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/scatterlist.h>
9#include <linux/swap.h>
10#include <linux/radix-tree.h>
11#include <linux/writeback.h>
12#include <linux/buffer_head.h>
13#include <linux/workqueue.h>
14#include <linux/kthread.h>
15#include <linux/slab.h>
16#include <linux/migrate.h>
17#include <linux/ratelimit.h>
18#include <linux/uuid.h>
19#include <linux/semaphore.h>
20#include <linux/error-injection.h>
21#include <linux/crc32c.h>
22#include <asm/unaligned.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
44#ifdef CONFIG_X86
45#include <asm/cpufeature.h>
46#endif
47
48#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
49 BTRFS_HEADER_FLAG_RELOC |\
50 BTRFS_SUPER_FLAG_ERROR |\
51 BTRFS_SUPER_FLAG_SEEDING |\
52 BTRFS_SUPER_FLAG_METADUMP |\
53 BTRFS_SUPER_FLAG_METADUMP_V2)
54
55static const struct extent_io_ops btree_extent_io_ops;
56static void end_workqueue_fn(struct btrfs_work *work);
57static void free_fs_root(struct btrfs_root *root);
58static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
59static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
60static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
61 struct btrfs_fs_info *fs_info);
62static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
63static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
64 struct extent_io_tree *dirty_pages,
65 int mark);
66static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
67 struct extent_io_tree *pinned_extents);
68static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
69static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
70
71/*
72 * btrfs_end_io_wq structs are used to do processing in task context when an IO
73 * is complete. This is used during reads to verify checksums, and it is used
74 * by writes to insert metadata for new file extents after IO is complete.
75 */
76struct btrfs_end_io_wq {
77 struct bio *bio;
78 bio_end_io_t *end_io;
79 void *private;
80 struct btrfs_fs_info *info;
81 blk_status_t status;
82 enum btrfs_wq_endio_type metadata;
83 struct btrfs_work work;
84};
85
86static struct kmem_cache *btrfs_end_io_wq_cache;
87
88int __init btrfs_end_io_wq_init(void)
89{
90 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
91 sizeof(struct btrfs_end_io_wq),
92 0,
93 SLAB_MEM_SPREAD,
94 NULL);
95 if (!btrfs_end_io_wq_cache)
96 return -ENOMEM;
97 return 0;
98}
99
100void __cold btrfs_end_io_wq_exit(void)
101{
102 kmem_cache_destroy(btrfs_end_io_wq_cache);
103}
104
105/*
106 * async submit bios are used to offload expensive checksumming
107 * onto the worker threads. They checksum file and metadata bios
108 * just before they are sent down the IO stack.
109 */
110struct async_submit_bio {
111 void *private_data;
112 struct btrfs_fs_info *fs_info;
113 struct bio *bio;
114 extent_submit_bio_start_t *submit_bio_start;
115 extent_submit_bio_done_t *submit_bio_done;
116 int mirror_num;
117 unsigned long bio_flags;
118 /*
119 * bio_offset is optional, can be used if the pages in the bio
120 * can't tell us where in the file the bio should go
121 */
122 u64 bio_offset;
123 struct btrfs_work work;
124 blk_status_t status;
125};
126
127/*
128 * Lockdep class keys for extent_buffer->lock's in this root. For a given
129 * eb, the lockdep key is determined by the btrfs_root it belongs to and
130 * the level the eb occupies in the tree.
131 *
132 * Different roots are used for different purposes and may nest inside each
133 * other and they require separate keysets. As lockdep keys should be
134 * static, assign keysets according to the purpose of the root as indicated
135 * by btrfs_root->objectid. This ensures that all special purpose roots
136 * have separate keysets.
137 *
138 * Lock-nesting across peer nodes is always done with the immediate parent
139 * node locked thus preventing deadlock. As lockdep doesn't know this, use
140 * subclass to avoid triggering lockdep warning in such cases.
141 *
142 * The key is set by the readpage_end_io_hook after the buffer has passed
143 * csum validation but before the pages are unlocked. It is also set by
144 * btrfs_init_new_buffer on freshly allocated blocks.
145 *
146 * We also add a check to make sure the highest level of the tree is the
147 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
148 * needs update as well.
149 */
150#ifdef CONFIG_DEBUG_LOCK_ALLOC
151# if BTRFS_MAX_LEVEL != 8
152# error
153# endif
154
155static struct btrfs_lockdep_keyset {
156 u64 id; /* root objectid */
157 const char *name_stem; /* lock name stem */
158 char names[BTRFS_MAX_LEVEL + 1][20];
159 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
160} btrfs_lockdep_keysets[] = {
161 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
162 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
163 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
164 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
165 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
166 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
167 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
168 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
169 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
170 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
171 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
172 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
173 { .id = 0, .name_stem = "tree" },
174};
175
176void __init btrfs_init_lockdep(void)
177{
178 int i, j;
179
180 /* initialize lockdep class names */
181 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
182 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
183
184 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
185 snprintf(ks->names[j], sizeof(ks->names[j]),
186 "btrfs-%s-%02d", ks->name_stem, j);
187 }
188}
189
190void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
191 int level)
192{
193 struct btrfs_lockdep_keyset *ks;
194
195 BUG_ON(level >= ARRAY_SIZE(ks->keys));
196
197 /* find the matching keyset, id 0 is the default entry */
198 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
199 if (ks->id == objectid)
200 break;
201
202 lockdep_set_class_and_name(&eb->lock,
203 &ks->keys[level], ks->names[level]);
204}
205
206#endif
207
208/*
209 * extents on the btree inode are pretty simple, there's one extent
210 * that covers the entire device
211 */
212struct extent_map *btree_get_extent(struct btrfs_inode *inode,
213 struct page *page, size_t pg_offset, u64 start, u64 len,
214 int create)
215{
216 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
217 struct extent_map_tree *em_tree = &inode->extent_tree;
218 struct extent_map *em;
219 int ret;
220
221 read_lock(&em_tree->lock);
222 em = lookup_extent_mapping(em_tree, start, len);
223 if (em) {
224 em->bdev = fs_info->fs_devices->latest_bdev;
225 read_unlock(&em_tree->lock);
226 goto out;
227 }
228 read_unlock(&em_tree->lock);
229
230 em = alloc_extent_map();
231 if (!em) {
232 em = ERR_PTR(-ENOMEM);
233 goto out;
234 }
235 em->start = 0;
236 em->len = (u64)-1;
237 em->block_len = (u64)-1;
238 em->block_start = 0;
239 em->bdev = fs_info->fs_devices->latest_bdev;
240
241 write_lock(&em_tree->lock);
242 ret = add_extent_mapping(em_tree, em, 0);
243 if (ret == -EEXIST) {
244 free_extent_map(em);
245 em = lookup_extent_mapping(em_tree, start, len);
246 if (!em)
247 em = ERR_PTR(-EIO);
248 } else if (ret) {
249 free_extent_map(em);
250 em = ERR_PTR(ret);
251 }
252 write_unlock(&em_tree->lock);
253
254out:
255 return em;
256}
257
258u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
259{
260 return crc32c(seed, data, len);
261}
262
263void btrfs_csum_final(u32 crc, u8 *result)
264{
265 put_unaligned_le32(~crc, result);
266}
267
268/*
269 * compute the csum for a btree block, and either verify it or write it
270 * into the csum field of the block.
271 */
272static int csum_tree_block(struct btrfs_fs_info *fs_info,
273 struct extent_buffer *buf,
274 int verify)
275{
276 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
277 char result[BTRFS_CSUM_SIZE];
278 unsigned long len;
279 unsigned long cur_len;
280 unsigned long offset = BTRFS_CSUM_SIZE;
281 char *kaddr;
282 unsigned long map_start;
283 unsigned long map_len;
284 int err;
285 u32 crc = ~(u32)0;
286
287 len = buf->len - offset;
288 while (len > 0) {
289 err = map_private_extent_buffer(buf, offset, 32,
290 &kaddr, &map_start, &map_len);
291 if (err)
292 return err;
293 cur_len = min(len, map_len - (offset - map_start));
294 crc = btrfs_csum_data(kaddr + offset - map_start,
295 crc, cur_len);
296 len -= cur_len;
297 offset += cur_len;
298 }
299 memset(result, 0, BTRFS_CSUM_SIZE);
300
301 btrfs_csum_final(crc, result);
302
303 if (verify) {
304 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
305 u32 val;
306 u32 found = 0;
307 memcpy(&found, result, csum_size);
308
309 read_extent_buffer(buf, &val, 0, csum_size);
310 btrfs_warn_rl(fs_info,
311 "%s checksum verify failed on %llu wanted %X found %X level %d",
312 fs_info->sb->s_id, buf->start,
313 val, found, btrfs_header_level(buf));
314 return -EUCLEAN;
315 }
316 } else {
317 write_extent_buffer(buf, result, 0, csum_size);
318 }
319
320 return 0;
321}
322
323/*
324 * we can't consider a given block up to date unless the transid of the
325 * block matches the transid in the parent node's pointer. This is how we
326 * detect blocks that either didn't get written at all or got written
327 * in the wrong place.
328 */
329static int verify_parent_transid(struct extent_io_tree *io_tree,
330 struct extent_buffer *eb, u64 parent_transid,
331 int atomic)
332{
333 struct extent_state *cached_state = NULL;
334 int ret;
335 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
336
337 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
338 return 0;
339
340 if (atomic)
341 return -EAGAIN;
342
343 if (need_lock) {
344 btrfs_tree_read_lock(eb);
345 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
346 }
347
348 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
349 &cached_state);
350 if (extent_buffer_uptodate(eb) &&
351 btrfs_header_generation(eb) == parent_transid) {
352 ret = 0;
353 goto out;
354 }
355 btrfs_err_rl(eb->fs_info,
356 "parent transid verify failed on %llu wanted %llu found %llu",
357 eb->start,
358 parent_transid, btrfs_header_generation(eb));
359 ret = 1;
360
361 /*
362 * Things reading via commit roots that don't have normal protection,
363 * like send, can have a really old block in cache that may point at a
364 * block that has been freed and re-allocated. So don't clear uptodate
365 * if we find an eb that is under IO (dirty/writeback) because we could
366 * end up reading in the stale data and then writing it back out and
367 * making everybody very sad.
368 */
369 if (!extent_buffer_under_io(eb))
370 clear_extent_buffer_uptodate(eb);
371out:
372 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
373 &cached_state);
374 if (need_lock)
375 btrfs_tree_read_unlock_blocking(eb);
376 return ret;
377}
378
379/*
380 * Return 0 if the superblock checksum type matches the checksum value of that
381 * algorithm. Pass the raw disk superblock data.
382 */
383static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
384 char *raw_disk_sb)
385{
386 struct btrfs_super_block *disk_sb =
387 (struct btrfs_super_block *)raw_disk_sb;
388 u16 csum_type = btrfs_super_csum_type(disk_sb);
389 int ret = 0;
390
391 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
392 u32 crc = ~(u32)0;
393 char result[sizeof(crc)];
394
395 /*
396 * The super_block structure does not span the whole
397 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
398 * is filled with zeros and is included in the checksum.
399 */
400 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
401 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
402 btrfs_csum_final(crc, result);
403
404 if (memcmp(raw_disk_sb, result, sizeof(result)))
405 ret = 1;
406 }
407
408 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
409 btrfs_err(fs_info, "unsupported checksum algorithm %u",
410 csum_type);
411 ret = 1;
412 }
413
414 return ret;
415}
416
417static int verify_level_key(struct btrfs_fs_info *fs_info,
418 struct extent_buffer *eb, int level,
419 struct btrfs_key *first_key)
420{
421 int found_level;
422 struct btrfs_key found_key;
423 int ret;
424
425 found_level = btrfs_header_level(eb);
426 if (found_level != level) {
427#ifdef CONFIG_BTRFS_DEBUG
428 WARN_ON(1);
429 btrfs_err(fs_info,
430"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
431 eb->start, level, found_level);
432#endif
433 return -EIO;
434 }
435
436 if (!first_key)
437 return 0;
438
439 /*
440 * For live tree block (new tree blocks in current transaction),
441 * we need proper lock context to avoid race, which is impossible here.
442 * So we only checks tree blocks which is read from disk, whose
443 * generation <= fs_info->last_trans_committed.
444 */
445 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
446 return 0;
447 if (found_level)
448 btrfs_node_key_to_cpu(eb, &found_key, 0);
449 else
450 btrfs_item_key_to_cpu(eb, &found_key, 0);
451 ret = btrfs_comp_cpu_keys(first_key, &found_key);
452
453#ifdef CONFIG_BTRFS_DEBUG
454 if (ret) {
455 WARN_ON(1);
456 btrfs_err(fs_info,
457"tree first key mismatch detected, bytenr=%llu key expected=(%llu, %u, %llu) has=(%llu, %u, %llu)",
458 eb->start, first_key->objectid, first_key->type,
459 first_key->offset, found_key.objectid,
460 found_key.type, found_key.offset);
461 }
462#endif
463 return ret;
464}
465
466/*
467 * helper to read a given tree block, doing retries as required when
468 * the checksums don't match and we have alternate mirrors to try.
469 *
470 * @parent_transid: expected transid, skip check if 0
471 * @level: expected level, mandatory check
472 * @first_key: expected key of first slot, skip check if NULL
473 */
474static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
475 struct extent_buffer *eb,
476 u64 parent_transid, int level,
477 struct btrfs_key *first_key)
478{
479 struct extent_io_tree *io_tree;
480 int failed = 0;
481 int ret;
482 int num_copies = 0;
483 int mirror_num = 0;
484 int failed_mirror = 0;
485
486 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
487 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
488 while (1) {
489 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
490 mirror_num);
491 if (!ret) {
492 if (verify_parent_transid(io_tree, eb,
493 parent_transid, 0))
494 ret = -EIO;
495 else if (verify_level_key(fs_info, eb, level,
496 first_key))
497 ret = -EUCLEAN;
498 else
499 break;
500 }
501
502 /*
503 * This buffer's crc is fine, but its contents are corrupted, so
504 * there is no reason to read the other copies, they won't be
505 * any less wrong.
506 */
507 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
508 ret == -EUCLEAN)
509 break;
510
511 num_copies = btrfs_num_copies(fs_info,
512 eb->start, eb->len);
513 if (num_copies == 1)
514 break;
515
516 if (!failed_mirror) {
517 failed = 1;
518 failed_mirror = eb->read_mirror;
519 }
520
521 mirror_num++;
522 if (mirror_num == failed_mirror)
523 mirror_num++;
524
525 if (mirror_num > num_copies)
526 break;
527 }
528
529 if (failed && !ret && failed_mirror)
530 repair_eb_io_failure(fs_info, eb, failed_mirror);
531
532 return ret;
533}
534
535/*
536 * checksum a dirty tree block before IO. This has extra checks to make sure
537 * we only fill in the checksum field in the first page of a multi-page block
538 */
539
540static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
541{
542 u64 start = page_offset(page);
543 u64 found_start;
544 struct extent_buffer *eb;
545
546 eb = (struct extent_buffer *)page->private;
547 if (page != eb->pages[0])
548 return 0;
549
550 found_start = btrfs_header_bytenr(eb);
551 /*
552 * Please do not consolidate these warnings into a single if.
553 * It is useful to know what went wrong.
554 */
555 if (WARN_ON(found_start != start))
556 return -EUCLEAN;
557 if (WARN_ON(!PageUptodate(page)))
558 return -EUCLEAN;
559
560 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
561 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
562
563 return csum_tree_block(fs_info, eb, 0);
564}
565
566static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
567 struct extent_buffer *eb)
568{
569 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
570 u8 fsid[BTRFS_FSID_SIZE];
571 int ret = 1;
572
573 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
574 while (fs_devices) {
575 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
576 ret = 0;
577 break;
578 }
579 fs_devices = fs_devices->seed;
580 }
581 return ret;
582}
583
584static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
585 u64 phy_offset, struct page *page,
586 u64 start, u64 end, int mirror)
587{
588 u64 found_start;
589 int found_level;
590 struct extent_buffer *eb;
591 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
592 struct btrfs_fs_info *fs_info = root->fs_info;
593 int ret = 0;
594 int reads_done;
595
596 if (!page->private)
597 goto out;
598
599 eb = (struct extent_buffer *)page->private;
600
601 /* the pending IO might have been the only thing that kept this buffer
602 * in memory. Make sure we have a ref for all this other checks
603 */
604 extent_buffer_get(eb);
605
606 reads_done = atomic_dec_and_test(&eb->io_pages);
607 if (!reads_done)
608 goto err;
609
610 eb->read_mirror = mirror;
611 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
612 ret = -EIO;
613 goto err;
614 }
615
616 found_start = btrfs_header_bytenr(eb);
617 if (found_start != eb->start) {
618 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
619 found_start, eb->start);
620 ret = -EIO;
621 goto err;
622 }
623 if (check_tree_block_fsid(fs_info, eb)) {
624 btrfs_err_rl(fs_info, "bad fsid on block %llu",
625 eb->start);
626 ret = -EIO;
627 goto err;
628 }
629 found_level = btrfs_header_level(eb);
630 if (found_level >= BTRFS_MAX_LEVEL) {
631 btrfs_err(fs_info, "bad tree block level %d",
632 (int)btrfs_header_level(eb));
633 ret = -EIO;
634 goto err;
635 }
636
637 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
638 eb, found_level);
639
640 ret = csum_tree_block(fs_info, eb, 1);
641 if (ret)
642 goto err;
643
644 /*
645 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 * that we don't try and read the other copies of this block, just
647 * return -EIO.
648 */
649 if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
650 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
651 ret = -EIO;
652 }
653
654 if (found_level > 0 && btrfs_check_node(fs_info, eb))
655 ret = -EIO;
656
657 if (!ret)
658 set_extent_buffer_uptodate(eb);
659err:
660 if (reads_done &&
661 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 btree_readahead_hook(eb, ret);
663
664 if (ret) {
665 /*
666 * our io error hook is going to dec the io pages
667 * again, we have to make sure it has something
668 * to decrement
669 */
670 atomic_inc(&eb->io_pages);
671 clear_extent_buffer_uptodate(eb);
672 }
673 free_extent_buffer(eb);
674out:
675 return ret;
676}
677
678static int btree_io_failed_hook(struct page *page, int failed_mirror)
679{
680 struct extent_buffer *eb;
681
682 eb = (struct extent_buffer *)page->private;
683 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
684 eb->read_mirror = failed_mirror;
685 atomic_dec(&eb->io_pages);
686 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
687 btree_readahead_hook(eb, -EIO);
688 return -EIO; /* we fixed nothing */
689}
690
691static void end_workqueue_bio(struct bio *bio)
692{
693 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
694 struct btrfs_fs_info *fs_info;
695 struct btrfs_workqueue *wq;
696 btrfs_work_func_t func;
697
698 fs_info = end_io_wq->info;
699 end_io_wq->status = bio->bi_status;
700
701 if (bio_op(bio) == REQ_OP_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
703 wq = fs_info->endio_meta_write_workers;
704 func = btrfs_endio_meta_write_helper;
705 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
706 wq = fs_info->endio_freespace_worker;
707 func = btrfs_freespace_write_helper;
708 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
709 wq = fs_info->endio_raid56_workers;
710 func = btrfs_endio_raid56_helper;
711 } else {
712 wq = fs_info->endio_write_workers;
713 func = btrfs_endio_write_helper;
714 }
715 } else {
716 if (unlikely(end_io_wq->metadata ==
717 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
718 wq = fs_info->endio_repair_workers;
719 func = btrfs_endio_repair_helper;
720 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
721 wq = fs_info->endio_raid56_workers;
722 func = btrfs_endio_raid56_helper;
723 } else if (end_io_wq->metadata) {
724 wq = fs_info->endio_meta_workers;
725 func = btrfs_endio_meta_helper;
726 } else {
727 wq = fs_info->endio_workers;
728 func = btrfs_endio_helper;
729 }
730 }
731
732 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
733 btrfs_queue_work(wq, &end_io_wq->work);
734}
735
736blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
737 enum btrfs_wq_endio_type metadata)
738{
739 struct btrfs_end_io_wq *end_io_wq;
740
741 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
742 if (!end_io_wq)
743 return BLK_STS_RESOURCE;
744
745 end_io_wq->private = bio->bi_private;
746 end_io_wq->end_io = bio->bi_end_io;
747 end_io_wq->info = info;
748 end_io_wq->status = 0;
749 end_io_wq->bio = bio;
750 end_io_wq->metadata = metadata;
751
752 bio->bi_private = end_io_wq;
753 bio->bi_end_io = end_workqueue_bio;
754 return 0;
755}
756
757static void run_one_async_start(struct btrfs_work *work)
758{
759 struct async_submit_bio *async;
760 blk_status_t ret;
761
762 async = container_of(work, struct async_submit_bio, work);
763 ret = async->submit_bio_start(async->private_data, async->bio,
764 async->bio_offset);
765 if (ret)
766 async->status = ret;
767}
768
769static void run_one_async_done(struct btrfs_work *work)
770{
771 struct async_submit_bio *async;
772
773 async = container_of(work, struct async_submit_bio, work);
774
775 /* If an error occurred we just want to clean up the bio and move on */
776 if (async->status) {
777 async->bio->bi_status = async->status;
778 bio_endio(async->bio);
779 return;
780 }
781
782 async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
783}
784
785static void run_one_async_free(struct btrfs_work *work)
786{
787 struct async_submit_bio *async;
788
789 async = container_of(work, struct async_submit_bio, work);
790 kfree(async);
791}
792
793blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
794 int mirror_num, unsigned long bio_flags,
795 u64 bio_offset, void *private_data,
796 extent_submit_bio_start_t *submit_bio_start,
797 extent_submit_bio_done_t *submit_bio_done)
798{
799 struct async_submit_bio *async;
800
801 async = kmalloc(sizeof(*async), GFP_NOFS);
802 if (!async)
803 return BLK_STS_RESOURCE;
804
805 async->private_data = private_data;
806 async->fs_info = fs_info;
807 async->bio = bio;
808 async->mirror_num = mirror_num;
809 async->submit_bio_start = submit_bio_start;
810 async->submit_bio_done = submit_bio_done;
811
812 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
813 run_one_async_done, run_one_async_free);
814
815 async->bio_flags = bio_flags;
816 async->bio_offset = bio_offset;
817
818 async->status = 0;
819
820 if (op_is_sync(bio->bi_opf))
821 btrfs_set_work_high_priority(&async->work);
822
823 btrfs_queue_work(fs_info->workers, &async->work);
824 return 0;
825}
826
827static blk_status_t btree_csum_one_bio(struct bio *bio)
828{
829 struct bio_vec *bvec;
830 struct btrfs_root *root;
831 int i, ret = 0;
832
833 ASSERT(!bio_flagged(bio, BIO_CLONED));
834 bio_for_each_segment_all(bvec, bio, i) {
835 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
836 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
837 if (ret)
838 break;
839 }
840
841 return errno_to_blk_status(ret);
842}
843
844static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
845 u64 bio_offset)
846{
847 /*
848 * when we're called for a write, we're already in the async
849 * submission context. Just jump into btrfs_map_bio
850 */
851 return btree_csum_one_bio(bio);
852}
853
854static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
855 int mirror_num)
856{
857 struct inode *inode = private_data;
858 blk_status_t ret;
859
860 /*
861 * when we're called for a write, we're already in the async
862 * submission context. Just jump into btrfs_map_bio
863 */
864 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
865 if (ret) {
866 bio->bi_status = ret;
867 bio_endio(bio);
868 }
869 return ret;
870}
871
872static int check_async_write(struct btrfs_inode *bi)
873{
874 if (atomic_read(&bi->sync_writers))
875 return 0;
876#ifdef CONFIG_X86
877 if (static_cpu_has(X86_FEATURE_XMM4_2))
878 return 0;
879#endif
880 return 1;
881}
882
883static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
884 int mirror_num, unsigned long bio_flags,
885 u64 bio_offset)
886{
887 struct inode *inode = private_data;
888 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
889 int async = check_async_write(BTRFS_I(inode));
890 blk_status_t ret;
891
892 if (bio_op(bio) != REQ_OP_WRITE) {
893 /*
894 * called for a read, do the setup so that checksum validation
895 * can happen in the async kernel threads
896 */
897 ret = btrfs_bio_wq_end_io(fs_info, bio,
898 BTRFS_WQ_ENDIO_METADATA);
899 if (ret)
900 goto out_w_error;
901 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
902 } else if (!async) {
903 ret = btree_csum_one_bio(bio);
904 if (ret)
905 goto out_w_error;
906 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
907 } else {
908 /*
909 * kthread helpers are used to submit writes so that
910 * checksumming can happen in parallel across all CPUs
911 */
912 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
913 bio_offset, private_data,
914 btree_submit_bio_start,
915 btree_submit_bio_done);
916 }
917
918 if (ret)
919 goto out_w_error;
920 return 0;
921
922out_w_error:
923 bio->bi_status = ret;
924 bio_endio(bio);
925 return ret;
926}
927
928#ifdef CONFIG_MIGRATION
929static int btree_migratepage(struct address_space *mapping,
930 struct page *newpage, struct page *page,
931 enum migrate_mode mode)
932{
933 /*
934 * we can't safely write a btree page from here,
935 * we haven't done the locking hook
936 */
937 if (PageDirty(page))
938 return -EAGAIN;
939 /*
940 * Buffers may be managed in a filesystem specific way.
941 * We must have no buffers or drop them.
942 */
943 if (page_has_private(page) &&
944 !try_to_release_page(page, GFP_KERNEL))
945 return -EAGAIN;
946 return migrate_page(mapping, newpage, page, mode);
947}
948#endif
949
950
951static int btree_writepages(struct address_space *mapping,
952 struct writeback_control *wbc)
953{
954 struct btrfs_fs_info *fs_info;
955 int ret;
956
957 if (wbc->sync_mode == WB_SYNC_NONE) {
958
959 if (wbc->for_kupdate)
960 return 0;
961
962 fs_info = BTRFS_I(mapping->host)->root->fs_info;
963 /* this is a bit racy, but that's ok */
964 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
965 BTRFS_DIRTY_METADATA_THRESH);
966 if (ret < 0)
967 return 0;
968 }
969 return btree_write_cache_pages(mapping, wbc);
970}
971
972static int btree_readpage(struct file *file, struct page *page)
973{
974 struct extent_io_tree *tree;
975 tree = &BTRFS_I(page->mapping->host)->io_tree;
976 return extent_read_full_page(tree, page, btree_get_extent, 0);
977}
978
979static int btree_releasepage(struct page *page, gfp_t gfp_flags)
980{
981 if (PageWriteback(page) || PageDirty(page))
982 return 0;
983
984 return try_release_extent_buffer(page);
985}
986
987static void btree_invalidatepage(struct page *page, unsigned int offset,
988 unsigned int length)
989{
990 struct extent_io_tree *tree;
991 tree = &BTRFS_I(page->mapping->host)->io_tree;
992 extent_invalidatepage(tree, page, offset);
993 btree_releasepage(page, GFP_NOFS);
994 if (PagePrivate(page)) {
995 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
996 "page private not zero on page %llu",
997 (unsigned long long)page_offset(page));
998 ClearPagePrivate(page);
999 set_page_private(page, 0);
1000 put_page(page);
1001 }
1002}
1003
1004static int btree_set_page_dirty(struct page *page)
1005{
1006#ifdef DEBUG
1007 struct extent_buffer *eb;
1008
1009 BUG_ON(!PagePrivate(page));
1010 eb = (struct extent_buffer *)page->private;
1011 BUG_ON(!eb);
1012 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1013 BUG_ON(!atomic_read(&eb->refs));
1014 btrfs_assert_tree_locked(eb);
1015#endif
1016 return __set_page_dirty_nobuffers(page);
1017}
1018
1019static const struct address_space_operations btree_aops = {
1020 .readpage = btree_readpage,
1021 .writepages = btree_writepages,
1022 .releasepage = btree_releasepage,
1023 .invalidatepage = btree_invalidatepage,
1024#ifdef CONFIG_MIGRATION
1025 .migratepage = btree_migratepage,
1026#endif
1027 .set_page_dirty = btree_set_page_dirty,
1028};
1029
1030void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1031{
1032 struct extent_buffer *buf = NULL;
1033 struct inode *btree_inode = fs_info->btree_inode;
1034
1035 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1036 if (IS_ERR(buf))
1037 return;
1038 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1039 buf, WAIT_NONE, 0);
1040 free_extent_buffer(buf);
1041}
1042
1043int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1044 int mirror_num, struct extent_buffer **eb)
1045{
1046 struct extent_buffer *buf = NULL;
1047 struct inode *btree_inode = fs_info->btree_inode;
1048 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1049 int ret;
1050
1051 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1052 if (IS_ERR(buf))
1053 return 0;
1054
1055 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1056
1057 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1058 mirror_num);
1059 if (ret) {
1060 free_extent_buffer(buf);
1061 return ret;
1062 }
1063
1064 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1065 free_extent_buffer(buf);
1066 return -EIO;
1067 } else if (extent_buffer_uptodate(buf)) {
1068 *eb = buf;
1069 } else {
1070 free_extent_buffer(buf);
1071 }
1072 return 0;
1073}
1074
1075struct extent_buffer *btrfs_find_create_tree_block(
1076 struct btrfs_fs_info *fs_info,
1077 u64 bytenr)
1078{
1079 if (btrfs_is_testing(fs_info))
1080 return alloc_test_extent_buffer(fs_info, bytenr);
1081 return alloc_extent_buffer(fs_info, bytenr);
1082}
1083
1084
1085int btrfs_write_tree_block(struct extent_buffer *buf)
1086{
1087 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1088 buf->start + buf->len - 1);
1089}
1090
1091void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1092{
1093 filemap_fdatawait_range(buf->pages[0]->mapping,
1094 buf->start, buf->start + buf->len - 1);
1095}
1096
1097/*
1098 * Read tree block at logical address @bytenr and do variant basic but critical
1099 * verification.
1100 *
1101 * @parent_transid: expected transid of this tree block, skip check if 0
1102 * @level: expected level, mandatory check
1103 * @first_key: expected key in slot 0, skip check if NULL
1104 */
1105struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1106 u64 parent_transid, int level,
1107 struct btrfs_key *first_key)
1108{
1109 struct extent_buffer *buf = NULL;
1110 int ret;
1111
1112 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1113 if (IS_ERR(buf))
1114 return buf;
1115
1116 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1117 level, first_key);
1118 if (ret) {
1119 free_extent_buffer(buf);
1120 return ERR_PTR(ret);
1121 }
1122 return buf;
1123
1124}
1125
1126void clean_tree_block(struct btrfs_fs_info *fs_info,
1127 struct extent_buffer *buf)
1128{
1129 if (btrfs_header_generation(buf) ==
1130 fs_info->running_transaction->transid) {
1131 btrfs_assert_tree_locked(buf);
1132
1133 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1134 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1135 -buf->len,
1136 fs_info->dirty_metadata_batch);
1137 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1138 btrfs_set_lock_blocking(buf);
1139 clear_extent_buffer_dirty(buf);
1140 }
1141 }
1142}
1143
1144static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1145{
1146 struct btrfs_subvolume_writers *writers;
1147 int ret;
1148
1149 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1150 if (!writers)
1151 return ERR_PTR(-ENOMEM);
1152
1153 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1154 if (ret < 0) {
1155 kfree(writers);
1156 return ERR_PTR(ret);
1157 }
1158
1159 init_waitqueue_head(&writers->wait);
1160 return writers;
1161}
1162
1163static void
1164btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1165{
1166 percpu_counter_destroy(&writers->counter);
1167 kfree(writers);
1168}
1169
1170static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1171 u64 objectid)
1172{
1173 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1174 root->node = NULL;
1175 root->commit_root = NULL;
1176 root->state = 0;
1177 root->orphan_cleanup_state = 0;
1178
1179 root->objectid = objectid;
1180 root->last_trans = 0;
1181 root->highest_objectid = 0;
1182 root->nr_delalloc_inodes = 0;
1183 root->nr_ordered_extents = 0;
1184 root->name = NULL;
1185 root->inode_tree = RB_ROOT;
1186 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1187 root->block_rsv = NULL;
1188 root->orphan_block_rsv = NULL;
1189
1190 INIT_LIST_HEAD(&root->dirty_list);
1191 INIT_LIST_HEAD(&root->root_list);
1192 INIT_LIST_HEAD(&root->delalloc_inodes);
1193 INIT_LIST_HEAD(&root->delalloc_root);
1194 INIT_LIST_HEAD(&root->ordered_extents);
1195 INIT_LIST_HEAD(&root->ordered_root);
1196 INIT_LIST_HEAD(&root->logged_list[0]);
1197 INIT_LIST_HEAD(&root->logged_list[1]);
1198 spin_lock_init(&root->orphan_lock);
1199 spin_lock_init(&root->inode_lock);
1200 spin_lock_init(&root->delalloc_lock);
1201 spin_lock_init(&root->ordered_extent_lock);
1202 spin_lock_init(&root->accounting_lock);
1203 spin_lock_init(&root->log_extents_lock[0]);
1204 spin_lock_init(&root->log_extents_lock[1]);
1205 spin_lock_init(&root->qgroup_meta_rsv_lock);
1206 mutex_init(&root->objectid_mutex);
1207 mutex_init(&root->log_mutex);
1208 mutex_init(&root->ordered_extent_mutex);
1209 mutex_init(&root->delalloc_mutex);
1210 init_waitqueue_head(&root->log_writer_wait);
1211 init_waitqueue_head(&root->log_commit_wait[0]);
1212 init_waitqueue_head(&root->log_commit_wait[1]);
1213 INIT_LIST_HEAD(&root->log_ctxs[0]);
1214 INIT_LIST_HEAD(&root->log_ctxs[1]);
1215 atomic_set(&root->log_commit[0], 0);
1216 atomic_set(&root->log_commit[1], 0);
1217 atomic_set(&root->log_writers, 0);
1218 atomic_set(&root->log_batch, 0);
1219 atomic_set(&root->orphan_inodes, 0);
1220 refcount_set(&root->refs, 1);
1221 atomic_set(&root->will_be_snapshotted, 0);
1222 root->log_transid = 0;
1223 root->log_transid_committed = -1;
1224 root->last_log_commit = 0;
1225 if (!dummy)
1226 extent_io_tree_init(&root->dirty_log_pages, NULL);
1227
1228 memset(&root->root_key, 0, sizeof(root->root_key));
1229 memset(&root->root_item, 0, sizeof(root->root_item));
1230 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1231 if (!dummy)
1232 root->defrag_trans_start = fs_info->generation;
1233 else
1234 root->defrag_trans_start = 0;
1235 root->root_key.objectid = objectid;
1236 root->anon_dev = 0;
1237
1238 spin_lock_init(&root->root_item_lock);
1239}
1240
1241static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1242 gfp_t flags)
1243{
1244 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1245 if (root)
1246 root->fs_info = fs_info;
1247 return root;
1248}
1249
1250#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1251/* Should only be used by the testing infrastructure */
1252struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1253{
1254 struct btrfs_root *root;
1255
1256 if (!fs_info)
1257 return ERR_PTR(-EINVAL);
1258
1259 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1260 if (!root)
1261 return ERR_PTR(-ENOMEM);
1262
1263 /* We don't use the stripesize in selftest, set it as sectorsize */
1264 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1265 root->alloc_bytenr = 0;
1266
1267 return root;
1268}
1269#endif
1270
1271struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1272 struct btrfs_fs_info *fs_info,
1273 u64 objectid)
1274{
1275 struct extent_buffer *leaf;
1276 struct btrfs_root *tree_root = fs_info->tree_root;
1277 struct btrfs_root *root;
1278 struct btrfs_key key;
1279 int ret = 0;
1280 uuid_le uuid = NULL_UUID_LE;
1281
1282 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1283 if (!root)
1284 return ERR_PTR(-ENOMEM);
1285
1286 __setup_root(root, fs_info, objectid);
1287 root->root_key.objectid = objectid;
1288 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1289 root->root_key.offset = 0;
1290
1291 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1292 if (IS_ERR(leaf)) {
1293 ret = PTR_ERR(leaf);
1294 leaf = NULL;
1295 goto fail;
1296 }
1297
1298 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1299 btrfs_set_header_bytenr(leaf, leaf->start);
1300 btrfs_set_header_generation(leaf, trans->transid);
1301 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1302 btrfs_set_header_owner(leaf, objectid);
1303 root->node = leaf;
1304
1305 write_extent_buffer_fsid(leaf, fs_info->fsid);
1306 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1307 btrfs_mark_buffer_dirty(leaf);
1308
1309 root->commit_root = btrfs_root_node(root);
1310 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1311
1312 root->root_item.flags = 0;
1313 root->root_item.byte_limit = 0;
1314 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1315 btrfs_set_root_generation(&root->root_item, trans->transid);
1316 btrfs_set_root_level(&root->root_item, 0);
1317 btrfs_set_root_refs(&root->root_item, 1);
1318 btrfs_set_root_used(&root->root_item, leaf->len);
1319 btrfs_set_root_last_snapshot(&root->root_item, 0);
1320 btrfs_set_root_dirid(&root->root_item, 0);
1321 if (is_fstree(objectid))
1322 uuid_le_gen(&uuid);
1323 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1324 root->root_item.drop_level = 0;
1325
1326 key.objectid = objectid;
1327 key.type = BTRFS_ROOT_ITEM_KEY;
1328 key.offset = 0;
1329 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1330 if (ret)
1331 goto fail;
1332
1333 btrfs_tree_unlock(leaf);
1334
1335 return root;
1336
1337fail:
1338 if (leaf) {
1339 btrfs_tree_unlock(leaf);
1340 free_extent_buffer(root->commit_root);
1341 free_extent_buffer(leaf);
1342 }
1343 kfree(root);
1344
1345 return ERR_PTR(ret);
1346}
1347
1348static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1349 struct btrfs_fs_info *fs_info)
1350{
1351 struct btrfs_root *root;
1352 struct extent_buffer *leaf;
1353
1354 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1355 if (!root)
1356 return ERR_PTR(-ENOMEM);
1357
1358 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1359
1360 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1361 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1362 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1363
1364 /*
1365 * DON'T set REF_COWS for log trees
1366 *
1367 * log trees do not get reference counted because they go away
1368 * before a real commit is actually done. They do store pointers
1369 * to file data extents, and those reference counts still get
1370 * updated (along with back refs to the log tree).
1371 */
1372
1373 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1374 NULL, 0, 0, 0);
1375 if (IS_ERR(leaf)) {
1376 kfree(root);
1377 return ERR_CAST(leaf);
1378 }
1379
1380 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1381 btrfs_set_header_bytenr(leaf, leaf->start);
1382 btrfs_set_header_generation(leaf, trans->transid);
1383 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1384 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1385 root->node = leaf;
1386
1387 write_extent_buffer_fsid(root->node, fs_info->fsid);
1388 btrfs_mark_buffer_dirty(root->node);
1389 btrfs_tree_unlock(root->node);
1390 return root;
1391}
1392
1393int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1394 struct btrfs_fs_info *fs_info)
1395{
1396 struct btrfs_root *log_root;
1397
1398 log_root = alloc_log_tree(trans, fs_info);
1399 if (IS_ERR(log_root))
1400 return PTR_ERR(log_root);
1401 WARN_ON(fs_info->log_root_tree);
1402 fs_info->log_root_tree = log_root;
1403 return 0;
1404}
1405
1406int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1407 struct btrfs_root *root)
1408{
1409 struct btrfs_fs_info *fs_info = root->fs_info;
1410 struct btrfs_root *log_root;
1411 struct btrfs_inode_item *inode_item;
1412
1413 log_root = alloc_log_tree(trans, fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1416
1417 log_root->last_trans = trans->transid;
1418 log_root->root_key.offset = root->root_key.objectid;
1419
1420 inode_item = &log_root->root_item.inode;
1421 btrfs_set_stack_inode_generation(inode_item, 1);
1422 btrfs_set_stack_inode_size(inode_item, 3);
1423 btrfs_set_stack_inode_nlink(inode_item, 1);
1424 btrfs_set_stack_inode_nbytes(inode_item,
1425 fs_info->nodesize);
1426 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1427
1428 btrfs_set_root_node(&log_root->root_item, log_root->node);
1429
1430 WARN_ON(root->log_root);
1431 root->log_root = log_root;
1432 root->log_transid = 0;
1433 root->log_transid_committed = -1;
1434 root->last_log_commit = 0;
1435 return 0;
1436}
1437
1438static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1439 struct btrfs_key *key)
1440{
1441 struct btrfs_root *root;
1442 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1443 struct btrfs_path *path;
1444 u64 generation;
1445 int ret;
1446 int level;
1447
1448 path = btrfs_alloc_path();
1449 if (!path)
1450 return ERR_PTR(-ENOMEM);
1451
1452 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1453 if (!root) {
1454 ret = -ENOMEM;
1455 goto alloc_fail;
1456 }
1457
1458 __setup_root(root, fs_info, key->objectid);
1459
1460 ret = btrfs_find_root(tree_root, key, path,
1461 &root->root_item, &root->root_key);
1462 if (ret) {
1463 if (ret > 0)
1464 ret = -ENOENT;
1465 goto find_fail;
1466 }
1467
1468 generation = btrfs_root_generation(&root->root_item);
1469 level = btrfs_root_level(&root->root_item);
1470 root->node = read_tree_block(fs_info,
1471 btrfs_root_bytenr(&root->root_item),
1472 generation, level, NULL);
1473 if (IS_ERR(root->node)) {
1474 ret = PTR_ERR(root->node);
1475 goto find_fail;
1476 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1477 ret = -EIO;
1478 free_extent_buffer(root->node);
1479 goto find_fail;
1480 }
1481 root->commit_root = btrfs_root_node(root);
1482out:
1483 btrfs_free_path(path);
1484 return root;
1485
1486find_fail:
1487 kfree(root);
1488alloc_fail:
1489 root = ERR_PTR(ret);
1490 goto out;
1491}
1492
1493struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1494 struct btrfs_key *location)
1495{
1496 struct btrfs_root *root;
1497
1498 root = btrfs_read_tree_root(tree_root, location);
1499 if (IS_ERR(root))
1500 return root;
1501
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1503 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1504 btrfs_check_and_init_root_item(&root->root_item);
1505 }
1506
1507 return root;
1508}
1509
1510int btrfs_init_fs_root(struct btrfs_root *root)
1511{
1512 int ret;
1513 struct btrfs_subvolume_writers *writers;
1514
1515 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1516 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1517 GFP_NOFS);
1518 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1519 ret = -ENOMEM;
1520 goto fail;
1521 }
1522
1523 writers = btrfs_alloc_subvolume_writers();
1524 if (IS_ERR(writers)) {
1525 ret = PTR_ERR(writers);
1526 goto fail;
1527 }
1528 root->subv_writers = writers;
1529
1530 btrfs_init_free_ino_ctl(root);
1531 spin_lock_init(&root->ino_cache_lock);
1532 init_waitqueue_head(&root->ino_cache_wait);
1533
1534 ret = get_anon_bdev(&root->anon_dev);
1535 if (ret)
1536 goto fail;
1537
1538 mutex_lock(&root->objectid_mutex);
1539 ret = btrfs_find_highest_objectid(root,
1540 &root->highest_objectid);
1541 if (ret) {
1542 mutex_unlock(&root->objectid_mutex);
1543 goto fail;
1544 }
1545
1546 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1547
1548 mutex_unlock(&root->objectid_mutex);
1549
1550 return 0;
1551fail:
1552 /* the caller is responsible to call free_fs_root */
1553 return ret;
1554}
1555
1556struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1557 u64 root_id)
1558{
1559 struct btrfs_root *root;
1560
1561 spin_lock(&fs_info->fs_roots_radix_lock);
1562 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1563 (unsigned long)root_id);
1564 spin_unlock(&fs_info->fs_roots_radix_lock);
1565 return root;
1566}
1567
1568int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1569 struct btrfs_root *root)
1570{
1571 int ret;
1572
1573 ret = radix_tree_preload(GFP_NOFS);
1574 if (ret)
1575 return ret;
1576
1577 spin_lock(&fs_info->fs_roots_radix_lock);
1578 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1579 (unsigned long)root->root_key.objectid,
1580 root);
1581 if (ret == 0)
1582 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1583 spin_unlock(&fs_info->fs_roots_radix_lock);
1584 radix_tree_preload_end();
1585
1586 return ret;
1587}
1588
1589struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1590 struct btrfs_key *location,
1591 bool check_ref)
1592{
1593 struct btrfs_root *root;
1594 struct btrfs_path *path;
1595 struct btrfs_key key;
1596 int ret;
1597
1598 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1599 return fs_info->tree_root;
1600 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1601 return fs_info->extent_root;
1602 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1603 return fs_info->chunk_root;
1604 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1605 return fs_info->dev_root;
1606 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1607 return fs_info->csum_root;
1608 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1609 return fs_info->quota_root ? fs_info->quota_root :
1610 ERR_PTR(-ENOENT);
1611 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1612 return fs_info->uuid_root ? fs_info->uuid_root :
1613 ERR_PTR(-ENOENT);
1614 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1615 return fs_info->free_space_root ? fs_info->free_space_root :
1616 ERR_PTR(-ENOENT);
1617again:
1618 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1619 if (root) {
1620 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1621 return ERR_PTR(-ENOENT);
1622 return root;
1623 }
1624
1625 root = btrfs_read_fs_root(fs_info->tree_root, location);
1626 if (IS_ERR(root))
1627 return root;
1628
1629 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1630 ret = -ENOENT;
1631 goto fail;
1632 }
1633
1634 ret = btrfs_init_fs_root(root);
1635 if (ret)
1636 goto fail;
1637
1638 path = btrfs_alloc_path();
1639 if (!path) {
1640 ret = -ENOMEM;
1641 goto fail;
1642 }
1643 key.objectid = BTRFS_ORPHAN_OBJECTID;
1644 key.type = BTRFS_ORPHAN_ITEM_KEY;
1645 key.offset = location->objectid;
1646
1647 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1648 btrfs_free_path(path);
1649 if (ret < 0)
1650 goto fail;
1651 if (ret == 0)
1652 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1653
1654 ret = btrfs_insert_fs_root(fs_info, root);
1655 if (ret) {
1656 if (ret == -EEXIST) {
1657 free_fs_root(root);
1658 goto again;
1659 }
1660 goto fail;
1661 }
1662 return root;
1663fail:
1664 free_fs_root(root);
1665 return ERR_PTR(ret);
1666}
1667
1668static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1669{
1670 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1671 int ret = 0;
1672 struct btrfs_device *device;
1673 struct backing_dev_info *bdi;
1674
1675 rcu_read_lock();
1676 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1677 if (!device->bdev)
1678 continue;
1679 bdi = device->bdev->bd_bdi;
1680 if (bdi_congested(bdi, bdi_bits)) {
1681 ret = 1;
1682 break;
1683 }
1684 }
1685 rcu_read_unlock();
1686 return ret;
1687}
1688
1689/*
1690 * called by the kthread helper functions to finally call the bio end_io
1691 * functions. This is where read checksum verification actually happens
1692 */
1693static void end_workqueue_fn(struct btrfs_work *work)
1694{
1695 struct bio *bio;
1696 struct btrfs_end_io_wq *end_io_wq;
1697
1698 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1699 bio = end_io_wq->bio;
1700
1701 bio->bi_status = end_io_wq->status;
1702 bio->bi_private = end_io_wq->private;
1703 bio->bi_end_io = end_io_wq->end_io;
1704 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1705 bio_endio(bio);
1706}
1707
1708static int cleaner_kthread(void *arg)
1709{
1710 struct btrfs_root *root = arg;
1711 struct btrfs_fs_info *fs_info = root->fs_info;
1712 int again;
1713 struct btrfs_trans_handle *trans;
1714
1715 do {
1716 again = 0;
1717
1718 /* Make the cleaner go to sleep early. */
1719 if (btrfs_need_cleaner_sleep(fs_info))
1720 goto sleep;
1721
1722 /*
1723 * Do not do anything if we might cause open_ctree() to block
1724 * before we have finished mounting the filesystem.
1725 */
1726 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1727 goto sleep;
1728
1729 if (!mutex_trylock(&fs_info->cleaner_mutex))
1730 goto sleep;
1731
1732 /*
1733 * Avoid the problem that we change the status of the fs
1734 * during the above check and trylock.
1735 */
1736 if (btrfs_need_cleaner_sleep(fs_info)) {
1737 mutex_unlock(&fs_info->cleaner_mutex);
1738 goto sleep;
1739 }
1740
1741 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1742 btrfs_run_delayed_iputs(fs_info);
1743 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1744
1745 again = btrfs_clean_one_deleted_snapshot(root);
1746 mutex_unlock(&fs_info->cleaner_mutex);
1747
1748 /*
1749 * The defragger has dealt with the R/O remount and umount,
1750 * needn't do anything special here.
1751 */
1752 btrfs_run_defrag_inodes(fs_info);
1753
1754 /*
1755 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1756 * with relocation (btrfs_relocate_chunk) and relocation
1757 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1758 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1759 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1760 * unused block groups.
1761 */
1762 btrfs_delete_unused_bgs(fs_info);
1763sleep:
1764 if (!again) {
1765 set_current_state(TASK_INTERRUPTIBLE);
1766 if (!kthread_should_stop())
1767 schedule();
1768 __set_current_state(TASK_RUNNING);
1769 }
1770 } while (!kthread_should_stop());
1771
1772 /*
1773 * Transaction kthread is stopped before us and wakes us up.
1774 * However we might have started a new transaction and COWed some
1775 * tree blocks when deleting unused block groups for example. So
1776 * make sure we commit the transaction we started to have a clean
1777 * shutdown when evicting the btree inode - if it has dirty pages
1778 * when we do the final iput() on it, eviction will trigger a
1779 * writeback for it which will fail with null pointer dereferences
1780 * since work queues and other resources were already released and
1781 * destroyed by the time the iput/eviction/writeback is made.
1782 */
1783 trans = btrfs_attach_transaction(root);
1784 if (IS_ERR(trans)) {
1785 if (PTR_ERR(trans) != -ENOENT)
1786 btrfs_err(fs_info,
1787 "cleaner transaction attach returned %ld",
1788 PTR_ERR(trans));
1789 } else {
1790 int ret;
1791
1792 ret = btrfs_commit_transaction(trans);
1793 if (ret)
1794 btrfs_err(fs_info,
1795 "cleaner open transaction commit returned %d",
1796 ret);
1797 }
1798
1799 return 0;
1800}
1801
1802static int transaction_kthread(void *arg)
1803{
1804 struct btrfs_root *root = arg;
1805 struct btrfs_fs_info *fs_info = root->fs_info;
1806 struct btrfs_trans_handle *trans;
1807 struct btrfs_transaction *cur;
1808 u64 transid;
1809 unsigned long now;
1810 unsigned long delay;
1811 bool cannot_commit;
1812
1813 do {
1814 cannot_commit = false;
1815 delay = HZ * fs_info->commit_interval;
1816 mutex_lock(&fs_info->transaction_kthread_mutex);
1817
1818 spin_lock(&fs_info->trans_lock);
1819 cur = fs_info->running_transaction;
1820 if (!cur) {
1821 spin_unlock(&fs_info->trans_lock);
1822 goto sleep;
1823 }
1824
1825 now = get_seconds();
1826 if (cur->state < TRANS_STATE_BLOCKED &&
1827 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1828 (now < cur->start_time ||
1829 now - cur->start_time < fs_info->commit_interval)) {
1830 spin_unlock(&fs_info->trans_lock);
1831 delay = HZ * 5;
1832 goto sleep;
1833 }
1834 transid = cur->transid;
1835 spin_unlock(&fs_info->trans_lock);
1836
1837 /* If the file system is aborted, this will always fail. */
1838 trans = btrfs_attach_transaction(root);
1839 if (IS_ERR(trans)) {
1840 if (PTR_ERR(trans) != -ENOENT)
1841 cannot_commit = true;
1842 goto sleep;
1843 }
1844 if (transid == trans->transid) {
1845 btrfs_commit_transaction(trans);
1846 } else {
1847 btrfs_end_transaction(trans);
1848 }
1849sleep:
1850 wake_up_process(fs_info->cleaner_kthread);
1851 mutex_unlock(&fs_info->transaction_kthread_mutex);
1852
1853 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1854 &fs_info->fs_state)))
1855 btrfs_cleanup_transaction(fs_info);
1856 if (!kthread_should_stop() &&
1857 (!btrfs_transaction_blocked(fs_info) ||
1858 cannot_commit))
1859 schedule_timeout_interruptible(delay);
1860 } while (!kthread_should_stop());
1861 return 0;
1862}
1863
1864/*
1865 * this will find the highest generation in the array of
1866 * root backups. The index of the highest array is returned,
1867 * or -1 if we can't find anything.
1868 *
1869 * We check to make sure the array is valid by comparing the
1870 * generation of the latest root in the array with the generation
1871 * in the super block. If they don't match we pitch it.
1872 */
1873static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1874{
1875 u64 cur;
1876 int newest_index = -1;
1877 struct btrfs_root_backup *root_backup;
1878 int i;
1879
1880 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1881 root_backup = info->super_copy->super_roots + i;
1882 cur = btrfs_backup_tree_root_gen(root_backup);
1883 if (cur == newest_gen)
1884 newest_index = i;
1885 }
1886
1887 /* check to see if we actually wrapped around */
1888 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1889 root_backup = info->super_copy->super_roots;
1890 cur = btrfs_backup_tree_root_gen(root_backup);
1891 if (cur == newest_gen)
1892 newest_index = 0;
1893 }
1894 return newest_index;
1895}
1896
1897
1898/*
1899 * find the oldest backup so we know where to store new entries
1900 * in the backup array. This will set the backup_root_index
1901 * field in the fs_info struct
1902 */
1903static void find_oldest_super_backup(struct btrfs_fs_info *info,
1904 u64 newest_gen)
1905{
1906 int newest_index = -1;
1907
1908 newest_index = find_newest_super_backup(info, newest_gen);
1909 /* if there was garbage in there, just move along */
1910 if (newest_index == -1) {
1911 info->backup_root_index = 0;
1912 } else {
1913 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1914 }
1915}
1916
1917/*
1918 * copy all the root pointers into the super backup array.
1919 * this will bump the backup pointer by one when it is
1920 * done
1921 */
1922static void backup_super_roots(struct btrfs_fs_info *info)
1923{
1924 int next_backup;
1925 struct btrfs_root_backup *root_backup;
1926 int last_backup;
1927
1928 next_backup = info->backup_root_index;
1929 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1930 BTRFS_NUM_BACKUP_ROOTS;
1931
1932 /*
1933 * just overwrite the last backup if we're at the same generation
1934 * this happens only at umount
1935 */
1936 root_backup = info->super_for_commit->super_roots + last_backup;
1937 if (btrfs_backup_tree_root_gen(root_backup) ==
1938 btrfs_header_generation(info->tree_root->node))
1939 next_backup = last_backup;
1940
1941 root_backup = info->super_for_commit->super_roots + next_backup;
1942
1943 /*
1944 * make sure all of our padding and empty slots get zero filled
1945 * regardless of which ones we use today
1946 */
1947 memset(root_backup, 0, sizeof(*root_backup));
1948
1949 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1950
1951 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1952 btrfs_set_backup_tree_root_gen(root_backup,
1953 btrfs_header_generation(info->tree_root->node));
1954
1955 btrfs_set_backup_tree_root_level(root_backup,
1956 btrfs_header_level(info->tree_root->node));
1957
1958 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1959 btrfs_set_backup_chunk_root_gen(root_backup,
1960 btrfs_header_generation(info->chunk_root->node));
1961 btrfs_set_backup_chunk_root_level(root_backup,
1962 btrfs_header_level(info->chunk_root->node));
1963
1964 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1965 btrfs_set_backup_extent_root_gen(root_backup,
1966 btrfs_header_generation(info->extent_root->node));
1967 btrfs_set_backup_extent_root_level(root_backup,
1968 btrfs_header_level(info->extent_root->node));
1969
1970 /*
1971 * we might commit during log recovery, which happens before we set
1972 * the fs_root. Make sure it is valid before we fill it in.
1973 */
1974 if (info->fs_root && info->fs_root->node) {
1975 btrfs_set_backup_fs_root(root_backup,
1976 info->fs_root->node->start);
1977 btrfs_set_backup_fs_root_gen(root_backup,
1978 btrfs_header_generation(info->fs_root->node));
1979 btrfs_set_backup_fs_root_level(root_backup,
1980 btrfs_header_level(info->fs_root->node));
1981 }
1982
1983 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1984 btrfs_set_backup_dev_root_gen(root_backup,
1985 btrfs_header_generation(info->dev_root->node));
1986 btrfs_set_backup_dev_root_level(root_backup,
1987 btrfs_header_level(info->dev_root->node));
1988
1989 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1990 btrfs_set_backup_csum_root_gen(root_backup,
1991 btrfs_header_generation(info->csum_root->node));
1992 btrfs_set_backup_csum_root_level(root_backup,
1993 btrfs_header_level(info->csum_root->node));
1994
1995 btrfs_set_backup_total_bytes(root_backup,
1996 btrfs_super_total_bytes(info->super_copy));
1997 btrfs_set_backup_bytes_used(root_backup,
1998 btrfs_super_bytes_used(info->super_copy));
1999 btrfs_set_backup_num_devices(root_backup,
2000 btrfs_super_num_devices(info->super_copy));
2001
2002 /*
2003 * if we don't copy this out to the super_copy, it won't get remembered
2004 * for the next commit
2005 */
2006 memcpy(&info->super_copy->super_roots,
2007 &info->super_for_commit->super_roots,
2008 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2009}
2010
2011/*
2012 * this copies info out of the root backup array and back into
2013 * the in-memory super block. It is meant to help iterate through
2014 * the array, so you send it the number of backups you've already
2015 * tried and the last backup index you used.
2016 *
2017 * this returns -1 when it has tried all the backups
2018 */
2019static noinline int next_root_backup(struct btrfs_fs_info *info,
2020 struct btrfs_super_block *super,
2021 int *num_backups_tried, int *backup_index)
2022{
2023 struct btrfs_root_backup *root_backup;
2024 int newest = *backup_index;
2025
2026 if (*num_backups_tried == 0) {
2027 u64 gen = btrfs_super_generation(super);
2028
2029 newest = find_newest_super_backup(info, gen);
2030 if (newest == -1)
2031 return -1;
2032
2033 *backup_index = newest;
2034 *num_backups_tried = 1;
2035 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2036 /* we've tried all the backups, all done */
2037 return -1;
2038 } else {
2039 /* jump to the next oldest backup */
2040 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2041 BTRFS_NUM_BACKUP_ROOTS;
2042 *backup_index = newest;
2043 *num_backups_tried += 1;
2044 }
2045 root_backup = super->super_roots + newest;
2046
2047 btrfs_set_super_generation(super,
2048 btrfs_backup_tree_root_gen(root_backup));
2049 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2050 btrfs_set_super_root_level(super,
2051 btrfs_backup_tree_root_level(root_backup));
2052 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2053
2054 /*
2055 * fixme: the total bytes and num_devices need to match or we should
2056 * need a fsck
2057 */
2058 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2059 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2060 return 0;
2061}
2062
2063/* helper to cleanup workers */
2064static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2065{
2066 btrfs_destroy_workqueue(fs_info->fixup_workers);
2067 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2068 btrfs_destroy_workqueue(fs_info->workers);
2069 btrfs_destroy_workqueue(fs_info->endio_workers);
2070 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2071 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2072 btrfs_destroy_workqueue(fs_info->rmw_workers);
2073 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2074 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2075 btrfs_destroy_workqueue(fs_info->submit_workers);
2076 btrfs_destroy_workqueue(fs_info->delayed_workers);
2077 btrfs_destroy_workqueue(fs_info->caching_workers);
2078 btrfs_destroy_workqueue(fs_info->readahead_workers);
2079 btrfs_destroy_workqueue(fs_info->flush_workers);
2080 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2081 btrfs_destroy_workqueue(fs_info->extent_workers);
2082 /*
2083 * Now that all other work queues are destroyed, we can safely destroy
2084 * the queues used for metadata I/O, since tasks from those other work
2085 * queues can do metadata I/O operations.
2086 */
2087 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2088 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2089}
2090
2091static void free_root_extent_buffers(struct btrfs_root *root)
2092{
2093 if (root) {
2094 free_extent_buffer(root->node);
2095 free_extent_buffer(root->commit_root);
2096 root->node = NULL;
2097 root->commit_root = NULL;
2098 }
2099}
2100
2101/* helper to cleanup tree roots */
2102static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2103{
2104 free_root_extent_buffers(info->tree_root);
2105
2106 free_root_extent_buffers(info->dev_root);
2107 free_root_extent_buffers(info->extent_root);
2108 free_root_extent_buffers(info->csum_root);
2109 free_root_extent_buffers(info->quota_root);
2110 free_root_extent_buffers(info->uuid_root);
2111 if (chunk_root)
2112 free_root_extent_buffers(info->chunk_root);
2113 free_root_extent_buffers(info->free_space_root);
2114}
2115
2116void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2117{
2118 int ret;
2119 struct btrfs_root *gang[8];
2120 int i;
2121
2122 while (!list_empty(&fs_info->dead_roots)) {
2123 gang[0] = list_entry(fs_info->dead_roots.next,
2124 struct btrfs_root, root_list);
2125 list_del(&gang[0]->root_list);
2126
2127 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2128 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2129 } else {
2130 free_extent_buffer(gang[0]->node);
2131 free_extent_buffer(gang[0]->commit_root);
2132 btrfs_put_fs_root(gang[0]);
2133 }
2134 }
2135
2136 while (1) {
2137 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2138 (void **)gang, 0,
2139 ARRAY_SIZE(gang));
2140 if (!ret)
2141 break;
2142 for (i = 0; i < ret; i++)
2143 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2144 }
2145
2146 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2147 btrfs_free_log_root_tree(NULL, fs_info);
2148 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2149 }
2150}
2151
2152static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2153{
2154 mutex_init(&fs_info->scrub_lock);
2155 atomic_set(&fs_info->scrubs_running, 0);
2156 atomic_set(&fs_info->scrub_pause_req, 0);
2157 atomic_set(&fs_info->scrubs_paused, 0);
2158 atomic_set(&fs_info->scrub_cancel_req, 0);
2159 init_waitqueue_head(&fs_info->scrub_pause_wait);
2160 fs_info->scrub_workers_refcnt = 0;
2161}
2162
2163static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2164{
2165 spin_lock_init(&fs_info->balance_lock);
2166 mutex_init(&fs_info->balance_mutex);
2167 atomic_set(&fs_info->balance_running, 0);
2168 atomic_set(&fs_info->balance_pause_req, 0);
2169 atomic_set(&fs_info->balance_cancel_req, 0);
2170 fs_info->balance_ctl = NULL;
2171 init_waitqueue_head(&fs_info->balance_wait_q);
2172}
2173
2174static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2175{
2176 struct inode *inode = fs_info->btree_inode;
2177
2178 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2179 set_nlink(inode, 1);
2180 /*
2181 * we set the i_size on the btree inode to the max possible int.
2182 * the real end of the address space is determined by all of
2183 * the devices in the system
2184 */
2185 inode->i_size = OFFSET_MAX;
2186 inode->i_mapping->a_ops = &btree_aops;
2187
2188 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2189 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2190 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2191 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2192
2193 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2194
2195 BTRFS_I(inode)->root = fs_info->tree_root;
2196 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2197 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2198 btrfs_insert_inode_hash(inode);
2199}
2200
2201static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2202{
2203 fs_info->dev_replace.lock_owner = 0;
2204 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2205 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2206 rwlock_init(&fs_info->dev_replace.lock);
2207 atomic_set(&fs_info->dev_replace.read_locks, 0);
2208 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2209 init_waitqueue_head(&fs_info->replace_wait);
2210 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2211}
2212
2213static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2214{
2215 spin_lock_init(&fs_info->qgroup_lock);
2216 mutex_init(&fs_info->qgroup_ioctl_lock);
2217 fs_info->qgroup_tree = RB_ROOT;
2218 fs_info->qgroup_op_tree = RB_ROOT;
2219 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2220 fs_info->qgroup_seq = 1;
2221 fs_info->qgroup_ulist = NULL;
2222 fs_info->qgroup_rescan_running = false;
2223 mutex_init(&fs_info->qgroup_rescan_lock);
2224}
2225
2226static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2227 struct btrfs_fs_devices *fs_devices)
2228{
2229 u32 max_active = fs_info->thread_pool_size;
2230 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2231
2232 fs_info->workers =
2233 btrfs_alloc_workqueue(fs_info, "worker",
2234 flags | WQ_HIGHPRI, max_active, 16);
2235
2236 fs_info->delalloc_workers =
2237 btrfs_alloc_workqueue(fs_info, "delalloc",
2238 flags, max_active, 2);
2239
2240 fs_info->flush_workers =
2241 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2242 flags, max_active, 0);
2243
2244 fs_info->caching_workers =
2245 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2246
2247 /*
2248 * a higher idle thresh on the submit workers makes it much more
2249 * likely that bios will be send down in a sane order to the
2250 * devices
2251 */
2252 fs_info->submit_workers =
2253 btrfs_alloc_workqueue(fs_info, "submit", flags,
2254 min_t(u64, fs_devices->num_devices,
2255 max_active), 64);
2256
2257 fs_info->fixup_workers =
2258 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2259
2260 /*
2261 * endios are largely parallel and should have a very
2262 * low idle thresh
2263 */
2264 fs_info->endio_workers =
2265 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2266 fs_info->endio_meta_workers =
2267 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2268 max_active, 4);
2269 fs_info->endio_meta_write_workers =
2270 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2271 max_active, 2);
2272 fs_info->endio_raid56_workers =
2273 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2274 max_active, 4);
2275 fs_info->endio_repair_workers =
2276 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2277 fs_info->rmw_workers =
2278 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2279 fs_info->endio_write_workers =
2280 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2281 max_active, 2);
2282 fs_info->endio_freespace_worker =
2283 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2284 max_active, 0);
2285 fs_info->delayed_workers =
2286 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2287 max_active, 0);
2288 fs_info->readahead_workers =
2289 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2290 max_active, 2);
2291 fs_info->qgroup_rescan_workers =
2292 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2293 fs_info->extent_workers =
2294 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2295 min_t(u64, fs_devices->num_devices,
2296 max_active), 8);
2297
2298 if (!(fs_info->workers && fs_info->delalloc_workers &&
2299 fs_info->submit_workers && fs_info->flush_workers &&
2300 fs_info->endio_workers && fs_info->endio_meta_workers &&
2301 fs_info->endio_meta_write_workers &&
2302 fs_info->endio_repair_workers &&
2303 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2304 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2305 fs_info->caching_workers && fs_info->readahead_workers &&
2306 fs_info->fixup_workers && fs_info->delayed_workers &&
2307 fs_info->extent_workers &&
2308 fs_info->qgroup_rescan_workers)) {
2309 return -ENOMEM;
2310 }
2311
2312 return 0;
2313}
2314
2315static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2316 struct btrfs_fs_devices *fs_devices)
2317{
2318 int ret;
2319 struct btrfs_root *log_tree_root;
2320 struct btrfs_super_block *disk_super = fs_info->super_copy;
2321 u64 bytenr = btrfs_super_log_root(disk_super);
2322 int level = btrfs_super_log_root_level(disk_super);
2323
2324 if (fs_devices->rw_devices == 0) {
2325 btrfs_warn(fs_info, "log replay required on RO media");
2326 return -EIO;
2327 }
2328
2329 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2330 if (!log_tree_root)
2331 return -ENOMEM;
2332
2333 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2334
2335 log_tree_root->node = read_tree_block(fs_info, bytenr,
2336 fs_info->generation + 1,
2337 level, NULL);
2338 if (IS_ERR(log_tree_root->node)) {
2339 btrfs_warn(fs_info, "failed to read log tree");
2340 ret = PTR_ERR(log_tree_root->node);
2341 kfree(log_tree_root);
2342 return ret;
2343 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2344 btrfs_err(fs_info, "failed to read log tree");
2345 free_extent_buffer(log_tree_root->node);
2346 kfree(log_tree_root);
2347 return -EIO;
2348 }
2349 /* returns with log_tree_root freed on success */
2350 ret = btrfs_recover_log_trees(log_tree_root);
2351 if (ret) {
2352 btrfs_handle_fs_error(fs_info, ret,
2353 "Failed to recover log tree");
2354 free_extent_buffer(log_tree_root->node);
2355 kfree(log_tree_root);
2356 return ret;
2357 }
2358
2359 if (sb_rdonly(fs_info->sb)) {
2360 ret = btrfs_commit_super(fs_info);
2361 if (ret)
2362 return ret;
2363 }
2364
2365 return 0;
2366}
2367
2368static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2369{
2370 struct btrfs_root *tree_root = fs_info->tree_root;
2371 struct btrfs_root *root;
2372 struct btrfs_key location;
2373 int ret;
2374
2375 BUG_ON(!fs_info->tree_root);
2376
2377 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2378 location.type = BTRFS_ROOT_ITEM_KEY;
2379 location.offset = 0;
2380
2381 root = btrfs_read_tree_root(tree_root, &location);
2382 if (IS_ERR(root)) {
2383 ret = PTR_ERR(root);
2384 goto out;
2385 }
2386 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2387 fs_info->extent_root = root;
2388
2389 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2390 root = btrfs_read_tree_root(tree_root, &location);
2391 if (IS_ERR(root)) {
2392 ret = PTR_ERR(root);
2393 goto out;
2394 }
2395 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2396 fs_info->dev_root = root;
2397 btrfs_init_devices_late(fs_info);
2398
2399 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2400 root = btrfs_read_tree_root(tree_root, &location);
2401 if (IS_ERR(root)) {
2402 ret = PTR_ERR(root);
2403 goto out;
2404 }
2405 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2406 fs_info->csum_root = root;
2407
2408 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2409 root = btrfs_read_tree_root(tree_root, &location);
2410 if (!IS_ERR(root)) {
2411 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2412 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2413 fs_info->quota_root = root;
2414 }
2415
2416 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2417 root = btrfs_read_tree_root(tree_root, &location);
2418 if (IS_ERR(root)) {
2419 ret = PTR_ERR(root);
2420 if (ret != -ENOENT)
2421 goto out;
2422 } else {
2423 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2424 fs_info->uuid_root = root;
2425 }
2426
2427 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2428 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2429 root = btrfs_read_tree_root(tree_root, &location);
2430 if (IS_ERR(root)) {
2431 ret = PTR_ERR(root);
2432 goto out;
2433 }
2434 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2435 fs_info->free_space_root = root;
2436 }
2437
2438 return 0;
2439out:
2440 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2441 location.objectid, ret);
2442 return ret;
2443}
2444
2445int open_ctree(struct super_block *sb,
2446 struct btrfs_fs_devices *fs_devices,
2447 char *options)
2448{
2449 u32 sectorsize;
2450 u32 nodesize;
2451 u32 stripesize;
2452 u64 generation;
2453 u64 features;
2454 struct btrfs_key location;
2455 struct buffer_head *bh;
2456 struct btrfs_super_block *disk_super;
2457 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2458 struct btrfs_root *tree_root;
2459 struct btrfs_root *chunk_root;
2460 int ret;
2461 int err = -EINVAL;
2462 int num_backups_tried = 0;
2463 int backup_index = 0;
2464 int clear_free_space_tree = 0;
2465 int level;
2466
2467 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2468 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2469 if (!tree_root || !chunk_root) {
2470 err = -ENOMEM;
2471 goto fail;
2472 }
2473
2474 ret = init_srcu_struct(&fs_info->subvol_srcu);
2475 if (ret) {
2476 err = ret;
2477 goto fail;
2478 }
2479
2480 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2481 if (ret) {
2482 err = ret;
2483 goto fail_srcu;
2484 }
2485 fs_info->dirty_metadata_batch = PAGE_SIZE *
2486 (1 + ilog2(nr_cpu_ids));
2487
2488 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2489 if (ret) {
2490 err = ret;
2491 goto fail_dirty_metadata_bytes;
2492 }
2493
2494 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2495 if (ret) {
2496 err = ret;
2497 goto fail_delalloc_bytes;
2498 }
2499
2500 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2501 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2502 INIT_LIST_HEAD(&fs_info->trans_list);
2503 INIT_LIST_HEAD(&fs_info->dead_roots);
2504 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2505 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2506 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2507 INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2508 spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2509 spin_lock_init(&fs_info->delalloc_root_lock);
2510 spin_lock_init(&fs_info->trans_lock);
2511 spin_lock_init(&fs_info->fs_roots_radix_lock);
2512 spin_lock_init(&fs_info->delayed_iput_lock);
2513 spin_lock_init(&fs_info->defrag_inodes_lock);
2514 spin_lock_init(&fs_info->tree_mod_seq_lock);
2515 spin_lock_init(&fs_info->super_lock);
2516 spin_lock_init(&fs_info->qgroup_op_lock);
2517 spin_lock_init(&fs_info->buffer_lock);
2518 spin_lock_init(&fs_info->unused_bgs_lock);
2519 rwlock_init(&fs_info->tree_mod_log_lock);
2520 mutex_init(&fs_info->unused_bg_unpin_mutex);
2521 mutex_init(&fs_info->delete_unused_bgs_mutex);
2522 mutex_init(&fs_info->reloc_mutex);
2523 mutex_init(&fs_info->delalloc_root_mutex);
2524 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2525 seqlock_init(&fs_info->profiles_lock);
2526
2527 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2528 INIT_LIST_HEAD(&fs_info->space_info);
2529 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2530 INIT_LIST_HEAD(&fs_info->unused_bgs);
2531 btrfs_mapping_init(&fs_info->mapping_tree);
2532 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2533 BTRFS_BLOCK_RSV_GLOBAL);
2534 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2535 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2536 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2537 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2538 BTRFS_BLOCK_RSV_DELOPS);
2539 atomic_set(&fs_info->async_delalloc_pages, 0);
2540 atomic_set(&fs_info->defrag_running, 0);
2541 atomic_set(&fs_info->qgroup_op_seq, 0);
2542 atomic_set(&fs_info->reada_works_cnt, 0);
2543 atomic64_set(&fs_info->tree_mod_seq, 0);
2544 fs_info->sb = sb;
2545 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2546 fs_info->metadata_ratio = 0;
2547 fs_info->defrag_inodes = RB_ROOT;
2548 atomic64_set(&fs_info->free_chunk_space, 0);
2549 fs_info->tree_mod_log = RB_ROOT;
2550 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2551 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2552 /* readahead state */
2553 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2554 spin_lock_init(&fs_info->reada_lock);
2555 btrfs_init_ref_verify(fs_info);
2556
2557 fs_info->thread_pool_size = min_t(unsigned long,
2558 num_online_cpus() + 2, 8);
2559
2560 INIT_LIST_HEAD(&fs_info->ordered_roots);
2561 spin_lock_init(&fs_info->ordered_root_lock);
2562
2563 fs_info->btree_inode = new_inode(sb);
2564 if (!fs_info->btree_inode) {
2565 err = -ENOMEM;
2566 goto fail_bio_counter;
2567 }
2568 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2569
2570 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2571 GFP_KERNEL);
2572 if (!fs_info->delayed_root) {
2573 err = -ENOMEM;
2574 goto fail_iput;
2575 }
2576 btrfs_init_delayed_root(fs_info->delayed_root);
2577
2578 btrfs_init_scrub(fs_info);
2579#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2580 fs_info->check_integrity_print_mask = 0;
2581#endif
2582 btrfs_init_balance(fs_info);
2583 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2584
2585 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2586 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2587
2588 btrfs_init_btree_inode(fs_info);
2589
2590 spin_lock_init(&fs_info->block_group_cache_lock);
2591 fs_info->block_group_cache_tree = RB_ROOT;
2592 fs_info->first_logical_byte = (u64)-1;
2593
2594 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2595 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2596 fs_info->pinned_extents = &fs_info->freed_extents[0];
2597 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2598
2599 mutex_init(&fs_info->ordered_operations_mutex);
2600 mutex_init(&fs_info->tree_log_mutex);
2601 mutex_init(&fs_info->chunk_mutex);
2602 mutex_init(&fs_info->transaction_kthread_mutex);
2603 mutex_init(&fs_info->cleaner_mutex);
2604 mutex_init(&fs_info->volume_mutex);
2605 mutex_init(&fs_info->ro_block_group_mutex);
2606 init_rwsem(&fs_info->commit_root_sem);
2607 init_rwsem(&fs_info->cleanup_work_sem);
2608 init_rwsem(&fs_info->subvol_sem);
2609 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2610
2611 btrfs_init_dev_replace_locks(fs_info);
2612 btrfs_init_qgroup(fs_info);
2613
2614 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2615 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2616
2617 init_waitqueue_head(&fs_info->transaction_throttle);
2618 init_waitqueue_head(&fs_info->transaction_wait);
2619 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2620 init_waitqueue_head(&fs_info->async_submit_wait);
2621
2622 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2623
2624 /* Usable values until the real ones are cached from the superblock */
2625 fs_info->nodesize = 4096;
2626 fs_info->sectorsize = 4096;
2627 fs_info->stripesize = 4096;
2628
2629 ret = btrfs_alloc_stripe_hash_table(fs_info);
2630 if (ret) {
2631 err = ret;
2632 goto fail_alloc;
2633 }
2634
2635 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2636
2637 invalidate_bdev(fs_devices->latest_bdev);
2638
2639 /*
2640 * Read super block and check the signature bytes only
2641 */
2642 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2643 if (IS_ERR(bh)) {
2644 err = PTR_ERR(bh);
2645 goto fail_alloc;
2646 }
2647
2648 /*
2649 * We want to check superblock checksum, the type is stored inside.
2650 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2651 */
2652 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2653 btrfs_err(fs_info, "superblock checksum mismatch");
2654 err = -EINVAL;
2655 brelse(bh);
2656 goto fail_alloc;
2657 }
2658
2659 /*
2660 * super_copy is zeroed at allocation time and we never touch the
2661 * following bytes up to INFO_SIZE, the checksum is calculated from
2662 * the whole block of INFO_SIZE
2663 */
2664 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2665 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2666 sizeof(*fs_info->super_for_commit));
2667 brelse(bh);
2668
2669 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2670
2671 ret = btrfs_check_super_valid(fs_info);
2672 if (ret) {
2673 btrfs_err(fs_info, "superblock contains fatal errors");
2674 err = -EINVAL;
2675 goto fail_alloc;
2676 }
2677
2678 disk_super = fs_info->super_copy;
2679 if (!btrfs_super_root(disk_super))
2680 goto fail_alloc;
2681
2682 /* check FS state, whether FS is broken. */
2683 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2684 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2685
2686 /*
2687 * run through our array of backup supers and setup
2688 * our ring pointer to the oldest one
2689 */
2690 generation = btrfs_super_generation(disk_super);
2691 find_oldest_super_backup(fs_info, generation);
2692
2693 /*
2694 * In the long term, we'll store the compression type in the super
2695 * block, and it'll be used for per file compression control.
2696 */
2697 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2698
2699 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2700 if (ret) {
2701 err = ret;
2702 goto fail_alloc;
2703 }
2704
2705 features = btrfs_super_incompat_flags(disk_super) &
2706 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2707 if (features) {
2708 btrfs_err(fs_info,
2709 "cannot mount because of unsupported optional features (%llx)",
2710 features);
2711 err = -EINVAL;
2712 goto fail_alloc;
2713 }
2714
2715 features = btrfs_super_incompat_flags(disk_super);
2716 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2717 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2718 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2719 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2720 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2721
2722 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2723 btrfs_info(fs_info, "has skinny extents");
2724
2725 /*
2726 * flag our filesystem as having big metadata blocks if
2727 * they are bigger than the page size
2728 */
2729 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2730 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2731 btrfs_info(fs_info,
2732 "flagging fs with big metadata feature");
2733 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2734 }
2735
2736 nodesize = btrfs_super_nodesize(disk_super);
2737 sectorsize = btrfs_super_sectorsize(disk_super);
2738 stripesize = sectorsize;
2739 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2740 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2741
2742 /* Cache block sizes */
2743 fs_info->nodesize = nodesize;
2744 fs_info->sectorsize = sectorsize;
2745 fs_info->stripesize = stripesize;
2746
2747 /*
2748 * mixed block groups end up with duplicate but slightly offset
2749 * extent buffers for the same range. It leads to corruptions
2750 */
2751 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2752 (sectorsize != nodesize)) {
2753 btrfs_err(fs_info,
2754"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2755 nodesize, sectorsize);
2756 goto fail_alloc;
2757 }
2758
2759 /*
2760 * Needn't use the lock because there is no other task which will
2761 * update the flag.
2762 */
2763 btrfs_set_super_incompat_flags(disk_super, features);
2764
2765 features = btrfs_super_compat_ro_flags(disk_super) &
2766 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2767 if (!sb_rdonly(sb) && features) {
2768 btrfs_err(fs_info,
2769 "cannot mount read-write because of unsupported optional features (%llx)",
2770 features);
2771 err = -EINVAL;
2772 goto fail_alloc;
2773 }
2774
2775 ret = btrfs_init_workqueues(fs_info, fs_devices);
2776 if (ret) {
2777 err = ret;
2778 goto fail_sb_buffer;
2779 }
2780
2781 sb->s_bdi->congested_fn = btrfs_congested_fn;
2782 sb->s_bdi->congested_data = fs_info;
2783 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2784 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2785 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2786 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2787
2788 sb->s_blocksize = sectorsize;
2789 sb->s_blocksize_bits = blksize_bits(sectorsize);
2790 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2791
2792 mutex_lock(&fs_info->chunk_mutex);
2793 ret = btrfs_read_sys_array(fs_info);
2794 mutex_unlock(&fs_info->chunk_mutex);
2795 if (ret) {
2796 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2797 goto fail_sb_buffer;
2798 }
2799
2800 generation = btrfs_super_chunk_root_generation(disk_super);
2801 level = btrfs_super_chunk_root_level(disk_super);
2802
2803 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2804
2805 chunk_root->node = read_tree_block(fs_info,
2806 btrfs_super_chunk_root(disk_super),
2807 generation, level, NULL);
2808 if (IS_ERR(chunk_root->node) ||
2809 !extent_buffer_uptodate(chunk_root->node)) {
2810 btrfs_err(fs_info, "failed to read chunk root");
2811 if (!IS_ERR(chunk_root->node))
2812 free_extent_buffer(chunk_root->node);
2813 chunk_root->node = NULL;
2814 goto fail_tree_roots;
2815 }
2816 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2817 chunk_root->commit_root = btrfs_root_node(chunk_root);
2818
2819 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2820 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2821
2822 ret = btrfs_read_chunk_tree(fs_info);
2823 if (ret) {
2824 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2825 goto fail_tree_roots;
2826 }
2827
2828 /*
2829 * Keep the devid that is marked to be the target device for the
2830 * device replace procedure
2831 */
2832 btrfs_free_extra_devids(fs_devices, 0);
2833
2834 if (!fs_devices->latest_bdev) {
2835 btrfs_err(fs_info, "failed to read devices");
2836 goto fail_tree_roots;
2837 }
2838
2839retry_root_backup:
2840 generation = btrfs_super_generation(disk_super);
2841 level = btrfs_super_root_level(disk_super);
2842
2843 tree_root->node = read_tree_block(fs_info,
2844 btrfs_super_root(disk_super),
2845 generation, level, NULL);
2846 if (IS_ERR(tree_root->node) ||
2847 !extent_buffer_uptodate(tree_root->node)) {
2848 btrfs_warn(fs_info, "failed to read tree root");
2849 if (!IS_ERR(tree_root->node))
2850 free_extent_buffer(tree_root->node);
2851 tree_root->node = NULL;
2852 goto recovery_tree_root;
2853 }
2854
2855 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2856 tree_root->commit_root = btrfs_root_node(tree_root);
2857 btrfs_set_root_refs(&tree_root->root_item, 1);
2858
2859 mutex_lock(&tree_root->objectid_mutex);
2860 ret = btrfs_find_highest_objectid(tree_root,
2861 &tree_root->highest_objectid);
2862 if (ret) {
2863 mutex_unlock(&tree_root->objectid_mutex);
2864 goto recovery_tree_root;
2865 }
2866
2867 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2868
2869 mutex_unlock(&tree_root->objectid_mutex);
2870
2871 ret = btrfs_read_roots(fs_info);
2872 if (ret)
2873 goto recovery_tree_root;
2874
2875 fs_info->generation = generation;
2876 fs_info->last_trans_committed = generation;
2877
2878 ret = btrfs_recover_balance(fs_info);
2879 if (ret) {
2880 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2881 goto fail_block_groups;
2882 }
2883
2884 ret = btrfs_init_dev_stats(fs_info);
2885 if (ret) {
2886 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2887 goto fail_block_groups;
2888 }
2889
2890 ret = btrfs_init_dev_replace(fs_info);
2891 if (ret) {
2892 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2893 goto fail_block_groups;
2894 }
2895
2896 btrfs_free_extra_devids(fs_devices, 1);
2897
2898 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2899 if (ret) {
2900 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2901 ret);
2902 goto fail_block_groups;
2903 }
2904
2905 ret = btrfs_sysfs_add_device(fs_devices);
2906 if (ret) {
2907 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2908 ret);
2909 goto fail_fsdev_sysfs;
2910 }
2911
2912 ret = btrfs_sysfs_add_mounted(fs_info);
2913 if (ret) {
2914 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2915 goto fail_fsdev_sysfs;
2916 }
2917
2918 ret = btrfs_init_space_info(fs_info);
2919 if (ret) {
2920 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2921 goto fail_sysfs;
2922 }
2923
2924 ret = btrfs_read_block_groups(fs_info);
2925 if (ret) {
2926 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2927 goto fail_sysfs;
2928 }
2929
2930 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
2931 btrfs_warn(fs_info,
2932 "writeable mount is not allowed due to too many missing devices");
2933 goto fail_sysfs;
2934 }
2935
2936 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2937 "btrfs-cleaner");
2938 if (IS_ERR(fs_info->cleaner_kthread))
2939 goto fail_sysfs;
2940
2941 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2942 tree_root,
2943 "btrfs-transaction");
2944 if (IS_ERR(fs_info->transaction_kthread))
2945 goto fail_cleaner;
2946
2947 if (!btrfs_test_opt(fs_info, NOSSD) &&
2948 !fs_info->fs_devices->rotating) {
2949 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2950 }
2951
2952 /*
2953 * Mount does not set all options immediately, we can do it now and do
2954 * not have to wait for transaction commit
2955 */
2956 btrfs_apply_pending_changes(fs_info);
2957
2958#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2959 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2960 ret = btrfsic_mount(fs_info, fs_devices,
2961 btrfs_test_opt(fs_info,
2962 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2963 1 : 0,
2964 fs_info->check_integrity_print_mask);
2965 if (ret)
2966 btrfs_warn(fs_info,
2967 "failed to initialize integrity check module: %d",
2968 ret);
2969 }
2970#endif
2971 ret = btrfs_read_qgroup_config(fs_info);
2972 if (ret)
2973 goto fail_trans_kthread;
2974
2975 if (btrfs_build_ref_tree(fs_info))
2976 btrfs_err(fs_info, "couldn't build ref tree");
2977
2978 /* do not make disk changes in broken FS or nologreplay is given */
2979 if (btrfs_super_log_root(disk_super) != 0 &&
2980 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2981 ret = btrfs_replay_log(fs_info, fs_devices);
2982 if (ret) {
2983 err = ret;
2984 goto fail_qgroup;
2985 }
2986 }
2987
2988 ret = btrfs_find_orphan_roots(fs_info);
2989 if (ret)
2990 goto fail_qgroup;
2991
2992 if (!sb_rdonly(sb)) {
2993 ret = btrfs_cleanup_fs_roots(fs_info);
2994 if (ret)
2995 goto fail_qgroup;
2996
2997 mutex_lock(&fs_info->cleaner_mutex);
2998 ret = btrfs_recover_relocation(tree_root);
2999 mutex_unlock(&fs_info->cleaner_mutex);
3000 if (ret < 0) {
3001 btrfs_warn(fs_info, "failed to recover relocation: %d",
3002 ret);
3003 err = -EINVAL;
3004 goto fail_qgroup;
3005 }
3006 }
3007
3008 location.objectid = BTRFS_FS_TREE_OBJECTID;
3009 location.type = BTRFS_ROOT_ITEM_KEY;
3010 location.offset = 0;
3011
3012 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3013 if (IS_ERR(fs_info->fs_root)) {
3014 err = PTR_ERR(fs_info->fs_root);
3015 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3016 goto fail_qgroup;
3017 }
3018
3019 if (sb_rdonly(sb))
3020 return 0;
3021
3022 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3023 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3024 clear_free_space_tree = 1;
3025 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3026 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3027 btrfs_warn(fs_info, "free space tree is invalid");
3028 clear_free_space_tree = 1;
3029 }
3030
3031 if (clear_free_space_tree) {
3032 btrfs_info(fs_info, "clearing free space tree");
3033 ret = btrfs_clear_free_space_tree(fs_info);
3034 if (ret) {
3035 btrfs_warn(fs_info,
3036 "failed to clear free space tree: %d", ret);
3037 close_ctree(fs_info);
3038 return ret;
3039 }
3040 }
3041
3042 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3043 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3044 btrfs_info(fs_info, "creating free space tree");
3045 ret = btrfs_create_free_space_tree(fs_info);
3046 if (ret) {
3047 btrfs_warn(fs_info,
3048 "failed to create free space tree: %d", ret);
3049 close_ctree(fs_info);
3050 return ret;
3051 }
3052 }
3053
3054 down_read(&fs_info->cleanup_work_sem);
3055 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3056 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3057 up_read(&fs_info->cleanup_work_sem);
3058 close_ctree(fs_info);
3059 return ret;
3060 }
3061 up_read(&fs_info->cleanup_work_sem);
3062
3063 ret = btrfs_resume_balance_async(fs_info);
3064 if (ret) {
3065 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3066 close_ctree(fs_info);
3067 return ret;
3068 }
3069
3070 ret = btrfs_resume_dev_replace_async(fs_info);
3071 if (ret) {
3072 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3073 close_ctree(fs_info);
3074 return ret;
3075 }
3076
3077 btrfs_qgroup_rescan_resume(fs_info);
3078
3079 if (!fs_info->uuid_root) {
3080 btrfs_info(fs_info, "creating UUID tree");
3081 ret = btrfs_create_uuid_tree(fs_info);
3082 if (ret) {
3083 btrfs_warn(fs_info,
3084 "failed to create the UUID tree: %d", ret);
3085 close_ctree(fs_info);
3086 return ret;
3087 }
3088 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3089 fs_info->generation !=
3090 btrfs_super_uuid_tree_generation(disk_super)) {
3091 btrfs_info(fs_info, "checking UUID tree");
3092 ret = btrfs_check_uuid_tree(fs_info);
3093 if (ret) {
3094 btrfs_warn(fs_info,
3095 "failed to check the UUID tree: %d", ret);
3096 close_ctree(fs_info);
3097 return ret;
3098 }
3099 } else {
3100 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3101 }
3102 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3103
3104 /*
3105 * backuproot only affect mount behavior, and if open_ctree succeeded,
3106 * no need to keep the flag
3107 */
3108 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3109
3110 return 0;
3111
3112fail_qgroup:
3113 btrfs_free_qgroup_config(fs_info);
3114fail_trans_kthread:
3115 kthread_stop(fs_info->transaction_kthread);
3116 btrfs_cleanup_transaction(fs_info);
3117 btrfs_free_fs_roots(fs_info);
3118fail_cleaner:
3119 kthread_stop(fs_info->cleaner_kthread);
3120
3121 /*
3122 * make sure we're done with the btree inode before we stop our
3123 * kthreads
3124 */
3125 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3126
3127fail_sysfs:
3128 btrfs_sysfs_remove_mounted(fs_info);
3129
3130fail_fsdev_sysfs:
3131 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3132
3133fail_block_groups:
3134 btrfs_put_block_group_cache(fs_info);
3135
3136fail_tree_roots:
3137 free_root_pointers(fs_info, 1);
3138 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3139
3140fail_sb_buffer:
3141 btrfs_stop_all_workers(fs_info);
3142 btrfs_free_block_groups(fs_info);
3143fail_alloc:
3144fail_iput:
3145 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3146
3147 iput(fs_info->btree_inode);
3148fail_bio_counter:
3149 percpu_counter_destroy(&fs_info->bio_counter);
3150fail_delalloc_bytes:
3151 percpu_counter_destroy(&fs_info->delalloc_bytes);
3152fail_dirty_metadata_bytes:
3153 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3154fail_srcu:
3155 cleanup_srcu_struct(&fs_info->subvol_srcu);
3156fail:
3157 btrfs_free_stripe_hash_table(fs_info);
3158 btrfs_close_devices(fs_info->fs_devices);
3159 return err;
3160
3161recovery_tree_root:
3162 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3163 goto fail_tree_roots;
3164
3165 free_root_pointers(fs_info, 0);
3166
3167 /* don't use the log in recovery mode, it won't be valid */
3168 btrfs_set_super_log_root(disk_super, 0);
3169
3170 /* we can't trust the free space cache either */
3171 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3172
3173 ret = next_root_backup(fs_info, fs_info->super_copy,
3174 &num_backups_tried, &backup_index);
3175 if (ret == -1)
3176 goto fail_block_groups;
3177 goto retry_root_backup;
3178}
3179ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3180
3181static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3182{
3183 if (uptodate) {
3184 set_buffer_uptodate(bh);
3185 } else {
3186 struct btrfs_device *device = (struct btrfs_device *)
3187 bh->b_private;
3188
3189 btrfs_warn_rl_in_rcu(device->fs_info,
3190 "lost page write due to IO error on %s",
3191 rcu_str_deref(device->name));
3192 /* note, we don't set_buffer_write_io_error because we have
3193 * our own ways of dealing with the IO errors
3194 */
3195 clear_buffer_uptodate(bh);
3196 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3197 }
3198 unlock_buffer(bh);
3199 put_bh(bh);
3200}
3201
3202int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3203 struct buffer_head **bh_ret)
3204{
3205 struct buffer_head *bh;
3206 struct btrfs_super_block *super;
3207 u64 bytenr;
3208
3209 bytenr = btrfs_sb_offset(copy_num);
3210 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3211 return -EINVAL;
3212
3213 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3214 /*
3215 * If we fail to read from the underlying devices, as of now
3216 * the best option we have is to mark it EIO.
3217 */
3218 if (!bh)
3219 return -EIO;
3220
3221 super = (struct btrfs_super_block *)bh->b_data;
3222 if (btrfs_super_bytenr(super) != bytenr ||
3223 btrfs_super_magic(super) != BTRFS_MAGIC) {
3224 brelse(bh);
3225 return -EINVAL;
3226 }
3227
3228 *bh_ret = bh;
3229 return 0;
3230}
3231
3232
3233struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3234{
3235 struct buffer_head *bh;
3236 struct buffer_head *latest = NULL;
3237 struct btrfs_super_block *super;
3238 int i;
3239 u64 transid = 0;
3240 int ret = -EINVAL;
3241
3242 /* we would like to check all the supers, but that would make
3243 * a btrfs mount succeed after a mkfs from a different FS.
3244 * So, we need to add a special mount option to scan for
3245 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3246 */
3247 for (i = 0; i < 1; i++) {
3248 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3249 if (ret)
3250 continue;
3251
3252 super = (struct btrfs_super_block *)bh->b_data;
3253
3254 if (!latest || btrfs_super_generation(super) > transid) {
3255 brelse(latest);
3256 latest = bh;
3257 transid = btrfs_super_generation(super);
3258 } else {
3259 brelse(bh);
3260 }
3261 }
3262
3263 if (!latest)
3264 return ERR_PTR(ret);
3265
3266 return latest;
3267}
3268
3269/*
3270 * Write superblock @sb to the @device. Do not wait for completion, all the
3271 * buffer heads we write are pinned.
3272 *
3273 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3274 * the expected device size at commit time. Note that max_mirrors must be
3275 * same for write and wait phases.
3276 *
3277 * Return number of errors when buffer head is not found or submission fails.
3278 */
3279static int write_dev_supers(struct btrfs_device *device,
3280 struct btrfs_super_block *sb, int max_mirrors)
3281{
3282 struct buffer_head *bh;
3283 int i;
3284 int ret;
3285 int errors = 0;
3286 u32 crc;
3287 u64 bytenr;
3288 int op_flags;
3289
3290 if (max_mirrors == 0)
3291 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3292
3293 for (i = 0; i < max_mirrors; i++) {
3294 bytenr = btrfs_sb_offset(i);
3295 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3296 device->commit_total_bytes)
3297 break;
3298
3299 btrfs_set_super_bytenr(sb, bytenr);
3300
3301 crc = ~(u32)0;
3302 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3303 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3304 btrfs_csum_final(crc, sb->csum);
3305
3306 /* One reference for us, and we leave it for the caller */
3307 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3308 BTRFS_SUPER_INFO_SIZE);
3309 if (!bh) {
3310 btrfs_err(device->fs_info,
3311 "couldn't get super buffer head for bytenr %llu",
3312 bytenr);
3313 errors++;
3314 continue;
3315 }
3316
3317 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3318
3319 /* one reference for submit_bh */
3320 get_bh(bh);
3321
3322 set_buffer_uptodate(bh);
3323 lock_buffer(bh);
3324 bh->b_end_io = btrfs_end_buffer_write_sync;
3325 bh->b_private = device;
3326
3327 /*
3328 * we fua the first super. The others we allow
3329 * to go down lazy.
3330 */
3331 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3332 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3333 op_flags |= REQ_FUA;
3334 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3335 if (ret)
3336 errors++;
3337 }
3338 return errors < i ? 0 : -1;
3339}
3340
3341/*
3342 * Wait for write completion of superblocks done by write_dev_supers,
3343 * @max_mirrors same for write and wait phases.
3344 *
3345 * Return number of errors when buffer head is not found or not marked up to
3346 * date.
3347 */
3348static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3349{
3350 struct buffer_head *bh;
3351 int i;
3352 int errors = 0;
3353 bool primary_failed = false;
3354 u64 bytenr;
3355
3356 if (max_mirrors == 0)
3357 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3358
3359 for (i = 0; i < max_mirrors; i++) {
3360 bytenr = btrfs_sb_offset(i);
3361 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3362 device->commit_total_bytes)
3363 break;
3364
3365 bh = __find_get_block(device->bdev,
3366 bytenr / BTRFS_BDEV_BLOCKSIZE,
3367 BTRFS_SUPER_INFO_SIZE);
3368 if (!bh) {
3369 errors++;
3370 if (i == 0)
3371 primary_failed = true;
3372 continue;
3373 }
3374 wait_on_buffer(bh);
3375 if (!buffer_uptodate(bh)) {
3376 errors++;
3377 if (i == 0)
3378 primary_failed = true;
3379 }
3380
3381 /* drop our reference */
3382 brelse(bh);
3383
3384 /* drop the reference from the writing run */
3385 brelse(bh);
3386 }
3387
3388 /* log error, force error return */
3389 if (primary_failed) {
3390 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3391 device->devid);
3392 return -1;
3393 }
3394
3395 return errors < i ? 0 : -1;
3396}
3397
3398/*
3399 * endio for the write_dev_flush, this will wake anyone waiting
3400 * for the barrier when it is done
3401 */
3402static void btrfs_end_empty_barrier(struct bio *bio)
3403{
3404 complete(bio->bi_private);
3405}
3406
3407/*
3408 * Submit a flush request to the device if it supports it. Error handling is
3409 * done in the waiting counterpart.
3410 */
3411static void write_dev_flush(struct btrfs_device *device)
3412{
3413 struct request_queue *q = bdev_get_queue(device->bdev);
3414 struct bio *bio = device->flush_bio;
3415
3416 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3417 return;
3418
3419 bio_reset(bio);
3420 bio->bi_end_io = btrfs_end_empty_barrier;
3421 bio_set_dev(bio, device->bdev);
3422 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3423 init_completion(&device->flush_wait);
3424 bio->bi_private = &device->flush_wait;
3425
3426 btrfsic_submit_bio(bio);
3427 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3428}
3429
3430/*
3431 * If the flush bio has been submitted by write_dev_flush, wait for it.
3432 */
3433static blk_status_t wait_dev_flush(struct btrfs_device *device)
3434{
3435 struct bio *bio = device->flush_bio;
3436
3437 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3438 return BLK_STS_OK;
3439
3440 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3441 wait_for_completion_io(&device->flush_wait);
3442
3443 return bio->bi_status;
3444}
3445
3446static int check_barrier_error(struct btrfs_fs_info *fs_info)
3447{
3448 if (!btrfs_check_rw_degradable(fs_info, NULL))
3449 return -EIO;
3450 return 0;
3451}
3452
3453/*
3454 * send an empty flush down to each device in parallel,
3455 * then wait for them
3456 */
3457static int barrier_all_devices(struct btrfs_fs_info *info)
3458{
3459 struct list_head *head;
3460 struct btrfs_device *dev;
3461 int errors_wait = 0;
3462 blk_status_t ret;
3463
3464 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3465 /* send down all the barriers */
3466 head = &info->fs_devices->devices;
3467 list_for_each_entry(dev, head, dev_list) {
3468 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3469 continue;
3470 if (!dev->bdev)
3471 continue;
3472 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3473 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3474 continue;
3475
3476 write_dev_flush(dev);
3477 dev->last_flush_error = BLK_STS_OK;
3478 }
3479
3480 /* wait for all the barriers */
3481 list_for_each_entry(dev, head, dev_list) {
3482 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3483 continue;
3484 if (!dev->bdev) {
3485 errors_wait++;
3486 continue;
3487 }
3488 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3489 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3490 continue;
3491
3492 ret = wait_dev_flush(dev);
3493 if (ret) {
3494 dev->last_flush_error = ret;
3495 btrfs_dev_stat_inc_and_print(dev,
3496 BTRFS_DEV_STAT_FLUSH_ERRS);
3497 errors_wait++;
3498 }
3499 }
3500
3501 if (errors_wait) {
3502 /*
3503 * At some point we need the status of all disks
3504 * to arrive at the volume status. So error checking
3505 * is being pushed to a separate loop.
3506 */
3507 return check_barrier_error(info);
3508 }
3509 return 0;
3510}
3511
3512int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3513{
3514 int raid_type;
3515 int min_tolerated = INT_MAX;
3516
3517 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3518 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3519 min_tolerated = min(min_tolerated,
3520 btrfs_raid_array[BTRFS_RAID_SINGLE].
3521 tolerated_failures);
3522
3523 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3524 if (raid_type == BTRFS_RAID_SINGLE)
3525 continue;
3526 if (!(flags & btrfs_raid_group[raid_type]))
3527 continue;
3528 min_tolerated = min(min_tolerated,
3529 btrfs_raid_array[raid_type].
3530 tolerated_failures);
3531 }
3532
3533 if (min_tolerated == INT_MAX) {
3534 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3535 min_tolerated = 0;
3536 }
3537
3538 return min_tolerated;
3539}
3540
3541int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3542{
3543 struct list_head *head;
3544 struct btrfs_device *dev;
3545 struct btrfs_super_block *sb;
3546 struct btrfs_dev_item *dev_item;
3547 int ret;
3548 int do_barriers;
3549 int max_errors;
3550 int total_errors = 0;
3551 u64 flags;
3552
3553 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3554
3555 /*
3556 * max_mirrors == 0 indicates we're from commit_transaction,
3557 * not from fsync where the tree roots in fs_info have not
3558 * been consistent on disk.
3559 */
3560 if (max_mirrors == 0)
3561 backup_super_roots(fs_info);
3562
3563 sb = fs_info->super_for_commit;
3564 dev_item = &sb->dev_item;
3565
3566 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3567 head = &fs_info->fs_devices->devices;
3568 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3569
3570 if (do_barriers) {
3571 ret = barrier_all_devices(fs_info);
3572 if (ret) {
3573 mutex_unlock(
3574 &fs_info->fs_devices->device_list_mutex);
3575 btrfs_handle_fs_error(fs_info, ret,
3576 "errors while submitting device barriers.");
3577 return ret;
3578 }
3579 }
3580
3581 list_for_each_entry(dev, head, dev_list) {
3582 if (!dev->bdev) {
3583 total_errors++;
3584 continue;
3585 }
3586 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3587 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3588 continue;
3589
3590 btrfs_set_stack_device_generation(dev_item, 0);
3591 btrfs_set_stack_device_type(dev_item, dev->type);
3592 btrfs_set_stack_device_id(dev_item, dev->devid);
3593 btrfs_set_stack_device_total_bytes(dev_item,
3594 dev->commit_total_bytes);
3595 btrfs_set_stack_device_bytes_used(dev_item,
3596 dev->commit_bytes_used);
3597 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3598 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3599 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3600 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3601 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3602
3603 flags = btrfs_super_flags(sb);
3604 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3605
3606 ret = write_dev_supers(dev, sb, max_mirrors);
3607 if (ret)
3608 total_errors++;
3609 }
3610 if (total_errors > max_errors) {
3611 btrfs_err(fs_info, "%d errors while writing supers",
3612 total_errors);
3613 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3614
3615 /* FUA is masked off if unsupported and can't be the reason */
3616 btrfs_handle_fs_error(fs_info, -EIO,
3617 "%d errors while writing supers",
3618 total_errors);
3619 return -EIO;
3620 }
3621
3622 total_errors = 0;
3623 list_for_each_entry(dev, head, dev_list) {
3624 if (!dev->bdev)
3625 continue;
3626 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3627 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3628 continue;
3629
3630 ret = wait_dev_supers(dev, max_mirrors);
3631 if (ret)
3632 total_errors++;
3633 }
3634 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3635 if (total_errors > max_errors) {
3636 btrfs_handle_fs_error(fs_info, -EIO,
3637 "%d errors while writing supers",
3638 total_errors);
3639 return -EIO;
3640 }
3641 return 0;
3642}
3643
3644/* Drop a fs root from the radix tree and free it. */
3645void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3646 struct btrfs_root *root)
3647{
3648 spin_lock(&fs_info->fs_roots_radix_lock);
3649 radix_tree_delete(&fs_info->fs_roots_radix,
3650 (unsigned long)root->root_key.objectid);
3651 spin_unlock(&fs_info->fs_roots_radix_lock);
3652
3653 if (btrfs_root_refs(&root->root_item) == 0)
3654 synchronize_srcu(&fs_info->subvol_srcu);
3655
3656 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3657 btrfs_free_log(NULL, root);
3658 if (root->reloc_root) {
3659 free_extent_buffer(root->reloc_root->node);
3660 free_extent_buffer(root->reloc_root->commit_root);
3661 btrfs_put_fs_root(root->reloc_root);
3662 root->reloc_root = NULL;
3663 }
3664 }
3665
3666 if (root->free_ino_pinned)
3667 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3668 if (root->free_ino_ctl)
3669 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3670 free_fs_root(root);
3671}
3672
3673static void free_fs_root(struct btrfs_root *root)
3674{
3675 iput(root->ino_cache_inode);
3676 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3677 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3678 root->orphan_block_rsv = NULL;
3679 if (root->anon_dev)
3680 free_anon_bdev(root->anon_dev);
3681 if (root->subv_writers)
3682 btrfs_free_subvolume_writers(root->subv_writers);
3683 free_extent_buffer(root->node);
3684 free_extent_buffer(root->commit_root);
3685 kfree(root->free_ino_ctl);
3686 kfree(root->free_ino_pinned);
3687 kfree(root->name);
3688 btrfs_put_fs_root(root);
3689}
3690
3691void btrfs_free_fs_root(struct btrfs_root *root)
3692{
3693 free_fs_root(root);
3694}
3695
3696int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3697{
3698 u64 root_objectid = 0;
3699 struct btrfs_root *gang[8];
3700 int i = 0;
3701 int err = 0;
3702 unsigned int ret = 0;
3703 int index;
3704
3705 while (1) {
3706 index = srcu_read_lock(&fs_info->subvol_srcu);
3707 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3708 (void **)gang, root_objectid,
3709 ARRAY_SIZE(gang));
3710 if (!ret) {
3711 srcu_read_unlock(&fs_info->subvol_srcu, index);
3712 break;
3713 }
3714 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3715
3716 for (i = 0; i < ret; i++) {
3717 /* Avoid to grab roots in dead_roots */
3718 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3719 gang[i] = NULL;
3720 continue;
3721 }
3722 /* grab all the search result for later use */
3723 gang[i] = btrfs_grab_fs_root(gang[i]);
3724 }
3725 srcu_read_unlock(&fs_info->subvol_srcu, index);
3726
3727 for (i = 0; i < ret; i++) {
3728 if (!gang[i])
3729 continue;
3730 root_objectid = gang[i]->root_key.objectid;
3731 err = btrfs_orphan_cleanup(gang[i]);
3732 if (err)
3733 break;
3734 btrfs_put_fs_root(gang[i]);
3735 }
3736 root_objectid++;
3737 }
3738
3739 /* release the uncleaned roots due to error */
3740 for (; i < ret; i++) {
3741 if (gang[i])
3742 btrfs_put_fs_root(gang[i]);
3743 }
3744 return err;
3745}
3746
3747int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3748{
3749 struct btrfs_root *root = fs_info->tree_root;
3750 struct btrfs_trans_handle *trans;
3751
3752 mutex_lock(&fs_info->cleaner_mutex);
3753 btrfs_run_delayed_iputs(fs_info);
3754 mutex_unlock(&fs_info->cleaner_mutex);
3755 wake_up_process(fs_info->cleaner_kthread);
3756
3757 /* wait until ongoing cleanup work done */
3758 down_write(&fs_info->cleanup_work_sem);
3759 up_write(&fs_info->cleanup_work_sem);
3760
3761 trans = btrfs_join_transaction(root);
3762 if (IS_ERR(trans))
3763 return PTR_ERR(trans);
3764 return btrfs_commit_transaction(trans);
3765}
3766
3767void close_ctree(struct btrfs_fs_info *fs_info)
3768{
3769 struct btrfs_root *root = fs_info->tree_root;
3770 int ret;
3771
3772 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3773
3774 /* wait for the qgroup rescan worker to stop */
3775 btrfs_qgroup_wait_for_completion(fs_info, false);
3776
3777 /* wait for the uuid_scan task to finish */
3778 down(&fs_info->uuid_tree_rescan_sem);
3779 /* avoid complains from lockdep et al., set sem back to initial state */
3780 up(&fs_info->uuid_tree_rescan_sem);
3781
3782 /* pause restriper - we want to resume on mount */
3783 btrfs_pause_balance(fs_info);
3784
3785 btrfs_dev_replace_suspend_for_unmount(fs_info);
3786
3787 btrfs_scrub_cancel(fs_info);
3788
3789 /* wait for any defraggers to finish */
3790 wait_event(fs_info->transaction_wait,
3791 (atomic_read(&fs_info->defrag_running) == 0));
3792
3793 /* clear out the rbtree of defraggable inodes */
3794 btrfs_cleanup_defrag_inodes(fs_info);
3795
3796 cancel_work_sync(&fs_info->async_reclaim_work);
3797
3798 if (!sb_rdonly(fs_info->sb)) {
3799 /*
3800 * If the cleaner thread is stopped and there are
3801 * block groups queued for removal, the deletion will be
3802 * skipped when we quit the cleaner thread.
3803 */
3804 btrfs_delete_unused_bgs(fs_info);
3805
3806 ret = btrfs_commit_super(fs_info);
3807 if (ret)
3808 btrfs_err(fs_info, "commit super ret %d", ret);
3809 }
3810
3811 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3812 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3813 btrfs_error_commit_super(fs_info);
3814
3815 kthread_stop(fs_info->transaction_kthread);
3816 kthread_stop(fs_info->cleaner_kthread);
3817
3818 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3819
3820 btrfs_free_qgroup_config(fs_info);
3821 ASSERT(list_empty(&fs_info->delalloc_roots));
3822
3823 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3824 btrfs_info(fs_info, "at unmount delalloc count %lld",
3825 percpu_counter_sum(&fs_info->delalloc_bytes));
3826 }
3827
3828 btrfs_sysfs_remove_mounted(fs_info);
3829 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3830
3831 btrfs_free_fs_roots(fs_info);
3832
3833 btrfs_put_block_group_cache(fs_info);
3834
3835 /*
3836 * we must make sure there is not any read request to
3837 * submit after we stopping all workers.
3838 */
3839 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3840 btrfs_stop_all_workers(fs_info);
3841
3842 btrfs_free_block_groups(fs_info);
3843
3844 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3845 free_root_pointers(fs_info, 1);
3846
3847 iput(fs_info->btree_inode);
3848
3849#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3850 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3851 btrfsic_unmount(fs_info->fs_devices);
3852#endif
3853
3854 btrfs_close_devices(fs_info->fs_devices);
3855 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3856
3857 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3858 percpu_counter_destroy(&fs_info->delalloc_bytes);
3859 percpu_counter_destroy(&fs_info->bio_counter);
3860 cleanup_srcu_struct(&fs_info->subvol_srcu);
3861
3862 btrfs_free_stripe_hash_table(fs_info);
3863 btrfs_free_ref_cache(fs_info);
3864
3865 __btrfs_free_block_rsv(root->orphan_block_rsv);
3866 root->orphan_block_rsv = NULL;
3867
3868 while (!list_empty(&fs_info->pinned_chunks)) {
3869 struct extent_map *em;
3870
3871 em = list_first_entry(&fs_info->pinned_chunks,
3872 struct extent_map, list);
3873 list_del_init(&em->list);
3874 free_extent_map(em);
3875 }
3876}
3877
3878int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3879 int atomic)
3880{
3881 int ret;
3882 struct inode *btree_inode = buf->pages[0]->mapping->host;
3883
3884 ret = extent_buffer_uptodate(buf);
3885 if (!ret)
3886 return ret;
3887
3888 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3889 parent_transid, atomic);
3890 if (ret == -EAGAIN)
3891 return ret;
3892 return !ret;
3893}
3894
3895void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3896{
3897 struct btrfs_fs_info *fs_info;
3898 struct btrfs_root *root;
3899 u64 transid = btrfs_header_generation(buf);
3900 int was_dirty;
3901
3902#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3903 /*
3904 * This is a fast path so only do this check if we have sanity tests
3905 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3906 * outside of the sanity tests.
3907 */
3908 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3909 return;
3910#endif
3911 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3912 fs_info = root->fs_info;
3913 btrfs_assert_tree_locked(buf);
3914 if (transid != fs_info->generation)
3915 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3916 buf->start, transid, fs_info->generation);
3917 was_dirty = set_extent_buffer_dirty(buf);
3918 if (!was_dirty)
3919 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3920 buf->len,
3921 fs_info->dirty_metadata_batch);
3922#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3923 /*
3924 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3925 * but item data not updated.
3926 * So here we should only check item pointers, not item data.
3927 */
3928 if (btrfs_header_level(buf) == 0 &&
3929 btrfs_check_leaf_relaxed(fs_info, buf)) {
3930 btrfs_print_leaf(buf);
3931 ASSERT(0);
3932 }
3933#endif
3934}
3935
3936static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
3937 int flush_delayed)
3938{
3939 /*
3940 * looks as though older kernels can get into trouble with
3941 * this code, they end up stuck in balance_dirty_pages forever
3942 */
3943 int ret;
3944
3945 if (current->flags & PF_MEMALLOC)
3946 return;
3947
3948 if (flush_delayed)
3949 btrfs_balance_delayed_items(fs_info);
3950
3951 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
3952 BTRFS_DIRTY_METADATA_THRESH);
3953 if (ret > 0) {
3954 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
3955 }
3956}
3957
3958void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
3959{
3960 __btrfs_btree_balance_dirty(fs_info, 1);
3961}
3962
3963void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
3964{
3965 __btrfs_btree_balance_dirty(fs_info, 0);
3966}
3967
3968int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
3969 struct btrfs_key *first_key)
3970{
3971 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3972 struct btrfs_fs_info *fs_info = root->fs_info;
3973
3974 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
3975 level, first_key);
3976}
3977
3978static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
3979{
3980 struct btrfs_super_block *sb = fs_info->super_copy;
3981 u64 nodesize = btrfs_super_nodesize(sb);
3982 u64 sectorsize = btrfs_super_sectorsize(sb);
3983 int ret = 0;
3984
3985 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
3986 btrfs_err(fs_info, "no valid FS found");
3987 ret = -EINVAL;
3988 }
3989 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
3990 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
3991 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
3992 ret = -EINVAL;
3993 }
3994 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3995 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
3996 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3997 ret = -EINVAL;
3998 }
3999 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4000 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4001 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4002 ret = -EINVAL;
4003 }
4004 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4005 btrfs_err(fs_info, "log_root level too big: %d >= %d",
4006 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4007 ret = -EINVAL;
4008 }
4009
4010 /*
4011 * Check sectorsize and nodesize first, other check will need it.
4012 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4013 */
4014 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4015 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4016 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4017 ret = -EINVAL;
4018 }
4019 /* Only PAGE SIZE is supported yet */
4020 if (sectorsize != PAGE_SIZE) {
4021 btrfs_err(fs_info,
4022 "sectorsize %llu not supported yet, only support %lu",
4023 sectorsize, PAGE_SIZE);
4024 ret = -EINVAL;
4025 }
4026 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4027 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4028 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4029 ret = -EINVAL;
4030 }
4031 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4032 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4033 le32_to_cpu(sb->__unused_leafsize), nodesize);
4034 ret = -EINVAL;
4035 }
4036
4037 /* Root alignment check */
4038 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4039 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4040 btrfs_super_root(sb));
4041 ret = -EINVAL;
4042 }
4043 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4044 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4045 btrfs_super_chunk_root(sb));
4046 ret = -EINVAL;
4047 }
4048 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4049 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4050 btrfs_super_log_root(sb));
4051 ret = -EINVAL;
4052 }
4053
4054 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
4055 btrfs_err(fs_info,
4056 "dev_item UUID does not match fsid: %pU != %pU",
4057 fs_info->fsid, sb->dev_item.fsid);
4058 ret = -EINVAL;
4059 }
4060
4061 /*
4062 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4063 * done later
4064 */
4065 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4066 btrfs_err(fs_info, "bytes_used is too small %llu",
4067 btrfs_super_bytes_used(sb));
4068 ret = -EINVAL;
4069 }
4070 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4071 btrfs_err(fs_info, "invalid stripesize %u",
4072 btrfs_super_stripesize(sb));
4073 ret = -EINVAL;
4074 }
4075 if (btrfs_super_num_devices(sb) > (1UL << 31))
4076 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4077 btrfs_super_num_devices(sb));
4078 if (btrfs_super_num_devices(sb) == 0) {
4079 btrfs_err(fs_info, "number of devices is 0");
4080 ret = -EINVAL;
4081 }
4082
4083 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4084 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4085 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4086 ret = -EINVAL;
4087 }
4088
4089 /*
4090 * Obvious sys_chunk_array corruptions, it must hold at least one key
4091 * and one chunk
4092 */
4093 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4094 btrfs_err(fs_info, "system chunk array too big %u > %u",
4095 btrfs_super_sys_array_size(sb),
4096 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4097 ret = -EINVAL;
4098 }
4099 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4100 + sizeof(struct btrfs_chunk)) {
4101 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4102 btrfs_super_sys_array_size(sb),
4103 sizeof(struct btrfs_disk_key)
4104 + sizeof(struct btrfs_chunk));
4105 ret = -EINVAL;
4106 }
4107
4108 /*
4109 * The generation is a global counter, we'll trust it more than the others
4110 * but it's still possible that it's the one that's wrong.
4111 */
4112 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4113 btrfs_warn(fs_info,
4114 "suspicious: generation < chunk_root_generation: %llu < %llu",
4115 btrfs_super_generation(sb),
4116 btrfs_super_chunk_root_generation(sb));
4117 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4118 && btrfs_super_cache_generation(sb) != (u64)-1)
4119 btrfs_warn(fs_info,
4120 "suspicious: generation < cache_generation: %llu < %llu",
4121 btrfs_super_generation(sb),
4122 btrfs_super_cache_generation(sb));
4123
4124 return ret;
4125}
4126
4127static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4128{
4129 /* cleanup FS via transaction */
4130 btrfs_cleanup_transaction(fs_info);
4131
4132 mutex_lock(&fs_info->cleaner_mutex);
4133 btrfs_run_delayed_iputs(fs_info);
4134 mutex_unlock(&fs_info->cleaner_mutex);
4135
4136 down_write(&fs_info->cleanup_work_sem);
4137 up_write(&fs_info->cleanup_work_sem);
4138}
4139
4140static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4141{
4142 struct btrfs_ordered_extent *ordered;
4143
4144 spin_lock(&root->ordered_extent_lock);
4145 /*
4146 * This will just short circuit the ordered completion stuff which will
4147 * make sure the ordered extent gets properly cleaned up.
4148 */
4149 list_for_each_entry(ordered, &root->ordered_extents,
4150 root_extent_list)
4151 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4152 spin_unlock(&root->ordered_extent_lock);
4153}
4154
4155static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4156{
4157 struct btrfs_root *root;
4158 struct list_head splice;
4159
4160 INIT_LIST_HEAD(&splice);
4161
4162 spin_lock(&fs_info->ordered_root_lock);
4163 list_splice_init(&fs_info->ordered_roots, &splice);
4164 while (!list_empty(&splice)) {
4165 root = list_first_entry(&splice, struct btrfs_root,
4166 ordered_root);
4167 list_move_tail(&root->ordered_root,
4168 &fs_info->ordered_roots);
4169
4170 spin_unlock(&fs_info->ordered_root_lock);
4171 btrfs_destroy_ordered_extents(root);
4172
4173 cond_resched();
4174 spin_lock(&fs_info->ordered_root_lock);
4175 }
4176 spin_unlock(&fs_info->ordered_root_lock);
4177}
4178
4179static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4180 struct btrfs_fs_info *fs_info)
4181{
4182 struct rb_node *node;
4183 struct btrfs_delayed_ref_root *delayed_refs;
4184 struct btrfs_delayed_ref_node *ref;
4185 int ret = 0;
4186
4187 delayed_refs = &trans->delayed_refs;
4188
4189 spin_lock(&delayed_refs->lock);
4190 if (atomic_read(&delayed_refs->num_entries) == 0) {
4191 spin_unlock(&delayed_refs->lock);
4192 btrfs_info(fs_info, "delayed_refs has NO entry");
4193 return ret;
4194 }
4195
4196 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4197 struct btrfs_delayed_ref_head *head;
4198 struct rb_node *n;
4199 bool pin_bytes = false;
4200
4201 head = rb_entry(node, struct btrfs_delayed_ref_head,
4202 href_node);
4203 if (!mutex_trylock(&head->mutex)) {
4204 refcount_inc(&head->refs);
4205 spin_unlock(&delayed_refs->lock);
4206
4207 mutex_lock(&head->mutex);
4208 mutex_unlock(&head->mutex);
4209 btrfs_put_delayed_ref_head(head);
4210 spin_lock(&delayed_refs->lock);
4211 continue;
4212 }
4213 spin_lock(&head->lock);
4214 while ((n = rb_first(&head->ref_tree)) != NULL) {
4215 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4216 ref_node);
4217 ref->in_tree = 0;
4218 rb_erase(&ref->ref_node, &head->ref_tree);
4219 RB_CLEAR_NODE(&ref->ref_node);
4220 if (!list_empty(&ref->add_list))
4221 list_del(&ref->add_list);
4222 atomic_dec(&delayed_refs->num_entries);
4223 btrfs_put_delayed_ref(ref);
4224 }
4225 if (head->must_insert_reserved)
4226 pin_bytes = true;
4227 btrfs_free_delayed_extent_op(head->extent_op);
4228 delayed_refs->num_heads--;
4229 if (head->processing == 0)
4230 delayed_refs->num_heads_ready--;
4231 atomic_dec(&delayed_refs->num_entries);
4232 rb_erase(&head->href_node, &delayed_refs->href_root);
4233 RB_CLEAR_NODE(&head->href_node);
4234 spin_unlock(&head->lock);
4235 spin_unlock(&delayed_refs->lock);
4236 mutex_unlock(&head->mutex);
4237
4238 if (pin_bytes)
4239 btrfs_pin_extent(fs_info, head->bytenr,
4240 head->num_bytes, 1);
4241 btrfs_put_delayed_ref_head(head);
4242 cond_resched();
4243 spin_lock(&delayed_refs->lock);
4244 }
4245
4246 spin_unlock(&delayed_refs->lock);
4247
4248 return ret;
4249}
4250
4251static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4252{
4253 struct btrfs_inode *btrfs_inode;
4254 struct list_head splice;
4255
4256 INIT_LIST_HEAD(&splice);
4257
4258 spin_lock(&root->delalloc_lock);
4259 list_splice_init(&root->delalloc_inodes, &splice);
4260
4261 while (!list_empty(&splice)) {
4262 struct inode *inode = NULL;
4263 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4264 delalloc_inodes);
4265 __btrfs_del_delalloc_inode(root, btrfs_inode);
4266 spin_unlock(&root->delalloc_lock);
4267
4268 /*
4269 * Make sure we get a live inode and that it'll not disappear
4270 * meanwhile.
4271 */
4272 inode = igrab(&btrfs_inode->vfs_inode);
4273 if (inode) {
4274 invalidate_inode_pages2(inode->i_mapping);
4275 iput(inode);
4276 }
4277 spin_lock(&root->delalloc_lock);
4278 }
4279 spin_unlock(&root->delalloc_lock);
4280}
4281
4282static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4283{
4284 struct btrfs_root *root;
4285 struct list_head splice;
4286
4287 INIT_LIST_HEAD(&splice);
4288
4289 spin_lock(&fs_info->delalloc_root_lock);
4290 list_splice_init(&fs_info->delalloc_roots, &splice);
4291 while (!list_empty(&splice)) {
4292 root = list_first_entry(&splice, struct btrfs_root,
4293 delalloc_root);
4294 root = btrfs_grab_fs_root(root);
4295 BUG_ON(!root);
4296 spin_unlock(&fs_info->delalloc_root_lock);
4297
4298 btrfs_destroy_delalloc_inodes(root);
4299 btrfs_put_fs_root(root);
4300
4301 spin_lock(&fs_info->delalloc_root_lock);
4302 }
4303 spin_unlock(&fs_info->delalloc_root_lock);
4304}
4305
4306static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4307 struct extent_io_tree *dirty_pages,
4308 int mark)
4309{
4310 int ret;
4311 struct extent_buffer *eb;
4312 u64 start = 0;
4313 u64 end;
4314
4315 while (1) {
4316 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4317 mark, NULL);
4318 if (ret)
4319 break;
4320
4321 clear_extent_bits(dirty_pages, start, end, mark);
4322 while (start <= end) {
4323 eb = find_extent_buffer(fs_info, start);
4324 start += fs_info->nodesize;
4325 if (!eb)
4326 continue;
4327 wait_on_extent_buffer_writeback(eb);
4328
4329 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4330 &eb->bflags))
4331 clear_extent_buffer_dirty(eb);
4332 free_extent_buffer_stale(eb);
4333 }
4334 }
4335
4336 return ret;
4337}
4338
4339static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4340 struct extent_io_tree *pinned_extents)
4341{
4342 struct extent_io_tree *unpin;
4343 u64 start;
4344 u64 end;
4345 int ret;
4346 bool loop = true;
4347
4348 unpin = pinned_extents;
4349again:
4350 while (1) {
4351 ret = find_first_extent_bit(unpin, 0, &start, &end,
4352 EXTENT_DIRTY, NULL);
4353 if (ret)
4354 break;
4355
4356 clear_extent_dirty(unpin, start, end);
4357 btrfs_error_unpin_extent_range(fs_info, start, end);
4358 cond_resched();
4359 }
4360
4361 if (loop) {
4362 if (unpin == &fs_info->freed_extents[0])
4363 unpin = &fs_info->freed_extents[1];
4364 else
4365 unpin = &fs_info->freed_extents[0];
4366 loop = false;
4367 goto again;
4368 }
4369
4370 return 0;
4371}
4372
4373static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4374{
4375 struct inode *inode;
4376
4377 inode = cache->io_ctl.inode;
4378 if (inode) {
4379 invalidate_inode_pages2(inode->i_mapping);
4380 BTRFS_I(inode)->generation = 0;
4381 cache->io_ctl.inode = NULL;
4382 iput(inode);
4383 }
4384 btrfs_put_block_group(cache);
4385}
4386
4387void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4388 struct btrfs_fs_info *fs_info)
4389{
4390 struct btrfs_block_group_cache *cache;
4391
4392 spin_lock(&cur_trans->dirty_bgs_lock);
4393 while (!list_empty(&cur_trans->dirty_bgs)) {
4394 cache = list_first_entry(&cur_trans->dirty_bgs,
4395 struct btrfs_block_group_cache,
4396 dirty_list);
4397
4398 if (!list_empty(&cache->io_list)) {
4399 spin_unlock(&cur_trans->dirty_bgs_lock);
4400 list_del_init(&cache->io_list);
4401 btrfs_cleanup_bg_io(cache);
4402 spin_lock(&cur_trans->dirty_bgs_lock);
4403 }
4404
4405 list_del_init(&cache->dirty_list);
4406 spin_lock(&cache->lock);
4407 cache->disk_cache_state = BTRFS_DC_ERROR;
4408 spin_unlock(&cache->lock);
4409
4410 spin_unlock(&cur_trans->dirty_bgs_lock);
4411 btrfs_put_block_group(cache);
4412 spin_lock(&cur_trans->dirty_bgs_lock);
4413 }
4414 spin_unlock(&cur_trans->dirty_bgs_lock);
4415
4416 /*
4417 * Refer to the definition of io_bgs member for details why it's safe
4418 * to use it without any locking
4419 */
4420 while (!list_empty(&cur_trans->io_bgs)) {
4421 cache = list_first_entry(&cur_trans->io_bgs,
4422 struct btrfs_block_group_cache,
4423 io_list);
4424
4425 list_del_init(&cache->io_list);
4426 spin_lock(&cache->lock);
4427 cache->disk_cache_state = BTRFS_DC_ERROR;
4428 spin_unlock(&cache->lock);
4429 btrfs_cleanup_bg_io(cache);
4430 }
4431}
4432
4433void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4434 struct btrfs_fs_info *fs_info)
4435{
4436 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4437 ASSERT(list_empty(&cur_trans->dirty_bgs));
4438 ASSERT(list_empty(&cur_trans->io_bgs));
4439
4440 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4441
4442 cur_trans->state = TRANS_STATE_COMMIT_START;
4443 wake_up(&fs_info->transaction_blocked_wait);
4444
4445 cur_trans->state = TRANS_STATE_UNBLOCKED;
4446 wake_up(&fs_info->transaction_wait);
4447
4448 btrfs_destroy_delayed_inodes(fs_info);
4449 btrfs_assert_delayed_root_empty(fs_info);
4450
4451 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4452 EXTENT_DIRTY);
4453 btrfs_destroy_pinned_extent(fs_info,
4454 fs_info->pinned_extents);
4455
4456 cur_trans->state =TRANS_STATE_COMPLETED;
4457 wake_up(&cur_trans->commit_wait);
4458}
4459
4460static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4461{
4462 struct btrfs_transaction *t;
4463
4464 mutex_lock(&fs_info->transaction_kthread_mutex);
4465
4466 spin_lock(&fs_info->trans_lock);
4467 while (!list_empty(&fs_info->trans_list)) {
4468 t = list_first_entry(&fs_info->trans_list,
4469 struct btrfs_transaction, list);
4470 if (t->state >= TRANS_STATE_COMMIT_START) {
4471 refcount_inc(&t->use_count);
4472 spin_unlock(&fs_info->trans_lock);
4473 btrfs_wait_for_commit(fs_info, t->transid);
4474 btrfs_put_transaction(t);
4475 spin_lock(&fs_info->trans_lock);
4476 continue;
4477 }
4478 if (t == fs_info->running_transaction) {
4479 t->state = TRANS_STATE_COMMIT_DOING;
4480 spin_unlock(&fs_info->trans_lock);
4481 /*
4482 * We wait for 0 num_writers since we don't hold a trans
4483 * handle open currently for this transaction.
4484 */
4485 wait_event(t->writer_wait,
4486 atomic_read(&t->num_writers) == 0);
4487 } else {
4488 spin_unlock(&fs_info->trans_lock);
4489 }
4490 btrfs_cleanup_one_transaction(t, fs_info);
4491
4492 spin_lock(&fs_info->trans_lock);
4493 if (t == fs_info->running_transaction)
4494 fs_info->running_transaction = NULL;
4495 list_del_init(&t->list);
4496 spin_unlock(&fs_info->trans_lock);
4497
4498 btrfs_put_transaction(t);
4499 trace_btrfs_transaction_commit(fs_info->tree_root);
4500 spin_lock(&fs_info->trans_lock);
4501 }
4502 spin_unlock(&fs_info->trans_lock);
4503 btrfs_destroy_all_ordered_extents(fs_info);
4504 btrfs_destroy_delayed_inodes(fs_info);
4505 btrfs_assert_delayed_root_empty(fs_info);
4506 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4507 btrfs_destroy_all_delalloc_inodes(fs_info);
4508 mutex_unlock(&fs_info->transaction_kthread_mutex);
4509
4510 return 0;
4511}
4512
4513static struct btrfs_fs_info *btree_fs_info(void *private_data)
4514{
4515 struct inode *inode = private_data;
4516 return btrfs_sb(inode->i_sb);
4517}
4518
4519static const struct extent_io_ops btree_extent_io_ops = {
4520 /* mandatory callbacks */
4521 .submit_bio_hook = btree_submit_bio_hook,
4522 .readpage_end_io_hook = btree_readpage_end_io_hook,
4523 /* note we're sharing with inode.c for the merge bio hook */
4524 .merge_bio_hook = btrfs_merge_bio_hook,
4525 .readpage_io_failed_hook = btree_io_failed_hook,
4526 .set_range_writeback = btrfs_set_range_writeback,
4527 .tree_fs_info = btree_fs_info,
4528
4529 /* optional callbacks */
4530};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/fs.h>
7#include <linux/blkdev.h>
8#include <linux/radix-tree.h>
9#include <linux/writeback.h>
10#include <linux/workqueue.h>
11#include <linux/kthread.h>
12#include <linux/slab.h>
13#include <linux/migrate.h>
14#include <linux/ratelimit.h>
15#include <linux/uuid.h>
16#include <linux/semaphore.h>
17#include <linux/error-injection.h>
18#include <linux/crc32c.h>
19#include <linux/sched/mm.h>
20#include <asm/unaligned.h>
21#include <crypto/hash.h>
22#include "ctree.h"
23#include "disk-io.h"
24#include "transaction.h"
25#include "btrfs_inode.h"
26#include "volumes.h"
27#include "print-tree.h"
28#include "locking.h"
29#include "tree-log.h"
30#include "free-space-cache.h"
31#include "free-space-tree.h"
32#include "check-integrity.h"
33#include "rcu-string.h"
34#include "dev-replace.h"
35#include "raid56.h"
36#include "sysfs.h"
37#include "qgroup.h"
38#include "compression.h"
39#include "tree-checker.h"
40#include "ref-verify.h"
41#include "block-group.h"
42#include "discard.h"
43#include "space-info.h"
44#include "zoned.h"
45#include "subpage.h"
46
47#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
48 BTRFS_HEADER_FLAG_RELOC |\
49 BTRFS_SUPER_FLAG_ERROR |\
50 BTRFS_SUPER_FLAG_SEEDING |\
51 BTRFS_SUPER_FLAG_METADUMP |\
52 BTRFS_SUPER_FLAG_METADUMP_V2)
53
54static void end_workqueue_fn(struct btrfs_work *work);
55static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_fs_info *fs_info);
58static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 struct extent_io_tree *dirty_pages,
61 int mark);
62static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *pinned_extents);
64static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66
67/*
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
71 */
72struct btrfs_end_io_wq {
73 struct bio *bio;
74 bio_end_io_t *end_io;
75 void *private;
76 struct btrfs_fs_info *info;
77 blk_status_t status;
78 enum btrfs_wq_endio_type metadata;
79 struct btrfs_work work;
80};
81
82static struct kmem_cache *btrfs_end_io_wq_cache;
83
84int __init btrfs_end_io_wq_init(void)
85{
86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq),
88 0,
89 SLAB_MEM_SPREAD,
90 NULL);
91 if (!btrfs_end_io_wq_cache)
92 return -ENOMEM;
93 return 0;
94}
95
96void __cold btrfs_end_io_wq_exit(void)
97{
98 kmem_cache_destroy(btrfs_end_io_wq_cache);
99}
100
101static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102{
103 if (fs_info->csum_shash)
104 crypto_free_shash(fs_info->csum_shash);
105}
106
107/*
108 * async submit bios are used to offload expensive checksumming
109 * onto the worker threads. They checksum file and metadata bios
110 * just before they are sent down the IO stack.
111 */
112struct async_submit_bio {
113 struct inode *inode;
114 struct bio *bio;
115 extent_submit_bio_start_t *submit_bio_start;
116 int mirror_num;
117
118 /* Optional parameter for submit_bio_start used by direct io */
119 u64 dio_file_offset;
120 struct btrfs_work work;
121 blk_status_t status;
122};
123
124/*
125 * Lockdep class keys for extent_buffer->lock's in this root. For a given
126 * eb, the lockdep key is determined by the btrfs_root it belongs to and
127 * the level the eb occupies in the tree.
128 *
129 * Different roots are used for different purposes and may nest inside each
130 * other and they require separate keysets. As lockdep keys should be
131 * static, assign keysets according to the purpose of the root as indicated
132 * by btrfs_root->root_key.objectid. This ensures that all special purpose
133 * roots have separate keysets.
134 *
135 * Lock-nesting across peer nodes is always done with the immediate parent
136 * node locked thus preventing deadlock. As lockdep doesn't know this, use
137 * subclass to avoid triggering lockdep warning in such cases.
138 *
139 * The key is set by the readpage_end_io_hook after the buffer has passed
140 * csum validation but before the pages are unlocked. It is also set by
141 * btrfs_init_new_buffer on freshly allocated blocks.
142 *
143 * We also add a check to make sure the highest level of the tree is the
144 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
145 * needs update as well.
146 */
147#ifdef CONFIG_DEBUG_LOCK_ALLOC
148# if BTRFS_MAX_LEVEL != 8
149# error
150# endif
151
152#define DEFINE_LEVEL(stem, level) \
153 .names[level] = "btrfs-" stem "-0" #level,
154
155#define DEFINE_NAME(stem) \
156 DEFINE_LEVEL(stem, 0) \
157 DEFINE_LEVEL(stem, 1) \
158 DEFINE_LEVEL(stem, 2) \
159 DEFINE_LEVEL(stem, 3) \
160 DEFINE_LEVEL(stem, 4) \
161 DEFINE_LEVEL(stem, 5) \
162 DEFINE_LEVEL(stem, 6) \
163 DEFINE_LEVEL(stem, 7)
164
165static struct btrfs_lockdep_keyset {
166 u64 id; /* root objectid */
167 /* Longest entry: btrfs-free-space-00 */
168 char names[BTRFS_MAX_LEVEL][20];
169 struct lock_class_key keys[BTRFS_MAX_LEVEL];
170} btrfs_lockdep_keysets[] = {
171 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") },
172 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") },
173 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") },
174 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") },
175 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
176 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
177 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
178 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
179 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
180 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
181 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
182 { .id = 0, DEFINE_NAME("tree") },
183};
184
185#undef DEFINE_LEVEL
186#undef DEFINE_NAME
187
188void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
189 int level)
190{
191 struct btrfs_lockdep_keyset *ks;
192
193 BUG_ON(level >= ARRAY_SIZE(ks->keys));
194
195 /* find the matching keyset, id 0 is the default entry */
196 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
197 if (ks->id == objectid)
198 break;
199
200 lockdep_set_class_and_name(&eb->lock,
201 &ks->keys[level], ks->names[level]);
202}
203
204#endif
205
206/*
207 * Compute the csum of a btree block and store the result to provided buffer.
208 */
209static void csum_tree_block(struct extent_buffer *buf, u8 *result)
210{
211 struct btrfs_fs_info *fs_info = buf->fs_info;
212 const int num_pages = num_extent_pages(buf);
213 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
214 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
215 char *kaddr;
216 int i;
217
218 shash->tfm = fs_info->csum_shash;
219 crypto_shash_init(shash);
220 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
221 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
222 first_page_part - BTRFS_CSUM_SIZE);
223
224 for (i = 1; i < num_pages; i++) {
225 kaddr = page_address(buf->pages[i]);
226 crypto_shash_update(shash, kaddr, PAGE_SIZE);
227 }
228 memset(result, 0, BTRFS_CSUM_SIZE);
229 crypto_shash_final(shash, result);
230}
231
232/*
233 * we can't consider a given block up to date unless the transid of the
234 * block matches the transid in the parent node's pointer. This is how we
235 * detect blocks that either didn't get written at all or got written
236 * in the wrong place.
237 */
238static int verify_parent_transid(struct extent_io_tree *io_tree,
239 struct extent_buffer *eb, u64 parent_transid,
240 int atomic)
241{
242 struct extent_state *cached_state = NULL;
243 int ret;
244
245 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
246 return 0;
247
248 if (atomic)
249 return -EAGAIN;
250
251 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
252 &cached_state);
253 if (extent_buffer_uptodate(eb) &&
254 btrfs_header_generation(eb) == parent_transid) {
255 ret = 0;
256 goto out;
257 }
258 btrfs_err_rl(eb->fs_info,
259 "parent transid verify failed on %llu wanted %llu found %llu",
260 eb->start,
261 parent_transid, btrfs_header_generation(eb));
262 ret = 1;
263 clear_extent_buffer_uptodate(eb);
264out:
265 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
266 &cached_state);
267 return ret;
268}
269
270static bool btrfs_supported_super_csum(u16 csum_type)
271{
272 switch (csum_type) {
273 case BTRFS_CSUM_TYPE_CRC32:
274 case BTRFS_CSUM_TYPE_XXHASH:
275 case BTRFS_CSUM_TYPE_SHA256:
276 case BTRFS_CSUM_TYPE_BLAKE2:
277 return true;
278 default:
279 return false;
280 }
281}
282
283/*
284 * Return 0 if the superblock checksum type matches the checksum value of that
285 * algorithm. Pass the raw disk superblock data.
286 */
287static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
288 char *raw_disk_sb)
289{
290 struct btrfs_super_block *disk_sb =
291 (struct btrfs_super_block *)raw_disk_sb;
292 char result[BTRFS_CSUM_SIZE];
293 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
294
295 shash->tfm = fs_info->csum_shash;
296
297 /*
298 * The super_block structure does not span the whole
299 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
300 * filled with zeros and is included in the checksum.
301 */
302 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
303 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
304
305 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
306 return 1;
307
308 return 0;
309}
310
311int btrfs_verify_level_key(struct extent_buffer *eb, int level,
312 struct btrfs_key *first_key, u64 parent_transid)
313{
314 struct btrfs_fs_info *fs_info = eb->fs_info;
315 int found_level;
316 struct btrfs_key found_key;
317 int ret;
318
319 found_level = btrfs_header_level(eb);
320 if (found_level != level) {
321 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
322 KERN_ERR "BTRFS: tree level check failed\n");
323 btrfs_err(fs_info,
324"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
325 eb->start, level, found_level);
326 return -EIO;
327 }
328
329 if (!first_key)
330 return 0;
331
332 /*
333 * For live tree block (new tree blocks in current transaction),
334 * we need proper lock context to avoid race, which is impossible here.
335 * So we only checks tree blocks which is read from disk, whose
336 * generation <= fs_info->last_trans_committed.
337 */
338 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
339 return 0;
340
341 /* We have @first_key, so this @eb must have at least one item */
342 if (btrfs_header_nritems(eb) == 0) {
343 btrfs_err(fs_info,
344 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
345 eb->start);
346 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
347 return -EUCLEAN;
348 }
349
350 if (found_level)
351 btrfs_node_key_to_cpu(eb, &found_key, 0);
352 else
353 btrfs_item_key_to_cpu(eb, &found_key, 0);
354 ret = btrfs_comp_cpu_keys(first_key, &found_key);
355
356 if (ret) {
357 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
358 KERN_ERR "BTRFS: tree first key check failed\n");
359 btrfs_err(fs_info,
360"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
361 eb->start, parent_transid, first_key->objectid,
362 first_key->type, first_key->offset,
363 found_key.objectid, found_key.type,
364 found_key.offset);
365 }
366 return ret;
367}
368
369/*
370 * helper to read a given tree block, doing retries as required when
371 * the checksums don't match and we have alternate mirrors to try.
372 *
373 * @parent_transid: expected transid, skip check if 0
374 * @level: expected level, mandatory check
375 * @first_key: expected key of first slot, skip check if NULL
376 */
377static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
378 u64 parent_transid, int level,
379 struct btrfs_key *first_key)
380{
381 struct btrfs_fs_info *fs_info = eb->fs_info;
382 struct extent_io_tree *io_tree;
383 int failed = 0;
384 int ret;
385 int num_copies = 0;
386 int mirror_num = 0;
387 int failed_mirror = 0;
388
389 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
390 while (1) {
391 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
392 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
393 if (!ret) {
394 if (verify_parent_transid(io_tree, eb,
395 parent_transid, 0))
396 ret = -EIO;
397 else if (btrfs_verify_level_key(eb, level,
398 first_key, parent_transid))
399 ret = -EUCLEAN;
400 else
401 break;
402 }
403
404 num_copies = btrfs_num_copies(fs_info,
405 eb->start, eb->len);
406 if (num_copies == 1)
407 break;
408
409 if (!failed_mirror) {
410 failed = 1;
411 failed_mirror = eb->read_mirror;
412 }
413
414 mirror_num++;
415 if (mirror_num == failed_mirror)
416 mirror_num++;
417
418 if (mirror_num > num_copies)
419 break;
420 }
421
422 if (failed && !ret && failed_mirror)
423 btrfs_repair_eb_io_failure(eb, failed_mirror);
424
425 return ret;
426}
427
428static int csum_one_extent_buffer(struct extent_buffer *eb)
429{
430 struct btrfs_fs_info *fs_info = eb->fs_info;
431 u8 result[BTRFS_CSUM_SIZE];
432 int ret;
433
434 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
435 offsetof(struct btrfs_header, fsid),
436 BTRFS_FSID_SIZE) == 0);
437 csum_tree_block(eb, result);
438
439 if (btrfs_header_level(eb))
440 ret = btrfs_check_node(eb);
441 else
442 ret = btrfs_check_leaf_full(eb);
443
444 if (ret < 0) {
445 btrfs_print_tree(eb, 0);
446 btrfs_err(fs_info,
447 "block=%llu write time tree block corruption detected",
448 eb->start);
449 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
450 return ret;
451 }
452 write_extent_buffer(eb, result, 0, fs_info->csum_size);
453
454 return 0;
455}
456
457/* Checksum all dirty extent buffers in one bio_vec */
458static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
459 struct bio_vec *bvec)
460{
461 struct page *page = bvec->bv_page;
462 u64 bvec_start = page_offset(page) + bvec->bv_offset;
463 u64 cur;
464 int ret = 0;
465
466 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
467 cur += fs_info->nodesize) {
468 struct extent_buffer *eb;
469 bool uptodate;
470
471 eb = find_extent_buffer(fs_info, cur);
472 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
473 fs_info->nodesize);
474
475 /* A dirty eb shouldn't disappear from buffer_radix */
476 if (WARN_ON(!eb))
477 return -EUCLEAN;
478
479 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
480 free_extent_buffer(eb);
481 return -EUCLEAN;
482 }
483 if (WARN_ON(!uptodate)) {
484 free_extent_buffer(eb);
485 return -EUCLEAN;
486 }
487
488 ret = csum_one_extent_buffer(eb);
489 free_extent_buffer(eb);
490 if (ret < 0)
491 return ret;
492 }
493 return ret;
494}
495
496/*
497 * Checksum a dirty tree block before IO. This has extra checks to make sure
498 * we only fill in the checksum field in the first page of a multi-page block.
499 * For subpage extent buffers we need bvec to also read the offset in the page.
500 */
501static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
502{
503 struct page *page = bvec->bv_page;
504 u64 start = page_offset(page);
505 u64 found_start;
506 struct extent_buffer *eb;
507
508 if (fs_info->sectorsize < PAGE_SIZE)
509 return csum_dirty_subpage_buffers(fs_info, bvec);
510
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
513 return 0;
514
515 found_start = btrfs_header_bytenr(eb);
516
517 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
518 WARN_ON(found_start != 0);
519 return 0;
520 }
521
522 /*
523 * Please do not consolidate these warnings into a single if.
524 * It is useful to know what went wrong.
525 */
526 if (WARN_ON(found_start != start))
527 return -EUCLEAN;
528 if (WARN_ON(!PageUptodate(page)))
529 return -EUCLEAN;
530
531 return csum_one_extent_buffer(eb);
532}
533
534static int check_tree_block_fsid(struct extent_buffer *eb)
535{
536 struct btrfs_fs_info *fs_info = eb->fs_info;
537 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
538 u8 fsid[BTRFS_FSID_SIZE];
539 u8 *metadata_uuid;
540
541 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
542 BTRFS_FSID_SIZE);
543 /*
544 * Checking the incompat flag is only valid for the current fs. For
545 * seed devices it's forbidden to have their uuid changed so reading
546 * ->fsid in this case is fine
547 */
548 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
549 metadata_uuid = fs_devices->metadata_uuid;
550 else
551 metadata_uuid = fs_devices->fsid;
552
553 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
554 return 0;
555
556 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
557 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
558 return 0;
559
560 return 1;
561}
562
563/* Do basic extent buffer checks at read time */
564static int validate_extent_buffer(struct extent_buffer *eb)
565{
566 struct btrfs_fs_info *fs_info = eb->fs_info;
567 u64 found_start;
568 const u32 csum_size = fs_info->csum_size;
569 u8 found_level;
570 u8 result[BTRFS_CSUM_SIZE];
571 const u8 *header_csum;
572 int ret = 0;
573
574 found_start = btrfs_header_bytenr(eb);
575 if (found_start != eb->start) {
576 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
577 eb->start, found_start);
578 ret = -EIO;
579 goto out;
580 }
581 if (check_tree_block_fsid(eb)) {
582 btrfs_err_rl(fs_info, "bad fsid on block %llu",
583 eb->start);
584 ret = -EIO;
585 goto out;
586 }
587 found_level = btrfs_header_level(eb);
588 if (found_level >= BTRFS_MAX_LEVEL) {
589 btrfs_err(fs_info, "bad tree block level %d on %llu",
590 (int)btrfs_header_level(eb), eb->start);
591 ret = -EIO;
592 goto out;
593 }
594
595 csum_tree_block(eb, result);
596 header_csum = page_address(eb->pages[0]) +
597 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
598
599 if (memcmp(result, header_csum, csum_size) != 0) {
600 btrfs_warn_rl(fs_info,
601 "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
602 eb->start,
603 CSUM_FMT_VALUE(csum_size, header_csum),
604 CSUM_FMT_VALUE(csum_size, result),
605 btrfs_header_level(eb));
606 ret = -EUCLEAN;
607 goto out;
608 }
609
610 /*
611 * If this is a leaf block and it is corrupt, set the corrupt bit so
612 * that we don't try and read the other copies of this block, just
613 * return -EIO.
614 */
615 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
616 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
617 ret = -EIO;
618 }
619
620 if (found_level > 0 && btrfs_check_node(eb))
621 ret = -EIO;
622
623 if (!ret)
624 set_extent_buffer_uptodate(eb);
625 else
626 btrfs_err(fs_info,
627 "block=%llu read time tree block corruption detected",
628 eb->start);
629out:
630 return ret;
631}
632
633static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
634 int mirror)
635{
636 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
637 struct extent_buffer *eb;
638 bool reads_done;
639 int ret = 0;
640
641 /*
642 * We don't allow bio merge for subpage metadata read, so we should
643 * only get one eb for each endio hook.
644 */
645 ASSERT(end == start + fs_info->nodesize - 1);
646 ASSERT(PagePrivate(page));
647
648 eb = find_extent_buffer(fs_info, start);
649 /*
650 * When we are reading one tree block, eb must have been inserted into
651 * the radix tree. If not, something is wrong.
652 */
653 ASSERT(eb);
654
655 reads_done = atomic_dec_and_test(&eb->io_pages);
656 /* Subpage read must finish in page read */
657 ASSERT(reads_done);
658
659 eb->read_mirror = mirror;
660 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
661 ret = -EIO;
662 goto err;
663 }
664 ret = validate_extent_buffer(eb);
665 if (ret < 0)
666 goto err;
667
668 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
669 btree_readahead_hook(eb, ret);
670
671 set_extent_buffer_uptodate(eb);
672
673 free_extent_buffer(eb);
674 return ret;
675err:
676 /*
677 * end_bio_extent_readpage decrements io_pages in case of error,
678 * make sure it has something to decrement.
679 */
680 atomic_inc(&eb->io_pages);
681 clear_extent_buffer_uptodate(eb);
682 free_extent_buffer(eb);
683 return ret;
684}
685
686int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
687 struct page *page, u64 start, u64 end,
688 int mirror)
689{
690 struct extent_buffer *eb;
691 int ret = 0;
692 int reads_done;
693
694 ASSERT(page->private);
695
696 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
697 return validate_subpage_buffer(page, start, end, mirror);
698
699 eb = (struct extent_buffer *)page->private;
700
701 /*
702 * The pending IO might have been the only thing that kept this buffer
703 * in memory. Make sure we have a ref for all this other checks
704 */
705 atomic_inc(&eb->refs);
706
707 reads_done = atomic_dec_and_test(&eb->io_pages);
708 if (!reads_done)
709 goto err;
710
711 eb->read_mirror = mirror;
712 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
713 ret = -EIO;
714 goto err;
715 }
716 ret = validate_extent_buffer(eb);
717err:
718 if (reads_done &&
719 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
720 btree_readahead_hook(eb, ret);
721
722 if (ret) {
723 /*
724 * our io error hook is going to dec the io pages
725 * again, we have to make sure it has something
726 * to decrement
727 */
728 atomic_inc(&eb->io_pages);
729 clear_extent_buffer_uptodate(eb);
730 }
731 free_extent_buffer(eb);
732
733 return ret;
734}
735
736static void end_workqueue_bio(struct bio *bio)
737{
738 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
739 struct btrfs_fs_info *fs_info;
740 struct btrfs_workqueue *wq;
741
742 fs_info = end_io_wq->info;
743 end_io_wq->status = bio->bi_status;
744
745 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
746 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
747 wq = fs_info->endio_meta_write_workers;
748 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
749 wq = fs_info->endio_freespace_worker;
750 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
751 wq = fs_info->endio_raid56_workers;
752 else
753 wq = fs_info->endio_write_workers;
754 } else {
755 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
756 wq = fs_info->endio_raid56_workers;
757 else if (end_io_wq->metadata)
758 wq = fs_info->endio_meta_workers;
759 else
760 wq = fs_info->endio_workers;
761 }
762
763 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
764 btrfs_queue_work(wq, &end_io_wq->work);
765}
766
767blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
768 enum btrfs_wq_endio_type metadata)
769{
770 struct btrfs_end_io_wq *end_io_wq;
771
772 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
773 if (!end_io_wq)
774 return BLK_STS_RESOURCE;
775
776 end_io_wq->private = bio->bi_private;
777 end_io_wq->end_io = bio->bi_end_io;
778 end_io_wq->info = info;
779 end_io_wq->status = 0;
780 end_io_wq->bio = bio;
781 end_io_wq->metadata = metadata;
782
783 bio->bi_private = end_io_wq;
784 bio->bi_end_io = end_workqueue_bio;
785 return 0;
786}
787
788static void run_one_async_start(struct btrfs_work *work)
789{
790 struct async_submit_bio *async;
791 blk_status_t ret;
792
793 async = container_of(work, struct async_submit_bio, work);
794 ret = async->submit_bio_start(async->inode, async->bio,
795 async->dio_file_offset);
796 if (ret)
797 async->status = ret;
798}
799
800/*
801 * In order to insert checksums into the metadata in large chunks, we wait
802 * until bio submission time. All the pages in the bio are checksummed and
803 * sums are attached onto the ordered extent record.
804 *
805 * At IO completion time the csums attached on the ordered extent record are
806 * inserted into the tree.
807 */
808static void run_one_async_done(struct btrfs_work *work)
809{
810 struct async_submit_bio *async;
811 struct inode *inode;
812 blk_status_t ret;
813
814 async = container_of(work, struct async_submit_bio, work);
815 inode = async->inode;
816
817 /* If an error occurred we just want to clean up the bio and move on */
818 if (async->status) {
819 async->bio->bi_status = async->status;
820 bio_endio(async->bio);
821 return;
822 }
823
824 /*
825 * All of the bios that pass through here are from async helpers.
826 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
827 * This changes nothing when cgroups aren't in use.
828 */
829 async->bio->bi_opf |= REQ_CGROUP_PUNT;
830 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
831 if (ret) {
832 async->bio->bi_status = ret;
833 bio_endio(async->bio);
834 }
835}
836
837static void run_one_async_free(struct btrfs_work *work)
838{
839 struct async_submit_bio *async;
840
841 async = container_of(work, struct async_submit_bio, work);
842 kfree(async);
843}
844
845blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
846 int mirror_num, unsigned long bio_flags,
847 u64 dio_file_offset,
848 extent_submit_bio_start_t *submit_bio_start)
849{
850 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
851 struct async_submit_bio *async;
852
853 async = kmalloc(sizeof(*async), GFP_NOFS);
854 if (!async)
855 return BLK_STS_RESOURCE;
856
857 async->inode = inode;
858 async->bio = bio;
859 async->mirror_num = mirror_num;
860 async->submit_bio_start = submit_bio_start;
861
862 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
863 run_one_async_free);
864
865 async->dio_file_offset = dio_file_offset;
866
867 async->status = 0;
868
869 if (op_is_sync(bio->bi_opf))
870 btrfs_set_work_high_priority(&async->work);
871
872 btrfs_queue_work(fs_info->workers, &async->work);
873 return 0;
874}
875
876static blk_status_t btree_csum_one_bio(struct bio *bio)
877{
878 struct bio_vec *bvec;
879 struct btrfs_root *root;
880 int ret = 0;
881 struct bvec_iter_all iter_all;
882
883 ASSERT(!bio_flagged(bio, BIO_CLONED));
884 bio_for_each_segment_all(bvec, bio, iter_all) {
885 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
886 ret = csum_dirty_buffer(root->fs_info, bvec);
887 if (ret)
888 break;
889 }
890
891 return errno_to_blk_status(ret);
892}
893
894static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
895 u64 dio_file_offset)
896{
897 /*
898 * when we're called for a write, we're already in the async
899 * submission context. Just jump into btrfs_map_bio
900 */
901 return btree_csum_one_bio(bio);
902}
903
904static bool should_async_write(struct btrfs_fs_info *fs_info,
905 struct btrfs_inode *bi)
906{
907 if (btrfs_is_zoned(fs_info))
908 return false;
909 if (atomic_read(&bi->sync_writers))
910 return false;
911 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
912 return false;
913 return true;
914}
915
916blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
917 int mirror_num, unsigned long bio_flags)
918{
919 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
920 blk_status_t ret;
921
922 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
923 /*
924 * called for a read, do the setup so that checksum validation
925 * can happen in the async kernel threads
926 */
927 ret = btrfs_bio_wq_end_io(fs_info, bio,
928 BTRFS_WQ_ENDIO_METADATA);
929 if (ret)
930 goto out_w_error;
931 ret = btrfs_map_bio(fs_info, bio, mirror_num);
932 } else if (!should_async_write(fs_info, BTRFS_I(inode))) {
933 ret = btree_csum_one_bio(bio);
934 if (ret)
935 goto out_w_error;
936 ret = btrfs_map_bio(fs_info, bio, mirror_num);
937 } else {
938 /*
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
941 */
942 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
943 0, btree_submit_bio_start);
944 }
945
946 if (ret)
947 goto out_w_error;
948 return 0;
949
950out_w_error:
951 bio->bi_status = ret;
952 bio_endio(bio);
953 return ret;
954}
955
956#ifdef CONFIG_MIGRATION
957static int btree_migratepage(struct address_space *mapping,
958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
960{
961 /*
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
964 */
965 if (PageDirty(page))
966 return -EAGAIN;
967 /*
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
970 */
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
973 return -EAGAIN;
974 return migrate_page(mapping, newpage, page, mode);
975}
976#endif
977
978
979static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
981{
982 struct btrfs_fs_info *fs_info;
983 int ret;
984
985 if (wbc->sync_mode == WB_SYNC_NONE) {
986
987 if (wbc->for_kupdate)
988 return 0;
989
990 fs_info = BTRFS_I(mapping->host)->root->fs_info;
991 /* this is a bit racy, but that's ok */
992 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
993 BTRFS_DIRTY_METADATA_THRESH,
994 fs_info->dirty_metadata_batch);
995 if (ret < 0)
996 return 0;
997 }
998 return btree_write_cache_pages(mapping, wbc);
999}
1000
1001static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1002{
1003 if (PageWriteback(page) || PageDirty(page))
1004 return 0;
1005
1006 return try_release_extent_buffer(page);
1007}
1008
1009static void btree_invalidatepage(struct page *page, unsigned int offset,
1010 unsigned int length)
1011{
1012 struct extent_io_tree *tree;
1013 tree = &BTRFS_I(page->mapping->host)->io_tree;
1014 extent_invalidatepage(tree, page, offset);
1015 btree_releasepage(page, GFP_NOFS);
1016 if (PagePrivate(page)) {
1017 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1018 "page private not zero on page %llu",
1019 (unsigned long long)page_offset(page));
1020 detach_page_private(page);
1021 }
1022}
1023
1024static int btree_set_page_dirty(struct page *page)
1025{
1026#ifdef DEBUG
1027 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1028 struct btrfs_subpage *subpage;
1029 struct extent_buffer *eb;
1030 int cur_bit = 0;
1031 u64 page_start = page_offset(page);
1032
1033 if (fs_info->sectorsize == PAGE_SIZE) {
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1040 return __set_page_dirty_nobuffers(page);
1041 }
1042 ASSERT(PagePrivate(page) && page->private);
1043 subpage = (struct btrfs_subpage *)page->private;
1044
1045 ASSERT(subpage->dirty_bitmap);
1046 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1047 unsigned long flags;
1048 u64 cur;
1049 u16 tmp = (1 << cur_bit);
1050
1051 spin_lock_irqsave(&subpage->lock, flags);
1052 if (!(tmp & subpage->dirty_bitmap)) {
1053 spin_unlock_irqrestore(&subpage->lock, flags);
1054 cur_bit++;
1055 continue;
1056 }
1057 spin_unlock_irqrestore(&subpage->lock, flags);
1058 cur = page_start + cur_bit * fs_info->sectorsize;
1059
1060 eb = find_extent_buffer(fs_info, cur);
1061 ASSERT(eb);
1062 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063 ASSERT(atomic_read(&eb->refs));
1064 btrfs_assert_tree_locked(eb);
1065 free_extent_buffer(eb);
1066
1067 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1068 }
1069#endif
1070 return __set_page_dirty_nobuffers(page);
1071}
1072
1073static const struct address_space_operations btree_aops = {
1074 .writepages = btree_writepages,
1075 .releasepage = btree_releasepage,
1076 .invalidatepage = btree_invalidatepage,
1077#ifdef CONFIG_MIGRATION
1078 .migratepage = btree_migratepage,
1079#endif
1080 .set_page_dirty = btree_set_page_dirty,
1081};
1082
1083struct extent_buffer *btrfs_find_create_tree_block(
1084 struct btrfs_fs_info *fs_info,
1085 u64 bytenr, u64 owner_root,
1086 int level)
1087{
1088 if (btrfs_is_testing(fs_info))
1089 return alloc_test_extent_buffer(fs_info, bytenr);
1090 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1091}
1092
1093/*
1094 * Read tree block at logical address @bytenr and do variant basic but critical
1095 * verification.
1096 *
1097 * @owner_root: the objectid of the root owner for this block.
1098 * @parent_transid: expected transid of this tree block, skip check if 0
1099 * @level: expected level, mandatory check
1100 * @first_key: expected key in slot 0, skip check if NULL
1101 */
1102struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1103 u64 owner_root, u64 parent_transid,
1104 int level, struct btrfs_key *first_key)
1105{
1106 struct extent_buffer *buf = NULL;
1107 int ret;
1108
1109 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1110 if (IS_ERR(buf))
1111 return buf;
1112
1113 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1114 level, first_key);
1115 if (ret) {
1116 free_extent_buffer_stale(buf);
1117 return ERR_PTR(ret);
1118 }
1119 return buf;
1120
1121}
1122
1123void btrfs_clean_tree_block(struct extent_buffer *buf)
1124{
1125 struct btrfs_fs_info *fs_info = buf->fs_info;
1126 if (btrfs_header_generation(buf) ==
1127 fs_info->running_transaction->transid) {
1128 btrfs_assert_tree_locked(buf);
1129
1130 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1131 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1132 -buf->len,
1133 fs_info->dirty_metadata_batch);
1134 clear_extent_buffer_dirty(buf);
1135 }
1136 }
1137}
1138
1139static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1140 u64 objectid)
1141{
1142 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1143 root->fs_info = fs_info;
1144 root->node = NULL;
1145 root->commit_root = NULL;
1146 root->state = 0;
1147 root->orphan_cleanup_state = 0;
1148
1149 root->last_trans = 0;
1150 root->free_objectid = 0;
1151 root->nr_delalloc_inodes = 0;
1152 root->nr_ordered_extents = 0;
1153 root->inode_tree = RB_ROOT;
1154 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1155 root->block_rsv = NULL;
1156
1157 INIT_LIST_HEAD(&root->dirty_list);
1158 INIT_LIST_HEAD(&root->root_list);
1159 INIT_LIST_HEAD(&root->delalloc_inodes);
1160 INIT_LIST_HEAD(&root->delalloc_root);
1161 INIT_LIST_HEAD(&root->ordered_extents);
1162 INIT_LIST_HEAD(&root->ordered_root);
1163 INIT_LIST_HEAD(&root->reloc_dirty_list);
1164 INIT_LIST_HEAD(&root->logged_list[0]);
1165 INIT_LIST_HEAD(&root->logged_list[1]);
1166 spin_lock_init(&root->inode_lock);
1167 spin_lock_init(&root->delalloc_lock);
1168 spin_lock_init(&root->ordered_extent_lock);
1169 spin_lock_init(&root->accounting_lock);
1170 spin_lock_init(&root->log_extents_lock[0]);
1171 spin_lock_init(&root->log_extents_lock[1]);
1172 spin_lock_init(&root->qgroup_meta_rsv_lock);
1173 mutex_init(&root->objectid_mutex);
1174 mutex_init(&root->log_mutex);
1175 mutex_init(&root->ordered_extent_mutex);
1176 mutex_init(&root->delalloc_mutex);
1177 init_waitqueue_head(&root->qgroup_flush_wait);
1178 init_waitqueue_head(&root->log_writer_wait);
1179 init_waitqueue_head(&root->log_commit_wait[0]);
1180 init_waitqueue_head(&root->log_commit_wait[1]);
1181 INIT_LIST_HEAD(&root->log_ctxs[0]);
1182 INIT_LIST_HEAD(&root->log_ctxs[1]);
1183 atomic_set(&root->log_commit[0], 0);
1184 atomic_set(&root->log_commit[1], 0);
1185 atomic_set(&root->log_writers, 0);
1186 atomic_set(&root->log_batch, 0);
1187 refcount_set(&root->refs, 1);
1188 atomic_set(&root->snapshot_force_cow, 0);
1189 atomic_set(&root->nr_swapfiles, 0);
1190 root->log_transid = 0;
1191 root->log_transid_committed = -1;
1192 root->last_log_commit = 0;
1193 if (!dummy) {
1194 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1195 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1196 extent_io_tree_init(fs_info, &root->log_csum_range,
1197 IO_TREE_LOG_CSUM_RANGE, NULL);
1198 }
1199
1200 memset(&root->root_key, 0, sizeof(root->root_key));
1201 memset(&root->root_item, 0, sizeof(root->root_item));
1202 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1203 root->root_key.objectid = objectid;
1204 root->anon_dev = 0;
1205
1206 spin_lock_init(&root->root_item_lock);
1207 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1208#ifdef CONFIG_BTRFS_DEBUG
1209 INIT_LIST_HEAD(&root->leak_list);
1210 spin_lock(&fs_info->fs_roots_radix_lock);
1211 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1212 spin_unlock(&fs_info->fs_roots_radix_lock);
1213#endif
1214}
1215
1216static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1217 u64 objectid, gfp_t flags)
1218{
1219 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1220 if (root)
1221 __setup_root(root, fs_info, objectid);
1222 return root;
1223}
1224
1225#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1226/* Should only be used by the testing infrastructure */
1227struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1228{
1229 struct btrfs_root *root;
1230
1231 if (!fs_info)
1232 return ERR_PTR(-EINVAL);
1233
1234 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1235 if (!root)
1236 return ERR_PTR(-ENOMEM);
1237
1238 /* We don't use the stripesize in selftest, set it as sectorsize */
1239 root->alloc_bytenr = 0;
1240
1241 return root;
1242}
1243#endif
1244
1245struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1246 u64 objectid)
1247{
1248 struct btrfs_fs_info *fs_info = trans->fs_info;
1249 struct extent_buffer *leaf;
1250 struct btrfs_root *tree_root = fs_info->tree_root;
1251 struct btrfs_root *root;
1252 struct btrfs_key key;
1253 unsigned int nofs_flag;
1254 int ret = 0;
1255
1256 /*
1257 * We're holding a transaction handle, so use a NOFS memory allocation
1258 * context to avoid deadlock if reclaim happens.
1259 */
1260 nofs_flag = memalloc_nofs_save();
1261 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1262 memalloc_nofs_restore(nofs_flag);
1263 if (!root)
1264 return ERR_PTR(-ENOMEM);
1265
1266 root->root_key.objectid = objectid;
1267 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1268 root->root_key.offset = 0;
1269
1270 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1271 BTRFS_NESTING_NORMAL);
1272 if (IS_ERR(leaf)) {
1273 ret = PTR_ERR(leaf);
1274 leaf = NULL;
1275 goto fail_unlock;
1276 }
1277
1278 root->node = leaf;
1279 btrfs_mark_buffer_dirty(leaf);
1280
1281 root->commit_root = btrfs_root_node(root);
1282 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1283
1284 btrfs_set_root_flags(&root->root_item, 0);
1285 btrfs_set_root_limit(&root->root_item, 0);
1286 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1287 btrfs_set_root_generation(&root->root_item, trans->transid);
1288 btrfs_set_root_level(&root->root_item, 0);
1289 btrfs_set_root_refs(&root->root_item, 1);
1290 btrfs_set_root_used(&root->root_item, leaf->len);
1291 btrfs_set_root_last_snapshot(&root->root_item, 0);
1292 btrfs_set_root_dirid(&root->root_item, 0);
1293 if (is_fstree(objectid))
1294 generate_random_guid(root->root_item.uuid);
1295 else
1296 export_guid(root->root_item.uuid, &guid_null);
1297 btrfs_set_root_drop_level(&root->root_item, 0);
1298
1299 btrfs_tree_unlock(leaf);
1300
1301 key.objectid = objectid;
1302 key.type = BTRFS_ROOT_ITEM_KEY;
1303 key.offset = 0;
1304 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1305 if (ret)
1306 goto fail;
1307
1308 return root;
1309
1310fail_unlock:
1311 if (leaf)
1312 btrfs_tree_unlock(leaf);
1313fail:
1314 btrfs_put_root(root);
1315
1316 return ERR_PTR(ret);
1317}
1318
1319static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1320 struct btrfs_fs_info *fs_info)
1321{
1322 struct btrfs_root *root;
1323
1324 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1325 if (!root)
1326 return ERR_PTR(-ENOMEM);
1327
1328 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1329 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1330 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1331
1332 return root;
1333}
1334
1335int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1336 struct btrfs_root *root)
1337{
1338 struct extent_buffer *leaf;
1339
1340 /*
1341 * DON'T set SHAREABLE bit for log trees.
1342 *
1343 * Log trees are not exposed to user space thus can't be snapshotted,
1344 * and they go away before a real commit is actually done.
1345 *
1346 * They do store pointers to file data extents, and those reference
1347 * counts still get updated (along with back refs to the log tree).
1348 */
1349
1350 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1351 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1352 if (IS_ERR(leaf))
1353 return PTR_ERR(leaf);
1354
1355 root->node = leaf;
1356
1357 btrfs_mark_buffer_dirty(root->node);
1358 btrfs_tree_unlock(root->node);
1359
1360 return 0;
1361}
1362
1363int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1364 struct btrfs_fs_info *fs_info)
1365{
1366 struct btrfs_root *log_root;
1367
1368 log_root = alloc_log_tree(trans, fs_info);
1369 if (IS_ERR(log_root))
1370 return PTR_ERR(log_root);
1371
1372 if (!btrfs_is_zoned(fs_info)) {
1373 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1374
1375 if (ret) {
1376 btrfs_put_root(log_root);
1377 return ret;
1378 }
1379 }
1380
1381 WARN_ON(fs_info->log_root_tree);
1382 fs_info->log_root_tree = log_root;
1383 return 0;
1384}
1385
1386int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root)
1388{
1389 struct btrfs_fs_info *fs_info = root->fs_info;
1390 struct btrfs_root *log_root;
1391 struct btrfs_inode_item *inode_item;
1392 int ret;
1393
1394 log_root = alloc_log_tree(trans, fs_info);
1395 if (IS_ERR(log_root))
1396 return PTR_ERR(log_root);
1397
1398 ret = btrfs_alloc_log_tree_node(trans, log_root);
1399 if (ret) {
1400 btrfs_put_root(log_root);
1401 return ret;
1402 }
1403
1404 log_root->last_trans = trans->transid;
1405 log_root->root_key.offset = root->root_key.objectid;
1406
1407 inode_item = &log_root->root_item.inode;
1408 btrfs_set_stack_inode_generation(inode_item, 1);
1409 btrfs_set_stack_inode_size(inode_item, 3);
1410 btrfs_set_stack_inode_nlink(inode_item, 1);
1411 btrfs_set_stack_inode_nbytes(inode_item,
1412 fs_info->nodesize);
1413 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1414
1415 btrfs_set_root_node(&log_root->root_item, log_root->node);
1416
1417 WARN_ON(root->log_root);
1418 root->log_root = log_root;
1419 root->log_transid = 0;
1420 root->log_transid_committed = -1;
1421 root->last_log_commit = 0;
1422 return 0;
1423}
1424
1425static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1426 struct btrfs_path *path,
1427 struct btrfs_key *key)
1428{
1429 struct btrfs_root *root;
1430 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1431 u64 generation;
1432 int ret;
1433 int level;
1434
1435 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1436 if (!root)
1437 return ERR_PTR(-ENOMEM);
1438
1439 ret = btrfs_find_root(tree_root, key, path,
1440 &root->root_item, &root->root_key);
1441 if (ret) {
1442 if (ret > 0)
1443 ret = -ENOENT;
1444 goto fail;
1445 }
1446
1447 generation = btrfs_root_generation(&root->root_item);
1448 level = btrfs_root_level(&root->root_item);
1449 root->node = read_tree_block(fs_info,
1450 btrfs_root_bytenr(&root->root_item),
1451 key->objectid, generation, level, NULL);
1452 if (IS_ERR(root->node)) {
1453 ret = PTR_ERR(root->node);
1454 root->node = NULL;
1455 goto fail;
1456 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1457 ret = -EIO;
1458 goto fail;
1459 }
1460 root->commit_root = btrfs_root_node(root);
1461 return root;
1462fail:
1463 btrfs_put_root(root);
1464 return ERR_PTR(ret);
1465}
1466
1467struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1468 struct btrfs_key *key)
1469{
1470 struct btrfs_root *root;
1471 struct btrfs_path *path;
1472
1473 path = btrfs_alloc_path();
1474 if (!path)
1475 return ERR_PTR(-ENOMEM);
1476 root = read_tree_root_path(tree_root, path, key);
1477 btrfs_free_path(path);
1478
1479 return root;
1480}
1481
1482/*
1483 * Initialize subvolume root in-memory structure
1484 *
1485 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1486 */
1487static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1488{
1489 int ret;
1490 unsigned int nofs_flag;
1491
1492 /*
1493 * We might be called under a transaction (e.g. indirect backref
1494 * resolution) which could deadlock if it triggers memory reclaim
1495 */
1496 nofs_flag = memalloc_nofs_save();
1497 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1498 memalloc_nofs_restore(nofs_flag);
1499 if (ret)
1500 goto fail;
1501
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1503 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1504 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1505 btrfs_check_and_init_root_item(&root->root_item);
1506 }
1507
1508 /*
1509 * Don't assign anonymous block device to roots that are not exposed to
1510 * userspace, the id pool is limited to 1M
1511 */
1512 if (is_fstree(root->root_key.objectid) &&
1513 btrfs_root_refs(&root->root_item) > 0) {
1514 if (!anon_dev) {
1515 ret = get_anon_bdev(&root->anon_dev);
1516 if (ret)
1517 goto fail;
1518 } else {
1519 root->anon_dev = anon_dev;
1520 }
1521 }
1522
1523 mutex_lock(&root->objectid_mutex);
1524 ret = btrfs_init_root_free_objectid(root);
1525 if (ret) {
1526 mutex_unlock(&root->objectid_mutex);
1527 goto fail;
1528 }
1529
1530 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1531
1532 mutex_unlock(&root->objectid_mutex);
1533
1534 return 0;
1535fail:
1536 /* The caller is responsible to call btrfs_free_fs_root */
1537 return ret;
1538}
1539
1540static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1541 u64 root_id)
1542{
1543 struct btrfs_root *root;
1544
1545 spin_lock(&fs_info->fs_roots_radix_lock);
1546 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1547 (unsigned long)root_id);
1548 if (root)
1549 root = btrfs_grab_root(root);
1550 spin_unlock(&fs_info->fs_roots_radix_lock);
1551 return root;
1552}
1553
1554static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1555 u64 objectid)
1556{
1557 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1558 return btrfs_grab_root(fs_info->tree_root);
1559 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1560 return btrfs_grab_root(fs_info->extent_root);
1561 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1562 return btrfs_grab_root(fs_info->chunk_root);
1563 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1564 return btrfs_grab_root(fs_info->dev_root);
1565 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1566 return btrfs_grab_root(fs_info->csum_root);
1567 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1568 return btrfs_grab_root(fs_info->quota_root) ?
1569 fs_info->quota_root : ERR_PTR(-ENOENT);
1570 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1571 return btrfs_grab_root(fs_info->uuid_root) ?
1572 fs_info->uuid_root : ERR_PTR(-ENOENT);
1573 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1574 return btrfs_grab_root(fs_info->free_space_root) ?
1575 fs_info->free_space_root : ERR_PTR(-ENOENT);
1576 return NULL;
1577}
1578
1579int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1580 struct btrfs_root *root)
1581{
1582 int ret;
1583
1584 ret = radix_tree_preload(GFP_NOFS);
1585 if (ret)
1586 return ret;
1587
1588 spin_lock(&fs_info->fs_roots_radix_lock);
1589 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1590 (unsigned long)root->root_key.objectid,
1591 root);
1592 if (ret == 0) {
1593 btrfs_grab_root(root);
1594 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1595 }
1596 spin_unlock(&fs_info->fs_roots_radix_lock);
1597 radix_tree_preload_end();
1598
1599 return ret;
1600}
1601
1602void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1603{
1604#ifdef CONFIG_BTRFS_DEBUG
1605 struct btrfs_root *root;
1606
1607 while (!list_empty(&fs_info->allocated_roots)) {
1608 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1609
1610 root = list_first_entry(&fs_info->allocated_roots,
1611 struct btrfs_root, leak_list);
1612 btrfs_err(fs_info, "leaked root %s refcount %d",
1613 btrfs_root_name(&root->root_key, buf),
1614 refcount_read(&root->refs));
1615 while (refcount_read(&root->refs) > 1)
1616 btrfs_put_root(root);
1617 btrfs_put_root(root);
1618 }
1619#endif
1620}
1621
1622void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1623{
1624 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1625 percpu_counter_destroy(&fs_info->delalloc_bytes);
1626 percpu_counter_destroy(&fs_info->ordered_bytes);
1627 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1628 btrfs_free_csum_hash(fs_info);
1629 btrfs_free_stripe_hash_table(fs_info);
1630 btrfs_free_ref_cache(fs_info);
1631 kfree(fs_info->balance_ctl);
1632 kfree(fs_info->delayed_root);
1633 btrfs_put_root(fs_info->extent_root);
1634 btrfs_put_root(fs_info->tree_root);
1635 btrfs_put_root(fs_info->chunk_root);
1636 btrfs_put_root(fs_info->dev_root);
1637 btrfs_put_root(fs_info->csum_root);
1638 btrfs_put_root(fs_info->quota_root);
1639 btrfs_put_root(fs_info->uuid_root);
1640 btrfs_put_root(fs_info->free_space_root);
1641 btrfs_put_root(fs_info->fs_root);
1642 btrfs_put_root(fs_info->data_reloc_root);
1643 btrfs_check_leaked_roots(fs_info);
1644 btrfs_extent_buffer_leak_debug_check(fs_info);
1645 kfree(fs_info->super_copy);
1646 kfree(fs_info->super_for_commit);
1647 kvfree(fs_info);
1648}
1649
1650
1651/*
1652 * Get an in-memory reference of a root structure.
1653 *
1654 * For essential trees like root/extent tree, we grab it from fs_info directly.
1655 * For subvolume trees, we check the cached filesystem roots first. If not
1656 * found, then read it from disk and add it to cached fs roots.
1657 *
1658 * Caller should release the root by calling btrfs_put_root() after the usage.
1659 *
1660 * NOTE: Reloc and log trees can't be read by this function as they share the
1661 * same root objectid.
1662 *
1663 * @objectid: root id
1664 * @anon_dev: preallocated anonymous block device number for new roots,
1665 * pass 0 for new allocation.
1666 * @check_ref: whether to check root item references, If true, return -ENOENT
1667 * for orphan roots
1668 */
1669static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1670 u64 objectid, dev_t anon_dev,
1671 bool check_ref)
1672{
1673 struct btrfs_root *root;
1674 struct btrfs_path *path;
1675 struct btrfs_key key;
1676 int ret;
1677
1678 root = btrfs_get_global_root(fs_info, objectid);
1679 if (root)
1680 return root;
1681again:
1682 root = btrfs_lookup_fs_root(fs_info, objectid);
1683 if (root) {
1684 /* Shouldn't get preallocated anon_dev for cached roots */
1685 ASSERT(!anon_dev);
1686 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1687 btrfs_put_root(root);
1688 return ERR_PTR(-ENOENT);
1689 }
1690 return root;
1691 }
1692
1693 key.objectid = objectid;
1694 key.type = BTRFS_ROOT_ITEM_KEY;
1695 key.offset = (u64)-1;
1696 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1697 if (IS_ERR(root))
1698 return root;
1699
1700 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1701 ret = -ENOENT;
1702 goto fail;
1703 }
1704
1705 ret = btrfs_init_fs_root(root, anon_dev);
1706 if (ret)
1707 goto fail;
1708
1709 path = btrfs_alloc_path();
1710 if (!path) {
1711 ret = -ENOMEM;
1712 goto fail;
1713 }
1714 key.objectid = BTRFS_ORPHAN_OBJECTID;
1715 key.type = BTRFS_ORPHAN_ITEM_KEY;
1716 key.offset = objectid;
1717
1718 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1719 btrfs_free_path(path);
1720 if (ret < 0)
1721 goto fail;
1722 if (ret == 0)
1723 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1724
1725 ret = btrfs_insert_fs_root(fs_info, root);
1726 if (ret) {
1727 btrfs_put_root(root);
1728 if (ret == -EEXIST)
1729 goto again;
1730 goto fail;
1731 }
1732 return root;
1733fail:
1734 btrfs_put_root(root);
1735 return ERR_PTR(ret);
1736}
1737
1738/*
1739 * Get in-memory reference of a root structure
1740 *
1741 * @objectid: tree objectid
1742 * @check_ref: if set, verify that the tree exists and the item has at least
1743 * one reference
1744 */
1745struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1746 u64 objectid, bool check_ref)
1747{
1748 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1749}
1750
1751/*
1752 * Get in-memory reference of a root structure, created as new, optionally pass
1753 * the anonymous block device id
1754 *
1755 * @objectid: tree objectid
1756 * @anon_dev: if zero, allocate a new anonymous block device or use the
1757 * parameter value
1758 */
1759struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1760 u64 objectid, dev_t anon_dev)
1761{
1762 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1763}
1764
1765/*
1766 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1767 * @fs_info: the fs_info
1768 * @objectid: the objectid we need to lookup
1769 *
1770 * This is exclusively used for backref walking, and exists specifically because
1771 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1772 * creation time, which means we may have to read the tree_root in order to look
1773 * up a fs root that is not in memory. If the root is not in memory we will
1774 * read the tree root commit root and look up the fs root from there. This is a
1775 * temporary root, it will not be inserted into the radix tree as it doesn't
1776 * have the most uptodate information, it'll simply be discarded once the
1777 * backref code is finished using the root.
1778 */
1779struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1780 struct btrfs_path *path,
1781 u64 objectid)
1782{
1783 struct btrfs_root *root;
1784 struct btrfs_key key;
1785
1786 ASSERT(path->search_commit_root && path->skip_locking);
1787
1788 /*
1789 * This can return -ENOENT if we ask for a root that doesn't exist, but
1790 * since this is called via the backref walking code we won't be looking
1791 * up a root that doesn't exist, unless there's corruption. So if root
1792 * != NULL just return it.
1793 */
1794 root = btrfs_get_global_root(fs_info, objectid);
1795 if (root)
1796 return root;
1797
1798 root = btrfs_lookup_fs_root(fs_info, objectid);
1799 if (root)
1800 return root;
1801
1802 key.objectid = objectid;
1803 key.type = BTRFS_ROOT_ITEM_KEY;
1804 key.offset = (u64)-1;
1805 root = read_tree_root_path(fs_info->tree_root, path, &key);
1806 btrfs_release_path(path);
1807
1808 return root;
1809}
1810
1811/*
1812 * called by the kthread helper functions to finally call the bio end_io
1813 * functions. This is where read checksum verification actually happens
1814 */
1815static void end_workqueue_fn(struct btrfs_work *work)
1816{
1817 struct bio *bio;
1818 struct btrfs_end_io_wq *end_io_wq;
1819
1820 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1821 bio = end_io_wq->bio;
1822
1823 bio->bi_status = end_io_wq->status;
1824 bio->bi_private = end_io_wq->private;
1825 bio->bi_end_io = end_io_wq->end_io;
1826 bio_endio(bio);
1827 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1828}
1829
1830static int cleaner_kthread(void *arg)
1831{
1832 struct btrfs_root *root = arg;
1833 struct btrfs_fs_info *fs_info = root->fs_info;
1834 int again;
1835
1836 while (1) {
1837 again = 0;
1838
1839 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1840
1841 /* Make the cleaner go to sleep early. */
1842 if (btrfs_need_cleaner_sleep(fs_info))
1843 goto sleep;
1844
1845 /*
1846 * Do not do anything if we might cause open_ctree() to block
1847 * before we have finished mounting the filesystem.
1848 */
1849 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1850 goto sleep;
1851
1852 if (!mutex_trylock(&fs_info->cleaner_mutex))
1853 goto sleep;
1854
1855 /*
1856 * Avoid the problem that we change the status of the fs
1857 * during the above check and trylock.
1858 */
1859 if (btrfs_need_cleaner_sleep(fs_info)) {
1860 mutex_unlock(&fs_info->cleaner_mutex);
1861 goto sleep;
1862 }
1863
1864 btrfs_run_delayed_iputs(fs_info);
1865
1866 again = btrfs_clean_one_deleted_snapshot(root);
1867 mutex_unlock(&fs_info->cleaner_mutex);
1868
1869 /*
1870 * The defragger has dealt with the R/O remount and umount,
1871 * needn't do anything special here.
1872 */
1873 btrfs_run_defrag_inodes(fs_info);
1874
1875 /*
1876 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1877 * with relocation (btrfs_relocate_chunk) and relocation
1878 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1879 * after acquiring fs_info->reclaim_bgs_lock. So we
1880 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1881 * unused block groups.
1882 */
1883 btrfs_delete_unused_bgs(fs_info);
1884
1885 /*
1886 * Reclaim block groups in the reclaim_bgs list after we deleted
1887 * all unused block_groups. This possibly gives us some more free
1888 * space.
1889 */
1890 btrfs_reclaim_bgs(fs_info);
1891sleep:
1892 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1893 if (kthread_should_park())
1894 kthread_parkme();
1895 if (kthread_should_stop())
1896 return 0;
1897 if (!again) {
1898 set_current_state(TASK_INTERRUPTIBLE);
1899 schedule();
1900 __set_current_state(TASK_RUNNING);
1901 }
1902 }
1903}
1904
1905static int transaction_kthread(void *arg)
1906{
1907 struct btrfs_root *root = arg;
1908 struct btrfs_fs_info *fs_info = root->fs_info;
1909 struct btrfs_trans_handle *trans;
1910 struct btrfs_transaction *cur;
1911 u64 transid;
1912 time64_t delta;
1913 unsigned long delay;
1914 bool cannot_commit;
1915
1916 do {
1917 cannot_commit = false;
1918 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1919 mutex_lock(&fs_info->transaction_kthread_mutex);
1920
1921 spin_lock(&fs_info->trans_lock);
1922 cur = fs_info->running_transaction;
1923 if (!cur) {
1924 spin_unlock(&fs_info->trans_lock);
1925 goto sleep;
1926 }
1927
1928 delta = ktime_get_seconds() - cur->start_time;
1929 if (cur->state < TRANS_STATE_COMMIT_START &&
1930 delta < fs_info->commit_interval) {
1931 spin_unlock(&fs_info->trans_lock);
1932 delay -= msecs_to_jiffies((delta - 1) * 1000);
1933 delay = min(delay,
1934 msecs_to_jiffies(fs_info->commit_interval * 1000));
1935 goto sleep;
1936 }
1937 transid = cur->transid;
1938 spin_unlock(&fs_info->trans_lock);
1939
1940 /* If the file system is aborted, this will always fail. */
1941 trans = btrfs_attach_transaction(root);
1942 if (IS_ERR(trans)) {
1943 if (PTR_ERR(trans) != -ENOENT)
1944 cannot_commit = true;
1945 goto sleep;
1946 }
1947 if (transid == trans->transid) {
1948 btrfs_commit_transaction(trans);
1949 } else {
1950 btrfs_end_transaction(trans);
1951 }
1952sleep:
1953 wake_up_process(fs_info->cleaner_kthread);
1954 mutex_unlock(&fs_info->transaction_kthread_mutex);
1955
1956 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1957 &fs_info->fs_state)))
1958 btrfs_cleanup_transaction(fs_info);
1959 if (!kthread_should_stop() &&
1960 (!btrfs_transaction_blocked(fs_info) ||
1961 cannot_commit))
1962 schedule_timeout_interruptible(delay);
1963 } while (!kthread_should_stop());
1964 return 0;
1965}
1966
1967/*
1968 * This will find the highest generation in the array of root backups. The
1969 * index of the highest array is returned, or -EINVAL if we can't find
1970 * anything.
1971 *
1972 * We check to make sure the array is valid by comparing the
1973 * generation of the latest root in the array with the generation
1974 * in the super block. If they don't match we pitch it.
1975 */
1976static int find_newest_super_backup(struct btrfs_fs_info *info)
1977{
1978 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1979 u64 cur;
1980 struct btrfs_root_backup *root_backup;
1981 int i;
1982
1983 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1984 root_backup = info->super_copy->super_roots + i;
1985 cur = btrfs_backup_tree_root_gen(root_backup);
1986 if (cur == newest_gen)
1987 return i;
1988 }
1989
1990 return -EINVAL;
1991}
1992
1993/*
1994 * copy all the root pointers into the super backup array.
1995 * this will bump the backup pointer by one when it is
1996 * done
1997 */
1998static void backup_super_roots(struct btrfs_fs_info *info)
1999{
2000 const int next_backup = info->backup_root_index;
2001 struct btrfs_root_backup *root_backup;
2002
2003 root_backup = info->super_for_commit->super_roots + next_backup;
2004
2005 /*
2006 * make sure all of our padding and empty slots get zero filled
2007 * regardless of which ones we use today
2008 */
2009 memset(root_backup, 0, sizeof(*root_backup));
2010
2011 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2012
2013 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2014 btrfs_set_backup_tree_root_gen(root_backup,
2015 btrfs_header_generation(info->tree_root->node));
2016
2017 btrfs_set_backup_tree_root_level(root_backup,
2018 btrfs_header_level(info->tree_root->node));
2019
2020 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2021 btrfs_set_backup_chunk_root_gen(root_backup,
2022 btrfs_header_generation(info->chunk_root->node));
2023 btrfs_set_backup_chunk_root_level(root_backup,
2024 btrfs_header_level(info->chunk_root->node));
2025
2026 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2027 btrfs_set_backup_extent_root_gen(root_backup,
2028 btrfs_header_generation(info->extent_root->node));
2029 btrfs_set_backup_extent_root_level(root_backup,
2030 btrfs_header_level(info->extent_root->node));
2031
2032 /*
2033 * we might commit during log recovery, which happens before we set
2034 * the fs_root. Make sure it is valid before we fill it in.
2035 */
2036 if (info->fs_root && info->fs_root->node) {
2037 btrfs_set_backup_fs_root(root_backup,
2038 info->fs_root->node->start);
2039 btrfs_set_backup_fs_root_gen(root_backup,
2040 btrfs_header_generation(info->fs_root->node));
2041 btrfs_set_backup_fs_root_level(root_backup,
2042 btrfs_header_level(info->fs_root->node));
2043 }
2044
2045 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2046 btrfs_set_backup_dev_root_gen(root_backup,
2047 btrfs_header_generation(info->dev_root->node));
2048 btrfs_set_backup_dev_root_level(root_backup,
2049 btrfs_header_level(info->dev_root->node));
2050
2051 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2052 btrfs_set_backup_csum_root_gen(root_backup,
2053 btrfs_header_generation(info->csum_root->node));
2054 btrfs_set_backup_csum_root_level(root_backup,
2055 btrfs_header_level(info->csum_root->node));
2056
2057 btrfs_set_backup_total_bytes(root_backup,
2058 btrfs_super_total_bytes(info->super_copy));
2059 btrfs_set_backup_bytes_used(root_backup,
2060 btrfs_super_bytes_used(info->super_copy));
2061 btrfs_set_backup_num_devices(root_backup,
2062 btrfs_super_num_devices(info->super_copy));
2063
2064 /*
2065 * if we don't copy this out to the super_copy, it won't get remembered
2066 * for the next commit
2067 */
2068 memcpy(&info->super_copy->super_roots,
2069 &info->super_for_commit->super_roots,
2070 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2071}
2072
2073/*
2074 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2075 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2076 *
2077 * fs_info - filesystem whose backup roots need to be read
2078 * priority - priority of backup root required
2079 *
2080 * Returns backup root index on success and -EINVAL otherwise.
2081 */
2082static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2083{
2084 int backup_index = find_newest_super_backup(fs_info);
2085 struct btrfs_super_block *super = fs_info->super_copy;
2086 struct btrfs_root_backup *root_backup;
2087
2088 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2089 if (priority == 0)
2090 return backup_index;
2091
2092 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2093 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2094 } else {
2095 return -EINVAL;
2096 }
2097
2098 root_backup = super->super_roots + backup_index;
2099
2100 btrfs_set_super_generation(super,
2101 btrfs_backup_tree_root_gen(root_backup));
2102 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2103 btrfs_set_super_root_level(super,
2104 btrfs_backup_tree_root_level(root_backup));
2105 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2106
2107 /*
2108 * Fixme: the total bytes and num_devices need to match or we should
2109 * need a fsck
2110 */
2111 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2112 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2113
2114 return backup_index;
2115}
2116
2117/* helper to cleanup workers */
2118static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2119{
2120 btrfs_destroy_workqueue(fs_info->fixup_workers);
2121 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2122 btrfs_destroy_workqueue(fs_info->workers);
2123 btrfs_destroy_workqueue(fs_info->endio_workers);
2124 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2125 btrfs_destroy_workqueue(fs_info->rmw_workers);
2126 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2127 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2128 btrfs_destroy_workqueue(fs_info->delayed_workers);
2129 btrfs_destroy_workqueue(fs_info->caching_workers);
2130 btrfs_destroy_workqueue(fs_info->readahead_workers);
2131 btrfs_destroy_workqueue(fs_info->flush_workers);
2132 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2133 if (fs_info->discard_ctl.discard_workers)
2134 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2135 /*
2136 * Now that all other work queues are destroyed, we can safely destroy
2137 * the queues used for metadata I/O, since tasks from those other work
2138 * queues can do metadata I/O operations.
2139 */
2140 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2141 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2142}
2143
2144static void free_root_extent_buffers(struct btrfs_root *root)
2145{
2146 if (root) {
2147 free_extent_buffer(root->node);
2148 free_extent_buffer(root->commit_root);
2149 root->node = NULL;
2150 root->commit_root = NULL;
2151 }
2152}
2153
2154/* helper to cleanup tree roots */
2155static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2156{
2157 free_root_extent_buffers(info->tree_root);
2158
2159 free_root_extent_buffers(info->dev_root);
2160 free_root_extent_buffers(info->extent_root);
2161 free_root_extent_buffers(info->csum_root);
2162 free_root_extent_buffers(info->quota_root);
2163 free_root_extent_buffers(info->uuid_root);
2164 free_root_extent_buffers(info->fs_root);
2165 free_root_extent_buffers(info->data_reloc_root);
2166 if (free_chunk_root)
2167 free_root_extent_buffers(info->chunk_root);
2168 free_root_extent_buffers(info->free_space_root);
2169}
2170
2171void btrfs_put_root(struct btrfs_root *root)
2172{
2173 if (!root)
2174 return;
2175
2176 if (refcount_dec_and_test(&root->refs)) {
2177 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2178 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2179 if (root->anon_dev)
2180 free_anon_bdev(root->anon_dev);
2181 btrfs_drew_lock_destroy(&root->snapshot_lock);
2182 free_root_extent_buffers(root);
2183#ifdef CONFIG_BTRFS_DEBUG
2184 spin_lock(&root->fs_info->fs_roots_radix_lock);
2185 list_del_init(&root->leak_list);
2186 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2187#endif
2188 kfree(root);
2189 }
2190}
2191
2192void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2193{
2194 int ret;
2195 struct btrfs_root *gang[8];
2196 int i;
2197
2198 while (!list_empty(&fs_info->dead_roots)) {
2199 gang[0] = list_entry(fs_info->dead_roots.next,
2200 struct btrfs_root, root_list);
2201 list_del(&gang[0]->root_list);
2202
2203 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2204 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2205 btrfs_put_root(gang[0]);
2206 }
2207
2208 while (1) {
2209 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2210 (void **)gang, 0,
2211 ARRAY_SIZE(gang));
2212 if (!ret)
2213 break;
2214 for (i = 0; i < ret; i++)
2215 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2216 }
2217}
2218
2219static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2220{
2221 mutex_init(&fs_info->scrub_lock);
2222 atomic_set(&fs_info->scrubs_running, 0);
2223 atomic_set(&fs_info->scrub_pause_req, 0);
2224 atomic_set(&fs_info->scrubs_paused, 0);
2225 atomic_set(&fs_info->scrub_cancel_req, 0);
2226 init_waitqueue_head(&fs_info->scrub_pause_wait);
2227 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2228}
2229
2230static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2231{
2232 spin_lock_init(&fs_info->balance_lock);
2233 mutex_init(&fs_info->balance_mutex);
2234 atomic_set(&fs_info->balance_pause_req, 0);
2235 atomic_set(&fs_info->balance_cancel_req, 0);
2236 fs_info->balance_ctl = NULL;
2237 init_waitqueue_head(&fs_info->balance_wait_q);
2238 atomic_set(&fs_info->reloc_cancel_req, 0);
2239}
2240
2241static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2242{
2243 struct inode *inode = fs_info->btree_inode;
2244
2245 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2246 set_nlink(inode, 1);
2247 /*
2248 * we set the i_size on the btree inode to the max possible int.
2249 * the real end of the address space is determined by all of
2250 * the devices in the system
2251 */
2252 inode->i_size = OFFSET_MAX;
2253 inode->i_mapping->a_ops = &btree_aops;
2254
2255 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2256 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2257 IO_TREE_BTREE_INODE_IO, inode);
2258 BTRFS_I(inode)->io_tree.track_uptodate = false;
2259 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2260
2261 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2262 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2263 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2264 btrfs_insert_inode_hash(inode);
2265}
2266
2267static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2268{
2269 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2270 init_rwsem(&fs_info->dev_replace.rwsem);
2271 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2272}
2273
2274static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2275{
2276 spin_lock_init(&fs_info->qgroup_lock);
2277 mutex_init(&fs_info->qgroup_ioctl_lock);
2278 fs_info->qgroup_tree = RB_ROOT;
2279 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2280 fs_info->qgroup_seq = 1;
2281 fs_info->qgroup_ulist = NULL;
2282 fs_info->qgroup_rescan_running = false;
2283 mutex_init(&fs_info->qgroup_rescan_lock);
2284}
2285
2286static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2287 struct btrfs_fs_devices *fs_devices)
2288{
2289 u32 max_active = fs_info->thread_pool_size;
2290 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2291
2292 fs_info->workers =
2293 btrfs_alloc_workqueue(fs_info, "worker",
2294 flags | WQ_HIGHPRI, max_active, 16);
2295
2296 fs_info->delalloc_workers =
2297 btrfs_alloc_workqueue(fs_info, "delalloc",
2298 flags, max_active, 2);
2299
2300 fs_info->flush_workers =
2301 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2302 flags, max_active, 0);
2303
2304 fs_info->caching_workers =
2305 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2306
2307 fs_info->fixup_workers =
2308 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2309
2310 /*
2311 * endios are largely parallel and should have a very
2312 * low idle thresh
2313 */
2314 fs_info->endio_workers =
2315 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2316 fs_info->endio_meta_workers =
2317 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2318 max_active, 4);
2319 fs_info->endio_meta_write_workers =
2320 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2321 max_active, 2);
2322 fs_info->endio_raid56_workers =
2323 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2324 max_active, 4);
2325 fs_info->rmw_workers =
2326 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2327 fs_info->endio_write_workers =
2328 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2329 max_active, 2);
2330 fs_info->endio_freespace_worker =
2331 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2332 max_active, 0);
2333 fs_info->delayed_workers =
2334 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2335 max_active, 0);
2336 fs_info->readahead_workers =
2337 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2338 max_active, 2);
2339 fs_info->qgroup_rescan_workers =
2340 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2341 fs_info->discard_ctl.discard_workers =
2342 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2343
2344 if (!(fs_info->workers && fs_info->delalloc_workers &&
2345 fs_info->flush_workers &&
2346 fs_info->endio_workers && fs_info->endio_meta_workers &&
2347 fs_info->endio_meta_write_workers &&
2348 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2349 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2350 fs_info->caching_workers && fs_info->readahead_workers &&
2351 fs_info->fixup_workers && fs_info->delayed_workers &&
2352 fs_info->qgroup_rescan_workers &&
2353 fs_info->discard_ctl.discard_workers)) {
2354 return -ENOMEM;
2355 }
2356
2357 return 0;
2358}
2359
2360static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2361{
2362 struct crypto_shash *csum_shash;
2363 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2364
2365 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2366
2367 if (IS_ERR(csum_shash)) {
2368 btrfs_err(fs_info, "error allocating %s hash for checksum",
2369 csum_driver);
2370 return PTR_ERR(csum_shash);
2371 }
2372
2373 fs_info->csum_shash = csum_shash;
2374
2375 return 0;
2376}
2377
2378static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2379 struct btrfs_fs_devices *fs_devices)
2380{
2381 int ret;
2382 struct btrfs_root *log_tree_root;
2383 struct btrfs_super_block *disk_super = fs_info->super_copy;
2384 u64 bytenr = btrfs_super_log_root(disk_super);
2385 int level = btrfs_super_log_root_level(disk_super);
2386
2387 if (fs_devices->rw_devices == 0) {
2388 btrfs_warn(fs_info, "log replay required on RO media");
2389 return -EIO;
2390 }
2391
2392 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2393 GFP_KERNEL);
2394 if (!log_tree_root)
2395 return -ENOMEM;
2396
2397 log_tree_root->node = read_tree_block(fs_info, bytenr,
2398 BTRFS_TREE_LOG_OBJECTID,
2399 fs_info->generation + 1, level,
2400 NULL);
2401 if (IS_ERR(log_tree_root->node)) {
2402 btrfs_warn(fs_info, "failed to read log tree");
2403 ret = PTR_ERR(log_tree_root->node);
2404 log_tree_root->node = NULL;
2405 btrfs_put_root(log_tree_root);
2406 return ret;
2407 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2408 btrfs_err(fs_info, "failed to read log tree");
2409 btrfs_put_root(log_tree_root);
2410 return -EIO;
2411 }
2412 /* returns with log_tree_root freed on success */
2413 ret = btrfs_recover_log_trees(log_tree_root);
2414 if (ret) {
2415 btrfs_handle_fs_error(fs_info, ret,
2416 "Failed to recover log tree");
2417 btrfs_put_root(log_tree_root);
2418 return ret;
2419 }
2420
2421 if (sb_rdonly(fs_info->sb)) {
2422 ret = btrfs_commit_super(fs_info);
2423 if (ret)
2424 return ret;
2425 }
2426
2427 return 0;
2428}
2429
2430static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2431{
2432 struct btrfs_root *tree_root = fs_info->tree_root;
2433 struct btrfs_root *root;
2434 struct btrfs_key location;
2435 int ret;
2436
2437 BUG_ON(!fs_info->tree_root);
2438
2439 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2440 location.type = BTRFS_ROOT_ITEM_KEY;
2441 location.offset = 0;
2442
2443 root = btrfs_read_tree_root(tree_root, &location);
2444 if (IS_ERR(root)) {
2445 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2446 ret = PTR_ERR(root);
2447 goto out;
2448 }
2449 } else {
2450 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2451 fs_info->extent_root = root;
2452 }
2453
2454 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2455 root = btrfs_read_tree_root(tree_root, &location);
2456 if (IS_ERR(root)) {
2457 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2458 ret = PTR_ERR(root);
2459 goto out;
2460 }
2461 } else {
2462 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2463 fs_info->dev_root = root;
2464 }
2465 /* Initialize fs_info for all devices in any case */
2466 btrfs_init_devices_late(fs_info);
2467
2468 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2469 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2470 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2471 root = btrfs_read_tree_root(tree_root, &location);
2472 if (IS_ERR(root)) {
2473 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2474 ret = PTR_ERR(root);
2475 goto out;
2476 }
2477 } else {
2478 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2479 fs_info->csum_root = root;
2480 }
2481 }
2482
2483 /*
2484 * This tree can share blocks with some other fs tree during relocation
2485 * and we need a proper setup by btrfs_get_fs_root
2486 */
2487 root = btrfs_get_fs_root(tree_root->fs_info,
2488 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2489 if (IS_ERR(root)) {
2490 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2491 ret = PTR_ERR(root);
2492 goto out;
2493 }
2494 } else {
2495 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2496 fs_info->data_reloc_root = root;
2497 }
2498
2499 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2500 root = btrfs_read_tree_root(tree_root, &location);
2501 if (!IS_ERR(root)) {
2502 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2503 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2504 fs_info->quota_root = root;
2505 }
2506
2507 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2508 root = btrfs_read_tree_root(tree_root, &location);
2509 if (IS_ERR(root)) {
2510 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2511 ret = PTR_ERR(root);
2512 if (ret != -ENOENT)
2513 goto out;
2514 }
2515 } else {
2516 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2517 fs_info->uuid_root = root;
2518 }
2519
2520 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2521 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2522 root = btrfs_read_tree_root(tree_root, &location);
2523 if (IS_ERR(root)) {
2524 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2525 ret = PTR_ERR(root);
2526 goto out;
2527 }
2528 } else {
2529 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2530 fs_info->free_space_root = root;
2531 }
2532 }
2533
2534 return 0;
2535out:
2536 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2537 location.objectid, ret);
2538 return ret;
2539}
2540
2541/*
2542 * Real super block validation
2543 * NOTE: super csum type and incompat features will not be checked here.
2544 *
2545 * @sb: super block to check
2546 * @mirror_num: the super block number to check its bytenr:
2547 * 0 the primary (1st) sb
2548 * 1, 2 2nd and 3rd backup copy
2549 * -1 skip bytenr check
2550 */
2551static int validate_super(struct btrfs_fs_info *fs_info,
2552 struct btrfs_super_block *sb, int mirror_num)
2553{
2554 u64 nodesize = btrfs_super_nodesize(sb);
2555 u64 sectorsize = btrfs_super_sectorsize(sb);
2556 int ret = 0;
2557
2558 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2559 btrfs_err(fs_info, "no valid FS found");
2560 ret = -EINVAL;
2561 }
2562 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2563 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2564 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2565 ret = -EINVAL;
2566 }
2567 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2568 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2569 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2570 ret = -EINVAL;
2571 }
2572 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2573 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2574 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2575 ret = -EINVAL;
2576 }
2577 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2578 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2579 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2580 ret = -EINVAL;
2581 }
2582
2583 /*
2584 * Check sectorsize and nodesize first, other check will need it.
2585 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2586 */
2587 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2588 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2589 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2590 ret = -EINVAL;
2591 }
2592
2593 /*
2594 * For 4K page size, we only support 4K sector size.
2595 * For 64K page size, we support read-write for 64K sector size, and
2596 * read-only for 4K sector size.
2597 */
2598 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2599 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2600 sectorsize != SZ_64K))) {
2601 btrfs_err(fs_info,
2602 "sectorsize %llu not yet supported for page size %lu",
2603 sectorsize, PAGE_SIZE);
2604 ret = -EINVAL;
2605 }
2606
2607 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2608 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2609 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2610 ret = -EINVAL;
2611 }
2612 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2613 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2614 le32_to_cpu(sb->__unused_leafsize), nodesize);
2615 ret = -EINVAL;
2616 }
2617
2618 /* Root alignment check */
2619 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2620 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2621 btrfs_super_root(sb));
2622 ret = -EINVAL;
2623 }
2624 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2625 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2626 btrfs_super_chunk_root(sb));
2627 ret = -EINVAL;
2628 }
2629 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2630 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2631 btrfs_super_log_root(sb));
2632 ret = -EINVAL;
2633 }
2634
2635 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2636 BTRFS_FSID_SIZE)) {
2637 btrfs_err(fs_info,
2638 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2639 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2640 ret = -EINVAL;
2641 }
2642
2643 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2644 memcmp(fs_info->fs_devices->metadata_uuid,
2645 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2646 btrfs_err(fs_info,
2647"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2648 fs_info->super_copy->metadata_uuid,
2649 fs_info->fs_devices->metadata_uuid);
2650 ret = -EINVAL;
2651 }
2652
2653 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2654 BTRFS_FSID_SIZE) != 0) {
2655 btrfs_err(fs_info,
2656 "dev_item UUID does not match metadata fsid: %pU != %pU",
2657 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2658 ret = -EINVAL;
2659 }
2660
2661 /*
2662 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2663 * done later
2664 */
2665 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2666 btrfs_err(fs_info, "bytes_used is too small %llu",
2667 btrfs_super_bytes_used(sb));
2668 ret = -EINVAL;
2669 }
2670 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2671 btrfs_err(fs_info, "invalid stripesize %u",
2672 btrfs_super_stripesize(sb));
2673 ret = -EINVAL;
2674 }
2675 if (btrfs_super_num_devices(sb) > (1UL << 31))
2676 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2677 btrfs_super_num_devices(sb));
2678 if (btrfs_super_num_devices(sb) == 0) {
2679 btrfs_err(fs_info, "number of devices is 0");
2680 ret = -EINVAL;
2681 }
2682
2683 if (mirror_num >= 0 &&
2684 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2685 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2686 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2687 ret = -EINVAL;
2688 }
2689
2690 /*
2691 * Obvious sys_chunk_array corruptions, it must hold at least one key
2692 * and one chunk
2693 */
2694 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2695 btrfs_err(fs_info, "system chunk array too big %u > %u",
2696 btrfs_super_sys_array_size(sb),
2697 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2698 ret = -EINVAL;
2699 }
2700 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2701 + sizeof(struct btrfs_chunk)) {
2702 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2703 btrfs_super_sys_array_size(sb),
2704 sizeof(struct btrfs_disk_key)
2705 + sizeof(struct btrfs_chunk));
2706 ret = -EINVAL;
2707 }
2708
2709 /*
2710 * The generation is a global counter, we'll trust it more than the others
2711 * but it's still possible that it's the one that's wrong.
2712 */
2713 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2714 btrfs_warn(fs_info,
2715 "suspicious: generation < chunk_root_generation: %llu < %llu",
2716 btrfs_super_generation(sb),
2717 btrfs_super_chunk_root_generation(sb));
2718 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2719 && btrfs_super_cache_generation(sb) != (u64)-1)
2720 btrfs_warn(fs_info,
2721 "suspicious: generation < cache_generation: %llu < %llu",
2722 btrfs_super_generation(sb),
2723 btrfs_super_cache_generation(sb));
2724
2725 return ret;
2726}
2727
2728/*
2729 * Validation of super block at mount time.
2730 * Some checks already done early at mount time, like csum type and incompat
2731 * flags will be skipped.
2732 */
2733static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2734{
2735 return validate_super(fs_info, fs_info->super_copy, 0);
2736}
2737
2738/*
2739 * Validation of super block at write time.
2740 * Some checks like bytenr check will be skipped as their values will be
2741 * overwritten soon.
2742 * Extra checks like csum type and incompat flags will be done here.
2743 */
2744static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2745 struct btrfs_super_block *sb)
2746{
2747 int ret;
2748
2749 ret = validate_super(fs_info, sb, -1);
2750 if (ret < 0)
2751 goto out;
2752 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2753 ret = -EUCLEAN;
2754 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2755 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2756 goto out;
2757 }
2758 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2759 ret = -EUCLEAN;
2760 btrfs_err(fs_info,
2761 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2762 btrfs_super_incompat_flags(sb),
2763 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2764 goto out;
2765 }
2766out:
2767 if (ret < 0)
2768 btrfs_err(fs_info,
2769 "super block corruption detected before writing it to disk");
2770 return ret;
2771}
2772
2773static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2774{
2775 int backup_index = find_newest_super_backup(fs_info);
2776 struct btrfs_super_block *sb = fs_info->super_copy;
2777 struct btrfs_root *tree_root = fs_info->tree_root;
2778 bool handle_error = false;
2779 int ret = 0;
2780 int i;
2781
2782 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2783 u64 generation;
2784 int level;
2785
2786 if (handle_error) {
2787 if (!IS_ERR(tree_root->node))
2788 free_extent_buffer(tree_root->node);
2789 tree_root->node = NULL;
2790
2791 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2792 break;
2793
2794 free_root_pointers(fs_info, 0);
2795
2796 /*
2797 * Don't use the log in recovery mode, it won't be
2798 * valid
2799 */
2800 btrfs_set_super_log_root(sb, 0);
2801
2802 /* We can't trust the free space cache either */
2803 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2804
2805 ret = read_backup_root(fs_info, i);
2806 backup_index = ret;
2807 if (ret < 0)
2808 return ret;
2809 }
2810 generation = btrfs_super_generation(sb);
2811 level = btrfs_super_root_level(sb);
2812 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2813 BTRFS_ROOT_TREE_OBJECTID,
2814 generation, level, NULL);
2815 if (IS_ERR(tree_root->node)) {
2816 handle_error = true;
2817 ret = PTR_ERR(tree_root->node);
2818 tree_root->node = NULL;
2819 btrfs_warn(fs_info, "couldn't read tree root");
2820 continue;
2821
2822 } else if (!extent_buffer_uptodate(tree_root->node)) {
2823 handle_error = true;
2824 ret = -EIO;
2825 btrfs_warn(fs_info, "error while reading tree root");
2826 continue;
2827 }
2828
2829 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2830 tree_root->commit_root = btrfs_root_node(tree_root);
2831 btrfs_set_root_refs(&tree_root->root_item, 1);
2832
2833 /*
2834 * No need to hold btrfs_root::objectid_mutex since the fs
2835 * hasn't been fully initialised and we are the only user
2836 */
2837 ret = btrfs_init_root_free_objectid(tree_root);
2838 if (ret < 0) {
2839 handle_error = true;
2840 continue;
2841 }
2842
2843 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2844
2845 ret = btrfs_read_roots(fs_info);
2846 if (ret < 0) {
2847 handle_error = true;
2848 continue;
2849 }
2850
2851 /* All successful */
2852 fs_info->generation = generation;
2853 fs_info->last_trans_committed = generation;
2854
2855 /* Always begin writing backup roots after the one being used */
2856 if (backup_index < 0) {
2857 fs_info->backup_root_index = 0;
2858 } else {
2859 fs_info->backup_root_index = backup_index + 1;
2860 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2861 }
2862 break;
2863 }
2864
2865 return ret;
2866}
2867
2868void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2869{
2870 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2871 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2872 INIT_LIST_HEAD(&fs_info->trans_list);
2873 INIT_LIST_HEAD(&fs_info->dead_roots);
2874 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2875 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2876 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2877 spin_lock_init(&fs_info->delalloc_root_lock);
2878 spin_lock_init(&fs_info->trans_lock);
2879 spin_lock_init(&fs_info->fs_roots_radix_lock);
2880 spin_lock_init(&fs_info->delayed_iput_lock);
2881 spin_lock_init(&fs_info->defrag_inodes_lock);
2882 spin_lock_init(&fs_info->super_lock);
2883 spin_lock_init(&fs_info->buffer_lock);
2884 spin_lock_init(&fs_info->unused_bgs_lock);
2885 spin_lock_init(&fs_info->treelog_bg_lock);
2886 rwlock_init(&fs_info->tree_mod_log_lock);
2887 mutex_init(&fs_info->unused_bg_unpin_mutex);
2888 mutex_init(&fs_info->reclaim_bgs_lock);
2889 mutex_init(&fs_info->reloc_mutex);
2890 mutex_init(&fs_info->delalloc_root_mutex);
2891 mutex_init(&fs_info->zoned_meta_io_lock);
2892 seqlock_init(&fs_info->profiles_lock);
2893
2894 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2895 INIT_LIST_HEAD(&fs_info->space_info);
2896 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2897 INIT_LIST_HEAD(&fs_info->unused_bgs);
2898 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2899#ifdef CONFIG_BTRFS_DEBUG
2900 INIT_LIST_HEAD(&fs_info->allocated_roots);
2901 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2902 spin_lock_init(&fs_info->eb_leak_lock);
2903#endif
2904 extent_map_tree_init(&fs_info->mapping_tree);
2905 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2906 BTRFS_BLOCK_RSV_GLOBAL);
2907 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2908 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2909 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2910 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2911 BTRFS_BLOCK_RSV_DELOPS);
2912 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2913 BTRFS_BLOCK_RSV_DELREFS);
2914
2915 atomic_set(&fs_info->async_delalloc_pages, 0);
2916 atomic_set(&fs_info->defrag_running, 0);
2917 atomic_set(&fs_info->reada_works_cnt, 0);
2918 atomic_set(&fs_info->nr_delayed_iputs, 0);
2919 atomic64_set(&fs_info->tree_mod_seq, 0);
2920 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2921 fs_info->metadata_ratio = 0;
2922 fs_info->defrag_inodes = RB_ROOT;
2923 atomic64_set(&fs_info->free_chunk_space, 0);
2924 fs_info->tree_mod_log = RB_ROOT;
2925 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2926 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2927 /* readahead state */
2928 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2929 spin_lock_init(&fs_info->reada_lock);
2930 btrfs_init_ref_verify(fs_info);
2931
2932 fs_info->thread_pool_size = min_t(unsigned long,
2933 num_online_cpus() + 2, 8);
2934
2935 INIT_LIST_HEAD(&fs_info->ordered_roots);
2936 spin_lock_init(&fs_info->ordered_root_lock);
2937
2938 btrfs_init_scrub(fs_info);
2939#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2940 fs_info->check_integrity_print_mask = 0;
2941#endif
2942 btrfs_init_balance(fs_info);
2943 btrfs_init_async_reclaim_work(fs_info);
2944
2945 spin_lock_init(&fs_info->block_group_cache_lock);
2946 fs_info->block_group_cache_tree = RB_ROOT;
2947 fs_info->first_logical_byte = (u64)-1;
2948
2949 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2950 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2951 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2952
2953 mutex_init(&fs_info->ordered_operations_mutex);
2954 mutex_init(&fs_info->tree_log_mutex);
2955 mutex_init(&fs_info->chunk_mutex);
2956 mutex_init(&fs_info->transaction_kthread_mutex);
2957 mutex_init(&fs_info->cleaner_mutex);
2958 mutex_init(&fs_info->ro_block_group_mutex);
2959 init_rwsem(&fs_info->commit_root_sem);
2960 init_rwsem(&fs_info->cleanup_work_sem);
2961 init_rwsem(&fs_info->subvol_sem);
2962 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2963
2964 btrfs_init_dev_replace_locks(fs_info);
2965 btrfs_init_qgroup(fs_info);
2966 btrfs_discard_init(fs_info);
2967
2968 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2969 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2970
2971 init_waitqueue_head(&fs_info->transaction_throttle);
2972 init_waitqueue_head(&fs_info->transaction_wait);
2973 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2974 init_waitqueue_head(&fs_info->async_submit_wait);
2975 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2976
2977 /* Usable values until the real ones are cached from the superblock */
2978 fs_info->nodesize = 4096;
2979 fs_info->sectorsize = 4096;
2980 fs_info->sectorsize_bits = ilog2(4096);
2981 fs_info->stripesize = 4096;
2982
2983 spin_lock_init(&fs_info->swapfile_pins_lock);
2984 fs_info->swapfile_pins = RB_ROOT;
2985
2986 spin_lock_init(&fs_info->send_reloc_lock);
2987 fs_info->send_in_progress = 0;
2988
2989 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2990 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2991}
2992
2993static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2994{
2995 int ret;
2996
2997 fs_info->sb = sb;
2998 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2999 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3000
3001 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3002 if (ret)
3003 return ret;
3004
3005 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3006 if (ret)
3007 return ret;
3008
3009 fs_info->dirty_metadata_batch = PAGE_SIZE *
3010 (1 + ilog2(nr_cpu_ids));
3011
3012 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3013 if (ret)
3014 return ret;
3015
3016 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3017 GFP_KERNEL);
3018 if (ret)
3019 return ret;
3020
3021 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3022 GFP_KERNEL);
3023 if (!fs_info->delayed_root)
3024 return -ENOMEM;
3025 btrfs_init_delayed_root(fs_info->delayed_root);
3026
3027 if (sb_rdonly(sb))
3028 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3029
3030 return btrfs_alloc_stripe_hash_table(fs_info);
3031}
3032
3033static int btrfs_uuid_rescan_kthread(void *data)
3034{
3035 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3036 int ret;
3037
3038 /*
3039 * 1st step is to iterate through the existing UUID tree and
3040 * to delete all entries that contain outdated data.
3041 * 2nd step is to add all missing entries to the UUID tree.
3042 */
3043 ret = btrfs_uuid_tree_iterate(fs_info);
3044 if (ret < 0) {
3045 if (ret != -EINTR)
3046 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3047 ret);
3048 up(&fs_info->uuid_tree_rescan_sem);
3049 return ret;
3050 }
3051 return btrfs_uuid_scan_kthread(data);
3052}
3053
3054static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3055{
3056 struct task_struct *task;
3057
3058 down(&fs_info->uuid_tree_rescan_sem);
3059 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3060 if (IS_ERR(task)) {
3061 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3062 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3063 up(&fs_info->uuid_tree_rescan_sem);
3064 return PTR_ERR(task);
3065 }
3066
3067 return 0;
3068}
3069
3070/*
3071 * Some options only have meaning at mount time and shouldn't persist across
3072 * remounts, or be displayed. Clear these at the end of mount and remount
3073 * code paths.
3074 */
3075void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3076{
3077 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3078 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3079}
3080
3081/*
3082 * Mounting logic specific to read-write file systems. Shared by open_ctree
3083 * and btrfs_remount when remounting from read-only to read-write.
3084 */
3085int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3086{
3087 int ret;
3088 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3089 bool clear_free_space_tree = false;
3090
3091 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3092 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3093 clear_free_space_tree = true;
3094 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3095 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3096 btrfs_warn(fs_info, "free space tree is invalid");
3097 clear_free_space_tree = true;
3098 }
3099
3100 if (clear_free_space_tree) {
3101 btrfs_info(fs_info, "clearing free space tree");
3102 ret = btrfs_clear_free_space_tree(fs_info);
3103 if (ret) {
3104 btrfs_warn(fs_info,
3105 "failed to clear free space tree: %d", ret);
3106 goto out;
3107 }
3108 }
3109
3110 /*
3111 * btrfs_find_orphan_roots() is responsible for finding all the dead
3112 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3113 * them into the fs_info->fs_roots_radix tree. This must be done before
3114 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3115 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3116 * item before the root's tree is deleted - this means that if we unmount
3117 * or crash before the deletion completes, on the next mount we will not
3118 * delete what remains of the tree because the orphan item does not
3119 * exists anymore, which is what tells us we have a pending deletion.
3120 */
3121 ret = btrfs_find_orphan_roots(fs_info);
3122 if (ret)
3123 goto out;
3124
3125 ret = btrfs_cleanup_fs_roots(fs_info);
3126 if (ret)
3127 goto out;
3128
3129 down_read(&fs_info->cleanup_work_sem);
3130 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3131 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3132 up_read(&fs_info->cleanup_work_sem);
3133 goto out;
3134 }
3135 up_read(&fs_info->cleanup_work_sem);
3136
3137 mutex_lock(&fs_info->cleaner_mutex);
3138 ret = btrfs_recover_relocation(fs_info->tree_root);
3139 mutex_unlock(&fs_info->cleaner_mutex);
3140 if (ret < 0) {
3141 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3142 goto out;
3143 }
3144
3145 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3146 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3147 btrfs_info(fs_info, "creating free space tree");
3148 ret = btrfs_create_free_space_tree(fs_info);
3149 if (ret) {
3150 btrfs_warn(fs_info,
3151 "failed to create free space tree: %d", ret);
3152 goto out;
3153 }
3154 }
3155
3156 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3157 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3158 if (ret)
3159 goto out;
3160 }
3161
3162 ret = btrfs_resume_balance_async(fs_info);
3163 if (ret)
3164 goto out;
3165
3166 ret = btrfs_resume_dev_replace_async(fs_info);
3167 if (ret) {
3168 btrfs_warn(fs_info, "failed to resume dev_replace");
3169 goto out;
3170 }
3171
3172 btrfs_qgroup_rescan_resume(fs_info);
3173
3174 if (!fs_info->uuid_root) {
3175 btrfs_info(fs_info, "creating UUID tree");
3176 ret = btrfs_create_uuid_tree(fs_info);
3177 if (ret) {
3178 btrfs_warn(fs_info,
3179 "failed to create the UUID tree %d", ret);
3180 goto out;
3181 }
3182 }
3183
3184out:
3185 return ret;
3186}
3187
3188int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3189 char *options)
3190{
3191 u32 sectorsize;
3192 u32 nodesize;
3193 u32 stripesize;
3194 u64 generation;
3195 u64 features;
3196 u16 csum_type;
3197 struct btrfs_super_block *disk_super;
3198 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3199 struct btrfs_root *tree_root;
3200 struct btrfs_root *chunk_root;
3201 int ret;
3202 int err = -EINVAL;
3203 int level;
3204
3205 ret = init_mount_fs_info(fs_info, sb);
3206 if (ret) {
3207 err = ret;
3208 goto fail;
3209 }
3210
3211 /* These need to be init'ed before we start creating inodes and such. */
3212 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3213 GFP_KERNEL);
3214 fs_info->tree_root = tree_root;
3215 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3216 GFP_KERNEL);
3217 fs_info->chunk_root = chunk_root;
3218 if (!tree_root || !chunk_root) {
3219 err = -ENOMEM;
3220 goto fail;
3221 }
3222
3223 fs_info->btree_inode = new_inode(sb);
3224 if (!fs_info->btree_inode) {
3225 err = -ENOMEM;
3226 goto fail;
3227 }
3228 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3229 btrfs_init_btree_inode(fs_info);
3230
3231 invalidate_bdev(fs_devices->latest_bdev);
3232
3233 /*
3234 * Read super block and check the signature bytes only
3235 */
3236 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
3237 if (IS_ERR(disk_super)) {
3238 err = PTR_ERR(disk_super);
3239 goto fail_alloc;
3240 }
3241
3242 /*
3243 * Verify the type first, if that or the checksum value are
3244 * corrupted, we'll find out
3245 */
3246 csum_type = btrfs_super_csum_type(disk_super);
3247 if (!btrfs_supported_super_csum(csum_type)) {
3248 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3249 csum_type);
3250 err = -EINVAL;
3251 btrfs_release_disk_super(disk_super);
3252 goto fail_alloc;
3253 }
3254
3255 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3256
3257 ret = btrfs_init_csum_hash(fs_info, csum_type);
3258 if (ret) {
3259 err = ret;
3260 btrfs_release_disk_super(disk_super);
3261 goto fail_alloc;
3262 }
3263
3264 /*
3265 * We want to check superblock checksum, the type is stored inside.
3266 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3267 */
3268 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3269 btrfs_err(fs_info, "superblock checksum mismatch");
3270 err = -EINVAL;
3271 btrfs_release_disk_super(disk_super);
3272 goto fail_alloc;
3273 }
3274
3275 /*
3276 * super_copy is zeroed at allocation time and we never touch the
3277 * following bytes up to INFO_SIZE, the checksum is calculated from
3278 * the whole block of INFO_SIZE
3279 */
3280 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3281 btrfs_release_disk_super(disk_super);
3282
3283 disk_super = fs_info->super_copy;
3284
3285
3286 features = btrfs_super_flags(disk_super);
3287 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3288 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3289 btrfs_set_super_flags(disk_super, features);
3290 btrfs_info(fs_info,
3291 "found metadata UUID change in progress flag, clearing");
3292 }
3293
3294 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3295 sizeof(*fs_info->super_for_commit));
3296
3297 ret = btrfs_validate_mount_super(fs_info);
3298 if (ret) {
3299 btrfs_err(fs_info, "superblock contains fatal errors");
3300 err = -EINVAL;
3301 goto fail_alloc;
3302 }
3303
3304 if (!btrfs_super_root(disk_super))
3305 goto fail_alloc;
3306
3307 /* check FS state, whether FS is broken. */
3308 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3309 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3310
3311 /*
3312 * In the long term, we'll store the compression type in the super
3313 * block, and it'll be used for per file compression control.
3314 */
3315 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3316
3317 /*
3318 * Flag our filesystem as having big metadata blocks if they are bigger
3319 * than the page size.
3320 */
3321 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3322 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3323 btrfs_info(fs_info,
3324 "flagging fs with big metadata feature");
3325 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3326 }
3327
3328 /* Set up fs_info before parsing mount options */
3329 nodesize = btrfs_super_nodesize(disk_super);
3330 sectorsize = btrfs_super_sectorsize(disk_super);
3331 stripesize = sectorsize;
3332 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3333 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3334
3335 fs_info->nodesize = nodesize;
3336 fs_info->sectorsize = sectorsize;
3337 fs_info->sectorsize_bits = ilog2(sectorsize);
3338 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3339 fs_info->stripesize = stripesize;
3340
3341 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3342 if (ret) {
3343 err = ret;
3344 goto fail_alloc;
3345 }
3346
3347 features = btrfs_super_incompat_flags(disk_super) &
3348 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3349 if (features) {
3350 btrfs_err(fs_info,
3351 "cannot mount because of unsupported optional features (%llx)",
3352 features);
3353 err = -EINVAL;
3354 goto fail_alloc;
3355 }
3356
3357 features = btrfs_super_incompat_flags(disk_super);
3358 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3359 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3360 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3361 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3362 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3363
3364 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3365 btrfs_info(fs_info, "has skinny extents");
3366
3367 /*
3368 * mixed block groups end up with duplicate but slightly offset
3369 * extent buffers for the same range. It leads to corruptions
3370 */
3371 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3372 (sectorsize != nodesize)) {
3373 btrfs_err(fs_info,
3374"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3375 nodesize, sectorsize);
3376 goto fail_alloc;
3377 }
3378
3379 /*
3380 * Needn't use the lock because there is no other task which will
3381 * update the flag.
3382 */
3383 btrfs_set_super_incompat_flags(disk_super, features);
3384
3385 features = btrfs_super_compat_ro_flags(disk_super) &
3386 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3387 if (!sb_rdonly(sb) && features) {
3388 btrfs_err(fs_info,
3389 "cannot mount read-write because of unsupported optional features (%llx)",
3390 features);
3391 err = -EINVAL;
3392 goto fail_alloc;
3393 }
3394
3395 /* For 4K sector size support, it's only read-only */
3396 if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) {
3397 if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) {
3398 btrfs_err(fs_info,
3399 "subpage sectorsize %u only supported read-only for page size %lu",
3400 sectorsize, PAGE_SIZE);
3401 err = -EINVAL;
3402 goto fail_alloc;
3403 }
3404 }
3405
3406 ret = btrfs_init_workqueues(fs_info, fs_devices);
3407 if (ret) {
3408 err = ret;
3409 goto fail_sb_buffer;
3410 }
3411
3412 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3413 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3414
3415 sb->s_blocksize = sectorsize;
3416 sb->s_blocksize_bits = blksize_bits(sectorsize);
3417 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3418
3419 mutex_lock(&fs_info->chunk_mutex);
3420 ret = btrfs_read_sys_array(fs_info);
3421 mutex_unlock(&fs_info->chunk_mutex);
3422 if (ret) {
3423 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3424 goto fail_sb_buffer;
3425 }
3426
3427 generation = btrfs_super_chunk_root_generation(disk_super);
3428 level = btrfs_super_chunk_root_level(disk_super);
3429
3430 chunk_root->node = read_tree_block(fs_info,
3431 btrfs_super_chunk_root(disk_super),
3432 BTRFS_CHUNK_TREE_OBJECTID,
3433 generation, level, NULL);
3434 if (IS_ERR(chunk_root->node) ||
3435 !extent_buffer_uptodate(chunk_root->node)) {
3436 btrfs_err(fs_info, "failed to read chunk root");
3437 if (!IS_ERR(chunk_root->node))
3438 free_extent_buffer(chunk_root->node);
3439 chunk_root->node = NULL;
3440 goto fail_tree_roots;
3441 }
3442 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3443 chunk_root->commit_root = btrfs_root_node(chunk_root);
3444
3445 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3446 offsetof(struct btrfs_header, chunk_tree_uuid),
3447 BTRFS_UUID_SIZE);
3448
3449 ret = btrfs_read_chunk_tree(fs_info);
3450 if (ret) {
3451 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3452 goto fail_tree_roots;
3453 }
3454
3455 /*
3456 * At this point we know all the devices that make this filesystem,
3457 * including the seed devices but we don't know yet if the replace
3458 * target is required. So free devices that are not part of this
3459 * filesystem but skip the replace target device which is checked
3460 * below in btrfs_init_dev_replace().
3461 */
3462 btrfs_free_extra_devids(fs_devices);
3463 if (!fs_devices->latest_bdev) {
3464 btrfs_err(fs_info, "failed to read devices");
3465 goto fail_tree_roots;
3466 }
3467
3468 ret = init_tree_roots(fs_info);
3469 if (ret)
3470 goto fail_tree_roots;
3471
3472 /*
3473 * Get zone type information of zoned block devices. This will also
3474 * handle emulation of a zoned filesystem if a regular device has the
3475 * zoned incompat feature flag set.
3476 */
3477 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3478 if (ret) {
3479 btrfs_err(fs_info,
3480 "zoned: failed to read device zone info: %d",
3481 ret);
3482 goto fail_block_groups;
3483 }
3484
3485 /*
3486 * If we have a uuid root and we're not being told to rescan we need to
3487 * check the generation here so we can set the
3488 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3489 * transaction during a balance or the log replay without updating the
3490 * uuid generation, and then if we crash we would rescan the uuid tree,
3491 * even though it was perfectly fine.
3492 */
3493 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3494 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3495 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3496
3497 ret = btrfs_verify_dev_extents(fs_info);
3498 if (ret) {
3499 btrfs_err(fs_info,
3500 "failed to verify dev extents against chunks: %d",
3501 ret);
3502 goto fail_block_groups;
3503 }
3504 ret = btrfs_recover_balance(fs_info);
3505 if (ret) {
3506 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3507 goto fail_block_groups;
3508 }
3509
3510 ret = btrfs_init_dev_stats(fs_info);
3511 if (ret) {
3512 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3513 goto fail_block_groups;
3514 }
3515
3516 ret = btrfs_init_dev_replace(fs_info);
3517 if (ret) {
3518 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3519 goto fail_block_groups;
3520 }
3521
3522 ret = btrfs_check_zoned_mode(fs_info);
3523 if (ret) {
3524 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3525 ret);
3526 goto fail_block_groups;
3527 }
3528
3529 ret = btrfs_sysfs_add_fsid(fs_devices);
3530 if (ret) {
3531 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3532 ret);
3533 goto fail_block_groups;
3534 }
3535
3536 ret = btrfs_sysfs_add_mounted(fs_info);
3537 if (ret) {
3538 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3539 goto fail_fsdev_sysfs;
3540 }
3541
3542 ret = btrfs_init_space_info(fs_info);
3543 if (ret) {
3544 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3545 goto fail_sysfs;
3546 }
3547
3548 ret = btrfs_read_block_groups(fs_info);
3549 if (ret) {
3550 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3551 goto fail_sysfs;
3552 }
3553
3554 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3555 btrfs_warn(fs_info,
3556 "writable mount is not allowed due to too many missing devices");
3557 goto fail_sysfs;
3558 }
3559
3560 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3561 "btrfs-cleaner");
3562 if (IS_ERR(fs_info->cleaner_kthread))
3563 goto fail_sysfs;
3564
3565 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3566 tree_root,
3567 "btrfs-transaction");
3568 if (IS_ERR(fs_info->transaction_kthread))
3569 goto fail_cleaner;
3570
3571 if (!btrfs_test_opt(fs_info, NOSSD) &&
3572 !fs_info->fs_devices->rotating) {
3573 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3574 }
3575
3576 /*
3577 * Mount does not set all options immediately, we can do it now and do
3578 * not have to wait for transaction commit
3579 */
3580 btrfs_apply_pending_changes(fs_info);
3581
3582#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3583 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3584 ret = btrfsic_mount(fs_info, fs_devices,
3585 btrfs_test_opt(fs_info,
3586 CHECK_INTEGRITY_DATA) ? 1 : 0,
3587 fs_info->check_integrity_print_mask);
3588 if (ret)
3589 btrfs_warn(fs_info,
3590 "failed to initialize integrity check module: %d",
3591 ret);
3592 }
3593#endif
3594 ret = btrfs_read_qgroup_config(fs_info);
3595 if (ret)
3596 goto fail_trans_kthread;
3597
3598 if (btrfs_build_ref_tree(fs_info))
3599 btrfs_err(fs_info, "couldn't build ref tree");
3600
3601 /* do not make disk changes in broken FS or nologreplay is given */
3602 if (btrfs_super_log_root(disk_super) != 0 &&
3603 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3604 btrfs_info(fs_info, "start tree-log replay");
3605 ret = btrfs_replay_log(fs_info, fs_devices);
3606 if (ret) {
3607 err = ret;
3608 goto fail_qgroup;
3609 }
3610 }
3611
3612 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3613 if (IS_ERR(fs_info->fs_root)) {
3614 err = PTR_ERR(fs_info->fs_root);
3615 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3616 fs_info->fs_root = NULL;
3617 goto fail_qgroup;
3618 }
3619
3620 if (sb_rdonly(sb))
3621 goto clear_oneshot;
3622
3623 ret = btrfs_start_pre_rw_mount(fs_info);
3624 if (ret) {
3625 close_ctree(fs_info);
3626 return ret;
3627 }
3628 btrfs_discard_resume(fs_info);
3629
3630 if (fs_info->uuid_root &&
3631 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3632 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3633 btrfs_info(fs_info, "checking UUID tree");
3634 ret = btrfs_check_uuid_tree(fs_info);
3635 if (ret) {
3636 btrfs_warn(fs_info,
3637 "failed to check the UUID tree: %d", ret);
3638 close_ctree(fs_info);
3639 return ret;
3640 }
3641 }
3642
3643 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3644
3645clear_oneshot:
3646 btrfs_clear_oneshot_options(fs_info);
3647 return 0;
3648
3649fail_qgroup:
3650 btrfs_free_qgroup_config(fs_info);
3651fail_trans_kthread:
3652 kthread_stop(fs_info->transaction_kthread);
3653 btrfs_cleanup_transaction(fs_info);
3654 btrfs_free_fs_roots(fs_info);
3655fail_cleaner:
3656 kthread_stop(fs_info->cleaner_kthread);
3657
3658 /*
3659 * make sure we're done with the btree inode before we stop our
3660 * kthreads
3661 */
3662 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3663
3664fail_sysfs:
3665 btrfs_sysfs_remove_mounted(fs_info);
3666
3667fail_fsdev_sysfs:
3668 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3669
3670fail_block_groups:
3671 btrfs_put_block_group_cache(fs_info);
3672
3673fail_tree_roots:
3674 if (fs_info->data_reloc_root)
3675 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3676 free_root_pointers(fs_info, true);
3677 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3678
3679fail_sb_buffer:
3680 btrfs_stop_all_workers(fs_info);
3681 btrfs_free_block_groups(fs_info);
3682fail_alloc:
3683 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3684
3685 iput(fs_info->btree_inode);
3686fail:
3687 btrfs_close_devices(fs_info->fs_devices);
3688 return err;
3689}
3690ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3691
3692static void btrfs_end_super_write(struct bio *bio)
3693{
3694 struct btrfs_device *device = bio->bi_private;
3695 struct bio_vec *bvec;
3696 struct bvec_iter_all iter_all;
3697 struct page *page;
3698
3699 bio_for_each_segment_all(bvec, bio, iter_all) {
3700 page = bvec->bv_page;
3701
3702 if (bio->bi_status) {
3703 btrfs_warn_rl_in_rcu(device->fs_info,
3704 "lost page write due to IO error on %s (%d)",
3705 rcu_str_deref(device->name),
3706 blk_status_to_errno(bio->bi_status));
3707 ClearPageUptodate(page);
3708 SetPageError(page);
3709 btrfs_dev_stat_inc_and_print(device,
3710 BTRFS_DEV_STAT_WRITE_ERRS);
3711 } else {
3712 SetPageUptodate(page);
3713 }
3714
3715 put_page(page);
3716 unlock_page(page);
3717 }
3718
3719 bio_put(bio);
3720}
3721
3722struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3723 int copy_num)
3724{
3725 struct btrfs_super_block *super;
3726 struct page *page;
3727 u64 bytenr, bytenr_orig;
3728 struct address_space *mapping = bdev->bd_inode->i_mapping;
3729 int ret;
3730
3731 bytenr_orig = btrfs_sb_offset(copy_num);
3732 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3733 if (ret == -ENOENT)
3734 return ERR_PTR(-EINVAL);
3735 else if (ret)
3736 return ERR_PTR(ret);
3737
3738 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3739 return ERR_PTR(-EINVAL);
3740
3741 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3742 if (IS_ERR(page))
3743 return ERR_CAST(page);
3744
3745 super = page_address(page);
3746 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3747 btrfs_release_disk_super(super);
3748 return ERR_PTR(-ENODATA);
3749 }
3750
3751 if (btrfs_super_bytenr(super) != bytenr_orig) {
3752 btrfs_release_disk_super(super);
3753 return ERR_PTR(-EINVAL);
3754 }
3755
3756 return super;
3757}
3758
3759
3760struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3761{
3762 struct btrfs_super_block *super, *latest = NULL;
3763 int i;
3764 u64 transid = 0;
3765
3766 /* we would like to check all the supers, but that would make
3767 * a btrfs mount succeed after a mkfs from a different FS.
3768 * So, we need to add a special mount option to scan for
3769 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3770 */
3771 for (i = 0; i < 1; i++) {
3772 super = btrfs_read_dev_one_super(bdev, i);
3773 if (IS_ERR(super))
3774 continue;
3775
3776 if (!latest || btrfs_super_generation(super) > transid) {
3777 if (latest)
3778 btrfs_release_disk_super(super);
3779
3780 latest = super;
3781 transid = btrfs_super_generation(super);
3782 }
3783 }
3784
3785 return super;
3786}
3787
3788/*
3789 * Write superblock @sb to the @device. Do not wait for completion, all the
3790 * pages we use for writing are locked.
3791 *
3792 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3793 * the expected device size at commit time. Note that max_mirrors must be
3794 * same for write and wait phases.
3795 *
3796 * Return number of errors when page is not found or submission fails.
3797 */
3798static int write_dev_supers(struct btrfs_device *device,
3799 struct btrfs_super_block *sb, int max_mirrors)
3800{
3801 struct btrfs_fs_info *fs_info = device->fs_info;
3802 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3803 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3804 int i;
3805 int errors = 0;
3806 int ret;
3807 u64 bytenr, bytenr_orig;
3808
3809 if (max_mirrors == 0)
3810 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3811
3812 shash->tfm = fs_info->csum_shash;
3813
3814 for (i = 0; i < max_mirrors; i++) {
3815 struct page *page;
3816 struct bio *bio;
3817 struct btrfs_super_block *disk_super;
3818
3819 bytenr_orig = btrfs_sb_offset(i);
3820 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3821 if (ret == -ENOENT) {
3822 continue;
3823 } else if (ret < 0) {
3824 btrfs_err(device->fs_info,
3825 "couldn't get super block location for mirror %d",
3826 i);
3827 errors++;
3828 continue;
3829 }
3830 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3831 device->commit_total_bytes)
3832 break;
3833
3834 btrfs_set_super_bytenr(sb, bytenr_orig);
3835
3836 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3837 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3838 sb->csum);
3839
3840 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3841 GFP_NOFS);
3842 if (!page) {
3843 btrfs_err(device->fs_info,
3844 "couldn't get super block page for bytenr %llu",
3845 bytenr);
3846 errors++;
3847 continue;
3848 }
3849
3850 /* Bump the refcount for wait_dev_supers() */
3851 get_page(page);
3852
3853 disk_super = page_address(page);
3854 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3855
3856 /*
3857 * Directly use bios here instead of relying on the page cache
3858 * to do I/O, so we don't lose the ability to do integrity
3859 * checking.
3860 */
3861 bio = bio_alloc(GFP_NOFS, 1);
3862 bio_set_dev(bio, device->bdev);
3863 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3864 bio->bi_private = device;
3865 bio->bi_end_io = btrfs_end_super_write;
3866 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3867 offset_in_page(bytenr));
3868
3869 /*
3870 * We FUA only the first super block. The others we allow to
3871 * go down lazy and there's a short window where the on-disk
3872 * copies might still contain the older version.
3873 */
3874 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3875 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3876 bio->bi_opf |= REQ_FUA;
3877
3878 btrfsic_submit_bio(bio);
3879 btrfs_advance_sb_log(device, i);
3880 }
3881 return errors < i ? 0 : -1;
3882}
3883
3884/*
3885 * Wait for write completion of superblocks done by write_dev_supers,
3886 * @max_mirrors same for write and wait phases.
3887 *
3888 * Return number of errors when page is not found or not marked up to
3889 * date.
3890 */
3891static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3892{
3893 int i;
3894 int errors = 0;
3895 bool primary_failed = false;
3896 int ret;
3897 u64 bytenr;
3898
3899 if (max_mirrors == 0)
3900 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3901
3902 for (i = 0; i < max_mirrors; i++) {
3903 struct page *page;
3904
3905 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3906 if (ret == -ENOENT) {
3907 break;
3908 } else if (ret < 0) {
3909 errors++;
3910 if (i == 0)
3911 primary_failed = true;
3912 continue;
3913 }
3914 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3915 device->commit_total_bytes)
3916 break;
3917
3918 page = find_get_page(device->bdev->bd_inode->i_mapping,
3919 bytenr >> PAGE_SHIFT);
3920 if (!page) {
3921 errors++;
3922 if (i == 0)
3923 primary_failed = true;
3924 continue;
3925 }
3926 /* Page is submitted locked and unlocked once the IO completes */
3927 wait_on_page_locked(page);
3928 if (PageError(page)) {
3929 errors++;
3930 if (i == 0)
3931 primary_failed = true;
3932 }
3933
3934 /* Drop our reference */
3935 put_page(page);
3936
3937 /* Drop the reference from the writing run */
3938 put_page(page);
3939 }
3940
3941 /* log error, force error return */
3942 if (primary_failed) {
3943 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3944 device->devid);
3945 return -1;
3946 }
3947
3948 return errors < i ? 0 : -1;
3949}
3950
3951/*
3952 * endio for the write_dev_flush, this will wake anyone waiting
3953 * for the barrier when it is done
3954 */
3955static void btrfs_end_empty_barrier(struct bio *bio)
3956{
3957 complete(bio->bi_private);
3958}
3959
3960/*
3961 * Submit a flush request to the device if it supports it. Error handling is
3962 * done in the waiting counterpart.
3963 */
3964static void write_dev_flush(struct btrfs_device *device)
3965{
3966 struct request_queue *q = bdev_get_queue(device->bdev);
3967 struct bio *bio = device->flush_bio;
3968
3969 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3970 return;
3971
3972 bio_reset(bio);
3973 bio->bi_end_io = btrfs_end_empty_barrier;
3974 bio_set_dev(bio, device->bdev);
3975 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3976 init_completion(&device->flush_wait);
3977 bio->bi_private = &device->flush_wait;
3978
3979 btrfsic_submit_bio(bio);
3980 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3981}
3982
3983/*
3984 * If the flush bio has been submitted by write_dev_flush, wait for it.
3985 */
3986static blk_status_t wait_dev_flush(struct btrfs_device *device)
3987{
3988 struct bio *bio = device->flush_bio;
3989
3990 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3991 return BLK_STS_OK;
3992
3993 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3994 wait_for_completion_io(&device->flush_wait);
3995
3996 return bio->bi_status;
3997}
3998
3999static int check_barrier_error(struct btrfs_fs_info *fs_info)
4000{
4001 if (!btrfs_check_rw_degradable(fs_info, NULL))
4002 return -EIO;
4003 return 0;
4004}
4005
4006/*
4007 * send an empty flush down to each device in parallel,
4008 * then wait for them
4009 */
4010static int barrier_all_devices(struct btrfs_fs_info *info)
4011{
4012 struct list_head *head;
4013 struct btrfs_device *dev;
4014 int errors_wait = 0;
4015 blk_status_t ret;
4016
4017 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4018 /* send down all the barriers */
4019 head = &info->fs_devices->devices;
4020 list_for_each_entry(dev, head, dev_list) {
4021 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4022 continue;
4023 if (!dev->bdev)
4024 continue;
4025 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4026 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4027 continue;
4028
4029 write_dev_flush(dev);
4030 dev->last_flush_error = BLK_STS_OK;
4031 }
4032
4033 /* wait for all the barriers */
4034 list_for_each_entry(dev, head, dev_list) {
4035 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4036 continue;
4037 if (!dev->bdev) {
4038 errors_wait++;
4039 continue;
4040 }
4041 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4042 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4043 continue;
4044
4045 ret = wait_dev_flush(dev);
4046 if (ret) {
4047 dev->last_flush_error = ret;
4048 btrfs_dev_stat_inc_and_print(dev,
4049 BTRFS_DEV_STAT_FLUSH_ERRS);
4050 errors_wait++;
4051 }
4052 }
4053
4054 if (errors_wait) {
4055 /*
4056 * At some point we need the status of all disks
4057 * to arrive at the volume status. So error checking
4058 * is being pushed to a separate loop.
4059 */
4060 return check_barrier_error(info);
4061 }
4062 return 0;
4063}
4064
4065int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4066{
4067 int raid_type;
4068 int min_tolerated = INT_MAX;
4069
4070 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4071 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4072 min_tolerated = min_t(int, min_tolerated,
4073 btrfs_raid_array[BTRFS_RAID_SINGLE].
4074 tolerated_failures);
4075
4076 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4077 if (raid_type == BTRFS_RAID_SINGLE)
4078 continue;
4079 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4080 continue;
4081 min_tolerated = min_t(int, min_tolerated,
4082 btrfs_raid_array[raid_type].
4083 tolerated_failures);
4084 }
4085
4086 if (min_tolerated == INT_MAX) {
4087 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4088 min_tolerated = 0;
4089 }
4090
4091 return min_tolerated;
4092}
4093
4094int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4095{
4096 struct list_head *head;
4097 struct btrfs_device *dev;
4098 struct btrfs_super_block *sb;
4099 struct btrfs_dev_item *dev_item;
4100 int ret;
4101 int do_barriers;
4102 int max_errors;
4103 int total_errors = 0;
4104 u64 flags;
4105
4106 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4107
4108 /*
4109 * max_mirrors == 0 indicates we're from commit_transaction,
4110 * not from fsync where the tree roots in fs_info have not
4111 * been consistent on disk.
4112 */
4113 if (max_mirrors == 0)
4114 backup_super_roots(fs_info);
4115
4116 sb = fs_info->super_for_commit;
4117 dev_item = &sb->dev_item;
4118
4119 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4120 head = &fs_info->fs_devices->devices;
4121 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4122
4123 if (do_barriers) {
4124 ret = barrier_all_devices(fs_info);
4125 if (ret) {
4126 mutex_unlock(
4127 &fs_info->fs_devices->device_list_mutex);
4128 btrfs_handle_fs_error(fs_info, ret,
4129 "errors while submitting device barriers.");
4130 return ret;
4131 }
4132 }
4133
4134 list_for_each_entry(dev, head, dev_list) {
4135 if (!dev->bdev) {
4136 total_errors++;
4137 continue;
4138 }
4139 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4140 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4141 continue;
4142
4143 btrfs_set_stack_device_generation(dev_item, 0);
4144 btrfs_set_stack_device_type(dev_item, dev->type);
4145 btrfs_set_stack_device_id(dev_item, dev->devid);
4146 btrfs_set_stack_device_total_bytes(dev_item,
4147 dev->commit_total_bytes);
4148 btrfs_set_stack_device_bytes_used(dev_item,
4149 dev->commit_bytes_used);
4150 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4151 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4152 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4153 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4154 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4155 BTRFS_FSID_SIZE);
4156
4157 flags = btrfs_super_flags(sb);
4158 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4159
4160 ret = btrfs_validate_write_super(fs_info, sb);
4161 if (ret < 0) {
4162 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4163 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4164 "unexpected superblock corruption detected");
4165 return -EUCLEAN;
4166 }
4167
4168 ret = write_dev_supers(dev, sb, max_mirrors);
4169 if (ret)
4170 total_errors++;
4171 }
4172 if (total_errors > max_errors) {
4173 btrfs_err(fs_info, "%d errors while writing supers",
4174 total_errors);
4175 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4176
4177 /* FUA is masked off if unsupported and can't be the reason */
4178 btrfs_handle_fs_error(fs_info, -EIO,
4179 "%d errors while writing supers",
4180 total_errors);
4181 return -EIO;
4182 }
4183
4184 total_errors = 0;
4185 list_for_each_entry(dev, head, dev_list) {
4186 if (!dev->bdev)
4187 continue;
4188 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4189 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4190 continue;
4191
4192 ret = wait_dev_supers(dev, max_mirrors);
4193 if (ret)
4194 total_errors++;
4195 }
4196 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4197 if (total_errors > max_errors) {
4198 btrfs_handle_fs_error(fs_info, -EIO,
4199 "%d errors while writing supers",
4200 total_errors);
4201 return -EIO;
4202 }
4203 return 0;
4204}
4205
4206/* Drop a fs root from the radix tree and free it. */
4207void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4208 struct btrfs_root *root)
4209{
4210 bool drop_ref = false;
4211
4212 spin_lock(&fs_info->fs_roots_radix_lock);
4213 radix_tree_delete(&fs_info->fs_roots_radix,
4214 (unsigned long)root->root_key.objectid);
4215 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4216 drop_ref = true;
4217 spin_unlock(&fs_info->fs_roots_radix_lock);
4218
4219 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4220 ASSERT(root->log_root == NULL);
4221 if (root->reloc_root) {
4222 btrfs_put_root(root->reloc_root);
4223 root->reloc_root = NULL;
4224 }
4225 }
4226
4227 if (drop_ref)
4228 btrfs_put_root(root);
4229}
4230
4231int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4232{
4233 u64 root_objectid = 0;
4234 struct btrfs_root *gang[8];
4235 int i = 0;
4236 int err = 0;
4237 unsigned int ret = 0;
4238
4239 while (1) {
4240 spin_lock(&fs_info->fs_roots_radix_lock);
4241 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4242 (void **)gang, root_objectid,
4243 ARRAY_SIZE(gang));
4244 if (!ret) {
4245 spin_unlock(&fs_info->fs_roots_radix_lock);
4246 break;
4247 }
4248 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4249
4250 for (i = 0; i < ret; i++) {
4251 /* Avoid to grab roots in dead_roots */
4252 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4253 gang[i] = NULL;
4254 continue;
4255 }
4256 /* grab all the search result for later use */
4257 gang[i] = btrfs_grab_root(gang[i]);
4258 }
4259 spin_unlock(&fs_info->fs_roots_radix_lock);
4260
4261 for (i = 0; i < ret; i++) {
4262 if (!gang[i])
4263 continue;
4264 root_objectid = gang[i]->root_key.objectid;
4265 err = btrfs_orphan_cleanup(gang[i]);
4266 if (err)
4267 break;
4268 btrfs_put_root(gang[i]);
4269 }
4270 root_objectid++;
4271 }
4272
4273 /* release the uncleaned roots due to error */
4274 for (; i < ret; i++) {
4275 if (gang[i])
4276 btrfs_put_root(gang[i]);
4277 }
4278 return err;
4279}
4280
4281int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4282{
4283 struct btrfs_root *root = fs_info->tree_root;
4284 struct btrfs_trans_handle *trans;
4285
4286 mutex_lock(&fs_info->cleaner_mutex);
4287 btrfs_run_delayed_iputs(fs_info);
4288 mutex_unlock(&fs_info->cleaner_mutex);
4289 wake_up_process(fs_info->cleaner_kthread);
4290
4291 /* wait until ongoing cleanup work done */
4292 down_write(&fs_info->cleanup_work_sem);
4293 up_write(&fs_info->cleanup_work_sem);
4294
4295 trans = btrfs_join_transaction(root);
4296 if (IS_ERR(trans))
4297 return PTR_ERR(trans);
4298 return btrfs_commit_transaction(trans);
4299}
4300
4301void __cold close_ctree(struct btrfs_fs_info *fs_info)
4302{
4303 int ret;
4304
4305 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4306 /*
4307 * We don't want the cleaner to start new transactions, add more delayed
4308 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4309 * because that frees the task_struct, and the transaction kthread might
4310 * still try to wake up the cleaner.
4311 */
4312 kthread_park(fs_info->cleaner_kthread);
4313
4314 /* wait for the qgroup rescan worker to stop */
4315 btrfs_qgroup_wait_for_completion(fs_info, false);
4316
4317 /* wait for the uuid_scan task to finish */
4318 down(&fs_info->uuid_tree_rescan_sem);
4319 /* avoid complains from lockdep et al., set sem back to initial state */
4320 up(&fs_info->uuid_tree_rescan_sem);
4321
4322 /* pause restriper - we want to resume on mount */
4323 btrfs_pause_balance(fs_info);
4324
4325 btrfs_dev_replace_suspend_for_unmount(fs_info);
4326
4327 btrfs_scrub_cancel(fs_info);
4328
4329 /* wait for any defraggers to finish */
4330 wait_event(fs_info->transaction_wait,
4331 (atomic_read(&fs_info->defrag_running) == 0));
4332
4333 /* clear out the rbtree of defraggable inodes */
4334 btrfs_cleanup_defrag_inodes(fs_info);
4335
4336 cancel_work_sync(&fs_info->async_reclaim_work);
4337 cancel_work_sync(&fs_info->async_data_reclaim_work);
4338 cancel_work_sync(&fs_info->preempt_reclaim_work);
4339
4340 cancel_work_sync(&fs_info->reclaim_bgs_work);
4341
4342 /* Cancel or finish ongoing discard work */
4343 btrfs_discard_cleanup(fs_info);
4344
4345 if (!sb_rdonly(fs_info->sb)) {
4346 /*
4347 * The cleaner kthread is stopped, so do one final pass over
4348 * unused block groups.
4349 */
4350 btrfs_delete_unused_bgs(fs_info);
4351
4352 /*
4353 * There might be existing delayed inode workers still running
4354 * and holding an empty delayed inode item. We must wait for
4355 * them to complete first because they can create a transaction.
4356 * This happens when someone calls btrfs_balance_delayed_items()
4357 * and then a transaction commit runs the same delayed nodes
4358 * before any delayed worker has done something with the nodes.
4359 * We must wait for any worker here and not at transaction
4360 * commit time since that could cause a deadlock.
4361 * This is a very rare case.
4362 */
4363 btrfs_flush_workqueue(fs_info->delayed_workers);
4364
4365 ret = btrfs_commit_super(fs_info);
4366 if (ret)
4367 btrfs_err(fs_info, "commit super ret %d", ret);
4368 }
4369
4370 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4371 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4372 btrfs_error_commit_super(fs_info);
4373
4374 kthread_stop(fs_info->transaction_kthread);
4375 kthread_stop(fs_info->cleaner_kthread);
4376
4377 ASSERT(list_empty(&fs_info->delayed_iputs));
4378 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4379
4380 if (btrfs_check_quota_leak(fs_info)) {
4381 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4382 btrfs_err(fs_info, "qgroup reserved space leaked");
4383 }
4384
4385 btrfs_free_qgroup_config(fs_info);
4386 ASSERT(list_empty(&fs_info->delalloc_roots));
4387
4388 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4389 btrfs_info(fs_info, "at unmount delalloc count %lld",
4390 percpu_counter_sum(&fs_info->delalloc_bytes));
4391 }
4392
4393 if (percpu_counter_sum(&fs_info->ordered_bytes))
4394 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4395 percpu_counter_sum(&fs_info->ordered_bytes));
4396
4397 btrfs_sysfs_remove_mounted(fs_info);
4398 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4399
4400 btrfs_put_block_group_cache(fs_info);
4401
4402 /*
4403 * we must make sure there is not any read request to
4404 * submit after we stopping all workers.
4405 */
4406 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4407 btrfs_stop_all_workers(fs_info);
4408
4409 /* We shouldn't have any transaction open at this point */
4410 ASSERT(list_empty(&fs_info->trans_list));
4411
4412 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4413 free_root_pointers(fs_info, true);
4414 btrfs_free_fs_roots(fs_info);
4415
4416 /*
4417 * We must free the block groups after dropping the fs_roots as we could
4418 * have had an IO error and have left over tree log blocks that aren't
4419 * cleaned up until the fs roots are freed. This makes the block group
4420 * accounting appear to be wrong because there's pending reserved bytes,
4421 * so make sure we do the block group cleanup afterwards.
4422 */
4423 btrfs_free_block_groups(fs_info);
4424
4425 iput(fs_info->btree_inode);
4426
4427#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4428 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4429 btrfsic_unmount(fs_info->fs_devices);
4430#endif
4431
4432 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4433 btrfs_close_devices(fs_info->fs_devices);
4434}
4435
4436int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4437 int atomic)
4438{
4439 int ret;
4440 struct inode *btree_inode = buf->pages[0]->mapping->host;
4441
4442 ret = extent_buffer_uptodate(buf);
4443 if (!ret)
4444 return ret;
4445
4446 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4447 parent_transid, atomic);
4448 if (ret == -EAGAIN)
4449 return ret;
4450 return !ret;
4451}
4452
4453void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4454{
4455 struct btrfs_fs_info *fs_info = buf->fs_info;
4456 u64 transid = btrfs_header_generation(buf);
4457 int was_dirty;
4458
4459#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4460 /*
4461 * This is a fast path so only do this check if we have sanity tests
4462 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4463 * outside of the sanity tests.
4464 */
4465 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4466 return;
4467#endif
4468 btrfs_assert_tree_locked(buf);
4469 if (transid != fs_info->generation)
4470 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4471 buf->start, transid, fs_info->generation);
4472 was_dirty = set_extent_buffer_dirty(buf);
4473 if (!was_dirty)
4474 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4475 buf->len,
4476 fs_info->dirty_metadata_batch);
4477#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4478 /*
4479 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4480 * but item data not updated.
4481 * So here we should only check item pointers, not item data.
4482 */
4483 if (btrfs_header_level(buf) == 0 &&
4484 btrfs_check_leaf_relaxed(buf)) {
4485 btrfs_print_leaf(buf);
4486 ASSERT(0);
4487 }
4488#endif
4489}
4490
4491static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4492 int flush_delayed)
4493{
4494 /*
4495 * looks as though older kernels can get into trouble with
4496 * this code, they end up stuck in balance_dirty_pages forever
4497 */
4498 int ret;
4499
4500 if (current->flags & PF_MEMALLOC)
4501 return;
4502
4503 if (flush_delayed)
4504 btrfs_balance_delayed_items(fs_info);
4505
4506 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4507 BTRFS_DIRTY_METADATA_THRESH,
4508 fs_info->dirty_metadata_batch);
4509 if (ret > 0) {
4510 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4511 }
4512}
4513
4514void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4515{
4516 __btrfs_btree_balance_dirty(fs_info, 1);
4517}
4518
4519void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4520{
4521 __btrfs_btree_balance_dirty(fs_info, 0);
4522}
4523
4524int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4525 struct btrfs_key *first_key)
4526{
4527 return btree_read_extent_buffer_pages(buf, parent_transid,
4528 level, first_key);
4529}
4530
4531static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4532{
4533 /* cleanup FS via transaction */
4534 btrfs_cleanup_transaction(fs_info);
4535
4536 mutex_lock(&fs_info->cleaner_mutex);
4537 btrfs_run_delayed_iputs(fs_info);
4538 mutex_unlock(&fs_info->cleaner_mutex);
4539
4540 down_write(&fs_info->cleanup_work_sem);
4541 up_write(&fs_info->cleanup_work_sem);
4542}
4543
4544static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4545{
4546 struct btrfs_root *gang[8];
4547 u64 root_objectid = 0;
4548 int ret;
4549
4550 spin_lock(&fs_info->fs_roots_radix_lock);
4551 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4552 (void **)gang, root_objectid,
4553 ARRAY_SIZE(gang))) != 0) {
4554 int i;
4555
4556 for (i = 0; i < ret; i++)
4557 gang[i] = btrfs_grab_root(gang[i]);
4558 spin_unlock(&fs_info->fs_roots_radix_lock);
4559
4560 for (i = 0; i < ret; i++) {
4561 if (!gang[i])
4562 continue;
4563 root_objectid = gang[i]->root_key.objectid;
4564 btrfs_free_log(NULL, gang[i]);
4565 btrfs_put_root(gang[i]);
4566 }
4567 root_objectid++;
4568 spin_lock(&fs_info->fs_roots_radix_lock);
4569 }
4570 spin_unlock(&fs_info->fs_roots_radix_lock);
4571 btrfs_free_log_root_tree(NULL, fs_info);
4572}
4573
4574static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4575{
4576 struct btrfs_ordered_extent *ordered;
4577
4578 spin_lock(&root->ordered_extent_lock);
4579 /*
4580 * This will just short circuit the ordered completion stuff which will
4581 * make sure the ordered extent gets properly cleaned up.
4582 */
4583 list_for_each_entry(ordered, &root->ordered_extents,
4584 root_extent_list)
4585 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4586 spin_unlock(&root->ordered_extent_lock);
4587}
4588
4589static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4590{
4591 struct btrfs_root *root;
4592 struct list_head splice;
4593
4594 INIT_LIST_HEAD(&splice);
4595
4596 spin_lock(&fs_info->ordered_root_lock);
4597 list_splice_init(&fs_info->ordered_roots, &splice);
4598 while (!list_empty(&splice)) {
4599 root = list_first_entry(&splice, struct btrfs_root,
4600 ordered_root);
4601 list_move_tail(&root->ordered_root,
4602 &fs_info->ordered_roots);
4603
4604 spin_unlock(&fs_info->ordered_root_lock);
4605 btrfs_destroy_ordered_extents(root);
4606
4607 cond_resched();
4608 spin_lock(&fs_info->ordered_root_lock);
4609 }
4610 spin_unlock(&fs_info->ordered_root_lock);
4611
4612 /*
4613 * We need this here because if we've been flipped read-only we won't
4614 * get sync() from the umount, so we need to make sure any ordered
4615 * extents that haven't had their dirty pages IO start writeout yet
4616 * actually get run and error out properly.
4617 */
4618 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4619}
4620
4621static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4622 struct btrfs_fs_info *fs_info)
4623{
4624 struct rb_node *node;
4625 struct btrfs_delayed_ref_root *delayed_refs;
4626 struct btrfs_delayed_ref_node *ref;
4627 int ret = 0;
4628
4629 delayed_refs = &trans->delayed_refs;
4630
4631 spin_lock(&delayed_refs->lock);
4632 if (atomic_read(&delayed_refs->num_entries) == 0) {
4633 spin_unlock(&delayed_refs->lock);
4634 btrfs_debug(fs_info, "delayed_refs has NO entry");
4635 return ret;
4636 }
4637
4638 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4639 struct btrfs_delayed_ref_head *head;
4640 struct rb_node *n;
4641 bool pin_bytes = false;
4642
4643 head = rb_entry(node, struct btrfs_delayed_ref_head,
4644 href_node);
4645 if (btrfs_delayed_ref_lock(delayed_refs, head))
4646 continue;
4647
4648 spin_lock(&head->lock);
4649 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4650 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4651 ref_node);
4652 ref->in_tree = 0;
4653 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4654 RB_CLEAR_NODE(&ref->ref_node);
4655 if (!list_empty(&ref->add_list))
4656 list_del(&ref->add_list);
4657 atomic_dec(&delayed_refs->num_entries);
4658 btrfs_put_delayed_ref(ref);
4659 }
4660 if (head->must_insert_reserved)
4661 pin_bytes = true;
4662 btrfs_free_delayed_extent_op(head->extent_op);
4663 btrfs_delete_ref_head(delayed_refs, head);
4664 spin_unlock(&head->lock);
4665 spin_unlock(&delayed_refs->lock);
4666 mutex_unlock(&head->mutex);
4667
4668 if (pin_bytes) {
4669 struct btrfs_block_group *cache;
4670
4671 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4672 BUG_ON(!cache);
4673
4674 spin_lock(&cache->space_info->lock);
4675 spin_lock(&cache->lock);
4676 cache->pinned += head->num_bytes;
4677 btrfs_space_info_update_bytes_pinned(fs_info,
4678 cache->space_info, head->num_bytes);
4679 cache->reserved -= head->num_bytes;
4680 cache->space_info->bytes_reserved -= head->num_bytes;
4681 spin_unlock(&cache->lock);
4682 spin_unlock(&cache->space_info->lock);
4683
4684 btrfs_put_block_group(cache);
4685
4686 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4687 head->bytenr + head->num_bytes - 1);
4688 }
4689 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4690 btrfs_put_delayed_ref_head(head);
4691 cond_resched();
4692 spin_lock(&delayed_refs->lock);
4693 }
4694 btrfs_qgroup_destroy_extent_records(trans);
4695
4696 spin_unlock(&delayed_refs->lock);
4697
4698 return ret;
4699}
4700
4701static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4702{
4703 struct btrfs_inode *btrfs_inode;
4704 struct list_head splice;
4705
4706 INIT_LIST_HEAD(&splice);
4707
4708 spin_lock(&root->delalloc_lock);
4709 list_splice_init(&root->delalloc_inodes, &splice);
4710
4711 while (!list_empty(&splice)) {
4712 struct inode *inode = NULL;
4713 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4714 delalloc_inodes);
4715 __btrfs_del_delalloc_inode(root, btrfs_inode);
4716 spin_unlock(&root->delalloc_lock);
4717
4718 /*
4719 * Make sure we get a live inode and that it'll not disappear
4720 * meanwhile.
4721 */
4722 inode = igrab(&btrfs_inode->vfs_inode);
4723 if (inode) {
4724 invalidate_inode_pages2(inode->i_mapping);
4725 iput(inode);
4726 }
4727 spin_lock(&root->delalloc_lock);
4728 }
4729 spin_unlock(&root->delalloc_lock);
4730}
4731
4732static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4733{
4734 struct btrfs_root *root;
4735 struct list_head splice;
4736
4737 INIT_LIST_HEAD(&splice);
4738
4739 spin_lock(&fs_info->delalloc_root_lock);
4740 list_splice_init(&fs_info->delalloc_roots, &splice);
4741 while (!list_empty(&splice)) {
4742 root = list_first_entry(&splice, struct btrfs_root,
4743 delalloc_root);
4744 root = btrfs_grab_root(root);
4745 BUG_ON(!root);
4746 spin_unlock(&fs_info->delalloc_root_lock);
4747
4748 btrfs_destroy_delalloc_inodes(root);
4749 btrfs_put_root(root);
4750
4751 spin_lock(&fs_info->delalloc_root_lock);
4752 }
4753 spin_unlock(&fs_info->delalloc_root_lock);
4754}
4755
4756static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4757 struct extent_io_tree *dirty_pages,
4758 int mark)
4759{
4760 int ret;
4761 struct extent_buffer *eb;
4762 u64 start = 0;
4763 u64 end;
4764
4765 while (1) {
4766 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4767 mark, NULL);
4768 if (ret)
4769 break;
4770
4771 clear_extent_bits(dirty_pages, start, end, mark);
4772 while (start <= end) {
4773 eb = find_extent_buffer(fs_info, start);
4774 start += fs_info->nodesize;
4775 if (!eb)
4776 continue;
4777 wait_on_extent_buffer_writeback(eb);
4778
4779 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4780 &eb->bflags))
4781 clear_extent_buffer_dirty(eb);
4782 free_extent_buffer_stale(eb);
4783 }
4784 }
4785
4786 return ret;
4787}
4788
4789static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4790 struct extent_io_tree *unpin)
4791{
4792 u64 start;
4793 u64 end;
4794 int ret;
4795
4796 while (1) {
4797 struct extent_state *cached_state = NULL;
4798
4799 /*
4800 * The btrfs_finish_extent_commit() may get the same range as
4801 * ours between find_first_extent_bit and clear_extent_dirty.
4802 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4803 * the same extent range.
4804 */
4805 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4806 ret = find_first_extent_bit(unpin, 0, &start, &end,
4807 EXTENT_DIRTY, &cached_state);
4808 if (ret) {
4809 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4810 break;
4811 }
4812
4813 clear_extent_dirty(unpin, start, end, &cached_state);
4814 free_extent_state(cached_state);
4815 btrfs_error_unpin_extent_range(fs_info, start, end);
4816 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4817 cond_resched();
4818 }
4819
4820 return 0;
4821}
4822
4823static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4824{
4825 struct inode *inode;
4826
4827 inode = cache->io_ctl.inode;
4828 if (inode) {
4829 invalidate_inode_pages2(inode->i_mapping);
4830 BTRFS_I(inode)->generation = 0;
4831 cache->io_ctl.inode = NULL;
4832 iput(inode);
4833 }
4834 ASSERT(cache->io_ctl.pages == NULL);
4835 btrfs_put_block_group(cache);
4836}
4837
4838void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4839 struct btrfs_fs_info *fs_info)
4840{
4841 struct btrfs_block_group *cache;
4842
4843 spin_lock(&cur_trans->dirty_bgs_lock);
4844 while (!list_empty(&cur_trans->dirty_bgs)) {
4845 cache = list_first_entry(&cur_trans->dirty_bgs,
4846 struct btrfs_block_group,
4847 dirty_list);
4848
4849 if (!list_empty(&cache->io_list)) {
4850 spin_unlock(&cur_trans->dirty_bgs_lock);
4851 list_del_init(&cache->io_list);
4852 btrfs_cleanup_bg_io(cache);
4853 spin_lock(&cur_trans->dirty_bgs_lock);
4854 }
4855
4856 list_del_init(&cache->dirty_list);
4857 spin_lock(&cache->lock);
4858 cache->disk_cache_state = BTRFS_DC_ERROR;
4859 spin_unlock(&cache->lock);
4860
4861 spin_unlock(&cur_trans->dirty_bgs_lock);
4862 btrfs_put_block_group(cache);
4863 btrfs_delayed_refs_rsv_release(fs_info, 1);
4864 spin_lock(&cur_trans->dirty_bgs_lock);
4865 }
4866 spin_unlock(&cur_trans->dirty_bgs_lock);
4867
4868 /*
4869 * Refer to the definition of io_bgs member for details why it's safe
4870 * to use it without any locking
4871 */
4872 while (!list_empty(&cur_trans->io_bgs)) {
4873 cache = list_first_entry(&cur_trans->io_bgs,
4874 struct btrfs_block_group,
4875 io_list);
4876
4877 list_del_init(&cache->io_list);
4878 spin_lock(&cache->lock);
4879 cache->disk_cache_state = BTRFS_DC_ERROR;
4880 spin_unlock(&cache->lock);
4881 btrfs_cleanup_bg_io(cache);
4882 }
4883}
4884
4885void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4886 struct btrfs_fs_info *fs_info)
4887{
4888 struct btrfs_device *dev, *tmp;
4889
4890 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4891 ASSERT(list_empty(&cur_trans->dirty_bgs));
4892 ASSERT(list_empty(&cur_trans->io_bgs));
4893
4894 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4895 post_commit_list) {
4896 list_del_init(&dev->post_commit_list);
4897 }
4898
4899 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4900
4901 cur_trans->state = TRANS_STATE_COMMIT_START;
4902 wake_up(&fs_info->transaction_blocked_wait);
4903
4904 cur_trans->state = TRANS_STATE_UNBLOCKED;
4905 wake_up(&fs_info->transaction_wait);
4906
4907 btrfs_destroy_delayed_inodes(fs_info);
4908
4909 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4910 EXTENT_DIRTY);
4911 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4912
4913 btrfs_free_redirty_list(cur_trans);
4914
4915 cur_trans->state =TRANS_STATE_COMPLETED;
4916 wake_up(&cur_trans->commit_wait);
4917}
4918
4919static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4920{
4921 struct btrfs_transaction *t;
4922
4923 mutex_lock(&fs_info->transaction_kthread_mutex);
4924
4925 spin_lock(&fs_info->trans_lock);
4926 while (!list_empty(&fs_info->trans_list)) {
4927 t = list_first_entry(&fs_info->trans_list,
4928 struct btrfs_transaction, list);
4929 if (t->state >= TRANS_STATE_COMMIT_START) {
4930 refcount_inc(&t->use_count);
4931 spin_unlock(&fs_info->trans_lock);
4932 btrfs_wait_for_commit(fs_info, t->transid);
4933 btrfs_put_transaction(t);
4934 spin_lock(&fs_info->trans_lock);
4935 continue;
4936 }
4937 if (t == fs_info->running_transaction) {
4938 t->state = TRANS_STATE_COMMIT_DOING;
4939 spin_unlock(&fs_info->trans_lock);
4940 /*
4941 * We wait for 0 num_writers since we don't hold a trans
4942 * handle open currently for this transaction.
4943 */
4944 wait_event(t->writer_wait,
4945 atomic_read(&t->num_writers) == 0);
4946 } else {
4947 spin_unlock(&fs_info->trans_lock);
4948 }
4949 btrfs_cleanup_one_transaction(t, fs_info);
4950
4951 spin_lock(&fs_info->trans_lock);
4952 if (t == fs_info->running_transaction)
4953 fs_info->running_transaction = NULL;
4954 list_del_init(&t->list);
4955 spin_unlock(&fs_info->trans_lock);
4956
4957 btrfs_put_transaction(t);
4958 trace_btrfs_transaction_commit(fs_info->tree_root);
4959 spin_lock(&fs_info->trans_lock);
4960 }
4961 spin_unlock(&fs_info->trans_lock);
4962 btrfs_destroy_all_ordered_extents(fs_info);
4963 btrfs_destroy_delayed_inodes(fs_info);
4964 btrfs_assert_delayed_root_empty(fs_info);
4965 btrfs_destroy_all_delalloc_inodes(fs_info);
4966 btrfs_drop_all_logs(fs_info);
4967 mutex_unlock(&fs_info->transaction_kthread_mutex);
4968
4969 return 0;
4970}
4971
4972int btrfs_init_root_free_objectid(struct btrfs_root *root)
4973{
4974 struct btrfs_path *path;
4975 int ret;
4976 struct extent_buffer *l;
4977 struct btrfs_key search_key;
4978 struct btrfs_key found_key;
4979 int slot;
4980
4981 path = btrfs_alloc_path();
4982 if (!path)
4983 return -ENOMEM;
4984
4985 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4986 search_key.type = -1;
4987 search_key.offset = (u64)-1;
4988 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4989 if (ret < 0)
4990 goto error;
4991 BUG_ON(ret == 0); /* Corruption */
4992 if (path->slots[0] > 0) {
4993 slot = path->slots[0] - 1;
4994 l = path->nodes[0];
4995 btrfs_item_key_to_cpu(l, &found_key, slot);
4996 root->free_objectid = max_t(u64, found_key.objectid + 1,
4997 BTRFS_FIRST_FREE_OBJECTID);
4998 } else {
4999 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5000 }
5001 ret = 0;
5002error:
5003 btrfs_free_path(path);
5004 return ret;
5005}
5006
5007int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5008{
5009 int ret;
5010 mutex_lock(&root->objectid_mutex);
5011
5012 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5013 btrfs_warn(root->fs_info,
5014 "the objectid of root %llu reaches its highest value",
5015 root->root_key.objectid);
5016 ret = -ENOSPC;
5017 goto out;
5018 }
5019
5020 *objectid = root->free_objectid++;
5021 ret = 0;
5022out:
5023 mutex_unlock(&root->objectid_mutex);
5024 return ret;
5025}