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 <asm/unaligned.h>
21#include <crypto/hash.h>
22#include "ctree.h"
23#include "disk-io.h"
24#include "transaction.h"
25#include "btrfs_inode.h"
26#include "volumes.h"
27#include "print-tree.h"
28#include "locking.h"
29#include "tree-log.h"
30#include "free-space-cache.h"
31#include "free-space-tree.h"
32#include "check-integrity.h"
33#include "rcu-string.h"
34#include "dev-replace.h"
35#include "raid56.h"
36#include "sysfs.h"
37#include "qgroup.h"
38#include "compression.h"
39#include "tree-checker.h"
40#include "ref-verify.h"
41#include "block-group.h"
42#include "discard.h"
43#include "space-info.h"
44#include "zoned.h"
45#include "subpage.h"
46
47#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
48 BTRFS_HEADER_FLAG_RELOC |\
49 BTRFS_SUPER_FLAG_ERROR |\
50 BTRFS_SUPER_FLAG_SEEDING |\
51 BTRFS_SUPER_FLAG_METADUMP |\
52 BTRFS_SUPER_FLAG_METADUMP_V2)
53
54static void end_workqueue_fn(struct btrfs_work *work);
55static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_fs_info *fs_info);
58static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 struct extent_io_tree *dirty_pages,
61 int mark);
62static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *pinned_extents);
64static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66
67/*
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
71 */
72struct btrfs_end_io_wq {
73 struct bio *bio;
74 bio_end_io_t *end_io;
75 void *private;
76 struct btrfs_fs_info *info;
77 blk_status_t status;
78 enum btrfs_wq_endio_type metadata;
79 struct btrfs_work work;
80};
81
82static struct kmem_cache *btrfs_end_io_wq_cache;
83
84int __init btrfs_end_io_wq_init(void)
85{
86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq),
88 0,
89 SLAB_MEM_SPREAD,
90 NULL);
91 if (!btrfs_end_io_wq_cache)
92 return -ENOMEM;
93 return 0;
94}
95
96void __cold btrfs_end_io_wq_exit(void)
97{
98 kmem_cache_destroy(btrfs_end_io_wq_cache);
99}
100
101static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102{
103 if (fs_info->csum_shash)
104 crypto_free_shash(fs_info->csum_shash);
105}
106
107/*
108 * async submit bios are used to offload expensive checksumming
109 * onto the worker threads. They checksum file and metadata bios
110 * just before they are sent down the IO stack.
111 */
112struct async_submit_bio {
113 struct inode *inode;
114 struct bio *bio;
115 extent_submit_bio_start_t *submit_bio_start;
116 int mirror_num;
117
118 /* Optional parameter for submit_bio_start used by direct io */
119 u64 dio_file_offset;
120 struct btrfs_work work;
121 blk_status_t status;
122};
123
124/*
125 * Lockdep class keys for extent_buffer->lock's in this root. For a given
126 * eb, the lockdep key is determined by the btrfs_root it belongs to and
127 * the level the eb occupies in the tree.
128 *
129 * Different roots are used for different purposes and may nest inside each
130 * other and they require separate keysets. As lockdep keys should be
131 * static, assign keysets according to the purpose of the root as indicated
132 * by btrfs_root->root_key.objectid. This ensures that all special purpose
133 * roots have separate keysets.
134 *
135 * Lock-nesting across peer nodes is always done with the immediate parent
136 * node locked thus preventing deadlock. As lockdep doesn't know this, use
137 * subclass to avoid triggering lockdep warning in such cases.
138 *
139 * The key is set by the readpage_end_io_hook after the buffer has passed
140 * csum validation but before the pages are unlocked. It is also set by
141 * btrfs_init_new_buffer on freshly allocated blocks.
142 *
143 * We also add a check to make sure the highest level of the tree is the
144 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
145 * needs update as well.
146 */
147#ifdef CONFIG_DEBUG_LOCK_ALLOC
148# if BTRFS_MAX_LEVEL != 8
149# error
150# endif
151
152#define DEFINE_LEVEL(stem, level) \
153 .names[level] = "btrfs-" stem "-0" #level,
154
155#define DEFINE_NAME(stem) \
156 DEFINE_LEVEL(stem, 0) \
157 DEFINE_LEVEL(stem, 1) \
158 DEFINE_LEVEL(stem, 2) \
159 DEFINE_LEVEL(stem, 3) \
160 DEFINE_LEVEL(stem, 4) \
161 DEFINE_LEVEL(stem, 5) \
162 DEFINE_LEVEL(stem, 6) \
163 DEFINE_LEVEL(stem, 7)
164
165static struct btrfs_lockdep_keyset {
166 u64 id; /* root objectid */
167 /* Longest entry: btrfs-free-space-00 */
168 char names[BTRFS_MAX_LEVEL][20];
169 struct lock_class_key keys[BTRFS_MAX_LEVEL];
170} btrfs_lockdep_keysets[] = {
171 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") },
172 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") },
173 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") },
174 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") },
175 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
176 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
177 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
178 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
179 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
180 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
181 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
182 { .id = 0, DEFINE_NAME("tree") },
183};
184
185#undef DEFINE_LEVEL
186#undef DEFINE_NAME
187
188void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
189 int level)
190{
191 struct btrfs_lockdep_keyset *ks;
192
193 BUG_ON(level >= ARRAY_SIZE(ks->keys));
194
195 /* find the matching keyset, id 0 is the default entry */
196 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
197 if (ks->id == objectid)
198 break;
199
200 lockdep_set_class_and_name(&eb->lock,
201 &ks->keys[level], ks->names[level]);
202}
203
204#endif
205
206/*
207 * Compute the csum of a btree block and store the result to provided buffer.
208 */
209static void csum_tree_block(struct extent_buffer *buf, u8 *result)
210{
211 struct btrfs_fs_info *fs_info = buf->fs_info;
212 const int num_pages = num_extent_pages(buf);
213 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
214 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
215 char *kaddr;
216 int i;
217
218 shash->tfm = fs_info->csum_shash;
219 crypto_shash_init(shash);
220 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
221 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
222 first_page_part - BTRFS_CSUM_SIZE);
223
224 for (i = 1; i < num_pages; i++) {
225 kaddr = page_address(buf->pages[i]);
226 crypto_shash_update(shash, kaddr, PAGE_SIZE);
227 }
228 memset(result, 0, BTRFS_CSUM_SIZE);
229 crypto_shash_final(shash, result);
230}
231
232/*
233 * we can't consider a given block up to date unless the transid of the
234 * block matches the transid in the parent node's pointer. This is how we
235 * detect blocks that either didn't get written at all or got written
236 * in the wrong place.
237 */
238static int verify_parent_transid(struct extent_io_tree *io_tree,
239 struct extent_buffer *eb, u64 parent_transid,
240 int atomic)
241{
242 struct extent_state *cached_state = NULL;
243 int ret;
244
245 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
246 return 0;
247
248 if (atomic)
249 return -EAGAIN;
250
251 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
252 &cached_state);
253 if (extent_buffer_uptodate(eb) &&
254 btrfs_header_generation(eb) == parent_transid) {
255 ret = 0;
256 goto out;
257 }
258 btrfs_err_rl(eb->fs_info,
259 "parent transid verify failed on %llu wanted %llu found %llu",
260 eb->start,
261 parent_transid, btrfs_header_generation(eb));
262 ret = 1;
263 clear_extent_buffer_uptodate(eb);
264out:
265 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
266 &cached_state);
267 return ret;
268}
269
270static bool btrfs_supported_super_csum(u16 csum_type)
271{
272 switch (csum_type) {
273 case BTRFS_CSUM_TYPE_CRC32:
274 case BTRFS_CSUM_TYPE_XXHASH:
275 case BTRFS_CSUM_TYPE_SHA256:
276 case BTRFS_CSUM_TYPE_BLAKE2:
277 return true;
278 default:
279 return false;
280 }
281}
282
283/*
284 * Return 0 if the superblock checksum type matches the checksum value of that
285 * algorithm. Pass the raw disk superblock data.
286 */
287static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
288 char *raw_disk_sb)
289{
290 struct btrfs_super_block *disk_sb =
291 (struct btrfs_super_block *)raw_disk_sb;
292 char result[BTRFS_CSUM_SIZE];
293 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
294
295 shash->tfm = fs_info->csum_shash;
296
297 /*
298 * The super_block structure does not span the whole
299 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
300 * filled with zeros and is included in the checksum.
301 */
302 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
303 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
304
305 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
306 return 1;
307
308 return 0;
309}
310
311int btrfs_verify_level_key(struct extent_buffer *eb, int level,
312 struct btrfs_key *first_key, u64 parent_transid)
313{
314 struct btrfs_fs_info *fs_info = eb->fs_info;
315 int found_level;
316 struct btrfs_key found_key;
317 int ret;
318
319 found_level = btrfs_header_level(eb);
320 if (found_level != level) {
321 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
322 KERN_ERR "BTRFS: tree level check failed\n");
323 btrfs_err(fs_info,
324"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
325 eb->start, level, found_level);
326 return -EIO;
327 }
328
329 if (!first_key)
330 return 0;
331
332 /*
333 * For live tree block (new tree blocks in current transaction),
334 * we need proper lock context to avoid race, which is impossible here.
335 * So we only checks tree blocks which is read from disk, whose
336 * generation <= fs_info->last_trans_committed.
337 */
338 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
339 return 0;
340
341 /* We have @first_key, so this @eb must have at least one item */
342 if (btrfs_header_nritems(eb) == 0) {
343 btrfs_err(fs_info,
344 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
345 eb->start);
346 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
347 return -EUCLEAN;
348 }
349
350 if (found_level)
351 btrfs_node_key_to_cpu(eb, &found_key, 0);
352 else
353 btrfs_item_key_to_cpu(eb, &found_key, 0);
354 ret = btrfs_comp_cpu_keys(first_key, &found_key);
355
356 if (ret) {
357 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
358 KERN_ERR "BTRFS: tree first key check failed\n");
359 btrfs_err(fs_info,
360"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
361 eb->start, parent_transid, first_key->objectid,
362 first_key->type, first_key->offset,
363 found_key.objectid, found_key.type,
364 found_key.offset);
365 }
366 return ret;
367}
368
369/*
370 * helper to read a given tree block, doing retries as required when
371 * the checksums don't match and we have alternate mirrors to try.
372 *
373 * @parent_transid: expected transid, skip check if 0
374 * @level: expected level, mandatory check
375 * @first_key: expected key of first slot, skip check if NULL
376 */
377static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
378 u64 parent_transid, int level,
379 struct btrfs_key *first_key)
380{
381 struct btrfs_fs_info *fs_info = eb->fs_info;
382 struct extent_io_tree *io_tree;
383 int failed = 0;
384 int ret;
385 int num_copies = 0;
386 int mirror_num = 0;
387 int failed_mirror = 0;
388
389 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
390 while (1) {
391 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
392 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
393 if (!ret) {
394 if (verify_parent_transid(io_tree, eb,
395 parent_transid, 0))
396 ret = -EIO;
397 else if (btrfs_verify_level_key(eb, level,
398 first_key, parent_transid))
399 ret = -EUCLEAN;
400 else
401 break;
402 }
403
404 num_copies = btrfs_num_copies(fs_info,
405 eb->start, eb->len);
406 if (num_copies == 1)
407 break;
408
409 if (!failed_mirror) {
410 failed = 1;
411 failed_mirror = eb->read_mirror;
412 }
413
414 mirror_num++;
415 if (mirror_num == failed_mirror)
416 mirror_num++;
417
418 if (mirror_num > num_copies)
419 break;
420 }
421
422 if (failed && !ret && failed_mirror)
423 btrfs_repair_eb_io_failure(eb, failed_mirror);
424
425 return ret;
426}
427
428static int csum_one_extent_buffer(struct extent_buffer *eb)
429{
430 struct btrfs_fs_info *fs_info = eb->fs_info;
431 u8 result[BTRFS_CSUM_SIZE];
432 int ret;
433
434 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
435 offsetof(struct btrfs_header, fsid),
436 BTRFS_FSID_SIZE) == 0);
437 csum_tree_block(eb, result);
438
439 if (btrfs_header_level(eb))
440 ret = btrfs_check_node(eb);
441 else
442 ret = btrfs_check_leaf_full(eb);
443
444 if (ret < 0) {
445 btrfs_print_tree(eb, 0);
446 btrfs_err(fs_info,
447 "block=%llu write time tree block corruption detected",
448 eb->start);
449 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
450 return ret;
451 }
452 write_extent_buffer(eb, result, 0, fs_info->csum_size);
453
454 return 0;
455}
456
457/* Checksum all dirty extent buffers in one bio_vec */
458static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
459 struct bio_vec *bvec)
460{
461 struct page *page = bvec->bv_page;
462 u64 bvec_start = page_offset(page) + bvec->bv_offset;
463 u64 cur;
464 int ret = 0;
465
466 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
467 cur += fs_info->nodesize) {
468 struct extent_buffer *eb;
469 bool uptodate;
470
471 eb = find_extent_buffer(fs_info, cur);
472 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
473 fs_info->nodesize);
474
475 /* A dirty eb shouldn't disappear from buffer_radix */
476 if (WARN_ON(!eb))
477 return -EUCLEAN;
478
479 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
480 free_extent_buffer(eb);
481 return -EUCLEAN;
482 }
483 if (WARN_ON(!uptodate)) {
484 free_extent_buffer(eb);
485 return -EUCLEAN;
486 }
487
488 ret = csum_one_extent_buffer(eb);
489 free_extent_buffer(eb);
490 if (ret < 0)
491 return ret;
492 }
493 return ret;
494}
495
496/*
497 * Checksum a dirty tree block before IO. This has extra checks to make sure
498 * we only fill in the checksum field in the first page of a multi-page block.
499 * For subpage extent buffers we need bvec to also read the offset in the page.
500 */
501static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
502{
503 struct page *page = bvec->bv_page;
504 u64 start = page_offset(page);
505 u64 found_start;
506 struct extent_buffer *eb;
507
508 if (fs_info->sectorsize < PAGE_SIZE)
509 return csum_dirty_subpage_buffers(fs_info, bvec);
510
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
513 return 0;
514
515 found_start = btrfs_header_bytenr(eb);
516
517 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
518 WARN_ON(found_start != 0);
519 return 0;
520 }
521
522 /*
523 * Please do not consolidate these warnings into a single if.
524 * It is useful to know what went wrong.
525 */
526 if (WARN_ON(found_start != start))
527 return -EUCLEAN;
528 if (WARN_ON(!PageUptodate(page)))
529 return -EUCLEAN;
530
531 return csum_one_extent_buffer(eb);
532}
533
534static int check_tree_block_fsid(struct extent_buffer *eb)
535{
536 struct btrfs_fs_info *fs_info = eb->fs_info;
537 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
538 u8 fsid[BTRFS_FSID_SIZE];
539 u8 *metadata_uuid;
540
541 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
542 BTRFS_FSID_SIZE);
543 /*
544 * Checking the incompat flag is only valid for the current fs. For
545 * seed devices it's forbidden to have their uuid changed so reading
546 * ->fsid in this case is fine
547 */
548 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
549 metadata_uuid = fs_devices->metadata_uuid;
550 else
551 metadata_uuid = fs_devices->fsid;
552
553 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
554 return 0;
555
556 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
557 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
558 return 0;
559
560 return 1;
561}
562
563/* Do basic extent buffer checks at read time */
564static int validate_extent_buffer(struct extent_buffer *eb)
565{
566 struct btrfs_fs_info *fs_info = eb->fs_info;
567 u64 found_start;
568 const u32 csum_size = fs_info->csum_size;
569 u8 found_level;
570 u8 result[BTRFS_CSUM_SIZE];
571 const u8 *header_csum;
572 int ret = 0;
573
574 found_start = btrfs_header_bytenr(eb);
575 if (found_start != eb->start) {
576 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
577 eb->start, found_start);
578 ret = -EIO;
579 goto out;
580 }
581 if (check_tree_block_fsid(eb)) {
582 btrfs_err_rl(fs_info, "bad fsid on block %llu",
583 eb->start);
584 ret = -EIO;
585 goto out;
586 }
587 found_level = btrfs_header_level(eb);
588 if (found_level >= BTRFS_MAX_LEVEL) {
589 btrfs_err(fs_info, "bad tree block level %d on %llu",
590 (int)btrfs_header_level(eb), eb->start);
591 ret = -EIO;
592 goto out;
593 }
594
595 csum_tree_block(eb, result);
596 header_csum = page_address(eb->pages[0]) +
597 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
598
599 if (memcmp(result, header_csum, csum_size) != 0) {
600 btrfs_warn_rl(fs_info,
601 "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
602 eb->start,
603 CSUM_FMT_VALUE(csum_size, header_csum),
604 CSUM_FMT_VALUE(csum_size, result),
605 btrfs_header_level(eb));
606 ret = -EUCLEAN;
607 goto out;
608 }
609
610 /*
611 * If this is a leaf block and it is corrupt, set the corrupt bit so
612 * that we don't try and read the other copies of this block, just
613 * return -EIO.
614 */
615 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
616 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
617 ret = -EIO;
618 }
619
620 if (found_level > 0 && btrfs_check_node(eb))
621 ret = -EIO;
622
623 if (!ret)
624 set_extent_buffer_uptodate(eb);
625 else
626 btrfs_err(fs_info,
627 "block=%llu read time tree block corruption detected",
628 eb->start);
629out:
630 return ret;
631}
632
633static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
634 int mirror)
635{
636 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
637 struct extent_buffer *eb;
638 bool reads_done;
639 int ret = 0;
640
641 /*
642 * We don't allow bio merge for subpage metadata read, so we should
643 * only get one eb for each endio hook.
644 */
645 ASSERT(end == start + fs_info->nodesize - 1);
646 ASSERT(PagePrivate(page));
647
648 eb = find_extent_buffer(fs_info, start);
649 /*
650 * When we are reading one tree block, eb must have been inserted into
651 * the radix tree. If not, something is wrong.
652 */
653 ASSERT(eb);
654
655 reads_done = atomic_dec_and_test(&eb->io_pages);
656 /* Subpage read must finish in page read */
657 ASSERT(reads_done);
658
659 eb->read_mirror = mirror;
660 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
661 ret = -EIO;
662 goto err;
663 }
664 ret = validate_extent_buffer(eb);
665 if (ret < 0)
666 goto err;
667
668 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
669 btree_readahead_hook(eb, ret);
670
671 set_extent_buffer_uptodate(eb);
672
673 free_extent_buffer(eb);
674 return ret;
675err:
676 /*
677 * end_bio_extent_readpage decrements io_pages in case of error,
678 * make sure it has something to decrement.
679 */
680 atomic_inc(&eb->io_pages);
681 clear_extent_buffer_uptodate(eb);
682 free_extent_buffer(eb);
683 return ret;
684}
685
686int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
687 struct page *page, u64 start, u64 end,
688 int mirror)
689{
690 struct extent_buffer *eb;
691 int ret = 0;
692 int reads_done;
693
694 ASSERT(page->private);
695
696 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
697 return validate_subpage_buffer(page, start, end, mirror);
698
699 eb = (struct extent_buffer *)page->private;
700
701 /*
702 * The pending IO might have been the only thing that kept this buffer
703 * in memory. Make sure we have a ref for all this other checks
704 */
705 atomic_inc(&eb->refs);
706
707 reads_done = atomic_dec_and_test(&eb->io_pages);
708 if (!reads_done)
709 goto err;
710
711 eb->read_mirror = mirror;
712 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
713 ret = -EIO;
714 goto err;
715 }
716 ret = validate_extent_buffer(eb);
717err:
718 if (reads_done &&
719 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
720 btree_readahead_hook(eb, ret);
721
722 if (ret) {
723 /*
724 * our io error hook is going to dec the io pages
725 * again, we have to make sure it has something
726 * to decrement
727 */
728 atomic_inc(&eb->io_pages);
729 clear_extent_buffer_uptodate(eb);
730 }
731 free_extent_buffer(eb);
732
733 return ret;
734}
735
736static void end_workqueue_bio(struct bio *bio)
737{
738 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
739 struct btrfs_fs_info *fs_info;
740 struct btrfs_workqueue *wq;
741
742 fs_info = end_io_wq->info;
743 end_io_wq->status = bio->bi_status;
744
745 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
746 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
747 wq = fs_info->endio_meta_write_workers;
748 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
749 wq = fs_info->endio_freespace_worker;
750 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
751 wq = fs_info->endio_raid56_workers;
752 else
753 wq = fs_info->endio_write_workers;
754 } else {
755 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
756 wq = fs_info->endio_raid56_workers;
757 else if (end_io_wq->metadata)
758 wq = fs_info->endio_meta_workers;
759 else
760 wq = fs_info->endio_workers;
761 }
762
763 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
764 btrfs_queue_work(wq, &end_io_wq->work);
765}
766
767blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
768 enum btrfs_wq_endio_type metadata)
769{
770 struct btrfs_end_io_wq *end_io_wq;
771
772 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
773 if (!end_io_wq)
774 return BLK_STS_RESOURCE;
775
776 end_io_wq->private = bio->bi_private;
777 end_io_wq->end_io = bio->bi_end_io;
778 end_io_wq->info = info;
779 end_io_wq->status = 0;
780 end_io_wq->bio = bio;
781 end_io_wq->metadata = metadata;
782
783 bio->bi_private = end_io_wq;
784 bio->bi_end_io = end_workqueue_bio;
785 return 0;
786}
787
788static void run_one_async_start(struct btrfs_work *work)
789{
790 struct async_submit_bio *async;
791 blk_status_t ret;
792
793 async = container_of(work, struct async_submit_bio, work);
794 ret = async->submit_bio_start(async->inode, async->bio,
795 async->dio_file_offset);
796 if (ret)
797 async->status = ret;
798}
799
800/*
801 * In order to insert checksums into the metadata in large chunks, we wait
802 * until bio submission time. All the pages in the bio are checksummed and
803 * sums are attached onto the ordered extent record.
804 *
805 * At IO completion time the csums attached on the ordered extent record are
806 * inserted into the tree.
807 */
808static void run_one_async_done(struct btrfs_work *work)
809{
810 struct async_submit_bio *async;
811 struct inode *inode;
812 blk_status_t ret;
813
814 async = container_of(work, struct async_submit_bio, work);
815 inode = async->inode;
816
817 /* If an error occurred we just want to clean up the bio and move on */
818 if (async->status) {
819 async->bio->bi_status = async->status;
820 bio_endio(async->bio);
821 return;
822 }
823
824 /*
825 * All of the bios that pass through here are from async helpers.
826 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
827 * This changes nothing when cgroups aren't in use.
828 */
829 async->bio->bi_opf |= REQ_CGROUP_PUNT;
830 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
831 if (ret) {
832 async->bio->bi_status = ret;
833 bio_endio(async->bio);
834 }
835}
836
837static void run_one_async_free(struct btrfs_work *work)
838{
839 struct async_submit_bio *async;
840
841 async = container_of(work, struct async_submit_bio, work);
842 kfree(async);
843}
844
845blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
846 int mirror_num, unsigned long bio_flags,
847 u64 dio_file_offset,
848 extent_submit_bio_start_t *submit_bio_start)
849{
850 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
851 struct async_submit_bio *async;
852
853 async = kmalloc(sizeof(*async), GFP_NOFS);
854 if (!async)
855 return BLK_STS_RESOURCE;
856
857 async->inode = inode;
858 async->bio = bio;
859 async->mirror_num = mirror_num;
860 async->submit_bio_start = submit_bio_start;
861
862 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
863 run_one_async_free);
864
865 async->dio_file_offset = dio_file_offset;
866
867 async->status = 0;
868
869 if (op_is_sync(bio->bi_opf))
870 btrfs_set_work_high_priority(&async->work);
871
872 btrfs_queue_work(fs_info->workers, &async->work);
873 return 0;
874}
875
876static blk_status_t btree_csum_one_bio(struct bio *bio)
877{
878 struct bio_vec *bvec;
879 struct btrfs_root *root;
880 int ret = 0;
881 struct bvec_iter_all iter_all;
882
883 ASSERT(!bio_flagged(bio, BIO_CLONED));
884 bio_for_each_segment_all(bvec, bio, iter_all) {
885 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
886 ret = csum_dirty_buffer(root->fs_info, bvec);
887 if (ret)
888 break;
889 }
890
891 return errno_to_blk_status(ret);
892}
893
894static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
895 u64 dio_file_offset)
896{
897 /*
898 * when we're called for a write, we're already in the async
899 * submission context. Just jump into btrfs_map_bio
900 */
901 return btree_csum_one_bio(bio);
902}
903
904static bool should_async_write(struct btrfs_fs_info *fs_info,
905 struct btrfs_inode *bi)
906{
907 if (btrfs_is_zoned(fs_info))
908 return false;
909 if (atomic_read(&bi->sync_writers))
910 return false;
911 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
912 return false;
913 return true;
914}
915
916blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
917 int mirror_num, unsigned long bio_flags)
918{
919 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
920 blk_status_t ret;
921
922 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
923 /*
924 * called for a read, do the setup so that checksum validation
925 * can happen in the async kernel threads
926 */
927 ret = btrfs_bio_wq_end_io(fs_info, bio,
928 BTRFS_WQ_ENDIO_METADATA);
929 if (ret)
930 goto out_w_error;
931 ret = btrfs_map_bio(fs_info, bio, mirror_num);
932 } else if (!should_async_write(fs_info, BTRFS_I(inode))) {
933 ret = btree_csum_one_bio(bio);
934 if (ret)
935 goto out_w_error;
936 ret = btrfs_map_bio(fs_info, bio, mirror_num);
937 } else {
938 /*
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
941 */
942 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
943 0, btree_submit_bio_start);
944 }
945
946 if (ret)
947 goto out_w_error;
948 return 0;
949
950out_w_error:
951 bio->bi_status = ret;
952 bio_endio(bio);
953 return ret;
954}
955
956#ifdef CONFIG_MIGRATION
957static int btree_migratepage(struct address_space *mapping,
958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
960{
961 /*
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
964 */
965 if (PageDirty(page))
966 return -EAGAIN;
967 /*
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
970 */
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
973 return -EAGAIN;
974 return migrate_page(mapping, newpage, page, mode);
975}
976#endif
977
978
979static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
981{
982 struct btrfs_fs_info *fs_info;
983 int ret;
984
985 if (wbc->sync_mode == WB_SYNC_NONE) {
986
987 if (wbc->for_kupdate)
988 return 0;
989
990 fs_info = BTRFS_I(mapping->host)->root->fs_info;
991 /* this is a bit racy, but that's ok */
992 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
993 BTRFS_DIRTY_METADATA_THRESH,
994 fs_info->dirty_metadata_batch);
995 if (ret < 0)
996 return 0;
997 }
998 return btree_write_cache_pages(mapping, wbc);
999}
1000
1001static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1002{
1003 if (PageWriteback(page) || PageDirty(page))
1004 return 0;
1005
1006 return try_release_extent_buffer(page);
1007}
1008
1009static void btree_invalidatepage(struct page *page, unsigned int offset,
1010 unsigned int length)
1011{
1012 struct extent_io_tree *tree;
1013 tree = &BTRFS_I(page->mapping->host)->io_tree;
1014 extent_invalidatepage(tree, page, offset);
1015 btree_releasepage(page, GFP_NOFS);
1016 if (PagePrivate(page)) {
1017 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1018 "page private not zero on page %llu",
1019 (unsigned long long)page_offset(page));
1020 detach_page_private(page);
1021 }
1022}
1023
1024static int btree_set_page_dirty(struct page *page)
1025{
1026#ifdef DEBUG
1027 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1028 struct btrfs_subpage *subpage;
1029 struct extent_buffer *eb;
1030 int cur_bit = 0;
1031 u64 page_start = page_offset(page);
1032
1033 if (fs_info->sectorsize == PAGE_SIZE) {
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1040 return __set_page_dirty_nobuffers(page);
1041 }
1042 ASSERT(PagePrivate(page) && page->private);
1043 subpage = (struct btrfs_subpage *)page->private;
1044
1045 ASSERT(subpage->dirty_bitmap);
1046 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1047 unsigned long flags;
1048 u64 cur;
1049 u16 tmp = (1 << cur_bit);
1050
1051 spin_lock_irqsave(&subpage->lock, flags);
1052 if (!(tmp & subpage->dirty_bitmap)) {
1053 spin_unlock_irqrestore(&subpage->lock, flags);
1054 cur_bit++;
1055 continue;
1056 }
1057 spin_unlock_irqrestore(&subpage->lock, flags);
1058 cur = page_start + cur_bit * fs_info->sectorsize;
1059
1060 eb = find_extent_buffer(fs_info, cur);
1061 ASSERT(eb);
1062 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063 ASSERT(atomic_read(&eb->refs));
1064 btrfs_assert_tree_locked(eb);
1065 free_extent_buffer(eb);
1066
1067 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1068 }
1069#endif
1070 return __set_page_dirty_nobuffers(page);
1071}
1072
1073static const struct address_space_operations btree_aops = {
1074 .writepages = btree_writepages,
1075 .releasepage = btree_releasepage,
1076 .invalidatepage = btree_invalidatepage,
1077#ifdef CONFIG_MIGRATION
1078 .migratepage = btree_migratepage,
1079#endif
1080 .set_page_dirty = btree_set_page_dirty,
1081};
1082
1083struct extent_buffer *btrfs_find_create_tree_block(
1084 struct btrfs_fs_info *fs_info,
1085 u64 bytenr, u64 owner_root,
1086 int level)
1087{
1088 if (btrfs_is_testing(fs_info))
1089 return alloc_test_extent_buffer(fs_info, bytenr);
1090 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1091}
1092
1093/*
1094 * Read tree block at logical address @bytenr and do variant basic but critical
1095 * verification.
1096 *
1097 * @owner_root: the objectid of the root owner for this block.
1098 * @parent_transid: expected transid of this tree block, skip check if 0
1099 * @level: expected level, mandatory check
1100 * @first_key: expected key in slot 0, skip check if NULL
1101 */
1102struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1103 u64 owner_root, u64 parent_transid,
1104 int level, struct btrfs_key *first_key)
1105{
1106 struct extent_buffer *buf = NULL;
1107 int ret;
1108
1109 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1110 if (IS_ERR(buf))
1111 return buf;
1112
1113 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1114 level, first_key);
1115 if (ret) {
1116 free_extent_buffer_stale(buf);
1117 return ERR_PTR(ret);
1118 }
1119 return buf;
1120
1121}
1122
1123void btrfs_clean_tree_block(struct extent_buffer *buf)
1124{
1125 struct btrfs_fs_info *fs_info = buf->fs_info;
1126 if (btrfs_header_generation(buf) ==
1127 fs_info->running_transaction->transid) {
1128 btrfs_assert_tree_locked(buf);
1129
1130 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1131 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1132 -buf->len,
1133 fs_info->dirty_metadata_batch);
1134 clear_extent_buffer_dirty(buf);
1135 }
1136 }
1137}
1138
1139static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1140 u64 objectid)
1141{
1142 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1143 root->fs_info = fs_info;
1144 root->node = NULL;
1145 root->commit_root = NULL;
1146 root->state = 0;
1147 root->orphan_cleanup_state = 0;
1148
1149 root->last_trans = 0;
1150 root->free_objectid = 0;
1151 root->nr_delalloc_inodes = 0;
1152 root->nr_ordered_extents = 0;
1153 root->inode_tree = RB_ROOT;
1154 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1155 root->block_rsv = NULL;
1156
1157 INIT_LIST_HEAD(&root->dirty_list);
1158 INIT_LIST_HEAD(&root->root_list);
1159 INIT_LIST_HEAD(&root->delalloc_inodes);
1160 INIT_LIST_HEAD(&root->delalloc_root);
1161 INIT_LIST_HEAD(&root->ordered_extents);
1162 INIT_LIST_HEAD(&root->ordered_root);
1163 INIT_LIST_HEAD(&root->reloc_dirty_list);
1164 INIT_LIST_HEAD(&root->logged_list[0]);
1165 INIT_LIST_HEAD(&root->logged_list[1]);
1166 spin_lock_init(&root->inode_lock);
1167 spin_lock_init(&root->delalloc_lock);
1168 spin_lock_init(&root->ordered_extent_lock);
1169 spin_lock_init(&root->accounting_lock);
1170 spin_lock_init(&root->log_extents_lock[0]);
1171 spin_lock_init(&root->log_extents_lock[1]);
1172 spin_lock_init(&root->qgroup_meta_rsv_lock);
1173 mutex_init(&root->objectid_mutex);
1174 mutex_init(&root->log_mutex);
1175 mutex_init(&root->ordered_extent_mutex);
1176 mutex_init(&root->delalloc_mutex);
1177 init_waitqueue_head(&root->qgroup_flush_wait);
1178 init_waitqueue_head(&root->log_writer_wait);
1179 init_waitqueue_head(&root->log_commit_wait[0]);
1180 init_waitqueue_head(&root->log_commit_wait[1]);
1181 INIT_LIST_HEAD(&root->log_ctxs[0]);
1182 INIT_LIST_HEAD(&root->log_ctxs[1]);
1183 atomic_set(&root->log_commit[0], 0);
1184 atomic_set(&root->log_commit[1], 0);
1185 atomic_set(&root->log_writers, 0);
1186 atomic_set(&root->log_batch, 0);
1187 refcount_set(&root->refs, 1);
1188 atomic_set(&root->snapshot_force_cow, 0);
1189 atomic_set(&root->nr_swapfiles, 0);
1190 root->log_transid = 0;
1191 root->log_transid_committed = -1;
1192 root->last_log_commit = 0;
1193 if (!dummy) {
1194 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1195 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1196 extent_io_tree_init(fs_info, &root->log_csum_range,
1197 IO_TREE_LOG_CSUM_RANGE, NULL);
1198 }
1199
1200 memset(&root->root_key, 0, sizeof(root->root_key));
1201 memset(&root->root_item, 0, sizeof(root->root_item));
1202 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1203 root->root_key.objectid = objectid;
1204 root->anon_dev = 0;
1205
1206 spin_lock_init(&root->root_item_lock);
1207 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1208#ifdef CONFIG_BTRFS_DEBUG
1209 INIT_LIST_HEAD(&root->leak_list);
1210 spin_lock(&fs_info->fs_roots_radix_lock);
1211 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1212 spin_unlock(&fs_info->fs_roots_radix_lock);
1213#endif
1214}
1215
1216static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1217 u64 objectid, gfp_t flags)
1218{
1219 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1220 if (root)
1221 __setup_root(root, fs_info, objectid);
1222 return root;
1223}
1224
1225#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1226/* Should only be used by the testing infrastructure */
1227struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1228{
1229 struct btrfs_root *root;
1230
1231 if (!fs_info)
1232 return ERR_PTR(-EINVAL);
1233
1234 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1235 if (!root)
1236 return ERR_PTR(-ENOMEM);
1237
1238 /* We don't use the stripesize in selftest, set it as sectorsize */
1239 root->alloc_bytenr = 0;
1240
1241 return root;
1242}
1243#endif
1244
1245struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1246 u64 objectid)
1247{
1248 struct btrfs_fs_info *fs_info = trans->fs_info;
1249 struct extent_buffer *leaf;
1250 struct btrfs_root *tree_root = fs_info->tree_root;
1251 struct btrfs_root *root;
1252 struct btrfs_key key;
1253 unsigned int nofs_flag;
1254 int ret = 0;
1255
1256 /*
1257 * We're holding a transaction handle, so use a NOFS memory allocation
1258 * context to avoid deadlock if reclaim happens.
1259 */
1260 nofs_flag = memalloc_nofs_save();
1261 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1262 memalloc_nofs_restore(nofs_flag);
1263 if (!root)
1264 return ERR_PTR(-ENOMEM);
1265
1266 root->root_key.objectid = objectid;
1267 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1268 root->root_key.offset = 0;
1269
1270 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1271 BTRFS_NESTING_NORMAL);
1272 if (IS_ERR(leaf)) {
1273 ret = PTR_ERR(leaf);
1274 leaf = NULL;
1275 goto fail_unlock;
1276 }
1277
1278 root->node = leaf;
1279 btrfs_mark_buffer_dirty(leaf);
1280
1281 root->commit_root = btrfs_root_node(root);
1282 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1283
1284 btrfs_set_root_flags(&root->root_item, 0);
1285 btrfs_set_root_limit(&root->root_item, 0);
1286 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1287 btrfs_set_root_generation(&root->root_item, trans->transid);
1288 btrfs_set_root_level(&root->root_item, 0);
1289 btrfs_set_root_refs(&root->root_item, 1);
1290 btrfs_set_root_used(&root->root_item, leaf->len);
1291 btrfs_set_root_last_snapshot(&root->root_item, 0);
1292 btrfs_set_root_dirid(&root->root_item, 0);
1293 if (is_fstree(objectid))
1294 generate_random_guid(root->root_item.uuid);
1295 else
1296 export_guid(root->root_item.uuid, &guid_null);
1297 btrfs_set_root_drop_level(&root->root_item, 0);
1298
1299 btrfs_tree_unlock(leaf);
1300
1301 key.objectid = objectid;
1302 key.type = BTRFS_ROOT_ITEM_KEY;
1303 key.offset = 0;
1304 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1305 if (ret)
1306 goto fail;
1307
1308 return root;
1309
1310fail_unlock:
1311 if (leaf)
1312 btrfs_tree_unlock(leaf);
1313fail:
1314 btrfs_put_root(root);
1315
1316 return ERR_PTR(ret);
1317}
1318
1319static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1320 struct btrfs_fs_info *fs_info)
1321{
1322 struct btrfs_root *root;
1323
1324 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1325 if (!root)
1326 return ERR_PTR(-ENOMEM);
1327
1328 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1329 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1330 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1331
1332 return root;
1333}
1334
1335int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1336 struct btrfs_root *root)
1337{
1338 struct extent_buffer *leaf;
1339
1340 /*
1341 * DON'T set SHAREABLE bit for log trees.
1342 *
1343 * Log trees are not exposed to user space thus can't be snapshotted,
1344 * and they go away before a real commit is actually done.
1345 *
1346 * They do store pointers to file data extents, and those reference
1347 * counts still get updated (along with back refs to the log tree).
1348 */
1349
1350 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1351 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1352 if (IS_ERR(leaf))
1353 return PTR_ERR(leaf);
1354
1355 root->node = leaf;
1356
1357 btrfs_mark_buffer_dirty(root->node);
1358 btrfs_tree_unlock(root->node);
1359
1360 return 0;
1361}
1362
1363int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1364 struct btrfs_fs_info *fs_info)
1365{
1366 struct btrfs_root *log_root;
1367
1368 log_root = alloc_log_tree(trans, fs_info);
1369 if (IS_ERR(log_root))
1370 return PTR_ERR(log_root);
1371
1372 if (!btrfs_is_zoned(fs_info)) {
1373 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1374
1375 if (ret) {
1376 btrfs_put_root(log_root);
1377 return ret;
1378 }
1379 }
1380
1381 WARN_ON(fs_info->log_root_tree);
1382 fs_info->log_root_tree = log_root;
1383 return 0;
1384}
1385
1386int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root)
1388{
1389 struct btrfs_fs_info *fs_info = root->fs_info;
1390 struct btrfs_root *log_root;
1391 struct btrfs_inode_item *inode_item;
1392 int ret;
1393
1394 log_root = alloc_log_tree(trans, fs_info);
1395 if (IS_ERR(log_root))
1396 return PTR_ERR(log_root);
1397
1398 ret = btrfs_alloc_log_tree_node(trans, log_root);
1399 if (ret) {
1400 btrfs_put_root(log_root);
1401 return ret;
1402 }
1403
1404 log_root->last_trans = trans->transid;
1405 log_root->root_key.offset = root->root_key.objectid;
1406
1407 inode_item = &log_root->root_item.inode;
1408 btrfs_set_stack_inode_generation(inode_item, 1);
1409 btrfs_set_stack_inode_size(inode_item, 3);
1410 btrfs_set_stack_inode_nlink(inode_item, 1);
1411 btrfs_set_stack_inode_nbytes(inode_item,
1412 fs_info->nodesize);
1413 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1414
1415 btrfs_set_root_node(&log_root->root_item, log_root->node);
1416
1417 WARN_ON(root->log_root);
1418 root->log_root = log_root;
1419 root->log_transid = 0;
1420 root->log_transid_committed = -1;
1421 root->last_log_commit = 0;
1422 return 0;
1423}
1424
1425static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1426 struct btrfs_path *path,
1427 struct btrfs_key *key)
1428{
1429 struct btrfs_root *root;
1430 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1431 u64 generation;
1432 int ret;
1433 int level;
1434
1435 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1436 if (!root)
1437 return ERR_PTR(-ENOMEM);
1438
1439 ret = btrfs_find_root(tree_root, key, path,
1440 &root->root_item, &root->root_key);
1441 if (ret) {
1442 if (ret > 0)
1443 ret = -ENOENT;
1444 goto fail;
1445 }
1446
1447 generation = btrfs_root_generation(&root->root_item);
1448 level = btrfs_root_level(&root->root_item);
1449 root->node = read_tree_block(fs_info,
1450 btrfs_root_bytenr(&root->root_item),
1451 key->objectid, generation, level, NULL);
1452 if (IS_ERR(root->node)) {
1453 ret = PTR_ERR(root->node);
1454 root->node = NULL;
1455 goto fail;
1456 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1457 ret = -EIO;
1458 goto fail;
1459 }
1460 root->commit_root = btrfs_root_node(root);
1461 return root;
1462fail:
1463 btrfs_put_root(root);
1464 return ERR_PTR(ret);
1465}
1466
1467struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1468 struct btrfs_key *key)
1469{
1470 struct btrfs_root *root;
1471 struct btrfs_path *path;
1472
1473 path = btrfs_alloc_path();
1474 if (!path)
1475 return ERR_PTR(-ENOMEM);
1476 root = read_tree_root_path(tree_root, path, key);
1477 btrfs_free_path(path);
1478
1479 return root;
1480}
1481
1482/*
1483 * Initialize subvolume root in-memory structure
1484 *
1485 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1486 */
1487static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1488{
1489 int ret;
1490 unsigned int nofs_flag;
1491
1492 /*
1493 * We might be called under a transaction (e.g. indirect backref
1494 * resolution) which could deadlock if it triggers memory reclaim
1495 */
1496 nofs_flag = memalloc_nofs_save();
1497 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1498 memalloc_nofs_restore(nofs_flag);
1499 if (ret)
1500 goto fail;
1501
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1503 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1504 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1505 btrfs_check_and_init_root_item(&root->root_item);
1506 }
1507
1508 /*
1509 * Don't assign anonymous block device to roots that are not exposed to
1510 * userspace, the id pool is limited to 1M
1511 */
1512 if (is_fstree(root->root_key.objectid) &&
1513 btrfs_root_refs(&root->root_item) > 0) {
1514 if (!anon_dev) {
1515 ret = get_anon_bdev(&root->anon_dev);
1516 if (ret)
1517 goto fail;
1518 } else {
1519 root->anon_dev = anon_dev;
1520 }
1521 }
1522
1523 mutex_lock(&root->objectid_mutex);
1524 ret = btrfs_init_root_free_objectid(root);
1525 if (ret) {
1526 mutex_unlock(&root->objectid_mutex);
1527 goto fail;
1528 }
1529
1530 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1531
1532 mutex_unlock(&root->objectid_mutex);
1533
1534 return 0;
1535fail:
1536 /* The caller is responsible to call btrfs_free_fs_root */
1537 return ret;
1538}
1539
1540static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1541 u64 root_id)
1542{
1543 struct btrfs_root *root;
1544
1545 spin_lock(&fs_info->fs_roots_radix_lock);
1546 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1547 (unsigned long)root_id);
1548 if (root)
1549 root = btrfs_grab_root(root);
1550 spin_unlock(&fs_info->fs_roots_radix_lock);
1551 return root;
1552}
1553
1554static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1555 u64 objectid)
1556{
1557 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1558 return btrfs_grab_root(fs_info->tree_root);
1559 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1560 return btrfs_grab_root(fs_info->extent_root);
1561 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1562 return btrfs_grab_root(fs_info->chunk_root);
1563 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1564 return btrfs_grab_root(fs_info->dev_root);
1565 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1566 return btrfs_grab_root(fs_info->csum_root);
1567 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1568 return btrfs_grab_root(fs_info->quota_root) ?
1569 fs_info->quota_root : ERR_PTR(-ENOENT);
1570 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1571 return btrfs_grab_root(fs_info->uuid_root) ?
1572 fs_info->uuid_root : ERR_PTR(-ENOENT);
1573 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1574 return btrfs_grab_root(fs_info->free_space_root) ?
1575 fs_info->free_space_root : ERR_PTR(-ENOENT);
1576 return NULL;
1577}
1578
1579int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1580 struct btrfs_root *root)
1581{
1582 int ret;
1583
1584 ret = radix_tree_preload(GFP_NOFS);
1585 if (ret)
1586 return ret;
1587
1588 spin_lock(&fs_info->fs_roots_radix_lock);
1589 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1590 (unsigned long)root->root_key.objectid,
1591 root);
1592 if (ret == 0) {
1593 btrfs_grab_root(root);
1594 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1595 }
1596 spin_unlock(&fs_info->fs_roots_radix_lock);
1597 radix_tree_preload_end();
1598
1599 return ret;
1600}
1601
1602void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1603{
1604#ifdef CONFIG_BTRFS_DEBUG
1605 struct btrfs_root *root;
1606
1607 while (!list_empty(&fs_info->allocated_roots)) {
1608 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1609
1610 root = list_first_entry(&fs_info->allocated_roots,
1611 struct btrfs_root, leak_list);
1612 btrfs_err(fs_info, "leaked root %s refcount %d",
1613 btrfs_root_name(&root->root_key, buf),
1614 refcount_read(&root->refs));
1615 while (refcount_read(&root->refs) > 1)
1616 btrfs_put_root(root);
1617 btrfs_put_root(root);
1618 }
1619#endif
1620}
1621
1622void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1623{
1624 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1625 percpu_counter_destroy(&fs_info->delalloc_bytes);
1626 percpu_counter_destroy(&fs_info->ordered_bytes);
1627 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1628 btrfs_free_csum_hash(fs_info);
1629 btrfs_free_stripe_hash_table(fs_info);
1630 btrfs_free_ref_cache(fs_info);
1631 kfree(fs_info->balance_ctl);
1632 kfree(fs_info->delayed_root);
1633 btrfs_put_root(fs_info->extent_root);
1634 btrfs_put_root(fs_info->tree_root);
1635 btrfs_put_root(fs_info->chunk_root);
1636 btrfs_put_root(fs_info->dev_root);
1637 btrfs_put_root(fs_info->csum_root);
1638 btrfs_put_root(fs_info->quota_root);
1639 btrfs_put_root(fs_info->uuid_root);
1640 btrfs_put_root(fs_info->free_space_root);
1641 btrfs_put_root(fs_info->fs_root);
1642 btrfs_put_root(fs_info->data_reloc_root);
1643 btrfs_check_leaked_roots(fs_info);
1644 btrfs_extent_buffer_leak_debug_check(fs_info);
1645 kfree(fs_info->super_copy);
1646 kfree(fs_info->super_for_commit);
1647 kvfree(fs_info);
1648}
1649
1650
1651/*
1652 * Get an in-memory reference of a root structure.
1653 *
1654 * For essential trees like root/extent tree, we grab it from fs_info directly.
1655 * For subvolume trees, we check the cached filesystem roots first. If not
1656 * found, then read it from disk and add it to cached fs roots.
1657 *
1658 * Caller should release the root by calling btrfs_put_root() after the usage.
1659 *
1660 * NOTE: Reloc and log trees can't be read by this function as they share the
1661 * same root objectid.
1662 *
1663 * @objectid: root id
1664 * @anon_dev: preallocated anonymous block device number for new roots,
1665 * pass 0 for new allocation.
1666 * @check_ref: whether to check root item references, If true, return -ENOENT
1667 * for orphan roots
1668 */
1669static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1670 u64 objectid, dev_t anon_dev,
1671 bool check_ref)
1672{
1673 struct btrfs_root *root;
1674 struct btrfs_path *path;
1675 struct btrfs_key key;
1676 int ret;
1677
1678 root = btrfs_get_global_root(fs_info, objectid);
1679 if (root)
1680 return root;
1681again:
1682 root = btrfs_lookup_fs_root(fs_info, objectid);
1683 if (root) {
1684 /* Shouldn't get preallocated anon_dev for cached roots */
1685 ASSERT(!anon_dev);
1686 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1687 btrfs_put_root(root);
1688 return ERR_PTR(-ENOENT);
1689 }
1690 return root;
1691 }
1692
1693 key.objectid = objectid;
1694 key.type = BTRFS_ROOT_ITEM_KEY;
1695 key.offset = (u64)-1;
1696 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1697 if (IS_ERR(root))
1698 return root;
1699
1700 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1701 ret = -ENOENT;
1702 goto fail;
1703 }
1704
1705 ret = btrfs_init_fs_root(root, anon_dev);
1706 if (ret)
1707 goto fail;
1708
1709 path = btrfs_alloc_path();
1710 if (!path) {
1711 ret = -ENOMEM;
1712 goto fail;
1713 }
1714 key.objectid = BTRFS_ORPHAN_OBJECTID;
1715 key.type = BTRFS_ORPHAN_ITEM_KEY;
1716 key.offset = objectid;
1717
1718 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1719 btrfs_free_path(path);
1720 if (ret < 0)
1721 goto fail;
1722 if (ret == 0)
1723 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1724
1725 ret = btrfs_insert_fs_root(fs_info, root);
1726 if (ret) {
1727 btrfs_put_root(root);
1728 if (ret == -EEXIST)
1729 goto again;
1730 goto fail;
1731 }
1732 return root;
1733fail:
1734 btrfs_put_root(root);
1735 return ERR_PTR(ret);
1736}
1737
1738/*
1739 * Get in-memory reference of a root structure
1740 *
1741 * @objectid: tree objectid
1742 * @check_ref: if set, verify that the tree exists and the item has at least
1743 * one reference
1744 */
1745struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1746 u64 objectid, bool check_ref)
1747{
1748 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1749}
1750
1751/*
1752 * Get in-memory reference of a root structure, created as new, optionally pass
1753 * the anonymous block device id
1754 *
1755 * @objectid: tree objectid
1756 * @anon_dev: if zero, allocate a new anonymous block device or use the
1757 * parameter value
1758 */
1759struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1760 u64 objectid, dev_t anon_dev)
1761{
1762 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1763}
1764
1765/*
1766 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1767 * @fs_info: the fs_info
1768 * @objectid: the objectid we need to lookup
1769 *
1770 * This is exclusively used for backref walking, and exists specifically because
1771 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1772 * creation time, which means we may have to read the tree_root in order to look
1773 * up a fs root that is not in memory. If the root is not in memory we will
1774 * read the tree root commit root and look up the fs root from there. This is a
1775 * temporary root, it will not be inserted into the radix tree as it doesn't
1776 * have the most uptodate information, it'll simply be discarded once the
1777 * backref code is finished using the root.
1778 */
1779struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1780 struct btrfs_path *path,
1781 u64 objectid)
1782{
1783 struct btrfs_root *root;
1784 struct btrfs_key key;
1785
1786 ASSERT(path->search_commit_root && path->skip_locking);
1787
1788 /*
1789 * This can return -ENOENT if we ask for a root that doesn't exist, but
1790 * since this is called via the backref walking code we won't be looking
1791 * up a root that doesn't exist, unless there's corruption. So if root
1792 * != NULL just return it.
1793 */
1794 root = btrfs_get_global_root(fs_info, objectid);
1795 if (root)
1796 return root;
1797
1798 root = btrfs_lookup_fs_root(fs_info, objectid);
1799 if (root)
1800 return root;
1801
1802 key.objectid = objectid;
1803 key.type = BTRFS_ROOT_ITEM_KEY;
1804 key.offset = (u64)-1;
1805 root = read_tree_root_path(fs_info->tree_root, path, &key);
1806 btrfs_release_path(path);
1807
1808 return root;
1809}
1810
1811/*
1812 * called by the kthread helper functions to finally call the bio end_io
1813 * functions. This is where read checksum verification actually happens
1814 */
1815static void end_workqueue_fn(struct btrfs_work *work)
1816{
1817 struct bio *bio;
1818 struct btrfs_end_io_wq *end_io_wq;
1819
1820 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1821 bio = end_io_wq->bio;
1822
1823 bio->bi_status = end_io_wq->status;
1824 bio->bi_private = end_io_wq->private;
1825 bio->bi_end_io = end_io_wq->end_io;
1826 bio_endio(bio);
1827 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1828}
1829
1830static int cleaner_kthread(void *arg)
1831{
1832 struct btrfs_root *root = arg;
1833 struct btrfs_fs_info *fs_info = root->fs_info;
1834 int again;
1835
1836 while (1) {
1837 again = 0;
1838
1839 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1840
1841 /* Make the cleaner go to sleep early. */
1842 if (btrfs_need_cleaner_sleep(fs_info))
1843 goto sleep;
1844
1845 /*
1846 * Do not do anything if we might cause open_ctree() to block
1847 * before we have finished mounting the filesystem.
1848 */
1849 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1850 goto sleep;
1851
1852 if (!mutex_trylock(&fs_info->cleaner_mutex))
1853 goto sleep;
1854
1855 /*
1856 * Avoid the problem that we change the status of the fs
1857 * during the above check and trylock.
1858 */
1859 if (btrfs_need_cleaner_sleep(fs_info)) {
1860 mutex_unlock(&fs_info->cleaner_mutex);
1861 goto sleep;
1862 }
1863
1864 btrfs_run_delayed_iputs(fs_info);
1865
1866 again = btrfs_clean_one_deleted_snapshot(root);
1867 mutex_unlock(&fs_info->cleaner_mutex);
1868
1869 /*
1870 * The defragger has dealt with the R/O remount and umount,
1871 * needn't do anything special here.
1872 */
1873 btrfs_run_defrag_inodes(fs_info);
1874
1875 /*
1876 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1877 * with relocation (btrfs_relocate_chunk) and relocation
1878 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1879 * after acquiring fs_info->reclaim_bgs_lock. So we
1880 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1881 * unused block groups.
1882 */
1883 btrfs_delete_unused_bgs(fs_info);
1884
1885 /*
1886 * Reclaim block groups in the reclaim_bgs list after we deleted
1887 * all unused block_groups. This possibly gives us some more free
1888 * space.
1889 */
1890 btrfs_reclaim_bgs(fs_info);
1891sleep:
1892 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1893 if (kthread_should_park())
1894 kthread_parkme();
1895 if (kthread_should_stop())
1896 return 0;
1897 if (!again) {
1898 set_current_state(TASK_INTERRUPTIBLE);
1899 schedule();
1900 __set_current_state(TASK_RUNNING);
1901 }
1902 }
1903}
1904
1905static int transaction_kthread(void *arg)
1906{
1907 struct btrfs_root *root = arg;
1908 struct btrfs_fs_info *fs_info = root->fs_info;
1909 struct btrfs_trans_handle *trans;
1910 struct btrfs_transaction *cur;
1911 u64 transid;
1912 time64_t delta;
1913 unsigned long delay;
1914 bool cannot_commit;
1915
1916 do {
1917 cannot_commit = false;
1918 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1919 mutex_lock(&fs_info->transaction_kthread_mutex);
1920
1921 spin_lock(&fs_info->trans_lock);
1922 cur = fs_info->running_transaction;
1923 if (!cur) {
1924 spin_unlock(&fs_info->trans_lock);
1925 goto sleep;
1926 }
1927
1928 delta = ktime_get_seconds() - cur->start_time;
1929 if (cur->state < TRANS_STATE_COMMIT_START &&
1930 delta < fs_info->commit_interval) {
1931 spin_unlock(&fs_info->trans_lock);
1932 delay -= msecs_to_jiffies((delta - 1) * 1000);
1933 delay = min(delay,
1934 msecs_to_jiffies(fs_info->commit_interval * 1000));
1935 goto sleep;
1936 }
1937 transid = cur->transid;
1938 spin_unlock(&fs_info->trans_lock);
1939
1940 /* If the file system is aborted, this will always fail. */
1941 trans = btrfs_attach_transaction(root);
1942 if (IS_ERR(trans)) {
1943 if (PTR_ERR(trans) != -ENOENT)
1944 cannot_commit = true;
1945 goto sleep;
1946 }
1947 if (transid == trans->transid) {
1948 btrfs_commit_transaction(trans);
1949 } else {
1950 btrfs_end_transaction(trans);
1951 }
1952sleep:
1953 wake_up_process(fs_info->cleaner_kthread);
1954 mutex_unlock(&fs_info->transaction_kthread_mutex);
1955
1956 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1957 &fs_info->fs_state)))
1958 btrfs_cleanup_transaction(fs_info);
1959 if (!kthread_should_stop() &&
1960 (!btrfs_transaction_blocked(fs_info) ||
1961 cannot_commit))
1962 schedule_timeout_interruptible(delay);
1963 } while (!kthread_should_stop());
1964 return 0;
1965}
1966
1967/*
1968 * This will find the highest generation in the array of root backups. The
1969 * index of the highest array is returned, or -EINVAL if we can't find
1970 * anything.
1971 *
1972 * We check to make sure the array is valid by comparing the
1973 * generation of the latest root in the array with the generation
1974 * in the super block. If they don't match we pitch it.
1975 */
1976static int find_newest_super_backup(struct btrfs_fs_info *info)
1977{
1978 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1979 u64 cur;
1980 struct btrfs_root_backup *root_backup;
1981 int i;
1982
1983 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1984 root_backup = info->super_copy->super_roots + i;
1985 cur = btrfs_backup_tree_root_gen(root_backup);
1986 if (cur == newest_gen)
1987 return i;
1988 }
1989
1990 return -EINVAL;
1991}
1992
1993/*
1994 * copy all the root pointers into the super backup array.
1995 * this will bump the backup pointer by one when it is
1996 * done
1997 */
1998static void backup_super_roots(struct btrfs_fs_info *info)
1999{
2000 const int next_backup = info->backup_root_index;
2001 struct btrfs_root_backup *root_backup;
2002
2003 root_backup = info->super_for_commit->super_roots + next_backup;
2004
2005 /*
2006 * make sure all of our padding and empty slots get zero filled
2007 * regardless of which ones we use today
2008 */
2009 memset(root_backup, 0, sizeof(*root_backup));
2010
2011 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2012
2013 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2014 btrfs_set_backup_tree_root_gen(root_backup,
2015 btrfs_header_generation(info->tree_root->node));
2016
2017 btrfs_set_backup_tree_root_level(root_backup,
2018 btrfs_header_level(info->tree_root->node));
2019
2020 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2021 btrfs_set_backup_chunk_root_gen(root_backup,
2022 btrfs_header_generation(info->chunk_root->node));
2023 btrfs_set_backup_chunk_root_level(root_backup,
2024 btrfs_header_level(info->chunk_root->node));
2025
2026 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2027 btrfs_set_backup_extent_root_gen(root_backup,
2028 btrfs_header_generation(info->extent_root->node));
2029 btrfs_set_backup_extent_root_level(root_backup,
2030 btrfs_header_level(info->extent_root->node));
2031
2032 /*
2033 * we might commit during log recovery, which happens before we set
2034 * the fs_root. Make sure it is valid before we fill it in.
2035 */
2036 if (info->fs_root && info->fs_root->node) {
2037 btrfs_set_backup_fs_root(root_backup,
2038 info->fs_root->node->start);
2039 btrfs_set_backup_fs_root_gen(root_backup,
2040 btrfs_header_generation(info->fs_root->node));
2041 btrfs_set_backup_fs_root_level(root_backup,
2042 btrfs_header_level(info->fs_root->node));
2043 }
2044
2045 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2046 btrfs_set_backup_dev_root_gen(root_backup,
2047 btrfs_header_generation(info->dev_root->node));
2048 btrfs_set_backup_dev_root_level(root_backup,
2049 btrfs_header_level(info->dev_root->node));
2050
2051 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2052 btrfs_set_backup_csum_root_gen(root_backup,
2053 btrfs_header_generation(info->csum_root->node));
2054 btrfs_set_backup_csum_root_level(root_backup,
2055 btrfs_header_level(info->csum_root->node));
2056
2057 btrfs_set_backup_total_bytes(root_backup,
2058 btrfs_super_total_bytes(info->super_copy));
2059 btrfs_set_backup_bytes_used(root_backup,
2060 btrfs_super_bytes_used(info->super_copy));
2061 btrfs_set_backup_num_devices(root_backup,
2062 btrfs_super_num_devices(info->super_copy));
2063
2064 /*
2065 * if we don't copy this out to the super_copy, it won't get remembered
2066 * for the next commit
2067 */
2068 memcpy(&info->super_copy->super_roots,
2069 &info->super_for_commit->super_roots,
2070 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2071}
2072
2073/*
2074 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2075 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2076 *
2077 * fs_info - filesystem whose backup roots need to be read
2078 * priority - priority of backup root required
2079 *
2080 * Returns backup root index on success and -EINVAL otherwise.
2081 */
2082static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2083{
2084 int backup_index = find_newest_super_backup(fs_info);
2085 struct btrfs_super_block *super = fs_info->super_copy;
2086 struct btrfs_root_backup *root_backup;
2087
2088 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2089 if (priority == 0)
2090 return backup_index;
2091
2092 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2093 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2094 } else {
2095 return -EINVAL;
2096 }
2097
2098 root_backup = super->super_roots + backup_index;
2099
2100 btrfs_set_super_generation(super,
2101 btrfs_backup_tree_root_gen(root_backup));
2102 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2103 btrfs_set_super_root_level(super,
2104 btrfs_backup_tree_root_level(root_backup));
2105 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2106
2107 /*
2108 * Fixme: the total bytes and num_devices need to match or we should
2109 * need a fsck
2110 */
2111 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2112 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2113
2114 return backup_index;
2115}
2116
2117/* helper to cleanup workers */
2118static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2119{
2120 btrfs_destroy_workqueue(fs_info->fixup_workers);
2121 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2122 btrfs_destroy_workqueue(fs_info->workers);
2123 btrfs_destroy_workqueue(fs_info->endio_workers);
2124 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2125 btrfs_destroy_workqueue(fs_info->rmw_workers);
2126 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2127 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2128 btrfs_destroy_workqueue(fs_info->delayed_workers);
2129 btrfs_destroy_workqueue(fs_info->caching_workers);
2130 btrfs_destroy_workqueue(fs_info->readahead_workers);
2131 btrfs_destroy_workqueue(fs_info->flush_workers);
2132 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2133 if (fs_info->discard_ctl.discard_workers)
2134 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2135 /*
2136 * Now that all other work queues are destroyed, we can safely destroy
2137 * the queues used for metadata I/O, since tasks from those other work
2138 * queues can do metadata I/O operations.
2139 */
2140 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2141 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2142}
2143
2144static void free_root_extent_buffers(struct btrfs_root *root)
2145{
2146 if (root) {
2147 free_extent_buffer(root->node);
2148 free_extent_buffer(root->commit_root);
2149 root->node = NULL;
2150 root->commit_root = NULL;
2151 }
2152}
2153
2154/* helper to cleanup tree roots */
2155static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2156{
2157 free_root_extent_buffers(info->tree_root);
2158
2159 free_root_extent_buffers(info->dev_root);
2160 free_root_extent_buffers(info->extent_root);
2161 free_root_extent_buffers(info->csum_root);
2162 free_root_extent_buffers(info->quota_root);
2163 free_root_extent_buffers(info->uuid_root);
2164 free_root_extent_buffers(info->fs_root);
2165 free_root_extent_buffers(info->data_reloc_root);
2166 if (free_chunk_root)
2167 free_root_extent_buffers(info->chunk_root);
2168 free_root_extent_buffers(info->free_space_root);
2169}
2170
2171void btrfs_put_root(struct btrfs_root *root)
2172{
2173 if (!root)
2174 return;
2175
2176 if (refcount_dec_and_test(&root->refs)) {
2177 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2178 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2179 if (root->anon_dev)
2180 free_anon_bdev(root->anon_dev);
2181 btrfs_drew_lock_destroy(&root->snapshot_lock);
2182 free_root_extent_buffers(root);
2183#ifdef CONFIG_BTRFS_DEBUG
2184 spin_lock(&root->fs_info->fs_roots_radix_lock);
2185 list_del_init(&root->leak_list);
2186 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2187#endif
2188 kfree(root);
2189 }
2190}
2191
2192void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2193{
2194 int ret;
2195 struct btrfs_root *gang[8];
2196 int i;
2197
2198 while (!list_empty(&fs_info->dead_roots)) {
2199 gang[0] = list_entry(fs_info->dead_roots.next,
2200 struct btrfs_root, root_list);
2201 list_del(&gang[0]->root_list);
2202
2203 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2204 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2205 btrfs_put_root(gang[0]);
2206 }
2207
2208 while (1) {
2209 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2210 (void **)gang, 0,
2211 ARRAY_SIZE(gang));
2212 if (!ret)
2213 break;
2214 for (i = 0; i < ret; i++)
2215 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2216 }
2217}
2218
2219static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2220{
2221 mutex_init(&fs_info->scrub_lock);
2222 atomic_set(&fs_info->scrubs_running, 0);
2223 atomic_set(&fs_info->scrub_pause_req, 0);
2224 atomic_set(&fs_info->scrubs_paused, 0);
2225 atomic_set(&fs_info->scrub_cancel_req, 0);
2226 init_waitqueue_head(&fs_info->scrub_pause_wait);
2227 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2228}
2229
2230static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2231{
2232 spin_lock_init(&fs_info->balance_lock);
2233 mutex_init(&fs_info->balance_mutex);
2234 atomic_set(&fs_info->balance_pause_req, 0);
2235 atomic_set(&fs_info->balance_cancel_req, 0);
2236 fs_info->balance_ctl = NULL;
2237 init_waitqueue_head(&fs_info->balance_wait_q);
2238 atomic_set(&fs_info->reloc_cancel_req, 0);
2239}
2240
2241static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2242{
2243 struct inode *inode = fs_info->btree_inode;
2244
2245 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2246 set_nlink(inode, 1);
2247 /*
2248 * we set the i_size on the btree inode to the max possible int.
2249 * the real end of the address space is determined by all of
2250 * the devices in the system
2251 */
2252 inode->i_size = OFFSET_MAX;
2253 inode->i_mapping->a_ops = &btree_aops;
2254
2255 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2256 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2257 IO_TREE_BTREE_INODE_IO, inode);
2258 BTRFS_I(inode)->io_tree.track_uptodate = false;
2259 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2260
2261 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2262 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2263 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2264 btrfs_insert_inode_hash(inode);
2265}
2266
2267static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2268{
2269 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2270 init_rwsem(&fs_info->dev_replace.rwsem);
2271 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2272}
2273
2274static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2275{
2276 spin_lock_init(&fs_info->qgroup_lock);
2277 mutex_init(&fs_info->qgroup_ioctl_lock);
2278 fs_info->qgroup_tree = RB_ROOT;
2279 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2280 fs_info->qgroup_seq = 1;
2281 fs_info->qgroup_ulist = NULL;
2282 fs_info->qgroup_rescan_running = false;
2283 mutex_init(&fs_info->qgroup_rescan_lock);
2284}
2285
2286static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2287 struct btrfs_fs_devices *fs_devices)
2288{
2289 u32 max_active = fs_info->thread_pool_size;
2290 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2291
2292 fs_info->workers =
2293 btrfs_alloc_workqueue(fs_info, "worker",
2294 flags | WQ_HIGHPRI, max_active, 16);
2295
2296 fs_info->delalloc_workers =
2297 btrfs_alloc_workqueue(fs_info, "delalloc",
2298 flags, max_active, 2);
2299
2300 fs_info->flush_workers =
2301 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2302 flags, max_active, 0);
2303
2304 fs_info->caching_workers =
2305 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2306
2307 fs_info->fixup_workers =
2308 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2309
2310 /*
2311 * endios are largely parallel and should have a very
2312 * low idle thresh
2313 */
2314 fs_info->endio_workers =
2315 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2316 fs_info->endio_meta_workers =
2317 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2318 max_active, 4);
2319 fs_info->endio_meta_write_workers =
2320 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2321 max_active, 2);
2322 fs_info->endio_raid56_workers =
2323 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2324 max_active, 4);
2325 fs_info->rmw_workers =
2326 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2327 fs_info->endio_write_workers =
2328 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2329 max_active, 2);
2330 fs_info->endio_freespace_worker =
2331 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2332 max_active, 0);
2333 fs_info->delayed_workers =
2334 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2335 max_active, 0);
2336 fs_info->readahead_workers =
2337 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2338 max_active, 2);
2339 fs_info->qgroup_rescan_workers =
2340 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2341 fs_info->discard_ctl.discard_workers =
2342 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2343
2344 if (!(fs_info->workers && fs_info->delalloc_workers &&
2345 fs_info->flush_workers &&
2346 fs_info->endio_workers && fs_info->endio_meta_workers &&
2347 fs_info->endio_meta_write_workers &&
2348 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2349 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2350 fs_info->caching_workers && fs_info->readahead_workers &&
2351 fs_info->fixup_workers && fs_info->delayed_workers &&
2352 fs_info->qgroup_rescan_workers &&
2353 fs_info->discard_ctl.discard_workers)) {
2354 return -ENOMEM;
2355 }
2356
2357 return 0;
2358}
2359
2360static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2361{
2362 struct crypto_shash *csum_shash;
2363 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2364
2365 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2366
2367 if (IS_ERR(csum_shash)) {
2368 btrfs_err(fs_info, "error allocating %s hash for checksum",
2369 csum_driver);
2370 return PTR_ERR(csum_shash);
2371 }
2372
2373 fs_info->csum_shash = csum_shash;
2374
2375 return 0;
2376}
2377
2378static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2379 struct btrfs_fs_devices *fs_devices)
2380{
2381 int ret;
2382 struct btrfs_root *log_tree_root;
2383 struct btrfs_super_block *disk_super = fs_info->super_copy;
2384 u64 bytenr = btrfs_super_log_root(disk_super);
2385 int level = btrfs_super_log_root_level(disk_super);
2386
2387 if (fs_devices->rw_devices == 0) {
2388 btrfs_warn(fs_info, "log replay required on RO media");
2389 return -EIO;
2390 }
2391
2392 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2393 GFP_KERNEL);
2394 if (!log_tree_root)
2395 return -ENOMEM;
2396
2397 log_tree_root->node = read_tree_block(fs_info, bytenr,
2398 BTRFS_TREE_LOG_OBJECTID,
2399 fs_info->generation + 1, level,
2400 NULL);
2401 if (IS_ERR(log_tree_root->node)) {
2402 btrfs_warn(fs_info, "failed to read log tree");
2403 ret = PTR_ERR(log_tree_root->node);
2404 log_tree_root->node = NULL;
2405 btrfs_put_root(log_tree_root);
2406 return ret;
2407 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2408 btrfs_err(fs_info, "failed to read log tree");
2409 btrfs_put_root(log_tree_root);
2410 return -EIO;
2411 }
2412 /* returns with log_tree_root freed on success */
2413 ret = btrfs_recover_log_trees(log_tree_root);
2414 if (ret) {
2415 btrfs_handle_fs_error(fs_info, ret,
2416 "Failed to recover log tree");
2417 btrfs_put_root(log_tree_root);
2418 return ret;
2419 }
2420
2421 if (sb_rdonly(fs_info->sb)) {
2422 ret = btrfs_commit_super(fs_info);
2423 if (ret)
2424 return ret;
2425 }
2426
2427 return 0;
2428}
2429
2430static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2431{
2432 struct btrfs_root *tree_root = fs_info->tree_root;
2433 struct btrfs_root *root;
2434 struct btrfs_key location;
2435 int ret;
2436
2437 BUG_ON(!fs_info->tree_root);
2438
2439 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2440 location.type = BTRFS_ROOT_ITEM_KEY;
2441 location.offset = 0;
2442
2443 root = btrfs_read_tree_root(tree_root, &location);
2444 if (IS_ERR(root)) {
2445 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2446 ret = PTR_ERR(root);
2447 goto out;
2448 }
2449 } else {
2450 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2451 fs_info->extent_root = root;
2452 }
2453
2454 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2455 root = btrfs_read_tree_root(tree_root, &location);
2456 if (IS_ERR(root)) {
2457 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2458 ret = PTR_ERR(root);
2459 goto out;
2460 }
2461 } else {
2462 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2463 fs_info->dev_root = root;
2464 }
2465 /* Initialize fs_info for all devices in any case */
2466 btrfs_init_devices_late(fs_info);
2467
2468 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2469 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2470 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2471 root = btrfs_read_tree_root(tree_root, &location);
2472 if (IS_ERR(root)) {
2473 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2474 ret = PTR_ERR(root);
2475 goto out;
2476 }
2477 } else {
2478 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2479 fs_info->csum_root = root;
2480 }
2481 }
2482
2483 /*
2484 * This tree can share blocks with some other fs tree during relocation
2485 * and we need a proper setup by btrfs_get_fs_root
2486 */
2487 root = btrfs_get_fs_root(tree_root->fs_info,
2488 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2489 if (IS_ERR(root)) {
2490 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2491 ret = PTR_ERR(root);
2492 goto out;
2493 }
2494 } else {
2495 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2496 fs_info->data_reloc_root = root;
2497 }
2498
2499 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2500 root = btrfs_read_tree_root(tree_root, &location);
2501 if (!IS_ERR(root)) {
2502 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2503 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2504 fs_info->quota_root = root;
2505 }
2506
2507 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2508 root = btrfs_read_tree_root(tree_root, &location);
2509 if (IS_ERR(root)) {
2510 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2511 ret = PTR_ERR(root);
2512 if (ret != -ENOENT)
2513 goto out;
2514 }
2515 } else {
2516 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2517 fs_info->uuid_root = root;
2518 }
2519
2520 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2521 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2522 root = btrfs_read_tree_root(tree_root, &location);
2523 if (IS_ERR(root)) {
2524 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2525 ret = PTR_ERR(root);
2526 goto out;
2527 }
2528 } else {
2529 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2530 fs_info->free_space_root = root;
2531 }
2532 }
2533
2534 return 0;
2535out:
2536 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2537 location.objectid, ret);
2538 return ret;
2539}
2540
2541/*
2542 * Real super block validation
2543 * NOTE: super csum type and incompat features will not be checked here.
2544 *
2545 * @sb: super block to check
2546 * @mirror_num: the super block number to check its bytenr:
2547 * 0 the primary (1st) sb
2548 * 1, 2 2nd and 3rd backup copy
2549 * -1 skip bytenr check
2550 */
2551static int validate_super(struct btrfs_fs_info *fs_info,
2552 struct btrfs_super_block *sb, int mirror_num)
2553{
2554 u64 nodesize = btrfs_super_nodesize(sb);
2555 u64 sectorsize = btrfs_super_sectorsize(sb);
2556 int ret = 0;
2557
2558 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2559 btrfs_err(fs_info, "no valid FS found");
2560 ret = -EINVAL;
2561 }
2562 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2563 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2564 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2565 ret = -EINVAL;
2566 }
2567 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2568 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2569 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2570 ret = -EINVAL;
2571 }
2572 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2573 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2574 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2575 ret = -EINVAL;
2576 }
2577 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2578 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2579 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2580 ret = -EINVAL;
2581 }
2582
2583 /*
2584 * Check sectorsize and nodesize first, other check will need it.
2585 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2586 */
2587 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2588 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2589 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2590 ret = -EINVAL;
2591 }
2592
2593 /*
2594 * For 4K page size, we only support 4K sector size.
2595 * For 64K page size, we support read-write for 64K sector size, and
2596 * read-only for 4K sector size.
2597 */
2598 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2599 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2600 sectorsize != SZ_64K))) {
2601 btrfs_err(fs_info,
2602 "sectorsize %llu not yet supported for page size %lu",
2603 sectorsize, PAGE_SIZE);
2604 ret = -EINVAL;
2605 }
2606
2607 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2608 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2609 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2610 ret = -EINVAL;
2611 }
2612 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2613 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2614 le32_to_cpu(sb->__unused_leafsize), nodesize);
2615 ret = -EINVAL;
2616 }
2617
2618 /* Root alignment check */
2619 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2620 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2621 btrfs_super_root(sb));
2622 ret = -EINVAL;
2623 }
2624 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2625 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2626 btrfs_super_chunk_root(sb));
2627 ret = -EINVAL;
2628 }
2629 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2630 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2631 btrfs_super_log_root(sb));
2632 ret = -EINVAL;
2633 }
2634
2635 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2636 BTRFS_FSID_SIZE)) {
2637 btrfs_err(fs_info,
2638 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2639 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2640 ret = -EINVAL;
2641 }
2642
2643 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2644 memcmp(fs_info->fs_devices->metadata_uuid,
2645 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2646 btrfs_err(fs_info,
2647"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2648 fs_info->super_copy->metadata_uuid,
2649 fs_info->fs_devices->metadata_uuid);
2650 ret = -EINVAL;
2651 }
2652
2653 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2654 BTRFS_FSID_SIZE) != 0) {
2655 btrfs_err(fs_info,
2656 "dev_item UUID does not match metadata fsid: %pU != %pU",
2657 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2658 ret = -EINVAL;
2659 }
2660
2661 /*
2662 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2663 * done later
2664 */
2665 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2666 btrfs_err(fs_info, "bytes_used is too small %llu",
2667 btrfs_super_bytes_used(sb));
2668 ret = -EINVAL;
2669 }
2670 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2671 btrfs_err(fs_info, "invalid stripesize %u",
2672 btrfs_super_stripesize(sb));
2673 ret = -EINVAL;
2674 }
2675 if (btrfs_super_num_devices(sb) > (1UL << 31))
2676 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2677 btrfs_super_num_devices(sb));
2678 if (btrfs_super_num_devices(sb) == 0) {
2679 btrfs_err(fs_info, "number of devices is 0");
2680 ret = -EINVAL;
2681 }
2682
2683 if (mirror_num >= 0 &&
2684 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2685 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2686 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2687 ret = -EINVAL;
2688 }
2689
2690 /*
2691 * Obvious sys_chunk_array corruptions, it must hold at least one key
2692 * and one chunk
2693 */
2694 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2695 btrfs_err(fs_info, "system chunk array too big %u > %u",
2696 btrfs_super_sys_array_size(sb),
2697 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2698 ret = -EINVAL;
2699 }
2700 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2701 + sizeof(struct btrfs_chunk)) {
2702 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2703 btrfs_super_sys_array_size(sb),
2704 sizeof(struct btrfs_disk_key)
2705 + sizeof(struct btrfs_chunk));
2706 ret = -EINVAL;
2707 }
2708
2709 /*
2710 * The generation is a global counter, we'll trust it more than the others
2711 * but it's still possible that it's the one that's wrong.
2712 */
2713 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2714 btrfs_warn(fs_info,
2715 "suspicious: generation < chunk_root_generation: %llu < %llu",
2716 btrfs_super_generation(sb),
2717 btrfs_super_chunk_root_generation(sb));
2718 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2719 && btrfs_super_cache_generation(sb) != (u64)-1)
2720 btrfs_warn(fs_info,
2721 "suspicious: generation < cache_generation: %llu < %llu",
2722 btrfs_super_generation(sb),
2723 btrfs_super_cache_generation(sb));
2724
2725 return ret;
2726}
2727
2728/*
2729 * Validation of super block at mount time.
2730 * Some checks already done early at mount time, like csum type and incompat
2731 * flags will be skipped.
2732 */
2733static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2734{
2735 return validate_super(fs_info, fs_info->super_copy, 0);
2736}
2737
2738/*
2739 * Validation of super block at write time.
2740 * Some checks like bytenr check will be skipped as their values will be
2741 * overwritten soon.
2742 * Extra checks like csum type and incompat flags will be done here.
2743 */
2744static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2745 struct btrfs_super_block *sb)
2746{
2747 int ret;
2748
2749 ret = validate_super(fs_info, sb, -1);
2750 if (ret < 0)
2751 goto out;
2752 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2753 ret = -EUCLEAN;
2754 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2755 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2756 goto out;
2757 }
2758 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2759 ret = -EUCLEAN;
2760 btrfs_err(fs_info,
2761 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2762 btrfs_super_incompat_flags(sb),
2763 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2764 goto out;
2765 }
2766out:
2767 if (ret < 0)
2768 btrfs_err(fs_info,
2769 "super block corruption detected before writing it to disk");
2770 return ret;
2771}
2772
2773static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2774{
2775 int backup_index = find_newest_super_backup(fs_info);
2776 struct btrfs_super_block *sb = fs_info->super_copy;
2777 struct btrfs_root *tree_root = fs_info->tree_root;
2778 bool handle_error = false;
2779 int ret = 0;
2780 int i;
2781
2782 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2783 u64 generation;
2784 int level;
2785
2786 if (handle_error) {
2787 if (!IS_ERR(tree_root->node))
2788 free_extent_buffer(tree_root->node);
2789 tree_root->node = NULL;
2790
2791 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2792 break;
2793
2794 free_root_pointers(fs_info, 0);
2795
2796 /*
2797 * Don't use the log in recovery mode, it won't be
2798 * valid
2799 */
2800 btrfs_set_super_log_root(sb, 0);
2801
2802 /* We can't trust the free space cache either */
2803 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2804
2805 ret = read_backup_root(fs_info, i);
2806 backup_index = ret;
2807 if (ret < 0)
2808 return ret;
2809 }
2810 generation = btrfs_super_generation(sb);
2811 level = btrfs_super_root_level(sb);
2812 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2813 BTRFS_ROOT_TREE_OBJECTID,
2814 generation, level, NULL);
2815 if (IS_ERR(tree_root->node)) {
2816 handle_error = true;
2817 ret = PTR_ERR(tree_root->node);
2818 tree_root->node = NULL;
2819 btrfs_warn(fs_info, "couldn't read tree root");
2820 continue;
2821
2822 } else if (!extent_buffer_uptodate(tree_root->node)) {
2823 handle_error = true;
2824 ret = -EIO;
2825 btrfs_warn(fs_info, "error while reading tree root");
2826 continue;
2827 }
2828
2829 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2830 tree_root->commit_root = btrfs_root_node(tree_root);
2831 btrfs_set_root_refs(&tree_root->root_item, 1);
2832
2833 /*
2834 * No need to hold btrfs_root::objectid_mutex since the fs
2835 * hasn't been fully initialised and we are the only user
2836 */
2837 ret = btrfs_init_root_free_objectid(tree_root);
2838 if (ret < 0) {
2839 handle_error = true;
2840 continue;
2841 }
2842
2843 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2844
2845 ret = btrfs_read_roots(fs_info);
2846 if (ret < 0) {
2847 handle_error = true;
2848 continue;
2849 }
2850
2851 /* All successful */
2852 fs_info->generation = generation;
2853 fs_info->last_trans_committed = generation;
2854
2855 /* Always begin writing backup roots after the one being used */
2856 if (backup_index < 0) {
2857 fs_info->backup_root_index = 0;
2858 } else {
2859 fs_info->backup_root_index = backup_index + 1;
2860 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2861 }
2862 break;
2863 }
2864
2865 return ret;
2866}
2867
2868void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2869{
2870 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2871 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2872 INIT_LIST_HEAD(&fs_info->trans_list);
2873 INIT_LIST_HEAD(&fs_info->dead_roots);
2874 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2875 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2876 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2877 spin_lock_init(&fs_info->delalloc_root_lock);
2878 spin_lock_init(&fs_info->trans_lock);
2879 spin_lock_init(&fs_info->fs_roots_radix_lock);
2880 spin_lock_init(&fs_info->delayed_iput_lock);
2881 spin_lock_init(&fs_info->defrag_inodes_lock);
2882 spin_lock_init(&fs_info->super_lock);
2883 spin_lock_init(&fs_info->buffer_lock);
2884 spin_lock_init(&fs_info->unused_bgs_lock);
2885 spin_lock_init(&fs_info->treelog_bg_lock);
2886 rwlock_init(&fs_info->tree_mod_log_lock);
2887 mutex_init(&fs_info->unused_bg_unpin_mutex);
2888 mutex_init(&fs_info->reclaim_bgs_lock);
2889 mutex_init(&fs_info->reloc_mutex);
2890 mutex_init(&fs_info->delalloc_root_mutex);
2891 mutex_init(&fs_info->zoned_meta_io_lock);
2892 seqlock_init(&fs_info->profiles_lock);
2893
2894 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2895 INIT_LIST_HEAD(&fs_info->space_info);
2896 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2897 INIT_LIST_HEAD(&fs_info->unused_bgs);
2898 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2899#ifdef CONFIG_BTRFS_DEBUG
2900 INIT_LIST_HEAD(&fs_info->allocated_roots);
2901 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2902 spin_lock_init(&fs_info->eb_leak_lock);
2903#endif
2904 extent_map_tree_init(&fs_info->mapping_tree);
2905 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2906 BTRFS_BLOCK_RSV_GLOBAL);
2907 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2908 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2909 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2910 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2911 BTRFS_BLOCK_RSV_DELOPS);
2912 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2913 BTRFS_BLOCK_RSV_DELREFS);
2914
2915 atomic_set(&fs_info->async_delalloc_pages, 0);
2916 atomic_set(&fs_info->defrag_running, 0);
2917 atomic_set(&fs_info->reada_works_cnt, 0);
2918 atomic_set(&fs_info->nr_delayed_iputs, 0);
2919 atomic64_set(&fs_info->tree_mod_seq, 0);
2920 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2921 fs_info->metadata_ratio = 0;
2922 fs_info->defrag_inodes = RB_ROOT;
2923 atomic64_set(&fs_info->free_chunk_space, 0);
2924 fs_info->tree_mod_log = RB_ROOT;
2925 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2926 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2927 /* readahead state */
2928 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2929 spin_lock_init(&fs_info->reada_lock);
2930 btrfs_init_ref_verify(fs_info);
2931
2932 fs_info->thread_pool_size = min_t(unsigned long,
2933 num_online_cpus() + 2, 8);
2934
2935 INIT_LIST_HEAD(&fs_info->ordered_roots);
2936 spin_lock_init(&fs_info->ordered_root_lock);
2937
2938 btrfs_init_scrub(fs_info);
2939#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2940 fs_info->check_integrity_print_mask = 0;
2941#endif
2942 btrfs_init_balance(fs_info);
2943 btrfs_init_async_reclaim_work(fs_info);
2944
2945 spin_lock_init(&fs_info->block_group_cache_lock);
2946 fs_info->block_group_cache_tree = RB_ROOT;
2947 fs_info->first_logical_byte = (u64)-1;
2948
2949 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2950 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2951 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2952
2953 mutex_init(&fs_info->ordered_operations_mutex);
2954 mutex_init(&fs_info->tree_log_mutex);
2955 mutex_init(&fs_info->chunk_mutex);
2956 mutex_init(&fs_info->transaction_kthread_mutex);
2957 mutex_init(&fs_info->cleaner_mutex);
2958 mutex_init(&fs_info->ro_block_group_mutex);
2959 init_rwsem(&fs_info->commit_root_sem);
2960 init_rwsem(&fs_info->cleanup_work_sem);
2961 init_rwsem(&fs_info->subvol_sem);
2962 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2963
2964 btrfs_init_dev_replace_locks(fs_info);
2965 btrfs_init_qgroup(fs_info);
2966 btrfs_discard_init(fs_info);
2967
2968 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2969 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2970
2971 init_waitqueue_head(&fs_info->transaction_throttle);
2972 init_waitqueue_head(&fs_info->transaction_wait);
2973 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2974 init_waitqueue_head(&fs_info->async_submit_wait);
2975 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2976
2977 /* Usable values until the real ones are cached from the superblock */
2978 fs_info->nodesize = 4096;
2979 fs_info->sectorsize = 4096;
2980 fs_info->sectorsize_bits = ilog2(4096);
2981 fs_info->stripesize = 4096;
2982
2983 spin_lock_init(&fs_info->swapfile_pins_lock);
2984 fs_info->swapfile_pins = RB_ROOT;
2985
2986 spin_lock_init(&fs_info->send_reloc_lock);
2987 fs_info->send_in_progress = 0;
2988
2989 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2990 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2991}
2992
2993static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2994{
2995 int ret;
2996
2997 fs_info->sb = sb;
2998 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2999 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3000
3001 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3002 if (ret)
3003 return ret;
3004
3005 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3006 if (ret)
3007 return ret;
3008
3009 fs_info->dirty_metadata_batch = PAGE_SIZE *
3010 (1 + ilog2(nr_cpu_ids));
3011
3012 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3013 if (ret)
3014 return ret;
3015
3016 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3017 GFP_KERNEL);
3018 if (ret)
3019 return ret;
3020
3021 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3022 GFP_KERNEL);
3023 if (!fs_info->delayed_root)
3024 return -ENOMEM;
3025 btrfs_init_delayed_root(fs_info->delayed_root);
3026
3027 if (sb_rdonly(sb))
3028 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3029
3030 return btrfs_alloc_stripe_hash_table(fs_info);
3031}
3032
3033static int btrfs_uuid_rescan_kthread(void *data)
3034{
3035 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3036 int ret;
3037
3038 /*
3039 * 1st step is to iterate through the existing UUID tree and
3040 * to delete all entries that contain outdated data.
3041 * 2nd step is to add all missing entries to the UUID tree.
3042 */
3043 ret = btrfs_uuid_tree_iterate(fs_info);
3044 if (ret < 0) {
3045 if (ret != -EINTR)
3046 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3047 ret);
3048 up(&fs_info->uuid_tree_rescan_sem);
3049 return ret;
3050 }
3051 return btrfs_uuid_scan_kthread(data);
3052}
3053
3054static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3055{
3056 struct task_struct *task;
3057
3058 down(&fs_info->uuid_tree_rescan_sem);
3059 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3060 if (IS_ERR(task)) {
3061 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3062 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3063 up(&fs_info->uuid_tree_rescan_sem);
3064 return PTR_ERR(task);
3065 }
3066
3067 return 0;
3068}
3069
3070/*
3071 * Some options only have meaning at mount time and shouldn't persist across
3072 * remounts, or be displayed. Clear these at the end of mount and remount
3073 * code paths.
3074 */
3075void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3076{
3077 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3078 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3079}
3080
3081/*
3082 * Mounting logic specific to read-write file systems. Shared by open_ctree
3083 * and btrfs_remount when remounting from read-only to read-write.
3084 */
3085int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3086{
3087 int ret;
3088 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3089 bool clear_free_space_tree = false;
3090
3091 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3092 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3093 clear_free_space_tree = true;
3094 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3095 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3096 btrfs_warn(fs_info, "free space tree is invalid");
3097 clear_free_space_tree = true;
3098 }
3099
3100 if (clear_free_space_tree) {
3101 btrfs_info(fs_info, "clearing free space tree");
3102 ret = btrfs_clear_free_space_tree(fs_info);
3103 if (ret) {
3104 btrfs_warn(fs_info,
3105 "failed to clear free space tree: %d", ret);
3106 goto out;
3107 }
3108 }
3109
3110 /*
3111 * btrfs_find_orphan_roots() is responsible for finding all the dead
3112 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3113 * them into the fs_info->fs_roots_radix tree. This must be done before
3114 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3115 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3116 * item before the root's tree is deleted - this means that if we unmount
3117 * or crash before the deletion completes, on the next mount we will not
3118 * delete what remains of the tree because the orphan item does not
3119 * exists anymore, which is what tells us we have a pending deletion.
3120 */
3121 ret = btrfs_find_orphan_roots(fs_info);
3122 if (ret)
3123 goto out;
3124
3125 ret = btrfs_cleanup_fs_roots(fs_info);
3126 if (ret)
3127 goto out;
3128
3129 down_read(&fs_info->cleanup_work_sem);
3130 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3131 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3132 up_read(&fs_info->cleanup_work_sem);
3133 goto out;
3134 }
3135 up_read(&fs_info->cleanup_work_sem);
3136
3137 mutex_lock(&fs_info->cleaner_mutex);
3138 ret = btrfs_recover_relocation(fs_info->tree_root);
3139 mutex_unlock(&fs_info->cleaner_mutex);
3140 if (ret < 0) {
3141 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3142 goto out;
3143 }
3144
3145 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3146 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3147 btrfs_info(fs_info, "creating free space tree");
3148 ret = btrfs_create_free_space_tree(fs_info);
3149 if (ret) {
3150 btrfs_warn(fs_info,
3151 "failed to create free space tree: %d", ret);
3152 goto out;
3153 }
3154 }
3155
3156 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3157 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3158 if (ret)
3159 goto out;
3160 }
3161
3162 ret = btrfs_resume_balance_async(fs_info);
3163 if (ret)
3164 goto out;
3165
3166 ret = btrfs_resume_dev_replace_async(fs_info);
3167 if (ret) {
3168 btrfs_warn(fs_info, "failed to resume dev_replace");
3169 goto out;
3170 }
3171
3172 btrfs_qgroup_rescan_resume(fs_info);
3173
3174 if (!fs_info->uuid_root) {
3175 btrfs_info(fs_info, "creating UUID tree");
3176 ret = btrfs_create_uuid_tree(fs_info);
3177 if (ret) {
3178 btrfs_warn(fs_info,
3179 "failed to create the UUID tree %d", ret);
3180 goto out;
3181 }
3182 }
3183
3184out:
3185 return ret;
3186}
3187
3188int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3189 char *options)
3190{
3191 u32 sectorsize;
3192 u32 nodesize;
3193 u32 stripesize;
3194 u64 generation;
3195 u64 features;
3196 u16 csum_type;
3197 struct btrfs_super_block *disk_super;
3198 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3199 struct btrfs_root *tree_root;
3200 struct btrfs_root *chunk_root;
3201 int ret;
3202 int err = -EINVAL;
3203 int level;
3204
3205 ret = init_mount_fs_info(fs_info, sb);
3206 if (ret) {
3207 err = ret;
3208 goto fail;
3209 }
3210
3211 /* These need to be init'ed before we start creating inodes and such. */
3212 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3213 GFP_KERNEL);
3214 fs_info->tree_root = tree_root;
3215 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3216 GFP_KERNEL);
3217 fs_info->chunk_root = chunk_root;
3218 if (!tree_root || !chunk_root) {
3219 err = -ENOMEM;
3220 goto fail;
3221 }
3222
3223 fs_info->btree_inode = new_inode(sb);
3224 if (!fs_info->btree_inode) {
3225 err = -ENOMEM;
3226 goto fail;
3227 }
3228 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3229 btrfs_init_btree_inode(fs_info);
3230
3231 invalidate_bdev(fs_devices->latest_bdev);
3232
3233 /*
3234 * Read super block and check the signature bytes only
3235 */
3236 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
3237 if (IS_ERR(disk_super)) {
3238 err = PTR_ERR(disk_super);
3239 goto fail_alloc;
3240 }
3241
3242 /*
3243 * Verify the type first, if that or the checksum value are
3244 * corrupted, we'll find out
3245 */
3246 csum_type = btrfs_super_csum_type(disk_super);
3247 if (!btrfs_supported_super_csum(csum_type)) {
3248 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3249 csum_type);
3250 err = -EINVAL;
3251 btrfs_release_disk_super(disk_super);
3252 goto fail_alloc;
3253 }
3254
3255 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3256
3257 ret = btrfs_init_csum_hash(fs_info, csum_type);
3258 if (ret) {
3259 err = ret;
3260 btrfs_release_disk_super(disk_super);
3261 goto fail_alloc;
3262 }
3263
3264 /*
3265 * We want to check superblock checksum, the type is stored inside.
3266 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3267 */
3268 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3269 btrfs_err(fs_info, "superblock checksum mismatch");
3270 err = -EINVAL;
3271 btrfs_release_disk_super(disk_super);
3272 goto fail_alloc;
3273 }
3274
3275 /*
3276 * super_copy is zeroed at allocation time and we never touch the
3277 * following bytes up to INFO_SIZE, the checksum is calculated from
3278 * the whole block of INFO_SIZE
3279 */
3280 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3281 btrfs_release_disk_super(disk_super);
3282
3283 disk_super = fs_info->super_copy;
3284
3285
3286 features = btrfs_super_flags(disk_super);
3287 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3288 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3289 btrfs_set_super_flags(disk_super, features);
3290 btrfs_info(fs_info,
3291 "found metadata UUID change in progress flag, clearing");
3292 }
3293
3294 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3295 sizeof(*fs_info->super_for_commit));
3296
3297 ret = btrfs_validate_mount_super(fs_info);
3298 if (ret) {
3299 btrfs_err(fs_info, "superblock contains fatal errors");
3300 err = -EINVAL;
3301 goto fail_alloc;
3302 }
3303
3304 if (!btrfs_super_root(disk_super))
3305 goto fail_alloc;
3306
3307 /* check FS state, whether FS is broken. */
3308 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3309 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3310
3311 /*
3312 * In the long term, we'll store the compression type in the super
3313 * block, and it'll be used for per file compression control.
3314 */
3315 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3316
3317 /*
3318 * Flag our filesystem as having big metadata blocks if they are bigger
3319 * than the page size.
3320 */
3321 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3322 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3323 btrfs_info(fs_info,
3324 "flagging fs with big metadata feature");
3325 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3326 }
3327
3328 /* Set up fs_info before parsing mount options */
3329 nodesize = btrfs_super_nodesize(disk_super);
3330 sectorsize = btrfs_super_sectorsize(disk_super);
3331 stripesize = sectorsize;
3332 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3333 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3334
3335 fs_info->nodesize = nodesize;
3336 fs_info->sectorsize = sectorsize;
3337 fs_info->sectorsize_bits = ilog2(sectorsize);
3338 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3339 fs_info->stripesize = stripesize;
3340
3341 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3342 if (ret) {
3343 err = ret;
3344 goto fail_alloc;
3345 }
3346
3347 features = btrfs_super_incompat_flags(disk_super) &
3348 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3349 if (features) {
3350 btrfs_err(fs_info,
3351 "cannot mount because of unsupported optional features (%llx)",
3352 features);
3353 err = -EINVAL;
3354 goto fail_alloc;
3355 }
3356
3357 features = btrfs_super_incompat_flags(disk_super);
3358 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3359 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3360 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3361 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3362 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3363
3364 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3365 btrfs_info(fs_info, "has skinny extents");
3366
3367 /*
3368 * mixed block groups end up with duplicate but slightly offset
3369 * extent buffers for the same range. It leads to corruptions
3370 */
3371 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3372 (sectorsize != nodesize)) {
3373 btrfs_err(fs_info,
3374"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3375 nodesize, sectorsize);
3376 goto fail_alloc;
3377 }
3378
3379 /*
3380 * Needn't use the lock because there is no other task which will
3381 * update the flag.
3382 */
3383 btrfs_set_super_incompat_flags(disk_super, features);
3384
3385 features = btrfs_super_compat_ro_flags(disk_super) &
3386 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3387 if (!sb_rdonly(sb) && features) {
3388 btrfs_err(fs_info,
3389 "cannot mount read-write because of unsupported optional features (%llx)",
3390 features);
3391 err = -EINVAL;
3392 goto fail_alloc;
3393 }
3394
3395 /* For 4K sector size support, it's only read-only */
3396 if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) {
3397 if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) {
3398 btrfs_err(fs_info,
3399 "subpage sectorsize %u only supported read-only for page size %lu",
3400 sectorsize, PAGE_SIZE);
3401 err = -EINVAL;
3402 goto fail_alloc;
3403 }
3404 }
3405
3406 ret = btrfs_init_workqueues(fs_info, fs_devices);
3407 if (ret) {
3408 err = ret;
3409 goto fail_sb_buffer;
3410 }
3411
3412 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3413 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3414
3415 sb->s_blocksize = sectorsize;
3416 sb->s_blocksize_bits = blksize_bits(sectorsize);
3417 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3418
3419 mutex_lock(&fs_info->chunk_mutex);
3420 ret = btrfs_read_sys_array(fs_info);
3421 mutex_unlock(&fs_info->chunk_mutex);
3422 if (ret) {
3423 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3424 goto fail_sb_buffer;
3425 }
3426
3427 generation = btrfs_super_chunk_root_generation(disk_super);
3428 level = btrfs_super_chunk_root_level(disk_super);
3429
3430 chunk_root->node = read_tree_block(fs_info,
3431 btrfs_super_chunk_root(disk_super),
3432 BTRFS_CHUNK_TREE_OBJECTID,
3433 generation, level, NULL);
3434 if (IS_ERR(chunk_root->node) ||
3435 !extent_buffer_uptodate(chunk_root->node)) {
3436 btrfs_err(fs_info, "failed to read chunk root");
3437 if (!IS_ERR(chunk_root->node))
3438 free_extent_buffer(chunk_root->node);
3439 chunk_root->node = NULL;
3440 goto fail_tree_roots;
3441 }
3442 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3443 chunk_root->commit_root = btrfs_root_node(chunk_root);
3444
3445 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3446 offsetof(struct btrfs_header, chunk_tree_uuid),
3447 BTRFS_UUID_SIZE);
3448
3449 ret = btrfs_read_chunk_tree(fs_info);
3450 if (ret) {
3451 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3452 goto fail_tree_roots;
3453 }
3454
3455 /*
3456 * At this point we know all the devices that make this filesystem,
3457 * including the seed devices but we don't know yet if the replace
3458 * target is required. So free devices that are not part of this
3459 * filesystem but skip the replace target device which is checked
3460 * below in btrfs_init_dev_replace().
3461 */
3462 btrfs_free_extra_devids(fs_devices);
3463 if (!fs_devices->latest_bdev) {
3464 btrfs_err(fs_info, "failed to read devices");
3465 goto fail_tree_roots;
3466 }
3467
3468 ret = init_tree_roots(fs_info);
3469 if (ret)
3470 goto fail_tree_roots;
3471
3472 /*
3473 * Get zone type information of zoned block devices. This will also
3474 * handle emulation of a zoned filesystem if a regular device has the
3475 * zoned incompat feature flag set.
3476 */
3477 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3478 if (ret) {
3479 btrfs_err(fs_info,
3480 "zoned: failed to read device zone info: %d",
3481 ret);
3482 goto fail_block_groups;
3483 }
3484
3485 /*
3486 * If we have a uuid root and we're not being told to rescan we need to
3487 * check the generation here so we can set the
3488 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3489 * transaction during a balance or the log replay without updating the
3490 * uuid generation, and then if we crash we would rescan the uuid tree,
3491 * even though it was perfectly fine.
3492 */
3493 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3494 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3495 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3496
3497 ret = btrfs_verify_dev_extents(fs_info);
3498 if (ret) {
3499 btrfs_err(fs_info,
3500 "failed to verify dev extents against chunks: %d",
3501 ret);
3502 goto fail_block_groups;
3503 }
3504 ret = btrfs_recover_balance(fs_info);
3505 if (ret) {
3506 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3507 goto fail_block_groups;
3508 }
3509
3510 ret = btrfs_init_dev_stats(fs_info);
3511 if (ret) {
3512 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3513 goto fail_block_groups;
3514 }
3515
3516 ret = btrfs_init_dev_replace(fs_info);
3517 if (ret) {
3518 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3519 goto fail_block_groups;
3520 }
3521
3522 ret = btrfs_check_zoned_mode(fs_info);
3523 if (ret) {
3524 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3525 ret);
3526 goto fail_block_groups;
3527 }
3528
3529 ret = btrfs_sysfs_add_fsid(fs_devices);
3530 if (ret) {
3531 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3532 ret);
3533 goto fail_block_groups;
3534 }
3535
3536 ret = btrfs_sysfs_add_mounted(fs_info);
3537 if (ret) {
3538 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3539 goto fail_fsdev_sysfs;
3540 }
3541
3542 ret = btrfs_init_space_info(fs_info);
3543 if (ret) {
3544 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3545 goto fail_sysfs;
3546 }
3547
3548 ret = btrfs_read_block_groups(fs_info);
3549 if (ret) {
3550 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3551 goto fail_sysfs;
3552 }
3553
3554 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3555 btrfs_warn(fs_info,
3556 "writable mount is not allowed due to too many missing devices");
3557 goto fail_sysfs;
3558 }
3559
3560 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3561 "btrfs-cleaner");
3562 if (IS_ERR(fs_info->cleaner_kthread))
3563 goto fail_sysfs;
3564
3565 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3566 tree_root,
3567 "btrfs-transaction");
3568 if (IS_ERR(fs_info->transaction_kthread))
3569 goto fail_cleaner;
3570
3571 if (!btrfs_test_opt(fs_info, NOSSD) &&
3572 !fs_info->fs_devices->rotating) {
3573 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3574 }
3575
3576 /*
3577 * Mount does not set all options immediately, we can do it now and do
3578 * not have to wait for transaction commit
3579 */
3580 btrfs_apply_pending_changes(fs_info);
3581
3582#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3583 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3584 ret = btrfsic_mount(fs_info, fs_devices,
3585 btrfs_test_opt(fs_info,
3586 CHECK_INTEGRITY_DATA) ? 1 : 0,
3587 fs_info->check_integrity_print_mask);
3588 if (ret)
3589 btrfs_warn(fs_info,
3590 "failed to initialize integrity check module: %d",
3591 ret);
3592 }
3593#endif
3594 ret = btrfs_read_qgroup_config(fs_info);
3595 if (ret)
3596 goto fail_trans_kthread;
3597
3598 if (btrfs_build_ref_tree(fs_info))
3599 btrfs_err(fs_info, "couldn't build ref tree");
3600
3601 /* do not make disk changes in broken FS or nologreplay is given */
3602 if (btrfs_super_log_root(disk_super) != 0 &&
3603 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3604 btrfs_info(fs_info, "start tree-log replay");
3605 ret = btrfs_replay_log(fs_info, fs_devices);
3606 if (ret) {
3607 err = ret;
3608 goto fail_qgroup;
3609 }
3610 }
3611
3612 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3613 if (IS_ERR(fs_info->fs_root)) {
3614 err = PTR_ERR(fs_info->fs_root);
3615 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3616 fs_info->fs_root = NULL;
3617 goto fail_qgroup;
3618 }
3619
3620 if (sb_rdonly(sb))
3621 goto clear_oneshot;
3622
3623 ret = btrfs_start_pre_rw_mount(fs_info);
3624 if (ret) {
3625 close_ctree(fs_info);
3626 return ret;
3627 }
3628 btrfs_discard_resume(fs_info);
3629
3630 if (fs_info->uuid_root &&
3631 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3632 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3633 btrfs_info(fs_info, "checking UUID tree");
3634 ret = btrfs_check_uuid_tree(fs_info);
3635 if (ret) {
3636 btrfs_warn(fs_info,
3637 "failed to check the UUID tree: %d", ret);
3638 close_ctree(fs_info);
3639 return ret;
3640 }
3641 }
3642
3643 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3644
3645clear_oneshot:
3646 btrfs_clear_oneshot_options(fs_info);
3647 return 0;
3648
3649fail_qgroup:
3650 btrfs_free_qgroup_config(fs_info);
3651fail_trans_kthread:
3652 kthread_stop(fs_info->transaction_kthread);
3653 btrfs_cleanup_transaction(fs_info);
3654 btrfs_free_fs_roots(fs_info);
3655fail_cleaner:
3656 kthread_stop(fs_info->cleaner_kthread);
3657
3658 /*
3659 * make sure we're done with the btree inode before we stop our
3660 * kthreads
3661 */
3662 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3663
3664fail_sysfs:
3665 btrfs_sysfs_remove_mounted(fs_info);
3666
3667fail_fsdev_sysfs:
3668 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3669
3670fail_block_groups:
3671 btrfs_put_block_group_cache(fs_info);
3672
3673fail_tree_roots:
3674 if (fs_info->data_reloc_root)
3675 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3676 free_root_pointers(fs_info, true);
3677 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3678
3679fail_sb_buffer:
3680 btrfs_stop_all_workers(fs_info);
3681 btrfs_free_block_groups(fs_info);
3682fail_alloc:
3683 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3684
3685 iput(fs_info->btree_inode);
3686fail:
3687 btrfs_close_devices(fs_info->fs_devices);
3688 return err;
3689}
3690ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3691
3692static void btrfs_end_super_write(struct bio *bio)
3693{
3694 struct btrfs_device *device = bio->bi_private;
3695 struct bio_vec *bvec;
3696 struct bvec_iter_all iter_all;
3697 struct page *page;
3698
3699 bio_for_each_segment_all(bvec, bio, iter_all) {
3700 page = bvec->bv_page;
3701
3702 if (bio->bi_status) {
3703 btrfs_warn_rl_in_rcu(device->fs_info,
3704 "lost page write due to IO error on %s (%d)",
3705 rcu_str_deref(device->name),
3706 blk_status_to_errno(bio->bi_status));
3707 ClearPageUptodate(page);
3708 SetPageError(page);
3709 btrfs_dev_stat_inc_and_print(device,
3710 BTRFS_DEV_STAT_WRITE_ERRS);
3711 } else {
3712 SetPageUptodate(page);
3713 }
3714
3715 put_page(page);
3716 unlock_page(page);
3717 }
3718
3719 bio_put(bio);
3720}
3721
3722struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3723 int copy_num)
3724{
3725 struct btrfs_super_block *super;
3726 struct page *page;
3727 u64 bytenr, bytenr_orig;
3728 struct address_space *mapping = bdev->bd_inode->i_mapping;
3729 int ret;
3730
3731 bytenr_orig = btrfs_sb_offset(copy_num);
3732 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3733 if (ret == -ENOENT)
3734 return ERR_PTR(-EINVAL);
3735 else if (ret)
3736 return ERR_PTR(ret);
3737
3738 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3739 return ERR_PTR(-EINVAL);
3740
3741 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3742 if (IS_ERR(page))
3743 return ERR_CAST(page);
3744
3745 super = page_address(page);
3746 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3747 btrfs_release_disk_super(super);
3748 return ERR_PTR(-ENODATA);
3749 }
3750
3751 if (btrfs_super_bytenr(super) != bytenr_orig) {
3752 btrfs_release_disk_super(super);
3753 return ERR_PTR(-EINVAL);
3754 }
3755
3756 return super;
3757}
3758
3759
3760struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3761{
3762 struct btrfs_super_block *super, *latest = NULL;
3763 int i;
3764 u64 transid = 0;
3765
3766 /* we would like to check all the supers, but that would make
3767 * a btrfs mount succeed after a mkfs from a different FS.
3768 * So, we need to add a special mount option to scan for
3769 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3770 */
3771 for (i = 0; i < 1; i++) {
3772 super = btrfs_read_dev_one_super(bdev, i);
3773 if (IS_ERR(super))
3774 continue;
3775
3776 if (!latest || btrfs_super_generation(super) > transid) {
3777 if (latest)
3778 btrfs_release_disk_super(super);
3779
3780 latest = super;
3781 transid = btrfs_super_generation(super);
3782 }
3783 }
3784
3785 return super;
3786}
3787
3788/*
3789 * Write superblock @sb to the @device. Do not wait for completion, all the
3790 * pages we use for writing are locked.
3791 *
3792 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3793 * the expected device size at commit time. Note that max_mirrors must be
3794 * same for write and wait phases.
3795 *
3796 * Return number of errors when page is not found or submission fails.
3797 */
3798static int write_dev_supers(struct btrfs_device *device,
3799 struct btrfs_super_block *sb, int max_mirrors)
3800{
3801 struct btrfs_fs_info *fs_info = device->fs_info;
3802 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3803 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3804 int i;
3805 int errors = 0;
3806 int ret;
3807 u64 bytenr, bytenr_orig;
3808
3809 if (max_mirrors == 0)
3810 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3811
3812 shash->tfm = fs_info->csum_shash;
3813
3814 for (i = 0; i < max_mirrors; i++) {
3815 struct page *page;
3816 struct bio *bio;
3817 struct btrfs_super_block *disk_super;
3818
3819 bytenr_orig = btrfs_sb_offset(i);
3820 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3821 if (ret == -ENOENT) {
3822 continue;
3823 } else if (ret < 0) {
3824 btrfs_err(device->fs_info,
3825 "couldn't get super block location for mirror %d",
3826 i);
3827 errors++;
3828 continue;
3829 }
3830 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3831 device->commit_total_bytes)
3832 break;
3833
3834 btrfs_set_super_bytenr(sb, bytenr_orig);
3835
3836 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3837 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3838 sb->csum);
3839
3840 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3841 GFP_NOFS);
3842 if (!page) {
3843 btrfs_err(device->fs_info,
3844 "couldn't get super block page for bytenr %llu",
3845 bytenr);
3846 errors++;
3847 continue;
3848 }
3849
3850 /* Bump the refcount for wait_dev_supers() */
3851 get_page(page);
3852
3853 disk_super = page_address(page);
3854 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3855
3856 /*
3857 * Directly use bios here instead of relying on the page cache
3858 * to do I/O, so we don't lose the ability to do integrity
3859 * checking.
3860 */
3861 bio = bio_alloc(GFP_NOFS, 1);
3862 bio_set_dev(bio, device->bdev);
3863 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3864 bio->bi_private = device;
3865 bio->bi_end_io = btrfs_end_super_write;
3866 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3867 offset_in_page(bytenr));
3868
3869 /*
3870 * We FUA only the first super block. The others we allow to
3871 * go down lazy and there's a short window where the on-disk
3872 * copies might still contain the older version.
3873 */
3874 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3875 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3876 bio->bi_opf |= REQ_FUA;
3877
3878 btrfsic_submit_bio(bio);
3879 btrfs_advance_sb_log(device, i);
3880 }
3881 return errors < i ? 0 : -1;
3882}
3883
3884/*
3885 * Wait for write completion of superblocks done by write_dev_supers,
3886 * @max_mirrors same for write and wait phases.
3887 *
3888 * Return number of errors when page is not found or not marked up to
3889 * date.
3890 */
3891static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3892{
3893 int i;
3894 int errors = 0;
3895 bool primary_failed = false;
3896 int ret;
3897 u64 bytenr;
3898
3899 if (max_mirrors == 0)
3900 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3901
3902 for (i = 0; i < max_mirrors; i++) {
3903 struct page *page;
3904
3905 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3906 if (ret == -ENOENT) {
3907 break;
3908 } else if (ret < 0) {
3909 errors++;
3910 if (i == 0)
3911 primary_failed = true;
3912 continue;
3913 }
3914 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3915 device->commit_total_bytes)
3916 break;
3917
3918 page = find_get_page(device->bdev->bd_inode->i_mapping,
3919 bytenr >> PAGE_SHIFT);
3920 if (!page) {
3921 errors++;
3922 if (i == 0)
3923 primary_failed = true;
3924 continue;
3925 }
3926 /* Page is submitted locked and unlocked once the IO completes */
3927 wait_on_page_locked(page);
3928 if (PageError(page)) {
3929 errors++;
3930 if (i == 0)
3931 primary_failed = true;
3932 }
3933
3934 /* Drop our reference */
3935 put_page(page);
3936
3937 /* Drop the reference from the writing run */
3938 put_page(page);
3939 }
3940
3941 /* log error, force error return */
3942 if (primary_failed) {
3943 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3944 device->devid);
3945 return -1;
3946 }
3947
3948 return errors < i ? 0 : -1;
3949}
3950
3951/*
3952 * endio for the write_dev_flush, this will wake anyone waiting
3953 * for the barrier when it is done
3954 */
3955static void btrfs_end_empty_barrier(struct bio *bio)
3956{
3957 complete(bio->bi_private);
3958}
3959
3960/*
3961 * Submit a flush request to the device if it supports it. Error handling is
3962 * done in the waiting counterpart.
3963 */
3964static void write_dev_flush(struct btrfs_device *device)
3965{
3966 struct request_queue *q = bdev_get_queue(device->bdev);
3967 struct bio *bio = device->flush_bio;
3968
3969 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3970 return;
3971
3972 bio_reset(bio);
3973 bio->bi_end_io = btrfs_end_empty_barrier;
3974 bio_set_dev(bio, device->bdev);
3975 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3976 init_completion(&device->flush_wait);
3977 bio->bi_private = &device->flush_wait;
3978
3979 btrfsic_submit_bio(bio);
3980 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3981}
3982
3983/*
3984 * If the flush bio has been submitted by write_dev_flush, wait for it.
3985 */
3986static blk_status_t wait_dev_flush(struct btrfs_device *device)
3987{
3988 struct bio *bio = device->flush_bio;
3989
3990 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3991 return BLK_STS_OK;
3992
3993 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3994 wait_for_completion_io(&device->flush_wait);
3995
3996 return bio->bi_status;
3997}
3998
3999static int check_barrier_error(struct btrfs_fs_info *fs_info)
4000{
4001 if (!btrfs_check_rw_degradable(fs_info, NULL))
4002 return -EIO;
4003 return 0;
4004}
4005
4006/*
4007 * send an empty flush down to each device in parallel,
4008 * then wait for them
4009 */
4010static int barrier_all_devices(struct btrfs_fs_info *info)
4011{
4012 struct list_head *head;
4013 struct btrfs_device *dev;
4014 int errors_wait = 0;
4015 blk_status_t ret;
4016
4017 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4018 /* send down all the barriers */
4019 head = &info->fs_devices->devices;
4020 list_for_each_entry(dev, head, dev_list) {
4021 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4022 continue;
4023 if (!dev->bdev)
4024 continue;
4025 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4026 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4027 continue;
4028
4029 write_dev_flush(dev);
4030 dev->last_flush_error = BLK_STS_OK;
4031 }
4032
4033 /* wait for all the barriers */
4034 list_for_each_entry(dev, head, dev_list) {
4035 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4036 continue;
4037 if (!dev->bdev) {
4038 errors_wait++;
4039 continue;
4040 }
4041 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4042 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4043 continue;
4044
4045 ret = wait_dev_flush(dev);
4046 if (ret) {
4047 dev->last_flush_error = ret;
4048 btrfs_dev_stat_inc_and_print(dev,
4049 BTRFS_DEV_STAT_FLUSH_ERRS);
4050 errors_wait++;
4051 }
4052 }
4053
4054 if (errors_wait) {
4055 /*
4056 * At some point we need the status of all disks
4057 * to arrive at the volume status. So error checking
4058 * is being pushed to a separate loop.
4059 */
4060 return check_barrier_error(info);
4061 }
4062 return 0;
4063}
4064
4065int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4066{
4067 int raid_type;
4068 int min_tolerated = INT_MAX;
4069
4070 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4071 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4072 min_tolerated = min_t(int, min_tolerated,
4073 btrfs_raid_array[BTRFS_RAID_SINGLE].
4074 tolerated_failures);
4075
4076 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4077 if (raid_type == BTRFS_RAID_SINGLE)
4078 continue;
4079 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4080 continue;
4081 min_tolerated = min_t(int, min_tolerated,
4082 btrfs_raid_array[raid_type].
4083 tolerated_failures);
4084 }
4085
4086 if (min_tolerated == INT_MAX) {
4087 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4088 min_tolerated = 0;
4089 }
4090
4091 return min_tolerated;
4092}
4093
4094int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4095{
4096 struct list_head *head;
4097 struct btrfs_device *dev;
4098 struct btrfs_super_block *sb;
4099 struct btrfs_dev_item *dev_item;
4100 int ret;
4101 int do_barriers;
4102 int max_errors;
4103 int total_errors = 0;
4104 u64 flags;
4105
4106 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4107
4108 /*
4109 * max_mirrors == 0 indicates we're from commit_transaction,
4110 * not from fsync where the tree roots in fs_info have not
4111 * been consistent on disk.
4112 */
4113 if (max_mirrors == 0)
4114 backup_super_roots(fs_info);
4115
4116 sb = fs_info->super_for_commit;
4117 dev_item = &sb->dev_item;
4118
4119 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4120 head = &fs_info->fs_devices->devices;
4121 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4122
4123 if (do_barriers) {
4124 ret = barrier_all_devices(fs_info);
4125 if (ret) {
4126 mutex_unlock(
4127 &fs_info->fs_devices->device_list_mutex);
4128 btrfs_handle_fs_error(fs_info, ret,
4129 "errors while submitting device barriers.");
4130 return ret;
4131 }
4132 }
4133
4134 list_for_each_entry(dev, head, dev_list) {
4135 if (!dev->bdev) {
4136 total_errors++;
4137 continue;
4138 }
4139 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4140 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4141 continue;
4142
4143 btrfs_set_stack_device_generation(dev_item, 0);
4144 btrfs_set_stack_device_type(dev_item, dev->type);
4145 btrfs_set_stack_device_id(dev_item, dev->devid);
4146 btrfs_set_stack_device_total_bytes(dev_item,
4147 dev->commit_total_bytes);
4148 btrfs_set_stack_device_bytes_used(dev_item,
4149 dev->commit_bytes_used);
4150 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4151 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4152 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4153 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4154 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4155 BTRFS_FSID_SIZE);
4156
4157 flags = btrfs_super_flags(sb);
4158 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4159
4160 ret = btrfs_validate_write_super(fs_info, sb);
4161 if (ret < 0) {
4162 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4163 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4164 "unexpected superblock corruption detected");
4165 return -EUCLEAN;
4166 }
4167
4168 ret = write_dev_supers(dev, sb, max_mirrors);
4169 if (ret)
4170 total_errors++;
4171 }
4172 if (total_errors > max_errors) {
4173 btrfs_err(fs_info, "%d errors while writing supers",
4174 total_errors);
4175 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4176
4177 /* FUA is masked off if unsupported and can't be the reason */
4178 btrfs_handle_fs_error(fs_info, -EIO,
4179 "%d errors while writing supers",
4180 total_errors);
4181 return -EIO;
4182 }
4183
4184 total_errors = 0;
4185 list_for_each_entry(dev, head, dev_list) {
4186 if (!dev->bdev)
4187 continue;
4188 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4189 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4190 continue;
4191
4192 ret = wait_dev_supers(dev, max_mirrors);
4193 if (ret)
4194 total_errors++;
4195 }
4196 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4197 if (total_errors > max_errors) {
4198 btrfs_handle_fs_error(fs_info, -EIO,
4199 "%d errors while writing supers",
4200 total_errors);
4201 return -EIO;
4202 }
4203 return 0;
4204}
4205
4206/* Drop a fs root from the radix tree and free it. */
4207void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4208 struct btrfs_root *root)
4209{
4210 bool drop_ref = false;
4211
4212 spin_lock(&fs_info->fs_roots_radix_lock);
4213 radix_tree_delete(&fs_info->fs_roots_radix,
4214 (unsigned long)root->root_key.objectid);
4215 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4216 drop_ref = true;
4217 spin_unlock(&fs_info->fs_roots_radix_lock);
4218
4219 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4220 ASSERT(root->log_root == NULL);
4221 if (root->reloc_root) {
4222 btrfs_put_root(root->reloc_root);
4223 root->reloc_root = NULL;
4224 }
4225 }
4226
4227 if (drop_ref)
4228 btrfs_put_root(root);
4229}
4230
4231int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4232{
4233 u64 root_objectid = 0;
4234 struct btrfs_root *gang[8];
4235 int i = 0;
4236 int err = 0;
4237 unsigned int ret = 0;
4238
4239 while (1) {
4240 spin_lock(&fs_info->fs_roots_radix_lock);
4241 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4242 (void **)gang, root_objectid,
4243 ARRAY_SIZE(gang));
4244 if (!ret) {
4245 spin_unlock(&fs_info->fs_roots_radix_lock);
4246 break;
4247 }
4248 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4249
4250 for (i = 0; i < ret; i++) {
4251 /* Avoid to grab roots in dead_roots */
4252 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4253 gang[i] = NULL;
4254 continue;
4255 }
4256 /* grab all the search result for later use */
4257 gang[i] = btrfs_grab_root(gang[i]);
4258 }
4259 spin_unlock(&fs_info->fs_roots_radix_lock);
4260
4261 for (i = 0; i < ret; i++) {
4262 if (!gang[i])
4263 continue;
4264 root_objectid = gang[i]->root_key.objectid;
4265 err = btrfs_orphan_cleanup(gang[i]);
4266 if (err)
4267 break;
4268 btrfs_put_root(gang[i]);
4269 }
4270 root_objectid++;
4271 }
4272
4273 /* release the uncleaned roots due to error */
4274 for (; i < ret; i++) {
4275 if (gang[i])
4276 btrfs_put_root(gang[i]);
4277 }
4278 return err;
4279}
4280
4281int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4282{
4283 struct btrfs_root *root = fs_info->tree_root;
4284 struct btrfs_trans_handle *trans;
4285
4286 mutex_lock(&fs_info->cleaner_mutex);
4287 btrfs_run_delayed_iputs(fs_info);
4288 mutex_unlock(&fs_info->cleaner_mutex);
4289 wake_up_process(fs_info->cleaner_kthread);
4290
4291 /* wait until ongoing cleanup work done */
4292 down_write(&fs_info->cleanup_work_sem);
4293 up_write(&fs_info->cleanup_work_sem);
4294
4295 trans = btrfs_join_transaction(root);
4296 if (IS_ERR(trans))
4297 return PTR_ERR(trans);
4298 return btrfs_commit_transaction(trans);
4299}
4300
4301void __cold close_ctree(struct btrfs_fs_info *fs_info)
4302{
4303 int ret;
4304
4305 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4306 /*
4307 * We don't want the cleaner to start new transactions, add more delayed
4308 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4309 * because that frees the task_struct, and the transaction kthread might
4310 * still try to wake up the cleaner.
4311 */
4312 kthread_park(fs_info->cleaner_kthread);
4313
4314 /* wait for the qgroup rescan worker to stop */
4315 btrfs_qgroup_wait_for_completion(fs_info, false);
4316
4317 /* wait for the uuid_scan task to finish */
4318 down(&fs_info->uuid_tree_rescan_sem);
4319 /* avoid complains from lockdep et al., set sem back to initial state */
4320 up(&fs_info->uuid_tree_rescan_sem);
4321
4322 /* pause restriper - we want to resume on mount */
4323 btrfs_pause_balance(fs_info);
4324
4325 btrfs_dev_replace_suspend_for_unmount(fs_info);
4326
4327 btrfs_scrub_cancel(fs_info);
4328
4329 /* wait for any defraggers to finish */
4330 wait_event(fs_info->transaction_wait,
4331 (atomic_read(&fs_info->defrag_running) == 0));
4332
4333 /* clear out the rbtree of defraggable inodes */
4334 btrfs_cleanup_defrag_inodes(fs_info);
4335
4336 cancel_work_sync(&fs_info->async_reclaim_work);
4337 cancel_work_sync(&fs_info->async_data_reclaim_work);
4338 cancel_work_sync(&fs_info->preempt_reclaim_work);
4339
4340 cancel_work_sync(&fs_info->reclaim_bgs_work);
4341
4342 /* Cancel or finish ongoing discard work */
4343 btrfs_discard_cleanup(fs_info);
4344
4345 if (!sb_rdonly(fs_info->sb)) {
4346 /*
4347 * The cleaner kthread is stopped, so do one final pass over
4348 * unused block groups.
4349 */
4350 btrfs_delete_unused_bgs(fs_info);
4351
4352 /*
4353 * There might be existing delayed inode workers still running
4354 * and holding an empty delayed inode item. We must wait for
4355 * them to complete first because they can create a transaction.
4356 * This happens when someone calls btrfs_balance_delayed_items()
4357 * and then a transaction commit runs the same delayed nodes
4358 * before any delayed worker has done something with the nodes.
4359 * We must wait for any worker here and not at transaction
4360 * commit time since that could cause a deadlock.
4361 * This is a very rare case.
4362 */
4363 btrfs_flush_workqueue(fs_info->delayed_workers);
4364
4365 ret = btrfs_commit_super(fs_info);
4366 if (ret)
4367 btrfs_err(fs_info, "commit super ret %d", ret);
4368 }
4369
4370 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4371 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4372 btrfs_error_commit_super(fs_info);
4373
4374 kthread_stop(fs_info->transaction_kthread);
4375 kthread_stop(fs_info->cleaner_kthread);
4376
4377 ASSERT(list_empty(&fs_info->delayed_iputs));
4378 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4379
4380 if (btrfs_check_quota_leak(fs_info)) {
4381 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4382 btrfs_err(fs_info, "qgroup reserved space leaked");
4383 }
4384
4385 btrfs_free_qgroup_config(fs_info);
4386 ASSERT(list_empty(&fs_info->delalloc_roots));
4387
4388 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4389 btrfs_info(fs_info, "at unmount delalloc count %lld",
4390 percpu_counter_sum(&fs_info->delalloc_bytes));
4391 }
4392
4393 if (percpu_counter_sum(&fs_info->ordered_bytes))
4394 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4395 percpu_counter_sum(&fs_info->ordered_bytes));
4396
4397 btrfs_sysfs_remove_mounted(fs_info);
4398 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4399
4400 btrfs_put_block_group_cache(fs_info);
4401
4402 /*
4403 * we must make sure there is not any read request to
4404 * submit after we stopping all workers.
4405 */
4406 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4407 btrfs_stop_all_workers(fs_info);
4408
4409 /* We shouldn't have any transaction open at this point */
4410 ASSERT(list_empty(&fs_info->trans_list));
4411
4412 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4413 free_root_pointers(fs_info, true);
4414 btrfs_free_fs_roots(fs_info);
4415
4416 /*
4417 * We must free the block groups after dropping the fs_roots as we could
4418 * have had an IO error and have left over tree log blocks that aren't
4419 * cleaned up until the fs roots are freed. This makes the block group
4420 * accounting appear to be wrong because there's pending reserved bytes,
4421 * so make sure we do the block group cleanup afterwards.
4422 */
4423 btrfs_free_block_groups(fs_info);
4424
4425 iput(fs_info->btree_inode);
4426
4427#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4428 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4429 btrfsic_unmount(fs_info->fs_devices);
4430#endif
4431
4432 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4433 btrfs_close_devices(fs_info->fs_devices);
4434}
4435
4436int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4437 int atomic)
4438{
4439 int ret;
4440 struct inode *btree_inode = buf->pages[0]->mapping->host;
4441
4442 ret = extent_buffer_uptodate(buf);
4443 if (!ret)
4444 return ret;
4445
4446 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4447 parent_transid, atomic);
4448 if (ret == -EAGAIN)
4449 return ret;
4450 return !ret;
4451}
4452
4453void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4454{
4455 struct btrfs_fs_info *fs_info = buf->fs_info;
4456 u64 transid = btrfs_header_generation(buf);
4457 int was_dirty;
4458
4459#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4460 /*
4461 * This is a fast path so only do this check if we have sanity tests
4462 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4463 * outside of the sanity tests.
4464 */
4465 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4466 return;
4467#endif
4468 btrfs_assert_tree_locked(buf);
4469 if (transid != fs_info->generation)
4470 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4471 buf->start, transid, fs_info->generation);
4472 was_dirty = set_extent_buffer_dirty(buf);
4473 if (!was_dirty)
4474 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4475 buf->len,
4476 fs_info->dirty_metadata_batch);
4477#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4478 /*
4479 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4480 * but item data not updated.
4481 * So here we should only check item pointers, not item data.
4482 */
4483 if (btrfs_header_level(buf) == 0 &&
4484 btrfs_check_leaf_relaxed(buf)) {
4485 btrfs_print_leaf(buf);
4486 ASSERT(0);
4487 }
4488#endif
4489}
4490
4491static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4492 int flush_delayed)
4493{
4494 /*
4495 * looks as though older kernels can get into trouble with
4496 * this code, they end up stuck in balance_dirty_pages forever
4497 */
4498 int ret;
4499
4500 if (current->flags & PF_MEMALLOC)
4501 return;
4502
4503 if (flush_delayed)
4504 btrfs_balance_delayed_items(fs_info);
4505
4506 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4507 BTRFS_DIRTY_METADATA_THRESH,
4508 fs_info->dirty_metadata_batch);
4509 if (ret > 0) {
4510 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4511 }
4512}
4513
4514void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4515{
4516 __btrfs_btree_balance_dirty(fs_info, 1);
4517}
4518
4519void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4520{
4521 __btrfs_btree_balance_dirty(fs_info, 0);
4522}
4523
4524int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4525 struct btrfs_key *first_key)
4526{
4527 return btree_read_extent_buffer_pages(buf, parent_transid,
4528 level, first_key);
4529}
4530
4531static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4532{
4533 /* cleanup FS via transaction */
4534 btrfs_cleanup_transaction(fs_info);
4535
4536 mutex_lock(&fs_info->cleaner_mutex);
4537 btrfs_run_delayed_iputs(fs_info);
4538 mutex_unlock(&fs_info->cleaner_mutex);
4539
4540 down_write(&fs_info->cleanup_work_sem);
4541 up_write(&fs_info->cleanup_work_sem);
4542}
4543
4544static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4545{
4546 struct btrfs_root *gang[8];
4547 u64 root_objectid = 0;
4548 int ret;
4549
4550 spin_lock(&fs_info->fs_roots_radix_lock);
4551 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4552 (void **)gang, root_objectid,
4553 ARRAY_SIZE(gang))) != 0) {
4554 int i;
4555
4556 for (i = 0; i < ret; i++)
4557 gang[i] = btrfs_grab_root(gang[i]);
4558 spin_unlock(&fs_info->fs_roots_radix_lock);
4559
4560 for (i = 0; i < ret; i++) {
4561 if (!gang[i])
4562 continue;
4563 root_objectid = gang[i]->root_key.objectid;
4564 btrfs_free_log(NULL, gang[i]);
4565 btrfs_put_root(gang[i]);
4566 }
4567 root_objectid++;
4568 spin_lock(&fs_info->fs_roots_radix_lock);
4569 }
4570 spin_unlock(&fs_info->fs_roots_radix_lock);
4571 btrfs_free_log_root_tree(NULL, fs_info);
4572}
4573
4574static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4575{
4576 struct btrfs_ordered_extent *ordered;
4577
4578 spin_lock(&root->ordered_extent_lock);
4579 /*
4580 * This will just short circuit the ordered completion stuff which will
4581 * make sure the ordered extent gets properly cleaned up.
4582 */
4583 list_for_each_entry(ordered, &root->ordered_extents,
4584 root_extent_list)
4585 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4586 spin_unlock(&root->ordered_extent_lock);
4587}
4588
4589static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4590{
4591 struct btrfs_root *root;
4592 struct list_head splice;
4593
4594 INIT_LIST_HEAD(&splice);
4595
4596 spin_lock(&fs_info->ordered_root_lock);
4597 list_splice_init(&fs_info->ordered_roots, &splice);
4598 while (!list_empty(&splice)) {
4599 root = list_first_entry(&splice, struct btrfs_root,
4600 ordered_root);
4601 list_move_tail(&root->ordered_root,
4602 &fs_info->ordered_roots);
4603
4604 spin_unlock(&fs_info->ordered_root_lock);
4605 btrfs_destroy_ordered_extents(root);
4606
4607 cond_resched();
4608 spin_lock(&fs_info->ordered_root_lock);
4609 }
4610 spin_unlock(&fs_info->ordered_root_lock);
4611
4612 /*
4613 * We need this here because if we've been flipped read-only we won't
4614 * get sync() from the umount, so we need to make sure any ordered
4615 * extents that haven't had their dirty pages IO start writeout yet
4616 * actually get run and error out properly.
4617 */
4618 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4619}
4620
4621static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4622 struct btrfs_fs_info *fs_info)
4623{
4624 struct rb_node *node;
4625 struct btrfs_delayed_ref_root *delayed_refs;
4626 struct btrfs_delayed_ref_node *ref;
4627 int ret = 0;
4628
4629 delayed_refs = &trans->delayed_refs;
4630
4631 spin_lock(&delayed_refs->lock);
4632 if (atomic_read(&delayed_refs->num_entries) == 0) {
4633 spin_unlock(&delayed_refs->lock);
4634 btrfs_debug(fs_info, "delayed_refs has NO entry");
4635 return ret;
4636 }
4637
4638 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4639 struct btrfs_delayed_ref_head *head;
4640 struct rb_node *n;
4641 bool pin_bytes = false;
4642
4643 head = rb_entry(node, struct btrfs_delayed_ref_head,
4644 href_node);
4645 if (btrfs_delayed_ref_lock(delayed_refs, head))
4646 continue;
4647
4648 spin_lock(&head->lock);
4649 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4650 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4651 ref_node);
4652 ref->in_tree = 0;
4653 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4654 RB_CLEAR_NODE(&ref->ref_node);
4655 if (!list_empty(&ref->add_list))
4656 list_del(&ref->add_list);
4657 atomic_dec(&delayed_refs->num_entries);
4658 btrfs_put_delayed_ref(ref);
4659 }
4660 if (head->must_insert_reserved)
4661 pin_bytes = true;
4662 btrfs_free_delayed_extent_op(head->extent_op);
4663 btrfs_delete_ref_head(delayed_refs, head);
4664 spin_unlock(&head->lock);
4665 spin_unlock(&delayed_refs->lock);
4666 mutex_unlock(&head->mutex);
4667
4668 if (pin_bytes) {
4669 struct btrfs_block_group *cache;
4670
4671 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4672 BUG_ON(!cache);
4673
4674 spin_lock(&cache->space_info->lock);
4675 spin_lock(&cache->lock);
4676 cache->pinned += head->num_bytes;
4677 btrfs_space_info_update_bytes_pinned(fs_info,
4678 cache->space_info, head->num_bytes);
4679 cache->reserved -= head->num_bytes;
4680 cache->space_info->bytes_reserved -= head->num_bytes;
4681 spin_unlock(&cache->lock);
4682 spin_unlock(&cache->space_info->lock);
4683
4684 btrfs_put_block_group(cache);
4685
4686 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4687 head->bytenr + head->num_bytes - 1);
4688 }
4689 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4690 btrfs_put_delayed_ref_head(head);
4691 cond_resched();
4692 spin_lock(&delayed_refs->lock);
4693 }
4694 btrfs_qgroup_destroy_extent_records(trans);
4695
4696 spin_unlock(&delayed_refs->lock);
4697
4698 return ret;
4699}
4700
4701static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4702{
4703 struct btrfs_inode *btrfs_inode;
4704 struct list_head splice;
4705
4706 INIT_LIST_HEAD(&splice);
4707
4708 spin_lock(&root->delalloc_lock);
4709 list_splice_init(&root->delalloc_inodes, &splice);
4710
4711 while (!list_empty(&splice)) {
4712 struct inode *inode = NULL;
4713 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4714 delalloc_inodes);
4715 __btrfs_del_delalloc_inode(root, btrfs_inode);
4716 spin_unlock(&root->delalloc_lock);
4717
4718 /*
4719 * Make sure we get a live inode and that it'll not disappear
4720 * meanwhile.
4721 */
4722 inode = igrab(&btrfs_inode->vfs_inode);
4723 if (inode) {
4724 invalidate_inode_pages2(inode->i_mapping);
4725 iput(inode);
4726 }
4727 spin_lock(&root->delalloc_lock);
4728 }
4729 spin_unlock(&root->delalloc_lock);
4730}
4731
4732static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4733{
4734 struct btrfs_root *root;
4735 struct list_head splice;
4736
4737 INIT_LIST_HEAD(&splice);
4738
4739 spin_lock(&fs_info->delalloc_root_lock);
4740 list_splice_init(&fs_info->delalloc_roots, &splice);
4741 while (!list_empty(&splice)) {
4742 root = list_first_entry(&splice, struct btrfs_root,
4743 delalloc_root);
4744 root = btrfs_grab_root(root);
4745 BUG_ON(!root);
4746 spin_unlock(&fs_info->delalloc_root_lock);
4747
4748 btrfs_destroy_delalloc_inodes(root);
4749 btrfs_put_root(root);
4750
4751 spin_lock(&fs_info->delalloc_root_lock);
4752 }
4753 spin_unlock(&fs_info->delalloc_root_lock);
4754}
4755
4756static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4757 struct extent_io_tree *dirty_pages,
4758 int mark)
4759{
4760 int ret;
4761 struct extent_buffer *eb;
4762 u64 start = 0;
4763 u64 end;
4764
4765 while (1) {
4766 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4767 mark, NULL);
4768 if (ret)
4769 break;
4770
4771 clear_extent_bits(dirty_pages, start, end, mark);
4772 while (start <= end) {
4773 eb = find_extent_buffer(fs_info, start);
4774 start += fs_info->nodesize;
4775 if (!eb)
4776 continue;
4777 wait_on_extent_buffer_writeback(eb);
4778
4779 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4780 &eb->bflags))
4781 clear_extent_buffer_dirty(eb);
4782 free_extent_buffer_stale(eb);
4783 }
4784 }
4785
4786 return ret;
4787}
4788
4789static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4790 struct extent_io_tree *unpin)
4791{
4792 u64 start;
4793 u64 end;
4794 int ret;
4795
4796 while (1) {
4797 struct extent_state *cached_state = NULL;
4798
4799 /*
4800 * The btrfs_finish_extent_commit() may get the same range as
4801 * ours between find_first_extent_bit and clear_extent_dirty.
4802 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4803 * the same extent range.
4804 */
4805 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4806 ret = find_first_extent_bit(unpin, 0, &start, &end,
4807 EXTENT_DIRTY, &cached_state);
4808 if (ret) {
4809 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4810 break;
4811 }
4812
4813 clear_extent_dirty(unpin, start, end, &cached_state);
4814 free_extent_state(cached_state);
4815 btrfs_error_unpin_extent_range(fs_info, start, end);
4816 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4817 cond_resched();
4818 }
4819
4820 return 0;
4821}
4822
4823static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4824{
4825 struct inode *inode;
4826
4827 inode = cache->io_ctl.inode;
4828 if (inode) {
4829 invalidate_inode_pages2(inode->i_mapping);
4830 BTRFS_I(inode)->generation = 0;
4831 cache->io_ctl.inode = NULL;
4832 iput(inode);
4833 }
4834 ASSERT(cache->io_ctl.pages == NULL);
4835 btrfs_put_block_group(cache);
4836}
4837
4838void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4839 struct btrfs_fs_info *fs_info)
4840{
4841 struct btrfs_block_group *cache;
4842
4843 spin_lock(&cur_trans->dirty_bgs_lock);
4844 while (!list_empty(&cur_trans->dirty_bgs)) {
4845 cache = list_first_entry(&cur_trans->dirty_bgs,
4846 struct btrfs_block_group,
4847 dirty_list);
4848
4849 if (!list_empty(&cache->io_list)) {
4850 spin_unlock(&cur_trans->dirty_bgs_lock);
4851 list_del_init(&cache->io_list);
4852 btrfs_cleanup_bg_io(cache);
4853 spin_lock(&cur_trans->dirty_bgs_lock);
4854 }
4855
4856 list_del_init(&cache->dirty_list);
4857 spin_lock(&cache->lock);
4858 cache->disk_cache_state = BTRFS_DC_ERROR;
4859 spin_unlock(&cache->lock);
4860
4861 spin_unlock(&cur_trans->dirty_bgs_lock);
4862 btrfs_put_block_group(cache);
4863 btrfs_delayed_refs_rsv_release(fs_info, 1);
4864 spin_lock(&cur_trans->dirty_bgs_lock);
4865 }
4866 spin_unlock(&cur_trans->dirty_bgs_lock);
4867
4868 /*
4869 * Refer to the definition of io_bgs member for details why it's safe
4870 * to use it without any locking
4871 */
4872 while (!list_empty(&cur_trans->io_bgs)) {
4873 cache = list_first_entry(&cur_trans->io_bgs,
4874 struct btrfs_block_group,
4875 io_list);
4876
4877 list_del_init(&cache->io_list);
4878 spin_lock(&cache->lock);
4879 cache->disk_cache_state = BTRFS_DC_ERROR;
4880 spin_unlock(&cache->lock);
4881 btrfs_cleanup_bg_io(cache);
4882 }
4883}
4884
4885void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4886 struct btrfs_fs_info *fs_info)
4887{
4888 struct btrfs_device *dev, *tmp;
4889
4890 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4891 ASSERT(list_empty(&cur_trans->dirty_bgs));
4892 ASSERT(list_empty(&cur_trans->io_bgs));
4893
4894 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4895 post_commit_list) {
4896 list_del_init(&dev->post_commit_list);
4897 }
4898
4899 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4900
4901 cur_trans->state = TRANS_STATE_COMMIT_START;
4902 wake_up(&fs_info->transaction_blocked_wait);
4903
4904 cur_trans->state = TRANS_STATE_UNBLOCKED;
4905 wake_up(&fs_info->transaction_wait);
4906
4907 btrfs_destroy_delayed_inodes(fs_info);
4908
4909 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4910 EXTENT_DIRTY);
4911 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4912
4913 btrfs_free_redirty_list(cur_trans);
4914
4915 cur_trans->state =TRANS_STATE_COMPLETED;
4916 wake_up(&cur_trans->commit_wait);
4917}
4918
4919static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4920{
4921 struct btrfs_transaction *t;
4922
4923 mutex_lock(&fs_info->transaction_kthread_mutex);
4924
4925 spin_lock(&fs_info->trans_lock);
4926 while (!list_empty(&fs_info->trans_list)) {
4927 t = list_first_entry(&fs_info->trans_list,
4928 struct btrfs_transaction, list);
4929 if (t->state >= TRANS_STATE_COMMIT_START) {
4930 refcount_inc(&t->use_count);
4931 spin_unlock(&fs_info->trans_lock);
4932 btrfs_wait_for_commit(fs_info, t->transid);
4933 btrfs_put_transaction(t);
4934 spin_lock(&fs_info->trans_lock);
4935 continue;
4936 }
4937 if (t == fs_info->running_transaction) {
4938 t->state = TRANS_STATE_COMMIT_DOING;
4939 spin_unlock(&fs_info->trans_lock);
4940 /*
4941 * We wait for 0 num_writers since we don't hold a trans
4942 * handle open currently for this transaction.
4943 */
4944 wait_event(t->writer_wait,
4945 atomic_read(&t->num_writers) == 0);
4946 } else {
4947 spin_unlock(&fs_info->trans_lock);
4948 }
4949 btrfs_cleanup_one_transaction(t, fs_info);
4950
4951 spin_lock(&fs_info->trans_lock);
4952 if (t == fs_info->running_transaction)
4953 fs_info->running_transaction = NULL;
4954 list_del_init(&t->list);
4955 spin_unlock(&fs_info->trans_lock);
4956
4957 btrfs_put_transaction(t);
4958 trace_btrfs_transaction_commit(fs_info->tree_root);
4959 spin_lock(&fs_info->trans_lock);
4960 }
4961 spin_unlock(&fs_info->trans_lock);
4962 btrfs_destroy_all_ordered_extents(fs_info);
4963 btrfs_destroy_delayed_inodes(fs_info);
4964 btrfs_assert_delayed_root_empty(fs_info);
4965 btrfs_destroy_all_delalloc_inodes(fs_info);
4966 btrfs_drop_all_logs(fs_info);
4967 mutex_unlock(&fs_info->transaction_kthread_mutex);
4968
4969 return 0;
4970}
4971
4972int btrfs_init_root_free_objectid(struct btrfs_root *root)
4973{
4974 struct btrfs_path *path;
4975 int ret;
4976 struct extent_buffer *l;
4977 struct btrfs_key search_key;
4978 struct btrfs_key found_key;
4979 int slot;
4980
4981 path = btrfs_alloc_path();
4982 if (!path)
4983 return -ENOMEM;
4984
4985 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4986 search_key.type = -1;
4987 search_key.offset = (u64)-1;
4988 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4989 if (ret < 0)
4990 goto error;
4991 BUG_ON(ret == 0); /* Corruption */
4992 if (path->slots[0] > 0) {
4993 slot = path->slots[0] - 1;
4994 l = path->nodes[0];
4995 btrfs_item_key_to_cpu(l, &found_key, slot);
4996 root->free_objectid = max_t(u64, found_key.objectid + 1,
4997 BTRFS_FIRST_FREE_OBJECTID);
4998 } else {
4999 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5000 }
5001 ret = 0;
5002error:
5003 btrfs_free_path(path);
5004 return ret;
5005}
5006
5007int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5008{
5009 int ret;
5010 mutex_lock(&root->objectid_mutex);
5011
5012 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5013 btrfs_warn(root->fs_info,
5014 "the objectid of root %llu reaches its highest value",
5015 root->root_key.objectid);
5016 ret = -ENOSPC;
5017 goto out;
5018 }
5019
5020 *objectid = root->free_objectid++;
5021 ret = 0;
5022out:
5023 mutex_unlock(&root->objectid_mutex);
5024 return ret;
5025}