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