1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
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>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "rcu-string.h"
33 #include "dev-replace.h"
37 #include "compression.h"
38 #include "tree-checker.h"
39 #include "ref-verify.h"
40 #include "block-group.h"
42 #include "space-info.h"
46 #include "accessors.h"
47 #include "extent-tree.h"
48 #include "root-tree.h"
50 #include "uuid-tree.h"
51 #include "relocation.h"
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)
62 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
63 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
65 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
67 if (fs_info->csum_shash)
68 crypto_free_shash(fs_info->csum_shash);
72 * Compute the csum of a btree block and store the result to provided buffer.
74 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
76 struct btrfs_fs_info *fs_info = buf->fs_info;
79 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
83 shash->tfm = fs_info->csum_shash;
84 crypto_shash_init(shash);
87 /* Pages are contiguous, handle them as a big one. */
89 first_page_part = fs_info->nodesize;
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);
97 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
98 first_page_part - BTRFS_CSUM_SIZE);
101 * Multiple single-page folios case would reach here.
103 * nodesize <= PAGE_SIZE and large folio all handled by above
104 * crypto_shash_update() already.
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);
110 memset(result, 0, BTRFS_CSUM_SIZE);
111 crypto_shash_final(shash, result);
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.
120 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
122 if (!extent_buffer_uptodate(eb))
125 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
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);
143 static bool btrfs_supported_super_csum(u16 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:
157 * Return 0 if the superblock checksum type matches the checksum value of that
158 * algorithm. Pass the raw disk superblock data.
160 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
161 const struct btrfs_super_block *disk_sb)
163 char result[BTRFS_CSUM_SIZE];
164 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
166 shash->tfm = fs_info->csum_shash;
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.
173 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
174 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
176 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
182 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
185 struct btrfs_fs_info *fs_info = eb->fs_info;
186 int num_folios = num_extent_folios(eb);
189 if (sb_rdonly(fs_info->sb))
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;
199 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
200 start, folio, offset_in_folio(folio, start),
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.
213 * @check: expected tree parentness check, see the comments of the
214 * structure for details.
216 int btrfs_read_extent_buffer(struct extent_buffer *eb,
217 struct btrfs_tree_parent_check *check)
219 struct btrfs_fs_info *fs_info = eb->fs_info;
224 int failed_mirror = 0;
229 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
230 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
234 num_copies = btrfs_num_copies(fs_info,
239 if (!failed_mirror) {
241 failed_mirror = eb->read_mirror;
245 if (mirror_num == failed_mirror)
248 if (mirror_num > num_copies)
252 if (failed && !ret && failed_mirror)
253 btrfs_repair_eb_io_failure(eb, failed_mirror);
259 * Checksum a dirty tree block before IO.
261 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
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);
267 u8 result[BTRFS_CSUM_SIZE];
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;
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.
281 if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
282 memzero_extent_buffer(eb, 0, eb->len);
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;
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);
297 if (btrfs_header_level(eb))
298 ret = btrfs_check_node(eb);
300 ret = btrfs_check_leaf(eb);
306 * Also check the generation, the eb reached here must be newer than
307 * last committed. Or something seriously wrong happened.
309 last_trans = btrfs_get_last_trans_committed(fs_info);
310 if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
313 "block=%llu bad generation, have %llu expect > %llu",
314 eb->start, btrfs_header_generation(eb), last_trans);
317 write_extent_buffer(eb, result, 0, fs_info->csum_size);
321 btrfs_print_tree(eb, 0);
322 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
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.
330 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
331 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
332 return errno_to_blk_status(ret);
335 static bool check_tree_block_fsid(struct extent_buffer *eb)
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];
341 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
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.
350 if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
353 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
354 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
360 /* Do basic extent buffer checks at read time */
361 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
362 struct btrfs_tree_parent_check *check)
364 struct btrfs_fs_info *fs_info = eb->fs_info;
366 const u32 csum_size = fs_info->csum_size;
368 u8 result[BTRFS_CSUM_SIZE];
369 const u8 *header_csum;
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);
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);
388 found_level = btrfs_header_level(eb);
389 if (found_level >= BTRFS_MAX_LEVEL) {
391 "bad tree block level, mirror %u level %d on logical %llu",
392 eb->read_mirror, btrfs_header_level(eb), eb->start);
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));
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));
412 if (found_level != check->level) {
414 "level verify failed on logical %llu mirror %u wanted %u found %u",
415 eb->start, eb->read_mirror, check->level, found_level);
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));
428 if (check->has_first_key) {
429 struct btrfs_key *expect_key = &check->first_key;
430 struct btrfs_key found_key;
433 btrfs_node_key_to_cpu(eb, &found_key, 0);
435 btrfs_item_key_to_cpu(eb, &found_key, 0);
436 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
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,
448 if (check->owner_root) {
449 ret = btrfs_check_eb_owner(eb, check->owner_root);
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
459 if (found_level == 0 && btrfs_check_leaf(eb)) {
460 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
464 if (found_level > 0 && btrfs_check_node(eb))
469 "read time tree block corruption detected on logical %llu mirror %u",
470 eb->start, eb->read_mirror);
475 #ifdef CONFIG_MIGRATION
476 static int btree_migrate_folio(struct address_space *mapping,
477 struct folio *dst, struct folio *src, enum migrate_mode mode)
480 * we can't safely write a btree page from here,
481 * we haven't done the locking hook
483 if (folio_test_dirty(src))
486 * Buffers may be managed in a filesystem specific way.
487 * We must have no buffers or drop them.
489 if (folio_get_private(src) &&
490 !filemap_release_folio(src, GFP_KERNEL))
492 return migrate_folio(mapping, dst, src, mode);
495 #define btree_migrate_folio NULL
498 static int btree_writepages(struct address_space *mapping,
499 struct writeback_control *wbc)
501 struct btrfs_fs_info *fs_info;
504 if (wbc->sync_mode == WB_SYNC_NONE) {
506 if (wbc->for_kupdate)
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);
517 return btree_write_cache_pages(mapping, wbc);
520 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
522 if (folio_test_writeback(folio) || folio_test_dirty(folio))
525 return try_release_extent_buffer(&folio->page);
528 static void btree_invalidate_folio(struct folio *folio, size_t offset,
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);
544 static bool btree_dirty_folio(struct address_space *mapping,
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;
552 u64 page_start = folio_pos(folio);
554 if (fs_info->sectorsize == PAGE_SIZE) {
555 eb = folio_get_private(folio);
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);
564 subpage = folio_get_private(folio);
566 for (cur_bit = spi->dirty_offset;
567 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
572 spin_lock_irqsave(&subpage->lock, flags);
573 if (!test_bit(cur_bit, subpage->bitmaps)) {
574 spin_unlock_irqrestore(&subpage->lock, flags);
577 spin_unlock_irqrestore(&subpage->lock, flags);
578 cur = page_start + cur_bit * fs_info->sectorsize;
580 eb = find_extent_buffer(fs_info, cur);
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);
587 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
589 return filemap_dirty_folio(mapping, folio);
592 #define btree_dirty_folio filemap_dirty_folio
595 static 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,
603 struct extent_buffer *btrfs_find_create_tree_block(
604 struct btrfs_fs_info *fs_info,
605 u64 bytenr, u64 owner_root,
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);
614 * Read tree block at logical address @bytenr and do variant basic but critical
617 * @check: expected tree parentness check, see comments of the
618 * structure for details.
620 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
621 struct btrfs_tree_parent_check *check)
623 struct extent_buffer *buf = NULL;
628 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
633 ret = btrfs_read_extent_buffer(buf, check);
635 free_extent_buffer_stale(buf);
638 if (btrfs_check_eb_owner(buf, check->owner_root)) {
639 free_extent_buffer_stale(buf);
640 return ERR_PTR(-EUCLEAN);
646 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
649 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
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;
657 root->commit_root = NULL;
659 RB_CLEAR_NODE(&root->rb_node);
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);
669 btrfs_init_root_block_rsv(root);
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);
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);
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);
721 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
722 u64 objectid, gfp_t flags)
724 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
726 __setup_root(root, fs_info, objectid);
730 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
731 /* Should only be used by the testing infrastructure */
732 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
734 struct btrfs_root *root;
737 return ERR_PTR(-EINVAL);
739 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
741 return ERR_PTR(-ENOMEM);
743 /* We don't use the stripesize in selftest, set it as sectorsize */
744 root->alloc_bytenr = 0;
750 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
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);
755 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
758 static int global_root_key_cmp(const void *k, const struct rb_node *node)
760 const struct btrfs_key *key = k;
761 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
763 return btrfs_comp_cpu_keys(key, &root->root_key);
766 int btrfs_global_root_insert(struct btrfs_root *root)
768 struct btrfs_fs_info *fs_info = root->fs_info;
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);
778 btrfs_warn(fs_info, "global root %llu %llu already exists",
779 root->root_key.objectid, root->root_key.offset);
784 void btrfs_global_root_delete(struct btrfs_root *root)
786 struct btrfs_fs_info *fs_info = root->fs_info;
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);
793 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
794 struct btrfs_key *key)
796 struct rb_node *node;
797 struct btrfs_root *root = NULL;
799 read_lock(&fs_info->global_root_lock);
800 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
802 root = container_of(node, struct btrfs_root, rb_node);
803 read_unlock(&fs_info->global_root_lock);
808 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
810 struct btrfs_block_group *block_group;
813 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
817 block_group = btrfs_lookup_block_group(fs_info, bytenr);
819 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
823 ret = block_group->global_root_id;
824 btrfs_put_block_group(block_group);
829 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
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),
837 return btrfs_global_root(fs_info, &key);
840 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
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),
848 return btrfs_global_root(fs_info, &key);
851 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
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);
858 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
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;
870 * We're holding a transaction handle, so use a NOFS memory allocation
871 * context to avoid deadlock if reclaim happens.
873 nofs_flag = memalloc_nofs_save();
874 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
875 memalloc_nofs_restore(nofs_flag);
877 return ERR_PTR(-ENOMEM);
879 root->root_key.objectid = objectid;
880 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
881 root->root_key.offset = 0;
883 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
884 0, BTRFS_NESTING_NORMAL);
892 btrfs_mark_buffer_dirty(trans, leaf);
894 root->commit_root = btrfs_root_node(root);
895 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
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);
909 export_guid(root->root_item.uuid, &guid_null);
910 btrfs_set_root_drop_level(&root->root_item, 0);
912 btrfs_tree_unlock(leaf);
914 key.objectid = objectid;
915 key.type = BTRFS_ROOT_ITEM_KEY;
917 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
924 btrfs_put_root(root);
929 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
930 struct btrfs_fs_info *fs_info)
932 struct btrfs_root *root;
934 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
936 return ERR_PTR(-ENOMEM);
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;
945 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
946 struct btrfs_root *root)
948 struct extent_buffer *leaf;
951 * DON'T set SHAREABLE bit for log trees.
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.
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).
960 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
961 NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
963 return PTR_ERR(leaf);
967 btrfs_mark_buffer_dirty(trans, root->node);
968 btrfs_tree_unlock(root->node);
973 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
974 struct btrfs_fs_info *fs_info)
976 struct btrfs_root *log_root;
978 log_root = alloc_log_tree(trans, fs_info);
979 if (IS_ERR(log_root))
980 return PTR_ERR(log_root);
982 if (!btrfs_is_zoned(fs_info)) {
983 int ret = btrfs_alloc_log_tree_node(trans, log_root);
986 btrfs_put_root(log_root);
991 WARN_ON(fs_info->log_root_tree);
992 fs_info->log_root_tree = log_root;
996 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
997 struct btrfs_root *root)
999 struct btrfs_fs_info *fs_info = root->fs_info;
1000 struct btrfs_root *log_root;
1001 struct btrfs_inode_item *inode_item;
1004 log_root = alloc_log_tree(trans, fs_info);
1005 if (IS_ERR(log_root))
1006 return PTR_ERR(log_root);
1008 ret = btrfs_alloc_log_tree_node(trans, log_root);
1010 btrfs_put_root(log_root);
1014 log_root->last_trans = trans->transid;
1015 log_root->root_key.offset = root->root_key.objectid;
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,
1023 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1025 btrfs_set_root_node(&log_root->root_item, log_root->node);
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);
1035 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1036 struct btrfs_path *path,
1037 struct btrfs_key *key)
1039 struct btrfs_root *root;
1040 struct btrfs_tree_parent_check check = { 0 };
1041 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1046 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1048 return ERR_PTR(-ENOMEM);
1050 ret = btrfs_find_root(tree_root, key, path,
1051 &root->root_item, &root->root_key);
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),
1065 if (IS_ERR(root->node)) {
1066 ret = PTR_ERR(root->node);
1070 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1076 * For real fs, and not log/reloc trees, root owner must
1077 * match its root node owner
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)) {
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);
1091 root->commit_root = btrfs_root_node(root);
1094 btrfs_put_root(root);
1095 return ERR_PTR(ret);
1098 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1099 struct btrfs_key *key)
1101 struct btrfs_root *root;
1102 struct btrfs_path *path;
1104 path = btrfs_alloc_path();
1106 return ERR_PTR(-ENOMEM);
1107 root = read_tree_root_path(tree_root, path, key);
1108 btrfs_free_path(path);
1114 * Initialize subvolume root in-memory structure
1116 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1118 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1122 btrfs_drew_lock_init(&root->snapshot_lock);
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);
1132 * Don't assign anonymous block device to roots that are not exposed to
1133 * userspace, the id pool is limited to 1M
1135 if (is_fstree(root->root_key.objectid) &&
1136 btrfs_root_refs(&root->root_item) > 0) {
1138 ret = get_anon_bdev(&root->anon_dev);
1142 root->anon_dev = anon_dev;
1146 mutex_lock(&root->objectid_mutex);
1147 ret = btrfs_init_root_free_objectid(root);
1149 mutex_unlock(&root->objectid_mutex);
1153 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1155 mutex_unlock(&root->objectid_mutex);
1159 /* The caller is responsible to call btrfs_free_fs_root */
1163 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1166 struct btrfs_root *root;
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);
1176 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1179 struct btrfs_key key = {
1180 .objectid = objectid,
1181 .type = BTRFS_ROOT_ITEM_KEY,
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);
1211 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1212 struct btrfs_root *root)
1216 ret = radix_tree_preload(GFP_NOFS);
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,
1225 btrfs_grab_root(root);
1226 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1228 spin_unlock(&fs_info->fs_roots_radix_lock);
1229 radix_tree_preload_end();
1234 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1236 #ifdef CONFIG_BTRFS_DEBUG
1237 struct btrfs_root *root;
1239 while (!list_empty(&fs_info->allocated_roots)) {
1240 char buf[BTRFS_ROOT_NAME_BUF_LEN];
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);
1254 static void free_global_roots(struct btrfs_fs_info *fs_info)
1256 struct btrfs_root *root;
1257 struct rb_node *node;
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);
1266 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
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);
1297 * Get an in-memory reference of a root structure.
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.
1303 * Caller should release the root by calling btrfs_put_root() after the usage.
1305 * NOTE: Reloc and log trees can't be read by this function as they share the
1306 * same root objectid.
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
1314 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1315 u64 objectid, dev_t *anon_dev,
1318 struct btrfs_root *root;
1319 struct btrfs_path *path;
1320 struct btrfs_key key;
1323 root = btrfs_get_global_root(fs_info, objectid);
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.
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.
1334 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1335 return ERR_PTR(-ENOENT);
1337 root = btrfs_lookup_fs_root(fs_info, objectid);
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.
1345 if (unlikely(anon_dev && *anon_dev)) {
1346 free_anon_bdev(*anon_dev);
1350 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1351 btrfs_put_root(root);
1352 return ERR_PTR(-ENOENT);
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);
1364 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1369 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1373 path = btrfs_alloc_path();
1378 key.objectid = BTRFS_ORPHAN_OBJECTID;
1379 key.type = BTRFS_ORPHAN_ITEM_KEY;
1380 key.offset = objectid;
1382 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1383 btrfs_free_path(path);
1387 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1389 ret = btrfs_insert_fs_root(fs_info, root);
1391 if (ret == -EEXIST) {
1392 btrfs_put_root(root);
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.
1405 if (anon_dev && *anon_dev)
1407 btrfs_put_root(root);
1408 return ERR_PTR(ret);
1412 * Get in-memory reference of a root structure
1414 * @objectid: tree objectid
1415 * @check_ref: if set, verify that the tree exists and the item has at least
1418 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1419 u64 objectid, bool check_ref)
1421 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1425 * Get in-memory reference of a root structure, created as new, optionally pass
1426 * the anonymous block device id
1428 * @objectid: tree objectid
1429 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1430 * parameter value if not NULL
1432 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1433 u64 objectid, dev_t *anon_dev)
1435 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1439 * Return a root for the given objectid.
1441 * @fs_info: the fs_info
1442 * @objectid: the objectid we need to lookup
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.
1453 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1454 struct btrfs_path *path,
1457 struct btrfs_root *root;
1458 struct btrfs_key key;
1460 ASSERT(path->search_commit_root && path->skip_locking);
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.
1468 root = btrfs_get_global_root(fs_info, objectid);
1472 root = btrfs_lookup_fs_root(fs_info, objectid);
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);
1485 static int cleaner_kthread(void *arg)
1487 struct btrfs_fs_info *fs_info = arg;
1493 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1495 /* Make the cleaner go to sleep early. */
1496 if (btrfs_need_cleaner_sleep(fs_info))
1500 * Do not do anything if we might cause open_ctree() to block
1501 * before we have finished mounting the filesystem.
1503 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1506 if (!mutex_trylock(&fs_info->cleaner_mutex))
1510 * Avoid the problem that we change the status of the fs
1511 * during the above check and trylock.
1513 if (btrfs_need_cleaner_sleep(fs_info)) {
1514 mutex_unlock(&fs_info->cleaner_mutex);
1518 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1519 btrfs_sysfs_feature_update(fs_info);
1521 btrfs_run_delayed_iputs(fs_info);
1523 again = btrfs_clean_one_deleted_snapshot(fs_info);
1524 mutex_unlock(&fs_info->cleaner_mutex);
1527 * The defragger has dealt with the R/O remount and umount,
1528 * needn't do anything special here.
1530 btrfs_run_defrag_inodes(fs_info);
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.
1540 btrfs_delete_unused_bgs(fs_info);
1543 * Reclaim block groups in the reclaim_bgs list after we deleted
1544 * all unused block_groups. This possibly gives us some more free
1547 btrfs_reclaim_bgs(fs_info);
1549 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1550 if (kthread_should_park())
1552 if (kthread_should_stop())
1555 set_current_state(TASK_INTERRUPTIBLE);
1557 __set_current_state(TASK_RUNNING);
1562 static int transaction_kthread(void *arg)
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;
1570 unsigned long delay;
1574 cannot_commit = false;
1575 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1576 mutex_lock(&fs_info->transaction_kthread_mutex);
1578 spin_lock(&fs_info->trans_lock);
1579 cur = fs_info->running_transaction;
1581 spin_unlock(&fs_info->trans_lock);
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);
1592 msecs_to_jiffies(fs_info->commit_interval * 1000));
1595 transid = cur->transid;
1596 spin_unlock(&fs_info->trans_lock);
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;
1605 if (transid == trans->transid) {
1606 btrfs_commit_transaction(trans);
1608 btrfs_end_transaction(trans);
1611 wake_up_process(fs_info->cleaner_kthread);
1612 mutex_unlock(&fs_info->transaction_kthread_mutex);
1614 if (BTRFS_FS_ERROR(fs_info))
1615 btrfs_cleanup_transaction(fs_info);
1616 if (!kthread_should_stop() &&
1617 (!btrfs_transaction_blocked(fs_info) ||
1619 schedule_timeout_interruptible(delay);
1620 } while (!kthread_should_stop());
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
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.
1633 static int find_newest_super_backup(struct btrfs_fs_info *info)
1635 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1637 struct btrfs_root_backup *root_backup;
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)
1651 * copy all the root pointers into the super backup array.
1652 * this will bump the backup pointer by one when it is
1655 static void backup_super_roots(struct btrfs_fs_info *info)
1657 const int next_backup = info->backup_root_index;
1658 struct btrfs_root_backup *root_backup;
1660 root_backup = info->super_for_commit->super_roots + next_backup;
1663 * make sure all of our padding and empty slots get zero filled
1664 * regardless of which ones we use today
1666 memset(root_backup, 0, sizeof(*root_backup));
1668 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
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));
1674 btrfs_set_backup_tree_root_level(root_backup,
1675 btrfs_header_level(info->tree_root->node));
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));
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);
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));
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));
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.
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));
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));
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));
1728 * if we don't copy this out to the super_copy, it won't get remembered
1729 * for the next commit
1731 memcpy(&info->super_copy->super_roots,
1732 &info->super_for_commit->super_roots,
1733 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
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
1740 * @fs_info: filesystem whose backup roots need to be read
1741 * @priority: priority of backup root required
1743 * Returns backup root index on success and -EINVAL otherwise.
1745 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
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;
1751 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1753 return backup_index;
1755 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1756 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1761 root_backup = super->super_roots + backup_index;
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));
1771 * Fixme: the total bytes and num_devices need to match or we should
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));
1777 return backup_index;
1780 /* helper to cleanup workers */
1781 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
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);
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.
1805 if (fs_info->endio_meta_workers)
1806 destroy_workqueue(fs_info->endio_meta_workers);
1809 static void free_root_extent_buffers(struct btrfs_root *root)
1812 free_extent_buffer(root->node);
1813 free_extent_buffer(root->commit_root);
1815 root->commit_root = NULL;
1819 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1821 struct btrfs_root *root, *tmp;
1823 rbtree_postorder_for_each_entry_safe(root, tmp,
1824 &fs_info->global_root_tree,
1826 free_root_extent_buffers(root);
1829 /* helper to cleanup tree roots */
1830 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1832 free_root_extent_buffers(info->tree_root);
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);
1846 void btrfs_put_root(struct btrfs_root *root)
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));
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);
1866 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1869 struct btrfs_root *gang[8];
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);
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]);
1883 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1888 for (i = 0; i < ret; i++)
1889 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1893 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
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);
1904 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
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);
1915 static int btrfs_init_btree_inode(struct super_block *sb)
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;
1922 inode = new_inode(sb);
1926 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1927 set_nlink(inode, 1);
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
1933 inode->i_size = OFFSET_MAX;
1934 inode->i_mapping->a_ops = &btree_aops;
1935 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
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);
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;
1953 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
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);
1960 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
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);
1973 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
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;
1980 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1982 fs_info->delalloc_workers =
1983 btrfs_alloc_workqueue(fs_info, "delalloc",
1984 flags, max_active, 2);
1986 fs_info->flush_workers =
1987 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1988 flags, max_active, 0);
1990 fs_info->caching_workers =
1991 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1993 fs_info->fixup_workers =
1994 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
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,
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,
2009 fs_info->delayed_workers =
2010 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2012 fs_info->qgroup_rescan_workers =
2013 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2015 fs_info->discard_ctl.discard_workers =
2016 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
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)) {
2033 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2035 struct crypto_shash *csum_shash;
2036 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2038 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2040 if (IS_ERR(csum_shash)) {
2041 btrfs_err(fs_info, "error allocating %s hash for checksum",
2043 return PTR_ERR(csum_shash);
2046 fs_info->csum_shash = csum_shash;
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.
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);
2058 case BTRFS_CSUM_TYPE_XXHASH:
2059 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2065 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2066 btrfs_super_csum_name(csum_type),
2067 crypto_shash_driver_name(csum_shash));
2071 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2072 struct btrfs_fs_devices *fs_devices)
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);
2081 if (fs_devices->rw_devices == 0) {
2082 btrfs_warn(fs_info, "log replay required on RO media");
2086 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
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);
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);
2108 /* returns with log_tree_root freed on success */
2109 ret = btrfs_recover_log_trees(log_tree_root);
2111 btrfs_handle_fs_error(fs_info, ret,
2112 "Failed to recover log tree");
2113 btrfs_put_root(log_tree_root);
2117 if (sb_rdonly(fs_info->sb)) {
2118 ret = btrfs_commit_super(fs_info);
2126 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2127 struct btrfs_path *path, u64 objectid,
2130 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2131 struct btrfs_root *root;
2132 u64 max_global_id = 0;
2134 struct btrfs_key key = {
2135 .objectid = objectid,
2136 .type = BTRFS_ROOT_ITEM_KEY,
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);
2149 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2153 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2154 ret = btrfs_next_leaf(tree_root, path);
2163 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2164 if (key.objectid != objectid)
2166 btrfs_release_path(path);
2169 * Just worry about this for extent tree, it'll be the same for
2172 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2173 max_global_id = max(max_global_id, key.offset);
2176 root = read_tree_root_path(tree_root, path, &key);
2178 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2179 ret = PTR_ERR(root);
2182 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2183 ret = btrfs_global_root_insert(root);
2185 btrfs_put_root(root);
2190 btrfs_release_path(path);
2192 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2193 fs_info->nr_global_roots = max_global_id + 1;
2195 if (!found || ret) {
2196 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2197 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2199 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2200 ret = ret ? ret : -ENOENT;
2203 btrfs_err(fs_info, "failed to load root %s", name);
2208 static int load_global_roots(struct btrfs_root *tree_root)
2210 struct btrfs_path *path;
2213 path = btrfs_alloc_path();
2217 ret = load_global_roots_objectid(tree_root, path,
2218 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2221 ret = load_global_roots_objectid(tree_root, path,
2222 BTRFS_CSUM_TREE_OBJECTID, "csum");
2225 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2227 ret = load_global_roots_objectid(tree_root, path,
2228 BTRFS_FREE_SPACE_TREE_OBJECTID,
2231 btrfs_free_path(path);
2235 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2237 struct btrfs_root *tree_root = fs_info->tree_root;
2238 struct btrfs_root *root;
2239 struct btrfs_key location;
2242 BUG_ON(!fs_info->tree_root);
2244 ret = load_global_roots(tree_root);
2248 location.type = BTRFS_ROOT_ITEM_KEY;
2249 location.offset = 0;
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);
2255 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2256 ret = PTR_ERR(root);
2260 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2261 fs_info->block_group_root = root;
2265 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2266 root = btrfs_read_tree_root(tree_root, &location);
2268 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2269 ret = PTR_ERR(root);
2273 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2274 fs_info->dev_root = root;
2276 /* Initialize fs_info for all devices in any case */
2277 ret = btrfs_init_devices_late(fs_info);
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
2285 root = btrfs_get_fs_root(tree_root->fs_info,
2286 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2288 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2289 ret = PTR_ERR(root);
2293 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2294 fs_info->data_reloc_root = root;
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;
2304 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2305 root = btrfs_read_tree_root(tree_root, &location);
2307 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2308 ret = PTR_ERR(root);
2313 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2314 fs_info->uuid_root = root;
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);
2321 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2322 ret = PTR_ERR(root);
2326 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2327 fs_info->stripe_root = root;
2333 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2334 location.objectid, ret);
2339 * Real super block validation
2340 * NOTE: super csum type and incompat features will not be checked here.
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
2348 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2349 struct btrfs_super_block *sb, int mirror_num)
2351 u64 nodesize = btrfs_super_nodesize(sb);
2352 u64 sectorsize = btrfs_super_sectorsize(sb);
2355 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2356 btrfs_err(fs_info, "no valid FS found");
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);
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);
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);
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);
2381 * Check sectorsize and nodesize first, other check will need it.
2382 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2384 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2385 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2386 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2391 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
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
2398 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2400 "sectorsize %llu not yet supported for page size %lu",
2401 sectorsize, PAGE_SIZE);
2405 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2406 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2407 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
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);
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));
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));
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));
2433 if (!fs_info->fs_devices->temp_fsid &&
2434 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2436 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2437 sb->fsid, fs_info->fs_devices->fsid);
2441 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2442 BTRFS_FSID_SIZE) != 0) {
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);
2449 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2450 BTRFS_FSID_SIZE) != 0) {
2452 "dev_item UUID does not match metadata fsid: %pU != %pU",
2453 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2458 * Artificial requirement for block-group-tree to force newer features
2459 * (free-space-tree, no-holes) so the test matrix is smaller.
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))) {
2465 "block-group-tree feature requires fres-space-tree and no-holes");
2470 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
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));
2478 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2479 btrfs_err(fs_info, "invalid stripesize %u",
2480 btrfs_super_stripesize(sb));
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");
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);
2499 * Obvious sys_chunk_array corruptions, it must hold at least one key
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);
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));
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.
2521 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
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)
2529 "suspicious: generation < cache_generation: %llu < %llu",
2530 btrfs_super_generation(sb),
2531 btrfs_super_cache_generation(sb));
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.
2541 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2543 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2547 * Validation of super block at write time.
2548 * Some checks like bytenr check will be skipped as their values will be
2550 * Extra checks like csum type and incompat flags will be done here.
2552 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2553 struct btrfs_super_block *sb)
2557 ret = btrfs_validate_super(fs_info, sb, -1);
2560 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2562 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2563 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2566 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
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);
2577 "super block corruption detected before writing it to disk");
2581 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2583 struct btrfs_tree_parent_check check = {
2586 .owner_root = root->root_key.objectid
2590 root->node = read_tree_block(root->fs_info, bytenr, &check);
2591 if (IS_ERR(root->node)) {
2592 ret = PTR_ERR(root->node);
2596 if (!extent_buffer_uptodate(root->node)) {
2597 free_extent_buffer(root->node);
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);
2608 static int load_important_roots(struct btrfs_fs_info *fs_info)
2610 struct btrfs_super_block *sb = fs_info->super_copy;
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);
2619 btrfs_warn(fs_info, "couldn't read tree root");
2625 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
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;
2634 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2636 if (!IS_ERR(tree_root->node))
2637 free_extent_buffer(tree_root->node);
2638 tree_root->node = NULL;
2640 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2643 free_root_pointers(fs_info, 0);
2646 * Don't use the log in recovery mode, it won't be
2649 btrfs_set_super_log_root(sb, 0);
2651 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2652 ret = read_backup_root(fs_info, i);
2658 ret = load_important_roots(fs_info);
2660 handle_error = true;
2665 * No need to hold btrfs_root::objectid_mutex since the fs
2666 * hasn't been fully initialised and we are the only user
2668 ret = btrfs_init_root_free_objectid(tree_root);
2670 handle_error = true;
2674 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2676 ret = btrfs_read_roots(fs_info);
2678 handle_error = true;
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;
2687 /* Always begin writing backup roots after the one being used */
2688 if (backup_index < 0) {
2689 fs_info->backup_root_index = 0;
2691 fs_info->backup_root_index = backup_index + 1;
2692 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2700 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
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);
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);
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);
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);
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);
2779 fs_info->thread_pool_size = min_t(unsigned long,
2780 num_online_cpus() + 2, 8);
2782 INIT_LIST_HEAD(&fs_info->ordered_roots);
2783 spin_lock_init(&fs_info->ordered_root_lock);
2785 btrfs_init_scrub(fs_info);
2786 btrfs_init_balance(fs_info);
2787 btrfs_init_async_reclaim_work(fs_info);
2789 rwlock_init(&fs_info->block_group_cache_lock);
2790 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2792 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2793 IO_TREE_FS_EXCLUDED_EXTENTS);
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);
2806 btrfs_init_dev_replace_locks(fs_info);
2807 btrfs_init_qgroup(fs_info);
2808 btrfs_discard_init(fs_info);
2810 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2811 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
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);
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;
2825 /* Default compress algorithm when user does -o compress */
2826 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2828 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2830 spin_lock_init(&fs_info->swapfile_pins_lock);
2831 fs_info->swapfile_pins = RB_ROOT;
2833 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2834 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2837 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2842 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2843 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2845 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2849 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2853 fs_info->dirty_metadata_batch = PAGE_SIZE *
2854 (1 + ilog2(nr_cpu_ids));
2856 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2860 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2865 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2867 if (!fs_info->delayed_root)
2869 btrfs_init_delayed_root(fs_info->delayed_root);
2872 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2874 return btrfs_alloc_stripe_hash_table(fs_info);
2877 static int btrfs_uuid_rescan_kthread(void *data)
2879 struct btrfs_fs_info *fs_info = data;
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.
2887 ret = btrfs_uuid_tree_iterate(fs_info);
2890 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2892 up(&fs_info->uuid_tree_rescan_sem);
2895 return btrfs_uuid_scan_kthread(data);
2898 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2900 struct task_struct *task;
2902 down(&fs_info->uuid_tree_rescan_sem);
2903 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
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);
2914 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2916 u64 root_objectid = 0;
2917 struct btrfs_root *gang[8];
2920 unsigned int ret = 0;
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,
2928 spin_unlock(&fs_info->fs_roots_radix_lock);
2931 root_objectid = gang[ret - 1]->root_key.objectid + 1;
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) {
2939 /* Grab all the search result for later use. */
2940 gang[i] = btrfs_grab_root(gang[i]);
2942 spin_unlock(&fs_info->fs_roots_radix_lock);
2944 for (i = 0; i < ret; i++) {
2947 root_objectid = gang[i]->root_key.objectid;
2948 err = btrfs_orphan_cleanup(gang[i]);
2951 btrfs_put_root(gang[i]);
2956 /* Release the uncleaned roots due to error. */
2957 for (; i < ret; i++) {
2959 btrfs_put_root(gang[i]);
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.
2968 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2971 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2972 bool rebuild_free_space_tree = false;
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))
2978 "'clear_cache' option is ignored with extent tree v2");
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;
2987 if (rebuild_free_space_tree) {
2988 btrfs_info(fs_info, "rebuilding free space tree");
2989 ret = btrfs_rebuild_free_space_tree(fs_info);
2992 "failed to rebuild free space tree: %d", ret);
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);
3003 "failed to disable free space tree: %d", ret);
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.
3019 ret = btrfs_find_orphan_roots(fs_info);
3023 ret = btrfs_cleanup_fs_roots(fs_info);
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);
3033 up_read(&fs_info->cleanup_work_sem);
3035 mutex_lock(&fs_info->cleaner_mutex);
3036 ret = btrfs_recover_relocation(fs_info);
3037 mutex_unlock(&fs_info->cleaner_mutex);
3039 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
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);
3049 "failed to create free space tree: %d", ret);
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);
3060 ret = btrfs_resume_balance_async(fs_info);
3064 ret = btrfs_resume_dev_replace_async(fs_info);
3066 btrfs_warn(fs_info, "failed to resume dev_replace");
3070 btrfs_qgroup_rescan_resume(fs_info);
3072 if (!fs_info->uuid_root) {
3073 btrfs_info(fs_info, "creating UUID tree");
3074 ret = btrfs_create_uuid_tree(fs_info);
3077 "failed to create the UUID tree %d", ret);
3087 * Do various sanity and dependency checks of different features.
3089 * @is_rw_mount: If the mount is read-write.
3091 * This is the place for less strict checks (like for subpage or artificial
3092 * feature dependencies).
3094 * For strict checks or possible corruption detection, see
3095 * btrfs_validate_super().
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.
3101 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
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);
3108 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3110 "cannot mount because of unknown incompat features (0x%llx)",
3115 /* Runtime limitation for mixed block groups. */
3116 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3117 (fs_info->sectorsize != fs_info->nodesize)) {
3119 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3120 fs_info->nodesize, fs_info->sectorsize);
3124 /* Mixed backref is an always-enabled feature. */
3125 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
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;
3134 * An ancient flag, which should really be marked deprecated.
3135 * Such runtime limitation doesn't really need a incompat flag.
3137 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3138 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3140 if (compat_ro_unsupp && is_rw_mount) {
3142 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
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.
3152 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3153 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3155 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3161 * Artificial limitations for block group tree, to force
3162 * block-group-tree to rely on no-holes and free-space-tree.
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))) {
3168 "block-group-tree feature requires no-holes and free-space-tree features");
3173 * Subpage runtime limitation on v1 cache.
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.
3179 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3181 "v1 space cache is not supported for page size %lu with sectorsize %u",
3182 PAGE_SIZE, fs_info->sectorsize);
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);
3194 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
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;
3209 ret = init_mount_fs_info(fs_info, sb);
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,
3216 fs_info->tree_root = tree_root;
3217 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3219 fs_info->chunk_root = chunk_root;
3220 if (!tree_root || !chunk_root) {
3225 ret = btrfs_init_btree_inode(sb);
3229 invalidate_bdev(fs_devices->latest_dev->bdev);
3232 * Read super block and check the signature bytes only
3234 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3235 if (IS_ERR(disk_super)) {
3236 ret = PTR_ERR(disk_super);
3240 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3242 * Verify the type first, if that or the checksum value are
3243 * corrupted, we'll find out
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",
3250 btrfs_release_disk_super(disk_super);
3254 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3256 ret = btrfs_init_csum_hash(fs_info, csum_type);
3258 btrfs_release_disk_super(disk_super);
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).
3266 if (btrfs_check_super_csum(fs_info, disk_super)) {
3267 btrfs_err(fs_info, "superblock checksum mismatch");
3269 btrfs_release_disk_super(disk_super);
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
3278 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3279 btrfs_release_disk_super(disk_super);
3281 disk_super = fs_info->super_copy;
3283 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3284 sizeof(*fs_info->super_for_commit));
3286 ret = btrfs_validate_mount_super(fs_info);
3288 btrfs_err(fs_info, "superblock contains fatal errors");
3293 if (!btrfs_super_root(disk_super)) {
3294 btrfs_err(fs_info, "invalid superblock tree root bytenr");
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);
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));
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;
3317 * Handle the space caching options appropriately now that we have the
3318 * super block loaded and validated.
3320 btrfs_set_free_space_cache_settings(fs_info);
3322 if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3327 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
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.
3335 fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3337 if (sectorsize < PAGE_SIZE) {
3338 struct btrfs_subpage_info *subpage_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) {
3348 btrfs_init_subpage_info(subpage_info, sectorsize);
3349 fs_info->subpage_info = subpage_info;
3352 ret = btrfs_init_workqueues(fs_info);
3354 goto fail_sb_buffer;
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);
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);
3363 mutex_lock(&fs_info->chunk_mutex);
3364 ret = btrfs_read_sys_array(fs_info);
3365 mutex_unlock(&fs_info->chunk_mutex);
3367 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3368 goto fail_sb_buffer;
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),
3376 btrfs_err(fs_info, "failed to read chunk root");
3377 goto fail_tree_roots;
3380 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3381 offsetof(struct btrfs_header, chunk_tree_uuid),
3384 ret = btrfs_read_chunk_tree(fs_info);
3386 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3387 goto fail_tree_roots;
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().
3397 btrfs_free_extra_devids(fs_devices);
3398 if (!fs_devices->latest_dev->bdev) {
3399 btrfs_err(fs_info, "failed to read devices");
3401 goto fail_tree_roots;
3404 ret = init_tree_roots(fs_info);
3406 goto fail_tree_roots;
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.
3413 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3416 "zoned: failed to read device zone info: %d", ret);
3417 goto fail_block_groups;
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.
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);
3432 ret = btrfs_verify_dev_extents(fs_info);
3435 "failed to verify dev extents against chunks: %d",
3437 goto fail_block_groups;
3439 ret = btrfs_recover_balance(fs_info);
3441 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3442 goto fail_block_groups;
3445 ret = btrfs_init_dev_stats(fs_info);
3447 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3448 goto fail_block_groups;
3451 ret = btrfs_init_dev_replace(fs_info);
3453 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3454 goto fail_block_groups;
3457 ret = btrfs_check_zoned_mode(fs_info);
3459 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3461 goto fail_block_groups;
3464 ret = btrfs_sysfs_add_fsid(fs_devices);
3466 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3468 goto fail_block_groups;
3471 ret = btrfs_sysfs_add_mounted(fs_info);
3473 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3474 goto fail_fsdev_sysfs;
3477 ret = btrfs_init_space_info(fs_info);
3479 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3483 ret = btrfs_read_block_groups(fs_info);
3485 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3489 btrfs_free_zone_cache(fs_info);
3491 btrfs_check_active_zone_reservation(fs_info);
3493 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3494 !btrfs_check_rw_degradable(fs_info, NULL)) {
3496 "writable mount is not allowed due to too many missing devices");
3501 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3503 if (IS_ERR(fs_info->cleaner_kthread)) {
3504 ret = PTR_ERR(fs_info->cleaner_kthread);
3508 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3510 "btrfs-transaction");
3511 if (IS_ERR(fs_info->transaction_kthread)) {
3512 ret = PTR_ERR(fs_info->transaction_kthread);
3516 ret = btrfs_read_qgroup_config(fs_info);
3518 goto fail_trans_kthread;
3520 if (btrfs_build_ref_tree(fs_info))
3521 btrfs_err(fs_info, "couldn't build ref tree");
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);
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;
3543 ret = btrfs_start_pre_rw_mount(fs_info);
3545 close_ctree(fs_info);
3548 btrfs_discard_resume(fs_info);
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);
3557 "failed to check the UUID tree: %d", ret);
3558 close_ctree(fs_info);
3563 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
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);
3572 btrfs_free_qgroup_config(fs_info);
3574 kthread_stop(fs_info->transaction_kthread);
3575 btrfs_cleanup_transaction(fs_info);
3576 btrfs_free_fs_roots(fs_info);
3578 kthread_stop(fs_info->cleaner_kthread);
3581 * make sure we're done with the btree inode before we stop our
3584 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3587 btrfs_sysfs_remove_mounted(fs_info);
3590 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3593 btrfs_put_block_group_cache(fs_info);
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);
3602 btrfs_stop_all_workers(fs_info);
3603 btrfs_free_block_groups(fs_info);
3605 btrfs_mapping_tree_free(fs_info);
3607 iput(fs_info->btree_inode);
3609 btrfs_close_devices(fs_info->fs_devices);
3613 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3615 static void btrfs_end_super_write(struct bio *bio)
3617 struct btrfs_device *device = bio->bi_private;
3618 struct bio_vec *bvec;
3619 struct bvec_iter_all iter_all;
3622 bio_for_each_segment_all(bvec, bio, iter_all) {
3623 page = bvec->bv_page;
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);
3632 btrfs_dev_stat_inc_and_print(device,
3633 BTRFS_DEV_STAT_WRITE_ERRS);
3635 SetPageUptodate(page);
3645 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3646 int copy_num, bool drop_cache)
3648 struct btrfs_super_block *super;
3650 u64 bytenr, bytenr_orig;
3651 struct address_space *mapping = bdev->bd_inode->i_mapping;
3654 bytenr_orig = btrfs_sb_offset(copy_num);
3655 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3657 return ERR_PTR(-EINVAL);
3659 return ERR_PTR(ret);
3661 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3662 return ERR_PTR(-EINVAL);
3665 /* This should only be called with the primary sb. */
3666 ASSERT(copy_num == 0);
3669 * Drop the page of the primary superblock, so later read will
3670 * always read from the device.
3672 invalidate_inode_pages2_range(mapping,
3673 bytenr >> PAGE_SHIFT,
3674 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3677 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3679 return ERR_CAST(page);
3681 super = page_address(page);
3682 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3683 btrfs_release_disk_super(super);
3684 return ERR_PTR(-ENODATA);
3687 if (btrfs_super_bytenr(super) != bytenr_orig) {
3688 btrfs_release_disk_super(super);
3689 return ERR_PTR(-EINVAL);
3696 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3698 struct btrfs_super_block *super, *latest = NULL;
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
3707 for (i = 0; i < 1; i++) {
3708 super = btrfs_read_dev_one_super(bdev, i, false);
3712 if (!latest || btrfs_super_generation(super) > transid) {
3714 btrfs_release_disk_super(super);
3717 transid = btrfs_super_generation(super);
3725 * Write superblock @sb to the @device. Do not wait for completion, all the
3726 * pages we use for writing are locked.
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.
3732 * Return number of errors when page is not found or submission fails.
3734 static int write_dev_supers(struct btrfs_device *device,
3735 struct btrfs_super_block *sb, int max_mirrors)
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);
3743 u64 bytenr, bytenr_orig;
3745 if (max_mirrors == 0)
3746 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3748 shash->tfm = fs_info->csum_shash;
3750 for (i = 0; i < max_mirrors; i++) {
3753 struct btrfs_super_block *disk_super;
3755 bytenr_orig = btrfs_sb_offset(i);
3756 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3757 if (ret == -ENOENT) {
3759 } else if (ret < 0) {
3760 btrfs_err(device->fs_info,
3761 "couldn't get super block location for mirror %d",
3766 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3767 device->commit_total_bytes)
3770 btrfs_set_super_bytenr(sb, bytenr_orig);
3772 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3773 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3776 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3779 btrfs_err(device->fs_info,
3780 "couldn't get super block page for bytenr %llu",
3786 /* Bump the refcount for wait_dev_supers() */
3789 disk_super = page_address(page);
3790 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
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
3797 bio = bio_alloc(device->bdev, 1,
3798 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
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));
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.
3811 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3812 bio->bi_opf |= REQ_FUA;
3815 if (btrfs_advance_sb_log(device, i))
3818 return errors < i ? 0 : -1;
3822 * Wait for write completion of superblocks done by write_dev_supers,
3823 * @max_mirrors same for write and wait phases.
3825 * Return number of errors when page is not found or not marked up to
3828 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3832 bool primary_failed = false;
3836 if (max_mirrors == 0)
3837 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3839 for (i = 0; i < max_mirrors; i++) {
3842 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3843 if (ret == -ENOENT) {
3845 } else if (ret < 0) {
3848 primary_failed = true;
3851 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3852 device->commit_total_bytes)
3855 page = find_get_page(device->bdev->bd_inode->i_mapping,
3856 bytenr >> PAGE_SHIFT);
3860 primary_failed = true;
3863 /* Page is submitted locked and unlocked once the IO completes */
3864 wait_on_page_locked(page);
3865 if (PageError(page)) {
3868 primary_failed = true;
3871 /* Drop our reference */
3874 /* Drop the reference from the writing run */
3878 /* log error, force error return */
3879 if (primary_failed) {
3880 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3885 return errors < i ? 0 : -1;
3889 * endio for the write_dev_flush, this will wake anyone waiting
3890 * for the barrier when it is done
3892 static void btrfs_end_empty_barrier(struct bio *bio)
3895 complete(bio->bi_private);
3899 * Submit a flush request to the device if it supports it. Error handling is
3900 * done in the waiting counterpart.
3902 static void write_dev_flush(struct btrfs_device *device)
3904 struct bio *bio = &device->flush_bio;
3906 device->last_flush_error = BLK_STS_OK;
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;
3914 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3918 * If the flush bio has been submitted by write_dev_flush, wait for it.
3919 * Return true for any error, and false otherwise.
3921 static bool wait_dev_flush(struct btrfs_device *device)
3923 struct bio *bio = &device->flush_bio;
3925 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3928 wait_for_completion_io(&device->flush_wait);
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);
3940 * send an empty flush down to each device in parallel,
3941 * then wait for them
3943 static int barrier_all_devices(struct btrfs_fs_info *info)
3945 struct list_head *head;
3946 struct btrfs_device *dev;
3947 int errors_wait = 0;
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))
3957 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3958 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3961 write_dev_flush(dev);
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))
3972 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3973 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3976 if (wait_dev_flush(dev))
3981 * Checks last_flush_error of disks in order to determine the device
3984 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3990 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3993 int min_tolerated = INT_MAX;
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);
4001 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4002 if (raid_type == BTRFS_RAID_SINGLE)
4004 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4006 min_tolerated = min_t(int, min_tolerated,
4007 btrfs_raid_array[raid_type].
4008 tolerated_failures);
4011 if (min_tolerated == INT_MAX) {
4012 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4016 return min_tolerated;
4019 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4021 struct list_head *head;
4022 struct btrfs_device *dev;
4023 struct btrfs_super_block *sb;
4024 struct btrfs_dev_item *dev_item;
4028 int total_errors = 0;
4031 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
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.
4038 if (max_mirrors == 0)
4039 backup_super_roots(fs_info);
4041 sb = fs_info->super_for_commit;
4042 dev_item = &sb->dev_item;
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;
4049 ret = barrier_all_devices(fs_info);
4052 &fs_info->fs_devices->device_list_mutex);
4053 btrfs_handle_fs_error(fs_info, ret,
4054 "errors while submitting device barriers.");
4059 list_for_each_entry(dev, head, dev_list) {
4064 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4065 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
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,
4082 flags = btrfs_super_flags(sb);
4083 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4085 ret = btrfs_validate_write_super(fs_info, sb);
4087 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4088 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4089 "unexpected superblock corruption detected");
4093 ret = write_dev_supers(dev, sb, max_mirrors);
4097 if (total_errors > max_errors) {
4098 btrfs_err(fs_info, "%d errors while writing supers",
4100 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
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",
4110 list_for_each_entry(dev, head, dev_list) {
4113 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4114 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4117 ret = wait_dev_supers(dev, max_mirrors);
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",
4131 /* Drop a fs root from the radix tree and free it. */
4132 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4133 struct btrfs_root *root)
4135 bool drop_ref = false;
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))
4142 spin_unlock(&fs_info->fs_roots_radix_lock);
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;
4153 btrfs_put_root(root);
4156 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4158 struct btrfs_root *root = fs_info->tree_root;
4159 struct btrfs_trans_handle *trans;
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);
4166 /* wait until ongoing cleanup work done */
4167 down_write(&fs_info->cleanup_work_sem);
4168 up_write(&fs_info->cleanup_work_sem);
4170 trans = btrfs_join_transaction(root);
4172 return PTR_ERR(trans);
4173 return btrfs_commit_transaction(trans);
4176 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4178 struct btrfs_transaction *trans;
4179 struct btrfs_transaction *tmp;
4182 if (list_empty(&fs_info->trans_list))
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.
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;
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;
4204 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4205 trans->transid, dirty_bytes);
4206 btrfs_cleanup_one_transaction(trans, fs_info);
4208 if (trans == fs_info->running_transaction)
4209 fs_info->running_transaction = NULL;
4210 list_del_init(&trans->list);
4212 btrfs_put_transaction(trans);
4213 trace_btrfs_transaction_commit(fs_info);
4218 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4222 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
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
4233 btrfs_wake_unfinished_drop(fs_info);
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.
4244 cancel_work_sync(&fs_info->reclaim_bgs_work);
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.
4251 kthread_park(fs_info->cleaner_kthread);
4253 /* wait for the qgroup rescan worker to stop */
4254 btrfs_qgroup_wait_for_completion(fs_info, false);
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);
4261 /* pause restriper - we want to resume on mount */
4262 btrfs_pause_balance(fs_info);
4264 btrfs_dev_replace_suspend_for_unmount(fs_info);
4266 btrfs_scrub_cancel(fs_info);
4268 /* wait for any defraggers to finish */
4269 wait_event(fs_info->transaction_wait,
4270 (atomic_read(&fs_info->defrag_running) == 0));
4272 /* clear out the rbtree of defraggable inodes */
4273 btrfs_cleanup_defrag_inodes(fs_info);
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.
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.
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).
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);
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);
4304 /* Cancel or finish ongoing discard work */
4305 btrfs_discard_cleanup(fs_info);
4307 if (!sb_rdonly(fs_info->sb)) {
4309 * The cleaner kthread is stopped, so do one final pass over
4310 * unused block groups.
4312 btrfs_delete_unused_bgs(fs_info);
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.
4325 btrfs_flush_workqueue(fs_info->delayed_workers);
4327 ret = btrfs_commit_super(fs_info);
4329 btrfs_err(fs_info, "commit super ret %d", ret);
4332 if (BTRFS_FS_ERROR(fs_info))
4333 btrfs_error_commit_super(fs_info);
4335 kthread_stop(fs_info->transaction_kthread);
4336 kthread_stop(fs_info->cleaner_kthread);
4338 ASSERT(list_empty(&fs_info->delayed_iputs));
4339 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
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");
4346 btrfs_free_qgroup_config(fs_info);
4347 ASSERT(list_empty(&fs_info->delalloc_roots));
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));
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));
4358 btrfs_sysfs_remove_mounted(fs_info);
4359 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4361 btrfs_put_block_group_cache(fs_info);
4364 * we must make sure there is not any read request to
4365 * submit after we stopping all workers.
4367 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4368 btrfs_stop_all_workers(fs_info);
4370 /* We shouldn't have any transaction open at this point */
4371 warn_about_uncommitted_trans(fs_info);
4373 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4374 free_root_pointers(fs_info, true);
4375 btrfs_free_fs_roots(fs_info);
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.
4384 btrfs_free_block_groups(fs_info);
4386 iput(fs_info->btree_inode);
4388 btrfs_mapping_tree_free(fs_info);
4389 btrfs_close_devices(fs_info->fs_devices);
4392 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4393 struct extent_buffer *buf)
4395 struct btrfs_fs_info *fs_info = buf->fs_info;
4396 u64 transid = btrfs_header_generation(buf);
4398 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
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.
4404 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
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);
4413 "dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4414 buf->start, transid, fs_info->generation);
4416 set_extent_buffer_dirty(buf);
4419 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4423 * looks as though older kernels can get into trouble with
4424 * this code, they end up stuck in balance_dirty_pages forever
4428 if (current->flags & PF_MEMALLOC)
4432 btrfs_balance_delayed_items(fs_info);
4434 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4435 BTRFS_DIRTY_METADATA_THRESH,
4436 fs_info->dirty_metadata_batch);
4438 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4442 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4444 __btrfs_btree_balance_dirty(fs_info, 1);
4447 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4449 __btrfs_btree_balance_dirty(fs_info, 0);
4452 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4454 /* cleanup FS via transaction */
4455 btrfs_cleanup_transaction(fs_info);
4457 mutex_lock(&fs_info->cleaner_mutex);
4458 btrfs_run_delayed_iputs(fs_info);
4459 mutex_unlock(&fs_info->cleaner_mutex);
4461 down_write(&fs_info->cleanup_work_sem);
4462 up_write(&fs_info->cleanup_work_sem);
4465 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4467 struct btrfs_root *gang[8];
4468 u64 root_objectid = 0;
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) {
4477 for (i = 0; i < ret; i++)
4478 gang[i] = btrfs_grab_root(gang[i]);
4479 spin_unlock(&fs_info->fs_roots_radix_lock);
4481 for (i = 0; i < ret; i++) {
4484 root_objectid = gang[i]->root_key.objectid;
4485 btrfs_free_log(NULL, gang[i]);
4486 btrfs_put_root(gang[i]);
4489 spin_lock(&fs_info->fs_roots_radix_lock);
4491 spin_unlock(&fs_info->fs_roots_radix_lock);
4492 btrfs_free_log_root_tree(NULL, fs_info);
4495 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4497 struct btrfs_ordered_extent *ordered;
4499 spin_lock(&root->ordered_extent_lock);
4501 * This will just short circuit the ordered completion stuff which will
4502 * make sure the ordered extent gets properly cleaned up.
4504 list_for_each_entry(ordered, &root->ordered_extents,
4506 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4507 spin_unlock(&root->ordered_extent_lock);
4510 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4512 struct btrfs_root *root;
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,
4520 list_move_tail(&root->ordered_root,
4521 &fs_info->ordered_roots);
4523 spin_unlock(&fs_info->ordered_root_lock);
4524 btrfs_destroy_ordered_extents(root);
4527 spin_lock(&fs_info->ordered_root_lock);
4529 spin_unlock(&fs_info->ordered_root_lock);
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.
4537 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4540 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4541 struct btrfs_fs_info *fs_info)
4543 struct rb_node *node;
4544 struct btrfs_delayed_ref_root *delayed_refs;
4545 struct btrfs_delayed_ref_node *ref;
4547 delayed_refs = &trans->delayed_refs;
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");
4556 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4557 struct btrfs_delayed_ref_head *head;
4559 bool pin_bytes = false;
4561 head = rb_entry(node, struct btrfs_delayed_ref_head,
4563 if (btrfs_delayed_ref_lock(delayed_refs, head))
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,
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);
4578 if (head->must_insert_reserved)
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);
4587 struct btrfs_block_group *cache;
4589 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
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);
4602 btrfs_put_block_group(cache);
4604 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4605 head->bytenr + head->num_bytes - 1);
4607 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4608 btrfs_put_delayed_ref_head(head);
4610 spin_lock(&delayed_refs->lock);
4612 btrfs_qgroup_destroy_extent_records(trans);
4614 spin_unlock(&delayed_refs->lock);
4617 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4619 struct btrfs_inode *btrfs_inode;
4622 spin_lock(&root->delalloc_lock);
4623 list_splice_init(&root->delalloc_inodes, &splice);
4625 while (!list_empty(&splice)) {
4626 struct inode *inode = NULL;
4627 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4629 __btrfs_del_delalloc_inode(root, btrfs_inode);
4630 spin_unlock(&root->delalloc_lock);
4633 * Make sure we get a live inode and that it'll not disappear
4636 inode = igrab(&btrfs_inode->vfs_inode);
4638 unsigned int nofs_flag;
4640 nofs_flag = memalloc_nofs_save();
4641 invalidate_inode_pages2(inode->i_mapping);
4642 memalloc_nofs_restore(nofs_flag);
4645 spin_lock(&root->delalloc_lock);
4647 spin_unlock(&root->delalloc_lock);
4650 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4652 struct btrfs_root *root;
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,
4660 root = btrfs_grab_root(root);
4662 spin_unlock(&fs_info->delalloc_root_lock);
4664 btrfs_destroy_delalloc_inodes(root);
4665 btrfs_put_root(root);
4667 spin_lock(&fs_info->delalloc_root_lock);
4669 spin_unlock(&fs_info->delalloc_root_lock);
4672 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4673 struct extent_io_tree *dirty_pages,
4676 struct extent_buffer *eb;
4680 while (find_first_extent_bit(dirty_pages, start, &start, &end,
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;
4689 btrfs_tree_lock(eb);
4690 wait_on_extent_buffer_writeback(eb);
4691 btrfs_clear_buffer_dirty(NULL, eb);
4692 btrfs_tree_unlock(eb);
4694 free_extent_buffer_stale(eb);
4699 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4700 struct extent_io_tree *unpin)
4706 struct extent_state *cached_state = NULL;
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.
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);
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);
4729 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4731 struct inode *inode;
4733 inode = cache->io_ctl.inode;
4735 unsigned int nofs_flag;
4737 nofs_flag = memalloc_nofs_save();
4738 invalidate_inode_pages2(inode->i_mapping);
4739 memalloc_nofs_restore(nofs_flag);
4741 BTRFS_I(inode)->generation = 0;
4742 cache->io_ctl.inode = NULL;
4745 ASSERT(cache->io_ctl.pages == NULL);
4746 btrfs_put_block_group(cache);
4749 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4750 struct btrfs_fs_info *fs_info)
4752 struct btrfs_block_group *cache;
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,
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);
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);
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);
4777 spin_unlock(&cur_trans->dirty_bgs_lock);
4780 * Refer to the definition of io_bgs member for details why it's safe
4781 * to use it without any locking
4783 while (!list_empty(&cur_trans->io_bgs)) {
4784 cache = list_first_entry(&cur_trans->io_bgs,
4785 struct btrfs_block_group,
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);
4796 static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4798 struct btrfs_root *gang[8];
4802 spin_lock(&fs_info->fs_roots_radix_lock);
4804 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4807 BTRFS_ROOT_TRANS_TAG);
4810 for (i = 0; i < ret; i++) {
4811 struct btrfs_root *root = gang[i];
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);
4819 spin_unlock(&fs_info->fs_roots_radix_lock);
4822 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4823 struct btrfs_fs_info *fs_info)
4825 struct btrfs_device *dev, *tmp;
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));
4831 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4833 list_del_init(&dev->post_commit_list);
4836 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4838 cur_trans->state = TRANS_STATE_COMMIT_START;
4839 wake_up(&fs_info->transaction_blocked_wait);
4841 cur_trans->state = TRANS_STATE_UNBLOCKED;
4842 wake_up(&fs_info->transaction_wait);
4844 btrfs_destroy_delayed_inodes(fs_info);
4846 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4848 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4850 btrfs_free_all_qgroup_pertrans(fs_info);
4852 cur_trans->state =TRANS_STATE_COMPLETED;
4853 wake_up(&cur_trans->commit_wait);
4856 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4858 struct btrfs_transaction *t;
4860 mutex_lock(&fs_info->transaction_kthread_mutex);
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);
4874 if (t == fs_info->running_transaction) {
4875 t->state = TRANS_STATE_COMMIT_DOING;
4876 spin_unlock(&fs_info->trans_lock);
4878 * We wait for 0 num_writers since we don't hold a trans
4879 * handle open currently for this transaction.
4881 wait_event(t->writer_wait,
4882 atomic_read(&t->num_writers) == 0);
4884 spin_unlock(&fs_info->trans_lock);
4886 btrfs_cleanup_one_transaction(t, fs_info);
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);
4894 btrfs_put_transaction(t);
4895 trace_btrfs_transaction_commit(fs_info);
4896 spin_lock(&fs_info->trans_lock);
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);
4909 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4911 struct btrfs_path *path;
4913 struct extent_buffer *l;
4914 struct btrfs_key search_key;
4915 struct btrfs_key found_key;
4918 path = btrfs_alloc_path();
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);
4928 BUG_ON(ret == 0); /* Corruption */
4929 if (path->slots[0] > 0) {
4930 slot = path->slots[0] - 1;
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);
4936 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4940 btrfs_free_path(path);
4944 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4947 mutex_lock(&root->objectid_mutex);
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);
4957 *objectid = root->free_objectid++;
4960 mutex_unlock(&root->objectid_mutex);