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