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/buffer_head.h>
11#include <linux/workqueue.h>
12#include <linux/kthread.h>
13#include <linux/slab.h>
14#include <linux/migrate.h>
15#include <linux/ratelimit.h>
16#include <linux/uuid.h>
17#include <linux/semaphore.h>
18#include <linux/error-injection.h>
19#include <linux/crc32c.h>
20#include <linux/sched/mm.h>
21#include <asm/unaligned.h>
22#include <crypto/hash.h>
23#include "ctree.h"
24#include "disk-io.h"
25#include "transaction.h"
26#include "btrfs_inode.h"
27#include "volumes.h"
28#include "print-tree.h"
29#include "locking.h"
30#include "tree-log.h"
31#include "free-space-cache.h"
32#include "free-space-tree.h"
33#include "inode-map.h"
34#include "check-integrity.h"
35#include "rcu-string.h"
36#include "dev-replace.h"
37#include "raid56.h"
38#include "sysfs.h"
39#include "qgroup.h"
40#include "compression.h"
41#include "tree-checker.h"
42#include "ref-verify.h"
43#include "block-group.h"
44
45#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
46 BTRFS_HEADER_FLAG_RELOC |\
47 BTRFS_SUPER_FLAG_ERROR |\
48 BTRFS_SUPER_FLAG_SEEDING |\
49 BTRFS_SUPER_FLAG_METADUMP |\
50 BTRFS_SUPER_FLAG_METADUMP_V2)
51
52static const struct extent_io_ops btree_extent_io_ops;
53static void end_workqueue_fn(struct btrfs_work *work);
54static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
60 int mark);
61static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
65
66/*
67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
68 * is complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
70 */
71struct btrfs_end_io_wq {
72 struct bio *bio;
73 bio_end_io_t *end_io;
74 void *private;
75 struct btrfs_fs_info *info;
76 blk_status_t status;
77 enum btrfs_wq_endio_type metadata;
78 struct btrfs_work work;
79};
80
81static struct kmem_cache *btrfs_end_io_wq_cache;
82
83int __init btrfs_end_io_wq_init(void)
84{
85 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 sizeof(struct btrfs_end_io_wq),
87 0,
88 SLAB_MEM_SPREAD,
89 NULL);
90 if (!btrfs_end_io_wq_cache)
91 return -ENOMEM;
92 return 0;
93}
94
95void __cold btrfs_end_io_wq_exit(void)
96{
97 kmem_cache_destroy(btrfs_end_io_wq_cache);
98}
99
100/*
101 * async submit bios are used to offload expensive checksumming
102 * onto the worker threads. They checksum file and metadata bios
103 * just before they are sent down the IO stack.
104 */
105struct async_submit_bio {
106 void *private_data;
107 struct bio *bio;
108 extent_submit_bio_start_t *submit_bio_start;
109 int mirror_num;
110 /*
111 * bio_offset is optional, can be used if the pages in the bio
112 * can't tell us where in the file the bio should go
113 */
114 u64 bio_offset;
115 struct btrfs_work work;
116 blk_status_t status;
117};
118
119/*
120 * Lockdep class keys for extent_buffer->lock's in this root. For a given
121 * eb, the lockdep key is determined by the btrfs_root it belongs to and
122 * the level the eb occupies in the tree.
123 *
124 * Different roots are used for different purposes and may nest inside each
125 * other and they require separate keysets. As lockdep keys should be
126 * static, assign keysets according to the purpose of the root as indicated
127 * by btrfs_root->root_key.objectid. This ensures that all special purpose
128 * roots have separate keysets.
129 *
130 * Lock-nesting across peer nodes is always done with the immediate parent
131 * node locked thus preventing deadlock. As lockdep doesn't know this, use
132 * subclass to avoid triggering lockdep warning in such cases.
133 *
134 * The key is set by the readpage_end_io_hook after the buffer has passed
135 * csum validation but before the pages are unlocked. It is also set by
136 * btrfs_init_new_buffer on freshly allocated blocks.
137 *
138 * We also add a check to make sure the highest level of the tree is the
139 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
140 * needs update as well.
141 */
142#ifdef CONFIG_DEBUG_LOCK_ALLOC
143# if BTRFS_MAX_LEVEL != 8
144# error
145# endif
146
147static struct btrfs_lockdep_keyset {
148 u64 id; /* root objectid */
149 const char *name_stem; /* lock name stem */
150 char names[BTRFS_MAX_LEVEL + 1][20];
151 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
152} btrfs_lockdep_keysets[] = {
153 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
154 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
155 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
156 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
157 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
158 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
159 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
160 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
161 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
162 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
163 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
164 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
165 { .id = 0, .name_stem = "tree" },
166};
167
168void __init btrfs_init_lockdep(void)
169{
170 int i, j;
171
172 /* initialize lockdep class names */
173 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
174 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
175
176 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
177 snprintf(ks->names[j], sizeof(ks->names[j]),
178 "btrfs-%s-%02d", ks->name_stem, j);
179 }
180}
181
182void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
183 int level)
184{
185 struct btrfs_lockdep_keyset *ks;
186
187 BUG_ON(level >= ARRAY_SIZE(ks->keys));
188
189 /* find the matching keyset, id 0 is the default entry */
190 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
191 if (ks->id == objectid)
192 break;
193
194 lockdep_set_class_and_name(&eb->lock,
195 &ks->keys[level], ks->names[level]);
196}
197
198#endif
199
200/*
201 * extents on the btree inode are pretty simple, there's one extent
202 * that covers the entire device
203 */
204struct extent_map *btree_get_extent(struct btrfs_inode *inode,
205 struct page *page, size_t pg_offset, u64 start, u64 len,
206 int create)
207{
208 struct btrfs_fs_info *fs_info = inode->root->fs_info;
209 struct extent_map_tree *em_tree = &inode->extent_tree;
210 struct extent_map *em;
211 int ret;
212
213 read_lock(&em_tree->lock);
214 em = lookup_extent_mapping(em_tree, start, len);
215 if (em) {
216 em->bdev = fs_info->fs_devices->latest_bdev;
217 read_unlock(&em_tree->lock);
218 goto out;
219 }
220 read_unlock(&em_tree->lock);
221
222 em = alloc_extent_map();
223 if (!em) {
224 em = ERR_PTR(-ENOMEM);
225 goto out;
226 }
227 em->start = 0;
228 em->len = (u64)-1;
229 em->block_len = (u64)-1;
230 em->block_start = 0;
231 em->bdev = fs_info->fs_devices->latest_bdev;
232
233 write_lock(&em_tree->lock);
234 ret = add_extent_mapping(em_tree, em, 0);
235 if (ret == -EEXIST) {
236 free_extent_map(em);
237 em = lookup_extent_mapping(em_tree, start, len);
238 if (!em)
239 em = ERR_PTR(-EIO);
240 } else if (ret) {
241 free_extent_map(em);
242 em = ERR_PTR(ret);
243 }
244 write_unlock(&em_tree->lock);
245
246out:
247 return em;
248}
249
250/*
251 * Compute the csum of a btree block and store the result to provided buffer.
252 *
253 * Returns error if the extent buffer cannot be mapped.
254 */
255static int csum_tree_block(struct extent_buffer *buf, u8 *result)
256{
257 struct btrfs_fs_info *fs_info = buf->fs_info;
258 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
259 unsigned long len;
260 unsigned long cur_len;
261 unsigned long offset = BTRFS_CSUM_SIZE;
262 char *kaddr;
263 unsigned long map_start;
264 unsigned long map_len;
265 int err;
266
267 shash->tfm = fs_info->csum_shash;
268 crypto_shash_init(shash);
269
270 len = buf->len - offset;
271
272 while (len > 0) {
273 /*
274 * Note: we don't need to check for the err == 1 case here, as
275 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
276 * and 'min_len = 32' and the currently implemented mapping
277 * algorithm we cannot cross a page boundary.
278 */
279 err = map_private_extent_buffer(buf, offset, 32,
280 &kaddr, &map_start, &map_len);
281 if (WARN_ON(err))
282 return err;
283 cur_len = min(len, map_len - (offset - map_start));
284 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
285 len -= cur_len;
286 offset += cur_len;
287 }
288 memset(result, 0, BTRFS_CSUM_SIZE);
289
290 crypto_shash_final(shash, result);
291
292 return 0;
293}
294
295/*
296 * we can't consider a given block up to date unless the transid of the
297 * block matches the transid in the parent node's pointer. This is how we
298 * detect blocks that either didn't get written at all or got written
299 * in the wrong place.
300 */
301static int verify_parent_transid(struct extent_io_tree *io_tree,
302 struct extent_buffer *eb, u64 parent_transid,
303 int atomic)
304{
305 struct extent_state *cached_state = NULL;
306 int ret;
307 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
308
309 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
310 return 0;
311
312 if (atomic)
313 return -EAGAIN;
314
315 if (need_lock) {
316 btrfs_tree_read_lock(eb);
317 btrfs_set_lock_blocking_read(eb);
318 }
319
320 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
321 &cached_state);
322 if (extent_buffer_uptodate(eb) &&
323 btrfs_header_generation(eb) == parent_transid) {
324 ret = 0;
325 goto out;
326 }
327 btrfs_err_rl(eb->fs_info,
328 "parent transid verify failed on %llu wanted %llu found %llu",
329 eb->start,
330 parent_transid, btrfs_header_generation(eb));
331 ret = 1;
332
333 /*
334 * Things reading via commit roots that don't have normal protection,
335 * like send, can have a really old block in cache that may point at a
336 * block that has been freed and re-allocated. So don't clear uptodate
337 * if we find an eb that is under IO (dirty/writeback) because we could
338 * end up reading in the stale data and then writing it back out and
339 * making everybody very sad.
340 */
341 if (!extent_buffer_under_io(eb))
342 clear_extent_buffer_uptodate(eb);
343out:
344 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
345 &cached_state);
346 if (need_lock)
347 btrfs_tree_read_unlock_blocking(eb);
348 return ret;
349}
350
351static bool btrfs_supported_super_csum(u16 csum_type)
352{
353 switch (csum_type) {
354 case BTRFS_CSUM_TYPE_CRC32:
355 return true;
356 default:
357 return false;
358 }
359}
360
361/*
362 * Return 0 if the superblock checksum type matches the checksum value of that
363 * algorithm. Pass the raw disk superblock data.
364 */
365static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
366 char *raw_disk_sb)
367{
368 struct btrfs_super_block *disk_sb =
369 (struct btrfs_super_block *)raw_disk_sb;
370 char result[BTRFS_CSUM_SIZE];
371 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
372
373 shash->tfm = fs_info->csum_shash;
374 crypto_shash_init(shash);
375
376 /*
377 * The super_block structure does not span the whole
378 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
379 * filled with zeros and is included in the checksum.
380 */
381 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
382 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
383 crypto_shash_final(shash, result);
384
385 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
386 return 1;
387
388 return 0;
389}
390
391int btrfs_verify_level_key(struct extent_buffer *eb, int level,
392 struct btrfs_key *first_key, u64 parent_transid)
393{
394 struct btrfs_fs_info *fs_info = eb->fs_info;
395 int found_level;
396 struct btrfs_key found_key;
397 int ret;
398
399 found_level = btrfs_header_level(eb);
400 if (found_level != level) {
401 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
402 KERN_ERR "BTRFS: tree level check failed\n");
403 btrfs_err(fs_info,
404"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
405 eb->start, level, found_level);
406 return -EIO;
407 }
408
409 if (!first_key)
410 return 0;
411
412 /*
413 * For live tree block (new tree blocks in current transaction),
414 * we need proper lock context to avoid race, which is impossible here.
415 * So we only checks tree blocks which is read from disk, whose
416 * generation <= fs_info->last_trans_committed.
417 */
418 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
419 return 0;
420
421 /* We have @first_key, so this @eb must have at least one item */
422 if (btrfs_header_nritems(eb) == 0) {
423 btrfs_err(fs_info,
424 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
425 eb->start);
426 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
427 return -EUCLEAN;
428 }
429
430 if (found_level)
431 btrfs_node_key_to_cpu(eb, &found_key, 0);
432 else
433 btrfs_item_key_to_cpu(eb, &found_key, 0);
434 ret = btrfs_comp_cpu_keys(first_key, &found_key);
435
436 if (ret) {
437 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
438 KERN_ERR "BTRFS: tree first key check failed\n");
439 btrfs_err(fs_info,
440"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
441 eb->start, parent_transid, first_key->objectid,
442 first_key->type, first_key->offset,
443 found_key.objectid, found_key.type,
444 found_key.offset);
445 }
446 return ret;
447}
448
449/*
450 * helper to read a given tree block, doing retries as required when
451 * the checksums don't match and we have alternate mirrors to try.
452 *
453 * @parent_transid: expected transid, skip check if 0
454 * @level: expected level, mandatory check
455 * @first_key: expected key of first slot, skip check if NULL
456 */
457static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
458 u64 parent_transid, int level,
459 struct btrfs_key *first_key)
460{
461 struct btrfs_fs_info *fs_info = eb->fs_info;
462 struct extent_io_tree *io_tree;
463 int failed = 0;
464 int ret;
465 int num_copies = 0;
466 int mirror_num = 0;
467 int failed_mirror = 0;
468
469 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
470 while (1) {
471 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
472 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
473 if (!ret) {
474 if (verify_parent_transid(io_tree, eb,
475 parent_transid, 0))
476 ret = -EIO;
477 else if (btrfs_verify_level_key(eb, level,
478 first_key, parent_transid))
479 ret = -EUCLEAN;
480 else
481 break;
482 }
483
484 num_copies = btrfs_num_copies(fs_info,
485 eb->start, eb->len);
486 if (num_copies == 1)
487 break;
488
489 if (!failed_mirror) {
490 failed = 1;
491 failed_mirror = eb->read_mirror;
492 }
493
494 mirror_num++;
495 if (mirror_num == failed_mirror)
496 mirror_num++;
497
498 if (mirror_num > num_copies)
499 break;
500 }
501
502 if (failed && !ret && failed_mirror)
503 btrfs_repair_eb_io_failure(eb, failed_mirror);
504
505 return ret;
506}
507
508/*
509 * checksum a dirty tree block before IO. This has extra checks to make sure
510 * we only fill in the checksum field in the first page of a multi-page block
511 */
512
513static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
514{
515 u64 start = page_offset(page);
516 u64 found_start;
517 u8 result[BTRFS_CSUM_SIZE];
518 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
519 struct extent_buffer *eb;
520 int ret;
521
522 eb = (struct extent_buffer *)page->private;
523 if (page != eb->pages[0])
524 return 0;
525
526 found_start = btrfs_header_bytenr(eb);
527 /*
528 * Please do not consolidate these warnings into a single if.
529 * It is useful to know what went wrong.
530 */
531 if (WARN_ON(found_start != start))
532 return -EUCLEAN;
533 if (WARN_ON(!PageUptodate(page)))
534 return -EUCLEAN;
535
536 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
537 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
538
539 if (csum_tree_block(eb, result))
540 return -EINVAL;
541
542 if (btrfs_header_level(eb))
543 ret = btrfs_check_node(eb);
544 else
545 ret = btrfs_check_leaf_full(eb);
546
547 if (ret < 0) {
548 btrfs_err(fs_info,
549 "block=%llu write time tree block corruption detected",
550 eb->start);
551 return ret;
552 }
553 write_extent_buffer(eb, result, 0, csum_size);
554
555 return 0;
556}
557
558static int check_tree_block_fsid(struct extent_buffer *eb)
559{
560 struct btrfs_fs_info *fs_info = eb->fs_info;
561 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
562 u8 fsid[BTRFS_FSID_SIZE];
563 int ret = 1;
564
565 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
566 while (fs_devices) {
567 u8 *metadata_uuid;
568
569 /*
570 * Checking the incompat flag is only valid for the current
571 * fs. For seed devices it's forbidden to have their uuid
572 * changed so reading ->fsid in this case is fine
573 */
574 if (fs_devices == fs_info->fs_devices &&
575 btrfs_fs_incompat(fs_info, METADATA_UUID))
576 metadata_uuid = fs_devices->metadata_uuid;
577 else
578 metadata_uuid = fs_devices->fsid;
579
580 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
581 ret = 0;
582 break;
583 }
584 fs_devices = fs_devices->seed;
585 }
586 return ret;
587}
588
589static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
590 u64 phy_offset, struct page *page,
591 u64 start, u64 end, int mirror)
592{
593 u64 found_start;
594 int found_level;
595 struct extent_buffer *eb;
596 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
597 struct btrfs_fs_info *fs_info = root->fs_info;
598 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
599 int ret = 0;
600 u8 result[BTRFS_CSUM_SIZE];
601 int reads_done;
602
603 if (!page->private)
604 goto out;
605
606 eb = (struct extent_buffer *)page->private;
607
608 /* the pending IO might have been the only thing that kept this buffer
609 * in memory. Make sure we have a ref for all this other checks
610 */
611 extent_buffer_get(eb);
612
613 reads_done = atomic_dec_and_test(&eb->io_pages);
614 if (!reads_done)
615 goto err;
616
617 eb->read_mirror = mirror;
618 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
619 ret = -EIO;
620 goto err;
621 }
622
623 found_start = btrfs_header_bytenr(eb);
624 if (found_start != eb->start) {
625 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
626 eb->start, found_start);
627 ret = -EIO;
628 goto err;
629 }
630 if (check_tree_block_fsid(eb)) {
631 btrfs_err_rl(fs_info, "bad fsid on block %llu",
632 eb->start);
633 ret = -EIO;
634 goto err;
635 }
636 found_level = btrfs_header_level(eb);
637 if (found_level >= BTRFS_MAX_LEVEL) {
638 btrfs_err(fs_info, "bad tree block level %d on %llu",
639 (int)btrfs_header_level(eb), eb->start);
640 ret = -EIO;
641 goto err;
642 }
643
644 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
645 eb, found_level);
646
647 ret = csum_tree_block(eb, result);
648 if (ret)
649 goto err;
650
651 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
652 u32 val;
653 u32 found = 0;
654
655 memcpy(&found, result, csum_size);
656
657 read_extent_buffer(eb, &val, 0, csum_size);
658 btrfs_warn_rl(fs_info,
659 "%s checksum verify failed on %llu wanted %x found %x level %d",
660 fs_info->sb->s_id, eb->start,
661 val, found, btrfs_header_level(eb));
662 ret = -EUCLEAN;
663 goto err;
664 }
665
666 /*
667 * If this is a leaf block and it is corrupt, set the corrupt bit so
668 * that we don't try and read the other copies of this block, just
669 * return -EIO.
670 */
671 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
672 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
673 ret = -EIO;
674 }
675
676 if (found_level > 0 && btrfs_check_node(eb))
677 ret = -EIO;
678
679 if (!ret)
680 set_extent_buffer_uptodate(eb);
681 else
682 btrfs_err(fs_info,
683 "block=%llu read time tree block corruption detected",
684 eb->start);
685err:
686 if (reads_done &&
687 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
688 btree_readahead_hook(eb, ret);
689
690 if (ret) {
691 /*
692 * our io error hook is going to dec the io pages
693 * again, we have to make sure it has something
694 * to decrement
695 */
696 atomic_inc(&eb->io_pages);
697 clear_extent_buffer_uptodate(eb);
698 }
699 free_extent_buffer(eb);
700out:
701 return ret;
702}
703
704static void end_workqueue_bio(struct bio *bio)
705{
706 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
707 struct btrfs_fs_info *fs_info;
708 struct btrfs_workqueue *wq;
709 btrfs_work_func_t func;
710
711 fs_info = end_io_wq->info;
712 end_io_wq->status = bio->bi_status;
713
714 if (bio_op(bio) == REQ_OP_WRITE) {
715 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
716 wq = fs_info->endio_meta_write_workers;
717 func = btrfs_endio_meta_write_helper;
718 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
719 wq = fs_info->endio_freespace_worker;
720 func = btrfs_freespace_write_helper;
721 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
722 wq = fs_info->endio_raid56_workers;
723 func = btrfs_endio_raid56_helper;
724 } else {
725 wq = fs_info->endio_write_workers;
726 func = btrfs_endio_write_helper;
727 }
728 } else {
729 if (unlikely(end_io_wq->metadata ==
730 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
731 wq = fs_info->endio_repair_workers;
732 func = btrfs_endio_repair_helper;
733 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
734 wq = fs_info->endio_raid56_workers;
735 func = btrfs_endio_raid56_helper;
736 } else if (end_io_wq->metadata) {
737 wq = fs_info->endio_meta_workers;
738 func = btrfs_endio_meta_helper;
739 } else {
740 wq = fs_info->endio_workers;
741 func = btrfs_endio_helper;
742 }
743 }
744
745 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
746 btrfs_queue_work(wq, &end_io_wq->work);
747}
748
749blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
750 enum btrfs_wq_endio_type metadata)
751{
752 struct btrfs_end_io_wq *end_io_wq;
753
754 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
755 if (!end_io_wq)
756 return BLK_STS_RESOURCE;
757
758 end_io_wq->private = bio->bi_private;
759 end_io_wq->end_io = bio->bi_end_io;
760 end_io_wq->info = info;
761 end_io_wq->status = 0;
762 end_io_wq->bio = bio;
763 end_io_wq->metadata = metadata;
764
765 bio->bi_private = end_io_wq;
766 bio->bi_end_io = end_workqueue_bio;
767 return 0;
768}
769
770static void run_one_async_start(struct btrfs_work *work)
771{
772 struct async_submit_bio *async;
773 blk_status_t ret;
774
775 async = container_of(work, struct async_submit_bio, work);
776 ret = async->submit_bio_start(async->private_data, async->bio,
777 async->bio_offset);
778 if (ret)
779 async->status = ret;
780}
781
782/*
783 * In order to insert checksums into the metadata in large chunks, we wait
784 * until bio submission time. All the pages in the bio are checksummed and
785 * sums are attached onto the ordered extent record.
786 *
787 * At IO completion time the csums attached on the ordered extent record are
788 * inserted into the tree.
789 */
790static void run_one_async_done(struct btrfs_work *work)
791{
792 struct async_submit_bio *async;
793 struct inode *inode;
794 blk_status_t ret;
795
796 async = container_of(work, struct async_submit_bio, work);
797 inode = async->private_data;
798
799 /* If an error occurred we just want to clean up the bio and move on */
800 if (async->status) {
801 async->bio->bi_status = async->status;
802 bio_endio(async->bio);
803 return;
804 }
805
806 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
807 async->mirror_num, 1);
808 if (ret) {
809 async->bio->bi_status = ret;
810 bio_endio(async->bio);
811 }
812}
813
814static void run_one_async_free(struct btrfs_work *work)
815{
816 struct async_submit_bio *async;
817
818 async = container_of(work, struct async_submit_bio, work);
819 kfree(async);
820}
821
822blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
823 int mirror_num, unsigned long bio_flags,
824 u64 bio_offset, void *private_data,
825 extent_submit_bio_start_t *submit_bio_start)
826{
827 struct async_submit_bio *async;
828
829 async = kmalloc(sizeof(*async), GFP_NOFS);
830 if (!async)
831 return BLK_STS_RESOURCE;
832
833 async->private_data = private_data;
834 async->bio = bio;
835 async->mirror_num = mirror_num;
836 async->submit_bio_start = submit_bio_start;
837
838 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
839 run_one_async_done, run_one_async_free);
840
841 async->bio_offset = bio_offset;
842
843 async->status = 0;
844
845 if (op_is_sync(bio->bi_opf))
846 btrfs_set_work_high_priority(&async->work);
847
848 btrfs_queue_work(fs_info->workers, &async->work);
849 return 0;
850}
851
852static blk_status_t btree_csum_one_bio(struct bio *bio)
853{
854 struct bio_vec *bvec;
855 struct btrfs_root *root;
856 int ret = 0;
857 struct bvec_iter_all iter_all;
858
859 ASSERT(!bio_flagged(bio, BIO_CLONED));
860 bio_for_each_segment_all(bvec, bio, iter_all) {
861 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
862 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
863 if (ret)
864 break;
865 }
866
867 return errno_to_blk_status(ret);
868}
869
870static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
871 u64 bio_offset)
872{
873 /*
874 * when we're called for a write, we're already in the async
875 * submission context. Just jump into btrfs_map_bio
876 */
877 return btree_csum_one_bio(bio);
878}
879
880static int check_async_write(struct btrfs_fs_info *fs_info,
881 struct btrfs_inode *bi)
882{
883 if (atomic_read(&bi->sync_writers))
884 return 0;
885 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
886 return 0;
887 return 1;
888}
889
890static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
891 int mirror_num,
892 unsigned long bio_flags)
893{
894 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
895 int async = check_async_write(fs_info, BTRFS_I(inode));
896 blk_status_t ret;
897
898 if (bio_op(bio) != REQ_OP_WRITE) {
899 /*
900 * called for a read, do the setup so that checksum validation
901 * can happen in the async kernel threads
902 */
903 ret = btrfs_bio_wq_end_io(fs_info, bio,
904 BTRFS_WQ_ENDIO_METADATA);
905 if (ret)
906 goto out_w_error;
907 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
908 } else if (!async) {
909 ret = btree_csum_one_bio(bio);
910 if (ret)
911 goto out_w_error;
912 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
913 } else {
914 /*
915 * kthread helpers are used to submit writes so that
916 * checksumming can happen in parallel across all CPUs
917 */
918 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
919 0, inode, btree_submit_bio_start);
920 }
921
922 if (ret)
923 goto out_w_error;
924 return 0;
925
926out_w_error:
927 bio->bi_status = ret;
928 bio_endio(bio);
929 return ret;
930}
931
932#ifdef CONFIG_MIGRATION
933static int btree_migratepage(struct address_space *mapping,
934 struct page *newpage, struct page *page,
935 enum migrate_mode mode)
936{
937 /*
938 * we can't safely write a btree page from here,
939 * we haven't done the locking hook
940 */
941 if (PageDirty(page))
942 return -EAGAIN;
943 /*
944 * Buffers may be managed in a filesystem specific way.
945 * We must have no buffers or drop them.
946 */
947 if (page_has_private(page) &&
948 !try_to_release_page(page, GFP_KERNEL))
949 return -EAGAIN;
950 return migrate_page(mapping, newpage, page, mode);
951}
952#endif
953
954
955static int btree_writepages(struct address_space *mapping,
956 struct writeback_control *wbc)
957{
958 struct btrfs_fs_info *fs_info;
959 int ret;
960
961 if (wbc->sync_mode == WB_SYNC_NONE) {
962
963 if (wbc->for_kupdate)
964 return 0;
965
966 fs_info = BTRFS_I(mapping->host)->root->fs_info;
967 /* this is a bit racy, but that's ok */
968 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
969 BTRFS_DIRTY_METADATA_THRESH,
970 fs_info->dirty_metadata_batch);
971 if (ret < 0)
972 return 0;
973 }
974 return btree_write_cache_pages(mapping, wbc);
975}
976
977static int btree_readpage(struct file *file, struct page *page)
978{
979 struct extent_io_tree *tree;
980 tree = &BTRFS_I(page->mapping->host)->io_tree;
981 return extent_read_full_page(tree, page, btree_get_extent, 0);
982}
983
984static int btree_releasepage(struct page *page, gfp_t gfp_flags)
985{
986 if (PageWriteback(page) || PageDirty(page))
987 return 0;
988
989 return try_release_extent_buffer(page);
990}
991
992static void btree_invalidatepage(struct page *page, unsigned int offset,
993 unsigned int length)
994{
995 struct extent_io_tree *tree;
996 tree = &BTRFS_I(page->mapping->host)->io_tree;
997 extent_invalidatepage(tree, page, offset);
998 btree_releasepage(page, GFP_NOFS);
999 if (PagePrivate(page)) {
1000 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1001 "page private not zero on page %llu",
1002 (unsigned long long)page_offset(page));
1003 ClearPagePrivate(page);
1004 set_page_private(page, 0);
1005 put_page(page);
1006 }
1007}
1008
1009static int btree_set_page_dirty(struct page *page)
1010{
1011#ifdef DEBUG
1012 struct extent_buffer *eb;
1013
1014 BUG_ON(!PagePrivate(page));
1015 eb = (struct extent_buffer *)page->private;
1016 BUG_ON(!eb);
1017 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1018 BUG_ON(!atomic_read(&eb->refs));
1019 btrfs_assert_tree_locked(eb);
1020#endif
1021 return __set_page_dirty_nobuffers(page);
1022}
1023
1024static const struct address_space_operations btree_aops = {
1025 .readpage = btree_readpage,
1026 .writepages = btree_writepages,
1027 .releasepage = btree_releasepage,
1028 .invalidatepage = btree_invalidatepage,
1029#ifdef CONFIG_MIGRATION
1030 .migratepage = btree_migratepage,
1031#endif
1032 .set_page_dirty = btree_set_page_dirty,
1033};
1034
1035void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1036{
1037 struct extent_buffer *buf = NULL;
1038 int ret;
1039
1040 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1041 if (IS_ERR(buf))
1042 return;
1043
1044 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1045 if (ret < 0)
1046 free_extent_buffer_stale(buf);
1047 else
1048 free_extent_buffer(buf);
1049}
1050
1051struct extent_buffer *btrfs_find_create_tree_block(
1052 struct btrfs_fs_info *fs_info,
1053 u64 bytenr)
1054{
1055 if (btrfs_is_testing(fs_info))
1056 return alloc_test_extent_buffer(fs_info, bytenr);
1057 return alloc_extent_buffer(fs_info, bytenr);
1058}
1059
1060/*
1061 * Read tree block at logical address @bytenr and do variant basic but critical
1062 * verification.
1063 *
1064 * @parent_transid: expected transid of this tree block, skip check if 0
1065 * @level: expected level, mandatory check
1066 * @first_key: expected key in slot 0, skip check if NULL
1067 */
1068struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1069 u64 parent_transid, int level,
1070 struct btrfs_key *first_key)
1071{
1072 struct extent_buffer *buf = NULL;
1073 int ret;
1074
1075 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1076 if (IS_ERR(buf))
1077 return buf;
1078
1079 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1080 level, first_key);
1081 if (ret) {
1082 free_extent_buffer_stale(buf);
1083 return ERR_PTR(ret);
1084 }
1085 return buf;
1086
1087}
1088
1089void btrfs_clean_tree_block(struct extent_buffer *buf)
1090{
1091 struct btrfs_fs_info *fs_info = buf->fs_info;
1092 if (btrfs_header_generation(buf) ==
1093 fs_info->running_transaction->transid) {
1094 btrfs_assert_tree_locked(buf);
1095
1096 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1097 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1098 -buf->len,
1099 fs_info->dirty_metadata_batch);
1100 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1101 btrfs_set_lock_blocking_write(buf);
1102 clear_extent_buffer_dirty(buf);
1103 }
1104 }
1105}
1106
1107static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1108{
1109 struct btrfs_subvolume_writers *writers;
1110 int ret;
1111
1112 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1113 if (!writers)
1114 return ERR_PTR(-ENOMEM);
1115
1116 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1117 if (ret < 0) {
1118 kfree(writers);
1119 return ERR_PTR(ret);
1120 }
1121
1122 init_waitqueue_head(&writers->wait);
1123 return writers;
1124}
1125
1126static void
1127btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1128{
1129 percpu_counter_destroy(&writers->counter);
1130 kfree(writers);
1131}
1132
1133static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1134 u64 objectid)
1135{
1136 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1137 root->node = NULL;
1138 root->commit_root = NULL;
1139 root->state = 0;
1140 root->orphan_cleanup_state = 0;
1141
1142 root->last_trans = 0;
1143 root->highest_objectid = 0;
1144 root->nr_delalloc_inodes = 0;
1145 root->nr_ordered_extents = 0;
1146 root->inode_tree = RB_ROOT;
1147 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1148 root->block_rsv = NULL;
1149
1150 INIT_LIST_HEAD(&root->dirty_list);
1151 INIT_LIST_HEAD(&root->root_list);
1152 INIT_LIST_HEAD(&root->delalloc_inodes);
1153 INIT_LIST_HEAD(&root->delalloc_root);
1154 INIT_LIST_HEAD(&root->ordered_extents);
1155 INIT_LIST_HEAD(&root->ordered_root);
1156 INIT_LIST_HEAD(&root->reloc_dirty_list);
1157 INIT_LIST_HEAD(&root->logged_list[0]);
1158 INIT_LIST_HEAD(&root->logged_list[1]);
1159 spin_lock_init(&root->inode_lock);
1160 spin_lock_init(&root->delalloc_lock);
1161 spin_lock_init(&root->ordered_extent_lock);
1162 spin_lock_init(&root->accounting_lock);
1163 spin_lock_init(&root->log_extents_lock[0]);
1164 spin_lock_init(&root->log_extents_lock[1]);
1165 spin_lock_init(&root->qgroup_meta_rsv_lock);
1166 mutex_init(&root->objectid_mutex);
1167 mutex_init(&root->log_mutex);
1168 mutex_init(&root->ordered_extent_mutex);
1169 mutex_init(&root->delalloc_mutex);
1170 init_waitqueue_head(&root->log_writer_wait);
1171 init_waitqueue_head(&root->log_commit_wait[0]);
1172 init_waitqueue_head(&root->log_commit_wait[1]);
1173 INIT_LIST_HEAD(&root->log_ctxs[0]);
1174 INIT_LIST_HEAD(&root->log_ctxs[1]);
1175 atomic_set(&root->log_commit[0], 0);
1176 atomic_set(&root->log_commit[1], 0);
1177 atomic_set(&root->log_writers, 0);
1178 atomic_set(&root->log_batch, 0);
1179 refcount_set(&root->refs, 1);
1180 atomic_set(&root->will_be_snapshotted, 0);
1181 atomic_set(&root->snapshot_force_cow, 0);
1182 atomic_set(&root->nr_swapfiles, 0);
1183 root->log_transid = 0;
1184 root->log_transid_committed = -1;
1185 root->last_log_commit = 0;
1186 if (!dummy)
1187 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1188 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1189
1190 memset(&root->root_key, 0, sizeof(root->root_key));
1191 memset(&root->root_item, 0, sizeof(root->root_item));
1192 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1193 if (!dummy)
1194 root->defrag_trans_start = fs_info->generation;
1195 else
1196 root->defrag_trans_start = 0;
1197 root->root_key.objectid = objectid;
1198 root->anon_dev = 0;
1199
1200 spin_lock_init(&root->root_item_lock);
1201 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1202}
1203
1204static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1205 gfp_t flags)
1206{
1207 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1208 if (root)
1209 root->fs_info = fs_info;
1210 return root;
1211}
1212
1213#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1214/* Should only be used by the testing infrastructure */
1215struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1216{
1217 struct btrfs_root *root;
1218
1219 if (!fs_info)
1220 return ERR_PTR(-EINVAL);
1221
1222 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1223 if (!root)
1224 return ERR_PTR(-ENOMEM);
1225
1226 /* We don't use the stripesize in selftest, set it as sectorsize */
1227 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1228 root->alloc_bytenr = 0;
1229
1230 return root;
1231}
1232#endif
1233
1234struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1235 u64 objectid)
1236{
1237 struct btrfs_fs_info *fs_info = trans->fs_info;
1238 struct extent_buffer *leaf;
1239 struct btrfs_root *tree_root = fs_info->tree_root;
1240 struct btrfs_root *root;
1241 struct btrfs_key key;
1242 unsigned int nofs_flag;
1243 int ret = 0;
1244 uuid_le uuid = NULL_UUID_LE;
1245
1246 /*
1247 * We're holding a transaction handle, so use a NOFS memory allocation
1248 * context to avoid deadlock if reclaim happens.
1249 */
1250 nofs_flag = memalloc_nofs_save();
1251 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1252 memalloc_nofs_restore(nofs_flag);
1253 if (!root)
1254 return ERR_PTR(-ENOMEM);
1255
1256 __setup_root(root, fs_info, objectid);
1257 root->root_key.objectid = objectid;
1258 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1259 root->root_key.offset = 0;
1260
1261 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1262 if (IS_ERR(leaf)) {
1263 ret = PTR_ERR(leaf);
1264 leaf = NULL;
1265 goto fail;
1266 }
1267
1268 root->node = leaf;
1269 btrfs_mark_buffer_dirty(leaf);
1270
1271 root->commit_root = btrfs_root_node(root);
1272 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1273
1274 root->root_item.flags = 0;
1275 root->root_item.byte_limit = 0;
1276 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1277 btrfs_set_root_generation(&root->root_item, trans->transid);
1278 btrfs_set_root_level(&root->root_item, 0);
1279 btrfs_set_root_refs(&root->root_item, 1);
1280 btrfs_set_root_used(&root->root_item, leaf->len);
1281 btrfs_set_root_last_snapshot(&root->root_item, 0);
1282 btrfs_set_root_dirid(&root->root_item, 0);
1283 if (is_fstree(objectid))
1284 uuid_le_gen(&uuid);
1285 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1286 root->root_item.drop_level = 0;
1287
1288 key.objectid = objectid;
1289 key.type = BTRFS_ROOT_ITEM_KEY;
1290 key.offset = 0;
1291 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1292 if (ret)
1293 goto fail;
1294
1295 btrfs_tree_unlock(leaf);
1296
1297 return root;
1298
1299fail:
1300 if (leaf) {
1301 btrfs_tree_unlock(leaf);
1302 free_extent_buffer(root->commit_root);
1303 free_extent_buffer(leaf);
1304 }
1305 kfree(root);
1306
1307 return ERR_PTR(ret);
1308}
1309
1310static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1311 struct btrfs_fs_info *fs_info)
1312{
1313 struct btrfs_root *root;
1314 struct extent_buffer *leaf;
1315
1316 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1317 if (!root)
1318 return ERR_PTR(-ENOMEM);
1319
1320 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1321
1322 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1323 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1324 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1325
1326 /*
1327 * DON'T set REF_COWS for log trees
1328 *
1329 * log trees do not get reference counted because they go away
1330 * before a real commit is actually done. They do store pointers
1331 * to file data extents, and those reference counts still get
1332 * updated (along with back refs to the log tree).
1333 */
1334
1335 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1336 NULL, 0, 0, 0);
1337 if (IS_ERR(leaf)) {
1338 kfree(root);
1339 return ERR_CAST(leaf);
1340 }
1341
1342 root->node = leaf;
1343
1344 btrfs_mark_buffer_dirty(root->node);
1345 btrfs_tree_unlock(root->node);
1346 return root;
1347}
1348
1349int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1350 struct btrfs_fs_info *fs_info)
1351{
1352 struct btrfs_root *log_root;
1353
1354 log_root = alloc_log_tree(trans, fs_info);
1355 if (IS_ERR(log_root))
1356 return PTR_ERR(log_root);
1357 WARN_ON(fs_info->log_root_tree);
1358 fs_info->log_root_tree = log_root;
1359 return 0;
1360}
1361
1362int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root)
1364{
1365 struct btrfs_fs_info *fs_info = root->fs_info;
1366 struct btrfs_root *log_root;
1367 struct btrfs_inode_item *inode_item;
1368
1369 log_root = alloc_log_tree(trans, fs_info);
1370 if (IS_ERR(log_root))
1371 return PTR_ERR(log_root);
1372
1373 log_root->last_trans = trans->transid;
1374 log_root->root_key.offset = root->root_key.objectid;
1375
1376 inode_item = &log_root->root_item.inode;
1377 btrfs_set_stack_inode_generation(inode_item, 1);
1378 btrfs_set_stack_inode_size(inode_item, 3);
1379 btrfs_set_stack_inode_nlink(inode_item, 1);
1380 btrfs_set_stack_inode_nbytes(inode_item,
1381 fs_info->nodesize);
1382 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1383
1384 btrfs_set_root_node(&log_root->root_item, log_root->node);
1385
1386 WARN_ON(root->log_root);
1387 root->log_root = log_root;
1388 root->log_transid = 0;
1389 root->log_transid_committed = -1;
1390 root->last_log_commit = 0;
1391 return 0;
1392}
1393
1394static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1395 struct btrfs_key *key)
1396{
1397 struct btrfs_root *root;
1398 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1399 struct btrfs_path *path;
1400 u64 generation;
1401 int ret;
1402 int level;
1403
1404 path = btrfs_alloc_path();
1405 if (!path)
1406 return ERR_PTR(-ENOMEM);
1407
1408 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1409 if (!root) {
1410 ret = -ENOMEM;
1411 goto alloc_fail;
1412 }
1413
1414 __setup_root(root, fs_info, key->objectid);
1415
1416 ret = btrfs_find_root(tree_root, key, path,
1417 &root->root_item, &root->root_key);
1418 if (ret) {
1419 if (ret > 0)
1420 ret = -ENOENT;
1421 goto find_fail;
1422 }
1423
1424 generation = btrfs_root_generation(&root->root_item);
1425 level = btrfs_root_level(&root->root_item);
1426 root->node = read_tree_block(fs_info,
1427 btrfs_root_bytenr(&root->root_item),
1428 generation, level, NULL);
1429 if (IS_ERR(root->node)) {
1430 ret = PTR_ERR(root->node);
1431 goto find_fail;
1432 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1433 ret = -EIO;
1434 free_extent_buffer(root->node);
1435 goto find_fail;
1436 }
1437 root->commit_root = btrfs_root_node(root);
1438out:
1439 btrfs_free_path(path);
1440 return root;
1441
1442find_fail:
1443 kfree(root);
1444alloc_fail:
1445 root = ERR_PTR(ret);
1446 goto out;
1447}
1448
1449struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1450 struct btrfs_key *location)
1451{
1452 struct btrfs_root *root;
1453
1454 root = btrfs_read_tree_root(tree_root, location);
1455 if (IS_ERR(root))
1456 return root;
1457
1458 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1459 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1460 btrfs_check_and_init_root_item(&root->root_item);
1461 }
1462
1463 return root;
1464}
1465
1466int btrfs_init_fs_root(struct btrfs_root *root)
1467{
1468 int ret;
1469 struct btrfs_subvolume_writers *writers;
1470
1471 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1472 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1473 GFP_NOFS);
1474 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1475 ret = -ENOMEM;
1476 goto fail;
1477 }
1478
1479 writers = btrfs_alloc_subvolume_writers();
1480 if (IS_ERR(writers)) {
1481 ret = PTR_ERR(writers);
1482 goto fail;
1483 }
1484 root->subv_writers = writers;
1485
1486 btrfs_init_free_ino_ctl(root);
1487 spin_lock_init(&root->ino_cache_lock);
1488 init_waitqueue_head(&root->ino_cache_wait);
1489
1490 ret = get_anon_bdev(&root->anon_dev);
1491 if (ret)
1492 goto fail;
1493
1494 mutex_lock(&root->objectid_mutex);
1495 ret = btrfs_find_highest_objectid(root,
1496 &root->highest_objectid);
1497 if (ret) {
1498 mutex_unlock(&root->objectid_mutex);
1499 goto fail;
1500 }
1501
1502 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1503
1504 mutex_unlock(&root->objectid_mutex);
1505
1506 return 0;
1507fail:
1508 /* The caller is responsible to call btrfs_free_fs_root */
1509 return ret;
1510}
1511
1512struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1513 u64 root_id)
1514{
1515 struct btrfs_root *root;
1516
1517 spin_lock(&fs_info->fs_roots_radix_lock);
1518 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1519 (unsigned long)root_id);
1520 spin_unlock(&fs_info->fs_roots_radix_lock);
1521 return root;
1522}
1523
1524int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1525 struct btrfs_root *root)
1526{
1527 int ret;
1528
1529 ret = radix_tree_preload(GFP_NOFS);
1530 if (ret)
1531 return ret;
1532
1533 spin_lock(&fs_info->fs_roots_radix_lock);
1534 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1535 (unsigned long)root->root_key.objectid,
1536 root);
1537 if (ret == 0)
1538 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1539 spin_unlock(&fs_info->fs_roots_radix_lock);
1540 radix_tree_preload_end();
1541
1542 return ret;
1543}
1544
1545struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1546 struct btrfs_key *location,
1547 bool check_ref)
1548{
1549 struct btrfs_root *root;
1550 struct btrfs_path *path;
1551 struct btrfs_key key;
1552 int ret;
1553
1554 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1555 return fs_info->tree_root;
1556 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1557 return fs_info->extent_root;
1558 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1559 return fs_info->chunk_root;
1560 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1561 return fs_info->dev_root;
1562 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1563 return fs_info->csum_root;
1564 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1565 return fs_info->quota_root ? fs_info->quota_root :
1566 ERR_PTR(-ENOENT);
1567 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1568 return fs_info->uuid_root ? fs_info->uuid_root :
1569 ERR_PTR(-ENOENT);
1570 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1571 return fs_info->free_space_root ? fs_info->free_space_root :
1572 ERR_PTR(-ENOENT);
1573again:
1574 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1575 if (root) {
1576 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1577 return ERR_PTR(-ENOENT);
1578 return root;
1579 }
1580
1581 root = btrfs_read_fs_root(fs_info->tree_root, location);
1582 if (IS_ERR(root))
1583 return root;
1584
1585 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1586 ret = -ENOENT;
1587 goto fail;
1588 }
1589
1590 ret = btrfs_init_fs_root(root);
1591 if (ret)
1592 goto fail;
1593
1594 path = btrfs_alloc_path();
1595 if (!path) {
1596 ret = -ENOMEM;
1597 goto fail;
1598 }
1599 key.objectid = BTRFS_ORPHAN_OBJECTID;
1600 key.type = BTRFS_ORPHAN_ITEM_KEY;
1601 key.offset = location->objectid;
1602
1603 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1604 btrfs_free_path(path);
1605 if (ret < 0)
1606 goto fail;
1607 if (ret == 0)
1608 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1609
1610 ret = btrfs_insert_fs_root(fs_info, root);
1611 if (ret) {
1612 if (ret == -EEXIST) {
1613 btrfs_free_fs_root(root);
1614 goto again;
1615 }
1616 goto fail;
1617 }
1618 return root;
1619fail:
1620 btrfs_free_fs_root(root);
1621 return ERR_PTR(ret);
1622}
1623
1624static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1625{
1626 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1627 int ret = 0;
1628 struct btrfs_device *device;
1629 struct backing_dev_info *bdi;
1630
1631 rcu_read_lock();
1632 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1633 if (!device->bdev)
1634 continue;
1635 bdi = device->bdev->bd_bdi;
1636 if (bdi_congested(bdi, bdi_bits)) {
1637 ret = 1;
1638 break;
1639 }
1640 }
1641 rcu_read_unlock();
1642 return ret;
1643}
1644
1645/*
1646 * called by the kthread helper functions to finally call the bio end_io
1647 * functions. This is where read checksum verification actually happens
1648 */
1649static void end_workqueue_fn(struct btrfs_work *work)
1650{
1651 struct bio *bio;
1652 struct btrfs_end_io_wq *end_io_wq;
1653
1654 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1655 bio = end_io_wq->bio;
1656
1657 bio->bi_status = end_io_wq->status;
1658 bio->bi_private = end_io_wq->private;
1659 bio->bi_end_io = end_io_wq->end_io;
1660 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1661 bio_endio(bio);
1662}
1663
1664static int cleaner_kthread(void *arg)
1665{
1666 struct btrfs_root *root = arg;
1667 struct btrfs_fs_info *fs_info = root->fs_info;
1668 int again;
1669
1670 while (1) {
1671 again = 0;
1672
1673 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1674
1675 /* Make the cleaner go to sleep early. */
1676 if (btrfs_need_cleaner_sleep(fs_info))
1677 goto sleep;
1678
1679 /*
1680 * Do not do anything if we might cause open_ctree() to block
1681 * before we have finished mounting the filesystem.
1682 */
1683 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1684 goto sleep;
1685
1686 if (!mutex_trylock(&fs_info->cleaner_mutex))
1687 goto sleep;
1688
1689 /*
1690 * Avoid the problem that we change the status of the fs
1691 * during the above check and trylock.
1692 */
1693 if (btrfs_need_cleaner_sleep(fs_info)) {
1694 mutex_unlock(&fs_info->cleaner_mutex);
1695 goto sleep;
1696 }
1697
1698 btrfs_run_delayed_iputs(fs_info);
1699
1700 again = btrfs_clean_one_deleted_snapshot(root);
1701 mutex_unlock(&fs_info->cleaner_mutex);
1702
1703 /*
1704 * The defragger has dealt with the R/O remount and umount,
1705 * needn't do anything special here.
1706 */
1707 btrfs_run_defrag_inodes(fs_info);
1708
1709 /*
1710 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1711 * with relocation (btrfs_relocate_chunk) and relocation
1712 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1713 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1714 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1715 * unused block groups.
1716 */
1717 btrfs_delete_unused_bgs(fs_info);
1718sleep:
1719 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1720 if (kthread_should_park())
1721 kthread_parkme();
1722 if (kthread_should_stop())
1723 return 0;
1724 if (!again) {
1725 set_current_state(TASK_INTERRUPTIBLE);
1726 schedule();
1727 __set_current_state(TASK_RUNNING);
1728 }
1729 }
1730}
1731
1732static int transaction_kthread(void *arg)
1733{
1734 struct btrfs_root *root = arg;
1735 struct btrfs_fs_info *fs_info = root->fs_info;
1736 struct btrfs_trans_handle *trans;
1737 struct btrfs_transaction *cur;
1738 u64 transid;
1739 time64_t now;
1740 unsigned long delay;
1741 bool cannot_commit;
1742
1743 do {
1744 cannot_commit = false;
1745 delay = HZ * fs_info->commit_interval;
1746 mutex_lock(&fs_info->transaction_kthread_mutex);
1747
1748 spin_lock(&fs_info->trans_lock);
1749 cur = fs_info->running_transaction;
1750 if (!cur) {
1751 spin_unlock(&fs_info->trans_lock);
1752 goto sleep;
1753 }
1754
1755 now = ktime_get_seconds();
1756 if (cur->state < TRANS_STATE_BLOCKED &&
1757 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1758 (now < cur->start_time ||
1759 now - cur->start_time < fs_info->commit_interval)) {
1760 spin_unlock(&fs_info->trans_lock);
1761 delay = HZ * 5;
1762 goto sleep;
1763 }
1764 transid = cur->transid;
1765 spin_unlock(&fs_info->trans_lock);
1766
1767 /* If the file system is aborted, this will always fail. */
1768 trans = btrfs_attach_transaction(root);
1769 if (IS_ERR(trans)) {
1770 if (PTR_ERR(trans) != -ENOENT)
1771 cannot_commit = true;
1772 goto sleep;
1773 }
1774 if (transid == trans->transid) {
1775 btrfs_commit_transaction(trans);
1776 } else {
1777 btrfs_end_transaction(trans);
1778 }
1779sleep:
1780 wake_up_process(fs_info->cleaner_kthread);
1781 mutex_unlock(&fs_info->transaction_kthread_mutex);
1782
1783 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1784 &fs_info->fs_state)))
1785 btrfs_cleanup_transaction(fs_info);
1786 if (!kthread_should_stop() &&
1787 (!btrfs_transaction_blocked(fs_info) ||
1788 cannot_commit))
1789 schedule_timeout_interruptible(delay);
1790 } while (!kthread_should_stop());
1791 return 0;
1792}
1793
1794/*
1795 * this will find the highest generation in the array of
1796 * root backups. The index of the highest array is returned,
1797 * or -1 if we can't find anything.
1798 *
1799 * We check to make sure the array is valid by comparing the
1800 * generation of the latest root in the array with the generation
1801 * in the super block. If they don't match we pitch it.
1802 */
1803static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1804{
1805 u64 cur;
1806 int newest_index = -1;
1807 struct btrfs_root_backup *root_backup;
1808 int i;
1809
1810 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811 root_backup = info->super_copy->super_roots + i;
1812 cur = btrfs_backup_tree_root_gen(root_backup);
1813 if (cur == newest_gen)
1814 newest_index = i;
1815 }
1816
1817 /* check to see if we actually wrapped around */
1818 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819 root_backup = info->super_copy->super_roots;
1820 cur = btrfs_backup_tree_root_gen(root_backup);
1821 if (cur == newest_gen)
1822 newest_index = 0;
1823 }
1824 return newest_index;
1825}
1826
1827
1828/*
1829 * find the oldest backup so we know where to store new entries
1830 * in the backup array. This will set the backup_root_index
1831 * field in the fs_info struct
1832 */
1833static void find_oldest_super_backup(struct btrfs_fs_info *info,
1834 u64 newest_gen)
1835{
1836 int newest_index = -1;
1837
1838 newest_index = find_newest_super_backup(info, newest_gen);
1839 /* if there was garbage in there, just move along */
1840 if (newest_index == -1) {
1841 info->backup_root_index = 0;
1842 } else {
1843 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1844 }
1845}
1846
1847/*
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1850 * done
1851 */
1852static void backup_super_roots(struct btrfs_fs_info *info)
1853{
1854 int next_backup;
1855 struct btrfs_root_backup *root_backup;
1856 int last_backup;
1857
1858 next_backup = info->backup_root_index;
1859 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860 BTRFS_NUM_BACKUP_ROOTS;
1861
1862 /*
1863 * just overwrite the last backup if we're at the same generation
1864 * this happens only at umount
1865 */
1866 root_backup = info->super_for_commit->super_roots + last_backup;
1867 if (btrfs_backup_tree_root_gen(root_backup) ==
1868 btrfs_header_generation(info->tree_root->node))
1869 next_backup = last_backup;
1870
1871 root_backup = info->super_for_commit->super_roots + next_backup;
1872
1873 /*
1874 * make sure all of our padding and empty slots get zero filled
1875 * regardless of which ones we use today
1876 */
1877 memset(root_backup, 0, sizeof(*root_backup));
1878
1879 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1880
1881 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882 btrfs_set_backup_tree_root_gen(root_backup,
1883 btrfs_header_generation(info->tree_root->node));
1884
1885 btrfs_set_backup_tree_root_level(root_backup,
1886 btrfs_header_level(info->tree_root->node));
1887
1888 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889 btrfs_set_backup_chunk_root_gen(root_backup,
1890 btrfs_header_generation(info->chunk_root->node));
1891 btrfs_set_backup_chunk_root_level(root_backup,
1892 btrfs_header_level(info->chunk_root->node));
1893
1894 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895 btrfs_set_backup_extent_root_gen(root_backup,
1896 btrfs_header_generation(info->extent_root->node));
1897 btrfs_set_backup_extent_root_level(root_backup,
1898 btrfs_header_level(info->extent_root->node));
1899
1900 /*
1901 * we might commit during log recovery, which happens before we set
1902 * the fs_root. Make sure it is valid before we fill it in.
1903 */
1904 if (info->fs_root && info->fs_root->node) {
1905 btrfs_set_backup_fs_root(root_backup,
1906 info->fs_root->node->start);
1907 btrfs_set_backup_fs_root_gen(root_backup,
1908 btrfs_header_generation(info->fs_root->node));
1909 btrfs_set_backup_fs_root_level(root_backup,
1910 btrfs_header_level(info->fs_root->node));
1911 }
1912
1913 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914 btrfs_set_backup_dev_root_gen(root_backup,
1915 btrfs_header_generation(info->dev_root->node));
1916 btrfs_set_backup_dev_root_level(root_backup,
1917 btrfs_header_level(info->dev_root->node));
1918
1919 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920 btrfs_set_backup_csum_root_gen(root_backup,
1921 btrfs_header_generation(info->csum_root->node));
1922 btrfs_set_backup_csum_root_level(root_backup,
1923 btrfs_header_level(info->csum_root->node));
1924
1925 btrfs_set_backup_total_bytes(root_backup,
1926 btrfs_super_total_bytes(info->super_copy));
1927 btrfs_set_backup_bytes_used(root_backup,
1928 btrfs_super_bytes_used(info->super_copy));
1929 btrfs_set_backup_num_devices(root_backup,
1930 btrfs_super_num_devices(info->super_copy));
1931
1932 /*
1933 * if we don't copy this out to the super_copy, it won't get remembered
1934 * for the next commit
1935 */
1936 memcpy(&info->super_copy->super_roots,
1937 &info->super_for_commit->super_roots,
1938 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1939}
1940
1941/*
1942 * this copies info out of the root backup array and back into
1943 * the in-memory super block. It is meant to help iterate through
1944 * the array, so you send it the number of backups you've already
1945 * tried and the last backup index you used.
1946 *
1947 * this returns -1 when it has tried all the backups
1948 */
1949static noinline int next_root_backup(struct btrfs_fs_info *info,
1950 struct btrfs_super_block *super,
1951 int *num_backups_tried, int *backup_index)
1952{
1953 struct btrfs_root_backup *root_backup;
1954 int newest = *backup_index;
1955
1956 if (*num_backups_tried == 0) {
1957 u64 gen = btrfs_super_generation(super);
1958
1959 newest = find_newest_super_backup(info, gen);
1960 if (newest == -1)
1961 return -1;
1962
1963 *backup_index = newest;
1964 *num_backups_tried = 1;
1965 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966 /* we've tried all the backups, all done */
1967 return -1;
1968 } else {
1969 /* jump to the next oldest backup */
1970 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 BTRFS_NUM_BACKUP_ROOTS;
1972 *backup_index = newest;
1973 *num_backups_tried += 1;
1974 }
1975 root_backup = super->super_roots + newest;
1976
1977 btrfs_set_super_generation(super,
1978 btrfs_backup_tree_root_gen(root_backup));
1979 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980 btrfs_set_super_root_level(super,
1981 btrfs_backup_tree_root_level(root_backup));
1982 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1983
1984 /*
1985 * fixme: the total bytes and num_devices need to match or we should
1986 * need a fsck
1987 */
1988 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1990 return 0;
1991}
1992
1993/* helper to cleanup workers */
1994static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1995{
1996 btrfs_destroy_workqueue(fs_info->fixup_workers);
1997 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1998 btrfs_destroy_workqueue(fs_info->workers);
1999 btrfs_destroy_workqueue(fs_info->endio_workers);
2000 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2001 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2002 btrfs_destroy_workqueue(fs_info->rmw_workers);
2003 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2004 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2005 btrfs_destroy_workqueue(fs_info->submit_workers);
2006 btrfs_destroy_workqueue(fs_info->delayed_workers);
2007 btrfs_destroy_workqueue(fs_info->caching_workers);
2008 btrfs_destroy_workqueue(fs_info->readahead_workers);
2009 btrfs_destroy_workqueue(fs_info->flush_workers);
2010 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2011 /*
2012 * Now that all other work queues are destroyed, we can safely destroy
2013 * the queues used for metadata I/O, since tasks from those other work
2014 * queues can do metadata I/O operations.
2015 */
2016 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2017 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2018}
2019
2020static void free_root_extent_buffers(struct btrfs_root *root)
2021{
2022 if (root) {
2023 free_extent_buffer(root->node);
2024 free_extent_buffer(root->commit_root);
2025 root->node = NULL;
2026 root->commit_root = NULL;
2027 }
2028}
2029
2030/* helper to cleanup tree roots */
2031static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2032{
2033 free_root_extent_buffers(info->tree_root);
2034
2035 free_root_extent_buffers(info->dev_root);
2036 free_root_extent_buffers(info->extent_root);
2037 free_root_extent_buffers(info->csum_root);
2038 free_root_extent_buffers(info->quota_root);
2039 free_root_extent_buffers(info->uuid_root);
2040 if (chunk_root)
2041 free_root_extent_buffers(info->chunk_root);
2042 free_root_extent_buffers(info->free_space_root);
2043}
2044
2045void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2046{
2047 int ret;
2048 struct btrfs_root *gang[8];
2049 int i;
2050
2051 while (!list_empty(&fs_info->dead_roots)) {
2052 gang[0] = list_entry(fs_info->dead_roots.next,
2053 struct btrfs_root, root_list);
2054 list_del(&gang[0]->root_list);
2055
2056 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2057 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2058 } else {
2059 free_extent_buffer(gang[0]->node);
2060 free_extent_buffer(gang[0]->commit_root);
2061 btrfs_put_fs_root(gang[0]);
2062 }
2063 }
2064
2065 while (1) {
2066 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2067 (void **)gang, 0,
2068 ARRAY_SIZE(gang));
2069 if (!ret)
2070 break;
2071 for (i = 0; i < ret; i++)
2072 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2073 }
2074
2075 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2076 btrfs_free_log_root_tree(NULL, fs_info);
2077 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
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 = 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 /*
2171 * a higher idle thresh on the submit workers makes it much more
2172 * likely that bios will be send down in a sane order to the
2173 * devices
2174 */
2175 fs_info->submit_workers =
2176 btrfs_alloc_workqueue(fs_info, "submit", flags,
2177 min_t(u64, fs_devices->num_devices,
2178 max_active), 64);
2179
2180 fs_info->fixup_workers =
2181 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2182
2183 /*
2184 * endios are largely parallel and should have a very
2185 * low idle thresh
2186 */
2187 fs_info->endio_workers =
2188 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2189 fs_info->endio_meta_workers =
2190 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2191 max_active, 4);
2192 fs_info->endio_meta_write_workers =
2193 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2194 max_active, 2);
2195 fs_info->endio_raid56_workers =
2196 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2197 max_active, 4);
2198 fs_info->endio_repair_workers =
2199 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2200 fs_info->rmw_workers =
2201 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2202 fs_info->endio_write_workers =
2203 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2204 max_active, 2);
2205 fs_info->endio_freespace_worker =
2206 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2207 max_active, 0);
2208 fs_info->delayed_workers =
2209 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2210 max_active, 0);
2211 fs_info->readahead_workers =
2212 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2213 max_active, 2);
2214 fs_info->qgroup_rescan_workers =
2215 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2216
2217 if (!(fs_info->workers && fs_info->delalloc_workers &&
2218 fs_info->submit_workers && fs_info->flush_workers &&
2219 fs_info->endio_workers && fs_info->endio_meta_workers &&
2220 fs_info->endio_meta_write_workers &&
2221 fs_info->endio_repair_workers &&
2222 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2223 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2224 fs_info->caching_workers && fs_info->readahead_workers &&
2225 fs_info->fixup_workers && fs_info->delayed_workers &&
2226 fs_info->qgroup_rescan_workers)) {
2227 return -ENOMEM;
2228 }
2229
2230 return 0;
2231}
2232
2233static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2234{
2235 struct crypto_shash *csum_shash;
2236 const char *csum_name = btrfs_super_csum_name(csum_type);
2237
2238 csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2239
2240 if (IS_ERR(csum_shash)) {
2241 btrfs_err(fs_info, "error allocating %s hash for checksum",
2242 csum_name);
2243 return PTR_ERR(csum_shash);
2244 }
2245
2246 fs_info->csum_shash = csum_shash;
2247
2248 return 0;
2249}
2250
2251static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2252{
2253 crypto_free_shash(fs_info->csum_shash);
2254}
2255
2256static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2257 struct btrfs_fs_devices *fs_devices)
2258{
2259 int ret;
2260 struct btrfs_root *log_tree_root;
2261 struct btrfs_super_block *disk_super = fs_info->super_copy;
2262 u64 bytenr = btrfs_super_log_root(disk_super);
2263 int level = btrfs_super_log_root_level(disk_super);
2264
2265 if (fs_devices->rw_devices == 0) {
2266 btrfs_warn(fs_info, "log replay required on RO media");
2267 return -EIO;
2268 }
2269
2270 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2271 if (!log_tree_root)
2272 return -ENOMEM;
2273
2274 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2275
2276 log_tree_root->node = read_tree_block(fs_info, bytenr,
2277 fs_info->generation + 1,
2278 level, NULL);
2279 if (IS_ERR(log_tree_root->node)) {
2280 btrfs_warn(fs_info, "failed to read log tree");
2281 ret = PTR_ERR(log_tree_root->node);
2282 kfree(log_tree_root);
2283 return ret;
2284 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2285 btrfs_err(fs_info, "failed to read log tree");
2286 free_extent_buffer(log_tree_root->node);
2287 kfree(log_tree_root);
2288 return -EIO;
2289 }
2290 /* returns with log_tree_root freed on success */
2291 ret = btrfs_recover_log_trees(log_tree_root);
2292 if (ret) {
2293 btrfs_handle_fs_error(fs_info, ret,
2294 "Failed to recover log tree");
2295 free_extent_buffer(log_tree_root->node);
2296 kfree(log_tree_root);
2297 return ret;
2298 }
2299
2300 if (sb_rdonly(fs_info->sb)) {
2301 ret = btrfs_commit_super(fs_info);
2302 if (ret)
2303 return ret;
2304 }
2305
2306 return 0;
2307}
2308
2309static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2310{
2311 struct btrfs_root *tree_root = fs_info->tree_root;
2312 struct btrfs_root *root;
2313 struct btrfs_key location;
2314 int ret;
2315
2316 BUG_ON(!fs_info->tree_root);
2317
2318 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2319 location.type = BTRFS_ROOT_ITEM_KEY;
2320 location.offset = 0;
2321
2322 root = btrfs_read_tree_root(tree_root, &location);
2323 if (IS_ERR(root)) {
2324 ret = PTR_ERR(root);
2325 goto out;
2326 }
2327 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2328 fs_info->extent_root = root;
2329
2330 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2331 root = btrfs_read_tree_root(tree_root, &location);
2332 if (IS_ERR(root)) {
2333 ret = PTR_ERR(root);
2334 goto out;
2335 }
2336 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2337 fs_info->dev_root = root;
2338 btrfs_init_devices_late(fs_info);
2339
2340 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2341 root = btrfs_read_tree_root(tree_root, &location);
2342 if (IS_ERR(root)) {
2343 ret = PTR_ERR(root);
2344 goto out;
2345 }
2346 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2347 fs_info->csum_root = root;
2348
2349 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2350 root = btrfs_read_tree_root(tree_root, &location);
2351 if (!IS_ERR(root)) {
2352 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2353 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2354 fs_info->quota_root = root;
2355 }
2356
2357 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2358 root = btrfs_read_tree_root(tree_root, &location);
2359 if (IS_ERR(root)) {
2360 ret = PTR_ERR(root);
2361 if (ret != -ENOENT)
2362 goto out;
2363 } else {
2364 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2365 fs_info->uuid_root = root;
2366 }
2367
2368 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2369 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2370 root = btrfs_read_tree_root(tree_root, &location);
2371 if (IS_ERR(root)) {
2372 ret = PTR_ERR(root);
2373 goto out;
2374 }
2375 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2376 fs_info->free_space_root = root;
2377 }
2378
2379 return 0;
2380out:
2381 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2382 location.objectid, ret);
2383 return ret;
2384}
2385
2386/*
2387 * Real super block validation
2388 * NOTE: super csum type and incompat features will not be checked here.
2389 *
2390 * @sb: super block to check
2391 * @mirror_num: the super block number to check its bytenr:
2392 * 0 the primary (1st) sb
2393 * 1, 2 2nd and 3rd backup copy
2394 * -1 skip bytenr check
2395 */
2396static int validate_super(struct btrfs_fs_info *fs_info,
2397 struct btrfs_super_block *sb, int mirror_num)
2398{
2399 u64 nodesize = btrfs_super_nodesize(sb);
2400 u64 sectorsize = btrfs_super_sectorsize(sb);
2401 int ret = 0;
2402
2403 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2404 btrfs_err(fs_info, "no valid FS found");
2405 ret = -EINVAL;
2406 }
2407 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2408 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2409 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2410 ret = -EINVAL;
2411 }
2412 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2413 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2414 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2415 ret = -EINVAL;
2416 }
2417 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2418 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2419 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2420 ret = -EINVAL;
2421 }
2422 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2423 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2424 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2425 ret = -EINVAL;
2426 }
2427
2428 /*
2429 * Check sectorsize and nodesize first, other check will need it.
2430 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2431 */
2432 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2433 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2434 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2435 ret = -EINVAL;
2436 }
2437 /* Only PAGE SIZE is supported yet */
2438 if (sectorsize != PAGE_SIZE) {
2439 btrfs_err(fs_info,
2440 "sectorsize %llu not supported yet, only support %lu",
2441 sectorsize, PAGE_SIZE);
2442 ret = -EINVAL;
2443 }
2444 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2445 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2446 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2447 ret = -EINVAL;
2448 }
2449 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2450 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2451 le32_to_cpu(sb->__unused_leafsize), nodesize);
2452 ret = -EINVAL;
2453 }
2454
2455 /* Root alignment check */
2456 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2457 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2458 btrfs_super_root(sb));
2459 ret = -EINVAL;
2460 }
2461 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2462 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2463 btrfs_super_chunk_root(sb));
2464 ret = -EINVAL;
2465 }
2466 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2467 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2468 btrfs_super_log_root(sb));
2469 ret = -EINVAL;
2470 }
2471
2472 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2473 BTRFS_FSID_SIZE) != 0) {
2474 btrfs_err(fs_info,
2475 "dev_item UUID does not match metadata fsid: %pU != %pU",
2476 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2477 ret = -EINVAL;
2478 }
2479
2480 /*
2481 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2482 * done later
2483 */
2484 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2485 btrfs_err(fs_info, "bytes_used is too small %llu",
2486 btrfs_super_bytes_used(sb));
2487 ret = -EINVAL;
2488 }
2489 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2490 btrfs_err(fs_info, "invalid stripesize %u",
2491 btrfs_super_stripesize(sb));
2492 ret = -EINVAL;
2493 }
2494 if (btrfs_super_num_devices(sb) > (1UL << 31))
2495 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2496 btrfs_super_num_devices(sb));
2497 if (btrfs_super_num_devices(sb) == 0) {
2498 btrfs_err(fs_info, "number of devices is 0");
2499 ret = -EINVAL;
2500 }
2501
2502 if (mirror_num >= 0 &&
2503 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2504 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2505 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2506 ret = -EINVAL;
2507 }
2508
2509 /*
2510 * Obvious sys_chunk_array corruptions, it must hold at least one key
2511 * and one chunk
2512 */
2513 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2514 btrfs_err(fs_info, "system chunk array too big %u > %u",
2515 btrfs_super_sys_array_size(sb),
2516 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2517 ret = -EINVAL;
2518 }
2519 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2520 + sizeof(struct btrfs_chunk)) {
2521 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2522 btrfs_super_sys_array_size(sb),
2523 sizeof(struct btrfs_disk_key)
2524 + sizeof(struct btrfs_chunk));
2525 ret = -EINVAL;
2526 }
2527
2528 /*
2529 * The generation is a global counter, we'll trust it more than the others
2530 * but it's still possible that it's the one that's wrong.
2531 */
2532 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2533 btrfs_warn(fs_info,
2534 "suspicious: generation < chunk_root_generation: %llu < %llu",
2535 btrfs_super_generation(sb),
2536 btrfs_super_chunk_root_generation(sb));
2537 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2538 && btrfs_super_cache_generation(sb) != (u64)-1)
2539 btrfs_warn(fs_info,
2540 "suspicious: generation < cache_generation: %llu < %llu",
2541 btrfs_super_generation(sb),
2542 btrfs_super_cache_generation(sb));
2543
2544 return ret;
2545}
2546
2547/*
2548 * Validation of super block at mount time.
2549 * Some checks already done early at mount time, like csum type and incompat
2550 * flags will be skipped.
2551 */
2552static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2553{
2554 return validate_super(fs_info, fs_info->super_copy, 0);
2555}
2556
2557/*
2558 * Validation of super block at write time.
2559 * Some checks like bytenr check will be skipped as their values will be
2560 * overwritten soon.
2561 * Extra checks like csum type and incompat flags will be done here.
2562 */
2563static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2564 struct btrfs_super_block *sb)
2565{
2566 int ret;
2567
2568 ret = validate_super(fs_info, sb, -1);
2569 if (ret < 0)
2570 goto out;
2571 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2572 ret = -EUCLEAN;
2573 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2574 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2575 goto out;
2576 }
2577 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2578 ret = -EUCLEAN;
2579 btrfs_err(fs_info,
2580 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2581 btrfs_super_incompat_flags(sb),
2582 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2583 goto out;
2584 }
2585out:
2586 if (ret < 0)
2587 btrfs_err(fs_info,
2588 "super block corruption detected before writing it to disk");
2589 return ret;
2590}
2591
2592int open_ctree(struct super_block *sb,
2593 struct btrfs_fs_devices *fs_devices,
2594 char *options)
2595{
2596 u32 sectorsize;
2597 u32 nodesize;
2598 u32 stripesize;
2599 u64 generation;
2600 u64 features;
2601 u16 csum_type;
2602 struct btrfs_key location;
2603 struct buffer_head *bh;
2604 struct btrfs_super_block *disk_super;
2605 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2606 struct btrfs_root *tree_root;
2607 struct btrfs_root *chunk_root;
2608 int ret;
2609 int err = -EINVAL;
2610 int num_backups_tried = 0;
2611 int backup_index = 0;
2612 int clear_free_space_tree = 0;
2613 int level;
2614
2615 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2616 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2617 if (!tree_root || !chunk_root) {
2618 err = -ENOMEM;
2619 goto fail;
2620 }
2621
2622 ret = init_srcu_struct(&fs_info->subvol_srcu);
2623 if (ret) {
2624 err = ret;
2625 goto fail;
2626 }
2627
2628 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2629 if (ret) {
2630 err = ret;
2631 goto fail_srcu;
2632 }
2633
2634 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2635 if (ret) {
2636 err = ret;
2637 goto fail_dio_bytes;
2638 }
2639 fs_info->dirty_metadata_batch = PAGE_SIZE *
2640 (1 + ilog2(nr_cpu_ids));
2641
2642 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2643 if (ret) {
2644 err = ret;
2645 goto fail_dirty_metadata_bytes;
2646 }
2647
2648 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2649 GFP_KERNEL);
2650 if (ret) {
2651 err = ret;
2652 goto fail_delalloc_bytes;
2653 }
2654
2655 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2656 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2657 INIT_LIST_HEAD(&fs_info->trans_list);
2658 INIT_LIST_HEAD(&fs_info->dead_roots);
2659 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2660 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2661 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2662 spin_lock_init(&fs_info->delalloc_root_lock);
2663 spin_lock_init(&fs_info->trans_lock);
2664 spin_lock_init(&fs_info->fs_roots_radix_lock);
2665 spin_lock_init(&fs_info->delayed_iput_lock);
2666 spin_lock_init(&fs_info->defrag_inodes_lock);
2667 spin_lock_init(&fs_info->tree_mod_seq_lock);
2668 spin_lock_init(&fs_info->super_lock);
2669 spin_lock_init(&fs_info->buffer_lock);
2670 spin_lock_init(&fs_info->unused_bgs_lock);
2671 rwlock_init(&fs_info->tree_mod_log_lock);
2672 mutex_init(&fs_info->unused_bg_unpin_mutex);
2673 mutex_init(&fs_info->delete_unused_bgs_mutex);
2674 mutex_init(&fs_info->reloc_mutex);
2675 mutex_init(&fs_info->delalloc_root_mutex);
2676 seqlock_init(&fs_info->profiles_lock);
2677
2678 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2679 INIT_LIST_HEAD(&fs_info->space_info);
2680 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2681 INIT_LIST_HEAD(&fs_info->unused_bgs);
2682 extent_map_tree_init(&fs_info->mapping_tree);
2683 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2684 BTRFS_BLOCK_RSV_GLOBAL);
2685 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2686 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2687 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2688 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2689 BTRFS_BLOCK_RSV_DELOPS);
2690 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2691 BTRFS_BLOCK_RSV_DELREFS);
2692
2693 atomic_set(&fs_info->async_delalloc_pages, 0);
2694 atomic_set(&fs_info->defrag_running, 0);
2695 atomic_set(&fs_info->reada_works_cnt, 0);
2696 atomic_set(&fs_info->nr_delayed_iputs, 0);
2697 atomic64_set(&fs_info->tree_mod_seq, 0);
2698 fs_info->sb = sb;
2699 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2700 fs_info->metadata_ratio = 0;
2701 fs_info->defrag_inodes = RB_ROOT;
2702 atomic64_set(&fs_info->free_chunk_space, 0);
2703 fs_info->tree_mod_log = RB_ROOT;
2704 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2705 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2706 /* readahead state */
2707 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2708 spin_lock_init(&fs_info->reada_lock);
2709 btrfs_init_ref_verify(fs_info);
2710
2711 fs_info->thread_pool_size = min_t(unsigned long,
2712 num_online_cpus() + 2, 8);
2713
2714 INIT_LIST_HEAD(&fs_info->ordered_roots);
2715 spin_lock_init(&fs_info->ordered_root_lock);
2716
2717 fs_info->btree_inode = new_inode(sb);
2718 if (!fs_info->btree_inode) {
2719 err = -ENOMEM;
2720 goto fail_bio_counter;
2721 }
2722 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2723
2724 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2725 GFP_KERNEL);
2726 if (!fs_info->delayed_root) {
2727 err = -ENOMEM;
2728 goto fail_iput;
2729 }
2730 btrfs_init_delayed_root(fs_info->delayed_root);
2731
2732 btrfs_init_scrub(fs_info);
2733#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2734 fs_info->check_integrity_print_mask = 0;
2735#endif
2736 btrfs_init_balance(fs_info);
2737 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2738
2739 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2740 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2741
2742 btrfs_init_btree_inode(fs_info);
2743
2744 spin_lock_init(&fs_info->block_group_cache_lock);
2745 fs_info->block_group_cache_tree = RB_ROOT;
2746 fs_info->first_logical_byte = (u64)-1;
2747
2748 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2749 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2750 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2751 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2752 fs_info->pinned_extents = &fs_info->freed_extents[0];
2753 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2754
2755 mutex_init(&fs_info->ordered_operations_mutex);
2756 mutex_init(&fs_info->tree_log_mutex);
2757 mutex_init(&fs_info->chunk_mutex);
2758 mutex_init(&fs_info->transaction_kthread_mutex);
2759 mutex_init(&fs_info->cleaner_mutex);
2760 mutex_init(&fs_info->ro_block_group_mutex);
2761 init_rwsem(&fs_info->commit_root_sem);
2762 init_rwsem(&fs_info->cleanup_work_sem);
2763 init_rwsem(&fs_info->subvol_sem);
2764 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2765
2766 btrfs_init_dev_replace_locks(fs_info);
2767 btrfs_init_qgroup(fs_info);
2768
2769 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2770 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2771
2772 init_waitqueue_head(&fs_info->transaction_throttle);
2773 init_waitqueue_head(&fs_info->transaction_wait);
2774 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2775 init_waitqueue_head(&fs_info->async_submit_wait);
2776 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2777
2778 /* Usable values until the real ones are cached from the superblock */
2779 fs_info->nodesize = 4096;
2780 fs_info->sectorsize = 4096;
2781 fs_info->stripesize = 4096;
2782
2783 spin_lock_init(&fs_info->swapfile_pins_lock);
2784 fs_info->swapfile_pins = RB_ROOT;
2785
2786 fs_info->send_in_progress = 0;
2787
2788 ret = btrfs_alloc_stripe_hash_table(fs_info);
2789 if (ret) {
2790 err = ret;
2791 goto fail_alloc;
2792 }
2793
2794 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2795
2796 invalidate_bdev(fs_devices->latest_bdev);
2797
2798 /*
2799 * Read super block and check the signature bytes only
2800 */
2801 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2802 if (IS_ERR(bh)) {
2803 err = PTR_ERR(bh);
2804 goto fail_alloc;
2805 }
2806
2807 /*
2808 * Verify the type first, if that or the the checksum value are
2809 * corrupted, we'll find out
2810 */
2811 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2812 if (!btrfs_supported_super_csum(csum_type)) {
2813 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2814 csum_type);
2815 err = -EINVAL;
2816 brelse(bh);
2817 goto fail_alloc;
2818 }
2819
2820 ret = btrfs_init_csum_hash(fs_info, csum_type);
2821 if (ret) {
2822 err = ret;
2823 goto fail_alloc;
2824 }
2825
2826 /*
2827 * We want to check superblock checksum, the type is stored inside.
2828 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2829 */
2830 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2831 btrfs_err(fs_info, "superblock checksum mismatch");
2832 err = -EINVAL;
2833 brelse(bh);
2834 goto fail_csum;
2835 }
2836
2837 /*
2838 * super_copy is zeroed at allocation time and we never touch the
2839 * following bytes up to INFO_SIZE, the checksum is calculated from
2840 * the whole block of INFO_SIZE
2841 */
2842 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2843 brelse(bh);
2844
2845 disk_super = fs_info->super_copy;
2846
2847 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2848 BTRFS_FSID_SIZE));
2849
2850 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2851 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2852 fs_info->super_copy->metadata_uuid,
2853 BTRFS_FSID_SIZE));
2854 }
2855
2856 features = btrfs_super_flags(disk_super);
2857 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2858 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2859 btrfs_set_super_flags(disk_super, features);
2860 btrfs_info(fs_info,
2861 "found metadata UUID change in progress flag, clearing");
2862 }
2863
2864 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2865 sizeof(*fs_info->super_for_commit));
2866
2867 ret = btrfs_validate_mount_super(fs_info);
2868 if (ret) {
2869 btrfs_err(fs_info, "superblock contains fatal errors");
2870 err = -EINVAL;
2871 goto fail_csum;
2872 }
2873
2874 if (!btrfs_super_root(disk_super))
2875 goto fail_csum;
2876
2877 /* check FS state, whether FS is broken. */
2878 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2879 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2880
2881 /*
2882 * run through our array of backup supers and setup
2883 * our ring pointer to the oldest one
2884 */
2885 generation = btrfs_super_generation(disk_super);
2886 find_oldest_super_backup(fs_info, generation);
2887
2888 /*
2889 * In the long term, we'll store the compression type in the super
2890 * block, and it'll be used for per file compression control.
2891 */
2892 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2893
2894 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2895 if (ret) {
2896 err = ret;
2897 goto fail_csum;
2898 }
2899
2900 features = btrfs_super_incompat_flags(disk_super) &
2901 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2902 if (features) {
2903 btrfs_err(fs_info,
2904 "cannot mount because of unsupported optional features (%llx)",
2905 features);
2906 err = -EINVAL;
2907 goto fail_csum;
2908 }
2909
2910 features = btrfs_super_incompat_flags(disk_super);
2911 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2912 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2913 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2914 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2915 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2916
2917 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2918 btrfs_info(fs_info, "has skinny extents");
2919
2920 /*
2921 * flag our filesystem as having big metadata blocks if
2922 * they are bigger than the page size
2923 */
2924 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2925 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2926 btrfs_info(fs_info,
2927 "flagging fs with big metadata feature");
2928 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2929 }
2930
2931 nodesize = btrfs_super_nodesize(disk_super);
2932 sectorsize = btrfs_super_sectorsize(disk_super);
2933 stripesize = sectorsize;
2934 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2935 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2936
2937 /* Cache block sizes */
2938 fs_info->nodesize = nodesize;
2939 fs_info->sectorsize = sectorsize;
2940 fs_info->stripesize = stripesize;
2941
2942 /*
2943 * mixed block groups end up with duplicate but slightly offset
2944 * extent buffers for the same range. It leads to corruptions
2945 */
2946 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2947 (sectorsize != nodesize)) {
2948 btrfs_err(fs_info,
2949"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2950 nodesize, sectorsize);
2951 goto fail_csum;
2952 }
2953
2954 /*
2955 * Needn't use the lock because there is no other task which will
2956 * update the flag.
2957 */
2958 btrfs_set_super_incompat_flags(disk_super, features);
2959
2960 features = btrfs_super_compat_ro_flags(disk_super) &
2961 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2962 if (!sb_rdonly(sb) && features) {
2963 btrfs_err(fs_info,
2964 "cannot mount read-write because of unsupported optional features (%llx)",
2965 features);
2966 err = -EINVAL;
2967 goto fail_csum;
2968 }
2969
2970 ret = btrfs_init_workqueues(fs_info, fs_devices);
2971 if (ret) {
2972 err = ret;
2973 goto fail_sb_buffer;
2974 }
2975
2976 sb->s_bdi->congested_fn = btrfs_congested_fn;
2977 sb->s_bdi->congested_data = fs_info;
2978 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2979 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2980 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2981 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2982
2983 sb->s_blocksize = sectorsize;
2984 sb->s_blocksize_bits = blksize_bits(sectorsize);
2985 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2986
2987 mutex_lock(&fs_info->chunk_mutex);
2988 ret = btrfs_read_sys_array(fs_info);
2989 mutex_unlock(&fs_info->chunk_mutex);
2990 if (ret) {
2991 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2992 goto fail_sb_buffer;
2993 }
2994
2995 generation = btrfs_super_chunk_root_generation(disk_super);
2996 level = btrfs_super_chunk_root_level(disk_super);
2997
2998 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2999
3000 chunk_root->node = read_tree_block(fs_info,
3001 btrfs_super_chunk_root(disk_super),
3002 generation, level, NULL);
3003 if (IS_ERR(chunk_root->node) ||
3004 !extent_buffer_uptodate(chunk_root->node)) {
3005 btrfs_err(fs_info, "failed to read chunk root");
3006 if (!IS_ERR(chunk_root->node))
3007 free_extent_buffer(chunk_root->node);
3008 chunk_root->node = NULL;
3009 goto fail_tree_roots;
3010 }
3011 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3012 chunk_root->commit_root = btrfs_root_node(chunk_root);
3013
3014 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3015 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3016
3017 ret = btrfs_read_chunk_tree(fs_info);
3018 if (ret) {
3019 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3020 goto fail_tree_roots;
3021 }
3022
3023 /*
3024 * Keep the devid that is marked to be the target device for the
3025 * device replace procedure
3026 */
3027 btrfs_free_extra_devids(fs_devices, 0);
3028
3029 if (!fs_devices->latest_bdev) {
3030 btrfs_err(fs_info, "failed to read devices");
3031 goto fail_tree_roots;
3032 }
3033
3034retry_root_backup:
3035 generation = btrfs_super_generation(disk_super);
3036 level = btrfs_super_root_level(disk_super);
3037
3038 tree_root->node = read_tree_block(fs_info,
3039 btrfs_super_root(disk_super),
3040 generation, level, NULL);
3041 if (IS_ERR(tree_root->node) ||
3042 !extent_buffer_uptodate(tree_root->node)) {
3043 btrfs_warn(fs_info, "failed to read tree root");
3044 if (!IS_ERR(tree_root->node))
3045 free_extent_buffer(tree_root->node);
3046 tree_root->node = NULL;
3047 goto recovery_tree_root;
3048 }
3049
3050 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3051 tree_root->commit_root = btrfs_root_node(tree_root);
3052 btrfs_set_root_refs(&tree_root->root_item, 1);
3053
3054 mutex_lock(&tree_root->objectid_mutex);
3055 ret = btrfs_find_highest_objectid(tree_root,
3056 &tree_root->highest_objectid);
3057 if (ret) {
3058 mutex_unlock(&tree_root->objectid_mutex);
3059 goto recovery_tree_root;
3060 }
3061
3062 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3063
3064 mutex_unlock(&tree_root->objectid_mutex);
3065
3066 ret = btrfs_read_roots(fs_info);
3067 if (ret)
3068 goto recovery_tree_root;
3069
3070 fs_info->generation = generation;
3071 fs_info->last_trans_committed = generation;
3072
3073 ret = btrfs_verify_dev_extents(fs_info);
3074 if (ret) {
3075 btrfs_err(fs_info,
3076 "failed to verify dev extents against chunks: %d",
3077 ret);
3078 goto fail_block_groups;
3079 }
3080 ret = btrfs_recover_balance(fs_info);
3081 if (ret) {
3082 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3083 goto fail_block_groups;
3084 }
3085
3086 ret = btrfs_init_dev_stats(fs_info);
3087 if (ret) {
3088 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3089 goto fail_block_groups;
3090 }
3091
3092 ret = btrfs_init_dev_replace(fs_info);
3093 if (ret) {
3094 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3095 goto fail_block_groups;
3096 }
3097
3098 btrfs_free_extra_devids(fs_devices, 1);
3099
3100 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3101 if (ret) {
3102 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3103 ret);
3104 goto fail_block_groups;
3105 }
3106
3107 ret = btrfs_sysfs_add_device(fs_devices);
3108 if (ret) {
3109 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3110 ret);
3111 goto fail_fsdev_sysfs;
3112 }
3113
3114 ret = btrfs_sysfs_add_mounted(fs_info);
3115 if (ret) {
3116 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3117 goto fail_fsdev_sysfs;
3118 }
3119
3120 ret = btrfs_init_space_info(fs_info);
3121 if (ret) {
3122 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3123 goto fail_sysfs;
3124 }
3125
3126 ret = btrfs_read_block_groups(fs_info);
3127 if (ret) {
3128 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3129 goto fail_sysfs;
3130 }
3131
3132 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3133 btrfs_warn(fs_info,
3134 "writable mount is not allowed due to too many missing devices");
3135 goto fail_sysfs;
3136 }
3137
3138 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3139 "btrfs-cleaner");
3140 if (IS_ERR(fs_info->cleaner_kthread))
3141 goto fail_sysfs;
3142
3143 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3144 tree_root,
3145 "btrfs-transaction");
3146 if (IS_ERR(fs_info->transaction_kthread))
3147 goto fail_cleaner;
3148
3149 if (!btrfs_test_opt(fs_info, NOSSD) &&
3150 !fs_info->fs_devices->rotating) {
3151 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3152 }
3153
3154 /*
3155 * Mount does not set all options immediately, we can do it now and do
3156 * not have to wait for transaction commit
3157 */
3158 btrfs_apply_pending_changes(fs_info);
3159
3160#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3161 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3162 ret = btrfsic_mount(fs_info, fs_devices,
3163 btrfs_test_opt(fs_info,
3164 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3165 1 : 0,
3166 fs_info->check_integrity_print_mask);
3167 if (ret)
3168 btrfs_warn(fs_info,
3169 "failed to initialize integrity check module: %d",
3170 ret);
3171 }
3172#endif
3173 ret = btrfs_read_qgroup_config(fs_info);
3174 if (ret)
3175 goto fail_trans_kthread;
3176
3177 if (btrfs_build_ref_tree(fs_info))
3178 btrfs_err(fs_info, "couldn't build ref tree");
3179
3180 /* do not make disk changes in broken FS or nologreplay is given */
3181 if (btrfs_super_log_root(disk_super) != 0 &&
3182 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3183 ret = btrfs_replay_log(fs_info, fs_devices);
3184 if (ret) {
3185 err = ret;
3186 goto fail_qgroup;
3187 }
3188 }
3189
3190 ret = btrfs_find_orphan_roots(fs_info);
3191 if (ret)
3192 goto fail_qgroup;
3193
3194 if (!sb_rdonly(sb)) {
3195 ret = btrfs_cleanup_fs_roots(fs_info);
3196 if (ret)
3197 goto fail_qgroup;
3198
3199 mutex_lock(&fs_info->cleaner_mutex);
3200 ret = btrfs_recover_relocation(tree_root);
3201 mutex_unlock(&fs_info->cleaner_mutex);
3202 if (ret < 0) {
3203 btrfs_warn(fs_info, "failed to recover relocation: %d",
3204 ret);
3205 err = -EINVAL;
3206 goto fail_qgroup;
3207 }
3208 }
3209
3210 location.objectid = BTRFS_FS_TREE_OBJECTID;
3211 location.type = BTRFS_ROOT_ITEM_KEY;
3212 location.offset = 0;
3213
3214 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3215 if (IS_ERR(fs_info->fs_root)) {
3216 err = PTR_ERR(fs_info->fs_root);
3217 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3218 goto fail_qgroup;
3219 }
3220
3221 if (sb_rdonly(sb))
3222 return 0;
3223
3224 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3225 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3226 clear_free_space_tree = 1;
3227 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3228 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3229 btrfs_warn(fs_info, "free space tree is invalid");
3230 clear_free_space_tree = 1;
3231 }
3232
3233 if (clear_free_space_tree) {
3234 btrfs_info(fs_info, "clearing free space tree");
3235 ret = btrfs_clear_free_space_tree(fs_info);
3236 if (ret) {
3237 btrfs_warn(fs_info,
3238 "failed to clear free space tree: %d", ret);
3239 close_ctree(fs_info);
3240 return ret;
3241 }
3242 }
3243
3244 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3245 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3246 btrfs_info(fs_info, "creating free space tree");
3247 ret = btrfs_create_free_space_tree(fs_info);
3248 if (ret) {
3249 btrfs_warn(fs_info,
3250 "failed to create free space tree: %d", ret);
3251 close_ctree(fs_info);
3252 return ret;
3253 }
3254 }
3255
3256 down_read(&fs_info->cleanup_work_sem);
3257 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3258 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3259 up_read(&fs_info->cleanup_work_sem);
3260 close_ctree(fs_info);
3261 return ret;
3262 }
3263 up_read(&fs_info->cleanup_work_sem);
3264
3265 ret = btrfs_resume_balance_async(fs_info);
3266 if (ret) {
3267 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3268 close_ctree(fs_info);
3269 return ret;
3270 }
3271
3272 ret = btrfs_resume_dev_replace_async(fs_info);
3273 if (ret) {
3274 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3275 close_ctree(fs_info);
3276 return ret;
3277 }
3278
3279 btrfs_qgroup_rescan_resume(fs_info);
3280
3281 if (!fs_info->uuid_root) {
3282 btrfs_info(fs_info, "creating UUID tree");
3283 ret = btrfs_create_uuid_tree(fs_info);
3284 if (ret) {
3285 btrfs_warn(fs_info,
3286 "failed to create the UUID tree: %d", ret);
3287 close_ctree(fs_info);
3288 return ret;
3289 }
3290 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3291 fs_info->generation !=
3292 btrfs_super_uuid_tree_generation(disk_super)) {
3293 btrfs_info(fs_info, "checking UUID tree");
3294 ret = btrfs_check_uuid_tree(fs_info);
3295 if (ret) {
3296 btrfs_warn(fs_info,
3297 "failed to check the UUID tree: %d", ret);
3298 close_ctree(fs_info);
3299 return ret;
3300 }
3301 } else {
3302 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3303 }
3304 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3305
3306 /*
3307 * backuproot only affect mount behavior, and if open_ctree succeeded,
3308 * no need to keep the flag
3309 */
3310 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3311
3312 return 0;
3313
3314fail_qgroup:
3315 btrfs_free_qgroup_config(fs_info);
3316fail_trans_kthread:
3317 kthread_stop(fs_info->transaction_kthread);
3318 btrfs_cleanup_transaction(fs_info);
3319 btrfs_free_fs_roots(fs_info);
3320fail_cleaner:
3321 kthread_stop(fs_info->cleaner_kthread);
3322
3323 /*
3324 * make sure we're done with the btree inode before we stop our
3325 * kthreads
3326 */
3327 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3328
3329fail_sysfs:
3330 btrfs_sysfs_remove_mounted(fs_info);
3331
3332fail_fsdev_sysfs:
3333 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3334
3335fail_block_groups:
3336 btrfs_put_block_group_cache(fs_info);
3337
3338fail_tree_roots:
3339 free_root_pointers(fs_info, 1);
3340 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3341
3342fail_sb_buffer:
3343 btrfs_stop_all_workers(fs_info);
3344 btrfs_free_block_groups(fs_info);
3345fail_csum:
3346 btrfs_free_csum_hash(fs_info);
3347fail_alloc:
3348fail_iput:
3349 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3350
3351 iput(fs_info->btree_inode);
3352fail_bio_counter:
3353 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3354fail_delalloc_bytes:
3355 percpu_counter_destroy(&fs_info->delalloc_bytes);
3356fail_dirty_metadata_bytes:
3357 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3358fail_dio_bytes:
3359 percpu_counter_destroy(&fs_info->dio_bytes);
3360fail_srcu:
3361 cleanup_srcu_struct(&fs_info->subvol_srcu);
3362fail:
3363 btrfs_free_stripe_hash_table(fs_info);
3364 btrfs_close_devices(fs_info->fs_devices);
3365 return err;
3366
3367recovery_tree_root:
3368 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3369 goto fail_tree_roots;
3370
3371 free_root_pointers(fs_info, 0);
3372
3373 /* don't use the log in recovery mode, it won't be valid */
3374 btrfs_set_super_log_root(disk_super, 0);
3375
3376 /* we can't trust the free space cache either */
3377 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3378
3379 ret = next_root_backup(fs_info, fs_info->super_copy,
3380 &num_backups_tried, &backup_index);
3381 if (ret == -1)
3382 goto fail_block_groups;
3383 goto retry_root_backup;
3384}
3385ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3386
3387static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3388{
3389 if (uptodate) {
3390 set_buffer_uptodate(bh);
3391 } else {
3392 struct btrfs_device *device = (struct btrfs_device *)
3393 bh->b_private;
3394
3395 btrfs_warn_rl_in_rcu(device->fs_info,
3396 "lost page write due to IO error on %s",
3397 rcu_str_deref(device->name));
3398 /* note, we don't set_buffer_write_io_error because we have
3399 * our own ways of dealing with the IO errors
3400 */
3401 clear_buffer_uptodate(bh);
3402 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3403 }
3404 unlock_buffer(bh);
3405 put_bh(bh);
3406}
3407
3408int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3409 struct buffer_head **bh_ret)
3410{
3411 struct buffer_head *bh;
3412 struct btrfs_super_block *super;
3413 u64 bytenr;
3414
3415 bytenr = btrfs_sb_offset(copy_num);
3416 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3417 return -EINVAL;
3418
3419 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3420 /*
3421 * If we fail to read from the underlying devices, as of now
3422 * the best option we have is to mark it EIO.
3423 */
3424 if (!bh)
3425 return -EIO;
3426
3427 super = (struct btrfs_super_block *)bh->b_data;
3428 if (btrfs_super_bytenr(super) != bytenr ||
3429 btrfs_super_magic(super) != BTRFS_MAGIC) {
3430 brelse(bh);
3431 return -EINVAL;
3432 }
3433
3434 *bh_ret = bh;
3435 return 0;
3436}
3437
3438
3439struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3440{
3441 struct buffer_head *bh;
3442 struct buffer_head *latest = NULL;
3443 struct btrfs_super_block *super;
3444 int i;
3445 u64 transid = 0;
3446 int ret = -EINVAL;
3447
3448 /* we would like to check all the supers, but that would make
3449 * a btrfs mount succeed after a mkfs from a different FS.
3450 * So, we need to add a special mount option to scan for
3451 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3452 */
3453 for (i = 0; i < 1; i++) {
3454 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3455 if (ret)
3456 continue;
3457
3458 super = (struct btrfs_super_block *)bh->b_data;
3459
3460 if (!latest || btrfs_super_generation(super) > transid) {
3461 brelse(latest);
3462 latest = bh;
3463 transid = btrfs_super_generation(super);
3464 } else {
3465 brelse(bh);
3466 }
3467 }
3468
3469 if (!latest)
3470 return ERR_PTR(ret);
3471
3472 return latest;
3473}
3474
3475/*
3476 * Write superblock @sb to the @device. Do not wait for completion, all the
3477 * buffer heads we write are pinned.
3478 *
3479 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3480 * the expected device size at commit time. Note that max_mirrors must be
3481 * same for write and wait phases.
3482 *
3483 * Return number of errors when buffer head is not found or submission fails.
3484 */
3485static int write_dev_supers(struct btrfs_device *device,
3486 struct btrfs_super_block *sb, int max_mirrors)
3487{
3488 struct btrfs_fs_info *fs_info = device->fs_info;
3489 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3490 struct buffer_head *bh;
3491 int i;
3492 int ret;
3493 int errors = 0;
3494 u64 bytenr;
3495 int op_flags;
3496
3497 if (max_mirrors == 0)
3498 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3499
3500 shash->tfm = fs_info->csum_shash;
3501
3502 for (i = 0; i < max_mirrors; i++) {
3503 bytenr = btrfs_sb_offset(i);
3504 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3505 device->commit_total_bytes)
3506 break;
3507
3508 btrfs_set_super_bytenr(sb, bytenr);
3509
3510 crypto_shash_init(shash);
3511 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3512 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3513 crypto_shash_final(shash, sb->csum);
3514
3515 /* One reference for us, and we leave it for the caller */
3516 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3517 BTRFS_SUPER_INFO_SIZE);
3518 if (!bh) {
3519 btrfs_err(device->fs_info,
3520 "couldn't get super buffer head for bytenr %llu",
3521 bytenr);
3522 errors++;
3523 continue;
3524 }
3525
3526 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3527
3528 /* one reference for submit_bh */
3529 get_bh(bh);
3530
3531 set_buffer_uptodate(bh);
3532 lock_buffer(bh);
3533 bh->b_end_io = btrfs_end_buffer_write_sync;
3534 bh->b_private = device;
3535
3536 /*
3537 * we fua the first super. The others we allow
3538 * to go down lazy.
3539 */
3540 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3541 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3542 op_flags |= REQ_FUA;
3543 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3544 if (ret)
3545 errors++;
3546 }
3547 return errors < i ? 0 : -1;
3548}
3549
3550/*
3551 * Wait for write completion of superblocks done by write_dev_supers,
3552 * @max_mirrors same for write and wait phases.
3553 *
3554 * Return number of errors when buffer head is not found or not marked up to
3555 * date.
3556 */
3557static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3558{
3559 struct buffer_head *bh;
3560 int i;
3561 int errors = 0;
3562 bool primary_failed = false;
3563 u64 bytenr;
3564
3565 if (max_mirrors == 0)
3566 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3567
3568 for (i = 0; i < max_mirrors; i++) {
3569 bytenr = btrfs_sb_offset(i);
3570 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3571 device->commit_total_bytes)
3572 break;
3573
3574 bh = __find_get_block(device->bdev,
3575 bytenr / BTRFS_BDEV_BLOCKSIZE,
3576 BTRFS_SUPER_INFO_SIZE);
3577 if (!bh) {
3578 errors++;
3579 if (i == 0)
3580 primary_failed = true;
3581 continue;
3582 }
3583 wait_on_buffer(bh);
3584 if (!buffer_uptodate(bh)) {
3585 errors++;
3586 if (i == 0)
3587 primary_failed = true;
3588 }
3589
3590 /* drop our reference */
3591 brelse(bh);
3592
3593 /* drop the reference from the writing run */
3594 brelse(bh);
3595 }
3596
3597 /* log error, force error return */
3598 if (primary_failed) {
3599 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3600 device->devid);
3601 return -1;
3602 }
3603
3604 return errors < i ? 0 : -1;
3605}
3606
3607/*
3608 * endio for the write_dev_flush, this will wake anyone waiting
3609 * for the barrier when it is done
3610 */
3611static void btrfs_end_empty_barrier(struct bio *bio)
3612{
3613 complete(bio->bi_private);
3614}
3615
3616/*
3617 * Submit a flush request to the device if it supports it. Error handling is
3618 * done in the waiting counterpart.
3619 */
3620static void write_dev_flush(struct btrfs_device *device)
3621{
3622 struct request_queue *q = bdev_get_queue(device->bdev);
3623 struct bio *bio = device->flush_bio;
3624
3625 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3626 return;
3627
3628 bio_reset(bio);
3629 bio->bi_end_io = btrfs_end_empty_barrier;
3630 bio_set_dev(bio, device->bdev);
3631 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3632 init_completion(&device->flush_wait);
3633 bio->bi_private = &device->flush_wait;
3634
3635 btrfsic_submit_bio(bio);
3636 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3637}
3638
3639/*
3640 * If the flush bio has been submitted by write_dev_flush, wait for it.
3641 */
3642static blk_status_t wait_dev_flush(struct btrfs_device *device)
3643{
3644 struct bio *bio = device->flush_bio;
3645
3646 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3647 return BLK_STS_OK;
3648
3649 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3650 wait_for_completion_io(&device->flush_wait);
3651
3652 return bio->bi_status;
3653}
3654
3655static int check_barrier_error(struct btrfs_fs_info *fs_info)
3656{
3657 if (!btrfs_check_rw_degradable(fs_info, NULL))
3658 return -EIO;
3659 return 0;
3660}
3661
3662/*
3663 * send an empty flush down to each device in parallel,
3664 * then wait for them
3665 */
3666static int barrier_all_devices(struct btrfs_fs_info *info)
3667{
3668 struct list_head *head;
3669 struct btrfs_device *dev;
3670 int errors_wait = 0;
3671 blk_status_t ret;
3672
3673 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3674 /* send down all the barriers */
3675 head = &info->fs_devices->devices;
3676 list_for_each_entry(dev, head, dev_list) {
3677 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3678 continue;
3679 if (!dev->bdev)
3680 continue;
3681 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3682 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3683 continue;
3684
3685 write_dev_flush(dev);
3686 dev->last_flush_error = BLK_STS_OK;
3687 }
3688
3689 /* wait for all the barriers */
3690 list_for_each_entry(dev, head, dev_list) {
3691 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3692 continue;
3693 if (!dev->bdev) {
3694 errors_wait++;
3695 continue;
3696 }
3697 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3698 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3699 continue;
3700
3701 ret = wait_dev_flush(dev);
3702 if (ret) {
3703 dev->last_flush_error = ret;
3704 btrfs_dev_stat_inc_and_print(dev,
3705 BTRFS_DEV_STAT_FLUSH_ERRS);
3706 errors_wait++;
3707 }
3708 }
3709
3710 if (errors_wait) {
3711 /*
3712 * At some point we need the status of all disks
3713 * to arrive at the volume status. So error checking
3714 * is being pushed to a separate loop.
3715 */
3716 return check_barrier_error(info);
3717 }
3718 return 0;
3719}
3720
3721int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3722{
3723 int raid_type;
3724 int min_tolerated = INT_MAX;
3725
3726 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3727 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3728 min_tolerated = min_t(int, min_tolerated,
3729 btrfs_raid_array[BTRFS_RAID_SINGLE].
3730 tolerated_failures);
3731
3732 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3733 if (raid_type == BTRFS_RAID_SINGLE)
3734 continue;
3735 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3736 continue;
3737 min_tolerated = min_t(int, min_tolerated,
3738 btrfs_raid_array[raid_type].
3739 tolerated_failures);
3740 }
3741
3742 if (min_tolerated == INT_MAX) {
3743 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3744 min_tolerated = 0;
3745 }
3746
3747 return min_tolerated;
3748}
3749
3750int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3751{
3752 struct list_head *head;
3753 struct btrfs_device *dev;
3754 struct btrfs_super_block *sb;
3755 struct btrfs_dev_item *dev_item;
3756 int ret;
3757 int do_barriers;
3758 int max_errors;
3759 int total_errors = 0;
3760 u64 flags;
3761
3762 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3763
3764 /*
3765 * max_mirrors == 0 indicates we're from commit_transaction,
3766 * not from fsync where the tree roots in fs_info have not
3767 * been consistent on disk.
3768 */
3769 if (max_mirrors == 0)
3770 backup_super_roots(fs_info);
3771
3772 sb = fs_info->super_for_commit;
3773 dev_item = &sb->dev_item;
3774
3775 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3776 head = &fs_info->fs_devices->devices;
3777 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3778
3779 if (do_barriers) {
3780 ret = barrier_all_devices(fs_info);
3781 if (ret) {
3782 mutex_unlock(
3783 &fs_info->fs_devices->device_list_mutex);
3784 btrfs_handle_fs_error(fs_info, ret,
3785 "errors while submitting device barriers.");
3786 return ret;
3787 }
3788 }
3789
3790 list_for_each_entry(dev, head, dev_list) {
3791 if (!dev->bdev) {
3792 total_errors++;
3793 continue;
3794 }
3795 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3796 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3797 continue;
3798
3799 btrfs_set_stack_device_generation(dev_item, 0);
3800 btrfs_set_stack_device_type(dev_item, dev->type);
3801 btrfs_set_stack_device_id(dev_item, dev->devid);
3802 btrfs_set_stack_device_total_bytes(dev_item,
3803 dev->commit_total_bytes);
3804 btrfs_set_stack_device_bytes_used(dev_item,
3805 dev->commit_bytes_used);
3806 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3807 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3808 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3809 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3810 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3811 BTRFS_FSID_SIZE);
3812
3813 flags = btrfs_super_flags(sb);
3814 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3815
3816 ret = btrfs_validate_write_super(fs_info, sb);
3817 if (ret < 0) {
3818 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3819 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3820 "unexpected superblock corruption detected");
3821 return -EUCLEAN;
3822 }
3823
3824 ret = write_dev_supers(dev, sb, max_mirrors);
3825 if (ret)
3826 total_errors++;
3827 }
3828 if (total_errors > max_errors) {
3829 btrfs_err(fs_info, "%d errors while writing supers",
3830 total_errors);
3831 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3832
3833 /* FUA is masked off if unsupported and can't be the reason */
3834 btrfs_handle_fs_error(fs_info, -EIO,
3835 "%d errors while writing supers",
3836 total_errors);
3837 return -EIO;
3838 }
3839
3840 total_errors = 0;
3841 list_for_each_entry(dev, head, dev_list) {
3842 if (!dev->bdev)
3843 continue;
3844 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3845 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3846 continue;
3847
3848 ret = wait_dev_supers(dev, max_mirrors);
3849 if (ret)
3850 total_errors++;
3851 }
3852 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3853 if (total_errors > max_errors) {
3854 btrfs_handle_fs_error(fs_info, -EIO,
3855 "%d errors while writing supers",
3856 total_errors);
3857 return -EIO;
3858 }
3859 return 0;
3860}
3861
3862/* Drop a fs root from the radix tree and free it. */
3863void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3864 struct btrfs_root *root)
3865{
3866 spin_lock(&fs_info->fs_roots_radix_lock);
3867 radix_tree_delete(&fs_info->fs_roots_radix,
3868 (unsigned long)root->root_key.objectid);
3869 spin_unlock(&fs_info->fs_roots_radix_lock);
3870
3871 if (btrfs_root_refs(&root->root_item) == 0)
3872 synchronize_srcu(&fs_info->subvol_srcu);
3873
3874 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3875 btrfs_free_log(NULL, root);
3876 if (root->reloc_root) {
3877 free_extent_buffer(root->reloc_root->node);
3878 free_extent_buffer(root->reloc_root->commit_root);
3879 btrfs_put_fs_root(root->reloc_root);
3880 root->reloc_root = NULL;
3881 }
3882 }
3883
3884 if (root->free_ino_pinned)
3885 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3886 if (root->free_ino_ctl)
3887 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3888 btrfs_free_fs_root(root);
3889}
3890
3891void btrfs_free_fs_root(struct btrfs_root *root)
3892{
3893 iput(root->ino_cache_inode);
3894 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3895 if (root->anon_dev)
3896 free_anon_bdev(root->anon_dev);
3897 if (root->subv_writers)
3898 btrfs_free_subvolume_writers(root->subv_writers);
3899 free_extent_buffer(root->node);
3900 free_extent_buffer(root->commit_root);
3901 kfree(root->free_ino_ctl);
3902 kfree(root->free_ino_pinned);
3903 btrfs_put_fs_root(root);
3904}
3905
3906int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3907{
3908 u64 root_objectid = 0;
3909 struct btrfs_root *gang[8];
3910 int i = 0;
3911 int err = 0;
3912 unsigned int ret = 0;
3913 int index;
3914
3915 while (1) {
3916 index = srcu_read_lock(&fs_info->subvol_srcu);
3917 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3918 (void **)gang, root_objectid,
3919 ARRAY_SIZE(gang));
3920 if (!ret) {
3921 srcu_read_unlock(&fs_info->subvol_srcu, index);
3922 break;
3923 }
3924 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3925
3926 for (i = 0; i < ret; i++) {
3927 /* Avoid to grab roots in dead_roots */
3928 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3929 gang[i] = NULL;
3930 continue;
3931 }
3932 /* grab all the search result for later use */
3933 gang[i] = btrfs_grab_fs_root(gang[i]);
3934 }
3935 srcu_read_unlock(&fs_info->subvol_srcu, index);
3936
3937 for (i = 0; i < ret; i++) {
3938 if (!gang[i])
3939 continue;
3940 root_objectid = gang[i]->root_key.objectid;
3941 err = btrfs_orphan_cleanup(gang[i]);
3942 if (err)
3943 break;
3944 btrfs_put_fs_root(gang[i]);
3945 }
3946 root_objectid++;
3947 }
3948
3949 /* release the uncleaned roots due to error */
3950 for (; i < ret; i++) {
3951 if (gang[i])
3952 btrfs_put_fs_root(gang[i]);
3953 }
3954 return err;
3955}
3956
3957int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3958{
3959 struct btrfs_root *root = fs_info->tree_root;
3960 struct btrfs_trans_handle *trans;
3961
3962 mutex_lock(&fs_info->cleaner_mutex);
3963 btrfs_run_delayed_iputs(fs_info);
3964 mutex_unlock(&fs_info->cleaner_mutex);
3965 wake_up_process(fs_info->cleaner_kthread);
3966
3967 /* wait until ongoing cleanup work done */
3968 down_write(&fs_info->cleanup_work_sem);
3969 up_write(&fs_info->cleanup_work_sem);
3970
3971 trans = btrfs_join_transaction(root);
3972 if (IS_ERR(trans))
3973 return PTR_ERR(trans);
3974 return btrfs_commit_transaction(trans);
3975}
3976
3977void close_ctree(struct btrfs_fs_info *fs_info)
3978{
3979 int ret;
3980
3981 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3982 /*
3983 * We don't want the cleaner to start new transactions, add more delayed
3984 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3985 * because that frees the task_struct, and the transaction kthread might
3986 * still try to wake up the cleaner.
3987 */
3988 kthread_park(fs_info->cleaner_kthread);
3989
3990 /* wait for the qgroup rescan worker to stop */
3991 btrfs_qgroup_wait_for_completion(fs_info, false);
3992
3993 /* wait for the uuid_scan task to finish */
3994 down(&fs_info->uuid_tree_rescan_sem);
3995 /* avoid complains from lockdep et al., set sem back to initial state */
3996 up(&fs_info->uuid_tree_rescan_sem);
3997
3998 /* pause restriper - we want to resume on mount */
3999 btrfs_pause_balance(fs_info);
4000
4001 btrfs_dev_replace_suspend_for_unmount(fs_info);
4002
4003 btrfs_scrub_cancel(fs_info);
4004
4005 /* wait for any defraggers to finish */
4006 wait_event(fs_info->transaction_wait,
4007 (atomic_read(&fs_info->defrag_running) == 0));
4008
4009 /* clear out the rbtree of defraggable inodes */
4010 btrfs_cleanup_defrag_inodes(fs_info);
4011
4012 cancel_work_sync(&fs_info->async_reclaim_work);
4013
4014 if (!sb_rdonly(fs_info->sb)) {
4015 /*
4016 * The cleaner kthread is stopped, so do one final pass over
4017 * unused block groups.
4018 */
4019 btrfs_delete_unused_bgs(fs_info);
4020
4021 ret = btrfs_commit_super(fs_info);
4022 if (ret)
4023 btrfs_err(fs_info, "commit super ret %d", ret);
4024 }
4025
4026 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4027 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4028 btrfs_error_commit_super(fs_info);
4029
4030 kthread_stop(fs_info->transaction_kthread);
4031 kthread_stop(fs_info->cleaner_kthread);
4032
4033 ASSERT(list_empty(&fs_info->delayed_iputs));
4034 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4035
4036 btrfs_free_qgroup_config(fs_info);
4037 ASSERT(list_empty(&fs_info->delalloc_roots));
4038
4039 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4040 btrfs_info(fs_info, "at unmount delalloc count %lld",
4041 percpu_counter_sum(&fs_info->delalloc_bytes));
4042 }
4043
4044 if (percpu_counter_sum(&fs_info->dio_bytes))
4045 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4046 percpu_counter_sum(&fs_info->dio_bytes));
4047
4048 btrfs_sysfs_remove_mounted(fs_info);
4049 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4050
4051 btrfs_free_fs_roots(fs_info);
4052
4053 btrfs_put_block_group_cache(fs_info);
4054
4055 /*
4056 * we must make sure there is not any read request to
4057 * submit after we stopping all workers.
4058 */
4059 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4060 btrfs_stop_all_workers(fs_info);
4061
4062 btrfs_free_block_groups(fs_info);
4063
4064 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4065 free_root_pointers(fs_info, 1);
4066
4067 iput(fs_info->btree_inode);
4068
4069#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4070 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4071 btrfsic_unmount(fs_info->fs_devices);
4072#endif
4073
4074 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4075 btrfs_close_devices(fs_info->fs_devices);
4076
4077 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4078 percpu_counter_destroy(&fs_info->delalloc_bytes);
4079 percpu_counter_destroy(&fs_info->dio_bytes);
4080 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4081 cleanup_srcu_struct(&fs_info->subvol_srcu);
4082
4083 btrfs_free_csum_hash(fs_info);
4084 btrfs_free_stripe_hash_table(fs_info);
4085 btrfs_free_ref_cache(fs_info);
4086}
4087
4088int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4089 int atomic)
4090{
4091 int ret;
4092 struct inode *btree_inode = buf->pages[0]->mapping->host;
4093
4094 ret = extent_buffer_uptodate(buf);
4095 if (!ret)
4096 return ret;
4097
4098 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4099 parent_transid, atomic);
4100 if (ret == -EAGAIN)
4101 return ret;
4102 return !ret;
4103}
4104
4105void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4106{
4107 struct btrfs_fs_info *fs_info;
4108 struct btrfs_root *root;
4109 u64 transid = btrfs_header_generation(buf);
4110 int was_dirty;
4111
4112#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4113 /*
4114 * This is a fast path so only do this check if we have sanity tests
4115 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4116 * outside of the sanity tests.
4117 */
4118 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4119 return;
4120#endif
4121 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4122 fs_info = root->fs_info;
4123 btrfs_assert_tree_locked(buf);
4124 if (transid != fs_info->generation)
4125 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4126 buf->start, transid, fs_info->generation);
4127 was_dirty = set_extent_buffer_dirty(buf);
4128 if (!was_dirty)
4129 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4130 buf->len,
4131 fs_info->dirty_metadata_batch);
4132#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4133 /*
4134 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4135 * but item data not updated.
4136 * So here we should only check item pointers, not item data.
4137 */
4138 if (btrfs_header_level(buf) == 0 &&
4139 btrfs_check_leaf_relaxed(buf)) {
4140 btrfs_print_leaf(buf);
4141 ASSERT(0);
4142 }
4143#endif
4144}
4145
4146static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4147 int flush_delayed)
4148{
4149 /*
4150 * looks as though older kernels can get into trouble with
4151 * this code, they end up stuck in balance_dirty_pages forever
4152 */
4153 int ret;
4154
4155 if (current->flags & PF_MEMALLOC)
4156 return;
4157
4158 if (flush_delayed)
4159 btrfs_balance_delayed_items(fs_info);
4160
4161 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4162 BTRFS_DIRTY_METADATA_THRESH,
4163 fs_info->dirty_metadata_batch);
4164 if (ret > 0) {
4165 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4166 }
4167}
4168
4169void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4170{
4171 __btrfs_btree_balance_dirty(fs_info, 1);
4172}
4173
4174void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4175{
4176 __btrfs_btree_balance_dirty(fs_info, 0);
4177}
4178
4179int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4180 struct btrfs_key *first_key)
4181{
4182 return btree_read_extent_buffer_pages(buf, parent_transid,
4183 level, first_key);
4184}
4185
4186static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4187{
4188 /* cleanup FS via transaction */
4189 btrfs_cleanup_transaction(fs_info);
4190
4191 mutex_lock(&fs_info->cleaner_mutex);
4192 btrfs_run_delayed_iputs(fs_info);
4193 mutex_unlock(&fs_info->cleaner_mutex);
4194
4195 down_write(&fs_info->cleanup_work_sem);
4196 up_write(&fs_info->cleanup_work_sem);
4197}
4198
4199static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4200{
4201 struct btrfs_ordered_extent *ordered;
4202
4203 spin_lock(&root->ordered_extent_lock);
4204 /*
4205 * This will just short circuit the ordered completion stuff which will
4206 * make sure the ordered extent gets properly cleaned up.
4207 */
4208 list_for_each_entry(ordered, &root->ordered_extents,
4209 root_extent_list)
4210 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4211 spin_unlock(&root->ordered_extent_lock);
4212}
4213
4214static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4215{
4216 struct btrfs_root *root;
4217 struct list_head splice;
4218
4219 INIT_LIST_HEAD(&splice);
4220
4221 spin_lock(&fs_info->ordered_root_lock);
4222 list_splice_init(&fs_info->ordered_roots, &splice);
4223 while (!list_empty(&splice)) {
4224 root = list_first_entry(&splice, struct btrfs_root,
4225 ordered_root);
4226 list_move_tail(&root->ordered_root,
4227 &fs_info->ordered_roots);
4228
4229 spin_unlock(&fs_info->ordered_root_lock);
4230 btrfs_destroy_ordered_extents(root);
4231
4232 cond_resched();
4233 spin_lock(&fs_info->ordered_root_lock);
4234 }
4235 spin_unlock(&fs_info->ordered_root_lock);
4236
4237 /*
4238 * We need this here because if we've been flipped read-only we won't
4239 * get sync() from the umount, so we need to make sure any ordered
4240 * extents that haven't had their dirty pages IO start writeout yet
4241 * actually get run and error out properly.
4242 */
4243 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4244}
4245
4246static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4247 struct btrfs_fs_info *fs_info)
4248{
4249 struct rb_node *node;
4250 struct btrfs_delayed_ref_root *delayed_refs;
4251 struct btrfs_delayed_ref_node *ref;
4252 int ret = 0;
4253
4254 delayed_refs = &trans->delayed_refs;
4255
4256 spin_lock(&delayed_refs->lock);
4257 if (atomic_read(&delayed_refs->num_entries) == 0) {
4258 spin_unlock(&delayed_refs->lock);
4259 btrfs_info(fs_info, "delayed_refs has NO entry");
4260 return ret;
4261 }
4262
4263 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4264 struct btrfs_delayed_ref_head *head;
4265 struct rb_node *n;
4266 bool pin_bytes = false;
4267
4268 head = rb_entry(node, struct btrfs_delayed_ref_head,
4269 href_node);
4270 if (btrfs_delayed_ref_lock(delayed_refs, head))
4271 continue;
4272
4273 spin_lock(&head->lock);
4274 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4275 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4276 ref_node);
4277 ref->in_tree = 0;
4278 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4279 RB_CLEAR_NODE(&ref->ref_node);
4280 if (!list_empty(&ref->add_list))
4281 list_del(&ref->add_list);
4282 atomic_dec(&delayed_refs->num_entries);
4283 btrfs_put_delayed_ref(ref);
4284 }
4285 if (head->must_insert_reserved)
4286 pin_bytes = true;
4287 btrfs_free_delayed_extent_op(head->extent_op);
4288 btrfs_delete_ref_head(delayed_refs, head);
4289 spin_unlock(&head->lock);
4290 spin_unlock(&delayed_refs->lock);
4291 mutex_unlock(&head->mutex);
4292
4293 if (pin_bytes)
4294 btrfs_pin_extent(fs_info, head->bytenr,
4295 head->num_bytes, 1);
4296 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4297 btrfs_put_delayed_ref_head(head);
4298 cond_resched();
4299 spin_lock(&delayed_refs->lock);
4300 }
4301
4302 spin_unlock(&delayed_refs->lock);
4303
4304 return ret;
4305}
4306
4307static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4308{
4309 struct btrfs_inode *btrfs_inode;
4310 struct list_head splice;
4311
4312 INIT_LIST_HEAD(&splice);
4313
4314 spin_lock(&root->delalloc_lock);
4315 list_splice_init(&root->delalloc_inodes, &splice);
4316
4317 while (!list_empty(&splice)) {
4318 struct inode *inode = NULL;
4319 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4320 delalloc_inodes);
4321 __btrfs_del_delalloc_inode(root, btrfs_inode);
4322 spin_unlock(&root->delalloc_lock);
4323
4324 /*
4325 * Make sure we get a live inode and that it'll not disappear
4326 * meanwhile.
4327 */
4328 inode = igrab(&btrfs_inode->vfs_inode);
4329 if (inode) {
4330 invalidate_inode_pages2(inode->i_mapping);
4331 iput(inode);
4332 }
4333 spin_lock(&root->delalloc_lock);
4334 }
4335 spin_unlock(&root->delalloc_lock);
4336}
4337
4338static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4339{
4340 struct btrfs_root *root;
4341 struct list_head splice;
4342
4343 INIT_LIST_HEAD(&splice);
4344
4345 spin_lock(&fs_info->delalloc_root_lock);
4346 list_splice_init(&fs_info->delalloc_roots, &splice);
4347 while (!list_empty(&splice)) {
4348 root = list_first_entry(&splice, struct btrfs_root,
4349 delalloc_root);
4350 root = btrfs_grab_fs_root(root);
4351 BUG_ON(!root);
4352 spin_unlock(&fs_info->delalloc_root_lock);
4353
4354 btrfs_destroy_delalloc_inodes(root);
4355 btrfs_put_fs_root(root);
4356
4357 spin_lock(&fs_info->delalloc_root_lock);
4358 }
4359 spin_unlock(&fs_info->delalloc_root_lock);
4360}
4361
4362static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4363 struct extent_io_tree *dirty_pages,
4364 int mark)
4365{
4366 int ret;
4367 struct extent_buffer *eb;
4368 u64 start = 0;
4369 u64 end;
4370
4371 while (1) {
4372 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4373 mark, NULL);
4374 if (ret)
4375 break;
4376
4377 clear_extent_bits(dirty_pages, start, end, mark);
4378 while (start <= end) {
4379 eb = find_extent_buffer(fs_info, start);
4380 start += fs_info->nodesize;
4381 if (!eb)
4382 continue;
4383 wait_on_extent_buffer_writeback(eb);
4384
4385 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4386 &eb->bflags))
4387 clear_extent_buffer_dirty(eb);
4388 free_extent_buffer_stale(eb);
4389 }
4390 }
4391
4392 return ret;
4393}
4394
4395static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4396 struct extent_io_tree *pinned_extents)
4397{
4398 struct extent_io_tree *unpin;
4399 u64 start;
4400 u64 end;
4401 int ret;
4402 bool loop = true;
4403
4404 unpin = pinned_extents;
4405again:
4406 while (1) {
4407 struct extent_state *cached_state = NULL;
4408
4409 /*
4410 * The btrfs_finish_extent_commit() may get the same range as
4411 * ours between find_first_extent_bit and clear_extent_dirty.
4412 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4413 * the same extent range.
4414 */
4415 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4416 ret = find_first_extent_bit(unpin, 0, &start, &end,
4417 EXTENT_DIRTY, &cached_state);
4418 if (ret) {
4419 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4420 break;
4421 }
4422
4423 clear_extent_dirty(unpin, start, end, &cached_state);
4424 free_extent_state(cached_state);
4425 btrfs_error_unpin_extent_range(fs_info, start, end);
4426 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4427 cond_resched();
4428 }
4429
4430 if (loop) {
4431 if (unpin == &fs_info->freed_extents[0])
4432 unpin = &fs_info->freed_extents[1];
4433 else
4434 unpin = &fs_info->freed_extents[0];
4435 loop = false;
4436 goto again;
4437 }
4438
4439 return 0;
4440}
4441
4442static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4443{
4444 struct inode *inode;
4445
4446 inode = cache->io_ctl.inode;
4447 if (inode) {
4448 invalidate_inode_pages2(inode->i_mapping);
4449 BTRFS_I(inode)->generation = 0;
4450 cache->io_ctl.inode = NULL;
4451 iput(inode);
4452 }
4453 btrfs_put_block_group(cache);
4454}
4455
4456void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4457 struct btrfs_fs_info *fs_info)
4458{
4459 struct btrfs_block_group_cache *cache;
4460
4461 spin_lock(&cur_trans->dirty_bgs_lock);
4462 while (!list_empty(&cur_trans->dirty_bgs)) {
4463 cache = list_first_entry(&cur_trans->dirty_bgs,
4464 struct btrfs_block_group_cache,
4465 dirty_list);
4466
4467 if (!list_empty(&cache->io_list)) {
4468 spin_unlock(&cur_trans->dirty_bgs_lock);
4469 list_del_init(&cache->io_list);
4470 btrfs_cleanup_bg_io(cache);
4471 spin_lock(&cur_trans->dirty_bgs_lock);
4472 }
4473
4474 list_del_init(&cache->dirty_list);
4475 spin_lock(&cache->lock);
4476 cache->disk_cache_state = BTRFS_DC_ERROR;
4477 spin_unlock(&cache->lock);
4478
4479 spin_unlock(&cur_trans->dirty_bgs_lock);
4480 btrfs_put_block_group(cache);
4481 btrfs_delayed_refs_rsv_release(fs_info, 1);
4482 spin_lock(&cur_trans->dirty_bgs_lock);
4483 }
4484 spin_unlock(&cur_trans->dirty_bgs_lock);
4485
4486 /*
4487 * Refer to the definition of io_bgs member for details why it's safe
4488 * to use it without any locking
4489 */
4490 while (!list_empty(&cur_trans->io_bgs)) {
4491 cache = list_first_entry(&cur_trans->io_bgs,
4492 struct btrfs_block_group_cache,
4493 io_list);
4494
4495 list_del_init(&cache->io_list);
4496 spin_lock(&cache->lock);
4497 cache->disk_cache_state = BTRFS_DC_ERROR;
4498 spin_unlock(&cache->lock);
4499 btrfs_cleanup_bg_io(cache);
4500 }
4501}
4502
4503void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4504 struct btrfs_fs_info *fs_info)
4505{
4506 struct btrfs_device *dev, *tmp;
4507
4508 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4509 ASSERT(list_empty(&cur_trans->dirty_bgs));
4510 ASSERT(list_empty(&cur_trans->io_bgs));
4511
4512 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4513 post_commit_list) {
4514 list_del_init(&dev->post_commit_list);
4515 }
4516
4517 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4518
4519 cur_trans->state = TRANS_STATE_COMMIT_START;
4520 wake_up(&fs_info->transaction_blocked_wait);
4521
4522 cur_trans->state = TRANS_STATE_UNBLOCKED;
4523 wake_up(&fs_info->transaction_wait);
4524
4525 btrfs_destroy_delayed_inodes(fs_info);
4526 btrfs_assert_delayed_root_empty(fs_info);
4527
4528 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4529 EXTENT_DIRTY);
4530 btrfs_destroy_pinned_extent(fs_info,
4531 fs_info->pinned_extents);
4532
4533 cur_trans->state =TRANS_STATE_COMPLETED;
4534 wake_up(&cur_trans->commit_wait);
4535}
4536
4537static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4538{
4539 struct btrfs_transaction *t;
4540
4541 mutex_lock(&fs_info->transaction_kthread_mutex);
4542
4543 spin_lock(&fs_info->trans_lock);
4544 while (!list_empty(&fs_info->trans_list)) {
4545 t = list_first_entry(&fs_info->trans_list,
4546 struct btrfs_transaction, list);
4547 if (t->state >= TRANS_STATE_COMMIT_START) {
4548 refcount_inc(&t->use_count);
4549 spin_unlock(&fs_info->trans_lock);
4550 btrfs_wait_for_commit(fs_info, t->transid);
4551 btrfs_put_transaction(t);
4552 spin_lock(&fs_info->trans_lock);
4553 continue;
4554 }
4555 if (t == fs_info->running_transaction) {
4556 t->state = TRANS_STATE_COMMIT_DOING;
4557 spin_unlock(&fs_info->trans_lock);
4558 /*
4559 * We wait for 0 num_writers since we don't hold a trans
4560 * handle open currently for this transaction.
4561 */
4562 wait_event(t->writer_wait,
4563 atomic_read(&t->num_writers) == 0);
4564 } else {
4565 spin_unlock(&fs_info->trans_lock);
4566 }
4567 btrfs_cleanup_one_transaction(t, fs_info);
4568
4569 spin_lock(&fs_info->trans_lock);
4570 if (t == fs_info->running_transaction)
4571 fs_info->running_transaction = NULL;
4572 list_del_init(&t->list);
4573 spin_unlock(&fs_info->trans_lock);
4574
4575 btrfs_put_transaction(t);
4576 trace_btrfs_transaction_commit(fs_info->tree_root);
4577 spin_lock(&fs_info->trans_lock);
4578 }
4579 spin_unlock(&fs_info->trans_lock);
4580 btrfs_destroy_all_ordered_extents(fs_info);
4581 btrfs_destroy_delayed_inodes(fs_info);
4582 btrfs_assert_delayed_root_empty(fs_info);
4583 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4584 btrfs_destroy_all_delalloc_inodes(fs_info);
4585 mutex_unlock(&fs_info->transaction_kthread_mutex);
4586
4587 return 0;
4588}
4589
4590static const struct extent_io_ops btree_extent_io_ops = {
4591 /* mandatory callbacks */
4592 .submit_bio_hook = btree_submit_bio_hook,
4593 .readpage_end_io_hook = btree_readpage_end_io_hook,
4594};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/fs.h>
7#include <linux/blkdev.h>
8#include <linux/radix-tree.h>
9#include <linux/writeback.h>
10#include <linux/workqueue.h>
11#include <linux/kthread.h>
12#include <linux/slab.h>
13#include <linux/migrate.h>
14#include <linux/ratelimit.h>
15#include <linux/uuid.h>
16#include <linux/semaphore.h>
17#include <linux/error-injection.h>
18#include <linux/crc32c.h>
19#include <linux/sched/mm.h>
20#include <linux/unaligned.h>
21#include <crypto/hash.h>
22#include "ctree.h"
23#include "disk-io.h"
24#include "transaction.h"
25#include "btrfs_inode.h"
26#include "bio.h"
27#include "print-tree.h"
28#include "locking.h"
29#include "tree-log.h"
30#include "free-space-cache.h"
31#include "free-space-tree.h"
32#include "dev-replace.h"
33#include "raid56.h"
34#include "sysfs.h"
35#include "qgroup.h"
36#include "compression.h"
37#include "tree-checker.h"
38#include "ref-verify.h"
39#include "block-group.h"
40#include "discard.h"
41#include "space-info.h"
42#include "zoned.h"
43#include "subpage.h"
44#include "fs.h"
45#include "accessors.h"
46#include "extent-tree.h"
47#include "root-tree.h"
48#include "defrag.h"
49#include "uuid-tree.h"
50#include "relocation.h"
51#include "scrub.h"
52#include "super.h"
53
54#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
55 BTRFS_HEADER_FLAG_RELOC |\
56 BTRFS_SUPER_FLAG_ERROR |\
57 BTRFS_SUPER_FLAG_SEEDING |\
58 BTRFS_SUPER_FLAG_METADUMP |\
59 BTRFS_SUPER_FLAG_METADUMP_V2)
60
61static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
62static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
63
64static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
65{
66 if (fs_info->csum_shash)
67 crypto_free_shash(fs_info->csum_shash);
68}
69
70/*
71 * Compute the csum of a btree block and store the result to provided buffer.
72 */
73static void csum_tree_block(struct extent_buffer *buf, u8 *result)
74{
75 struct btrfs_fs_info *fs_info = buf->fs_info;
76 int num_pages;
77 u32 first_page_part;
78 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
79 char *kaddr;
80 int i;
81
82 shash->tfm = fs_info->csum_shash;
83 crypto_shash_init(shash);
84
85 if (buf->addr) {
86 /* Pages are contiguous, handle them as a big one. */
87 kaddr = buf->addr;
88 first_page_part = fs_info->nodesize;
89 num_pages = 1;
90 } else {
91 kaddr = folio_address(buf->folios[0]);
92 first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
93 num_pages = num_extent_pages(buf);
94 }
95
96 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
97 first_page_part - BTRFS_CSUM_SIZE);
98
99 /*
100 * Multiple single-page folios case would reach here.
101 *
102 * nodesize <= PAGE_SIZE and large folio all handled by above
103 * crypto_shash_update() already.
104 */
105 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
106 kaddr = folio_address(buf->folios[i]);
107 crypto_shash_update(shash, kaddr, PAGE_SIZE);
108 }
109 memset(result, 0, BTRFS_CSUM_SIZE);
110 crypto_shash_final(shash, result);
111}
112
113/*
114 * we can't consider a given block up to date unless the transid of the
115 * block matches the transid in the parent node's pointer. This is how we
116 * detect blocks that either didn't get written at all or got written
117 * in the wrong place.
118 */
119int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
120{
121 if (!extent_buffer_uptodate(eb))
122 return 0;
123
124 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
125 return 1;
126
127 if (atomic)
128 return -EAGAIN;
129
130 if (!extent_buffer_uptodate(eb) ||
131 btrfs_header_generation(eb) != parent_transid) {
132 btrfs_err_rl(eb->fs_info,
133"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
134 eb->start, eb->read_mirror,
135 parent_transid, btrfs_header_generation(eb));
136 clear_extent_buffer_uptodate(eb);
137 return 0;
138 }
139 return 1;
140}
141
142static bool btrfs_supported_super_csum(u16 csum_type)
143{
144 switch (csum_type) {
145 case BTRFS_CSUM_TYPE_CRC32:
146 case BTRFS_CSUM_TYPE_XXHASH:
147 case BTRFS_CSUM_TYPE_SHA256:
148 case BTRFS_CSUM_TYPE_BLAKE2:
149 return true;
150 default:
151 return false;
152 }
153}
154
155/*
156 * Return 0 if the superblock checksum type matches the checksum value of that
157 * algorithm. Pass the raw disk superblock data.
158 */
159int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
160 const struct btrfs_super_block *disk_sb)
161{
162 char result[BTRFS_CSUM_SIZE];
163 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
164
165 shash->tfm = fs_info->csum_shash;
166
167 /*
168 * The super_block structure does not span the whole
169 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
170 * filled with zeros and is included in the checksum.
171 */
172 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
173 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
174
175 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
176 return 1;
177
178 return 0;
179}
180
181static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
182 int mirror_num)
183{
184 struct btrfs_fs_info *fs_info = eb->fs_info;
185 int num_folios = num_extent_folios(eb);
186 int ret = 0;
187
188 if (sb_rdonly(fs_info->sb))
189 return -EROFS;
190
191 for (int i = 0; i < num_folios; i++) {
192 struct folio *folio = eb->folios[i];
193 u64 start = max_t(u64, eb->start, folio_pos(folio));
194 u64 end = min_t(u64, eb->start + eb->len,
195 folio_pos(folio) + eb->folio_size);
196 u32 len = end - start;
197
198 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
199 start, folio, offset_in_folio(folio, start),
200 mirror_num);
201 if (ret)
202 break;
203 }
204
205 return ret;
206}
207
208/*
209 * helper to read a given tree block, doing retries as required when
210 * the checksums don't match and we have alternate mirrors to try.
211 *
212 * @check: expected tree parentness check, see the comments of the
213 * structure for details.
214 */
215int btrfs_read_extent_buffer(struct extent_buffer *eb,
216 const struct btrfs_tree_parent_check *check)
217{
218 struct btrfs_fs_info *fs_info = eb->fs_info;
219 int failed = 0;
220 int ret;
221 int num_copies = 0;
222 int mirror_num = 0;
223 int failed_mirror = 0;
224
225 ASSERT(check);
226
227 while (1) {
228 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
229 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
230 if (!ret)
231 break;
232
233 num_copies = btrfs_num_copies(fs_info,
234 eb->start, eb->len);
235 if (num_copies == 1)
236 break;
237
238 if (!failed_mirror) {
239 failed = 1;
240 failed_mirror = eb->read_mirror;
241 }
242
243 mirror_num++;
244 if (mirror_num == failed_mirror)
245 mirror_num++;
246
247 if (mirror_num > num_copies)
248 break;
249 }
250
251 if (failed && !ret && failed_mirror)
252 btrfs_repair_eb_io_failure(eb, failed_mirror);
253
254 return ret;
255}
256
257/*
258 * Checksum a dirty tree block before IO.
259 */
260blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
261{
262 struct extent_buffer *eb = bbio->private;
263 struct btrfs_fs_info *fs_info = eb->fs_info;
264 u64 found_start = btrfs_header_bytenr(eb);
265 u64 last_trans;
266 u8 result[BTRFS_CSUM_SIZE];
267 int ret;
268
269 /* Btree blocks are always contiguous on disk. */
270 if (WARN_ON_ONCE(bbio->file_offset != eb->start))
271 return BLK_STS_IOERR;
272 if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
273 return BLK_STS_IOERR;
274
275 /*
276 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
277 * checksum it but zero-out its content. This is done to preserve
278 * ordering of I/O without unnecessarily writing out data.
279 */
280 if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
281 memzero_extent_buffer(eb, 0, eb->len);
282 return BLK_STS_OK;
283 }
284
285 if (WARN_ON_ONCE(found_start != eb->start))
286 return BLK_STS_IOERR;
287 if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
288 eb->start, eb->len)))
289 return BLK_STS_IOERR;
290
291 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
292 offsetof(struct btrfs_header, fsid),
293 BTRFS_FSID_SIZE) == 0);
294 csum_tree_block(eb, result);
295
296 if (btrfs_header_level(eb))
297 ret = btrfs_check_node(eb);
298 else
299 ret = btrfs_check_leaf(eb);
300
301 if (ret < 0)
302 goto error;
303
304 /*
305 * Also check the generation, the eb reached here must be newer than
306 * last committed. Or something seriously wrong happened.
307 */
308 last_trans = btrfs_get_last_trans_committed(fs_info);
309 if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
310 ret = -EUCLEAN;
311 btrfs_err(fs_info,
312 "block=%llu bad generation, have %llu expect > %llu",
313 eb->start, btrfs_header_generation(eb), last_trans);
314 goto error;
315 }
316 write_extent_buffer(eb, result, 0, fs_info->csum_size);
317 return BLK_STS_OK;
318
319error:
320 btrfs_print_tree(eb, 0);
321 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
322 eb->start);
323 /*
324 * Be noisy if this is an extent buffer from a log tree. We don't abort
325 * a transaction in case there's a bad log tree extent buffer, we just
326 * fallback to a transaction commit. Still we want to know when there is
327 * a bad log tree extent buffer, as that may signal a bug somewhere.
328 */
329 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
330 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
331 return errno_to_blk_status(ret);
332}
333
334static bool check_tree_block_fsid(struct extent_buffer *eb)
335{
336 struct btrfs_fs_info *fs_info = eb->fs_info;
337 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
338 u8 fsid[BTRFS_FSID_SIZE];
339
340 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
341 BTRFS_FSID_SIZE);
342
343 /*
344 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
345 * This is then overwritten by metadata_uuid if it is present in the
346 * device_list_add(). The same true for a seed device as well. So use of
347 * fs_devices::metadata_uuid is appropriate here.
348 */
349 if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
350 return false;
351
352 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
353 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
354 return false;
355
356 return true;
357}
358
359/* Do basic extent buffer checks at read time */
360int btrfs_validate_extent_buffer(struct extent_buffer *eb,
361 const struct btrfs_tree_parent_check *check)
362{
363 struct btrfs_fs_info *fs_info = eb->fs_info;
364 u64 found_start;
365 const u32 csum_size = fs_info->csum_size;
366 u8 found_level;
367 u8 result[BTRFS_CSUM_SIZE];
368 const u8 *header_csum;
369 int ret = 0;
370 const bool ignore_csum = btrfs_test_opt(fs_info, IGNOREMETACSUMS);
371
372 ASSERT(check);
373
374 found_start = btrfs_header_bytenr(eb);
375 if (found_start != eb->start) {
376 btrfs_err_rl(fs_info,
377 "bad tree block start, mirror %u want %llu have %llu",
378 eb->read_mirror, eb->start, found_start);
379 ret = -EIO;
380 goto out;
381 }
382 if (check_tree_block_fsid(eb)) {
383 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
384 eb->start, eb->read_mirror);
385 ret = -EIO;
386 goto out;
387 }
388 found_level = btrfs_header_level(eb);
389 if (found_level >= BTRFS_MAX_LEVEL) {
390 btrfs_err(fs_info,
391 "bad tree block level, mirror %u level %d on logical %llu",
392 eb->read_mirror, btrfs_header_level(eb), eb->start);
393 ret = -EIO;
394 goto out;
395 }
396
397 csum_tree_block(eb, result);
398 header_csum = folio_address(eb->folios[0]) +
399 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
400
401 if (memcmp(result, header_csum, csum_size) != 0) {
402 btrfs_warn_rl(fs_info,
403"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d%s",
404 eb->start, eb->read_mirror,
405 CSUM_FMT_VALUE(csum_size, header_csum),
406 CSUM_FMT_VALUE(csum_size, result),
407 btrfs_header_level(eb),
408 ignore_csum ? ", ignored" : "");
409 if (!ignore_csum) {
410 ret = -EUCLEAN;
411 goto out;
412 }
413 }
414
415 if (found_level != check->level) {
416 btrfs_err(fs_info,
417 "level verify failed on logical %llu mirror %u wanted %u found %u",
418 eb->start, eb->read_mirror, check->level, found_level);
419 ret = -EIO;
420 goto out;
421 }
422 if (unlikely(check->transid &&
423 btrfs_header_generation(eb) != check->transid)) {
424 btrfs_err_rl(eb->fs_info,
425"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
426 eb->start, eb->read_mirror, check->transid,
427 btrfs_header_generation(eb));
428 ret = -EIO;
429 goto out;
430 }
431 if (check->has_first_key) {
432 const struct btrfs_key *expect_key = &check->first_key;
433 struct btrfs_key found_key;
434
435 if (found_level)
436 btrfs_node_key_to_cpu(eb, &found_key, 0);
437 else
438 btrfs_item_key_to_cpu(eb, &found_key, 0);
439 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
440 btrfs_err(fs_info,
441"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
442 eb->start, check->transid,
443 expect_key->objectid,
444 expect_key->type, expect_key->offset,
445 found_key.objectid, found_key.type,
446 found_key.offset);
447 ret = -EUCLEAN;
448 goto out;
449 }
450 }
451 if (check->owner_root) {
452 ret = btrfs_check_eb_owner(eb, check->owner_root);
453 if (ret < 0)
454 goto out;
455 }
456
457 /*
458 * If this is a leaf block and it is corrupt, set the corrupt bit so
459 * that we don't try and read the other copies of this block, just
460 * return -EIO.
461 */
462 if (found_level == 0 && btrfs_check_leaf(eb)) {
463 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
464 ret = -EIO;
465 }
466
467 if (found_level > 0 && btrfs_check_node(eb))
468 ret = -EIO;
469
470 if (ret)
471 btrfs_err(fs_info,
472 "read time tree block corruption detected on logical %llu mirror %u",
473 eb->start, eb->read_mirror);
474out:
475 return ret;
476}
477
478#ifdef CONFIG_MIGRATION
479static int btree_migrate_folio(struct address_space *mapping,
480 struct folio *dst, struct folio *src, enum migrate_mode mode)
481{
482 /*
483 * we can't safely write a btree page from here,
484 * we haven't done the locking hook
485 */
486 if (folio_test_dirty(src))
487 return -EAGAIN;
488 /*
489 * Buffers may be managed in a filesystem specific way.
490 * We must have no buffers or drop them.
491 */
492 if (folio_get_private(src) &&
493 !filemap_release_folio(src, GFP_KERNEL))
494 return -EAGAIN;
495 return migrate_folio(mapping, dst, src, mode);
496}
497#else
498#define btree_migrate_folio NULL
499#endif
500
501static int btree_writepages(struct address_space *mapping,
502 struct writeback_control *wbc)
503{
504 int ret;
505
506 if (wbc->sync_mode == WB_SYNC_NONE) {
507 struct btrfs_fs_info *fs_info;
508
509 if (wbc->for_kupdate)
510 return 0;
511
512 fs_info = inode_to_fs_info(mapping->host);
513 /* this is a bit racy, but that's ok */
514 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
515 BTRFS_DIRTY_METADATA_THRESH,
516 fs_info->dirty_metadata_batch);
517 if (ret < 0)
518 return 0;
519 }
520 return btree_write_cache_pages(mapping, wbc);
521}
522
523static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
524{
525 if (folio_test_writeback(folio) || folio_test_dirty(folio))
526 return false;
527
528 return try_release_extent_buffer(folio);
529}
530
531static void btree_invalidate_folio(struct folio *folio, size_t offset,
532 size_t length)
533{
534 struct extent_io_tree *tree;
535
536 tree = &folio_to_inode(folio)->io_tree;
537 extent_invalidate_folio(tree, folio, offset);
538 btree_release_folio(folio, GFP_NOFS);
539 if (folio_get_private(folio)) {
540 btrfs_warn(folio_to_fs_info(folio),
541 "folio private not zero on folio %llu",
542 (unsigned long long)folio_pos(folio));
543 folio_detach_private(folio);
544 }
545}
546
547#ifdef DEBUG
548static bool btree_dirty_folio(struct address_space *mapping,
549 struct folio *folio)
550{
551 struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
552 struct btrfs_subpage_info *spi = fs_info->subpage_info;
553 struct btrfs_subpage *subpage;
554 struct extent_buffer *eb;
555 int cur_bit = 0;
556 u64 page_start = folio_pos(folio);
557
558 if (fs_info->sectorsize == PAGE_SIZE) {
559 eb = folio_get_private(folio);
560 BUG_ON(!eb);
561 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
562 BUG_ON(!atomic_read(&eb->refs));
563 btrfs_assert_tree_write_locked(eb);
564 return filemap_dirty_folio(mapping, folio);
565 }
566
567 ASSERT(spi);
568 subpage = folio_get_private(folio);
569
570 for (cur_bit = spi->dirty_offset;
571 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
572 cur_bit++) {
573 unsigned long flags;
574 u64 cur;
575
576 spin_lock_irqsave(&subpage->lock, flags);
577 if (!test_bit(cur_bit, subpage->bitmaps)) {
578 spin_unlock_irqrestore(&subpage->lock, flags);
579 continue;
580 }
581 spin_unlock_irqrestore(&subpage->lock, flags);
582 cur = page_start + cur_bit * fs_info->sectorsize;
583
584 eb = find_extent_buffer(fs_info, cur);
585 ASSERT(eb);
586 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
587 ASSERT(atomic_read(&eb->refs));
588 btrfs_assert_tree_write_locked(eb);
589 free_extent_buffer(eb);
590
591 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
592 }
593 return filemap_dirty_folio(mapping, folio);
594}
595#else
596#define btree_dirty_folio filemap_dirty_folio
597#endif
598
599static const struct address_space_operations btree_aops = {
600 .writepages = btree_writepages,
601 .release_folio = btree_release_folio,
602 .invalidate_folio = btree_invalidate_folio,
603 .migrate_folio = btree_migrate_folio,
604 .dirty_folio = btree_dirty_folio,
605};
606
607struct extent_buffer *btrfs_find_create_tree_block(
608 struct btrfs_fs_info *fs_info,
609 u64 bytenr, u64 owner_root,
610 int level)
611{
612 if (btrfs_is_testing(fs_info))
613 return alloc_test_extent_buffer(fs_info, bytenr);
614 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
615}
616
617/*
618 * Read tree block at logical address @bytenr and do variant basic but critical
619 * verification.
620 *
621 * @check: expected tree parentness check, see comments of the
622 * structure for details.
623 */
624struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
625 struct btrfs_tree_parent_check *check)
626{
627 struct extent_buffer *buf = NULL;
628 int ret;
629
630 ASSERT(check);
631
632 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
633 check->level);
634 if (IS_ERR(buf))
635 return buf;
636
637 ret = btrfs_read_extent_buffer(buf, check);
638 if (ret) {
639 free_extent_buffer_stale(buf);
640 return ERR_PTR(ret);
641 }
642 return buf;
643
644}
645
646static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
647 u64 objectid)
648{
649 bool dummy = btrfs_is_testing(fs_info);
650
651 memset(&root->root_key, 0, sizeof(root->root_key));
652 memset(&root->root_item, 0, sizeof(root->root_item));
653 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
654 root->fs_info = fs_info;
655 root->root_key.objectid = objectid;
656 root->node = NULL;
657 root->commit_root = NULL;
658 root->state = 0;
659 RB_CLEAR_NODE(&root->rb_node);
660
661 btrfs_set_root_last_trans(root, 0);
662 root->free_objectid = 0;
663 root->nr_delalloc_inodes = 0;
664 root->nr_ordered_extents = 0;
665 xa_init(&root->inodes);
666 xa_init(&root->delayed_nodes);
667
668 btrfs_init_root_block_rsv(root);
669
670 INIT_LIST_HEAD(&root->dirty_list);
671 INIT_LIST_HEAD(&root->root_list);
672 INIT_LIST_HEAD(&root->delalloc_inodes);
673 INIT_LIST_HEAD(&root->delalloc_root);
674 INIT_LIST_HEAD(&root->ordered_extents);
675 INIT_LIST_HEAD(&root->ordered_root);
676 INIT_LIST_HEAD(&root->reloc_dirty_list);
677 spin_lock_init(&root->delalloc_lock);
678 spin_lock_init(&root->ordered_extent_lock);
679 spin_lock_init(&root->accounting_lock);
680 spin_lock_init(&root->qgroup_meta_rsv_lock);
681 mutex_init(&root->objectid_mutex);
682 mutex_init(&root->log_mutex);
683 mutex_init(&root->ordered_extent_mutex);
684 mutex_init(&root->delalloc_mutex);
685 init_waitqueue_head(&root->qgroup_flush_wait);
686 init_waitqueue_head(&root->log_writer_wait);
687 init_waitqueue_head(&root->log_commit_wait[0]);
688 init_waitqueue_head(&root->log_commit_wait[1]);
689 INIT_LIST_HEAD(&root->log_ctxs[0]);
690 INIT_LIST_HEAD(&root->log_ctxs[1]);
691 atomic_set(&root->log_commit[0], 0);
692 atomic_set(&root->log_commit[1], 0);
693 atomic_set(&root->log_writers, 0);
694 atomic_set(&root->log_batch, 0);
695 refcount_set(&root->refs, 1);
696 atomic_set(&root->snapshot_force_cow, 0);
697 atomic_set(&root->nr_swapfiles, 0);
698 btrfs_set_root_log_transid(root, 0);
699 root->log_transid_committed = -1;
700 btrfs_set_root_last_log_commit(root, 0);
701 root->anon_dev = 0;
702 if (!dummy) {
703 extent_io_tree_init(fs_info, &root->dirty_log_pages,
704 IO_TREE_ROOT_DIRTY_LOG_PAGES);
705 extent_io_tree_init(fs_info, &root->log_csum_range,
706 IO_TREE_LOG_CSUM_RANGE);
707 }
708
709 spin_lock_init(&root->root_item_lock);
710 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
711#ifdef CONFIG_BTRFS_DEBUG
712 INIT_LIST_HEAD(&root->leak_list);
713 spin_lock(&fs_info->fs_roots_radix_lock);
714 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
715 spin_unlock(&fs_info->fs_roots_radix_lock);
716#endif
717}
718
719static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
720 u64 objectid, gfp_t flags)
721{
722 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
723 if (root)
724 __setup_root(root, fs_info, objectid);
725 return root;
726}
727
728#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
729/* Should only be used by the testing infrastructure */
730struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
731{
732 struct btrfs_root *root;
733
734 if (!fs_info)
735 return ERR_PTR(-EINVAL);
736
737 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
738 if (!root)
739 return ERR_PTR(-ENOMEM);
740
741 /* We don't use the stripesize in selftest, set it as sectorsize */
742 root->alloc_bytenr = 0;
743
744 return root;
745}
746#endif
747
748static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
749{
750 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
751 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
752
753 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
754}
755
756static int global_root_key_cmp(const void *k, const struct rb_node *node)
757{
758 const struct btrfs_key *key = k;
759 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
760
761 return btrfs_comp_cpu_keys(key, &root->root_key);
762}
763
764int btrfs_global_root_insert(struct btrfs_root *root)
765{
766 struct btrfs_fs_info *fs_info = root->fs_info;
767 struct rb_node *tmp;
768 int ret = 0;
769
770 write_lock(&fs_info->global_root_lock);
771 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
772 write_unlock(&fs_info->global_root_lock);
773
774 if (tmp) {
775 ret = -EEXIST;
776 btrfs_warn(fs_info, "global root %llu %llu already exists",
777 btrfs_root_id(root), root->root_key.offset);
778 }
779 return ret;
780}
781
782void btrfs_global_root_delete(struct btrfs_root *root)
783{
784 struct btrfs_fs_info *fs_info = root->fs_info;
785
786 write_lock(&fs_info->global_root_lock);
787 rb_erase(&root->rb_node, &fs_info->global_root_tree);
788 write_unlock(&fs_info->global_root_lock);
789}
790
791struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
792 struct btrfs_key *key)
793{
794 struct rb_node *node;
795 struct btrfs_root *root = NULL;
796
797 read_lock(&fs_info->global_root_lock);
798 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
799 if (node)
800 root = container_of(node, struct btrfs_root, rb_node);
801 read_unlock(&fs_info->global_root_lock);
802
803 return root;
804}
805
806static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
807{
808 struct btrfs_block_group *block_group;
809 u64 ret;
810
811 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
812 return 0;
813
814 if (bytenr)
815 block_group = btrfs_lookup_block_group(fs_info, bytenr);
816 else
817 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
818 ASSERT(block_group);
819 if (!block_group)
820 return 0;
821 ret = block_group->global_root_id;
822 btrfs_put_block_group(block_group);
823
824 return ret;
825}
826
827struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
828{
829 struct btrfs_key key = {
830 .objectid = BTRFS_CSUM_TREE_OBJECTID,
831 .type = BTRFS_ROOT_ITEM_KEY,
832 .offset = btrfs_global_root_id(fs_info, bytenr),
833 };
834
835 return btrfs_global_root(fs_info, &key);
836}
837
838struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
839{
840 struct btrfs_key key = {
841 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
842 .type = BTRFS_ROOT_ITEM_KEY,
843 .offset = btrfs_global_root_id(fs_info, bytenr),
844 };
845
846 return btrfs_global_root(fs_info, &key);
847}
848
849struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
850 u64 objectid)
851{
852 struct btrfs_fs_info *fs_info = trans->fs_info;
853 struct extent_buffer *leaf;
854 struct btrfs_root *tree_root = fs_info->tree_root;
855 struct btrfs_root *root;
856 struct btrfs_key key;
857 unsigned int nofs_flag;
858 int ret = 0;
859
860 /*
861 * We're holding a transaction handle, so use a NOFS memory allocation
862 * context to avoid deadlock if reclaim happens.
863 */
864 nofs_flag = memalloc_nofs_save();
865 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
866 memalloc_nofs_restore(nofs_flag);
867 if (!root)
868 return ERR_PTR(-ENOMEM);
869
870 root->root_key.objectid = objectid;
871 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
872 root->root_key.offset = 0;
873
874 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
875 0, BTRFS_NESTING_NORMAL);
876 if (IS_ERR(leaf)) {
877 ret = PTR_ERR(leaf);
878 leaf = NULL;
879 goto fail;
880 }
881
882 root->node = leaf;
883 btrfs_mark_buffer_dirty(trans, leaf);
884
885 root->commit_root = btrfs_root_node(root);
886 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
887
888 btrfs_set_root_flags(&root->root_item, 0);
889 btrfs_set_root_limit(&root->root_item, 0);
890 btrfs_set_root_bytenr(&root->root_item, leaf->start);
891 btrfs_set_root_generation(&root->root_item, trans->transid);
892 btrfs_set_root_level(&root->root_item, 0);
893 btrfs_set_root_refs(&root->root_item, 1);
894 btrfs_set_root_used(&root->root_item, leaf->len);
895 btrfs_set_root_last_snapshot(&root->root_item, 0);
896 btrfs_set_root_dirid(&root->root_item, 0);
897 if (is_fstree(objectid))
898 generate_random_guid(root->root_item.uuid);
899 else
900 export_guid(root->root_item.uuid, &guid_null);
901 btrfs_set_root_drop_level(&root->root_item, 0);
902
903 btrfs_tree_unlock(leaf);
904
905 key.objectid = objectid;
906 key.type = BTRFS_ROOT_ITEM_KEY;
907 key.offset = 0;
908 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
909 if (ret)
910 goto fail;
911
912 return root;
913
914fail:
915 btrfs_put_root(root);
916
917 return ERR_PTR(ret);
918}
919
920static struct btrfs_root *alloc_log_tree(struct btrfs_fs_info *fs_info)
921{
922 struct btrfs_root *root;
923
924 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
925 if (!root)
926 return ERR_PTR(-ENOMEM);
927
928 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
929 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
930 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
931
932 return root;
933}
934
935int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
936 struct btrfs_root *root)
937{
938 struct extent_buffer *leaf;
939
940 /*
941 * DON'T set SHAREABLE bit for log trees.
942 *
943 * Log trees are not exposed to user space thus can't be snapshotted,
944 * and they go away before a real commit is actually done.
945 *
946 * They do store pointers to file data extents, and those reference
947 * counts still get updated (along with back refs to the log tree).
948 */
949
950 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
951 NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
952 if (IS_ERR(leaf))
953 return PTR_ERR(leaf);
954
955 root->node = leaf;
956
957 btrfs_mark_buffer_dirty(trans, root->node);
958 btrfs_tree_unlock(root->node);
959
960 return 0;
961}
962
963int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
964 struct btrfs_fs_info *fs_info)
965{
966 struct btrfs_root *log_root;
967
968 log_root = alloc_log_tree(fs_info);
969 if (IS_ERR(log_root))
970 return PTR_ERR(log_root);
971
972 if (!btrfs_is_zoned(fs_info)) {
973 int ret = btrfs_alloc_log_tree_node(trans, log_root);
974
975 if (ret) {
976 btrfs_put_root(log_root);
977 return ret;
978 }
979 }
980
981 WARN_ON(fs_info->log_root_tree);
982 fs_info->log_root_tree = log_root;
983 return 0;
984}
985
986int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
987 struct btrfs_root *root)
988{
989 struct btrfs_fs_info *fs_info = root->fs_info;
990 struct btrfs_root *log_root;
991 struct btrfs_inode_item *inode_item;
992 int ret;
993
994 log_root = alloc_log_tree(fs_info);
995 if (IS_ERR(log_root))
996 return PTR_ERR(log_root);
997
998 ret = btrfs_alloc_log_tree_node(trans, log_root);
999 if (ret) {
1000 btrfs_put_root(log_root);
1001 return ret;
1002 }
1003
1004 btrfs_set_root_last_trans(log_root, trans->transid);
1005 log_root->root_key.offset = btrfs_root_id(root);
1006
1007 inode_item = &log_root->root_item.inode;
1008 btrfs_set_stack_inode_generation(inode_item, 1);
1009 btrfs_set_stack_inode_size(inode_item, 3);
1010 btrfs_set_stack_inode_nlink(inode_item, 1);
1011 btrfs_set_stack_inode_nbytes(inode_item,
1012 fs_info->nodesize);
1013 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1014
1015 btrfs_set_root_node(&log_root->root_item, log_root->node);
1016
1017 WARN_ON(root->log_root);
1018 root->log_root = log_root;
1019 btrfs_set_root_log_transid(root, 0);
1020 root->log_transid_committed = -1;
1021 btrfs_set_root_last_log_commit(root, 0);
1022 return 0;
1023}
1024
1025static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1026 struct btrfs_path *path,
1027 const struct btrfs_key *key)
1028{
1029 struct btrfs_root *root;
1030 struct btrfs_tree_parent_check check = { 0 };
1031 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1032 u64 generation;
1033 int ret;
1034 int level;
1035
1036 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1037 if (!root)
1038 return ERR_PTR(-ENOMEM);
1039
1040 ret = btrfs_find_root(tree_root, key, path,
1041 &root->root_item, &root->root_key);
1042 if (ret) {
1043 if (ret > 0)
1044 ret = -ENOENT;
1045 goto fail;
1046 }
1047
1048 generation = btrfs_root_generation(&root->root_item);
1049 level = btrfs_root_level(&root->root_item);
1050 check.level = level;
1051 check.transid = generation;
1052 check.owner_root = key->objectid;
1053 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1054 &check);
1055 if (IS_ERR(root->node)) {
1056 ret = PTR_ERR(root->node);
1057 root->node = NULL;
1058 goto fail;
1059 }
1060 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1061 ret = -EIO;
1062 goto fail;
1063 }
1064
1065 /*
1066 * For real fs, and not log/reloc trees, root owner must
1067 * match its root node owner
1068 */
1069 if (!btrfs_is_testing(fs_info) &&
1070 btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1071 btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID &&
1072 btrfs_root_id(root) != btrfs_header_owner(root->node)) {
1073 btrfs_crit(fs_info,
1074"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1075 btrfs_root_id(root), root->node->start,
1076 btrfs_header_owner(root->node),
1077 btrfs_root_id(root));
1078 ret = -EUCLEAN;
1079 goto fail;
1080 }
1081 root->commit_root = btrfs_root_node(root);
1082 return root;
1083fail:
1084 btrfs_put_root(root);
1085 return ERR_PTR(ret);
1086}
1087
1088struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1089 const struct btrfs_key *key)
1090{
1091 struct btrfs_root *root;
1092 struct btrfs_path *path;
1093
1094 path = btrfs_alloc_path();
1095 if (!path)
1096 return ERR_PTR(-ENOMEM);
1097 root = read_tree_root_path(tree_root, path, key);
1098 btrfs_free_path(path);
1099
1100 return root;
1101}
1102
1103/*
1104 * Initialize subvolume root in-memory structure
1105 *
1106 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1107 */
1108static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1109{
1110 int ret;
1111
1112 btrfs_drew_lock_init(&root->snapshot_lock);
1113
1114 if (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1115 !btrfs_is_data_reloc_root(root) &&
1116 is_fstree(btrfs_root_id(root))) {
1117 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1118 btrfs_check_and_init_root_item(&root->root_item);
1119 }
1120
1121 /*
1122 * Don't assign anonymous block device to roots that are not exposed to
1123 * userspace, the id pool is limited to 1M
1124 */
1125 if (is_fstree(btrfs_root_id(root)) &&
1126 btrfs_root_refs(&root->root_item) > 0) {
1127 if (!anon_dev) {
1128 ret = get_anon_bdev(&root->anon_dev);
1129 if (ret)
1130 goto fail;
1131 } else {
1132 root->anon_dev = anon_dev;
1133 }
1134 }
1135
1136 mutex_lock(&root->objectid_mutex);
1137 ret = btrfs_init_root_free_objectid(root);
1138 if (ret) {
1139 mutex_unlock(&root->objectid_mutex);
1140 goto fail;
1141 }
1142
1143 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1144
1145 mutex_unlock(&root->objectid_mutex);
1146
1147 return 0;
1148fail:
1149 /* The caller is responsible to call btrfs_free_fs_root */
1150 return ret;
1151}
1152
1153static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1154 u64 root_id)
1155{
1156 struct btrfs_root *root;
1157
1158 spin_lock(&fs_info->fs_roots_radix_lock);
1159 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1160 (unsigned long)root_id);
1161 root = btrfs_grab_root(root);
1162 spin_unlock(&fs_info->fs_roots_radix_lock);
1163 return root;
1164}
1165
1166static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1167 u64 objectid)
1168{
1169 struct btrfs_key key = {
1170 .objectid = objectid,
1171 .type = BTRFS_ROOT_ITEM_KEY,
1172 .offset = 0,
1173 };
1174
1175 switch (objectid) {
1176 case BTRFS_ROOT_TREE_OBJECTID:
1177 return btrfs_grab_root(fs_info->tree_root);
1178 case BTRFS_EXTENT_TREE_OBJECTID:
1179 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1180 case BTRFS_CHUNK_TREE_OBJECTID:
1181 return btrfs_grab_root(fs_info->chunk_root);
1182 case BTRFS_DEV_TREE_OBJECTID:
1183 return btrfs_grab_root(fs_info->dev_root);
1184 case BTRFS_CSUM_TREE_OBJECTID:
1185 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1186 case BTRFS_QUOTA_TREE_OBJECTID:
1187 return btrfs_grab_root(fs_info->quota_root);
1188 case BTRFS_UUID_TREE_OBJECTID:
1189 return btrfs_grab_root(fs_info->uuid_root);
1190 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1191 return btrfs_grab_root(fs_info->block_group_root);
1192 case BTRFS_FREE_SPACE_TREE_OBJECTID:
1193 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1194 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1195 return btrfs_grab_root(fs_info->stripe_root);
1196 default:
1197 return NULL;
1198 }
1199}
1200
1201int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1202 struct btrfs_root *root)
1203{
1204 int ret;
1205
1206 ret = radix_tree_preload(GFP_NOFS);
1207 if (ret)
1208 return ret;
1209
1210 spin_lock(&fs_info->fs_roots_radix_lock);
1211 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1212 (unsigned long)btrfs_root_id(root),
1213 root);
1214 if (ret == 0) {
1215 btrfs_grab_root(root);
1216 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1217 }
1218 spin_unlock(&fs_info->fs_roots_radix_lock);
1219 radix_tree_preload_end();
1220
1221 return ret;
1222}
1223
1224void btrfs_check_leaked_roots(const struct btrfs_fs_info *fs_info)
1225{
1226#ifdef CONFIG_BTRFS_DEBUG
1227 struct btrfs_root *root;
1228
1229 while (!list_empty(&fs_info->allocated_roots)) {
1230 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1231
1232 root = list_first_entry(&fs_info->allocated_roots,
1233 struct btrfs_root, leak_list);
1234 btrfs_err(fs_info, "leaked root %s refcount %d",
1235 btrfs_root_name(&root->root_key, buf),
1236 refcount_read(&root->refs));
1237 WARN_ON_ONCE(1);
1238 while (refcount_read(&root->refs) > 1)
1239 btrfs_put_root(root);
1240 btrfs_put_root(root);
1241 }
1242#endif
1243}
1244
1245static void free_global_roots(struct btrfs_fs_info *fs_info)
1246{
1247 struct btrfs_root *root;
1248 struct rb_node *node;
1249
1250 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1251 root = rb_entry(node, struct btrfs_root, rb_node);
1252 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1253 btrfs_put_root(root);
1254 }
1255}
1256
1257void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1258{
1259 struct percpu_counter *em_counter = &fs_info->evictable_extent_maps;
1260
1261 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1262 percpu_counter_destroy(&fs_info->delalloc_bytes);
1263 percpu_counter_destroy(&fs_info->ordered_bytes);
1264 if (percpu_counter_initialized(em_counter))
1265 ASSERT(percpu_counter_sum_positive(em_counter) == 0);
1266 percpu_counter_destroy(em_counter);
1267 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1268 btrfs_free_csum_hash(fs_info);
1269 btrfs_free_stripe_hash_table(fs_info);
1270 btrfs_free_ref_cache(fs_info);
1271 kfree(fs_info->balance_ctl);
1272 kfree(fs_info->delayed_root);
1273 free_global_roots(fs_info);
1274 btrfs_put_root(fs_info->tree_root);
1275 btrfs_put_root(fs_info->chunk_root);
1276 btrfs_put_root(fs_info->dev_root);
1277 btrfs_put_root(fs_info->quota_root);
1278 btrfs_put_root(fs_info->uuid_root);
1279 btrfs_put_root(fs_info->fs_root);
1280 btrfs_put_root(fs_info->data_reloc_root);
1281 btrfs_put_root(fs_info->block_group_root);
1282 btrfs_put_root(fs_info->stripe_root);
1283 btrfs_check_leaked_roots(fs_info);
1284 btrfs_extent_buffer_leak_debug_check(fs_info);
1285 kfree(fs_info->super_copy);
1286 kfree(fs_info->super_for_commit);
1287 kvfree(fs_info);
1288}
1289
1290
1291/*
1292 * Get an in-memory reference of a root structure.
1293 *
1294 * For essential trees like root/extent tree, we grab it from fs_info directly.
1295 * For subvolume trees, we check the cached filesystem roots first. If not
1296 * found, then read it from disk and add it to cached fs roots.
1297 *
1298 * Caller should release the root by calling btrfs_put_root() after the usage.
1299 *
1300 * NOTE: Reloc and log trees can't be read by this function as they share the
1301 * same root objectid.
1302 *
1303 * @objectid: root id
1304 * @anon_dev: preallocated anonymous block device number for new roots,
1305 * pass NULL for a new allocation.
1306 * @check_ref: whether to check root item references, If true, return -ENOENT
1307 * for orphan roots
1308 */
1309static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1310 u64 objectid, dev_t *anon_dev,
1311 bool check_ref)
1312{
1313 struct btrfs_root *root;
1314 struct btrfs_path *path;
1315 struct btrfs_key key;
1316 int ret;
1317
1318 root = btrfs_get_global_root(fs_info, objectid);
1319 if (root)
1320 return root;
1321
1322 /*
1323 * If we're called for non-subvolume trees, and above function didn't
1324 * find one, do not try to read it from disk.
1325 *
1326 * This is namely for free-space-tree and quota tree, which can change
1327 * at runtime and should only be grabbed from fs_info.
1328 */
1329 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1330 return ERR_PTR(-ENOENT);
1331again:
1332 root = btrfs_lookup_fs_root(fs_info, objectid);
1333 if (root) {
1334 /*
1335 * Some other caller may have read out the newly inserted
1336 * subvolume already (for things like backref walk etc). Not
1337 * that common but still possible. In that case, we just need
1338 * to free the anon_dev.
1339 */
1340 if (unlikely(anon_dev && *anon_dev)) {
1341 free_anon_bdev(*anon_dev);
1342 *anon_dev = 0;
1343 }
1344
1345 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1346 btrfs_put_root(root);
1347 return ERR_PTR(-ENOENT);
1348 }
1349 return root;
1350 }
1351
1352 key.objectid = objectid;
1353 key.type = BTRFS_ROOT_ITEM_KEY;
1354 key.offset = (u64)-1;
1355 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1356 if (IS_ERR(root))
1357 return root;
1358
1359 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1360 ret = -ENOENT;
1361 goto fail;
1362 }
1363
1364 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1365 if (ret)
1366 goto fail;
1367
1368 path = btrfs_alloc_path();
1369 if (!path) {
1370 ret = -ENOMEM;
1371 goto fail;
1372 }
1373 key.objectid = BTRFS_ORPHAN_OBJECTID;
1374 key.type = BTRFS_ORPHAN_ITEM_KEY;
1375 key.offset = objectid;
1376
1377 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1378 btrfs_free_path(path);
1379 if (ret < 0)
1380 goto fail;
1381 if (ret == 0)
1382 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1383
1384 ret = btrfs_insert_fs_root(fs_info, root);
1385 if (ret) {
1386 if (ret == -EEXIST) {
1387 btrfs_put_root(root);
1388 goto again;
1389 }
1390 goto fail;
1391 }
1392 return root;
1393fail:
1394 /*
1395 * If our caller provided us an anonymous device, then it's his
1396 * responsibility to free it in case we fail. So we have to set our
1397 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1398 * and once again by our caller.
1399 */
1400 if (anon_dev && *anon_dev)
1401 root->anon_dev = 0;
1402 btrfs_put_root(root);
1403 return ERR_PTR(ret);
1404}
1405
1406/*
1407 * Get in-memory reference of a root structure
1408 *
1409 * @objectid: tree objectid
1410 * @check_ref: if set, verify that the tree exists and the item has at least
1411 * one reference
1412 */
1413struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1414 u64 objectid, bool check_ref)
1415{
1416 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1417}
1418
1419/*
1420 * Get in-memory reference of a root structure, created as new, optionally pass
1421 * the anonymous block device id
1422 *
1423 * @objectid: tree objectid
1424 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1425 * parameter value if not NULL
1426 */
1427struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1428 u64 objectid, dev_t *anon_dev)
1429{
1430 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1431}
1432
1433/*
1434 * Return a root for the given objectid.
1435 *
1436 * @fs_info: the fs_info
1437 * @objectid: the objectid we need to lookup
1438 *
1439 * This is exclusively used for backref walking, and exists specifically because
1440 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1441 * creation time, which means we may have to read the tree_root in order to look
1442 * up a fs root that is not in memory. If the root is not in memory we will
1443 * read the tree root commit root and look up the fs root from there. This is a
1444 * temporary root, it will not be inserted into the radix tree as it doesn't
1445 * have the most uptodate information, it'll simply be discarded once the
1446 * backref code is finished using the root.
1447 */
1448struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1449 struct btrfs_path *path,
1450 u64 objectid)
1451{
1452 struct btrfs_root *root;
1453 struct btrfs_key key;
1454
1455 ASSERT(path->search_commit_root && path->skip_locking);
1456
1457 /*
1458 * This can return -ENOENT if we ask for a root that doesn't exist, but
1459 * since this is called via the backref walking code we won't be looking
1460 * up a root that doesn't exist, unless there's corruption. So if root
1461 * != NULL just return it.
1462 */
1463 root = btrfs_get_global_root(fs_info, objectid);
1464 if (root)
1465 return root;
1466
1467 root = btrfs_lookup_fs_root(fs_info, objectid);
1468 if (root)
1469 return root;
1470
1471 key.objectid = objectid;
1472 key.type = BTRFS_ROOT_ITEM_KEY;
1473 key.offset = (u64)-1;
1474 root = read_tree_root_path(fs_info->tree_root, path, &key);
1475 btrfs_release_path(path);
1476
1477 return root;
1478}
1479
1480static int cleaner_kthread(void *arg)
1481{
1482 struct btrfs_fs_info *fs_info = arg;
1483 int again;
1484
1485 while (1) {
1486 again = 0;
1487
1488 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1489
1490 /* Make the cleaner go to sleep early. */
1491 if (btrfs_need_cleaner_sleep(fs_info))
1492 goto sleep;
1493
1494 /*
1495 * Do not do anything if we might cause open_ctree() to block
1496 * before we have finished mounting the filesystem.
1497 */
1498 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1499 goto sleep;
1500
1501 if (!mutex_trylock(&fs_info->cleaner_mutex))
1502 goto sleep;
1503
1504 /*
1505 * Avoid the problem that we change the status of the fs
1506 * during the above check and trylock.
1507 */
1508 if (btrfs_need_cleaner_sleep(fs_info)) {
1509 mutex_unlock(&fs_info->cleaner_mutex);
1510 goto sleep;
1511 }
1512
1513 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1514 btrfs_sysfs_feature_update(fs_info);
1515
1516 btrfs_run_delayed_iputs(fs_info);
1517
1518 again = btrfs_clean_one_deleted_snapshot(fs_info);
1519 mutex_unlock(&fs_info->cleaner_mutex);
1520
1521 /*
1522 * The defragger has dealt with the R/O remount and umount,
1523 * needn't do anything special here.
1524 */
1525 btrfs_run_defrag_inodes(fs_info);
1526
1527 /*
1528 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1529 * with relocation (btrfs_relocate_chunk) and relocation
1530 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1531 * after acquiring fs_info->reclaim_bgs_lock. So we
1532 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1533 * unused block groups.
1534 */
1535 btrfs_delete_unused_bgs(fs_info);
1536
1537 /*
1538 * Reclaim block groups in the reclaim_bgs list after we deleted
1539 * all unused block_groups. This possibly gives us some more free
1540 * space.
1541 */
1542 btrfs_reclaim_bgs(fs_info);
1543sleep:
1544 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1545 if (kthread_should_park())
1546 kthread_parkme();
1547 if (kthread_should_stop())
1548 return 0;
1549 if (!again) {
1550 set_current_state(TASK_INTERRUPTIBLE);
1551 schedule();
1552 __set_current_state(TASK_RUNNING);
1553 }
1554 }
1555}
1556
1557static int transaction_kthread(void *arg)
1558{
1559 struct btrfs_root *root = arg;
1560 struct btrfs_fs_info *fs_info = root->fs_info;
1561 struct btrfs_trans_handle *trans;
1562 struct btrfs_transaction *cur;
1563 u64 transid;
1564 time64_t delta;
1565 unsigned long delay;
1566 bool cannot_commit;
1567
1568 do {
1569 cannot_commit = false;
1570 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1571 mutex_lock(&fs_info->transaction_kthread_mutex);
1572
1573 spin_lock(&fs_info->trans_lock);
1574 cur = fs_info->running_transaction;
1575 if (!cur) {
1576 spin_unlock(&fs_info->trans_lock);
1577 goto sleep;
1578 }
1579
1580 delta = ktime_get_seconds() - cur->start_time;
1581 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1582 cur->state < TRANS_STATE_COMMIT_PREP &&
1583 delta < fs_info->commit_interval) {
1584 spin_unlock(&fs_info->trans_lock);
1585 delay -= msecs_to_jiffies((delta - 1) * 1000);
1586 delay = min(delay,
1587 msecs_to_jiffies(fs_info->commit_interval * 1000));
1588 goto sleep;
1589 }
1590 transid = cur->transid;
1591 spin_unlock(&fs_info->trans_lock);
1592
1593 /* If the file system is aborted, this will always fail. */
1594 trans = btrfs_attach_transaction(root);
1595 if (IS_ERR(trans)) {
1596 if (PTR_ERR(trans) != -ENOENT)
1597 cannot_commit = true;
1598 goto sleep;
1599 }
1600 if (transid == trans->transid) {
1601 btrfs_commit_transaction(trans);
1602 } else {
1603 btrfs_end_transaction(trans);
1604 }
1605sleep:
1606 wake_up_process(fs_info->cleaner_kthread);
1607 mutex_unlock(&fs_info->transaction_kthread_mutex);
1608
1609 if (BTRFS_FS_ERROR(fs_info))
1610 btrfs_cleanup_transaction(fs_info);
1611 if (!kthread_should_stop() &&
1612 (!btrfs_transaction_blocked(fs_info) ||
1613 cannot_commit))
1614 schedule_timeout_interruptible(delay);
1615 } while (!kthread_should_stop());
1616 return 0;
1617}
1618
1619/*
1620 * This will find the highest generation in the array of root backups. The
1621 * index of the highest array is returned, or -EINVAL if we can't find
1622 * anything.
1623 *
1624 * We check to make sure the array is valid by comparing the
1625 * generation of the latest root in the array with the generation
1626 * in the super block. If they don't match we pitch it.
1627 */
1628static int find_newest_super_backup(struct btrfs_fs_info *info)
1629{
1630 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1631 u64 cur;
1632 struct btrfs_root_backup *root_backup;
1633 int i;
1634
1635 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1636 root_backup = info->super_copy->super_roots + i;
1637 cur = btrfs_backup_tree_root_gen(root_backup);
1638 if (cur == newest_gen)
1639 return i;
1640 }
1641
1642 return -EINVAL;
1643}
1644
1645/*
1646 * copy all the root pointers into the super backup array.
1647 * this will bump the backup pointer by one when it is
1648 * done
1649 */
1650static void backup_super_roots(struct btrfs_fs_info *info)
1651{
1652 const int next_backup = info->backup_root_index;
1653 struct btrfs_root_backup *root_backup;
1654
1655 root_backup = info->super_for_commit->super_roots + next_backup;
1656
1657 /*
1658 * make sure all of our padding and empty slots get zero filled
1659 * regardless of which ones we use today
1660 */
1661 memset(root_backup, 0, sizeof(*root_backup));
1662
1663 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1664
1665 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1666 btrfs_set_backup_tree_root_gen(root_backup,
1667 btrfs_header_generation(info->tree_root->node));
1668
1669 btrfs_set_backup_tree_root_level(root_backup,
1670 btrfs_header_level(info->tree_root->node));
1671
1672 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1673 btrfs_set_backup_chunk_root_gen(root_backup,
1674 btrfs_header_generation(info->chunk_root->node));
1675 btrfs_set_backup_chunk_root_level(root_backup,
1676 btrfs_header_level(info->chunk_root->node));
1677
1678 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1679 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1680 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1681
1682 btrfs_set_backup_extent_root(root_backup,
1683 extent_root->node->start);
1684 btrfs_set_backup_extent_root_gen(root_backup,
1685 btrfs_header_generation(extent_root->node));
1686 btrfs_set_backup_extent_root_level(root_backup,
1687 btrfs_header_level(extent_root->node));
1688
1689 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1690 btrfs_set_backup_csum_root_gen(root_backup,
1691 btrfs_header_generation(csum_root->node));
1692 btrfs_set_backup_csum_root_level(root_backup,
1693 btrfs_header_level(csum_root->node));
1694 }
1695
1696 /*
1697 * we might commit during log recovery, which happens before we set
1698 * the fs_root. Make sure it is valid before we fill it in.
1699 */
1700 if (info->fs_root && info->fs_root->node) {
1701 btrfs_set_backup_fs_root(root_backup,
1702 info->fs_root->node->start);
1703 btrfs_set_backup_fs_root_gen(root_backup,
1704 btrfs_header_generation(info->fs_root->node));
1705 btrfs_set_backup_fs_root_level(root_backup,
1706 btrfs_header_level(info->fs_root->node));
1707 }
1708
1709 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1710 btrfs_set_backup_dev_root_gen(root_backup,
1711 btrfs_header_generation(info->dev_root->node));
1712 btrfs_set_backup_dev_root_level(root_backup,
1713 btrfs_header_level(info->dev_root->node));
1714
1715 btrfs_set_backup_total_bytes(root_backup,
1716 btrfs_super_total_bytes(info->super_copy));
1717 btrfs_set_backup_bytes_used(root_backup,
1718 btrfs_super_bytes_used(info->super_copy));
1719 btrfs_set_backup_num_devices(root_backup,
1720 btrfs_super_num_devices(info->super_copy));
1721
1722 /*
1723 * if we don't copy this out to the super_copy, it won't get remembered
1724 * for the next commit
1725 */
1726 memcpy(&info->super_copy->super_roots,
1727 &info->super_for_commit->super_roots,
1728 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1729}
1730
1731/*
1732 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1733 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1734 *
1735 * @fs_info: filesystem whose backup roots need to be read
1736 * @priority: priority of backup root required
1737 *
1738 * Returns backup root index on success and -EINVAL otherwise.
1739 */
1740static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1741{
1742 int backup_index = find_newest_super_backup(fs_info);
1743 struct btrfs_super_block *super = fs_info->super_copy;
1744 struct btrfs_root_backup *root_backup;
1745
1746 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1747 if (priority == 0)
1748 return backup_index;
1749
1750 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1751 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1752 } else {
1753 return -EINVAL;
1754 }
1755
1756 root_backup = super->super_roots + backup_index;
1757
1758 btrfs_set_super_generation(super,
1759 btrfs_backup_tree_root_gen(root_backup));
1760 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1761 btrfs_set_super_root_level(super,
1762 btrfs_backup_tree_root_level(root_backup));
1763 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1764
1765 /*
1766 * Fixme: the total bytes and num_devices need to match or we should
1767 * need a fsck
1768 */
1769 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1770 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1771
1772 return backup_index;
1773}
1774
1775/* helper to cleanup workers */
1776static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1777{
1778 btrfs_destroy_workqueue(fs_info->fixup_workers);
1779 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1780 btrfs_destroy_workqueue(fs_info->workers);
1781 if (fs_info->endio_workers)
1782 destroy_workqueue(fs_info->endio_workers);
1783 if (fs_info->rmw_workers)
1784 destroy_workqueue(fs_info->rmw_workers);
1785 if (fs_info->compressed_write_workers)
1786 destroy_workqueue(fs_info->compressed_write_workers);
1787 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1788 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1789 btrfs_destroy_workqueue(fs_info->delayed_workers);
1790 btrfs_destroy_workqueue(fs_info->caching_workers);
1791 btrfs_destroy_workqueue(fs_info->flush_workers);
1792 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1793 if (fs_info->discard_ctl.discard_workers)
1794 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1795 /*
1796 * Now that all other work queues are destroyed, we can safely destroy
1797 * the queues used for metadata I/O, since tasks from those other work
1798 * queues can do metadata I/O operations.
1799 */
1800 if (fs_info->endio_meta_workers)
1801 destroy_workqueue(fs_info->endio_meta_workers);
1802}
1803
1804static void free_root_extent_buffers(struct btrfs_root *root)
1805{
1806 if (root) {
1807 free_extent_buffer(root->node);
1808 free_extent_buffer(root->commit_root);
1809 root->node = NULL;
1810 root->commit_root = NULL;
1811 }
1812}
1813
1814static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1815{
1816 struct btrfs_root *root, *tmp;
1817
1818 rbtree_postorder_for_each_entry_safe(root, tmp,
1819 &fs_info->global_root_tree,
1820 rb_node)
1821 free_root_extent_buffers(root);
1822}
1823
1824/* helper to cleanup tree roots */
1825static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1826{
1827 free_root_extent_buffers(info->tree_root);
1828
1829 free_global_root_pointers(info);
1830 free_root_extent_buffers(info->dev_root);
1831 free_root_extent_buffers(info->quota_root);
1832 free_root_extent_buffers(info->uuid_root);
1833 free_root_extent_buffers(info->fs_root);
1834 free_root_extent_buffers(info->data_reloc_root);
1835 free_root_extent_buffers(info->block_group_root);
1836 free_root_extent_buffers(info->stripe_root);
1837 if (free_chunk_root)
1838 free_root_extent_buffers(info->chunk_root);
1839}
1840
1841void btrfs_put_root(struct btrfs_root *root)
1842{
1843 if (!root)
1844 return;
1845
1846 if (refcount_dec_and_test(&root->refs)) {
1847 if (WARN_ON(!xa_empty(&root->inodes)))
1848 xa_destroy(&root->inodes);
1849 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1850 if (root->anon_dev)
1851 free_anon_bdev(root->anon_dev);
1852 free_root_extent_buffers(root);
1853#ifdef CONFIG_BTRFS_DEBUG
1854 spin_lock(&root->fs_info->fs_roots_radix_lock);
1855 list_del_init(&root->leak_list);
1856 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1857#endif
1858 kfree(root);
1859 }
1860}
1861
1862void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1863{
1864 int ret;
1865 struct btrfs_root *gang[8];
1866 int i;
1867
1868 while (!list_empty(&fs_info->dead_roots)) {
1869 gang[0] = list_entry(fs_info->dead_roots.next,
1870 struct btrfs_root, root_list);
1871 list_del(&gang[0]->root_list);
1872
1873 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1874 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1875 btrfs_put_root(gang[0]);
1876 }
1877
1878 while (1) {
1879 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1880 (void **)gang, 0,
1881 ARRAY_SIZE(gang));
1882 if (!ret)
1883 break;
1884 for (i = 0; i < ret; i++)
1885 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1886 }
1887}
1888
1889static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1890{
1891 mutex_init(&fs_info->scrub_lock);
1892 atomic_set(&fs_info->scrubs_running, 0);
1893 atomic_set(&fs_info->scrub_pause_req, 0);
1894 atomic_set(&fs_info->scrubs_paused, 0);
1895 atomic_set(&fs_info->scrub_cancel_req, 0);
1896 init_waitqueue_head(&fs_info->scrub_pause_wait);
1897 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1898}
1899
1900static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1901{
1902 spin_lock_init(&fs_info->balance_lock);
1903 mutex_init(&fs_info->balance_mutex);
1904 atomic_set(&fs_info->balance_pause_req, 0);
1905 atomic_set(&fs_info->balance_cancel_req, 0);
1906 fs_info->balance_ctl = NULL;
1907 init_waitqueue_head(&fs_info->balance_wait_q);
1908 atomic_set(&fs_info->reloc_cancel_req, 0);
1909}
1910
1911static int btrfs_init_btree_inode(struct super_block *sb)
1912{
1913 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1914 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1915 fs_info->tree_root);
1916 struct inode *inode;
1917
1918 inode = new_inode(sb);
1919 if (!inode)
1920 return -ENOMEM;
1921
1922 btrfs_set_inode_number(BTRFS_I(inode), BTRFS_BTREE_INODE_OBJECTID);
1923 set_nlink(inode, 1);
1924 /*
1925 * we set the i_size on the btree inode to the max possible int.
1926 * the real end of the address space is determined by all of
1927 * the devices in the system
1928 */
1929 inode->i_size = OFFSET_MAX;
1930 inode->i_mapping->a_ops = &btree_aops;
1931 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1932
1933 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1934 IO_TREE_BTREE_INODE_IO);
1935 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1936
1937 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1938 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1939 __insert_inode_hash(inode, hash);
1940 fs_info->btree_inode = inode;
1941
1942 return 0;
1943}
1944
1945static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1946{
1947 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1948 init_rwsem(&fs_info->dev_replace.rwsem);
1949 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1950}
1951
1952static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1953{
1954 spin_lock_init(&fs_info->qgroup_lock);
1955 mutex_init(&fs_info->qgroup_ioctl_lock);
1956 fs_info->qgroup_tree = RB_ROOT;
1957 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1958 fs_info->qgroup_seq = 1;
1959 fs_info->qgroup_ulist = NULL;
1960 fs_info->qgroup_rescan_running = false;
1961 fs_info->qgroup_drop_subtree_thres = BTRFS_QGROUP_DROP_SUBTREE_THRES_DEFAULT;
1962 mutex_init(&fs_info->qgroup_rescan_lock);
1963}
1964
1965static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1966{
1967 u32 max_active = fs_info->thread_pool_size;
1968 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1969 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1970
1971 fs_info->workers =
1972 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1973
1974 fs_info->delalloc_workers =
1975 btrfs_alloc_workqueue(fs_info, "delalloc",
1976 flags, max_active, 2);
1977
1978 fs_info->flush_workers =
1979 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1980 flags, max_active, 0);
1981
1982 fs_info->caching_workers =
1983 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1984
1985 fs_info->fixup_workers =
1986 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1987
1988 fs_info->endio_workers =
1989 alloc_workqueue("btrfs-endio", flags, max_active);
1990 fs_info->endio_meta_workers =
1991 alloc_workqueue("btrfs-endio-meta", flags, max_active);
1992 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1993 fs_info->endio_write_workers =
1994 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1995 max_active, 2);
1996 fs_info->compressed_write_workers =
1997 alloc_workqueue("btrfs-compressed-write", flags, max_active);
1998 fs_info->endio_freespace_worker =
1999 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2000 max_active, 0);
2001 fs_info->delayed_workers =
2002 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2003 max_active, 0);
2004 fs_info->qgroup_rescan_workers =
2005 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2006 ordered_flags);
2007 fs_info->discard_ctl.discard_workers =
2008 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2009
2010 if (!(fs_info->workers &&
2011 fs_info->delalloc_workers && fs_info->flush_workers &&
2012 fs_info->endio_workers && fs_info->endio_meta_workers &&
2013 fs_info->compressed_write_workers &&
2014 fs_info->endio_write_workers &&
2015 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2016 fs_info->caching_workers && fs_info->fixup_workers &&
2017 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2018 fs_info->discard_ctl.discard_workers)) {
2019 return -ENOMEM;
2020 }
2021
2022 return 0;
2023}
2024
2025static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2026{
2027 struct crypto_shash *csum_shash;
2028 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2029
2030 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2031
2032 if (IS_ERR(csum_shash)) {
2033 btrfs_err(fs_info, "error allocating %s hash for checksum",
2034 csum_driver);
2035 return PTR_ERR(csum_shash);
2036 }
2037
2038 fs_info->csum_shash = csum_shash;
2039
2040 /*
2041 * Check if the checksum implementation is a fast accelerated one.
2042 * As-is this is a bit of a hack and should be replaced once the csum
2043 * implementations provide that information themselves.
2044 */
2045 switch (csum_type) {
2046 case BTRFS_CSUM_TYPE_CRC32:
2047 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2048 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2049 break;
2050 case BTRFS_CSUM_TYPE_XXHASH:
2051 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2052 break;
2053 default:
2054 break;
2055 }
2056
2057 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2058 btrfs_super_csum_name(csum_type),
2059 crypto_shash_driver_name(csum_shash));
2060 return 0;
2061}
2062
2063static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2064 struct btrfs_fs_devices *fs_devices)
2065{
2066 int ret;
2067 struct btrfs_tree_parent_check check = { 0 };
2068 struct btrfs_root *log_tree_root;
2069 struct btrfs_super_block *disk_super = fs_info->super_copy;
2070 u64 bytenr = btrfs_super_log_root(disk_super);
2071 int level = btrfs_super_log_root_level(disk_super);
2072
2073 if (fs_devices->rw_devices == 0) {
2074 btrfs_warn(fs_info, "log replay required on RO media");
2075 return -EIO;
2076 }
2077
2078 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2079 GFP_KERNEL);
2080 if (!log_tree_root)
2081 return -ENOMEM;
2082
2083 check.level = level;
2084 check.transid = fs_info->generation + 1;
2085 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2086 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2087 if (IS_ERR(log_tree_root->node)) {
2088 btrfs_warn(fs_info, "failed to read log tree");
2089 ret = PTR_ERR(log_tree_root->node);
2090 log_tree_root->node = NULL;
2091 btrfs_put_root(log_tree_root);
2092 return ret;
2093 }
2094 if (!extent_buffer_uptodate(log_tree_root->node)) {
2095 btrfs_err(fs_info, "failed to read log tree");
2096 btrfs_put_root(log_tree_root);
2097 return -EIO;
2098 }
2099
2100 /* returns with log_tree_root freed on success */
2101 ret = btrfs_recover_log_trees(log_tree_root);
2102 if (ret) {
2103 btrfs_handle_fs_error(fs_info, ret,
2104 "Failed to recover log tree");
2105 btrfs_put_root(log_tree_root);
2106 return ret;
2107 }
2108
2109 if (sb_rdonly(fs_info->sb)) {
2110 ret = btrfs_commit_super(fs_info);
2111 if (ret)
2112 return ret;
2113 }
2114
2115 return 0;
2116}
2117
2118static int load_global_roots_objectid(struct btrfs_root *tree_root,
2119 struct btrfs_path *path, u64 objectid,
2120 const char *name)
2121{
2122 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2123 struct btrfs_root *root;
2124 u64 max_global_id = 0;
2125 int ret;
2126 struct btrfs_key key = {
2127 .objectid = objectid,
2128 .type = BTRFS_ROOT_ITEM_KEY,
2129 .offset = 0,
2130 };
2131 bool found = false;
2132
2133 /* If we have IGNOREDATACSUMS skip loading these roots. */
2134 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2135 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2136 set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2137 return 0;
2138 }
2139
2140 while (1) {
2141 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2142 if (ret < 0)
2143 break;
2144
2145 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2146 ret = btrfs_next_leaf(tree_root, path);
2147 if (ret) {
2148 if (ret > 0)
2149 ret = 0;
2150 break;
2151 }
2152 }
2153 ret = 0;
2154
2155 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2156 if (key.objectid != objectid)
2157 break;
2158 btrfs_release_path(path);
2159
2160 /*
2161 * Just worry about this for extent tree, it'll be the same for
2162 * everybody.
2163 */
2164 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2165 max_global_id = max(max_global_id, key.offset);
2166
2167 found = true;
2168 root = read_tree_root_path(tree_root, path, &key);
2169 if (IS_ERR(root)) {
2170 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2171 ret = PTR_ERR(root);
2172 break;
2173 }
2174 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2175 ret = btrfs_global_root_insert(root);
2176 if (ret) {
2177 btrfs_put_root(root);
2178 break;
2179 }
2180 key.offset++;
2181 }
2182 btrfs_release_path(path);
2183
2184 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2185 fs_info->nr_global_roots = max_global_id + 1;
2186
2187 if (!found || ret) {
2188 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2189 set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2190
2191 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2192 ret = ret ? ret : -ENOENT;
2193 else
2194 ret = 0;
2195 btrfs_err(fs_info, "failed to load root %s", name);
2196 }
2197 return ret;
2198}
2199
2200static int load_global_roots(struct btrfs_root *tree_root)
2201{
2202 struct btrfs_path *path;
2203 int ret = 0;
2204
2205 path = btrfs_alloc_path();
2206 if (!path)
2207 return -ENOMEM;
2208
2209 ret = load_global_roots_objectid(tree_root, path,
2210 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2211 if (ret)
2212 goto out;
2213 ret = load_global_roots_objectid(tree_root, path,
2214 BTRFS_CSUM_TREE_OBJECTID, "csum");
2215 if (ret)
2216 goto out;
2217 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2218 goto out;
2219 ret = load_global_roots_objectid(tree_root, path,
2220 BTRFS_FREE_SPACE_TREE_OBJECTID,
2221 "free space");
2222out:
2223 btrfs_free_path(path);
2224 return ret;
2225}
2226
2227static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2228{
2229 struct btrfs_root *tree_root = fs_info->tree_root;
2230 struct btrfs_root *root;
2231 struct btrfs_key location;
2232 int ret;
2233
2234 ASSERT(fs_info->tree_root);
2235
2236 ret = load_global_roots(tree_root);
2237 if (ret)
2238 return ret;
2239
2240 location.type = BTRFS_ROOT_ITEM_KEY;
2241 location.offset = 0;
2242
2243 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2244 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2245 root = btrfs_read_tree_root(tree_root, &location);
2246 if (IS_ERR(root)) {
2247 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2248 ret = PTR_ERR(root);
2249 goto out;
2250 }
2251 } else {
2252 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2253 fs_info->block_group_root = root;
2254 }
2255 }
2256
2257 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2258 root = btrfs_read_tree_root(tree_root, &location);
2259 if (IS_ERR(root)) {
2260 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2261 ret = PTR_ERR(root);
2262 goto out;
2263 }
2264 } else {
2265 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2266 fs_info->dev_root = root;
2267 }
2268 /* Initialize fs_info for all devices in any case */
2269 ret = btrfs_init_devices_late(fs_info);
2270 if (ret)
2271 goto out;
2272
2273 /*
2274 * This tree can share blocks with some other fs tree during relocation
2275 * and we need a proper setup by btrfs_get_fs_root
2276 */
2277 root = btrfs_get_fs_root(tree_root->fs_info,
2278 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2279 if (IS_ERR(root)) {
2280 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2281 ret = PTR_ERR(root);
2282 goto out;
2283 }
2284 } else {
2285 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2286 fs_info->data_reloc_root = root;
2287 }
2288
2289 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2290 root = btrfs_read_tree_root(tree_root, &location);
2291 if (!IS_ERR(root)) {
2292 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2293 fs_info->quota_root = root;
2294 }
2295
2296 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2297 root = btrfs_read_tree_root(tree_root, &location);
2298 if (IS_ERR(root)) {
2299 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2300 ret = PTR_ERR(root);
2301 if (ret != -ENOENT)
2302 goto out;
2303 }
2304 } else {
2305 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2306 fs_info->uuid_root = root;
2307 }
2308
2309 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2310 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2311 root = btrfs_read_tree_root(tree_root, &location);
2312 if (IS_ERR(root)) {
2313 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2314 ret = PTR_ERR(root);
2315 goto out;
2316 }
2317 } else {
2318 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2319 fs_info->stripe_root = root;
2320 }
2321 }
2322
2323 return 0;
2324out:
2325 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2326 location.objectid, ret);
2327 return ret;
2328}
2329
2330/*
2331 * Real super block validation
2332 * NOTE: super csum type and incompat features will not be checked here.
2333 *
2334 * @sb: super block to check
2335 * @mirror_num: the super block number to check its bytenr:
2336 * 0 the primary (1st) sb
2337 * 1, 2 2nd and 3rd backup copy
2338 * -1 skip bytenr check
2339 */
2340int btrfs_validate_super(const struct btrfs_fs_info *fs_info,
2341 const struct btrfs_super_block *sb, int mirror_num)
2342{
2343 u64 nodesize = btrfs_super_nodesize(sb);
2344 u64 sectorsize = btrfs_super_sectorsize(sb);
2345 int ret = 0;
2346 const bool ignore_flags = btrfs_test_opt(fs_info, IGNORESUPERFLAGS);
2347
2348 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2349 btrfs_err(fs_info, "no valid FS found");
2350 ret = -EINVAL;
2351 }
2352 if ((btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)) {
2353 if (!ignore_flags) {
2354 btrfs_err(fs_info,
2355 "unrecognized or unsupported super flag 0x%llx",
2356 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2357 ret = -EINVAL;
2358 } else {
2359 btrfs_info(fs_info,
2360 "unrecognized or unsupported super flags: 0x%llx, ignored",
2361 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2362 }
2363 }
2364 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2365 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2366 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2367 ret = -EINVAL;
2368 }
2369 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2370 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2371 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2372 ret = -EINVAL;
2373 }
2374 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2375 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2376 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2377 ret = -EINVAL;
2378 }
2379
2380 /*
2381 * Check sectorsize and nodesize first, other check will need it.
2382 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2383 */
2384 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2385 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2386 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2387 ret = -EINVAL;
2388 }
2389
2390 /*
2391 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2392 *
2393 * We can support 16K sectorsize with 64K page size without problem,
2394 * but such sectorsize/pagesize combination doesn't make much sense.
2395 * 4K will be our future standard, PAGE_SIZE is supported from the very
2396 * beginning.
2397 */
2398 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2399 btrfs_err(fs_info,
2400 "sectorsize %llu not yet supported for page size %lu",
2401 sectorsize, PAGE_SIZE);
2402 ret = -EINVAL;
2403 }
2404
2405 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2406 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2407 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2408 ret = -EINVAL;
2409 }
2410 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2411 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2412 le32_to_cpu(sb->__unused_leafsize), nodesize);
2413 ret = -EINVAL;
2414 }
2415
2416 /* Root alignment check */
2417 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2418 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2419 btrfs_super_root(sb));
2420 ret = -EINVAL;
2421 }
2422 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2423 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2424 btrfs_super_chunk_root(sb));
2425 ret = -EINVAL;
2426 }
2427 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2428 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2429 btrfs_super_log_root(sb));
2430 ret = -EINVAL;
2431 }
2432
2433 if (!fs_info->fs_devices->temp_fsid &&
2434 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2435 btrfs_err(fs_info,
2436 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2437 sb->fsid, fs_info->fs_devices->fsid);
2438 ret = -EINVAL;
2439 }
2440
2441 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2442 BTRFS_FSID_SIZE) != 0) {
2443 btrfs_err(fs_info,
2444"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2445 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2446 ret = -EINVAL;
2447 }
2448
2449 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2450 BTRFS_FSID_SIZE) != 0) {
2451 btrfs_err(fs_info,
2452 "dev_item UUID does not match metadata fsid: %pU != %pU",
2453 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2454 ret = -EINVAL;
2455 }
2456
2457 /*
2458 * Artificial requirement for block-group-tree to force newer features
2459 * (free-space-tree, no-holes) so the test matrix is smaller.
2460 */
2461 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2462 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2463 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2464 btrfs_err(fs_info,
2465 "block-group-tree feature requires free-space-tree and no-holes");
2466 ret = -EINVAL;
2467 }
2468
2469 /*
2470 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2471 * done later
2472 */
2473 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2474 btrfs_err(fs_info, "bytes_used is too small %llu",
2475 btrfs_super_bytes_used(sb));
2476 ret = -EINVAL;
2477 }
2478 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2479 btrfs_err(fs_info, "invalid stripesize %u",
2480 btrfs_super_stripesize(sb));
2481 ret = -EINVAL;
2482 }
2483 if (btrfs_super_num_devices(sb) > (1UL << 31))
2484 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2485 btrfs_super_num_devices(sb));
2486 if (btrfs_super_num_devices(sb) == 0) {
2487 btrfs_err(fs_info, "number of devices is 0");
2488 ret = -EINVAL;
2489 }
2490
2491 if (mirror_num >= 0 &&
2492 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2493 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2494 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2495 ret = -EINVAL;
2496 }
2497
2498 /*
2499 * Obvious sys_chunk_array corruptions, it must hold at least one key
2500 * and one chunk
2501 */
2502 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2503 btrfs_err(fs_info, "system chunk array too big %u > %u",
2504 btrfs_super_sys_array_size(sb),
2505 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2506 ret = -EINVAL;
2507 }
2508 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2509 + sizeof(struct btrfs_chunk)) {
2510 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2511 btrfs_super_sys_array_size(sb),
2512 sizeof(struct btrfs_disk_key)
2513 + sizeof(struct btrfs_chunk));
2514 ret = -EINVAL;
2515 }
2516
2517 /*
2518 * The generation is a global counter, we'll trust it more than the others
2519 * but it's still possible that it's the one that's wrong.
2520 */
2521 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2522 btrfs_warn(fs_info,
2523 "suspicious: generation < chunk_root_generation: %llu < %llu",
2524 btrfs_super_generation(sb),
2525 btrfs_super_chunk_root_generation(sb));
2526 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2527 && btrfs_super_cache_generation(sb) != (u64)-1)
2528 btrfs_warn(fs_info,
2529 "suspicious: generation < cache_generation: %llu < %llu",
2530 btrfs_super_generation(sb),
2531 btrfs_super_cache_generation(sb));
2532
2533 return ret;
2534}
2535
2536/*
2537 * Validation of super block at mount time.
2538 * Some checks already done early at mount time, like csum type and incompat
2539 * flags will be skipped.
2540 */
2541static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2542{
2543 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2544}
2545
2546/*
2547 * Validation of super block at write time.
2548 * Some checks like bytenr check will be skipped as their values will be
2549 * overwritten soon.
2550 * Extra checks like csum type and incompat flags will be done here.
2551 */
2552static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2553 struct btrfs_super_block *sb)
2554{
2555 int ret;
2556
2557 ret = btrfs_validate_super(fs_info, sb, -1);
2558 if (ret < 0)
2559 goto out;
2560 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2561 ret = -EUCLEAN;
2562 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2563 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2564 goto out;
2565 }
2566 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2567 ret = -EUCLEAN;
2568 btrfs_err(fs_info,
2569 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2570 btrfs_super_incompat_flags(sb),
2571 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2572 goto out;
2573 }
2574out:
2575 if (ret < 0)
2576 btrfs_err(fs_info,
2577 "super block corruption detected before writing it to disk");
2578 return ret;
2579}
2580
2581static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2582{
2583 struct btrfs_tree_parent_check check = {
2584 .level = level,
2585 .transid = gen,
2586 .owner_root = btrfs_root_id(root)
2587 };
2588 int ret = 0;
2589
2590 root->node = read_tree_block(root->fs_info, bytenr, &check);
2591 if (IS_ERR(root->node)) {
2592 ret = PTR_ERR(root->node);
2593 root->node = NULL;
2594 return ret;
2595 }
2596 if (!extent_buffer_uptodate(root->node)) {
2597 free_extent_buffer(root->node);
2598 root->node = NULL;
2599 return -EIO;
2600 }
2601
2602 btrfs_set_root_node(&root->root_item, root->node);
2603 root->commit_root = btrfs_root_node(root);
2604 btrfs_set_root_refs(&root->root_item, 1);
2605 return ret;
2606}
2607
2608static int load_important_roots(struct btrfs_fs_info *fs_info)
2609{
2610 struct btrfs_super_block *sb = fs_info->super_copy;
2611 u64 gen, bytenr;
2612 int level, ret;
2613
2614 bytenr = btrfs_super_root(sb);
2615 gen = btrfs_super_generation(sb);
2616 level = btrfs_super_root_level(sb);
2617 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2618 if (ret) {
2619 btrfs_warn(fs_info, "couldn't read tree root");
2620 return ret;
2621 }
2622 return 0;
2623}
2624
2625static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2626{
2627 int backup_index = find_newest_super_backup(fs_info);
2628 struct btrfs_super_block *sb = fs_info->super_copy;
2629 struct btrfs_root *tree_root = fs_info->tree_root;
2630 bool handle_error = false;
2631 int ret = 0;
2632 int i;
2633
2634 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2635 if (handle_error) {
2636 if (!IS_ERR(tree_root->node))
2637 free_extent_buffer(tree_root->node);
2638 tree_root->node = NULL;
2639
2640 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2641 break;
2642
2643 free_root_pointers(fs_info, 0);
2644
2645 /*
2646 * Don't use the log in recovery mode, it won't be
2647 * valid
2648 */
2649 btrfs_set_super_log_root(sb, 0);
2650
2651 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2652 ret = read_backup_root(fs_info, i);
2653 backup_index = ret;
2654 if (ret < 0)
2655 return ret;
2656 }
2657
2658 ret = load_important_roots(fs_info);
2659 if (ret) {
2660 handle_error = true;
2661 continue;
2662 }
2663
2664 /*
2665 * No need to hold btrfs_root::objectid_mutex since the fs
2666 * hasn't been fully initialised and we are the only user
2667 */
2668 ret = btrfs_init_root_free_objectid(tree_root);
2669 if (ret < 0) {
2670 handle_error = true;
2671 continue;
2672 }
2673
2674 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2675
2676 ret = btrfs_read_roots(fs_info);
2677 if (ret < 0) {
2678 handle_error = true;
2679 continue;
2680 }
2681
2682 /* All successful */
2683 fs_info->generation = btrfs_header_generation(tree_root->node);
2684 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2685 fs_info->last_reloc_trans = 0;
2686
2687 /* Always begin writing backup roots after the one being used */
2688 if (backup_index < 0) {
2689 fs_info->backup_root_index = 0;
2690 } else {
2691 fs_info->backup_root_index = backup_index + 1;
2692 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2693 }
2694 break;
2695 }
2696
2697 return ret;
2698}
2699
2700void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2701{
2702 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2703 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2704 INIT_LIST_HEAD(&fs_info->trans_list);
2705 INIT_LIST_HEAD(&fs_info->dead_roots);
2706 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2707 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2708 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2709 spin_lock_init(&fs_info->delalloc_root_lock);
2710 spin_lock_init(&fs_info->trans_lock);
2711 spin_lock_init(&fs_info->fs_roots_radix_lock);
2712 spin_lock_init(&fs_info->delayed_iput_lock);
2713 spin_lock_init(&fs_info->defrag_inodes_lock);
2714 spin_lock_init(&fs_info->super_lock);
2715 spin_lock_init(&fs_info->buffer_lock);
2716 spin_lock_init(&fs_info->unused_bgs_lock);
2717 spin_lock_init(&fs_info->treelog_bg_lock);
2718 spin_lock_init(&fs_info->zone_active_bgs_lock);
2719 spin_lock_init(&fs_info->relocation_bg_lock);
2720 rwlock_init(&fs_info->tree_mod_log_lock);
2721 rwlock_init(&fs_info->global_root_lock);
2722 mutex_init(&fs_info->unused_bg_unpin_mutex);
2723 mutex_init(&fs_info->reclaim_bgs_lock);
2724 mutex_init(&fs_info->reloc_mutex);
2725 mutex_init(&fs_info->delalloc_root_mutex);
2726 mutex_init(&fs_info->zoned_meta_io_lock);
2727 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2728 seqlock_init(&fs_info->profiles_lock);
2729
2730 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2731 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2732 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2733 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2734 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2735 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2736 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2737 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2738 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2739 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2740 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2741 BTRFS_LOCKDEP_TRANS_COMPLETED);
2742
2743 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2744 INIT_LIST_HEAD(&fs_info->space_info);
2745 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2746 INIT_LIST_HEAD(&fs_info->unused_bgs);
2747 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2748 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2749#ifdef CONFIG_BTRFS_DEBUG
2750 INIT_LIST_HEAD(&fs_info->allocated_roots);
2751 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2752 spin_lock_init(&fs_info->eb_leak_lock);
2753#endif
2754 fs_info->mapping_tree = RB_ROOT_CACHED;
2755 rwlock_init(&fs_info->mapping_tree_lock);
2756 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2757 BTRFS_BLOCK_RSV_GLOBAL);
2758 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2759 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2760 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2761 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2762 BTRFS_BLOCK_RSV_DELOPS);
2763 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2764 BTRFS_BLOCK_RSV_DELREFS);
2765
2766 atomic_set(&fs_info->async_delalloc_pages, 0);
2767 atomic_set(&fs_info->defrag_running, 0);
2768 atomic_set(&fs_info->nr_delayed_iputs, 0);
2769 atomic64_set(&fs_info->tree_mod_seq, 0);
2770 fs_info->global_root_tree = RB_ROOT;
2771 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2772 fs_info->metadata_ratio = 0;
2773 fs_info->defrag_inodes = RB_ROOT;
2774 atomic64_set(&fs_info->free_chunk_space, 0);
2775 fs_info->tree_mod_log = RB_ROOT;
2776 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2777 btrfs_init_ref_verify(fs_info);
2778
2779 fs_info->thread_pool_size = min_t(unsigned long,
2780 num_online_cpus() + 2, 8);
2781
2782 INIT_LIST_HEAD(&fs_info->ordered_roots);
2783 spin_lock_init(&fs_info->ordered_root_lock);
2784
2785 btrfs_init_scrub(fs_info);
2786 btrfs_init_balance(fs_info);
2787 btrfs_init_async_reclaim_work(fs_info);
2788 btrfs_init_extent_map_shrinker_work(fs_info);
2789
2790 rwlock_init(&fs_info->block_group_cache_lock);
2791 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2792
2793 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2794 IO_TREE_FS_EXCLUDED_EXTENTS);
2795
2796 mutex_init(&fs_info->ordered_operations_mutex);
2797 mutex_init(&fs_info->tree_log_mutex);
2798 mutex_init(&fs_info->chunk_mutex);
2799 mutex_init(&fs_info->transaction_kthread_mutex);
2800 mutex_init(&fs_info->cleaner_mutex);
2801 mutex_init(&fs_info->ro_block_group_mutex);
2802 init_rwsem(&fs_info->commit_root_sem);
2803 init_rwsem(&fs_info->cleanup_work_sem);
2804 init_rwsem(&fs_info->subvol_sem);
2805 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2806
2807 btrfs_init_dev_replace_locks(fs_info);
2808 btrfs_init_qgroup(fs_info);
2809 btrfs_discard_init(fs_info);
2810
2811 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2812 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2813
2814 init_waitqueue_head(&fs_info->transaction_throttle);
2815 init_waitqueue_head(&fs_info->transaction_wait);
2816 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2817 init_waitqueue_head(&fs_info->async_submit_wait);
2818 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2819
2820 /* Usable values until the real ones are cached from the superblock */
2821 fs_info->nodesize = 4096;
2822 fs_info->sectorsize = 4096;
2823 fs_info->sectorsize_bits = ilog2(4096);
2824 fs_info->stripesize = 4096;
2825
2826 /* Default compress algorithm when user does -o compress */
2827 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2828
2829 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2830
2831 spin_lock_init(&fs_info->swapfile_pins_lock);
2832 fs_info->swapfile_pins = RB_ROOT;
2833
2834 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2835 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2836}
2837
2838static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2839{
2840 int ret;
2841
2842 fs_info->sb = sb;
2843 /* Temporary fixed values for block size until we read the superblock. */
2844 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2845 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2846
2847 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2848 if (ret)
2849 return ret;
2850
2851 ret = percpu_counter_init(&fs_info->evictable_extent_maps, 0, GFP_KERNEL);
2852 if (ret)
2853 return ret;
2854
2855 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2856 if (ret)
2857 return ret;
2858
2859 fs_info->dirty_metadata_batch = PAGE_SIZE *
2860 (1 + ilog2(nr_cpu_ids));
2861
2862 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2863 if (ret)
2864 return ret;
2865
2866 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2867 GFP_KERNEL);
2868 if (ret)
2869 return ret;
2870
2871 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2872 GFP_KERNEL);
2873 if (!fs_info->delayed_root)
2874 return -ENOMEM;
2875 btrfs_init_delayed_root(fs_info->delayed_root);
2876
2877 if (sb_rdonly(sb))
2878 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2879 if (btrfs_test_opt(fs_info, IGNOREMETACSUMS))
2880 set_bit(BTRFS_FS_STATE_SKIP_META_CSUMS, &fs_info->fs_state);
2881
2882 return btrfs_alloc_stripe_hash_table(fs_info);
2883}
2884
2885static int btrfs_uuid_rescan_kthread(void *data)
2886{
2887 struct btrfs_fs_info *fs_info = data;
2888 int ret;
2889
2890 /*
2891 * 1st step is to iterate through the existing UUID tree and
2892 * to delete all entries that contain outdated data.
2893 * 2nd step is to add all missing entries to the UUID tree.
2894 */
2895 ret = btrfs_uuid_tree_iterate(fs_info);
2896 if (ret < 0) {
2897 if (ret != -EINTR)
2898 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2899 ret);
2900 up(&fs_info->uuid_tree_rescan_sem);
2901 return ret;
2902 }
2903 return btrfs_uuid_scan_kthread(data);
2904}
2905
2906static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2907{
2908 struct task_struct *task;
2909
2910 down(&fs_info->uuid_tree_rescan_sem);
2911 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2912 if (IS_ERR(task)) {
2913 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2914 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2915 up(&fs_info->uuid_tree_rescan_sem);
2916 return PTR_ERR(task);
2917 }
2918
2919 return 0;
2920}
2921
2922static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2923{
2924 u64 root_objectid = 0;
2925 struct btrfs_root *gang[8];
2926 int ret = 0;
2927
2928 while (1) {
2929 unsigned int found;
2930
2931 spin_lock(&fs_info->fs_roots_radix_lock);
2932 found = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2933 (void **)gang, root_objectid,
2934 ARRAY_SIZE(gang));
2935 if (!found) {
2936 spin_unlock(&fs_info->fs_roots_radix_lock);
2937 break;
2938 }
2939 root_objectid = btrfs_root_id(gang[found - 1]) + 1;
2940
2941 for (int i = 0; i < found; i++) {
2942 /* Avoid to grab roots in dead_roots. */
2943 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2944 gang[i] = NULL;
2945 continue;
2946 }
2947 /* Grab all the search result for later use. */
2948 gang[i] = btrfs_grab_root(gang[i]);
2949 }
2950 spin_unlock(&fs_info->fs_roots_radix_lock);
2951
2952 for (int i = 0; i < found; i++) {
2953 if (!gang[i])
2954 continue;
2955 root_objectid = btrfs_root_id(gang[i]);
2956 /*
2957 * Continue to release the remaining roots after the first
2958 * error without cleanup and preserve the first error
2959 * for the return.
2960 */
2961 if (!ret)
2962 ret = btrfs_orphan_cleanup(gang[i]);
2963 btrfs_put_root(gang[i]);
2964 }
2965 if (ret)
2966 break;
2967
2968 root_objectid++;
2969 }
2970 return ret;
2971}
2972
2973/*
2974 * Mounting logic specific to read-write file systems. Shared by open_ctree
2975 * and btrfs_remount when remounting from read-only to read-write.
2976 */
2977int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2978{
2979 int ret;
2980 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2981 bool rebuild_free_space_tree = false;
2982
2983 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2984 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2985 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2986 btrfs_warn(fs_info,
2987 "'clear_cache' option is ignored with extent tree v2");
2988 else
2989 rebuild_free_space_tree = true;
2990 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2991 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2992 btrfs_warn(fs_info, "free space tree is invalid");
2993 rebuild_free_space_tree = true;
2994 }
2995
2996 if (rebuild_free_space_tree) {
2997 btrfs_info(fs_info, "rebuilding free space tree");
2998 ret = btrfs_rebuild_free_space_tree(fs_info);
2999 if (ret) {
3000 btrfs_warn(fs_info,
3001 "failed to rebuild free space tree: %d", ret);
3002 goto out;
3003 }
3004 }
3005
3006 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3007 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3008 btrfs_info(fs_info, "disabling free space tree");
3009 ret = btrfs_delete_free_space_tree(fs_info);
3010 if (ret) {
3011 btrfs_warn(fs_info,
3012 "failed to disable free space tree: %d", ret);
3013 goto out;
3014 }
3015 }
3016
3017 /*
3018 * btrfs_find_orphan_roots() is responsible for finding all the dead
3019 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3020 * them into the fs_info->fs_roots_radix tree. This must be done before
3021 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3022 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3023 * item before the root's tree is deleted - this means that if we unmount
3024 * or crash before the deletion completes, on the next mount we will not
3025 * delete what remains of the tree because the orphan item does not
3026 * exists anymore, which is what tells us we have a pending deletion.
3027 */
3028 ret = btrfs_find_orphan_roots(fs_info);
3029 if (ret)
3030 goto out;
3031
3032 ret = btrfs_cleanup_fs_roots(fs_info);
3033 if (ret)
3034 goto out;
3035
3036 down_read(&fs_info->cleanup_work_sem);
3037 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3038 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3039 up_read(&fs_info->cleanup_work_sem);
3040 goto out;
3041 }
3042 up_read(&fs_info->cleanup_work_sem);
3043
3044 mutex_lock(&fs_info->cleaner_mutex);
3045 ret = btrfs_recover_relocation(fs_info);
3046 mutex_unlock(&fs_info->cleaner_mutex);
3047 if (ret < 0) {
3048 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3049 goto out;
3050 }
3051
3052 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3053 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3054 btrfs_info(fs_info, "creating free space tree");
3055 ret = btrfs_create_free_space_tree(fs_info);
3056 if (ret) {
3057 btrfs_warn(fs_info,
3058 "failed to create free space tree: %d", ret);
3059 goto out;
3060 }
3061 }
3062
3063 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3064 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3065 if (ret)
3066 goto out;
3067 }
3068
3069 ret = btrfs_resume_balance_async(fs_info);
3070 if (ret)
3071 goto out;
3072
3073 ret = btrfs_resume_dev_replace_async(fs_info);
3074 if (ret) {
3075 btrfs_warn(fs_info, "failed to resume dev_replace");
3076 goto out;
3077 }
3078
3079 btrfs_qgroup_rescan_resume(fs_info);
3080
3081 if (!fs_info->uuid_root) {
3082 btrfs_info(fs_info, "creating UUID tree");
3083 ret = btrfs_create_uuid_tree(fs_info);
3084 if (ret) {
3085 btrfs_warn(fs_info,
3086 "failed to create the UUID tree %d", ret);
3087 goto out;
3088 }
3089 }
3090
3091out:
3092 return ret;
3093}
3094
3095/*
3096 * Do various sanity and dependency checks of different features.
3097 *
3098 * @is_rw_mount: If the mount is read-write.
3099 *
3100 * This is the place for less strict checks (like for subpage or artificial
3101 * feature dependencies).
3102 *
3103 * For strict checks or possible corruption detection, see
3104 * btrfs_validate_super().
3105 *
3106 * This should be called after btrfs_parse_options(), as some mount options
3107 * (space cache related) can modify on-disk format like free space tree and
3108 * screw up certain feature dependencies.
3109 */
3110int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3111{
3112 struct btrfs_super_block *disk_super = fs_info->super_copy;
3113 u64 incompat = btrfs_super_incompat_flags(disk_super);
3114 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3115 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3116
3117 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3118 btrfs_err(fs_info,
3119 "cannot mount because of unknown incompat features (0x%llx)",
3120 incompat);
3121 return -EINVAL;
3122 }
3123
3124 /* Runtime limitation for mixed block groups. */
3125 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3126 (fs_info->sectorsize != fs_info->nodesize)) {
3127 btrfs_err(fs_info,
3128"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3129 fs_info->nodesize, fs_info->sectorsize);
3130 return -EINVAL;
3131 }
3132
3133 /* Mixed backref is an always-enabled feature. */
3134 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3135
3136 /* Set compression related flags just in case. */
3137 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3138 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3139 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3140 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3141
3142 /*
3143 * An ancient flag, which should really be marked deprecated.
3144 * Such runtime limitation doesn't really need a incompat flag.
3145 */
3146 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3147 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3148
3149 if (compat_ro_unsupp && is_rw_mount) {
3150 btrfs_err(fs_info,
3151 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3152 compat_ro);
3153 return -EINVAL;
3154 }
3155
3156 /*
3157 * We have unsupported RO compat features, although RO mounted, we
3158 * should not cause any metadata writes, including log replay.
3159 * Or we could screw up whatever the new feature requires.
3160 */
3161 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3162 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3163 btrfs_err(fs_info,
3164"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3165 compat_ro);
3166 return -EINVAL;
3167 }
3168
3169 /*
3170 * Artificial limitations for block group tree, to force
3171 * block-group-tree to rely on no-holes and free-space-tree.
3172 */
3173 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3174 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3175 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3176 btrfs_err(fs_info,
3177"block-group-tree feature requires no-holes and free-space-tree features");
3178 return -EINVAL;
3179 }
3180
3181 /*
3182 * Subpage runtime limitation on v1 cache.
3183 *
3184 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3185 * we're already defaulting to v2 cache, no need to bother v1 as it's
3186 * going to be deprecated anyway.
3187 */
3188 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3189 btrfs_warn(fs_info,
3190 "v1 space cache is not supported for page size %lu with sectorsize %u",
3191 PAGE_SIZE, fs_info->sectorsize);
3192 return -EINVAL;
3193 }
3194
3195 /* This can be called by remount, we need to protect the super block. */
3196 spin_lock(&fs_info->super_lock);
3197 btrfs_set_super_incompat_flags(disk_super, incompat);
3198 spin_unlock(&fs_info->super_lock);
3199
3200 return 0;
3201}
3202
3203int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices)
3204{
3205 u32 sectorsize;
3206 u32 nodesize;
3207 u32 stripesize;
3208 u64 generation;
3209 u16 csum_type;
3210 struct btrfs_super_block *disk_super;
3211 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3212 struct btrfs_root *tree_root;
3213 struct btrfs_root *chunk_root;
3214 int ret;
3215 int level;
3216
3217 ret = init_mount_fs_info(fs_info, sb);
3218 if (ret)
3219 goto fail;
3220
3221 /* These need to be init'ed before we start creating inodes and such. */
3222 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3223 GFP_KERNEL);
3224 fs_info->tree_root = tree_root;
3225 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3226 GFP_KERNEL);
3227 fs_info->chunk_root = chunk_root;
3228 if (!tree_root || !chunk_root) {
3229 ret = -ENOMEM;
3230 goto fail;
3231 }
3232
3233 ret = btrfs_init_btree_inode(sb);
3234 if (ret)
3235 goto fail;
3236
3237 invalidate_bdev(fs_devices->latest_dev->bdev);
3238
3239 /*
3240 * Read super block and check the signature bytes only
3241 */
3242 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3243 if (IS_ERR(disk_super)) {
3244 ret = PTR_ERR(disk_super);
3245 goto fail_alloc;
3246 }
3247
3248 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3249 /*
3250 * Verify the type first, if that or the checksum value are
3251 * corrupted, we'll find out
3252 */
3253 csum_type = btrfs_super_csum_type(disk_super);
3254 if (!btrfs_supported_super_csum(csum_type)) {
3255 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3256 csum_type);
3257 ret = -EINVAL;
3258 btrfs_release_disk_super(disk_super);
3259 goto fail_alloc;
3260 }
3261
3262 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3263
3264 ret = btrfs_init_csum_hash(fs_info, csum_type);
3265 if (ret) {
3266 btrfs_release_disk_super(disk_super);
3267 goto fail_alloc;
3268 }
3269
3270 /*
3271 * We want to check superblock checksum, the type is stored inside.
3272 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3273 */
3274 if (btrfs_check_super_csum(fs_info, disk_super)) {
3275 btrfs_err(fs_info, "superblock checksum mismatch");
3276 ret = -EINVAL;
3277 btrfs_release_disk_super(disk_super);
3278 goto fail_alloc;
3279 }
3280
3281 /*
3282 * super_copy is zeroed at allocation time and we never touch the
3283 * following bytes up to INFO_SIZE, the checksum is calculated from
3284 * the whole block of INFO_SIZE
3285 */
3286 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3287 btrfs_release_disk_super(disk_super);
3288
3289 disk_super = fs_info->super_copy;
3290
3291 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3292 sizeof(*fs_info->super_for_commit));
3293
3294 ret = btrfs_validate_mount_super(fs_info);
3295 if (ret) {
3296 btrfs_err(fs_info, "superblock contains fatal errors");
3297 ret = -EINVAL;
3298 goto fail_alloc;
3299 }
3300
3301 if (!btrfs_super_root(disk_super)) {
3302 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3303 ret = -EINVAL;
3304 goto fail_alloc;
3305 }
3306
3307 /* check FS state, whether FS is broken. */
3308 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3309 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3310
3311 /* Set up fs_info before parsing mount options */
3312 nodesize = btrfs_super_nodesize(disk_super);
3313 sectorsize = btrfs_super_sectorsize(disk_super);
3314 stripesize = sectorsize;
3315 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3316 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3317
3318 fs_info->nodesize = nodesize;
3319 fs_info->sectorsize = sectorsize;
3320 fs_info->sectorsize_bits = ilog2(sectorsize);
3321 fs_info->sectors_per_page = (PAGE_SIZE >> fs_info->sectorsize_bits);
3322 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3323 fs_info->stripesize = stripesize;
3324
3325 /*
3326 * Handle the space caching options appropriately now that we have the
3327 * super block loaded and validated.
3328 */
3329 btrfs_set_free_space_cache_settings(fs_info);
3330
3331 if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3332 ret = -EINVAL;
3333 goto fail_alloc;
3334 }
3335
3336 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3337 if (ret < 0)
3338 goto fail_alloc;
3339
3340 /*
3341 * At this point our mount options are validated, if we set ->max_inline
3342 * to something non-standard make sure we truncate it to sectorsize.
3343 */
3344 fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3345
3346 if (sectorsize < PAGE_SIZE)
3347 btrfs_warn(fs_info,
3348 "read-write for sector size %u with page size %lu is experimental",
3349 sectorsize, PAGE_SIZE);
3350
3351 ret = btrfs_init_workqueues(fs_info);
3352 if (ret)
3353 goto fail_sb_buffer;
3354
3355 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3356 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3357
3358 /* Update the values for the current filesystem. */
3359 sb->s_blocksize = sectorsize;
3360 sb->s_blocksize_bits = blksize_bits(sectorsize);
3361 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3362
3363 mutex_lock(&fs_info->chunk_mutex);
3364 ret = btrfs_read_sys_array(fs_info);
3365 mutex_unlock(&fs_info->chunk_mutex);
3366 if (ret) {
3367 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3368 goto fail_sb_buffer;
3369 }
3370
3371 generation = btrfs_super_chunk_root_generation(disk_super);
3372 level = btrfs_super_chunk_root_level(disk_super);
3373 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3374 generation, level);
3375 if (ret) {
3376 btrfs_err(fs_info, "failed to read chunk root");
3377 goto fail_tree_roots;
3378 }
3379
3380 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3381 offsetof(struct btrfs_header, chunk_tree_uuid),
3382 BTRFS_UUID_SIZE);
3383
3384 ret = btrfs_read_chunk_tree(fs_info);
3385 if (ret) {
3386 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3387 goto fail_tree_roots;
3388 }
3389
3390 /*
3391 * At this point we know all the devices that make this filesystem,
3392 * including the seed devices but we don't know yet if the replace
3393 * target is required. So free devices that are not part of this
3394 * filesystem but skip the replace target device which is checked
3395 * below in btrfs_init_dev_replace().
3396 */
3397 btrfs_free_extra_devids(fs_devices);
3398 if (!fs_devices->latest_dev->bdev) {
3399 btrfs_err(fs_info, "failed to read devices");
3400 ret = -EIO;
3401 goto fail_tree_roots;
3402 }
3403
3404 ret = init_tree_roots(fs_info);
3405 if (ret)
3406 goto fail_tree_roots;
3407
3408 /*
3409 * Get zone type information of zoned block devices. This will also
3410 * handle emulation of a zoned filesystem if a regular device has the
3411 * zoned incompat feature flag set.
3412 */
3413 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3414 if (ret) {
3415 btrfs_err(fs_info,
3416 "zoned: failed to read device zone info: %d", ret);
3417 goto fail_block_groups;
3418 }
3419
3420 /*
3421 * If we have a uuid root and we're not being told to rescan we need to
3422 * check the generation here so we can set the
3423 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3424 * transaction during a balance or the log replay without updating the
3425 * uuid generation, and then if we crash we would rescan the uuid tree,
3426 * even though it was perfectly fine.
3427 */
3428 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3429 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3430 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3431
3432 ret = btrfs_verify_dev_extents(fs_info);
3433 if (ret) {
3434 btrfs_err(fs_info,
3435 "failed to verify dev extents against chunks: %d",
3436 ret);
3437 goto fail_block_groups;
3438 }
3439 ret = btrfs_recover_balance(fs_info);
3440 if (ret) {
3441 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3442 goto fail_block_groups;
3443 }
3444
3445 ret = btrfs_init_dev_stats(fs_info);
3446 if (ret) {
3447 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3448 goto fail_block_groups;
3449 }
3450
3451 ret = btrfs_init_dev_replace(fs_info);
3452 if (ret) {
3453 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3454 goto fail_block_groups;
3455 }
3456
3457 ret = btrfs_check_zoned_mode(fs_info);
3458 if (ret) {
3459 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3460 ret);
3461 goto fail_block_groups;
3462 }
3463
3464 ret = btrfs_sysfs_add_fsid(fs_devices);
3465 if (ret) {
3466 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3467 ret);
3468 goto fail_block_groups;
3469 }
3470
3471 ret = btrfs_sysfs_add_mounted(fs_info);
3472 if (ret) {
3473 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3474 goto fail_fsdev_sysfs;
3475 }
3476
3477 ret = btrfs_init_space_info(fs_info);
3478 if (ret) {
3479 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3480 goto fail_sysfs;
3481 }
3482
3483 ret = btrfs_read_block_groups(fs_info);
3484 if (ret) {
3485 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3486 goto fail_sysfs;
3487 }
3488
3489 btrfs_free_zone_cache(fs_info);
3490
3491 btrfs_check_active_zone_reservation(fs_info);
3492
3493 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3494 !btrfs_check_rw_degradable(fs_info, NULL)) {
3495 btrfs_warn(fs_info,
3496 "writable mount is not allowed due to too many missing devices");
3497 ret = -EINVAL;
3498 goto fail_sysfs;
3499 }
3500
3501 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3502 "btrfs-cleaner");
3503 if (IS_ERR(fs_info->cleaner_kthread)) {
3504 ret = PTR_ERR(fs_info->cleaner_kthread);
3505 goto fail_sysfs;
3506 }
3507
3508 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3509 tree_root,
3510 "btrfs-transaction");
3511 if (IS_ERR(fs_info->transaction_kthread)) {
3512 ret = PTR_ERR(fs_info->transaction_kthread);
3513 goto fail_cleaner;
3514 }
3515
3516 ret = btrfs_read_qgroup_config(fs_info);
3517 if (ret)
3518 goto fail_trans_kthread;
3519
3520 if (btrfs_build_ref_tree(fs_info))
3521 btrfs_err(fs_info, "couldn't build ref tree");
3522
3523 /* do not make disk changes in broken FS or nologreplay is given */
3524 if (btrfs_super_log_root(disk_super) != 0 &&
3525 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3526 btrfs_info(fs_info, "start tree-log replay");
3527 ret = btrfs_replay_log(fs_info, fs_devices);
3528 if (ret)
3529 goto fail_qgroup;
3530 }
3531
3532 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3533 if (IS_ERR(fs_info->fs_root)) {
3534 ret = PTR_ERR(fs_info->fs_root);
3535 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3536 fs_info->fs_root = NULL;
3537 goto fail_qgroup;
3538 }
3539
3540 if (sb_rdonly(sb))
3541 return 0;
3542
3543 ret = btrfs_start_pre_rw_mount(fs_info);
3544 if (ret) {
3545 close_ctree(fs_info);
3546 return ret;
3547 }
3548 btrfs_discard_resume(fs_info);
3549
3550 if (fs_info->uuid_root &&
3551 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3552 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3553 btrfs_info(fs_info, "checking UUID tree");
3554 ret = btrfs_check_uuid_tree(fs_info);
3555 if (ret) {
3556 btrfs_warn(fs_info,
3557 "failed to check the UUID tree: %d", ret);
3558 close_ctree(fs_info);
3559 return ret;
3560 }
3561 }
3562
3563 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3564
3565 /* Kick the cleaner thread so it'll start deleting snapshots. */
3566 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3567 wake_up_process(fs_info->cleaner_kthread);
3568
3569 return 0;
3570
3571fail_qgroup:
3572 btrfs_free_qgroup_config(fs_info);
3573fail_trans_kthread:
3574 kthread_stop(fs_info->transaction_kthread);
3575 btrfs_cleanup_transaction(fs_info);
3576 btrfs_free_fs_roots(fs_info);
3577fail_cleaner:
3578 kthread_stop(fs_info->cleaner_kthread);
3579
3580 /*
3581 * make sure we're done with the btree inode before we stop our
3582 * kthreads
3583 */
3584 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3585
3586fail_sysfs:
3587 btrfs_sysfs_remove_mounted(fs_info);
3588
3589fail_fsdev_sysfs:
3590 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3591
3592fail_block_groups:
3593 btrfs_put_block_group_cache(fs_info);
3594
3595fail_tree_roots:
3596 if (fs_info->data_reloc_root)
3597 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3598 free_root_pointers(fs_info, true);
3599 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3600
3601fail_sb_buffer:
3602 btrfs_stop_all_workers(fs_info);
3603 btrfs_free_block_groups(fs_info);
3604fail_alloc:
3605 btrfs_mapping_tree_free(fs_info);
3606
3607 iput(fs_info->btree_inode);
3608fail:
3609 btrfs_close_devices(fs_info->fs_devices);
3610 ASSERT(ret < 0);
3611 return ret;
3612}
3613ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3614
3615static void btrfs_end_super_write(struct bio *bio)
3616{
3617 struct btrfs_device *device = bio->bi_private;
3618 struct folio_iter fi;
3619
3620 bio_for_each_folio_all(fi, bio) {
3621 if (bio->bi_status) {
3622 btrfs_warn_rl_in_rcu(device->fs_info,
3623 "lost super block write due to IO error on %s (%d)",
3624 btrfs_dev_name(device),
3625 blk_status_to_errno(bio->bi_status));
3626 btrfs_dev_stat_inc_and_print(device,
3627 BTRFS_DEV_STAT_WRITE_ERRS);
3628 /* Ensure failure if the primary sb fails. */
3629 if (bio->bi_opf & REQ_FUA)
3630 atomic_add(BTRFS_SUPER_PRIMARY_WRITE_ERROR,
3631 &device->sb_write_errors);
3632 else
3633 atomic_inc(&device->sb_write_errors);
3634 }
3635 folio_unlock(fi.folio);
3636 folio_put(fi.folio);
3637 }
3638
3639 bio_put(bio);
3640}
3641
3642struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3643 int copy_num, bool drop_cache)
3644{
3645 struct btrfs_super_block *super;
3646 struct page *page;
3647 u64 bytenr, bytenr_orig;
3648 struct address_space *mapping = bdev->bd_mapping;
3649 int ret;
3650
3651 bytenr_orig = btrfs_sb_offset(copy_num);
3652 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3653 if (ret == -ENOENT)
3654 return ERR_PTR(-EINVAL);
3655 else if (ret)
3656 return ERR_PTR(ret);
3657
3658 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3659 return ERR_PTR(-EINVAL);
3660
3661 if (drop_cache) {
3662 /* This should only be called with the primary sb. */
3663 ASSERT(copy_num == 0);
3664
3665 /*
3666 * Drop the page of the primary superblock, so later read will
3667 * always read from the device.
3668 */
3669 invalidate_inode_pages2_range(mapping,
3670 bytenr >> PAGE_SHIFT,
3671 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3672 }
3673
3674 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3675 if (IS_ERR(page))
3676 return ERR_CAST(page);
3677
3678 super = page_address(page);
3679 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3680 btrfs_release_disk_super(super);
3681 return ERR_PTR(-ENODATA);
3682 }
3683
3684 if (btrfs_super_bytenr(super) != bytenr_orig) {
3685 btrfs_release_disk_super(super);
3686 return ERR_PTR(-EINVAL);
3687 }
3688
3689 return super;
3690}
3691
3692
3693struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3694{
3695 struct btrfs_super_block *super, *latest = NULL;
3696 int i;
3697 u64 transid = 0;
3698
3699 /* we would like to check all the supers, but that would make
3700 * a btrfs mount succeed after a mkfs from a different FS.
3701 * So, we need to add a special mount option to scan for
3702 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3703 */
3704 for (i = 0; i < 1; i++) {
3705 super = btrfs_read_dev_one_super(bdev, i, false);
3706 if (IS_ERR(super))
3707 continue;
3708
3709 if (!latest || btrfs_super_generation(super) > transid) {
3710 if (latest)
3711 btrfs_release_disk_super(super);
3712
3713 latest = super;
3714 transid = btrfs_super_generation(super);
3715 }
3716 }
3717
3718 return super;
3719}
3720
3721/*
3722 * Write superblock @sb to the @device. Do not wait for completion, all the
3723 * folios we use for writing are locked.
3724 *
3725 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3726 * the expected device size at commit time. Note that max_mirrors must be
3727 * same for write and wait phases.
3728 *
3729 * Return number of errors when folio is not found or submission fails.
3730 */
3731static int write_dev_supers(struct btrfs_device *device,
3732 struct btrfs_super_block *sb, int max_mirrors)
3733{
3734 struct btrfs_fs_info *fs_info = device->fs_info;
3735 struct address_space *mapping = device->bdev->bd_mapping;
3736 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3737 int i;
3738 int ret;
3739 u64 bytenr, bytenr_orig;
3740
3741 atomic_set(&device->sb_write_errors, 0);
3742
3743 if (max_mirrors == 0)
3744 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3745
3746 shash->tfm = fs_info->csum_shash;
3747
3748 for (i = 0; i < max_mirrors; i++) {
3749 struct folio *folio;
3750 struct bio *bio;
3751 struct btrfs_super_block *disk_super;
3752 size_t offset;
3753
3754 bytenr_orig = btrfs_sb_offset(i);
3755 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3756 if (ret == -ENOENT) {
3757 continue;
3758 } else if (ret < 0) {
3759 btrfs_err(device->fs_info,
3760 "couldn't get super block location for mirror %d",
3761 i);
3762 atomic_inc(&device->sb_write_errors);
3763 continue;
3764 }
3765 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3766 device->commit_total_bytes)
3767 break;
3768
3769 btrfs_set_super_bytenr(sb, bytenr_orig);
3770
3771 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3772 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3773 sb->csum);
3774
3775 folio = __filemap_get_folio(mapping, bytenr >> PAGE_SHIFT,
3776 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3777 GFP_NOFS);
3778 if (IS_ERR(folio)) {
3779 btrfs_err(device->fs_info,
3780 "couldn't get super block page for bytenr %llu",
3781 bytenr);
3782 atomic_inc(&device->sb_write_errors);
3783 continue;
3784 }
3785 ASSERT(folio_order(folio) == 0);
3786
3787 offset = offset_in_folio(folio, bytenr);
3788 disk_super = folio_address(folio) + offset;
3789 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3790
3791 /*
3792 * Directly use bios here instead of relying on the page cache
3793 * to do I/O, so we don't lose the ability to do integrity
3794 * checking.
3795 */
3796 bio = bio_alloc(device->bdev, 1,
3797 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3798 GFP_NOFS);
3799 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3800 bio->bi_private = device;
3801 bio->bi_end_io = btrfs_end_super_write;
3802 bio_add_folio_nofail(bio, folio, BTRFS_SUPER_INFO_SIZE, offset);
3803
3804 /*
3805 * We FUA only the first super block. The others we allow to
3806 * go down lazy and there's a short window where the on-disk
3807 * copies might still contain the older version.
3808 */
3809 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3810 bio->bi_opf |= REQ_FUA;
3811 submit_bio(bio);
3812
3813 if (btrfs_advance_sb_log(device, i))
3814 atomic_inc(&device->sb_write_errors);
3815 }
3816 return atomic_read(&device->sb_write_errors) < i ? 0 : -1;
3817}
3818
3819/*
3820 * Wait for write completion of superblocks done by write_dev_supers,
3821 * @max_mirrors same for write and wait phases.
3822 *
3823 * Return -1 if primary super block write failed or when there were no super block
3824 * copies written. Otherwise 0.
3825 */
3826static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3827{
3828 int i;
3829 int errors = 0;
3830 bool primary_failed = false;
3831 int ret;
3832 u64 bytenr;
3833
3834 if (max_mirrors == 0)
3835 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3836
3837 for (i = 0; i < max_mirrors; i++) {
3838 struct folio *folio;
3839
3840 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3841 if (ret == -ENOENT) {
3842 break;
3843 } else if (ret < 0) {
3844 errors++;
3845 if (i == 0)
3846 primary_failed = true;
3847 continue;
3848 }
3849 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3850 device->commit_total_bytes)
3851 break;
3852
3853 folio = filemap_get_folio(device->bdev->bd_mapping,
3854 bytenr >> PAGE_SHIFT);
3855 /* If the folio has been removed, then we know it completed. */
3856 if (IS_ERR(folio))
3857 continue;
3858 ASSERT(folio_order(folio) == 0);
3859
3860 /* Folio will be unlocked once the write completes. */
3861 folio_wait_locked(folio);
3862 folio_put(folio);
3863 }
3864
3865 errors += atomic_read(&device->sb_write_errors);
3866 if (errors >= BTRFS_SUPER_PRIMARY_WRITE_ERROR)
3867 primary_failed = true;
3868 if (primary_failed) {
3869 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3870 device->devid);
3871 return -1;
3872 }
3873
3874 return errors < i ? 0 : -1;
3875}
3876
3877/*
3878 * endio for the write_dev_flush, this will wake anyone waiting
3879 * for the barrier when it is done
3880 */
3881static void btrfs_end_empty_barrier(struct bio *bio)
3882{
3883 bio_uninit(bio);
3884 complete(bio->bi_private);
3885}
3886
3887/*
3888 * Submit a flush request to the device if it supports it. Error handling is
3889 * done in the waiting counterpart.
3890 */
3891static void write_dev_flush(struct btrfs_device *device)
3892{
3893 struct bio *bio = &device->flush_bio;
3894
3895 device->last_flush_error = BLK_STS_OK;
3896
3897 bio_init(bio, device->bdev, NULL, 0,
3898 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3899 bio->bi_end_io = btrfs_end_empty_barrier;
3900 init_completion(&device->flush_wait);
3901 bio->bi_private = &device->flush_wait;
3902 submit_bio(bio);
3903 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3904}
3905
3906/*
3907 * If the flush bio has been submitted by write_dev_flush, wait for it.
3908 * Return true for any error, and false otherwise.
3909 */
3910static bool wait_dev_flush(struct btrfs_device *device)
3911{
3912 struct bio *bio = &device->flush_bio;
3913
3914 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3915 return false;
3916
3917 wait_for_completion_io(&device->flush_wait);
3918
3919 if (bio->bi_status) {
3920 device->last_flush_error = bio->bi_status;
3921 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3922 return true;
3923 }
3924
3925 return false;
3926}
3927
3928/*
3929 * send an empty flush down to each device in parallel,
3930 * then wait for them
3931 */
3932static int barrier_all_devices(struct btrfs_fs_info *info)
3933{
3934 struct list_head *head;
3935 struct btrfs_device *dev;
3936 int errors_wait = 0;
3937
3938 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3939 /* send down all the barriers */
3940 head = &info->fs_devices->devices;
3941 list_for_each_entry(dev, head, dev_list) {
3942 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3943 continue;
3944 if (!dev->bdev)
3945 continue;
3946 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3947 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3948 continue;
3949
3950 write_dev_flush(dev);
3951 }
3952
3953 /* wait for all the barriers */
3954 list_for_each_entry(dev, head, dev_list) {
3955 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3956 continue;
3957 if (!dev->bdev) {
3958 errors_wait++;
3959 continue;
3960 }
3961 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3962 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3963 continue;
3964
3965 if (wait_dev_flush(dev))
3966 errors_wait++;
3967 }
3968
3969 /*
3970 * Checks last_flush_error of disks in order to determine the device
3971 * state.
3972 */
3973 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3974 return -EIO;
3975
3976 return 0;
3977}
3978
3979int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3980{
3981 int raid_type;
3982 int min_tolerated = INT_MAX;
3983
3984 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3985 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3986 min_tolerated = min_t(int, min_tolerated,
3987 btrfs_raid_array[BTRFS_RAID_SINGLE].
3988 tolerated_failures);
3989
3990 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3991 if (raid_type == BTRFS_RAID_SINGLE)
3992 continue;
3993 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3994 continue;
3995 min_tolerated = min_t(int, min_tolerated,
3996 btrfs_raid_array[raid_type].
3997 tolerated_failures);
3998 }
3999
4000 if (min_tolerated == INT_MAX) {
4001 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4002 min_tolerated = 0;
4003 }
4004
4005 return min_tolerated;
4006}
4007
4008int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4009{
4010 struct list_head *head;
4011 struct btrfs_device *dev;
4012 struct btrfs_super_block *sb;
4013 struct btrfs_dev_item *dev_item;
4014 int ret;
4015 int do_barriers;
4016 int max_errors;
4017 int total_errors = 0;
4018 u64 flags;
4019
4020 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4021
4022 /*
4023 * max_mirrors == 0 indicates we're from commit_transaction,
4024 * not from fsync where the tree roots in fs_info have not
4025 * been consistent on disk.
4026 */
4027 if (max_mirrors == 0)
4028 backup_super_roots(fs_info);
4029
4030 sb = fs_info->super_for_commit;
4031 dev_item = &sb->dev_item;
4032
4033 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4034 head = &fs_info->fs_devices->devices;
4035 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4036
4037 if (do_barriers) {
4038 ret = barrier_all_devices(fs_info);
4039 if (ret) {
4040 mutex_unlock(
4041 &fs_info->fs_devices->device_list_mutex);
4042 btrfs_handle_fs_error(fs_info, ret,
4043 "errors while submitting device barriers.");
4044 return ret;
4045 }
4046 }
4047
4048 list_for_each_entry(dev, head, dev_list) {
4049 if (!dev->bdev) {
4050 total_errors++;
4051 continue;
4052 }
4053 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4054 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4055 continue;
4056
4057 btrfs_set_stack_device_generation(dev_item, 0);
4058 btrfs_set_stack_device_type(dev_item, dev->type);
4059 btrfs_set_stack_device_id(dev_item, dev->devid);
4060 btrfs_set_stack_device_total_bytes(dev_item,
4061 dev->commit_total_bytes);
4062 btrfs_set_stack_device_bytes_used(dev_item,
4063 dev->commit_bytes_used);
4064 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4065 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4066 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4067 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4068 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4069 BTRFS_FSID_SIZE);
4070
4071 flags = btrfs_super_flags(sb);
4072 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4073
4074 ret = btrfs_validate_write_super(fs_info, sb);
4075 if (ret < 0) {
4076 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4077 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4078 "unexpected superblock corruption detected");
4079 return -EUCLEAN;
4080 }
4081
4082 ret = write_dev_supers(dev, sb, max_mirrors);
4083 if (ret)
4084 total_errors++;
4085 }
4086 if (total_errors > max_errors) {
4087 btrfs_err(fs_info, "%d errors while writing supers",
4088 total_errors);
4089 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4090
4091 /* FUA is masked off if unsupported and can't be the reason */
4092 btrfs_handle_fs_error(fs_info, -EIO,
4093 "%d errors while writing supers",
4094 total_errors);
4095 return -EIO;
4096 }
4097
4098 total_errors = 0;
4099 list_for_each_entry(dev, head, dev_list) {
4100 if (!dev->bdev)
4101 continue;
4102 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4103 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4104 continue;
4105
4106 ret = wait_dev_supers(dev, max_mirrors);
4107 if (ret)
4108 total_errors++;
4109 }
4110 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4111 if (total_errors > max_errors) {
4112 btrfs_handle_fs_error(fs_info, -EIO,
4113 "%d errors while writing supers",
4114 total_errors);
4115 return -EIO;
4116 }
4117 return 0;
4118}
4119
4120/* Drop a fs root from the radix tree and free it. */
4121void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4122 struct btrfs_root *root)
4123{
4124 bool drop_ref = false;
4125
4126 spin_lock(&fs_info->fs_roots_radix_lock);
4127 radix_tree_delete(&fs_info->fs_roots_radix,
4128 (unsigned long)btrfs_root_id(root));
4129 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4130 drop_ref = true;
4131 spin_unlock(&fs_info->fs_roots_radix_lock);
4132
4133 if (BTRFS_FS_ERROR(fs_info)) {
4134 ASSERT(root->log_root == NULL);
4135 if (root->reloc_root) {
4136 btrfs_put_root(root->reloc_root);
4137 root->reloc_root = NULL;
4138 }
4139 }
4140
4141 if (drop_ref)
4142 btrfs_put_root(root);
4143}
4144
4145int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4146{
4147 mutex_lock(&fs_info->cleaner_mutex);
4148 btrfs_run_delayed_iputs(fs_info);
4149 mutex_unlock(&fs_info->cleaner_mutex);
4150 wake_up_process(fs_info->cleaner_kthread);
4151
4152 /* wait until ongoing cleanup work done */
4153 down_write(&fs_info->cleanup_work_sem);
4154 up_write(&fs_info->cleanup_work_sem);
4155
4156 return btrfs_commit_current_transaction(fs_info->tree_root);
4157}
4158
4159static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4160{
4161 struct btrfs_transaction *trans;
4162 struct btrfs_transaction *tmp;
4163 bool found = false;
4164
4165 /*
4166 * This function is only called at the very end of close_ctree(),
4167 * thus no other running transaction, no need to take trans_lock.
4168 */
4169 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4170 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4171 struct extent_state *cached = NULL;
4172 u64 dirty_bytes = 0;
4173 u64 cur = 0;
4174 u64 found_start;
4175 u64 found_end;
4176
4177 found = true;
4178 while (find_first_extent_bit(&trans->dirty_pages, cur,
4179 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4180 dirty_bytes += found_end + 1 - found_start;
4181 cur = found_end + 1;
4182 }
4183 btrfs_warn(fs_info,
4184 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4185 trans->transid, dirty_bytes);
4186 btrfs_cleanup_one_transaction(trans);
4187
4188 if (trans == fs_info->running_transaction)
4189 fs_info->running_transaction = NULL;
4190 list_del_init(&trans->list);
4191
4192 btrfs_put_transaction(trans);
4193 trace_btrfs_transaction_commit(fs_info);
4194 }
4195 ASSERT(!found);
4196}
4197
4198void __cold close_ctree(struct btrfs_fs_info *fs_info)
4199{
4200 int ret;
4201
4202 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4203
4204 /*
4205 * If we had UNFINISHED_DROPS we could still be processing them, so
4206 * clear that bit and wake up relocation so it can stop.
4207 * We must do this before stopping the block group reclaim task, because
4208 * at btrfs_relocate_block_group() we wait for this bit, and after the
4209 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4210 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4211 * return 1.
4212 */
4213 btrfs_wake_unfinished_drop(fs_info);
4214
4215 /*
4216 * We may have the reclaim task running and relocating a data block group,
4217 * in which case it may create delayed iputs. So stop it before we park
4218 * the cleaner kthread otherwise we can get new delayed iputs after
4219 * parking the cleaner, and that can make the async reclaim task to hang
4220 * if it's waiting for delayed iputs to complete, since the cleaner is
4221 * parked and can not run delayed iputs - this will make us hang when
4222 * trying to stop the async reclaim task.
4223 */
4224 cancel_work_sync(&fs_info->reclaim_bgs_work);
4225 /*
4226 * We don't want the cleaner to start new transactions, add more delayed
4227 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4228 * because that frees the task_struct, and the transaction kthread might
4229 * still try to wake up the cleaner.
4230 */
4231 kthread_park(fs_info->cleaner_kthread);
4232
4233 /* wait for the qgroup rescan worker to stop */
4234 btrfs_qgroup_wait_for_completion(fs_info, false);
4235
4236 /* wait for the uuid_scan task to finish */
4237 down(&fs_info->uuid_tree_rescan_sem);
4238 /* avoid complains from lockdep et al., set sem back to initial state */
4239 up(&fs_info->uuid_tree_rescan_sem);
4240
4241 /* pause restriper - we want to resume on mount */
4242 btrfs_pause_balance(fs_info);
4243
4244 btrfs_dev_replace_suspend_for_unmount(fs_info);
4245
4246 btrfs_scrub_cancel(fs_info);
4247
4248 /* wait for any defraggers to finish */
4249 wait_event(fs_info->transaction_wait,
4250 (atomic_read(&fs_info->defrag_running) == 0));
4251
4252 /* clear out the rbtree of defraggable inodes */
4253 btrfs_cleanup_defrag_inodes(fs_info);
4254
4255 /*
4256 * Wait for any fixup workers to complete.
4257 * If we don't wait for them here and they are still running by the time
4258 * we call kthread_stop() against the cleaner kthread further below, we
4259 * get an use-after-free on the cleaner because the fixup worker adds an
4260 * inode to the list of delayed iputs and then attempts to wakeup the
4261 * cleaner kthread, which was already stopped and destroyed. We parked
4262 * already the cleaner, but below we run all pending delayed iputs.
4263 */
4264 btrfs_flush_workqueue(fs_info->fixup_workers);
4265 /*
4266 * Similar case here, we have to wait for delalloc workers before we
4267 * proceed below and stop the cleaner kthread, otherwise we trigger a
4268 * use-after-tree on the cleaner kthread task_struct when a delalloc
4269 * worker running submit_compressed_extents() adds a delayed iput, which
4270 * does a wake up on the cleaner kthread, which was already freed below
4271 * when we call kthread_stop().
4272 */
4273 btrfs_flush_workqueue(fs_info->delalloc_workers);
4274
4275 /*
4276 * After we parked the cleaner kthread, ordered extents may have
4277 * completed and created new delayed iputs. If one of the async reclaim
4278 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4279 * can hang forever trying to stop it, because if a delayed iput is
4280 * added after it ran btrfs_run_delayed_iputs() and before it called
4281 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4282 * no one else to run iputs.
4283 *
4284 * So wait for all ongoing ordered extents to complete and then run
4285 * delayed iputs. This works because once we reach this point no one
4286 * can either create new ordered extents nor create delayed iputs
4287 * through some other means.
4288 *
4289 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4290 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4291 * but the delayed iput for the respective inode is made only when doing
4292 * the final btrfs_put_ordered_extent() (which must happen at
4293 * btrfs_finish_ordered_io() when we are unmounting).
4294 */
4295 btrfs_flush_workqueue(fs_info->endio_write_workers);
4296 /* Ordered extents for free space inodes. */
4297 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4298 btrfs_run_delayed_iputs(fs_info);
4299
4300 cancel_work_sync(&fs_info->async_reclaim_work);
4301 cancel_work_sync(&fs_info->async_data_reclaim_work);
4302 cancel_work_sync(&fs_info->preempt_reclaim_work);
4303 cancel_work_sync(&fs_info->em_shrinker_work);
4304
4305 /* Cancel or finish ongoing discard work */
4306 btrfs_discard_cleanup(fs_info);
4307
4308 if (!sb_rdonly(fs_info->sb)) {
4309 /*
4310 * The cleaner kthread is stopped, so do one final pass over
4311 * unused block groups.
4312 */
4313 btrfs_delete_unused_bgs(fs_info);
4314
4315 /*
4316 * There might be existing delayed inode workers still running
4317 * and holding an empty delayed inode item. We must wait for
4318 * them to complete first because they can create a transaction.
4319 * This happens when someone calls btrfs_balance_delayed_items()
4320 * and then a transaction commit runs the same delayed nodes
4321 * before any delayed worker has done something with the nodes.
4322 * We must wait for any worker here and not at transaction
4323 * commit time since that could cause a deadlock.
4324 * This is a very rare case.
4325 */
4326 btrfs_flush_workqueue(fs_info->delayed_workers);
4327
4328 ret = btrfs_commit_super(fs_info);
4329 if (ret)
4330 btrfs_err(fs_info, "commit super ret %d", ret);
4331 }
4332
4333 if (BTRFS_FS_ERROR(fs_info))
4334 btrfs_error_commit_super(fs_info);
4335
4336 kthread_stop(fs_info->transaction_kthread);
4337 kthread_stop(fs_info->cleaner_kthread);
4338
4339 ASSERT(list_empty(&fs_info->delayed_iputs));
4340 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4341
4342 if (btrfs_check_quota_leak(fs_info)) {
4343 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4344 btrfs_err(fs_info, "qgroup reserved space leaked");
4345 }
4346
4347 btrfs_free_qgroup_config(fs_info);
4348 ASSERT(list_empty(&fs_info->delalloc_roots));
4349
4350 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4351 btrfs_info(fs_info, "at unmount delalloc count %lld",
4352 percpu_counter_sum(&fs_info->delalloc_bytes));
4353 }
4354
4355 if (percpu_counter_sum(&fs_info->ordered_bytes))
4356 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4357 percpu_counter_sum(&fs_info->ordered_bytes));
4358
4359 btrfs_sysfs_remove_mounted(fs_info);
4360 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4361
4362 btrfs_put_block_group_cache(fs_info);
4363
4364 /*
4365 * we must make sure there is not any read request to
4366 * submit after we stopping all workers.
4367 */
4368 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4369 btrfs_stop_all_workers(fs_info);
4370
4371 /* We shouldn't have any transaction open at this point */
4372 warn_about_uncommitted_trans(fs_info);
4373
4374 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4375 free_root_pointers(fs_info, true);
4376 btrfs_free_fs_roots(fs_info);
4377
4378 /*
4379 * We must free the block groups after dropping the fs_roots as we could
4380 * have had an IO error and have left over tree log blocks that aren't
4381 * cleaned up until the fs roots are freed. This makes the block group
4382 * accounting appear to be wrong because there's pending reserved bytes,
4383 * so make sure we do the block group cleanup afterwards.
4384 */
4385 btrfs_free_block_groups(fs_info);
4386
4387 iput(fs_info->btree_inode);
4388
4389 btrfs_mapping_tree_free(fs_info);
4390 btrfs_close_devices(fs_info->fs_devices);
4391}
4392
4393void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4394 struct extent_buffer *buf)
4395{
4396 struct btrfs_fs_info *fs_info = buf->fs_info;
4397 u64 transid = btrfs_header_generation(buf);
4398
4399#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4400 /*
4401 * This is a fast path so only do this check if we have sanity tests
4402 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4403 * outside of the sanity tests.
4404 */
4405 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4406 return;
4407#endif
4408 /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4409 ASSERT(trans->transid == fs_info->generation);
4410 btrfs_assert_tree_write_locked(buf);
4411 if (unlikely(transid != fs_info->generation)) {
4412 btrfs_abort_transaction(trans, -EUCLEAN);
4413 btrfs_crit(fs_info,
4414"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4415 buf->start, transid, fs_info->generation);
4416 }
4417 set_extent_buffer_dirty(buf);
4418}
4419
4420static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4421 int flush_delayed)
4422{
4423 /*
4424 * looks as though older kernels can get into trouble with
4425 * this code, they end up stuck in balance_dirty_pages forever
4426 */
4427 int ret;
4428
4429 if (current->flags & PF_MEMALLOC)
4430 return;
4431
4432 if (flush_delayed)
4433 btrfs_balance_delayed_items(fs_info);
4434
4435 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4436 BTRFS_DIRTY_METADATA_THRESH,
4437 fs_info->dirty_metadata_batch);
4438 if (ret > 0) {
4439 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4440 }
4441}
4442
4443void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4444{
4445 __btrfs_btree_balance_dirty(fs_info, 1);
4446}
4447
4448void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4449{
4450 __btrfs_btree_balance_dirty(fs_info, 0);
4451}
4452
4453static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4454{
4455 /* cleanup FS via transaction */
4456 btrfs_cleanup_transaction(fs_info);
4457
4458 mutex_lock(&fs_info->cleaner_mutex);
4459 btrfs_run_delayed_iputs(fs_info);
4460 mutex_unlock(&fs_info->cleaner_mutex);
4461
4462 down_write(&fs_info->cleanup_work_sem);
4463 up_write(&fs_info->cleanup_work_sem);
4464}
4465
4466static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4467{
4468 struct btrfs_root *gang[8];
4469 u64 root_objectid = 0;
4470 int ret;
4471
4472 spin_lock(&fs_info->fs_roots_radix_lock);
4473 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4474 (void **)gang, root_objectid,
4475 ARRAY_SIZE(gang))) != 0) {
4476 int i;
4477
4478 for (i = 0; i < ret; i++)
4479 gang[i] = btrfs_grab_root(gang[i]);
4480 spin_unlock(&fs_info->fs_roots_radix_lock);
4481
4482 for (i = 0; i < ret; i++) {
4483 if (!gang[i])
4484 continue;
4485 root_objectid = btrfs_root_id(gang[i]);
4486 btrfs_free_log(NULL, gang[i]);
4487 btrfs_put_root(gang[i]);
4488 }
4489 root_objectid++;
4490 spin_lock(&fs_info->fs_roots_radix_lock);
4491 }
4492 spin_unlock(&fs_info->fs_roots_radix_lock);
4493 btrfs_free_log_root_tree(NULL, fs_info);
4494}
4495
4496static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4497{
4498 struct btrfs_ordered_extent *ordered;
4499
4500 spin_lock(&root->ordered_extent_lock);
4501 /*
4502 * This will just short circuit the ordered completion stuff which will
4503 * make sure the ordered extent gets properly cleaned up.
4504 */
4505 list_for_each_entry(ordered, &root->ordered_extents,
4506 root_extent_list)
4507 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4508 spin_unlock(&root->ordered_extent_lock);
4509}
4510
4511static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4512{
4513 struct btrfs_root *root;
4514 LIST_HEAD(splice);
4515
4516 spin_lock(&fs_info->ordered_root_lock);
4517 list_splice_init(&fs_info->ordered_roots, &splice);
4518 while (!list_empty(&splice)) {
4519 root = list_first_entry(&splice, struct btrfs_root,
4520 ordered_root);
4521 list_move_tail(&root->ordered_root,
4522 &fs_info->ordered_roots);
4523
4524 spin_unlock(&fs_info->ordered_root_lock);
4525 btrfs_destroy_ordered_extents(root);
4526
4527 cond_resched();
4528 spin_lock(&fs_info->ordered_root_lock);
4529 }
4530 spin_unlock(&fs_info->ordered_root_lock);
4531
4532 /*
4533 * We need this here because if we've been flipped read-only we won't
4534 * get sync() from the umount, so we need to make sure any ordered
4535 * extents that haven't had their dirty pages IO start writeout yet
4536 * actually get run and error out properly.
4537 */
4538 btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
4539}
4540
4541static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4542{
4543 struct btrfs_inode *btrfs_inode;
4544 LIST_HEAD(splice);
4545
4546 spin_lock(&root->delalloc_lock);
4547 list_splice_init(&root->delalloc_inodes, &splice);
4548
4549 while (!list_empty(&splice)) {
4550 struct inode *inode = NULL;
4551 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4552 delalloc_inodes);
4553 btrfs_del_delalloc_inode(btrfs_inode);
4554 spin_unlock(&root->delalloc_lock);
4555
4556 /*
4557 * Make sure we get a live inode and that it'll not disappear
4558 * meanwhile.
4559 */
4560 inode = igrab(&btrfs_inode->vfs_inode);
4561 if (inode) {
4562 unsigned int nofs_flag;
4563
4564 nofs_flag = memalloc_nofs_save();
4565 invalidate_inode_pages2(inode->i_mapping);
4566 memalloc_nofs_restore(nofs_flag);
4567 iput(inode);
4568 }
4569 spin_lock(&root->delalloc_lock);
4570 }
4571 spin_unlock(&root->delalloc_lock);
4572}
4573
4574static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4575{
4576 struct btrfs_root *root;
4577 LIST_HEAD(splice);
4578
4579 spin_lock(&fs_info->delalloc_root_lock);
4580 list_splice_init(&fs_info->delalloc_roots, &splice);
4581 while (!list_empty(&splice)) {
4582 root = list_first_entry(&splice, struct btrfs_root,
4583 delalloc_root);
4584 root = btrfs_grab_root(root);
4585 BUG_ON(!root);
4586 spin_unlock(&fs_info->delalloc_root_lock);
4587
4588 btrfs_destroy_delalloc_inodes(root);
4589 btrfs_put_root(root);
4590
4591 spin_lock(&fs_info->delalloc_root_lock);
4592 }
4593 spin_unlock(&fs_info->delalloc_root_lock);
4594}
4595
4596static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4597 struct extent_io_tree *dirty_pages,
4598 int mark)
4599{
4600 struct extent_buffer *eb;
4601 u64 start = 0;
4602 u64 end;
4603
4604 while (find_first_extent_bit(dirty_pages, start, &start, &end,
4605 mark, NULL)) {
4606 clear_extent_bits(dirty_pages, start, end, mark);
4607 while (start <= end) {
4608 eb = find_extent_buffer(fs_info, start);
4609 start += fs_info->nodesize;
4610 if (!eb)
4611 continue;
4612
4613 btrfs_tree_lock(eb);
4614 wait_on_extent_buffer_writeback(eb);
4615 btrfs_clear_buffer_dirty(NULL, eb);
4616 btrfs_tree_unlock(eb);
4617
4618 free_extent_buffer_stale(eb);
4619 }
4620 }
4621}
4622
4623static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4624 struct extent_io_tree *unpin)
4625{
4626 u64 start;
4627 u64 end;
4628
4629 while (1) {
4630 struct extent_state *cached_state = NULL;
4631
4632 /*
4633 * The btrfs_finish_extent_commit() may get the same range as
4634 * ours between find_first_extent_bit and clear_extent_dirty.
4635 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4636 * the same extent range.
4637 */
4638 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4639 if (!find_first_extent_bit(unpin, 0, &start, &end,
4640 EXTENT_DIRTY, &cached_state)) {
4641 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4642 break;
4643 }
4644
4645 clear_extent_dirty(unpin, start, end, &cached_state);
4646 free_extent_state(cached_state);
4647 btrfs_error_unpin_extent_range(fs_info, start, end);
4648 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4649 cond_resched();
4650 }
4651}
4652
4653static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4654{
4655 struct inode *inode;
4656
4657 inode = cache->io_ctl.inode;
4658 if (inode) {
4659 unsigned int nofs_flag;
4660
4661 nofs_flag = memalloc_nofs_save();
4662 invalidate_inode_pages2(inode->i_mapping);
4663 memalloc_nofs_restore(nofs_flag);
4664
4665 BTRFS_I(inode)->generation = 0;
4666 cache->io_ctl.inode = NULL;
4667 iput(inode);
4668 }
4669 ASSERT(cache->io_ctl.pages == NULL);
4670 btrfs_put_block_group(cache);
4671}
4672
4673void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4674 struct btrfs_fs_info *fs_info)
4675{
4676 struct btrfs_block_group *cache;
4677
4678 spin_lock(&cur_trans->dirty_bgs_lock);
4679 while (!list_empty(&cur_trans->dirty_bgs)) {
4680 cache = list_first_entry(&cur_trans->dirty_bgs,
4681 struct btrfs_block_group,
4682 dirty_list);
4683
4684 if (!list_empty(&cache->io_list)) {
4685 spin_unlock(&cur_trans->dirty_bgs_lock);
4686 list_del_init(&cache->io_list);
4687 btrfs_cleanup_bg_io(cache);
4688 spin_lock(&cur_trans->dirty_bgs_lock);
4689 }
4690
4691 list_del_init(&cache->dirty_list);
4692 spin_lock(&cache->lock);
4693 cache->disk_cache_state = BTRFS_DC_ERROR;
4694 spin_unlock(&cache->lock);
4695
4696 spin_unlock(&cur_trans->dirty_bgs_lock);
4697 btrfs_put_block_group(cache);
4698 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4699 spin_lock(&cur_trans->dirty_bgs_lock);
4700 }
4701 spin_unlock(&cur_trans->dirty_bgs_lock);
4702
4703 /*
4704 * Refer to the definition of io_bgs member for details why it's safe
4705 * to use it without any locking
4706 */
4707 while (!list_empty(&cur_trans->io_bgs)) {
4708 cache = list_first_entry(&cur_trans->io_bgs,
4709 struct btrfs_block_group,
4710 io_list);
4711
4712 list_del_init(&cache->io_list);
4713 spin_lock(&cache->lock);
4714 cache->disk_cache_state = BTRFS_DC_ERROR;
4715 spin_unlock(&cache->lock);
4716 btrfs_cleanup_bg_io(cache);
4717 }
4718}
4719
4720static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4721{
4722 struct btrfs_root *gang[8];
4723 int i;
4724 int ret;
4725
4726 spin_lock(&fs_info->fs_roots_radix_lock);
4727 while (1) {
4728 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4729 (void **)gang, 0,
4730 ARRAY_SIZE(gang),
4731 BTRFS_ROOT_TRANS_TAG);
4732 if (ret == 0)
4733 break;
4734 for (i = 0; i < ret; i++) {
4735 struct btrfs_root *root = gang[i];
4736
4737 btrfs_qgroup_free_meta_all_pertrans(root);
4738 radix_tree_tag_clear(&fs_info->fs_roots_radix,
4739 (unsigned long)btrfs_root_id(root),
4740 BTRFS_ROOT_TRANS_TAG);
4741 }
4742 }
4743 spin_unlock(&fs_info->fs_roots_radix_lock);
4744}
4745
4746void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans)
4747{
4748 struct btrfs_fs_info *fs_info = cur_trans->fs_info;
4749 struct btrfs_device *dev, *tmp;
4750
4751 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4752 ASSERT(list_empty(&cur_trans->dirty_bgs));
4753 ASSERT(list_empty(&cur_trans->io_bgs));
4754
4755 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4756 post_commit_list) {
4757 list_del_init(&dev->post_commit_list);
4758 }
4759
4760 btrfs_destroy_delayed_refs(cur_trans);
4761
4762 cur_trans->state = TRANS_STATE_COMMIT_START;
4763 wake_up(&fs_info->transaction_blocked_wait);
4764
4765 cur_trans->state = TRANS_STATE_UNBLOCKED;
4766 wake_up(&fs_info->transaction_wait);
4767
4768 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4769 EXTENT_DIRTY);
4770 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4771
4772 cur_trans->state =TRANS_STATE_COMPLETED;
4773 wake_up(&cur_trans->commit_wait);
4774}
4775
4776static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4777{
4778 struct btrfs_transaction *t;
4779
4780 mutex_lock(&fs_info->transaction_kthread_mutex);
4781
4782 spin_lock(&fs_info->trans_lock);
4783 while (!list_empty(&fs_info->trans_list)) {
4784 t = list_first_entry(&fs_info->trans_list,
4785 struct btrfs_transaction, list);
4786 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4787 refcount_inc(&t->use_count);
4788 spin_unlock(&fs_info->trans_lock);
4789 btrfs_wait_for_commit(fs_info, t->transid);
4790 btrfs_put_transaction(t);
4791 spin_lock(&fs_info->trans_lock);
4792 continue;
4793 }
4794 if (t == fs_info->running_transaction) {
4795 t->state = TRANS_STATE_COMMIT_DOING;
4796 spin_unlock(&fs_info->trans_lock);
4797 /*
4798 * We wait for 0 num_writers since we don't hold a trans
4799 * handle open currently for this transaction.
4800 */
4801 wait_event(t->writer_wait,
4802 atomic_read(&t->num_writers) == 0);
4803 } else {
4804 spin_unlock(&fs_info->trans_lock);
4805 }
4806 btrfs_cleanup_one_transaction(t);
4807
4808 spin_lock(&fs_info->trans_lock);
4809 if (t == fs_info->running_transaction)
4810 fs_info->running_transaction = NULL;
4811 list_del_init(&t->list);
4812 spin_unlock(&fs_info->trans_lock);
4813
4814 btrfs_put_transaction(t);
4815 trace_btrfs_transaction_commit(fs_info);
4816 spin_lock(&fs_info->trans_lock);
4817 }
4818 spin_unlock(&fs_info->trans_lock);
4819 btrfs_destroy_all_ordered_extents(fs_info);
4820 btrfs_destroy_delayed_inodes(fs_info);
4821 btrfs_assert_delayed_root_empty(fs_info);
4822 btrfs_destroy_all_delalloc_inodes(fs_info);
4823 btrfs_drop_all_logs(fs_info);
4824 btrfs_free_all_qgroup_pertrans(fs_info);
4825 mutex_unlock(&fs_info->transaction_kthread_mutex);
4826
4827 return 0;
4828}
4829
4830int btrfs_init_root_free_objectid(struct btrfs_root *root)
4831{
4832 struct btrfs_path *path;
4833 int ret;
4834 struct extent_buffer *l;
4835 struct btrfs_key search_key;
4836 struct btrfs_key found_key;
4837 int slot;
4838
4839 path = btrfs_alloc_path();
4840 if (!path)
4841 return -ENOMEM;
4842
4843 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4844 search_key.type = -1;
4845 search_key.offset = (u64)-1;
4846 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4847 if (ret < 0)
4848 goto error;
4849 if (ret == 0) {
4850 /*
4851 * Key with offset -1 found, there would have to exist a root
4852 * with such id, but this is out of valid range.
4853 */
4854 ret = -EUCLEAN;
4855 goto error;
4856 }
4857 if (path->slots[0] > 0) {
4858 slot = path->slots[0] - 1;
4859 l = path->nodes[0];
4860 btrfs_item_key_to_cpu(l, &found_key, slot);
4861 root->free_objectid = max_t(u64, found_key.objectid + 1,
4862 BTRFS_FIRST_FREE_OBJECTID);
4863 } else {
4864 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4865 }
4866 ret = 0;
4867error:
4868 btrfs_free_path(path);
4869 return ret;
4870}
4871
4872int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4873{
4874 int ret;
4875 mutex_lock(&root->objectid_mutex);
4876
4877 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4878 btrfs_warn(root->fs_info,
4879 "the objectid of root %llu reaches its highest value",
4880 btrfs_root_id(root));
4881 ret = -ENOSPC;
4882 goto out;
4883 }
4884
4885 *objectid = root->free_objectid++;
4886 ret = 0;
4887out:
4888 mutex_unlock(&root->objectid_mutex);
4889 return ret;
4890}