Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/fs.h>
7#include <linux/blkdev.h>
8#include <linux/radix-tree.h>
9#include <linux/writeback.h>
10#include <linux/workqueue.h>
11#include <linux/kthread.h>
12#include <linux/slab.h>
13#include <linux/migrate.h>
14#include <linux/ratelimit.h>
15#include <linux/uuid.h>
16#include <linux/semaphore.h>
17#include <linux/error-injection.h>
18#include <linux/crc32c.h>
19#include <linux/sched/mm.h>
20#include <asm/unaligned.h>
21#include <crypto/hash.h>
22#include "ctree.h"
23#include "disk-io.h"
24#include "transaction.h"
25#include "btrfs_inode.h"
26#include "bio.h"
27#include "print-tree.h"
28#include "locking.h"
29#include "tree-log.h"
30#include "free-space-cache.h"
31#include "free-space-tree.h"
32#include "rcu-string.h"
33#include "dev-replace.h"
34#include "raid56.h"
35#include "sysfs.h"
36#include "qgroup.h"
37#include "compression.h"
38#include "tree-checker.h"
39#include "ref-verify.h"
40#include "block-group.h"
41#include "discard.h"
42#include "space-info.h"
43#include "zoned.h"
44#include "subpage.h"
45#include "fs.h"
46#include "accessors.h"
47#include "extent-tree.h"
48#include "root-tree.h"
49#include "defrag.h"
50#include "uuid-tree.h"
51#include "relocation.h"
52#include "scrub.h"
53#include "super.h"
54
55#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
56 BTRFS_HEADER_FLAG_RELOC |\
57 BTRFS_SUPER_FLAG_ERROR |\
58 BTRFS_SUPER_FLAG_SEEDING |\
59 BTRFS_SUPER_FLAG_METADUMP |\
60 BTRFS_SUPER_FLAG_METADUMP_V2)
61
62static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
63static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
64
65static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
66{
67 if (fs_info->csum_shash)
68 crypto_free_shash(fs_info->csum_shash);
69}
70
71/*
72 * Compute the csum of a btree block and store the result to provided buffer.
73 */
74static void csum_tree_block(struct extent_buffer *buf, u8 *result)
75{
76 struct btrfs_fs_info *fs_info = buf->fs_info;
77 int num_pages;
78 u32 first_page_part;
79 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
80 char *kaddr;
81 int i;
82
83 shash->tfm = fs_info->csum_shash;
84 crypto_shash_init(shash);
85
86 if (buf->addr) {
87 /* Pages are contiguous, handle them as a big one. */
88 kaddr = buf->addr;
89 first_page_part = fs_info->nodesize;
90 num_pages = 1;
91 } else {
92 kaddr = folio_address(buf->folios[0]);
93 first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
94 num_pages = num_extent_pages(buf);
95 }
96
97 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
98 first_page_part - BTRFS_CSUM_SIZE);
99
100 /*
101 * Multiple single-page folios case would reach here.
102 *
103 * nodesize <= PAGE_SIZE and large folio all handled by above
104 * crypto_shash_update() already.
105 */
106 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
107 kaddr = folio_address(buf->folios[i]);
108 crypto_shash_update(shash, kaddr, PAGE_SIZE);
109 }
110 memset(result, 0, BTRFS_CSUM_SIZE);
111 crypto_shash_final(shash, result);
112}
113
114/*
115 * we can't consider a given block up to date unless the transid of the
116 * block matches the transid in the parent node's pointer. This is how we
117 * detect blocks that either didn't get written at all or got written
118 * in the wrong place.
119 */
120int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
121{
122 if (!extent_buffer_uptodate(eb))
123 return 0;
124
125 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
126 return 1;
127
128 if (atomic)
129 return -EAGAIN;
130
131 if (!extent_buffer_uptodate(eb) ||
132 btrfs_header_generation(eb) != parent_transid) {
133 btrfs_err_rl(eb->fs_info,
134"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
135 eb->start, eb->read_mirror,
136 parent_transid, btrfs_header_generation(eb));
137 clear_extent_buffer_uptodate(eb);
138 return 0;
139 }
140 return 1;
141}
142
143static bool btrfs_supported_super_csum(u16 csum_type)
144{
145 switch (csum_type) {
146 case BTRFS_CSUM_TYPE_CRC32:
147 case BTRFS_CSUM_TYPE_XXHASH:
148 case BTRFS_CSUM_TYPE_SHA256:
149 case BTRFS_CSUM_TYPE_BLAKE2:
150 return true;
151 default:
152 return false;
153 }
154}
155
156/*
157 * Return 0 if the superblock checksum type matches the checksum value of that
158 * algorithm. Pass the raw disk superblock data.
159 */
160int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
161 const struct btrfs_super_block *disk_sb)
162{
163 char result[BTRFS_CSUM_SIZE];
164 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
165
166 shash->tfm = fs_info->csum_shash;
167
168 /*
169 * The super_block structure does not span the whole
170 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
171 * filled with zeros and is included in the checksum.
172 */
173 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
174 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
175
176 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
177 return 1;
178
179 return 0;
180}
181
182static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
183 int mirror_num)
184{
185 struct btrfs_fs_info *fs_info = eb->fs_info;
186 int num_folios = num_extent_folios(eb);
187 int ret = 0;
188
189 if (sb_rdonly(fs_info->sb))
190 return -EROFS;
191
192 for (int i = 0; i < num_folios; i++) {
193 struct folio *folio = eb->folios[i];
194 u64 start = max_t(u64, eb->start, folio_pos(folio));
195 u64 end = min_t(u64, eb->start + eb->len,
196 folio_pos(folio) + folio_size(folio));
197 u32 len = end - start;
198
199 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
200 start, folio, offset_in_folio(folio, start),
201 mirror_num);
202 if (ret)
203 break;
204 }
205
206 return ret;
207}
208
209/*
210 * helper to read a given tree block, doing retries as required when
211 * the checksums don't match and we have alternate mirrors to try.
212 *
213 * @check: expected tree parentness check, see the comments of the
214 * structure for details.
215 */
216int btrfs_read_extent_buffer(struct extent_buffer *eb,
217 struct btrfs_tree_parent_check *check)
218{
219 struct btrfs_fs_info *fs_info = eb->fs_info;
220 int failed = 0;
221 int ret;
222 int num_copies = 0;
223 int mirror_num = 0;
224 int failed_mirror = 0;
225
226 ASSERT(check);
227
228 while (1) {
229 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
230 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
231 if (!ret)
232 break;
233
234 num_copies = btrfs_num_copies(fs_info,
235 eb->start, eb->len);
236 if (num_copies == 1)
237 break;
238
239 if (!failed_mirror) {
240 failed = 1;
241 failed_mirror = eb->read_mirror;
242 }
243
244 mirror_num++;
245 if (mirror_num == failed_mirror)
246 mirror_num++;
247
248 if (mirror_num > num_copies)
249 break;
250 }
251
252 if (failed && !ret && failed_mirror)
253 btrfs_repair_eb_io_failure(eb, failed_mirror);
254
255 return ret;
256}
257
258/*
259 * Checksum a dirty tree block before IO.
260 */
261blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
262{
263 struct extent_buffer *eb = bbio->private;
264 struct btrfs_fs_info *fs_info = eb->fs_info;
265 u64 found_start = btrfs_header_bytenr(eb);
266 u64 last_trans;
267 u8 result[BTRFS_CSUM_SIZE];
268 int ret;
269
270 /* Btree blocks are always contiguous on disk. */
271 if (WARN_ON_ONCE(bbio->file_offset != eb->start))
272 return BLK_STS_IOERR;
273 if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
274 return BLK_STS_IOERR;
275
276 /*
277 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
278 * checksum it but zero-out its content. This is done to preserve
279 * ordering of I/O without unnecessarily writing out data.
280 */
281 if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
282 memzero_extent_buffer(eb, 0, eb->len);
283 return BLK_STS_OK;
284 }
285
286 if (WARN_ON_ONCE(found_start != eb->start))
287 return BLK_STS_IOERR;
288 if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
289 eb->start, eb->len)))
290 return BLK_STS_IOERR;
291
292 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
293 offsetof(struct btrfs_header, fsid),
294 BTRFS_FSID_SIZE) == 0);
295 csum_tree_block(eb, result);
296
297 if (btrfs_header_level(eb))
298 ret = btrfs_check_node(eb);
299 else
300 ret = btrfs_check_leaf(eb);
301
302 if (ret < 0)
303 goto error;
304
305 /*
306 * Also check the generation, the eb reached here must be newer than
307 * last committed. Or something seriously wrong happened.
308 */
309 last_trans = btrfs_get_last_trans_committed(fs_info);
310 if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
311 ret = -EUCLEAN;
312 btrfs_err(fs_info,
313 "block=%llu bad generation, have %llu expect > %llu",
314 eb->start, btrfs_header_generation(eb), last_trans);
315 goto error;
316 }
317 write_extent_buffer(eb, result, 0, fs_info->csum_size);
318 return BLK_STS_OK;
319
320error:
321 btrfs_print_tree(eb, 0);
322 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
323 eb->start);
324 /*
325 * Be noisy if this is an extent buffer from a log tree. We don't abort
326 * a transaction in case there's a bad log tree extent buffer, we just
327 * fallback to a transaction commit. Still we want to know when there is
328 * a bad log tree extent buffer, as that may signal a bug somewhere.
329 */
330 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
331 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
332 return errno_to_blk_status(ret);
333}
334
335static bool check_tree_block_fsid(struct extent_buffer *eb)
336{
337 struct btrfs_fs_info *fs_info = eb->fs_info;
338 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
339 u8 fsid[BTRFS_FSID_SIZE];
340
341 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
342 BTRFS_FSID_SIZE);
343
344 /*
345 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
346 * This is then overwritten by metadata_uuid if it is present in the
347 * device_list_add(). The same true for a seed device as well. So use of
348 * fs_devices::metadata_uuid is appropriate here.
349 */
350 if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
351 return false;
352
353 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
354 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
355 return false;
356
357 return true;
358}
359
360/* Do basic extent buffer checks at read time */
361int btrfs_validate_extent_buffer(struct extent_buffer *eb,
362 struct btrfs_tree_parent_check *check)
363{
364 struct btrfs_fs_info *fs_info = eb->fs_info;
365 u64 found_start;
366 const u32 csum_size = fs_info->csum_size;
367 u8 found_level;
368 u8 result[BTRFS_CSUM_SIZE];
369 const u8 *header_csum;
370 int ret = 0;
371
372 ASSERT(check);
373
374 found_start = btrfs_header_bytenr(eb);
375 if (found_start != eb->start) {
376 btrfs_err_rl(fs_info,
377 "bad tree block start, mirror %u want %llu have %llu",
378 eb->read_mirror, eb->start, found_start);
379 ret = -EIO;
380 goto out;
381 }
382 if (check_tree_block_fsid(eb)) {
383 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
384 eb->start, eb->read_mirror);
385 ret = -EIO;
386 goto out;
387 }
388 found_level = btrfs_header_level(eb);
389 if (found_level >= BTRFS_MAX_LEVEL) {
390 btrfs_err(fs_info,
391 "bad tree block level, mirror %u level %d on logical %llu",
392 eb->read_mirror, btrfs_header_level(eb), eb->start);
393 ret = -EIO;
394 goto out;
395 }
396
397 csum_tree_block(eb, result);
398 header_csum = folio_address(eb->folios[0]) +
399 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
400
401 if (memcmp(result, header_csum, csum_size) != 0) {
402 btrfs_warn_rl(fs_info,
403"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
404 eb->start, eb->read_mirror,
405 CSUM_FMT_VALUE(csum_size, header_csum),
406 CSUM_FMT_VALUE(csum_size, result),
407 btrfs_header_level(eb));
408 ret = -EUCLEAN;
409 goto out;
410 }
411
412 if (found_level != check->level) {
413 btrfs_err(fs_info,
414 "level verify failed on logical %llu mirror %u wanted %u found %u",
415 eb->start, eb->read_mirror, check->level, found_level);
416 ret = -EIO;
417 goto out;
418 }
419 if (unlikely(check->transid &&
420 btrfs_header_generation(eb) != check->transid)) {
421 btrfs_err_rl(eb->fs_info,
422"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
423 eb->start, eb->read_mirror, check->transid,
424 btrfs_header_generation(eb));
425 ret = -EIO;
426 goto out;
427 }
428 if (check->has_first_key) {
429 struct btrfs_key *expect_key = &check->first_key;
430 struct btrfs_key found_key;
431
432 if (found_level)
433 btrfs_node_key_to_cpu(eb, &found_key, 0);
434 else
435 btrfs_item_key_to_cpu(eb, &found_key, 0);
436 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
437 btrfs_err(fs_info,
438"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
439 eb->start, check->transid,
440 expect_key->objectid,
441 expect_key->type, expect_key->offset,
442 found_key.objectid, found_key.type,
443 found_key.offset);
444 ret = -EUCLEAN;
445 goto out;
446 }
447 }
448 if (check->owner_root) {
449 ret = btrfs_check_eb_owner(eb, check->owner_root);
450 if (ret < 0)
451 goto out;
452 }
453
454 /*
455 * If this is a leaf block and it is corrupt, set the corrupt bit so
456 * that we don't try and read the other copies of this block, just
457 * return -EIO.
458 */
459 if (found_level == 0 && btrfs_check_leaf(eb)) {
460 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
461 ret = -EIO;
462 }
463
464 if (found_level > 0 && btrfs_check_node(eb))
465 ret = -EIO;
466
467 if (ret)
468 btrfs_err(fs_info,
469 "read time tree block corruption detected on logical %llu mirror %u",
470 eb->start, eb->read_mirror);
471out:
472 return ret;
473}
474
475#ifdef CONFIG_MIGRATION
476static int btree_migrate_folio(struct address_space *mapping,
477 struct folio *dst, struct folio *src, enum migrate_mode mode)
478{
479 /*
480 * we can't safely write a btree page from here,
481 * we haven't done the locking hook
482 */
483 if (folio_test_dirty(src))
484 return -EAGAIN;
485 /*
486 * Buffers may be managed in a filesystem specific way.
487 * We must have no buffers or drop them.
488 */
489 if (folio_get_private(src) &&
490 !filemap_release_folio(src, GFP_KERNEL))
491 return -EAGAIN;
492 return migrate_folio(mapping, dst, src, mode);
493}
494#else
495#define btree_migrate_folio NULL
496#endif
497
498static int btree_writepages(struct address_space *mapping,
499 struct writeback_control *wbc)
500{
501 struct btrfs_fs_info *fs_info;
502 int ret;
503
504 if (wbc->sync_mode == WB_SYNC_NONE) {
505
506 if (wbc->for_kupdate)
507 return 0;
508
509 fs_info = BTRFS_I(mapping->host)->root->fs_info;
510 /* this is a bit racy, but that's ok */
511 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
512 BTRFS_DIRTY_METADATA_THRESH,
513 fs_info->dirty_metadata_batch);
514 if (ret < 0)
515 return 0;
516 }
517 return btree_write_cache_pages(mapping, wbc);
518}
519
520static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
521{
522 if (folio_test_writeback(folio) || folio_test_dirty(folio))
523 return false;
524
525 return try_release_extent_buffer(&folio->page);
526}
527
528static void btree_invalidate_folio(struct folio *folio, size_t offset,
529 size_t length)
530{
531 struct extent_io_tree *tree;
532 tree = &BTRFS_I(folio->mapping->host)->io_tree;
533 extent_invalidate_folio(tree, folio, offset);
534 btree_release_folio(folio, GFP_NOFS);
535 if (folio_get_private(folio)) {
536 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
537 "folio private not zero on folio %llu",
538 (unsigned long long)folio_pos(folio));
539 folio_detach_private(folio);
540 }
541}
542
543#ifdef DEBUG
544static bool btree_dirty_folio(struct address_space *mapping,
545 struct folio *folio)
546{
547 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
548 struct btrfs_subpage_info *spi = fs_info->subpage_info;
549 struct btrfs_subpage *subpage;
550 struct extent_buffer *eb;
551 int cur_bit = 0;
552 u64 page_start = folio_pos(folio);
553
554 if (fs_info->sectorsize == PAGE_SIZE) {
555 eb = folio_get_private(folio);
556 BUG_ON(!eb);
557 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
558 BUG_ON(!atomic_read(&eb->refs));
559 btrfs_assert_tree_write_locked(eb);
560 return filemap_dirty_folio(mapping, folio);
561 }
562
563 ASSERT(spi);
564 subpage = folio_get_private(folio);
565
566 for (cur_bit = spi->dirty_offset;
567 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
568 cur_bit++) {
569 unsigned long flags;
570 u64 cur;
571
572 spin_lock_irqsave(&subpage->lock, flags);
573 if (!test_bit(cur_bit, subpage->bitmaps)) {
574 spin_unlock_irqrestore(&subpage->lock, flags);
575 continue;
576 }
577 spin_unlock_irqrestore(&subpage->lock, flags);
578 cur = page_start + cur_bit * fs_info->sectorsize;
579
580 eb = find_extent_buffer(fs_info, cur);
581 ASSERT(eb);
582 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
583 ASSERT(atomic_read(&eb->refs));
584 btrfs_assert_tree_write_locked(eb);
585 free_extent_buffer(eb);
586
587 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
588 }
589 return filemap_dirty_folio(mapping, folio);
590}
591#else
592#define btree_dirty_folio filemap_dirty_folio
593#endif
594
595static const struct address_space_operations btree_aops = {
596 .writepages = btree_writepages,
597 .release_folio = btree_release_folio,
598 .invalidate_folio = btree_invalidate_folio,
599 .migrate_folio = btree_migrate_folio,
600 .dirty_folio = btree_dirty_folio,
601};
602
603struct extent_buffer *btrfs_find_create_tree_block(
604 struct btrfs_fs_info *fs_info,
605 u64 bytenr, u64 owner_root,
606 int level)
607{
608 if (btrfs_is_testing(fs_info))
609 return alloc_test_extent_buffer(fs_info, bytenr);
610 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
611}
612
613/*
614 * Read tree block at logical address @bytenr and do variant basic but critical
615 * verification.
616 *
617 * @check: expected tree parentness check, see comments of the
618 * structure for details.
619 */
620struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
621 struct btrfs_tree_parent_check *check)
622{
623 struct extent_buffer *buf = NULL;
624 int ret;
625
626 ASSERT(check);
627
628 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
629 check->level);
630 if (IS_ERR(buf))
631 return buf;
632
633 ret = btrfs_read_extent_buffer(buf, check);
634 if (ret) {
635 free_extent_buffer_stale(buf);
636 return ERR_PTR(ret);
637 }
638 if (btrfs_check_eb_owner(buf, check->owner_root)) {
639 free_extent_buffer_stale(buf);
640 return ERR_PTR(-EUCLEAN);
641 }
642 return buf;
643
644}
645
646static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
647 u64 objectid)
648{
649 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
650
651 memset(&root->root_key, 0, sizeof(root->root_key));
652 memset(&root->root_item, 0, sizeof(root->root_item));
653 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
654 root->fs_info = fs_info;
655 root->root_key.objectid = objectid;
656 root->node = NULL;
657 root->commit_root = NULL;
658 root->state = 0;
659 RB_CLEAR_NODE(&root->rb_node);
660
661 root->last_trans = 0;
662 root->free_objectid = 0;
663 root->nr_delalloc_inodes = 0;
664 root->nr_ordered_extents = 0;
665 root->inode_tree = RB_ROOT;
666 /* GFP flags are compatible with XA_FLAGS_*. */
667 xa_init_flags(&root->delayed_nodes, GFP_ATOMIC);
668
669 btrfs_init_root_block_rsv(root);
670
671 INIT_LIST_HEAD(&root->dirty_list);
672 INIT_LIST_HEAD(&root->root_list);
673 INIT_LIST_HEAD(&root->delalloc_inodes);
674 INIT_LIST_HEAD(&root->delalloc_root);
675 INIT_LIST_HEAD(&root->ordered_extents);
676 INIT_LIST_HEAD(&root->ordered_root);
677 INIT_LIST_HEAD(&root->reloc_dirty_list);
678 spin_lock_init(&root->inode_lock);
679 spin_lock_init(&root->delalloc_lock);
680 spin_lock_init(&root->ordered_extent_lock);
681 spin_lock_init(&root->accounting_lock);
682 spin_lock_init(&root->qgroup_meta_rsv_lock);
683 mutex_init(&root->objectid_mutex);
684 mutex_init(&root->log_mutex);
685 mutex_init(&root->ordered_extent_mutex);
686 mutex_init(&root->delalloc_mutex);
687 init_waitqueue_head(&root->qgroup_flush_wait);
688 init_waitqueue_head(&root->log_writer_wait);
689 init_waitqueue_head(&root->log_commit_wait[0]);
690 init_waitqueue_head(&root->log_commit_wait[1]);
691 INIT_LIST_HEAD(&root->log_ctxs[0]);
692 INIT_LIST_HEAD(&root->log_ctxs[1]);
693 atomic_set(&root->log_commit[0], 0);
694 atomic_set(&root->log_commit[1], 0);
695 atomic_set(&root->log_writers, 0);
696 atomic_set(&root->log_batch, 0);
697 refcount_set(&root->refs, 1);
698 atomic_set(&root->snapshot_force_cow, 0);
699 atomic_set(&root->nr_swapfiles, 0);
700 btrfs_set_root_log_transid(root, 0);
701 root->log_transid_committed = -1;
702 btrfs_set_root_last_log_commit(root, 0);
703 root->anon_dev = 0;
704 if (!dummy) {
705 extent_io_tree_init(fs_info, &root->dirty_log_pages,
706 IO_TREE_ROOT_DIRTY_LOG_PAGES);
707 extent_io_tree_init(fs_info, &root->log_csum_range,
708 IO_TREE_LOG_CSUM_RANGE);
709 }
710
711 spin_lock_init(&root->root_item_lock);
712 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
713#ifdef CONFIG_BTRFS_DEBUG
714 INIT_LIST_HEAD(&root->leak_list);
715 spin_lock(&fs_info->fs_roots_radix_lock);
716 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
717 spin_unlock(&fs_info->fs_roots_radix_lock);
718#endif
719}
720
721static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
722 u64 objectid, gfp_t flags)
723{
724 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
725 if (root)
726 __setup_root(root, fs_info, objectid);
727 return root;
728}
729
730#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
731/* Should only be used by the testing infrastructure */
732struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
733{
734 struct btrfs_root *root;
735
736 if (!fs_info)
737 return ERR_PTR(-EINVAL);
738
739 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
740 if (!root)
741 return ERR_PTR(-ENOMEM);
742
743 /* We don't use the stripesize in selftest, set it as sectorsize */
744 root->alloc_bytenr = 0;
745
746 return root;
747}
748#endif
749
750static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
751{
752 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
753 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
754
755 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
756}
757
758static int global_root_key_cmp(const void *k, const struct rb_node *node)
759{
760 const struct btrfs_key *key = k;
761 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
762
763 return btrfs_comp_cpu_keys(key, &root->root_key);
764}
765
766int btrfs_global_root_insert(struct btrfs_root *root)
767{
768 struct btrfs_fs_info *fs_info = root->fs_info;
769 struct rb_node *tmp;
770 int ret = 0;
771
772 write_lock(&fs_info->global_root_lock);
773 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
774 write_unlock(&fs_info->global_root_lock);
775
776 if (tmp) {
777 ret = -EEXIST;
778 btrfs_warn(fs_info, "global root %llu %llu already exists",
779 root->root_key.objectid, root->root_key.offset);
780 }
781 return ret;
782}
783
784void btrfs_global_root_delete(struct btrfs_root *root)
785{
786 struct btrfs_fs_info *fs_info = root->fs_info;
787
788 write_lock(&fs_info->global_root_lock);
789 rb_erase(&root->rb_node, &fs_info->global_root_tree);
790 write_unlock(&fs_info->global_root_lock);
791}
792
793struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
794 struct btrfs_key *key)
795{
796 struct rb_node *node;
797 struct btrfs_root *root = NULL;
798
799 read_lock(&fs_info->global_root_lock);
800 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
801 if (node)
802 root = container_of(node, struct btrfs_root, rb_node);
803 read_unlock(&fs_info->global_root_lock);
804
805 return root;
806}
807
808static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
809{
810 struct btrfs_block_group *block_group;
811 u64 ret;
812
813 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
814 return 0;
815
816 if (bytenr)
817 block_group = btrfs_lookup_block_group(fs_info, bytenr);
818 else
819 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
820 ASSERT(block_group);
821 if (!block_group)
822 return 0;
823 ret = block_group->global_root_id;
824 btrfs_put_block_group(block_group);
825
826 return ret;
827}
828
829struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
830{
831 struct btrfs_key key = {
832 .objectid = BTRFS_CSUM_TREE_OBJECTID,
833 .type = BTRFS_ROOT_ITEM_KEY,
834 .offset = btrfs_global_root_id(fs_info, bytenr),
835 };
836
837 return btrfs_global_root(fs_info, &key);
838}
839
840struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
841{
842 struct btrfs_key key = {
843 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
844 .type = BTRFS_ROOT_ITEM_KEY,
845 .offset = btrfs_global_root_id(fs_info, bytenr),
846 };
847
848 return btrfs_global_root(fs_info, &key);
849}
850
851struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
852{
853 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
854 return fs_info->block_group_root;
855 return btrfs_extent_root(fs_info, 0);
856}
857
858struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
859 u64 objectid)
860{
861 struct btrfs_fs_info *fs_info = trans->fs_info;
862 struct extent_buffer *leaf;
863 struct btrfs_root *tree_root = fs_info->tree_root;
864 struct btrfs_root *root;
865 struct btrfs_key key;
866 unsigned int nofs_flag;
867 int ret = 0;
868
869 /*
870 * We're holding a transaction handle, so use a NOFS memory allocation
871 * context to avoid deadlock if reclaim happens.
872 */
873 nofs_flag = memalloc_nofs_save();
874 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
875 memalloc_nofs_restore(nofs_flag);
876 if (!root)
877 return ERR_PTR(-ENOMEM);
878
879 root->root_key.objectid = objectid;
880 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
881 root->root_key.offset = 0;
882
883 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
884 0, BTRFS_NESTING_NORMAL);
885 if (IS_ERR(leaf)) {
886 ret = PTR_ERR(leaf);
887 leaf = NULL;
888 goto fail;
889 }
890
891 root->node = leaf;
892 btrfs_mark_buffer_dirty(trans, leaf);
893
894 root->commit_root = btrfs_root_node(root);
895 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
896
897 btrfs_set_root_flags(&root->root_item, 0);
898 btrfs_set_root_limit(&root->root_item, 0);
899 btrfs_set_root_bytenr(&root->root_item, leaf->start);
900 btrfs_set_root_generation(&root->root_item, trans->transid);
901 btrfs_set_root_level(&root->root_item, 0);
902 btrfs_set_root_refs(&root->root_item, 1);
903 btrfs_set_root_used(&root->root_item, leaf->len);
904 btrfs_set_root_last_snapshot(&root->root_item, 0);
905 btrfs_set_root_dirid(&root->root_item, 0);
906 if (is_fstree(objectid))
907 generate_random_guid(root->root_item.uuid);
908 else
909 export_guid(root->root_item.uuid, &guid_null);
910 btrfs_set_root_drop_level(&root->root_item, 0);
911
912 btrfs_tree_unlock(leaf);
913
914 key.objectid = objectid;
915 key.type = BTRFS_ROOT_ITEM_KEY;
916 key.offset = 0;
917 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
918 if (ret)
919 goto fail;
920
921 return root;
922
923fail:
924 btrfs_put_root(root);
925
926 return ERR_PTR(ret);
927}
928
929static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
930 struct btrfs_fs_info *fs_info)
931{
932 struct btrfs_root *root;
933
934 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
935 if (!root)
936 return ERR_PTR(-ENOMEM);
937
938 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
939 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
940 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
941
942 return root;
943}
944
945int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
946 struct btrfs_root *root)
947{
948 struct extent_buffer *leaf;
949
950 /*
951 * DON'T set SHAREABLE bit for log trees.
952 *
953 * Log trees are not exposed to user space thus can't be snapshotted,
954 * and they go away before a real commit is actually done.
955 *
956 * They do store pointers to file data extents, and those reference
957 * counts still get updated (along with back refs to the log tree).
958 */
959
960 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
961 NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
962 if (IS_ERR(leaf))
963 return PTR_ERR(leaf);
964
965 root->node = leaf;
966
967 btrfs_mark_buffer_dirty(trans, root->node);
968 btrfs_tree_unlock(root->node);
969
970 return 0;
971}
972
973int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
974 struct btrfs_fs_info *fs_info)
975{
976 struct btrfs_root *log_root;
977
978 log_root = alloc_log_tree(trans, fs_info);
979 if (IS_ERR(log_root))
980 return PTR_ERR(log_root);
981
982 if (!btrfs_is_zoned(fs_info)) {
983 int ret = btrfs_alloc_log_tree_node(trans, log_root);
984
985 if (ret) {
986 btrfs_put_root(log_root);
987 return ret;
988 }
989 }
990
991 WARN_ON(fs_info->log_root_tree);
992 fs_info->log_root_tree = log_root;
993 return 0;
994}
995
996int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
997 struct btrfs_root *root)
998{
999 struct btrfs_fs_info *fs_info = root->fs_info;
1000 struct btrfs_root *log_root;
1001 struct btrfs_inode_item *inode_item;
1002 int ret;
1003
1004 log_root = alloc_log_tree(trans, fs_info);
1005 if (IS_ERR(log_root))
1006 return PTR_ERR(log_root);
1007
1008 ret = btrfs_alloc_log_tree_node(trans, log_root);
1009 if (ret) {
1010 btrfs_put_root(log_root);
1011 return ret;
1012 }
1013
1014 log_root->last_trans = trans->transid;
1015 log_root->root_key.offset = root->root_key.objectid;
1016
1017 inode_item = &log_root->root_item.inode;
1018 btrfs_set_stack_inode_generation(inode_item, 1);
1019 btrfs_set_stack_inode_size(inode_item, 3);
1020 btrfs_set_stack_inode_nlink(inode_item, 1);
1021 btrfs_set_stack_inode_nbytes(inode_item,
1022 fs_info->nodesize);
1023 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1024
1025 btrfs_set_root_node(&log_root->root_item, log_root->node);
1026
1027 WARN_ON(root->log_root);
1028 root->log_root = log_root;
1029 btrfs_set_root_log_transid(root, 0);
1030 root->log_transid_committed = -1;
1031 btrfs_set_root_last_log_commit(root, 0);
1032 return 0;
1033}
1034
1035static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1036 struct btrfs_path *path,
1037 struct btrfs_key *key)
1038{
1039 struct btrfs_root *root;
1040 struct btrfs_tree_parent_check check = { 0 };
1041 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1042 u64 generation;
1043 int ret;
1044 int level;
1045
1046 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1047 if (!root)
1048 return ERR_PTR(-ENOMEM);
1049
1050 ret = btrfs_find_root(tree_root, key, path,
1051 &root->root_item, &root->root_key);
1052 if (ret) {
1053 if (ret > 0)
1054 ret = -ENOENT;
1055 goto fail;
1056 }
1057
1058 generation = btrfs_root_generation(&root->root_item);
1059 level = btrfs_root_level(&root->root_item);
1060 check.level = level;
1061 check.transid = generation;
1062 check.owner_root = key->objectid;
1063 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1064 &check);
1065 if (IS_ERR(root->node)) {
1066 ret = PTR_ERR(root->node);
1067 root->node = NULL;
1068 goto fail;
1069 }
1070 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1071 ret = -EIO;
1072 goto fail;
1073 }
1074
1075 /*
1076 * For real fs, and not log/reloc trees, root owner must
1077 * match its root node owner
1078 */
1079 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1080 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1081 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1082 root->root_key.objectid != btrfs_header_owner(root->node)) {
1083 btrfs_crit(fs_info,
1084"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1085 root->root_key.objectid, root->node->start,
1086 btrfs_header_owner(root->node),
1087 root->root_key.objectid);
1088 ret = -EUCLEAN;
1089 goto fail;
1090 }
1091 root->commit_root = btrfs_root_node(root);
1092 return root;
1093fail:
1094 btrfs_put_root(root);
1095 return ERR_PTR(ret);
1096}
1097
1098struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1099 struct btrfs_key *key)
1100{
1101 struct btrfs_root *root;
1102 struct btrfs_path *path;
1103
1104 path = btrfs_alloc_path();
1105 if (!path)
1106 return ERR_PTR(-ENOMEM);
1107 root = read_tree_root_path(tree_root, path, key);
1108 btrfs_free_path(path);
1109
1110 return root;
1111}
1112
1113/*
1114 * Initialize subvolume root in-memory structure
1115 *
1116 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1117 */
1118static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1119{
1120 int ret;
1121
1122 btrfs_drew_lock_init(&root->snapshot_lock);
1123
1124 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1125 !btrfs_is_data_reloc_root(root) &&
1126 is_fstree(root->root_key.objectid)) {
1127 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1128 btrfs_check_and_init_root_item(&root->root_item);
1129 }
1130
1131 /*
1132 * Don't assign anonymous block device to roots that are not exposed to
1133 * userspace, the id pool is limited to 1M
1134 */
1135 if (is_fstree(root->root_key.objectid) &&
1136 btrfs_root_refs(&root->root_item) > 0) {
1137 if (!anon_dev) {
1138 ret = get_anon_bdev(&root->anon_dev);
1139 if (ret)
1140 goto fail;
1141 } else {
1142 root->anon_dev = anon_dev;
1143 }
1144 }
1145
1146 mutex_lock(&root->objectid_mutex);
1147 ret = btrfs_init_root_free_objectid(root);
1148 if (ret) {
1149 mutex_unlock(&root->objectid_mutex);
1150 goto fail;
1151 }
1152
1153 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1154
1155 mutex_unlock(&root->objectid_mutex);
1156
1157 return 0;
1158fail:
1159 /* The caller is responsible to call btrfs_free_fs_root */
1160 return ret;
1161}
1162
1163static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1164 u64 root_id)
1165{
1166 struct btrfs_root *root;
1167
1168 spin_lock(&fs_info->fs_roots_radix_lock);
1169 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170 (unsigned long)root_id);
1171 root = btrfs_grab_root(root);
1172 spin_unlock(&fs_info->fs_roots_radix_lock);
1173 return root;
1174}
1175
1176static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1177 u64 objectid)
1178{
1179 struct btrfs_key key = {
1180 .objectid = objectid,
1181 .type = BTRFS_ROOT_ITEM_KEY,
1182 .offset = 0,
1183 };
1184
1185 switch (objectid) {
1186 case BTRFS_ROOT_TREE_OBJECTID:
1187 return btrfs_grab_root(fs_info->tree_root);
1188 case BTRFS_EXTENT_TREE_OBJECTID:
1189 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1190 case BTRFS_CHUNK_TREE_OBJECTID:
1191 return btrfs_grab_root(fs_info->chunk_root);
1192 case BTRFS_DEV_TREE_OBJECTID:
1193 return btrfs_grab_root(fs_info->dev_root);
1194 case BTRFS_CSUM_TREE_OBJECTID:
1195 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1196 case BTRFS_QUOTA_TREE_OBJECTID:
1197 return btrfs_grab_root(fs_info->quota_root);
1198 case BTRFS_UUID_TREE_OBJECTID:
1199 return btrfs_grab_root(fs_info->uuid_root);
1200 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1201 return btrfs_grab_root(fs_info->block_group_root);
1202 case BTRFS_FREE_SPACE_TREE_OBJECTID:
1203 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1204 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1205 return btrfs_grab_root(fs_info->stripe_root);
1206 default:
1207 return NULL;
1208 }
1209}
1210
1211int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1212 struct btrfs_root *root)
1213{
1214 int ret;
1215
1216 ret = radix_tree_preload(GFP_NOFS);
1217 if (ret)
1218 return ret;
1219
1220 spin_lock(&fs_info->fs_roots_radix_lock);
1221 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222 (unsigned long)root->root_key.objectid,
1223 root);
1224 if (ret == 0) {
1225 btrfs_grab_root(root);
1226 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1227 }
1228 spin_unlock(&fs_info->fs_roots_radix_lock);
1229 radix_tree_preload_end();
1230
1231 return ret;
1232}
1233
1234void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1235{
1236#ifdef CONFIG_BTRFS_DEBUG
1237 struct btrfs_root *root;
1238
1239 while (!list_empty(&fs_info->allocated_roots)) {
1240 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1241
1242 root = list_first_entry(&fs_info->allocated_roots,
1243 struct btrfs_root, leak_list);
1244 btrfs_err(fs_info, "leaked root %s refcount %d",
1245 btrfs_root_name(&root->root_key, buf),
1246 refcount_read(&root->refs));
1247 while (refcount_read(&root->refs) > 1)
1248 btrfs_put_root(root);
1249 btrfs_put_root(root);
1250 }
1251#endif
1252}
1253
1254static void free_global_roots(struct btrfs_fs_info *fs_info)
1255{
1256 struct btrfs_root *root;
1257 struct rb_node *node;
1258
1259 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1260 root = rb_entry(node, struct btrfs_root, rb_node);
1261 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1262 btrfs_put_root(root);
1263 }
1264}
1265
1266void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1267{
1268 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1269 percpu_counter_destroy(&fs_info->delalloc_bytes);
1270 percpu_counter_destroy(&fs_info->ordered_bytes);
1271 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1272 btrfs_free_csum_hash(fs_info);
1273 btrfs_free_stripe_hash_table(fs_info);
1274 btrfs_free_ref_cache(fs_info);
1275 kfree(fs_info->balance_ctl);
1276 kfree(fs_info->delayed_root);
1277 free_global_roots(fs_info);
1278 btrfs_put_root(fs_info->tree_root);
1279 btrfs_put_root(fs_info->chunk_root);
1280 btrfs_put_root(fs_info->dev_root);
1281 btrfs_put_root(fs_info->quota_root);
1282 btrfs_put_root(fs_info->uuid_root);
1283 btrfs_put_root(fs_info->fs_root);
1284 btrfs_put_root(fs_info->data_reloc_root);
1285 btrfs_put_root(fs_info->block_group_root);
1286 btrfs_put_root(fs_info->stripe_root);
1287 btrfs_check_leaked_roots(fs_info);
1288 btrfs_extent_buffer_leak_debug_check(fs_info);
1289 kfree(fs_info->super_copy);
1290 kfree(fs_info->super_for_commit);
1291 kfree(fs_info->subpage_info);
1292 kvfree(fs_info);
1293}
1294
1295
1296/*
1297 * Get an in-memory reference of a root structure.
1298 *
1299 * For essential trees like root/extent tree, we grab it from fs_info directly.
1300 * For subvolume trees, we check the cached filesystem roots first. If not
1301 * found, then read it from disk and add it to cached fs roots.
1302 *
1303 * Caller should release the root by calling btrfs_put_root() after the usage.
1304 *
1305 * NOTE: Reloc and log trees can't be read by this function as they share the
1306 * same root objectid.
1307 *
1308 * @objectid: root id
1309 * @anon_dev: preallocated anonymous block device number for new roots,
1310 * pass NULL for a new allocation.
1311 * @check_ref: whether to check root item references, If true, return -ENOENT
1312 * for orphan roots
1313 */
1314static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1315 u64 objectid, dev_t *anon_dev,
1316 bool check_ref)
1317{
1318 struct btrfs_root *root;
1319 struct btrfs_path *path;
1320 struct btrfs_key key;
1321 int ret;
1322
1323 root = btrfs_get_global_root(fs_info, objectid);
1324 if (root)
1325 return root;
1326
1327 /*
1328 * If we're called for non-subvolume trees, and above function didn't
1329 * find one, do not try to read it from disk.
1330 *
1331 * This is namely for free-space-tree and quota tree, which can change
1332 * at runtime and should only be grabbed from fs_info.
1333 */
1334 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1335 return ERR_PTR(-ENOENT);
1336again:
1337 root = btrfs_lookup_fs_root(fs_info, objectid);
1338 if (root) {
1339 /*
1340 * Some other caller may have read out the newly inserted
1341 * subvolume already (for things like backref walk etc). Not
1342 * that common but still possible. In that case, we just need
1343 * to free the anon_dev.
1344 */
1345 if (unlikely(anon_dev && *anon_dev)) {
1346 free_anon_bdev(*anon_dev);
1347 *anon_dev = 0;
1348 }
1349
1350 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1351 btrfs_put_root(root);
1352 return ERR_PTR(-ENOENT);
1353 }
1354 return root;
1355 }
1356
1357 key.objectid = objectid;
1358 key.type = BTRFS_ROOT_ITEM_KEY;
1359 key.offset = (u64)-1;
1360 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1361 if (IS_ERR(root))
1362 return root;
1363
1364 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1365 ret = -ENOENT;
1366 goto fail;
1367 }
1368
1369 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1370 if (ret)
1371 goto fail;
1372
1373 path = btrfs_alloc_path();
1374 if (!path) {
1375 ret = -ENOMEM;
1376 goto fail;
1377 }
1378 key.objectid = BTRFS_ORPHAN_OBJECTID;
1379 key.type = BTRFS_ORPHAN_ITEM_KEY;
1380 key.offset = objectid;
1381
1382 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1383 btrfs_free_path(path);
1384 if (ret < 0)
1385 goto fail;
1386 if (ret == 0)
1387 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1388
1389 ret = btrfs_insert_fs_root(fs_info, root);
1390 if (ret) {
1391 if (ret == -EEXIST) {
1392 btrfs_put_root(root);
1393 goto again;
1394 }
1395 goto fail;
1396 }
1397 return root;
1398fail:
1399 /*
1400 * If our caller provided us an anonymous device, then it's his
1401 * responsibility to free it in case we fail. So we have to set our
1402 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1403 * and once again by our caller.
1404 */
1405 if (anon_dev && *anon_dev)
1406 root->anon_dev = 0;
1407 btrfs_put_root(root);
1408 return ERR_PTR(ret);
1409}
1410
1411/*
1412 * Get in-memory reference of a root structure
1413 *
1414 * @objectid: tree objectid
1415 * @check_ref: if set, verify that the tree exists and the item has at least
1416 * one reference
1417 */
1418struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1419 u64 objectid, bool check_ref)
1420{
1421 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1422}
1423
1424/*
1425 * Get in-memory reference of a root structure, created as new, optionally pass
1426 * the anonymous block device id
1427 *
1428 * @objectid: tree objectid
1429 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1430 * parameter value if not NULL
1431 */
1432struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1433 u64 objectid, dev_t *anon_dev)
1434{
1435 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1436}
1437
1438/*
1439 * Return a root for the given objectid.
1440 *
1441 * @fs_info: the fs_info
1442 * @objectid: the objectid we need to lookup
1443 *
1444 * This is exclusively used for backref walking, and exists specifically because
1445 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1446 * creation time, which means we may have to read the tree_root in order to look
1447 * up a fs root that is not in memory. If the root is not in memory we will
1448 * read the tree root commit root and look up the fs root from there. This is a
1449 * temporary root, it will not be inserted into the radix tree as it doesn't
1450 * have the most uptodate information, it'll simply be discarded once the
1451 * backref code is finished using the root.
1452 */
1453struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1454 struct btrfs_path *path,
1455 u64 objectid)
1456{
1457 struct btrfs_root *root;
1458 struct btrfs_key key;
1459
1460 ASSERT(path->search_commit_root && path->skip_locking);
1461
1462 /*
1463 * This can return -ENOENT if we ask for a root that doesn't exist, but
1464 * since this is called via the backref walking code we won't be looking
1465 * up a root that doesn't exist, unless there's corruption. So if root
1466 * != NULL just return it.
1467 */
1468 root = btrfs_get_global_root(fs_info, objectid);
1469 if (root)
1470 return root;
1471
1472 root = btrfs_lookup_fs_root(fs_info, objectid);
1473 if (root)
1474 return root;
1475
1476 key.objectid = objectid;
1477 key.type = BTRFS_ROOT_ITEM_KEY;
1478 key.offset = (u64)-1;
1479 root = read_tree_root_path(fs_info->tree_root, path, &key);
1480 btrfs_release_path(path);
1481
1482 return root;
1483}
1484
1485static int cleaner_kthread(void *arg)
1486{
1487 struct btrfs_fs_info *fs_info = arg;
1488 int again;
1489
1490 while (1) {
1491 again = 0;
1492
1493 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1494
1495 /* Make the cleaner go to sleep early. */
1496 if (btrfs_need_cleaner_sleep(fs_info))
1497 goto sleep;
1498
1499 /*
1500 * Do not do anything if we might cause open_ctree() to block
1501 * before we have finished mounting the filesystem.
1502 */
1503 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1504 goto sleep;
1505
1506 if (!mutex_trylock(&fs_info->cleaner_mutex))
1507 goto sleep;
1508
1509 /*
1510 * Avoid the problem that we change the status of the fs
1511 * during the above check and trylock.
1512 */
1513 if (btrfs_need_cleaner_sleep(fs_info)) {
1514 mutex_unlock(&fs_info->cleaner_mutex);
1515 goto sleep;
1516 }
1517
1518 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1519 btrfs_sysfs_feature_update(fs_info);
1520
1521 btrfs_run_delayed_iputs(fs_info);
1522
1523 again = btrfs_clean_one_deleted_snapshot(fs_info);
1524 mutex_unlock(&fs_info->cleaner_mutex);
1525
1526 /*
1527 * The defragger has dealt with the R/O remount and umount,
1528 * needn't do anything special here.
1529 */
1530 btrfs_run_defrag_inodes(fs_info);
1531
1532 /*
1533 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1534 * with relocation (btrfs_relocate_chunk) and relocation
1535 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1536 * after acquiring fs_info->reclaim_bgs_lock. So we
1537 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1538 * unused block groups.
1539 */
1540 btrfs_delete_unused_bgs(fs_info);
1541
1542 /*
1543 * Reclaim block groups in the reclaim_bgs list after we deleted
1544 * all unused block_groups. This possibly gives us some more free
1545 * space.
1546 */
1547 btrfs_reclaim_bgs(fs_info);
1548sleep:
1549 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1550 if (kthread_should_park())
1551 kthread_parkme();
1552 if (kthread_should_stop())
1553 return 0;
1554 if (!again) {
1555 set_current_state(TASK_INTERRUPTIBLE);
1556 schedule();
1557 __set_current_state(TASK_RUNNING);
1558 }
1559 }
1560}
1561
1562static int transaction_kthread(void *arg)
1563{
1564 struct btrfs_root *root = arg;
1565 struct btrfs_fs_info *fs_info = root->fs_info;
1566 struct btrfs_trans_handle *trans;
1567 struct btrfs_transaction *cur;
1568 u64 transid;
1569 time64_t delta;
1570 unsigned long delay;
1571 bool cannot_commit;
1572
1573 do {
1574 cannot_commit = false;
1575 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1576 mutex_lock(&fs_info->transaction_kthread_mutex);
1577
1578 spin_lock(&fs_info->trans_lock);
1579 cur = fs_info->running_transaction;
1580 if (!cur) {
1581 spin_unlock(&fs_info->trans_lock);
1582 goto sleep;
1583 }
1584
1585 delta = ktime_get_seconds() - cur->start_time;
1586 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1587 cur->state < TRANS_STATE_COMMIT_PREP &&
1588 delta < fs_info->commit_interval) {
1589 spin_unlock(&fs_info->trans_lock);
1590 delay -= msecs_to_jiffies((delta - 1) * 1000);
1591 delay = min(delay,
1592 msecs_to_jiffies(fs_info->commit_interval * 1000));
1593 goto sleep;
1594 }
1595 transid = cur->transid;
1596 spin_unlock(&fs_info->trans_lock);
1597
1598 /* If the file system is aborted, this will always fail. */
1599 trans = btrfs_attach_transaction(root);
1600 if (IS_ERR(trans)) {
1601 if (PTR_ERR(trans) != -ENOENT)
1602 cannot_commit = true;
1603 goto sleep;
1604 }
1605 if (transid == trans->transid) {
1606 btrfs_commit_transaction(trans);
1607 } else {
1608 btrfs_end_transaction(trans);
1609 }
1610sleep:
1611 wake_up_process(fs_info->cleaner_kthread);
1612 mutex_unlock(&fs_info->transaction_kthread_mutex);
1613
1614 if (BTRFS_FS_ERROR(fs_info))
1615 btrfs_cleanup_transaction(fs_info);
1616 if (!kthread_should_stop() &&
1617 (!btrfs_transaction_blocked(fs_info) ||
1618 cannot_commit))
1619 schedule_timeout_interruptible(delay);
1620 } while (!kthread_should_stop());
1621 return 0;
1622}
1623
1624/*
1625 * This will find the highest generation in the array of root backups. The
1626 * index of the highest array is returned, or -EINVAL if we can't find
1627 * anything.
1628 *
1629 * We check to make sure the array is valid by comparing the
1630 * generation of the latest root in the array with the generation
1631 * in the super block. If they don't match we pitch it.
1632 */
1633static int find_newest_super_backup(struct btrfs_fs_info *info)
1634{
1635 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1636 u64 cur;
1637 struct btrfs_root_backup *root_backup;
1638 int i;
1639
1640 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1641 root_backup = info->super_copy->super_roots + i;
1642 cur = btrfs_backup_tree_root_gen(root_backup);
1643 if (cur == newest_gen)
1644 return i;
1645 }
1646
1647 return -EINVAL;
1648}
1649
1650/*
1651 * copy all the root pointers into the super backup array.
1652 * this will bump the backup pointer by one when it is
1653 * done
1654 */
1655static void backup_super_roots(struct btrfs_fs_info *info)
1656{
1657 const int next_backup = info->backup_root_index;
1658 struct btrfs_root_backup *root_backup;
1659
1660 root_backup = info->super_for_commit->super_roots + next_backup;
1661
1662 /*
1663 * make sure all of our padding and empty slots get zero filled
1664 * regardless of which ones we use today
1665 */
1666 memset(root_backup, 0, sizeof(*root_backup));
1667
1668 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1669
1670 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1671 btrfs_set_backup_tree_root_gen(root_backup,
1672 btrfs_header_generation(info->tree_root->node));
1673
1674 btrfs_set_backup_tree_root_level(root_backup,
1675 btrfs_header_level(info->tree_root->node));
1676
1677 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1678 btrfs_set_backup_chunk_root_gen(root_backup,
1679 btrfs_header_generation(info->chunk_root->node));
1680 btrfs_set_backup_chunk_root_level(root_backup,
1681 btrfs_header_level(info->chunk_root->node));
1682
1683 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1684 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1685 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1686
1687 btrfs_set_backup_extent_root(root_backup,
1688 extent_root->node->start);
1689 btrfs_set_backup_extent_root_gen(root_backup,
1690 btrfs_header_generation(extent_root->node));
1691 btrfs_set_backup_extent_root_level(root_backup,
1692 btrfs_header_level(extent_root->node));
1693
1694 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1695 btrfs_set_backup_csum_root_gen(root_backup,
1696 btrfs_header_generation(csum_root->node));
1697 btrfs_set_backup_csum_root_level(root_backup,
1698 btrfs_header_level(csum_root->node));
1699 }
1700
1701 /*
1702 * we might commit during log recovery, which happens before we set
1703 * the fs_root. Make sure it is valid before we fill it in.
1704 */
1705 if (info->fs_root && info->fs_root->node) {
1706 btrfs_set_backup_fs_root(root_backup,
1707 info->fs_root->node->start);
1708 btrfs_set_backup_fs_root_gen(root_backup,
1709 btrfs_header_generation(info->fs_root->node));
1710 btrfs_set_backup_fs_root_level(root_backup,
1711 btrfs_header_level(info->fs_root->node));
1712 }
1713
1714 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1715 btrfs_set_backup_dev_root_gen(root_backup,
1716 btrfs_header_generation(info->dev_root->node));
1717 btrfs_set_backup_dev_root_level(root_backup,
1718 btrfs_header_level(info->dev_root->node));
1719
1720 btrfs_set_backup_total_bytes(root_backup,
1721 btrfs_super_total_bytes(info->super_copy));
1722 btrfs_set_backup_bytes_used(root_backup,
1723 btrfs_super_bytes_used(info->super_copy));
1724 btrfs_set_backup_num_devices(root_backup,
1725 btrfs_super_num_devices(info->super_copy));
1726
1727 /*
1728 * if we don't copy this out to the super_copy, it won't get remembered
1729 * for the next commit
1730 */
1731 memcpy(&info->super_copy->super_roots,
1732 &info->super_for_commit->super_roots,
1733 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1734}
1735
1736/*
1737 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1738 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1739 *
1740 * @fs_info: filesystem whose backup roots need to be read
1741 * @priority: priority of backup root required
1742 *
1743 * Returns backup root index on success and -EINVAL otherwise.
1744 */
1745static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1746{
1747 int backup_index = find_newest_super_backup(fs_info);
1748 struct btrfs_super_block *super = fs_info->super_copy;
1749 struct btrfs_root_backup *root_backup;
1750
1751 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1752 if (priority == 0)
1753 return backup_index;
1754
1755 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1756 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1757 } else {
1758 return -EINVAL;
1759 }
1760
1761 root_backup = super->super_roots + backup_index;
1762
1763 btrfs_set_super_generation(super,
1764 btrfs_backup_tree_root_gen(root_backup));
1765 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1766 btrfs_set_super_root_level(super,
1767 btrfs_backup_tree_root_level(root_backup));
1768 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1769
1770 /*
1771 * Fixme: the total bytes and num_devices need to match or we should
1772 * need a fsck
1773 */
1774 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1775 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1776
1777 return backup_index;
1778}
1779
1780/* helper to cleanup workers */
1781static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1782{
1783 btrfs_destroy_workqueue(fs_info->fixup_workers);
1784 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1785 btrfs_destroy_workqueue(fs_info->workers);
1786 if (fs_info->endio_workers)
1787 destroy_workqueue(fs_info->endio_workers);
1788 if (fs_info->rmw_workers)
1789 destroy_workqueue(fs_info->rmw_workers);
1790 if (fs_info->compressed_write_workers)
1791 destroy_workqueue(fs_info->compressed_write_workers);
1792 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1793 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1794 btrfs_destroy_workqueue(fs_info->delayed_workers);
1795 btrfs_destroy_workqueue(fs_info->caching_workers);
1796 btrfs_destroy_workqueue(fs_info->flush_workers);
1797 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1798 if (fs_info->discard_ctl.discard_workers)
1799 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1800 /*
1801 * Now that all other work queues are destroyed, we can safely destroy
1802 * the queues used for metadata I/O, since tasks from those other work
1803 * queues can do metadata I/O operations.
1804 */
1805 if (fs_info->endio_meta_workers)
1806 destroy_workqueue(fs_info->endio_meta_workers);
1807}
1808
1809static void free_root_extent_buffers(struct btrfs_root *root)
1810{
1811 if (root) {
1812 free_extent_buffer(root->node);
1813 free_extent_buffer(root->commit_root);
1814 root->node = NULL;
1815 root->commit_root = NULL;
1816 }
1817}
1818
1819static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1820{
1821 struct btrfs_root *root, *tmp;
1822
1823 rbtree_postorder_for_each_entry_safe(root, tmp,
1824 &fs_info->global_root_tree,
1825 rb_node)
1826 free_root_extent_buffers(root);
1827}
1828
1829/* helper to cleanup tree roots */
1830static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1831{
1832 free_root_extent_buffers(info->tree_root);
1833
1834 free_global_root_pointers(info);
1835 free_root_extent_buffers(info->dev_root);
1836 free_root_extent_buffers(info->quota_root);
1837 free_root_extent_buffers(info->uuid_root);
1838 free_root_extent_buffers(info->fs_root);
1839 free_root_extent_buffers(info->data_reloc_root);
1840 free_root_extent_buffers(info->block_group_root);
1841 free_root_extent_buffers(info->stripe_root);
1842 if (free_chunk_root)
1843 free_root_extent_buffers(info->chunk_root);
1844}
1845
1846void btrfs_put_root(struct btrfs_root *root)
1847{
1848 if (!root)
1849 return;
1850
1851 if (refcount_dec_and_test(&root->refs)) {
1852 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1853 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1854 if (root->anon_dev)
1855 free_anon_bdev(root->anon_dev);
1856 free_root_extent_buffers(root);
1857#ifdef CONFIG_BTRFS_DEBUG
1858 spin_lock(&root->fs_info->fs_roots_radix_lock);
1859 list_del_init(&root->leak_list);
1860 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1861#endif
1862 kfree(root);
1863 }
1864}
1865
1866void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1867{
1868 int ret;
1869 struct btrfs_root *gang[8];
1870 int i;
1871
1872 while (!list_empty(&fs_info->dead_roots)) {
1873 gang[0] = list_entry(fs_info->dead_roots.next,
1874 struct btrfs_root, root_list);
1875 list_del(&gang[0]->root_list);
1876
1877 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1878 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1879 btrfs_put_root(gang[0]);
1880 }
1881
1882 while (1) {
1883 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1884 (void **)gang, 0,
1885 ARRAY_SIZE(gang));
1886 if (!ret)
1887 break;
1888 for (i = 0; i < ret; i++)
1889 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1890 }
1891}
1892
1893static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1894{
1895 mutex_init(&fs_info->scrub_lock);
1896 atomic_set(&fs_info->scrubs_running, 0);
1897 atomic_set(&fs_info->scrub_pause_req, 0);
1898 atomic_set(&fs_info->scrubs_paused, 0);
1899 atomic_set(&fs_info->scrub_cancel_req, 0);
1900 init_waitqueue_head(&fs_info->scrub_pause_wait);
1901 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1902}
1903
1904static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1905{
1906 spin_lock_init(&fs_info->balance_lock);
1907 mutex_init(&fs_info->balance_mutex);
1908 atomic_set(&fs_info->balance_pause_req, 0);
1909 atomic_set(&fs_info->balance_cancel_req, 0);
1910 fs_info->balance_ctl = NULL;
1911 init_waitqueue_head(&fs_info->balance_wait_q);
1912 atomic_set(&fs_info->reloc_cancel_req, 0);
1913}
1914
1915static int btrfs_init_btree_inode(struct super_block *sb)
1916{
1917 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1918 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1919 fs_info->tree_root);
1920 struct inode *inode;
1921
1922 inode = new_inode(sb);
1923 if (!inode)
1924 return -ENOMEM;
1925
1926 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1927 set_nlink(inode, 1);
1928 /*
1929 * we set the i_size on the btree inode to the max possible int.
1930 * the real end of the address space is determined by all of
1931 * the devices in the system
1932 */
1933 inode->i_size = OFFSET_MAX;
1934 inode->i_mapping->a_ops = &btree_aops;
1935 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1936
1937 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1938 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1939 IO_TREE_BTREE_INODE_IO);
1940 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1941
1942 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1943 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1944 BTRFS_I(inode)->location.type = 0;
1945 BTRFS_I(inode)->location.offset = 0;
1946 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1947 __insert_inode_hash(inode, hash);
1948 fs_info->btree_inode = inode;
1949
1950 return 0;
1951}
1952
1953static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1954{
1955 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1956 init_rwsem(&fs_info->dev_replace.rwsem);
1957 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1958}
1959
1960static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1961{
1962 spin_lock_init(&fs_info->qgroup_lock);
1963 mutex_init(&fs_info->qgroup_ioctl_lock);
1964 fs_info->qgroup_tree = RB_ROOT;
1965 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1966 fs_info->qgroup_seq = 1;
1967 fs_info->qgroup_ulist = NULL;
1968 fs_info->qgroup_rescan_running = false;
1969 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1970 mutex_init(&fs_info->qgroup_rescan_lock);
1971}
1972
1973static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1974{
1975 u32 max_active = fs_info->thread_pool_size;
1976 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1977 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1978
1979 fs_info->workers =
1980 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1981
1982 fs_info->delalloc_workers =
1983 btrfs_alloc_workqueue(fs_info, "delalloc",
1984 flags, max_active, 2);
1985
1986 fs_info->flush_workers =
1987 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1988 flags, max_active, 0);
1989
1990 fs_info->caching_workers =
1991 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1992
1993 fs_info->fixup_workers =
1994 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1995
1996 fs_info->endio_workers =
1997 alloc_workqueue("btrfs-endio", flags, max_active);
1998 fs_info->endio_meta_workers =
1999 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2000 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2001 fs_info->endio_write_workers =
2002 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2003 max_active, 2);
2004 fs_info->compressed_write_workers =
2005 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2006 fs_info->endio_freespace_worker =
2007 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2008 max_active, 0);
2009 fs_info->delayed_workers =
2010 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2011 max_active, 0);
2012 fs_info->qgroup_rescan_workers =
2013 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2014 ordered_flags);
2015 fs_info->discard_ctl.discard_workers =
2016 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2017
2018 if (!(fs_info->workers &&
2019 fs_info->delalloc_workers && fs_info->flush_workers &&
2020 fs_info->endio_workers && fs_info->endio_meta_workers &&
2021 fs_info->compressed_write_workers &&
2022 fs_info->endio_write_workers &&
2023 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2024 fs_info->caching_workers && fs_info->fixup_workers &&
2025 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2026 fs_info->discard_ctl.discard_workers)) {
2027 return -ENOMEM;
2028 }
2029
2030 return 0;
2031}
2032
2033static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2034{
2035 struct crypto_shash *csum_shash;
2036 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2037
2038 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2039
2040 if (IS_ERR(csum_shash)) {
2041 btrfs_err(fs_info, "error allocating %s hash for checksum",
2042 csum_driver);
2043 return PTR_ERR(csum_shash);
2044 }
2045
2046 fs_info->csum_shash = csum_shash;
2047
2048 /*
2049 * Check if the checksum implementation is a fast accelerated one.
2050 * As-is this is a bit of a hack and should be replaced once the csum
2051 * implementations provide that information themselves.
2052 */
2053 switch (csum_type) {
2054 case BTRFS_CSUM_TYPE_CRC32:
2055 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2056 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2057 break;
2058 case BTRFS_CSUM_TYPE_XXHASH:
2059 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2060 break;
2061 default:
2062 break;
2063 }
2064
2065 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2066 btrfs_super_csum_name(csum_type),
2067 crypto_shash_driver_name(csum_shash));
2068 return 0;
2069}
2070
2071static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2072 struct btrfs_fs_devices *fs_devices)
2073{
2074 int ret;
2075 struct btrfs_tree_parent_check check = { 0 };
2076 struct btrfs_root *log_tree_root;
2077 struct btrfs_super_block *disk_super = fs_info->super_copy;
2078 u64 bytenr = btrfs_super_log_root(disk_super);
2079 int level = btrfs_super_log_root_level(disk_super);
2080
2081 if (fs_devices->rw_devices == 0) {
2082 btrfs_warn(fs_info, "log replay required on RO media");
2083 return -EIO;
2084 }
2085
2086 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2087 GFP_KERNEL);
2088 if (!log_tree_root)
2089 return -ENOMEM;
2090
2091 check.level = level;
2092 check.transid = fs_info->generation + 1;
2093 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2094 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2095 if (IS_ERR(log_tree_root->node)) {
2096 btrfs_warn(fs_info, "failed to read log tree");
2097 ret = PTR_ERR(log_tree_root->node);
2098 log_tree_root->node = NULL;
2099 btrfs_put_root(log_tree_root);
2100 return ret;
2101 }
2102 if (!extent_buffer_uptodate(log_tree_root->node)) {
2103 btrfs_err(fs_info, "failed to read log tree");
2104 btrfs_put_root(log_tree_root);
2105 return -EIO;
2106 }
2107
2108 /* returns with log_tree_root freed on success */
2109 ret = btrfs_recover_log_trees(log_tree_root);
2110 if (ret) {
2111 btrfs_handle_fs_error(fs_info, ret,
2112 "Failed to recover log tree");
2113 btrfs_put_root(log_tree_root);
2114 return ret;
2115 }
2116
2117 if (sb_rdonly(fs_info->sb)) {
2118 ret = btrfs_commit_super(fs_info);
2119 if (ret)
2120 return ret;
2121 }
2122
2123 return 0;
2124}
2125
2126static int load_global_roots_objectid(struct btrfs_root *tree_root,
2127 struct btrfs_path *path, u64 objectid,
2128 const char *name)
2129{
2130 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2131 struct btrfs_root *root;
2132 u64 max_global_id = 0;
2133 int ret;
2134 struct btrfs_key key = {
2135 .objectid = objectid,
2136 .type = BTRFS_ROOT_ITEM_KEY,
2137 .offset = 0,
2138 };
2139 bool found = false;
2140
2141 /* If we have IGNOREDATACSUMS skip loading these roots. */
2142 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2143 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2144 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2145 return 0;
2146 }
2147
2148 while (1) {
2149 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2150 if (ret < 0)
2151 break;
2152
2153 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2154 ret = btrfs_next_leaf(tree_root, path);
2155 if (ret) {
2156 if (ret > 0)
2157 ret = 0;
2158 break;
2159 }
2160 }
2161 ret = 0;
2162
2163 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2164 if (key.objectid != objectid)
2165 break;
2166 btrfs_release_path(path);
2167
2168 /*
2169 * Just worry about this for extent tree, it'll be the same for
2170 * everybody.
2171 */
2172 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2173 max_global_id = max(max_global_id, key.offset);
2174
2175 found = true;
2176 root = read_tree_root_path(tree_root, path, &key);
2177 if (IS_ERR(root)) {
2178 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2179 ret = PTR_ERR(root);
2180 break;
2181 }
2182 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2183 ret = btrfs_global_root_insert(root);
2184 if (ret) {
2185 btrfs_put_root(root);
2186 break;
2187 }
2188 key.offset++;
2189 }
2190 btrfs_release_path(path);
2191
2192 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2193 fs_info->nr_global_roots = max_global_id + 1;
2194
2195 if (!found || ret) {
2196 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2197 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2198
2199 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2200 ret = ret ? ret : -ENOENT;
2201 else
2202 ret = 0;
2203 btrfs_err(fs_info, "failed to load root %s", name);
2204 }
2205 return ret;
2206}
2207
2208static int load_global_roots(struct btrfs_root *tree_root)
2209{
2210 struct btrfs_path *path;
2211 int ret = 0;
2212
2213 path = btrfs_alloc_path();
2214 if (!path)
2215 return -ENOMEM;
2216
2217 ret = load_global_roots_objectid(tree_root, path,
2218 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2219 if (ret)
2220 goto out;
2221 ret = load_global_roots_objectid(tree_root, path,
2222 BTRFS_CSUM_TREE_OBJECTID, "csum");
2223 if (ret)
2224 goto out;
2225 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2226 goto out;
2227 ret = load_global_roots_objectid(tree_root, path,
2228 BTRFS_FREE_SPACE_TREE_OBJECTID,
2229 "free space");
2230out:
2231 btrfs_free_path(path);
2232 return ret;
2233}
2234
2235static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2236{
2237 struct btrfs_root *tree_root = fs_info->tree_root;
2238 struct btrfs_root *root;
2239 struct btrfs_key location;
2240 int ret;
2241
2242 BUG_ON(!fs_info->tree_root);
2243
2244 ret = load_global_roots(tree_root);
2245 if (ret)
2246 return ret;
2247
2248 location.type = BTRFS_ROOT_ITEM_KEY;
2249 location.offset = 0;
2250
2251 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2252 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2253 root = btrfs_read_tree_root(tree_root, &location);
2254 if (IS_ERR(root)) {
2255 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2256 ret = PTR_ERR(root);
2257 goto out;
2258 }
2259 } else {
2260 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2261 fs_info->block_group_root = root;
2262 }
2263 }
2264
2265 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2266 root = btrfs_read_tree_root(tree_root, &location);
2267 if (IS_ERR(root)) {
2268 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2269 ret = PTR_ERR(root);
2270 goto out;
2271 }
2272 } else {
2273 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2274 fs_info->dev_root = root;
2275 }
2276 /* Initialize fs_info for all devices in any case */
2277 ret = btrfs_init_devices_late(fs_info);
2278 if (ret)
2279 goto out;
2280
2281 /*
2282 * This tree can share blocks with some other fs tree during relocation
2283 * and we need a proper setup by btrfs_get_fs_root
2284 */
2285 root = btrfs_get_fs_root(tree_root->fs_info,
2286 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2287 if (IS_ERR(root)) {
2288 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2289 ret = PTR_ERR(root);
2290 goto out;
2291 }
2292 } else {
2293 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2294 fs_info->data_reloc_root = root;
2295 }
2296
2297 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2298 root = btrfs_read_tree_root(tree_root, &location);
2299 if (!IS_ERR(root)) {
2300 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2301 fs_info->quota_root = root;
2302 }
2303
2304 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2305 root = btrfs_read_tree_root(tree_root, &location);
2306 if (IS_ERR(root)) {
2307 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2308 ret = PTR_ERR(root);
2309 if (ret != -ENOENT)
2310 goto out;
2311 }
2312 } else {
2313 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2314 fs_info->uuid_root = root;
2315 }
2316
2317 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2318 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2319 root = btrfs_read_tree_root(tree_root, &location);
2320 if (IS_ERR(root)) {
2321 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2322 ret = PTR_ERR(root);
2323 goto out;
2324 }
2325 } else {
2326 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2327 fs_info->stripe_root = root;
2328 }
2329 }
2330
2331 return 0;
2332out:
2333 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2334 location.objectid, ret);
2335 return ret;
2336}
2337
2338/*
2339 * Real super block validation
2340 * NOTE: super csum type and incompat features will not be checked here.
2341 *
2342 * @sb: super block to check
2343 * @mirror_num: the super block number to check its bytenr:
2344 * 0 the primary (1st) sb
2345 * 1, 2 2nd and 3rd backup copy
2346 * -1 skip bytenr check
2347 */
2348int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2349 struct btrfs_super_block *sb, int mirror_num)
2350{
2351 u64 nodesize = btrfs_super_nodesize(sb);
2352 u64 sectorsize = btrfs_super_sectorsize(sb);
2353 int ret = 0;
2354
2355 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2356 btrfs_err(fs_info, "no valid FS found");
2357 ret = -EINVAL;
2358 }
2359 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2360 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2361 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2362 ret = -EINVAL;
2363 }
2364 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2365 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2366 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2367 ret = -EINVAL;
2368 }
2369 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2370 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2371 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2372 ret = -EINVAL;
2373 }
2374 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2375 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2376 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2377 ret = -EINVAL;
2378 }
2379
2380 /*
2381 * Check sectorsize and nodesize first, other check will need it.
2382 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2383 */
2384 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2385 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2386 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2387 ret = -EINVAL;
2388 }
2389
2390 /*
2391 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2392 *
2393 * We can support 16K sectorsize with 64K page size without problem,
2394 * but such sectorsize/pagesize combination doesn't make much sense.
2395 * 4K will be our future standard, PAGE_SIZE is supported from the very
2396 * beginning.
2397 */
2398 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2399 btrfs_err(fs_info,
2400 "sectorsize %llu not yet supported for page size %lu",
2401 sectorsize, PAGE_SIZE);
2402 ret = -EINVAL;
2403 }
2404
2405 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2406 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2407 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2408 ret = -EINVAL;
2409 }
2410 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2411 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2412 le32_to_cpu(sb->__unused_leafsize), nodesize);
2413 ret = -EINVAL;
2414 }
2415
2416 /* Root alignment check */
2417 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2418 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2419 btrfs_super_root(sb));
2420 ret = -EINVAL;
2421 }
2422 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2423 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2424 btrfs_super_chunk_root(sb));
2425 ret = -EINVAL;
2426 }
2427 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2428 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2429 btrfs_super_log_root(sb));
2430 ret = -EINVAL;
2431 }
2432
2433 if (!fs_info->fs_devices->temp_fsid &&
2434 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2435 btrfs_err(fs_info,
2436 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2437 sb->fsid, fs_info->fs_devices->fsid);
2438 ret = -EINVAL;
2439 }
2440
2441 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2442 BTRFS_FSID_SIZE) != 0) {
2443 btrfs_err(fs_info,
2444"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2445 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2446 ret = -EINVAL;
2447 }
2448
2449 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2450 BTRFS_FSID_SIZE) != 0) {
2451 btrfs_err(fs_info,
2452 "dev_item UUID does not match metadata fsid: %pU != %pU",
2453 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2454 ret = -EINVAL;
2455 }
2456
2457 /*
2458 * Artificial requirement for block-group-tree to force newer features
2459 * (free-space-tree, no-holes) so the test matrix is smaller.
2460 */
2461 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2462 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2463 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2464 btrfs_err(fs_info,
2465 "block-group-tree feature requires fres-space-tree and no-holes");
2466 ret = -EINVAL;
2467 }
2468
2469 /*
2470 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2471 * done later
2472 */
2473 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2474 btrfs_err(fs_info, "bytes_used is too small %llu",
2475 btrfs_super_bytes_used(sb));
2476 ret = -EINVAL;
2477 }
2478 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2479 btrfs_err(fs_info, "invalid stripesize %u",
2480 btrfs_super_stripesize(sb));
2481 ret = -EINVAL;
2482 }
2483 if (btrfs_super_num_devices(sb) > (1UL << 31))
2484 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2485 btrfs_super_num_devices(sb));
2486 if (btrfs_super_num_devices(sb) == 0) {
2487 btrfs_err(fs_info, "number of devices is 0");
2488 ret = -EINVAL;
2489 }
2490
2491 if (mirror_num >= 0 &&
2492 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2493 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2494 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2495 ret = -EINVAL;
2496 }
2497
2498 /*
2499 * Obvious sys_chunk_array corruptions, it must hold at least one key
2500 * and one chunk
2501 */
2502 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2503 btrfs_err(fs_info, "system chunk array too big %u > %u",
2504 btrfs_super_sys_array_size(sb),
2505 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2506 ret = -EINVAL;
2507 }
2508 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2509 + sizeof(struct btrfs_chunk)) {
2510 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2511 btrfs_super_sys_array_size(sb),
2512 sizeof(struct btrfs_disk_key)
2513 + sizeof(struct btrfs_chunk));
2514 ret = -EINVAL;
2515 }
2516
2517 /*
2518 * The generation is a global counter, we'll trust it more than the others
2519 * but it's still possible that it's the one that's wrong.
2520 */
2521 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2522 btrfs_warn(fs_info,
2523 "suspicious: generation < chunk_root_generation: %llu < %llu",
2524 btrfs_super_generation(sb),
2525 btrfs_super_chunk_root_generation(sb));
2526 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2527 && btrfs_super_cache_generation(sb) != (u64)-1)
2528 btrfs_warn(fs_info,
2529 "suspicious: generation < cache_generation: %llu < %llu",
2530 btrfs_super_generation(sb),
2531 btrfs_super_cache_generation(sb));
2532
2533 return ret;
2534}
2535
2536/*
2537 * Validation of super block at mount time.
2538 * Some checks already done early at mount time, like csum type and incompat
2539 * flags will be skipped.
2540 */
2541static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2542{
2543 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2544}
2545
2546/*
2547 * Validation of super block at write time.
2548 * Some checks like bytenr check will be skipped as their values will be
2549 * overwritten soon.
2550 * Extra checks like csum type and incompat flags will be done here.
2551 */
2552static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2553 struct btrfs_super_block *sb)
2554{
2555 int ret;
2556
2557 ret = btrfs_validate_super(fs_info, sb, -1);
2558 if (ret < 0)
2559 goto out;
2560 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2561 ret = -EUCLEAN;
2562 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2563 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2564 goto out;
2565 }
2566 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2567 ret = -EUCLEAN;
2568 btrfs_err(fs_info,
2569 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2570 btrfs_super_incompat_flags(sb),
2571 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2572 goto out;
2573 }
2574out:
2575 if (ret < 0)
2576 btrfs_err(fs_info,
2577 "super block corruption detected before writing it to disk");
2578 return ret;
2579}
2580
2581static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2582{
2583 struct btrfs_tree_parent_check check = {
2584 .level = level,
2585 .transid = gen,
2586 .owner_root = root->root_key.objectid
2587 };
2588 int ret = 0;
2589
2590 root->node = read_tree_block(root->fs_info, bytenr, &check);
2591 if (IS_ERR(root->node)) {
2592 ret = PTR_ERR(root->node);
2593 root->node = NULL;
2594 return ret;
2595 }
2596 if (!extent_buffer_uptodate(root->node)) {
2597 free_extent_buffer(root->node);
2598 root->node = NULL;
2599 return -EIO;
2600 }
2601
2602 btrfs_set_root_node(&root->root_item, root->node);
2603 root->commit_root = btrfs_root_node(root);
2604 btrfs_set_root_refs(&root->root_item, 1);
2605 return ret;
2606}
2607
2608static int load_important_roots(struct btrfs_fs_info *fs_info)
2609{
2610 struct btrfs_super_block *sb = fs_info->super_copy;
2611 u64 gen, bytenr;
2612 int level, ret;
2613
2614 bytenr = btrfs_super_root(sb);
2615 gen = btrfs_super_generation(sb);
2616 level = btrfs_super_root_level(sb);
2617 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2618 if (ret) {
2619 btrfs_warn(fs_info, "couldn't read tree root");
2620 return ret;
2621 }
2622 return 0;
2623}
2624
2625static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2626{
2627 int backup_index = find_newest_super_backup(fs_info);
2628 struct btrfs_super_block *sb = fs_info->super_copy;
2629 struct btrfs_root *tree_root = fs_info->tree_root;
2630 bool handle_error = false;
2631 int ret = 0;
2632 int i;
2633
2634 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2635 if (handle_error) {
2636 if (!IS_ERR(tree_root->node))
2637 free_extent_buffer(tree_root->node);
2638 tree_root->node = NULL;
2639
2640 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2641 break;
2642
2643 free_root_pointers(fs_info, 0);
2644
2645 /*
2646 * Don't use the log in recovery mode, it won't be
2647 * valid
2648 */
2649 btrfs_set_super_log_root(sb, 0);
2650
2651 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2652 ret = read_backup_root(fs_info, i);
2653 backup_index = ret;
2654 if (ret < 0)
2655 return ret;
2656 }
2657
2658 ret = load_important_roots(fs_info);
2659 if (ret) {
2660 handle_error = true;
2661 continue;
2662 }
2663
2664 /*
2665 * No need to hold btrfs_root::objectid_mutex since the fs
2666 * hasn't been fully initialised and we are the only user
2667 */
2668 ret = btrfs_init_root_free_objectid(tree_root);
2669 if (ret < 0) {
2670 handle_error = true;
2671 continue;
2672 }
2673
2674 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2675
2676 ret = btrfs_read_roots(fs_info);
2677 if (ret < 0) {
2678 handle_error = true;
2679 continue;
2680 }
2681
2682 /* All successful */
2683 fs_info->generation = btrfs_header_generation(tree_root->node);
2684 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2685 fs_info->last_reloc_trans = 0;
2686
2687 /* Always begin writing backup roots after the one being used */
2688 if (backup_index < 0) {
2689 fs_info->backup_root_index = 0;
2690 } else {
2691 fs_info->backup_root_index = backup_index + 1;
2692 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2693 }
2694 break;
2695 }
2696
2697 return ret;
2698}
2699
2700void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2701{
2702 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2703 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2704 INIT_LIST_HEAD(&fs_info->trans_list);
2705 INIT_LIST_HEAD(&fs_info->dead_roots);
2706 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2707 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2708 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2709 spin_lock_init(&fs_info->delalloc_root_lock);
2710 spin_lock_init(&fs_info->trans_lock);
2711 spin_lock_init(&fs_info->fs_roots_radix_lock);
2712 spin_lock_init(&fs_info->delayed_iput_lock);
2713 spin_lock_init(&fs_info->defrag_inodes_lock);
2714 spin_lock_init(&fs_info->super_lock);
2715 spin_lock_init(&fs_info->buffer_lock);
2716 spin_lock_init(&fs_info->unused_bgs_lock);
2717 spin_lock_init(&fs_info->treelog_bg_lock);
2718 spin_lock_init(&fs_info->zone_active_bgs_lock);
2719 spin_lock_init(&fs_info->relocation_bg_lock);
2720 rwlock_init(&fs_info->tree_mod_log_lock);
2721 rwlock_init(&fs_info->global_root_lock);
2722 mutex_init(&fs_info->unused_bg_unpin_mutex);
2723 mutex_init(&fs_info->reclaim_bgs_lock);
2724 mutex_init(&fs_info->reloc_mutex);
2725 mutex_init(&fs_info->delalloc_root_mutex);
2726 mutex_init(&fs_info->zoned_meta_io_lock);
2727 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2728 seqlock_init(&fs_info->profiles_lock);
2729
2730 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2731 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2732 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2733 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2734 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2735 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2736 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2737 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2738 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2739 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2740 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2741 BTRFS_LOCKDEP_TRANS_COMPLETED);
2742
2743 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2744 INIT_LIST_HEAD(&fs_info->space_info);
2745 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2746 INIT_LIST_HEAD(&fs_info->unused_bgs);
2747 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2748 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2749#ifdef CONFIG_BTRFS_DEBUG
2750 INIT_LIST_HEAD(&fs_info->allocated_roots);
2751 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2752 spin_lock_init(&fs_info->eb_leak_lock);
2753#endif
2754 fs_info->mapping_tree = RB_ROOT_CACHED;
2755 rwlock_init(&fs_info->mapping_tree_lock);
2756 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2757 BTRFS_BLOCK_RSV_GLOBAL);
2758 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2759 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2760 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2761 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2762 BTRFS_BLOCK_RSV_DELOPS);
2763 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2764 BTRFS_BLOCK_RSV_DELREFS);
2765
2766 atomic_set(&fs_info->async_delalloc_pages, 0);
2767 atomic_set(&fs_info->defrag_running, 0);
2768 atomic_set(&fs_info->nr_delayed_iputs, 0);
2769 atomic64_set(&fs_info->tree_mod_seq, 0);
2770 fs_info->global_root_tree = RB_ROOT;
2771 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2772 fs_info->metadata_ratio = 0;
2773 fs_info->defrag_inodes = RB_ROOT;
2774 atomic64_set(&fs_info->free_chunk_space, 0);
2775 fs_info->tree_mod_log = RB_ROOT;
2776 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2777 btrfs_init_ref_verify(fs_info);
2778
2779 fs_info->thread_pool_size = min_t(unsigned long,
2780 num_online_cpus() + 2, 8);
2781
2782 INIT_LIST_HEAD(&fs_info->ordered_roots);
2783 spin_lock_init(&fs_info->ordered_root_lock);
2784
2785 btrfs_init_scrub(fs_info);
2786 btrfs_init_balance(fs_info);
2787 btrfs_init_async_reclaim_work(fs_info);
2788
2789 rwlock_init(&fs_info->block_group_cache_lock);
2790 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2791
2792 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2793 IO_TREE_FS_EXCLUDED_EXTENTS);
2794
2795 mutex_init(&fs_info->ordered_operations_mutex);
2796 mutex_init(&fs_info->tree_log_mutex);
2797 mutex_init(&fs_info->chunk_mutex);
2798 mutex_init(&fs_info->transaction_kthread_mutex);
2799 mutex_init(&fs_info->cleaner_mutex);
2800 mutex_init(&fs_info->ro_block_group_mutex);
2801 init_rwsem(&fs_info->commit_root_sem);
2802 init_rwsem(&fs_info->cleanup_work_sem);
2803 init_rwsem(&fs_info->subvol_sem);
2804 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2805
2806 btrfs_init_dev_replace_locks(fs_info);
2807 btrfs_init_qgroup(fs_info);
2808 btrfs_discard_init(fs_info);
2809
2810 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2811 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2812
2813 init_waitqueue_head(&fs_info->transaction_throttle);
2814 init_waitqueue_head(&fs_info->transaction_wait);
2815 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2816 init_waitqueue_head(&fs_info->async_submit_wait);
2817 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2818
2819 /* Usable values until the real ones are cached from the superblock */
2820 fs_info->nodesize = 4096;
2821 fs_info->sectorsize = 4096;
2822 fs_info->sectorsize_bits = ilog2(4096);
2823 fs_info->stripesize = 4096;
2824
2825 /* Default compress algorithm when user does -o compress */
2826 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2827
2828 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2829
2830 spin_lock_init(&fs_info->swapfile_pins_lock);
2831 fs_info->swapfile_pins = RB_ROOT;
2832
2833 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2834 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2835}
2836
2837static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2838{
2839 int ret;
2840
2841 fs_info->sb = sb;
2842 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2843 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2844
2845 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2846 if (ret)
2847 return ret;
2848
2849 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2850 if (ret)
2851 return ret;
2852
2853 fs_info->dirty_metadata_batch = PAGE_SIZE *
2854 (1 + ilog2(nr_cpu_ids));
2855
2856 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2857 if (ret)
2858 return ret;
2859
2860 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2861 GFP_KERNEL);
2862 if (ret)
2863 return ret;
2864
2865 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2866 GFP_KERNEL);
2867 if (!fs_info->delayed_root)
2868 return -ENOMEM;
2869 btrfs_init_delayed_root(fs_info->delayed_root);
2870
2871 if (sb_rdonly(sb))
2872 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2873
2874 return btrfs_alloc_stripe_hash_table(fs_info);
2875}
2876
2877static int btrfs_uuid_rescan_kthread(void *data)
2878{
2879 struct btrfs_fs_info *fs_info = data;
2880 int ret;
2881
2882 /*
2883 * 1st step is to iterate through the existing UUID tree and
2884 * to delete all entries that contain outdated data.
2885 * 2nd step is to add all missing entries to the UUID tree.
2886 */
2887 ret = btrfs_uuid_tree_iterate(fs_info);
2888 if (ret < 0) {
2889 if (ret != -EINTR)
2890 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2891 ret);
2892 up(&fs_info->uuid_tree_rescan_sem);
2893 return ret;
2894 }
2895 return btrfs_uuid_scan_kthread(data);
2896}
2897
2898static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2899{
2900 struct task_struct *task;
2901
2902 down(&fs_info->uuid_tree_rescan_sem);
2903 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2904 if (IS_ERR(task)) {
2905 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2906 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2907 up(&fs_info->uuid_tree_rescan_sem);
2908 return PTR_ERR(task);
2909 }
2910
2911 return 0;
2912}
2913
2914static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2915{
2916 u64 root_objectid = 0;
2917 struct btrfs_root *gang[8];
2918 int i = 0;
2919 int err = 0;
2920 unsigned int ret = 0;
2921
2922 while (1) {
2923 spin_lock(&fs_info->fs_roots_radix_lock);
2924 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2925 (void **)gang, root_objectid,
2926 ARRAY_SIZE(gang));
2927 if (!ret) {
2928 spin_unlock(&fs_info->fs_roots_radix_lock);
2929 break;
2930 }
2931 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2932
2933 for (i = 0; i < ret; i++) {
2934 /* Avoid to grab roots in dead_roots. */
2935 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2936 gang[i] = NULL;
2937 continue;
2938 }
2939 /* Grab all the search result for later use. */
2940 gang[i] = btrfs_grab_root(gang[i]);
2941 }
2942 spin_unlock(&fs_info->fs_roots_radix_lock);
2943
2944 for (i = 0; i < ret; i++) {
2945 if (!gang[i])
2946 continue;
2947 root_objectid = gang[i]->root_key.objectid;
2948 err = btrfs_orphan_cleanup(gang[i]);
2949 if (err)
2950 goto out;
2951 btrfs_put_root(gang[i]);
2952 }
2953 root_objectid++;
2954 }
2955out:
2956 /* Release the uncleaned roots due to error. */
2957 for (; i < ret; i++) {
2958 if (gang[i])
2959 btrfs_put_root(gang[i]);
2960 }
2961 return err;
2962}
2963
2964/*
2965 * Mounting logic specific to read-write file systems. Shared by open_ctree
2966 * and btrfs_remount when remounting from read-only to read-write.
2967 */
2968int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2969{
2970 int ret;
2971 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2972 bool rebuild_free_space_tree = false;
2973
2974 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2975 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2976 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2977 btrfs_warn(fs_info,
2978 "'clear_cache' option is ignored with extent tree v2");
2979 else
2980 rebuild_free_space_tree = true;
2981 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2982 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2983 btrfs_warn(fs_info, "free space tree is invalid");
2984 rebuild_free_space_tree = true;
2985 }
2986
2987 if (rebuild_free_space_tree) {
2988 btrfs_info(fs_info, "rebuilding free space tree");
2989 ret = btrfs_rebuild_free_space_tree(fs_info);
2990 if (ret) {
2991 btrfs_warn(fs_info,
2992 "failed to rebuild free space tree: %d", ret);
2993 goto out;
2994 }
2995 }
2996
2997 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2998 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
2999 btrfs_info(fs_info, "disabling free space tree");
3000 ret = btrfs_delete_free_space_tree(fs_info);
3001 if (ret) {
3002 btrfs_warn(fs_info,
3003 "failed to disable free space tree: %d", ret);
3004 goto out;
3005 }
3006 }
3007
3008 /*
3009 * btrfs_find_orphan_roots() is responsible for finding all the dead
3010 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3011 * them into the fs_info->fs_roots_radix tree. This must be done before
3012 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3013 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3014 * item before the root's tree is deleted - this means that if we unmount
3015 * or crash before the deletion completes, on the next mount we will not
3016 * delete what remains of the tree because the orphan item does not
3017 * exists anymore, which is what tells us we have a pending deletion.
3018 */
3019 ret = btrfs_find_orphan_roots(fs_info);
3020 if (ret)
3021 goto out;
3022
3023 ret = btrfs_cleanup_fs_roots(fs_info);
3024 if (ret)
3025 goto out;
3026
3027 down_read(&fs_info->cleanup_work_sem);
3028 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3029 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3030 up_read(&fs_info->cleanup_work_sem);
3031 goto out;
3032 }
3033 up_read(&fs_info->cleanup_work_sem);
3034
3035 mutex_lock(&fs_info->cleaner_mutex);
3036 ret = btrfs_recover_relocation(fs_info);
3037 mutex_unlock(&fs_info->cleaner_mutex);
3038 if (ret < 0) {
3039 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3040 goto out;
3041 }
3042
3043 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3044 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3045 btrfs_info(fs_info, "creating free space tree");
3046 ret = btrfs_create_free_space_tree(fs_info);
3047 if (ret) {
3048 btrfs_warn(fs_info,
3049 "failed to create free space tree: %d", ret);
3050 goto out;
3051 }
3052 }
3053
3054 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3055 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3056 if (ret)
3057 goto out;
3058 }
3059
3060 ret = btrfs_resume_balance_async(fs_info);
3061 if (ret)
3062 goto out;
3063
3064 ret = btrfs_resume_dev_replace_async(fs_info);
3065 if (ret) {
3066 btrfs_warn(fs_info, "failed to resume dev_replace");
3067 goto out;
3068 }
3069
3070 btrfs_qgroup_rescan_resume(fs_info);
3071
3072 if (!fs_info->uuid_root) {
3073 btrfs_info(fs_info, "creating UUID tree");
3074 ret = btrfs_create_uuid_tree(fs_info);
3075 if (ret) {
3076 btrfs_warn(fs_info,
3077 "failed to create the UUID tree %d", ret);
3078 goto out;
3079 }
3080 }
3081
3082out:
3083 return ret;
3084}
3085
3086/*
3087 * Do various sanity and dependency checks of different features.
3088 *
3089 * @is_rw_mount: If the mount is read-write.
3090 *
3091 * This is the place for less strict checks (like for subpage or artificial
3092 * feature dependencies).
3093 *
3094 * For strict checks or possible corruption detection, see
3095 * btrfs_validate_super().
3096 *
3097 * This should be called after btrfs_parse_options(), as some mount options
3098 * (space cache related) can modify on-disk format like free space tree and
3099 * screw up certain feature dependencies.
3100 */
3101int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3102{
3103 struct btrfs_super_block *disk_super = fs_info->super_copy;
3104 u64 incompat = btrfs_super_incompat_flags(disk_super);
3105 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3106 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3107
3108 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3109 btrfs_err(fs_info,
3110 "cannot mount because of unknown incompat features (0x%llx)",
3111 incompat);
3112 return -EINVAL;
3113 }
3114
3115 /* Runtime limitation for mixed block groups. */
3116 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3117 (fs_info->sectorsize != fs_info->nodesize)) {
3118 btrfs_err(fs_info,
3119"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3120 fs_info->nodesize, fs_info->sectorsize);
3121 return -EINVAL;
3122 }
3123
3124 /* Mixed backref is an always-enabled feature. */
3125 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3126
3127 /* Set compression related flags just in case. */
3128 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3129 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3130 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3131 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3132
3133 /*
3134 * An ancient flag, which should really be marked deprecated.
3135 * Such runtime limitation doesn't really need a incompat flag.
3136 */
3137 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3138 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3139
3140 if (compat_ro_unsupp && is_rw_mount) {
3141 btrfs_err(fs_info,
3142 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3143 compat_ro);
3144 return -EINVAL;
3145 }
3146
3147 /*
3148 * We have unsupported RO compat features, although RO mounted, we
3149 * should not cause any metadata writes, including log replay.
3150 * Or we could screw up whatever the new feature requires.
3151 */
3152 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3153 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3154 btrfs_err(fs_info,
3155"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3156 compat_ro);
3157 return -EINVAL;
3158 }
3159
3160 /*
3161 * Artificial limitations for block group tree, to force
3162 * block-group-tree to rely on no-holes and free-space-tree.
3163 */
3164 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3165 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3166 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3167 btrfs_err(fs_info,
3168"block-group-tree feature requires no-holes and free-space-tree features");
3169 return -EINVAL;
3170 }
3171
3172 /*
3173 * Subpage runtime limitation on v1 cache.
3174 *
3175 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3176 * we're already defaulting to v2 cache, no need to bother v1 as it's
3177 * going to be deprecated anyway.
3178 */
3179 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3180 btrfs_warn(fs_info,
3181 "v1 space cache is not supported for page size %lu with sectorsize %u",
3182 PAGE_SIZE, fs_info->sectorsize);
3183 return -EINVAL;
3184 }
3185
3186 /* This can be called by remount, we need to protect the super block. */
3187 spin_lock(&fs_info->super_lock);
3188 btrfs_set_super_incompat_flags(disk_super, incompat);
3189 spin_unlock(&fs_info->super_lock);
3190
3191 return 0;
3192}
3193
3194int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3195 char *options)
3196{
3197 u32 sectorsize;
3198 u32 nodesize;
3199 u32 stripesize;
3200 u64 generation;
3201 u16 csum_type;
3202 struct btrfs_super_block *disk_super;
3203 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3204 struct btrfs_root *tree_root;
3205 struct btrfs_root *chunk_root;
3206 int ret;
3207 int level;
3208
3209 ret = init_mount_fs_info(fs_info, sb);
3210 if (ret)
3211 goto fail;
3212
3213 /* These need to be init'ed before we start creating inodes and such. */
3214 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3215 GFP_KERNEL);
3216 fs_info->tree_root = tree_root;
3217 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3218 GFP_KERNEL);
3219 fs_info->chunk_root = chunk_root;
3220 if (!tree_root || !chunk_root) {
3221 ret = -ENOMEM;
3222 goto fail;
3223 }
3224
3225 ret = btrfs_init_btree_inode(sb);
3226 if (ret)
3227 goto fail;
3228
3229 invalidate_bdev(fs_devices->latest_dev->bdev);
3230
3231 /*
3232 * Read super block and check the signature bytes only
3233 */
3234 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3235 if (IS_ERR(disk_super)) {
3236 ret = PTR_ERR(disk_super);
3237 goto fail_alloc;
3238 }
3239
3240 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3241 /*
3242 * Verify the type first, if that or the checksum value are
3243 * corrupted, we'll find out
3244 */
3245 csum_type = btrfs_super_csum_type(disk_super);
3246 if (!btrfs_supported_super_csum(csum_type)) {
3247 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3248 csum_type);
3249 ret = -EINVAL;
3250 btrfs_release_disk_super(disk_super);
3251 goto fail_alloc;
3252 }
3253
3254 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3255
3256 ret = btrfs_init_csum_hash(fs_info, csum_type);
3257 if (ret) {
3258 btrfs_release_disk_super(disk_super);
3259 goto fail_alloc;
3260 }
3261
3262 /*
3263 * We want to check superblock checksum, the type is stored inside.
3264 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3265 */
3266 if (btrfs_check_super_csum(fs_info, disk_super)) {
3267 btrfs_err(fs_info, "superblock checksum mismatch");
3268 ret = -EINVAL;
3269 btrfs_release_disk_super(disk_super);
3270 goto fail_alloc;
3271 }
3272
3273 /*
3274 * super_copy is zeroed at allocation time and we never touch the
3275 * following bytes up to INFO_SIZE, the checksum is calculated from
3276 * the whole block of INFO_SIZE
3277 */
3278 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3279 btrfs_release_disk_super(disk_super);
3280
3281 disk_super = fs_info->super_copy;
3282
3283 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3284 sizeof(*fs_info->super_for_commit));
3285
3286 ret = btrfs_validate_mount_super(fs_info);
3287 if (ret) {
3288 btrfs_err(fs_info, "superblock contains fatal errors");
3289 ret = -EINVAL;
3290 goto fail_alloc;
3291 }
3292
3293 if (!btrfs_super_root(disk_super)) {
3294 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3295 ret = -EINVAL;
3296 goto fail_alloc;
3297 }
3298
3299 /* check FS state, whether FS is broken. */
3300 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3301 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3302
3303 /* Set up fs_info before parsing mount options */
3304 nodesize = btrfs_super_nodesize(disk_super);
3305 sectorsize = btrfs_super_sectorsize(disk_super);
3306 stripesize = sectorsize;
3307 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3308 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3309
3310 fs_info->nodesize = nodesize;
3311 fs_info->sectorsize = sectorsize;
3312 fs_info->sectorsize_bits = ilog2(sectorsize);
3313 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3314 fs_info->stripesize = stripesize;
3315
3316 /*
3317 * Handle the space caching options appropriately now that we have the
3318 * super block loaded and validated.
3319 */
3320 btrfs_set_free_space_cache_settings(fs_info);
3321
3322 if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3323 ret = -EINVAL;
3324 goto fail_alloc;
3325 }
3326
3327 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3328 if (ret < 0)
3329 goto fail_alloc;
3330
3331 /*
3332 * At this point our mount options are validated, if we set ->max_inline
3333 * to something non-standard make sure we truncate it to sectorsize.
3334 */
3335 fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3336
3337 if (sectorsize < PAGE_SIZE) {
3338 struct btrfs_subpage_info *subpage_info;
3339
3340 btrfs_warn(fs_info,
3341 "read-write for sector size %u with page size %lu is experimental",
3342 sectorsize, PAGE_SIZE);
3343 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3344 if (!subpage_info) {
3345 ret = -ENOMEM;
3346 goto fail_alloc;
3347 }
3348 btrfs_init_subpage_info(subpage_info, sectorsize);
3349 fs_info->subpage_info = subpage_info;
3350 }
3351
3352 ret = btrfs_init_workqueues(fs_info);
3353 if (ret)
3354 goto fail_sb_buffer;
3355
3356 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3357 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3358
3359 sb->s_blocksize = sectorsize;
3360 sb->s_blocksize_bits = blksize_bits(sectorsize);
3361 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3362
3363 mutex_lock(&fs_info->chunk_mutex);
3364 ret = btrfs_read_sys_array(fs_info);
3365 mutex_unlock(&fs_info->chunk_mutex);
3366 if (ret) {
3367 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3368 goto fail_sb_buffer;
3369 }
3370
3371 generation = btrfs_super_chunk_root_generation(disk_super);
3372 level = btrfs_super_chunk_root_level(disk_super);
3373 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3374 generation, level);
3375 if (ret) {
3376 btrfs_err(fs_info, "failed to read chunk root");
3377 goto fail_tree_roots;
3378 }
3379
3380 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3381 offsetof(struct btrfs_header, chunk_tree_uuid),
3382 BTRFS_UUID_SIZE);
3383
3384 ret = btrfs_read_chunk_tree(fs_info);
3385 if (ret) {
3386 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3387 goto fail_tree_roots;
3388 }
3389
3390 /*
3391 * At this point we know all the devices that make this filesystem,
3392 * including the seed devices but we don't know yet if the replace
3393 * target is required. So free devices that are not part of this
3394 * filesystem but skip the replace target device which is checked
3395 * below in btrfs_init_dev_replace().
3396 */
3397 btrfs_free_extra_devids(fs_devices);
3398 if (!fs_devices->latest_dev->bdev) {
3399 btrfs_err(fs_info, "failed to read devices");
3400 ret = -EIO;
3401 goto fail_tree_roots;
3402 }
3403
3404 ret = init_tree_roots(fs_info);
3405 if (ret)
3406 goto fail_tree_roots;
3407
3408 /*
3409 * Get zone type information of zoned block devices. This will also
3410 * handle emulation of a zoned filesystem if a regular device has the
3411 * zoned incompat feature flag set.
3412 */
3413 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3414 if (ret) {
3415 btrfs_err(fs_info,
3416 "zoned: failed to read device zone info: %d", ret);
3417 goto fail_block_groups;
3418 }
3419
3420 /*
3421 * If we have a uuid root and we're not being told to rescan we need to
3422 * check the generation here so we can set the
3423 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3424 * transaction during a balance or the log replay without updating the
3425 * uuid generation, and then if we crash we would rescan the uuid tree,
3426 * even though it was perfectly fine.
3427 */
3428 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3429 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3430 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3431
3432 ret = btrfs_verify_dev_extents(fs_info);
3433 if (ret) {
3434 btrfs_err(fs_info,
3435 "failed to verify dev extents against chunks: %d",
3436 ret);
3437 goto fail_block_groups;
3438 }
3439 ret = btrfs_recover_balance(fs_info);
3440 if (ret) {
3441 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3442 goto fail_block_groups;
3443 }
3444
3445 ret = btrfs_init_dev_stats(fs_info);
3446 if (ret) {
3447 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3448 goto fail_block_groups;
3449 }
3450
3451 ret = btrfs_init_dev_replace(fs_info);
3452 if (ret) {
3453 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3454 goto fail_block_groups;
3455 }
3456
3457 ret = btrfs_check_zoned_mode(fs_info);
3458 if (ret) {
3459 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3460 ret);
3461 goto fail_block_groups;
3462 }
3463
3464 ret = btrfs_sysfs_add_fsid(fs_devices);
3465 if (ret) {
3466 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3467 ret);
3468 goto fail_block_groups;
3469 }
3470
3471 ret = btrfs_sysfs_add_mounted(fs_info);
3472 if (ret) {
3473 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3474 goto fail_fsdev_sysfs;
3475 }
3476
3477 ret = btrfs_init_space_info(fs_info);
3478 if (ret) {
3479 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3480 goto fail_sysfs;
3481 }
3482
3483 ret = btrfs_read_block_groups(fs_info);
3484 if (ret) {
3485 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3486 goto fail_sysfs;
3487 }
3488
3489 btrfs_free_zone_cache(fs_info);
3490
3491 btrfs_check_active_zone_reservation(fs_info);
3492
3493 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3494 !btrfs_check_rw_degradable(fs_info, NULL)) {
3495 btrfs_warn(fs_info,
3496 "writable mount is not allowed due to too many missing devices");
3497 ret = -EINVAL;
3498 goto fail_sysfs;
3499 }
3500
3501 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3502 "btrfs-cleaner");
3503 if (IS_ERR(fs_info->cleaner_kthread)) {
3504 ret = PTR_ERR(fs_info->cleaner_kthread);
3505 goto fail_sysfs;
3506 }
3507
3508 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3509 tree_root,
3510 "btrfs-transaction");
3511 if (IS_ERR(fs_info->transaction_kthread)) {
3512 ret = PTR_ERR(fs_info->transaction_kthread);
3513 goto fail_cleaner;
3514 }
3515
3516 ret = btrfs_read_qgroup_config(fs_info);
3517 if (ret)
3518 goto fail_trans_kthread;
3519
3520 if (btrfs_build_ref_tree(fs_info))
3521 btrfs_err(fs_info, "couldn't build ref tree");
3522
3523 /* do not make disk changes in broken FS or nologreplay is given */
3524 if (btrfs_super_log_root(disk_super) != 0 &&
3525 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3526 btrfs_info(fs_info, "start tree-log replay");
3527 ret = btrfs_replay_log(fs_info, fs_devices);
3528 if (ret)
3529 goto fail_qgroup;
3530 }
3531
3532 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3533 if (IS_ERR(fs_info->fs_root)) {
3534 ret = PTR_ERR(fs_info->fs_root);
3535 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3536 fs_info->fs_root = NULL;
3537 goto fail_qgroup;
3538 }
3539
3540 if (sb_rdonly(sb))
3541 return 0;
3542
3543 ret = btrfs_start_pre_rw_mount(fs_info);
3544 if (ret) {
3545 close_ctree(fs_info);
3546 return ret;
3547 }
3548 btrfs_discard_resume(fs_info);
3549
3550 if (fs_info->uuid_root &&
3551 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3552 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3553 btrfs_info(fs_info, "checking UUID tree");
3554 ret = btrfs_check_uuid_tree(fs_info);
3555 if (ret) {
3556 btrfs_warn(fs_info,
3557 "failed to check the UUID tree: %d", ret);
3558 close_ctree(fs_info);
3559 return ret;
3560 }
3561 }
3562
3563 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3564
3565 /* Kick the cleaner thread so it'll start deleting snapshots. */
3566 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3567 wake_up_process(fs_info->cleaner_kthread);
3568
3569 return 0;
3570
3571fail_qgroup:
3572 btrfs_free_qgroup_config(fs_info);
3573fail_trans_kthread:
3574 kthread_stop(fs_info->transaction_kthread);
3575 btrfs_cleanup_transaction(fs_info);
3576 btrfs_free_fs_roots(fs_info);
3577fail_cleaner:
3578 kthread_stop(fs_info->cleaner_kthread);
3579
3580 /*
3581 * make sure we're done with the btree inode before we stop our
3582 * kthreads
3583 */
3584 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3585
3586fail_sysfs:
3587 btrfs_sysfs_remove_mounted(fs_info);
3588
3589fail_fsdev_sysfs:
3590 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3591
3592fail_block_groups:
3593 btrfs_put_block_group_cache(fs_info);
3594
3595fail_tree_roots:
3596 if (fs_info->data_reloc_root)
3597 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3598 free_root_pointers(fs_info, true);
3599 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3600
3601fail_sb_buffer:
3602 btrfs_stop_all_workers(fs_info);
3603 btrfs_free_block_groups(fs_info);
3604fail_alloc:
3605 btrfs_mapping_tree_free(fs_info);
3606
3607 iput(fs_info->btree_inode);
3608fail:
3609 btrfs_close_devices(fs_info->fs_devices);
3610 ASSERT(ret < 0);
3611 return ret;
3612}
3613ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3614
3615static void btrfs_end_super_write(struct bio *bio)
3616{
3617 struct btrfs_device *device = bio->bi_private;
3618 struct bio_vec *bvec;
3619 struct bvec_iter_all iter_all;
3620 struct page *page;
3621
3622 bio_for_each_segment_all(bvec, bio, iter_all) {
3623 page = bvec->bv_page;
3624
3625 if (bio->bi_status) {
3626 btrfs_warn_rl_in_rcu(device->fs_info,
3627 "lost page write due to IO error on %s (%d)",
3628 btrfs_dev_name(device),
3629 blk_status_to_errno(bio->bi_status));
3630 ClearPageUptodate(page);
3631 SetPageError(page);
3632 btrfs_dev_stat_inc_and_print(device,
3633 BTRFS_DEV_STAT_WRITE_ERRS);
3634 } else {
3635 SetPageUptodate(page);
3636 }
3637
3638 put_page(page);
3639 unlock_page(page);
3640 }
3641
3642 bio_put(bio);
3643}
3644
3645struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3646 int copy_num, bool drop_cache)
3647{
3648 struct btrfs_super_block *super;
3649 struct page *page;
3650 u64 bytenr, bytenr_orig;
3651 struct address_space *mapping = bdev->bd_inode->i_mapping;
3652 int ret;
3653
3654 bytenr_orig = btrfs_sb_offset(copy_num);
3655 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3656 if (ret == -ENOENT)
3657 return ERR_PTR(-EINVAL);
3658 else if (ret)
3659 return ERR_PTR(ret);
3660
3661 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3662 return ERR_PTR(-EINVAL);
3663
3664 if (drop_cache) {
3665 /* This should only be called with the primary sb. */
3666 ASSERT(copy_num == 0);
3667
3668 /*
3669 * Drop the page of the primary superblock, so later read will
3670 * always read from the device.
3671 */
3672 invalidate_inode_pages2_range(mapping,
3673 bytenr >> PAGE_SHIFT,
3674 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3675 }
3676
3677 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3678 if (IS_ERR(page))
3679 return ERR_CAST(page);
3680
3681 super = page_address(page);
3682 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3683 btrfs_release_disk_super(super);
3684 return ERR_PTR(-ENODATA);
3685 }
3686
3687 if (btrfs_super_bytenr(super) != bytenr_orig) {
3688 btrfs_release_disk_super(super);
3689 return ERR_PTR(-EINVAL);
3690 }
3691
3692 return super;
3693}
3694
3695
3696struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3697{
3698 struct btrfs_super_block *super, *latest = NULL;
3699 int i;
3700 u64 transid = 0;
3701
3702 /* we would like to check all the supers, but that would make
3703 * a btrfs mount succeed after a mkfs from a different FS.
3704 * So, we need to add a special mount option to scan for
3705 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3706 */
3707 for (i = 0; i < 1; i++) {
3708 super = btrfs_read_dev_one_super(bdev, i, false);
3709 if (IS_ERR(super))
3710 continue;
3711
3712 if (!latest || btrfs_super_generation(super) > transid) {
3713 if (latest)
3714 btrfs_release_disk_super(super);
3715
3716 latest = super;
3717 transid = btrfs_super_generation(super);
3718 }
3719 }
3720
3721 return super;
3722}
3723
3724/*
3725 * Write superblock @sb to the @device. Do not wait for completion, all the
3726 * pages we use for writing are locked.
3727 *
3728 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3729 * the expected device size at commit time. Note that max_mirrors must be
3730 * same for write and wait phases.
3731 *
3732 * Return number of errors when page is not found or submission fails.
3733 */
3734static int write_dev_supers(struct btrfs_device *device,
3735 struct btrfs_super_block *sb, int max_mirrors)
3736{
3737 struct btrfs_fs_info *fs_info = device->fs_info;
3738 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3739 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3740 int i;
3741 int errors = 0;
3742 int ret;
3743 u64 bytenr, bytenr_orig;
3744
3745 if (max_mirrors == 0)
3746 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3747
3748 shash->tfm = fs_info->csum_shash;
3749
3750 for (i = 0; i < max_mirrors; i++) {
3751 struct page *page;
3752 struct bio *bio;
3753 struct btrfs_super_block *disk_super;
3754
3755 bytenr_orig = btrfs_sb_offset(i);
3756 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3757 if (ret == -ENOENT) {
3758 continue;
3759 } else if (ret < 0) {
3760 btrfs_err(device->fs_info,
3761 "couldn't get super block location for mirror %d",
3762 i);
3763 errors++;
3764 continue;
3765 }
3766 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3767 device->commit_total_bytes)
3768 break;
3769
3770 btrfs_set_super_bytenr(sb, bytenr_orig);
3771
3772 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3773 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3774 sb->csum);
3775
3776 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3777 GFP_NOFS);
3778 if (!page) {
3779 btrfs_err(device->fs_info,
3780 "couldn't get super block page for bytenr %llu",
3781 bytenr);
3782 errors++;
3783 continue;
3784 }
3785
3786 /* Bump the refcount for wait_dev_supers() */
3787 get_page(page);
3788
3789 disk_super = page_address(page);
3790 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3791
3792 /*
3793 * Directly use bios here instead of relying on the page cache
3794 * to do I/O, so we don't lose the ability to do integrity
3795 * checking.
3796 */
3797 bio = bio_alloc(device->bdev, 1,
3798 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3799 GFP_NOFS);
3800 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3801 bio->bi_private = device;
3802 bio->bi_end_io = btrfs_end_super_write;
3803 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3804 offset_in_page(bytenr));
3805
3806 /*
3807 * We FUA only the first super block. The others we allow to
3808 * go down lazy and there's a short window where the on-disk
3809 * copies might still contain the older version.
3810 */
3811 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3812 bio->bi_opf |= REQ_FUA;
3813 submit_bio(bio);
3814
3815 if (btrfs_advance_sb_log(device, i))
3816 errors++;
3817 }
3818 return errors < i ? 0 : -1;
3819}
3820
3821/*
3822 * Wait for write completion of superblocks done by write_dev_supers,
3823 * @max_mirrors same for write and wait phases.
3824 *
3825 * Return number of errors when page is not found or not marked up to
3826 * date.
3827 */
3828static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3829{
3830 int i;
3831 int errors = 0;
3832 bool primary_failed = false;
3833 int ret;
3834 u64 bytenr;
3835
3836 if (max_mirrors == 0)
3837 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3838
3839 for (i = 0; i < max_mirrors; i++) {
3840 struct page *page;
3841
3842 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3843 if (ret == -ENOENT) {
3844 break;
3845 } else if (ret < 0) {
3846 errors++;
3847 if (i == 0)
3848 primary_failed = true;
3849 continue;
3850 }
3851 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3852 device->commit_total_bytes)
3853 break;
3854
3855 page = find_get_page(device->bdev->bd_inode->i_mapping,
3856 bytenr >> PAGE_SHIFT);
3857 if (!page) {
3858 errors++;
3859 if (i == 0)
3860 primary_failed = true;
3861 continue;
3862 }
3863 /* Page is submitted locked and unlocked once the IO completes */
3864 wait_on_page_locked(page);
3865 if (PageError(page)) {
3866 errors++;
3867 if (i == 0)
3868 primary_failed = true;
3869 }
3870
3871 /* Drop our reference */
3872 put_page(page);
3873
3874 /* Drop the reference from the writing run */
3875 put_page(page);
3876 }
3877
3878 /* log error, force error return */
3879 if (primary_failed) {
3880 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3881 device->devid);
3882 return -1;
3883 }
3884
3885 return errors < i ? 0 : -1;
3886}
3887
3888/*
3889 * endio for the write_dev_flush, this will wake anyone waiting
3890 * for the barrier when it is done
3891 */
3892static void btrfs_end_empty_barrier(struct bio *bio)
3893{
3894 bio_uninit(bio);
3895 complete(bio->bi_private);
3896}
3897
3898/*
3899 * Submit a flush request to the device if it supports it. Error handling is
3900 * done in the waiting counterpart.
3901 */
3902static void write_dev_flush(struct btrfs_device *device)
3903{
3904 struct bio *bio = &device->flush_bio;
3905
3906 device->last_flush_error = BLK_STS_OK;
3907
3908 bio_init(bio, device->bdev, NULL, 0,
3909 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3910 bio->bi_end_io = btrfs_end_empty_barrier;
3911 init_completion(&device->flush_wait);
3912 bio->bi_private = &device->flush_wait;
3913 submit_bio(bio);
3914 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3915}
3916
3917/*
3918 * If the flush bio has been submitted by write_dev_flush, wait for it.
3919 * Return true for any error, and false otherwise.
3920 */
3921static bool wait_dev_flush(struct btrfs_device *device)
3922{
3923 struct bio *bio = &device->flush_bio;
3924
3925 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3926 return false;
3927
3928 wait_for_completion_io(&device->flush_wait);
3929
3930 if (bio->bi_status) {
3931 device->last_flush_error = bio->bi_status;
3932 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3933 return true;
3934 }
3935
3936 return false;
3937}
3938
3939/*
3940 * send an empty flush down to each device in parallel,
3941 * then wait for them
3942 */
3943static int barrier_all_devices(struct btrfs_fs_info *info)
3944{
3945 struct list_head *head;
3946 struct btrfs_device *dev;
3947 int errors_wait = 0;
3948
3949 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3950 /* send down all the barriers */
3951 head = &info->fs_devices->devices;
3952 list_for_each_entry(dev, head, dev_list) {
3953 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3954 continue;
3955 if (!dev->bdev)
3956 continue;
3957 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3958 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3959 continue;
3960
3961 write_dev_flush(dev);
3962 }
3963
3964 /* wait for all the barriers */
3965 list_for_each_entry(dev, head, dev_list) {
3966 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3967 continue;
3968 if (!dev->bdev) {
3969 errors_wait++;
3970 continue;
3971 }
3972 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3973 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3974 continue;
3975
3976 if (wait_dev_flush(dev))
3977 errors_wait++;
3978 }
3979
3980 /*
3981 * Checks last_flush_error of disks in order to determine the device
3982 * state.
3983 */
3984 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3985 return -EIO;
3986
3987 return 0;
3988}
3989
3990int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3991{
3992 int raid_type;
3993 int min_tolerated = INT_MAX;
3994
3995 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3996 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3997 min_tolerated = min_t(int, min_tolerated,
3998 btrfs_raid_array[BTRFS_RAID_SINGLE].
3999 tolerated_failures);
4000
4001 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4002 if (raid_type == BTRFS_RAID_SINGLE)
4003 continue;
4004 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4005 continue;
4006 min_tolerated = min_t(int, min_tolerated,
4007 btrfs_raid_array[raid_type].
4008 tolerated_failures);
4009 }
4010
4011 if (min_tolerated == INT_MAX) {
4012 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4013 min_tolerated = 0;
4014 }
4015
4016 return min_tolerated;
4017}
4018
4019int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4020{
4021 struct list_head *head;
4022 struct btrfs_device *dev;
4023 struct btrfs_super_block *sb;
4024 struct btrfs_dev_item *dev_item;
4025 int ret;
4026 int do_barriers;
4027 int max_errors;
4028 int total_errors = 0;
4029 u64 flags;
4030
4031 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4032
4033 /*
4034 * max_mirrors == 0 indicates we're from commit_transaction,
4035 * not from fsync where the tree roots in fs_info have not
4036 * been consistent on disk.
4037 */
4038 if (max_mirrors == 0)
4039 backup_super_roots(fs_info);
4040
4041 sb = fs_info->super_for_commit;
4042 dev_item = &sb->dev_item;
4043
4044 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4045 head = &fs_info->fs_devices->devices;
4046 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4047
4048 if (do_barriers) {
4049 ret = barrier_all_devices(fs_info);
4050 if (ret) {
4051 mutex_unlock(
4052 &fs_info->fs_devices->device_list_mutex);
4053 btrfs_handle_fs_error(fs_info, ret,
4054 "errors while submitting device barriers.");
4055 return ret;
4056 }
4057 }
4058
4059 list_for_each_entry(dev, head, dev_list) {
4060 if (!dev->bdev) {
4061 total_errors++;
4062 continue;
4063 }
4064 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4065 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4066 continue;
4067
4068 btrfs_set_stack_device_generation(dev_item, 0);
4069 btrfs_set_stack_device_type(dev_item, dev->type);
4070 btrfs_set_stack_device_id(dev_item, dev->devid);
4071 btrfs_set_stack_device_total_bytes(dev_item,
4072 dev->commit_total_bytes);
4073 btrfs_set_stack_device_bytes_used(dev_item,
4074 dev->commit_bytes_used);
4075 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4076 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4077 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4078 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4079 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4080 BTRFS_FSID_SIZE);
4081
4082 flags = btrfs_super_flags(sb);
4083 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4084
4085 ret = btrfs_validate_write_super(fs_info, sb);
4086 if (ret < 0) {
4087 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4088 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4089 "unexpected superblock corruption detected");
4090 return -EUCLEAN;
4091 }
4092
4093 ret = write_dev_supers(dev, sb, max_mirrors);
4094 if (ret)
4095 total_errors++;
4096 }
4097 if (total_errors > max_errors) {
4098 btrfs_err(fs_info, "%d errors while writing supers",
4099 total_errors);
4100 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4101
4102 /* FUA is masked off if unsupported and can't be the reason */
4103 btrfs_handle_fs_error(fs_info, -EIO,
4104 "%d errors while writing supers",
4105 total_errors);
4106 return -EIO;
4107 }
4108
4109 total_errors = 0;
4110 list_for_each_entry(dev, head, dev_list) {
4111 if (!dev->bdev)
4112 continue;
4113 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4114 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4115 continue;
4116
4117 ret = wait_dev_supers(dev, max_mirrors);
4118 if (ret)
4119 total_errors++;
4120 }
4121 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4122 if (total_errors > max_errors) {
4123 btrfs_handle_fs_error(fs_info, -EIO,
4124 "%d errors while writing supers",
4125 total_errors);
4126 return -EIO;
4127 }
4128 return 0;
4129}
4130
4131/* Drop a fs root from the radix tree and free it. */
4132void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4133 struct btrfs_root *root)
4134{
4135 bool drop_ref = false;
4136
4137 spin_lock(&fs_info->fs_roots_radix_lock);
4138 radix_tree_delete(&fs_info->fs_roots_radix,
4139 (unsigned long)root->root_key.objectid);
4140 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4141 drop_ref = true;
4142 spin_unlock(&fs_info->fs_roots_radix_lock);
4143
4144 if (BTRFS_FS_ERROR(fs_info)) {
4145 ASSERT(root->log_root == NULL);
4146 if (root->reloc_root) {
4147 btrfs_put_root(root->reloc_root);
4148 root->reloc_root = NULL;
4149 }
4150 }
4151
4152 if (drop_ref)
4153 btrfs_put_root(root);
4154}
4155
4156int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4157{
4158 struct btrfs_root *root = fs_info->tree_root;
4159 struct btrfs_trans_handle *trans;
4160
4161 mutex_lock(&fs_info->cleaner_mutex);
4162 btrfs_run_delayed_iputs(fs_info);
4163 mutex_unlock(&fs_info->cleaner_mutex);
4164 wake_up_process(fs_info->cleaner_kthread);
4165
4166 /* wait until ongoing cleanup work done */
4167 down_write(&fs_info->cleanup_work_sem);
4168 up_write(&fs_info->cleanup_work_sem);
4169
4170 trans = btrfs_join_transaction(root);
4171 if (IS_ERR(trans))
4172 return PTR_ERR(trans);
4173 return btrfs_commit_transaction(trans);
4174}
4175
4176static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4177{
4178 struct btrfs_transaction *trans;
4179 struct btrfs_transaction *tmp;
4180 bool found = false;
4181
4182 if (list_empty(&fs_info->trans_list))
4183 return;
4184
4185 /*
4186 * This function is only called at the very end of close_ctree(),
4187 * thus no other running transaction, no need to take trans_lock.
4188 */
4189 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4190 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4191 struct extent_state *cached = NULL;
4192 u64 dirty_bytes = 0;
4193 u64 cur = 0;
4194 u64 found_start;
4195 u64 found_end;
4196
4197 found = true;
4198 while (find_first_extent_bit(&trans->dirty_pages, cur,
4199 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4200 dirty_bytes += found_end + 1 - found_start;
4201 cur = found_end + 1;
4202 }
4203 btrfs_warn(fs_info,
4204 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4205 trans->transid, dirty_bytes);
4206 btrfs_cleanup_one_transaction(trans, fs_info);
4207
4208 if (trans == fs_info->running_transaction)
4209 fs_info->running_transaction = NULL;
4210 list_del_init(&trans->list);
4211
4212 btrfs_put_transaction(trans);
4213 trace_btrfs_transaction_commit(fs_info);
4214 }
4215 ASSERT(!found);
4216}
4217
4218void __cold close_ctree(struct btrfs_fs_info *fs_info)
4219{
4220 int ret;
4221
4222 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4223
4224 /*
4225 * If we had UNFINISHED_DROPS we could still be processing them, so
4226 * clear that bit and wake up relocation so it can stop.
4227 * We must do this before stopping the block group reclaim task, because
4228 * at btrfs_relocate_block_group() we wait for this bit, and after the
4229 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4230 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4231 * return 1.
4232 */
4233 btrfs_wake_unfinished_drop(fs_info);
4234
4235 /*
4236 * We may have the reclaim task running and relocating a data block group,
4237 * in which case it may create delayed iputs. So stop it before we park
4238 * the cleaner kthread otherwise we can get new delayed iputs after
4239 * parking the cleaner, and that can make the async reclaim task to hang
4240 * if it's waiting for delayed iputs to complete, since the cleaner is
4241 * parked and can not run delayed iputs - this will make us hang when
4242 * trying to stop the async reclaim task.
4243 */
4244 cancel_work_sync(&fs_info->reclaim_bgs_work);
4245 /*
4246 * We don't want the cleaner to start new transactions, add more delayed
4247 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4248 * because that frees the task_struct, and the transaction kthread might
4249 * still try to wake up the cleaner.
4250 */
4251 kthread_park(fs_info->cleaner_kthread);
4252
4253 /* wait for the qgroup rescan worker to stop */
4254 btrfs_qgroup_wait_for_completion(fs_info, false);
4255
4256 /* wait for the uuid_scan task to finish */
4257 down(&fs_info->uuid_tree_rescan_sem);
4258 /* avoid complains from lockdep et al., set sem back to initial state */
4259 up(&fs_info->uuid_tree_rescan_sem);
4260
4261 /* pause restriper - we want to resume on mount */
4262 btrfs_pause_balance(fs_info);
4263
4264 btrfs_dev_replace_suspend_for_unmount(fs_info);
4265
4266 btrfs_scrub_cancel(fs_info);
4267
4268 /* wait for any defraggers to finish */
4269 wait_event(fs_info->transaction_wait,
4270 (atomic_read(&fs_info->defrag_running) == 0));
4271
4272 /* clear out the rbtree of defraggable inodes */
4273 btrfs_cleanup_defrag_inodes(fs_info);
4274
4275 /*
4276 * After we parked the cleaner kthread, ordered extents may have
4277 * completed and created new delayed iputs. If one of the async reclaim
4278 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4279 * can hang forever trying to stop it, because if a delayed iput is
4280 * added after it ran btrfs_run_delayed_iputs() and before it called
4281 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4282 * no one else to run iputs.
4283 *
4284 * So wait for all ongoing ordered extents to complete and then run
4285 * delayed iputs. This works because once we reach this point no one
4286 * can either create new ordered extents nor create delayed iputs
4287 * through some other means.
4288 *
4289 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4290 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4291 * but the delayed iput for the respective inode is made only when doing
4292 * the final btrfs_put_ordered_extent() (which must happen at
4293 * btrfs_finish_ordered_io() when we are unmounting).
4294 */
4295 btrfs_flush_workqueue(fs_info->endio_write_workers);
4296 /* Ordered extents for free space inodes. */
4297 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4298 btrfs_run_delayed_iputs(fs_info);
4299
4300 cancel_work_sync(&fs_info->async_reclaim_work);
4301 cancel_work_sync(&fs_info->async_data_reclaim_work);
4302 cancel_work_sync(&fs_info->preempt_reclaim_work);
4303
4304 /* Cancel or finish ongoing discard work */
4305 btrfs_discard_cleanup(fs_info);
4306
4307 if (!sb_rdonly(fs_info->sb)) {
4308 /*
4309 * The cleaner kthread is stopped, so do one final pass over
4310 * unused block groups.
4311 */
4312 btrfs_delete_unused_bgs(fs_info);
4313
4314 /*
4315 * There might be existing delayed inode workers still running
4316 * and holding an empty delayed inode item. We must wait for
4317 * them to complete first because they can create a transaction.
4318 * This happens when someone calls btrfs_balance_delayed_items()
4319 * and then a transaction commit runs the same delayed nodes
4320 * before any delayed worker has done something with the nodes.
4321 * We must wait for any worker here and not at transaction
4322 * commit time since that could cause a deadlock.
4323 * This is a very rare case.
4324 */
4325 btrfs_flush_workqueue(fs_info->delayed_workers);
4326
4327 ret = btrfs_commit_super(fs_info);
4328 if (ret)
4329 btrfs_err(fs_info, "commit super ret %d", ret);
4330 }
4331
4332 if (BTRFS_FS_ERROR(fs_info))
4333 btrfs_error_commit_super(fs_info);
4334
4335 kthread_stop(fs_info->transaction_kthread);
4336 kthread_stop(fs_info->cleaner_kthread);
4337
4338 ASSERT(list_empty(&fs_info->delayed_iputs));
4339 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4340
4341 if (btrfs_check_quota_leak(fs_info)) {
4342 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4343 btrfs_err(fs_info, "qgroup reserved space leaked");
4344 }
4345
4346 btrfs_free_qgroup_config(fs_info);
4347 ASSERT(list_empty(&fs_info->delalloc_roots));
4348
4349 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4350 btrfs_info(fs_info, "at unmount delalloc count %lld",
4351 percpu_counter_sum(&fs_info->delalloc_bytes));
4352 }
4353
4354 if (percpu_counter_sum(&fs_info->ordered_bytes))
4355 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4356 percpu_counter_sum(&fs_info->ordered_bytes));
4357
4358 btrfs_sysfs_remove_mounted(fs_info);
4359 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4360
4361 btrfs_put_block_group_cache(fs_info);
4362
4363 /*
4364 * we must make sure there is not any read request to
4365 * submit after we stopping all workers.
4366 */
4367 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4368 btrfs_stop_all_workers(fs_info);
4369
4370 /* We shouldn't have any transaction open at this point */
4371 warn_about_uncommitted_trans(fs_info);
4372
4373 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4374 free_root_pointers(fs_info, true);
4375 btrfs_free_fs_roots(fs_info);
4376
4377 /*
4378 * We must free the block groups after dropping the fs_roots as we could
4379 * have had an IO error and have left over tree log blocks that aren't
4380 * cleaned up until the fs roots are freed. This makes the block group
4381 * accounting appear to be wrong because there's pending reserved bytes,
4382 * so make sure we do the block group cleanup afterwards.
4383 */
4384 btrfs_free_block_groups(fs_info);
4385
4386 iput(fs_info->btree_inode);
4387
4388 btrfs_mapping_tree_free(fs_info);
4389 btrfs_close_devices(fs_info->fs_devices);
4390}
4391
4392void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4393 struct extent_buffer *buf)
4394{
4395 struct btrfs_fs_info *fs_info = buf->fs_info;
4396 u64 transid = btrfs_header_generation(buf);
4397
4398#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4399 /*
4400 * This is a fast path so only do this check if we have sanity tests
4401 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4402 * outside of the sanity tests.
4403 */
4404 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4405 return;
4406#endif
4407 /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4408 ASSERT(trans->transid == fs_info->generation);
4409 btrfs_assert_tree_write_locked(buf);
4410 if (unlikely(transid != fs_info->generation)) {
4411 btrfs_abort_transaction(trans, -EUCLEAN);
4412 btrfs_crit(fs_info,
4413"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4414 buf->start, transid, fs_info->generation);
4415 }
4416 set_extent_buffer_dirty(buf);
4417}
4418
4419static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4420 int flush_delayed)
4421{
4422 /*
4423 * looks as though older kernels can get into trouble with
4424 * this code, they end up stuck in balance_dirty_pages forever
4425 */
4426 int ret;
4427
4428 if (current->flags & PF_MEMALLOC)
4429 return;
4430
4431 if (flush_delayed)
4432 btrfs_balance_delayed_items(fs_info);
4433
4434 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4435 BTRFS_DIRTY_METADATA_THRESH,
4436 fs_info->dirty_metadata_batch);
4437 if (ret > 0) {
4438 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4439 }
4440}
4441
4442void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4443{
4444 __btrfs_btree_balance_dirty(fs_info, 1);
4445}
4446
4447void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4448{
4449 __btrfs_btree_balance_dirty(fs_info, 0);
4450}
4451
4452static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4453{
4454 /* cleanup FS via transaction */
4455 btrfs_cleanup_transaction(fs_info);
4456
4457 mutex_lock(&fs_info->cleaner_mutex);
4458 btrfs_run_delayed_iputs(fs_info);
4459 mutex_unlock(&fs_info->cleaner_mutex);
4460
4461 down_write(&fs_info->cleanup_work_sem);
4462 up_write(&fs_info->cleanup_work_sem);
4463}
4464
4465static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4466{
4467 struct btrfs_root *gang[8];
4468 u64 root_objectid = 0;
4469 int ret;
4470
4471 spin_lock(&fs_info->fs_roots_radix_lock);
4472 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4473 (void **)gang, root_objectid,
4474 ARRAY_SIZE(gang))) != 0) {
4475 int i;
4476
4477 for (i = 0; i < ret; i++)
4478 gang[i] = btrfs_grab_root(gang[i]);
4479 spin_unlock(&fs_info->fs_roots_radix_lock);
4480
4481 for (i = 0; i < ret; i++) {
4482 if (!gang[i])
4483 continue;
4484 root_objectid = gang[i]->root_key.objectid;
4485 btrfs_free_log(NULL, gang[i]);
4486 btrfs_put_root(gang[i]);
4487 }
4488 root_objectid++;
4489 spin_lock(&fs_info->fs_roots_radix_lock);
4490 }
4491 spin_unlock(&fs_info->fs_roots_radix_lock);
4492 btrfs_free_log_root_tree(NULL, fs_info);
4493}
4494
4495static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4496{
4497 struct btrfs_ordered_extent *ordered;
4498
4499 spin_lock(&root->ordered_extent_lock);
4500 /*
4501 * This will just short circuit the ordered completion stuff which will
4502 * make sure the ordered extent gets properly cleaned up.
4503 */
4504 list_for_each_entry(ordered, &root->ordered_extents,
4505 root_extent_list)
4506 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4507 spin_unlock(&root->ordered_extent_lock);
4508}
4509
4510static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4511{
4512 struct btrfs_root *root;
4513 LIST_HEAD(splice);
4514
4515 spin_lock(&fs_info->ordered_root_lock);
4516 list_splice_init(&fs_info->ordered_roots, &splice);
4517 while (!list_empty(&splice)) {
4518 root = list_first_entry(&splice, struct btrfs_root,
4519 ordered_root);
4520 list_move_tail(&root->ordered_root,
4521 &fs_info->ordered_roots);
4522
4523 spin_unlock(&fs_info->ordered_root_lock);
4524 btrfs_destroy_ordered_extents(root);
4525
4526 cond_resched();
4527 spin_lock(&fs_info->ordered_root_lock);
4528 }
4529 spin_unlock(&fs_info->ordered_root_lock);
4530
4531 /*
4532 * We need this here because if we've been flipped read-only we won't
4533 * get sync() from the umount, so we need to make sure any ordered
4534 * extents that haven't had their dirty pages IO start writeout yet
4535 * actually get run and error out properly.
4536 */
4537 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4538}
4539
4540static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4541 struct btrfs_fs_info *fs_info)
4542{
4543 struct rb_node *node;
4544 struct btrfs_delayed_ref_root *delayed_refs;
4545 struct btrfs_delayed_ref_node *ref;
4546
4547 delayed_refs = &trans->delayed_refs;
4548
4549 spin_lock(&delayed_refs->lock);
4550 if (atomic_read(&delayed_refs->num_entries) == 0) {
4551 spin_unlock(&delayed_refs->lock);
4552 btrfs_debug(fs_info, "delayed_refs has NO entry");
4553 return;
4554 }
4555
4556 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4557 struct btrfs_delayed_ref_head *head;
4558 struct rb_node *n;
4559 bool pin_bytes = false;
4560
4561 head = rb_entry(node, struct btrfs_delayed_ref_head,
4562 href_node);
4563 if (btrfs_delayed_ref_lock(delayed_refs, head))
4564 continue;
4565
4566 spin_lock(&head->lock);
4567 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4568 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4569 ref_node);
4570 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4571 RB_CLEAR_NODE(&ref->ref_node);
4572 if (!list_empty(&ref->add_list))
4573 list_del(&ref->add_list);
4574 atomic_dec(&delayed_refs->num_entries);
4575 btrfs_put_delayed_ref(ref);
4576 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4577 }
4578 if (head->must_insert_reserved)
4579 pin_bytes = true;
4580 btrfs_free_delayed_extent_op(head->extent_op);
4581 btrfs_delete_ref_head(delayed_refs, head);
4582 spin_unlock(&head->lock);
4583 spin_unlock(&delayed_refs->lock);
4584 mutex_unlock(&head->mutex);
4585
4586 if (pin_bytes) {
4587 struct btrfs_block_group *cache;
4588
4589 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4590 BUG_ON(!cache);
4591
4592 spin_lock(&cache->space_info->lock);
4593 spin_lock(&cache->lock);
4594 cache->pinned += head->num_bytes;
4595 btrfs_space_info_update_bytes_pinned(fs_info,
4596 cache->space_info, head->num_bytes);
4597 cache->reserved -= head->num_bytes;
4598 cache->space_info->bytes_reserved -= head->num_bytes;
4599 spin_unlock(&cache->lock);
4600 spin_unlock(&cache->space_info->lock);
4601
4602 btrfs_put_block_group(cache);
4603
4604 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4605 head->bytenr + head->num_bytes - 1);
4606 }
4607 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4608 btrfs_put_delayed_ref_head(head);
4609 cond_resched();
4610 spin_lock(&delayed_refs->lock);
4611 }
4612 btrfs_qgroup_destroy_extent_records(trans);
4613
4614 spin_unlock(&delayed_refs->lock);
4615}
4616
4617static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4618{
4619 struct btrfs_inode *btrfs_inode;
4620 LIST_HEAD(splice);
4621
4622 spin_lock(&root->delalloc_lock);
4623 list_splice_init(&root->delalloc_inodes, &splice);
4624
4625 while (!list_empty(&splice)) {
4626 struct inode *inode = NULL;
4627 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4628 delalloc_inodes);
4629 __btrfs_del_delalloc_inode(root, btrfs_inode);
4630 spin_unlock(&root->delalloc_lock);
4631
4632 /*
4633 * Make sure we get a live inode and that it'll not disappear
4634 * meanwhile.
4635 */
4636 inode = igrab(&btrfs_inode->vfs_inode);
4637 if (inode) {
4638 unsigned int nofs_flag;
4639
4640 nofs_flag = memalloc_nofs_save();
4641 invalidate_inode_pages2(inode->i_mapping);
4642 memalloc_nofs_restore(nofs_flag);
4643 iput(inode);
4644 }
4645 spin_lock(&root->delalloc_lock);
4646 }
4647 spin_unlock(&root->delalloc_lock);
4648}
4649
4650static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4651{
4652 struct btrfs_root *root;
4653 LIST_HEAD(splice);
4654
4655 spin_lock(&fs_info->delalloc_root_lock);
4656 list_splice_init(&fs_info->delalloc_roots, &splice);
4657 while (!list_empty(&splice)) {
4658 root = list_first_entry(&splice, struct btrfs_root,
4659 delalloc_root);
4660 root = btrfs_grab_root(root);
4661 BUG_ON(!root);
4662 spin_unlock(&fs_info->delalloc_root_lock);
4663
4664 btrfs_destroy_delalloc_inodes(root);
4665 btrfs_put_root(root);
4666
4667 spin_lock(&fs_info->delalloc_root_lock);
4668 }
4669 spin_unlock(&fs_info->delalloc_root_lock);
4670}
4671
4672static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4673 struct extent_io_tree *dirty_pages,
4674 int mark)
4675{
4676 struct extent_buffer *eb;
4677 u64 start = 0;
4678 u64 end;
4679
4680 while (find_first_extent_bit(dirty_pages, start, &start, &end,
4681 mark, NULL)) {
4682 clear_extent_bits(dirty_pages, start, end, mark);
4683 while (start <= end) {
4684 eb = find_extent_buffer(fs_info, start);
4685 start += fs_info->nodesize;
4686 if (!eb)
4687 continue;
4688
4689 btrfs_tree_lock(eb);
4690 wait_on_extent_buffer_writeback(eb);
4691 btrfs_clear_buffer_dirty(NULL, eb);
4692 btrfs_tree_unlock(eb);
4693
4694 free_extent_buffer_stale(eb);
4695 }
4696 }
4697}
4698
4699static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4700 struct extent_io_tree *unpin)
4701{
4702 u64 start;
4703 u64 end;
4704
4705 while (1) {
4706 struct extent_state *cached_state = NULL;
4707
4708 /*
4709 * The btrfs_finish_extent_commit() may get the same range as
4710 * ours between find_first_extent_bit and clear_extent_dirty.
4711 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4712 * the same extent range.
4713 */
4714 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4715 if (!find_first_extent_bit(unpin, 0, &start, &end,
4716 EXTENT_DIRTY, &cached_state)) {
4717 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4718 break;
4719 }
4720
4721 clear_extent_dirty(unpin, start, end, &cached_state);
4722 free_extent_state(cached_state);
4723 btrfs_error_unpin_extent_range(fs_info, start, end);
4724 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4725 cond_resched();
4726 }
4727}
4728
4729static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4730{
4731 struct inode *inode;
4732
4733 inode = cache->io_ctl.inode;
4734 if (inode) {
4735 unsigned int nofs_flag;
4736
4737 nofs_flag = memalloc_nofs_save();
4738 invalidate_inode_pages2(inode->i_mapping);
4739 memalloc_nofs_restore(nofs_flag);
4740
4741 BTRFS_I(inode)->generation = 0;
4742 cache->io_ctl.inode = NULL;
4743 iput(inode);
4744 }
4745 ASSERT(cache->io_ctl.pages == NULL);
4746 btrfs_put_block_group(cache);
4747}
4748
4749void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4750 struct btrfs_fs_info *fs_info)
4751{
4752 struct btrfs_block_group *cache;
4753
4754 spin_lock(&cur_trans->dirty_bgs_lock);
4755 while (!list_empty(&cur_trans->dirty_bgs)) {
4756 cache = list_first_entry(&cur_trans->dirty_bgs,
4757 struct btrfs_block_group,
4758 dirty_list);
4759
4760 if (!list_empty(&cache->io_list)) {
4761 spin_unlock(&cur_trans->dirty_bgs_lock);
4762 list_del_init(&cache->io_list);
4763 btrfs_cleanup_bg_io(cache);
4764 spin_lock(&cur_trans->dirty_bgs_lock);
4765 }
4766
4767 list_del_init(&cache->dirty_list);
4768 spin_lock(&cache->lock);
4769 cache->disk_cache_state = BTRFS_DC_ERROR;
4770 spin_unlock(&cache->lock);
4771
4772 spin_unlock(&cur_trans->dirty_bgs_lock);
4773 btrfs_put_block_group(cache);
4774 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4775 spin_lock(&cur_trans->dirty_bgs_lock);
4776 }
4777 spin_unlock(&cur_trans->dirty_bgs_lock);
4778
4779 /*
4780 * Refer to the definition of io_bgs member for details why it's safe
4781 * to use it without any locking
4782 */
4783 while (!list_empty(&cur_trans->io_bgs)) {
4784 cache = list_first_entry(&cur_trans->io_bgs,
4785 struct btrfs_block_group,
4786 io_list);
4787
4788 list_del_init(&cache->io_list);
4789 spin_lock(&cache->lock);
4790 cache->disk_cache_state = BTRFS_DC_ERROR;
4791 spin_unlock(&cache->lock);
4792 btrfs_cleanup_bg_io(cache);
4793 }
4794}
4795
4796static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4797{
4798 struct btrfs_root *gang[8];
4799 int i;
4800 int ret;
4801
4802 spin_lock(&fs_info->fs_roots_radix_lock);
4803 while (1) {
4804 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4805 (void **)gang, 0,
4806 ARRAY_SIZE(gang),
4807 BTRFS_ROOT_TRANS_TAG);
4808 if (ret == 0)
4809 break;
4810 for (i = 0; i < ret; i++) {
4811 struct btrfs_root *root = gang[i];
4812
4813 btrfs_qgroup_free_meta_all_pertrans(root);
4814 radix_tree_tag_clear(&fs_info->fs_roots_radix,
4815 (unsigned long)root->root_key.objectid,
4816 BTRFS_ROOT_TRANS_TAG);
4817 }
4818 }
4819 spin_unlock(&fs_info->fs_roots_radix_lock);
4820}
4821
4822void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4823 struct btrfs_fs_info *fs_info)
4824{
4825 struct btrfs_device *dev, *tmp;
4826
4827 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4828 ASSERT(list_empty(&cur_trans->dirty_bgs));
4829 ASSERT(list_empty(&cur_trans->io_bgs));
4830
4831 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4832 post_commit_list) {
4833 list_del_init(&dev->post_commit_list);
4834 }
4835
4836 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4837
4838 cur_trans->state = TRANS_STATE_COMMIT_START;
4839 wake_up(&fs_info->transaction_blocked_wait);
4840
4841 cur_trans->state = TRANS_STATE_UNBLOCKED;
4842 wake_up(&fs_info->transaction_wait);
4843
4844 btrfs_destroy_delayed_inodes(fs_info);
4845
4846 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4847 EXTENT_DIRTY);
4848 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4849
4850 btrfs_free_all_qgroup_pertrans(fs_info);
4851
4852 cur_trans->state =TRANS_STATE_COMPLETED;
4853 wake_up(&cur_trans->commit_wait);
4854}
4855
4856static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4857{
4858 struct btrfs_transaction *t;
4859
4860 mutex_lock(&fs_info->transaction_kthread_mutex);
4861
4862 spin_lock(&fs_info->trans_lock);
4863 while (!list_empty(&fs_info->trans_list)) {
4864 t = list_first_entry(&fs_info->trans_list,
4865 struct btrfs_transaction, list);
4866 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4867 refcount_inc(&t->use_count);
4868 spin_unlock(&fs_info->trans_lock);
4869 btrfs_wait_for_commit(fs_info, t->transid);
4870 btrfs_put_transaction(t);
4871 spin_lock(&fs_info->trans_lock);
4872 continue;
4873 }
4874 if (t == fs_info->running_transaction) {
4875 t->state = TRANS_STATE_COMMIT_DOING;
4876 spin_unlock(&fs_info->trans_lock);
4877 /*
4878 * We wait for 0 num_writers since we don't hold a trans
4879 * handle open currently for this transaction.
4880 */
4881 wait_event(t->writer_wait,
4882 atomic_read(&t->num_writers) == 0);
4883 } else {
4884 spin_unlock(&fs_info->trans_lock);
4885 }
4886 btrfs_cleanup_one_transaction(t, fs_info);
4887
4888 spin_lock(&fs_info->trans_lock);
4889 if (t == fs_info->running_transaction)
4890 fs_info->running_transaction = NULL;
4891 list_del_init(&t->list);
4892 spin_unlock(&fs_info->trans_lock);
4893
4894 btrfs_put_transaction(t);
4895 trace_btrfs_transaction_commit(fs_info);
4896 spin_lock(&fs_info->trans_lock);
4897 }
4898 spin_unlock(&fs_info->trans_lock);
4899 btrfs_destroy_all_ordered_extents(fs_info);
4900 btrfs_destroy_delayed_inodes(fs_info);
4901 btrfs_assert_delayed_root_empty(fs_info);
4902 btrfs_destroy_all_delalloc_inodes(fs_info);
4903 btrfs_drop_all_logs(fs_info);
4904 mutex_unlock(&fs_info->transaction_kthread_mutex);
4905
4906 return 0;
4907}
4908
4909int btrfs_init_root_free_objectid(struct btrfs_root *root)
4910{
4911 struct btrfs_path *path;
4912 int ret;
4913 struct extent_buffer *l;
4914 struct btrfs_key search_key;
4915 struct btrfs_key found_key;
4916 int slot;
4917
4918 path = btrfs_alloc_path();
4919 if (!path)
4920 return -ENOMEM;
4921
4922 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4923 search_key.type = -1;
4924 search_key.offset = (u64)-1;
4925 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4926 if (ret < 0)
4927 goto error;
4928 BUG_ON(ret == 0); /* Corruption */
4929 if (path->slots[0] > 0) {
4930 slot = path->slots[0] - 1;
4931 l = path->nodes[0];
4932 btrfs_item_key_to_cpu(l, &found_key, slot);
4933 root->free_objectid = max_t(u64, found_key.objectid + 1,
4934 BTRFS_FIRST_FREE_OBJECTID);
4935 } else {
4936 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4937 }
4938 ret = 0;
4939error:
4940 btrfs_free_path(path);
4941 return ret;
4942}
4943
4944int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4945{
4946 int ret;
4947 mutex_lock(&root->objectid_mutex);
4948
4949 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4950 btrfs_warn(root->fs_info,
4951 "the objectid of root %llu reaches its highest value",
4952 root->root_key.objectid);
4953 ret = -ENOSPC;
4954 goto out;
4955 }
4956
4957 *objectid = root->free_objectid++;
4958 ret = 0;
4959out:
4960 mutex_unlock(&root->objectid_mutex);
4961 return ret;
4962}
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/slab.h>
29#include <linux/migrate.h>
30#include <linux/ratelimit.h>
31#include <linux/uuid.h>
32#include <linux/semaphore.h>
33#include <asm/unaligned.h>
34#include "ctree.h"
35#include "disk-io.h"
36#include "hash.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "locking.h"
42#include "tree-log.h"
43#include "free-space-cache.h"
44#include "free-space-tree.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48#include "dev-replace.h"
49#include "raid56.h"
50#include "sysfs.h"
51#include "qgroup.h"
52#include "compression.h"
53
54#ifdef CONFIG_X86
55#include <asm/cpufeature.h>
56#endif
57
58#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
63
64static const struct extent_io_ops btree_extent_io_ops;
65static void end_workqueue_fn(struct btrfs_work *work);
66static void free_fs_root(struct btrfs_root *root);
67static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
68 int read_only);
69static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
70static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
71 struct btrfs_root *root);
72static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
73static int btrfs_destroy_marked_extents(struct btrfs_root *root,
74 struct extent_io_tree *dirty_pages,
75 int mark);
76static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
77 struct extent_io_tree *pinned_extents);
78static int btrfs_cleanup_transaction(struct btrfs_root *root);
79static void btrfs_error_commit_super(struct btrfs_root *root);
80
81/*
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
85 */
86struct btrfs_end_io_wq {
87 struct bio *bio;
88 bio_end_io_t *end_io;
89 void *private;
90 struct btrfs_fs_info *info;
91 int error;
92 enum btrfs_wq_endio_type metadata;
93 struct list_head list;
94 struct btrfs_work work;
95};
96
97static struct kmem_cache *btrfs_end_io_wq_cache;
98
99int __init btrfs_end_io_wq_init(void)
100{
101 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq),
103 0,
104 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
105 NULL);
106 if (!btrfs_end_io_wq_cache)
107 return -ENOMEM;
108 return 0;
109}
110
111void btrfs_end_io_wq_exit(void)
112{
113 kmem_cache_destroy(btrfs_end_io_wq_cache);
114}
115
116/*
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
120 */
121struct async_submit_bio {
122 struct inode *inode;
123 struct bio *bio;
124 struct list_head list;
125 extent_submit_bio_hook_t *submit_bio_start;
126 extent_submit_bio_hook_t *submit_bio_done;
127 int rw;
128 int mirror_num;
129 unsigned long bio_flags;
130 /*
131 * bio_offset is optional, can be used if the pages in the bio
132 * can't tell us where in the file the bio should go
133 */
134 u64 bio_offset;
135 struct btrfs_work work;
136 int error;
137};
138
139/*
140 * Lockdep class keys for extent_buffer->lock's in this root. For a given
141 * eb, the lockdep key is determined by the btrfs_root it belongs to and
142 * the level the eb occupies in the tree.
143 *
144 * Different roots are used for different purposes and may nest inside each
145 * other and they require separate keysets. As lockdep keys should be
146 * static, assign keysets according to the purpose of the root as indicated
147 * by btrfs_root->objectid. This ensures that all special purpose roots
148 * have separate keysets.
149 *
150 * Lock-nesting across peer nodes is always done with the immediate parent
151 * node locked thus preventing deadlock. As lockdep doesn't know this, use
152 * subclass to avoid triggering lockdep warning in such cases.
153 *
154 * The key is set by the readpage_end_io_hook after the buffer has passed
155 * csum validation but before the pages are unlocked. It is also set by
156 * btrfs_init_new_buffer on freshly allocated blocks.
157 *
158 * We also add a check to make sure the highest level of the tree is the
159 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
160 * needs update as well.
161 */
162#ifdef CONFIG_DEBUG_LOCK_ALLOC
163# if BTRFS_MAX_LEVEL != 8
164# error
165# endif
166
167static struct btrfs_lockdep_keyset {
168 u64 id; /* root objectid */
169 const char *name_stem; /* lock name stem */
170 char names[BTRFS_MAX_LEVEL + 1][20];
171 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
172} btrfs_lockdep_keysets[] = {
173 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
174 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
175 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
176 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
177 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
178 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
179 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
180 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
181 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
182 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
183 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
184 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
185 { .id = 0, .name_stem = "tree" },
186};
187
188void __init btrfs_init_lockdep(void)
189{
190 int i, j;
191
192 /* initialize lockdep class names */
193 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
194 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
195
196 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
197 snprintf(ks->names[j], sizeof(ks->names[j]),
198 "btrfs-%s-%02d", ks->name_stem, j);
199 }
200}
201
202void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
203 int level)
204{
205 struct btrfs_lockdep_keyset *ks;
206
207 BUG_ON(level >= ARRAY_SIZE(ks->keys));
208
209 /* find the matching keyset, id 0 is the default entry */
210 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
211 if (ks->id == objectid)
212 break;
213
214 lockdep_set_class_and_name(&eb->lock,
215 &ks->keys[level], ks->names[level]);
216}
217
218#endif
219
220/*
221 * extents on the btree inode are pretty simple, there's one extent
222 * that covers the entire device
223 */
224static struct extent_map *btree_get_extent(struct inode *inode,
225 struct page *page, size_t pg_offset, u64 start, u64 len,
226 int create)
227{
228 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
229 struct extent_map *em;
230 int ret;
231
232 read_lock(&em_tree->lock);
233 em = lookup_extent_mapping(em_tree, start, len);
234 if (em) {
235 em->bdev =
236 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
237 read_unlock(&em_tree->lock);
238 goto out;
239 }
240 read_unlock(&em_tree->lock);
241
242 em = alloc_extent_map();
243 if (!em) {
244 em = ERR_PTR(-ENOMEM);
245 goto out;
246 }
247 em->start = 0;
248 em->len = (u64)-1;
249 em->block_len = (u64)-1;
250 em->block_start = 0;
251 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
252
253 write_lock(&em_tree->lock);
254 ret = add_extent_mapping(em_tree, em, 0);
255 if (ret == -EEXIST) {
256 free_extent_map(em);
257 em = lookup_extent_mapping(em_tree, start, len);
258 if (!em)
259 em = ERR_PTR(-EIO);
260 } else if (ret) {
261 free_extent_map(em);
262 em = ERR_PTR(ret);
263 }
264 write_unlock(&em_tree->lock);
265
266out:
267 return em;
268}
269
270u32 btrfs_csum_data(char *data, u32 seed, size_t len)
271{
272 return btrfs_crc32c(seed, data, len);
273}
274
275void btrfs_csum_final(u32 crc, char *result)
276{
277 put_unaligned_le32(~crc, result);
278}
279
280/*
281 * compute the csum for a btree block, and either verify it or write it
282 * into the csum field of the block.
283 */
284static int csum_tree_block(struct btrfs_fs_info *fs_info,
285 struct extent_buffer *buf,
286 int verify)
287{
288 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
289 char *result = NULL;
290 unsigned long len;
291 unsigned long cur_len;
292 unsigned long offset = BTRFS_CSUM_SIZE;
293 char *kaddr;
294 unsigned long map_start;
295 unsigned long map_len;
296 int err;
297 u32 crc = ~(u32)0;
298 unsigned long inline_result;
299
300 len = buf->len - offset;
301 while (len > 0) {
302 err = map_private_extent_buffer(buf, offset, 32,
303 &kaddr, &map_start, &map_len);
304 if (err)
305 return err;
306 cur_len = min(len, map_len - (offset - map_start));
307 crc = btrfs_csum_data(kaddr + offset - map_start,
308 crc, cur_len);
309 len -= cur_len;
310 offset += cur_len;
311 }
312 if (csum_size > sizeof(inline_result)) {
313 result = kzalloc(csum_size, GFP_NOFS);
314 if (!result)
315 return -ENOMEM;
316 } else {
317 result = (char *)&inline_result;
318 }
319
320 btrfs_csum_final(crc, result);
321
322 if (verify) {
323 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
324 u32 val;
325 u32 found = 0;
326 memcpy(&found, result, csum_size);
327
328 read_extent_buffer(buf, &val, 0, csum_size);
329 btrfs_warn_rl(fs_info,
330 "%s checksum verify failed on %llu wanted %X found %X "
331 "level %d",
332 fs_info->sb->s_id, buf->start,
333 val, found, btrfs_header_level(buf));
334 if (result != (char *)&inline_result)
335 kfree(result);
336 return -EUCLEAN;
337 }
338 } else {
339 write_extent_buffer(buf, result, 0, csum_size);
340 }
341 if (result != (char *)&inline_result)
342 kfree(result);
343 return 0;
344}
345
346/*
347 * we can't consider a given block up to date unless the transid of the
348 * block matches the transid in the parent node's pointer. This is how we
349 * detect blocks that either didn't get written at all or got written
350 * in the wrong place.
351 */
352static int verify_parent_transid(struct extent_io_tree *io_tree,
353 struct extent_buffer *eb, u64 parent_transid,
354 int atomic)
355{
356 struct extent_state *cached_state = NULL;
357 int ret;
358 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
359
360 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
361 return 0;
362
363 if (atomic)
364 return -EAGAIN;
365
366 if (need_lock) {
367 btrfs_tree_read_lock(eb);
368 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
369 }
370
371 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
372 &cached_state);
373 if (extent_buffer_uptodate(eb) &&
374 btrfs_header_generation(eb) == parent_transid) {
375 ret = 0;
376 goto out;
377 }
378 btrfs_err_rl(eb->fs_info,
379 "parent transid verify failed on %llu wanted %llu found %llu",
380 eb->start,
381 parent_transid, btrfs_header_generation(eb));
382 ret = 1;
383
384 /*
385 * Things reading via commit roots that don't have normal protection,
386 * like send, can have a really old block in cache that may point at a
387 * block that has been free'd and re-allocated. So don't clear uptodate
388 * if we find an eb that is under IO (dirty/writeback) because we could
389 * end up reading in the stale data and then writing it back out and
390 * making everybody very sad.
391 */
392 if (!extent_buffer_under_io(eb))
393 clear_extent_buffer_uptodate(eb);
394out:
395 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
396 &cached_state, GFP_NOFS);
397 if (need_lock)
398 btrfs_tree_read_unlock_blocking(eb);
399 return ret;
400}
401
402/*
403 * Return 0 if the superblock checksum type matches the checksum value of that
404 * algorithm. Pass the raw disk superblock data.
405 */
406static int btrfs_check_super_csum(char *raw_disk_sb)
407{
408 struct btrfs_super_block *disk_sb =
409 (struct btrfs_super_block *)raw_disk_sb;
410 u16 csum_type = btrfs_super_csum_type(disk_sb);
411 int ret = 0;
412
413 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
414 u32 crc = ~(u32)0;
415 const int csum_size = sizeof(crc);
416 char result[csum_size];
417
418 /*
419 * The super_block structure does not span the whole
420 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
421 * is filled with zeros and is included in the checkum.
422 */
423 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
424 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
425 btrfs_csum_final(crc, result);
426
427 if (memcmp(raw_disk_sb, result, csum_size))
428 ret = 1;
429 }
430
431 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
432 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
433 csum_type);
434 ret = 1;
435 }
436
437 return ret;
438}
439
440/*
441 * helper to read a given tree block, doing retries as required when
442 * the checksums don't match and we have alternate mirrors to try.
443 */
444static int btree_read_extent_buffer_pages(struct btrfs_root *root,
445 struct extent_buffer *eb,
446 u64 start, u64 parent_transid)
447{
448 struct extent_io_tree *io_tree;
449 int failed = 0;
450 int ret;
451 int num_copies = 0;
452 int mirror_num = 0;
453 int failed_mirror = 0;
454
455 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
456 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
457 while (1) {
458 ret = read_extent_buffer_pages(io_tree, eb, start,
459 WAIT_COMPLETE,
460 btree_get_extent, mirror_num);
461 if (!ret) {
462 if (!verify_parent_transid(io_tree, eb,
463 parent_transid, 0))
464 break;
465 else
466 ret = -EIO;
467 }
468
469 /*
470 * This buffer's crc is fine, but its contents are corrupted, so
471 * there is no reason to read the other copies, they won't be
472 * any less wrong.
473 */
474 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
475 break;
476
477 num_copies = btrfs_num_copies(root->fs_info,
478 eb->start, eb->len);
479 if (num_copies == 1)
480 break;
481
482 if (!failed_mirror) {
483 failed = 1;
484 failed_mirror = eb->read_mirror;
485 }
486
487 mirror_num++;
488 if (mirror_num == failed_mirror)
489 mirror_num++;
490
491 if (mirror_num > num_copies)
492 break;
493 }
494
495 if (failed && !ret && failed_mirror)
496 repair_eb_io_failure(root, eb, failed_mirror);
497
498 return ret;
499}
500
501/*
502 * checksum a dirty tree block before IO. This has extra checks to make sure
503 * we only fill in the checksum field in the first page of a multi-page block
504 */
505
506static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
507{
508 u64 start = page_offset(page);
509 u64 found_start;
510 struct extent_buffer *eb;
511
512 eb = (struct extent_buffer *)page->private;
513 if (page != eb->pages[0])
514 return 0;
515
516 found_start = btrfs_header_bytenr(eb);
517 /*
518 * Please do not consolidate these warnings into a single if.
519 * It is useful to know what went wrong.
520 */
521 if (WARN_ON(found_start != start))
522 return -EUCLEAN;
523 if (WARN_ON(!PageUptodate(page)))
524 return -EUCLEAN;
525
526 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
527 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
528
529 return csum_tree_block(fs_info, eb, 0);
530}
531
532static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
533 struct extent_buffer *eb)
534{
535 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
536 u8 fsid[BTRFS_UUID_SIZE];
537 int ret = 1;
538
539 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
540 while (fs_devices) {
541 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
542 ret = 0;
543 break;
544 }
545 fs_devices = fs_devices->seed;
546 }
547 return ret;
548}
549
550#define CORRUPT(reason, eb, root, slot) \
551 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
552 "root=%llu, slot=%d", reason, \
553 btrfs_header_bytenr(eb), root->objectid, slot)
554
555static noinline int check_leaf(struct btrfs_root *root,
556 struct extent_buffer *leaf)
557{
558 struct btrfs_key key;
559 struct btrfs_key leaf_key;
560 u32 nritems = btrfs_header_nritems(leaf);
561 int slot;
562
563 if (nritems == 0)
564 return 0;
565
566 /* Check the 0 item */
567 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
568 BTRFS_LEAF_DATA_SIZE(root)) {
569 CORRUPT("invalid item offset size pair", leaf, root, 0);
570 return -EIO;
571 }
572
573 /*
574 * Check to make sure each items keys are in the correct order and their
575 * offsets make sense. We only have to loop through nritems-1 because
576 * we check the current slot against the next slot, which verifies the
577 * next slot's offset+size makes sense and that the current's slot
578 * offset is correct.
579 */
580 for (slot = 0; slot < nritems - 1; slot++) {
581 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
582 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
583
584 /* Make sure the keys are in the right order */
585 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
586 CORRUPT("bad key order", leaf, root, slot);
587 return -EIO;
588 }
589
590 /*
591 * Make sure the offset and ends are right, remember that the
592 * item data starts at the end of the leaf and grows towards the
593 * front.
594 */
595 if (btrfs_item_offset_nr(leaf, slot) !=
596 btrfs_item_end_nr(leaf, slot + 1)) {
597 CORRUPT("slot offset bad", leaf, root, slot);
598 return -EIO;
599 }
600
601 /*
602 * Check to make sure that we don't point outside of the leaf,
603 * just incase all the items are consistent to eachother, but
604 * all point outside of the leaf.
605 */
606 if (btrfs_item_end_nr(leaf, slot) >
607 BTRFS_LEAF_DATA_SIZE(root)) {
608 CORRUPT("slot end outside of leaf", leaf, root, slot);
609 return -EIO;
610 }
611 }
612
613 return 0;
614}
615
616static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
617 u64 phy_offset, struct page *page,
618 u64 start, u64 end, int mirror)
619{
620 u64 found_start;
621 int found_level;
622 struct extent_buffer *eb;
623 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
624 struct btrfs_fs_info *fs_info = root->fs_info;
625 int ret = 0;
626 int reads_done;
627
628 if (!page->private)
629 goto out;
630
631 eb = (struct extent_buffer *)page->private;
632
633 /* the pending IO might have been the only thing that kept this buffer
634 * in memory. Make sure we have a ref for all this other checks
635 */
636 extent_buffer_get(eb);
637
638 reads_done = atomic_dec_and_test(&eb->io_pages);
639 if (!reads_done)
640 goto err;
641
642 eb->read_mirror = mirror;
643 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
644 ret = -EIO;
645 goto err;
646 }
647
648 found_start = btrfs_header_bytenr(eb);
649 if (found_start != eb->start) {
650 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
651 found_start, eb->start);
652 ret = -EIO;
653 goto err;
654 }
655 if (check_tree_block_fsid(fs_info, eb)) {
656 btrfs_err_rl(fs_info, "bad fsid on block %llu",
657 eb->start);
658 ret = -EIO;
659 goto err;
660 }
661 found_level = btrfs_header_level(eb);
662 if (found_level >= BTRFS_MAX_LEVEL) {
663 btrfs_err(fs_info, "bad tree block level %d",
664 (int)btrfs_header_level(eb));
665 ret = -EIO;
666 goto err;
667 }
668
669 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
670 eb, found_level);
671
672 ret = csum_tree_block(fs_info, eb, 1);
673 if (ret)
674 goto err;
675
676 /*
677 * If this is a leaf block and it is corrupt, set the corrupt bit so
678 * that we don't try and read the other copies of this block, just
679 * return -EIO.
680 */
681 if (found_level == 0 && check_leaf(root, eb)) {
682 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
683 ret = -EIO;
684 }
685
686 if (!ret)
687 set_extent_buffer_uptodate(eb);
688err:
689 if (reads_done &&
690 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
691 btree_readahead_hook(fs_info, eb, eb->start, ret);
692
693 if (ret) {
694 /*
695 * our io error hook is going to dec the io pages
696 * again, we have to make sure it has something
697 * to decrement
698 */
699 atomic_inc(&eb->io_pages);
700 clear_extent_buffer_uptodate(eb);
701 }
702 free_extent_buffer(eb);
703out:
704 return ret;
705}
706
707static int btree_io_failed_hook(struct page *page, int failed_mirror)
708{
709 struct extent_buffer *eb;
710
711 eb = (struct extent_buffer *)page->private;
712 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
713 eb->read_mirror = failed_mirror;
714 atomic_dec(&eb->io_pages);
715 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
716 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
717 return -EIO; /* we fixed nothing */
718}
719
720static void end_workqueue_bio(struct bio *bio)
721{
722 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
723 struct btrfs_fs_info *fs_info;
724 struct btrfs_workqueue *wq;
725 btrfs_work_func_t func;
726
727 fs_info = end_io_wq->info;
728 end_io_wq->error = bio->bi_error;
729
730 if (bio->bi_rw & REQ_WRITE) {
731 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
732 wq = fs_info->endio_meta_write_workers;
733 func = btrfs_endio_meta_write_helper;
734 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
735 wq = fs_info->endio_freespace_worker;
736 func = btrfs_freespace_write_helper;
737 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
738 wq = fs_info->endio_raid56_workers;
739 func = btrfs_endio_raid56_helper;
740 } else {
741 wq = fs_info->endio_write_workers;
742 func = btrfs_endio_write_helper;
743 }
744 } else {
745 if (unlikely(end_io_wq->metadata ==
746 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
747 wq = fs_info->endio_repair_workers;
748 func = btrfs_endio_repair_helper;
749 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
750 wq = fs_info->endio_raid56_workers;
751 func = btrfs_endio_raid56_helper;
752 } else if (end_io_wq->metadata) {
753 wq = fs_info->endio_meta_workers;
754 func = btrfs_endio_meta_helper;
755 } else {
756 wq = fs_info->endio_workers;
757 func = btrfs_endio_helper;
758 }
759 }
760
761 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
762 btrfs_queue_work(wq, &end_io_wq->work);
763}
764
765int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
766 enum btrfs_wq_endio_type metadata)
767{
768 struct btrfs_end_io_wq *end_io_wq;
769
770 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
771 if (!end_io_wq)
772 return -ENOMEM;
773
774 end_io_wq->private = bio->bi_private;
775 end_io_wq->end_io = bio->bi_end_io;
776 end_io_wq->info = info;
777 end_io_wq->error = 0;
778 end_io_wq->bio = bio;
779 end_io_wq->metadata = metadata;
780
781 bio->bi_private = end_io_wq;
782 bio->bi_end_io = end_workqueue_bio;
783 return 0;
784}
785
786unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
787{
788 unsigned long limit = min_t(unsigned long,
789 info->thread_pool_size,
790 info->fs_devices->open_devices);
791 return 256 * limit;
792}
793
794static void run_one_async_start(struct btrfs_work *work)
795{
796 struct async_submit_bio *async;
797 int ret;
798
799 async = container_of(work, struct async_submit_bio, work);
800 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
801 async->mirror_num, async->bio_flags,
802 async->bio_offset);
803 if (ret)
804 async->error = ret;
805}
806
807static void run_one_async_done(struct btrfs_work *work)
808{
809 struct btrfs_fs_info *fs_info;
810 struct async_submit_bio *async;
811 int limit;
812
813 async = container_of(work, struct async_submit_bio, work);
814 fs_info = BTRFS_I(async->inode)->root->fs_info;
815
816 limit = btrfs_async_submit_limit(fs_info);
817 limit = limit * 2 / 3;
818
819 /*
820 * atomic_dec_return implies a barrier for waitqueue_active
821 */
822 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
823 waitqueue_active(&fs_info->async_submit_wait))
824 wake_up(&fs_info->async_submit_wait);
825
826 /* If an error occurred we just want to clean up the bio and move on */
827 if (async->error) {
828 async->bio->bi_error = async->error;
829 bio_endio(async->bio);
830 return;
831 }
832
833 async->submit_bio_done(async->inode, async->rw, async->bio,
834 async->mirror_num, async->bio_flags,
835 async->bio_offset);
836}
837
838static void run_one_async_free(struct btrfs_work *work)
839{
840 struct async_submit_bio *async;
841
842 async = container_of(work, struct async_submit_bio, work);
843 kfree(async);
844}
845
846int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
847 int rw, struct bio *bio, int mirror_num,
848 unsigned long bio_flags,
849 u64 bio_offset,
850 extent_submit_bio_hook_t *submit_bio_start,
851 extent_submit_bio_hook_t *submit_bio_done)
852{
853 struct async_submit_bio *async;
854
855 async = kmalloc(sizeof(*async), GFP_NOFS);
856 if (!async)
857 return -ENOMEM;
858
859 async->inode = inode;
860 async->rw = rw;
861 async->bio = bio;
862 async->mirror_num = mirror_num;
863 async->submit_bio_start = submit_bio_start;
864 async->submit_bio_done = submit_bio_done;
865
866 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
867 run_one_async_done, run_one_async_free);
868
869 async->bio_flags = bio_flags;
870 async->bio_offset = bio_offset;
871
872 async->error = 0;
873
874 atomic_inc(&fs_info->nr_async_submits);
875
876 if (rw & REQ_SYNC)
877 btrfs_set_work_high_priority(&async->work);
878
879 btrfs_queue_work(fs_info->workers, &async->work);
880
881 while (atomic_read(&fs_info->async_submit_draining) &&
882 atomic_read(&fs_info->nr_async_submits)) {
883 wait_event(fs_info->async_submit_wait,
884 (atomic_read(&fs_info->nr_async_submits) == 0));
885 }
886
887 return 0;
888}
889
890static int btree_csum_one_bio(struct bio *bio)
891{
892 struct bio_vec *bvec;
893 struct btrfs_root *root;
894 int i, ret = 0;
895
896 bio_for_each_segment_all(bvec, bio, i) {
897 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
898 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
899 if (ret)
900 break;
901 }
902
903 return ret;
904}
905
906static int __btree_submit_bio_start(struct inode *inode, int rw,
907 struct bio *bio, int mirror_num,
908 unsigned long bio_flags,
909 u64 bio_offset)
910{
911 /*
912 * when we're called for a write, we're already in the async
913 * submission context. Just jump into btrfs_map_bio
914 */
915 return btree_csum_one_bio(bio);
916}
917
918static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
919 int mirror_num, unsigned long bio_flags,
920 u64 bio_offset)
921{
922 int ret;
923
924 /*
925 * when we're called for a write, we're already in the async
926 * submission context. Just jump into btrfs_map_bio
927 */
928 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
929 if (ret) {
930 bio->bi_error = ret;
931 bio_endio(bio);
932 }
933 return ret;
934}
935
936static int check_async_write(struct inode *inode, unsigned long bio_flags)
937{
938 if (bio_flags & EXTENT_BIO_TREE_LOG)
939 return 0;
940#ifdef CONFIG_X86
941 if (static_cpu_has(X86_FEATURE_XMM4_2))
942 return 0;
943#endif
944 return 1;
945}
946
947static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
948 int mirror_num, unsigned long bio_flags,
949 u64 bio_offset)
950{
951 int async = check_async_write(inode, bio_flags);
952 int ret;
953
954 if (!(rw & REQ_WRITE)) {
955 /*
956 * called for a read, do the setup so that checksum validation
957 * can happen in the async kernel threads
958 */
959 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
960 bio, BTRFS_WQ_ENDIO_METADATA);
961 if (ret)
962 goto out_w_error;
963 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
964 mirror_num, 0);
965 } else if (!async) {
966 ret = btree_csum_one_bio(bio);
967 if (ret)
968 goto out_w_error;
969 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
970 mirror_num, 0);
971 } else {
972 /*
973 * kthread helpers are used to submit writes so that
974 * checksumming can happen in parallel across all CPUs
975 */
976 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
977 inode, rw, bio, mirror_num, 0,
978 bio_offset,
979 __btree_submit_bio_start,
980 __btree_submit_bio_done);
981 }
982
983 if (ret)
984 goto out_w_error;
985 return 0;
986
987out_w_error:
988 bio->bi_error = ret;
989 bio_endio(bio);
990 return ret;
991}
992
993#ifdef CONFIG_MIGRATION
994static int btree_migratepage(struct address_space *mapping,
995 struct page *newpage, struct page *page,
996 enum migrate_mode mode)
997{
998 /*
999 * we can't safely write a btree page from here,
1000 * we haven't done the locking hook
1001 */
1002 if (PageDirty(page))
1003 return -EAGAIN;
1004 /*
1005 * Buffers may be managed in a filesystem specific way.
1006 * We must have no buffers or drop them.
1007 */
1008 if (page_has_private(page) &&
1009 !try_to_release_page(page, GFP_KERNEL))
1010 return -EAGAIN;
1011 return migrate_page(mapping, newpage, page, mode);
1012}
1013#endif
1014
1015
1016static int btree_writepages(struct address_space *mapping,
1017 struct writeback_control *wbc)
1018{
1019 struct btrfs_fs_info *fs_info;
1020 int ret;
1021
1022 if (wbc->sync_mode == WB_SYNC_NONE) {
1023
1024 if (wbc->for_kupdate)
1025 return 0;
1026
1027 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1028 /* this is a bit racy, but that's ok */
1029 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1030 BTRFS_DIRTY_METADATA_THRESH);
1031 if (ret < 0)
1032 return 0;
1033 }
1034 return btree_write_cache_pages(mapping, wbc);
1035}
1036
1037static int btree_readpage(struct file *file, struct page *page)
1038{
1039 struct extent_io_tree *tree;
1040 tree = &BTRFS_I(page->mapping->host)->io_tree;
1041 return extent_read_full_page(tree, page, btree_get_extent, 0);
1042}
1043
1044static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1045{
1046 if (PageWriteback(page) || PageDirty(page))
1047 return 0;
1048
1049 return try_release_extent_buffer(page);
1050}
1051
1052static void btree_invalidatepage(struct page *page, unsigned int offset,
1053 unsigned int length)
1054{
1055 struct extent_io_tree *tree;
1056 tree = &BTRFS_I(page->mapping->host)->io_tree;
1057 extent_invalidatepage(tree, page, offset);
1058 btree_releasepage(page, GFP_NOFS);
1059 if (PagePrivate(page)) {
1060 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1061 "page private not zero on page %llu",
1062 (unsigned long long)page_offset(page));
1063 ClearPagePrivate(page);
1064 set_page_private(page, 0);
1065 put_page(page);
1066 }
1067}
1068
1069static int btree_set_page_dirty(struct page *page)
1070{
1071#ifdef DEBUG
1072 struct extent_buffer *eb;
1073
1074 BUG_ON(!PagePrivate(page));
1075 eb = (struct extent_buffer *)page->private;
1076 BUG_ON(!eb);
1077 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1078 BUG_ON(!atomic_read(&eb->refs));
1079 btrfs_assert_tree_locked(eb);
1080#endif
1081 return __set_page_dirty_nobuffers(page);
1082}
1083
1084static const struct address_space_operations btree_aops = {
1085 .readpage = btree_readpage,
1086 .writepages = btree_writepages,
1087 .releasepage = btree_releasepage,
1088 .invalidatepage = btree_invalidatepage,
1089#ifdef CONFIG_MIGRATION
1090 .migratepage = btree_migratepage,
1091#endif
1092 .set_page_dirty = btree_set_page_dirty,
1093};
1094
1095void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1096{
1097 struct extent_buffer *buf = NULL;
1098 struct inode *btree_inode = root->fs_info->btree_inode;
1099
1100 buf = btrfs_find_create_tree_block(root, bytenr);
1101 if (!buf)
1102 return;
1103 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1104 buf, 0, WAIT_NONE, btree_get_extent, 0);
1105 free_extent_buffer(buf);
1106}
1107
1108int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1109 int mirror_num, struct extent_buffer **eb)
1110{
1111 struct extent_buffer *buf = NULL;
1112 struct inode *btree_inode = root->fs_info->btree_inode;
1113 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1114 int ret;
1115
1116 buf = btrfs_find_create_tree_block(root, bytenr);
1117 if (!buf)
1118 return 0;
1119
1120 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1121
1122 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1123 btree_get_extent, mirror_num);
1124 if (ret) {
1125 free_extent_buffer(buf);
1126 return ret;
1127 }
1128
1129 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1130 free_extent_buffer(buf);
1131 return -EIO;
1132 } else if (extent_buffer_uptodate(buf)) {
1133 *eb = buf;
1134 } else {
1135 free_extent_buffer(buf);
1136 }
1137 return 0;
1138}
1139
1140struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1141 u64 bytenr)
1142{
1143 return find_extent_buffer(fs_info, bytenr);
1144}
1145
1146struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1147 u64 bytenr)
1148{
1149 if (btrfs_test_is_dummy_root(root))
1150 return alloc_test_extent_buffer(root->fs_info, bytenr);
1151 return alloc_extent_buffer(root->fs_info, bytenr);
1152}
1153
1154
1155int btrfs_write_tree_block(struct extent_buffer *buf)
1156{
1157 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1158 buf->start + buf->len - 1);
1159}
1160
1161int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1162{
1163 return filemap_fdatawait_range(buf->pages[0]->mapping,
1164 buf->start, buf->start + buf->len - 1);
1165}
1166
1167struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1168 u64 parent_transid)
1169{
1170 struct extent_buffer *buf = NULL;
1171 int ret;
1172
1173 buf = btrfs_find_create_tree_block(root, bytenr);
1174 if (!buf)
1175 return ERR_PTR(-ENOMEM);
1176
1177 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1178 if (ret) {
1179 free_extent_buffer(buf);
1180 return ERR_PTR(ret);
1181 }
1182 return buf;
1183
1184}
1185
1186void clean_tree_block(struct btrfs_trans_handle *trans,
1187 struct btrfs_fs_info *fs_info,
1188 struct extent_buffer *buf)
1189{
1190 if (btrfs_header_generation(buf) ==
1191 fs_info->running_transaction->transid) {
1192 btrfs_assert_tree_locked(buf);
1193
1194 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1195 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1196 -buf->len,
1197 fs_info->dirty_metadata_batch);
1198 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1199 btrfs_set_lock_blocking(buf);
1200 clear_extent_buffer_dirty(buf);
1201 }
1202 }
1203}
1204
1205static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1206{
1207 struct btrfs_subvolume_writers *writers;
1208 int ret;
1209
1210 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1211 if (!writers)
1212 return ERR_PTR(-ENOMEM);
1213
1214 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1215 if (ret < 0) {
1216 kfree(writers);
1217 return ERR_PTR(ret);
1218 }
1219
1220 init_waitqueue_head(&writers->wait);
1221 return writers;
1222}
1223
1224static void
1225btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1226{
1227 percpu_counter_destroy(&writers->counter);
1228 kfree(writers);
1229}
1230
1231static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1232 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1233 u64 objectid)
1234{
1235 root->node = NULL;
1236 root->commit_root = NULL;
1237 root->sectorsize = sectorsize;
1238 root->nodesize = nodesize;
1239 root->stripesize = stripesize;
1240 root->state = 0;
1241 root->orphan_cleanup_state = 0;
1242
1243 root->objectid = objectid;
1244 root->last_trans = 0;
1245 root->highest_objectid = 0;
1246 root->nr_delalloc_inodes = 0;
1247 root->nr_ordered_extents = 0;
1248 root->name = NULL;
1249 root->inode_tree = RB_ROOT;
1250 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1251 root->block_rsv = NULL;
1252 root->orphan_block_rsv = NULL;
1253
1254 INIT_LIST_HEAD(&root->dirty_list);
1255 INIT_LIST_HEAD(&root->root_list);
1256 INIT_LIST_HEAD(&root->delalloc_inodes);
1257 INIT_LIST_HEAD(&root->delalloc_root);
1258 INIT_LIST_HEAD(&root->ordered_extents);
1259 INIT_LIST_HEAD(&root->ordered_root);
1260 INIT_LIST_HEAD(&root->logged_list[0]);
1261 INIT_LIST_HEAD(&root->logged_list[1]);
1262 spin_lock_init(&root->orphan_lock);
1263 spin_lock_init(&root->inode_lock);
1264 spin_lock_init(&root->delalloc_lock);
1265 spin_lock_init(&root->ordered_extent_lock);
1266 spin_lock_init(&root->accounting_lock);
1267 spin_lock_init(&root->log_extents_lock[0]);
1268 spin_lock_init(&root->log_extents_lock[1]);
1269 mutex_init(&root->objectid_mutex);
1270 mutex_init(&root->log_mutex);
1271 mutex_init(&root->ordered_extent_mutex);
1272 mutex_init(&root->delalloc_mutex);
1273 init_waitqueue_head(&root->log_writer_wait);
1274 init_waitqueue_head(&root->log_commit_wait[0]);
1275 init_waitqueue_head(&root->log_commit_wait[1]);
1276 INIT_LIST_HEAD(&root->log_ctxs[0]);
1277 INIT_LIST_HEAD(&root->log_ctxs[1]);
1278 atomic_set(&root->log_commit[0], 0);
1279 atomic_set(&root->log_commit[1], 0);
1280 atomic_set(&root->log_writers, 0);
1281 atomic_set(&root->log_batch, 0);
1282 atomic_set(&root->orphan_inodes, 0);
1283 atomic_set(&root->refs, 1);
1284 atomic_set(&root->will_be_snapshoted, 0);
1285 atomic_set(&root->qgroup_meta_rsv, 0);
1286 root->log_transid = 0;
1287 root->log_transid_committed = -1;
1288 root->last_log_commit = 0;
1289 if (fs_info)
1290 extent_io_tree_init(&root->dirty_log_pages,
1291 fs_info->btree_inode->i_mapping);
1292
1293 memset(&root->root_key, 0, sizeof(root->root_key));
1294 memset(&root->root_item, 0, sizeof(root->root_item));
1295 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1296 if (fs_info)
1297 root->defrag_trans_start = fs_info->generation;
1298 else
1299 root->defrag_trans_start = 0;
1300 root->root_key.objectid = objectid;
1301 root->anon_dev = 0;
1302
1303 spin_lock_init(&root->root_item_lock);
1304}
1305
1306static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1307 gfp_t flags)
1308{
1309 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1310 if (root)
1311 root->fs_info = fs_info;
1312 return root;
1313}
1314
1315#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1316/* Should only be used by the testing infrastructure */
1317struct btrfs_root *btrfs_alloc_dummy_root(void)
1318{
1319 struct btrfs_root *root;
1320
1321 root = btrfs_alloc_root(NULL, GFP_KERNEL);
1322 if (!root)
1323 return ERR_PTR(-ENOMEM);
1324 __setup_root(4096, 4096, 4096, root, NULL, 1);
1325 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1326 root->alloc_bytenr = 0;
1327
1328 return root;
1329}
1330#endif
1331
1332struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1333 struct btrfs_fs_info *fs_info,
1334 u64 objectid)
1335{
1336 struct extent_buffer *leaf;
1337 struct btrfs_root *tree_root = fs_info->tree_root;
1338 struct btrfs_root *root;
1339 struct btrfs_key key;
1340 int ret = 0;
1341 uuid_le uuid;
1342
1343 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1344 if (!root)
1345 return ERR_PTR(-ENOMEM);
1346
1347 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1348 tree_root->stripesize, root, fs_info, objectid);
1349 root->root_key.objectid = objectid;
1350 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1351 root->root_key.offset = 0;
1352
1353 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1354 if (IS_ERR(leaf)) {
1355 ret = PTR_ERR(leaf);
1356 leaf = NULL;
1357 goto fail;
1358 }
1359
1360 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1361 btrfs_set_header_bytenr(leaf, leaf->start);
1362 btrfs_set_header_generation(leaf, trans->transid);
1363 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1364 btrfs_set_header_owner(leaf, objectid);
1365 root->node = leaf;
1366
1367 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1368 BTRFS_FSID_SIZE);
1369 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1370 btrfs_header_chunk_tree_uuid(leaf),
1371 BTRFS_UUID_SIZE);
1372 btrfs_mark_buffer_dirty(leaf);
1373
1374 root->commit_root = btrfs_root_node(root);
1375 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1376
1377 root->root_item.flags = 0;
1378 root->root_item.byte_limit = 0;
1379 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1380 btrfs_set_root_generation(&root->root_item, trans->transid);
1381 btrfs_set_root_level(&root->root_item, 0);
1382 btrfs_set_root_refs(&root->root_item, 1);
1383 btrfs_set_root_used(&root->root_item, leaf->len);
1384 btrfs_set_root_last_snapshot(&root->root_item, 0);
1385 btrfs_set_root_dirid(&root->root_item, 0);
1386 uuid_le_gen(&uuid);
1387 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1388 root->root_item.drop_level = 0;
1389
1390 key.objectid = objectid;
1391 key.type = BTRFS_ROOT_ITEM_KEY;
1392 key.offset = 0;
1393 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1394 if (ret)
1395 goto fail;
1396
1397 btrfs_tree_unlock(leaf);
1398
1399 return root;
1400
1401fail:
1402 if (leaf) {
1403 btrfs_tree_unlock(leaf);
1404 free_extent_buffer(root->commit_root);
1405 free_extent_buffer(leaf);
1406 }
1407 kfree(root);
1408
1409 return ERR_PTR(ret);
1410}
1411
1412static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1413 struct btrfs_fs_info *fs_info)
1414{
1415 struct btrfs_root *root;
1416 struct btrfs_root *tree_root = fs_info->tree_root;
1417 struct extent_buffer *leaf;
1418
1419 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1420 if (!root)
1421 return ERR_PTR(-ENOMEM);
1422
1423 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1424 tree_root->stripesize, root, fs_info,
1425 BTRFS_TREE_LOG_OBJECTID);
1426
1427 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1428 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1429 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1430
1431 /*
1432 * DON'T set REF_COWS for log trees
1433 *
1434 * log trees do not get reference counted because they go away
1435 * before a real commit is actually done. They do store pointers
1436 * to file data extents, and those reference counts still get
1437 * updated (along with back refs to the log tree).
1438 */
1439
1440 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1441 NULL, 0, 0, 0);
1442 if (IS_ERR(leaf)) {
1443 kfree(root);
1444 return ERR_CAST(leaf);
1445 }
1446
1447 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1448 btrfs_set_header_bytenr(leaf, leaf->start);
1449 btrfs_set_header_generation(leaf, trans->transid);
1450 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1451 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1452 root->node = leaf;
1453
1454 write_extent_buffer(root->node, root->fs_info->fsid,
1455 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1456 btrfs_mark_buffer_dirty(root->node);
1457 btrfs_tree_unlock(root->node);
1458 return root;
1459}
1460
1461int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1462 struct btrfs_fs_info *fs_info)
1463{
1464 struct btrfs_root *log_root;
1465
1466 log_root = alloc_log_tree(trans, fs_info);
1467 if (IS_ERR(log_root))
1468 return PTR_ERR(log_root);
1469 WARN_ON(fs_info->log_root_tree);
1470 fs_info->log_root_tree = log_root;
1471 return 0;
1472}
1473
1474int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1475 struct btrfs_root *root)
1476{
1477 struct btrfs_root *log_root;
1478 struct btrfs_inode_item *inode_item;
1479
1480 log_root = alloc_log_tree(trans, root->fs_info);
1481 if (IS_ERR(log_root))
1482 return PTR_ERR(log_root);
1483
1484 log_root->last_trans = trans->transid;
1485 log_root->root_key.offset = root->root_key.objectid;
1486
1487 inode_item = &log_root->root_item.inode;
1488 btrfs_set_stack_inode_generation(inode_item, 1);
1489 btrfs_set_stack_inode_size(inode_item, 3);
1490 btrfs_set_stack_inode_nlink(inode_item, 1);
1491 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1492 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1493
1494 btrfs_set_root_node(&log_root->root_item, log_root->node);
1495
1496 WARN_ON(root->log_root);
1497 root->log_root = log_root;
1498 root->log_transid = 0;
1499 root->log_transid_committed = -1;
1500 root->last_log_commit = 0;
1501 return 0;
1502}
1503
1504static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1505 struct btrfs_key *key)
1506{
1507 struct btrfs_root *root;
1508 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1509 struct btrfs_path *path;
1510 u64 generation;
1511 int ret;
1512
1513 path = btrfs_alloc_path();
1514 if (!path)
1515 return ERR_PTR(-ENOMEM);
1516
1517 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1518 if (!root) {
1519 ret = -ENOMEM;
1520 goto alloc_fail;
1521 }
1522
1523 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1524 tree_root->stripesize, root, fs_info, key->objectid);
1525
1526 ret = btrfs_find_root(tree_root, key, path,
1527 &root->root_item, &root->root_key);
1528 if (ret) {
1529 if (ret > 0)
1530 ret = -ENOENT;
1531 goto find_fail;
1532 }
1533
1534 generation = btrfs_root_generation(&root->root_item);
1535 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1536 generation);
1537 if (IS_ERR(root->node)) {
1538 ret = PTR_ERR(root->node);
1539 goto find_fail;
1540 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1541 ret = -EIO;
1542 free_extent_buffer(root->node);
1543 goto find_fail;
1544 }
1545 root->commit_root = btrfs_root_node(root);
1546out:
1547 btrfs_free_path(path);
1548 return root;
1549
1550find_fail:
1551 kfree(root);
1552alloc_fail:
1553 root = ERR_PTR(ret);
1554 goto out;
1555}
1556
1557struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1558 struct btrfs_key *location)
1559{
1560 struct btrfs_root *root;
1561
1562 root = btrfs_read_tree_root(tree_root, location);
1563 if (IS_ERR(root))
1564 return root;
1565
1566 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1567 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1568 btrfs_check_and_init_root_item(&root->root_item);
1569 }
1570
1571 return root;
1572}
1573
1574int btrfs_init_fs_root(struct btrfs_root *root)
1575{
1576 int ret;
1577 struct btrfs_subvolume_writers *writers;
1578
1579 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1580 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1581 GFP_NOFS);
1582 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1583 ret = -ENOMEM;
1584 goto fail;
1585 }
1586
1587 writers = btrfs_alloc_subvolume_writers();
1588 if (IS_ERR(writers)) {
1589 ret = PTR_ERR(writers);
1590 goto fail;
1591 }
1592 root->subv_writers = writers;
1593
1594 btrfs_init_free_ino_ctl(root);
1595 spin_lock_init(&root->ino_cache_lock);
1596 init_waitqueue_head(&root->ino_cache_wait);
1597
1598 ret = get_anon_bdev(&root->anon_dev);
1599 if (ret)
1600 goto free_writers;
1601
1602 mutex_lock(&root->objectid_mutex);
1603 ret = btrfs_find_highest_objectid(root,
1604 &root->highest_objectid);
1605 if (ret) {
1606 mutex_unlock(&root->objectid_mutex);
1607 goto free_root_dev;
1608 }
1609
1610 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1611
1612 mutex_unlock(&root->objectid_mutex);
1613
1614 return 0;
1615
1616free_root_dev:
1617 free_anon_bdev(root->anon_dev);
1618free_writers:
1619 btrfs_free_subvolume_writers(root->subv_writers);
1620fail:
1621 kfree(root->free_ino_ctl);
1622 kfree(root->free_ino_pinned);
1623 return ret;
1624}
1625
1626static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1627 u64 root_id)
1628{
1629 struct btrfs_root *root;
1630
1631 spin_lock(&fs_info->fs_roots_radix_lock);
1632 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1633 (unsigned long)root_id);
1634 spin_unlock(&fs_info->fs_roots_radix_lock);
1635 return root;
1636}
1637
1638int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1639 struct btrfs_root *root)
1640{
1641 int ret;
1642
1643 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1644 if (ret)
1645 return ret;
1646
1647 spin_lock(&fs_info->fs_roots_radix_lock);
1648 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1649 (unsigned long)root->root_key.objectid,
1650 root);
1651 if (ret == 0)
1652 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1653 spin_unlock(&fs_info->fs_roots_radix_lock);
1654 radix_tree_preload_end();
1655
1656 return ret;
1657}
1658
1659struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1660 struct btrfs_key *location,
1661 bool check_ref)
1662{
1663 struct btrfs_root *root;
1664 struct btrfs_path *path;
1665 struct btrfs_key key;
1666 int ret;
1667
1668 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1669 return fs_info->tree_root;
1670 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1671 return fs_info->extent_root;
1672 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1673 return fs_info->chunk_root;
1674 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1675 return fs_info->dev_root;
1676 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1677 return fs_info->csum_root;
1678 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1679 return fs_info->quota_root ? fs_info->quota_root :
1680 ERR_PTR(-ENOENT);
1681 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1682 return fs_info->uuid_root ? fs_info->uuid_root :
1683 ERR_PTR(-ENOENT);
1684 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1685 return fs_info->free_space_root ? fs_info->free_space_root :
1686 ERR_PTR(-ENOENT);
1687again:
1688 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1689 if (root) {
1690 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1691 return ERR_PTR(-ENOENT);
1692 return root;
1693 }
1694
1695 root = btrfs_read_fs_root(fs_info->tree_root, location);
1696 if (IS_ERR(root))
1697 return root;
1698
1699 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1700 ret = -ENOENT;
1701 goto fail;
1702 }
1703
1704 ret = btrfs_init_fs_root(root);
1705 if (ret)
1706 goto fail;
1707
1708 path = btrfs_alloc_path();
1709 if (!path) {
1710 ret = -ENOMEM;
1711 goto fail;
1712 }
1713 key.objectid = BTRFS_ORPHAN_OBJECTID;
1714 key.type = BTRFS_ORPHAN_ITEM_KEY;
1715 key.offset = location->objectid;
1716
1717 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1718 btrfs_free_path(path);
1719 if (ret < 0)
1720 goto fail;
1721 if (ret == 0)
1722 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1723
1724 ret = btrfs_insert_fs_root(fs_info, root);
1725 if (ret) {
1726 if (ret == -EEXIST) {
1727 free_fs_root(root);
1728 goto again;
1729 }
1730 goto fail;
1731 }
1732 return root;
1733fail:
1734 free_fs_root(root);
1735 return ERR_PTR(ret);
1736}
1737
1738static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1739{
1740 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1741 int ret = 0;
1742 struct btrfs_device *device;
1743 struct backing_dev_info *bdi;
1744
1745 rcu_read_lock();
1746 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1747 if (!device->bdev)
1748 continue;
1749 bdi = blk_get_backing_dev_info(device->bdev);
1750 if (bdi_congested(bdi, bdi_bits)) {
1751 ret = 1;
1752 break;
1753 }
1754 }
1755 rcu_read_unlock();
1756 return ret;
1757}
1758
1759static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1760{
1761 int err;
1762
1763 err = bdi_setup_and_register(bdi, "btrfs");
1764 if (err)
1765 return err;
1766
1767 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1768 bdi->congested_fn = btrfs_congested_fn;
1769 bdi->congested_data = info;
1770 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1771 return 0;
1772}
1773
1774/*
1775 * called by the kthread helper functions to finally call the bio end_io
1776 * functions. This is where read checksum verification actually happens
1777 */
1778static void end_workqueue_fn(struct btrfs_work *work)
1779{
1780 struct bio *bio;
1781 struct btrfs_end_io_wq *end_io_wq;
1782
1783 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1784 bio = end_io_wq->bio;
1785
1786 bio->bi_error = end_io_wq->error;
1787 bio->bi_private = end_io_wq->private;
1788 bio->bi_end_io = end_io_wq->end_io;
1789 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1790 bio_endio(bio);
1791}
1792
1793static int cleaner_kthread(void *arg)
1794{
1795 struct btrfs_root *root = arg;
1796 int again;
1797 struct btrfs_trans_handle *trans;
1798
1799 do {
1800 again = 0;
1801
1802 /* Make the cleaner go to sleep early. */
1803 if (btrfs_need_cleaner_sleep(root))
1804 goto sleep;
1805
1806 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1807 goto sleep;
1808
1809 /*
1810 * Avoid the problem that we change the status of the fs
1811 * during the above check and trylock.
1812 */
1813 if (btrfs_need_cleaner_sleep(root)) {
1814 mutex_unlock(&root->fs_info->cleaner_mutex);
1815 goto sleep;
1816 }
1817
1818 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1819 btrfs_run_delayed_iputs(root);
1820 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1821
1822 again = btrfs_clean_one_deleted_snapshot(root);
1823 mutex_unlock(&root->fs_info->cleaner_mutex);
1824
1825 /*
1826 * The defragger has dealt with the R/O remount and umount,
1827 * needn't do anything special here.
1828 */
1829 btrfs_run_defrag_inodes(root->fs_info);
1830
1831 /*
1832 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1833 * with relocation (btrfs_relocate_chunk) and relocation
1834 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1835 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1836 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1837 * unused block groups.
1838 */
1839 btrfs_delete_unused_bgs(root->fs_info);
1840sleep:
1841 if (!again) {
1842 set_current_state(TASK_INTERRUPTIBLE);
1843 if (!kthread_should_stop())
1844 schedule();
1845 __set_current_state(TASK_RUNNING);
1846 }
1847 } while (!kthread_should_stop());
1848
1849 /*
1850 * Transaction kthread is stopped before us and wakes us up.
1851 * However we might have started a new transaction and COWed some
1852 * tree blocks when deleting unused block groups for example. So
1853 * make sure we commit the transaction we started to have a clean
1854 * shutdown when evicting the btree inode - if it has dirty pages
1855 * when we do the final iput() on it, eviction will trigger a
1856 * writeback for it which will fail with null pointer dereferences
1857 * since work queues and other resources were already released and
1858 * destroyed by the time the iput/eviction/writeback is made.
1859 */
1860 trans = btrfs_attach_transaction(root);
1861 if (IS_ERR(trans)) {
1862 if (PTR_ERR(trans) != -ENOENT)
1863 btrfs_err(root->fs_info,
1864 "cleaner transaction attach returned %ld",
1865 PTR_ERR(trans));
1866 } else {
1867 int ret;
1868
1869 ret = btrfs_commit_transaction(trans, root);
1870 if (ret)
1871 btrfs_err(root->fs_info,
1872 "cleaner open transaction commit returned %d",
1873 ret);
1874 }
1875
1876 return 0;
1877}
1878
1879static int transaction_kthread(void *arg)
1880{
1881 struct btrfs_root *root = arg;
1882 struct btrfs_trans_handle *trans;
1883 struct btrfs_transaction *cur;
1884 u64 transid;
1885 unsigned long now;
1886 unsigned long delay;
1887 bool cannot_commit;
1888
1889 do {
1890 cannot_commit = false;
1891 delay = HZ * root->fs_info->commit_interval;
1892 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1893
1894 spin_lock(&root->fs_info->trans_lock);
1895 cur = root->fs_info->running_transaction;
1896 if (!cur) {
1897 spin_unlock(&root->fs_info->trans_lock);
1898 goto sleep;
1899 }
1900
1901 now = get_seconds();
1902 if (cur->state < TRANS_STATE_BLOCKED &&
1903 (now < cur->start_time ||
1904 now - cur->start_time < root->fs_info->commit_interval)) {
1905 spin_unlock(&root->fs_info->trans_lock);
1906 delay = HZ * 5;
1907 goto sleep;
1908 }
1909 transid = cur->transid;
1910 spin_unlock(&root->fs_info->trans_lock);
1911
1912 /* If the file system is aborted, this will always fail. */
1913 trans = btrfs_attach_transaction(root);
1914 if (IS_ERR(trans)) {
1915 if (PTR_ERR(trans) != -ENOENT)
1916 cannot_commit = true;
1917 goto sleep;
1918 }
1919 if (transid == trans->transid) {
1920 btrfs_commit_transaction(trans, root);
1921 } else {
1922 btrfs_end_transaction(trans, root);
1923 }
1924sleep:
1925 wake_up_process(root->fs_info->cleaner_kthread);
1926 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1927
1928 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1929 &root->fs_info->fs_state)))
1930 btrfs_cleanup_transaction(root);
1931 set_current_state(TASK_INTERRUPTIBLE);
1932 if (!kthread_should_stop() &&
1933 (!btrfs_transaction_blocked(root->fs_info) ||
1934 cannot_commit))
1935 schedule_timeout(delay);
1936 __set_current_state(TASK_RUNNING);
1937 } while (!kthread_should_stop());
1938 return 0;
1939}
1940
1941/*
1942 * this will find the highest generation in the array of
1943 * root backups. The index of the highest array is returned,
1944 * or -1 if we can't find anything.
1945 *
1946 * We check to make sure the array is valid by comparing the
1947 * generation of the latest root in the array with the generation
1948 * in the super block. If they don't match we pitch it.
1949 */
1950static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1951{
1952 u64 cur;
1953 int newest_index = -1;
1954 struct btrfs_root_backup *root_backup;
1955 int i;
1956
1957 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1958 root_backup = info->super_copy->super_roots + i;
1959 cur = btrfs_backup_tree_root_gen(root_backup);
1960 if (cur == newest_gen)
1961 newest_index = i;
1962 }
1963
1964 /* check to see if we actually wrapped around */
1965 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1966 root_backup = info->super_copy->super_roots;
1967 cur = btrfs_backup_tree_root_gen(root_backup);
1968 if (cur == newest_gen)
1969 newest_index = 0;
1970 }
1971 return newest_index;
1972}
1973
1974
1975/*
1976 * find the oldest backup so we know where to store new entries
1977 * in the backup array. This will set the backup_root_index
1978 * field in the fs_info struct
1979 */
1980static void find_oldest_super_backup(struct btrfs_fs_info *info,
1981 u64 newest_gen)
1982{
1983 int newest_index = -1;
1984
1985 newest_index = find_newest_super_backup(info, newest_gen);
1986 /* if there was garbage in there, just move along */
1987 if (newest_index == -1) {
1988 info->backup_root_index = 0;
1989 } else {
1990 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1991 }
1992}
1993
1994/*
1995 * copy all the root pointers into the super backup array.
1996 * this will bump the backup pointer by one when it is
1997 * done
1998 */
1999static void backup_super_roots(struct btrfs_fs_info *info)
2000{
2001 int next_backup;
2002 struct btrfs_root_backup *root_backup;
2003 int last_backup;
2004
2005 next_backup = info->backup_root_index;
2006 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2007 BTRFS_NUM_BACKUP_ROOTS;
2008
2009 /*
2010 * just overwrite the last backup if we're at the same generation
2011 * this happens only at umount
2012 */
2013 root_backup = info->super_for_commit->super_roots + last_backup;
2014 if (btrfs_backup_tree_root_gen(root_backup) ==
2015 btrfs_header_generation(info->tree_root->node))
2016 next_backup = last_backup;
2017
2018 root_backup = info->super_for_commit->super_roots + next_backup;
2019
2020 /*
2021 * make sure all of our padding and empty slots get zero filled
2022 * regardless of which ones we use today
2023 */
2024 memset(root_backup, 0, sizeof(*root_backup));
2025
2026 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2027
2028 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2029 btrfs_set_backup_tree_root_gen(root_backup,
2030 btrfs_header_generation(info->tree_root->node));
2031
2032 btrfs_set_backup_tree_root_level(root_backup,
2033 btrfs_header_level(info->tree_root->node));
2034
2035 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2036 btrfs_set_backup_chunk_root_gen(root_backup,
2037 btrfs_header_generation(info->chunk_root->node));
2038 btrfs_set_backup_chunk_root_level(root_backup,
2039 btrfs_header_level(info->chunk_root->node));
2040
2041 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2042 btrfs_set_backup_extent_root_gen(root_backup,
2043 btrfs_header_generation(info->extent_root->node));
2044 btrfs_set_backup_extent_root_level(root_backup,
2045 btrfs_header_level(info->extent_root->node));
2046
2047 /*
2048 * we might commit during log recovery, which happens before we set
2049 * the fs_root. Make sure it is valid before we fill it in.
2050 */
2051 if (info->fs_root && info->fs_root->node) {
2052 btrfs_set_backup_fs_root(root_backup,
2053 info->fs_root->node->start);
2054 btrfs_set_backup_fs_root_gen(root_backup,
2055 btrfs_header_generation(info->fs_root->node));
2056 btrfs_set_backup_fs_root_level(root_backup,
2057 btrfs_header_level(info->fs_root->node));
2058 }
2059
2060 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2061 btrfs_set_backup_dev_root_gen(root_backup,
2062 btrfs_header_generation(info->dev_root->node));
2063 btrfs_set_backup_dev_root_level(root_backup,
2064 btrfs_header_level(info->dev_root->node));
2065
2066 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2067 btrfs_set_backup_csum_root_gen(root_backup,
2068 btrfs_header_generation(info->csum_root->node));
2069 btrfs_set_backup_csum_root_level(root_backup,
2070 btrfs_header_level(info->csum_root->node));
2071
2072 btrfs_set_backup_total_bytes(root_backup,
2073 btrfs_super_total_bytes(info->super_copy));
2074 btrfs_set_backup_bytes_used(root_backup,
2075 btrfs_super_bytes_used(info->super_copy));
2076 btrfs_set_backup_num_devices(root_backup,
2077 btrfs_super_num_devices(info->super_copy));
2078
2079 /*
2080 * if we don't copy this out to the super_copy, it won't get remembered
2081 * for the next commit
2082 */
2083 memcpy(&info->super_copy->super_roots,
2084 &info->super_for_commit->super_roots,
2085 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2086}
2087
2088/*
2089 * this copies info out of the root backup array and back into
2090 * the in-memory super block. It is meant to help iterate through
2091 * the array, so you send it the number of backups you've already
2092 * tried and the last backup index you used.
2093 *
2094 * this returns -1 when it has tried all the backups
2095 */
2096static noinline int next_root_backup(struct btrfs_fs_info *info,
2097 struct btrfs_super_block *super,
2098 int *num_backups_tried, int *backup_index)
2099{
2100 struct btrfs_root_backup *root_backup;
2101 int newest = *backup_index;
2102
2103 if (*num_backups_tried == 0) {
2104 u64 gen = btrfs_super_generation(super);
2105
2106 newest = find_newest_super_backup(info, gen);
2107 if (newest == -1)
2108 return -1;
2109
2110 *backup_index = newest;
2111 *num_backups_tried = 1;
2112 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2113 /* we've tried all the backups, all done */
2114 return -1;
2115 } else {
2116 /* jump to the next oldest backup */
2117 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2118 BTRFS_NUM_BACKUP_ROOTS;
2119 *backup_index = newest;
2120 *num_backups_tried += 1;
2121 }
2122 root_backup = super->super_roots + newest;
2123
2124 btrfs_set_super_generation(super,
2125 btrfs_backup_tree_root_gen(root_backup));
2126 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2127 btrfs_set_super_root_level(super,
2128 btrfs_backup_tree_root_level(root_backup));
2129 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2130
2131 /*
2132 * fixme: the total bytes and num_devices need to match or we should
2133 * need a fsck
2134 */
2135 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2136 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2137 return 0;
2138}
2139
2140/* helper to cleanup workers */
2141static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2142{
2143 btrfs_destroy_workqueue(fs_info->fixup_workers);
2144 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2145 btrfs_destroy_workqueue(fs_info->workers);
2146 btrfs_destroy_workqueue(fs_info->endio_workers);
2147 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2148 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2149 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2150 btrfs_destroy_workqueue(fs_info->rmw_workers);
2151 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2152 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2153 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2154 btrfs_destroy_workqueue(fs_info->submit_workers);
2155 btrfs_destroy_workqueue(fs_info->delayed_workers);
2156 btrfs_destroy_workqueue(fs_info->caching_workers);
2157 btrfs_destroy_workqueue(fs_info->readahead_workers);
2158 btrfs_destroy_workqueue(fs_info->flush_workers);
2159 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2160 btrfs_destroy_workqueue(fs_info->extent_workers);
2161}
2162
2163static void free_root_extent_buffers(struct btrfs_root *root)
2164{
2165 if (root) {
2166 free_extent_buffer(root->node);
2167 free_extent_buffer(root->commit_root);
2168 root->node = NULL;
2169 root->commit_root = NULL;
2170 }
2171}
2172
2173/* helper to cleanup tree roots */
2174static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2175{
2176 free_root_extent_buffers(info->tree_root);
2177
2178 free_root_extent_buffers(info->dev_root);
2179 free_root_extent_buffers(info->extent_root);
2180 free_root_extent_buffers(info->csum_root);
2181 free_root_extent_buffers(info->quota_root);
2182 free_root_extent_buffers(info->uuid_root);
2183 if (chunk_root)
2184 free_root_extent_buffers(info->chunk_root);
2185 free_root_extent_buffers(info->free_space_root);
2186}
2187
2188void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2189{
2190 int ret;
2191 struct btrfs_root *gang[8];
2192 int i;
2193
2194 while (!list_empty(&fs_info->dead_roots)) {
2195 gang[0] = list_entry(fs_info->dead_roots.next,
2196 struct btrfs_root, root_list);
2197 list_del(&gang[0]->root_list);
2198
2199 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2200 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2201 } else {
2202 free_extent_buffer(gang[0]->node);
2203 free_extent_buffer(gang[0]->commit_root);
2204 btrfs_put_fs_root(gang[0]);
2205 }
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 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2219 btrfs_free_log_root_tree(NULL, fs_info);
2220 btrfs_destroy_pinned_extent(fs_info->tree_root,
2221 fs_info->pinned_extents);
2222 }
2223}
2224
2225static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2226{
2227 mutex_init(&fs_info->scrub_lock);
2228 atomic_set(&fs_info->scrubs_running, 0);
2229 atomic_set(&fs_info->scrub_pause_req, 0);
2230 atomic_set(&fs_info->scrubs_paused, 0);
2231 atomic_set(&fs_info->scrub_cancel_req, 0);
2232 init_waitqueue_head(&fs_info->scrub_pause_wait);
2233 fs_info->scrub_workers_refcnt = 0;
2234}
2235
2236static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2237{
2238 spin_lock_init(&fs_info->balance_lock);
2239 mutex_init(&fs_info->balance_mutex);
2240 atomic_set(&fs_info->balance_running, 0);
2241 atomic_set(&fs_info->balance_pause_req, 0);
2242 atomic_set(&fs_info->balance_cancel_req, 0);
2243 fs_info->balance_ctl = NULL;
2244 init_waitqueue_head(&fs_info->balance_wait_q);
2245}
2246
2247static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2248 struct btrfs_root *tree_root)
2249{
2250 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2251 set_nlink(fs_info->btree_inode, 1);
2252 /*
2253 * we set the i_size on the btree inode to the max possible int.
2254 * the real end of the address space is determined by all of
2255 * the devices in the system
2256 */
2257 fs_info->btree_inode->i_size = OFFSET_MAX;
2258 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2259
2260 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2261 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2262 fs_info->btree_inode->i_mapping);
2263 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2264 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2265
2266 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2267
2268 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2269 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2270 sizeof(struct btrfs_key));
2271 set_bit(BTRFS_INODE_DUMMY,
2272 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2273 btrfs_insert_inode_hash(fs_info->btree_inode);
2274}
2275
2276static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2277{
2278 fs_info->dev_replace.lock_owner = 0;
2279 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2280 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2281 rwlock_init(&fs_info->dev_replace.lock);
2282 atomic_set(&fs_info->dev_replace.read_locks, 0);
2283 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2284 init_waitqueue_head(&fs_info->replace_wait);
2285 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2286}
2287
2288static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2289{
2290 spin_lock_init(&fs_info->qgroup_lock);
2291 mutex_init(&fs_info->qgroup_ioctl_lock);
2292 fs_info->qgroup_tree = RB_ROOT;
2293 fs_info->qgroup_op_tree = RB_ROOT;
2294 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2295 fs_info->qgroup_seq = 1;
2296 fs_info->quota_enabled = 0;
2297 fs_info->pending_quota_state = 0;
2298 fs_info->qgroup_ulist = NULL;
2299 mutex_init(&fs_info->qgroup_rescan_lock);
2300}
2301
2302static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2303 struct btrfs_fs_devices *fs_devices)
2304{
2305 int max_active = fs_info->thread_pool_size;
2306 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2307
2308 fs_info->workers =
2309 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2310 max_active, 16);
2311
2312 fs_info->delalloc_workers =
2313 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2314
2315 fs_info->flush_workers =
2316 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2317
2318 fs_info->caching_workers =
2319 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2320
2321 /*
2322 * a higher idle thresh on the submit workers makes it much more
2323 * likely that bios will be send down in a sane order to the
2324 * devices
2325 */
2326 fs_info->submit_workers =
2327 btrfs_alloc_workqueue("submit", flags,
2328 min_t(u64, fs_devices->num_devices,
2329 max_active), 64);
2330
2331 fs_info->fixup_workers =
2332 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2333
2334 /*
2335 * endios are largely parallel and should have a very
2336 * low idle thresh
2337 */
2338 fs_info->endio_workers =
2339 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2340 fs_info->endio_meta_workers =
2341 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2342 fs_info->endio_meta_write_workers =
2343 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2344 fs_info->endio_raid56_workers =
2345 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2346 fs_info->endio_repair_workers =
2347 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2348 fs_info->rmw_workers =
2349 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2350 fs_info->endio_write_workers =
2351 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2352 fs_info->endio_freespace_worker =
2353 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2354 fs_info->delayed_workers =
2355 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2356 fs_info->readahead_workers =
2357 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2358 fs_info->qgroup_rescan_workers =
2359 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2360 fs_info->extent_workers =
2361 btrfs_alloc_workqueue("extent-refs", flags,
2362 min_t(u64, fs_devices->num_devices,
2363 max_active), 8);
2364
2365 if (!(fs_info->workers && fs_info->delalloc_workers &&
2366 fs_info->submit_workers && fs_info->flush_workers &&
2367 fs_info->endio_workers && fs_info->endio_meta_workers &&
2368 fs_info->endio_meta_write_workers &&
2369 fs_info->endio_repair_workers &&
2370 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2371 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2372 fs_info->caching_workers && fs_info->readahead_workers &&
2373 fs_info->fixup_workers && fs_info->delayed_workers &&
2374 fs_info->extent_workers &&
2375 fs_info->qgroup_rescan_workers)) {
2376 return -ENOMEM;
2377 }
2378
2379 return 0;
2380}
2381
2382static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2383 struct btrfs_fs_devices *fs_devices)
2384{
2385 int ret;
2386 struct btrfs_root *tree_root = fs_info->tree_root;
2387 struct btrfs_root *log_tree_root;
2388 struct btrfs_super_block *disk_super = fs_info->super_copy;
2389 u64 bytenr = btrfs_super_log_root(disk_super);
2390
2391 if (fs_devices->rw_devices == 0) {
2392 btrfs_warn(fs_info, "log replay required on RO media");
2393 return -EIO;
2394 }
2395
2396 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2397 if (!log_tree_root)
2398 return -ENOMEM;
2399
2400 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2401 tree_root->stripesize, log_tree_root, fs_info,
2402 BTRFS_TREE_LOG_OBJECTID);
2403
2404 log_tree_root->node = read_tree_block(tree_root, bytenr,
2405 fs_info->generation + 1);
2406 if (IS_ERR(log_tree_root->node)) {
2407 btrfs_warn(fs_info, "failed to read log tree");
2408 ret = PTR_ERR(log_tree_root->node);
2409 kfree(log_tree_root);
2410 return ret;
2411 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2412 btrfs_err(fs_info, "failed to read log tree");
2413 free_extent_buffer(log_tree_root->node);
2414 kfree(log_tree_root);
2415 return -EIO;
2416 }
2417 /* returns with log_tree_root freed on success */
2418 ret = btrfs_recover_log_trees(log_tree_root);
2419 if (ret) {
2420 btrfs_std_error(tree_root->fs_info, ret,
2421 "Failed to recover log tree");
2422 free_extent_buffer(log_tree_root->node);
2423 kfree(log_tree_root);
2424 return ret;
2425 }
2426
2427 if (fs_info->sb->s_flags & MS_RDONLY) {
2428 ret = btrfs_commit_super(tree_root);
2429 if (ret)
2430 return ret;
2431 }
2432
2433 return 0;
2434}
2435
2436static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2437 struct btrfs_root *tree_root)
2438{
2439 struct btrfs_root *root;
2440 struct btrfs_key location;
2441 int ret;
2442
2443 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2444 location.type = BTRFS_ROOT_ITEM_KEY;
2445 location.offset = 0;
2446
2447 root = btrfs_read_tree_root(tree_root, &location);
2448 if (IS_ERR(root))
2449 return PTR_ERR(root);
2450 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2451 fs_info->extent_root = root;
2452
2453 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2454 root = btrfs_read_tree_root(tree_root, &location);
2455 if (IS_ERR(root))
2456 return PTR_ERR(root);
2457 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2458 fs_info->dev_root = root;
2459 btrfs_init_devices_late(fs_info);
2460
2461 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2462 root = btrfs_read_tree_root(tree_root, &location);
2463 if (IS_ERR(root))
2464 return PTR_ERR(root);
2465 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2466 fs_info->csum_root = root;
2467
2468 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2469 root = btrfs_read_tree_root(tree_root, &location);
2470 if (!IS_ERR(root)) {
2471 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2472 fs_info->quota_enabled = 1;
2473 fs_info->pending_quota_state = 1;
2474 fs_info->quota_root = root;
2475 }
2476
2477 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2478 root = btrfs_read_tree_root(tree_root, &location);
2479 if (IS_ERR(root)) {
2480 ret = PTR_ERR(root);
2481 if (ret != -ENOENT)
2482 return ret;
2483 } else {
2484 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2485 fs_info->uuid_root = root;
2486 }
2487
2488 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2489 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2490 root = btrfs_read_tree_root(tree_root, &location);
2491 if (IS_ERR(root))
2492 return PTR_ERR(root);
2493 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2494 fs_info->free_space_root = root;
2495 }
2496
2497 return 0;
2498}
2499
2500int open_ctree(struct super_block *sb,
2501 struct btrfs_fs_devices *fs_devices,
2502 char *options)
2503{
2504 u32 sectorsize;
2505 u32 nodesize;
2506 u32 stripesize;
2507 u64 generation;
2508 u64 features;
2509 struct btrfs_key location;
2510 struct buffer_head *bh;
2511 struct btrfs_super_block *disk_super;
2512 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2513 struct btrfs_root *tree_root;
2514 struct btrfs_root *chunk_root;
2515 int ret;
2516 int err = -EINVAL;
2517 int num_backups_tried = 0;
2518 int backup_index = 0;
2519 int max_active;
2520
2521 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2522 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2523 if (!tree_root || !chunk_root) {
2524 err = -ENOMEM;
2525 goto fail;
2526 }
2527
2528 ret = init_srcu_struct(&fs_info->subvol_srcu);
2529 if (ret) {
2530 err = ret;
2531 goto fail;
2532 }
2533
2534 ret = setup_bdi(fs_info, &fs_info->bdi);
2535 if (ret) {
2536 err = ret;
2537 goto fail_srcu;
2538 }
2539
2540 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2541 if (ret) {
2542 err = ret;
2543 goto fail_bdi;
2544 }
2545 fs_info->dirty_metadata_batch = PAGE_SIZE *
2546 (1 + ilog2(nr_cpu_ids));
2547
2548 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2549 if (ret) {
2550 err = ret;
2551 goto fail_dirty_metadata_bytes;
2552 }
2553
2554 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2555 if (ret) {
2556 err = ret;
2557 goto fail_delalloc_bytes;
2558 }
2559
2560 fs_info->btree_inode = new_inode(sb);
2561 if (!fs_info->btree_inode) {
2562 err = -ENOMEM;
2563 goto fail_bio_counter;
2564 }
2565
2566 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2567
2568 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2569 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2570 INIT_LIST_HEAD(&fs_info->trans_list);
2571 INIT_LIST_HEAD(&fs_info->dead_roots);
2572 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2573 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2574 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2575 spin_lock_init(&fs_info->delalloc_root_lock);
2576 spin_lock_init(&fs_info->trans_lock);
2577 spin_lock_init(&fs_info->fs_roots_radix_lock);
2578 spin_lock_init(&fs_info->delayed_iput_lock);
2579 spin_lock_init(&fs_info->defrag_inodes_lock);
2580 spin_lock_init(&fs_info->free_chunk_lock);
2581 spin_lock_init(&fs_info->tree_mod_seq_lock);
2582 spin_lock_init(&fs_info->super_lock);
2583 spin_lock_init(&fs_info->qgroup_op_lock);
2584 spin_lock_init(&fs_info->buffer_lock);
2585 spin_lock_init(&fs_info->unused_bgs_lock);
2586 rwlock_init(&fs_info->tree_mod_log_lock);
2587 mutex_init(&fs_info->unused_bg_unpin_mutex);
2588 mutex_init(&fs_info->delete_unused_bgs_mutex);
2589 mutex_init(&fs_info->reloc_mutex);
2590 mutex_init(&fs_info->delalloc_root_mutex);
2591 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2592 seqlock_init(&fs_info->profiles_lock);
2593
2594 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2595 INIT_LIST_HEAD(&fs_info->space_info);
2596 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2597 INIT_LIST_HEAD(&fs_info->unused_bgs);
2598 btrfs_mapping_init(&fs_info->mapping_tree);
2599 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2600 BTRFS_BLOCK_RSV_GLOBAL);
2601 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2602 BTRFS_BLOCK_RSV_DELALLOC);
2603 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2604 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2605 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2606 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2607 BTRFS_BLOCK_RSV_DELOPS);
2608 atomic_set(&fs_info->nr_async_submits, 0);
2609 atomic_set(&fs_info->async_delalloc_pages, 0);
2610 atomic_set(&fs_info->async_submit_draining, 0);
2611 atomic_set(&fs_info->nr_async_bios, 0);
2612 atomic_set(&fs_info->defrag_running, 0);
2613 atomic_set(&fs_info->qgroup_op_seq, 0);
2614 atomic_set(&fs_info->reada_works_cnt, 0);
2615 atomic64_set(&fs_info->tree_mod_seq, 0);
2616 fs_info->sb = sb;
2617 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2618 fs_info->metadata_ratio = 0;
2619 fs_info->defrag_inodes = RB_ROOT;
2620 fs_info->free_chunk_space = 0;
2621 fs_info->tree_mod_log = RB_ROOT;
2622 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2623 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2624 /* readahead state */
2625 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2626 spin_lock_init(&fs_info->reada_lock);
2627
2628 fs_info->thread_pool_size = min_t(unsigned long,
2629 num_online_cpus() + 2, 8);
2630
2631 INIT_LIST_HEAD(&fs_info->ordered_roots);
2632 spin_lock_init(&fs_info->ordered_root_lock);
2633 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2634 GFP_KERNEL);
2635 if (!fs_info->delayed_root) {
2636 err = -ENOMEM;
2637 goto fail_iput;
2638 }
2639 btrfs_init_delayed_root(fs_info->delayed_root);
2640
2641 btrfs_init_scrub(fs_info);
2642#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2643 fs_info->check_integrity_print_mask = 0;
2644#endif
2645 btrfs_init_balance(fs_info);
2646 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2647
2648 sb->s_blocksize = 4096;
2649 sb->s_blocksize_bits = blksize_bits(4096);
2650 sb->s_bdi = &fs_info->bdi;
2651
2652 btrfs_init_btree_inode(fs_info, tree_root);
2653
2654 spin_lock_init(&fs_info->block_group_cache_lock);
2655 fs_info->block_group_cache_tree = RB_ROOT;
2656 fs_info->first_logical_byte = (u64)-1;
2657
2658 extent_io_tree_init(&fs_info->freed_extents[0],
2659 fs_info->btree_inode->i_mapping);
2660 extent_io_tree_init(&fs_info->freed_extents[1],
2661 fs_info->btree_inode->i_mapping);
2662 fs_info->pinned_extents = &fs_info->freed_extents[0];
2663 fs_info->do_barriers = 1;
2664
2665
2666 mutex_init(&fs_info->ordered_operations_mutex);
2667 mutex_init(&fs_info->tree_log_mutex);
2668 mutex_init(&fs_info->chunk_mutex);
2669 mutex_init(&fs_info->transaction_kthread_mutex);
2670 mutex_init(&fs_info->cleaner_mutex);
2671 mutex_init(&fs_info->volume_mutex);
2672 mutex_init(&fs_info->ro_block_group_mutex);
2673 init_rwsem(&fs_info->commit_root_sem);
2674 init_rwsem(&fs_info->cleanup_work_sem);
2675 init_rwsem(&fs_info->subvol_sem);
2676 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2677
2678 btrfs_init_dev_replace_locks(fs_info);
2679 btrfs_init_qgroup(fs_info);
2680
2681 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2682 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2683
2684 init_waitqueue_head(&fs_info->transaction_throttle);
2685 init_waitqueue_head(&fs_info->transaction_wait);
2686 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2687 init_waitqueue_head(&fs_info->async_submit_wait);
2688
2689 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2690
2691 ret = btrfs_alloc_stripe_hash_table(fs_info);
2692 if (ret) {
2693 err = ret;
2694 goto fail_alloc;
2695 }
2696
2697 __setup_root(4096, 4096, 4096, tree_root,
2698 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2699
2700 invalidate_bdev(fs_devices->latest_bdev);
2701
2702 /*
2703 * Read super block and check the signature bytes only
2704 */
2705 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2706 if (IS_ERR(bh)) {
2707 err = PTR_ERR(bh);
2708 goto fail_alloc;
2709 }
2710
2711 /*
2712 * We want to check superblock checksum, the type is stored inside.
2713 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2714 */
2715 if (btrfs_check_super_csum(bh->b_data)) {
2716 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2717 err = -EINVAL;
2718 brelse(bh);
2719 goto fail_alloc;
2720 }
2721
2722 /*
2723 * super_copy is zeroed at allocation time and we never touch the
2724 * following bytes up to INFO_SIZE, the checksum is calculated from
2725 * the whole block of INFO_SIZE
2726 */
2727 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2728 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2729 sizeof(*fs_info->super_for_commit));
2730 brelse(bh);
2731
2732 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2733
2734 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2735 if (ret) {
2736 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2737 err = -EINVAL;
2738 goto fail_alloc;
2739 }
2740
2741 disk_super = fs_info->super_copy;
2742 if (!btrfs_super_root(disk_super))
2743 goto fail_alloc;
2744
2745 /* check FS state, whether FS is broken. */
2746 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2747 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2748
2749 /*
2750 * run through our array of backup supers and setup
2751 * our ring pointer to the oldest one
2752 */
2753 generation = btrfs_super_generation(disk_super);
2754 find_oldest_super_backup(fs_info, generation);
2755
2756 /*
2757 * In the long term, we'll store the compression type in the super
2758 * block, and it'll be used for per file compression control.
2759 */
2760 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2761
2762 ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2763 if (ret) {
2764 err = ret;
2765 goto fail_alloc;
2766 }
2767
2768 features = btrfs_super_incompat_flags(disk_super) &
2769 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2770 if (features) {
2771 printk(KERN_ERR "BTRFS: couldn't mount because of "
2772 "unsupported optional features (%Lx).\n",
2773 features);
2774 err = -EINVAL;
2775 goto fail_alloc;
2776 }
2777
2778 features = btrfs_super_incompat_flags(disk_super);
2779 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2780 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2781 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2782
2783 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2784 printk(KERN_INFO "BTRFS: has skinny extents\n");
2785
2786 /*
2787 * flag our filesystem as having big metadata blocks if
2788 * they are bigger than the page size
2789 */
2790 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2791 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2792 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2793 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2794 }
2795
2796 nodesize = btrfs_super_nodesize(disk_super);
2797 sectorsize = btrfs_super_sectorsize(disk_super);
2798 stripesize = btrfs_super_stripesize(disk_super);
2799 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2800 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2801
2802 /*
2803 * mixed block groups end up with duplicate but slightly offset
2804 * extent buffers for the same range. It leads to corruptions
2805 */
2806 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2807 (sectorsize != nodesize)) {
2808 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2809 "are not allowed for mixed block groups on %s\n",
2810 sb->s_id);
2811 goto fail_alloc;
2812 }
2813
2814 /*
2815 * Needn't use the lock because there is no other task which will
2816 * update the flag.
2817 */
2818 btrfs_set_super_incompat_flags(disk_super, features);
2819
2820 features = btrfs_super_compat_ro_flags(disk_super) &
2821 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2822 if (!(sb->s_flags & MS_RDONLY) && features) {
2823 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2824 "unsupported option features (%Lx).\n",
2825 features);
2826 err = -EINVAL;
2827 goto fail_alloc;
2828 }
2829
2830 max_active = fs_info->thread_pool_size;
2831
2832 ret = btrfs_init_workqueues(fs_info, fs_devices);
2833 if (ret) {
2834 err = ret;
2835 goto fail_sb_buffer;
2836 }
2837
2838 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2839 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2840 SZ_4M / PAGE_SIZE);
2841
2842 tree_root->nodesize = nodesize;
2843 tree_root->sectorsize = sectorsize;
2844 tree_root->stripesize = stripesize;
2845
2846 sb->s_blocksize = sectorsize;
2847 sb->s_blocksize_bits = blksize_bits(sectorsize);
2848
2849 mutex_lock(&fs_info->chunk_mutex);
2850 ret = btrfs_read_sys_array(tree_root);
2851 mutex_unlock(&fs_info->chunk_mutex);
2852 if (ret) {
2853 printk(KERN_ERR "BTRFS: failed to read the system "
2854 "array on %s\n", sb->s_id);
2855 goto fail_sb_buffer;
2856 }
2857
2858 generation = btrfs_super_chunk_root_generation(disk_super);
2859
2860 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2861 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2862
2863 chunk_root->node = read_tree_block(chunk_root,
2864 btrfs_super_chunk_root(disk_super),
2865 generation);
2866 if (IS_ERR(chunk_root->node) ||
2867 !extent_buffer_uptodate(chunk_root->node)) {
2868 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2869 sb->s_id);
2870 if (!IS_ERR(chunk_root->node))
2871 free_extent_buffer(chunk_root->node);
2872 chunk_root->node = NULL;
2873 goto fail_tree_roots;
2874 }
2875 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2876 chunk_root->commit_root = btrfs_root_node(chunk_root);
2877
2878 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2879 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2880
2881 ret = btrfs_read_chunk_tree(chunk_root);
2882 if (ret) {
2883 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2884 sb->s_id);
2885 goto fail_tree_roots;
2886 }
2887
2888 /*
2889 * keep the device that is marked to be the target device for the
2890 * dev_replace procedure
2891 */
2892 btrfs_close_extra_devices(fs_devices, 0);
2893
2894 if (!fs_devices->latest_bdev) {
2895 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2896 sb->s_id);
2897 goto fail_tree_roots;
2898 }
2899
2900retry_root_backup:
2901 generation = btrfs_super_generation(disk_super);
2902
2903 tree_root->node = read_tree_block(tree_root,
2904 btrfs_super_root(disk_super),
2905 generation);
2906 if (IS_ERR(tree_root->node) ||
2907 !extent_buffer_uptodate(tree_root->node)) {
2908 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2909 sb->s_id);
2910 if (!IS_ERR(tree_root->node))
2911 free_extent_buffer(tree_root->node);
2912 tree_root->node = NULL;
2913 goto recovery_tree_root;
2914 }
2915
2916 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2917 tree_root->commit_root = btrfs_root_node(tree_root);
2918 btrfs_set_root_refs(&tree_root->root_item, 1);
2919
2920 mutex_lock(&tree_root->objectid_mutex);
2921 ret = btrfs_find_highest_objectid(tree_root,
2922 &tree_root->highest_objectid);
2923 if (ret) {
2924 mutex_unlock(&tree_root->objectid_mutex);
2925 goto recovery_tree_root;
2926 }
2927
2928 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2929
2930 mutex_unlock(&tree_root->objectid_mutex);
2931
2932 ret = btrfs_read_roots(fs_info, tree_root);
2933 if (ret)
2934 goto recovery_tree_root;
2935
2936 fs_info->generation = generation;
2937 fs_info->last_trans_committed = generation;
2938
2939 ret = btrfs_recover_balance(fs_info);
2940 if (ret) {
2941 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2942 goto fail_block_groups;
2943 }
2944
2945 ret = btrfs_init_dev_stats(fs_info);
2946 if (ret) {
2947 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2948 ret);
2949 goto fail_block_groups;
2950 }
2951
2952 ret = btrfs_init_dev_replace(fs_info);
2953 if (ret) {
2954 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2955 goto fail_block_groups;
2956 }
2957
2958 btrfs_close_extra_devices(fs_devices, 1);
2959
2960 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2961 if (ret) {
2962 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2963 goto fail_block_groups;
2964 }
2965
2966 ret = btrfs_sysfs_add_device(fs_devices);
2967 if (ret) {
2968 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2969 goto fail_fsdev_sysfs;
2970 }
2971
2972 ret = btrfs_sysfs_add_mounted(fs_info);
2973 if (ret) {
2974 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2975 goto fail_fsdev_sysfs;
2976 }
2977
2978 ret = btrfs_init_space_info(fs_info);
2979 if (ret) {
2980 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2981 goto fail_sysfs;
2982 }
2983
2984 ret = btrfs_read_block_groups(fs_info->extent_root);
2985 if (ret) {
2986 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2987 goto fail_sysfs;
2988 }
2989 fs_info->num_tolerated_disk_barrier_failures =
2990 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2991 if (fs_info->fs_devices->missing_devices >
2992 fs_info->num_tolerated_disk_barrier_failures &&
2993 !(sb->s_flags & MS_RDONLY)) {
2994 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2995 fs_info->fs_devices->missing_devices,
2996 fs_info->num_tolerated_disk_barrier_failures);
2997 goto fail_sysfs;
2998 }
2999
3000 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3001 "btrfs-cleaner");
3002 if (IS_ERR(fs_info->cleaner_kthread))
3003 goto fail_sysfs;
3004
3005 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3006 tree_root,
3007 "btrfs-transaction");
3008 if (IS_ERR(fs_info->transaction_kthread))
3009 goto fail_cleaner;
3010
3011 if (!btrfs_test_opt(tree_root, SSD) &&
3012 !btrfs_test_opt(tree_root, NOSSD) &&
3013 !fs_info->fs_devices->rotating) {
3014 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
3015 "mode\n");
3016 btrfs_set_opt(fs_info->mount_opt, SSD);
3017 }
3018
3019 /*
3020 * Mount does not set all options immediatelly, we can do it now and do
3021 * not have to wait for transaction commit
3022 */
3023 btrfs_apply_pending_changes(fs_info);
3024
3025#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3026 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3027 ret = btrfsic_mount(tree_root, fs_devices,
3028 btrfs_test_opt(tree_root,
3029 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3030 1 : 0,
3031 fs_info->check_integrity_print_mask);
3032 if (ret)
3033 printk(KERN_WARNING "BTRFS: failed to initialize"
3034 " integrity check module %s\n", sb->s_id);
3035 }
3036#endif
3037 ret = btrfs_read_qgroup_config(fs_info);
3038 if (ret)
3039 goto fail_trans_kthread;
3040
3041 /* do not make disk changes in broken FS or nologreplay is given */
3042 if (btrfs_super_log_root(disk_super) != 0 &&
3043 !btrfs_test_opt(tree_root, NOLOGREPLAY)) {
3044 ret = btrfs_replay_log(fs_info, fs_devices);
3045 if (ret) {
3046 err = ret;
3047 goto fail_qgroup;
3048 }
3049 }
3050
3051 ret = btrfs_find_orphan_roots(tree_root);
3052 if (ret)
3053 goto fail_qgroup;
3054
3055 if (!(sb->s_flags & MS_RDONLY)) {
3056 ret = btrfs_cleanup_fs_roots(fs_info);
3057 if (ret)
3058 goto fail_qgroup;
3059
3060 mutex_lock(&fs_info->cleaner_mutex);
3061 ret = btrfs_recover_relocation(tree_root);
3062 mutex_unlock(&fs_info->cleaner_mutex);
3063 if (ret < 0) {
3064 printk(KERN_WARNING
3065 "BTRFS: failed to recover relocation\n");
3066 err = -EINVAL;
3067 goto fail_qgroup;
3068 }
3069 }
3070
3071 location.objectid = BTRFS_FS_TREE_OBJECTID;
3072 location.type = BTRFS_ROOT_ITEM_KEY;
3073 location.offset = 0;
3074
3075 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3076 if (IS_ERR(fs_info->fs_root)) {
3077 err = PTR_ERR(fs_info->fs_root);
3078 goto fail_qgroup;
3079 }
3080
3081 if (sb->s_flags & MS_RDONLY)
3082 return 0;
3083
3084 if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
3085 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3086 pr_info("BTRFS: creating free space tree\n");
3087 ret = btrfs_create_free_space_tree(fs_info);
3088 if (ret) {
3089 pr_warn("BTRFS: failed to create free space tree %d\n",
3090 ret);
3091 close_ctree(tree_root);
3092 return ret;
3093 }
3094 }
3095
3096 down_read(&fs_info->cleanup_work_sem);
3097 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3098 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3099 up_read(&fs_info->cleanup_work_sem);
3100 close_ctree(tree_root);
3101 return ret;
3102 }
3103 up_read(&fs_info->cleanup_work_sem);
3104
3105 ret = btrfs_resume_balance_async(fs_info);
3106 if (ret) {
3107 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3108 close_ctree(tree_root);
3109 return ret;
3110 }
3111
3112 ret = btrfs_resume_dev_replace_async(fs_info);
3113 if (ret) {
3114 pr_warn("BTRFS: failed to resume dev_replace\n");
3115 close_ctree(tree_root);
3116 return ret;
3117 }
3118
3119 btrfs_qgroup_rescan_resume(fs_info);
3120
3121 if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
3122 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3123 pr_info("BTRFS: clearing free space tree\n");
3124 ret = btrfs_clear_free_space_tree(fs_info);
3125 if (ret) {
3126 pr_warn("BTRFS: failed to clear free space tree %d\n",
3127 ret);
3128 close_ctree(tree_root);
3129 return ret;
3130 }
3131 }
3132
3133 if (!fs_info->uuid_root) {
3134 pr_info("BTRFS: creating UUID tree\n");
3135 ret = btrfs_create_uuid_tree(fs_info);
3136 if (ret) {
3137 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3138 ret);
3139 close_ctree(tree_root);
3140 return ret;
3141 }
3142 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3143 fs_info->generation !=
3144 btrfs_super_uuid_tree_generation(disk_super)) {
3145 pr_info("BTRFS: checking UUID tree\n");
3146 ret = btrfs_check_uuid_tree(fs_info);
3147 if (ret) {
3148 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3149 ret);
3150 close_ctree(tree_root);
3151 return ret;
3152 }
3153 } else {
3154 fs_info->update_uuid_tree_gen = 1;
3155 }
3156
3157 fs_info->open = 1;
3158
3159 /*
3160 * backuproot only affect mount behavior, and if open_ctree succeeded,
3161 * no need to keep the flag
3162 */
3163 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3164
3165 return 0;
3166
3167fail_qgroup:
3168 btrfs_free_qgroup_config(fs_info);
3169fail_trans_kthread:
3170 kthread_stop(fs_info->transaction_kthread);
3171 btrfs_cleanup_transaction(fs_info->tree_root);
3172 btrfs_free_fs_roots(fs_info);
3173fail_cleaner:
3174 kthread_stop(fs_info->cleaner_kthread);
3175
3176 /*
3177 * make sure we're done with the btree inode before we stop our
3178 * kthreads
3179 */
3180 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3181
3182fail_sysfs:
3183 btrfs_sysfs_remove_mounted(fs_info);
3184
3185fail_fsdev_sysfs:
3186 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3187
3188fail_block_groups:
3189 btrfs_put_block_group_cache(fs_info);
3190 btrfs_free_block_groups(fs_info);
3191
3192fail_tree_roots:
3193 free_root_pointers(fs_info, 1);
3194 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3195
3196fail_sb_buffer:
3197 btrfs_stop_all_workers(fs_info);
3198fail_alloc:
3199fail_iput:
3200 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3201
3202 iput(fs_info->btree_inode);
3203fail_bio_counter:
3204 percpu_counter_destroy(&fs_info->bio_counter);
3205fail_delalloc_bytes:
3206 percpu_counter_destroy(&fs_info->delalloc_bytes);
3207fail_dirty_metadata_bytes:
3208 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3209fail_bdi:
3210 bdi_destroy(&fs_info->bdi);
3211fail_srcu:
3212 cleanup_srcu_struct(&fs_info->subvol_srcu);
3213fail:
3214 btrfs_free_stripe_hash_table(fs_info);
3215 btrfs_close_devices(fs_info->fs_devices);
3216 return err;
3217
3218recovery_tree_root:
3219 if (!btrfs_test_opt(tree_root, USEBACKUPROOT))
3220 goto fail_tree_roots;
3221
3222 free_root_pointers(fs_info, 0);
3223
3224 /* don't use the log in recovery mode, it won't be valid */
3225 btrfs_set_super_log_root(disk_super, 0);
3226
3227 /* we can't trust the free space cache either */
3228 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3229
3230 ret = next_root_backup(fs_info, fs_info->super_copy,
3231 &num_backups_tried, &backup_index);
3232 if (ret == -1)
3233 goto fail_block_groups;
3234 goto retry_root_backup;
3235}
3236
3237static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3238{
3239 if (uptodate) {
3240 set_buffer_uptodate(bh);
3241 } else {
3242 struct btrfs_device *device = (struct btrfs_device *)
3243 bh->b_private;
3244
3245 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3246 "lost page write due to IO error on %s",
3247 rcu_str_deref(device->name));
3248 /* note, we dont' set_buffer_write_io_error because we have
3249 * our own ways of dealing with the IO errors
3250 */
3251 clear_buffer_uptodate(bh);
3252 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3253 }
3254 unlock_buffer(bh);
3255 put_bh(bh);
3256}
3257
3258int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3259 struct buffer_head **bh_ret)
3260{
3261 struct buffer_head *bh;
3262 struct btrfs_super_block *super;
3263 u64 bytenr;
3264
3265 bytenr = btrfs_sb_offset(copy_num);
3266 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3267 return -EINVAL;
3268
3269 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3270 /*
3271 * If we fail to read from the underlying devices, as of now
3272 * the best option we have is to mark it EIO.
3273 */
3274 if (!bh)
3275 return -EIO;
3276
3277 super = (struct btrfs_super_block *)bh->b_data;
3278 if (btrfs_super_bytenr(super) != bytenr ||
3279 btrfs_super_magic(super) != BTRFS_MAGIC) {
3280 brelse(bh);
3281 return -EINVAL;
3282 }
3283
3284 *bh_ret = bh;
3285 return 0;
3286}
3287
3288
3289struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3290{
3291 struct buffer_head *bh;
3292 struct buffer_head *latest = NULL;
3293 struct btrfs_super_block *super;
3294 int i;
3295 u64 transid = 0;
3296 int ret = -EINVAL;
3297
3298 /* we would like to check all the supers, but that would make
3299 * a btrfs mount succeed after a mkfs from a different FS.
3300 * So, we need to add a special mount option to scan for
3301 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3302 */
3303 for (i = 0; i < 1; i++) {
3304 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3305 if (ret)
3306 continue;
3307
3308 super = (struct btrfs_super_block *)bh->b_data;
3309
3310 if (!latest || btrfs_super_generation(super) > transid) {
3311 brelse(latest);
3312 latest = bh;
3313 transid = btrfs_super_generation(super);
3314 } else {
3315 brelse(bh);
3316 }
3317 }
3318
3319 if (!latest)
3320 return ERR_PTR(ret);
3321
3322 return latest;
3323}
3324
3325/*
3326 * this should be called twice, once with wait == 0 and
3327 * once with wait == 1. When wait == 0 is done, all the buffer heads
3328 * we write are pinned.
3329 *
3330 * They are released when wait == 1 is done.
3331 * max_mirrors must be the same for both runs, and it indicates how
3332 * many supers on this one device should be written.
3333 *
3334 * max_mirrors == 0 means to write them all.
3335 */
3336static int write_dev_supers(struct btrfs_device *device,
3337 struct btrfs_super_block *sb,
3338 int do_barriers, int wait, int max_mirrors)
3339{
3340 struct buffer_head *bh;
3341 int i;
3342 int ret;
3343 int errors = 0;
3344 u32 crc;
3345 u64 bytenr;
3346
3347 if (max_mirrors == 0)
3348 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3349
3350 for (i = 0; i < max_mirrors; i++) {
3351 bytenr = btrfs_sb_offset(i);
3352 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3353 device->commit_total_bytes)
3354 break;
3355
3356 if (wait) {
3357 bh = __find_get_block(device->bdev, bytenr / 4096,
3358 BTRFS_SUPER_INFO_SIZE);
3359 if (!bh) {
3360 errors++;
3361 continue;
3362 }
3363 wait_on_buffer(bh);
3364 if (!buffer_uptodate(bh))
3365 errors++;
3366
3367 /* drop our reference */
3368 brelse(bh);
3369
3370 /* drop the reference from the wait == 0 run */
3371 brelse(bh);
3372 continue;
3373 } else {
3374 btrfs_set_super_bytenr(sb, bytenr);
3375
3376 crc = ~(u32)0;
3377 crc = btrfs_csum_data((char *)sb +
3378 BTRFS_CSUM_SIZE, crc,
3379 BTRFS_SUPER_INFO_SIZE -
3380 BTRFS_CSUM_SIZE);
3381 btrfs_csum_final(crc, sb->csum);
3382
3383 /*
3384 * one reference for us, and we leave it for the
3385 * caller
3386 */
3387 bh = __getblk(device->bdev, bytenr / 4096,
3388 BTRFS_SUPER_INFO_SIZE);
3389 if (!bh) {
3390 btrfs_err(device->dev_root->fs_info,
3391 "couldn't get super buffer head for bytenr %llu",
3392 bytenr);
3393 errors++;
3394 continue;
3395 }
3396
3397 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3398
3399 /* one reference for submit_bh */
3400 get_bh(bh);
3401
3402 set_buffer_uptodate(bh);
3403 lock_buffer(bh);
3404 bh->b_end_io = btrfs_end_buffer_write_sync;
3405 bh->b_private = device;
3406 }
3407
3408 /*
3409 * we fua the first super. The others we allow
3410 * to go down lazy.
3411 */
3412 if (i == 0)
3413 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3414 else
3415 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3416 if (ret)
3417 errors++;
3418 }
3419 return errors < i ? 0 : -1;
3420}
3421
3422/*
3423 * endio for the write_dev_flush, this will wake anyone waiting
3424 * for the barrier when it is done
3425 */
3426static void btrfs_end_empty_barrier(struct bio *bio)
3427{
3428 if (bio->bi_private)
3429 complete(bio->bi_private);
3430 bio_put(bio);
3431}
3432
3433/*
3434 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3435 * sent down. With wait == 1, it waits for the previous flush.
3436 *
3437 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3438 * capable
3439 */
3440static int write_dev_flush(struct btrfs_device *device, int wait)
3441{
3442 struct bio *bio;
3443 int ret = 0;
3444
3445 if (device->nobarriers)
3446 return 0;
3447
3448 if (wait) {
3449 bio = device->flush_bio;
3450 if (!bio)
3451 return 0;
3452
3453 wait_for_completion(&device->flush_wait);
3454
3455 if (bio->bi_error) {
3456 ret = bio->bi_error;
3457 btrfs_dev_stat_inc_and_print(device,
3458 BTRFS_DEV_STAT_FLUSH_ERRS);
3459 }
3460
3461 /* drop the reference from the wait == 0 run */
3462 bio_put(bio);
3463 device->flush_bio = NULL;
3464
3465 return ret;
3466 }
3467
3468 /*
3469 * one reference for us, and we leave it for the
3470 * caller
3471 */
3472 device->flush_bio = NULL;
3473 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3474 if (!bio)
3475 return -ENOMEM;
3476
3477 bio->bi_end_io = btrfs_end_empty_barrier;
3478 bio->bi_bdev = device->bdev;
3479 init_completion(&device->flush_wait);
3480 bio->bi_private = &device->flush_wait;
3481 device->flush_bio = bio;
3482
3483 bio_get(bio);
3484 btrfsic_submit_bio(WRITE_FLUSH, bio);
3485
3486 return 0;
3487}
3488
3489/*
3490 * send an empty flush down to each device in parallel,
3491 * then wait for them
3492 */
3493static int barrier_all_devices(struct btrfs_fs_info *info)
3494{
3495 struct list_head *head;
3496 struct btrfs_device *dev;
3497 int errors_send = 0;
3498 int errors_wait = 0;
3499 int ret;
3500
3501 /* send down all the barriers */
3502 head = &info->fs_devices->devices;
3503 list_for_each_entry_rcu(dev, head, dev_list) {
3504 if (dev->missing)
3505 continue;
3506 if (!dev->bdev) {
3507 errors_send++;
3508 continue;
3509 }
3510 if (!dev->in_fs_metadata || !dev->writeable)
3511 continue;
3512
3513 ret = write_dev_flush(dev, 0);
3514 if (ret)
3515 errors_send++;
3516 }
3517
3518 /* wait for all the barriers */
3519 list_for_each_entry_rcu(dev, head, dev_list) {
3520 if (dev->missing)
3521 continue;
3522 if (!dev->bdev) {
3523 errors_wait++;
3524 continue;
3525 }
3526 if (!dev->in_fs_metadata || !dev->writeable)
3527 continue;
3528
3529 ret = write_dev_flush(dev, 1);
3530 if (ret)
3531 errors_wait++;
3532 }
3533 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3534 errors_wait > info->num_tolerated_disk_barrier_failures)
3535 return -EIO;
3536 return 0;
3537}
3538
3539int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3540{
3541 int raid_type;
3542 int min_tolerated = INT_MAX;
3543
3544 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3545 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3546 min_tolerated = min(min_tolerated,
3547 btrfs_raid_array[BTRFS_RAID_SINGLE].
3548 tolerated_failures);
3549
3550 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3551 if (raid_type == BTRFS_RAID_SINGLE)
3552 continue;
3553 if (!(flags & btrfs_raid_group[raid_type]))
3554 continue;
3555 min_tolerated = min(min_tolerated,
3556 btrfs_raid_array[raid_type].
3557 tolerated_failures);
3558 }
3559
3560 if (min_tolerated == INT_MAX) {
3561 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3562 min_tolerated = 0;
3563 }
3564
3565 return min_tolerated;
3566}
3567
3568int btrfs_calc_num_tolerated_disk_barrier_failures(
3569 struct btrfs_fs_info *fs_info)
3570{
3571 struct btrfs_ioctl_space_info space;
3572 struct btrfs_space_info *sinfo;
3573 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3574 BTRFS_BLOCK_GROUP_SYSTEM,
3575 BTRFS_BLOCK_GROUP_METADATA,
3576 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3577 int i;
3578 int c;
3579 int num_tolerated_disk_barrier_failures =
3580 (int)fs_info->fs_devices->num_devices;
3581
3582 for (i = 0; i < ARRAY_SIZE(types); i++) {
3583 struct btrfs_space_info *tmp;
3584
3585 sinfo = NULL;
3586 rcu_read_lock();
3587 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3588 if (tmp->flags == types[i]) {
3589 sinfo = tmp;
3590 break;
3591 }
3592 }
3593 rcu_read_unlock();
3594
3595 if (!sinfo)
3596 continue;
3597
3598 down_read(&sinfo->groups_sem);
3599 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3600 u64 flags;
3601
3602 if (list_empty(&sinfo->block_groups[c]))
3603 continue;
3604
3605 btrfs_get_block_group_info(&sinfo->block_groups[c],
3606 &space);
3607 if (space.total_bytes == 0 || space.used_bytes == 0)
3608 continue;
3609 flags = space.flags;
3610
3611 num_tolerated_disk_barrier_failures = min(
3612 num_tolerated_disk_barrier_failures,
3613 btrfs_get_num_tolerated_disk_barrier_failures(
3614 flags));
3615 }
3616 up_read(&sinfo->groups_sem);
3617 }
3618
3619 return num_tolerated_disk_barrier_failures;
3620}
3621
3622static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3623{
3624 struct list_head *head;
3625 struct btrfs_device *dev;
3626 struct btrfs_super_block *sb;
3627 struct btrfs_dev_item *dev_item;
3628 int ret;
3629 int do_barriers;
3630 int max_errors;
3631 int total_errors = 0;
3632 u64 flags;
3633
3634 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3635 backup_super_roots(root->fs_info);
3636
3637 sb = root->fs_info->super_for_commit;
3638 dev_item = &sb->dev_item;
3639
3640 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3641 head = &root->fs_info->fs_devices->devices;
3642 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3643
3644 if (do_barriers) {
3645 ret = barrier_all_devices(root->fs_info);
3646 if (ret) {
3647 mutex_unlock(
3648 &root->fs_info->fs_devices->device_list_mutex);
3649 btrfs_std_error(root->fs_info, ret,
3650 "errors while submitting device barriers.");
3651 return ret;
3652 }
3653 }
3654
3655 list_for_each_entry_rcu(dev, head, dev_list) {
3656 if (!dev->bdev) {
3657 total_errors++;
3658 continue;
3659 }
3660 if (!dev->in_fs_metadata || !dev->writeable)
3661 continue;
3662
3663 btrfs_set_stack_device_generation(dev_item, 0);
3664 btrfs_set_stack_device_type(dev_item, dev->type);
3665 btrfs_set_stack_device_id(dev_item, dev->devid);
3666 btrfs_set_stack_device_total_bytes(dev_item,
3667 dev->commit_total_bytes);
3668 btrfs_set_stack_device_bytes_used(dev_item,
3669 dev->commit_bytes_used);
3670 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3671 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3672 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3673 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3674 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3675
3676 flags = btrfs_super_flags(sb);
3677 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3678
3679 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3680 if (ret)
3681 total_errors++;
3682 }
3683 if (total_errors > max_errors) {
3684 btrfs_err(root->fs_info, "%d errors while writing supers",
3685 total_errors);
3686 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3687
3688 /* FUA is masked off if unsupported and can't be the reason */
3689 btrfs_std_error(root->fs_info, -EIO,
3690 "%d errors while writing supers", total_errors);
3691 return -EIO;
3692 }
3693
3694 total_errors = 0;
3695 list_for_each_entry_rcu(dev, head, dev_list) {
3696 if (!dev->bdev)
3697 continue;
3698 if (!dev->in_fs_metadata || !dev->writeable)
3699 continue;
3700
3701 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3702 if (ret)
3703 total_errors++;
3704 }
3705 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3706 if (total_errors > max_errors) {
3707 btrfs_std_error(root->fs_info, -EIO,
3708 "%d errors while writing supers", total_errors);
3709 return -EIO;
3710 }
3711 return 0;
3712}
3713
3714int write_ctree_super(struct btrfs_trans_handle *trans,
3715 struct btrfs_root *root, int max_mirrors)
3716{
3717 return write_all_supers(root, max_mirrors);
3718}
3719
3720/* Drop a fs root from the radix tree and free it. */
3721void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3722 struct btrfs_root *root)
3723{
3724 spin_lock(&fs_info->fs_roots_radix_lock);
3725 radix_tree_delete(&fs_info->fs_roots_radix,
3726 (unsigned long)root->root_key.objectid);
3727 spin_unlock(&fs_info->fs_roots_radix_lock);
3728
3729 if (btrfs_root_refs(&root->root_item) == 0)
3730 synchronize_srcu(&fs_info->subvol_srcu);
3731
3732 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3733 btrfs_free_log(NULL, root);
3734
3735 if (root->free_ino_pinned)
3736 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3737 if (root->free_ino_ctl)
3738 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3739 free_fs_root(root);
3740}
3741
3742static void free_fs_root(struct btrfs_root *root)
3743{
3744 iput(root->ino_cache_inode);
3745 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3746 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3747 root->orphan_block_rsv = NULL;
3748 if (root->anon_dev)
3749 free_anon_bdev(root->anon_dev);
3750 if (root->subv_writers)
3751 btrfs_free_subvolume_writers(root->subv_writers);
3752 free_extent_buffer(root->node);
3753 free_extent_buffer(root->commit_root);
3754 kfree(root->free_ino_ctl);
3755 kfree(root->free_ino_pinned);
3756 kfree(root->name);
3757 btrfs_put_fs_root(root);
3758}
3759
3760void btrfs_free_fs_root(struct btrfs_root *root)
3761{
3762 free_fs_root(root);
3763}
3764
3765int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3766{
3767 u64 root_objectid = 0;
3768 struct btrfs_root *gang[8];
3769 int i = 0;
3770 int err = 0;
3771 unsigned int ret = 0;
3772 int index;
3773
3774 while (1) {
3775 index = srcu_read_lock(&fs_info->subvol_srcu);
3776 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3777 (void **)gang, root_objectid,
3778 ARRAY_SIZE(gang));
3779 if (!ret) {
3780 srcu_read_unlock(&fs_info->subvol_srcu, index);
3781 break;
3782 }
3783 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3784
3785 for (i = 0; i < ret; i++) {
3786 /* Avoid to grab roots in dead_roots */
3787 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3788 gang[i] = NULL;
3789 continue;
3790 }
3791 /* grab all the search result for later use */
3792 gang[i] = btrfs_grab_fs_root(gang[i]);
3793 }
3794 srcu_read_unlock(&fs_info->subvol_srcu, index);
3795
3796 for (i = 0; i < ret; i++) {
3797 if (!gang[i])
3798 continue;
3799 root_objectid = gang[i]->root_key.objectid;
3800 err = btrfs_orphan_cleanup(gang[i]);
3801 if (err)
3802 break;
3803 btrfs_put_fs_root(gang[i]);
3804 }
3805 root_objectid++;
3806 }
3807
3808 /* release the uncleaned roots due to error */
3809 for (; i < ret; i++) {
3810 if (gang[i])
3811 btrfs_put_fs_root(gang[i]);
3812 }
3813 return err;
3814}
3815
3816int btrfs_commit_super(struct btrfs_root *root)
3817{
3818 struct btrfs_trans_handle *trans;
3819
3820 mutex_lock(&root->fs_info->cleaner_mutex);
3821 btrfs_run_delayed_iputs(root);
3822 mutex_unlock(&root->fs_info->cleaner_mutex);
3823 wake_up_process(root->fs_info->cleaner_kthread);
3824
3825 /* wait until ongoing cleanup work done */
3826 down_write(&root->fs_info->cleanup_work_sem);
3827 up_write(&root->fs_info->cleanup_work_sem);
3828
3829 trans = btrfs_join_transaction(root);
3830 if (IS_ERR(trans))
3831 return PTR_ERR(trans);
3832 return btrfs_commit_transaction(trans, root);
3833}
3834
3835void close_ctree(struct btrfs_root *root)
3836{
3837 struct btrfs_fs_info *fs_info = root->fs_info;
3838 int ret;
3839
3840 fs_info->closing = 1;
3841 smp_mb();
3842
3843 /* wait for the qgroup rescan worker to stop */
3844 btrfs_qgroup_wait_for_completion(fs_info);
3845
3846 /* wait for the uuid_scan task to finish */
3847 down(&fs_info->uuid_tree_rescan_sem);
3848 /* avoid complains from lockdep et al., set sem back to initial state */
3849 up(&fs_info->uuid_tree_rescan_sem);
3850
3851 /* pause restriper - we want to resume on mount */
3852 btrfs_pause_balance(fs_info);
3853
3854 btrfs_dev_replace_suspend_for_unmount(fs_info);
3855
3856 btrfs_scrub_cancel(fs_info);
3857
3858 /* wait for any defraggers to finish */
3859 wait_event(fs_info->transaction_wait,
3860 (atomic_read(&fs_info->defrag_running) == 0));
3861
3862 /* clear out the rbtree of defraggable inodes */
3863 btrfs_cleanup_defrag_inodes(fs_info);
3864
3865 cancel_work_sync(&fs_info->async_reclaim_work);
3866
3867 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3868 /*
3869 * If the cleaner thread is stopped and there are
3870 * block groups queued for removal, the deletion will be
3871 * skipped when we quit the cleaner thread.
3872 */
3873 btrfs_delete_unused_bgs(root->fs_info);
3874
3875 ret = btrfs_commit_super(root);
3876 if (ret)
3877 btrfs_err(fs_info, "commit super ret %d", ret);
3878 }
3879
3880 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3881 btrfs_error_commit_super(root);
3882
3883 kthread_stop(fs_info->transaction_kthread);
3884 kthread_stop(fs_info->cleaner_kthread);
3885
3886 fs_info->closing = 2;
3887 smp_mb();
3888
3889 btrfs_free_qgroup_config(fs_info);
3890
3891 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3892 btrfs_info(fs_info, "at unmount delalloc count %lld",
3893 percpu_counter_sum(&fs_info->delalloc_bytes));
3894 }
3895
3896 btrfs_sysfs_remove_mounted(fs_info);
3897 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3898
3899 btrfs_free_fs_roots(fs_info);
3900
3901 btrfs_put_block_group_cache(fs_info);
3902
3903 btrfs_free_block_groups(fs_info);
3904
3905 /*
3906 * we must make sure there is not any read request to
3907 * submit after we stopping all workers.
3908 */
3909 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3910 btrfs_stop_all_workers(fs_info);
3911
3912 fs_info->open = 0;
3913 free_root_pointers(fs_info, 1);
3914
3915 iput(fs_info->btree_inode);
3916
3917#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3918 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3919 btrfsic_unmount(root, fs_info->fs_devices);
3920#endif
3921
3922 btrfs_close_devices(fs_info->fs_devices);
3923 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3924
3925 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3926 percpu_counter_destroy(&fs_info->delalloc_bytes);
3927 percpu_counter_destroy(&fs_info->bio_counter);
3928 bdi_destroy(&fs_info->bdi);
3929 cleanup_srcu_struct(&fs_info->subvol_srcu);
3930
3931 btrfs_free_stripe_hash_table(fs_info);
3932
3933 __btrfs_free_block_rsv(root->orphan_block_rsv);
3934 root->orphan_block_rsv = NULL;
3935
3936 lock_chunks(root);
3937 while (!list_empty(&fs_info->pinned_chunks)) {
3938 struct extent_map *em;
3939
3940 em = list_first_entry(&fs_info->pinned_chunks,
3941 struct extent_map, list);
3942 list_del_init(&em->list);
3943 free_extent_map(em);
3944 }
3945 unlock_chunks(root);
3946}
3947
3948int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3949 int atomic)
3950{
3951 int ret;
3952 struct inode *btree_inode = buf->pages[0]->mapping->host;
3953
3954 ret = extent_buffer_uptodate(buf);
3955 if (!ret)
3956 return ret;
3957
3958 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3959 parent_transid, atomic);
3960 if (ret == -EAGAIN)
3961 return ret;
3962 return !ret;
3963}
3964
3965void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3966{
3967 struct btrfs_root *root;
3968 u64 transid = btrfs_header_generation(buf);
3969 int was_dirty;
3970
3971#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3972 /*
3973 * This is a fast path so only do this check if we have sanity tests
3974 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3975 * outside of the sanity tests.
3976 */
3977 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3978 return;
3979#endif
3980 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3981 btrfs_assert_tree_locked(buf);
3982 if (transid != root->fs_info->generation)
3983 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3984 "found %llu running %llu\n",
3985 buf->start, transid, root->fs_info->generation);
3986 was_dirty = set_extent_buffer_dirty(buf);
3987 if (!was_dirty)
3988 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3989 buf->len,
3990 root->fs_info->dirty_metadata_batch);
3991#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3992 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3993 btrfs_print_leaf(root, buf);
3994 ASSERT(0);
3995 }
3996#endif
3997}
3998
3999static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4000 int flush_delayed)
4001{
4002 /*
4003 * looks as though older kernels can get into trouble with
4004 * this code, they end up stuck in balance_dirty_pages forever
4005 */
4006 int ret;
4007
4008 if (current->flags & PF_MEMALLOC)
4009 return;
4010
4011 if (flush_delayed)
4012 btrfs_balance_delayed_items(root);
4013
4014 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4015 BTRFS_DIRTY_METADATA_THRESH);
4016 if (ret > 0) {
4017 balance_dirty_pages_ratelimited(
4018 root->fs_info->btree_inode->i_mapping);
4019 }
4020}
4021
4022void btrfs_btree_balance_dirty(struct btrfs_root *root)
4023{
4024 __btrfs_btree_balance_dirty(root, 1);
4025}
4026
4027void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4028{
4029 __btrfs_btree_balance_dirty(root, 0);
4030}
4031
4032int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4033{
4034 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4035 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4036}
4037
4038static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4039 int read_only)
4040{
4041 struct btrfs_super_block *sb = fs_info->super_copy;
4042 u64 nodesize = btrfs_super_nodesize(sb);
4043 u64 sectorsize = btrfs_super_sectorsize(sb);
4044 int ret = 0;
4045
4046 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4047 printk(KERN_ERR "BTRFS: no valid FS found\n");
4048 ret = -EINVAL;
4049 }
4050 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4051 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4052 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4053 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4054 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4055 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4056 ret = -EINVAL;
4057 }
4058 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4059 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4060 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4061 ret = -EINVAL;
4062 }
4063 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4064 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4065 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4066 ret = -EINVAL;
4067 }
4068
4069 /*
4070 * Check sectorsize and nodesize first, other check will need it.
4071 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4072 */
4073 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4074 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4075 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4076 ret = -EINVAL;
4077 }
4078 /* Only PAGE SIZE is supported yet */
4079 if (sectorsize != PAGE_SIZE) {
4080 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4081 sectorsize, PAGE_SIZE);
4082 ret = -EINVAL;
4083 }
4084 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4085 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4086 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4087 ret = -EINVAL;
4088 }
4089 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4090 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4091 le32_to_cpu(sb->__unused_leafsize),
4092 nodesize);
4093 ret = -EINVAL;
4094 }
4095
4096 /* Root alignment check */
4097 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4098 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4099 btrfs_super_root(sb));
4100 ret = -EINVAL;
4101 }
4102 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4103 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4104 btrfs_super_chunk_root(sb));
4105 ret = -EINVAL;
4106 }
4107 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4108 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4109 btrfs_super_log_root(sb));
4110 ret = -EINVAL;
4111 }
4112
4113 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4114 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4115 fs_info->fsid, sb->dev_item.fsid);
4116 ret = -EINVAL;
4117 }
4118
4119 /*
4120 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4121 * done later
4122 */
4123 if (btrfs_super_num_devices(sb) > (1UL << 31))
4124 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4125 btrfs_super_num_devices(sb));
4126 if (btrfs_super_num_devices(sb) == 0) {
4127 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4128 ret = -EINVAL;
4129 }
4130
4131 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4132 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4133 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4134 ret = -EINVAL;
4135 }
4136
4137 /*
4138 * Obvious sys_chunk_array corruptions, it must hold at least one key
4139 * and one chunk
4140 */
4141 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4142 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4143 btrfs_super_sys_array_size(sb),
4144 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4145 ret = -EINVAL;
4146 }
4147 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4148 + sizeof(struct btrfs_chunk)) {
4149 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4150 btrfs_super_sys_array_size(sb),
4151 sizeof(struct btrfs_disk_key)
4152 + sizeof(struct btrfs_chunk));
4153 ret = -EINVAL;
4154 }
4155
4156 /*
4157 * The generation is a global counter, we'll trust it more than the others
4158 * but it's still possible that it's the one that's wrong.
4159 */
4160 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4161 printk(KERN_WARNING
4162 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4163 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4164 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4165 && btrfs_super_cache_generation(sb) != (u64)-1)
4166 printk(KERN_WARNING
4167 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4168 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4169
4170 return ret;
4171}
4172
4173static void btrfs_error_commit_super(struct btrfs_root *root)
4174{
4175 mutex_lock(&root->fs_info->cleaner_mutex);
4176 btrfs_run_delayed_iputs(root);
4177 mutex_unlock(&root->fs_info->cleaner_mutex);
4178
4179 down_write(&root->fs_info->cleanup_work_sem);
4180 up_write(&root->fs_info->cleanup_work_sem);
4181
4182 /* cleanup FS via transaction */
4183 btrfs_cleanup_transaction(root);
4184}
4185
4186static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4187{
4188 struct btrfs_ordered_extent *ordered;
4189
4190 spin_lock(&root->ordered_extent_lock);
4191 /*
4192 * This will just short circuit the ordered completion stuff which will
4193 * make sure the ordered extent gets properly cleaned up.
4194 */
4195 list_for_each_entry(ordered, &root->ordered_extents,
4196 root_extent_list)
4197 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4198 spin_unlock(&root->ordered_extent_lock);
4199}
4200
4201static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4202{
4203 struct btrfs_root *root;
4204 struct list_head splice;
4205
4206 INIT_LIST_HEAD(&splice);
4207
4208 spin_lock(&fs_info->ordered_root_lock);
4209 list_splice_init(&fs_info->ordered_roots, &splice);
4210 while (!list_empty(&splice)) {
4211 root = list_first_entry(&splice, struct btrfs_root,
4212 ordered_root);
4213 list_move_tail(&root->ordered_root,
4214 &fs_info->ordered_roots);
4215
4216 spin_unlock(&fs_info->ordered_root_lock);
4217 btrfs_destroy_ordered_extents(root);
4218
4219 cond_resched();
4220 spin_lock(&fs_info->ordered_root_lock);
4221 }
4222 spin_unlock(&fs_info->ordered_root_lock);
4223}
4224
4225static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4226 struct btrfs_root *root)
4227{
4228 struct rb_node *node;
4229 struct btrfs_delayed_ref_root *delayed_refs;
4230 struct btrfs_delayed_ref_node *ref;
4231 int ret = 0;
4232
4233 delayed_refs = &trans->delayed_refs;
4234
4235 spin_lock(&delayed_refs->lock);
4236 if (atomic_read(&delayed_refs->num_entries) == 0) {
4237 spin_unlock(&delayed_refs->lock);
4238 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4239 return ret;
4240 }
4241
4242 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4243 struct btrfs_delayed_ref_head *head;
4244 struct btrfs_delayed_ref_node *tmp;
4245 bool pin_bytes = false;
4246
4247 head = rb_entry(node, struct btrfs_delayed_ref_head,
4248 href_node);
4249 if (!mutex_trylock(&head->mutex)) {
4250 atomic_inc(&head->node.refs);
4251 spin_unlock(&delayed_refs->lock);
4252
4253 mutex_lock(&head->mutex);
4254 mutex_unlock(&head->mutex);
4255 btrfs_put_delayed_ref(&head->node);
4256 spin_lock(&delayed_refs->lock);
4257 continue;
4258 }
4259 spin_lock(&head->lock);
4260 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4261 list) {
4262 ref->in_tree = 0;
4263 list_del(&ref->list);
4264 atomic_dec(&delayed_refs->num_entries);
4265 btrfs_put_delayed_ref(ref);
4266 }
4267 if (head->must_insert_reserved)
4268 pin_bytes = true;
4269 btrfs_free_delayed_extent_op(head->extent_op);
4270 delayed_refs->num_heads--;
4271 if (head->processing == 0)
4272 delayed_refs->num_heads_ready--;
4273 atomic_dec(&delayed_refs->num_entries);
4274 head->node.in_tree = 0;
4275 rb_erase(&head->href_node, &delayed_refs->href_root);
4276 spin_unlock(&head->lock);
4277 spin_unlock(&delayed_refs->lock);
4278 mutex_unlock(&head->mutex);
4279
4280 if (pin_bytes)
4281 btrfs_pin_extent(root, head->node.bytenr,
4282 head->node.num_bytes, 1);
4283 btrfs_put_delayed_ref(&head->node);
4284 cond_resched();
4285 spin_lock(&delayed_refs->lock);
4286 }
4287
4288 spin_unlock(&delayed_refs->lock);
4289
4290 return ret;
4291}
4292
4293static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4294{
4295 struct btrfs_inode *btrfs_inode;
4296 struct list_head splice;
4297
4298 INIT_LIST_HEAD(&splice);
4299
4300 spin_lock(&root->delalloc_lock);
4301 list_splice_init(&root->delalloc_inodes, &splice);
4302
4303 while (!list_empty(&splice)) {
4304 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4305 delalloc_inodes);
4306
4307 list_del_init(&btrfs_inode->delalloc_inodes);
4308 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4309 &btrfs_inode->runtime_flags);
4310 spin_unlock(&root->delalloc_lock);
4311
4312 btrfs_invalidate_inodes(btrfs_inode->root);
4313
4314 spin_lock(&root->delalloc_lock);
4315 }
4316
4317 spin_unlock(&root->delalloc_lock);
4318}
4319
4320static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4321{
4322 struct btrfs_root *root;
4323 struct list_head splice;
4324
4325 INIT_LIST_HEAD(&splice);
4326
4327 spin_lock(&fs_info->delalloc_root_lock);
4328 list_splice_init(&fs_info->delalloc_roots, &splice);
4329 while (!list_empty(&splice)) {
4330 root = list_first_entry(&splice, struct btrfs_root,
4331 delalloc_root);
4332 list_del_init(&root->delalloc_root);
4333 root = btrfs_grab_fs_root(root);
4334 BUG_ON(!root);
4335 spin_unlock(&fs_info->delalloc_root_lock);
4336
4337 btrfs_destroy_delalloc_inodes(root);
4338 btrfs_put_fs_root(root);
4339
4340 spin_lock(&fs_info->delalloc_root_lock);
4341 }
4342 spin_unlock(&fs_info->delalloc_root_lock);
4343}
4344
4345static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4346 struct extent_io_tree *dirty_pages,
4347 int mark)
4348{
4349 int ret;
4350 struct extent_buffer *eb;
4351 u64 start = 0;
4352 u64 end;
4353
4354 while (1) {
4355 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4356 mark, NULL);
4357 if (ret)
4358 break;
4359
4360 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4361 while (start <= end) {
4362 eb = btrfs_find_tree_block(root->fs_info, start);
4363 start += root->nodesize;
4364 if (!eb)
4365 continue;
4366 wait_on_extent_buffer_writeback(eb);
4367
4368 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4369 &eb->bflags))
4370 clear_extent_buffer_dirty(eb);
4371 free_extent_buffer_stale(eb);
4372 }
4373 }
4374
4375 return ret;
4376}
4377
4378static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4379 struct extent_io_tree *pinned_extents)
4380{
4381 struct extent_io_tree *unpin;
4382 u64 start;
4383 u64 end;
4384 int ret;
4385 bool loop = true;
4386
4387 unpin = pinned_extents;
4388again:
4389 while (1) {
4390 ret = find_first_extent_bit(unpin, 0, &start, &end,
4391 EXTENT_DIRTY, NULL);
4392 if (ret)
4393 break;
4394
4395 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4396 btrfs_error_unpin_extent_range(root, start, end);
4397 cond_resched();
4398 }
4399
4400 if (loop) {
4401 if (unpin == &root->fs_info->freed_extents[0])
4402 unpin = &root->fs_info->freed_extents[1];
4403 else
4404 unpin = &root->fs_info->freed_extents[0];
4405 loop = false;
4406 goto again;
4407 }
4408
4409 return 0;
4410}
4411
4412void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4413 struct btrfs_root *root)
4414{
4415 btrfs_destroy_delayed_refs(cur_trans, root);
4416
4417 cur_trans->state = TRANS_STATE_COMMIT_START;
4418 wake_up(&root->fs_info->transaction_blocked_wait);
4419
4420 cur_trans->state = TRANS_STATE_UNBLOCKED;
4421 wake_up(&root->fs_info->transaction_wait);
4422
4423 btrfs_destroy_delayed_inodes(root);
4424 btrfs_assert_delayed_root_empty(root);
4425
4426 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4427 EXTENT_DIRTY);
4428 btrfs_destroy_pinned_extent(root,
4429 root->fs_info->pinned_extents);
4430
4431 cur_trans->state =TRANS_STATE_COMPLETED;
4432 wake_up(&cur_trans->commit_wait);
4433
4434 /*
4435 memset(cur_trans, 0, sizeof(*cur_trans));
4436 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4437 */
4438}
4439
4440static int btrfs_cleanup_transaction(struct btrfs_root *root)
4441{
4442 struct btrfs_transaction *t;
4443
4444 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4445
4446 spin_lock(&root->fs_info->trans_lock);
4447 while (!list_empty(&root->fs_info->trans_list)) {
4448 t = list_first_entry(&root->fs_info->trans_list,
4449 struct btrfs_transaction, list);
4450 if (t->state >= TRANS_STATE_COMMIT_START) {
4451 atomic_inc(&t->use_count);
4452 spin_unlock(&root->fs_info->trans_lock);
4453 btrfs_wait_for_commit(root, t->transid);
4454 btrfs_put_transaction(t);
4455 spin_lock(&root->fs_info->trans_lock);
4456 continue;
4457 }
4458 if (t == root->fs_info->running_transaction) {
4459 t->state = TRANS_STATE_COMMIT_DOING;
4460 spin_unlock(&root->fs_info->trans_lock);
4461 /*
4462 * We wait for 0 num_writers since we don't hold a trans
4463 * handle open currently for this transaction.
4464 */
4465 wait_event(t->writer_wait,
4466 atomic_read(&t->num_writers) == 0);
4467 } else {
4468 spin_unlock(&root->fs_info->trans_lock);
4469 }
4470 btrfs_cleanup_one_transaction(t, root);
4471
4472 spin_lock(&root->fs_info->trans_lock);
4473 if (t == root->fs_info->running_transaction)
4474 root->fs_info->running_transaction = NULL;
4475 list_del_init(&t->list);
4476 spin_unlock(&root->fs_info->trans_lock);
4477
4478 btrfs_put_transaction(t);
4479 trace_btrfs_transaction_commit(root);
4480 spin_lock(&root->fs_info->trans_lock);
4481 }
4482 spin_unlock(&root->fs_info->trans_lock);
4483 btrfs_destroy_all_ordered_extents(root->fs_info);
4484 btrfs_destroy_delayed_inodes(root);
4485 btrfs_assert_delayed_root_empty(root);
4486 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4487 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4488 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4489
4490 return 0;
4491}
4492
4493static const struct extent_io_ops btree_extent_io_ops = {
4494 .readpage_end_io_hook = btree_readpage_end_io_hook,
4495 .readpage_io_failed_hook = btree_io_failed_hook,
4496 .submit_bio_hook = btree_submit_bio_hook,
4497 /* note we're sharing with inode.c for the merge bio hook */
4498 .merge_bio_hook = btrfs_merge_bio_hook,
4499};