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 "dev-replace.h"
36 #include "compression.h"
37 #include "tree-checker.h"
38 #include "ref-verify.h"
39 #include "block-group.h"
41 #include "space-info.h"
45 #include "accessors.h"
46 #include "extent-tree.h"
47 #include "root-tree.h"
49 #include "uuid-tree.h"
50 #include "relocation.h"
54 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
55 BTRFS_HEADER_FLAG_RELOC |\
56 BTRFS_SUPER_FLAG_ERROR |\
57 BTRFS_SUPER_FLAG_SEEDING |\
58 BTRFS_SUPER_FLAG_METADUMP |\
59 BTRFS_SUPER_FLAG_METADUMP_V2)
61 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
62 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
64 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
66 if (fs_info->csum_shash)
67 crypto_free_shash(fs_info->csum_shash);
71 * Compute the csum of a btree block and store the result to provided buffer.
73 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
75 struct btrfs_fs_info *fs_info = buf->fs_info;
78 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
82 shash->tfm = fs_info->csum_shash;
83 crypto_shash_init(shash);
86 /* Pages are contiguous, handle them as a big one. */
88 first_page_part = fs_info->nodesize;
91 kaddr = folio_address(buf->folios[0]);
92 first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
93 num_pages = num_extent_pages(buf);
96 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
97 first_page_part - BTRFS_CSUM_SIZE);
100 * Multiple single-page folios case would reach here.
102 * nodesize <= PAGE_SIZE and large folio all handled by above
103 * crypto_shash_update() already.
105 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
106 kaddr = folio_address(buf->folios[i]);
107 crypto_shash_update(shash, kaddr, PAGE_SIZE);
109 memset(result, 0, BTRFS_CSUM_SIZE);
110 crypto_shash_final(shash, result);
114 * we can't consider a given block up to date unless the transid of the
115 * block matches the transid in the parent node's pointer. This is how we
116 * detect blocks that either didn't get written at all or got written
117 * in the wrong place.
119 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
121 if (!extent_buffer_uptodate(eb))
124 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
130 if (!extent_buffer_uptodate(eb) ||
131 btrfs_header_generation(eb) != parent_transid) {
132 btrfs_err_rl(eb->fs_info,
133 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
134 eb->start, eb->read_mirror,
135 parent_transid, btrfs_header_generation(eb));
136 clear_extent_buffer_uptodate(eb);
142 static bool btrfs_supported_super_csum(u16 csum_type)
145 case BTRFS_CSUM_TYPE_CRC32:
146 case BTRFS_CSUM_TYPE_XXHASH:
147 case BTRFS_CSUM_TYPE_SHA256:
148 case BTRFS_CSUM_TYPE_BLAKE2:
156 * Return 0 if the superblock checksum type matches the checksum value of that
157 * algorithm. Pass the raw disk superblock data.
159 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
160 const struct btrfs_super_block *disk_sb)
162 char result[BTRFS_CSUM_SIZE];
163 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
165 shash->tfm = fs_info->csum_shash;
168 * The super_block structure does not span the whole
169 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
170 * filled with zeros and is included in the checksum.
172 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
173 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
175 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
181 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
184 struct btrfs_fs_info *fs_info = eb->fs_info;
185 int num_folios = num_extent_folios(eb);
188 if (sb_rdonly(fs_info->sb))
191 for (int i = 0; i < num_folios; i++) {
192 struct folio *folio = eb->folios[i];
193 u64 start = max_t(u64, eb->start, folio_pos(folio));
194 u64 end = min_t(u64, eb->start + eb->len,
195 folio_pos(folio) + eb->folio_size);
196 u32 len = end - start;
198 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
199 start, folio, offset_in_folio(folio, start),
209 * helper to read a given tree block, doing retries as required when
210 * the checksums don't match and we have alternate mirrors to try.
212 * @check: expected tree parentness check, see the comments of the
213 * structure for details.
215 int btrfs_read_extent_buffer(struct extent_buffer *eb,
216 struct btrfs_tree_parent_check *check)
218 struct btrfs_fs_info *fs_info = eb->fs_info;
223 int failed_mirror = 0;
228 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
229 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
233 num_copies = btrfs_num_copies(fs_info,
238 if (!failed_mirror) {
240 failed_mirror = eb->read_mirror;
244 if (mirror_num == failed_mirror)
247 if (mirror_num > num_copies)
251 if (failed && !ret && failed_mirror)
252 btrfs_repair_eb_io_failure(eb, failed_mirror);
258 * Checksum a dirty tree block before IO.
260 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
262 struct extent_buffer *eb = bbio->private;
263 struct btrfs_fs_info *fs_info = eb->fs_info;
264 u64 found_start = btrfs_header_bytenr(eb);
266 u8 result[BTRFS_CSUM_SIZE];
269 /* Btree blocks are always contiguous on disk. */
270 if (WARN_ON_ONCE(bbio->file_offset != eb->start))
271 return BLK_STS_IOERR;
272 if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
273 return BLK_STS_IOERR;
276 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
277 * checksum it but zero-out its content. This is done to preserve
278 * ordering of I/O without unnecessarily writing out data.
280 if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
281 memzero_extent_buffer(eb, 0, eb->len);
285 if (WARN_ON_ONCE(found_start != eb->start))
286 return BLK_STS_IOERR;
287 if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
288 eb->start, eb->len)))
289 return BLK_STS_IOERR;
291 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
292 offsetof(struct btrfs_header, fsid),
293 BTRFS_FSID_SIZE) == 0);
294 csum_tree_block(eb, result);
296 if (btrfs_header_level(eb))
297 ret = btrfs_check_node(eb);
299 ret = btrfs_check_leaf(eb);
305 * Also check the generation, the eb reached here must be newer than
306 * last committed. Or something seriously wrong happened.
308 last_trans = btrfs_get_last_trans_committed(fs_info);
309 if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
312 "block=%llu bad generation, have %llu expect > %llu",
313 eb->start, btrfs_header_generation(eb), last_trans);
316 write_extent_buffer(eb, result, 0, fs_info->csum_size);
320 btrfs_print_tree(eb, 0);
321 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
324 * Be noisy if this is an extent buffer from a log tree. We don't abort
325 * a transaction in case there's a bad log tree extent buffer, we just
326 * fallback to a transaction commit. Still we want to know when there is
327 * a bad log tree extent buffer, as that may signal a bug somewhere.
329 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
330 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
331 return errno_to_blk_status(ret);
334 static bool check_tree_block_fsid(struct extent_buffer *eb)
336 struct btrfs_fs_info *fs_info = eb->fs_info;
337 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
338 u8 fsid[BTRFS_FSID_SIZE];
340 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
344 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
345 * This is then overwritten by metadata_uuid if it is present in the
346 * device_list_add(). The same true for a seed device as well. So use of
347 * fs_devices::metadata_uuid is appropriate here.
349 if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
352 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
353 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
359 /* Do basic extent buffer checks at read time */
360 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
361 struct btrfs_tree_parent_check *check)
363 struct btrfs_fs_info *fs_info = eb->fs_info;
365 const u32 csum_size = fs_info->csum_size;
367 u8 result[BTRFS_CSUM_SIZE];
368 const u8 *header_csum;
373 found_start = btrfs_header_bytenr(eb);
374 if (found_start != eb->start) {
375 btrfs_err_rl(fs_info,
376 "bad tree block start, mirror %u want %llu have %llu",
377 eb->read_mirror, eb->start, found_start);
381 if (check_tree_block_fsid(eb)) {
382 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
383 eb->start, eb->read_mirror);
387 found_level = btrfs_header_level(eb);
388 if (found_level >= BTRFS_MAX_LEVEL) {
390 "bad tree block level, mirror %u level %d on logical %llu",
391 eb->read_mirror, btrfs_header_level(eb), eb->start);
396 csum_tree_block(eb, result);
397 header_csum = folio_address(eb->folios[0]) +
398 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
400 if (memcmp(result, header_csum, csum_size) != 0) {
401 btrfs_warn_rl(fs_info,
402 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
403 eb->start, eb->read_mirror,
404 CSUM_FMT_VALUE(csum_size, header_csum),
405 CSUM_FMT_VALUE(csum_size, result),
406 btrfs_header_level(eb));
411 if (found_level != check->level) {
413 "level verify failed on logical %llu mirror %u wanted %u found %u",
414 eb->start, eb->read_mirror, check->level, found_level);
418 if (unlikely(check->transid &&
419 btrfs_header_generation(eb) != check->transid)) {
420 btrfs_err_rl(eb->fs_info,
421 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
422 eb->start, eb->read_mirror, check->transid,
423 btrfs_header_generation(eb));
427 if (check->has_first_key) {
428 struct btrfs_key *expect_key = &check->first_key;
429 struct btrfs_key found_key;
432 btrfs_node_key_to_cpu(eb, &found_key, 0);
434 btrfs_item_key_to_cpu(eb, &found_key, 0);
435 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
437 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
438 eb->start, check->transid,
439 expect_key->objectid,
440 expect_key->type, expect_key->offset,
441 found_key.objectid, found_key.type,
447 if (check->owner_root) {
448 ret = btrfs_check_eb_owner(eb, check->owner_root);
454 * If this is a leaf block and it is corrupt, set the corrupt bit so
455 * that we don't try and read the other copies of this block, just
458 if (found_level == 0 && btrfs_check_leaf(eb)) {
459 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
463 if (found_level > 0 && btrfs_check_node(eb))
468 "read time tree block corruption detected on logical %llu mirror %u",
469 eb->start, eb->read_mirror);
474 #ifdef CONFIG_MIGRATION
475 static int btree_migrate_folio(struct address_space *mapping,
476 struct folio *dst, struct folio *src, enum migrate_mode mode)
479 * we can't safely write a btree page from here,
480 * we haven't done the locking hook
482 if (folio_test_dirty(src))
485 * Buffers may be managed in a filesystem specific way.
486 * We must have no buffers or drop them.
488 if (folio_get_private(src) &&
489 !filemap_release_folio(src, GFP_KERNEL))
491 return migrate_folio(mapping, dst, src, mode);
494 #define btree_migrate_folio NULL
497 static int btree_writepages(struct address_space *mapping,
498 struct writeback_control *wbc)
502 if (wbc->sync_mode == WB_SYNC_NONE) {
503 struct btrfs_fs_info *fs_info;
505 if (wbc->for_kupdate)
508 fs_info = inode_to_fs_info(mapping->host);
509 /* this is a bit racy, but that's ok */
510 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
511 BTRFS_DIRTY_METADATA_THRESH,
512 fs_info->dirty_metadata_batch);
516 return btree_write_cache_pages(mapping, wbc);
519 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
521 if (folio_test_writeback(folio) || folio_test_dirty(folio))
524 return try_release_extent_buffer(&folio->page);
527 static void btree_invalidate_folio(struct folio *folio, size_t offset,
530 struct extent_io_tree *tree;
532 tree = &folio_to_inode(folio)->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(folio_to_fs_info(folio),
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 = inode_to_fs_info(mapping->host);
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));
1248 while (refcount_read(&root->refs) > 1)
1249 btrfs_put_root(root);
1250 btrfs_put_root(root);
1255 static void free_global_roots(struct btrfs_fs_info *fs_info)
1257 struct btrfs_root *root;
1258 struct rb_node *node;
1260 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1261 root = rb_entry(node, struct btrfs_root, rb_node);
1262 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1263 btrfs_put_root(root);
1267 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1269 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1270 percpu_counter_destroy(&fs_info->delalloc_bytes);
1271 percpu_counter_destroy(&fs_info->ordered_bytes);
1272 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1273 btrfs_free_csum_hash(fs_info);
1274 btrfs_free_stripe_hash_table(fs_info);
1275 btrfs_free_ref_cache(fs_info);
1276 kfree(fs_info->balance_ctl);
1277 kfree(fs_info->delayed_root);
1278 free_global_roots(fs_info);
1279 btrfs_put_root(fs_info->tree_root);
1280 btrfs_put_root(fs_info->chunk_root);
1281 btrfs_put_root(fs_info->dev_root);
1282 btrfs_put_root(fs_info->quota_root);
1283 btrfs_put_root(fs_info->uuid_root);
1284 btrfs_put_root(fs_info->fs_root);
1285 btrfs_put_root(fs_info->data_reloc_root);
1286 btrfs_put_root(fs_info->block_group_root);
1287 btrfs_put_root(fs_info->stripe_root);
1288 btrfs_check_leaked_roots(fs_info);
1289 btrfs_extent_buffer_leak_debug_check(fs_info);
1290 kfree(fs_info->super_copy);
1291 kfree(fs_info->super_for_commit);
1292 kfree(fs_info->subpage_info);
1298 * Get an in-memory reference of a root structure.
1300 * For essential trees like root/extent tree, we grab it from fs_info directly.
1301 * For subvolume trees, we check the cached filesystem roots first. If not
1302 * found, then read it from disk and add it to cached fs roots.
1304 * Caller should release the root by calling btrfs_put_root() after the usage.
1306 * NOTE: Reloc and log trees can't be read by this function as they share the
1307 * same root objectid.
1309 * @objectid: root id
1310 * @anon_dev: preallocated anonymous block device number for new roots,
1311 * pass NULL for a new allocation.
1312 * @check_ref: whether to check root item references, If true, return -ENOENT
1315 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1316 u64 objectid, dev_t *anon_dev,
1319 struct btrfs_root *root;
1320 struct btrfs_path *path;
1321 struct btrfs_key key;
1324 root = btrfs_get_global_root(fs_info, objectid);
1329 * If we're called for non-subvolume trees, and above function didn't
1330 * find one, do not try to read it from disk.
1332 * This is namely for free-space-tree and quota tree, which can change
1333 * at runtime and should only be grabbed from fs_info.
1335 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1336 return ERR_PTR(-ENOENT);
1338 root = btrfs_lookup_fs_root(fs_info, objectid);
1341 * Some other caller may have read out the newly inserted
1342 * subvolume already (for things like backref walk etc). Not
1343 * that common but still possible. In that case, we just need
1344 * to free the anon_dev.
1346 if (unlikely(anon_dev && *anon_dev)) {
1347 free_anon_bdev(*anon_dev);
1351 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1352 btrfs_put_root(root);
1353 return ERR_PTR(-ENOENT);
1358 key.objectid = objectid;
1359 key.type = BTRFS_ROOT_ITEM_KEY;
1360 key.offset = (u64)-1;
1361 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1365 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1370 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1374 path = btrfs_alloc_path();
1379 key.objectid = BTRFS_ORPHAN_OBJECTID;
1380 key.type = BTRFS_ORPHAN_ITEM_KEY;
1381 key.offset = objectid;
1383 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1384 btrfs_free_path(path);
1388 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1390 ret = btrfs_insert_fs_root(fs_info, root);
1392 if (ret == -EEXIST) {
1393 btrfs_put_root(root);
1401 * If our caller provided us an anonymous device, then it's his
1402 * responsibility to free it in case we fail. So we have to set our
1403 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1404 * and once again by our caller.
1406 if (anon_dev && *anon_dev)
1408 btrfs_put_root(root);
1409 return ERR_PTR(ret);
1413 * Get in-memory reference of a root structure
1415 * @objectid: tree objectid
1416 * @check_ref: if set, verify that the tree exists and the item has at least
1419 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1420 u64 objectid, bool check_ref)
1422 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1426 * Get in-memory reference of a root structure, created as new, optionally pass
1427 * the anonymous block device id
1429 * @objectid: tree objectid
1430 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1431 * parameter value if not NULL
1433 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1434 u64 objectid, dev_t *anon_dev)
1436 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1440 * Return a root for the given objectid.
1442 * @fs_info: the fs_info
1443 * @objectid: the objectid we need to lookup
1445 * This is exclusively used for backref walking, and exists specifically because
1446 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1447 * creation time, which means we may have to read the tree_root in order to look
1448 * up a fs root that is not in memory. If the root is not in memory we will
1449 * read the tree root commit root and look up the fs root from there. This is a
1450 * temporary root, it will not be inserted into the radix tree as it doesn't
1451 * have the most uptodate information, it'll simply be discarded once the
1452 * backref code is finished using the root.
1454 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1455 struct btrfs_path *path,
1458 struct btrfs_root *root;
1459 struct btrfs_key key;
1461 ASSERT(path->search_commit_root && path->skip_locking);
1464 * This can return -ENOENT if we ask for a root that doesn't exist, but
1465 * since this is called via the backref walking code we won't be looking
1466 * up a root that doesn't exist, unless there's corruption. So if root
1467 * != NULL just return it.
1469 root = btrfs_get_global_root(fs_info, objectid);
1473 root = btrfs_lookup_fs_root(fs_info, objectid);
1477 key.objectid = objectid;
1478 key.type = BTRFS_ROOT_ITEM_KEY;
1479 key.offset = (u64)-1;
1480 root = read_tree_root_path(fs_info->tree_root, path, &key);
1481 btrfs_release_path(path);
1486 static int cleaner_kthread(void *arg)
1488 struct btrfs_fs_info *fs_info = arg;
1494 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1496 /* Make the cleaner go to sleep early. */
1497 if (btrfs_need_cleaner_sleep(fs_info))
1501 * Do not do anything if we might cause open_ctree() to block
1502 * before we have finished mounting the filesystem.
1504 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1507 if (!mutex_trylock(&fs_info->cleaner_mutex))
1511 * Avoid the problem that we change the status of the fs
1512 * during the above check and trylock.
1514 if (btrfs_need_cleaner_sleep(fs_info)) {
1515 mutex_unlock(&fs_info->cleaner_mutex);
1519 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1520 btrfs_sysfs_feature_update(fs_info);
1522 btrfs_run_delayed_iputs(fs_info);
1524 again = btrfs_clean_one_deleted_snapshot(fs_info);
1525 mutex_unlock(&fs_info->cleaner_mutex);
1528 * The defragger has dealt with the R/O remount and umount,
1529 * needn't do anything special here.
1531 btrfs_run_defrag_inodes(fs_info);
1534 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1535 * with relocation (btrfs_relocate_chunk) and relocation
1536 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1537 * after acquiring fs_info->reclaim_bgs_lock. So we
1538 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1539 * unused block groups.
1541 btrfs_delete_unused_bgs(fs_info);
1544 * Reclaim block groups in the reclaim_bgs list after we deleted
1545 * all unused block_groups. This possibly gives us some more free
1548 btrfs_reclaim_bgs(fs_info);
1550 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1551 if (kthread_should_park())
1553 if (kthread_should_stop())
1556 set_current_state(TASK_INTERRUPTIBLE);
1558 __set_current_state(TASK_RUNNING);
1563 static int transaction_kthread(void *arg)
1565 struct btrfs_root *root = arg;
1566 struct btrfs_fs_info *fs_info = root->fs_info;
1567 struct btrfs_trans_handle *trans;
1568 struct btrfs_transaction *cur;
1571 unsigned long delay;
1575 cannot_commit = false;
1576 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1577 mutex_lock(&fs_info->transaction_kthread_mutex);
1579 spin_lock(&fs_info->trans_lock);
1580 cur = fs_info->running_transaction;
1582 spin_unlock(&fs_info->trans_lock);
1586 delta = ktime_get_seconds() - cur->start_time;
1587 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1588 cur->state < TRANS_STATE_COMMIT_PREP &&
1589 delta < fs_info->commit_interval) {
1590 spin_unlock(&fs_info->trans_lock);
1591 delay -= msecs_to_jiffies((delta - 1) * 1000);
1593 msecs_to_jiffies(fs_info->commit_interval * 1000));
1596 transid = cur->transid;
1597 spin_unlock(&fs_info->trans_lock);
1599 /* If the file system is aborted, this will always fail. */
1600 trans = btrfs_attach_transaction(root);
1601 if (IS_ERR(trans)) {
1602 if (PTR_ERR(trans) != -ENOENT)
1603 cannot_commit = true;
1606 if (transid == trans->transid) {
1607 btrfs_commit_transaction(trans);
1609 btrfs_end_transaction(trans);
1612 wake_up_process(fs_info->cleaner_kthread);
1613 mutex_unlock(&fs_info->transaction_kthread_mutex);
1615 if (BTRFS_FS_ERROR(fs_info))
1616 btrfs_cleanup_transaction(fs_info);
1617 if (!kthread_should_stop() &&
1618 (!btrfs_transaction_blocked(fs_info) ||
1620 schedule_timeout_interruptible(delay);
1621 } while (!kthread_should_stop());
1626 * This will find the highest generation in the array of root backups. The
1627 * index of the highest array is returned, or -EINVAL if we can't find
1630 * We check to make sure the array is valid by comparing the
1631 * generation of the latest root in the array with the generation
1632 * in the super block. If they don't match we pitch it.
1634 static int find_newest_super_backup(struct btrfs_fs_info *info)
1636 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1638 struct btrfs_root_backup *root_backup;
1641 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1642 root_backup = info->super_copy->super_roots + i;
1643 cur = btrfs_backup_tree_root_gen(root_backup);
1644 if (cur == newest_gen)
1652 * copy all the root pointers into the super backup array.
1653 * this will bump the backup pointer by one when it is
1656 static void backup_super_roots(struct btrfs_fs_info *info)
1658 const int next_backup = info->backup_root_index;
1659 struct btrfs_root_backup *root_backup;
1661 root_backup = info->super_for_commit->super_roots + next_backup;
1664 * make sure all of our padding and empty slots get zero filled
1665 * regardless of which ones we use today
1667 memset(root_backup, 0, sizeof(*root_backup));
1669 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1671 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1672 btrfs_set_backup_tree_root_gen(root_backup,
1673 btrfs_header_generation(info->tree_root->node));
1675 btrfs_set_backup_tree_root_level(root_backup,
1676 btrfs_header_level(info->tree_root->node));
1678 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1679 btrfs_set_backup_chunk_root_gen(root_backup,
1680 btrfs_header_generation(info->chunk_root->node));
1681 btrfs_set_backup_chunk_root_level(root_backup,
1682 btrfs_header_level(info->chunk_root->node));
1684 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1685 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1686 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1688 btrfs_set_backup_extent_root(root_backup,
1689 extent_root->node->start);
1690 btrfs_set_backup_extent_root_gen(root_backup,
1691 btrfs_header_generation(extent_root->node));
1692 btrfs_set_backup_extent_root_level(root_backup,
1693 btrfs_header_level(extent_root->node));
1695 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1696 btrfs_set_backup_csum_root_gen(root_backup,
1697 btrfs_header_generation(csum_root->node));
1698 btrfs_set_backup_csum_root_level(root_backup,
1699 btrfs_header_level(csum_root->node));
1703 * we might commit during log recovery, which happens before we set
1704 * the fs_root. Make sure it is valid before we fill it in.
1706 if (info->fs_root && info->fs_root->node) {
1707 btrfs_set_backup_fs_root(root_backup,
1708 info->fs_root->node->start);
1709 btrfs_set_backup_fs_root_gen(root_backup,
1710 btrfs_header_generation(info->fs_root->node));
1711 btrfs_set_backup_fs_root_level(root_backup,
1712 btrfs_header_level(info->fs_root->node));
1715 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1716 btrfs_set_backup_dev_root_gen(root_backup,
1717 btrfs_header_generation(info->dev_root->node));
1718 btrfs_set_backup_dev_root_level(root_backup,
1719 btrfs_header_level(info->dev_root->node));
1721 btrfs_set_backup_total_bytes(root_backup,
1722 btrfs_super_total_bytes(info->super_copy));
1723 btrfs_set_backup_bytes_used(root_backup,
1724 btrfs_super_bytes_used(info->super_copy));
1725 btrfs_set_backup_num_devices(root_backup,
1726 btrfs_super_num_devices(info->super_copy));
1729 * if we don't copy this out to the super_copy, it won't get remembered
1730 * for the next commit
1732 memcpy(&info->super_copy->super_roots,
1733 &info->super_for_commit->super_roots,
1734 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1738 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1739 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1741 * @fs_info: filesystem whose backup roots need to be read
1742 * @priority: priority of backup root required
1744 * Returns backup root index on success and -EINVAL otherwise.
1746 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1748 int backup_index = find_newest_super_backup(fs_info);
1749 struct btrfs_super_block *super = fs_info->super_copy;
1750 struct btrfs_root_backup *root_backup;
1752 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1754 return backup_index;
1756 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1757 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1762 root_backup = super->super_roots + backup_index;
1764 btrfs_set_super_generation(super,
1765 btrfs_backup_tree_root_gen(root_backup));
1766 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1767 btrfs_set_super_root_level(super,
1768 btrfs_backup_tree_root_level(root_backup));
1769 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1772 * Fixme: the total bytes and num_devices need to match or we should
1775 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1776 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1778 return backup_index;
1781 /* helper to cleanup workers */
1782 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1784 btrfs_destroy_workqueue(fs_info->fixup_workers);
1785 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1786 btrfs_destroy_workqueue(fs_info->workers);
1787 if (fs_info->endio_workers)
1788 destroy_workqueue(fs_info->endio_workers);
1789 if (fs_info->rmw_workers)
1790 destroy_workqueue(fs_info->rmw_workers);
1791 if (fs_info->compressed_write_workers)
1792 destroy_workqueue(fs_info->compressed_write_workers);
1793 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1794 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1795 btrfs_destroy_workqueue(fs_info->delayed_workers);
1796 btrfs_destroy_workqueue(fs_info->caching_workers);
1797 btrfs_destroy_workqueue(fs_info->flush_workers);
1798 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1799 if (fs_info->discard_ctl.discard_workers)
1800 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1802 * Now that all other work queues are destroyed, we can safely destroy
1803 * the queues used for metadata I/O, since tasks from those other work
1804 * queues can do metadata I/O operations.
1806 if (fs_info->endio_meta_workers)
1807 destroy_workqueue(fs_info->endio_meta_workers);
1810 static void free_root_extent_buffers(struct btrfs_root *root)
1813 free_extent_buffer(root->node);
1814 free_extent_buffer(root->commit_root);
1816 root->commit_root = NULL;
1820 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1822 struct btrfs_root *root, *tmp;
1824 rbtree_postorder_for_each_entry_safe(root, tmp,
1825 &fs_info->global_root_tree,
1827 free_root_extent_buffers(root);
1830 /* helper to cleanup tree roots */
1831 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1833 free_root_extent_buffers(info->tree_root);
1835 free_global_root_pointers(info);
1836 free_root_extent_buffers(info->dev_root);
1837 free_root_extent_buffers(info->quota_root);
1838 free_root_extent_buffers(info->uuid_root);
1839 free_root_extent_buffers(info->fs_root);
1840 free_root_extent_buffers(info->data_reloc_root);
1841 free_root_extent_buffers(info->block_group_root);
1842 free_root_extent_buffers(info->stripe_root);
1843 if (free_chunk_root)
1844 free_root_extent_buffers(info->chunk_root);
1847 void btrfs_put_root(struct btrfs_root *root)
1852 if (refcount_dec_and_test(&root->refs)) {
1853 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1854 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1856 free_anon_bdev(root->anon_dev);
1857 free_root_extent_buffers(root);
1858 #ifdef CONFIG_BTRFS_DEBUG
1859 spin_lock(&root->fs_info->fs_roots_radix_lock);
1860 list_del_init(&root->leak_list);
1861 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1867 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1870 struct btrfs_root *gang[8];
1873 while (!list_empty(&fs_info->dead_roots)) {
1874 gang[0] = list_entry(fs_info->dead_roots.next,
1875 struct btrfs_root, root_list);
1876 list_del(&gang[0]->root_list);
1878 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1879 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1880 btrfs_put_root(gang[0]);
1884 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1889 for (i = 0; i < ret; i++)
1890 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1894 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1896 mutex_init(&fs_info->scrub_lock);
1897 atomic_set(&fs_info->scrubs_running, 0);
1898 atomic_set(&fs_info->scrub_pause_req, 0);
1899 atomic_set(&fs_info->scrubs_paused, 0);
1900 atomic_set(&fs_info->scrub_cancel_req, 0);
1901 init_waitqueue_head(&fs_info->scrub_pause_wait);
1902 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1905 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1907 spin_lock_init(&fs_info->balance_lock);
1908 mutex_init(&fs_info->balance_mutex);
1909 atomic_set(&fs_info->balance_pause_req, 0);
1910 atomic_set(&fs_info->balance_cancel_req, 0);
1911 fs_info->balance_ctl = NULL;
1912 init_waitqueue_head(&fs_info->balance_wait_q);
1913 atomic_set(&fs_info->reloc_cancel_req, 0);
1916 static int btrfs_init_btree_inode(struct super_block *sb)
1918 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1919 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1920 fs_info->tree_root);
1921 struct inode *inode;
1923 inode = new_inode(sb);
1927 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1928 set_nlink(inode, 1);
1930 * we set the i_size on the btree inode to the max possible int.
1931 * the real end of the address space is determined by all of
1932 * the devices in the system
1934 inode->i_size = OFFSET_MAX;
1935 inode->i_mapping->a_ops = &btree_aops;
1936 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1938 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1939 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1940 IO_TREE_BTREE_INODE_IO);
1941 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1943 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1944 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1945 BTRFS_I(inode)->location.type = 0;
1946 BTRFS_I(inode)->location.offset = 0;
1947 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1948 __insert_inode_hash(inode, hash);
1949 fs_info->btree_inode = inode;
1954 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1956 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1957 init_rwsem(&fs_info->dev_replace.rwsem);
1958 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1961 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1963 spin_lock_init(&fs_info->qgroup_lock);
1964 mutex_init(&fs_info->qgroup_ioctl_lock);
1965 fs_info->qgroup_tree = RB_ROOT;
1966 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1967 fs_info->qgroup_seq = 1;
1968 fs_info->qgroup_ulist = NULL;
1969 fs_info->qgroup_rescan_running = false;
1970 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1971 mutex_init(&fs_info->qgroup_rescan_lock);
1974 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1976 u32 max_active = fs_info->thread_pool_size;
1977 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1978 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1981 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1983 fs_info->delalloc_workers =
1984 btrfs_alloc_workqueue(fs_info, "delalloc",
1985 flags, max_active, 2);
1987 fs_info->flush_workers =
1988 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1989 flags, max_active, 0);
1991 fs_info->caching_workers =
1992 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1994 fs_info->fixup_workers =
1995 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1997 fs_info->endio_workers =
1998 alloc_workqueue("btrfs-endio", flags, max_active);
1999 fs_info->endio_meta_workers =
2000 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2001 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2002 fs_info->endio_write_workers =
2003 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2005 fs_info->compressed_write_workers =
2006 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2007 fs_info->endio_freespace_worker =
2008 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2010 fs_info->delayed_workers =
2011 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2013 fs_info->qgroup_rescan_workers =
2014 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2016 fs_info->discard_ctl.discard_workers =
2017 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2019 if (!(fs_info->workers &&
2020 fs_info->delalloc_workers && fs_info->flush_workers &&
2021 fs_info->endio_workers && fs_info->endio_meta_workers &&
2022 fs_info->compressed_write_workers &&
2023 fs_info->endio_write_workers &&
2024 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2025 fs_info->caching_workers && fs_info->fixup_workers &&
2026 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2027 fs_info->discard_ctl.discard_workers)) {
2034 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2036 struct crypto_shash *csum_shash;
2037 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2039 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2041 if (IS_ERR(csum_shash)) {
2042 btrfs_err(fs_info, "error allocating %s hash for checksum",
2044 return PTR_ERR(csum_shash);
2047 fs_info->csum_shash = csum_shash;
2050 * Check if the checksum implementation is a fast accelerated one.
2051 * As-is this is a bit of a hack and should be replaced once the csum
2052 * implementations provide that information themselves.
2054 switch (csum_type) {
2055 case BTRFS_CSUM_TYPE_CRC32:
2056 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2057 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2059 case BTRFS_CSUM_TYPE_XXHASH:
2060 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2066 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2067 btrfs_super_csum_name(csum_type),
2068 crypto_shash_driver_name(csum_shash));
2072 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2073 struct btrfs_fs_devices *fs_devices)
2076 struct btrfs_tree_parent_check check = { 0 };
2077 struct btrfs_root *log_tree_root;
2078 struct btrfs_super_block *disk_super = fs_info->super_copy;
2079 u64 bytenr = btrfs_super_log_root(disk_super);
2080 int level = btrfs_super_log_root_level(disk_super);
2082 if (fs_devices->rw_devices == 0) {
2083 btrfs_warn(fs_info, "log replay required on RO media");
2087 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2092 check.level = level;
2093 check.transid = fs_info->generation + 1;
2094 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2095 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2096 if (IS_ERR(log_tree_root->node)) {
2097 btrfs_warn(fs_info, "failed to read log tree");
2098 ret = PTR_ERR(log_tree_root->node);
2099 log_tree_root->node = NULL;
2100 btrfs_put_root(log_tree_root);
2103 if (!extent_buffer_uptodate(log_tree_root->node)) {
2104 btrfs_err(fs_info, "failed to read log tree");
2105 btrfs_put_root(log_tree_root);
2109 /* returns with log_tree_root freed on success */
2110 ret = btrfs_recover_log_trees(log_tree_root);
2112 btrfs_handle_fs_error(fs_info, ret,
2113 "Failed to recover log tree");
2114 btrfs_put_root(log_tree_root);
2118 if (sb_rdonly(fs_info->sb)) {
2119 ret = btrfs_commit_super(fs_info);
2127 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2128 struct btrfs_path *path, u64 objectid,
2131 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2132 struct btrfs_root *root;
2133 u64 max_global_id = 0;
2135 struct btrfs_key key = {
2136 .objectid = objectid,
2137 .type = BTRFS_ROOT_ITEM_KEY,
2142 /* If we have IGNOREDATACSUMS skip loading these roots. */
2143 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2144 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2145 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2150 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2154 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2155 ret = btrfs_next_leaf(tree_root, path);
2164 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2165 if (key.objectid != objectid)
2167 btrfs_release_path(path);
2170 * Just worry about this for extent tree, it'll be the same for
2173 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2174 max_global_id = max(max_global_id, key.offset);
2177 root = read_tree_root_path(tree_root, path, &key);
2179 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2180 ret = PTR_ERR(root);
2183 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2184 ret = btrfs_global_root_insert(root);
2186 btrfs_put_root(root);
2191 btrfs_release_path(path);
2193 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2194 fs_info->nr_global_roots = max_global_id + 1;
2196 if (!found || ret) {
2197 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2198 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2200 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2201 ret = ret ? ret : -ENOENT;
2204 btrfs_err(fs_info, "failed to load root %s", name);
2209 static int load_global_roots(struct btrfs_root *tree_root)
2211 struct btrfs_path *path;
2214 path = btrfs_alloc_path();
2218 ret = load_global_roots_objectid(tree_root, path,
2219 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2222 ret = load_global_roots_objectid(tree_root, path,
2223 BTRFS_CSUM_TREE_OBJECTID, "csum");
2226 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2228 ret = load_global_roots_objectid(tree_root, path,
2229 BTRFS_FREE_SPACE_TREE_OBJECTID,
2232 btrfs_free_path(path);
2236 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2238 struct btrfs_root *tree_root = fs_info->tree_root;
2239 struct btrfs_root *root;
2240 struct btrfs_key location;
2243 ASSERT(fs_info->tree_root);
2245 ret = load_global_roots(tree_root);
2249 location.type = BTRFS_ROOT_ITEM_KEY;
2250 location.offset = 0;
2252 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2253 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2254 root = btrfs_read_tree_root(tree_root, &location);
2256 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2257 ret = PTR_ERR(root);
2261 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2262 fs_info->block_group_root = root;
2266 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2267 root = btrfs_read_tree_root(tree_root, &location);
2269 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2270 ret = PTR_ERR(root);
2274 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2275 fs_info->dev_root = root;
2277 /* Initialize fs_info for all devices in any case */
2278 ret = btrfs_init_devices_late(fs_info);
2283 * This tree can share blocks with some other fs tree during relocation
2284 * and we need a proper setup by btrfs_get_fs_root
2286 root = btrfs_get_fs_root(tree_root->fs_info,
2287 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2289 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2290 ret = PTR_ERR(root);
2294 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2295 fs_info->data_reloc_root = root;
2298 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2299 root = btrfs_read_tree_root(tree_root, &location);
2300 if (!IS_ERR(root)) {
2301 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2302 fs_info->quota_root = root;
2305 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2306 root = btrfs_read_tree_root(tree_root, &location);
2308 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2309 ret = PTR_ERR(root);
2314 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2315 fs_info->uuid_root = root;
2318 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2319 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2320 root = btrfs_read_tree_root(tree_root, &location);
2322 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2323 ret = PTR_ERR(root);
2327 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2328 fs_info->stripe_root = root;
2334 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2335 location.objectid, ret);
2340 * Real super block validation
2341 * NOTE: super csum type and incompat features will not be checked here.
2343 * @sb: super block to check
2344 * @mirror_num: the super block number to check its bytenr:
2345 * 0 the primary (1st) sb
2346 * 1, 2 2nd and 3rd backup copy
2347 * -1 skip bytenr check
2349 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2350 struct btrfs_super_block *sb, int mirror_num)
2352 u64 nodesize = btrfs_super_nodesize(sb);
2353 u64 sectorsize = btrfs_super_sectorsize(sb);
2356 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2357 btrfs_err(fs_info, "no valid FS found");
2360 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2361 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2362 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2365 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2366 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2367 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2370 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2371 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2372 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2375 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2376 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2377 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2382 * Check sectorsize and nodesize first, other check will need it.
2383 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2385 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2386 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2387 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2392 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2394 * We can support 16K sectorsize with 64K page size without problem,
2395 * but such sectorsize/pagesize combination doesn't make much sense.
2396 * 4K will be our future standard, PAGE_SIZE is supported from the very
2399 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2401 "sectorsize %llu not yet supported for page size %lu",
2402 sectorsize, PAGE_SIZE);
2406 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2407 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2408 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2411 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2412 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2413 le32_to_cpu(sb->__unused_leafsize), nodesize);
2417 /* Root alignment check */
2418 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2419 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2420 btrfs_super_root(sb));
2423 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2424 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2425 btrfs_super_chunk_root(sb));
2428 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2429 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2430 btrfs_super_log_root(sb));
2434 if (!fs_info->fs_devices->temp_fsid &&
2435 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2437 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2438 sb->fsid, fs_info->fs_devices->fsid);
2442 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2443 BTRFS_FSID_SIZE) != 0) {
2445 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2446 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2450 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2451 BTRFS_FSID_SIZE) != 0) {
2453 "dev_item UUID does not match metadata fsid: %pU != %pU",
2454 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2459 * Artificial requirement for block-group-tree to force newer features
2460 * (free-space-tree, no-holes) so the test matrix is smaller.
2462 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2463 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2464 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2466 "block-group-tree feature requires fres-space-tree and no-holes");
2471 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2474 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2475 btrfs_err(fs_info, "bytes_used is too small %llu",
2476 btrfs_super_bytes_used(sb));
2479 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2480 btrfs_err(fs_info, "invalid stripesize %u",
2481 btrfs_super_stripesize(sb));
2484 if (btrfs_super_num_devices(sb) > (1UL << 31))
2485 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2486 btrfs_super_num_devices(sb));
2487 if (btrfs_super_num_devices(sb) == 0) {
2488 btrfs_err(fs_info, "number of devices is 0");
2492 if (mirror_num >= 0 &&
2493 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2494 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2495 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2500 * Obvious sys_chunk_array corruptions, it must hold at least one key
2503 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2504 btrfs_err(fs_info, "system chunk array too big %u > %u",
2505 btrfs_super_sys_array_size(sb),
2506 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2509 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2510 + sizeof(struct btrfs_chunk)) {
2511 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2512 btrfs_super_sys_array_size(sb),
2513 sizeof(struct btrfs_disk_key)
2514 + sizeof(struct btrfs_chunk));
2519 * The generation is a global counter, we'll trust it more than the others
2520 * but it's still possible that it's the one that's wrong.
2522 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2524 "suspicious: generation < chunk_root_generation: %llu < %llu",
2525 btrfs_super_generation(sb),
2526 btrfs_super_chunk_root_generation(sb));
2527 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2528 && btrfs_super_cache_generation(sb) != (u64)-1)
2530 "suspicious: generation < cache_generation: %llu < %llu",
2531 btrfs_super_generation(sb),
2532 btrfs_super_cache_generation(sb));
2538 * Validation of super block at mount time.
2539 * Some checks already done early at mount time, like csum type and incompat
2540 * flags will be skipped.
2542 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2544 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2548 * Validation of super block at write time.
2549 * Some checks like bytenr check will be skipped as their values will be
2551 * Extra checks like csum type and incompat flags will be done here.
2553 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2554 struct btrfs_super_block *sb)
2558 ret = btrfs_validate_super(fs_info, sb, -1);
2561 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2563 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2564 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2567 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2570 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2571 btrfs_super_incompat_flags(sb),
2572 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2578 "super block corruption detected before writing it to disk");
2582 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2584 struct btrfs_tree_parent_check check = {
2587 .owner_root = root->root_key.objectid
2591 root->node = read_tree_block(root->fs_info, bytenr, &check);
2592 if (IS_ERR(root->node)) {
2593 ret = PTR_ERR(root->node);
2597 if (!extent_buffer_uptodate(root->node)) {
2598 free_extent_buffer(root->node);
2603 btrfs_set_root_node(&root->root_item, root->node);
2604 root->commit_root = btrfs_root_node(root);
2605 btrfs_set_root_refs(&root->root_item, 1);
2609 static int load_important_roots(struct btrfs_fs_info *fs_info)
2611 struct btrfs_super_block *sb = fs_info->super_copy;
2615 bytenr = btrfs_super_root(sb);
2616 gen = btrfs_super_generation(sb);
2617 level = btrfs_super_root_level(sb);
2618 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2620 btrfs_warn(fs_info, "couldn't read tree root");
2626 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2628 int backup_index = find_newest_super_backup(fs_info);
2629 struct btrfs_super_block *sb = fs_info->super_copy;
2630 struct btrfs_root *tree_root = fs_info->tree_root;
2631 bool handle_error = false;
2635 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2637 if (!IS_ERR(tree_root->node))
2638 free_extent_buffer(tree_root->node);
2639 tree_root->node = NULL;
2641 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2644 free_root_pointers(fs_info, 0);
2647 * Don't use the log in recovery mode, it won't be
2650 btrfs_set_super_log_root(sb, 0);
2652 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2653 ret = read_backup_root(fs_info, i);
2659 ret = load_important_roots(fs_info);
2661 handle_error = true;
2666 * No need to hold btrfs_root::objectid_mutex since the fs
2667 * hasn't been fully initialised and we are the only user
2669 ret = btrfs_init_root_free_objectid(tree_root);
2671 handle_error = true;
2675 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2677 ret = btrfs_read_roots(fs_info);
2679 handle_error = true;
2683 /* All successful */
2684 fs_info->generation = btrfs_header_generation(tree_root->node);
2685 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2686 fs_info->last_reloc_trans = 0;
2688 /* Always begin writing backup roots after the one being used */
2689 if (backup_index < 0) {
2690 fs_info->backup_root_index = 0;
2692 fs_info->backup_root_index = backup_index + 1;
2693 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2701 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2703 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2704 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2705 INIT_LIST_HEAD(&fs_info->trans_list);
2706 INIT_LIST_HEAD(&fs_info->dead_roots);
2707 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2708 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2709 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2710 spin_lock_init(&fs_info->delalloc_root_lock);
2711 spin_lock_init(&fs_info->trans_lock);
2712 spin_lock_init(&fs_info->fs_roots_radix_lock);
2713 spin_lock_init(&fs_info->delayed_iput_lock);
2714 spin_lock_init(&fs_info->defrag_inodes_lock);
2715 spin_lock_init(&fs_info->super_lock);
2716 spin_lock_init(&fs_info->buffer_lock);
2717 spin_lock_init(&fs_info->unused_bgs_lock);
2718 spin_lock_init(&fs_info->treelog_bg_lock);
2719 spin_lock_init(&fs_info->zone_active_bgs_lock);
2720 spin_lock_init(&fs_info->relocation_bg_lock);
2721 rwlock_init(&fs_info->tree_mod_log_lock);
2722 rwlock_init(&fs_info->global_root_lock);
2723 mutex_init(&fs_info->unused_bg_unpin_mutex);
2724 mutex_init(&fs_info->reclaim_bgs_lock);
2725 mutex_init(&fs_info->reloc_mutex);
2726 mutex_init(&fs_info->delalloc_root_mutex);
2727 mutex_init(&fs_info->zoned_meta_io_lock);
2728 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2729 seqlock_init(&fs_info->profiles_lock);
2731 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2732 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2733 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2734 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2735 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2736 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2737 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2738 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2739 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2740 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2741 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2742 BTRFS_LOCKDEP_TRANS_COMPLETED);
2744 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2745 INIT_LIST_HEAD(&fs_info->space_info);
2746 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2747 INIT_LIST_HEAD(&fs_info->unused_bgs);
2748 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2749 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2750 #ifdef CONFIG_BTRFS_DEBUG
2751 INIT_LIST_HEAD(&fs_info->allocated_roots);
2752 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2753 spin_lock_init(&fs_info->eb_leak_lock);
2755 fs_info->mapping_tree = RB_ROOT_CACHED;
2756 rwlock_init(&fs_info->mapping_tree_lock);
2757 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2758 BTRFS_BLOCK_RSV_GLOBAL);
2759 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2760 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2761 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2762 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2763 BTRFS_BLOCK_RSV_DELOPS);
2764 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2765 BTRFS_BLOCK_RSV_DELREFS);
2767 atomic_set(&fs_info->async_delalloc_pages, 0);
2768 atomic_set(&fs_info->defrag_running, 0);
2769 atomic_set(&fs_info->nr_delayed_iputs, 0);
2770 atomic64_set(&fs_info->tree_mod_seq, 0);
2771 fs_info->global_root_tree = RB_ROOT;
2772 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2773 fs_info->metadata_ratio = 0;
2774 fs_info->defrag_inodes = RB_ROOT;
2775 atomic64_set(&fs_info->free_chunk_space, 0);
2776 fs_info->tree_mod_log = RB_ROOT;
2777 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2778 btrfs_init_ref_verify(fs_info);
2780 fs_info->thread_pool_size = min_t(unsigned long,
2781 num_online_cpus() + 2, 8);
2783 INIT_LIST_HEAD(&fs_info->ordered_roots);
2784 spin_lock_init(&fs_info->ordered_root_lock);
2786 btrfs_init_scrub(fs_info);
2787 btrfs_init_balance(fs_info);
2788 btrfs_init_async_reclaim_work(fs_info);
2790 rwlock_init(&fs_info->block_group_cache_lock);
2791 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2793 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2794 IO_TREE_FS_EXCLUDED_EXTENTS);
2796 mutex_init(&fs_info->ordered_operations_mutex);
2797 mutex_init(&fs_info->tree_log_mutex);
2798 mutex_init(&fs_info->chunk_mutex);
2799 mutex_init(&fs_info->transaction_kthread_mutex);
2800 mutex_init(&fs_info->cleaner_mutex);
2801 mutex_init(&fs_info->ro_block_group_mutex);
2802 init_rwsem(&fs_info->commit_root_sem);
2803 init_rwsem(&fs_info->cleanup_work_sem);
2804 init_rwsem(&fs_info->subvol_sem);
2805 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2807 btrfs_init_dev_replace_locks(fs_info);
2808 btrfs_init_qgroup(fs_info);
2809 btrfs_discard_init(fs_info);
2811 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2812 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2814 init_waitqueue_head(&fs_info->transaction_throttle);
2815 init_waitqueue_head(&fs_info->transaction_wait);
2816 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2817 init_waitqueue_head(&fs_info->async_submit_wait);
2818 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2820 /* Usable values until the real ones are cached from the superblock */
2821 fs_info->nodesize = 4096;
2822 fs_info->sectorsize = 4096;
2823 fs_info->sectorsize_bits = ilog2(4096);
2824 fs_info->stripesize = 4096;
2826 /* Default compress algorithm when user does -o compress */
2827 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2829 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2831 spin_lock_init(&fs_info->swapfile_pins_lock);
2832 fs_info->swapfile_pins = RB_ROOT;
2834 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2835 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2838 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2843 /* Temporary fixed values for block size until we read the superblock. */
2844 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2845 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2847 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2851 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2855 fs_info->dirty_metadata_batch = PAGE_SIZE *
2856 (1 + ilog2(nr_cpu_ids));
2858 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2862 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2867 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2869 if (!fs_info->delayed_root)
2871 btrfs_init_delayed_root(fs_info->delayed_root);
2874 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2876 return btrfs_alloc_stripe_hash_table(fs_info);
2879 static int btrfs_uuid_rescan_kthread(void *data)
2881 struct btrfs_fs_info *fs_info = data;
2885 * 1st step is to iterate through the existing UUID tree and
2886 * to delete all entries that contain outdated data.
2887 * 2nd step is to add all missing entries to the UUID tree.
2889 ret = btrfs_uuid_tree_iterate(fs_info);
2892 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2894 up(&fs_info->uuid_tree_rescan_sem);
2897 return btrfs_uuid_scan_kthread(data);
2900 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2902 struct task_struct *task;
2904 down(&fs_info->uuid_tree_rescan_sem);
2905 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2907 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2908 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2909 up(&fs_info->uuid_tree_rescan_sem);
2910 return PTR_ERR(task);
2916 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2918 u64 root_objectid = 0;
2919 struct btrfs_root *gang[8];
2922 unsigned int ret = 0;
2925 spin_lock(&fs_info->fs_roots_radix_lock);
2926 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2927 (void **)gang, root_objectid,
2930 spin_unlock(&fs_info->fs_roots_radix_lock);
2933 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2935 for (i = 0; i < ret; i++) {
2936 /* Avoid to grab roots in dead_roots. */
2937 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2941 /* Grab all the search result for later use. */
2942 gang[i] = btrfs_grab_root(gang[i]);
2944 spin_unlock(&fs_info->fs_roots_radix_lock);
2946 for (i = 0; i < ret; i++) {
2949 root_objectid = gang[i]->root_key.objectid;
2950 err = btrfs_orphan_cleanup(gang[i]);
2953 btrfs_put_root(gang[i]);
2958 /* Release the uncleaned roots due to error. */
2959 for (; i < ret; i++) {
2961 btrfs_put_root(gang[i]);
2967 * Mounting logic specific to read-write file systems. Shared by open_ctree
2968 * and btrfs_remount when remounting from read-only to read-write.
2970 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2973 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2974 bool rebuild_free_space_tree = false;
2976 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2977 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2978 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2980 "'clear_cache' option is ignored with extent tree v2");
2982 rebuild_free_space_tree = true;
2983 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2984 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2985 btrfs_warn(fs_info, "free space tree is invalid");
2986 rebuild_free_space_tree = true;
2989 if (rebuild_free_space_tree) {
2990 btrfs_info(fs_info, "rebuilding free space tree");
2991 ret = btrfs_rebuild_free_space_tree(fs_info);
2994 "failed to rebuild free space tree: %d", ret);
2999 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3000 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3001 btrfs_info(fs_info, "disabling free space tree");
3002 ret = btrfs_delete_free_space_tree(fs_info);
3005 "failed to disable free space tree: %d", ret);
3011 * btrfs_find_orphan_roots() is responsible for finding all the dead
3012 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3013 * them into the fs_info->fs_roots_radix tree. This must be done before
3014 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3015 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3016 * item before the root's tree is deleted - this means that if we unmount
3017 * or crash before the deletion completes, on the next mount we will not
3018 * delete what remains of the tree because the orphan item does not
3019 * exists anymore, which is what tells us we have a pending deletion.
3021 ret = btrfs_find_orphan_roots(fs_info);
3025 ret = btrfs_cleanup_fs_roots(fs_info);
3029 down_read(&fs_info->cleanup_work_sem);
3030 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3031 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3032 up_read(&fs_info->cleanup_work_sem);
3035 up_read(&fs_info->cleanup_work_sem);
3037 mutex_lock(&fs_info->cleaner_mutex);
3038 ret = btrfs_recover_relocation(fs_info);
3039 mutex_unlock(&fs_info->cleaner_mutex);
3041 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3045 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3046 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3047 btrfs_info(fs_info, "creating free space tree");
3048 ret = btrfs_create_free_space_tree(fs_info);
3051 "failed to create free space tree: %d", ret);
3056 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3057 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3062 ret = btrfs_resume_balance_async(fs_info);
3066 ret = btrfs_resume_dev_replace_async(fs_info);
3068 btrfs_warn(fs_info, "failed to resume dev_replace");
3072 btrfs_qgroup_rescan_resume(fs_info);
3074 if (!fs_info->uuid_root) {
3075 btrfs_info(fs_info, "creating UUID tree");
3076 ret = btrfs_create_uuid_tree(fs_info);
3079 "failed to create the UUID tree %d", ret);
3089 * Do various sanity and dependency checks of different features.
3091 * @is_rw_mount: If the mount is read-write.
3093 * This is the place for less strict checks (like for subpage or artificial
3094 * feature dependencies).
3096 * For strict checks or possible corruption detection, see
3097 * btrfs_validate_super().
3099 * This should be called after btrfs_parse_options(), as some mount options
3100 * (space cache related) can modify on-disk format like free space tree and
3101 * screw up certain feature dependencies.
3103 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3105 struct btrfs_super_block *disk_super = fs_info->super_copy;
3106 u64 incompat = btrfs_super_incompat_flags(disk_super);
3107 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3108 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3110 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3112 "cannot mount because of unknown incompat features (0x%llx)",
3117 /* Runtime limitation for mixed block groups. */
3118 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3119 (fs_info->sectorsize != fs_info->nodesize)) {
3121 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3122 fs_info->nodesize, fs_info->sectorsize);
3126 /* Mixed backref is an always-enabled feature. */
3127 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3129 /* Set compression related flags just in case. */
3130 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3131 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3132 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3133 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3136 * An ancient flag, which should really be marked deprecated.
3137 * Such runtime limitation doesn't really need a incompat flag.
3139 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3140 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3142 if (compat_ro_unsupp && is_rw_mount) {
3144 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3150 * We have unsupported RO compat features, although RO mounted, we
3151 * should not cause any metadata writes, including log replay.
3152 * Or we could screw up whatever the new feature requires.
3154 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3155 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3157 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3163 * Artificial limitations for block group tree, to force
3164 * block-group-tree to rely on no-holes and free-space-tree.
3166 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3167 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3168 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3170 "block-group-tree feature requires no-holes and free-space-tree features");
3175 * Subpage runtime limitation on v1 cache.
3177 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3178 * we're already defaulting to v2 cache, no need to bother v1 as it's
3179 * going to be deprecated anyway.
3181 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3183 "v1 space cache is not supported for page size %lu with sectorsize %u",
3184 PAGE_SIZE, fs_info->sectorsize);
3188 /* This can be called by remount, we need to protect the super block. */
3189 spin_lock(&fs_info->super_lock);
3190 btrfs_set_super_incompat_flags(disk_super, incompat);
3191 spin_unlock(&fs_info->super_lock);
3196 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3204 struct btrfs_super_block *disk_super;
3205 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3206 struct btrfs_root *tree_root;
3207 struct btrfs_root *chunk_root;
3211 ret = init_mount_fs_info(fs_info, sb);
3215 /* These need to be init'ed before we start creating inodes and such. */
3216 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3218 fs_info->tree_root = tree_root;
3219 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3221 fs_info->chunk_root = chunk_root;
3222 if (!tree_root || !chunk_root) {
3227 ret = btrfs_init_btree_inode(sb);
3231 invalidate_bdev(fs_devices->latest_dev->bdev);
3234 * Read super block and check the signature bytes only
3236 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3237 if (IS_ERR(disk_super)) {
3238 ret = PTR_ERR(disk_super);
3242 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3244 * Verify the type first, if that or the checksum value are
3245 * corrupted, we'll find out
3247 csum_type = btrfs_super_csum_type(disk_super);
3248 if (!btrfs_supported_super_csum(csum_type)) {
3249 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3252 btrfs_release_disk_super(disk_super);
3256 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3258 ret = btrfs_init_csum_hash(fs_info, csum_type);
3260 btrfs_release_disk_super(disk_super);
3265 * We want to check superblock checksum, the type is stored inside.
3266 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3268 if (btrfs_check_super_csum(fs_info, disk_super)) {
3269 btrfs_err(fs_info, "superblock checksum mismatch");
3271 btrfs_release_disk_super(disk_super);
3276 * super_copy is zeroed at allocation time and we never touch the
3277 * following bytes up to INFO_SIZE, the checksum is calculated from
3278 * the whole block of INFO_SIZE
3280 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3281 btrfs_release_disk_super(disk_super);
3283 disk_super = fs_info->super_copy;
3285 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3286 sizeof(*fs_info->super_for_commit));
3288 ret = btrfs_validate_mount_super(fs_info);
3290 btrfs_err(fs_info, "superblock contains fatal errors");
3295 if (!btrfs_super_root(disk_super)) {
3296 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3301 /* check FS state, whether FS is broken. */
3302 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3303 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3305 /* Set up fs_info before parsing mount options */
3306 nodesize = btrfs_super_nodesize(disk_super);
3307 sectorsize = btrfs_super_sectorsize(disk_super);
3308 stripesize = sectorsize;
3309 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3310 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3312 fs_info->nodesize = nodesize;
3313 fs_info->sectorsize = sectorsize;
3314 fs_info->sectorsize_bits = ilog2(sectorsize);
3315 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3316 fs_info->stripesize = stripesize;
3319 * Handle the space caching options appropriately now that we have the
3320 * super block loaded and validated.
3322 btrfs_set_free_space_cache_settings(fs_info);
3324 if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3329 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3334 * At this point our mount options are validated, if we set ->max_inline
3335 * to something non-standard make sure we truncate it to sectorsize.
3337 fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3339 if (sectorsize < PAGE_SIZE) {
3340 struct btrfs_subpage_info *subpage_info;
3343 "read-write for sector size %u with page size %lu is experimental",
3344 sectorsize, PAGE_SIZE);
3345 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3346 if (!subpage_info) {
3350 btrfs_init_subpage_info(subpage_info, sectorsize);
3351 fs_info->subpage_info = subpage_info;
3354 ret = btrfs_init_workqueues(fs_info);
3356 goto fail_sb_buffer;
3358 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3359 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3361 /* Update the values for the current filesystem. */
3362 sb->s_blocksize = sectorsize;
3363 sb->s_blocksize_bits = blksize_bits(sectorsize);
3364 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3366 mutex_lock(&fs_info->chunk_mutex);
3367 ret = btrfs_read_sys_array(fs_info);
3368 mutex_unlock(&fs_info->chunk_mutex);
3370 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3371 goto fail_sb_buffer;
3374 generation = btrfs_super_chunk_root_generation(disk_super);
3375 level = btrfs_super_chunk_root_level(disk_super);
3376 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3379 btrfs_err(fs_info, "failed to read chunk root");
3380 goto fail_tree_roots;
3383 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3384 offsetof(struct btrfs_header, chunk_tree_uuid),
3387 ret = btrfs_read_chunk_tree(fs_info);
3389 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3390 goto fail_tree_roots;
3394 * At this point we know all the devices that make this filesystem,
3395 * including the seed devices but we don't know yet if the replace
3396 * target is required. So free devices that are not part of this
3397 * filesystem but skip the replace target device which is checked
3398 * below in btrfs_init_dev_replace().
3400 btrfs_free_extra_devids(fs_devices);
3401 if (!fs_devices->latest_dev->bdev) {
3402 btrfs_err(fs_info, "failed to read devices");
3404 goto fail_tree_roots;
3407 ret = init_tree_roots(fs_info);
3409 goto fail_tree_roots;
3412 * Get zone type information of zoned block devices. This will also
3413 * handle emulation of a zoned filesystem if a regular device has the
3414 * zoned incompat feature flag set.
3416 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3419 "zoned: failed to read device zone info: %d", ret);
3420 goto fail_block_groups;
3424 * If we have a uuid root and we're not being told to rescan we need to
3425 * check the generation here so we can set the
3426 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3427 * transaction during a balance or the log replay without updating the
3428 * uuid generation, and then if we crash we would rescan the uuid tree,
3429 * even though it was perfectly fine.
3431 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3432 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3433 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3435 ret = btrfs_verify_dev_extents(fs_info);
3438 "failed to verify dev extents against chunks: %d",
3440 goto fail_block_groups;
3442 ret = btrfs_recover_balance(fs_info);
3444 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3445 goto fail_block_groups;
3448 ret = btrfs_init_dev_stats(fs_info);
3450 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3451 goto fail_block_groups;
3454 ret = btrfs_init_dev_replace(fs_info);
3456 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3457 goto fail_block_groups;
3460 ret = btrfs_check_zoned_mode(fs_info);
3462 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3464 goto fail_block_groups;
3467 ret = btrfs_sysfs_add_fsid(fs_devices);
3469 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3471 goto fail_block_groups;
3474 ret = btrfs_sysfs_add_mounted(fs_info);
3476 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3477 goto fail_fsdev_sysfs;
3480 ret = btrfs_init_space_info(fs_info);
3482 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3486 ret = btrfs_read_block_groups(fs_info);
3488 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3492 btrfs_free_zone_cache(fs_info);
3494 btrfs_check_active_zone_reservation(fs_info);
3496 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3497 !btrfs_check_rw_degradable(fs_info, NULL)) {
3499 "writable mount is not allowed due to too many missing devices");
3504 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3506 if (IS_ERR(fs_info->cleaner_kthread)) {
3507 ret = PTR_ERR(fs_info->cleaner_kthread);
3511 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3513 "btrfs-transaction");
3514 if (IS_ERR(fs_info->transaction_kthread)) {
3515 ret = PTR_ERR(fs_info->transaction_kthread);
3519 ret = btrfs_read_qgroup_config(fs_info);
3521 goto fail_trans_kthread;
3523 if (btrfs_build_ref_tree(fs_info))
3524 btrfs_err(fs_info, "couldn't build ref tree");
3526 /* do not make disk changes in broken FS or nologreplay is given */
3527 if (btrfs_super_log_root(disk_super) != 0 &&
3528 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3529 btrfs_info(fs_info, "start tree-log replay");
3530 ret = btrfs_replay_log(fs_info, fs_devices);
3535 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3536 if (IS_ERR(fs_info->fs_root)) {
3537 ret = PTR_ERR(fs_info->fs_root);
3538 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3539 fs_info->fs_root = NULL;
3546 ret = btrfs_start_pre_rw_mount(fs_info);
3548 close_ctree(fs_info);
3551 btrfs_discard_resume(fs_info);
3553 if (fs_info->uuid_root &&
3554 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3555 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3556 btrfs_info(fs_info, "checking UUID tree");
3557 ret = btrfs_check_uuid_tree(fs_info);
3560 "failed to check the UUID tree: %d", ret);
3561 close_ctree(fs_info);
3566 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3568 /* Kick the cleaner thread so it'll start deleting snapshots. */
3569 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3570 wake_up_process(fs_info->cleaner_kthread);
3575 btrfs_free_qgroup_config(fs_info);
3577 kthread_stop(fs_info->transaction_kthread);
3578 btrfs_cleanup_transaction(fs_info);
3579 btrfs_free_fs_roots(fs_info);
3581 kthread_stop(fs_info->cleaner_kthread);
3584 * make sure we're done with the btree inode before we stop our
3587 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3590 btrfs_sysfs_remove_mounted(fs_info);
3593 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3596 btrfs_put_block_group_cache(fs_info);
3599 if (fs_info->data_reloc_root)
3600 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3601 free_root_pointers(fs_info, true);
3602 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3605 btrfs_stop_all_workers(fs_info);
3606 btrfs_free_block_groups(fs_info);
3608 btrfs_mapping_tree_free(fs_info);
3610 iput(fs_info->btree_inode);
3612 btrfs_close_devices(fs_info->fs_devices);
3616 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3618 static void btrfs_end_super_write(struct bio *bio)
3620 struct btrfs_device *device = bio->bi_private;
3621 struct bio_vec *bvec;
3622 struct bvec_iter_all iter_all;
3625 bio_for_each_segment_all(bvec, bio, iter_all) {
3626 page = bvec->bv_page;
3628 if (bio->bi_status) {
3629 btrfs_warn_rl_in_rcu(device->fs_info,
3630 "lost page write due to IO error on %s (%d)",
3631 btrfs_dev_name(device),
3632 blk_status_to_errno(bio->bi_status));
3633 ClearPageUptodate(page);
3635 btrfs_dev_stat_inc_and_print(device,
3636 BTRFS_DEV_STAT_WRITE_ERRS);
3638 SetPageUptodate(page);
3648 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3649 int copy_num, bool drop_cache)
3651 struct btrfs_super_block *super;
3653 u64 bytenr, bytenr_orig;
3654 struct address_space *mapping = bdev->bd_inode->i_mapping;
3657 bytenr_orig = btrfs_sb_offset(copy_num);
3658 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3660 return ERR_PTR(-EINVAL);
3662 return ERR_PTR(ret);
3664 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3665 return ERR_PTR(-EINVAL);
3668 /* This should only be called with the primary sb. */
3669 ASSERT(copy_num == 0);
3672 * Drop the page of the primary superblock, so later read will
3673 * always read from the device.
3675 invalidate_inode_pages2_range(mapping,
3676 bytenr >> PAGE_SHIFT,
3677 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3680 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3682 return ERR_CAST(page);
3684 super = page_address(page);
3685 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3686 btrfs_release_disk_super(super);
3687 return ERR_PTR(-ENODATA);
3690 if (btrfs_super_bytenr(super) != bytenr_orig) {
3691 btrfs_release_disk_super(super);
3692 return ERR_PTR(-EINVAL);
3699 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3701 struct btrfs_super_block *super, *latest = NULL;
3705 /* we would like to check all the supers, but that would make
3706 * a btrfs mount succeed after a mkfs from a different FS.
3707 * So, we need to add a special mount option to scan for
3708 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3710 for (i = 0; i < 1; i++) {
3711 super = btrfs_read_dev_one_super(bdev, i, false);
3715 if (!latest || btrfs_super_generation(super) > transid) {
3717 btrfs_release_disk_super(super);
3720 transid = btrfs_super_generation(super);
3728 * Write superblock @sb to the @device. Do not wait for completion, all the
3729 * pages we use for writing are locked.
3731 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3732 * the expected device size at commit time. Note that max_mirrors must be
3733 * same for write and wait phases.
3735 * Return number of errors when page is not found or submission fails.
3737 static int write_dev_supers(struct btrfs_device *device,
3738 struct btrfs_super_block *sb, int max_mirrors)
3740 struct btrfs_fs_info *fs_info = device->fs_info;
3741 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3742 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3746 u64 bytenr, bytenr_orig;
3748 if (max_mirrors == 0)
3749 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3751 shash->tfm = fs_info->csum_shash;
3753 for (i = 0; i < max_mirrors; i++) {
3756 struct btrfs_super_block *disk_super;
3758 bytenr_orig = btrfs_sb_offset(i);
3759 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3760 if (ret == -ENOENT) {
3762 } else if (ret < 0) {
3763 btrfs_err(device->fs_info,
3764 "couldn't get super block location for mirror %d",
3769 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3770 device->commit_total_bytes)
3773 btrfs_set_super_bytenr(sb, bytenr_orig);
3775 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3776 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3779 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3782 btrfs_err(device->fs_info,
3783 "couldn't get super block page for bytenr %llu",
3789 /* Bump the refcount for wait_dev_supers() */
3792 disk_super = page_address(page);
3793 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3796 * Directly use bios here instead of relying on the page cache
3797 * to do I/O, so we don't lose the ability to do integrity
3800 bio = bio_alloc(device->bdev, 1,
3801 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3803 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3804 bio->bi_private = device;
3805 bio->bi_end_io = btrfs_end_super_write;
3806 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3807 offset_in_page(bytenr));
3810 * We FUA only the first super block. The others we allow to
3811 * go down lazy and there's a short window where the on-disk
3812 * copies might still contain the older version.
3814 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3815 bio->bi_opf |= REQ_FUA;
3818 if (btrfs_advance_sb_log(device, i))
3821 return errors < i ? 0 : -1;
3825 * Wait for write completion of superblocks done by write_dev_supers,
3826 * @max_mirrors same for write and wait phases.
3828 * Return number of errors when page is not found or not marked up to
3831 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3835 bool primary_failed = false;
3839 if (max_mirrors == 0)
3840 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3842 for (i = 0; i < max_mirrors; i++) {
3845 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3846 if (ret == -ENOENT) {
3848 } else if (ret < 0) {
3851 primary_failed = true;
3854 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3855 device->commit_total_bytes)
3858 page = find_get_page(device->bdev->bd_inode->i_mapping,
3859 bytenr >> PAGE_SHIFT);
3863 primary_failed = true;
3866 /* Page is submitted locked and unlocked once the IO completes */
3867 wait_on_page_locked(page);
3868 if (PageError(page)) {
3871 primary_failed = true;
3874 /* Drop our reference */
3877 /* Drop the reference from the writing run */
3881 /* log error, force error return */
3882 if (primary_failed) {
3883 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3888 return errors < i ? 0 : -1;
3892 * endio for the write_dev_flush, this will wake anyone waiting
3893 * for the barrier when it is done
3895 static void btrfs_end_empty_barrier(struct bio *bio)
3898 complete(bio->bi_private);
3902 * Submit a flush request to the device if it supports it. Error handling is
3903 * done in the waiting counterpart.
3905 static void write_dev_flush(struct btrfs_device *device)
3907 struct bio *bio = &device->flush_bio;
3909 device->last_flush_error = BLK_STS_OK;
3911 bio_init(bio, device->bdev, NULL, 0,
3912 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3913 bio->bi_end_io = btrfs_end_empty_barrier;
3914 init_completion(&device->flush_wait);
3915 bio->bi_private = &device->flush_wait;
3917 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3921 * If the flush bio has been submitted by write_dev_flush, wait for it.
3922 * Return true for any error, and false otherwise.
3924 static bool wait_dev_flush(struct btrfs_device *device)
3926 struct bio *bio = &device->flush_bio;
3928 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3931 wait_for_completion_io(&device->flush_wait);
3933 if (bio->bi_status) {
3934 device->last_flush_error = bio->bi_status;
3935 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3943 * send an empty flush down to each device in parallel,
3944 * then wait for them
3946 static int barrier_all_devices(struct btrfs_fs_info *info)
3948 struct list_head *head;
3949 struct btrfs_device *dev;
3950 int errors_wait = 0;
3952 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3953 /* send down all the barriers */
3954 head = &info->fs_devices->devices;
3955 list_for_each_entry(dev, head, dev_list) {
3956 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3960 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3961 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3964 write_dev_flush(dev);
3967 /* wait for all the barriers */
3968 list_for_each_entry(dev, head, dev_list) {
3969 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3975 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3976 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3979 if (wait_dev_flush(dev))
3984 * Checks last_flush_error of disks in order to determine the device
3987 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3993 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3996 int min_tolerated = INT_MAX;
3998 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3999 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4000 min_tolerated = min_t(int, min_tolerated,
4001 btrfs_raid_array[BTRFS_RAID_SINGLE].
4002 tolerated_failures);
4004 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4005 if (raid_type == BTRFS_RAID_SINGLE)
4007 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4009 min_tolerated = min_t(int, min_tolerated,
4010 btrfs_raid_array[raid_type].
4011 tolerated_failures);
4014 if (min_tolerated == INT_MAX) {
4015 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4019 return min_tolerated;
4022 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4024 struct list_head *head;
4025 struct btrfs_device *dev;
4026 struct btrfs_super_block *sb;
4027 struct btrfs_dev_item *dev_item;
4031 int total_errors = 0;
4034 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4037 * max_mirrors == 0 indicates we're from commit_transaction,
4038 * not from fsync where the tree roots in fs_info have not
4039 * been consistent on disk.
4041 if (max_mirrors == 0)
4042 backup_super_roots(fs_info);
4044 sb = fs_info->super_for_commit;
4045 dev_item = &sb->dev_item;
4047 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4048 head = &fs_info->fs_devices->devices;
4049 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4052 ret = barrier_all_devices(fs_info);
4055 &fs_info->fs_devices->device_list_mutex);
4056 btrfs_handle_fs_error(fs_info, ret,
4057 "errors while submitting device barriers.");
4062 list_for_each_entry(dev, head, dev_list) {
4067 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4068 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4071 btrfs_set_stack_device_generation(dev_item, 0);
4072 btrfs_set_stack_device_type(dev_item, dev->type);
4073 btrfs_set_stack_device_id(dev_item, dev->devid);
4074 btrfs_set_stack_device_total_bytes(dev_item,
4075 dev->commit_total_bytes);
4076 btrfs_set_stack_device_bytes_used(dev_item,
4077 dev->commit_bytes_used);
4078 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4079 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4080 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4081 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4082 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4085 flags = btrfs_super_flags(sb);
4086 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4088 ret = btrfs_validate_write_super(fs_info, sb);
4090 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4091 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4092 "unexpected superblock corruption detected");
4096 ret = write_dev_supers(dev, sb, max_mirrors);
4100 if (total_errors > max_errors) {
4101 btrfs_err(fs_info, "%d errors while writing supers",
4103 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4105 /* FUA is masked off if unsupported and can't be the reason */
4106 btrfs_handle_fs_error(fs_info, -EIO,
4107 "%d errors while writing supers",
4113 list_for_each_entry(dev, head, dev_list) {
4116 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4117 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4120 ret = wait_dev_supers(dev, max_mirrors);
4124 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4125 if (total_errors > max_errors) {
4126 btrfs_handle_fs_error(fs_info, -EIO,
4127 "%d errors while writing supers",
4134 /* Drop a fs root from the radix tree and free it. */
4135 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4136 struct btrfs_root *root)
4138 bool drop_ref = false;
4140 spin_lock(&fs_info->fs_roots_radix_lock);
4141 radix_tree_delete(&fs_info->fs_roots_radix,
4142 (unsigned long)root->root_key.objectid);
4143 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4145 spin_unlock(&fs_info->fs_roots_radix_lock);
4147 if (BTRFS_FS_ERROR(fs_info)) {
4148 ASSERT(root->log_root == NULL);
4149 if (root->reloc_root) {
4150 btrfs_put_root(root->reloc_root);
4151 root->reloc_root = NULL;
4156 btrfs_put_root(root);
4159 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4161 struct btrfs_root *root = fs_info->tree_root;
4162 struct btrfs_trans_handle *trans;
4164 mutex_lock(&fs_info->cleaner_mutex);
4165 btrfs_run_delayed_iputs(fs_info);
4166 mutex_unlock(&fs_info->cleaner_mutex);
4167 wake_up_process(fs_info->cleaner_kthread);
4169 /* wait until ongoing cleanup work done */
4170 down_write(&fs_info->cleanup_work_sem);
4171 up_write(&fs_info->cleanup_work_sem);
4173 trans = btrfs_join_transaction(root);
4175 return PTR_ERR(trans);
4176 return btrfs_commit_transaction(trans);
4179 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4181 struct btrfs_transaction *trans;
4182 struct btrfs_transaction *tmp;
4185 if (list_empty(&fs_info->trans_list))
4189 * This function is only called at the very end of close_ctree(),
4190 * thus no other running transaction, no need to take trans_lock.
4192 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4193 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4194 struct extent_state *cached = NULL;
4195 u64 dirty_bytes = 0;
4201 while (find_first_extent_bit(&trans->dirty_pages, cur,
4202 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4203 dirty_bytes += found_end + 1 - found_start;
4204 cur = found_end + 1;
4207 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4208 trans->transid, dirty_bytes);
4209 btrfs_cleanup_one_transaction(trans, fs_info);
4211 if (trans == fs_info->running_transaction)
4212 fs_info->running_transaction = NULL;
4213 list_del_init(&trans->list);
4215 btrfs_put_transaction(trans);
4216 trace_btrfs_transaction_commit(fs_info);
4221 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4225 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4228 * If we had UNFINISHED_DROPS we could still be processing them, so
4229 * clear that bit and wake up relocation so it can stop.
4230 * We must do this before stopping the block group reclaim task, because
4231 * at btrfs_relocate_block_group() we wait for this bit, and after the
4232 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4233 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4236 btrfs_wake_unfinished_drop(fs_info);
4239 * We may have the reclaim task running and relocating a data block group,
4240 * in which case it may create delayed iputs. So stop it before we park
4241 * the cleaner kthread otherwise we can get new delayed iputs after
4242 * parking the cleaner, and that can make the async reclaim task to hang
4243 * if it's waiting for delayed iputs to complete, since the cleaner is
4244 * parked and can not run delayed iputs - this will make us hang when
4245 * trying to stop the async reclaim task.
4247 cancel_work_sync(&fs_info->reclaim_bgs_work);
4249 * We don't want the cleaner to start new transactions, add more delayed
4250 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4251 * because that frees the task_struct, and the transaction kthread might
4252 * still try to wake up the cleaner.
4254 kthread_park(fs_info->cleaner_kthread);
4256 /* wait for the qgroup rescan worker to stop */
4257 btrfs_qgroup_wait_for_completion(fs_info, false);
4259 /* wait for the uuid_scan task to finish */
4260 down(&fs_info->uuid_tree_rescan_sem);
4261 /* avoid complains from lockdep et al., set sem back to initial state */
4262 up(&fs_info->uuid_tree_rescan_sem);
4264 /* pause restriper - we want to resume on mount */
4265 btrfs_pause_balance(fs_info);
4267 btrfs_dev_replace_suspend_for_unmount(fs_info);
4269 btrfs_scrub_cancel(fs_info);
4271 /* wait for any defraggers to finish */
4272 wait_event(fs_info->transaction_wait,
4273 (atomic_read(&fs_info->defrag_running) == 0));
4275 /* clear out the rbtree of defraggable inodes */
4276 btrfs_cleanup_defrag_inodes(fs_info);
4279 * After we parked the cleaner kthread, ordered extents may have
4280 * completed and created new delayed iputs. If one of the async reclaim
4281 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4282 * can hang forever trying to stop it, because if a delayed iput is
4283 * added after it ran btrfs_run_delayed_iputs() and before it called
4284 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4285 * no one else to run iputs.
4287 * So wait for all ongoing ordered extents to complete and then run
4288 * delayed iputs. This works because once we reach this point no one
4289 * can either create new ordered extents nor create delayed iputs
4290 * through some other means.
4292 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4293 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4294 * but the delayed iput for the respective inode is made only when doing
4295 * the final btrfs_put_ordered_extent() (which must happen at
4296 * btrfs_finish_ordered_io() when we are unmounting).
4298 btrfs_flush_workqueue(fs_info->endio_write_workers);
4299 /* Ordered extents for free space inodes. */
4300 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4301 btrfs_run_delayed_iputs(fs_info);
4303 cancel_work_sync(&fs_info->async_reclaim_work);
4304 cancel_work_sync(&fs_info->async_data_reclaim_work);
4305 cancel_work_sync(&fs_info->preempt_reclaim_work);
4307 /* Cancel or finish ongoing discard work */
4308 btrfs_discard_cleanup(fs_info);
4310 if (!sb_rdonly(fs_info->sb)) {
4312 * The cleaner kthread is stopped, so do one final pass over
4313 * unused block groups.
4315 btrfs_delete_unused_bgs(fs_info);
4318 * There might be existing delayed inode workers still running
4319 * and holding an empty delayed inode item. We must wait for
4320 * them to complete first because they can create a transaction.
4321 * This happens when someone calls btrfs_balance_delayed_items()
4322 * and then a transaction commit runs the same delayed nodes
4323 * before any delayed worker has done something with the nodes.
4324 * We must wait for any worker here and not at transaction
4325 * commit time since that could cause a deadlock.
4326 * This is a very rare case.
4328 btrfs_flush_workqueue(fs_info->delayed_workers);
4330 ret = btrfs_commit_super(fs_info);
4332 btrfs_err(fs_info, "commit super ret %d", ret);
4335 if (BTRFS_FS_ERROR(fs_info))
4336 btrfs_error_commit_super(fs_info);
4338 kthread_stop(fs_info->transaction_kthread);
4339 kthread_stop(fs_info->cleaner_kthread);
4341 ASSERT(list_empty(&fs_info->delayed_iputs));
4342 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4344 if (btrfs_check_quota_leak(fs_info)) {
4345 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4346 btrfs_err(fs_info, "qgroup reserved space leaked");
4349 btrfs_free_qgroup_config(fs_info);
4350 ASSERT(list_empty(&fs_info->delalloc_roots));
4352 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4353 btrfs_info(fs_info, "at unmount delalloc count %lld",
4354 percpu_counter_sum(&fs_info->delalloc_bytes));
4357 if (percpu_counter_sum(&fs_info->ordered_bytes))
4358 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4359 percpu_counter_sum(&fs_info->ordered_bytes));
4361 btrfs_sysfs_remove_mounted(fs_info);
4362 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4364 btrfs_put_block_group_cache(fs_info);
4367 * we must make sure there is not any read request to
4368 * submit after we stopping all workers.
4370 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4371 btrfs_stop_all_workers(fs_info);
4373 /* We shouldn't have any transaction open at this point */
4374 warn_about_uncommitted_trans(fs_info);
4376 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4377 free_root_pointers(fs_info, true);
4378 btrfs_free_fs_roots(fs_info);
4381 * We must free the block groups after dropping the fs_roots as we could
4382 * have had an IO error and have left over tree log blocks that aren't
4383 * cleaned up until the fs roots are freed. This makes the block group
4384 * accounting appear to be wrong because there's pending reserved bytes,
4385 * so make sure we do the block group cleanup afterwards.
4387 btrfs_free_block_groups(fs_info);
4389 iput(fs_info->btree_inode);
4391 btrfs_mapping_tree_free(fs_info);
4392 btrfs_close_devices(fs_info->fs_devices);
4395 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4396 struct extent_buffer *buf)
4398 struct btrfs_fs_info *fs_info = buf->fs_info;
4399 u64 transid = btrfs_header_generation(buf);
4401 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4403 * This is a fast path so only do this check if we have sanity tests
4404 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4405 * outside of the sanity tests.
4407 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4410 /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4411 ASSERT(trans->transid == fs_info->generation);
4412 btrfs_assert_tree_write_locked(buf);
4413 if (unlikely(transid != fs_info->generation)) {
4414 btrfs_abort_transaction(trans, -EUCLEAN);
4416 "dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4417 buf->start, transid, fs_info->generation);
4419 set_extent_buffer_dirty(buf);
4422 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4426 * looks as though older kernels can get into trouble with
4427 * this code, they end up stuck in balance_dirty_pages forever
4431 if (current->flags & PF_MEMALLOC)
4435 btrfs_balance_delayed_items(fs_info);
4437 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4438 BTRFS_DIRTY_METADATA_THRESH,
4439 fs_info->dirty_metadata_batch);
4441 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4445 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4447 __btrfs_btree_balance_dirty(fs_info, 1);
4450 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4452 __btrfs_btree_balance_dirty(fs_info, 0);
4455 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4457 /* cleanup FS via transaction */
4458 btrfs_cleanup_transaction(fs_info);
4460 mutex_lock(&fs_info->cleaner_mutex);
4461 btrfs_run_delayed_iputs(fs_info);
4462 mutex_unlock(&fs_info->cleaner_mutex);
4464 down_write(&fs_info->cleanup_work_sem);
4465 up_write(&fs_info->cleanup_work_sem);
4468 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4470 struct btrfs_root *gang[8];
4471 u64 root_objectid = 0;
4474 spin_lock(&fs_info->fs_roots_radix_lock);
4475 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4476 (void **)gang, root_objectid,
4477 ARRAY_SIZE(gang))) != 0) {
4480 for (i = 0; i < ret; i++)
4481 gang[i] = btrfs_grab_root(gang[i]);
4482 spin_unlock(&fs_info->fs_roots_radix_lock);
4484 for (i = 0; i < ret; i++) {
4487 root_objectid = gang[i]->root_key.objectid;
4488 btrfs_free_log(NULL, gang[i]);
4489 btrfs_put_root(gang[i]);
4492 spin_lock(&fs_info->fs_roots_radix_lock);
4494 spin_unlock(&fs_info->fs_roots_radix_lock);
4495 btrfs_free_log_root_tree(NULL, fs_info);
4498 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4500 struct btrfs_ordered_extent *ordered;
4502 spin_lock(&root->ordered_extent_lock);
4504 * This will just short circuit the ordered completion stuff which will
4505 * make sure the ordered extent gets properly cleaned up.
4507 list_for_each_entry(ordered, &root->ordered_extents,
4509 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4510 spin_unlock(&root->ordered_extent_lock);
4513 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4515 struct btrfs_root *root;
4518 spin_lock(&fs_info->ordered_root_lock);
4519 list_splice_init(&fs_info->ordered_roots, &splice);
4520 while (!list_empty(&splice)) {
4521 root = list_first_entry(&splice, struct btrfs_root,
4523 list_move_tail(&root->ordered_root,
4524 &fs_info->ordered_roots);
4526 spin_unlock(&fs_info->ordered_root_lock);
4527 btrfs_destroy_ordered_extents(root);
4530 spin_lock(&fs_info->ordered_root_lock);
4532 spin_unlock(&fs_info->ordered_root_lock);
4535 * We need this here because if we've been flipped read-only we won't
4536 * get sync() from the umount, so we need to make sure any ordered
4537 * extents that haven't had their dirty pages IO start writeout yet
4538 * actually get run and error out properly.
4540 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4543 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4544 struct btrfs_fs_info *fs_info)
4546 struct rb_node *node;
4547 struct btrfs_delayed_ref_root *delayed_refs;
4548 struct btrfs_delayed_ref_node *ref;
4550 delayed_refs = &trans->delayed_refs;
4552 spin_lock(&delayed_refs->lock);
4553 if (atomic_read(&delayed_refs->num_entries) == 0) {
4554 spin_unlock(&delayed_refs->lock);
4555 btrfs_debug(fs_info, "delayed_refs has NO entry");
4559 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4560 struct btrfs_delayed_ref_head *head;
4562 bool pin_bytes = false;
4564 head = rb_entry(node, struct btrfs_delayed_ref_head,
4566 if (btrfs_delayed_ref_lock(delayed_refs, head))
4569 spin_lock(&head->lock);
4570 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4571 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4573 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4574 RB_CLEAR_NODE(&ref->ref_node);
4575 if (!list_empty(&ref->add_list))
4576 list_del(&ref->add_list);
4577 atomic_dec(&delayed_refs->num_entries);
4578 btrfs_put_delayed_ref(ref);
4579 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4581 if (head->must_insert_reserved)
4583 btrfs_free_delayed_extent_op(head->extent_op);
4584 btrfs_delete_ref_head(delayed_refs, head);
4585 spin_unlock(&head->lock);
4586 spin_unlock(&delayed_refs->lock);
4587 mutex_unlock(&head->mutex);
4590 struct btrfs_block_group *cache;
4592 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4595 spin_lock(&cache->space_info->lock);
4596 spin_lock(&cache->lock);
4597 cache->pinned += head->num_bytes;
4598 btrfs_space_info_update_bytes_pinned(fs_info,
4599 cache->space_info, head->num_bytes);
4600 cache->reserved -= head->num_bytes;
4601 cache->space_info->bytes_reserved -= head->num_bytes;
4602 spin_unlock(&cache->lock);
4603 spin_unlock(&cache->space_info->lock);
4605 btrfs_put_block_group(cache);
4607 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4608 head->bytenr + head->num_bytes - 1);
4610 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4611 btrfs_put_delayed_ref_head(head);
4613 spin_lock(&delayed_refs->lock);
4615 btrfs_qgroup_destroy_extent_records(trans);
4617 spin_unlock(&delayed_refs->lock);
4620 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4622 struct btrfs_inode *btrfs_inode;
4625 spin_lock(&root->delalloc_lock);
4626 list_splice_init(&root->delalloc_inodes, &splice);
4628 while (!list_empty(&splice)) {
4629 struct inode *inode = NULL;
4630 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4632 btrfs_del_delalloc_inode(btrfs_inode);
4633 spin_unlock(&root->delalloc_lock);
4636 * Make sure we get a live inode and that it'll not disappear
4639 inode = igrab(&btrfs_inode->vfs_inode);
4641 unsigned int nofs_flag;
4643 nofs_flag = memalloc_nofs_save();
4644 invalidate_inode_pages2(inode->i_mapping);
4645 memalloc_nofs_restore(nofs_flag);
4648 spin_lock(&root->delalloc_lock);
4650 spin_unlock(&root->delalloc_lock);
4653 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4655 struct btrfs_root *root;
4658 spin_lock(&fs_info->delalloc_root_lock);
4659 list_splice_init(&fs_info->delalloc_roots, &splice);
4660 while (!list_empty(&splice)) {
4661 root = list_first_entry(&splice, struct btrfs_root,
4663 root = btrfs_grab_root(root);
4665 spin_unlock(&fs_info->delalloc_root_lock);
4667 btrfs_destroy_delalloc_inodes(root);
4668 btrfs_put_root(root);
4670 spin_lock(&fs_info->delalloc_root_lock);
4672 spin_unlock(&fs_info->delalloc_root_lock);
4675 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4676 struct extent_io_tree *dirty_pages,
4679 struct extent_buffer *eb;
4683 while (find_first_extent_bit(dirty_pages, start, &start, &end,
4685 clear_extent_bits(dirty_pages, start, end, mark);
4686 while (start <= end) {
4687 eb = find_extent_buffer(fs_info, start);
4688 start += fs_info->nodesize;
4692 btrfs_tree_lock(eb);
4693 wait_on_extent_buffer_writeback(eb);
4694 btrfs_clear_buffer_dirty(NULL, eb);
4695 btrfs_tree_unlock(eb);
4697 free_extent_buffer_stale(eb);
4702 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4703 struct extent_io_tree *unpin)
4709 struct extent_state *cached_state = NULL;
4712 * The btrfs_finish_extent_commit() may get the same range as
4713 * ours between find_first_extent_bit and clear_extent_dirty.
4714 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4715 * the same extent range.
4717 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4718 if (!find_first_extent_bit(unpin, 0, &start, &end,
4719 EXTENT_DIRTY, &cached_state)) {
4720 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4724 clear_extent_dirty(unpin, start, end, &cached_state);
4725 free_extent_state(cached_state);
4726 btrfs_error_unpin_extent_range(fs_info, start, end);
4727 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4732 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4734 struct inode *inode;
4736 inode = cache->io_ctl.inode;
4738 unsigned int nofs_flag;
4740 nofs_flag = memalloc_nofs_save();
4741 invalidate_inode_pages2(inode->i_mapping);
4742 memalloc_nofs_restore(nofs_flag);
4744 BTRFS_I(inode)->generation = 0;
4745 cache->io_ctl.inode = NULL;
4748 ASSERT(cache->io_ctl.pages == NULL);
4749 btrfs_put_block_group(cache);
4752 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4753 struct btrfs_fs_info *fs_info)
4755 struct btrfs_block_group *cache;
4757 spin_lock(&cur_trans->dirty_bgs_lock);
4758 while (!list_empty(&cur_trans->dirty_bgs)) {
4759 cache = list_first_entry(&cur_trans->dirty_bgs,
4760 struct btrfs_block_group,
4763 if (!list_empty(&cache->io_list)) {
4764 spin_unlock(&cur_trans->dirty_bgs_lock);
4765 list_del_init(&cache->io_list);
4766 btrfs_cleanup_bg_io(cache);
4767 spin_lock(&cur_trans->dirty_bgs_lock);
4770 list_del_init(&cache->dirty_list);
4771 spin_lock(&cache->lock);
4772 cache->disk_cache_state = BTRFS_DC_ERROR;
4773 spin_unlock(&cache->lock);
4775 spin_unlock(&cur_trans->dirty_bgs_lock);
4776 btrfs_put_block_group(cache);
4777 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4778 spin_lock(&cur_trans->dirty_bgs_lock);
4780 spin_unlock(&cur_trans->dirty_bgs_lock);
4783 * Refer to the definition of io_bgs member for details why it's safe
4784 * to use it without any locking
4786 while (!list_empty(&cur_trans->io_bgs)) {
4787 cache = list_first_entry(&cur_trans->io_bgs,
4788 struct btrfs_block_group,
4791 list_del_init(&cache->io_list);
4792 spin_lock(&cache->lock);
4793 cache->disk_cache_state = BTRFS_DC_ERROR;
4794 spin_unlock(&cache->lock);
4795 btrfs_cleanup_bg_io(cache);
4799 static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4801 struct btrfs_root *gang[8];
4805 spin_lock(&fs_info->fs_roots_radix_lock);
4807 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4810 BTRFS_ROOT_TRANS_TAG);
4813 for (i = 0; i < ret; i++) {
4814 struct btrfs_root *root = gang[i];
4816 btrfs_qgroup_free_meta_all_pertrans(root);
4817 radix_tree_tag_clear(&fs_info->fs_roots_radix,
4818 (unsigned long)root->root_key.objectid,
4819 BTRFS_ROOT_TRANS_TAG);
4822 spin_unlock(&fs_info->fs_roots_radix_lock);
4825 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4826 struct btrfs_fs_info *fs_info)
4828 struct btrfs_device *dev, *tmp;
4830 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4831 ASSERT(list_empty(&cur_trans->dirty_bgs));
4832 ASSERT(list_empty(&cur_trans->io_bgs));
4834 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4836 list_del_init(&dev->post_commit_list);
4839 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4841 cur_trans->state = TRANS_STATE_COMMIT_START;
4842 wake_up(&fs_info->transaction_blocked_wait);
4844 cur_trans->state = TRANS_STATE_UNBLOCKED;
4845 wake_up(&fs_info->transaction_wait);
4847 btrfs_destroy_delayed_inodes(fs_info);
4849 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4851 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4853 btrfs_free_all_qgroup_pertrans(fs_info);
4855 cur_trans->state =TRANS_STATE_COMPLETED;
4856 wake_up(&cur_trans->commit_wait);
4859 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4861 struct btrfs_transaction *t;
4863 mutex_lock(&fs_info->transaction_kthread_mutex);
4865 spin_lock(&fs_info->trans_lock);
4866 while (!list_empty(&fs_info->trans_list)) {
4867 t = list_first_entry(&fs_info->trans_list,
4868 struct btrfs_transaction, list);
4869 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4870 refcount_inc(&t->use_count);
4871 spin_unlock(&fs_info->trans_lock);
4872 btrfs_wait_for_commit(fs_info, t->transid);
4873 btrfs_put_transaction(t);
4874 spin_lock(&fs_info->trans_lock);
4877 if (t == fs_info->running_transaction) {
4878 t->state = TRANS_STATE_COMMIT_DOING;
4879 spin_unlock(&fs_info->trans_lock);
4881 * We wait for 0 num_writers since we don't hold a trans
4882 * handle open currently for this transaction.
4884 wait_event(t->writer_wait,
4885 atomic_read(&t->num_writers) == 0);
4887 spin_unlock(&fs_info->trans_lock);
4889 btrfs_cleanup_one_transaction(t, fs_info);
4891 spin_lock(&fs_info->trans_lock);
4892 if (t == fs_info->running_transaction)
4893 fs_info->running_transaction = NULL;
4894 list_del_init(&t->list);
4895 spin_unlock(&fs_info->trans_lock);
4897 btrfs_put_transaction(t);
4898 trace_btrfs_transaction_commit(fs_info);
4899 spin_lock(&fs_info->trans_lock);
4901 spin_unlock(&fs_info->trans_lock);
4902 btrfs_destroy_all_ordered_extents(fs_info);
4903 btrfs_destroy_delayed_inodes(fs_info);
4904 btrfs_assert_delayed_root_empty(fs_info);
4905 btrfs_destroy_all_delalloc_inodes(fs_info);
4906 btrfs_drop_all_logs(fs_info);
4907 mutex_unlock(&fs_info->transaction_kthread_mutex);
4912 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4914 struct btrfs_path *path;
4916 struct extent_buffer *l;
4917 struct btrfs_key search_key;
4918 struct btrfs_key found_key;
4921 path = btrfs_alloc_path();
4925 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4926 search_key.type = -1;
4927 search_key.offset = (u64)-1;
4928 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4933 * Key with offset -1 found, there would have to exist a root
4934 * with such id, but this is out of valid range.
4939 if (path->slots[0] > 0) {
4940 slot = path->slots[0] - 1;
4942 btrfs_item_key_to_cpu(l, &found_key, slot);
4943 root->free_objectid = max_t(u64, found_key.objectid + 1,
4944 BTRFS_FIRST_FREE_OBJECTID);
4946 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4950 btrfs_free_path(path);
4954 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4957 mutex_lock(&root->objectid_mutex);
4959 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4960 btrfs_warn(root->fs_info,
4961 "the objectid of root %llu reaches its highest value",
4962 root->root_key.objectid);
4967 *objectid = root->free_objectid++;
4970 mutex_unlock(&root->objectid_mutex);