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 "dev-replace.h"
33#include "raid56.h"
34#include "sysfs.h"
35#include "qgroup.h"
36#include "compression.h"
37#include "tree-checker.h"
38#include "ref-verify.h"
39#include "block-group.h"
40#include "discard.h"
41#include "space-info.h"
42#include "zoned.h"
43#include "subpage.h"
44#include "fs.h"
45#include "accessors.h"
46#include "extent-tree.h"
47#include "root-tree.h"
48#include "defrag.h"
49#include "uuid-tree.h"
50#include "relocation.h"
51#include "scrub.h"
52#include "super.h"
53
54#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
55 BTRFS_HEADER_FLAG_RELOC |\
56 BTRFS_SUPER_FLAG_ERROR |\
57 BTRFS_SUPER_FLAG_SEEDING |\
58 BTRFS_SUPER_FLAG_METADUMP |\
59 BTRFS_SUPER_FLAG_METADUMP_V2)
60
61static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
62static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
63
64static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
65{
66 if (fs_info->csum_shash)
67 crypto_free_shash(fs_info->csum_shash);
68}
69
70/*
71 * Compute the csum of a btree block and store the result to provided buffer.
72 */
73static void csum_tree_block(struct extent_buffer *buf, u8 *result)
74{
75 struct btrfs_fs_info *fs_info = buf->fs_info;
76 int num_pages;
77 u32 first_page_part;
78 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
79 char *kaddr;
80 int i;
81
82 shash->tfm = fs_info->csum_shash;
83 crypto_shash_init(shash);
84
85 if (buf->addr) {
86 /* Pages are contiguous, handle them as a big one. */
87 kaddr = buf->addr;
88 first_page_part = fs_info->nodesize;
89 num_pages = 1;
90 } else {
91 kaddr = folio_address(buf->folios[0]);
92 first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
93 num_pages = num_extent_pages(buf);
94 }
95
96 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
97 first_page_part - BTRFS_CSUM_SIZE);
98
99 /*
100 * Multiple single-page folios case would reach here.
101 *
102 * nodesize <= PAGE_SIZE and large folio all handled by above
103 * crypto_shash_update() already.
104 */
105 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
106 kaddr = folio_address(buf->folios[i]);
107 crypto_shash_update(shash, kaddr, PAGE_SIZE);
108 }
109 memset(result, 0, BTRFS_CSUM_SIZE);
110 crypto_shash_final(shash, result);
111}
112
113/*
114 * we can't consider a given block up to date unless the transid of the
115 * block matches the transid in the parent node's pointer. This is how we
116 * detect blocks that either didn't get written at all or got written
117 * in the wrong place.
118 */
119int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
120{
121 if (!extent_buffer_uptodate(eb))
122 return 0;
123
124 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
125 return 1;
126
127 if (atomic)
128 return -EAGAIN;
129
130 if (!extent_buffer_uptodate(eb) ||
131 btrfs_header_generation(eb) != parent_transid) {
132 btrfs_err_rl(eb->fs_info,
133"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
134 eb->start, eb->read_mirror,
135 parent_transid, btrfs_header_generation(eb));
136 clear_extent_buffer_uptodate(eb);
137 return 0;
138 }
139 return 1;
140}
141
142static bool btrfs_supported_super_csum(u16 csum_type)
143{
144 switch (csum_type) {
145 case BTRFS_CSUM_TYPE_CRC32:
146 case BTRFS_CSUM_TYPE_XXHASH:
147 case BTRFS_CSUM_TYPE_SHA256:
148 case BTRFS_CSUM_TYPE_BLAKE2:
149 return true;
150 default:
151 return false;
152 }
153}
154
155/*
156 * Return 0 if the superblock checksum type matches the checksum value of that
157 * algorithm. Pass the raw disk superblock data.
158 */
159int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
160 const struct btrfs_super_block *disk_sb)
161{
162 char result[BTRFS_CSUM_SIZE];
163 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
164
165 shash->tfm = fs_info->csum_shash;
166
167 /*
168 * The super_block structure does not span the whole
169 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
170 * filled with zeros and is included in the checksum.
171 */
172 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
173 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
174
175 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
176 return 1;
177
178 return 0;
179}
180
181static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
182 int mirror_num)
183{
184 struct btrfs_fs_info *fs_info = eb->fs_info;
185 int num_folios = num_extent_folios(eb);
186 int ret = 0;
187
188 if (sb_rdonly(fs_info->sb))
189 return -EROFS;
190
191 for (int i = 0; i < num_folios; i++) {
192 struct folio *folio = eb->folios[i];
193 u64 start = max_t(u64, eb->start, folio_pos(folio));
194 u64 end = min_t(u64, eb->start + eb->len,
195 folio_pos(folio) + eb->folio_size);
196 u32 len = end - start;
197
198 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
199 start, folio, offset_in_folio(folio, start),
200 mirror_num);
201 if (ret)
202 break;
203 }
204
205 return ret;
206}
207
208/*
209 * helper to read a given tree block, doing retries as required when
210 * the checksums don't match and we have alternate mirrors to try.
211 *
212 * @check: expected tree parentness check, see the comments of the
213 * structure for details.
214 */
215int btrfs_read_extent_buffer(struct extent_buffer *eb,
216 struct btrfs_tree_parent_check *check)
217{
218 struct btrfs_fs_info *fs_info = eb->fs_info;
219 int failed = 0;
220 int ret;
221 int num_copies = 0;
222 int mirror_num = 0;
223 int failed_mirror = 0;
224
225 ASSERT(check);
226
227 while (1) {
228 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
229 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
230 if (!ret)
231 break;
232
233 num_copies = btrfs_num_copies(fs_info,
234 eb->start, eb->len);
235 if (num_copies == 1)
236 break;
237
238 if (!failed_mirror) {
239 failed = 1;
240 failed_mirror = eb->read_mirror;
241 }
242
243 mirror_num++;
244 if (mirror_num == failed_mirror)
245 mirror_num++;
246
247 if (mirror_num > num_copies)
248 break;
249 }
250
251 if (failed && !ret && failed_mirror)
252 btrfs_repair_eb_io_failure(eb, failed_mirror);
253
254 return ret;
255}
256
257/*
258 * Checksum a dirty tree block before IO.
259 */
260blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
261{
262 struct extent_buffer *eb = bbio->private;
263 struct btrfs_fs_info *fs_info = eb->fs_info;
264 u64 found_start = btrfs_header_bytenr(eb);
265 u64 last_trans;
266 u8 result[BTRFS_CSUM_SIZE];
267 int ret;
268
269 /* Btree blocks are always contiguous on disk. */
270 if (WARN_ON_ONCE(bbio->file_offset != eb->start))
271 return BLK_STS_IOERR;
272 if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
273 return BLK_STS_IOERR;
274
275 /*
276 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
277 * checksum it but zero-out its content. This is done to preserve
278 * ordering of I/O without unnecessarily writing out data.
279 */
280 if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
281 memzero_extent_buffer(eb, 0, eb->len);
282 return BLK_STS_OK;
283 }
284
285 if (WARN_ON_ONCE(found_start != eb->start))
286 return BLK_STS_IOERR;
287 if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
288 eb->start, eb->len)))
289 return BLK_STS_IOERR;
290
291 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
292 offsetof(struct btrfs_header, fsid),
293 BTRFS_FSID_SIZE) == 0);
294 csum_tree_block(eb, result);
295
296 if (btrfs_header_level(eb))
297 ret = btrfs_check_node(eb);
298 else
299 ret = btrfs_check_leaf(eb);
300
301 if (ret < 0)
302 goto error;
303
304 /*
305 * Also check the generation, the eb reached here must be newer than
306 * last committed. Or something seriously wrong happened.
307 */
308 last_trans = btrfs_get_last_trans_committed(fs_info);
309 if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
310 ret = -EUCLEAN;
311 btrfs_err(fs_info,
312 "block=%llu bad generation, have %llu expect > %llu",
313 eb->start, btrfs_header_generation(eb), last_trans);
314 goto error;
315 }
316 write_extent_buffer(eb, result, 0, fs_info->csum_size);
317 return BLK_STS_OK;
318
319error:
320 btrfs_print_tree(eb, 0);
321 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
322 eb->start);
323 /*
324 * Be noisy if this is an extent buffer from a log tree. We don't abort
325 * a transaction in case there's a bad log tree extent buffer, we just
326 * fallback to a transaction commit. Still we want to know when there is
327 * a bad log tree extent buffer, as that may signal a bug somewhere.
328 */
329 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
330 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
331 return errno_to_blk_status(ret);
332}
333
334static bool check_tree_block_fsid(struct extent_buffer *eb)
335{
336 struct btrfs_fs_info *fs_info = eb->fs_info;
337 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
338 u8 fsid[BTRFS_FSID_SIZE];
339
340 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
341 BTRFS_FSID_SIZE);
342
343 /*
344 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
345 * This is then overwritten by metadata_uuid if it is present in the
346 * device_list_add(). The same true for a seed device as well. So use of
347 * fs_devices::metadata_uuid is appropriate here.
348 */
349 if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
350 return false;
351
352 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
353 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
354 return false;
355
356 return true;
357}
358
359/* Do basic extent buffer checks at read time */
360int btrfs_validate_extent_buffer(struct extent_buffer *eb,
361 struct btrfs_tree_parent_check *check)
362{
363 struct btrfs_fs_info *fs_info = eb->fs_info;
364 u64 found_start;
365 const u32 csum_size = fs_info->csum_size;
366 u8 found_level;
367 u8 result[BTRFS_CSUM_SIZE];
368 const u8 *header_csum;
369 int ret = 0;
370
371 ASSERT(check);
372
373 found_start = btrfs_header_bytenr(eb);
374 if (found_start != eb->start) {
375 btrfs_err_rl(fs_info,
376 "bad tree block start, mirror %u want %llu have %llu",
377 eb->read_mirror, eb->start, found_start);
378 ret = -EIO;
379 goto out;
380 }
381 if (check_tree_block_fsid(eb)) {
382 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
383 eb->start, eb->read_mirror);
384 ret = -EIO;
385 goto out;
386 }
387 found_level = btrfs_header_level(eb);
388 if (found_level >= BTRFS_MAX_LEVEL) {
389 btrfs_err(fs_info,
390 "bad tree block level, mirror %u level %d on logical %llu",
391 eb->read_mirror, btrfs_header_level(eb), eb->start);
392 ret = -EIO;
393 goto out;
394 }
395
396 csum_tree_block(eb, result);
397 header_csum = folio_address(eb->folios[0]) +
398 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
399
400 if (memcmp(result, header_csum, csum_size) != 0) {
401 btrfs_warn_rl(fs_info,
402"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
403 eb->start, eb->read_mirror,
404 CSUM_FMT_VALUE(csum_size, header_csum),
405 CSUM_FMT_VALUE(csum_size, result),
406 btrfs_header_level(eb));
407 ret = -EUCLEAN;
408 goto out;
409 }
410
411 if (found_level != check->level) {
412 btrfs_err(fs_info,
413 "level verify failed on logical %llu mirror %u wanted %u found %u",
414 eb->start, eb->read_mirror, check->level, found_level);
415 ret = -EIO;
416 goto out;
417 }
418 if (unlikely(check->transid &&
419 btrfs_header_generation(eb) != check->transid)) {
420 btrfs_err_rl(eb->fs_info,
421"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
422 eb->start, eb->read_mirror, check->transid,
423 btrfs_header_generation(eb));
424 ret = -EIO;
425 goto out;
426 }
427 if (check->has_first_key) {
428 struct btrfs_key *expect_key = &check->first_key;
429 struct btrfs_key found_key;
430
431 if (found_level)
432 btrfs_node_key_to_cpu(eb, &found_key, 0);
433 else
434 btrfs_item_key_to_cpu(eb, &found_key, 0);
435 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
436 btrfs_err(fs_info,
437"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
438 eb->start, check->transid,
439 expect_key->objectid,
440 expect_key->type, expect_key->offset,
441 found_key.objectid, found_key.type,
442 found_key.offset);
443 ret = -EUCLEAN;
444 goto out;
445 }
446 }
447 if (check->owner_root) {
448 ret = btrfs_check_eb_owner(eb, check->owner_root);
449 if (ret < 0)
450 goto out;
451 }
452
453 /*
454 * If this is a leaf block and it is corrupt, set the corrupt bit so
455 * that we don't try and read the other copies of this block, just
456 * return -EIO.
457 */
458 if (found_level == 0 && btrfs_check_leaf(eb)) {
459 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
460 ret = -EIO;
461 }
462
463 if (found_level > 0 && btrfs_check_node(eb))
464 ret = -EIO;
465
466 if (ret)
467 btrfs_err(fs_info,
468 "read time tree block corruption detected on logical %llu mirror %u",
469 eb->start, eb->read_mirror);
470out:
471 return ret;
472}
473
474#ifdef CONFIG_MIGRATION
475static int btree_migrate_folio(struct address_space *mapping,
476 struct folio *dst, struct folio *src, enum migrate_mode mode)
477{
478 /*
479 * we can't safely write a btree page from here,
480 * we haven't done the locking hook
481 */
482 if (folio_test_dirty(src))
483 return -EAGAIN;
484 /*
485 * Buffers may be managed in a filesystem specific way.
486 * We must have no buffers or drop them.
487 */
488 if (folio_get_private(src) &&
489 !filemap_release_folio(src, GFP_KERNEL))
490 return -EAGAIN;
491 return migrate_folio(mapping, dst, src, mode);
492}
493#else
494#define btree_migrate_folio NULL
495#endif
496
497static int btree_writepages(struct address_space *mapping,
498 struct writeback_control *wbc)
499{
500 int ret;
501
502 if (wbc->sync_mode == WB_SYNC_NONE) {
503 struct btrfs_fs_info *fs_info;
504
505 if (wbc->for_kupdate)
506 return 0;
507
508 fs_info = inode_to_fs_info(mapping->host);
509 /* this is a bit racy, but that's ok */
510 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
511 BTRFS_DIRTY_METADATA_THRESH,
512 fs_info->dirty_metadata_batch);
513 if (ret < 0)
514 return 0;
515 }
516 return btree_write_cache_pages(mapping, wbc);
517}
518
519static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
520{
521 if (folio_test_writeback(folio) || folio_test_dirty(folio))
522 return false;
523
524 return try_release_extent_buffer(&folio->page);
525}
526
527static void btree_invalidate_folio(struct folio *folio, size_t offset,
528 size_t length)
529{
530 struct extent_io_tree *tree;
531
532 tree = &folio_to_inode(folio)->io_tree;
533 extent_invalidate_folio(tree, folio, offset);
534 btree_release_folio(folio, GFP_NOFS);
535 if (folio_get_private(folio)) {
536 btrfs_warn(folio_to_fs_info(folio),
537 "folio private not zero on folio %llu",
538 (unsigned long long)folio_pos(folio));
539 folio_detach_private(folio);
540 }
541}
542
543#ifdef DEBUG
544static bool btree_dirty_folio(struct address_space *mapping,
545 struct folio *folio)
546{
547 struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
548 struct btrfs_subpage_info *spi = fs_info->subpage_info;
549 struct btrfs_subpage *subpage;
550 struct extent_buffer *eb;
551 int cur_bit = 0;
552 u64 page_start = folio_pos(folio);
553
554 if (fs_info->sectorsize == PAGE_SIZE) {
555 eb = folio_get_private(folio);
556 BUG_ON(!eb);
557 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
558 BUG_ON(!atomic_read(&eb->refs));
559 btrfs_assert_tree_write_locked(eb);
560 return filemap_dirty_folio(mapping, folio);
561 }
562
563 ASSERT(spi);
564 subpage = folio_get_private(folio);
565
566 for (cur_bit = spi->dirty_offset;
567 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
568 cur_bit++) {
569 unsigned long flags;
570 u64 cur;
571
572 spin_lock_irqsave(&subpage->lock, flags);
573 if (!test_bit(cur_bit, subpage->bitmaps)) {
574 spin_unlock_irqrestore(&subpage->lock, flags);
575 continue;
576 }
577 spin_unlock_irqrestore(&subpage->lock, flags);
578 cur = page_start + cur_bit * fs_info->sectorsize;
579
580 eb = find_extent_buffer(fs_info, cur);
581 ASSERT(eb);
582 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
583 ASSERT(atomic_read(&eb->refs));
584 btrfs_assert_tree_write_locked(eb);
585 free_extent_buffer(eb);
586
587 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
588 }
589 return filemap_dirty_folio(mapping, folio);
590}
591#else
592#define btree_dirty_folio filemap_dirty_folio
593#endif
594
595static const struct address_space_operations btree_aops = {
596 .writepages = btree_writepages,
597 .release_folio = btree_release_folio,
598 .invalidate_folio = btree_invalidate_folio,
599 .migrate_folio = btree_migrate_folio,
600 .dirty_folio = btree_dirty_folio,
601};
602
603struct extent_buffer *btrfs_find_create_tree_block(
604 struct btrfs_fs_info *fs_info,
605 u64 bytenr, u64 owner_root,
606 int level)
607{
608 if (btrfs_is_testing(fs_info))
609 return alloc_test_extent_buffer(fs_info, bytenr);
610 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
611}
612
613/*
614 * Read tree block at logical address @bytenr and do variant basic but critical
615 * verification.
616 *
617 * @check: expected tree parentness check, see comments of the
618 * structure for details.
619 */
620struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
621 struct btrfs_tree_parent_check *check)
622{
623 struct extent_buffer *buf = NULL;
624 int ret;
625
626 ASSERT(check);
627
628 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
629 check->level);
630 if (IS_ERR(buf))
631 return buf;
632
633 ret = btrfs_read_extent_buffer(buf, check);
634 if (ret) {
635 free_extent_buffer_stale(buf);
636 return ERR_PTR(ret);
637 }
638 if (btrfs_check_eb_owner(buf, check->owner_root)) {
639 free_extent_buffer_stale(buf);
640 return ERR_PTR(-EUCLEAN);
641 }
642 return buf;
643
644}
645
646static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
647 u64 objectid)
648{
649 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
650
651 memset(&root->root_key, 0, sizeof(root->root_key));
652 memset(&root->root_item, 0, sizeof(root->root_item));
653 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
654 root->fs_info = fs_info;
655 root->root_key.objectid = objectid;
656 root->node = NULL;
657 root->commit_root = NULL;
658 root->state = 0;
659 RB_CLEAR_NODE(&root->rb_node);
660
661 root->last_trans = 0;
662 root->free_objectid = 0;
663 root->nr_delalloc_inodes = 0;
664 root->nr_ordered_extents = 0;
665 root->inode_tree = RB_ROOT;
666 /* GFP flags are compatible with XA_FLAGS_*. */
667 xa_init_flags(&root->delayed_nodes, GFP_ATOMIC);
668
669 btrfs_init_root_block_rsv(root);
670
671 INIT_LIST_HEAD(&root->dirty_list);
672 INIT_LIST_HEAD(&root->root_list);
673 INIT_LIST_HEAD(&root->delalloc_inodes);
674 INIT_LIST_HEAD(&root->delalloc_root);
675 INIT_LIST_HEAD(&root->ordered_extents);
676 INIT_LIST_HEAD(&root->ordered_root);
677 INIT_LIST_HEAD(&root->reloc_dirty_list);
678 spin_lock_init(&root->inode_lock);
679 spin_lock_init(&root->delalloc_lock);
680 spin_lock_init(&root->ordered_extent_lock);
681 spin_lock_init(&root->accounting_lock);
682 spin_lock_init(&root->qgroup_meta_rsv_lock);
683 mutex_init(&root->objectid_mutex);
684 mutex_init(&root->log_mutex);
685 mutex_init(&root->ordered_extent_mutex);
686 mutex_init(&root->delalloc_mutex);
687 init_waitqueue_head(&root->qgroup_flush_wait);
688 init_waitqueue_head(&root->log_writer_wait);
689 init_waitqueue_head(&root->log_commit_wait[0]);
690 init_waitqueue_head(&root->log_commit_wait[1]);
691 INIT_LIST_HEAD(&root->log_ctxs[0]);
692 INIT_LIST_HEAD(&root->log_ctxs[1]);
693 atomic_set(&root->log_commit[0], 0);
694 atomic_set(&root->log_commit[1], 0);
695 atomic_set(&root->log_writers, 0);
696 atomic_set(&root->log_batch, 0);
697 refcount_set(&root->refs, 1);
698 atomic_set(&root->snapshot_force_cow, 0);
699 atomic_set(&root->nr_swapfiles, 0);
700 btrfs_set_root_log_transid(root, 0);
701 root->log_transid_committed = -1;
702 btrfs_set_root_last_log_commit(root, 0);
703 root->anon_dev = 0;
704 if (!dummy) {
705 extent_io_tree_init(fs_info, &root->dirty_log_pages,
706 IO_TREE_ROOT_DIRTY_LOG_PAGES);
707 extent_io_tree_init(fs_info, &root->log_csum_range,
708 IO_TREE_LOG_CSUM_RANGE);
709 }
710
711 spin_lock_init(&root->root_item_lock);
712 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
713#ifdef CONFIG_BTRFS_DEBUG
714 INIT_LIST_HEAD(&root->leak_list);
715 spin_lock(&fs_info->fs_roots_radix_lock);
716 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
717 spin_unlock(&fs_info->fs_roots_radix_lock);
718#endif
719}
720
721static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
722 u64 objectid, gfp_t flags)
723{
724 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
725 if (root)
726 __setup_root(root, fs_info, objectid);
727 return root;
728}
729
730#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
731/* Should only be used by the testing infrastructure */
732struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
733{
734 struct btrfs_root *root;
735
736 if (!fs_info)
737 return ERR_PTR(-EINVAL);
738
739 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
740 if (!root)
741 return ERR_PTR(-ENOMEM);
742
743 /* We don't use the stripesize in selftest, set it as sectorsize */
744 root->alloc_bytenr = 0;
745
746 return root;
747}
748#endif
749
750static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
751{
752 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
753 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
754
755 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
756}
757
758static int global_root_key_cmp(const void *k, const struct rb_node *node)
759{
760 const struct btrfs_key *key = k;
761 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
762
763 return btrfs_comp_cpu_keys(key, &root->root_key);
764}
765
766int btrfs_global_root_insert(struct btrfs_root *root)
767{
768 struct btrfs_fs_info *fs_info = root->fs_info;
769 struct rb_node *tmp;
770 int ret = 0;
771
772 write_lock(&fs_info->global_root_lock);
773 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
774 write_unlock(&fs_info->global_root_lock);
775
776 if (tmp) {
777 ret = -EEXIST;
778 btrfs_warn(fs_info, "global root %llu %llu already exists",
779 root->root_key.objectid, root->root_key.offset);
780 }
781 return ret;
782}
783
784void btrfs_global_root_delete(struct btrfs_root *root)
785{
786 struct btrfs_fs_info *fs_info = root->fs_info;
787
788 write_lock(&fs_info->global_root_lock);
789 rb_erase(&root->rb_node, &fs_info->global_root_tree);
790 write_unlock(&fs_info->global_root_lock);
791}
792
793struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
794 struct btrfs_key *key)
795{
796 struct rb_node *node;
797 struct btrfs_root *root = NULL;
798
799 read_lock(&fs_info->global_root_lock);
800 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
801 if (node)
802 root = container_of(node, struct btrfs_root, rb_node);
803 read_unlock(&fs_info->global_root_lock);
804
805 return root;
806}
807
808static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
809{
810 struct btrfs_block_group *block_group;
811 u64 ret;
812
813 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
814 return 0;
815
816 if (bytenr)
817 block_group = btrfs_lookup_block_group(fs_info, bytenr);
818 else
819 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
820 ASSERT(block_group);
821 if (!block_group)
822 return 0;
823 ret = block_group->global_root_id;
824 btrfs_put_block_group(block_group);
825
826 return ret;
827}
828
829struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
830{
831 struct btrfs_key key = {
832 .objectid = BTRFS_CSUM_TREE_OBJECTID,
833 .type = BTRFS_ROOT_ITEM_KEY,
834 .offset = btrfs_global_root_id(fs_info, bytenr),
835 };
836
837 return btrfs_global_root(fs_info, &key);
838}
839
840struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
841{
842 struct btrfs_key key = {
843 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
844 .type = BTRFS_ROOT_ITEM_KEY,
845 .offset = btrfs_global_root_id(fs_info, bytenr),
846 };
847
848 return btrfs_global_root(fs_info, &key);
849}
850
851struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
852{
853 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
854 return fs_info->block_group_root;
855 return btrfs_extent_root(fs_info, 0);
856}
857
858struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
859 u64 objectid)
860{
861 struct btrfs_fs_info *fs_info = trans->fs_info;
862 struct extent_buffer *leaf;
863 struct btrfs_root *tree_root = fs_info->tree_root;
864 struct btrfs_root *root;
865 struct btrfs_key key;
866 unsigned int nofs_flag;
867 int ret = 0;
868
869 /*
870 * We're holding a transaction handle, so use a NOFS memory allocation
871 * context to avoid deadlock if reclaim happens.
872 */
873 nofs_flag = memalloc_nofs_save();
874 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
875 memalloc_nofs_restore(nofs_flag);
876 if (!root)
877 return ERR_PTR(-ENOMEM);
878
879 root->root_key.objectid = objectid;
880 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
881 root->root_key.offset = 0;
882
883 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
884 0, BTRFS_NESTING_NORMAL);
885 if (IS_ERR(leaf)) {
886 ret = PTR_ERR(leaf);
887 leaf = NULL;
888 goto fail;
889 }
890
891 root->node = leaf;
892 btrfs_mark_buffer_dirty(trans, leaf);
893
894 root->commit_root = btrfs_root_node(root);
895 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
896
897 btrfs_set_root_flags(&root->root_item, 0);
898 btrfs_set_root_limit(&root->root_item, 0);
899 btrfs_set_root_bytenr(&root->root_item, leaf->start);
900 btrfs_set_root_generation(&root->root_item, trans->transid);
901 btrfs_set_root_level(&root->root_item, 0);
902 btrfs_set_root_refs(&root->root_item, 1);
903 btrfs_set_root_used(&root->root_item, leaf->len);
904 btrfs_set_root_last_snapshot(&root->root_item, 0);
905 btrfs_set_root_dirid(&root->root_item, 0);
906 if (is_fstree(objectid))
907 generate_random_guid(root->root_item.uuid);
908 else
909 export_guid(root->root_item.uuid, &guid_null);
910 btrfs_set_root_drop_level(&root->root_item, 0);
911
912 btrfs_tree_unlock(leaf);
913
914 key.objectid = objectid;
915 key.type = BTRFS_ROOT_ITEM_KEY;
916 key.offset = 0;
917 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
918 if (ret)
919 goto fail;
920
921 return root;
922
923fail:
924 btrfs_put_root(root);
925
926 return ERR_PTR(ret);
927}
928
929static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
930 struct btrfs_fs_info *fs_info)
931{
932 struct btrfs_root *root;
933
934 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
935 if (!root)
936 return ERR_PTR(-ENOMEM);
937
938 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
939 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
940 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
941
942 return root;
943}
944
945int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
946 struct btrfs_root *root)
947{
948 struct extent_buffer *leaf;
949
950 /*
951 * DON'T set SHAREABLE bit for log trees.
952 *
953 * Log trees are not exposed to user space thus can't be snapshotted,
954 * and they go away before a real commit is actually done.
955 *
956 * They do store pointers to file data extents, and those reference
957 * counts still get updated (along with back refs to the log tree).
958 */
959
960 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
961 NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
962 if (IS_ERR(leaf))
963 return PTR_ERR(leaf);
964
965 root->node = leaf;
966
967 btrfs_mark_buffer_dirty(trans, root->node);
968 btrfs_tree_unlock(root->node);
969
970 return 0;
971}
972
973int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
974 struct btrfs_fs_info *fs_info)
975{
976 struct btrfs_root *log_root;
977
978 log_root = alloc_log_tree(trans, fs_info);
979 if (IS_ERR(log_root))
980 return PTR_ERR(log_root);
981
982 if (!btrfs_is_zoned(fs_info)) {
983 int ret = btrfs_alloc_log_tree_node(trans, log_root);
984
985 if (ret) {
986 btrfs_put_root(log_root);
987 return ret;
988 }
989 }
990
991 WARN_ON(fs_info->log_root_tree);
992 fs_info->log_root_tree = log_root;
993 return 0;
994}
995
996int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
997 struct btrfs_root *root)
998{
999 struct btrfs_fs_info *fs_info = root->fs_info;
1000 struct btrfs_root *log_root;
1001 struct btrfs_inode_item *inode_item;
1002 int ret;
1003
1004 log_root = alloc_log_tree(trans, fs_info);
1005 if (IS_ERR(log_root))
1006 return PTR_ERR(log_root);
1007
1008 ret = btrfs_alloc_log_tree_node(trans, log_root);
1009 if (ret) {
1010 btrfs_put_root(log_root);
1011 return ret;
1012 }
1013
1014 log_root->last_trans = trans->transid;
1015 log_root->root_key.offset = root->root_key.objectid;
1016
1017 inode_item = &log_root->root_item.inode;
1018 btrfs_set_stack_inode_generation(inode_item, 1);
1019 btrfs_set_stack_inode_size(inode_item, 3);
1020 btrfs_set_stack_inode_nlink(inode_item, 1);
1021 btrfs_set_stack_inode_nbytes(inode_item,
1022 fs_info->nodesize);
1023 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1024
1025 btrfs_set_root_node(&log_root->root_item, log_root->node);
1026
1027 WARN_ON(root->log_root);
1028 root->log_root = log_root;
1029 btrfs_set_root_log_transid(root, 0);
1030 root->log_transid_committed = -1;
1031 btrfs_set_root_last_log_commit(root, 0);
1032 return 0;
1033}
1034
1035static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1036 struct btrfs_path *path,
1037 struct btrfs_key *key)
1038{
1039 struct btrfs_root *root;
1040 struct btrfs_tree_parent_check check = { 0 };
1041 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1042 u64 generation;
1043 int ret;
1044 int level;
1045
1046 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1047 if (!root)
1048 return ERR_PTR(-ENOMEM);
1049
1050 ret = btrfs_find_root(tree_root, key, path,
1051 &root->root_item, &root->root_key);
1052 if (ret) {
1053 if (ret > 0)
1054 ret = -ENOENT;
1055 goto fail;
1056 }
1057
1058 generation = btrfs_root_generation(&root->root_item);
1059 level = btrfs_root_level(&root->root_item);
1060 check.level = level;
1061 check.transid = generation;
1062 check.owner_root = key->objectid;
1063 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1064 &check);
1065 if (IS_ERR(root->node)) {
1066 ret = PTR_ERR(root->node);
1067 root->node = NULL;
1068 goto fail;
1069 }
1070 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1071 ret = -EIO;
1072 goto fail;
1073 }
1074
1075 /*
1076 * For real fs, and not log/reloc trees, root owner must
1077 * match its root node owner
1078 */
1079 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1080 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1081 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1082 root->root_key.objectid != btrfs_header_owner(root->node)) {
1083 btrfs_crit(fs_info,
1084"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1085 root->root_key.objectid, root->node->start,
1086 btrfs_header_owner(root->node),
1087 root->root_key.objectid);
1088 ret = -EUCLEAN;
1089 goto fail;
1090 }
1091 root->commit_root = btrfs_root_node(root);
1092 return root;
1093fail:
1094 btrfs_put_root(root);
1095 return ERR_PTR(ret);
1096}
1097
1098struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1099 struct btrfs_key *key)
1100{
1101 struct btrfs_root *root;
1102 struct btrfs_path *path;
1103
1104 path = btrfs_alloc_path();
1105 if (!path)
1106 return ERR_PTR(-ENOMEM);
1107 root = read_tree_root_path(tree_root, path, key);
1108 btrfs_free_path(path);
1109
1110 return root;
1111}
1112
1113/*
1114 * Initialize subvolume root in-memory structure
1115 *
1116 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1117 */
1118static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1119{
1120 int ret;
1121
1122 btrfs_drew_lock_init(&root->snapshot_lock);
1123
1124 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1125 !btrfs_is_data_reloc_root(root) &&
1126 is_fstree(root->root_key.objectid)) {
1127 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1128 btrfs_check_and_init_root_item(&root->root_item);
1129 }
1130
1131 /*
1132 * Don't assign anonymous block device to roots that are not exposed to
1133 * userspace, the id pool is limited to 1M
1134 */
1135 if (is_fstree(root->root_key.objectid) &&
1136 btrfs_root_refs(&root->root_item) > 0) {
1137 if (!anon_dev) {
1138 ret = get_anon_bdev(&root->anon_dev);
1139 if (ret)
1140 goto fail;
1141 } else {
1142 root->anon_dev = anon_dev;
1143 }
1144 }
1145
1146 mutex_lock(&root->objectid_mutex);
1147 ret = btrfs_init_root_free_objectid(root);
1148 if (ret) {
1149 mutex_unlock(&root->objectid_mutex);
1150 goto fail;
1151 }
1152
1153 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1154
1155 mutex_unlock(&root->objectid_mutex);
1156
1157 return 0;
1158fail:
1159 /* The caller is responsible to call btrfs_free_fs_root */
1160 return ret;
1161}
1162
1163static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1164 u64 root_id)
1165{
1166 struct btrfs_root *root;
1167
1168 spin_lock(&fs_info->fs_roots_radix_lock);
1169 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170 (unsigned long)root_id);
1171 root = btrfs_grab_root(root);
1172 spin_unlock(&fs_info->fs_roots_radix_lock);
1173 return root;
1174}
1175
1176static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1177 u64 objectid)
1178{
1179 struct btrfs_key key = {
1180 .objectid = objectid,
1181 .type = BTRFS_ROOT_ITEM_KEY,
1182 .offset = 0,
1183 };
1184
1185 switch (objectid) {
1186 case BTRFS_ROOT_TREE_OBJECTID:
1187 return btrfs_grab_root(fs_info->tree_root);
1188 case BTRFS_EXTENT_TREE_OBJECTID:
1189 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1190 case BTRFS_CHUNK_TREE_OBJECTID:
1191 return btrfs_grab_root(fs_info->chunk_root);
1192 case BTRFS_DEV_TREE_OBJECTID:
1193 return btrfs_grab_root(fs_info->dev_root);
1194 case BTRFS_CSUM_TREE_OBJECTID:
1195 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1196 case BTRFS_QUOTA_TREE_OBJECTID:
1197 return btrfs_grab_root(fs_info->quota_root);
1198 case BTRFS_UUID_TREE_OBJECTID:
1199 return btrfs_grab_root(fs_info->uuid_root);
1200 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1201 return btrfs_grab_root(fs_info->block_group_root);
1202 case BTRFS_FREE_SPACE_TREE_OBJECTID:
1203 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1204 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1205 return btrfs_grab_root(fs_info->stripe_root);
1206 default:
1207 return NULL;
1208 }
1209}
1210
1211int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1212 struct btrfs_root *root)
1213{
1214 int ret;
1215
1216 ret = radix_tree_preload(GFP_NOFS);
1217 if (ret)
1218 return ret;
1219
1220 spin_lock(&fs_info->fs_roots_radix_lock);
1221 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222 (unsigned long)root->root_key.objectid,
1223 root);
1224 if (ret == 0) {
1225 btrfs_grab_root(root);
1226 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1227 }
1228 spin_unlock(&fs_info->fs_roots_radix_lock);
1229 radix_tree_preload_end();
1230
1231 return ret;
1232}
1233
1234void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1235{
1236#ifdef CONFIG_BTRFS_DEBUG
1237 struct btrfs_root *root;
1238
1239 while (!list_empty(&fs_info->allocated_roots)) {
1240 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1241
1242 root = list_first_entry(&fs_info->allocated_roots,
1243 struct btrfs_root, leak_list);
1244 btrfs_err(fs_info, "leaked root %s refcount %d",
1245 btrfs_root_name(&root->root_key, buf),
1246 refcount_read(&root->refs));
1247 WARN_ON_ONCE(1);
1248 while (refcount_read(&root->refs) > 1)
1249 btrfs_put_root(root);
1250 btrfs_put_root(root);
1251 }
1252#endif
1253}
1254
1255static void free_global_roots(struct btrfs_fs_info *fs_info)
1256{
1257 struct btrfs_root *root;
1258 struct rb_node *node;
1259
1260 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1261 root = rb_entry(node, struct btrfs_root, rb_node);
1262 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1263 btrfs_put_root(root);
1264 }
1265}
1266
1267void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1268{
1269 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1270 percpu_counter_destroy(&fs_info->delalloc_bytes);
1271 percpu_counter_destroy(&fs_info->ordered_bytes);
1272 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1273 btrfs_free_csum_hash(fs_info);
1274 btrfs_free_stripe_hash_table(fs_info);
1275 btrfs_free_ref_cache(fs_info);
1276 kfree(fs_info->balance_ctl);
1277 kfree(fs_info->delayed_root);
1278 free_global_roots(fs_info);
1279 btrfs_put_root(fs_info->tree_root);
1280 btrfs_put_root(fs_info->chunk_root);
1281 btrfs_put_root(fs_info->dev_root);
1282 btrfs_put_root(fs_info->quota_root);
1283 btrfs_put_root(fs_info->uuid_root);
1284 btrfs_put_root(fs_info->fs_root);
1285 btrfs_put_root(fs_info->data_reloc_root);
1286 btrfs_put_root(fs_info->block_group_root);
1287 btrfs_put_root(fs_info->stripe_root);
1288 btrfs_check_leaked_roots(fs_info);
1289 btrfs_extent_buffer_leak_debug_check(fs_info);
1290 kfree(fs_info->super_copy);
1291 kfree(fs_info->super_for_commit);
1292 kfree(fs_info->subpage_info);
1293 kvfree(fs_info);
1294}
1295
1296
1297/*
1298 * Get an in-memory reference of a root structure.
1299 *
1300 * For essential trees like root/extent tree, we grab it from fs_info directly.
1301 * For subvolume trees, we check the cached filesystem roots first. If not
1302 * found, then read it from disk and add it to cached fs roots.
1303 *
1304 * Caller should release the root by calling btrfs_put_root() after the usage.
1305 *
1306 * NOTE: Reloc and log trees can't be read by this function as they share the
1307 * same root objectid.
1308 *
1309 * @objectid: root id
1310 * @anon_dev: preallocated anonymous block device number for new roots,
1311 * pass NULL for a new allocation.
1312 * @check_ref: whether to check root item references, If true, return -ENOENT
1313 * for orphan roots
1314 */
1315static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1316 u64 objectid, dev_t *anon_dev,
1317 bool check_ref)
1318{
1319 struct btrfs_root *root;
1320 struct btrfs_path *path;
1321 struct btrfs_key key;
1322 int ret;
1323
1324 root = btrfs_get_global_root(fs_info, objectid);
1325 if (root)
1326 return root;
1327
1328 /*
1329 * If we're called for non-subvolume trees, and above function didn't
1330 * find one, do not try to read it from disk.
1331 *
1332 * This is namely for free-space-tree and quota tree, which can change
1333 * at runtime and should only be grabbed from fs_info.
1334 */
1335 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1336 return ERR_PTR(-ENOENT);
1337again:
1338 root = btrfs_lookup_fs_root(fs_info, objectid);
1339 if (root) {
1340 /*
1341 * Some other caller may have read out the newly inserted
1342 * subvolume already (for things like backref walk etc). Not
1343 * that common but still possible. In that case, we just need
1344 * to free the anon_dev.
1345 */
1346 if (unlikely(anon_dev && *anon_dev)) {
1347 free_anon_bdev(*anon_dev);
1348 *anon_dev = 0;
1349 }
1350
1351 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1352 btrfs_put_root(root);
1353 return ERR_PTR(-ENOENT);
1354 }
1355 return root;
1356 }
1357
1358 key.objectid = objectid;
1359 key.type = BTRFS_ROOT_ITEM_KEY;
1360 key.offset = (u64)-1;
1361 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1362 if (IS_ERR(root))
1363 return root;
1364
1365 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1366 ret = -ENOENT;
1367 goto fail;
1368 }
1369
1370 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1371 if (ret)
1372 goto fail;
1373
1374 path = btrfs_alloc_path();
1375 if (!path) {
1376 ret = -ENOMEM;
1377 goto fail;
1378 }
1379 key.objectid = BTRFS_ORPHAN_OBJECTID;
1380 key.type = BTRFS_ORPHAN_ITEM_KEY;
1381 key.offset = objectid;
1382
1383 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1384 btrfs_free_path(path);
1385 if (ret < 0)
1386 goto fail;
1387 if (ret == 0)
1388 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1389
1390 ret = btrfs_insert_fs_root(fs_info, root);
1391 if (ret) {
1392 if (ret == -EEXIST) {
1393 btrfs_put_root(root);
1394 goto again;
1395 }
1396 goto fail;
1397 }
1398 return root;
1399fail:
1400 /*
1401 * If our caller provided us an anonymous device, then it's his
1402 * responsibility to free it in case we fail. So we have to set our
1403 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1404 * and once again by our caller.
1405 */
1406 if (anon_dev && *anon_dev)
1407 root->anon_dev = 0;
1408 btrfs_put_root(root);
1409 return ERR_PTR(ret);
1410}
1411
1412/*
1413 * Get in-memory reference of a root structure
1414 *
1415 * @objectid: tree objectid
1416 * @check_ref: if set, verify that the tree exists and the item has at least
1417 * one reference
1418 */
1419struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1420 u64 objectid, bool check_ref)
1421{
1422 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1423}
1424
1425/*
1426 * Get in-memory reference of a root structure, created as new, optionally pass
1427 * the anonymous block device id
1428 *
1429 * @objectid: tree objectid
1430 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1431 * parameter value if not NULL
1432 */
1433struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1434 u64 objectid, dev_t *anon_dev)
1435{
1436 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1437}
1438
1439/*
1440 * Return a root for the given objectid.
1441 *
1442 * @fs_info: the fs_info
1443 * @objectid: the objectid we need to lookup
1444 *
1445 * This is exclusively used for backref walking, and exists specifically because
1446 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1447 * creation time, which means we may have to read the tree_root in order to look
1448 * up a fs root that is not in memory. If the root is not in memory we will
1449 * read the tree root commit root and look up the fs root from there. This is a
1450 * temporary root, it will not be inserted into the radix tree as it doesn't
1451 * have the most uptodate information, it'll simply be discarded once the
1452 * backref code is finished using the root.
1453 */
1454struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1455 struct btrfs_path *path,
1456 u64 objectid)
1457{
1458 struct btrfs_root *root;
1459 struct btrfs_key key;
1460
1461 ASSERT(path->search_commit_root && path->skip_locking);
1462
1463 /*
1464 * This can return -ENOENT if we ask for a root that doesn't exist, but
1465 * since this is called via the backref walking code we won't be looking
1466 * up a root that doesn't exist, unless there's corruption. So if root
1467 * != NULL just return it.
1468 */
1469 root = btrfs_get_global_root(fs_info, objectid);
1470 if (root)
1471 return root;
1472
1473 root = btrfs_lookup_fs_root(fs_info, objectid);
1474 if (root)
1475 return root;
1476
1477 key.objectid = objectid;
1478 key.type = BTRFS_ROOT_ITEM_KEY;
1479 key.offset = (u64)-1;
1480 root = read_tree_root_path(fs_info->tree_root, path, &key);
1481 btrfs_release_path(path);
1482
1483 return root;
1484}
1485
1486static int cleaner_kthread(void *arg)
1487{
1488 struct btrfs_fs_info *fs_info = arg;
1489 int again;
1490
1491 while (1) {
1492 again = 0;
1493
1494 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1495
1496 /* Make the cleaner go to sleep early. */
1497 if (btrfs_need_cleaner_sleep(fs_info))
1498 goto sleep;
1499
1500 /*
1501 * Do not do anything if we might cause open_ctree() to block
1502 * before we have finished mounting the filesystem.
1503 */
1504 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1505 goto sleep;
1506
1507 if (!mutex_trylock(&fs_info->cleaner_mutex))
1508 goto sleep;
1509
1510 /*
1511 * Avoid the problem that we change the status of the fs
1512 * during the above check and trylock.
1513 */
1514 if (btrfs_need_cleaner_sleep(fs_info)) {
1515 mutex_unlock(&fs_info->cleaner_mutex);
1516 goto sleep;
1517 }
1518
1519 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1520 btrfs_sysfs_feature_update(fs_info);
1521
1522 btrfs_run_delayed_iputs(fs_info);
1523
1524 again = btrfs_clean_one_deleted_snapshot(fs_info);
1525 mutex_unlock(&fs_info->cleaner_mutex);
1526
1527 /*
1528 * The defragger has dealt with the R/O remount and umount,
1529 * needn't do anything special here.
1530 */
1531 btrfs_run_defrag_inodes(fs_info);
1532
1533 /*
1534 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1535 * with relocation (btrfs_relocate_chunk) and relocation
1536 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1537 * after acquiring fs_info->reclaim_bgs_lock. So we
1538 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1539 * unused block groups.
1540 */
1541 btrfs_delete_unused_bgs(fs_info);
1542
1543 /*
1544 * Reclaim block groups in the reclaim_bgs list after we deleted
1545 * all unused block_groups. This possibly gives us some more free
1546 * space.
1547 */
1548 btrfs_reclaim_bgs(fs_info);
1549sleep:
1550 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1551 if (kthread_should_park())
1552 kthread_parkme();
1553 if (kthread_should_stop())
1554 return 0;
1555 if (!again) {
1556 set_current_state(TASK_INTERRUPTIBLE);
1557 schedule();
1558 __set_current_state(TASK_RUNNING);
1559 }
1560 }
1561}
1562
1563static int transaction_kthread(void *arg)
1564{
1565 struct btrfs_root *root = arg;
1566 struct btrfs_fs_info *fs_info = root->fs_info;
1567 struct btrfs_trans_handle *trans;
1568 struct btrfs_transaction *cur;
1569 u64 transid;
1570 time64_t delta;
1571 unsigned long delay;
1572 bool cannot_commit;
1573
1574 do {
1575 cannot_commit = false;
1576 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1577 mutex_lock(&fs_info->transaction_kthread_mutex);
1578
1579 spin_lock(&fs_info->trans_lock);
1580 cur = fs_info->running_transaction;
1581 if (!cur) {
1582 spin_unlock(&fs_info->trans_lock);
1583 goto sleep;
1584 }
1585
1586 delta = ktime_get_seconds() - cur->start_time;
1587 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1588 cur->state < TRANS_STATE_COMMIT_PREP &&
1589 delta < fs_info->commit_interval) {
1590 spin_unlock(&fs_info->trans_lock);
1591 delay -= msecs_to_jiffies((delta - 1) * 1000);
1592 delay = min(delay,
1593 msecs_to_jiffies(fs_info->commit_interval * 1000));
1594 goto sleep;
1595 }
1596 transid = cur->transid;
1597 spin_unlock(&fs_info->trans_lock);
1598
1599 /* If the file system is aborted, this will always fail. */
1600 trans = btrfs_attach_transaction(root);
1601 if (IS_ERR(trans)) {
1602 if (PTR_ERR(trans) != -ENOENT)
1603 cannot_commit = true;
1604 goto sleep;
1605 }
1606 if (transid == trans->transid) {
1607 btrfs_commit_transaction(trans);
1608 } else {
1609 btrfs_end_transaction(trans);
1610 }
1611sleep:
1612 wake_up_process(fs_info->cleaner_kthread);
1613 mutex_unlock(&fs_info->transaction_kthread_mutex);
1614
1615 if (BTRFS_FS_ERROR(fs_info))
1616 btrfs_cleanup_transaction(fs_info);
1617 if (!kthread_should_stop() &&
1618 (!btrfs_transaction_blocked(fs_info) ||
1619 cannot_commit))
1620 schedule_timeout_interruptible(delay);
1621 } while (!kthread_should_stop());
1622 return 0;
1623}
1624
1625/*
1626 * This will find the highest generation in the array of root backups. The
1627 * index of the highest array is returned, or -EINVAL if we can't find
1628 * anything.
1629 *
1630 * We check to make sure the array is valid by comparing the
1631 * generation of the latest root in the array with the generation
1632 * in the super block. If they don't match we pitch it.
1633 */
1634static int find_newest_super_backup(struct btrfs_fs_info *info)
1635{
1636 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1637 u64 cur;
1638 struct btrfs_root_backup *root_backup;
1639 int i;
1640
1641 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1642 root_backup = info->super_copy->super_roots + i;
1643 cur = btrfs_backup_tree_root_gen(root_backup);
1644 if (cur == newest_gen)
1645 return i;
1646 }
1647
1648 return -EINVAL;
1649}
1650
1651/*
1652 * copy all the root pointers into the super backup array.
1653 * this will bump the backup pointer by one when it is
1654 * done
1655 */
1656static void backup_super_roots(struct btrfs_fs_info *info)
1657{
1658 const int next_backup = info->backup_root_index;
1659 struct btrfs_root_backup *root_backup;
1660
1661 root_backup = info->super_for_commit->super_roots + next_backup;
1662
1663 /*
1664 * make sure all of our padding and empty slots get zero filled
1665 * regardless of which ones we use today
1666 */
1667 memset(root_backup, 0, sizeof(*root_backup));
1668
1669 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1670
1671 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1672 btrfs_set_backup_tree_root_gen(root_backup,
1673 btrfs_header_generation(info->tree_root->node));
1674
1675 btrfs_set_backup_tree_root_level(root_backup,
1676 btrfs_header_level(info->tree_root->node));
1677
1678 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1679 btrfs_set_backup_chunk_root_gen(root_backup,
1680 btrfs_header_generation(info->chunk_root->node));
1681 btrfs_set_backup_chunk_root_level(root_backup,
1682 btrfs_header_level(info->chunk_root->node));
1683
1684 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1685 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1686 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1687
1688 btrfs_set_backup_extent_root(root_backup,
1689 extent_root->node->start);
1690 btrfs_set_backup_extent_root_gen(root_backup,
1691 btrfs_header_generation(extent_root->node));
1692 btrfs_set_backup_extent_root_level(root_backup,
1693 btrfs_header_level(extent_root->node));
1694
1695 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1696 btrfs_set_backup_csum_root_gen(root_backup,
1697 btrfs_header_generation(csum_root->node));
1698 btrfs_set_backup_csum_root_level(root_backup,
1699 btrfs_header_level(csum_root->node));
1700 }
1701
1702 /*
1703 * we might commit during log recovery, which happens before we set
1704 * the fs_root. Make sure it is valid before we fill it in.
1705 */
1706 if (info->fs_root && info->fs_root->node) {
1707 btrfs_set_backup_fs_root(root_backup,
1708 info->fs_root->node->start);
1709 btrfs_set_backup_fs_root_gen(root_backup,
1710 btrfs_header_generation(info->fs_root->node));
1711 btrfs_set_backup_fs_root_level(root_backup,
1712 btrfs_header_level(info->fs_root->node));
1713 }
1714
1715 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1716 btrfs_set_backup_dev_root_gen(root_backup,
1717 btrfs_header_generation(info->dev_root->node));
1718 btrfs_set_backup_dev_root_level(root_backup,
1719 btrfs_header_level(info->dev_root->node));
1720
1721 btrfs_set_backup_total_bytes(root_backup,
1722 btrfs_super_total_bytes(info->super_copy));
1723 btrfs_set_backup_bytes_used(root_backup,
1724 btrfs_super_bytes_used(info->super_copy));
1725 btrfs_set_backup_num_devices(root_backup,
1726 btrfs_super_num_devices(info->super_copy));
1727
1728 /*
1729 * if we don't copy this out to the super_copy, it won't get remembered
1730 * for the next commit
1731 */
1732 memcpy(&info->super_copy->super_roots,
1733 &info->super_for_commit->super_roots,
1734 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1735}
1736
1737/*
1738 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1739 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1740 *
1741 * @fs_info: filesystem whose backup roots need to be read
1742 * @priority: priority of backup root required
1743 *
1744 * Returns backup root index on success and -EINVAL otherwise.
1745 */
1746static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1747{
1748 int backup_index = find_newest_super_backup(fs_info);
1749 struct btrfs_super_block *super = fs_info->super_copy;
1750 struct btrfs_root_backup *root_backup;
1751
1752 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1753 if (priority == 0)
1754 return backup_index;
1755
1756 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1757 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1758 } else {
1759 return -EINVAL;
1760 }
1761
1762 root_backup = super->super_roots + backup_index;
1763
1764 btrfs_set_super_generation(super,
1765 btrfs_backup_tree_root_gen(root_backup));
1766 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1767 btrfs_set_super_root_level(super,
1768 btrfs_backup_tree_root_level(root_backup));
1769 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1770
1771 /*
1772 * Fixme: the total bytes and num_devices need to match or we should
1773 * need a fsck
1774 */
1775 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1776 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1777
1778 return backup_index;
1779}
1780
1781/* helper to cleanup workers */
1782static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1783{
1784 btrfs_destroy_workqueue(fs_info->fixup_workers);
1785 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1786 btrfs_destroy_workqueue(fs_info->workers);
1787 if (fs_info->endio_workers)
1788 destroy_workqueue(fs_info->endio_workers);
1789 if (fs_info->rmw_workers)
1790 destroy_workqueue(fs_info->rmw_workers);
1791 if (fs_info->compressed_write_workers)
1792 destroy_workqueue(fs_info->compressed_write_workers);
1793 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1794 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1795 btrfs_destroy_workqueue(fs_info->delayed_workers);
1796 btrfs_destroy_workqueue(fs_info->caching_workers);
1797 btrfs_destroy_workqueue(fs_info->flush_workers);
1798 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1799 if (fs_info->discard_ctl.discard_workers)
1800 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1801 /*
1802 * Now that all other work queues are destroyed, we can safely destroy
1803 * the queues used for metadata I/O, since tasks from those other work
1804 * queues can do metadata I/O operations.
1805 */
1806 if (fs_info->endio_meta_workers)
1807 destroy_workqueue(fs_info->endio_meta_workers);
1808}
1809
1810static void free_root_extent_buffers(struct btrfs_root *root)
1811{
1812 if (root) {
1813 free_extent_buffer(root->node);
1814 free_extent_buffer(root->commit_root);
1815 root->node = NULL;
1816 root->commit_root = NULL;
1817 }
1818}
1819
1820static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1821{
1822 struct btrfs_root *root, *tmp;
1823
1824 rbtree_postorder_for_each_entry_safe(root, tmp,
1825 &fs_info->global_root_tree,
1826 rb_node)
1827 free_root_extent_buffers(root);
1828}
1829
1830/* helper to cleanup tree roots */
1831static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1832{
1833 free_root_extent_buffers(info->tree_root);
1834
1835 free_global_root_pointers(info);
1836 free_root_extent_buffers(info->dev_root);
1837 free_root_extent_buffers(info->quota_root);
1838 free_root_extent_buffers(info->uuid_root);
1839 free_root_extent_buffers(info->fs_root);
1840 free_root_extent_buffers(info->data_reloc_root);
1841 free_root_extent_buffers(info->block_group_root);
1842 free_root_extent_buffers(info->stripe_root);
1843 if (free_chunk_root)
1844 free_root_extent_buffers(info->chunk_root);
1845}
1846
1847void btrfs_put_root(struct btrfs_root *root)
1848{
1849 if (!root)
1850 return;
1851
1852 if (refcount_dec_and_test(&root->refs)) {
1853 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1854 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1855 if (root->anon_dev)
1856 free_anon_bdev(root->anon_dev);
1857 free_root_extent_buffers(root);
1858#ifdef CONFIG_BTRFS_DEBUG
1859 spin_lock(&root->fs_info->fs_roots_radix_lock);
1860 list_del_init(&root->leak_list);
1861 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1862#endif
1863 kfree(root);
1864 }
1865}
1866
1867void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1868{
1869 int ret;
1870 struct btrfs_root *gang[8];
1871 int i;
1872
1873 while (!list_empty(&fs_info->dead_roots)) {
1874 gang[0] = list_entry(fs_info->dead_roots.next,
1875 struct btrfs_root, root_list);
1876 list_del(&gang[0]->root_list);
1877
1878 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1879 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1880 btrfs_put_root(gang[0]);
1881 }
1882
1883 while (1) {
1884 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1885 (void **)gang, 0,
1886 ARRAY_SIZE(gang));
1887 if (!ret)
1888 break;
1889 for (i = 0; i < ret; i++)
1890 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1891 }
1892}
1893
1894static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1895{
1896 mutex_init(&fs_info->scrub_lock);
1897 atomic_set(&fs_info->scrubs_running, 0);
1898 atomic_set(&fs_info->scrub_pause_req, 0);
1899 atomic_set(&fs_info->scrubs_paused, 0);
1900 atomic_set(&fs_info->scrub_cancel_req, 0);
1901 init_waitqueue_head(&fs_info->scrub_pause_wait);
1902 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1903}
1904
1905static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1906{
1907 spin_lock_init(&fs_info->balance_lock);
1908 mutex_init(&fs_info->balance_mutex);
1909 atomic_set(&fs_info->balance_pause_req, 0);
1910 atomic_set(&fs_info->balance_cancel_req, 0);
1911 fs_info->balance_ctl = NULL;
1912 init_waitqueue_head(&fs_info->balance_wait_q);
1913 atomic_set(&fs_info->reloc_cancel_req, 0);
1914}
1915
1916static int btrfs_init_btree_inode(struct super_block *sb)
1917{
1918 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1919 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1920 fs_info->tree_root);
1921 struct inode *inode;
1922
1923 inode = new_inode(sb);
1924 if (!inode)
1925 return -ENOMEM;
1926
1927 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1928 set_nlink(inode, 1);
1929 /*
1930 * we set the i_size on the btree inode to the max possible int.
1931 * the real end of the address space is determined by all of
1932 * the devices in the system
1933 */
1934 inode->i_size = OFFSET_MAX;
1935 inode->i_mapping->a_ops = &btree_aops;
1936 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1937
1938 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1939 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1940 IO_TREE_BTREE_INODE_IO);
1941 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1942
1943 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1944 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1945 BTRFS_I(inode)->location.type = 0;
1946 BTRFS_I(inode)->location.offset = 0;
1947 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1948 __insert_inode_hash(inode, hash);
1949 fs_info->btree_inode = inode;
1950
1951 return 0;
1952}
1953
1954static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1955{
1956 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1957 init_rwsem(&fs_info->dev_replace.rwsem);
1958 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1959}
1960
1961static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1962{
1963 spin_lock_init(&fs_info->qgroup_lock);
1964 mutex_init(&fs_info->qgroup_ioctl_lock);
1965 fs_info->qgroup_tree = RB_ROOT;
1966 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1967 fs_info->qgroup_seq = 1;
1968 fs_info->qgroup_ulist = NULL;
1969 fs_info->qgroup_rescan_running = false;
1970 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1971 mutex_init(&fs_info->qgroup_rescan_lock);
1972}
1973
1974static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1975{
1976 u32 max_active = fs_info->thread_pool_size;
1977 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1978 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1979
1980 fs_info->workers =
1981 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1982
1983 fs_info->delalloc_workers =
1984 btrfs_alloc_workqueue(fs_info, "delalloc",
1985 flags, max_active, 2);
1986
1987 fs_info->flush_workers =
1988 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1989 flags, max_active, 0);
1990
1991 fs_info->caching_workers =
1992 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1993
1994 fs_info->fixup_workers =
1995 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1996
1997 fs_info->endio_workers =
1998 alloc_workqueue("btrfs-endio", flags, max_active);
1999 fs_info->endio_meta_workers =
2000 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2001 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2002 fs_info->endio_write_workers =
2003 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2004 max_active, 2);
2005 fs_info->compressed_write_workers =
2006 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2007 fs_info->endio_freespace_worker =
2008 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2009 max_active, 0);
2010 fs_info->delayed_workers =
2011 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2012 max_active, 0);
2013 fs_info->qgroup_rescan_workers =
2014 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2015 ordered_flags);
2016 fs_info->discard_ctl.discard_workers =
2017 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2018
2019 if (!(fs_info->workers &&
2020 fs_info->delalloc_workers && fs_info->flush_workers &&
2021 fs_info->endio_workers && fs_info->endio_meta_workers &&
2022 fs_info->compressed_write_workers &&
2023 fs_info->endio_write_workers &&
2024 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2025 fs_info->caching_workers && fs_info->fixup_workers &&
2026 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2027 fs_info->discard_ctl.discard_workers)) {
2028 return -ENOMEM;
2029 }
2030
2031 return 0;
2032}
2033
2034static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2035{
2036 struct crypto_shash *csum_shash;
2037 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2038
2039 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2040
2041 if (IS_ERR(csum_shash)) {
2042 btrfs_err(fs_info, "error allocating %s hash for checksum",
2043 csum_driver);
2044 return PTR_ERR(csum_shash);
2045 }
2046
2047 fs_info->csum_shash = csum_shash;
2048
2049 /*
2050 * Check if the checksum implementation is a fast accelerated one.
2051 * As-is this is a bit of a hack and should be replaced once the csum
2052 * implementations provide that information themselves.
2053 */
2054 switch (csum_type) {
2055 case BTRFS_CSUM_TYPE_CRC32:
2056 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2057 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2058 break;
2059 case BTRFS_CSUM_TYPE_XXHASH:
2060 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2061 break;
2062 default:
2063 break;
2064 }
2065
2066 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2067 btrfs_super_csum_name(csum_type),
2068 crypto_shash_driver_name(csum_shash));
2069 return 0;
2070}
2071
2072static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2073 struct btrfs_fs_devices *fs_devices)
2074{
2075 int ret;
2076 struct btrfs_tree_parent_check check = { 0 };
2077 struct btrfs_root *log_tree_root;
2078 struct btrfs_super_block *disk_super = fs_info->super_copy;
2079 u64 bytenr = btrfs_super_log_root(disk_super);
2080 int level = btrfs_super_log_root_level(disk_super);
2081
2082 if (fs_devices->rw_devices == 0) {
2083 btrfs_warn(fs_info, "log replay required on RO media");
2084 return -EIO;
2085 }
2086
2087 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2088 GFP_KERNEL);
2089 if (!log_tree_root)
2090 return -ENOMEM;
2091
2092 check.level = level;
2093 check.transid = fs_info->generation + 1;
2094 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2095 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2096 if (IS_ERR(log_tree_root->node)) {
2097 btrfs_warn(fs_info, "failed to read log tree");
2098 ret = PTR_ERR(log_tree_root->node);
2099 log_tree_root->node = NULL;
2100 btrfs_put_root(log_tree_root);
2101 return ret;
2102 }
2103 if (!extent_buffer_uptodate(log_tree_root->node)) {
2104 btrfs_err(fs_info, "failed to read log tree");
2105 btrfs_put_root(log_tree_root);
2106 return -EIO;
2107 }
2108
2109 /* returns with log_tree_root freed on success */
2110 ret = btrfs_recover_log_trees(log_tree_root);
2111 if (ret) {
2112 btrfs_handle_fs_error(fs_info, ret,
2113 "Failed to recover log tree");
2114 btrfs_put_root(log_tree_root);
2115 return ret;
2116 }
2117
2118 if (sb_rdonly(fs_info->sb)) {
2119 ret = btrfs_commit_super(fs_info);
2120 if (ret)
2121 return ret;
2122 }
2123
2124 return 0;
2125}
2126
2127static int load_global_roots_objectid(struct btrfs_root *tree_root,
2128 struct btrfs_path *path, u64 objectid,
2129 const char *name)
2130{
2131 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2132 struct btrfs_root *root;
2133 u64 max_global_id = 0;
2134 int ret;
2135 struct btrfs_key key = {
2136 .objectid = objectid,
2137 .type = BTRFS_ROOT_ITEM_KEY,
2138 .offset = 0,
2139 };
2140 bool found = false;
2141
2142 /* If we have IGNOREDATACSUMS skip loading these roots. */
2143 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2144 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2145 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2146 return 0;
2147 }
2148
2149 while (1) {
2150 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2151 if (ret < 0)
2152 break;
2153
2154 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2155 ret = btrfs_next_leaf(tree_root, path);
2156 if (ret) {
2157 if (ret > 0)
2158 ret = 0;
2159 break;
2160 }
2161 }
2162 ret = 0;
2163
2164 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2165 if (key.objectid != objectid)
2166 break;
2167 btrfs_release_path(path);
2168
2169 /*
2170 * Just worry about this for extent tree, it'll be the same for
2171 * everybody.
2172 */
2173 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2174 max_global_id = max(max_global_id, key.offset);
2175
2176 found = true;
2177 root = read_tree_root_path(tree_root, path, &key);
2178 if (IS_ERR(root)) {
2179 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2180 ret = PTR_ERR(root);
2181 break;
2182 }
2183 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2184 ret = btrfs_global_root_insert(root);
2185 if (ret) {
2186 btrfs_put_root(root);
2187 break;
2188 }
2189 key.offset++;
2190 }
2191 btrfs_release_path(path);
2192
2193 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2194 fs_info->nr_global_roots = max_global_id + 1;
2195
2196 if (!found || ret) {
2197 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2198 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2199
2200 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2201 ret = ret ? ret : -ENOENT;
2202 else
2203 ret = 0;
2204 btrfs_err(fs_info, "failed to load root %s", name);
2205 }
2206 return ret;
2207}
2208
2209static int load_global_roots(struct btrfs_root *tree_root)
2210{
2211 struct btrfs_path *path;
2212 int ret = 0;
2213
2214 path = btrfs_alloc_path();
2215 if (!path)
2216 return -ENOMEM;
2217
2218 ret = load_global_roots_objectid(tree_root, path,
2219 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2220 if (ret)
2221 goto out;
2222 ret = load_global_roots_objectid(tree_root, path,
2223 BTRFS_CSUM_TREE_OBJECTID, "csum");
2224 if (ret)
2225 goto out;
2226 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2227 goto out;
2228 ret = load_global_roots_objectid(tree_root, path,
2229 BTRFS_FREE_SPACE_TREE_OBJECTID,
2230 "free space");
2231out:
2232 btrfs_free_path(path);
2233 return ret;
2234}
2235
2236static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2237{
2238 struct btrfs_root *tree_root = fs_info->tree_root;
2239 struct btrfs_root *root;
2240 struct btrfs_key location;
2241 int ret;
2242
2243 ASSERT(fs_info->tree_root);
2244
2245 ret = load_global_roots(tree_root);
2246 if (ret)
2247 return ret;
2248
2249 location.type = BTRFS_ROOT_ITEM_KEY;
2250 location.offset = 0;
2251
2252 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2253 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2254 root = btrfs_read_tree_root(tree_root, &location);
2255 if (IS_ERR(root)) {
2256 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2257 ret = PTR_ERR(root);
2258 goto out;
2259 }
2260 } else {
2261 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2262 fs_info->block_group_root = root;
2263 }
2264 }
2265
2266 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2267 root = btrfs_read_tree_root(tree_root, &location);
2268 if (IS_ERR(root)) {
2269 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2270 ret = PTR_ERR(root);
2271 goto out;
2272 }
2273 } else {
2274 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2275 fs_info->dev_root = root;
2276 }
2277 /* Initialize fs_info for all devices in any case */
2278 ret = btrfs_init_devices_late(fs_info);
2279 if (ret)
2280 goto out;
2281
2282 /*
2283 * This tree can share blocks with some other fs tree during relocation
2284 * and we need a proper setup by btrfs_get_fs_root
2285 */
2286 root = btrfs_get_fs_root(tree_root->fs_info,
2287 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2288 if (IS_ERR(root)) {
2289 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2290 ret = PTR_ERR(root);
2291 goto out;
2292 }
2293 } else {
2294 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2295 fs_info->data_reloc_root = root;
2296 }
2297
2298 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2299 root = btrfs_read_tree_root(tree_root, &location);
2300 if (!IS_ERR(root)) {
2301 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2302 fs_info->quota_root = root;
2303 }
2304
2305 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2306 root = btrfs_read_tree_root(tree_root, &location);
2307 if (IS_ERR(root)) {
2308 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2309 ret = PTR_ERR(root);
2310 if (ret != -ENOENT)
2311 goto out;
2312 }
2313 } else {
2314 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2315 fs_info->uuid_root = root;
2316 }
2317
2318 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2319 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2320 root = btrfs_read_tree_root(tree_root, &location);
2321 if (IS_ERR(root)) {
2322 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2323 ret = PTR_ERR(root);
2324 goto out;
2325 }
2326 } else {
2327 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2328 fs_info->stripe_root = root;
2329 }
2330 }
2331
2332 return 0;
2333out:
2334 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2335 location.objectid, ret);
2336 return ret;
2337}
2338
2339/*
2340 * Real super block validation
2341 * NOTE: super csum type and incompat features will not be checked here.
2342 *
2343 * @sb: super block to check
2344 * @mirror_num: the super block number to check its bytenr:
2345 * 0 the primary (1st) sb
2346 * 1, 2 2nd and 3rd backup copy
2347 * -1 skip bytenr check
2348 */
2349int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2350 struct btrfs_super_block *sb, int mirror_num)
2351{
2352 u64 nodesize = btrfs_super_nodesize(sb);
2353 u64 sectorsize = btrfs_super_sectorsize(sb);
2354 int ret = 0;
2355
2356 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2357 btrfs_err(fs_info, "no valid FS found");
2358 ret = -EINVAL;
2359 }
2360 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2361 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2362 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2363 ret = -EINVAL;
2364 }
2365 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2366 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2367 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2368 ret = -EINVAL;
2369 }
2370 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2371 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2372 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2373 ret = -EINVAL;
2374 }
2375 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2376 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2377 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2378 ret = -EINVAL;
2379 }
2380
2381 /*
2382 * Check sectorsize and nodesize first, other check will need it.
2383 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2384 */
2385 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2386 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2387 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2388 ret = -EINVAL;
2389 }
2390
2391 /*
2392 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2393 *
2394 * We can support 16K sectorsize with 64K page size without problem,
2395 * but such sectorsize/pagesize combination doesn't make much sense.
2396 * 4K will be our future standard, PAGE_SIZE is supported from the very
2397 * beginning.
2398 */
2399 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2400 btrfs_err(fs_info,
2401 "sectorsize %llu not yet supported for page size %lu",
2402 sectorsize, PAGE_SIZE);
2403 ret = -EINVAL;
2404 }
2405
2406 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2407 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2408 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2409 ret = -EINVAL;
2410 }
2411 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2412 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2413 le32_to_cpu(sb->__unused_leafsize), nodesize);
2414 ret = -EINVAL;
2415 }
2416
2417 /* Root alignment check */
2418 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2419 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2420 btrfs_super_root(sb));
2421 ret = -EINVAL;
2422 }
2423 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2424 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2425 btrfs_super_chunk_root(sb));
2426 ret = -EINVAL;
2427 }
2428 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2429 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2430 btrfs_super_log_root(sb));
2431 ret = -EINVAL;
2432 }
2433
2434 if (!fs_info->fs_devices->temp_fsid &&
2435 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2436 btrfs_err(fs_info,
2437 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2438 sb->fsid, fs_info->fs_devices->fsid);
2439 ret = -EINVAL;
2440 }
2441
2442 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2443 BTRFS_FSID_SIZE) != 0) {
2444 btrfs_err(fs_info,
2445"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2446 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2447 ret = -EINVAL;
2448 }
2449
2450 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2451 BTRFS_FSID_SIZE) != 0) {
2452 btrfs_err(fs_info,
2453 "dev_item UUID does not match metadata fsid: %pU != %pU",
2454 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2455 ret = -EINVAL;
2456 }
2457
2458 /*
2459 * Artificial requirement for block-group-tree to force newer features
2460 * (free-space-tree, no-holes) so the test matrix is smaller.
2461 */
2462 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2463 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2464 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2465 btrfs_err(fs_info,
2466 "block-group-tree feature requires fres-space-tree and no-holes");
2467 ret = -EINVAL;
2468 }
2469
2470 /*
2471 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2472 * done later
2473 */
2474 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2475 btrfs_err(fs_info, "bytes_used is too small %llu",
2476 btrfs_super_bytes_used(sb));
2477 ret = -EINVAL;
2478 }
2479 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2480 btrfs_err(fs_info, "invalid stripesize %u",
2481 btrfs_super_stripesize(sb));
2482 ret = -EINVAL;
2483 }
2484 if (btrfs_super_num_devices(sb) > (1UL << 31))
2485 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2486 btrfs_super_num_devices(sb));
2487 if (btrfs_super_num_devices(sb) == 0) {
2488 btrfs_err(fs_info, "number of devices is 0");
2489 ret = -EINVAL;
2490 }
2491
2492 if (mirror_num >= 0 &&
2493 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2494 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2495 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2496 ret = -EINVAL;
2497 }
2498
2499 /*
2500 * Obvious sys_chunk_array corruptions, it must hold at least one key
2501 * and one chunk
2502 */
2503 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2504 btrfs_err(fs_info, "system chunk array too big %u > %u",
2505 btrfs_super_sys_array_size(sb),
2506 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2507 ret = -EINVAL;
2508 }
2509 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2510 + sizeof(struct btrfs_chunk)) {
2511 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2512 btrfs_super_sys_array_size(sb),
2513 sizeof(struct btrfs_disk_key)
2514 + sizeof(struct btrfs_chunk));
2515 ret = -EINVAL;
2516 }
2517
2518 /*
2519 * The generation is a global counter, we'll trust it more than the others
2520 * but it's still possible that it's the one that's wrong.
2521 */
2522 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2523 btrfs_warn(fs_info,
2524 "suspicious: generation < chunk_root_generation: %llu < %llu",
2525 btrfs_super_generation(sb),
2526 btrfs_super_chunk_root_generation(sb));
2527 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2528 && btrfs_super_cache_generation(sb) != (u64)-1)
2529 btrfs_warn(fs_info,
2530 "suspicious: generation < cache_generation: %llu < %llu",
2531 btrfs_super_generation(sb),
2532 btrfs_super_cache_generation(sb));
2533
2534 return ret;
2535}
2536
2537/*
2538 * Validation of super block at mount time.
2539 * Some checks already done early at mount time, like csum type and incompat
2540 * flags will be skipped.
2541 */
2542static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2543{
2544 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2545}
2546
2547/*
2548 * Validation of super block at write time.
2549 * Some checks like bytenr check will be skipped as their values will be
2550 * overwritten soon.
2551 * Extra checks like csum type and incompat flags will be done here.
2552 */
2553static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2554 struct btrfs_super_block *sb)
2555{
2556 int ret;
2557
2558 ret = btrfs_validate_super(fs_info, sb, -1);
2559 if (ret < 0)
2560 goto out;
2561 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2562 ret = -EUCLEAN;
2563 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2564 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2565 goto out;
2566 }
2567 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2568 ret = -EUCLEAN;
2569 btrfs_err(fs_info,
2570 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2571 btrfs_super_incompat_flags(sb),
2572 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2573 goto out;
2574 }
2575out:
2576 if (ret < 0)
2577 btrfs_err(fs_info,
2578 "super block corruption detected before writing it to disk");
2579 return ret;
2580}
2581
2582static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2583{
2584 struct btrfs_tree_parent_check check = {
2585 .level = level,
2586 .transid = gen,
2587 .owner_root = root->root_key.objectid
2588 };
2589 int ret = 0;
2590
2591 root->node = read_tree_block(root->fs_info, bytenr, &check);
2592 if (IS_ERR(root->node)) {
2593 ret = PTR_ERR(root->node);
2594 root->node = NULL;
2595 return ret;
2596 }
2597 if (!extent_buffer_uptodate(root->node)) {
2598 free_extent_buffer(root->node);
2599 root->node = NULL;
2600 return -EIO;
2601 }
2602
2603 btrfs_set_root_node(&root->root_item, root->node);
2604 root->commit_root = btrfs_root_node(root);
2605 btrfs_set_root_refs(&root->root_item, 1);
2606 return ret;
2607}
2608
2609static int load_important_roots(struct btrfs_fs_info *fs_info)
2610{
2611 struct btrfs_super_block *sb = fs_info->super_copy;
2612 u64 gen, bytenr;
2613 int level, ret;
2614
2615 bytenr = btrfs_super_root(sb);
2616 gen = btrfs_super_generation(sb);
2617 level = btrfs_super_root_level(sb);
2618 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2619 if (ret) {
2620 btrfs_warn(fs_info, "couldn't read tree root");
2621 return ret;
2622 }
2623 return 0;
2624}
2625
2626static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2627{
2628 int backup_index = find_newest_super_backup(fs_info);
2629 struct btrfs_super_block *sb = fs_info->super_copy;
2630 struct btrfs_root *tree_root = fs_info->tree_root;
2631 bool handle_error = false;
2632 int ret = 0;
2633 int i;
2634
2635 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2636 if (handle_error) {
2637 if (!IS_ERR(tree_root->node))
2638 free_extent_buffer(tree_root->node);
2639 tree_root->node = NULL;
2640
2641 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2642 break;
2643
2644 free_root_pointers(fs_info, 0);
2645
2646 /*
2647 * Don't use the log in recovery mode, it won't be
2648 * valid
2649 */
2650 btrfs_set_super_log_root(sb, 0);
2651
2652 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2653 ret = read_backup_root(fs_info, i);
2654 backup_index = ret;
2655 if (ret < 0)
2656 return ret;
2657 }
2658
2659 ret = load_important_roots(fs_info);
2660 if (ret) {
2661 handle_error = true;
2662 continue;
2663 }
2664
2665 /*
2666 * No need to hold btrfs_root::objectid_mutex since the fs
2667 * hasn't been fully initialised and we are the only user
2668 */
2669 ret = btrfs_init_root_free_objectid(tree_root);
2670 if (ret < 0) {
2671 handle_error = true;
2672 continue;
2673 }
2674
2675 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2676
2677 ret = btrfs_read_roots(fs_info);
2678 if (ret < 0) {
2679 handle_error = true;
2680 continue;
2681 }
2682
2683 /* All successful */
2684 fs_info->generation = btrfs_header_generation(tree_root->node);
2685 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2686 fs_info->last_reloc_trans = 0;
2687
2688 /* Always begin writing backup roots after the one being used */
2689 if (backup_index < 0) {
2690 fs_info->backup_root_index = 0;
2691 } else {
2692 fs_info->backup_root_index = backup_index + 1;
2693 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2694 }
2695 break;
2696 }
2697
2698 return ret;
2699}
2700
2701void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2702{
2703 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2704 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2705 INIT_LIST_HEAD(&fs_info->trans_list);
2706 INIT_LIST_HEAD(&fs_info->dead_roots);
2707 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2708 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2709 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2710 spin_lock_init(&fs_info->delalloc_root_lock);
2711 spin_lock_init(&fs_info->trans_lock);
2712 spin_lock_init(&fs_info->fs_roots_radix_lock);
2713 spin_lock_init(&fs_info->delayed_iput_lock);
2714 spin_lock_init(&fs_info->defrag_inodes_lock);
2715 spin_lock_init(&fs_info->super_lock);
2716 spin_lock_init(&fs_info->buffer_lock);
2717 spin_lock_init(&fs_info->unused_bgs_lock);
2718 spin_lock_init(&fs_info->treelog_bg_lock);
2719 spin_lock_init(&fs_info->zone_active_bgs_lock);
2720 spin_lock_init(&fs_info->relocation_bg_lock);
2721 rwlock_init(&fs_info->tree_mod_log_lock);
2722 rwlock_init(&fs_info->global_root_lock);
2723 mutex_init(&fs_info->unused_bg_unpin_mutex);
2724 mutex_init(&fs_info->reclaim_bgs_lock);
2725 mutex_init(&fs_info->reloc_mutex);
2726 mutex_init(&fs_info->delalloc_root_mutex);
2727 mutex_init(&fs_info->zoned_meta_io_lock);
2728 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2729 seqlock_init(&fs_info->profiles_lock);
2730
2731 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2732 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2733 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2734 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2735 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2736 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2737 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2738 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2739 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2740 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2741 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2742 BTRFS_LOCKDEP_TRANS_COMPLETED);
2743
2744 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2745 INIT_LIST_HEAD(&fs_info->space_info);
2746 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2747 INIT_LIST_HEAD(&fs_info->unused_bgs);
2748 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2749 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2750#ifdef CONFIG_BTRFS_DEBUG
2751 INIT_LIST_HEAD(&fs_info->allocated_roots);
2752 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2753 spin_lock_init(&fs_info->eb_leak_lock);
2754#endif
2755 fs_info->mapping_tree = RB_ROOT_CACHED;
2756 rwlock_init(&fs_info->mapping_tree_lock);
2757 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2758 BTRFS_BLOCK_RSV_GLOBAL);
2759 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2760 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2761 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2762 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2763 BTRFS_BLOCK_RSV_DELOPS);
2764 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2765 BTRFS_BLOCK_RSV_DELREFS);
2766
2767 atomic_set(&fs_info->async_delalloc_pages, 0);
2768 atomic_set(&fs_info->defrag_running, 0);
2769 atomic_set(&fs_info->nr_delayed_iputs, 0);
2770 atomic64_set(&fs_info->tree_mod_seq, 0);
2771 fs_info->global_root_tree = RB_ROOT;
2772 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2773 fs_info->metadata_ratio = 0;
2774 fs_info->defrag_inodes = RB_ROOT;
2775 atomic64_set(&fs_info->free_chunk_space, 0);
2776 fs_info->tree_mod_log = RB_ROOT;
2777 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2778 btrfs_init_ref_verify(fs_info);
2779
2780 fs_info->thread_pool_size = min_t(unsigned long,
2781 num_online_cpus() + 2, 8);
2782
2783 INIT_LIST_HEAD(&fs_info->ordered_roots);
2784 spin_lock_init(&fs_info->ordered_root_lock);
2785
2786 btrfs_init_scrub(fs_info);
2787 btrfs_init_balance(fs_info);
2788 btrfs_init_async_reclaim_work(fs_info);
2789
2790 rwlock_init(&fs_info->block_group_cache_lock);
2791 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2792
2793 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2794 IO_TREE_FS_EXCLUDED_EXTENTS);
2795
2796 mutex_init(&fs_info->ordered_operations_mutex);
2797 mutex_init(&fs_info->tree_log_mutex);
2798 mutex_init(&fs_info->chunk_mutex);
2799 mutex_init(&fs_info->transaction_kthread_mutex);
2800 mutex_init(&fs_info->cleaner_mutex);
2801 mutex_init(&fs_info->ro_block_group_mutex);
2802 init_rwsem(&fs_info->commit_root_sem);
2803 init_rwsem(&fs_info->cleanup_work_sem);
2804 init_rwsem(&fs_info->subvol_sem);
2805 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2806
2807 btrfs_init_dev_replace_locks(fs_info);
2808 btrfs_init_qgroup(fs_info);
2809 btrfs_discard_init(fs_info);
2810
2811 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2812 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2813
2814 init_waitqueue_head(&fs_info->transaction_throttle);
2815 init_waitqueue_head(&fs_info->transaction_wait);
2816 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2817 init_waitqueue_head(&fs_info->async_submit_wait);
2818 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2819
2820 /* Usable values until the real ones are cached from the superblock */
2821 fs_info->nodesize = 4096;
2822 fs_info->sectorsize = 4096;
2823 fs_info->sectorsize_bits = ilog2(4096);
2824 fs_info->stripesize = 4096;
2825
2826 /* Default compress algorithm when user does -o compress */
2827 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2828
2829 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2830
2831 spin_lock_init(&fs_info->swapfile_pins_lock);
2832 fs_info->swapfile_pins = RB_ROOT;
2833
2834 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2835 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2836}
2837
2838static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2839{
2840 int ret;
2841
2842 fs_info->sb = sb;
2843 /* Temporary fixed values for block size until we read the superblock. */
2844 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2845 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2846
2847 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2848 if (ret)
2849 return ret;
2850
2851 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2852 if (ret)
2853 return ret;
2854
2855 fs_info->dirty_metadata_batch = PAGE_SIZE *
2856 (1 + ilog2(nr_cpu_ids));
2857
2858 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2859 if (ret)
2860 return ret;
2861
2862 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2863 GFP_KERNEL);
2864 if (ret)
2865 return ret;
2866
2867 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2868 GFP_KERNEL);
2869 if (!fs_info->delayed_root)
2870 return -ENOMEM;
2871 btrfs_init_delayed_root(fs_info->delayed_root);
2872
2873 if (sb_rdonly(sb))
2874 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2875
2876 return btrfs_alloc_stripe_hash_table(fs_info);
2877}
2878
2879static int btrfs_uuid_rescan_kthread(void *data)
2880{
2881 struct btrfs_fs_info *fs_info = data;
2882 int ret;
2883
2884 /*
2885 * 1st step is to iterate through the existing UUID tree and
2886 * to delete all entries that contain outdated data.
2887 * 2nd step is to add all missing entries to the UUID tree.
2888 */
2889 ret = btrfs_uuid_tree_iterate(fs_info);
2890 if (ret < 0) {
2891 if (ret != -EINTR)
2892 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2893 ret);
2894 up(&fs_info->uuid_tree_rescan_sem);
2895 return ret;
2896 }
2897 return btrfs_uuid_scan_kthread(data);
2898}
2899
2900static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2901{
2902 struct task_struct *task;
2903
2904 down(&fs_info->uuid_tree_rescan_sem);
2905 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2906 if (IS_ERR(task)) {
2907 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2908 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2909 up(&fs_info->uuid_tree_rescan_sem);
2910 return PTR_ERR(task);
2911 }
2912
2913 return 0;
2914}
2915
2916static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2917{
2918 u64 root_objectid = 0;
2919 struct btrfs_root *gang[8];
2920 int i = 0;
2921 int err = 0;
2922 unsigned int ret = 0;
2923
2924 while (1) {
2925 spin_lock(&fs_info->fs_roots_radix_lock);
2926 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2927 (void **)gang, root_objectid,
2928 ARRAY_SIZE(gang));
2929 if (!ret) {
2930 spin_unlock(&fs_info->fs_roots_radix_lock);
2931 break;
2932 }
2933 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2934
2935 for (i = 0; i < ret; i++) {
2936 /* Avoid to grab roots in dead_roots. */
2937 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2938 gang[i] = NULL;
2939 continue;
2940 }
2941 /* Grab all the search result for later use. */
2942 gang[i] = btrfs_grab_root(gang[i]);
2943 }
2944 spin_unlock(&fs_info->fs_roots_radix_lock);
2945
2946 for (i = 0; i < ret; i++) {
2947 if (!gang[i])
2948 continue;
2949 root_objectid = gang[i]->root_key.objectid;
2950 err = btrfs_orphan_cleanup(gang[i]);
2951 if (err)
2952 goto out;
2953 btrfs_put_root(gang[i]);
2954 }
2955 root_objectid++;
2956 }
2957out:
2958 /* Release the uncleaned roots due to error. */
2959 for (; i < ret; i++) {
2960 if (gang[i])
2961 btrfs_put_root(gang[i]);
2962 }
2963 return err;
2964}
2965
2966/*
2967 * Mounting logic specific to read-write file systems. Shared by open_ctree
2968 * and btrfs_remount when remounting from read-only to read-write.
2969 */
2970int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2971{
2972 int ret;
2973 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2974 bool rebuild_free_space_tree = false;
2975
2976 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2977 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2978 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2979 btrfs_warn(fs_info,
2980 "'clear_cache' option is ignored with extent tree v2");
2981 else
2982 rebuild_free_space_tree = true;
2983 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2984 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2985 btrfs_warn(fs_info, "free space tree is invalid");
2986 rebuild_free_space_tree = true;
2987 }
2988
2989 if (rebuild_free_space_tree) {
2990 btrfs_info(fs_info, "rebuilding free space tree");
2991 ret = btrfs_rebuild_free_space_tree(fs_info);
2992 if (ret) {
2993 btrfs_warn(fs_info,
2994 "failed to rebuild free space tree: %d", ret);
2995 goto out;
2996 }
2997 }
2998
2999 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3000 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3001 btrfs_info(fs_info, "disabling free space tree");
3002 ret = btrfs_delete_free_space_tree(fs_info);
3003 if (ret) {
3004 btrfs_warn(fs_info,
3005 "failed to disable free space tree: %d", ret);
3006 goto out;
3007 }
3008 }
3009
3010 /*
3011 * btrfs_find_orphan_roots() is responsible for finding all the dead
3012 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3013 * them into the fs_info->fs_roots_radix tree. This must be done before
3014 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3015 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3016 * item before the root's tree is deleted - this means that if we unmount
3017 * or crash before the deletion completes, on the next mount we will not
3018 * delete what remains of the tree because the orphan item does not
3019 * exists anymore, which is what tells us we have a pending deletion.
3020 */
3021 ret = btrfs_find_orphan_roots(fs_info);
3022 if (ret)
3023 goto out;
3024
3025 ret = btrfs_cleanup_fs_roots(fs_info);
3026 if (ret)
3027 goto out;
3028
3029 down_read(&fs_info->cleanup_work_sem);
3030 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3031 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3032 up_read(&fs_info->cleanup_work_sem);
3033 goto out;
3034 }
3035 up_read(&fs_info->cleanup_work_sem);
3036
3037 mutex_lock(&fs_info->cleaner_mutex);
3038 ret = btrfs_recover_relocation(fs_info);
3039 mutex_unlock(&fs_info->cleaner_mutex);
3040 if (ret < 0) {
3041 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3042 goto out;
3043 }
3044
3045 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3046 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3047 btrfs_info(fs_info, "creating free space tree");
3048 ret = btrfs_create_free_space_tree(fs_info);
3049 if (ret) {
3050 btrfs_warn(fs_info,
3051 "failed to create free space tree: %d", ret);
3052 goto out;
3053 }
3054 }
3055
3056 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3057 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3058 if (ret)
3059 goto out;
3060 }
3061
3062 ret = btrfs_resume_balance_async(fs_info);
3063 if (ret)
3064 goto out;
3065
3066 ret = btrfs_resume_dev_replace_async(fs_info);
3067 if (ret) {
3068 btrfs_warn(fs_info, "failed to resume dev_replace");
3069 goto out;
3070 }
3071
3072 btrfs_qgroup_rescan_resume(fs_info);
3073
3074 if (!fs_info->uuid_root) {
3075 btrfs_info(fs_info, "creating UUID tree");
3076 ret = btrfs_create_uuid_tree(fs_info);
3077 if (ret) {
3078 btrfs_warn(fs_info,
3079 "failed to create the UUID tree %d", ret);
3080 goto out;
3081 }
3082 }
3083
3084out:
3085 return ret;
3086}
3087
3088/*
3089 * Do various sanity and dependency checks of different features.
3090 *
3091 * @is_rw_mount: If the mount is read-write.
3092 *
3093 * This is the place for less strict checks (like for subpage or artificial
3094 * feature dependencies).
3095 *
3096 * For strict checks or possible corruption detection, see
3097 * btrfs_validate_super().
3098 *
3099 * This should be called after btrfs_parse_options(), as some mount options
3100 * (space cache related) can modify on-disk format like free space tree and
3101 * screw up certain feature dependencies.
3102 */
3103int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3104{
3105 struct btrfs_super_block *disk_super = fs_info->super_copy;
3106 u64 incompat = btrfs_super_incompat_flags(disk_super);
3107 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3108 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3109
3110 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3111 btrfs_err(fs_info,
3112 "cannot mount because of unknown incompat features (0x%llx)",
3113 incompat);
3114 return -EINVAL;
3115 }
3116
3117 /* Runtime limitation for mixed block groups. */
3118 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3119 (fs_info->sectorsize != fs_info->nodesize)) {
3120 btrfs_err(fs_info,
3121"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3122 fs_info->nodesize, fs_info->sectorsize);
3123 return -EINVAL;
3124 }
3125
3126 /* Mixed backref is an always-enabled feature. */
3127 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3128
3129 /* Set compression related flags just in case. */
3130 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3131 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3132 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3133 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3134
3135 /*
3136 * An ancient flag, which should really be marked deprecated.
3137 * Such runtime limitation doesn't really need a incompat flag.
3138 */
3139 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3140 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3141
3142 if (compat_ro_unsupp && is_rw_mount) {
3143 btrfs_err(fs_info,
3144 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3145 compat_ro);
3146 return -EINVAL;
3147 }
3148
3149 /*
3150 * We have unsupported RO compat features, although RO mounted, we
3151 * should not cause any metadata writes, including log replay.
3152 * Or we could screw up whatever the new feature requires.
3153 */
3154 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3155 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3156 btrfs_err(fs_info,
3157"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3158 compat_ro);
3159 return -EINVAL;
3160 }
3161
3162 /*
3163 * Artificial limitations for block group tree, to force
3164 * block-group-tree to rely on no-holes and free-space-tree.
3165 */
3166 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3167 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3168 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3169 btrfs_err(fs_info,
3170"block-group-tree feature requires no-holes and free-space-tree features");
3171 return -EINVAL;
3172 }
3173
3174 /*
3175 * Subpage runtime limitation on v1 cache.
3176 *
3177 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3178 * we're already defaulting to v2 cache, no need to bother v1 as it's
3179 * going to be deprecated anyway.
3180 */
3181 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3182 btrfs_warn(fs_info,
3183 "v1 space cache is not supported for page size %lu with sectorsize %u",
3184 PAGE_SIZE, fs_info->sectorsize);
3185 return -EINVAL;
3186 }
3187
3188 /* This can be called by remount, we need to protect the super block. */
3189 spin_lock(&fs_info->super_lock);
3190 btrfs_set_super_incompat_flags(disk_super, incompat);
3191 spin_unlock(&fs_info->super_lock);
3192
3193 return 0;
3194}
3195
3196int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3197 char *options)
3198{
3199 u32 sectorsize;
3200 u32 nodesize;
3201 u32 stripesize;
3202 u64 generation;
3203 u16 csum_type;
3204 struct btrfs_super_block *disk_super;
3205 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3206 struct btrfs_root *tree_root;
3207 struct btrfs_root *chunk_root;
3208 int ret;
3209 int level;
3210
3211 ret = init_mount_fs_info(fs_info, sb);
3212 if (ret)
3213 goto fail;
3214
3215 /* These need to be init'ed before we start creating inodes and such. */
3216 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3217 GFP_KERNEL);
3218 fs_info->tree_root = tree_root;
3219 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3220 GFP_KERNEL);
3221 fs_info->chunk_root = chunk_root;
3222 if (!tree_root || !chunk_root) {
3223 ret = -ENOMEM;
3224 goto fail;
3225 }
3226
3227 ret = btrfs_init_btree_inode(sb);
3228 if (ret)
3229 goto fail;
3230
3231 invalidate_bdev(fs_devices->latest_dev->bdev);
3232
3233 /*
3234 * Read super block and check the signature bytes only
3235 */
3236 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3237 if (IS_ERR(disk_super)) {
3238 ret = PTR_ERR(disk_super);
3239 goto fail_alloc;
3240 }
3241
3242 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3243 /*
3244 * Verify the type first, if that or the checksum value are
3245 * corrupted, we'll find out
3246 */
3247 csum_type = btrfs_super_csum_type(disk_super);
3248 if (!btrfs_supported_super_csum(csum_type)) {
3249 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3250 csum_type);
3251 ret = -EINVAL;
3252 btrfs_release_disk_super(disk_super);
3253 goto fail_alloc;
3254 }
3255
3256 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3257
3258 ret = btrfs_init_csum_hash(fs_info, csum_type);
3259 if (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, disk_super)) {
3269 btrfs_err(fs_info, "superblock checksum mismatch");
3270 ret = -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 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3286 sizeof(*fs_info->super_for_commit));
3287
3288 ret = btrfs_validate_mount_super(fs_info);
3289 if (ret) {
3290 btrfs_err(fs_info, "superblock contains fatal errors");
3291 ret = -EINVAL;
3292 goto fail_alloc;
3293 }
3294
3295 if (!btrfs_super_root(disk_super)) {
3296 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3297 ret = -EINVAL;
3298 goto fail_alloc;
3299 }
3300
3301 /* check FS state, whether FS is broken. */
3302 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3303 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3304
3305 /* Set up fs_info before parsing mount options */
3306 nodesize = btrfs_super_nodesize(disk_super);
3307 sectorsize = btrfs_super_sectorsize(disk_super);
3308 stripesize = sectorsize;
3309 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3310 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3311
3312 fs_info->nodesize = nodesize;
3313 fs_info->sectorsize = sectorsize;
3314 fs_info->sectorsize_bits = ilog2(sectorsize);
3315 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3316 fs_info->stripesize = stripesize;
3317
3318 /*
3319 * Handle the space caching options appropriately now that we have the
3320 * super block loaded and validated.
3321 */
3322 btrfs_set_free_space_cache_settings(fs_info);
3323
3324 if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3325 ret = -EINVAL;
3326 goto fail_alloc;
3327 }
3328
3329 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3330 if (ret < 0)
3331 goto fail_alloc;
3332
3333 /*
3334 * At this point our mount options are validated, if we set ->max_inline
3335 * to something non-standard make sure we truncate it to sectorsize.
3336 */
3337 fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3338
3339 if (sectorsize < PAGE_SIZE) {
3340 struct btrfs_subpage_info *subpage_info;
3341
3342 btrfs_warn(fs_info,
3343 "read-write for sector size %u with page size %lu is experimental",
3344 sectorsize, PAGE_SIZE);
3345 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3346 if (!subpage_info) {
3347 ret = -ENOMEM;
3348 goto fail_alloc;
3349 }
3350 btrfs_init_subpage_info(subpage_info, sectorsize);
3351 fs_info->subpage_info = subpage_info;
3352 }
3353
3354 ret = btrfs_init_workqueues(fs_info);
3355 if (ret)
3356 goto fail_sb_buffer;
3357
3358 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3359 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3360
3361 /* Update the values for the current filesystem. */
3362 sb->s_blocksize = sectorsize;
3363 sb->s_blocksize_bits = blksize_bits(sectorsize);
3364 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3365
3366 mutex_lock(&fs_info->chunk_mutex);
3367 ret = btrfs_read_sys_array(fs_info);
3368 mutex_unlock(&fs_info->chunk_mutex);
3369 if (ret) {
3370 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3371 goto fail_sb_buffer;
3372 }
3373
3374 generation = btrfs_super_chunk_root_generation(disk_super);
3375 level = btrfs_super_chunk_root_level(disk_super);
3376 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3377 generation, level);
3378 if (ret) {
3379 btrfs_err(fs_info, "failed to read chunk root");
3380 goto fail_tree_roots;
3381 }
3382
3383 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3384 offsetof(struct btrfs_header, chunk_tree_uuid),
3385 BTRFS_UUID_SIZE);
3386
3387 ret = btrfs_read_chunk_tree(fs_info);
3388 if (ret) {
3389 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3390 goto fail_tree_roots;
3391 }
3392
3393 /*
3394 * At this point we know all the devices that make this filesystem,
3395 * including the seed devices but we don't know yet if the replace
3396 * target is required. So free devices that are not part of this
3397 * filesystem but skip the replace target device which is checked
3398 * below in btrfs_init_dev_replace().
3399 */
3400 btrfs_free_extra_devids(fs_devices);
3401 if (!fs_devices->latest_dev->bdev) {
3402 btrfs_err(fs_info, "failed to read devices");
3403 ret = -EIO;
3404 goto fail_tree_roots;
3405 }
3406
3407 ret = init_tree_roots(fs_info);
3408 if (ret)
3409 goto fail_tree_roots;
3410
3411 /*
3412 * Get zone type information of zoned block devices. This will also
3413 * handle emulation of a zoned filesystem if a regular device has the
3414 * zoned incompat feature flag set.
3415 */
3416 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3417 if (ret) {
3418 btrfs_err(fs_info,
3419 "zoned: failed to read device zone info: %d", ret);
3420 goto fail_block_groups;
3421 }
3422
3423 /*
3424 * If we have a uuid root and we're not being told to rescan we need to
3425 * check the generation here so we can set the
3426 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3427 * transaction during a balance or the log replay without updating the
3428 * uuid generation, and then if we crash we would rescan the uuid tree,
3429 * even though it was perfectly fine.
3430 */
3431 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3432 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3433 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3434
3435 ret = btrfs_verify_dev_extents(fs_info);
3436 if (ret) {
3437 btrfs_err(fs_info,
3438 "failed to verify dev extents against chunks: %d",
3439 ret);
3440 goto fail_block_groups;
3441 }
3442 ret = btrfs_recover_balance(fs_info);
3443 if (ret) {
3444 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3445 goto fail_block_groups;
3446 }
3447
3448 ret = btrfs_init_dev_stats(fs_info);
3449 if (ret) {
3450 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3451 goto fail_block_groups;
3452 }
3453
3454 ret = btrfs_init_dev_replace(fs_info);
3455 if (ret) {
3456 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3457 goto fail_block_groups;
3458 }
3459
3460 ret = btrfs_check_zoned_mode(fs_info);
3461 if (ret) {
3462 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3463 ret);
3464 goto fail_block_groups;
3465 }
3466
3467 ret = btrfs_sysfs_add_fsid(fs_devices);
3468 if (ret) {
3469 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3470 ret);
3471 goto fail_block_groups;
3472 }
3473
3474 ret = btrfs_sysfs_add_mounted(fs_info);
3475 if (ret) {
3476 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3477 goto fail_fsdev_sysfs;
3478 }
3479
3480 ret = btrfs_init_space_info(fs_info);
3481 if (ret) {
3482 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3483 goto fail_sysfs;
3484 }
3485
3486 ret = btrfs_read_block_groups(fs_info);
3487 if (ret) {
3488 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3489 goto fail_sysfs;
3490 }
3491
3492 btrfs_free_zone_cache(fs_info);
3493
3494 btrfs_check_active_zone_reservation(fs_info);
3495
3496 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3497 !btrfs_check_rw_degradable(fs_info, NULL)) {
3498 btrfs_warn(fs_info,
3499 "writable mount is not allowed due to too many missing devices");
3500 ret = -EINVAL;
3501 goto fail_sysfs;
3502 }
3503
3504 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3505 "btrfs-cleaner");
3506 if (IS_ERR(fs_info->cleaner_kthread)) {
3507 ret = PTR_ERR(fs_info->cleaner_kthread);
3508 goto fail_sysfs;
3509 }
3510
3511 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3512 tree_root,
3513 "btrfs-transaction");
3514 if (IS_ERR(fs_info->transaction_kthread)) {
3515 ret = PTR_ERR(fs_info->transaction_kthread);
3516 goto fail_cleaner;
3517 }
3518
3519 ret = btrfs_read_qgroup_config(fs_info);
3520 if (ret)
3521 goto fail_trans_kthread;
3522
3523 if (btrfs_build_ref_tree(fs_info))
3524 btrfs_err(fs_info, "couldn't build ref tree");
3525
3526 /* do not make disk changes in broken FS or nologreplay is given */
3527 if (btrfs_super_log_root(disk_super) != 0 &&
3528 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3529 btrfs_info(fs_info, "start tree-log replay");
3530 ret = btrfs_replay_log(fs_info, fs_devices);
3531 if (ret)
3532 goto fail_qgroup;
3533 }
3534
3535 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3536 if (IS_ERR(fs_info->fs_root)) {
3537 ret = PTR_ERR(fs_info->fs_root);
3538 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3539 fs_info->fs_root = NULL;
3540 goto fail_qgroup;
3541 }
3542
3543 if (sb_rdonly(sb))
3544 return 0;
3545
3546 ret = btrfs_start_pre_rw_mount(fs_info);
3547 if (ret) {
3548 close_ctree(fs_info);
3549 return ret;
3550 }
3551 btrfs_discard_resume(fs_info);
3552
3553 if (fs_info->uuid_root &&
3554 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3555 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3556 btrfs_info(fs_info, "checking UUID tree");
3557 ret = btrfs_check_uuid_tree(fs_info);
3558 if (ret) {
3559 btrfs_warn(fs_info,
3560 "failed to check the UUID tree: %d", ret);
3561 close_ctree(fs_info);
3562 return ret;
3563 }
3564 }
3565
3566 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3567
3568 /* Kick the cleaner thread so it'll start deleting snapshots. */
3569 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3570 wake_up_process(fs_info->cleaner_kthread);
3571
3572 return 0;
3573
3574fail_qgroup:
3575 btrfs_free_qgroup_config(fs_info);
3576fail_trans_kthread:
3577 kthread_stop(fs_info->transaction_kthread);
3578 btrfs_cleanup_transaction(fs_info);
3579 btrfs_free_fs_roots(fs_info);
3580fail_cleaner:
3581 kthread_stop(fs_info->cleaner_kthread);
3582
3583 /*
3584 * make sure we're done with the btree inode before we stop our
3585 * kthreads
3586 */
3587 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3588
3589fail_sysfs:
3590 btrfs_sysfs_remove_mounted(fs_info);
3591
3592fail_fsdev_sysfs:
3593 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3594
3595fail_block_groups:
3596 btrfs_put_block_group_cache(fs_info);
3597
3598fail_tree_roots:
3599 if (fs_info->data_reloc_root)
3600 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3601 free_root_pointers(fs_info, true);
3602 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3603
3604fail_sb_buffer:
3605 btrfs_stop_all_workers(fs_info);
3606 btrfs_free_block_groups(fs_info);
3607fail_alloc:
3608 btrfs_mapping_tree_free(fs_info);
3609
3610 iput(fs_info->btree_inode);
3611fail:
3612 btrfs_close_devices(fs_info->fs_devices);
3613 ASSERT(ret < 0);
3614 return ret;
3615}
3616ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3617
3618static void btrfs_end_super_write(struct bio *bio)
3619{
3620 struct btrfs_device *device = bio->bi_private;
3621 struct bio_vec *bvec;
3622 struct bvec_iter_all iter_all;
3623 struct page *page;
3624
3625 bio_for_each_segment_all(bvec, bio, iter_all) {
3626 page = bvec->bv_page;
3627
3628 if (bio->bi_status) {
3629 btrfs_warn_rl_in_rcu(device->fs_info,
3630 "lost page write due to IO error on %s (%d)",
3631 btrfs_dev_name(device),
3632 blk_status_to_errno(bio->bi_status));
3633 ClearPageUptodate(page);
3634 SetPageError(page);
3635 btrfs_dev_stat_inc_and_print(device,
3636 BTRFS_DEV_STAT_WRITE_ERRS);
3637 } else {
3638 SetPageUptodate(page);
3639 }
3640
3641 put_page(page);
3642 unlock_page(page);
3643 }
3644
3645 bio_put(bio);
3646}
3647
3648struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3649 int copy_num, bool drop_cache)
3650{
3651 struct btrfs_super_block *super;
3652 struct page *page;
3653 u64 bytenr, bytenr_orig;
3654 struct address_space *mapping = bdev->bd_inode->i_mapping;
3655 int ret;
3656
3657 bytenr_orig = btrfs_sb_offset(copy_num);
3658 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3659 if (ret == -ENOENT)
3660 return ERR_PTR(-EINVAL);
3661 else if (ret)
3662 return ERR_PTR(ret);
3663
3664 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3665 return ERR_PTR(-EINVAL);
3666
3667 if (drop_cache) {
3668 /* This should only be called with the primary sb. */
3669 ASSERT(copy_num == 0);
3670
3671 /*
3672 * Drop the page of the primary superblock, so later read will
3673 * always read from the device.
3674 */
3675 invalidate_inode_pages2_range(mapping,
3676 bytenr >> PAGE_SHIFT,
3677 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3678 }
3679
3680 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3681 if (IS_ERR(page))
3682 return ERR_CAST(page);
3683
3684 super = page_address(page);
3685 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3686 btrfs_release_disk_super(super);
3687 return ERR_PTR(-ENODATA);
3688 }
3689
3690 if (btrfs_super_bytenr(super) != bytenr_orig) {
3691 btrfs_release_disk_super(super);
3692 return ERR_PTR(-EINVAL);
3693 }
3694
3695 return super;
3696}
3697
3698
3699struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3700{
3701 struct btrfs_super_block *super, *latest = NULL;
3702 int i;
3703 u64 transid = 0;
3704
3705 /* we would like to check all the supers, but that would make
3706 * a btrfs mount succeed after a mkfs from a different FS.
3707 * So, we need to add a special mount option to scan for
3708 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3709 */
3710 for (i = 0; i < 1; i++) {
3711 super = btrfs_read_dev_one_super(bdev, i, false);
3712 if (IS_ERR(super))
3713 continue;
3714
3715 if (!latest || btrfs_super_generation(super) > transid) {
3716 if (latest)
3717 btrfs_release_disk_super(super);
3718
3719 latest = super;
3720 transid = btrfs_super_generation(super);
3721 }
3722 }
3723
3724 return super;
3725}
3726
3727/*
3728 * Write superblock @sb to the @device. Do not wait for completion, all the
3729 * pages we use for writing are locked.
3730 *
3731 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3732 * the expected device size at commit time. Note that max_mirrors must be
3733 * same for write and wait phases.
3734 *
3735 * Return number of errors when page is not found or submission fails.
3736 */
3737static int write_dev_supers(struct btrfs_device *device,
3738 struct btrfs_super_block *sb, int max_mirrors)
3739{
3740 struct btrfs_fs_info *fs_info = device->fs_info;
3741 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3742 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3743 int i;
3744 int errors = 0;
3745 int ret;
3746 u64 bytenr, bytenr_orig;
3747
3748 if (max_mirrors == 0)
3749 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3750
3751 shash->tfm = fs_info->csum_shash;
3752
3753 for (i = 0; i < max_mirrors; i++) {
3754 struct page *page;
3755 struct bio *bio;
3756 struct btrfs_super_block *disk_super;
3757
3758 bytenr_orig = btrfs_sb_offset(i);
3759 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3760 if (ret == -ENOENT) {
3761 continue;
3762 } else if (ret < 0) {
3763 btrfs_err(device->fs_info,
3764 "couldn't get super block location for mirror %d",
3765 i);
3766 errors++;
3767 continue;
3768 }
3769 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3770 device->commit_total_bytes)
3771 break;
3772
3773 btrfs_set_super_bytenr(sb, bytenr_orig);
3774
3775 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3776 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3777 sb->csum);
3778
3779 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3780 GFP_NOFS);
3781 if (!page) {
3782 btrfs_err(device->fs_info,
3783 "couldn't get super block page for bytenr %llu",
3784 bytenr);
3785 errors++;
3786 continue;
3787 }
3788
3789 /* Bump the refcount for wait_dev_supers() */
3790 get_page(page);
3791
3792 disk_super = page_address(page);
3793 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3794
3795 /*
3796 * Directly use bios here instead of relying on the page cache
3797 * to do I/O, so we don't lose the ability to do integrity
3798 * checking.
3799 */
3800 bio = bio_alloc(device->bdev, 1,
3801 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3802 GFP_NOFS);
3803 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3804 bio->bi_private = device;
3805 bio->bi_end_io = btrfs_end_super_write;
3806 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3807 offset_in_page(bytenr));
3808
3809 /*
3810 * We FUA only the first super block. The others we allow to
3811 * go down lazy and there's a short window where the on-disk
3812 * copies might still contain the older version.
3813 */
3814 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3815 bio->bi_opf |= REQ_FUA;
3816 submit_bio(bio);
3817
3818 if (btrfs_advance_sb_log(device, i))
3819 errors++;
3820 }
3821 return errors < i ? 0 : -1;
3822}
3823
3824/*
3825 * Wait for write completion of superblocks done by write_dev_supers,
3826 * @max_mirrors same for write and wait phases.
3827 *
3828 * Return number of errors when page is not found or not marked up to
3829 * date.
3830 */
3831static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3832{
3833 int i;
3834 int errors = 0;
3835 bool primary_failed = false;
3836 int ret;
3837 u64 bytenr;
3838
3839 if (max_mirrors == 0)
3840 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3841
3842 for (i = 0; i < max_mirrors; i++) {
3843 struct page *page;
3844
3845 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3846 if (ret == -ENOENT) {
3847 break;
3848 } else if (ret < 0) {
3849 errors++;
3850 if (i == 0)
3851 primary_failed = true;
3852 continue;
3853 }
3854 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3855 device->commit_total_bytes)
3856 break;
3857
3858 page = find_get_page(device->bdev->bd_inode->i_mapping,
3859 bytenr >> PAGE_SHIFT);
3860 if (!page) {
3861 errors++;
3862 if (i == 0)
3863 primary_failed = true;
3864 continue;
3865 }
3866 /* Page is submitted locked and unlocked once the IO completes */
3867 wait_on_page_locked(page);
3868 if (PageError(page)) {
3869 errors++;
3870 if (i == 0)
3871 primary_failed = true;
3872 }
3873
3874 /* Drop our reference */
3875 put_page(page);
3876
3877 /* Drop the reference from the writing run */
3878 put_page(page);
3879 }
3880
3881 /* log error, force error return */
3882 if (primary_failed) {
3883 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3884 device->devid);
3885 return -1;
3886 }
3887
3888 return errors < i ? 0 : -1;
3889}
3890
3891/*
3892 * endio for the write_dev_flush, this will wake anyone waiting
3893 * for the barrier when it is done
3894 */
3895static void btrfs_end_empty_barrier(struct bio *bio)
3896{
3897 bio_uninit(bio);
3898 complete(bio->bi_private);
3899}
3900
3901/*
3902 * Submit a flush request to the device if it supports it. Error handling is
3903 * done in the waiting counterpart.
3904 */
3905static void write_dev_flush(struct btrfs_device *device)
3906{
3907 struct bio *bio = &device->flush_bio;
3908
3909 device->last_flush_error = BLK_STS_OK;
3910
3911 bio_init(bio, device->bdev, NULL, 0,
3912 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3913 bio->bi_end_io = btrfs_end_empty_barrier;
3914 init_completion(&device->flush_wait);
3915 bio->bi_private = &device->flush_wait;
3916 submit_bio(bio);
3917 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3918}
3919
3920/*
3921 * If the flush bio has been submitted by write_dev_flush, wait for it.
3922 * Return true for any error, and false otherwise.
3923 */
3924static bool wait_dev_flush(struct btrfs_device *device)
3925{
3926 struct bio *bio = &device->flush_bio;
3927
3928 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3929 return false;
3930
3931 wait_for_completion_io(&device->flush_wait);
3932
3933 if (bio->bi_status) {
3934 device->last_flush_error = bio->bi_status;
3935 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3936 return true;
3937 }
3938
3939 return false;
3940}
3941
3942/*
3943 * send an empty flush down to each device in parallel,
3944 * then wait for them
3945 */
3946static int barrier_all_devices(struct btrfs_fs_info *info)
3947{
3948 struct list_head *head;
3949 struct btrfs_device *dev;
3950 int errors_wait = 0;
3951
3952 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3953 /* send down all the barriers */
3954 head = &info->fs_devices->devices;
3955 list_for_each_entry(dev, head, dev_list) {
3956 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3957 continue;
3958 if (!dev->bdev)
3959 continue;
3960 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3961 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3962 continue;
3963
3964 write_dev_flush(dev);
3965 }
3966
3967 /* wait for all the barriers */
3968 list_for_each_entry(dev, head, dev_list) {
3969 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3970 continue;
3971 if (!dev->bdev) {
3972 errors_wait++;
3973 continue;
3974 }
3975 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3976 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3977 continue;
3978
3979 if (wait_dev_flush(dev))
3980 errors_wait++;
3981 }
3982
3983 /*
3984 * Checks last_flush_error of disks in order to determine the device
3985 * state.
3986 */
3987 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3988 return -EIO;
3989
3990 return 0;
3991}
3992
3993int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3994{
3995 int raid_type;
3996 int min_tolerated = INT_MAX;
3997
3998 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3999 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4000 min_tolerated = min_t(int, min_tolerated,
4001 btrfs_raid_array[BTRFS_RAID_SINGLE].
4002 tolerated_failures);
4003
4004 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4005 if (raid_type == BTRFS_RAID_SINGLE)
4006 continue;
4007 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4008 continue;
4009 min_tolerated = min_t(int, min_tolerated,
4010 btrfs_raid_array[raid_type].
4011 tolerated_failures);
4012 }
4013
4014 if (min_tolerated == INT_MAX) {
4015 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4016 min_tolerated = 0;
4017 }
4018
4019 return min_tolerated;
4020}
4021
4022int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4023{
4024 struct list_head *head;
4025 struct btrfs_device *dev;
4026 struct btrfs_super_block *sb;
4027 struct btrfs_dev_item *dev_item;
4028 int ret;
4029 int do_barriers;
4030 int max_errors;
4031 int total_errors = 0;
4032 u64 flags;
4033
4034 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4035
4036 /*
4037 * max_mirrors == 0 indicates we're from commit_transaction,
4038 * not from fsync where the tree roots in fs_info have not
4039 * been consistent on disk.
4040 */
4041 if (max_mirrors == 0)
4042 backup_super_roots(fs_info);
4043
4044 sb = fs_info->super_for_commit;
4045 dev_item = &sb->dev_item;
4046
4047 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4048 head = &fs_info->fs_devices->devices;
4049 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4050
4051 if (do_barriers) {
4052 ret = barrier_all_devices(fs_info);
4053 if (ret) {
4054 mutex_unlock(
4055 &fs_info->fs_devices->device_list_mutex);
4056 btrfs_handle_fs_error(fs_info, ret,
4057 "errors while submitting device barriers.");
4058 return ret;
4059 }
4060 }
4061
4062 list_for_each_entry(dev, head, dev_list) {
4063 if (!dev->bdev) {
4064 total_errors++;
4065 continue;
4066 }
4067 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4068 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4069 continue;
4070
4071 btrfs_set_stack_device_generation(dev_item, 0);
4072 btrfs_set_stack_device_type(dev_item, dev->type);
4073 btrfs_set_stack_device_id(dev_item, dev->devid);
4074 btrfs_set_stack_device_total_bytes(dev_item,
4075 dev->commit_total_bytes);
4076 btrfs_set_stack_device_bytes_used(dev_item,
4077 dev->commit_bytes_used);
4078 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4079 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4080 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4081 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4082 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4083 BTRFS_FSID_SIZE);
4084
4085 flags = btrfs_super_flags(sb);
4086 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4087
4088 ret = btrfs_validate_write_super(fs_info, sb);
4089 if (ret < 0) {
4090 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4091 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4092 "unexpected superblock corruption detected");
4093 return -EUCLEAN;
4094 }
4095
4096 ret = write_dev_supers(dev, sb, max_mirrors);
4097 if (ret)
4098 total_errors++;
4099 }
4100 if (total_errors > max_errors) {
4101 btrfs_err(fs_info, "%d errors while writing supers",
4102 total_errors);
4103 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4104
4105 /* FUA is masked off if unsupported and can't be the reason */
4106 btrfs_handle_fs_error(fs_info, -EIO,
4107 "%d errors while writing supers",
4108 total_errors);
4109 return -EIO;
4110 }
4111
4112 total_errors = 0;
4113 list_for_each_entry(dev, head, dev_list) {
4114 if (!dev->bdev)
4115 continue;
4116 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4117 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4118 continue;
4119
4120 ret = wait_dev_supers(dev, max_mirrors);
4121 if (ret)
4122 total_errors++;
4123 }
4124 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4125 if (total_errors > max_errors) {
4126 btrfs_handle_fs_error(fs_info, -EIO,
4127 "%d errors while writing supers",
4128 total_errors);
4129 return -EIO;
4130 }
4131 return 0;
4132}
4133
4134/* Drop a fs root from the radix tree and free it. */
4135void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4136 struct btrfs_root *root)
4137{
4138 bool drop_ref = false;
4139
4140 spin_lock(&fs_info->fs_roots_radix_lock);
4141 radix_tree_delete(&fs_info->fs_roots_radix,
4142 (unsigned long)root->root_key.objectid);
4143 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4144 drop_ref = true;
4145 spin_unlock(&fs_info->fs_roots_radix_lock);
4146
4147 if (BTRFS_FS_ERROR(fs_info)) {
4148 ASSERT(root->log_root == NULL);
4149 if (root->reloc_root) {
4150 btrfs_put_root(root->reloc_root);
4151 root->reloc_root = NULL;
4152 }
4153 }
4154
4155 if (drop_ref)
4156 btrfs_put_root(root);
4157}
4158
4159int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4160{
4161 struct btrfs_root *root = fs_info->tree_root;
4162 struct btrfs_trans_handle *trans;
4163
4164 mutex_lock(&fs_info->cleaner_mutex);
4165 btrfs_run_delayed_iputs(fs_info);
4166 mutex_unlock(&fs_info->cleaner_mutex);
4167 wake_up_process(fs_info->cleaner_kthread);
4168
4169 /* wait until ongoing cleanup work done */
4170 down_write(&fs_info->cleanup_work_sem);
4171 up_write(&fs_info->cleanup_work_sem);
4172
4173 trans = btrfs_join_transaction(root);
4174 if (IS_ERR(trans))
4175 return PTR_ERR(trans);
4176 return btrfs_commit_transaction(trans);
4177}
4178
4179static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4180{
4181 struct btrfs_transaction *trans;
4182 struct btrfs_transaction *tmp;
4183 bool found = false;
4184
4185 if (list_empty(&fs_info->trans_list))
4186 return;
4187
4188 /*
4189 * This function is only called at the very end of close_ctree(),
4190 * thus no other running transaction, no need to take trans_lock.
4191 */
4192 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4193 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4194 struct extent_state *cached = NULL;
4195 u64 dirty_bytes = 0;
4196 u64 cur = 0;
4197 u64 found_start;
4198 u64 found_end;
4199
4200 found = true;
4201 while (find_first_extent_bit(&trans->dirty_pages, cur,
4202 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4203 dirty_bytes += found_end + 1 - found_start;
4204 cur = found_end + 1;
4205 }
4206 btrfs_warn(fs_info,
4207 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4208 trans->transid, dirty_bytes);
4209 btrfs_cleanup_one_transaction(trans, fs_info);
4210
4211 if (trans == fs_info->running_transaction)
4212 fs_info->running_transaction = NULL;
4213 list_del_init(&trans->list);
4214
4215 btrfs_put_transaction(trans);
4216 trace_btrfs_transaction_commit(fs_info);
4217 }
4218 ASSERT(!found);
4219}
4220
4221void __cold close_ctree(struct btrfs_fs_info *fs_info)
4222{
4223 int ret;
4224
4225 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4226
4227 /*
4228 * If we had UNFINISHED_DROPS we could still be processing them, so
4229 * clear that bit and wake up relocation so it can stop.
4230 * We must do this before stopping the block group reclaim task, because
4231 * at btrfs_relocate_block_group() we wait for this bit, and after the
4232 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4233 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4234 * return 1.
4235 */
4236 btrfs_wake_unfinished_drop(fs_info);
4237
4238 /*
4239 * We may have the reclaim task running and relocating a data block group,
4240 * in which case it may create delayed iputs. So stop it before we park
4241 * the cleaner kthread otherwise we can get new delayed iputs after
4242 * parking the cleaner, and that can make the async reclaim task to hang
4243 * if it's waiting for delayed iputs to complete, since the cleaner is
4244 * parked and can not run delayed iputs - this will make us hang when
4245 * trying to stop the async reclaim task.
4246 */
4247 cancel_work_sync(&fs_info->reclaim_bgs_work);
4248 /*
4249 * We don't want the cleaner to start new transactions, add more delayed
4250 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4251 * because that frees the task_struct, and the transaction kthread might
4252 * still try to wake up the cleaner.
4253 */
4254 kthread_park(fs_info->cleaner_kthread);
4255
4256 /* wait for the qgroup rescan worker to stop */
4257 btrfs_qgroup_wait_for_completion(fs_info, false);
4258
4259 /* wait for the uuid_scan task to finish */
4260 down(&fs_info->uuid_tree_rescan_sem);
4261 /* avoid complains from lockdep et al., set sem back to initial state */
4262 up(&fs_info->uuid_tree_rescan_sem);
4263
4264 /* pause restriper - we want to resume on mount */
4265 btrfs_pause_balance(fs_info);
4266
4267 btrfs_dev_replace_suspend_for_unmount(fs_info);
4268
4269 btrfs_scrub_cancel(fs_info);
4270
4271 /* wait for any defraggers to finish */
4272 wait_event(fs_info->transaction_wait,
4273 (atomic_read(&fs_info->defrag_running) == 0));
4274
4275 /* clear out the rbtree of defraggable inodes */
4276 btrfs_cleanup_defrag_inodes(fs_info);
4277
4278 /*
4279 * After we parked the cleaner kthread, ordered extents may have
4280 * completed and created new delayed iputs. If one of the async reclaim
4281 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4282 * can hang forever trying to stop it, because if a delayed iput is
4283 * added after it ran btrfs_run_delayed_iputs() and before it called
4284 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4285 * no one else to run iputs.
4286 *
4287 * So wait for all ongoing ordered extents to complete and then run
4288 * delayed iputs. This works because once we reach this point no one
4289 * can either create new ordered extents nor create delayed iputs
4290 * through some other means.
4291 *
4292 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4293 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4294 * but the delayed iput for the respective inode is made only when doing
4295 * the final btrfs_put_ordered_extent() (which must happen at
4296 * btrfs_finish_ordered_io() when we are unmounting).
4297 */
4298 btrfs_flush_workqueue(fs_info->endio_write_workers);
4299 /* Ordered extents for free space inodes. */
4300 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4301 btrfs_run_delayed_iputs(fs_info);
4302
4303 cancel_work_sync(&fs_info->async_reclaim_work);
4304 cancel_work_sync(&fs_info->async_data_reclaim_work);
4305 cancel_work_sync(&fs_info->preempt_reclaim_work);
4306
4307 /* Cancel or finish ongoing discard work */
4308 btrfs_discard_cleanup(fs_info);
4309
4310 if (!sb_rdonly(fs_info->sb)) {
4311 /*
4312 * The cleaner kthread is stopped, so do one final pass over
4313 * unused block groups.
4314 */
4315 btrfs_delete_unused_bgs(fs_info);
4316
4317 /*
4318 * There might be existing delayed inode workers still running
4319 * and holding an empty delayed inode item. We must wait for
4320 * them to complete first because they can create a transaction.
4321 * This happens when someone calls btrfs_balance_delayed_items()
4322 * and then a transaction commit runs the same delayed nodes
4323 * before any delayed worker has done something with the nodes.
4324 * We must wait for any worker here and not at transaction
4325 * commit time since that could cause a deadlock.
4326 * This is a very rare case.
4327 */
4328 btrfs_flush_workqueue(fs_info->delayed_workers);
4329
4330 ret = btrfs_commit_super(fs_info);
4331 if (ret)
4332 btrfs_err(fs_info, "commit super ret %d", ret);
4333 }
4334
4335 if (BTRFS_FS_ERROR(fs_info))
4336 btrfs_error_commit_super(fs_info);
4337
4338 kthread_stop(fs_info->transaction_kthread);
4339 kthread_stop(fs_info->cleaner_kthread);
4340
4341 ASSERT(list_empty(&fs_info->delayed_iputs));
4342 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4343
4344 if (btrfs_check_quota_leak(fs_info)) {
4345 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4346 btrfs_err(fs_info, "qgroup reserved space leaked");
4347 }
4348
4349 btrfs_free_qgroup_config(fs_info);
4350 ASSERT(list_empty(&fs_info->delalloc_roots));
4351
4352 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4353 btrfs_info(fs_info, "at unmount delalloc count %lld",
4354 percpu_counter_sum(&fs_info->delalloc_bytes));
4355 }
4356
4357 if (percpu_counter_sum(&fs_info->ordered_bytes))
4358 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4359 percpu_counter_sum(&fs_info->ordered_bytes));
4360
4361 btrfs_sysfs_remove_mounted(fs_info);
4362 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4363
4364 btrfs_put_block_group_cache(fs_info);
4365
4366 /*
4367 * we must make sure there is not any read request to
4368 * submit after we stopping all workers.
4369 */
4370 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4371 btrfs_stop_all_workers(fs_info);
4372
4373 /* We shouldn't have any transaction open at this point */
4374 warn_about_uncommitted_trans(fs_info);
4375
4376 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4377 free_root_pointers(fs_info, true);
4378 btrfs_free_fs_roots(fs_info);
4379
4380 /*
4381 * We must free the block groups after dropping the fs_roots as we could
4382 * have had an IO error and have left over tree log blocks that aren't
4383 * cleaned up until the fs roots are freed. This makes the block group
4384 * accounting appear to be wrong because there's pending reserved bytes,
4385 * so make sure we do the block group cleanup afterwards.
4386 */
4387 btrfs_free_block_groups(fs_info);
4388
4389 iput(fs_info->btree_inode);
4390
4391 btrfs_mapping_tree_free(fs_info);
4392 btrfs_close_devices(fs_info->fs_devices);
4393}
4394
4395void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4396 struct extent_buffer *buf)
4397{
4398 struct btrfs_fs_info *fs_info = buf->fs_info;
4399 u64 transid = btrfs_header_generation(buf);
4400
4401#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4402 /*
4403 * This is a fast path so only do this check if we have sanity tests
4404 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4405 * outside of the sanity tests.
4406 */
4407 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4408 return;
4409#endif
4410 /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4411 ASSERT(trans->transid == fs_info->generation);
4412 btrfs_assert_tree_write_locked(buf);
4413 if (unlikely(transid != fs_info->generation)) {
4414 btrfs_abort_transaction(trans, -EUCLEAN);
4415 btrfs_crit(fs_info,
4416"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4417 buf->start, transid, fs_info->generation);
4418 }
4419 set_extent_buffer_dirty(buf);
4420}
4421
4422static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4423 int flush_delayed)
4424{
4425 /*
4426 * looks as though older kernels can get into trouble with
4427 * this code, they end up stuck in balance_dirty_pages forever
4428 */
4429 int ret;
4430
4431 if (current->flags & PF_MEMALLOC)
4432 return;
4433
4434 if (flush_delayed)
4435 btrfs_balance_delayed_items(fs_info);
4436
4437 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4438 BTRFS_DIRTY_METADATA_THRESH,
4439 fs_info->dirty_metadata_batch);
4440 if (ret > 0) {
4441 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4442 }
4443}
4444
4445void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4446{
4447 __btrfs_btree_balance_dirty(fs_info, 1);
4448}
4449
4450void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4451{
4452 __btrfs_btree_balance_dirty(fs_info, 0);
4453}
4454
4455static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4456{
4457 /* cleanup FS via transaction */
4458 btrfs_cleanup_transaction(fs_info);
4459
4460 mutex_lock(&fs_info->cleaner_mutex);
4461 btrfs_run_delayed_iputs(fs_info);
4462 mutex_unlock(&fs_info->cleaner_mutex);
4463
4464 down_write(&fs_info->cleanup_work_sem);
4465 up_write(&fs_info->cleanup_work_sem);
4466}
4467
4468static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4469{
4470 struct btrfs_root *gang[8];
4471 u64 root_objectid = 0;
4472 int ret;
4473
4474 spin_lock(&fs_info->fs_roots_radix_lock);
4475 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4476 (void **)gang, root_objectid,
4477 ARRAY_SIZE(gang))) != 0) {
4478 int i;
4479
4480 for (i = 0; i < ret; i++)
4481 gang[i] = btrfs_grab_root(gang[i]);
4482 spin_unlock(&fs_info->fs_roots_radix_lock);
4483
4484 for (i = 0; i < ret; i++) {
4485 if (!gang[i])
4486 continue;
4487 root_objectid = gang[i]->root_key.objectid;
4488 btrfs_free_log(NULL, gang[i]);
4489 btrfs_put_root(gang[i]);
4490 }
4491 root_objectid++;
4492 spin_lock(&fs_info->fs_roots_radix_lock);
4493 }
4494 spin_unlock(&fs_info->fs_roots_radix_lock);
4495 btrfs_free_log_root_tree(NULL, fs_info);
4496}
4497
4498static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4499{
4500 struct btrfs_ordered_extent *ordered;
4501
4502 spin_lock(&root->ordered_extent_lock);
4503 /*
4504 * This will just short circuit the ordered completion stuff which will
4505 * make sure the ordered extent gets properly cleaned up.
4506 */
4507 list_for_each_entry(ordered, &root->ordered_extents,
4508 root_extent_list)
4509 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4510 spin_unlock(&root->ordered_extent_lock);
4511}
4512
4513static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4514{
4515 struct btrfs_root *root;
4516 LIST_HEAD(splice);
4517
4518 spin_lock(&fs_info->ordered_root_lock);
4519 list_splice_init(&fs_info->ordered_roots, &splice);
4520 while (!list_empty(&splice)) {
4521 root = list_first_entry(&splice, struct btrfs_root,
4522 ordered_root);
4523 list_move_tail(&root->ordered_root,
4524 &fs_info->ordered_roots);
4525
4526 spin_unlock(&fs_info->ordered_root_lock);
4527 btrfs_destroy_ordered_extents(root);
4528
4529 cond_resched();
4530 spin_lock(&fs_info->ordered_root_lock);
4531 }
4532 spin_unlock(&fs_info->ordered_root_lock);
4533
4534 /*
4535 * We need this here because if we've been flipped read-only we won't
4536 * get sync() from the umount, so we need to make sure any ordered
4537 * extents that haven't had their dirty pages IO start writeout yet
4538 * actually get run and error out properly.
4539 */
4540 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4541}
4542
4543static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4544 struct btrfs_fs_info *fs_info)
4545{
4546 struct rb_node *node;
4547 struct btrfs_delayed_ref_root *delayed_refs;
4548 struct btrfs_delayed_ref_node *ref;
4549
4550 delayed_refs = &trans->delayed_refs;
4551
4552 spin_lock(&delayed_refs->lock);
4553 if (atomic_read(&delayed_refs->num_entries) == 0) {
4554 spin_unlock(&delayed_refs->lock);
4555 btrfs_debug(fs_info, "delayed_refs has NO entry");
4556 return;
4557 }
4558
4559 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4560 struct btrfs_delayed_ref_head *head;
4561 struct rb_node *n;
4562 bool pin_bytes = false;
4563
4564 head = rb_entry(node, struct btrfs_delayed_ref_head,
4565 href_node);
4566 if (btrfs_delayed_ref_lock(delayed_refs, head))
4567 continue;
4568
4569 spin_lock(&head->lock);
4570 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4571 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4572 ref_node);
4573 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4574 RB_CLEAR_NODE(&ref->ref_node);
4575 if (!list_empty(&ref->add_list))
4576 list_del(&ref->add_list);
4577 atomic_dec(&delayed_refs->num_entries);
4578 btrfs_put_delayed_ref(ref);
4579 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4580 }
4581 if (head->must_insert_reserved)
4582 pin_bytes = true;
4583 btrfs_free_delayed_extent_op(head->extent_op);
4584 btrfs_delete_ref_head(delayed_refs, head);
4585 spin_unlock(&head->lock);
4586 spin_unlock(&delayed_refs->lock);
4587 mutex_unlock(&head->mutex);
4588
4589 if (pin_bytes) {
4590 struct btrfs_block_group *cache;
4591
4592 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4593 BUG_ON(!cache);
4594
4595 spin_lock(&cache->space_info->lock);
4596 spin_lock(&cache->lock);
4597 cache->pinned += head->num_bytes;
4598 btrfs_space_info_update_bytes_pinned(fs_info,
4599 cache->space_info, head->num_bytes);
4600 cache->reserved -= head->num_bytes;
4601 cache->space_info->bytes_reserved -= head->num_bytes;
4602 spin_unlock(&cache->lock);
4603 spin_unlock(&cache->space_info->lock);
4604
4605 btrfs_put_block_group(cache);
4606
4607 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4608 head->bytenr + head->num_bytes - 1);
4609 }
4610 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4611 btrfs_put_delayed_ref_head(head);
4612 cond_resched();
4613 spin_lock(&delayed_refs->lock);
4614 }
4615 btrfs_qgroup_destroy_extent_records(trans);
4616
4617 spin_unlock(&delayed_refs->lock);
4618}
4619
4620static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4621{
4622 struct btrfs_inode *btrfs_inode;
4623 LIST_HEAD(splice);
4624
4625 spin_lock(&root->delalloc_lock);
4626 list_splice_init(&root->delalloc_inodes, &splice);
4627
4628 while (!list_empty(&splice)) {
4629 struct inode *inode = NULL;
4630 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4631 delalloc_inodes);
4632 btrfs_del_delalloc_inode(btrfs_inode);
4633 spin_unlock(&root->delalloc_lock);
4634
4635 /*
4636 * Make sure we get a live inode and that it'll not disappear
4637 * meanwhile.
4638 */
4639 inode = igrab(&btrfs_inode->vfs_inode);
4640 if (inode) {
4641 unsigned int nofs_flag;
4642
4643 nofs_flag = memalloc_nofs_save();
4644 invalidate_inode_pages2(inode->i_mapping);
4645 memalloc_nofs_restore(nofs_flag);
4646 iput(inode);
4647 }
4648 spin_lock(&root->delalloc_lock);
4649 }
4650 spin_unlock(&root->delalloc_lock);
4651}
4652
4653static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4654{
4655 struct btrfs_root *root;
4656 LIST_HEAD(splice);
4657
4658 spin_lock(&fs_info->delalloc_root_lock);
4659 list_splice_init(&fs_info->delalloc_roots, &splice);
4660 while (!list_empty(&splice)) {
4661 root = list_first_entry(&splice, struct btrfs_root,
4662 delalloc_root);
4663 root = btrfs_grab_root(root);
4664 BUG_ON(!root);
4665 spin_unlock(&fs_info->delalloc_root_lock);
4666
4667 btrfs_destroy_delalloc_inodes(root);
4668 btrfs_put_root(root);
4669
4670 spin_lock(&fs_info->delalloc_root_lock);
4671 }
4672 spin_unlock(&fs_info->delalloc_root_lock);
4673}
4674
4675static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4676 struct extent_io_tree *dirty_pages,
4677 int mark)
4678{
4679 struct extent_buffer *eb;
4680 u64 start = 0;
4681 u64 end;
4682
4683 while (find_first_extent_bit(dirty_pages, start, &start, &end,
4684 mark, NULL)) {
4685 clear_extent_bits(dirty_pages, start, end, mark);
4686 while (start <= end) {
4687 eb = find_extent_buffer(fs_info, start);
4688 start += fs_info->nodesize;
4689 if (!eb)
4690 continue;
4691
4692 btrfs_tree_lock(eb);
4693 wait_on_extent_buffer_writeback(eb);
4694 btrfs_clear_buffer_dirty(NULL, eb);
4695 btrfs_tree_unlock(eb);
4696
4697 free_extent_buffer_stale(eb);
4698 }
4699 }
4700}
4701
4702static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4703 struct extent_io_tree *unpin)
4704{
4705 u64 start;
4706 u64 end;
4707
4708 while (1) {
4709 struct extent_state *cached_state = NULL;
4710
4711 /*
4712 * The btrfs_finish_extent_commit() may get the same range as
4713 * ours between find_first_extent_bit and clear_extent_dirty.
4714 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4715 * the same extent range.
4716 */
4717 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4718 if (!find_first_extent_bit(unpin, 0, &start, &end,
4719 EXTENT_DIRTY, &cached_state)) {
4720 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4721 break;
4722 }
4723
4724 clear_extent_dirty(unpin, start, end, &cached_state);
4725 free_extent_state(cached_state);
4726 btrfs_error_unpin_extent_range(fs_info, start, end);
4727 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4728 cond_resched();
4729 }
4730}
4731
4732static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4733{
4734 struct inode *inode;
4735
4736 inode = cache->io_ctl.inode;
4737 if (inode) {
4738 unsigned int nofs_flag;
4739
4740 nofs_flag = memalloc_nofs_save();
4741 invalidate_inode_pages2(inode->i_mapping);
4742 memalloc_nofs_restore(nofs_flag);
4743
4744 BTRFS_I(inode)->generation = 0;
4745 cache->io_ctl.inode = NULL;
4746 iput(inode);
4747 }
4748 ASSERT(cache->io_ctl.pages == NULL);
4749 btrfs_put_block_group(cache);
4750}
4751
4752void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4753 struct btrfs_fs_info *fs_info)
4754{
4755 struct btrfs_block_group *cache;
4756
4757 spin_lock(&cur_trans->dirty_bgs_lock);
4758 while (!list_empty(&cur_trans->dirty_bgs)) {
4759 cache = list_first_entry(&cur_trans->dirty_bgs,
4760 struct btrfs_block_group,
4761 dirty_list);
4762
4763 if (!list_empty(&cache->io_list)) {
4764 spin_unlock(&cur_trans->dirty_bgs_lock);
4765 list_del_init(&cache->io_list);
4766 btrfs_cleanup_bg_io(cache);
4767 spin_lock(&cur_trans->dirty_bgs_lock);
4768 }
4769
4770 list_del_init(&cache->dirty_list);
4771 spin_lock(&cache->lock);
4772 cache->disk_cache_state = BTRFS_DC_ERROR;
4773 spin_unlock(&cache->lock);
4774
4775 spin_unlock(&cur_trans->dirty_bgs_lock);
4776 btrfs_put_block_group(cache);
4777 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4778 spin_lock(&cur_trans->dirty_bgs_lock);
4779 }
4780 spin_unlock(&cur_trans->dirty_bgs_lock);
4781
4782 /*
4783 * Refer to the definition of io_bgs member for details why it's safe
4784 * to use it without any locking
4785 */
4786 while (!list_empty(&cur_trans->io_bgs)) {
4787 cache = list_first_entry(&cur_trans->io_bgs,
4788 struct btrfs_block_group,
4789 io_list);
4790
4791 list_del_init(&cache->io_list);
4792 spin_lock(&cache->lock);
4793 cache->disk_cache_state = BTRFS_DC_ERROR;
4794 spin_unlock(&cache->lock);
4795 btrfs_cleanup_bg_io(cache);
4796 }
4797}
4798
4799static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4800{
4801 struct btrfs_root *gang[8];
4802 int i;
4803 int ret;
4804
4805 spin_lock(&fs_info->fs_roots_radix_lock);
4806 while (1) {
4807 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4808 (void **)gang, 0,
4809 ARRAY_SIZE(gang),
4810 BTRFS_ROOT_TRANS_TAG);
4811 if (ret == 0)
4812 break;
4813 for (i = 0; i < ret; i++) {
4814 struct btrfs_root *root = gang[i];
4815
4816 btrfs_qgroup_free_meta_all_pertrans(root);
4817 radix_tree_tag_clear(&fs_info->fs_roots_radix,
4818 (unsigned long)root->root_key.objectid,
4819 BTRFS_ROOT_TRANS_TAG);
4820 }
4821 }
4822 spin_unlock(&fs_info->fs_roots_radix_lock);
4823}
4824
4825void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4826 struct btrfs_fs_info *fs_info)
4827{
4828 struct btrfs_device *dev, *tmp;
4829
4830 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4831 ASSERT(list_empty(&cur_trans->dirty_bgs));
4832 ASSERT(list_empty(&cur_trans->io_bgs));
4833
4834 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4835 post_commit_list) {
4836 list_del_init(&dev->post_commit_list);
4837 }
4838
4839 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4840
4841 cur_trans->state = TRANS_STATE_COMMIT_START;
4842 wake_up(&fs_info->transaction_blocked_wait);
4843
4844 cur_trans->state = TRANS_STATE_UNBLOCKED;
4845 wake_up(&fs_info->transaction_wait);
4846
4847 btrfs_destroy_delayed_inodes(fs_info);
4848
4849 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4850 EXTENT_DIRTY);
4851 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4852
4853 btrfs_free_all_qgroup_pertrans(fs_info);
4854
4855 cur_trans->state =TRANS_STATE_COMPLETED;
4856 wake_up(&cur_trans->commit_wait);
4857}
4858
4859static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4860{
4861 struct btrfs_transaction *t;
4862
4863 mutex_lock(&fs_info->transaction_kthread_mutex);
4864
4865 spin_lock(&fs_info->trans_lock);
4866 while (!list_empty(&fs_info->trans_list)) {
4867 t = list_first_entry(&fs_info->trans_list,
4868 struct btrfs_transaction, list);
4869 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4870 refcount_inc(&t->use_count);
4871 spin_unlock(&fs_info->trans_lock);
4872 btrfs_wait_for_commit(fs_info, t->transid);
4873 btrfs_put_transaction(t);
4874 spin_lock(&fs_info->trans_lock);
4875 continue;
4876 }
4877 if (t == fs_info->running_transaction) {
4878 t->state = TRANS_STATE_COMMIT_DOING;
4879 spin_unlock(&fs_info->trans_lock);
4880 /*
4881 * We wait for 0 num_writers since we don't hold a trans
4882 * handle open currently for this transaction.
4883 */
4884 wait_event(t->writer_wait,
4885 atomic_read(&t->num_writers) == 0);
4886 } else {
4887 spin_unlock(&fs_info->trans_lock);
4888 }
4889 btrfs_cleanup_one_transaction(t, fs_info);
4890
4891 spin_lock(&fs_info->trans_lock);
4892 if (t == fs_info->running_transaction)
4893 fs_info->running_transaction = NULL;
4894 list_del_init(&t->list);
4895 spin_unlock(&fs_info->trans_lock);
4896
4897 btrfs_put_transaction(t);
4898 trace_btrfs_transaction_commit(fs_info);
4899 spin_lock(&fs_info->trans_lock);
4900 }
4901 spin_unlock(&fs_info->trans_lock);
4902 btrfs_destroy_all_ordered_extents(fs_info);
4903 btrfs_destroy_delayed_inodes(fs_info);
4904 btrfs_assert_delayed_root_empty(fs_info);
4905 btrfs_destroy_all_delalloc_inodes(fs_info);
4906 btrfs_drop_all_logs(fs_info);
4907 mutex_unlock(&fs_info->transaction_kthread_mutex);
4908
4909 return 0;
4910}
4911
4912int btrfs_init_root_free_objectid(struct btrfs_root *root)
4913{
4914 struct btrfs_path *path;
4915 int ret;
4916 struct extent_buffer *l;
4917 struct btrfs_key search_key;
4918 struct btrfs_key found_key;
4919 int slot;
4920
4921 path = btrfs_alloc_path();
4922 if (!path)
4923 return -ENOMEM;
4924
4925 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4926 search_key.type = -1;
4927 search_key.offset = (u64)-1;
4928 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4929 if (ret < 0)
4930 goto error;
4931 if (ret == 0) {
4932 /*
4933 * Key with offset -1 found, there would have to exist a root
4934 * with such id, but this is out of valid range.
4935 */
4936 ret = -EUCLEAN;
4937 goto error;
4938 }
4939 if (path->slots[0] > 0) {
4940 slot = path->slots[0] - 1;
4941 l = path->nodes[0];
4942 btrfs_item_key_to_cpu(l, &found_key, slot);
4943 root->free_objectid = max_t(u64, found_key.objectid + 1,
4944 BTRFS_FIRST_FREE_OBJECTID);
4945 } else {
4946 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4947 }
4948 ret = 0;
4949error:
4950 btrfs_free_path(path);
4951 return ret;
4952}
4953
4954int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4955{
4956 int ret;
4957 mutex_lock(&root->objectid_mutex);
4958
4959 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4960 btrfs_warn(root->fs_info,
4961 "the objectid of root %llu reaches its highest value",
4962 root->root_key.objectid);
4963 ret = -ENOSPC;
4964 goto out;
4965 }
4966
4967 *objectid = root->free_objectid++;
4968 ret = 0;
4969out:
4970 mutex_unlock(&root->objectid_mutex);
4971 return ret;
4972}
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 "inode-map.h"
33#include "check-integrity.h"
34#include "rcu-string.h"
35#include "dev-replace.h"
36#include "raid56.h"
37#include "sysfs.h"
38#include "qgroup.h"
39#include "compression.h"
40#include "tree-checker.h"
41#include "ref-verify.h"
42#include "block-group.h"
43#include "discard.h"
44#include "space-info.h"
45
46#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
52
53static const struct extent_io_ops btree_extent_io_ops;
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 void *private_data;
114 struct bio *bio;
115 extent_submit_bio_start_t *submit_bio_start;
116 int mirror_num;
117 /*
118 * bio_offset is optional, can be used if the pages in the bio
119 * can't tell us where in the file the bio should go
120 */
121 u64 bio_offset;
122 struct btrfs_work work;
123 blk_status_t status;
124};
125
126/*
127 * Lockdep class keys for extent_buffer->lock's in this root. For a given
128 * eb, the lockdep key is determined by the btrfs_root it belongs to and
129 * the level the eb occupies in the tree.
130 *
131 * Different roots are used for different purposes and may nest inside each
132 * other and they require separate keysets. As lockdep keys should be
133 * static, assign keysets according to the purpose of the root as indicated
134 * by btrfs_root->root_key.objectid. This ensures that all special purpose
135 * roots have separate keysets.
136 *
137 * Lock-nesting across peer nodes is always done with the immediate parent
138 * node locked thus preventing deadlock. As lockdep doesn't know this, use
139 * subclass to avoid triggering lockdep warning in such cases.
140 *
141 * The key is set by the readpage_end_io_hook after the buffer has passed
142 * csum validation but before the pages are unlocked. It is also set by
143 * btrfs_init_new_buffer on freshly allocated blocks.
144 *
145 * We also add a check to make sure the highest level of the tree is the
146 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
147 * needs update as well.
148 */
149#ifdef CONFIG_DEBUG_LOCK_ALLOC
150# if BTRFS_MAX_LEVEL != 8
151# error
152# endif
153
154static struct btrfs_lockdep_keyset {
155 u64 id; /* root objectid */
156 const char *name_stem; /* lock name stem */
157 char names[BTRFS_MAX_LEVEL + 1][20];
158 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
159} btrfs_lockdep_keysets[] = {
160 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
161 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
162 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
163 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
164 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
165 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
166 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
167 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
168 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
169 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
170 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
171 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
172 { .id = 0, .name_stem = "tree" },
173};
174
175void __init btrfs_init_lockdep(void)
176{
177 int i, j;
178
179 /* initialize lockdep class names */
180 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
181 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
182
183 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
184 snprintf(ks->names[j], sizeof(ks->names[j]),
185 "btrfs-%s-%02d", ks->name_stem, j);
186 }
187}
188
189void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
190 int level)
191{
192 struct btrfs_lockdep_keyset *ks;
193
194 BUG_ON(level >= ARRAY_SIZE(ks->keys));
195
196 /* find the matching keyset, id 0 is the default entry */
197 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
198 if (ks->id == objectid)
199 break;
200
201 lockdep_set_class_and_name(&eb->lock,
202 &ks->keys[level], ks->names[level]);
203}
204
205#endif
206
207/*
208 * extents on the btree inode are pretty simple, there's one extent
209 * that covers the entire device
210 */
211struct extent_map *btree_get_extent(struct btrfs_inode *inode,
212 struct page *page, size_t pg_offset,
213 u64 start, u64 len)
214{
215 struct extent_map_tree *em_tree = &inode->extent_tree;
216 struct extent_map *em;
217 int ret;
218
219 read_lock(&em_tree->lock);
220 em = lookup_extent_mapping(em_tree, start, len);
221 if (em) {
222 read_unlock(&em_tree->lock);
223 goto out;
224 }
225 read_unlock(&em_tree->lock);
226
227 em = alloc_extent_map();
228 if (!em) {
229 em = ERR_PTR(-ENOMEM);
230 goto out;
231 }
232 em->start = 0;
233 em->len = (u64)-1;
234 em->block_len = (u64)-1;
235 em->block_start = 0;
236
237 write_lock(&em_tree->lock);
238 ret = add_extent_mapping(em_tree, em, 0);
239 if (ret == -EEXIST) {
240 free_extent_map(em);
241 em = lookup_extent_mapping(em_tree, start, len);
242 if (!em)
243 em = ERR_PTR(-EIO);
244 } else if (ret) {
245 free_extent_map(em);
246 em = ERR_PTR(ret);
247 }
248 write_unlock(&em_tree->lock);
249
250out:
251 return em;
252}
253
254/*
255 * Compute the csum of a btree block and store the result to provided buffer.
256 */
257static void csum_tree_block(struct extent_buffer *buf, u8 *result)
258{
259 struct btrfs_fs_info *fs_info = buf->fs_info;
260 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
261 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
262 char *kaddr;
263 int i;
264
265 shash->tfm = fs_info->csum_shash;
266 crypto_shash_init(shash);
267 kaddr = page_address(buf->pages[0]);
268 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
269 PAGE_SIZE - BTRFS_CSUM_SIZE);
270
271 for (i = 1; i < num_pages; i++) {
272 kaddr = page_address(buf->pages[i]);
273 crypto_shash_update(shash, kaddr, PAGE_SIZE);
274 }
275 memset(result, 0, BTRFS_CSUM_SIZE);
276 crypto_shash_final(shash, result);
277}
278
279/*
280 * we can't consider a given block up to date unless the transid of the
281 * block matches the transid in the parent node's pointer. This is how we
282 * detect blocks that either didn't get written at all or got written
283 * in the wrong place.
284 */
285static int verify_parent_transid(struct extent_io_tree *io_tree,
286 struct extent_buffer *eb, u64 parent_transid,
287 int atomic)
288{
289 struct extent_state *cached_state = NULL;
290 int ret;
291 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
292
293 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
294 return 0;
295
296 if (atomic)
297 return -EAGAIN;
298
299 if (need_lock) {
300 btrfs_tree_read_lock(eb);
301 btrfs_set_lock_blocking_read(eb);
302 }
303
304 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
305 &cached_state);
306 if (extent_buffer_uptodate(eb) &&
307 btrfs_header_generation(eb) == parent_transid) {
308 ret = 0;
309 goto out;
310 }
311 btrfs_err_rl(eb->fs_info,
312 "parent transid verify failed on %llu wanted %llu found %llu",
313 eb->start,
314 parent_transid, btrfs_header_generation(eb));
315 ret = 1;
316
317 /*
318 * Things reading via commit roots that don't have normal protection,
319 * like send, can have a really old block in cache that may point at a
320 * block that has been freed and re-allocated. So don't clear uptodate
321 * if we find an eb that is under IO (dirty/writeback) because we could
322 * end up reading in the stale data and then writing it back out and
323 * making everybody very sad.
324 */
325 if (!extent_buffer_under_io(eb))
326 clear_extent_buffer_uptodate(eb);
327out:
328 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
329 &cached_state);
330 if (need_lock)
331 btrfs_tree_read_unlock_blocking(eb);
332 return ret;
333}
334
335static bool btrfs_supported_super_csum(u16 csum_type)
336{
337 switch (csum_type) {
338 case BTRFS_CSUM_TYPE_CRC32:
339 case BTRFS_CSUM_TYPE_XXHASH:
340 case BTRFS_CSUM_TYPE_SHA256:
341 case BTRFS_CSUM_TYPE_BLAKE2:
342 return true;
343 default:
344 return false;
345 }
346}
347
348/*
349 * Return 0 if the superblock checksum type matches the checksum value of that
350 * algorithm. Pass the raw disk superblock data.
351 */
352static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
353 char *raw_disk_sb)
354{
355 struct btrfs_super_block *disk_sb =
356 (struct btrfs_super_block *)raw_disk_sb;
357 char result[BTRFS_CSUM_SIZE];
358 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
359
360 shash->tfm = fs_info->csum_shash;
361
362 /*
363 * The super_block structure does not span the whole
364 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
365 * filled with zeros and is included in the checksum.
366 */
367 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
368 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
369
370 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
371 return 1;
372
373 return 0;
374}
375
376int btrfs_verify_level_key(struct extent_buffer *eb, int level,
377 struct btrfs_key *first_key, u64 parent_transid)
378{
379 struct btrfs_fs_info *fs_info = eb->fs_info;
380 int found_level;
381 struct btrfs_key found_key;
382 int ret;
383
384 found_level = btrfs_header_level(eb);
385 if (found_level != level) {
386 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
387 KERN_ERR "BTRFS: tree level check failed\n");
388 btrfs_err(fs_info,
389"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
390 eb->start, level, found_level);
391 return -EIO;
392 }
393
394 if (!first_key)
395 return 0;
396
397 /*
398 * For live tree block (new tree blocks in current transaction),
399 * we need proper lock context to avoid race, which is impossible here.
400 * So we only checks tree blocks which is read from disk, whose
401 * generation <= fs_info->last_trans_committed.
402 */
403 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
404 return 0;
405
406 /* We have @first_key, so this @eb must have at least one item */
407 if (btrfs_header_nritems(eb) == 0) {
408 btrfs_err(fs_info,
409 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
410 eb->start);
411 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
412 return -EUCLEAN;
413 }
414
415 if (found_level)
416 btrfs_node_key_to_cpu(eb, &found_key, 0);
417 else
418 btrfs_item_key_to_cpu(eb, &found_key, 0);
419 ret = btrfs_comp_cpu_keys(first_key, &found_key);
420
421 if (ret) {
422 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
423 KERN_ERR "BTRFS: tree first key check failed\n");
424 btrfs_err(fs_info,
425"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
426 eb->start, parent_transid, first_key->objectid,
427 first_key->type, first_key->offset,
428 found_key.objectid, found_key.type,
429 found_key.offset);
430 }
431 return ret;
432}
433
434/*
435 * helper to read a given tree block, doing retries as required when
436 * the checksums don't match and we have alternate mirrors to try.
437 *
438 * @parent_transid: expected transid, skip check if 0
439 * @level: expected level, mandatory check
440 * @first_key: expected key of first slot, skip check if NULL
441 */
442static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
443 u64 parent_transid, int level,
444 struct btrfs_key *first_key)
445{
446 struct btrfs_fs_info *fs_info = eb->fs_info;
447 struct extent_io_tree *io_tree;
448 int failed = 0;
449 int ret;
450 int num_copies = 0;
451 int mirror_num = 0;
452 int failed_mirror = 0;
453
454 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
455 while (1) {
456 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
457 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
458 if (!ret) {
459 if (verify_parent_transid(io_tree, eb,
460 parent_transid, 0))
461 ret = -EIO;
462 else if (btrfs_verify_level_key(eb, level,
463 first_key, parent_transid))
464 ret = -EUCLEAN;
465 else
466 break;
467 }
468
469 num_copies = btrfs_num_copies(fs_info,
470 eb->start, eb->len);
471 if (num_copies == 1)
472 break;
473
474 if (!failed_mirror) {
475 failed = 1;
476 failed_mirror = eb->read_mirror;
477 }
478
479 mirror_num++;
480 if (mirror_num == failed_mirror)
481 mirror_num++;
482
483 if (mirror_num > num_copies)
484 break;
485 }
486
487 if (failed && !ret && failed_mirror)
488 btrfs_repair_eb_io_failure(eb, failed_mirror);
489
490 return ret;
491}
492
493/*
494 * checksum a dirty tree block before IO. This has extra checks to make sure
495 * we only fill in the checksum field in the first page of a multi-page block
496 */
497
498static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
499{
500 u64 start = page_offset(page);
501 u64 found_start;
502 u8 result[BTRFS_CSUM_SIZE];
503 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
504 struct extent_buffer *eb;
505 int ret;
506
507 eb = (struct extent_buffer *)page->private;
508 if (page != eb->pages[0])
509 return 0;
510
511 found_start = btrfs_header_bytenr(eb);
512 /*
513 * Please do not consolidate these warnings into a single if.
514 * It is useful to know what went wrong.
515 */
516 if (WARN_ON(found_start != start))
517 return -EUCLEAN;
518 if (WARN_ON(!PageUptodate(page)))
519 return -EUCLEAN;
520
521 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
522 offsetof(struct btrfs_header, fsid),
523 BTRFS_FSID_SIZE) == 0);
524
525 csum_tree_block(eb, result);
526
527 if (btrfs_header_level(eb))
528 ret = btrfs_check_node(eb);
529 else
530 ret = btrfs_check_leaf_full(eb);
531
532 if (ret < 0) {
533 btrfs_print_tree(eb, 0);
534 btrfs_err(fs_info,
535 "block=%llu write time tree block corruption detected",
536 eb->start);
537 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
538 return ret;
539 }
540 write_extent_buffer(eb, result, 0, csum_size);
541
542 return 0;
543}
544
545static int check_tree_block_fsid(struct extent_buffer *eb)
546{
547 struct btrfs_fs_info *fs_info = eb->fs_info;
548 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
549 u8 fsid[BTRFS_FSID_SIZE];
550 int ret = 1;
551
552 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
553 BTRFS_FSID_SIZE);
554 while (fs_devices) {
555 u8 *metadata_uuid;
556
557 /*
558 * Checking the incompat flag is only valid for the current
559 * fs. For seed devices it's forbidden to have their uuid
560 * changed so reading ->fsid in this case is fine
561 */
562 if (fs_devices == fs_info->fs_devices &&
563 btrfs_fs_incompat(fs_info, METADATA_UUID))
564 metadata_uuid = fs_devices->metadata_uuid;
565 else
566 metadata_uuid = fs_devices->fsid;
567
568 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
569 ret = 0;
570 break;
571 }
572 fs_devices = fs_devices->seed;
573 }
574 return ret;
575}
576
577static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
578 u64 phy_offset, struct page *page,
579 u64 start, u64 end, int mirror)
580{
581 u64 found_start;
582 int found_level;
583 struct extent_buffer *eb;
584 struct btrfs_fs_info *fs_info;
585 u16 csum_size;
586 int ret = 0;
587 u8 result[BTRFS_CSUM_SIZE];
588 int reads_done;
589
590 if (!page->private)
591 goto out;
592
593 eb = (struct extent_buffer *)page->private;
594 fs_info = eb->fs_info;
595 csum_size = btrfs_super_csum_size(fs_info->super_copy);
596
597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
599 */
600 atomic_inc(&eb->refs);
601
602 reads_done = atomic_dec_and_test(&eb->io_pages);
603 if (!reads_done)
604 goto err;
605
606 eb->read_mirror = mirror;
607 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
608 ret = -EIO;
609 goto err;
610 }
611
612 found_start = btrfs_header_bytenr(eb);
613 if (found_start != eb->start) {
614 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
615 eb->start, found_start);
616 ret = -EIO;
617 goto err;
618 }
619 if (check_tree_block_fsid(eb)) {
620 btrfs_err_rl(fs_info, "bad fsid on block %llu",
621 eb->start);
622 ret = -EIO;
623 goto err;
624 }
625 found_level = btrfs_header_level(eb);
626 if (found_level >= BTRFS_MAX_LEVEL) {
627 btrfs_err(fs_info, "bad tree block level %d on %llu",
628 (int)btrfs_header_level(eb), eb->start);
629 ret = -EIO;
630 goto err;
631 }
632
633 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
634 eb, found_level);
635
636 csum_tree_block(eb, result);
637
638 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
639 u8 val[BTRFS_CSUM_SIZE] = { 0 };
640
641 read_extent_buffer(eb, &val, 0, csum_size);
642 btrfs_warn_rl(fs_info,
643 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
644 fs_info->sb->s_id, eb->start,
645 CSUM_FMT_VALUE(csum_size, val),
646 CSUM_FMT_VALUE(csum_size, result),
647 btrfs_header_level(eb));
648 ret = -EUCLEAN;
649 goto err;
650 }
651
652 /*
653 * If this is a leaf block and it is corrupt, set the corrupt bit so
654 * that we don't try and read the other copies of this block, just
655 * return -EIO.
656 */
657 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
658 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
659 ret = -EIO;
660 }
661
662 if (found_level > 0 && btrfs_check_node(eb))
663 ret = -EIO;
664
665 if (!ret)
666 set_extent_buffer_uptodate(eb);
667 else
668 btrfs_err(fs_info,
669 "block=%llu read time tree block corruption detected",
670 eb->start);
671err:
672 if (reads_done &&
673 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
674 btree_readahead_hook(eb, ret);
675
676 if (ret) {
677 /*
678 * our io error hook is going to dec the io pages
679 * again, we have to make sure it has something
680 * to decrement
681 */
682 atomic_inc(&eb->io_pages);
683 clear_extent_buffer_uptodate(eb);
684 }
685 free_extent_buffer(eb);
686out:
687 return ret;
688}
689
690static void end_workqueue_bio(struct bio *bio)
691{
692 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
693 struct btrfs_fs_info *fs_info;
694 struct btrfs_workqueue *wq;
695
696 fs_info = end_io_wq->info;
697 end_io_wq->status = bio->bi_status;
698
699 if (bio_op(bio) == REQ_OP_WRITE) {
700 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
701 wq = fs_info->endio_meta_write_workers;
702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
703 wq = fs_info->endio_freespace_worker;
704 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
705 wq = fs_info->endio_raid56_workers;
706 else
707 wq = fs_info->endio_write_workers;
708 } else {
709 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
710 wq = fs_info->endio_raid56_workers;
711 else if (end_io_wq->metadata)
712 wq = fs_info->endio_meta_workers;
713 else
714 wq = fs_info->endio_workers;
715 }
716
717 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
718 btrfs_queue_work(wq, &end_io_wq->work);
719}
720
721blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
722 enum btrfs_wq_endio_type metadata)
723{
724 struct btrfs_end_io_wq *end_io_wq;
725
726 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
727 if (!end_io_wq)
728 return BLK_STS_RESOURCE;
729
730 end_io_wq->private = bio->bi_private;
731 end_io_wq->end_io = bio->bi_end_io;
732 end_io_wq->info = info;
733 end_io_wq->status = 0;
734 end_io_wq->bio = bio;
735 end_io_wq->metadata = metadata;
736
737 bio->bi_private = end_io_wq;
738 bio->bi_end_io = end_workqueue_bio;
739 return 0;
740}
741
742static void run_one_async_start(struct btrfs_work *work)
743{
744 struct async_submit_bio *async;
745 blk_status_t ret;
746
747 async = container_of(work, struct async_submit_bio, work);
748 ret = async->submit_bio_start(async->private_data, async->bio,
749 async->bio_offset);
750 if (ret)
751 async->status = ret;
752}
753
754/*
755 * In order to insert checksums into the metadata in large chunks, we wait
756 * until bio submission time. All the pages in the bio are checksummed and
757 * sums are attached onto the ordered extent record.
758 *
759 * At IO completion time the csums attached on the ordered extent record are
760 * inserted into the tree.
761 */
762static void run_one_async_done(struct btrfs_work *work)
763{
764 struct async_submit_bio *async;
765 struct inode *inode;
766 blk_status_t ret;
767
768 async = container_of(work, struct async_submit_bio, work);
769 inode = async->private_data;
770
771 /* If an error occurred we just want to clean up the bio and move on */
772 if (async->status) {
773 async->bio->bi_status = async->status;
774 bio_endio(async->bio);
775 return;
776 }
777
778 /*
779 * All of the bios that pass through here are from async helpers.
780 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
781 * This changes nothing when cgroups aren't in use.
782 */
783 async->bio->bi_opf |= REQ_CGROUP_PUNT;
784 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
785 if (ret) {
786 async->bio->bi_status = ret;
787 bio_endio(async->bio);
788 }
789}
790
791static void run_one_async_free(struct btrfs_work *work)
792{
793 struct async_submit_bio *async;
794
795 async = container_of(work, struct async_submit_bio, work);
796 kfree(async);
797}
798
799blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
800 int mirror_num, unsigned long bio_flags,
801 u64 bio_offset, void *private_data,
802 extent_submit_bio_start_t *submit_bio_start)
803{
804 struct async_submit_bio *async;
805
806 async = kmalloc(sizeof(*async), GFP_NOFS);
807 if (!async)
808 return BLK_STS_RESOURCE;
809
810 async->private_data = private_data;
811 async->bio = bio;
812 async->mirror_num = mirror_num;
813 async->submit_bio_start = submit_bio_start;
814
815 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
816 run_one_async_free);
817
818 async->bio_offset = bio_offset;
819
820 async->status = 0;
821
822 if (op_is_sync(bio->bi_opf))
823 btrfs_set_work_high_priority(&async->work);
824
825 btrfs_queue_work(fs_info->workers, &async->work);
826 return 0;
827}
828
829static blk_status_t btree_csum_one_bio(struct bio *bio)
830{
831 struct bio_vec *bvec;
832 struct btrfs_root *root;
833 int ret = 0;
834 struct bvec_iter_all iter_all;
835
836 ASSERT(!bio_flagged(bio, BIO_CLONED));
837 bio_for_each_segment_all(bvec, bio, iter_all) {
838 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
839 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
840 if (ret)
841 break;
842 }
843
844 return errno_to_blk_status(ret);
845}
846
847static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
848 u64 bio_offset)
849{
850 /*
851 * when we're called for a write, we're already in the async
852 * submission context. Just jump into btrfs_map_bio
853 */
854 return btree_csum_one_bio(bio);
855}
856
857static int check_async_write(struct btrfs_fs_info *fs_info,
858 struct btrfs_inode *bi)
859{
860 if (atomic_read(&bi->sync_writers))
861 return 0;
862 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
863 return 0;
864 return 1;
865}
866
867static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
868 int mirror_num,
869 unsigned long bio_flags)
870{
871 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
872 int async = check_async_write(fs_info, BTRFS_I(inode));
873 blk_status_t ret;
874
875 if (bio_op(bio) != REQ_OP_WRITE) {
876 /*
877 * called for a read, do the setup so that checksum validation
878 * can happen in the async kernel threads
879 */
880 ret = btrfs_bio_wq_end_io(fs_info, bio,
881 BTRFS_WQ_ENDIO_METADATA);
882 if (ret)
883 goto out_w_error;
884 ret = btrfs_map_bio(fs_info, bio, mirror_num);
885 } else if (!async) {
886 ret = btree_csum_one_bio(bio);
887 if (ret)
888 goto out_w_error;
889 ret = btrfs_map_bio(fs_info, bio, mirror_num);
890 } else {
891 /*
892 * kthread helpers are used to submit writes so that
893 * checksumming can happen in parallel across all CPUs
894 */
895 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
896 0, inode, btree_submit_bio_start);
897 }
898
899 if (ret)
900 goto out_w_error;
901 return 0;
902
903out_w_error:
904 bio->bi_status = ret;
905 bio_endio(bio);
906 return ret;
907}
908
909#ifdef CONFIG_MIGRATION
910static int btree_migratepage(struct address_space *mapping,
911 struct page *newpage, struct page *page,
912 enum migrate_mode mode)
913{
914 /*
915 * we can't safely write a btree page from here,
916 * we haven't done the locking hook
917 */
918 if (PageDirty(page))
919 return -EAGAIN;
920 /*
921 * Buffers may be managed in a filesystem specific way.
922 * We must have no buffers or drop them.
923 */
924 if (page_has_private(page) &&
925 !try_to_release_page(page, GFP_KERNEL))
926 return -EAGAIN;
927 return migrate_page(mapping, newpage, page, mode);
928}
929#endif
930
931
932static int btree_writepages(struct address_space *mapping,
933 struct writeback_control *wbc)
934{
935 struct btrfs_fs_info *fs_info;
936 int ret;
937
938 if (wbc->sync_mode == WB_SYNC_NONE) {
939
940 if (wbc->for_kupdate)
941 return 0;
942
943 fs_info = BTRFS_I(mapping->host)->root->fs_info;
944 /* this is a bit racy, but that's ok */
945 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
946 BTRFS_DIRTY_METADATA_THRESH,
947 fs_info->dirty_metadata_batch);
948 if (ret < 0)
949 return 0;
950 }
951 return btree_write_cache_pages(mapping, wbc);
952}
953
954static int btree_readpage(struct file *file, struct page *page)
955{
956 return extent_read_full_page(page, btree_get_extent, 0);
957}
958
959static int btree_releasepage(struct page *page, gfp_t gfp_flags)
960{
961 if (PageWriteback(page) || PageDirty(page))
962 return 0;
963
964 return try_release_extent_buffer(page);
965}
966
967static void btree_invalidatepage(struct page *page, unsigned int offset,
968 unsigned int length)
969{
970 struct extent_io_tree *tree;
971 tree = &BTRFS_I(page->mapping->host)->io_tree;
972 extent_invalidatepage(tree, page, offset);
973 btree_releasepage(page, GFP_NOFS);
974 if (PagePrivate(page)) {
975 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
976 "page private not zero on page %llu",
977 (unsigned long long)page_offset(page));
978 detach_page_private(page);
979 }
980}
981
982static int btree_set_page_dirty(struct page *page)
983{
984#ifdef DEBUG
985 struct extent_buffer *eb;
986
987 BUG_ON(!PagePrivate(page));
988 eb = (struct extent_buffer *)page->private;
989 BUG_ON(!eb);
990 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
991 BUG_ON(!atomic_read(&eb->refs));
992 btrfs_assert_tree_locked(eb);
993#endif
994 return __set_page_dirty_nobuffers(page);
995}
996
997static const struct address_space_operations btree_aops = {
998 .readpage = btree_readpage,
999 .writepages = btree_writepages,
1000 .releasepage = btree_releasepage,
1001 .invalidatepage = btree_invalidatepage,
1002#ifdef CONFIG_MIGRATION
1003 .migratepage = btree_migratepage,
1004#endif
1005 .set_page_dirty = btree_set_page_dirty,
1006};
1007
1008void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1009{
1010 struct extent_buffer *buf = NULL;
1011 int ret;
1012
1013 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1014 if (IS_ERR(buf))
1015 return;
1016
1017 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1018 if (ret < 0)
1019 free_extent_buffer_stale(buf);
1020 else
1021 free_extent_buffer(buf);
1022}
1023
1024struct extent_buffer *btrfs_find_create_tree_block(
1025 struct btrfs_fs_info *fs_info,
1026 u64 bytenr)
1027{
1028 if (btrfs_is_testing(fs_info))
1029 return alloc_test_extent_buffer(fs_info, bytenr);
1030 return alloc_extent_buffer(fs_info, bytenr);
1031}
1032
1033/*
1034 * Read tree block at logical address @bytenr and do variant basic but critical
1035 * verification.
1036 *
1037 * @parent_transid: expected transid of this tree block, skip check if 0
1038 * @level: expected level, mandatory check
1039 * @first_key: expected key in slot 0, skip check if NULL
1040 */
1041struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1042 u64 parent_transid, int level,
1043 struct btrfs_key *first_key)
1044{
1045 struct extent_buffer *buf = NULL;
1046 int ret;
1047
1048 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1049 if (IS_ERR(buf))
1050 return buf;
1051
1052 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1053 level, first_key);
1054 if (ret) {
1055 free_extent_buffer_stale(buf);
1056 return ERR_PTR(ret);
1057 }
1058 return buf;
1059
1060}
1061
1062void btrfs_clean_tree_block(struct extent_buffer *buf)
1063{
1064 struct btrfs_fs_info *fs_info = buf->fs_info;
1065 if (btrfs_header_generation(buf) ==
1066 fs_info->running_transaction->transid) {
1067 btrfs_assert_tree_locked(buf);
1068
1069 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1070 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1071 -buf->len,
1072 fs_info->dirty_metadata_batch);
1073 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1074 btrfs_set_lock_blocking_write(buf);
1075 clear_extent_buffer_dirty(buf);
1076 }
1077 }
1078}
1079
1080static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1081 u64 objectid)
1082{
1083 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1084 root->fs_info = fs_info;
1085 root->node = NULL;
1086 root->commit_root = NULL;
1087 root->state = 0;
1088 root->orphan_cleanup_state = 0;
1089
1090 root->last_trans = 0;
1091 root->highest_objectid = 0;
1092 root->nr_delalloc_inodes = 0;
1093 root->nr_ordered_extents = 0;
1094 root->inode_tree = RB_ROOT;
1095 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1096 root->block_rsv = NULL;
1097
1098 INIT_LIST_HEAD(&root->dirty_list);
1099 INIT_LIST_HEAD(&root->root_list);
1100 INIT_LIST_HEAD(&root->delalloc_inodes);
1101 INIT_LIST_HEAD(&root->delalloc_root);
1102 INIT_LIST_HEAD(&root->ordered_extents);
1103 INIT_LIST_HEAD(&root->ordered_root);
1104 INIT_LIST_HEAD(&root->reloc_dirty_list);
1105 INIT_LIST_HEAD(&root->logged_list[0]);
1106 INIT_LIST_HEAD(&root->logged_list[1]);
1107 spin_lock_init(&root->inode_lock);
1108 spin_lock_init(&root->delalloc_lock);
1109 spin_lock_init(&root->ordered_extent_lock);
1110 spin_lock_init(&root->accounting_lock);
1111 spin_lock_init(&root->log_extents_lock[0]);
1112 spin_lock_init(&root->log_extents_lock[1]);
1113 spin_lock_init(&root->qgroup_meta_rsv_lock);
1114 mutex_init(&root->objectid_mutex);
1115 mutex_init(&root->log_mutex);
1116 mutex_init(&root->ordered_extent_mutex);
1117 mutex_init(&root->delalloc_mutex);
1118 init_waitqueue_head(&root->qgroup_flush_wait);
1119 init_waitqueue_head(&root->log_writer_wait);
1120 init_waitqueue_head(&root->log_commit_wait[0]);
1121 init_waitqueue_head(&root->log_commit_wait[1]);
1122 INIT_LIST_HEAD(&root->log_ctxs[0]);
1123 INIT_LIST_HEAD(&root->log_ctxs[1]);
1124 atomic_set(&root->log_commit[0], 0);
1125 atomic_set(&root->log_commit[1], 0);
1126 atomic_set(&root->log_writers, 0);
1127 atomic_set(&root->log_batch, 0);
1128 refcount_set(&root->refs, 1);
1129 atomic_set(&root->snapshot_force_cow, 0);
1130 atomic_set(&root->nr_swapfiles, 0);
1131 root->log_transid = 0;
1132 root->log_transid_committed = -1;
1133 root->last_log_commit = 0;
1134 if (!dummy) {
1135 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1136 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1137 extent_io_tree_init(fs_info, &root->log_csum_range,
1138 IO_TREE_LOG_CSUM_RANGE, NULL);
1139 }
1140
1141 memset(&root->root_key, 0, sizeof(root->root_key));
1142 memset(&root->root_item, 0, sizeof(root->root_item));
1143 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1144 root->root_key.objectid = objectid;
1145 root->anon_dev = 0;
1146
1147 spin_lock_init(&root->root_item_lock);
1148 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1149#ifdef CONFIG_BTRFS_DEBUG
1150 INIT_LIST_HEAD(&root->leak_list);
1151 spin_lock(&fs_info->fs_roots_radix_lock);
1152 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1153 spin_unlock(&fs_info->fs_roots_radix_lock);
1154#endif
1155}
1156
1157static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1158 u64 objectid, gfp_t flags)
1159{
1160 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1161 if (root)
1162 __setup_root(root, fs_info, objectid);
1163 return root;
1164}
1165
1166#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1167/* Should only be used by the testing infrastructure */
1168struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1169{
1170 struct btrfs_root *root;
1171
1172 if (!fs_info)
1173 return ERR_PTR(-EINVAL);
1174
1175 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1176 if (!root)
1177 return ERR_PTR(-ENOMEM);
1178
1179 /* We don't use the stripesize in selftest, set it as sectorsize */
1180 root->alloc_bytenr = 0;
1181
1182 return root;
1183}
1184#endif
1185
1186struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1187 u64 objectid)
1188{
1189 struct btrfs_fs_info *fs_info = trans->fs_info;
1190 struct extent_buffer *leaf;
1191 struct btrfs_root *tree_root = fs_info->tree_root;
1192 struct btrfs_root *root;
1193 struct btrfs_key key;
1194 unsigned int nofs_flag;
1195 int ret = 0;
1196
1197 /*
1198 * We're holding a transaction handle, so use a NOFS memory allocation
1199 * context to avoid deadlock if reclaim happens.
1200 */
1201 nofs_flag = memalloc_nofs_save();
1202 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1203 memalloc_nofs_restore(nofs_flag);
1204 if (!root)
1205 return ERR_PTR(-ENOMEM);
1206
1207 root->root_key.objectid = objectid;
1208 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1209 root->root_key.offset = 0;
1210
1211 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1212 if (IS_ERR(leaf)) {
1213 ret = PTR_ERR(leaf);
1214 leaf = NULL;
1215 goto fail;
1216 }
1217
1218 root->node = leaf;
1219 btrfs_mark_buffer_dirty(leaf);
1220
1221 root->commit_root = btrfs_root_node(root);
1222 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1223
1224 root->root_item.flags = 0;
1225 root->root_item.byte_limit = 0;
1226 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1227 btrfs_set_root_generation(&root->root_item, trans->transid);
1228 btrfs_set_root_level(&root->root_item, 0);
1229 btrfs_set_root_refs(&root->root_item, 1);
1230 btrfs_set_root_used(&root->root_item, leaf->len);
1231 btrfs_set_root_last_snapshot(&root->root_item, 0);
1232 btrfs_set_root_dirid(&root->root_item, 0);
1233 if (is_fstree(objectid))
1234 generate_random_guid(root->root_item.uuid);
1235 else
1236 export_guid(root->root_item.uuid, &guid_null);
1237 root->root_item.drop_level = 0;
1238
1239 key.objectid = objectid;
1240 key.type = BTRFS_ROOT_ITEM_KEY;
1241 key.offset = 0;
1242 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1243 if (ret)
1244 goto fail;
1245
1246 btrfs_tree_unlock(leaf);
1247
1248 return root;
1249
1250fail:
1251 if (leaf)
1252 btrfs_tree_unlock(leaf);
1253 btrfs_put_root(root);
1254
1255 return ERR_PTR(ret);
1256}
1257
1258static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1259 struct btrfs_fs_info *fs_info)
1260{
1261 struct btrfs_root *root;
1262 struct extent_buffer *leaf;
1263
1264 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1265 if (!root)
1266 return ERR_PTR(-ENOMEM);
1267
1268 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1269 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1270 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1271
1272 /*
1273 * DON'T set SHAREABLE bit for log trees.
1274 *
1275 * Log trees are not exposed to user space thus can't be snapshotted,
1276 * and they go away before a real commit is actually done.
1277 *
1278 * They do store pointers to file data extents, and those reference
1279 * counts still get updated (along with back refs to the log tree).
1280 */
1281
1282 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1283 NULL, 0, 0, 0);
1284 if (IS_ERR(leaf)) {
1285 btrfs_put_root(root);
1286 return ERR_CAST(leaf);
1287 }
1288
1289 root->node = leaf;
1290
1291 btrfs_mark_buffer_dirty(root->node);
1292 btrfs_tree_unlock(root->node);
1293 return root;
1294}
1295
1296int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1297 struct btrfs_fs_info *fs_info)
1298{
1299 struct btrfs_root *log_root;
1300
1301 log_root = alloc_log_tree(trans, fs_info);
1302 if (IS_ERR(log_root))
1303 return PTR_ERR(log_root);
1304 WARN_ON(fs_info->log_root_tree);
1305 fs_info->log_root_tree = log_root;
1306 return 0;
1307}
1308
1309int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1310 struct btrfs_root *root)
1311{
1312 struct btrfs_fs_info *fs_info = root->fs_info;
1313 struct btrfs_root *log_root;
1314 struct btrfs_inode_item *inode_item;
1315
1316 log_root = alloc_log_tree(trans, fs_info);
1317 if (IS_ERR(log_root))
1318 return PTR_ERR(log_root);
1319
1320 log_root->last_trans = trans->transid;
1321 log_root->root_key.offset = root->root_key.objectid;
1322
1323 inode_item = &log_root->root_item.inode;
1324 btrfs_set_stack_inode_generation(inode_item, 1);
1325 btrfs_set_stack_inode_size(inode_item, 3);
1326 btrfs_set_stack_inode_nlink(inode_item, 1);
1327 btrfs_set_stack_inode_nbytes(inode_item,
1328 fs_info->nodesize);
1329 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1330
1331 btrfs_set_root_node(&log_root->root_item, log_root->node);
1332
1333 WARN_ON(root->log_root);
1334 root->log_root = log_root;
1335 root->log_transid = 0;
1336 root->log_transid_committed = -1;
1337 root->last_log_commit = 0;
1338 return 0;
1339}
1340
1341struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1342 struct btrfs_key *key)
1343{
1344 struct btrfs_root *root;
1345 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1346 struct btrfs_path *path;
1347 u64 generation;
1348 int ret;
1349 int level;
1350
1351 path = btrfs_alloc_path();
1352 if (!path)
1353 return ERR_PTR(-ENOMEM);
1354
1355 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1356 if (!root) {
1357 ret = -ENOMEM;
1358 goto alloc_fail;
1359 }
1360
1361 ret = btrfs_find_root(tree_root, key, path,
1362 &root->root_item, &root->root_key);
1363 if (ret) {
1364 if (ret > 0)
1365 ret = -ENOENT;
1366 goto find_fail;
1367 }
1368
1369 generation = btrfs_root_generation(&root->root_item);
1370 level = btrfs_root_level(&root->root_item);
1371 root->node = read_tree_block(fs_info,
1372 btrfs_root_bytenr(&root->root_item),
1373 generation, level, NULL);
1374 if (IS_ERR(root->node)) {
1375 ret = PTR_ERR(root->node);
1376 root->node = NULL;
1377 goto find_fail;
1378 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1379 ret = -EIO;
1380 goto find_fail;
1381 }
1382 root->commit_root = btrfs_root_node(root);
1383out:
1384 btrfs_free_path(path);
1385 return root;
1386
1387find_fail:
1388 btrfs_put_root(root);
1389alloc_fail:
1390 root = ERR_PTR(ret);
1391 goto out;
1392}
1393
1394/*
1395 * Initialize subvolume root in-memory structure
1396 *
1397 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1398 */
1399static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1400{
1401 int ret;
1402 unsigned int nofs_flag;
1403
1404 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1405 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1406 GFP_NOFS);
1407 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1408 ret = -ENOMEM;
1409 goto fail;
1410 }
1411
1412 /*
1413 * We might be called under a transaction (e.g. indirect backref
1414 * resolution) which could deadlock if it triggers memory reclaim
1415 */
1416 nofs_flag = memalloc_nofs_save();
1417 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1418 memalloc_nofs_restore(nofs_flag);
1419 if (ret)
1420 goto fail;
1421
1422 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1423 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1424 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1425 btrfs_check_and_init_root_item(&root->root_item);
1426 }
1427
1428 btrfs_init_free_ino_ctl(root);
1429 spin_lock_init(&root->ino_cache_lock);
1430 init_waitqueue_head(&root->ino_cache_wait);
1431
1432 /*
1433 * Don't assign anonymous block device to roots that are not exposed to
1434 * userspace, the id pool is limited to 1M
1435 */
1436 if (is_fstree(root->root_key.objectid) &&
1437 btrfs_root_refs(&root->root_item) > 0) {
1438 if (!anon_dev) {
1439 ret = get_anon_bdev(&root->anon_dev);
1440 if (ret)
1441 goto fail;
1442 } else {
1443 root->anon_dev = anon_dev;
1444 }
1445 }
1446
1447 mutex_lock(&root->objectid_mutex);
1448 ret = btrfs_find_highest_objectid(root,
1449 &root->highest_objectid);
1450 if (ret) {
1451 mutex_unlock(&root->objectid_mutex);
1452 goto fail;
1453 }
1454
1455 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1456
1457 mutex_unlock(&root->objectid_mutex);
1458
1459 return 0;
1460fail:
1461 /* The caller is responsible to call btrfs_free_fs_root */
1462 return ret;
1463}
1464
1465static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1466 u64 root_id)
1467{
1468 struct btrfs_root *root;
1469
1470 spin_lock(&fs_info->fs_roots_radix_lock);
1471 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1472 (unsigned long)root_id);
1473 if (root)
1474 root = btrfs_grab_root(root);
1475 spin_unlock(&fs_info->fs_roots_radix_lock);
1476 return root;
1477}
1478
1479int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1480 struct btrfs_root *root)
1481{
1482 int ret;
1483
1484 ret = radix_tree_preload(GFP_NOFS);
1485 if (ret)
1486 return ret;
1487
1488 spin_lock(&fs_info->fs_roots_radix_lock);
1489 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1490 (unsigned long)root->root_key.objectid,
1491 root);
1492 if (ret == 0) {
1493 btrfs_grab_root(root);
1494 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1495 }
1496 spin_unlock(&fs_info->fs_roots_radix_lock);
1497 radix_tree_preload_end();
1498
1499 return ret;
1500}
1501
1502void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1503{
1504#ifdef CONFIG_BTRFS_DEBUG
1505 struct btrfs_root *root;
1506
1507 while (!list_empty(&fs_info->allocated_roots)) {
1508 root = list_first_entry(&fs_info->allocated_roots,
1509 struct btrfs_root, leak_list);
1510 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1511 root->root_key.objectid, root->root_key.offset,
1512 refcount_read(&root->refs));
1513 while (refcount_read(&root->refs) > 1)
1514 btrfs_put_root(root);
1515 btrfs_put_root(root);
1516 }
1517#endif
1518}
1519
1520void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1521{
1522 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1523 percpu_counter_destroy(&fs_info->delalloc_bytes);
1524 percpu_counter_destroy(&fs_info->dio_bytes);
1525 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1526 btrfs_free_csum_hash(fs_info);
1527 btrfs_free_stripe_hash_table(fs_info);
1528 btrfs_free_ref_cache(fs_info);
1529 kfree(fs_info->balance_ctl);
1530 kfree(fs_info->delayed_root);
1531 btrfs_put_root(fs_info->extent_root);
1532 btrfs_put_root(fs_info->tree_root);
1533 btrfs_put_root(fs_info->chunk_root);
1534 btrfs_put_root(fs_info->dev_root);
1535 btrfs_put_root(fs_info->csum_root);
1536 btrfs_put_root(fs_info->quota_root);
1537 btrfs_put_root(fs_info->uuid_root);
1538 btrfs_put_root(fs_info->free_space_root);
1539 btrfs_put_root(fs_info->fs_root);
1540 btrfs_put_root(fs_info->data_reloc_root);
1541 btrfs_check_leaked_roots(fs_info);
1542 btrfs_extent_buffer_leak_debug_check(fs_info);
1543 kfree(fs_info->super_copy);
1544 kfree(fs_info->super_for_commit);
1545 kvfree(fs_info);
1546}
1547
1548
1549/*
1550 * Get an in-memory reference of a root structure.
1551 *
1552 * For essential trees like root/extent tree, we grab it from fs_info directly.
1553 * For subvolume trees, we check the cached filesystem roots first. If not
1554 * found, then read it from disk and add it to cached fs roots.
1555 *
1556 * Caller should release the root by calling btrfs_put_root() after the usage.
1557 *
1558 * NOTE: Reloc and log trees can't be read by this function as they share the
1559 * same root objectid.
1560 *
1561 * @objectid: root id
1562 * @anon_dev: preallocated anonymous block device number for new roots,
1563 * pass 0 for new allocation.
1564 * @check_ref: whether to check root item references, If true, return -ENOENT
1565 * for orphan roots
1566 */
1567static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1568 u64 objectid, dev_t anon_dev,
1569 bool check_ref)
1570{
1571 struct btrfs_root *root;
1572 struct btrfs_path *path;
1573 struct btrfs_key key;
1574 int ret;
1575
1576 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1577 return btrfs_grab_root(fs_info->tree_root);
1578 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1579 return btrfs_grab_root(fs_info->extent_root);
1580 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1581 return btrfs_grab_root(fs_info->chunk_root);
1582 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1583 return btrfs_grab_root(fs_info->dev_root);
1584 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1585 return btrfs_grab_root(fs_info->csum_root);
1586 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1587 return btrfs_grab_root(fs_info->quota_root) ?
1588 fs_info->quota_root : ERR_PTR(-ENOENT);
1589 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1590 return btrfs_grab_root(fs_info->uuid_root) ?
1591 fs_info->uuid_root : ERR_PTR(-ENOENT);
1592 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1593 return btrfs_grab_root(fs_info->free_space_root) ?
1594 fs_info->free_space_root : ERR_PTR(-ENOENT);
1595again:
1596 root = btrfs_lookup_fs_root(fs_info, objectid);
1597 if (root) {
1598 /* Shouldn't get preallocated anon_dev for cached roots */
1599 ASSERT(!anon_dev);
1600 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1601 btrfs_put_root(root);
1602 return ERR_PTR(-ENOENT);
1603 }
1604 return root;
1605 }
1606
1607 key.objectid = objectid;
1608 key.type = BTRFS_ROOT_ITEM_KEY;
1609 key.offset = (u64)-1;
1610 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1611 if (IS_ERR(root))
1612 return root;
1613
1614 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1615 ret = -ENOENT;
1616 goto fail;
1617 }
1618
1619 ret = btrfs_init_fs_root(root, anon_dev);
1620 if (ret)
1621 goto fail;
1622
1623 path = btrfs_alloc_path();
1624 if (!path) {
1625 ret = -ENOMEM;
1626 goto fail;
1627 }
1628 key.objectid = BTRFS_ORPHAN_OBJECTID;
1629 key.type = BTRFS_ORPHAN_ITEM_KEY;
1630 key.offset = objectid;
1631
1632 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1633 btrfs_free_path(path);
1634 if (ret < 0)
1635 goto fail;
1636 if (ret == 0)
1637 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1638
1639 ret = btrfs_insert_fs_root(fs_info, root);
1640 if (ret) {
1641 btrfs_put_root(root);
1642 if (ret == -EEXIST)
1643 goto again;
1644 goto fail;
1645 }
1646 return root;
1647fail:
1648 btrfs_put_root(root);
1649 return ERR_PTR(ret);
1650}
1651
1652/*
1653 * Get in-memory reference of a root structure
1654 *
1655 * @objectid: tree objectid
1656 * @check_ref: if set, verify that the tree exists and the item has at least
1657 * one reference
1658 */
1659struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1660 u64 objectid, bool check_ref)
1661{
1662 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1663}
1664
1665/*
1666 * Get in-memory reference of a root structure, created as new, optionally pass
1667 * the anonymous block device id
1668 *
1669 * @objectid: tree objectid
1670 * @anon_dev: if zero, allocate a new anonymous block device or use the
1671 * parameter value
1672 */
1673struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1674 u64 objectid, dev_t anon_dev)
1675{
1676 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1677}
1678
1679/*
1680 * called by the kthread helper functions to finally call the bio end_io
1681 * functions. This is where read checksum verification actually happens
1682 */
1683static void end_workqueue_fn(struct btrfs_work *work)
1684{
1685 struct bio *bio;
1686 struct btrfs_end_io_wq *end_io_wq;
1687
1688 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1689 bio = end_io_wq->bio;
1690
1691 bio->bi_status = end_io_wq->status;
1692 bio->bi_private = end_io_wq->private;
1693 bio->bi_end_io = end_io_wq->end_io;
1694 bio_endio(bio);
1695 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1696}
1697
1698static int cleaner_kthread(void *arg)
1699{
1700 struct btrfs_root *root = arg;
1701 struct btrfs_fs_info *fs_info = root->fs_info;
1702 int again;
1703
1704 while (1) {
1705 again = 0;
1706
1707 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1708
1709 /* Make the cleaner go to sleep early. */
1710 if (btrfs_need_cleaner_sleep(fs_info))
1711 goto sleep;
1712
1713 /*
1714 * Do not do anything if we might cause open_ctree() to block
1715 * before we have finished mounting the filesystem.
1716 */
1717 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1718 goto sleep;
1719
1720 if (!mutex_trylock(&fs_info->cleaner_mutex))
1721 goto sleep;
1722
1723 /*
1724 * Avoid the problem that we change the status of the fs
1725 * during the above check and trylock.
1726 */
1727 if (btrfs_need_cleaner_sleep(fs_info)) {
1728 mutex_unlock(&fs_info->cleaner_mutex);
1729 goto sleep;
1730 }
1731
1732 btrfs_run_delayed_iputs(fs_info);
1733
1734 again = btrfs_clean_one_deleted_snapshot(root);
1735 mutex_unlock(&fs_info->cleaner_mutex);
1736
1737 /*
1738 * The defragger has dealt with the R/O remount and umount,
1739 * needn't do anything special here.
1740 */
1741 btrfs_run_defrag_inodes(fs_info);
1742
1743 /*
1744 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1745 * with relocation (btrfs_relocate_chunk) and relocation
1746 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1747 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1748 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1749 * unused block groups.
1750 */
1751 btrfs_delete_unused_bgs(fs_info);
1752sleep:
1753 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1754 if (kthread_should_park())
1755 kthread_parkme();
1756 if (kthread_should_stop())
1757 return 0;
1758 if (!again) {
1759 set_current_state(TASK_INTERRUPTIBLE);
1760 schedule();
1761 __set_current_state(TASK_RUNNING);
1762 }
1763 }
1764}
1765
1766static int transaction_kthread(void *arg)
1767{
1768 struct btrfs_root *root = arg;
1769 struct btrfs_fs_info *fs_info = root->fs_info;
1770 struct btrfs_trans_handle *trans;
1771 struct btrfs_transaction *cur;
1772 u64 transid;
1773 time64_t now;
1774 unsigned long delay;
1775 bool cannot_commit;
1776
1777 do {
1778 cannot_commit = false;
1779 delay = HZ * fs_info->commit_interval;
1780 mutex_lock(&fs_info->transaction_kthread_mutex);
1781
1782 spin_lock(&fs_info->trans_lock);
1783 cur = fs_info->running_transaction;
1784 if (!cur) {
1785 spin_unlock(&fs_info->trans_lock);
1786 goto sleep;
1787 }
1788
1789 now = ktime_get_seconds();
1790 if (cur->state < TRANS_STATE_COMMIT_START &&
1791 (now < cur->start_time ||
1792 now - cur->start_time < fs_info->commit_interval)) {
1793 spin_unlock(&fs_info->trans_lock);
1794 delay = HZ * 5;
1795 goto sleep;
1796 }
1797 transid = cur->transid;
1798 spin_unlock(&fs_info->trans_lock);
1799
1800 /* If the file system is aborted, this will always fail. */
1801 trans = btrfs_attach_transaction(root);
1802 if (IS_ERR(trans)) {
1803 if (PTR_ERR(trans) != -ENOENT)
1804 cannot_commit = true;
1805 goto sleep;
1806 }
1807 if (transid == trans->transid) {
1808 btrfs_commit_transaction(trans);
1809 } else {
1810 btrfs_end_transaction(trans);
1811 }
1812sleep:
1813 wake_up_process(fs_info->cleaner_kthread);
1814 mutex_unlock(&fs_info->transaction_kthread_mutex);
1815
1816 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1817 &fs_info->fs_state)))
1818 btrfs_cleanup_transaction(fs_info);
1819 if (!kthread_should_stop() &&
1820 (!btrfs_transaction_blocked(fs_info) ||
1821 cannot_commit))
1822 schedule_timeout_interruptible(delay);
1823 } while (!kthread_should_stop());
1824 return 0;
1825}
1826
1827/*
1828 * This will find the highest generation in the array of root backups. The
1829 * index of the highest array is returned, or -EINVAL if we can't find
1830 * anything.
1831 *
1832 * We check to make sure the array is valid by comparing the
1833 * generation of the latest root in the array with the generation
1834 * in the super block. If they don't match we pitch it.
1835 */
1836static int find_newest_super_backup(struct btrfs_fs_info *info)
1837{
1838 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1839 u64 cur;
1840 struct btrfs_root_backup *root_backup;
1841 int i;
1842
1843 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1844 root_backup = info->super_copy->super_roots + i;
1845 cur = btrfs_backup_tree_root_gen(root_backup);
1846 if (cur == newest_gen)
1847 return i;
1848 }
1849
1850 return -EINVAL;
1851}
1852
1853/*
1854 * copy all the root pointers into the super backup array.
1855 * this will bump the backup pointer by one when it is
1856 * done
1857 */
1858static void backup_super_roots(struct btrfs_fs_info *info)
1859{
1860 const int next_backup = info->backup_root_index;
1861 struct btrfs_root_backup *root_backup;
1862
1863 root_backup = info->super_for_commit->super_roots + next_backup;
1864
1865 /*
1866 * make sure all of our padding and empty slots get zero filled
1867 * regardless of which ones we use today
1868 */
1869 memset(root_backup, 0, sizeof(*root_backup));
1870
1871 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1872
1873 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1874 btrfs_set_backup_tree_root_gen(root_backup,
1875 btrfs_header_generation(info->tree_root->node));
1876
1877 btrfs_set_backup_tree_root_level(root_backup,
1878 btrfs_header_level(info->tree_root->node));
1879
1880 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1881 btrfs_set_backup_chunk_root_gen(root_backup,
1882 btrfs_header_generation(info->chunk_root->node));
1883 btrfs_set_backup_chunk_root_level(root_backup,
1884 btrfs_header_level(info->chunk_root->node));
1885
1886 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1887 btrfs_set_backup_extent_root_gen(root_backup,
1888 btrfs_header_generation(info->extent_root->node));
1889 btrfs_set_backup_extent_root_level(root_backup,
1890 btrfs_header_level(info->extent_root->node));
1891
1892 /*
1893 * we might commit during log recovery, which happens before we set
1894 * the fs_root. Make sure it is valid before we fill it in.
1895 */
1896 if (info->fs_root && info->fs_root->node) {
1897 btrfs_set_backup_fs_root(root_backup,
1898 info->fs_root->node->start);
1899 btrfs_set_backup_fs_root_gen(root_backup,
1900 btrfs_header_generation(info->fs_root->node));
1901 btrfs_set_backup_fs_root_level(root_backup,
1902 btrfs_header_level(info->fs_root->node));
1903 }
1904
1905 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1906 btrfs_set_backup_dev_root_gen(root_backup,
1907 btrfs_header_generation(info->dev_root->node));
1908 btrfs_set_backup_dev_root_level(root_backup,
1909 btrfs_header_level(info->dev_root->node));
1910
1911 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1912 btrfs_set_backup_csum_root_gen(root_backup,
1913 btrfs_header_generation(info->csum_root->node));
1914 btrfs_set_backup_csum_root_level(root_backup,
1915 btrfs_header_level(info->csum_root->node));
1916
1917 btrfs_set_backup_total_bytes(root_backup,
1918 btrfs_super_total_bytes(info->super_copy));
1919 btrfs_set_backup_bytes_used(root_backup,
1920 btrfs_super_bytes_used(info->super_copy));
1921 btrfs_set_backup_num_devices(root_backup,
1922 btrfs_super_num_devices(info->super_copy));
1923
1924 /*
1925 * if we don't copy this out to the super_copy, it won't get remembered
1926 * for the next commit
1927 */
1928 memcpy(&info->super_copy->super_roots,
1929 &info->super_for_commit->super_roots,
1930 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1931}
1932
1933/*
1934 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1935 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1936 *
1937 * fs_info - filesystem whose backup roots need to be read
1938 * priority - priority of backup root required
1939 *
1940 * Returns backup root index on success and -EINVAL otherwise.
1941 */
1942static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1943{
1944 int backup_index = find_newest_super_backup(fs_info);
1945 struct btrfs_super_block *super = fs_info->super_copy;
1946 struct btrfs_root_backup *root_backup;
1947
1948 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1949 if (priority == 0)
1950 return backup_index;
1951
1952 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1953 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1954 } else {
1955 return -EINVAL;
1956 }
1957
1958 root_backup = super->super_roots + backup_index;
1959
1960 btrfs_set_super_generation(super,
1961 btrfs_backup_tree_root_gen(root_backup));
1962 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1963 btrfs_set_super_root_level(super,
1964 btrfs_backup_tree_root_level(root_backup));
1965 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1966
1967 /*
1968 * Fixme: the total bytes and num_devices need to match or we should
1969 * need a fsck
1970 */
1971 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1972 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1973
1974 return backup_index;
1975}
1976
1977/* helper to cleanup workers */
1978static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1979{
1980 btrfs_destroy_workqueue(fs_info->fixup_workers);
1981 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1982 btrfs_destroy_workqueue(fs_info->workers);
1983 btrfs_destroy_workqueue(fs_info->endio_workers);
1984 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1985 btrfs_destroy_workqueue(fs_info->rmw_workers);
1986 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1987 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1988 btrfs_destroy_workqueue(fs_info->delayed_workers);
1989 btrfs_destroy_workqueue(fs_info->caching_workers);
1990 btrfs_destroy_workqueue(fs_info->readahead_workers);
1991 btrfs_destroy_workqueue(fs_info->flush_workers);
1992 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1993 if (fs_info->discard_ctl.discard_workers)
1994 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1995 /*
1996 * Now that all other work queues are destroyed, we can safely destroy
1997 * the queues used for metadata I/O, since tasks from those other work
1998 * queues can do metadata I/O operations.
1999 */
2000 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2001 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2002}
2003
2004static void free_root_extent_buffers(struct btrfs_root *root)
2005{
2006 if (root) {
2007 free_extent_buffer(root->node);
2008 free_extent_buffer(root->commit_root);
2009 root->node = NULL;
2010 root->commit_root = NULL;
2011 }
2012}
2013
2014/* helper to cleanup tree roots */
2015static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2016{
2017 free_root_extent_buffers(info->tree_root);
2018
2019 free_root_extent_buffers(info->dev_root);
2020 free_root_extent_buffers(info->extent_root);
2021 free_root_extent_buffers(info->csum_root);
2022 free_root_extent_buffers(info->quota_root);
2023 free_root_extent_buffers(info->uuid_root);
2024 free_root_extent_buffers(info->fs_root);
2025 free_root_extent_buffers(info->data_reloc_root);
2026 if (free_chunk_root)
2027 free_root_extent_buffers(info->chunk_root);
2028 free_root_extent_buffers(info->free_space_root);
2029}
2030
2031void btrfs_put_root(struct btrfs_root *root)
2032{
2033 if (!root)
2034 return;
2035
2036 if (refcount_dec_and_test(&root->refs)) {
2037 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2038 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2039 if (root->anon_dev)
2040 free_anon_bdev(root->anon_dev);
2041 btrfs_drew_lock_destroy(&root->snapshot_lock);
2042 free_root_extent_buffers(root);
2043 kfree(root->free_ino_ctl);
2044 kfree(root->free_ino_pinned);
2045#ifdef CONFIG_BTRFS_DEBUG
2046 spin_lock(&root->fs_info->fs_roots_radix_lock);
2047 list_del_init(&root->leak_list);
2048 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2049#endif
2050 kfree(root);
2051 }
2052}
2053
2054void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2055{
2056 int ret;
2057 struct btrfs_root *gang[8];
2058 int i;
2059
2060 while (!list_empty(&fs_info->dead_roots)) {
2061 gang[0] = list_entry(fs_info->dead_roots.next,
2062 struct btrfs_root, root_list);
2063 list_del(&gang[0]->root_list);
2064
2065 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2066 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2067 btrfs_put_root(gang[0]);
2068 }
2069
2070 while (1) {
2071 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2072 (void **)gang, 0,
2073 ARRAY_SIZE(gang));
2074 if (!ret)
2075 break;
2076 for (i = 0; i < ret; i++)
2077 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2078 }
2079}
2080
2081static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2082{
2083 mutex_init(&fs_info->scrub_lock);
2084 atomic_set(&fs_info->scrubs_running, 0);
2085 atomic_set(&fs_info->scrub_pause_req, 0);
2086 atomic_set(&fs_info->scrubs_paused, 0);
2087 atomic_set(&fs_info->scrub_cancel_req, 0);
2088 init_waitqueue_head(&fs_info->scrub_pause_wait);
2089 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2090}
2091
2092static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2093{
2094 spin_lock_init(&fs_info->balance_lock);
2095 mutex_init(&fs_info->balance_mutex);
2096 atomic_set(&fs_info->balance_pause_req, 0);
2097 atomic_set(&fs_info->balance_cancel_req, 0);
2098 fs_info->balance_ctl = NULL;
2099 init_waitqueue_head(&fs_info->balance_wait_q);
2100}
2101
2102static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2103{
2104 struct inode *inode = fs_info->btree_inode;
2105
2106 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2107 set_nlink(inode, 1);
2108 /*
2109 * we set the i_size on the btree inode to the max possible int.
2110 * the real end of the address space is determined by all of
2111 * the devices in the system
2112 */
2113 inode->i_size = OFFSET_MAX;
2114 inode->i_mapping->a_ops = &btree_aops;
2115
2116 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2117 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2118 IO_TREE_INODE_IO, inode);
2119 BTRFS_I(inode)->io_tree.track_uptodate = false;
2120 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2121
2122 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2123
2124 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2125 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2126 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2127 btrfs_insert_inode_hash(inode);
2128}
2129
2130static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2131{
2132 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2133 init_rwsem(&fs_info->dev_replace.rwsem);
2134 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2135}
2136
2137static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2138{
2139 spin_lock_init(&fs_info->qgroup_lock);
2140 mutex_init(&fs_info->qgroup_ioctl_lock);
2141 fs_info->qgroup_tree = RB_ROOT;
2142 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2143 fs_info->qgroup_seq = 1;
2144 fs_info->qgroup_ulist = NULL;
2145 fs_info->qgroup_rescan_running = false;
2146 mutex_init(&fs_info->qgroup_rescan_lock);
2147}
2148
2149static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2150 struct btrfs_fs_devices *fs_devices)
2151{
2152 u32 max_active = fs_info->thread_pool_size;
2153 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2154
2155 fs_info->workers =
2156 btrfs_alloc_workqueue(fs_info, "worker",
2157 flags | WQ_HIGHPRI, max_active, 16);
2158
2159 fs_info->delalloc_workers =
2160 btrfs_alloc_workqueue(fs_info, "delalloc",
2161 flags, max_active, 2);
2162
2163 fs_info->flush_workers =
2164 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2165 flags, max_active, 0);
2166
2167 fs_info->caching_workers =
2168 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2169
2170 fs_info->fixup_workers =
2171 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2172
2173 /*
2174 * endios are largely parallel and should have a very
2175 * low idle thresh
2176 */
2177 fs_info->endio_workers =
2178 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2179 fs_info->endio_meta_workers =
2180 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2181 max_active, 4);
2182 fs_info->endio_meta_write_workers =
2183 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2184 max_active, 2);
2185 fs_info->endio_raid56_workers =
2186 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2187 max_active, 4);
2188 fs_info->rmw_workers =
2189 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2190 fs_info->endio_write_workers =
2191 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2192 max_active, 2);
2193 fs_info->endio_freespace_worker =
2194 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2195 max_active, 0);
2196 fs_info->delayed_workers =
2197 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2198 max_active, 0);
2199 fs_info->readahead_workers =
2200 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2201 max_active, 2);
2202 fs_info->qgroup_rescan_workers =
2203 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2204 fs_info->discard_ctl.discard_workers =
2205 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2206
2207 if (!(fs_info->workers && fs_info->delalloc_workers &&
2208 fs_info->flush_workers &&
2209 fs_info->endio_workers && fs_info->endio_meta_workers &&
2210 fs_info->endio_meta_write_workers &&
2211 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2212 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2213 fs_info->caching_workers && fs_info->readahead_workers &&
2214 fs_info->fixup_workers && fs_info->delayed_workers &&
2215 fs_info->qgroup_rescan_workers &&
2216 fs_info->discard_ctl.discard_workers)) {
2217 return -ENOMEM;
2218 }
2219
2220 return 0;
2221}
2222
2223static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2224{
2225 struct crypto_shash *csum_shash;
2226 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2227
2228 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2229
2230 if (IS_ERR(csum_shash)) {
2231 btrfs_err(fs_info, "error allocating %s hash for checksum",
2232 csum_driver);
2233 return PTR_ERR(csum_shash);
2234 }
2235
2236 fs_info->csum_shash = csum_shash;
2237
2238 return 0;
2239}
2240
2241static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2242 struct btrfs_fs_devices *fs_devices)
2243{
2244 int ret;
2245 struct btrfs_root *log_tree_root;
2246 struct btrfs_super_block *disk_super = fs_info->super_copy;
2247 u64 bytenr = btrfs_super_log_root(disk_super);
2248 int level = btrfs_super_log_root_level(disk_super);
2249
2250 if (fs_devices->rw_devices == 0) {
2251 btrfs_warn(fs_info, "log replay required on RO media");
2252 return -EIO;
2253 }
2254
2255 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2256 GFP_KERNEL);
2257 if (!log_tree_root)
2258 return -ENOMEM;
2259
2260 log_tree_root->node = read_tree_block(fs_info, bytenr,
2261 fs_info->generation + 1,
2262 level, NULL);
2263 if (IS_ERR(log_tree_root->node)) {
2264 btrfs_warn(fs_info, "failed to read log tree");
2265 ret = PTR_ERR(log_tree_root->node);
2266 log_tree_root->node = NULL;
2267 btrfs_put_root(log_tree_root);
2268 return ret;
2269 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2270 btrfs_err(fs_info, "failed to read log tree");
2271 btrfs_put_root(log_tree_root);
2272 return -EIO;
2273 }
2274 /* returns with log_tree_root freed on success */
2275 ret = btrfs_recover_log_trees(log_tree_root);
2276 if (ret) {
2277 btrfs_handle_fs_error(fs_info, ret,
2278 "Failed to recover log tree");
2279 btrfs_put_root(log_tree_root);
2280 return ret;
2281 }
2282
2283 if (sb_rdonly(fs_info->sb)) {
2284 ret = btrfs_commit_super(fs_info);
2285 if (ret)
2286 return ret;
2287 }
2288
2289 return 0;
2290}
2291
2292static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2293{
2294 struct btrfs_root *tree_root = fs_info->tree_root;
2295 struct btrfs_root *root;
2296 struct btrfs_key location;
2297 int ret;
2298
2299 BUG_ON(!fs_info->tree_root);
2300
2301 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2302 location.type = BTRFS_ROOT_ITEM_KEY;
2303 location.offset = 0;
2304
2305 root = btrfs_read_tree_root(tree_root, &location);
2306 if (IS_ERR(root)) {
2307 ret = PTR_ERR(root);
2308 goto out;
2309 }
2310 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2311 fs_info->extent_root = root;
2312
2313 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2314 root = btrfs_read_tree_root(tree_root, &location);
2315 if (IS_ERR(root)) {
2316 ret = PTR_ERR(root);
2317 goto out;
2318 }
2319 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2320 fs_info->dev_root = root;
2321 btrfs_init_devices_late(fs_info);
2322
2323 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2324 root = btrfs_read_tree_root(tree_root, &location);
2325 if (IS_ERR(root)) {
2326 ret = PTR_ERR(root);
2327 goto out;
2328 }
2329 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2330 fs_info->csum_root = root;
2331
2332 /*
2333 * This tree can share blocks with some other fs tree during relocation
2334 * and we need a proper setup by btrfs_get_fs_root
2335 */
2336 root = btrfs_get_fs_root(tree_root->fs_info,
2337 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2338 if (IS_ERR(root)) {
2339 ret = PTR_ERR(root);
2340 goto out;
2341 }
2342 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2343 fs_info->data_reloc_root = root;
2344
2345 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2346 root = btrfs_read_tree_root(tree_root, &location);
2347 if (!IS_ERR(root)) {
2348 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2349 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2350 fs_info->quota_root = root;
2351 }
2352
2353 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2354 root = btrfs_read_tree_root(tree_root, &location);
2355 if (IS_ERR(root)) {
2356 ret = PTR_ERR(root);
2357 if (ret != -ENOENT)
2358 goto out;
2359 } else {
2360 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2361 fs_info->uuid_root = root;
2362 }
2363
2364 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2365 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2366 root = btrfs_read_tree_root(tree_root, &location);
2367 if (IS_ERR(root)) {
2368 ret = PTR_ERR(root);
2369 goto out;
2370 }
2371 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2372 fs_info->free_space_root = root;
2373 }
2374
2375 return 0;
2376out:
2377 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2378 location.objectid, ret);
2379 return ret;
2380}
2381
2382/*
2383 * Real super block validation
2384 * NOTE: super csum type and incompat features will not be checked here.
2385 *
2386 * @sb: super block to check
2387 * @mirror_num: the super block number to check its bytenr:
2388 * 0 the primary (1st) sb
2389 * 1, 2 2nd and 3rd backup copy
2390 * -1 skip bytenr check
2391 */
2392static int validate_super(struct btrfs_fs_info *fs_info,
2393 struct btrfs_super_block *sb, int mirror_num)
2394{
2395 u64 nodesize = btrfs_super_nodesize(sb);
2396 u64 sectorsize = btrfs_super_sectorsize(sb);
2397 int ret = 0;
2398
2399 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2400 btrfs_err(fs_info, "no valid FS found");
2401 ret = -EINVAL;
2402 }
2403 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2404 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2405 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2406 ret = -EINVAL;
2407 }
2408 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2409 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2410 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2411 ret = -EINVAL;
2412 }
2413 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2414 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2415 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2416 ret = -EINVAL;
2417 }
2418 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2419 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2420 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2421 ret = -EINVAL;
2422 }
2423
2424 /*
2425 * Check sectorsize and nodesize first, other check will need it.
2426 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2427 */
2428 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2429 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2430 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2431 ret = -EINVAL;
2432 }
2433 /* Only PAGE SIZE is supported yet */
2434 if (sectorsize != PAGE_SIZE) {
2435 btrfs_err(fs_info,
2436 "sectorsize %llu not supported yet, only support %lu",
2437 sectorsize, PAGE_SIZE);
2438 ret = -EINVAL;
2439 }
2440 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2441 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2442 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2443 ret = -EINVAL;
2444 }
2445 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2446 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2447 le32_to_cpu(sb->__unused_leafsize), nodesize);
2448 ret = -EINVAL;
2449 }
2450
2451 /* Root alignment check */
2452 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2453 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2454 btrfs_super_root(sb));
2455 ret = -EINVAL;
2456 }
2457 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2458 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2459 btrfs_super_chunk_root(sb));
2460 ret = -EINVAL;
2461 }
2462 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2463 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2464 btrfs_super_log_root(sb));
2465 ret = -EINVAL;
2466 }
2467
2468 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2469 BTRFS_FSID_SIZE) != 0) {
2470 btrfs_err(fs_info,
2471 "dev_item UUID does not match metadata fsid: %pU != %pU",
2472 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2473 ret = -EINVAL;
2474 }
2475
2476 /*
2477 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2478 * done later
2479 */
2480 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2481 btrfs_err(fs_info, "bytes_used is too small %llu",
2482 btrfs_super_bytes_used(sb));
2483 ret = -EINVAL;
2484 }
2485 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2486 btrfs_err(fs_info, "invalid stripesize %u",
2487 btrfs_super_stripesize(sb));
2488 ret = -EINVAL;
2489 }
2490 if (btrfs_super_num_devices(sb) > (1UL << 31))
2491 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2492 btrfs_super_num_devices(sb));
2493 if (btrfs_super_num_devices(sb) == 0) {
2494 btrfs_err(fs_info, "number of devices is 0");
2495 ret = -EINVAL;
2496 }
2497
2498 if (mirror_num >= 0 &&
2499 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2500 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2501 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2502 ret = -EINVAL;
2503 }
2504
2505 /*
2506 * Obvious sys_chunk_array corruptions, it must hold at least one key
2507 * and one chunk
2508 */
2509 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2510 btrfs_err(fs_info, "system chunk array too big %u > %u",
2511 btrfs_super_sys_array_size(sb),
2512 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2513 ret = -EINVAL;
2514 }
2515 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2516 + sizeof(struct btrfs_chunk)) {
2517 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2518 btrfs_super_sys_array_size(sb),
2519 sizeof(struct btrfs_disk_key)
2520 + sizeof(struct btrfs_chunk));
2521 ret = -EINVAL;
2522 }
2523
2524 /*
2525 * The generation is a global counter, we'll trust it more than the others
2526 * but it's still possible that it's the one that's wrong.
2527 */
2528 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2529 btrfs_warn(fs_info,
2530 "suspicious: generation < chunk_root_generation: %llu < %llu",
2531 btrfs_super_generation(sb),
2532 btrfs_super_chunk_root_generation(sb));
2533 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2534 && btrfs_super_cache_generation(sb) != (u64)-1)
2535 btrfs_warn(fs_info,
2536 "suspicious: generation < cache_generation: %llu < %llu",
2537 btrfs_super_generation(sb),
2538 btrfs_super_cache_generation(sb));
2539
2540 return ret;
2541}
2542
2543/*
2544 * Validation of super block at mount time.
2545 * Some checks already done early at mount time, like csum type and incompat
2546 * flags will be skipped.
2547 */
2548static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2549{
2550 return validate_super(fs_info, fs_info->super_copy, 0);
2551}
2552
2553/*
2554 * Validation of super block at write time.
2555 * Some checks like bytenr check will be skipped as their values will be
2556 * overwritten soon.
2557 * Extra checks like csum type and incompat flags will be done here.
2558 */
2559static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2560 struct btrfs_super_block *sb)
2561{
2562 int ret;
2563
2564 ret = validate_super(fs_info, sb, -1);
2565 if (ret < 0)
2566 goto out;
2567 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2568 ret = -EUCLEAN;
2569 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2570 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2571 goto out;
2572 }
2573 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2574 ret = -EUCLEAN;
2575 btrfs_err(fs_info,
2576 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2577 btrfs_super_incompat_flags(sb),
2578 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2579 goto out;
2580 }
2581out:
2582 if (ret < 0)
2583 btrfs_err(fs_info,
2584 "super block corruption detected before writing it to disk");
2585 return ret;
2586}
2587
2588static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2589{
2590 int backup_index = find_newest_super_backup(fs_info);
2591 struct btrfs_super_block *sb = fs_info->super_copy;
2592 struct btrfs_root *tree_root = fs_info->tree_root;
2593 bool handle_error = false;
2594 int ret = 0;
2595 int i;
2596
2597 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2598 u64 generation;
2599 int level;
2600
2601 if (handle_error) {
2602 if (!IS_ERR(tree_root->node))
2603 free_extent_buffer(tree_root->node);
2604 tree_root->node = NULL;
2605
2606 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2607 break;
2608
2609 free_root_pointers(fs_info, 0);
2610
2611 /*
2612 * Don't use the log in recovery mode, it won't be
2613 * valid
2614 */
2615 btrfs_set_super_log_root(sb, 0);
2616
2617 /* We can't trust the free space cache either */
2618 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2619
2620 ret = read_backup_root(fs_info, i);
2621 backup_index = ret;
2622 if (ret < 0)
2623 return ret;
2624 }
2625 generation = btrfs_super_generation(sb);
2626 level = btrfs_super_root_level(sb);
2627 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2628 generation, level, NULL);
2629 if (IS_ERR(tree_root->node) ||
2630 !extent_buffer_uptodate(tree_root->node)) {
2631 handle_error = true;
2632
2633 if (IS_ERR(tree_root->node)) {
2634 ret = PTR_ERR(tree_root->node);
2635 tree_root->node = NULL;
2636 } else if (!extent_buffer_uptodate(tree_root->node)) {
2637 ret = -EUCLEAN;
2638 }
2639
2640 btrfs_warn(fs_info, "failed to read tree root");
2641 continue;
2642 }
2643
2644 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2645 tree_root->commit_root = btrfs_root_node(tree_root);
2646 btrfs_set_root_refs(&tree_root->root_item, 1);
2647
2648 /*
2649 * No need to hold btrfs_root::objectid_mutex since the fs
2650 * hasn't been fully initialised and we are the only user
2651 */
2652 ret = btrfs_find_highest_objectid(tree_root,
2653 &tree_root->highest_objectid);
2654 if (ret < 0) {
2655 handle_error = true;
2656 continue;
2657 }
2658
2659 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2660
2661 ret = btrfs_read_roots(fs_info);
2662 if (ret < 0) {
2663 handle_error = true;
2664 continue;
2665 }
2666
2667 /* All successful */
2668 fs_info->generation = generation;
2669 fs_info->last_trans_committed = generation;
2670
2671 /* Always begin writing backup roots after the one being used */
2672 if (backup_index < 0) {
2673 fs_info->backup_root_index = 0;
2674 } else {
2675 fs_info->backup_root_index = backup_index + 1;
2676 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2677 }
2678 break;
2679 }
2680
2681 return ret;
2682}
2683
2684void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2685{
2686 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2687 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2688 INIT_LIST_HEAD(&fs_info->trans_list);
2689 INIT_LIST_HEAD(&fs_info->dead_roots);
2690 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2691 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2692 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2693 spin_lock_init(&fs_info->delalloc_root_lock);
2694 spin_lock_init(&fs_info->trans_lock);
2695 spin_lock_init(&fs_info->fs_roots_radix_lock);
2696 spin_lock_init(&fs_info->delayed_iput_lock);
2697 spin_lock_init(&fs_info->defrag_inodes_lock);
2698 spin_lock_init(&fs_info->super_lock);
2699 spin_lock_init(&fs_info->buffer_lock);
2700 spin_lock_init(&fs_info->unused_bgs_lock);
2701 rwlock_init(&fs_info->tree_mod_log_lock);
2702 mutex_init(&fs_info->unused_bg_unpin_mutex);
2703 mutex_init(&fs_info->delete_unused_bgs_mutex);
2704 mutex_init(&fs_info->reloc_mutex);
2705 mutex_init(&fs_info->delalloc_root_mutex);
2706 seqlock_init(&fs_info->profiles_lock);
2707
2708 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2709 INIT_LIST_HEAD(&fs_info->space_info);
2710 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2711 INIT_LIST_HEAD(&fs_info->unused_bgs);
2712#ifdef CONFIG_BTRFS_DEBUG
2713 INIT_LIST_HEAD(&fs_info->allocated_roots);
2714 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2715 spin_lock_init(&fs_info->eb_leak_lock);
2716#endif
2717 extent_map_tree_init(&fs_info->mapping_tree);
2718 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2719 BTRFS_BLOCK_RSV_GLOBAL);
2720 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2721 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2722 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2723 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2724 BTRFS_BLOCK_RSV_DELOPS);
2725 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2726 BTRFS_BLOCK_RSV_DELREFS);
2727
2728 atomic_set(&fs_info->async_delalloc_pages, 0);
2729 atomic_set(&fs_info->defrag_running, 0);
2730 atomic_set(&fs_info->reada_works_cnt, 0);
2731 atomic_set(&fs_info->nr_delayed_iputs, 0);
2732 atomic64_set(&fs_info->tree_mod_seq, 0);
2733 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2734 fs_info->metadata_ratio = 0;
2735 fs_info->defrag_inodes = RB_ROOT;
2736 atomic64_set(&fs_info->free_chunk_space, 0);
2737 fs_info->tree_mod_log = RB_ROOT;
2738 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2739 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2740 /* readahead state */
2741 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2742 spin_lock_init(&fs_info->reada_lock);
2743 btrfs_init_ref_verify(fs_info);
2744
2745 fs_info->thread_pool_size = min_t(unsigned long,
2746 num_online_cpus() + 2, 8);
2747
2748 INIT_LIST_HEAD(&fs_info->ordered_roots);
2749 spin_lock_init(&fs_info->ordered_root_lock);
2750
2751 btrfs_init_scrub(fs_info);
2752#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2753 fs_info->check_integrity_print_mask = 0;
2754#endif
2755 btrfs_init_balance(fs_info);
2756 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2757
2758 spin_lock_init(&fs_info->block_group_cache_lock);
2759 fs_info->block_group_cache_tree = RB_ROOT;
2760 fs_info->first_logical_byte = (u64)-1;
2761
2762 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2763 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2764 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2765
2766 mutex_init(&fs_info->ordered_operations_mutex);
2767 mutex_init(&fs_info->tree_log_mutex);
2768 mutex_init(&fs_info->chunk_mutex);
2769 mutex_init(&fs_info->transaction_kthread_mutex);
2770 mutex_init(&fs_info->cleaner_mutex);
2771 mutex_init(&fs_info->ro_block_group_mutex);
2772 init_rwsem(&fs_info->commit_root_sem);
2773 init_rwsem(&fs_info->cleanup_work_sem);
2774 init_rwsem(&fs_info->subvol_sem);
2775 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2776
2777 btrfs_init_dev_replace_locks(fs_info);
2778 btrfs_init_qgroup(fs_info);
2779 btrfs_discard_init(fs_info);
2780
2781 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2782 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2783
2784 init_waitqueue_head(&fs_info->transaction_throttle);
2785 init_waitqueue_head(&fs_info->transaction_wait);
2786 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2787 init_waitqueue_head(&fs_info->async_submit_wait);
2788 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2789
2790 /* Usable values until the real ones are cached from the superblock */
2791 fs_info->nodesize = 4096;
2792 fs_info->sectorsize = 4096;
2793 fs_info->stripesize = 4096;
2794
2795 spin_lock_init(&fs_info->swapfile_pins_lock);
2796 fs_info->swapfile_pins = RB_ROOT;
2797
2798 fs_info->send_in_progress = 0;
2799}
2800
2801static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2802{
2803 int ret;
2804
2805 fs_info->sb = sb;
2806 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2807 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2808
2809 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2810 if (ret)
2811 return ret;
2812
2813 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2814 if (ret)
2815 return ret;
2816
2817 fs_info->dirty_metadata_batch = PAGE_SIZE *
2818 (1 + ilog2(nr_cpu_ids));
2819
2820 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2821 if (ret)
2822 return ret;
2823
2824 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2825 GFP_KERNEL);
2826 if (ret)
2827 return ret;
2828
2829 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2830 GFP_KERNEL);
2831 if (!fs_info->delayed_root)
2832 return -ENOMEM;
2833 btrfs_init_delayed_root(fs_info->delayed_root);
2834
2835 return btrfs_alloc_stripe_hash_table(fs_info);
2836}
2837
2838static int btrfs_uuid_rescan_kthread(void *data)
2839{
2840 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2841 int ret;
2842
2843 /*
2844 * 1st step is to iterate through the existing UUID tree and
2845 * to delete all entries that contain outdated data.
2846 * 2nd step is to add all missing entries to the UUID tree.
2847 */
2848 ret = btrfs_uuid_tree_iterate(fs_info);
2849 if (ret < 0) {
2850 if (ret != -EINTR)
2851 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2852 ret);
2853 up(&fs_info->uuid_tree_rescan_sem);
2854 return ret;
2855 }
2856 return btrfs_uuid_scan_kthread(data);
2857}
2858
2859static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2860{
2861 struct task_struct *task;
2862
2863 down(&fs_info->uuid_tree_rescan_sem);
2864 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2865 if (IS_ERR(task)) {
2866 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2867 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2868 up(&fs_info->uuid_tree_rescan_sem);
2869 return PTR_ERR(task);
2870 }
2871
2872 return 0;
2873}
2874
2875int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2876 char *options)
2877{
2878 u32 sectorsize;
2879 u32 nodesize;
2880 u32 stripesize;
2881 u64 generation;
2882 u64 features;
2883 u16 csum_type;
2884 struct btrfs_super_block *disk_super;
2885 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2886 struct btrfs_root *tree_root;
2887 struct btrfs_root *chunk_root;
2888 int ret;
2889 int err = -EINVAL;
2890 int clear_free_space_tree = 0;
2891 int level;
2892
2893 ret = init_mount_fs_info(fs_info, sb);
2894 if (ret) {
2895 err = ret;
2896 goto fail;
2897 }
2898
2899 /* These need to be init'ed before we start creating inodes and such. */
2900 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2901 GFP_KERNEL);
2902 fs_info->tree_root = tree_root;
2903 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2904 GFP_KERNEL);
2905 fs_info->chunk_root = chunk_root;
2906 if (!tree_root || !chunk_root) {
2907 err = -ENOMEM;
2908 goto fail;
2909 }
2910
2911 fs_info->btree_inode = new_inode(sb);
2912 if (!fs_info->btree_inode) {
2913 err = -ENOMEM;
2914 goto fail;
2915 }
2916 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2917 btrfs_init_btree_inode(fs_info);
2918
2919 invalidate_bdev(fs_devices->latest_bdev);
2920
2921 /*
2922 * Read super block and check the signature bytes only
2923 */
2924 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2925 if (IS_ERR(disk_super)) {
2926 err = PTR_ERR(disk_super);
2927 goto fail_alloc;
2928 }
2929
2930 /*
2931 * Verify the type first, if that or the the checksum value are
2932 * corrupted, we'll find out
2933 */
2934 csum_type = btrfs_super_csum_type(disk_super);
2935 if (!btrfs_supported_super_csum(csum_type)) {
2936 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2937 csum_type);
2938 err = -EINVAL;
2939 btrfs_release_disk_super(disk_super);
2940 goto fail_alloc;
2941 }
2942
2943 ret = btrfs_init_csum_hash(fs_info, csum_type);
2944 if (ret) {
2945 err = ret;
2946 btrfs_release_disk_super(disk_super);
2947 goto fail_alloc;
2948 }
2949
2950 /*
2951 * We want to check superblock checksum, the type is stored inside.
2952 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2953 */
2954 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2955 btrfs_err(fs_info, "superblock checksum mismatch");
2956 err = -EINVAL;
2957 btrfs_release_disk_super(disk_super);
2958 goto fail_alloc;
2959 }
2960
2961 /*
2962 * super_copy is zeroed at allocation time and we never touch the
2963 * following bytes up to INFO_SIZE, the checksum is calculated from
2964 * the whole block of INFO_SIZE
2965 */
2966 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2967 btrfs_release_disk_super(disk_super);
2968
2969 disk_super = fs_info->super_copy;
2970
2971 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2972 BTRFS_FSID_SIZE));
2973
2974 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2975 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2976 fs_info->super_copy->metadata_uuid,
2977 BTRFS_FSID_SIZE));
2978 }
2979
2980 features = btrfs_super_flags(disk_super);
2981 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2982 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2983 btrfs_set_super_flags(disk_super, features);
2984 btrfs_info(fs_info,
2985 "found metadata UUID change in progress flag, clearing");
2986 }
2987
2988 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2989 sizeof(*fs_info->super_for_commit));
2990
2991 ret = btrfs_validate_mount_super(fs_info);
2992 if (ret) {
2993 btrfs_err(fs_info, "superblock contains fatal errors");
2994 err = -EINVAL;
2995 goto fail_alloc;
2996 }
2997
2998 if (!btrfs_super_root(disk_super))
2999 goto fail_alloc;
3000
3001 /* check FS state, whether FS is broken. */
3002 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3003 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3004
3005 /*
3006 * In the long term, we'll store the compression type in the super
3007 * block, and it'll be used for per file compression control.
3008 */
3009 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3010
3011 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3012 if (ret) {
3013 err = ret;
3014 goto fail_alloc;
3015 }
3016
3017 features = btrfs_super_incompat_flags(disk_super) &
3018 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3019 if (features) {
3020 btrfs_err(fs_info,
3021 "cannot mount because of unsupported optional features (%llx)",
3022 features);
3023 err = -EINVAL;
3024 goto fail_alloc;
3025 }
3026
3027 features = btrfs_super_incompat_flags(disk_super);
3028 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3029 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3030 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3031 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3032 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3033
3034 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3035 btrfs_info(fs_info, "has skinny extents");
3036
3037 /*
3038 * flag our filesystem as having big metadata blocks if
3039 * they are bigger than the page size
3040 */
3041 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3042 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3043 btrfs_info(fs_info,
3044 "flagging fs with big metadata feature");
3045 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3046 }
3047
3048 nodesize = btrfs_super_nodesize(disk_super);
3049 sectorsize = btrfs_super_sectorsize(disk_super);
3050 stripesize = sectorsize;
3051 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3052 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3053
3054 /* Cache block sizes */
3055 fs_info->nodesize = nodesize;
3056 fs_info->sectorsize = sectorsize;
3057 fs_info->stripesize = stripesize;
3058
3059 /*
3060 * mixed block groups end up with duplicate but slightly offset
3061 * extent buffers for the same range. It leads to corruptions
3062 */
3063 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3064 (sectorsize != nodesize)) {
3065 btrfs_err(fs_info,
3066"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3067 nodesize, sectorsize);
3068 goto fail_alloc;
3069 }
3070
3071 /*
3072 * Needn't use the lock because there is no other task which will
3073 * update the flag.
3074 */
3075 btrfs_set_super_incompat_flags(disk_super, features);
3076
3077 features = btrfs_super_compat_ro_flags(disk_super) &
3078 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3079 if (!sb_rdonly(sb) && features) {
3080 btrfs_err(fs_info,
3081 "cannot mount read-write because of unsupported optional features (%llx)",
3082 features);
3083 err = -EINVAL;
3084 goto fail_alloc;
3085 }
3086
3087 ret = btrfs_init_workqueues(fs_info, fs_devices);
3088 if (ret) {
3089 err = ret;
3090 goto fail_sb_buffer;
3091 }
3092
3093 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3094 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3095 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3096 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3097
3098 sb->s_blocksize = sectorsize;
3099 sb->s_blocksize_bits = blksize_bits(sectorsize);
3100 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3101
3102 mutex_lock(&fs_info->chunk_mutex);
3103 ret = btrfs_read_sys_array(fs_info);
3104 mutex_unlock(&fs_info->chunk_mutex);
3105 if (ret) {
3106 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3107 goto fail_sb_buffer;
3108 }
3109
3110 generation = btrfs_super_chunk_root_generation(disk_super);
3111 level = btrfs_super_chunk_root_level(disk_super);
3112
3113 chunk_root->node = read_tree_block(fs_info,
3114 btrfs_super_chunk_root(disk_super),
3115 generation, level, NULL);
3116 if (IS_ERR(chunk_root->node) ||
3117 !extent_buffer_uptodate(chunk_root->node)) {
3118 btrfs_err(fs_info, "failed to read chunk root");
3119 if (!IS_ERR(chunk_root->node))
3120 free_extent_buffer(chunk_root->node);
3121 chunk_root->node = NULL;
3122 goto fail_tree_roots;
3123 }
3124 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3125 chunk_root->commit_root = btrfs_root_node(chunk_root);
3126
3127 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3128 offsetof(struct btrfs_header, chunk_tree_uuid),
3129 BTRFS_UUID_SIZE);
3130
3131 ret = btrfs_read_chunk_tree(fs_info);
3132 if (ret) {
3133 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3134 goto fail_tree_roots;
3135 }
3136
3137 /*
3138 * Keep the devid that is marked to be the target device for the
3139 * device replace procedure
3140 */
3141 btrfs_free_extra_devids(fs_devices, 0);
3142
3143 if (!fs_devices->latest_bdev) {
3144 btrfs_err(fs_info, "failed to read devices");
3145 goto fail_tree_roots;
3146 }
3147
3148 ret = init_tree_roots(fs_info);
3149 if (ret)
3150 goto fail_tree_roots;
3151
3152 /*
3153 * If we have a uuid root and we're not being told to rescan we need to
3154 * check the generation here so we can set the
3155 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3156 * transaction during a balance or the log replay without updating the
3157 * uuid generation, and then if we crash we would rescan the uuid tree,
3158 * even though it was perfectly fine.
3159 */
3160 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3161 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3162 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3163
3164 ret = btrfs_verify_dev_extents(fs_info);
3165 if (ret) {
3166 btrfs_err(fs_info,
3167 "failed to verify dev extents against chunks: %d",
3168 ret);
3169 goto fail_block_groups;
3170 }
3171 ret = btrfs_recover_balance(fs_info);
3172 if (ret) {
3173 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3174 goto fail_block_groups;
3175 }
3176
3177 ret = btrfs_init_dev_stats(fs_info);
3178 if (ret) {
3179 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3180 goto fail_block_groups;
3181 }
3182
3183 ret = btrfs_init_dev_replace(fs_info);
3184 if (ret) {
3185 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3186 goto fail_block_groups;
3187 }
3188
3189 btrfs_free_extra_devids(fs_devices, 1);
3190
3191 ret = btrfs_sysfs_add_fsid(fs_devices);
3192 if (ret) {
3193 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3194 ret);
3195 goto fail_block_groups;
3196 }
3197
3198 ret = btrfs_sysfs_add_mounted(fs_info);
3199 if (ret) {
3200 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3201 goto fail_fsdev_sysfs;
3202 }
3203
3204 ret = btrfs_init_space_info(fs_info);
3205 if (ret) {
3206 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3207 goto fail_sysfs;
3208 }
3209
3210 ret = btrfs_read_block_groups(fs_info);
3211 if (ret) {
3212 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3213 goto fail_sysfs;
3214 }
3215
3216 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3217 btrfs_warn(fs_info,
3218 "writable mount is not allowed due to too many missing devices");
3219 goto fail_sysfs;
3220 }
3221
3222 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3223 "btrfs-cleaner");
3224 if (IS_ERR(fs_info->cleaner_kthread))
3225 goto fail_sysfs;
3226
3227 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3228 tree_root,
3229 "btrfs-transaction");
3230 if (IS_ERR(fs_info->transaction_kthread))
3231 goto fail_cleaner;
3232
3233 if (!btrfs_test_opt(fs_info, NOSSD) &&
3234 !fs_info->fs_devices->rotating) {
3235 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3236 }
3237
3238 /*
3239 * Mount does not set all options immediately, we can do it now and do
3240 * not have to wait for transaction commit
3241 */
3242 btrfs_apply_pending_changes(fs_info);
3243
3244#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3245 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3246 ret = btrfsic_mount(fs_info, fs_devices,
3247 btrfs_test_opt(fs_info,
3248 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3249 1 : 0,
3250 fs_info->check_integrity_print_mask);
3251 if (ret)
3252 btrfs_warn(fs_info,
3253 "failed to initialize integrity check module: %d",
3254 ret);
3255 }
3256#endif
3257 ret = btrfs_read_qgroup_config(fs_info);
3258 if (ret)
3259 goto fail_trans_kthread;
3260
3261 if (btrfs_build_ref_tree(fs_info))
3262 btrfs_err(fs_info, "couldn't build ref tree");
3263
3264 /* do not make disk changes in broken FS or nologreplay is given */
3265 if (btrfs_super_log_root(disk_super) != 0 &&
3266 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3267 btrfs_info(fs_info, "start tree-log replay");
3268 ret = btrfs_replay_log(fs_info, fs_devices);
3269 if (ret) {
3270 err = ret;
3271 goto fail_qgroup;
3272 }
3273 }
3274
3275 ret = btrfs_find_orphan_roots(fs_info);
3276 if (ret)
3277 goto fail_qgroup;
3278
3279 if (!sb_rdonly(sb)) {
3280 ret = btrfs_cleanup_fs_roots(fs_info);
3281 if (ret)
3282 goto fail_qgroup;
3283
3284 mutex_lock(&fs_info->cleaner_mutex);
3285 ret = btrfs_recover_relocation(tree_root);
3286 mutex_unlock(&fs_info->cleaner_mutex);
3287 if (ret < 0) {
3288 btrfs_warn(fs_info, "failed to recover relocation: %d",
3289 ret);
3290 err = -EINVAL;
3291 goto fail_qgroup;
3292 }
3293 }
3294
3295 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3296 if (IS_ERR(fs_info->fs_root)) {
3297 err = PTR_ERR(fs_info->fs_root);
3298 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3299 fs_info->fs_root = NULL;
3300 goto fail_qgroup;
3301 }
3302
3303 if (sb_rdonly(sb))
3304 return 0;
3305
3306 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3307 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3308 clear_free_space_tree = 1;
3309 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3310 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3311 btrfs_warn(fs_info, "free space tree is invalid");
3312 clear_free_space_tree = 1;
3313 }
3314
3315 if (clear_free_space_tree) {
3316 btrfs_info(fs_info, "clearing free space tree");
3317 ret = btrfs_clear_free_space_tree(fs_info);
3318 if (ret) {
3319 btrfs_warn(fs_info,
3320 "failed to clear free space tree: %d", ret);
3321 close_ctree(fs_info);
3322 return ret;
3323 }
3324 }
3325
3326 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3327 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3328 btrfs_info(fs_info, "creating free space tree");
3329 ret = btrfs_create_free_space_tree(fs_info);
3330 if (ret) {
3331 btrfs_warn(fs_info,
3332 "failed to create free space tree: %d", ret);
3333 close_ctree(fs_info);
3334 return ret;
3335 }
3336 }
3337
3338 down_read(&fs_info->cleanup_work_sem);
3339 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3340 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3341 up_read(&fs_info->cleanup_work_sem);
3342 close_ctree(fs_info);
3343 return ret;
3344 }
3345 up_read(&fs_info->cleanup_work_sem);
3346
3347 ret = btrfs_resume_balance_async(fs_info);
3348 if (ret) {
3349 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3350 close_ctree(fs_info);
3351 return ret;
3352 }
3353
3354 ret = btrfs_resume_dev_replace_async(fs_info);
3355 if (ret) {
3356 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3357 close_ctree(fs_info);
3358 return ret;
3359 }
3360
3361 btrfs_qgroup_rescan_resume(fs_info);
3362 btrfs_discard_resume(fs_info);
3363
3364 if (!fs_info->uuid_root) {
3365 btrfs_info(fs_info, "creating UUID tree");
3366 ret = btrfs_create_uuid_tree(fs_info);
3367 if (ret) {
3368 btrfs_warn(fs_info,
3369 "failed to create the UUID tree: %d", ret);
3370 close_ctree(fs_info);
3371 return ret;
3372 }
3373 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3374 fs_info->generation !=
3375 btrfs_super_uuid_tree_generation(disk_super)) {
3376 btrfs_info(fs_info, "checking UUID tree");
3377 ret = btrfs_check_uuid_tree(fs_info);
3378 if (ret) {
3379 btrfs_warn(fs_info,
3380 "failed to check the UUID tree: %d", ret);
3381 close_ctree(fs_info);
3382 return ret;
3383 }
3384 }
3385 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3386
3387 /*
3388 * backuproot only affect mount behavior, and if open_ctree succeeded,
3389 * no need to keep the flag
3390 */
3391 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3392
3393 return 0;
3394
3395fail_qgroup:
3396 btrfs_free_qgroup_config(fs_info);
3397fail_trans_kthread:
3398 kthread_stop(fs_info->transaction_kthread);
3399 btrfs_cleanup_transaction(fs_info);
3400 btrfs_free_fs_roots(fs_info);
3401fail_cleaner:
3402 kthread_stop(fs_info->cleaner_kthread);
3403
3404 /*
3405 * make sure we're done with the btree inode before we stop our
3406 * kthreads
3407 */
3408 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3409
3410fail_sysfs:
3411 btrfs_sysfs_remove_mounted(fs_info);
3412
3413fail_fsdev_sysfs:
3414 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3415
3416fail_block_groups:
3417 btrfs_put_block_group_cache(fs_info);
3418
3419fail_tree_roots:
3420 if (fs_info->data_reloc_root)
3421 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3422 free_root_pointers(fs_info, true);
3423 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3424
3425fail_sb_buffer:
3426 btrfs_stop_all_workers(fs_info);
3427 btrfs_free_block_groups(fs_info);
3428fail_alloc:
3429 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3430
3431 iput(fs_info->btree_inode);
3432fail:
3433 btrfs_close_devices(fs_info->fs_devices);
3434 return err;
3435}
3436ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3437
3438static void btrfs_end_super_write(struct bio *bio)
3439{
3440 struct btrfs_device *device = bio->bi_private;
3441 struct bio_vec *bvec;
3442 struct bvec_iter_all iter_all;
3443 struct page *page;
3444
3445 bio_for_each_segment_all(bvec, bio, iter_all) {
3446 page = bvec->bv_page;
3447
3448 if (bio->bi_status) {
3449 btrfs_warn_rl_in_rcu(device->fs_info,
3450 "lost page write due to IO error on %s (%d)",
3451 rcu_str_deref(device->name),
3452 blk_status_to_errno(bio->bi_status));
3453 ClearPageUptodate(page);
3454 SetPageError(page);
3455 btrfs_dev_stat_inc_and_print(device,
3456 BTRFS_DEV_STAT_WRITE_ERRS);
3457 } else {
3458 SetPageUptodate(page);
3459 }
3460
3461 put_page(page);
3462 unlock_page(page);
3463 }
3464
3465 bio_put(bio);
3466}
3467
3468struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3469 int copy_num)
3470{
3471 struct btrfs_super_block *super;
3472 struct page *page;
3473 u64 bytenr;
3474 struct address_space *mapping = bdev->bd_inode->i_mapping;
3475
3476 bytenr = btrfs_sb_offset(copy_num);
3477 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3478 return ERR_PTR(-EINVAL);
3479
3480 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3481 if (IS_ERR(page))
3482 return ERR_CAST(page);
3483
3484 super = page_address(page);
3485 if (btrfs_super_bytenr(super) != bytenr ||
3486 btrfs_super_magic(super) != BTRFS_MAGIC) {
3487 btrfs_release_disk_super(super);
3488 return ERR_PTR(-EINVAL);
3489 }
3490
3491 return super;
3492}
3493
3494
3495struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3496{
3497 struct btrfs_super_block *super, *latest = NULL;
3498 int i;
3499 u64 transid = 0;
3500
3501 /* we would like to check all the supers, but that would make
3502 * a btrfs mount succeed after a mkfs from a different FS.
3503 * So, we need to add a special mount option to scan for
3504 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3505 */
3506 for (i = 0; i < 1; i++) {
3507 super = btrfs_read_dev_one_super(bdev, i);
3508 if (IS_ERR(super))
3509 continue;
3510
3511 if (!latest || btrfs_super_generation(super) > transid) {
3512 if (latest)
3513 btrfs_release_disk_super(super);
3514
3515 latest = super;
3516 transid = btrfs_super_generation(super);
3517 }
3518 }
3519
3520 return super;
3521}
3522
3523/*
3524 * Write superblock @sb to the @device. Do not wait for completion, all the
3525 * pages we use for writing are locked.
3526 *
3527 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3528 * the expected device size at commit time. Note that max_mirrors must be
3529 * same for write and wait phases.
3530 *
3531 * Return number of errors when page is not found or submission fails.
3532 */
3533static int write_dev_supers(struct btrfs_device *device,
3534 struct btrfs_super_block *sb, int max_mirrors)
3535{
3536 struct btrfs_fs_info *fs_info = device->fs_info;
3537 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3538 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3539 int i;
3540 int errors = 0;
3541 u64 bytenr;
3542
3543 if (max_mirrors == 0)
3544 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3545
3546 shash->tfm = fs_info->csum_shash;
3547
3548 for (i = 0; i < max_mirrors; i++) {
3549 struct page *page;
3550 struct bio *bio;
3551 struct btrfs_super_block *disk_super;
3552
3553 bytenr = btrfs_sb_offset(i);
3554 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3555 device->commit_total_bytes)
3556 break;
3557
3558 btrfs_set_super_bytenr(sb, bytenr);
3559
3560 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3561 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3562 sb->csum);
3563
3564 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3565 GFP_NOFS);
3566 if (!page) {
3567 btrfs_err(device->fs_info,
3568 "couldn't get super block page for bytenr %llu",
3569 bytenr);
3570 errors++;
3571 continue;
3572 }
3573
3574 /* Bump the refcount for wait_dev_supers() */
3575 get_page(page);
3576
3577 disk_super = page_address(page);
3578 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3579
3580 /*
3581 * Directly use bios here instead of relying on the page cache
3582 * to do I/O, so we don't lose the ability to do integrity
3583 * checking.
3584 */
3585 bio = bio_alloc(GFP_NOFS, 1);
3586 bio_set_dev(bio, device->bdev);
3587 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3588 bio->bi_private = device;
3589 bio->bi_end_io = btrfs_end_super_write;
3590 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3591 offset_in_page(bytenr));
3592
3593 /*
3594 * We FUA only the first super block. The others we allow to
3595 * go down lazy and there's a short window where the on-disk
3596 * copies might still contain the older version.
3597 */
3598 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3599 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3600 bio->bi_opf |= REQ_FUA;
3601
3602 btrfsic_submit_bio(bio);
3603 }
3604 return errors < i ? 0 : -1;
3605}
3606
3607/*
3608 * Wait for write completion of superblocks done by write_dev_supers,
3609 * @max_mirrors same for write and wait phases.
3610 *
3611 * Return number of errors when page is not found or not marked up to
3612 * date.
3613 */
3614static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3615{
3616 int i;
3617 int errors = 0;
3618 bool primary_failed = false;
3619 u64 bytenr;
3620
3621 if (max_mirrors == 0)
3622 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3623
3624 for (i = 0; i < max_mirrors; i++) {
3625 struct page *page;
3626
3627 bytenr = btrfs_sb_offset(i);
3628 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3629 device->commit_total_bytes)
3630 break;
3631
3632 page = find_get_page(device->bdev->bd_inode->i_mapping,
3633 bytenr >> PAGE_SHIFT);
3634 if (!page) {
3635 errors++;
3636 if (i == 0)
3637 primary_failed = true;
3638 continue;
3639 }
3640 /* Page is submitted locked and unlocked once the IO completes */
3641 wait_on_page_locked(page);
3642 if (PageError(page)) {
3643 errors++;
3644 if (i == 0)
3645 primary_failed = true;
3646 }
3647
3648 /* Drop our reference */
3649 put_page(page);
3650
3651 /* Drop the reference from the writing run */
3652 put_page(page);
3653 }
3654
3655 /* log error, force error return */
3656 if (primary_failed) {
3657 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3658 device->devid);
3659 return -1;
3660 }
3661
3662 return errors < i ? 0 : -1;
3663}
3664
3665/*
3666 * endio for the write_dev_flush, this will wake anyone waiting
3667 * for the barrier when it is done
3668 */
3669static void btrfs_end_empty_barrier(struct bio *bio)
3670{
3671 complete(bio->bi_private);
3672}
3673
3674/*
3675 * Submit a flush request to the device if it supports it. Error handling is
3676 * done in the waiting counterpart.
3677 */
3678static void write_dev_flush(struct btrfs_device *device)
3679{
3680 struct request_queue *q = bdev_get_queue(device->bdev);
3681 struct bio *bio = device->flush_bio;
3682
3683 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3684 return;
3685
3686 bio_reset(bio);
3687 bio->bi_end_io = btrfs_end_empty_barrier;
3688 bio_set_dev(bio, device->bdev);
3689 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3690 init_completion(&device->flush_wait);
3691 bio->bi_private = &device->flush_wait;
3692
3693 btrfsic_submit_bio(bio);
3694 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3695}
3696
3697/*
3698 * If the flush bio has been submitted by write_dev_flush, wait for it.
3699 */
3700static blk_status_t wait_dev_flush(struct btrfs_device *device)
3701{
3702 struct bio *bio = device->flush_bio;
3703
3704 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3705 return BLK_STS_OK;
3706
3707 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3708 wait_for_completion_io(&device->flush_wait);
3709
3710 return bio->bi_status;
3711}
3712
3713static int check_barrier_error(struct btrfs_fs_info *fs_info)
3714{
3715 if (!btrfs_check_rw_degradable(fs_info, NULL))
3716 return -EIO;
3717 return 0;
3718}
3719
3720/*
3721 * send an empty flush down to each device in parallel,
3722 * then wait for them
3723 */
3724static int barrier_all_devices(struct btrfs_fs_info *info)
3725{
3726 struct list_head *head;
3727 struct btrfs_device *dev;
3728 int errors_wait = 0;
3729 blk_status_t ret;
3730
3731 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3732 /* send down all the barriers */
3733 head = &info->fs_devices->devices;
3734 list_for_each_entry(dev, head, dev_list) {
3735 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3736 continue;
3737 if (!dev->bdev)
3738 continue;
3739 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3740 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3741 continue;
3742
3743 write_dev_flush(dev);
3744 dev->last_flush_error = BLK_STS_OK;
3745 }
3746
3747 /* wait for all the barriers */
3748 list_for_each_entry(dev, head, dev_list) {
3749 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3750 continue;
3751 if (!dev->bdev) {
3752 errors_wait++;
3753 continue;
3754 }
3755 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3756 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3757 continue;
3758
3759 ret = wait_dev_flush(dev);
3760 if (ret) {
3761 dev->last_flush_error = ret;
3762 btrfs_dev_stat_inc_and_print(dev,
3763 BTRFS_DEV_STAT_FLUSH_ERRS);
3764 errors_wait++;
3765 }
3766 }
3767
3768 if (errors_wait) {
3769 /*
3770 * At some point we need the status of all disks
3771 * to arrive at the volume status. So error checking
3772 * is being pushed to a separate loop.
3773 */
3774 return check_barrier_error(info);
3775 }
3776 return 0;
3777}
3778
3779int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3780{
3781 int raid_type;
3782 int min_tolerated = INT_MAX;
3783
3784 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3785 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3786 min_tolerated = min_t(int, min_tolerated,
3787 btrfs_raid_array[BTRFS_RAID_SINGLE].
3788 tolerated_failures);
3789
3790 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3791 if (raid_type == BTRFS_RAID_SINGLE)
3792 continue;
3793 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3794 continue;
3795 min_tolerated = min_t(int, min_tolerated,
3796 btrfs_raid_array[raid_type].
3797 tolerated_failures);
3798 }
3799
3800 if (min_tolerated == INT_MAX) {
3801 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3802 min_tolerated = 0;
3803 }
3804
3805 return min_tolerated;
3806}
3807
3808int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3809{
3810 struct list_head *head;
3811 struct btrfs_device *dev;
3812 struct btrfs_super_block *sb;
3813 struct btrfs_dev_item *dev_item;
3814 int ret;
3815 int do_barriers;
3816 int max_errors;
3817 int total_errors = 0;
3818 u64 flags;
3819
3820 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3821
3822 /*
3823 * max_mirrors == 0 indicates we're from commit_transaction,
3824 * not from fsync where the tree roots in fs_info have not
3825 * been consistent on disk.
3826 */
3827 if (max_mirrors == 0)
3828 backup_super_roots(fs_info);
3829
3830 sb = fs_info->super_for_commit;
3831 dev_item = &sb->dev_item;
3832
3833 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3834 head = &fs_info->fs_devices->devices;
3835 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3836
3837 if (do_barriers) {
3838 ret = barrier_all_devices(fs_info);
3839 if (ret) {
3840 mutex_unlock(
3841 &fs_info->fs_devices->device_list_mutex);
3842 btrfs_handle_fs_error(fs_info, ret,
3843 "errors while submitting device barriers.");
3844 return ret;
3845 }
3846 }
3847
3848 list_for_each_entry(dev, head, dev_list) {
3849 if (!dev->bdev) {
3850 total_errors++;
3851 continue;
3852 }
3853 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3854 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3855 continue;
3856
3857 btrfs_set_stack_device_generation(dev_item, 0);
3858 btrfs_set_stack_device_type(dev_item, dev->type);
3859 btrfs_set_stack_device_id(dev_item, dev->devid);
3860 btrfs_set_stack_device_total_bytes(dev_item,
3861 dev->commit_total_bytes);
3862 btrfs_set_stack_device_bytes_used(dev_item,
3863 dev->commit_bytes_used);
3864 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3865 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3866 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3867 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3868 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3869 BTRFS_FSID_SIZE);
3870
3871 flags = btrfs_super_flags(sb);
3872 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3873
3874 ret = btrfs_validate_write_super(fs_info, sb);
3875 if (ret < 0) {
3876 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3877 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3878 "unexpected superblock corruption detected");
3879 return -EUCLEAN;
3880 }
3881
3882 ret = write_dev_supers(dev, sb, max_mirrors);
3883 if (ret)
3884 total_errors++;
3885 }
3886 if (total_errors > max_errors) {
3887 btrfs_err(fs_info, "%d errors while writing supers",
3888 total_errors);
3889 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3890
3891 /* FUA is masked off if unsupported and can't be the reason */
3892 btrfs_handle_fs_error(fs_info, -EIO,
3893 "%d errors while writing supers",
3894 total_errors);
3895 return -EIO;
3896 }
3897
3898 total_errors = 0;
3899 list_for_each_entry(dev, head, dev_list) {
3900 if (!dev->bdev)
3901 continue;
3902 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3903 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3904 continue;
3905
3906 ret = wait_dev_supers(dev, max_mirrors);
3907 if (ret)
3908 total_errors++;
3909 }
3910 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3911 if (total_errors > max_errors) {
3912 btrfs_handle_fs_error(fs_info, -EIO,
3913 "%d errors while writing supers",
3914 total_errors);
3915 return -EIO;
3916 }
3917 return 0;
3918}
3919
3920/* Drop a fs root from the radix tree and free it. */
3921void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3922 struct btrfs_root *root)
3923{
3924 bool drop_ref = false;
3925
3926 spin_lock(&fs_info->fs_roots_radix_lock);
3927 radix_tree_delete(&fs_info->fs_roots_radix,
3928 (unsigned long)root->root_key.objectid);
3929 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3930 drop_ref = true;
3931 spin_unlock(&fs_info->fs_roots_radix_lock);
3932
3933 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3934 ASSERT(root->log_root == NULL);
3935 if (root->reloc_root) {
3936 btrfs_put_root(root->reloc_root);
3937 root->reloc_root = NULL;
3938 }
3939 }
3940
3941 if (root->free_ino_pinned)
3942 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3943 if (root->free_ino_ctl)
3944 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3945 if (root->ino_cache_inode) {
3946 iput(root->ino_cache_inode);
3947 root->ino_cache_inode = NULL;
3948 }
3949 if (drop_ref)
3950 btrfs_put_root(root);
3951}
3952
3953int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3954{
3955 u64 root_objectid = 0;
3956 struct btrfs_root *gang[8];
3957 int i = 0;
3958 int err = 0;
3959 unsigned int ret = 0;
3960
3961 while (1) {
3962 spin_lock(&fs_info->fs_roots_radix_lock);
3963 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3964 (void **)gang, root_objectid,
3965 ARRAY_SIZE(gang));
3966 if (!ret) {
3967 spin_unlock(&fs_info->fs_roots_radix_lock);
3968 break;
3969 }
3970 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3971
3972 for (i = 0; i < ret; i++) {
3973 /* Avoid to grab roots in dead_roots */
3974 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3975 gang[i] = NULL;
3976 continue;
3977 }
3978 /* grab all the search result for later use */
3979 gang[i] = btrfs_grab_root(gang[i]);
3980 }
3981 spin_unlock(&fs_info->fs_roots_radix_lock);
3982
3983 for (i = 0; i < ret; i++) {
3984 if (!gang[i])
3985 continue;
3986 root_objectid = gang[i]->root_key.objectid;
3987 err = btrfs_orphan_cleanup(gang[i]);
3988 if (err)
3989 break;
3990 btrfs_put_root(gang[i]);
3991 }
3992 root_objectid++;
3993 }
3994
3995 /* release the uncleaned roots due to error */
3996 for (; i < ret; i++) {
3997 if (gang[i])
3998 btrfs_put_root(gang[i]);
3999 }
4000 return err;
4001}
4002
4003int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4004{
4005 struct btrfs_root *root = fs_info->tree_root;
4006 struct btrfs_trans_handle *trans;
4007
4008 mutex_lock(&fs_info->cleaner_mutex);
4009 btrfs_run_delayed_iputs(fs_info);
4010 mutex_unlock(&fs_info->cleaner_mutex);
4011 wake_up_process(fs_info->cleaner_kthread);
4012
4013 /* wait until ongoing cleanup work done */
4014 down_write(&fs_info->cleanup_work_sem);
4015 up_write(&fs_info->cleanup_work_sem);
4016
4017 trans = btrfs_join_transaction(root);
4018 if (IS_ERR(trans))
4019 return PTR_ERR(trans);
4020 return btrfs_commit_transaction(trans);
4021}
4022
4023void __cold close_ctree(struct btrfs_fs_info *fs_info)
4024{
4025 int ret;
4026
4027 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4028 /*
4029 * We don't want the cleaner to start new transactions, add more delayed
4030 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4031 * because that frees the task_struct, and the transaction kthread might
4032 * still try to wake up the cleaner.
4033 */
4034 kthread_park(fs_info->cleaner_kthread);
4035
4036 /* wait for the qgroup rescan worker to stop */
4037 btrfs_qgroup_wait_for_completion(fs_info, false);
4038
4039 /* wait for the uuid_scan task to finish */
4040 down(&fs_info->uuid_tree_rescan_sem);
4041 /* avoid complains from lockdep et al., set sem back to initial state */
4042 up(&fs_info->uuid_tree_rescan_sem);
4043
4044 /* pause restriper - we want to resume on mount */
4045 btrfs_pause_balance(fs_info);
4046
4047 btrfs_dev_replace_suspend_for_unmount(fs_info);
4048
4049 btrfs_scrub_cancel(fs_info);
4050
4051 /* wait for any defraggers to finish */
4052 wait_event(fs_info->transaction_wait,
4053 (atomic_read(&fs_info->defrag_running) == 0));
4054
4055 /* clear out the rbtree of defraggable inodes */
4056 btrfs_cleanup_defrag_inodes(fs_info);
4057
4058 cancel_work_sync(&fs_info->async_reclaim_work);
4059
4060 /* Cancel or finish ongoing discard work */
4061 btrfs_discard_cleanup(fs_info);
4062
4063 if (!sb_rdonly(fs_info->sb)) {
4064 /*
4065 * The cleaner kthread is stopped, so do one final pass over
4066 * unused block groups.
4067 */
4068 btrfs_delete_unused_bgs(fs_info);
4069
4070 /*
4071 * There might be existing delayed inode workers still running
4072 * and holding an empty delayed inode item. We must wait for
4073 * them to complete first because they can create a transaction.
4074 * This happens when someone calls btrfs_balance_delayed_items()
4075 * and then a transaction commit runs the same delayed nodes
4076 * before any delayed worker has done something with the nodes.
4077 * We must wait for any worker here and not at transaction
4078 * commit time since that could cause a deadlock.
4079 * This is a very rare case.
4080 */
4081 btrfs_flush_workqueue(fs_info->delayed_workers);
4082
4083 ret = btrfs_commit_super(fs_info);
4084 if (ret)
4085 btrfs_err(fs_info, "commit super ret %d", ret);
4086 }
4087
4088 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4089 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4090 btrfs_error_commit_super(fs_info);
4091
4092 kthread_stop(fs_info->transaction_kthread);
4093 kthread_stop(fs_info->cleaner_kthread);
4094
4095 ASSERT(list_empty(&fs_info->delayed_iputs));
4096 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4097
4098 if (btrfs_check_quota_leak(fs_info)) {
4099 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4100 btrfs_err(fs_info, "qgroup reserved space leaked");
4101 }
4102
4103 btrfs_free_qgroup_config(fs_info);
4104 ASSERT(list_empty(&fs_info->delalloc_roots));
4105
4106 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4107 btrfs_info(fs_info, "at unmount delalloc count %lld",
4108 percpu_counter_sum(&fs_info->delalloc_bytes));
4109 }
4110
4111 if (percpu_counter_sum(&fs_info->dio_bytes))
4112 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4113 percpu_counter_sum(&fs_info->dio_bytes));
4114
4115 btrfs_sysfs_remove_mounted(fs_info);
4116 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4117
4118 btrfs_put_block_group_cache(fs_info);
4119
4120 /*
4121 * we must make sure there is not any read request to
4122 * submit after we stopping all workers.
4123 */
4124 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4125 btrfs_stop_all_workers(fs_info);
4126
4127 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4128 free_root_pointers(fs_info, true);
4129 btrfs_free_fs_roots(fs_info);
4130
4131 /*
4132 * We must free the block groups after dropping the fs_roots as we could
4133 * have had an IO error and have left over tree log blocks that aren't
4134 * cleaned up until the fs roots are freed. This makes the block group
4135 * accounting appear to be wrong because there's pending reserved bytes,
4136 * so make sure we do the block group cleanup afterwards.
4137 */
4138 btrfs_free_block_groups(fs_info);
4139
4140 iput(fs_info->btree_inode);
4141
4142#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4143 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4144 btrfsic_unmount(fs_info->fs_devices);
4145#endif
4146
4147 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4148 btrfs_close_devices(fs_info->fs_devices);
4149}
4150
4151int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4152 int atomic)
4153{
4154 int ret;
4155 struct inode *btree_inode = buf->pages[0]->mapping->host;
4156
4157 ret = extent_buffer_uptodate(buf);
4158 if (!ret)
4159 return ret;
4160
4161 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4162 parent_transid, atomic);
4163 if (ret == -EAGAIN)
4164 return ret;
4165 return !ret;
4166}
4167
4168void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4169{
4170 struct btrfs_fs_info *fs_info;
4171 struct btrfs_root *root;
4172 u64 transid = btrfs_header_generation(buf);
4173 int was_dirty;
4174
4175#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4176 /*
4177 * This is a fast path so only do this check if we have sanity tests
4178 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4179 * outside of the sanity tests.
4180 */
4181 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4182 return;
4183#endif
4184 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4185 fs_info = root->fs_info;
4186 btrfs_assert_tree_locked(buf);
4187 if (transid != fs_info->generation)
4188 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4189 buf->start, transid, fs_info->generation);
4190 was_dirty = set_extent_buffer_dirty(buf);
4191 if (!was_dirty)
4192 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4193 buf->len,
4194 fs_info->dirty_metadata_batch);
4195#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4196 /*
4197 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4198 * but item data not updated.
4199 * So here we should only check item pointers, not item data.
4200 */
4201 if (btrfs_header_level(buf) == 0 &&
4202 btrfs_check_leaf_relaxed(buf)) {
4203 btrfs_print_leaf(buf);
4204 ASSERT(0);
4205 }
4206#endif
4207}
4208
4209static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4210 int flush_delayed)
4211{
4212 /*
4213 * looks as though older kernels can get into trouble with
4214 * this code, they end up stuck in balance_dirty_pages forever
4215 */
4216 int ret;
4217
4218 if (current->flags & PF_MEMALLOC)
4219 return;
4220
4221 if (flush_delayed)
4222 btrfs_balance_delayed_items(fs_info);
4223
4224 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4225 BTRFS_DIRTY_METADATA_THRESH,
4226 fs_info->dirty_metadata_batch);
4227 if (ret > 0) {
4228 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4229 }
4230}
4231
4232void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4233{
4234 __btrfs_btree_balance_dirty(fs_info, 1);
4235}
4236
4237void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4238{
4239 __btrfs_btree_balance_dirty(fs_info, 0);
4240}
4241
4242int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4243 struct btrfs_key *first_key)
4244{
4245 return btree_read_extent_buffer_pages(buf, parent_transid,
4246 level, first_key);
4247}
4248
4249static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4250{
4251 /* cleanup FS via transaction */
4252 btrfs_cleanup_transaction(fs_info);
4253
4254 mutex_lock(&fs_info->cleaner_mutex);
4255 btrfs_run_delayed_iputs(fs_info);
4256 mutex_unlock(&fs_info->cleaner_mutex);
4257
4258 down_write(&fs_info->cleanup_work_sem);
4259 up_write(&fs_info->cleanup_work_sem);
4260}
4261
4262static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4263{
4264 struct btrfs_root *gang[8];
4265 u64 root_objectid = 0;
4266 int ret;
4267
4268 spin_lock(&fs_info->fs_roots_radix_lock);
4269 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4270 (void **)gang, root_objectid,
4271 ARRAY_SIZE(gang))) != 0) {
4272 int i;
4273
4274 for (i = 0; i < ret; i++)
4275 gang[i] = btrfs_grab_root(gang[i]);
4276 spin_unlock(&fs_info->fs_roots_radix_lock);
4277
4278 for (i = 0; i < ret; i++) {
4279 if (!gang[i])
4280 continue;
4281 root_objectid = gang[i]->root_key.objectid;
4282 btrfs_free_log(NULL, gang[i]);
4283 btrfs_put_root(gang[i]);
4284 }
4285 root_objectid++;
4286 spin_lock(&fs_info->fs_roots_radix_lock);
4287 }
4288 spin_unlock(&fs_info->fs_roots_radix_lock);
4289 btrfs_free_log_root_tree(NULL, fs_info);
4290}
4291
4292static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4293{
4294 struct btrfs_ordered_extent *ordered;
4295
4296 spin_lock(&root->ordered_extent_lock);
4297 /*
4298 * This will just short circuit the ordered completion stuff which will
4299 * make sure the ordered extent gets properly cleaned up.
4300 */
4301 list_for_each_entry(ordered, &root->ordered_extents,
4302 root_extent_list)
4303 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4304 spin_unlock(&root->ordered_extent_lock);
4305}
4306
4307static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4308{
4309 struct btrfs_root *root;
4310 struct list_head splice;
4311
4312 INIT_LIST_HEAD(&splice);
4313
4314 spin_lock(&fs_info->ordered_root_lock);
4315 list_splice_init(&fs_info->ordered_roots, &splice);
4316 while (!list_empty(&splice)) {
4317 root = list_first_entry(&splice, struct btrfs_root,
4318 ordered_root);
4319 list_move_tail(&root->ordered_root,
4320 &fs_info->ordered_roots);
4321
4322 spin_unlock(&fs_info->ordered_root_lock);
4323 btrfs_destroy_ordered_extents(root);
4324
4325 cond_resched();
4326 spin_lock(&fs_info->ordered_root_lock);
4327 }
4328 spin_unlock(&fs_info->ordered_root_lock);
4329
4330 /*
4331 * We need this here because if we've been flipped read-only we won't
4332 * get sync() from the umount, so we need to make sure any ordered
4333 * extents that haven't had their dirty pages IO start writeout yet
4334 * actually get run and error out properly.
4335 */
4336 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4337}
4338
4339static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4340 struct btrfs_fs_info *fs_info)
4341{
4342 struct rb_node *node;
4343 struct btrfs_delayed_ref_root *delayed_refs;
4344 struct btrfs_delayed_ref_node *ref;
4345 int ret = 0;
4346
4347 delayed_refs = &trans->delayed_refs;
4348
4349 spin_lock(&delayed_refs->lock);
4350 if (atomic_read(&delayed_refs->num_entries) == 0) {
4351 spin_unlock(&delayed_refs->lock);
4352 btrfs_debug(fs_info, "delayed_refs has NO entry");
4353 return ret;
4354 }
4355
4356 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4357 struct btrfs_delayed_ref_head *head;
4358 struct rb_node *n;
4359 bool pin_bytes = false;
4360
4361 head = rb_entry(node, struct btrfs_delayed_ref_head,
4362 href_node);
4363 if (btrfs_delayed_ref_lock(delayed_refs, head))
4364 continue;
4365
4366 spin_lock(&head->lock);
4367 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4368 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4369 ref_node);
4370 ref->in_tree = 0;
4371 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4372 RB_CLEAR_NODE(&ref->ref_node);
4373 if (!list_empty(&ref->add_list))
4374 list_del(&ref->add_list);
4375 atomic_dec(&delayed_refs->num_entries);
4376 btrfs_put_delayed_ref(ref);
4377 }
4378 if (head->must_insert_reserved)
4379 pin_bytes = true;
4380 btrfs_free_delayed_extent_op(head->extent_op);
4381 btrfs_delete_ref_head(delayed_refs, head);
4382 spin_unlock(&head->lock);
4383 spin_unlock(&delayed_refs->lock);
4384 mutex_unlock(&head->mutex);
4385
4386 if (pin_bytes) {
4387 struct btrfs_block_group *cache;
4388
4389 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4390 BUG_ON(!cache);
4391
4392 spin_lock(&cache->space_info->lock);
4393 spin_lock(&cache->lock);
4394 cache->pinned += head->num_bytes;
4395 btrfs_space_info_update_bytes_pinned(fs_info,
4396 cache->space_info, head->num_bytes);
4397 cache->reserved -= head->num_bytes;
4398 cache->space_info->bytes_reserved -= head->num_bytes;
4399 spin_unlock(&cache->lock);
4400 spin_unlock(&cache->space_info->lock);
4401 percpu_counter_add_batch(
4402 &cache->space_info->total_bytes_pinned,
4403 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4404
4405 btrfs_put_block_group(cache);
4406
4407 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4408 head->bytenr + head->num_bytes - 1);
4409 }
4410 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4411 btrfs_put_delayed_ref_head(head);
4412 cond_resched();
4413 spin_lock(&delayed_refs->lock);
4414 }
4415 btrfs_qgroup_destroy_extent_records(trans);
4416
4417 spin_unlock(&delayed_refs->lock);
4418
4419 return ret;
4420}
4421
4422static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4423{
4424 struct btrfs_inode *btrfs_inode;
4425 struct list_head splice;
4426
4427 INIT_LIST_HEAD(&splice);
4428
4429 spin_lock(&root->delalloc_lock);
4430 list_splice_init(&root->delalloc_inodes, &splice);
4431
4432 while (!list_empty(&splice)) {
4433 struct inode *inode = NULL;
4434 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4435 delalloc_inodes);
4436 __btrfs_del_delalloc_inode(root, btrfs_inode);
4437 spin_unlock(&root->delalloc_lock);
4438
4439 /*
4440 * Make sure we get a live inode and that it'll not disappear
4441 * meanwhile.
4442 */
4443 inode = igrab(&btrfs_inode->vfs_inode);
4444 if (inode) {
4445 invalidate_inode_pages2(inode->i_mapping);
4446 iput(inode);
4447 }
4448 spin_lock(&root->delalloc_lock);
4449 }
4450 spin_unlock(&root->delalloc_lock);
4451}
4452
4453static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4454{
4455 struct btrfs_root *root;
4456 struct list_head splice;
4457
4458 INIT_LIST_HEAD(&splice);
4459
4460 spin_lock(&fs_info->delalloc_root_lock);
4461 list_splice_init(&fs_info->delalloc_roots, &splice);
4462 while (!list_empty(&splice)) {
4463 root = list_first_entry(&splice, struct btrfs_root,
4464 delalloc_root);
4465 root = btrfs_grab_root(root);
4466 BUG_ON(!root);
4467 spin_unlock(&fs_info->delalloc_root_lock);
4468
4469 btrfs_destroy_delalloc_inodes(root);
4470 btrfs_put_root(root);
4471
4472 spin_lock(&fs_info->delalloc_root_lock);
4473 }
4474 spin_unlock(&fs_info->delalloc_root_lock);
4475}
4476
4477static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4478 struct extent_io_tree *dirty_pages,
4479 int mark)
4480{
4481 int ret;
4482 struct extent_buffer *eb;
4483 u64 start = 0;
4484 u64 end;
4485
4486 while (1) {
4487 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4488 mark, NULL);
4489 if (ret)
4490 break;
4491
4492 clear_extent_bits(dirty_pages, start, end, mark);
4493 while (start <= end) {
4494 eb = find_extent_buffer(fs_info, start);
4495 start += fs_info->nodesize;
4496 if (!eb)
4497 continue;
4498 wait_on_extent_buffer_writeback(eb);
4499
4500 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4501 &eb->bflags))
4502 clear_extent_buffer_dirty(eb);
4503 free_extent_buffer_stale(eb);
4504 }
4505 }
4506
4507 return ret;
4508}
4509
4510static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4511 struct extent_io_tree *unpin)
4512{
4513 u64 start;
4514 u64 end;
4515 int ret;
4516
4517 while (1) {
4518 struct extent_state *cached_state = NULL;
4519
4520 /*
4521 * The btrfs_finish_extent_commit() may get the same range as
4522 * ours between find_first_extent_bit and clear_extent_dirty.
4523 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4524 * the same extent range.
4525 */
4526 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4527 ret = find_first_extent_bit(unpin, 0, &start, &end,
4528 EXTENT_DIRTY, &cached_state);
4529 if (ret) {
4530 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4531 break;
4532 }
4533
4534 clear_extent_dirty(unpin, start, end, &cached_state);
4535 free_extent_state(cached_state);
4536 btrfs_error_unpin_extent_range(fs_info, start, end);
4537 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4538 cond_resched();
4539 }
4540
4541 return 0;
4542}
4543
4544static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4545{
4546 struct inode *inode;
4547
4548 inode = cache->io_ctl.inode;
4549 if (inode) {
4550 invalidate_inode_pages2(inode->i_mapping);
4551 BTRFS_I(inode)->generation = 0;
4552 cache->io_ctl.inode = NULL;
4553 iput(inode);
4554 }
4555 ASSERT(cache->io_ctl.pages == NULL);
4556 btrfs_put_block_group(cache);
4557}
4558
4559void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4560 struct btrfs_fs_info *fs_info)
4561{
4562 struct btrfs_block_group *cache;
4563
4564 spin_lock(&cur_trans->dirty_bgs_lock);
4565 while (!list_empty(&cur_trans->dirty_bgs)) {
4566 cache = list_first_entry(&cur_trans->dirty_bgs,
4567 struct btrfs_block_group,
4568 dirty_list);
4569
4570 if (!list_empty(&cache->io_list)) {
4571 spin_unlock(&cur_trans->dirty_bgs_lock);
4572 list_del_init(&cache->io_list);
4573 btrfs_cleanup_bg_io(cache);
4574 spin_lock(&cur_trans->dirty_bgs_lock);
4575 }
4576
4577 list_del_init(&cache->dirty_list);
4578 spin_lock(&cache->lock);
4579 cache->disk_cache_state = BTRFS_DC_ERROR;
4580 spin_unlock(&cache->lock);
4581
4582 spin_unlock(&cur_trans->dirty_bgs_lock);
4583 btrfs_put_block_group(cache);
4584 btrfs_delayed_refs_rsv_release(fs_info, 1);
4585 spin_lock(&cur_trans->dirty_bgs_lock);
4586 }
4587 spin_unlock(&cur_trans->dirty_bgs_lock);
4588
4589 /*
4590 * Refer to the definition of io_bgs member for details why it's safe
4591 * to use it without any locking
4592 */
4593 while (!list_empty(&cur_trans->io_bgs)) {
4594 cache = list_first_entry(&cur_trans->io_bgs,
4595 struct btrfs_block_group,
4596 io_list);
4597
4598 list_del_init(&cache->io_list);
4599 spin_lock(&cache->lock);
4600 cache->disk_cache_state = BTRFS_DC_ERROR;
4601 spin_unlock(&cache->lock);
4602 btrfs_cleanup_bg_io(cache);
4603 }
4604}
4605
4606void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4607 struct btrfs_fs_info *fs_info)
4608{
4609 struct btrfs_device *dev, *tmp;
4610
4611 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4612 ASSERT(list_empty(&cur_trans->dirty_bgs));
4613 ASSERT(list_empty(&cur_trans->io_bgs));
4614
4615 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4616 post_commit_list) {
4617 list_del_init(&dev->post_commit_list);
4618 }
4619
4620 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4621
4622 cur_trans->state = TRANS_STATE_COMMIT_START;
4623 wake_up(&fs_info->transaction_blocked_wait);
4624
4625 cur_trans->state = TRANS_STATE_UNBLOCKED;
4626 wake_up(&fs_info->transaction_wait);
4627
4628 btrfs_destroy_delayed_inodes(fs_info);
4629
4630 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4631 EXTENT_DIRTY);
4632 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4633
4634 cur_trans->state =TRANS_STATE_COMPLETED;
4635 wake_up(&cur_trans->commit_wait);
4636}
4637
4638static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4639{
4640 struct btrfs_transaction *t;
4641
4642 mutex_lock(&fs_info->transaction_kthread_mutex);
4643
4644 spin_lock(&fs_info->trans_lock);
4645 while (!list_empty(&fs_info->trans_list)) {
4646 t = list_first_entry(&fs_info->trans_list,
4647 struct btrfs_transaction, list);
4648 if (t->state >= TRANS_STATE_COMMIT_START) {
4649 refcount_inc(&t->use_count);
4650 spin_unlock(&fs_info->trans_lock);
4651 btrfs_wait_for_commit(fs_info, t->transid);
4652 btrfs_put_transaction(t);
4653 spin_lock(&fs_info->trans_lock);
4654 continue;
4655 }
4656 if (t == fs_info->running_transaction) {
4657 t->state = TRANS_STATE_COMMIT_DOING;
4658 spin_unlock(&fs_info->trans_lock);
4659 /*
4660 * We wait for 0 num_writers since we don't hold a trans
4661 * handle open currently for this transaction.
4662 */
4663 wait_event(t->writer_wait,
4664 atomic_read(&t->num_writers) == 0);
4665 } else {
4666 spin_unlock(&fs_info->trans_lock);
4667 }
4668 btrfs_cleanup_one_transaction(t, fs_info);
4669
4670 spin_lock(&fs_info->trans_lock);
4671 if (t == fs_info->running_transaction)
4672 fs_info->running_transaction = NULL;
4673 list_del_init(&t->list);
4674 spin_unlock(&fs_info->trans_lock);
4675
4676 btrfs_put_transaction(t);
4677 trace_btrfs_transaction_commit(fs_info->tree_root);
4678 spin_lock(&fs_info->trans_lock);
4679 }
4680 spin_unlock(&fs_info->trans_lock);
4681 btrfs_destroy_all_ordered_extents(fs_info);
4682 btrfs_destroy_delayed_inodes(fs_info);
4683 btrfs_assert_delayed_root_empty(fs_info);
4684 btrfs_destroy_all_delalloc_inodes(fs_info);
4685 btrfs_drop_all_logs(fs_info);
4686 mutex_unlock(&fs_info->transaction_kthread_mutex);
4687
4688 return 0;
4689}
4690
4691static const struct extent_io_ops btree_extent_io_ops = {
4692 /* mandatory callbacks */
4693 .submit_bio_hook = btree_submit_bio_hook,
4694 .readpage_end_io_hook = btree_readpage_end_io_hook,
4695};