "failed to create free space tree: %d", ret);
3152 goto out;
3153 }
3154 }
3155
3156 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3157 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3158 if (ret)
3159 goto out;
3160 }
3161
3162 ret = btrfs_resume_balance_async(fs_info);
3163 if (ret)
3164 goto out;
3165
3166 ret = btrfs_resume_dev_replace_async(fs_info);
3167 if (ret) {
3168 btrfs_warn(fs_info, "failed to resume dev_replace");
3169 goto out;
3170 }
3171
3172 btrfs_qgroup_rescan_resume(fs_info);
3173
3174 if (!fs_info->uuid_root) {
3175 btrfs_info(fs_info, "creating UUID tree");
3176 ret = btrfs_create_uuid_tree(fs_info);
3177 if (ret) {
3178 btrfs_warn(fs_info,
3179 "failed to create the UUID tree %d", ret);
3180 goto out;
3181 }
3182 }
3183
3184out:
3185 return ret;
3186}
3187
3188int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3189 char *options)
3190{
3191 u32 sectorsize;
3192 u32 nodesize;
3193 u32 stripesize;
3194 u64 generation;
3195 u64 features;
3196 u16 csum_type;
3197 struct btrfs_super_block *disk_super;
3198 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3199 struct btrfs_root *tree_root;
3200 struct btrfs_root *chunk_root;
3201 int ret;
3202 int err = -EINVAL;
3203 int level;
3204
3205 ret = init_mount_fs_info(fs_info, sb);
3206 if (ret) {
3207 err = ret;
3208 goto fail;
3209 }
3210
3211 /* These need to be init'ed before we start creating inodes and such. */
3212 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3213 GFP_KERNEL);
3214 fs_info->tree_root = tree_root;
3215 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3216 GFP_KERNEL);
3217 fs_info->chunk_root = chunk_root;
3218 if (!tree_root || !chunk_root) {
3219 err = -ENOMEM;
3220 goto fail;
3221 }
3222
3223 fs_info->btree_inode = new_inode(sb);
3224 if (!fs_info->btree_inode) {
3225 err = -ENOMEM;
3226 goto fail;
3227 }
3228 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3229 btrfs_init_btree_inode(fs_info);
3230
3231 invalidate_bdev(fs_devices->latest_bdev);
3232
3233 /*
3234 * Read super block and check the signature bytes only
3235 */
3236 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
3237 if (IS_ERR(disk_super)) {
3238 err = PTR_ERR(disk_super);
3239 goto fail_alloc;
3240 }
3241
3242 /*
3243 * Verify the type first, if that or the checksum value are
3244 * corrupted, we'll find out
3245 */
3246 csum_type = btrfs_super_csum_type(disk_super);
3247 if (!btrfs_supported_super_csum(csum_type)) {
3248 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3249 csum_type);
3250 err = -EINVAL;
3251 btrfs_release_disk_super(disk_super);
3252 goto fail_alloc;
3253 }
3254
3255 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3256
3257 ret = btrfs_init_csum_hash(fs_info, csum_type);
3258 if (ret) {
3259 err = ret;
3260 btrfs_release_disk_super(disk_super);
3261 goto fail_alloc;
3262 }
3263
3264 /*
3265 * We want to check superblock checksum, the type is stored inside.
3266 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3267 */
3268 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3269 btrfs_err(fs_info, "superblock checksum mismatch");
3270 err = -EINVAL;
3271 btrfs_release_disk_super(disk_super);
3272 goto fail_alloc;
3273 }
3274
3275 /*
3276 * super_copy is zeroed at allocation time and we never touch the
3277 * following bytes up to INFO_SIZE, the checksum is calculated from
3278 * the whole block of INFO_SIZE
3279 */
3280 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3281 btrfs_release_disk_super(disk_super);
3282
3283 disk_super = fs_info->super_copy;
3284
3285
3286 features = btrfs_super_flags(disk_super);
3287 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3288 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3289 btrfs_set_super_flags(disk_super, features);
3290 btrfs_info(fs_info,
3291 "found metadata UUID change in progress flag, clearing");
3292 }
3293
3294 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3295 sizeof(*fs_info->super_for_commit));
3296
3297 ret = btrfs_validate_mount_super(fs_info);
3298 if (ret) {
3299 btrfs_err(fs_info, "superblock contains fatal errors");
3300 err = -EINVAL;
3301 goto fail_alloc;
3302 }
3303
3304 if (!btrfs_super_root(disk_super))
3305 goto fail_alloc;
3306
3307 /* check FS state, whether FS is broken. */
3308 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3309 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3310
3311 /*
3312 * In the long term, we'll store the compression type in the super
3313 * block, and it'll be used for per file compression control.
3314 */
3315 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3316
3317 /*
3318 * Flag our filesystem as having big metadata blocks if they are bigger
3319 * than the page size.
3320 */
3321 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3322 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3323 btrfs_info(fs_info,
3324 "flagging fs with big metadata feature");
3325 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3326 }
3327
3328 /* Set up fs_info before parsing mount options */
3329 nodesize = btrfs_super_nodesize(disk_super);
3330 sectorsize = btrfs_super_sectorsize(disk_super);
3331 stripesize = sectorsize;
3332 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3333 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3334
3335 fs_info->nodesize = nodesize;
3336 fs_info->sectorsize = sectorsize;
3337 fs_info->sectorsize_bits = ilog2(sectorsize);
3338 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3339 fs_info->stripesize = stripesize;
3340
3341 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3342 if (ret) {
3343 err = ret;
3344 goto fail_alloc;
3345 }
3346
3347 features = btrfs_super_incompat_flags(disk_super) &
3348 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3349 if (features) {
3350 btrfs_err(fs_info,
3351 "cannot mount because of unsupported optional features (%llx)",
3352 features);
3353 err = -EINVAL;
3354 goto fail_alloc;
3355 }
3356
3357 features = btrfs_super_incompat_flags(disk_super);
3358 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3359 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3360 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3361 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3362 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3363
3364 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3365 btrfs_info(fs_info, "has skinny extents");
3366
3367 /*
3368 * mixed block groups end up with duplicate but slightly offset
3369 * extent buffers for the same range. It leads to corruptions
3370 */
3371 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3372 (sectorsize != nodesize)) {
3373 btrfs_err(fs_info,
3374"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3375 nodesize, sectorsize);
3376 goto fail_alloc;
3377 }
3378
3379 /*
3380 * Needn't use the lock because there is no other task which will
3381 * update the flag.
3382 */
3383 btrfs_set_super_incompat_flags(disk_super, features);
3384
3385 features = btrfs_super_compat_ro_flags(disk_super) &
3386 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3387 if (!sb_rdonly(sb) && features) {
3388 btrfs_err(fs_info,
3389 "cannot mount read-write because of unsupported optional features (%llx)",
3390 features);
3391 err = -EINVAL;
3392 goto fail_alloc;
3393 }
3394
3395 /* For 4K sector size support, it's only read-only */
3396 if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) {
3397 if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) {
3398 btrfs_err(fs_info,
3399 "subpage sectorsize %u only supported read-only for page size %lu",
3400 sectorsize, PAGE_SIZE);
3401 err = -EINVAL;
3402 goto fail_alloc;
3403 }
3404 }
3405
3406 ret = btrfs_init_workqueues(fs_info, fs_devices);
3407 if (ret) {
3408 err = ret;
3409 goto fail_sb_buffer;
3410 }
3411
3412 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3413 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3414
3415 sb->s_blocksize = sectorsize;
3416 sb->s_blocksize_bits = blksize_bits(sectorsize);
3417 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3418
3419 mutex_lock(&fs_info->chunk_mutex);
3420 ret = btrfs_read_sys_array(fs_info);
3421 mutex_unlock(&fs_info->chunk_mutex);
3422 if (ret) {
3423 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3424 goto fail_sb_buffer;
3425 }
3426
3427 generation = btrfs_super_chunk_root_generation(disk_super);
3428 level = btrfs_super_chunk_root_level(disk_super);
3429
3430 chunk_root->node = read_tree_block(fs_info,
3431 btrfs_super_chunk_root(disk_super),
3432 BTRFS_CHUNK_TREE_OBJECTID,
3433 generation, level, NULL);
3434 if (IS_ERR(chunk_root->node) ||
3435 !extent_buffer_uptodate(chunk_root->node)) {
3436 btrfs_err(fs_info, "failed to read chunk root");
3437 if (!IS_ERR(chunk_root->node))
3438 free_extent_buffer(chunk_root->node);
3439 chunk_root->node = NULL;
3440 goto fail_tree_roots;
3441 }
3442 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3443 chunk_root->commit_root = btrfs_root_node(chunk_root);
3444
3445 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3446 offsetof(struct btrfs_header, chunk_tree_uuid),
3447 BTRFS_UUID_SIZE);
3448
3449 ret = btrfs_read_chunk_tree(fs_info);
3450 if (ret) {
3451 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3452 goto fail_tree_roots;
3453 }
3454
3455 /*
3456 * At this point we know all the devices that make this filesystem,
3457 * including the seed devices but we don't know yet if the replace
3458 * target is required. So free devices that are not part of this
3459 * filesystem but skip the replace target device which is checked
3460 * below in btrfs_init_dev_replace().
3461 */
3462 btrfs_free_extra_devids(fs_devices);
3463 if (!fs_devices->latest_bdev) {
3464 btrfs_err(fs_info, "failed to read devices");
3465 goto fail_tree_roots;
3466 }
3467
3468 ret = init_tree_roots(fs_info);
3469 if (ret)
3470 goto fail_tree_roots;
3471
3472 /*
3473 * Get zone type information of zoned block devices. This will also
3474 * handle emulation of a zoned filesystem if a regular device has the
3475 * zoned incompat feature flag set.
3476 */
3477 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3478 if (ret) {
3479 btrfs_err(fs_info,
3480 "zoned: failed to read device zone info: %d",
3481 ret);
3482 goto fail_block_groups;
3483 }
3484
3485 /*
3486 * If we have a uuid root and we're not being told to rescan we need to
3487 * check the generation here so we can set the
3488 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3489 * transaction during a balance or the log replay without updating the
3490 * uuid generation, and then if we crash we would rescan the uuid tree,
3491 * even though it was perfectly fine.
3492 */
3493 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3494 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3495 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3496
3497 ret = btrfs_verify_dev_extents(fs_info);
3498 if (ret) {
3499 btrfs_err(fs_info,
3500 "failed to verify dev extents against chunks: %d",
3501 ret);
3502 goto fail_block_groups;
3503 }
3504 ret = btrfs_recover_balance(fs_info);
3505 if (ret) {
3506 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3507 goto fail_block_groups;
3508 }
3509
3510 ret = btrfs_init_dev_stats(fs_info);
3511 if (ret) {
3512 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3513 goto fail_block_groups;
3514 }
3515
3516 ret = btrfs_init_dev_replace(fs_info);
3517 if (ret) {
3518 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3519 goto fail_block_groups;
3520 }
3521
3522 ret = btrfs_check_zoned_mode(fs_info);
3523 if (ret) {
3524 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3525 ret);
3526 goto fail_block_groups;
3527 }
3528
3529 ret = btrfs_sysfs_add_fsid(fs_devices);
3530 if (ret) {
3531 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3532 ret);
3533 goto fail_block_groups;
3534 }
3535
3536 ret = btrfs_sysfs_add_mounted(fs_info);
3537 if (ret) {
3538 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3539 goto fail_fsdev_sysfs;
3540 }
3541
3542 ret = btrfs_init_space_info(fs_info);
3543 if (ret) {
3544 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3545 goto fail_sysfs;
3546 }
3547
3548 ret = btrfs_read_block_groups(fs_info);
3549 if (ret) {
3550 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3551 goto fail_sysfs;
3552 }
3553
3554 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3555 btrfs_warn(fs_info,
3556 "writable mount is not allowed due to too many missing devices");
3557 goto fail_sysfs;
3558 }
3559
3560 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3561 "btrfs-cleaner");
3562 if (IS_ERR(fs_info->cleaner_kthread))
3563 goto fail_sysfs;
3564
3565 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3566 tree_root,
3567 "btrfs-transaction");
3568 if (IS_ERR(fs_info->transaction_kthread))
3569 goto fail_cleaner;
3570
3571 if (!btrfs_test_opt(fs_info, NOSSD) &&
3572 !fs_info->fs_devices->rotating) {
3573 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3574 }
3575
3576 /*
3577 * Mount does not set all options immediately, we can do it now and do
3578 * not have to wait for transaction commit
3579 */
3580 btrfs_apply_pending_changes(fs_info);
3581
3582#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3583 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3584 ret = btrfsic_mount(fs_info, fs_devices,
3585 btrfs_test_opt(fs_info,
3586 CHECK_INTEGRITY_DATA) ? 1 : 0,
3587 fs_info->check_integrity_print_mask);
3588 if (ret)
3589 btrfs_warn(fs_info,
3590 "failed to initialize integrity check module: %d",
3591 ret);
3592 }
3593#endif
3594 ret = btrfs_read_qgroup_config(fs_info);
3595 if (ret)
3596 goto fail_trans_kthread;
3597
3598 if (btrfs_build_ref_tree(fs_info))
3599 btrfs_err(fs_info, "couldn't build ref tree");
3600
3601 /* do not make disk changes in broken FS or nologreplay is given */
3602 if (btrfs_super_log_root(disk_super) != 0 &&
3603 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3604 btrfs_info(fs_info, "start tree-log replay");
3605 ret = btrfs_replay_log(fs_info, fs_devices);
3606 if (ret) {
3607 err = ret;
3608 goto fail_qgroup;
3609 }
3610 }
3611
3612 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3613 if (IS_ERR(fs_info->fs_root)) {
3614 err = PTR_ERR(fs_info->fs_root);
3615 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3616 fs_info->fs_root = NULL;
3617 goto fail_qgroup;
3618 }
3619
3620 if (sb_rdonly(sb))
3621 goto clear_oneshot;
3622
3623 ret = btrfs_start_pre_rw_mount(fs_info);
3624 if (ret) {
3625 close_ctree(fs_info);
3626 return ret;
3627 }
3628 btrfs_discard_resume(fs_info);
3629
3630 if (fs_info->uuid_root &&
3631 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3632 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3633 btrfs_info(fs_info, "checking UUID tree");
3634 ret = btrfs_check_uuid_tree(fs_info);
3635 if (ret) {
3636 btrfs_warn(fs_info,
3637 "failed to check the UUID tree: %d", ret);
3638 close_ctree(fs_info);
3639 return ret;
3640 }
3641 }
3642
3643 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3644
3645clear_oneshot:
3646 btrfs_clear_oneshot_options(fs_info);
3647 return 0;
3648
3649fail_qgroup:
3650 btrfs_free_qgroup_config(fs_info);
3651fail_trans_kthread:
3652 kthread_stop(fs_info->transaction_kthread);
3653 btrfs_cleanup_transaction(fs_info);
3654 btrfs_free_fs_roots(fs_info);
3655fail_cleaner:
3656 kthread_stop(fs_info->cleaner_kthread);
3657
3658 /*
3659 * make sure we're done with the btree inode before we stop our
3660 * kthreads
3661 */
3662 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3663
3664fail_sysfs:
3665 btrfs_sysfs_remove_mounted(fs_info);
3666
3667fail_fsdev_sysfs:
3668 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3669
3670fail_block_groups:
3671 btrfs_put_block_group_cache(fs_info);
3672
3673fail_tree_roots:
3674 if (fs_info->data_reloc_root)
3675 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3676 free_root_pointers(fs_info, true);
3677 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3678
3679fail_sb_buffer:
3680 btrfs_stop_all_workers(fs_info);
3681 btrfs_free_block_groups(fs_info);
3682fail_alloc:
3683 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3684
3685 iput(fs_info->btree_inode);
3686fail:
3687 btrfs_close_devices(fs_info->fs_devices);
3688 return err;
3689}
3690ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3691
3692static void btrfs_end_super_write(struct bio *bio)
3693{
3694 struct btrfs_device *device = bio->bi_private;
3695 struct bio_vec *bvec;
3696 struct bvec_iter_all iter_all;
3697 struct page *page;
3698
3699 bio_for_each_segment_all(bvec, bio, iter_all) {
3700 page = bvec->bv_page;
3701
3702 if (bio->bi_status) {
3703 btrfs_warn_rl_in_rcu(device->fs_info,
3704 "lost page write due to IO error on %s (%d)",
3705 rcu_str_deref(device->name),
3706 blk_status_to_errno(bio->bi_status));
3707 ClearPageUptodate(page);
3708 SetPageError(page);
3709 btrfs_dev_stat_inc_and_print(device,
3710 BTRFS_DEV_STAT_WRITE_ERRS);
3711 } else {
3712 SetPageUptodate(page);
3713 }
3714
3715 put_page(page);
3716 unlock_page(page);
3717 }
3718
3719 bio_put(bio);
3720}
3721
3722struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3723 int copy_num)
3724{
3725 struct btrfs_super_block *super;
3726 struct page *page;
3727 u64 bytenr, bytenr_orig;
3728 struct address_space *mapping = bdev->bd_inode->i_mapping;
3729 int ret;
3730
3731 bytenr_orig = btrfs_sb_offset(copy_num);
3732 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3733 if (ret == -ENOENT)
3734 return ERR_PTR(-EINVAL);
3735 else if (ret)
3736 return ERR_PTR(ret);
3737
3738 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3739 return ERR_PTR(-EINVAL);
3740
3741 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3742 if (IS_ERR(page))
3743 return ERR_CAST(page);
3744
3745 super = page_address(page);
3746 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3747 btrfs_release_disk_super(super);
3748 return ERR_PTR(-ENODATA);
3749 }
3750
3751 if (btrfs_super_bytenr(super) != bytenr_orig) {
3752 btrfs_release_disk_super(super);
3753 return ERR_PTR(-EINVAL);
3754 }
3755
3756 return super;
3757}
3758
3759
3760struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3761{
3762 struct btrfs_super_block *super, *latest = NULL;
3763 int i;
3764 u64 transid = 0;
3765
3766 /* we would like to check all the supers, but that would make
3767 * a btrfs mount succeed after a mkfs from a different FS.
3768 * So, we need to add a special mount option to scan for
3769 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3770 */
3771 for (i = 0; i < 1; i++) {
3772 super = btrfs_read_dev_one_super(bdev, i);
3773 if (IS_ERR(super))
3774 continue;
3775
3776 if (!latest || btrfs_super_generation(super) > transid) {
3777 if (latest)
3778 btrfs_release_disk_super(super);
3779
3780 latest = super;
3781 transid = btrfs_super_generation(super);
3782 }
3783 }
3784
3785 return super;
3786}
3787
3788/*
3789 * Write superblock @sb to the @device. Do not wait for completion, all the
3790 * pages we use for writing are locked.
3791 *
3792 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3793 * the expected device size at commit time. Note that max_mirrors must be
3794 * same for write and wait phases.
3795 *
3796 * Return number of errors when page is not found or submission fails.
3797 */
3798static int write_dev_supers(struct btrfs_device *device,
3799 struct btrfs_super_block *sb, int max_mirrors)
3800{
3801 struct btrfs_fs_info *fs_info = device->fs_info;
3802 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3803 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3804 int i;
3805 int errors = 0;
3806 int ret;
3807 u64 bytenr, bytenr_orig;
3808
3809 if (max_mirrors == 0)
3810 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3811
3812 shash->tfm = fs_info->csum_shash;
3813
3814 for (i = 0; i < max_mirrors; i++) {
3815 struct page *page;
3816 struct bio *bio;
3817 struct btrfs_super_block *disk_super;
3818
3819 bytenr_orig = btrfs_sb_offset(i);
3820 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3821 if (ret == -ENOENT) {
3822 continue;
3823 } else if (ret < 0) {
3824 btrfs_err(device->fs_info,
3825 "couldn't get super block location for mirror %d",
3826 i);
3827 errors++;
3828 continue;
3829 }
3830 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3831 device->commit_total_bytes)
3832 break;
3833
3834 btrfs_set_super_bytenr(sb, bytenr_orig);
3835
3836 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3837 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3838 sb->csum);
3839
3840 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3841 GFP_NOFS);
3842 if (!page) {
3843 btrfs_err(device->fs_info,
3844 "couldn't get super block page for bytenr %llu",
3845 bytenr);
3846 errors++;
3847 continue;
3848 }
3849
3850 /* Bump the refcount for wait_dev_supers() */
3851 get_page(page);
3852
3853 disk_super = page_address(page);
3854 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3855
3856 /*
3857 * Directly use bios here instead of relying on the page cache
3858 * to do I/O, so we don't lose the ability to do integrity
3859 * checking.
3860 */
3861 bio = bio_alloc(GFP_NOFS, 1);
3862 bio_set_dev(bio, device->bdev);
3863 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3864 bio->bi_private = device;
3865 bio->bi_end_io = btrfs_end_super_write;
3866 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3867 offset_in_page(bytenr));
3868
3869 /*
3870 * We FUA only the first super block. The others we allow to
3871 * go down lazy and there's a short window where the on-disk
3872 * copies might still contain the older version.
3873 */
3874 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3875 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3876 bio->bi_opf |= REQ_FUA;
3877
3878 btrfsic_submit_bio(bio);
3879 btrfs_advance_sb_log(device, i);
3880 }
3881 return errors < i ? 0 : -1;
3882}
3883
3884/*
3885 * Wait for write completion of superblocks done by write_dev_supers,
3886 * @max_mirrors same for write and wait phases.
3887 *
3888 * Return number of errors when page is not found or not marked up to
3889 * date.
3890 */
3891static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3892{
3893 int i;
3894 int errors = 0;
3895 bool primary_failed = false;
3896 int ret;
3897 u64 bytenr;
3898
3899 if (max_mirrors == 0)
3900 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3901
3902 for (i = 0; i < max_mirrors; i++) {
3903 struct page *page;
3904
3905 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3906 if (ret == -ENOENT) {
3907 break;
3908 } else if (ret < 0) {
3909 errors++;
3910 if (i == 0)
3911 primary_failed = true;
3912 continue;
3913 }
3914 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3915 device->commit_total_bytes)
3916 break;
3917
3918 page = find_get_page(device->bdev->bd_inode->i_mapping,
3919 bytenr >> PAGE_SHIFT);
3920 if (!page) {
3921 errors++;
3922 if (i == 0)
3923 primary_failed = true;
3924 continue;
3925 }
3926 /* Page is submitted locked and unlocked once the IO completes */
3927 wait_on_page_locked(page);
3928 if (PageError(page)) {
3929 errors++;
3930 if (i == 0)
3931 primary_failed = true;
3932 }
3933
3934 /* Drop our reference */
3935 put_page(page);
3936
3937 /* Drop the reference from the writing run */
3938 put_page(page);
3939 }
3940
3941 /* log error, force error return */
3942 if (primary_failed) {
3943 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3944 device->devid);
3945 return -1;
3946 }
3947
3948 return errors < i ? 0 : -1;
3949}
3950
3951/*
3952 * endio for the write_dev_flush, this will wake anyone waiting
3953 * for the barrier when it is done
3954 */
3955static void btrfs_end_empty_barrier(struct bio *bio)
3956{
3957 complete(bio->bi_private);
3958}
3959
3960/*
3961 * Submit a flush request to the device if it supports it. Error handling is
3962 * done in the waiting counterpart.
3963 */
3964static void write_dev_flush(struct btrfs_device *device)
3965{
3966 struct request_queue *q = bdev_get_queue(device->bdev);
3967 struct bio *bio = device->flush_bio;
3968
3969 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3970 return;
3971
3972 bio_reset(bio);
3973 bio->bi_end_io = btrfs_end_empty_barrier;
3974 bio_set_dev(bio, device->bdev);
3975 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3976 init_completion(&device->flush_wait);
3977 bio->bi_private = &device->flush_wait;
3978
3979 btrfsic_submit_bio(bio);
3980 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3981}
3982
3983/*
3984 * If the flush bio has been submitted by write_dev_flush, wait for it.
3985 */
3986static blk_status_t wait_dev_flush(struct btrfs_device *device)
3987{
3988 struct bio *bio = device->flush_bio;
3989
3990 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3991 return BLK_STS_OK;
3992
3993 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3994 wait_for_completion_io(&device->flush_wait);
3995
3996 return bio->bi_status;
3997}
3998
3999static int check_barrier_error(struct btrfs_fs_info *fs_info)
4000{
4001 if (!btrfs_check_rw_degradable(fs_info, NULL))
4002 return -EIO;
4003 return 0;
4004}
4005
4006/*
4007 * send an empty flush down to each device in parallel,
4008 * then wait for them
4009 */
4010static int barrier_all_devices(struct btrfs_fs_info *info)
4011{
4012 struct list_head *head;
4013 struct btrfs_device *dev;
4014 int errors_wait = 0;
4015 blk_status_t ret;
4016
4017 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4018 /* send down all the barriers */
4019 head = &info->fs_devices->devices;
4020 list_for_each_entry(dev, head, dev_list) {
4021 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4022 continue;
4023 if (!dev->bdev)
4024 continue;
4025 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4026 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4027 continue;
4028
4029 write_dev_flush(dev);
4030 dev->last_flush_error = BLK_STS_OK;
4031 }
4032
4033 /* wait for all the barriers */
4034 list_for_each_entry(dev, head, dev_list) {
4035 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4036 continue;
4037 if (!dev->bdev) {
4038 errors_wait++;
4039 continue;
4040 }
4041 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4042 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4043 continue;
4044
4045 ret = wait_dev_flush(dev);
4046 if (ret) {
4047 dev->last_flush_error = ret;
4048 btrfs_dev_stat_inc_and_print(dev,
4049 BTRFS_DEV_STAT_FLUSH_ERRS);
4050 errors_wait++;
4051 }
4052 }
4053
4054 if (errors_wait) {
4055 /*
4056 * At some point we need the status of all disks
4057 * to arrive at the volume status. So error checking
4058 * is being pushed to a separate loop.
4059 */
4060 return check_barrier_error(info);
4061 }
4062 return 0;
4063}
4064
4065int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4066{
4067 int raid_type;
4068 int min_tolerated = INT_MAX;
4069
4070 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4071 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4072 min_tolerated = min_t(int, min_tolerated,
4073 btrfs_raid_array[BTRFS_RAID_SINGLE].
4074 tolerated_failures);
4075
4076 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4077 if (raid_type == BTRFS_RAID_SINGLE)
4078 continue;
4079 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4080 continue;
4081 min_tolerated = min_t(int, min_tolerated,
4082 btrfs_raid_array[raid_type].
4083 tolerated_failures);
4084 }
4085
4086 if (min_tolerated == INT_MAX) {
4087 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4088 min_tolerated = 0;
4089 }
4090
4091 return min_tolerated;
4092}
4093
4094int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4095{
4096 struct list_head *head;
4097 struct btrfs_device *dev;
4098 struct btrfs_super_block *sb;
4099 struct btrfs_dev_item *dev_item;
4100 int ret;
4101 int do_barriers;
4102 int max_errors;
4103 int total_errors = 0;
4104 u64 flags;
4105
4106 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4107
4108 /*
4109 * max_mirrors == 0 indicates we're from commit_transaction,
4110 * not from fsync where the tree roots in fs_info have not
4111 * been consistent on disk.
4112 */
4113 if (max_mirrors == 0)
4114 backup_super_roots(fs_info);
4115
4116 sb = fs_info->super_for_commit;
4117 dev_item = &sb->dev_item;
4118
4119 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4120 head = &fs_info->fs_devices->devices;
4121 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4122
4123 if (do_barriers) {
4124 ret = barrier_all_devices(fs_info);
4125 if (ret) {
4126 mutex_unlock(
4127 &fs_info->fs_devices->device_list_mutex);
4128 btrfs_handle_fs_error(fs_info, ret,
4129 "errors while submitting device barriers.");
4130 return ret;
4131 }
4132 }
4133
4134 list_for_each_entry(dev, head, dev_list) {
4135 if (!dev->bdev) {
4136 total_errors++;
4137 continue;
4138 }
4139 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4140 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4141 continue;
4142
4143 btrfs_set_stack_device_generation(dev_item, 0);
4144 btrfs_set_stack_device_type(dev_item, dev->type);
4145 btrfs_set_stack_device_id(dev_item, dev->devid);
4146 btrfs_set_stack_device_total_bytes(dev_item,
4147 dev->commit_total_bytes);
4148 btrfs_set_stack_device_bytes_used(dev_item,
4149 dev->commit_bytes_used);
4150 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4151 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4152 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4153 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4154 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4155 BTRFS_FSID_SIZE);
4156
4157 flags = btrfs_super_flags(sb);
4158 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4159
4160 ret = btrfs_validate_write_super(fs_info, sb);
4161 if (ret < 0) {
4162 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4163 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4164 "unexpected superblock corruption detected");
4165 return -EUCLEAN;
4166 }
4167
4168 ret = write_dev_supers(dev, sb, max_mirrors);
4169 if (ret)
4170 total_errors++;
4171 }
4172 if (total_errors > max_errors) {
4173 btrfs_err(fs_info, "%d errors while writing supers",
4174 total_errors);
4175 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4176
4177 /* FUA is masked off if unsupported and can't be the reason */
4178 btrfs_handle_fs_error(fs_info, -EIO,
4179 "%d errors while writing supers",
4180 total_errors);
4181 return -EIO;
4182 }
4183
4184 total_errors = 0;
4185 list_for_each_entry(dev, head, dev_list) {
4186 if (!dev->bdev)
4187 continue;
4188 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4189 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4190 continue;
4191
4192 ret = wait_dev_supers(dev, max_mirrors);
4193 if (ret)
4194 total_errors++;
4195 }
4196 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4197 if (total_errors > max_errors) {
4198 btrfs_handle_fs_error(fs_info, -EIO,
4199 "%d errors while writing supers",
4200 total_errors);
4201 return -EIO;
4202 }
4203 return 0;
4204}
4205
4206/* Drop a fs root from the radix tree and free it. */
4207void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4208 struct btrfs_root *root)
4209{
4210 bool drop_ref = false;
4211
4212 spin_lock(&fs_info->fs_roots_radix_lock);
4213 radix_tree_delete(&fs_info->fs_roots_radix,
4214 (unsigned long)root->root_key.objectid);
4215 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4216 drop_ref = true;
4217 spin_unlock(&fs_info->fs_roots_radix_lock);
4218
4219 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4220 ASSERT(root->log_root == NULL);
4221 if (root->reloc_root) {
4222 btrfs_put_root(root->reloc_root);
4223 root->reloc_root = NULL;
4224 }
4225 }
4226
4227 if (drop_ref)
4228 btrfs_put_root(root);
4229}
4230
4231int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4232{
4233 u64 root_objectid = 0;
4234 struct btrfs_root *gang[8];
4235 int i = 0;
4236 int err = 0;
4237 unsigned int ret = 0;
4238
4239 while (1) {
4240 spin_lock(&fs_info->fs_roots_radix_lock);
4241 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4242 (void **)gang, root_objectid,
4243 ARRAY_SIZE(gang));
4244 if (!ret) {
4245 spin_unlock(&fs_info->fs_roots_radix_lock);
4246 break;
4247 }
4248 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4249
4250 for (i = 0; i < ret; i++) {
4251 /* Avoid to grab roots in dead_roots */
4252 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4253 gang[i] = NULL;
4254 continue;
4255 }
4256 /* grab all the search result for later use */
4257 gang[i] = btrfs_grab_root(gang[i]);
4258 }
4259 spin_unlock(&fs_info->fs_roots_radix_lock);
4260
4261 for (i = 0; i < ret; i++) {
4262 if (!gang[i])
4263 continue;
4264 root_objectid = gang[i]->root_key.objectid;
4265 err = btrfs_orphan_cleanup(gang[i]);
4266 if (err)
4267 break;
4268 btrfs_put_root(gang[i]);
4269 }
4270 root_objectid++;
4271 }
4272
4273 /* release the uncleaned roots due to error */
4274 for (; i < ret; i++) {
4275 if (gang[i])
4276 btrfs_put_root(gang[i]);
4277 }
4278 return err;
4279}
4280
4281int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4282{
4283 struct btrfs_root *root = fs_info->tree_root;
4284 struct btrfs_trans_handle *trans;
4285
4286 mutex_lock(&fs_info->cleaner_mutex);
4287 btrfs_run_delayed_iputs(fs_info);
4288 mutex_unlock(&fs_info->cleaner_mutex);
4289 wake_up_process(fs_info->cleaner_kthread);
4290
4291 /* wait until ongoing cleanup work done */
4292 down_write(&fs_info->cleanup_work_sem);
4293 up_write(&fs_info->cleanup_work_sem);
4294
4295 trans = btrfs_join_transaction(root);
4296 if (IS_ERR(trans))
4297 return PTR_ERR(trans);
4298 return btrfs_commit_transaction(trans);
4299}
4300
4301void __cold close_ctree(struct btrfs_fs_info *fs_info)
4302{
4303 int ret;
4304
4305 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4306 /*
4307 * We don't want the cleaner to start new transactions, add more delayed
4308 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4309 * because that frees the task_struct, and the transaction kthread might
4310 * still try to wake up the cleaner.
4311 */
4312 kthread_park(fs_info->cleaner_kthread);
4313
4314 /* wait for the qgroup rescan worker to stop */
4315 btrfs_qgroup_wait_for_completion(fs_info, false);
4316
4317 /* wait for the uuid_scan task to finish */
4318 down(&fs_info->uuid_tree_rescan_sem);
4319 /* avoid complains from lockdep et al., set sem back to initial state */
4320 up(&fs_info->uuid_tree_rescan_sem);
4321
4322 /* pause restriper - we want to resume on mount */
4323 btrfs_pause_balance(fs_info);
4324
4325 btrfs_dev_replace_suspend_for_unmount(fs_info);
4326
4327 btrfs_scrub_cancel(fs_info);
4328
4329 /* wait for any defraggers to finish */
4330 wait_event(fs_info->transaction_wait,
4331 (atomic_read(&fs_info->defrag_running) == 0));
4332
4333 /* clear out the rbtree of defraggable inodes */
4334 btrfs_cleanup_defrag_inodes(fs_info);
4335
4336 cancel_work_sync(&fs_info->async_reclaim_work);
4337 cancel_work_sync(&fs_info->async_data_reclaim_work);
4338 cancel_work_sync(&fs_info->preempt_reclaim_work);
4339
4340 cancel_work_sync(&fs_info->reclaim_bgs_work);
4341
4342 /* Cancel or finish ongoing discard work */
4343 btrfs_discard_cleanup(fs_info);
4344
4345 if (!sb_rdonly(fs_info->sb)) {
4346 /*
4347 * The cleaner kthread is stopped, so do one final pass over
4348 * unused block groups.
4349 */
4350 btrfs_delete_unused_bgs(fs_info);
4351
4352 /*
4353 * There might be existing delayed inode workers still running
4354 * and holding an empty delayed inode item. We must wait for
4355 * them to complete first because they can create a transaction.
4356 * This happens when someone calls btrfs_balance_delayed_items()
4357 * and then a transaction commit runs the same delayed nodes
4358 * before any delayed worker has done something with the nodes.
4359 * We must wait for any worker here and not at transaction
4360 * commit time since that could cause a deadlock.
4361 * This is a very rare case.
4362 */
4363 btrfs_flush_workqueue(fs_info->delayed_workers);
4364
4365 ret = btrfs_commit_super(fs_info);
4366 if (ret)
4367 btrfs_err(fs_info, "commit super ret %d", ret);
4368 }
4369
4370 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4371 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4372 btrfs_error_commit_super(fs_info);
4373
4374 kthread_stop(fs_info->transaction_kthread);
4375 kthread_stop(fs_info->cleaner_kthread);
4376
4377 ASSERT(list_empty(&fs_info->delayed_iputs));
4378 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4379
4380 if (btrfs_check_quota_leak(fs_info)) {
4381 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4382 btrfs_err(fs_info, "qgroup reserved space leaked");
4383 }
4384
4385 btrfs_free_qgroup_config(fs_info);
4386 ASSERT(list_empty(&fs_info->delalloc_roots));
4387
4388 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4389 btrfs_info(fs_info, "at unmount delalloc count %lld",
4390 percpu_counter_sum(&fs_info->delalloc_bytes));
4391 }
4392
4393 if (percpu_counter_sum(&fs_info->ordered_bytes))
4394 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4395 percpu_counter_sum(&fs_info->ordered_bytes));
4396
4397 btrfs_sysfs_remove_mounted(fs_info);
4398 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4399
4400 btrfs_put_block_group_cache(fs_info);
4401
4402 /*
4403 * we must make sure there is not any read request to
4404 * submit after we stopping all workers.
4405 */
4406 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4407 btrfs_stop_all_workers(fs_info);
4408
4409 /* We shouldn't have any transaction open at this point */
4410 ASSERT(list_empty(&fs_info->trans_list));
4411
4412 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4413 free_root_pointers(fs_info, true);
4414 btrfs_free_fs_roots(fs_info);
4415
4416 /*
4417 * We must free the block groups after dropping the fs_roots as we could
4418 * have had an IO error and have left over tree log blocks that aren't
4419 * cleaned up until the fs roots are freed. This makes the block group
4420 * accounting appear to be wrong because there's pending reserved bytes,
4421 * so make sure we do the block group cleanup afterwards.
4422 */
4423 btrfs_free_block_groups(fs_info);
4424
4425 iput(fs_info->btree_inode);
4426
4427#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4428 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4429 btrfsic_unmount(fs_info->fs_devices);
4430#endif
4431
4432 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4433 btrfs_close_devices(fs_info->fs_devices);
4434}
4435
4436int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4437 int atomic)
4438{
4439 int ret;
4440 struct inode *btree_inode = buf->pages[0]->mapping->host;
4441
4442 ret = extent_buffer_uptodate(buf);
4443 if (!ret)
4444 return ret;
4445
4446 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4447 parent_transid, atomic);
4448 if (ret == -EAGAIN)
4449 return ret;
4450 return !ret;
4451}
4452
4453void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4454{
4455 struct btrfs_fs_info *fs_info = buf->fs_info;
4456 u64 transid = btrfs_header_generation(buf);
4457 int was_dirty;
4458
4459#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4460 /*
4461 * This is a fast path so only do this check if we have sanity tests
4462 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4463 * outside of the sanity tests.
4464 */
4465 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4466 return;
4467#endif
4468 btrfs_assert_tree_locked(buf);
4469 if (transid != fs_info->generation)
4470 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4471 buf->start, transid, fs_info->generation);
4472 was_dirty = set_extent_buffer_dirty(buf);
4473 if (!was_dirty)
4474 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4475 buf->len,
4476 fs_info->dirty_metadata_batch);
4477#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4478 /*
4479 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4480 * but item data not updated.
4481 * So here we should only check item pointers, not item data.
4482 */
4483 if (btrfs_header_level(buf) == 0 &&
4484 btrfs_check_leaf_relaxed(buf)) {
4485 btrfs_print_leaf(buf);
4486 ASSERT(0);
4487 }
4488#endif
4489}
4490
4491static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4492 int flush_delayed)
4493{
4494 /*
4495 * looks as though older kernels can get into trouble with
4496 * this code, they end up stuck in balance_dirty_pages forever
4497 */
4498 int ret;
4499
4500 if (current->flags & PF_MEMALLOC)
4501 return;
4502
4503 if (flush_delayed)
4504 btrfs_balance_delayed_items(fs_info);
4505
4506 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4507 BTRFS_DIRTY_METADATA_THRESH,
4508 fs_info->dirty_metadata_batch);
4509 if (ret > 0) {
4510 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4511 }
4512}
4513
4514void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4515{
4516 __btrfs_btree_balance_dirty(fs_info, 1);
4517}
4518
4519void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4520{
4521 __btrfs_btree_balance_dirty(fs_info, 0);
4522}
4523
4524int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4525 struct btrfs_key *first_key)
4526{
4527 return btree_read_extent_buffer_pages(buf, parent_transid,
4528 level, first_key);
4529}
4530
4531static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4532{
4533 /* cleanup FS via transaction */
4534 btrfs_cleanup_transaction(fs_info);
4535
4536 mutex_lock(&fs_info->cleaner_mutex);
4537 btrfs_run_delayed_iputs(fs_info);
4538 mutex_unlock(&fs_info->cleaner_mutex);
4539
4540 down_write(&fs_info->cleanup_work_sem);
4541 up_write(&fs_info->cleanup_work_sem);
4542}
4543
4544static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4545{
4546 struct btrfs_root *gang[8];
4547 u64 root_objectid = 0;
4548 int ret;
4549
4550 spin_lock(&fs_info->fs_roots_radix_lock);
4551 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4552 (void **)gang, root_objectid,
4553 ARRAY_SIZE(gang))) != 0) {
4554 int i;
4555
4556 for (i = 0; i < ret; i++)
4557 gang[i] = btrfs_grab_root(gang[i]);
4558 spin_unlock(&fs_info->fs_roots_radix_lock);
4559
4560 for (i = 0; i < ret; i++) {
4561 if (!gang[i])
4562 continue;
4563 root_objectid = gang[i]->root_key.objectid;
4564 btrfs_free_log(NULL, gang[i]);
4565 btrfs_put_root(gang[i]);
4566 }
4567 root_objectid++;
4568 spin_lock(&fs_info->fs_roots_radix_lock);
4569 }
4570 spin_unlock(&fs_info->fs_roots_radix_lock);
4571 btrfs_free_log_root_tree(NULL, fs_info);
4572}
4573
4574static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4575{
4576 struct btrfs_ordered_extent *ordered;
4577
4578 spin_lock(&root->ordered_extent_lock);
4579 /*
4580 * This will just short circuit the ordered completion stuff which will
4581 * make sure the ordered extent gets properly cleaned up.
4582 */
4583 list_for_each_entry(ordered, &root->ordered_extents,
4584 root_extent_list)
4585 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4586 spin_unlock(&root->ordered_extent_lock);
4587}
4588
4589static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4590{
4591 struct btrfs_root *root;
4592 struct list_head splice;
4593
4594 INIT_LIST_HEAD(&splice);
4595
4596 spin_lock(&fs_info->ordered_root_lock);
4597 list_splice_init(&fs_info->ordered_roots, &splice);
4598 while (!list_empty(&splice)) {
4599 root = list_first_entry(&splice, struct btrfs_root,
4600 ordered_root);
4601 list_move_tail(&root->ordered_root,
4602 &fs_info->ordered_roots);
4603
4604 spin_unlock(&fs_info->ordered_root_lock);
4605 btrfs_destroy_ordered_extents(root);
4606
4607 cond_resched();
4608 spin_lock(&fs_info->ordered_root_lock);
4609 }
4610 spin_unlock(&fs_info->ordered_root_lock);
4611
4612 /*
4613 * We need this here because if we've been flipped read-only we won't
4614 * get sync() from the umount, so we need to make sure any ordered
4615 * extents that haven't had their dirty pages IO start writeout yet
4616 * actually get run and error out properly.
4617 */
4618 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4619}
4620
4621static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4622 struct btrfs_fs_info *fs_info)
4623{
4624 struct rb_node *node;
4625 struct btrfs_delayed_ref_root *delayed_refs;
4626 struct btrfs_delayed_ref_node *ref;
4627 int ret = 0;
4628
4629 delayed_refs = &trans->delayed_refs;
4630
4631 spin_lock(&delayed_refs->lock);
4632 if (atomic_read(&delayed_refs->num_entries) == 0) {
4633 spin_unlock(&delayed_refs->lock);
4634 btrfs_debug(fs_info, "delayed_refs has NO entry");
4635 return ret;
4636 }
4637
4638 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4639 struct btrfs_delayed_ref_head *head;
4640 struct rb_node *n;
4641 bool pin_bytes = false;
4642
4643 head = rb_entry(node, struct btrfs_delayed_ref_head,
4644 href_node);
4645 if (btrfs_delayed_ref_lock(delayed_refs, head))
4646 continue;
4647
4648 spin_lock(&head->lock);
4649 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4650 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4651 ref_node);
4652 ref->in_tree = 0;
4653 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4654 RB_CLEAR_NODE(&ref->ref_node);
4655 if (!list_empty(&ref->add_list))
4656 list_del(&ref->add_list);
4657 atomic_dec(&delayed_refs->num_entries);
4658 btrfs_put_delayed_ref(ref);
4659 }
4660 if (head->must_insert_reserved)
4661 pin_bytes = true;
4662 btrfs_free_delayed_extent_op(head->extent_op);
4663 btrfs_delete_ref_head(delayed_refs, head);
4664 spin_unlock(&head->lock);
4665 spin_unlock(&delayed_refs->lock);
4666 mutex_unlock(&head->mutex);
4667
4668 if (pin_bytes) {
4669 struct btrfs_block_group *cache;
4670
4671 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4672 BUG_ON(!cache);
4673
4674 spin_lock(&cache->space_info->lock);
4675 spin_lock(&cache->lock);
4676 cache->pinned += head->num_bytes;
4677 btrfs_space_info_update_bytes_pinned(fs_info,
4678 cache->space_info, head->num_bytes);
4679 cache->reserved -= head->num_bytes;
4680 cache->space_info->bytes_reserved -= head->num_bytes;
4681 spin_unlock(&cache->lock);
4682 spin_unlock(&cache->space_info->lock);
4683
4684 btrfs_put_block_group(cache);
4685
4686 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4687 head->bytenr + head->num_bytes - 1);
4688 }
4689 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4690 btrfs_put_delayed_ref_head(head);
4691 cond_resched();
4692 spin_lock(&delayed_refs->lock);
4693 }
4694 btrfs_qgroup_destroy_extent_records(trans);
4695
4696 spin_unlock(&delayed_refs->lock);
4697
4698 return ret;
4699}
4700
4701static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4702{
4703 struct btrfs_inode *btrfs_inode;
4704 struct list_head splice;
4705
4706 INIT_LIST_HEAD(&splice);
4707
4708 spin_lock(&root->delalloc_lock);
4709 list_splice_init(&root->delalloc_inodes, &splice);
4710
4711 while (!list_empty(&splice)) {
4712 struct inode *inode = NULL;
4713 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4714 delalloc_inodes);
4715 __btrfs_del_delalloc_inode(root, btrfs_inode);
4716 spin_unlock(&root->delalloc_lock);
4717
4718 /*
4719 * Make sure we get a live inode and that it'll not disappear
4720 * meanwhile.
4721 */
4722 inode = igrab(&btrfs_inode->vfs_inode);
4723 if (inode) {
4724 invalidate_inode_pages2(inode->i_mapping);
4725 iput(inode);
4726 }
4727 spin_lock(&root->delalloc_lock);
4728 }
4729 spin_unlock(&root->delalloc_lock);
4730}
4731
4732static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4733{
4734 struct btrfs_root *root;
4735 struct list_head splice;
4736
4737 INIT_LIST_HEAD(&splice);
4738
4739 spin_lock(&fs_info->delalloc_root_lock);
4740 list_splice_init(&fs_info->delalloc_roots, &splice);
4741 while (!list_empty(&splice)) {
4742 root = list_first_entry(&splice, struct btrfs_root,
4743 delalloc_root);
4744 root = btrfs_grab_root(root);
4745 BUG_ON(!root);
4746 spin_unlock(&fs_info->delalloc_root_lock);
4747
4748 btrfs_destroy_delalloc_inodes(root);
4749 btrfs_put_root(root);
4750
4751 spin_lock(&fs_info->delalloc_root_lock);
4752 }
4753 spin_unlock(&fs_info->delalloc_root_lock);
4754}
4755
4756static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4757 struct extent_io_tree *dirty_pages,
4758 int mark)
4759{
4760 int ret;
4761 struct extent_buffer *eb;
4762 u64 start = 0;
4763 u64 end;
4764
4765 while (1) {
4766 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4767 mark, NULL);
4768 if (ret)
4769 break;
4770
4771 clear_extent_bits(dirty_pages, start, end, mark);
4772 while (start <= end) {
4773 eb = find_extent_buffer(fs_info, start);
4774 start += fs_info->nodesize;
4775 if (!eb)
4776 continue;
4777 wait_on_extent_buffer_writeback(eb);
4778
4779 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4780 &eb->bflags))
4781 clear_extent_buffer_dirty(eb);
4782 free_extent_buffer_stale(eb);
4783 }
4784 }
4785
4786 return ret;
4787}
4788
4789static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4790 struct extent_io_tree *unpin)
4791{
4792 u64 start;
4793 u64 end;
4794 int ret;
4795
4796 while (1) {
4797 struct extent_state *cached_state = NULL;
4798
4799 /*
4800 * The btrfs_finish_extent_commit() may get the same range as
4801 * ours between find_first_extent_bit and clear_extent_dirty.
4802 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4803 * the same extent range.
4804 */
4805 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4806 ret = find_first_extent_bit(unpin, 0, &start, &end,
4807 EXTENT_DIRTY, &cached_state);
4808 if (ret) {
4809 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4810 break;
4811 }
4812
4813 clear_extent_dirty(unpin, start, end, &cached_state);
4814 free_extent_state(cached_state);
4815 btrfs_error_unpin_extent_range(fs_info, start, end);
4816 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4817 cond_resched();
4818 }
4819
4820 return 0;
4821}
4822
4823static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4824{
4825 struct inode *inode;
4826
4827 inode = cache->io_ctl.inode;
4828 if (inode) {
4829 invalidate_inode_pages2(inode->i_mapping);
4830 BTRFS_I(inode)->generation = 0;
4831 cache->io_ctl.inode = NULL;
4832 iput(inode);
4833 }
4834 ASSERT(cache->io_ctl.pages == NULL);
4835 btrfs_put_block_group(cache);
4836}
4837
4838void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4839 struct btrfs_fs_info *fs_info)
4840{
4841 struct btrfs_block_group *cache;
4842
4843 spin_lock(&cur_trans->dirty_bgs_lock);
4844 while (!list_empty(&cur_trans->dirty_bgs)) {
4845 cache = list_first_entry(&cur_trans->dirty_bgs,
4846 struct btrfs_block_group,
4847 dirty_list);
4848
4849 if (!list_empty(&cache->io_list)) {
4850 spin_unlock(&cur_trans->dirty_bgs_lock);
4851 list_del_init(&cache->io_list);
4852 btrfs_cleanup_bg_io(cache);
4853 spin_lock(&cur_trans->dirty_bgs_lock);
4854 }
4855
4856 list_del_init(&cache->dirty_list);
4857 spin_lock(&cache->lock);
4858 cache->disk_cache_state = BTRFS_DC_ERROR;
4859 spin_unlock(&cache->lock);
4860
4861 spin_unlock(&cur_trans->dirty_bgs_lock);
4862 btrfs_put_block_group(cache);
4863 btrfs_delayed_refs_rsv_release(fs_info, 1);
4864 spin_lock(&cur_trans->dirty_bgs_lock);
4865 }
4866 spin_unlock(&cur_trans->dirty_bgs_lock);
4867
4868 /*
4869 * Refer to the definition of io_bgs member for details why it's safe
4870 * to use it without any locking
4871 */
4872 while (!list_empty(&cur_trans->io_bgs)) {
4873 cache = list_first_entry(&cur_trans->io_bgs,
4874 struct btrfs_block_group,
4875 io_list);
4876
4877 list_del_init(&cache->io_list);
4878 spin_lock(&cache->lock);
4879 cache->disk_cache_state = BTRFS_DC_ERROR;
4880 spin_unlock(&cache->lock);
4881 btrfs_cleanup_bg_io(cache);
4882 }
4883}
4884
4885void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4886 struct btrfs_fs_info *fs_info)
4887{
4888 struct btrfs_device *dev, *tmp;
4889
4890 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4891 ASSERT(list_empty(&cur_trans->dirty_bgs));
4892 ASSERT(list_empty(&cur_trans->io_bgs));
4893
4894 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4895 post_commit_list) {
4896 list_del_init(&dev->post_commit_list);
4897 }
4898
4899 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4900
4901 cur_trans->state = TRANS_STATE_COMMIT_START;
4902 wake_up(&fs_info->transaction_blocked_wait);
4903
4904 cur_trans->state = TRANS_STATE_UNBLOCKED;
4905 wake_up(&fs_info->transaction_wait);
4906
4907 btrfs_destroy_delayed_inodes(fs_info);
4908
4909 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4910 EXTENT_DIRTY);
4911 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4912
4913 btrfs_free_redirty_list(cur_trans);
4914
4915 cur_trans->state =TRANS_STATE_COMPLETED;
4916 wake_up(&cur_trans->commit_wait);
4917}
4918
4919static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4920{
4921 struct btrfs_transaction *t;
4922
4923 mutex_lock(&fs_info->transaction_kthread_mutex);
4924
4925 spin_lock(&fs_info->trans_lock);
4926 while (!list_empty(&fs_info->trans_list)) {
4927 t = list_first_entry(&fs_info->trans_list,
4928 struct btrfs_transaction, list);
4929 if (t->state >= TRANS_STATE_COMMIT_START) {
4930 refcount_inc(&t->use_count);
4931 spin_unlock(&fs_info->trans_lock);
4932 btrfs_wait_for_commit(fs_info, t->transid);
4933 btrfs_put_transaction(t);
4934 spin_lock(&fs_info->trans_lock);
4935 continue;
4936 }
4937 if (t == fs_info->running_transaction) {
4938 t->state = TRANS_STATE_COMMIT_DOING;
4939 spin_unlock(&fs_info->trans_lock);
4940 /*
4941 * We wait for 0 num_writers since we don't hold a trans
4942 * handle open currently for this transaction.
4943 */
4944 wait_event(t->writer_wait,
4945 atomic_read(&t->num_writers) == 0);
4946 } else {
4947 spin_unlock(&fs_info->trans_lock);
4948 }
4949 btrfs_cleanup_one_transaction(t, fs_info);
4950
4951 spin_lock(&fs_info->trans_lock);
4952 if (t == fs_info->running_transaction)
4953 fs_info->running_transaction = NULL;
4954 list_del_init(&t->list);
4955 spin_unlock(&fs_info->trans_lock);
4956
4957 btrfs_put_transaction(t);
4958 trace_btrfs_transaction_commit(fs_info->tree_root);
4959 spin_lock(&fs_info->trans_lock);
4960 }
4961 spin_unlock(&fs_info->trans_lock);
4962 btrfs_destroy_all_ordered_extents(fs_info);
4963 btrfs_destroy_delayed_inodes(fs_info);
4964 btrfs_assert_delayed_root_empty(fs_info);
4965 btrfs_destroy_all_delalloc_inodes(fs_info);
4966 btrfs_drop_all_logs(fs_info);
4967 mutex_unlock(&fs_info->transaction_kthread_mutex);
4968
4969 return 0;
4970}
4971
4972int btrfs_init_root_free_objectid(struct btrfs_root *root)
4973{
4974 struct btrfs_path *path;
4975 int ret;
4976 struct extent_buffer *l;
4977 struct btrfs_key search_key;
4978 struct btrfs_key found_key;
4979 int slot;
4980
4981 path = btrfs_alloc_path();
4982 if (!path)
4983 return -ENOMEM;
4984
4985 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4986 search_key.type = -1;
4987 search_key.offset = (u64)-1;
4988 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4989 if (ret < 0)
4990 goto error;
4991 BUG_ON(ret == 0); /* Corruption */
4992 if (path->slots[0] > 0) {
4993 slot = path->slots[0] - 1;
4994 l = path->nodes[0];
4995 btrfs_item_key_to_cpu(l, &found_key, slot);
4996 root->free_objectid = max_t(u64, found_key.objectid + 1,
4997 BTRFS_FIRST_FREE_OBJECTID);
4998 } else {
4999 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5000 }
5001 ret = 0;
5002error:
5003 btrfs_free_path(path);
5004 return ret;
5005}
5006
5007int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5008{
5009 int ret;
5010 mutex_lock(&root->objectid_mutex);
5011
5012 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5013 btrfs_warn(root->fs_info,
5014 "the objectid of root %llu reaches its highest value",
5015 root->root_key.objectid);
5016 ret = -ENOSPC;
5017 goto out;
5018 }
5019
5020 *objectid = root->free_objectid++;
5021 ret = 0;
5022out:
5023 mutex_unlock(&root->objectid_mutex);
5024 return ret;
5025}