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 "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41 #include "block-group.h"
43 #include "space-info.h"
47 #include "accessors.h"
48 #include "extent-tree.h"
49 #include "root-tree.h"
51 #include "uuid-tree.h"
52 #include "relocation.h"
56 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
57 BTRFS_HEADER_FLAG_RELOC |\
58 BTRFS_SUPER_FLAG_ERROR |\
59 BTRFS_SUPER_FLAG_SEEDING |\
60 BTRFS_SUPER_FLAG_METADUMP |\
61 BTRFS_SUPER_FLAG_METADUMP_V2)
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
68 if (fs_info->csum_shash)
69 crypto_free_shash(fs_info->csum_shash);
73 * Compute the csum of a btree block and store the result to provided buffer.
75 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
77 struct btrfs_fs_info *fs_info = buf->fs_info;
78 const int num_pages = num_extent_pages(buf);
79 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
80 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
84 shash->tfm = fs_info->csum_shash;
85 crypto_shash_init(shash);
86 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
87 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
88 first_page_part - BTRFS_CSUM_SIZE);
90 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
91 kaddr = page_address(buf->pages[i]);
92 crypto_shash_update(shash, kaddr, PAGE_SIZE);
94 memset(result, 0, BTRFS_CSUM_SIZE);
95 crypto_shash_final(shash, result);
99 * we can't consider a given block up to date unless the transid of the
100 * block matches the transid in the parent node's pointer. This is how we
101 * detect blocks that either didn't get written at all or got written
102 * in the wrong place.
104 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
106 if (!extent_buffer_uptodate(eb))
109 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
115 if (!extent_buffer_uptodate(eb) ||
116 btrfs_header_generation(eb) != parent_transid) {
117 btrfs_err_rl(eb->fs_info,
118 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
119 eb->start, eb->read_mirror,
120 parent_transid, btrfs_header_generation(eb));
121 clear_extent_buffer_uptodate(eb);
127 static bool btrfs_supported_super_csum(u16 csum_type)
130 case BTRFS_CSUM_TYPE_CRC32:
131 case BTRFS_CSUM_TYPE_XXHASH:
132 case BTRFS_CSUM_TYPE_SHA256:
133 case BTRFS_CSUM_TYPE_BLAKE2:
141 * Return 0 if the superblock checksum type matches the checksum value of that
142 * algorithm. Pass the raw disk superblock data.
144 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
145 const struct btrfs_super_block *disk_sb)
147 char result[BTRFS_CSUM_SIZE];
148 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
150 shash->tfm = fs_info->csum_shash;
153 * The super_block structure does not span the whole
154 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
155 * filled with zeros and is included in the checksum.
157 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
158 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
160 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
166 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
169 struct btrfs_fs_info *fs_info = eb->fs_info;
170 int i, num_pages = num_extent_pages(eb);
173 if (sb_rdonly(fs_info->sb))
176 for (i = 0; i < num_pages; i++) {
177 struct page *p = eb->pages[i];
178 u64 start = max_t(u64, eb->start, page_offset(p));
179 u64 end = min_t(u64, eb->start + eb->len, page_offset(p) + PAGE_SIZE);
180 u32 len = end - start;
182 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
183 start, p, offset_in_page(start), mirror_num);
192 * helper to read a given tree block, doing retries as required when
193 * the checksums don't match and we have alternate mirrors to try.
195 * @check: expected tree parentness check, see the comments of the
196 * structure for details.
198 int btrfs_read_extent_buffer(struct extent_buffer *eb,
199 struct btrfs_tree_parent_check *check)
201 struct btrfs_fs_info *fs_info = eb->fs_info;
206 int failed_mirror = 0;
211 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
212 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
216 num_copies = btrfs_num_copies(fs_info,
221 if (!failed_mirror) {
223 failed_mirror = eb->read_mirror;
227 if (mirror_num == failed_mirror)
230 if (mirror_num > num_copies)
234 if (failed && !ret && failed_mirror)
235 btrfs_repair_eb_io_failure(eb, failed_mirror);
241 * Checksum a dirty tree block before IO.
243 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
245 struct extent_buffer *eb = bbio->private;
246 struct btrfs_fs_info *fs_info = eb->fs_info;
247 u64 found_start = btrfs_header_bytenr(eb);
248 u8 result[BTRFS_CSUM_SIZE];
251 /* Btree blocks are always contiguous on disk. */
252 if (WARN_ON_ONCE(bbio->file_offset != eb->start))
253 return BLK_STS_IOERR;
254 if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
255 return BLK_STS_IOERR;
257 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
258 WARN_ON_ONCE(found_start != 0);
262 if (WARN_ON_ONCE(found_start != eb->start))
263 return BLK_STS_IOERR;
264 if (WARN_ON(!btrfs_page_test_uptodate(fs_info, eb->pages[0], eb->start,
266 return BLK_STS_IOERR;
268 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
269 offsetof(struct btrfs_header, fsid),
270 BTRFS_FSID_SIZE) == 0);
271 csum_tree_block(eb, result);
273 if (btrfs_header_level(eb))
274 ret = btrfs_check_node(eb);
276 ret = btrfs_check_leaf(eb);
282 * Also check the generation, the eb reached here must be newer than
283 * last committed. Or something seriously wrong happened.
285 if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
288 "block=%llu bad generation, have %llu expect > %llu",
289 eb->start, btrfs_header_generation(eb),
290 fs_info->last_trans_committed);
293 write_extent_buffer(eb, result, 0, fs_info->csum_size);
297 btrfs_print_tree(eb, 0);
298 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
301 * Be noisy if this is an extent buffer from a log tree. We don't abort
302 * a transaction in case there's a bad log tree extent buffer, we just
303 * fallback to a transaction commit. Still we want to know when there is
304 * a bad log tree extent buffer, as that may signal a bug somewhere.
306 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
307 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
308 return errno_to_blk_status(ret);
311 static bool check_tree_block_fsid(struct extent_buffer *eb)
313 struct btrfs_fs_info *fs_info = eb->fs_info;
314 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
315 u8 fsid[BTRFS_FSID_SIZE];
318 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
321 * Checking the incompat flag is only valid for the current fs. For
322 * seed devices it's forbidden to have their uuid changed so reading
323 * ->fsid in this case is fine
325 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
326 metadata_uuid = fs_devices->metadata_uuid;
328 metadata_uuid = fs_devices->fsid;
330 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
333 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
334 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
340 /* Do basic extent buffer checks at read time */
341 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
342 struct btrfs_tree_parent_check *check)
344 struct btrfs_fs_info *fs_info = eb->fs_info;
346 const u32 csum_size = fs_info->csum_size;
348 u8 result[BTRFS_CSUM_SIZE];
349 const u8 *header_csum;
354 found_start = btrfs_header_bytenr(eb);
355 if (found_start != eb->start) {
356 btrfs_err_rl(fs_info,
357 "bad tree block start, mirror %u want %llu have %llu",
358 eb->read_mirror, eb->start, found_start);
362 if (check_tree_block_fsid(eb)) {
363 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
364 eb->start, eb->read_mirror);
368 found_level = btrfs_header_level(eb);
369 if (found_level >= BTRFS_MAX_LEVEL) {
371 "bad tree block level, mirror %u level %d on logical %llu",
372 eb->read_mirror, btrfs_header_level(eb), eb->start);
377 csum_tree_block(eb, result);
378 header_csum = page_address(eb->pages[0]) +
379 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
381 if (memcmp(result, header_csum, csum_size) != 0) {
382 btrfs_warn_rl(fs_info,
383 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
384 eb->start, eb->read_mirror,
385 CSUM_FMT_VALUE(csum_size, header_csum),
386 CSUM_FMT_VALUE(csum_size, result),
387 btrfs_header_level(eb));
392 if (found_level != check->level) {
394 "level verify failed on logical %llu mirror %u wanted %u found %u",
395 eb->start, eb->read_mirror, check->level, found_level);
399 if (unlikely(check->transid &&
400 btrfs_header_generation(eb) != check->transid)) {
401 btrfs_err_rl(eb->fs_info,
402 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
403 eb->start, eb->read_mirror, check->transid,
404 btrfs_header_generation(eb));
408 if (check->has_first_key) {
409 struct btrfs_key *expect_key = &check->first_key;
410 struct btrfs_key found_key;
413 btrfs_node_key_to_cpu(eb, &found_key, 0);
415 btrfs_item_key_to_cpu(eb, &found_key, 0);
416 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
418 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
419 eb->start, check->transid,
420 expect_key->objectid,
421 expect_key->type, expect_key->offset,
422 found_key.objectid, found_key.type,
428 if (check->owner_root) {
429 ret = btrfs_check_eb_owner(eb, check->owner_root);
435 * If this is a leaf block and it is corrupt, set the corrupt bit so
436 * that we don't try and read the other copies of this block, just
439 if (found_level == 0 && btrfs_check_leaf(eb)) {
440 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
444 if (found_level > 0 && btrfs_check_node(eb))
449 "read time tree block corruption detected on logical %llu mirror %u",
450 eb->start, eb->read_mirror);
455 #ifdef CONFIG_MIGRATION
456 static int btree_migrate_folio(struct address_space *mapping,
457 struct folio *dst, struct folio *src, enum migrate_mode mode)
460 * we can't safely write a btree page from here,
461 * we haven't done the locking hook
463 if (folio_test_dirty(src))
466 * Buffers may be managed in a filesystem specific way.
467 * We must have no buffers or drop them.
469 if (folio_get_private(src) &&
470 !filemap_release_folio(src, GFP_KERNEL))
472 return migrate_folio(mapping, dst, src, mode);
475 #define btree_migrate_folio NULL
478 static int btree_writepages(struct address_space *mapping,
479 struct writeback_control *wbc)
481 struct btrfs_fs_info *fs_info;
484 if (wbc->sync_mode == WB_SYNC_NONE) {
486 if (wbc->for_kupdate)
489 fs_info = BTRFS_I(mapping->host)->root->fs_info;
490 /* this is a bit racy, but that's ok */
491 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
492 BTRFS_DIRTY_METADATA_THRESH,
493 fs_info->dirty_metadata_batch);
497 return btree_write_cache_pages(mapping, wbc);
500 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
502 if (folio_test_writeback(folio) || folio_test_dirty(folio))
505 return try_release_extent_buffer(&folio->page);
508 static void btree_invalidate_folio(struct folio *folio, size_t offset,
511 struct extent_io_tree *tree;
512 tree = &BTRFS_I(folio->mapping->host)->io_tree;
513 extent_invalidate_folio(tree, folio, offset);
514 btree_release_folio(folio, GFP_NOFS);
515 if (folio_get_private(folio)) {
516 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
517 "folio private not zero on folio %llu",
518 (unsigned long long)folio_pos(folio));
519 folio_detach_private(folio);
524 static bool btree_dirty_folio(struct address_space *mapping,
527 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
528 struct btrfs_subpage *subpage;
529 struct extent_buffer *eb;
531 u64 page_start = folio_pos(folio);
533 if (fs_info->sectorsize == PAGE_SIZE) {
534 eb = folio_get_private(folio);
536 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
537 BUG_ON(!atomic_read(&eb->refs));
538 btrfs_assert_tree_write_locked(eb);
539 return filemap_dirty_folio(mapping, folio);
541 subpage = folio_get_private(folio);
543 ASSERT(subpage->dirty_bitmap);
544 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
547 u16 tmp = (1 << cur_bit);
549 spin_lock_irqsave(&subpage->lock, flags);
550 if (!(tmp & subpage->dirty_bitmap)) {
551 spin_unlock_irqrestore(&subpage->lock, flags);
555 spin_unlock_irqrestore(&subpage->lock, flags);
556 cur = page_start + cur_bit * fs_info->sectorsize;
558 eb = find_extent_buffer(fs_info, cur);
560 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
561 ASSERT(atomic_read(&eb->refs));
562 btrfs_assert_tree_write_locked(eb);
563 free_extent_buffer(eb);
565 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
567 return filemap_dirty_folio(mapping, folio);
570 #define btree_dirty_folio filemap_dirty_folio
573 static const struct address_space_operations btree_aops = {
574 .writepages = btree_writepages,
575 .release_folio = btree_release_folio,
576 .invalidate_folio = btree_invalidate_folio,
577 .migrate_folio = btree_migrate_folio,
578 .dirty_folio = btree_dirty_folio,
581 struct extent_buffer *btrfs_find_create_tree_block(
582 struct btrfs_fs_info *fs_info,
583 u64 bytenr, u64 owner_root,
586 if (btrfs_is_testing(fs_info))
587 return alloc_test_extent_buffer(fs_info, bytenr);
588 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
592 * Read tree block at logical address @bytenr and do variant basic but critical
595 * @check: expected tree parentness check, see comments of the
596 * structure for details.
598 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
599 struct btrfs_tree_parent_check *check)
601 struct extent_buffer *buf = NULL;
606 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
611 ret = btrfs_read_extent_buffer(buf, check);
613 free_extent_buffer_stale(buf);
616 if (btrfs_check_eb_owner(buf, check->owner_root)) {
617 free_extent_buffer_stale(buf);
618 return ERR_PTR(-EUCLEAN);
624 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
627 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
629 memset(&root->root_key, 0, sizeof(root->root_key));
630 memset(&root->root_item, 0, sizeof(root->root_item));
631 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
632 root->fs_info = fs_info;
633 root->root_key.objectid = objectid;
635 root->commit_root = NULL;
637 RB_CLEAR_NODE(&root->rb_node);
639 root->last_trans = 0;
640 root->free_objectid = 0;
641 root->nr_delalloc_inodes = 0;
642 root->nr_ordered_extents = 0;
643 root->inode_tree = RB_ROOT;
644 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
646 btrfs_init_root_block_rsv(root);
648 INIT_LIST_HEAD(&root->dirty_list);
649 INIT_LIST_HEAD(&root->root_list);
650 INIT_LIST_HEAD(&root->delalloc_inodes);
651 INIT_LIST_HEAD(&root->delalloc_root);
652 INIT_LIST_HEAD(&root->ordered_extents);
653 INIT_LIST_HEAD(&root->ordered_root);
654 INIT_LIST_HEAD(&root->reloc_dirty_list);
655 INIT_LIST_HEAD(&root->logged_list[0]);
656 INIT_LIST_HEAD(&root->logged_list[1]);
657 spin_lock_init(&root->inode_lock);
658 spin_lock_init(&root->delalloc_lock);
659 spin_lock_init(&root->ordered_extent_lock);
660 spin_lock_init(&root->accounting_lock);
661 spin_lock_init(&root->log_extents_lock[0]);
662 spin_lock_init(&root->log_extents_lock[1]);
663 spin_lock_init(&root->qgroup_meta_rsv_lock);
664 mutex_init(&root->objectid_mutex);
665 mutex_init(&root->log_mutex);
666 mutex_init(&root->ordered_extent_mutex);
667 mutex_init(&root->delalloc_mutex);
668 init_waitqueue_head(&root->qgroup_flush_wait);
669 init_waitqueue_head(&root->log_writer_wait);
670 init_waitqueue_head(&root->log_commit_wait[0]);
671 init_waitqueue_head(&root->log_commit_wait[1]);
672 INIT_LIST_HEAD(&root->log_ctxs[0]);
673 INIT_LIST_HEAD(&root->log_ctxs[1]);
674 atomic_set(&root->log_commit[0], 0);
675 atomic_set(&root->log_commit[1], 0);
676 atomic_set(&root->log_writers, 0);
677 atomic_set(&root->log_batch, 0);
678 refcount_set(&root->refs, 1);
679 atomic_set(&root->snapshot_force_cow, 0);
680 atomic_set(&root->nr_swapfiles, 0);
681 root->log_transid = 0;
682 root->log_transid_committed = -1;
683 root->last_log_commit = 0;
686 extent_io_tree_init(fs_info, &root->dirty_log_pages,
687 IO_TREE_ROOT_DIRTY_LOG_PAGES);
688 extent_io_tree_init(fs_info, &root->log_csum_range,
689 IO_TREE_LOG_CSUM_RANGE);
692 spin_lock_init(&root->root_item_lock);
693 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
694 #ifdef CONFIG_BTRFS_DEBUG
695 INIT_LIST_HEAD(&root->leak_list);
696 spin_lock(&fs_info->fs_roots_radix_lock);
697 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
698 spin_unlock(&fs_info->fs_roots_radix_lock);
702 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
703 u64 objectid, gfp_t flags)
705 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
707 __setup_root(root, fs_info, objectid);
711 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
712 /* Should only be used by the testing infrastructure */
713 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
715 struct btrfs_root *root;
718 return ERR_PTR(-EINVAL);
720 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
722 return ERR_PTR(-ENOMEM);
724 /* We don't use the stripesize in selftest, set it as sectorsize */
725 root->alloc_bytenr = 0;
731 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
733 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
734 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
736 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
739 static int global_root_key_cmp(const void *k, const struct rb_node *node)
741 const struct btrfs_key *key = k;
742 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
744 return btrfs_comp_cpu_keys(key, &root->root_key);
747 int btrfs_global_root_insert(struct btrfs_root *root)
749 struct btrfs_fs_info *fs_info = root->fs_info;
753 write_lock(&fs_info->global_root_lock);
754 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
755 write_unlock(&fs_info->global_root_lock);
759 btrfs_warn(fs_info, "global root %llu %llu already exists",
760 root->root_key.objectid, root->root_key.offset);
765 void btrfs_global_root_delete(struct btrfs_root *root)
767 struct btrfs_fs_info *fs_info = root->fs_info;
769 write_lock(&fs_info->global_root_lock);
770 rb_erase(&root->rb_node, &fs_info->global_root_tree);
771 write_unlock(&fs_info->global_root_lock);
774 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
775 struct btrfs_key *key)
777 struct rb_node *node;
778 struct btrfs_root *root = NULL;
780 read_lock(&fs_info->global_root_lock);
781 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
783 root = container_of(node, struct btrfs_root, rb_node);
784 read_unlock(&fs_info->global_root_lock);
789 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
791 struct btrfs_block_group *block_group;
794 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
798 block_group = btrfs_lookup_block_group(fs_info, bytenr);
800 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
804 ret = block_group->global_root_id;
805 btrfs_put_block_group(block_group);
810 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
812 struct btrfs_key key = {
813 .objectid = BTRFS_CSUM_TREE_OBJECTID,
814 .type = BTRFS_ROOT_ITEM_KEY,
815 .offset = btrfs_global_root_id(fs_info, bytenr),
818 return btrfs_global_root(fs_info, &key);
821 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
823 struct btrfs_key key = {
824 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
825 .type = BTRFS_ROOT_ITEM_KEY,
826 .offset = btrfs_global_root_id(fs_info, bytenr),
829 return btrfs_global_root(fs_info, &key);
832 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
834 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
835 return fs_info->block_group_root;
836 return btrfs_extent_root(fs_info, 0);
839 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
842 struct btrfs_fs_info *fs_info = trans->fs_info;
843 struct extent_buffer *leaf;
844 struct btrfs_root *tree_root = fs_info->tree_root;
845 struct btrfs_root *root;
846 struct btrfs_key key;
847 unsigned int nofs_flag;
851 * We're holding a transaction handle, so use a NOFS memory allocation
852 * context to avoid deadlock if reclaim happens.
854 nofs_flag = memalloc_nofs_save();
855 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
856 memalloc_nofs_restore(nofs_flag);
858 return ERR_PTR(-ENOMEM);
860 root->root_key.objectid = objectid;
861 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
862 root->root_key.offset = 0;
864 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
865 BTRFS_NESTING_NORMAL);
873 btrfs_mark_buffer_dirty(leaf);
875 root->commit_root = btrfs_root_node(root);
876 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
878 btrfs_set_root_flags(&root->root_item, 0);
879 btrfs_set_root_limit(&root->root_item, 0);
880 btrfs_set_root_bytenr(&root->root_item, leaf->start);
881 btrfs_set_root_generation(&root->root_item, trans->transid);
882 btrfs_set_root_level(&root->root_item, 0);
883 btrfs_set_root_refs(&root->root_item, 1);
884 btrfs_set_root_used(&root->root_item, leaf->len);
885 btrfs_set_root_last_snapshot(&root->root_item, 0);
886 btrfs_set_root_dirid(&root->root_item, 0);
887 if (is_fstree(objectid))
888 generate_random_guid(root->root_item.uuid);
890 export_guid(root->root_item.uuid, &guid_null);
891 btrfs_set_root_drop_level(&root->root_item, 0);
893 btrfs_tree_unlock(leaf);
895 key.objectid = objectid;
896 key.type = BTRFS_ROOT_ITEM_KEY;
898 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
905 btrfs_put_root(root);
910 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
911 struct btrfs_fs_info *fs_info)
913 struct btrfs_root *root;
915 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
917 return ERR_PTR(-ENOMEM);
919 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
920 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
921 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
926 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
927 struct btrfs_root *root)
929 struct extent_buffer *leaf;
932 * DON'T set SHAREABLE bit for log trees.
934 * Log trees are not exposed to user space thus can't be snapshotted,
935 * and they go away before a real commit is actually done.
937 * They do store pointers to file data extents, and those reference
938 * counts still get updated (along with back refs to the log tree).
941 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
942 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
944 return PTR_ERR(leaf);
948 btrfs_mark_buffer_dirty(root->node);
949 btrfs_tree_unlock(root->node);
954 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
955 struct btrfs_fs_info *fs_info)
957 struct btrfs_root *log_root;
959 log_root = alloc_log_tree(trans, fs_info);
960 if (IS_ERR(log_root))
961 return PTR_ERR(log_root);
963 if (!btrfs_is_zoned(fs_info)) {
964 int ret = btrfs_alloc_log_tree_node(trans, log_root);
967 btrfs_put_root(log_root);
972 WARN_ON(fs_info->log_root_tree);
973 fs_info->log_root_tree = log_root;
977 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
978 struct btrfs_root *root)
980 struct btrfs_fs_info *fs_info = root->fs_info;
981 struct btrfs_root *log_root;
982 struct btrfs_inode_item *inode_item;
985 log_root = alloc_log_tree(trans, fs_info);
986 if (IS_ERR(log_root))
987 return PTR_ERR(log_root);
989 ret = btrfs_alloc_log_tree_node(trans, log_root);
991 btrfs_put_root(log_root);
995 log_root->last_trans = trans->transid;
996 log_root->root_key.offset = root->root_key.objectid;
998 inode_item = &log_root->root_item.inode;
999 btrfs_set_stack_inode_generation(inode_item, 1);
1000 btrfs_set_stack_inode_size(inode_item, 3);
1001 btrfs_set_stack_inode_nlink(inode_item, 1);
1002 btrfs_set_stack_inode_nbytes(inode_item,
1004 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1006 btrfs_set_root_node(&log_root->root_item, log_root->node);
1008 WARN_ON(root->log_root);
1009 root->log_root = log_root;
1010 root->log_transid = 0;
1011 root->log_transid_committed = -1;
1012 root->last_log_commit = 0;
1016 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1017 struct btrfs_path *path,
1018 struct btrfs_key *key)
1020 struct btrfs_root *root;
1021 struct btrfs_tree_parent_check check = { 0 };
1022 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1027 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1029 return ERR_PTR(-ENOMEM);
1031 ret = btrfs_find_root(tree_root, key, path,
1032 &root->root_item, &root->root_key);
1039 generation = btrfs_root_generation(&root->root_item);
1040 level = btrfs_root_level(&root->root_item);
1041 check.level = level;
1042 check.transid = generation;
1043 check.owner_root = key->objectid;
1044 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1046 if (IS_ERR(root->node)) {
1047 ret = PTR_ERR(root->node);
1051 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1057 * For real fs, and not log/reloc trees, root owner must
1058 * match its root node owner
1060 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1061 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1062 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1063 root->root_key.objectid != btrfs_header_owner(root->node)) {
1065 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1066 root->root_key.objectid, root->node->start,
1067 btrfs_header_owner(root->node),
1068 root->root_key.objectid);
1072 root->commit_root = btrfs_root_node(root);
1075 btrfs_put_root(root);
1076 return ERR_PTR(ret);
1079 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1080 struct btrfs_key *key)
1082 struct btrfs_root *root;
1083 struct btrfs_path *path;
1085 path = btrfs_alloc_path();
1087 return ERR_PTR(-ENOMEM);
1088 root = read_tree_root_path(tree_root, path, key);
1089 btrfs_free_path(path);
1095 * Initialize subvolume root in-memory structure
1097 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1099 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1103 btrfs_drew_lock_init(&root->snapshot_lock);
1105 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1106 !btrfs_is_data_reloc_root(root)) {
1107 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1108 btrfs_check_and_init_root_item(&root->root_item);
1112 * Don't assign anonymous block device to roots that are not exposed to
1113 * userspace, the id pool is limited to 1M
1115 if (is_fstree(root->root_key.objectid) &&
1116 btrfs_root_refs(&root->root_item) > 0) {
1118 ret = get_anon_bdev(&root->anon_dev);
1122 root->anon_dev = anon_dev;
1126 mutex_lock(&root->objectid_mutex);
1127 ret = btrfs_init_root_free_objectid(root);
1129 mutex_unlock(&root->objectid_mutex);
1133 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1135 mutex_unlock(&root->objectid_mutex);
1139 /* The caller is responsible to call btrfs_free_fs_root */
1143 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1146 struct btrfs_root *root;
1148 spin_lock(&fs_info->fs_roots_radix_lock);
1149 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1150 (unsigned long)root_id);
1151 root = btrfs_grab_root(root);
1152 spin_unlock(&fs_info->fs_roots_radix_lock);
1156 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1159 struct btrfs_key key = {
1160 .objectid = objectid,
1161 .type = BTRFS_ROOT_ITEM_KEY,
1166 case BTRFS_ROOT_TREE_OBJECTID:
1167 return btrfs_grab_root(fs_info->tree_root);
1168 case BTRFS_EXTENT_TREE_OBJECTID:
1169 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1170 case BTRFS_CHUNK_TREE_OBJECTID:
1171 return btrfs_grab_root(fs_info->chunk_root);
1172 case BTRFS_DEV_TREE_OBJECTID:
1173 return btrfs_grab_root(fs_info->dev_root);
1174 case BTRFS_CSUM_TREE_OBJECTID:
1175 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1176 case BTRFS_QUOTA_TREE_OBJECTID:
1177 return btrfs_grab_root(fs_info->quota_root);
1178 case BTRFS_UUID_TREE_OBJECTID:
1179 return btrfs_grab_root(fs_info->uuid_root);
1180 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1181 return btrfs_grab_root(fs_info->block_group_root);
1182 case BTRFS_FREE_SPACE_TREE_OBJECTID:
1183 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1189 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1190 struct btrfs_root *root)
1194 ret = radix_tree_preload(GFP_NOFS);
1198 spin_lock(&fs_info->fs_roots_radix_lock);
1199 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1200 (unsigned long)root->root_key.objectid,
1203 btrfs_grab_root(root);
1204 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1206 spin_unlock(&fs_info->fs_roots_radix_lock);
1207 radix_tree_preload_end();
1212 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1214 #ifdef CONFIG_BTRFS_DEBUG
1215 struct btrfs_root *root;
1217 while (!list_empty(&fs_info->allocated_roots)) {
1218 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1220 root = list_first_entry(&fs_info->allocated_roots,
1221 struct btrfs_root, leak_list);
1222 btrfs_err(fs_info, "leaked root %s refcount %d",
1223 btrfs_root_name(&root->root_key, buf),
1224 refcount_read(&root->refs));
1225 while (refcount_read(&root->refs) > 1)
1226 btrfs_put_root(root);
1227 btrfs_put_root(root);
1232 static void free_global_roots(struct btrfs_fs_info *fs_info)
1234 struct btrfs_root *root;
1235 struct rb_node *node;
1237 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1238 root = rb_entry(node, struct btrfs_root, rb_node);
1239 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1240 btrfs_put_root(root);
1244 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1246 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1247 percpu_counter_destroy(&fs_info->delalloc_bytes);
1248 percpu_counter_destroy(&fs_info->ordered_bytes);
1249 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1250 btrfs_free_csum_hash(fs_info);
1251 btrfs_free_stripe_hash_table(fs_info);
1252 btrfs_free_ref_cache(fs_info);
1253 kfree(fs_info->balance_ctl);
1254 kfree(fs_info->delayed_root);
1255 free_global_roots(fs_info);
1256 btrfs_put_root(fs_info->tree_root);
1257 btrfs_put_root(fs_info->chunk_root);
1258 btrfs_put_root(fs_info->dev_root);
1259 btrfs_put_root(fs_info->quota_root);
1260 btrfs_put_root(fs_info->uuid_root);
1261 btrfs_put_root(fs_info->fs_root);
1262 btrfs_put_root(fs_info->data_reloc_root);
1263 btrfs_put_root(fs_info->block_group_root);
1264 btrfs_check_leaked_roots(fs_info);
1265 btrfs_extent_buffer_leak_debug_check(fs_info);
1266 kfree(fs_info->super_copy);
1267 kfree(fs_info->super_for_commit);
1268 kfree(fs_info->subpage_info);
1274 * Get an in-memory reference of a root structure.
1276 * For essential trees like root/extent tree, we grab it from fs_info directly.
1277 * For subvolume trees, we check the cached filesystem roots first. If not
1278 * found, then read it from disk and add it to cached fs roots.
1280 * Caller should release the root by calling btrfs_put_root() after the usage.
1282 * NOTE: Reloc and log trees can't be read by this function as they share the
1283 * same root objectid.
1285 * @objectid: root id
1286 * @anon_dev: preallocated anonymous block device number for new roots,
1287 * pass 0 for new allocation.
1288 * @check_ref: whether to check root item references, If true, return -ENOENT
1291 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1292 u64 objectid, dev_t anon_dev,
1295 struct btrfs_root *root;
1296 struct btrfs_path *path;
1297 struct btrfs_key key;
1300 root = btrfs_get_global_root(fs_info, objectid);
1304 root = btrfs_lookup_fs_root(fs_info, objectid);
1306 /* Shouldn't get preallocated anon_dev for cached roots */
1308 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1309 btrfs_put_root(root);
1310 return ERR_PTR(-ENOENT);
1315 key.objectid = objectid;
1316 key.type = BTRFS_ROOT_ITEM_KEY;
1317 key.offset = (u64)-1;
1318 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1322 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1327 ret = btrfs_init_fs_root(root, anon_dev);
1331 path = btrfs_alloc_path();
1336 key.objectid = BTRFS_ORPHAN_OBJECTID;
1337 key.type = BTRFS_ORPHAN_ITEM_KEY;
1338 key.offset = objectid;
1340 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1341 btrfs_free_path(path);
1345 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1347 ret = btrfs_insert_fs_root(fs_info, root);
1349 if (ret == -EEXIST) {
1350 btrfs_put_root(root);
1358 * If our caller provided us an anonymous device, then it's his
1359 * responsibility to free it in case we fail. So we have to set our
1360 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1361 * and once again by our caller.
1365 btrfs_put_root(root);
1366 return ERR_PTR(ret);
1370 * Get in-memory reference of a root structure
1372 * @objectid: tree objectid
1373 * @check_ref: if set, verify that the tree exists and the item has at least
1376 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1377 u64 objectid, bool check_ref)
1379 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1383 * Get in-memory reference of a root structure, created as new, optionally pass
1384 * the anonymous block device id
1386 * @objectid: tree objectid
1387 * @anon_dev: if zero, allocate a new anonymous block device or use the
1390 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1391 u64 objectid, dev_t anon_dev)
1393 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1397 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1398 * @fs_info: the fs_info
1399 * @objectid: the objectid we need to lookup
1401 * This is exclusively used for backref walking, and exists specifically because
1402 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1403 * creation time, which means we may have to read the tree_root in order to look
1404 * up a fs root that is not in memory. If the root is not in memory we will
1405 * read the tree root commit root and look up the fs root from there. This is a
1406 * temporary root, it will not be inserted into the radix tree as it doesn't
1407 * have the most uptodate information, it'll simply be discarded once the
1408 * backref code is finished using the root.
1410 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1411 struct btrfs_path *path,
1414 struct btrfs_root *root;
1415 struct btrfs_key key;
1417 ASSERT(path->search_commit_root && path->skip_locking);
1420 * This can return -ENOENT if we ask for a root that doesn't exist, but
1421 * since this is called via the backref walking code we won't be looking
1422 * up a root that doesn't exist, unless there's corruption. So if root
1423 * != NULL just return it.
1425 root = btrfs_get_global_root(fs_info, objectid);
1429 root = btrfs_lookup_fs_root(fs_info, objectid);
1433 key.objectid = objectid;
1434 key.type = BTRFS_ROOT_ITEM_KEY;
1435 key.offset = (u64)-1;
1436 root = read_tree_root_path(fs_info->tree_root, path, &key);
1437 btrfs_release_path(path);
1442 static int cleaner_kthread(void *arg)
1444 struct btrfs_fs_info *fs_info = arg;
1450 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1452 /* Make the cleaner go to sleep early. */
1453 if (btrfs_need_cleaner_sleep(fs_info))
1457 * Do not do anything if we might cause open_ctree() to block
1458 * before we have finished mounting the filesystem.
1460 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1463 if (!mutex_trylock(&fs_info->cleaner_mutex))
1467 * Avoid the problem that we change the status of the fs
1468 * during the above check and trylock.
1470 if (btrfs_need_cleaner_sleep(fs_info)) {
1471 mutex_unlock(&fs_info->cleaner_mutex);
1475 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1476 btrfs_sysfs_feature_update(fs_info);
1478 btrfs_run_delayed_iputs(fs_info);
1480 again = btrfs_clean_one_deleted_snapshot(fs_info);
1481 mutex_unlock(&fs_info->cleaner_mutex);
1484 * The defragger has dealt with the R/O remount and umount,
1485 * needn't do anything special here.
1487 btrfs_run_defrag_inodes(fs_info);
1490 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1491 * with relocation (btrfs_relocate_chunk) and relocation
1492 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1493 * after acquiring fs_info->reclaim_bgs_lock. So we
1494 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1495 * unused block groups.
1497 btrfs_delete_unused_bgs(fs_info);
1500 * Reclaim block groups in the reclaim_bgs list after we deleted
1501 * all unused block_groups. This possibly gives us some more free
1504 btrfs_reclaim_bgs(fs_info);
1506 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1507 if (kthread_should_park())
1509 if (kthread_should_stop())
1512 set_current_state(TASK_INTERRUPTIBLE);
1514 __set_current_state(TASK_RUNNING);
1519 static int transaction_kthread(void *arg)
1521 struct btrfs_root *root = arg;
1522 struct btrfs_fs_info *fs_info = root->fs_info;
1523 struct btrfs_trans_handle *trans;
1524 struct btrfs_transaction *cur;
1527 unsigned long delay;
1531 cannot_commit = false;
1532 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1533 mutex_lock(&fs_info->transaction_kthread_mutex);
1535 spin_lock(&fs_info->trans_lock);
1536 cur = fs_info->running_transaction;
1538 spin_unlock(&fs_info->trans_lock);
1542 delta = ktime_get_seconds() - cur->start_time;
1543 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1544 cur->state < TRANS_STATE_COMMIT_START &&
1545 delta < fs_info->commit_interval) {
1546 spin_unlock(&fs_info->trans_lock);
1547 delay -= msecs_to_jiffies((delta - 1) * 1000);
1549 msecs_to_jiffies(fs_info->commit_interval * 1000));
1552 transid = cur->transid;
1553 spin_unlock(&fs_info->trans_lock);
1555 /* If the file system is aborted, this will always fail. */
1556 trans = btrfs_attach_transaction(root);
1557 if (IS_ERR(trans)) {
1558 if (PTR_ERR(trans) != -ENOENT)
1559 cannot_commit = true;
1562 if (transid == trans->transid) {
1563 btrfs_commit_transaction(trans);
1565 btrfs_end_transaction(trans);
1568 wake_up_process(fs_info->cleaner_kthread);
1569 mutex_unlock(&fs_info->transaction_kthread_mutex);
1571 if (BTRFS_FS_ERROR(fs_info))
1572 btrfs_cleanup_transaction(fs_info);
1573 if (!kthread_should_stop() &&
1574 (!btrfs_transaction_blocked(fs_info) ||
1576 schedule_timeout_interruptible(delay);
1577 } while (!kthread_should_stop());
1582 * This will find the highest generation in the array of root backups. The
1583 * index of the highest array is returned, or -EINVAL if we can't find
1586 * We check to make sure the array is valid by comparing the
1587 * generation of the latest root in the array with the generation
1588 * in the super block. If they don't match we pitch it.
1590 static int find_newest_super_backup(struct btrfs_fs_info *info)
1592 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1594 struct btrfs_root_backup *root_backup;
1597 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1598 root_backup = info->super_copy->super_roots + i;
1599 cur = btrfs_backup_tree_root_gen(root_backup);
1600 if (cur == newest_gen)
1608 * copy all the root pointers into the super backup array.
1609 * this will bump the backup pointer by one when it is
1612 static void backup_super_roots(struct btrfs_fs_info *info)
1614 const int next_backup = info->backup_root_index;
1615 struct btrfs_root_backup *root_backup;
1617 root_backup = info->super_for_commit->super_roots + next_backup;
1620 * make sure all of our padding and empty slots get zero filled
1621 * regardless of which ones we use today
1623 memset(root_backup, 0, sizeof(*root_backup));
1625 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1627 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1628 btrfs_set_backup_tree_root_gen(root_backup,
1629 btrfs_header_generation(info->tree_root->node));
1631 btrfs_set_backup_tree_root_level(root_backup,
1632 btrfs_header_level(info->tree_root->node));
1634 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1635 btrfs_set_backup_chunk_root_gen(root_backup,
1636 btrfs_header_generation(info->chunk_root->node));
1637 btrfs_set_backup_chunk_root_level(root_backup,
1638 btrfs_header_level(info->chunk_root->node));
1640 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1641 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1642 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1644 btrfs_set_backup_extent_root(root_backup,
1645 extent_root->node->start);
1646 btrfs_set_backup_extent_root_gen(root_backup,
1647 btrfs_header_generation(extent_root->node));
1648 btrfs_set_backup_extent_root_level(root_backup,
1649 btrfs_header_level(extent_root->node));
1651 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1652 btrfs_set_backup_csum_root_gen(root_backup,
1653 btrfs_header_generation(csum_root->node));
1654 btrfs_set_backup_csum_root_level(root_backup,
1655 btrfs_header_level(csum_root->node));
1659 * we might commit during log recovery, which happens before we set
1660 * the fs_root. Make sure it is valid before we fill it in.
1662 if (info->fs_root && info->fs_root->node) {
1663 btrfs_set_backup_fs_root(root_backup,
1664 info->fs_root->node->start);
1665 btrfs_set_backup_fs_root_gen(root_backup,
1666 btrfs_header_generation(info->fs_root->node));
1667 btrfs_set_backup_fs_root_level(root_backup,
1668 btrfs_header_level(info->fs_root->node));
1671 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1672 btrfs_set_backup_dev_root_gen(root_backup,
1673 btrfs_header_generation(info->dev_root->node));
1674 btrfs_set_backup_dev_root_level(root_backup,
1675 btrfs_header_level(info->dev_root->node));
1677 btrfs_set_backup_total_bytes(root_backup,
1678 btrfs_super_total_bytes(info->super_copy));
1679 btrfs_set_backup_bytes_used(root_backup,
1680 btrfs_super_bytes_used(info->super_copy));
1681 btrfs_set_backup_num_devices(root_backup,
1682 btrfs_super_num_devices(info->super_copy));
1685 * if we don't copy this out to the super_copy, it won't get remembered
1686 * for the next commit
1688 memcpy(&info->super_copy->super_roots,
1689 &info->super_for_commit->super_roots,
1690 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1694 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1695 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1697 * fs_info - filesystem whose backup roots need to be read
1698 * priority - priority of backup root required
1700 * Returns backup root index on success and -EINVAL otherwise.
1702 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1704 int backup_index = find_newest_super_backup(fs_info);
1705 struct btrfs_super_block *super = fs_info->super_copy;
1706 struct btrfs_root_backup *root_backup;
1708 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1710 return backup_index;
1712 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1713 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1718 root_backup = super->super_roots + backup_index;
1720 btrfs_set_super_generation(super,
1721 btrfs_backup_tree_root_gen(root_backup));
1722 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1723 btrfs_set_super_root_level(super,
1724 btrfs_backup_tree_root_level(root_backup));
1725 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1728 * Fixme: the total bytes and num_devices need to match or we should
1731 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1732 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1734 return backup_index;
1737 /* helper to cleanup workers */
1738 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1740 btrfs_destroy_workqueue(fs_info->fixup_workers);
1741 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1742 btrfs_destroy_workqueue(fs_info->workers);
1743 if (fs_info->endio_workers)
1744 destroy_workqueue(fs_info->endio_workers);
1745 if (fs_info->rmw_workers)
1746 destroy_workqueue(fs_info->rmw_workers);
1747 if (fs_info->compressed_write_workers)
1748 destroy_workqueue(fs_info->compressed_write_workers);
1749 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1750 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1751 btrfs_destroy_workqueue(fs_info->delayed_workers);
1752 btrfs_destroy_workqueue(fs_info->caching_workers);
1753 btrfs_destroy_workqueue(fs_info->flush_workers);
1754 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1755 if (fs_info->discard_ctl.discard_workers)
1756 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1758 * Now that all other work queues are destroyed, we can safely destroy
1759 * the queues used for metadata I/O, since tasks from those other work
1760 * queues can do metadata I/O operations.
1762 if (fs_info->endio_meta_workers)
1763 destroy_workqueue(fs_info->endio_meta_workers);
1766 static void free_root_extent_buffers(struct btrfs_root *root)
1769 free_extent_buffer(root->node);
1770 free_extent_buffer(root->commit_root);
1772 root->commit_root = NULL;
1776 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1778 struct btrfs_root *root, *tmp;
1780 rbtree_postorder_for_each_entry_safe(root, tmp,
1781 &fs_info->global_root_tree,
1783 free_root_extent_buffers(root);
1786 /* helper to cleanup tree roots */
1787 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1789 free_root_extent_buffers(info->tree_root);
1791 free_global_root_pointers(info);
1792 free_root_extent_buffers(info->dev_root);
1793 free_root_extent_buffers(info->quota_root);
1794 free_root_extent_buffers(info->uuid_root);
1795 free_root_extent_buffers(info->fs_root);
1796 free_root_extent_buffers(info->data_reloc_root);
1797 free_root_extent_buffers(info->block_group_root);
1798 if (free_chunk_root)
1799 free_root_extent_buffers(info->chunk_root);
1802 void btrfs_put_root(struct btrfs_root *root)
1807 if (refcount_dec_and_test(&root->refs)) {
1808 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1809 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1811 free_anon_bdev(root->anon_dev);
1812 free_root_extent_buffers(root);
1813 #ifdef CONFIG_BTRFS_DEBUG
1814 spin_lock(&root->fs_info->fs_roots_radix_lock);
1815 list_del_init(&root->leak_list);
1816 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1822 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1825 struct btrfs_root *gang[8];
1828 while (!list_empty(&fs_info->dead_roots)) {
1829 gang[0] = list_entry(fs_info->dead_roots.next,
1830 struct btrfs_root, root_list);
1831 list_del(&gang[0]->root_list);
1833 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1834 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1835 btrfs_put_root(gang[0]);
1839 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1844 for (i = 0; i < ret; i++)
1845 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1849 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1851 mutex_init(&fs_info->scrub_lock);
1852 atomic_set(&fs_info->scrubs_running, 0);
1853 atomic_set(&fs_info->scrub_pause_req, 0);
1854 atomic_set(&fs_info->scrubs_paused, 0);
1855 atomic_set(&fs_info->scrub_cancel_req, 0);
1856 init_waitqueue_head(&fs_info->scrub_pause_wait);
1857 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1860 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1862 spin_lock_init(&fs_info->balance_lock);
1863 mutex_init(&fs_info->balance_mutex);
1864 atomic_set(&fs_info->balance_pause_req, 0);
1865 atomic_set(&fs_info->balance_cancel_req, 0);
1866 fs_info->balance_ctl = NULL;
1867 init_waitqueue_head(&fs_info->balance_wait_q);
1868 atomic_set(&fs_info->reloc_cancel_req, 0);
1871 static int btrfs_init_btree_inode(struct super_block *sb)
1873 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1874 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1875 fs_info->tree_root);
1876 struct inode *inode;
1878 inode = new_inode(sb);
1882 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1883 set_nlink(inode, 1);
1885 * we set the i_size on the btree inode to the max possible int.
1886 * the real end of the address space is determined by all of
1887 * the devices in the system
1889 inode->i_size = OFFSET_MAX;
1890 inode->i_mapping->a_ops = &btree_aops;
1891 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1893 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1894 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1895 IO_TREE_BTREE_INODE_IO);
1896 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1898 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1899 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1900 BTRFS_I(inode)->location.type = 0;
1901 BTRFS_I(inode)->location.offset = 0;
1902 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1903 __insert_inode_hash(inode, hash);
1904 fs_info->btree_inode = inode;
1909 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1911 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1912 init_rwsem(&fs_info->dev_replace.rwsem);
1913 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1916 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1918 spin_lock_init(&fs_info->qgroup_lock);
1919 mutex_init(&fs_info->qgroup_ioctl_lock);
1920 fs_info->qgroup_tree = RB_ROOT;
1921 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1922 fs_info->qgroup_seq = 1;
1923 fs_info->qgroup_ulist = NULL;
1924 fs_info->qgroup_rescan_running = false;
1925 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1926 mutex_init(&fs_info->qgroup_rescan_lock);
1929 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1931 u32 max_active = fs_info->thread_pool_size;
1932 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1933 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1936 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1938 fs_info->delalloc_workers =
1939 btrfs_alloc_workqueue(fs_info, "delalloc",
1940 flags, max_active, 2);
1942 fs_info->flush_workers =
1943 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1944 flags, max_active, 0);
1946 fs_info->caching_workers =
1947 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1949 fs_info->fixup_workers =
1950 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1952 fs_info->endio_workers =
1953 alloc_workqueue("btrfs-endio", flags, max_active);
1954 fs_info->endio_meta_workers =
1955 alloc_workqueue("btrfs-endio-meta", flags, max_active);
1956 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1957 fs_info->endio_write_workers =
1958 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1960 fs_info->compressed_write_workers =
1961 alloc_workqueue("btrfs-compressed-write", flags, max_active);
1962 fs_info->endio_freespace_worker =
1963 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
1965 fs_info->delayed_workers =
1966 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
1968 fs_info->qgroup_rescan_workers =
1969 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
1971 fs_info->discard_ctl.discard_workers =
1972 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
1974 if (!(fs_info->workers &&
1975 fs_info->delalloc_workers && fs_info->flush_workers &&
1976 fs_info->endio_workers && fs_info->endio_meta_workers &&
1977 fs_info->compressed_write_workers &&
1978 fs_info->endio_write_workers &&
1979 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
1980 fs_info->caching_workers && fs_info->fixup_workers &&
1981 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
1982 fs_info->discard_ctl.discard_workers)) {
1989 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
1991 struct crypto_shash *csum_shash;
1992 const char *csum_driver = btrfs_super_csum_driver(csum_type);
1994 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
1996 if (IS_ERR(csum_shash)) {
1997 btrfs_err(fs_info, "error allocating %s hash for checksum",
1999 return PTR_ERR(csum_shash);
2002 fs_info->csum_shash = csum_shash;
2005 * Check if the checksum implementation is a fast accelerated one.
2006 * As-is this is a bit of a hack and should be replaced once the csum
2007 * implementations provide that information themselves.
2009 switch (csum_type) {
2010 case BTRFS_CSUM_TYPE_CRC32:
2011 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2012 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2014 case BTRFS_CSUM_TYPE_XXHASH:
2015 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2021 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2022 btrfs_super_csum_name(csum_type),
2023 crypto_shash_driver_name(csum_shash));
2027 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2028 struct btrfs_fs_devices *fs_devices)
2031 struct btrfs_tree_parent_check check = { 0 };
2032 struct btrfs_root *log_tree_root;
2033 struct btrfs_super_block *disk_super = fs_info->super_copy;
2034 u64 bytenr = btrfs_super_log_root(disk_super);
2035 int level = btrfs_super_log_root_level(disk_super);
2037 if (fs_devices->rw_devices == 0) {
2038 btrfs_warn(fs_info, "log replay required on RO media");
2042 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2047 check.level = level;
2048 check.transid = fs_info->generation + 1;
2049 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2050 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2051 if (IS_ERR(log_tree_root->node)) {
2052 btrfs_warn(fs_info, "failed to read log tree");
2053 ret = PTR_ERR(log_tree_root->node);
2054 log_tree_root->node = NULL;
2055 btrfs_put_root(log_tree_root);
2058 if (!extent_buffer_uptodate(log_tree_root->node)) {
2059 btrfs_err(fs_info, "failed to read log tree");
2060 btrfs_put_root(log_tree_root);
2064 /* returns with log_tree_root freed on success */
2065 ret = btrfs_recover_log_trees(log_tree_root);
2067 btrfs_handle_fs_error(fs_info, ret,
2068 "Failed to recover log tree");
2069 btrfs_put_root(log_tree_root);
2073 if (sb_rdonly(fs_info->sb)) {
2074 ret = btrfs_commit_super(fs_info);
2082 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2083 struct btrfs_path *path, u64 objectid,
2086 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2087 struct btrfs_root *root;
2088 u64 max_global_id = 0;
2090 struct btrfs_key key = {
2091 .objectid = objectid,
2092 .type = BTRFS_ROOT_ITEM_KEY,
2097 /* If we have IGNOREDATACSUMS skip loading these roots. */
2098 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2099 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2100 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2105 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2109 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2110 ret = btrfs_next_leaf(tree_root, path);
2119 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2120 if (key.objectid != objectid)
2122 btrfs_release_path(path);
2125 * Just worry about this for extent tree, it'll be the same for
2128 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2129 max_global_id = max(max_global_id, key.offset);
2132 root = read_tree_root_path(tree_root, path, &key);
2134 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2135 ret = PTR_ERR(root);
2138 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2139 ret = btrfs_global_root_insert(root);
2141 btrfs_put_root(root);
2146 btrfs_release_path(path);
2148 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2149 fs_info->nr_global_roots = max_global_id + 1;
2151 if (!found || ret) {
2152 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2153 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2155 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2156 ret = ret ? ret : -ENOENT;
2159 btrfs_err(fs_info, "failed to load root %s", name);
2164 static int load_global_roots(struct btrfs_root *tree_root)
2166 struct btrfs_path *path;
2169 path = btrfs_alloc_path();
2173 ret = load_global_roots_objectid(tree_root, path,
2174 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2177 ret = load_global_roots_objectid(tree_root, path,
2178 BTRFS_CSUM_TREE_OBJECTID, "csum");
2181 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2183 ret = load_global_roots_objectid(tree_root, path,
2184 BTRFS_FREE_SPACE_TREE_OBJECTID,
2187 btrfs_free_path(path);
2191 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2193 struct btrfs_root *tree_root = fs_info->tree_root;
2194 struct btrfs_root *root;
2195 struct btrfs_key location;
2198 BUG_ON(!fs_info->tree_root);
2200 ret = load_global_roots(tree_root);
2204 location.type = BTRFS_ROOT_ITEM_KEY;
2205 location.offset = 0;
2207 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2208 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2209 root = btrfs_read_tree_root(tree_root, &location);
2211 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2212 ret = PTR_ERR(root);
2216 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2217 fs_info->block_group_root = root;
2221 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2222 root = btrfs_read_tree_root(tree_root, &location);
2224 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2225 ret = PTR_ERR(root);
2229 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2230 fs_info->dev_root = root;
2232 /* Initialize fs_info for all devices in any case */
2233 ret = btrfs_init_devices_late(fs_info);
2238 * This tree can share blocks with some other fs tree during relocation
2239 * and we need a proper setup by btrfs_get_fs_root
2241 root = btrfs_get_fs_root(tree_root->fs_info,
2242 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2244 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2245 ret = PTR_ERR(root);
2249 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2250 fs_info->data_reloc_root = root;
2253 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2254 root = btrfs_read_tree_root(tree_root, &location);
2255 if (!IS_ERR(root)) {
2256 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2257 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2258 fs_info->quota_root = root;
2261 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2262 root = btrfs_read_tree_root(tree_root, &location);
2264 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2265 ret = PTR_ERR(root);
2270 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2271 fs_info->uuid_root = root;
2276 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2277 location.objectid, ret);
2282 * Real super block validation
2283 * NOTE: super csum type and incompat features will not be checked here.
2285 * @sb: super block to check
2286 * @mirror_num: the super block number to check its bytenr:
2287 * 0 the primary (1st) sb
2288 * 1, 2 2nd and 3rd backup copy
2289 * -1 skip bytenr check
2291 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2292 struct btrfs_super_block *sb, int mirror_num)
2294 u64 nodesize = btrfs_super_nodesize(sb);
2295 u64 sectorsize = btrfs_super_sectorsize(sb);
2298 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2299 btrfs_err(fs_info, "no valid FS found");
2302 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2303 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2304 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2307 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2308 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2309 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2312 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2313 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2314 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2317 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2318 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2319 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2324 * Check sectorsize and nodesize first, other check will need it.
2325 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2327 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2328 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2329 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2334 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2336 * We can support 16K sectorsize with 64K page size without problem,
2337 * but such sectorsize/pagesize combination doesn't make much sense.
2338 * 4K will be our future standard, PAGE_SIZE is supported from the very
2341 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2343 "sectorsize %llu not yet supported for page size %lu",
2344 sectorsize, PAGE_SIZE);
2348 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2349 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2350 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2353 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2354 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2355 le32_to_cpu(sb->__unused_leafsize), nodesize);
2359 /* Root alignment check */
2360 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2361 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2362 btrfs_super_root(sb));
2365 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2366 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2367 btrfs_super_chunk_root(sb));
2370 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2371 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2372 btrfs_super_log_root(sb));
2376 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2379 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2380 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2384 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2385 memcmp(fs_info->fs_devices->metadata_uuid,
2386 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2388 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2389 fs_info->super_copy->metadata_uuid,
2390 fs_info->fs_devices->metadata_uuid);
2394 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2395 BTRFS_FSID_SIZE) != 0) {
2397 "dev_item UUID does not match metadata fsid: %pU != %pU",
2398 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2403 * Artificial requirement for block-group-tree to force newer features
2404 * (free-space-tree, no-holes) so the test matrix is smaller.
2406 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2407 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2408 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2410 "block-group-tree feature requires fres-space-tree and no-holes");
2415 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2418 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2419 btrfs_err(fs_info, "bytes_used is too small %llu",
2420 btrfs_super_bytes_used(sb));
2423 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2424 btrfs_err(fs_info, "invalid stripesize %u",
2425 btrfs_super_stripesize(sb));
2428 if (btrfs_super_num_devices(sb) > (1UL << 31))
2429 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2430 btrfs_super_num_devices(sb));
2431 if (btrfs_super_num_devices(sb) == 0) {
2432 btrfs_err(fs_info, "number of devices is 0");
2436 if (mirror_num >= 0 &&
2437 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2438 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2439 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2444 * Obvious sys_chunk_array corruptions, it must hold at least one key
2447 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2448 btrfs_err(fs_info, "system chunk array too big %u > %u",
2449 btrfs_super_sys_array_size(sb),
2450 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2453 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2454 + sizeof(struct btrfs_chunk)) {
2455 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2456 btrfs_super_sys_array_size(sb),
2457 sizeof(struct btrfs_disk_key)
2458 + sizeof(struct btrfs_chunk));
2463 * The generation is a global counter, we'll trust it more than the others
2464 * but it's still possible that it's the one that's wrong.
2466 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2468 "suspicious: generation < chunk_root_generation: %llu < %llu",
2469 btrfs_super_generation(sb),
2470 btrfs_super_chunk_root_generation(sb));
2471 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2472 && btrfs_super_cache_generation(sb) != (u64)-1)
2474 "suspicious: generation < cache_generation: %llu < %llu",
2475 btrfs_super_generation(sb),
2476 btrfs_super_cache_generation(sb));
2482 * Validation of super block at mount time.
2483 * Some checks already done early at mount time, like csum type and incompat
2484 * flags will be skipped.
2486 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2488 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2492 * Validation of super block at write time.
2493 * Some checks like bytenr check will be skipped as their values will be
2495 * Extra checks like csum type and incompat flags will be done here.
2497 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2498 struct btrfs_super_block *sb)
2502 ret = btrfs_validate_super(fs_info, sb, -1);
2505 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2507 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2508 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2511 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2514 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2515 btrfs_super_incompat_flags(sb),
2516 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2522 "super block corruption detected before writing it to disk");
2526 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2528 struct btrfs_tree_parent_check check = {
2531 .owner_root = root->root_key.objectid
2535 root->node = read_tree_block(root->fs_info, bytenr, &check);
2536 if (IS_ERR(root->node)) {
2537 ret = PTR_ERR(root->node);
2541 if (!extent_buffer_uptodate(root->node)) {
2542 free_extent_buffer(root->node);
2547 btrfs_set_root_node(&root->root_item, root->node);
2548 root->commit_root = btrfs_root_node(root);
2549 btrfs_set_root_refs(&root->root_item, 1);
2553 static int load_important_roots(struct btrfs_fs_info *fs_info)
2555 struct btrfs_super_block *sb = fs_info->super_copy;
2559 bytenr = btrfs_super_root(sb);
2560 gen = btrfs_super_generation(sb);
2561 level = btrfs_super_root_level(sb);
2562 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2564 btrfs_warn(fs_info, "couldn't read tree root");
2570 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2572 int backup_index = find_newest_super_backup(fs_info);
2573 struct btrfs_super_block *sb = fs_info->super_copy;
2574 struct btrfs_root *tree_root = fs_info->tree_root;
2575 bool handle_error = false;
2579 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2581 if (!IS_ERR(tree_root->node))
2582 free_extent_buffer(tree_root->node);
2583 tree_root->node = NULL;
2585 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2588 free_root_pointers(fs_info, 0);
2591 * Don't use the log in recovery mode, it won't be
2594 btrfs_set_super_log_root(sb, 0);
2596 /* We can't trust the free space cache either */
2597 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2599 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2600 ret = read_backup_root(fs_info, i);
2606 ret = load_important_roots(fs_info);
2608 handle_error = true;
2613 * No need to hold btrfs_root::objectid_mutex since the fs
2614 * hasn't been fully initialised and we are the only user
2616 ret = btrfs_init_root_free_objectid(tree_root);
2618 handle_error = true;
2622 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2624 ret = btrfs_read_roots(fs_info);
2626 handle_error = true;
2630 /* All successful */
2631 fs_info->generation = btrfs_header_generation(tree_root->node);
2632 fs_info->last_trans_committed = fs_info->generation;
2633 fs_info->last_reloc_trans = 0;
2635 /* Always begin writing backup roots after the one being used */
2636 if (backup_index < 0) {
2637 fs_info->backup_root_index = 0;
2639 fs_info->backup_root_index = backup_index + 1;
2640 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2648 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2650 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2651 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2652 INIT_LIST_HEAD(&fs_info->trans_list);
2653 INIT_LIST_HEAD(&fs_info->dead_roots);
2654 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2655 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2656 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2657 spin_lock_init(&fs_info->delalloc_root_lock);
2658 spin_lock_init(&fs_info->trans_lock);
2659 spin_lock_init(&fs_info->fs_roots_radix_lock);
2660 spin_lock_init(&fs_info->delayed_iput_lock);
2661 spin_lock_init(&fs_info->defrag_inodes_lock);
2662 spin_lock_init(&fs_info->super_lock);
2663 spin_lock_init(&fs_info->buffer_lock);
2664 spin_lock_init(&fs_info->unused_bgs_lock);
2665 spin_lock_init(&fs_info->treelog_bg_lock);
2666 spin_lock_init(&fs_info->zone_active_bgs_lock);
2667 spin_lock_init(&fs_info->relocation_bg_lock);
2668 rwlock_init(&fs_info->tree_mod_log_lock);
2669 rwlock_init(&fs_info->global_root_lock);
2670 mutex_init(&fs_info->unused_bg_unpin_mutex);
2671 mutex_init(&fs_info->reclaim_bgs_lock);
2672 mutex_init(&fs_info->reloc_mutex);
2673 mutex_init(&fs_info->delalloc_root_mutex);
2674 mutex_init(&fs_info->zoned_meta_io_lock);
2675 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2676 seqlock_init(&fs_info->profiles_lock);
2678 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2679 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2680 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2681 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2682 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
2683 BTRFS_LOCKDEP_TRANS_COMMIT_START);
2684 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2685 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2686 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2687 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2688 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2689 BTRFS_LOCKDEP_TRANS_COMPLETED);
2691 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2692 INIT_LIST_HEAD(&fs_info->space_info);
2693 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2694 INIT_LIST_HEAD(&fs_info->unused_bgs);
2695 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2696 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2697 #ifdef CONFIG_BTRFS_DEBUG
2698 INIT_LIST_HEAD(&fs_info->allocated_roots);
2699 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2700 spin_lock_init(&fs_info->eb_leak_lock);
2702 extent_map_tree_init(&fs_info->mapping_tree);
2703 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2704 BTRFS_BLOCK_RSV_GLOBAL);
2705 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2706 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2707 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2708 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2709 BTRFS_BLOCK_RSV_DELOPS);
2710 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2711 BTRFS_BLOCK_RSV_DELREFS);
2713 atomic_set(&fs_info->async_delalloc_pages, 0);
2714 atomic_set(&fs_info->defrag_running, 0);
2715 atomic_set(&fs_info->nr_delayed_iputs, 0);
2716 atomic64_set(&fs_info->tree_mod_seq, 0);
2717 fs_info->global_root_tree = RB_ROOT;
2718 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2719 fs_info->metadata_ratio = 0;
2720 fs_info->defrag_inodes = RB_ROOT;
2721 atomic64_set(&fs_info->free_chunk_space, 0);
2722 fs_info->tree_mod_log = RB_ROOT;
2723 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2724 btrfs_init_ref_verify(fs_info);
2726 fs_info->thread_pool_size = min_t(unsigned long,
2727 num_online_cpus() + 2, 8);
2729 INIT_LIST_HEAD(&fs_info->ordered_roots);
2730 spin_lock_init(&fs_info->ordered_root_lock);
2732 btrfs_init_scrub(fs_info);
2733 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2734 fs_info->check_integrity_print_mask = 0;
2736 btrfs_init_balance(fs_info);
2737 btrfs_init_async_reclaim_work(fs_info);
2739 rwlock_init(&fs_info->block_group_cache_lock);
2740 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2742 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2743 IO_TREE_FS_EXCLUDED_EXTENTS);
2745 mutex_init(&fs_info->ordered_operations_mutex);
2746 mutex_init(&fs_info->tree_log_mutex);
2747 mutex_init(&fs_info->chunk_mutex);
2748 mutex_init(&fs_info->transaction_kthread_mutex);
2749 mutex_init(&fs_info->cleaner_mutex);
2750 mutex_init(&fs_info->ro_block_group_mutex);
2751 init_rwsem(&fs_info->commit_root_sem);
2752 init_rwsem(&fs_info->cleanup_work_sem);
2753 init_rwsem(&fs_info->subvol_sem);
2754 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2756 btrfs_init_dev_replace_locks(fs_info);
2757 btrfs_init_qgroup(fs_info);
2758 btrfs_discard_init(fs_info);
2760 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2761 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2763 init_waitqueue_head(&fs_info->transaction_throttle);
2764 init_waitqueue_head(&fs_info->transaction_wait);
2765 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2766 init_waitqueue_head(&fs_info->async_submit_wait);
2767 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2769 /* Usable values until the real ones are cached from the superblock */
2770 fs_info->nodesize = 4096;
2771 fs_info->sectorsize = 4096;
2772 fs_info->sectorsize_bits = ilog2(4096);
2773 fs_info->stripesize = 4096;
2775 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2777 spin_lock_init(&fs_info->swapfile_pins_lock);
2778 fs_info->swapfile_pins = RB_ROOT;
2780 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2781 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2784 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2789 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2790 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2792 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2796 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2800 fs_info->dirty_metadata_batch = PAGE_SIZE *
2801 (1 + ilog2(nr_cpu_ids));
2803 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2807 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2812 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2814 if (!fs_info->delayed_root)
2816 btrfs_init_delayed_root(fs_info->delayed_root);
2819 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2821 return btrfs_alloc_stripe_hash_table(fs_info);
2824 static int btrfs_uuid_rescan_kthread(void *data)
2826 struct btrfs_fs_info *fs_info = data;
2830 * 1st step is to iterate through the existing UUID tree and
2831 * to delete all entries that contain outdated data.
2832 * 2nd step is to add all missing entries to the UUID tree.
2834 ret = btrfs_uuid_tree_iterate(fs_info);
2837 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2839 up(&fs_info->uuid_tree_rescan_sem);
2842 return btrfs_uuid_scan_kthread(data);
2845 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2847 struct task_struct *task;
2849 down(&fs_info->uuid_tree_rescan_sem);
2850 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2852 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2853 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2854 up(&fs_info->uuid_tree_rescan_sem);
2855 return PTR_ERR(task);
2862 * Some options only have meaning at mount time and shouldn't persist across
2863 * remounts, or be displayed. Clear these at the end of mount and remount
2866 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
2868 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
2869 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
2873 * Mounting logic specific to read-write file systems. Shared by open_ctree
2874 * and btrfs_remount when remounting from read-only to read-write.
2876 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2879 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2880 bool rebuild_free_space_tree = false;
2882 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2883 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2884 rebuild_free_space_tree = true;
2885 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2886 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2887 btrfs_warn(fs_info, "free space tree is invalid");
2888 rebuild_free_space_tree = true;
2891 if (rebuild_free_space_tree) {
2892 btrfs_info(fs_info, "rebuilding free space tree");
2893 ret = btrfs_rebuild_free_space_tree(fs_info);
2896 "failed to rebuild free space tree: %d", ret);
2901 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2902 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
2903 btrfs_info(fs_info, "disabling free space tree");
2904 ret = btrfs_delete_free_space_tree(fs_info);
2907 "failed to disable free space tree: %d", ret);
2913 * btrfs_find_orphan_roots() is responsible for finding all the dead
2914 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
2915 * them into the fs_info->fs_roots_radix tree. This must be done before
2916 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
2917 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
2918 * item before the root's tree is deleted - this means that if we unmount
2919 * or crash before the deletion completes, on the next mount we will not
2920 * delete what remains of the tree because the orphan item does not
2921 * exists anymore, which is what tells us we have a pending deletion.
2923 ret = btrfs_find_orphan_roots(fs_info);
2927 ret = btrfs_cleanup_fs_roots(fs_info);
2931 down_read(&fs_info->cleanup_work_sem);
2932 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2933 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2934 up_read(&fs_info->cleanup_work_sem);
2937 up_read(&fs_info->cleanup_work_sem);
2939 mutex_lock(&fs_info->cleaner_mutex);
2940 ret = btrfs_recover_relocation(fs_info);
2941 mutex_unlock(&fs_info->cleaner_mutex);
2943 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
2947 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
2948 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2949 btrfs_info(fs_info, "creating free space tree");
2950 ret = btrfs_create_free_space_tree(fs_info);
2953 "failed to create free space tree: %d", ret);
2958 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
2959 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
2964 ret = btrfs_resume_balance_async(fs_info);
2968 ret = btrfs_resume_dev_replace_async(fs_info);
2970 btrfs_warn(fs_info, "failed to resume dev_replace");
2974 btrfs_qgroup_rescan_resume(fs_info);
2976 if (!fs_info->uuid_root) {
2977 btrfs_info(fs_info, "creating UUID tree");
2978 ret = btrfs_create_uuid_tree(fs_info);
2981 "failed to create the UUID tree %d", ret);
2991 * Do various sanity and dependency checks of different features.
2993 * @is_rw_mount: If the mount is read-write.
2995 * This is the place for less strict checks (like for subpage or artificial
2996 * feature dependencies).
2998 * For strict checks or possible corruption detection, see
2999 * btrfs_validate_super().
3001 * This should be called after btrfs_parse_options(), as some mount options
3002 * (space cache related) can modify on-disk format like free space tree and
3003 * screw up certain feature dependencies.
3005 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3007 struct btrfs_super_block *disk_super = fs_info->super_copy;
3008 u64 incompat = btrfs_super_incompat_flags(disk_super);
3009 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3010 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3012 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3014 "cannot mount because of unknown incompat features (0x%llx)",
3019 /* Runtime limitation for mixed block groups. */
3020 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3021 (fs_info->sectorsize != fs_info->nodesize)) {
3023 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3024 fs_info->nodesize, fs_info->sectorsize);
3028 /* Mixed backref is an always-enabled feature. */
3029 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3031 /* Set compression related flags just in case. */
3032 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3033 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3034 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3035 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3038 * An ancient flag, which should really be marked deprecated.
3039 * Such runtime limitation doesn't really need a incompat flag.
3041 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3042 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3044 if (compat_ro_unsupp && is_rw_mount) {
3046 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3052 * We have unsupported RO compat features, although RO mounted, we
3053 * should not cause any metadata writes, including log replay.
3054 * Or we could screw up whatever the new feature requires.
3056 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3057 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3059 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3065 * Artificial limitations for block group tree, to force
3066 * block-group-tree to rely on no-holes and free-space-tree.
3068 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3069 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3070 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3072 "block-group-tree feature requires no-holes and free-space-tree features");
3077 * Subpage runtime limitation on v1 cache.
3079 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3080 * we're already defaulting to v2 cache, no need to bother v1 as it's
3081 * going to be deprecated anyway.
3083 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3085 "v1 space cache is not supported for page size %lu with sectorsize %u",
3086 PAGE_SIZE, fs_info->sectorsize);
3090 /* This can be called by remount, we need to protect the super block. */
3091 spin_lock(&fs_info->super_lock);
3092 btrfs_set_super_incompat_flags(disk_super, incompat);
3093 spin_unlock(&fs_info->super_lock);
3098 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3107 struct btrfs_super_block *disk_super;
3108 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3109 struct btrfs_root *tree_root;
3110 struct btrfs_root *chunk_root;
3114 ret = init_mount_fs_info(fs_info, sb);
3118 /* These need to be init'ed before we start creating inodes and such. */
3119 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3121 fs_info->tree_root = tree_root;
3122 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3124 fs_info->chunk_root = chunk_root;
3125 if (!tree_root || !chunk_root) {
3130 ret = btrfs_init_btree_inode(sb);
3134 invalidate_bdev(fs_devices->latest_dev->bdev);
3137 * Read super block and check the signature bytes only
3139 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3140 if (IS_ERR(disk_super)) {
3141 ret = PTR_ERR(disk_super);
3146 * Verify the type first, if that or the checksum value are
3147 * corrupted, we'll find out
3149 csum_type = btrfs_super_csum_type(disk_super);
3150 if (!btrfs_supported_super_csum(csum_type)) {
3151 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3154 btrfs_release_disk_super(disk_super);
3158 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3160 ret = btrfs_init_csum_hash(fs_info, csum_type);
3162 btrfs_release_disk_super(disk_super);
3167 * We want to check superblock checksum, the type is stored inside.
3168 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3170 if (btrfs_check_super_csum(fs_info, disk_super)) {
3171 btrfs_err(fs_info, "superblock checksum mismatch");
3173 btrfs_release_disk_super(disk_super);
3178 * super_copy is zeroed at allocation time and we never touch the
3179 * following bytes up to INFO_SIZE, the checksum is calculated from
3180 * the whole block of INFO_SIZE
3182 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3183 btrfs_release_disk_super(disk_super);
3185 disk_super = fs_info->super_copy;
3188 features = btrfs_super_flags(disk_super);
3189 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3190 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3191 btrfs_set_super_flags(disk_super, features);
3193 "found metadata UUID change in progress flag, clearing");
3196 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3197 sizeof(*fs_info->super_for_commit));
3199 ret = btrfs_validate_mount_super(fs_info);
3201 btrfs_err(fs_info, "superblock contains fatal errors");
3206 if (!btrfs_super_root(disk_super)) {
3207 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3212 /* check FS state, whether FS is broken. */
3213 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3214 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3217 * In the long term, we'll store the compression type in the super
3218 * block, and it'll be used for per file compression control.
3220 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3223 /* Set up fs_info before parsing mount options */
3224 nodesize = btrfs_super_nodesize(disk_super);
3225 sectorsize = btrfs_super_sectorsize(disk_super);
3226 stripesize = sectorsize;
3227 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3228 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3230 fs_info->nodesize = nodesize;
3231 fs_info->sectorsize = sectorsize;
3232 fs_info->sectorsize_bits = ilog2(sectorsize);
3233 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3234 fs_info->stripesize = stripesize;
3236 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3240 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3244 if (sectorsize < PAGE_SIZE) {
3245 struct btrfs_subpage_info *subpage_info;
3248 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3249 * going to be deprecated.
3251 * Force to use v2 cache for subpage case.
3253 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3254 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3255 "forcing free space tree for sector size %u with page size %lu",
3256 sectorsize, PAGE_SIZE);
3259 "read-write for sector size %u with page size %lu is experimental",
3260 sectorsize, PAGE_SIZE);
3261 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3262 if (!subpage_info) {
3266 btrfs_init_subpage_info(subpage_info, sectorsize);
3267 fs_info->subpage_info = subpage_info;
3270 ret = btrfs_init_workqueues(fs_info);
3272 goto fail_sb_buffer;
3274 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3275 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3277 sb->s_blocksize = sectorsize;
3278 sb->s_blocksize_bits = blksize_bits(sectorsize);
3279 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3281 mutex_lock(&fs_info->chunk_mutex);
3282 ret = btrfs_read_sys_array(fs_info);
3283 mutex_unlock(&fs_info->chunk_mutex);
3285 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3286 goto fail_sb_buffer;
3289 generation = btrfs_super_chunk_root_generation(disk_super);
3290 level = btrfs_super_chunk_root_level(disk_super);
3291 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3294 btrfs_err(fs_info, "failed to read chunk root");
3295 goto fail_tree_roots;
3298 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3299 offsetof(struct btrfs_header, chunk_tree_uuid),
3302 ret = btrfs_read_chunk_tree(fs_info);
3304 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3305 goto fail_tree_roots;
3309 * At this point we know all the devices that make this filesystem,
3310 * including the seed devices but we don't know yet if the replace
3311 * target is required. So free devices that are not part of this
3312 * filesystem but skip the replace target device which is checked
3313 * below in btrfs_init_dev_replace().
3315 btrfs_free_extra_devids(fs_devices);
3316 if (!fs_devices->latest_dev->bdev) {
3317 btrfs_err(fs_info, "failed to read devices");
3319 goto fail_tree_roots;
3322 ret = init_tree_roots(fs_info);
3324 goto fail_tree_roots;
3327 * Get zone type information of zoned block devices. This will also
3328 * handle emulation of a zoned filesystem if a regular device has the
3329 * zoned incompat feature flag set.
3331 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3334 "zoned: failed to read device zone info: %d", ret);
3335 goto fail_block_groups;
3339 * If we have a uuid root and we're not being told to rescan we need to
3340 * check the generation here so we can set the
3341 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3342 * transaction during a balance or the log replay without updating the
3343 * uuid generation, and then if we crash we would rescan the uuid tree,
3344 * even though it was perfectly fine.
3346 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3347 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3348 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3350 ret = btrfs_verify_dev_extents(fs_info);
3353 "failed to verify dev extents against chunks: %d",
3355 goto fail_block_groups;
3357 ret = btrfs_recover_balance(fs_info);
3359 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3360 goto fail_block_groups;
3363 ret = btrfs_init_dev_stats(fs_info);
3365 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3366 goto fail_block_groups;
3369 ret = btrfs_init_dev_replace(fs_info);
3371 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3372 goto fail_block_groups;
3375 ret = btrfs_check_zoned_mode(fs_info);
3377 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3379 goto fail_block_groups;
3382 ret = btrfs_sysfs_add_fsid(fs_devices);
3384 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3386 goto fail_block_groups;
3389 ret = btrfs_sysfs_add_mounted(fs_info);
3391 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3392 goto fail_fsdev_sysfs;
3395 ret = btrfs_init_space_info(fs_info);
3397 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3401 ret = btrfs_read_block_groups(fs_info);
3403 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3407 btrfs_free_zone_cache(fs_info);
3409 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3410 !btrfs_check_rw_degradable(fs_info, NULL)) {
3412 "writable mount is not allowed due to too many missing devices");
3417 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3419 if (IS_ERR(fs_info->cleaner_kthread)) {
3420 ret = PTR_ERR(fs_info->cleaner_kthread);
3424 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3426 "btrfs-transaction");
3427 if (IS_ERR(fs_info->transaction_kthread)) {
3428 ret = PTR_ERR(fs_info->transaction_kthread);
3432 if (!btrfs_test_opt(fs_info, NOSSD) &&
3433 !fs_info->fs_devices->rotating) {
3434 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3438 * For devices supporting discard turn on discard=async automatically,
3439 * unless it's already set or disabled. This could be turned off by
3440 * nodiscard for the same mount.
3442 if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3443 btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3444 btrfs_test_opt(fs_info, NODISCARD)) &&
3445 fs_info->fs_devices->discardable) {
3446 btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3447 "auto enabling async discard");
3450 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3451 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3452 ret = btrfsic_mount(fs_info, fs_devices,
3453 btrfs_test_opt(fs_info,
3454 CHECK_INTEGRITY_DATA) ? 1 : 0,
3455 fs_info->check_integrity_print_mask);
3458 "failed to initialize integrity check module: %d",
3462 ret = btrfs_read_qgroup_config(fs_info);
3464 goto fail_trans_kthread;
3466 if (btrfs_build_ref_tree(fs_info))
3467 btrfs_err(fs_info, "couldn't build ref tree");
3469 /* do not make disk changes in broken FS or nologreplay is given */
3470 if (btrfs_super_log_root(disk_super) != 0 &&
3471 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3472 btrfs_info(fs_info, "start tree-log replay");
3473 ret = btrfs_replay_log(fs_info, fs_devices);
3478 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3479 if (IS_ERR(fs_info->fs_root)) {
3480 ret = PTR_ERR(fs_info->fs_root);
3481 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3482 fs_info->fs_root = NULL;
3489 ret = btrfs_start_pre_rw_mount(fs_info);
3491 close_ctree(fs_info);
3494 btrfs_discard_resume(fs_info);
3496 if (fs_info->uuid_root &&
3497 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3498 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3499 btrfs_info(fs_info, "checking UUID tree");
3500 ret = btrfs_check_uuid_tree(fs_info);
3503 "failed to check the UUID tree: %d", ret);
3504 close_ctree(fs_info);
3509 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3511 /* Kick the cleaner thread so it'll start deleting snapshots. */
3512 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3513 wake_up_process(fs_info->cleaner_kthread);
3516 btrfs_clear_oneshot_options(fs_info);
3520 btrfs_free_qgroup_config(fs_info);
3522 kthread_stop(fs_info->transaction_kthread);
3523 btrfs_cleanup_transaction(fs_info);
3524 btrfs_free_fs_roots(fs_info);
3526 kthread_stop(fs_info->cleaner_kthread);
3529 * make sure we're done with the btree inode before we stop our
3532 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3535 btrfs_sysfs_remove_mounted(fs_info);
3538 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3541 btrfs_put_block_group_cache(fs_info);
3544 if (fs_info->data_reloc_root)
3545 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3546 free_root_pointers(fs_info, true);
3547 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3550 btrfs_stop_all_workers(fs_info);
3551 btrfs_free_block_groups(fs_info);
3553 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3555 iput(fs_info->btree_inode);
3557 btrfs_close_devices(fs_info->fs_devices);
3561 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3563 static void btrfs_end_super_write(struct bio *bio)
3565 struct btrfs_device *device = bio->bi_private;
3566 struct bio_vec *bvec;
3567 struct bvec_iter_all iter_all;
3570 bio_for_each_segment_all(bvec, bio, iter_all) {
3571 page = bvec->bv_page;
3573 if (bio->bi_status) {
3574 btrfs_warn_rl_in_rcu(device->fs_info,
3575 "lost page write due to IO error on %s (%d)",
3576 btrfs_dev_name(device),
3577 blk_status_to_errno(bio->bi_status));
3578 ClearPageUptodate(page);
3580 btrfs_dev_stat_inc_and_print(device,
3581 BTRFS_DEV_STAT_WRITE_ERRS);
3583 SetPageUptodate(page);
3593 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3594 int copy_num, bool drop_cache)
3596 struct btrfs_super_block *super;
3598 u64 bytenr, bytenr_orig;
3599 struct address_space *mapping = bdev->bd_inode->i_mapping;
3602 bytenr_orig = btrfs_sb_offset(copy_num);
3603 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3605 return ERR_PTR(-EINVAL);
3607 return ERR_PTR(ret);
3609 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3610 return ERR_PTR(-EINVAL);
3613 /* This should only be called with the primary sb. */
3614 ASSERT(copy_num == 0);
3617 * Drop the page of the primary superblock, so later read will
3618 * always read from the device.
3620 invalidate_inode_pages2_range(mapping,
3621 bytenr >> PAGE_SHIFT,
3622 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3625 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3627 return ERR_CAST(page);
3629 super = page_address(page);
3630 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3631 btrfs_release_disk_super(super);
3632 return ERR_PTR(-ENODATA);
3635 if (btrfs_super_bytenr(super) != bytenr_orig) {
3636 btrfs_release_disk_super(super);
3637 return ERR_PTR(-EINVAL);
3644 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3646 struct btrfs_super_block *super, *latest = NULL;
3650 /* we would like to check all the supers, but that would make
3651 * a btrfs mount succeed after a mkfs from a different FS.
3652 * So, we need to add a special mount option to scan for
3653 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3655 for (i = 0; i < 1; i++) {
3656 super = btrfs_read_dev_one_super(bdev, i, false);
3660 if (!latest || btrfs_super_generation(super) > transid) {
3662 btrfs_release_disk_super(super);
3665 transid = btrfs_super_generation(super);
3673 * Write superblock @sb to the @device. Do not wait for completion, all the
3674 * pages we use for writing are locked.
3676 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3677 * the expected device size at commit time. Note that max_mirrors must be
3678 * same for write and wait phases.
3680 * Return number of errors when page is not found or submission fails.
3682 static int write_dev_supers(struct btrfs_device *device,
3683 struct btrfs_super_block *sb, int max_mirrors)
3685 struct btrfs_fs_info *fs_info = device->fs_info;
3686 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3687 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3691 u64 bytenr, bytenr_orig;
3693 if (max_mirrors == 0)
3694 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3696 shash->tfm = fs_info->csum_shash;
3698 for (i = 0; i < max_mirrors; i++) {
3701 struct btrfs_super_block *disk_super;
3703 bytenr_orig = btrfs_sb_offset(i);
3704 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3705 if (ret == -ENOENT) {
3707 } else if (ret < 0) {
3708 btrfs_err(device->fs_info,
3709 "couldn't get super block location for mirror %d",
3714 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3715 device->commit_total_bytes)
3718 btrfs_set_super_bytenr(sb, bytenr_orig);
3720 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3721 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3724 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3727 btrfs_err(device->fs_info,
3728 "couldn't get super block page for bytenr %llu",
3734 /* Bump the refcount for wait_dev_supers() */
3737 disk_super = page_address(page);
3738 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3741 * Directly use bios here instead of relying on the page cache
3742 * to do I/O, so we don't lose the ability to do integrity
3745 bio = bio_alloc(device->bdev, 1,
3746 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3748 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3749 bio->bi_private = device;
3750 bio->bi_end_io = btrfs_end_super_write;
3751 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3752 offset_in_page(bytenr));
3755 * We FUA only the first super block. The others we allow to
3756 * go down lazy and there's a short window where the on-disk
3757 * copies might still contain the older version.
3759 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3760 bio->bi_opf |= REQ_FUA;
3762 btrfsic_check_bio(bio);
3765 if (btrfs_advance_sb_log(device, i))
3768 return errors < i ? 0 : -1;
3772 * Wait for write completion of superblocks done by write_dev_supers,
3773 * @max_mirrors same for write and wait phases.
3775 * Return number of errors when page is not found or not marked up to
3778 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3782 bool primary_failed = false;
3786 if (max_mirrors == 0)
3787 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3789 for (i = 0; i < max_mirrors; i++) {
3792 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3793 if (ret == -ENOENT) {
3795 } else if (ret < 0) {
3798 primary_failed = true;
3801 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3802 device->commit_total_bytes)
3805 page = find_get_page(device->bdev->bd_inode->i_mapping,
3806 bytenr >> PAGE_SHIFT);
3810 primary_failed = true;
3813 /* Page is submitted locked and unlocked once the IO completes */
3814 wait_on_page_locked(page);
3815 if (PageError(page)) {
3818 primary_failed = true;
3821 /* Drop our reference */
3824 /* Drop the reference from the writing run */
3828 /* log error, force error return */
3829 if (primary_failed) {
3830 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3835 return errors < i ? 0 : -1;
3839 * endio for the write_dev_flush, this will wake anyone waiting
3840 * for the barrier when it is done
3842 static void btrfs_end_empty_barrier(struct bio *bio)
3845 complete(bio->bi_private);
3849 * Submit a flush request to the device if it supports it. Error handling is
3850 * done in the waiting counterpart.
3852 static void write_dev_flush(struct btrfs_device *device)
3854 struct bio *bio = &device->flush_bio;
3856 device->last_flush_error = BLK_STS_OK;
3858 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3860 * When a disk has write caching disabled, we skip submission of a bio
3861 * with flush and sync requests before writing the superblock, since
3862 * it's not needed. However when the integrity checker is enabled, this
3863 * results in reports that there are metadata blocks referred by a
3864 * superblock that were not properly flushed. So don't skip the bio
3865 * submission only when the integrity checker is enabled for the sake
3866 * of simplicity, since this is a debug tool and not meant for use in
3869 if (!bdev_write_cache(device->bdev))
3873 bio_init(bio, device->bdev, NULL, 0,
3874 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3875 bio->bi_end_io = btrfs_end_empty_barrier;
3876 init_completion(&device->flush_wait);
3877 bio->bi_private = &device->flush_wait;
3879 btrfsic_check_bio(bio);
3881 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3885 * If the flush bio has been submitted by write_dev_flush, wait for it.
3886 * Return true for any error, and false otherwise.
3888 static bool wait_dev_flush(struct btrfs_device *device)
3890 struct bio *bio = &device->flush_bio;
3892 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3895 wait_for_completion_io(&device->flush_wait);
3897 if (bio->bi_status) {
3898 device->last_flush_error = bio->bi_status;
3899 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3907 * send an empty flush down to each device in parallel,
3908 * then wait for them
3910 static int barrier_all_devices(struct btrfs_fs_info *info)
3912 struct list_head *head;
3913 struct btrfs_device *dev;
3914 int errors_wait = 0;
3916 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3917 /* send down all the barriers */
3918 head = &info->fs_devices->devices;
3919 list_for_each_entry(dev, head, dev_list) {
3920 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3924 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3925 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3928 write_dev_flush(dev);
3931 /* wait for all the barriers */
3932 list_for_each_entry(dev, head, dev_list) {
3933 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3939 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3940 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3943 if (wait_dev_flush(dev))
3948 * Checks last_flush_error of disks in order to determine the device
3951 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3957 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3960 int min_tolerated = INT_MAX;
3962 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3963 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3964 min_tolerated = min_t(int, min_tolerated,
3965 btrfs_raid_array[BTRFS_RAID_SINGLE].
3966 tolerated_failures);
3968 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3969 if (raid_type == BTRFS_RAID_SINGLE)
3971 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3973 min_tolerated = min_t(int, min_tolerated,
3974 btrfs_raid_array[raid_type].
3975 tolerated_failures);
3978 if (min_tolerated == INT_MAX) {
3979 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3983 return min_tolerated;
3986 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3988 struct list_head *head;
3989 struct btrfs_device *dev;
3990 struct btrfs_super_block *sb;
3991 struct btrfs_dev_item *dev_item;
3995 int total_errors = 0;
3998 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4001 * max_mirrors == 0 indicates we're from commit_transaction,
4002 * not from fsync where the tree roots in fs_info have not
4003 * been consistent on disk.
4005 if (max_mirrors == 0)
4006 backup_super_roots(fs_info);
4008 sb = fs_info->super_for_commit;
4009 dev_item = &sb->dev_item;
4011 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4012 head = &fs_info->fs_devices->devices;
4013 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4016 ret = barrier_all_devices(fs_info);
4019 &fs_info->fs_devices->device_list_mutex);
4020 btrfs_handle_fs_error(fs_info, ret,
4021 "errors while submitting device barriers.");
4026 list_for_each_entry(dev, head, dev_list) {
4031 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4032 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4035 btrfs_set_stack_device_generation(dev_item, 0);
4036 btrfs_set_stack_device_type(dev_item, dev->type);
4037 btrfs_set_stack_device_id(dev_item, dev->devid);
4038 btrfs_set_stack_device_total_bytes(dev_item,
4039 dev->commit_total_bytes);
4040 btrfs_set_stack_device_bytes_used(dev_item,
4041 dev->commit_bytes_used);
4042 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4043 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4044 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4045 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4046 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4049 flags = btrfs_super_flags(sb);
4050 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4052 ret = btrfs_validate_write_super(fs_info, sb);
4054 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4055 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4056 "unexpected superblock corruption detected");
4060 ret = write_dev_supers(dev, sb, max_mirrors);
4064 if (total_errors > max_errors) {
4065 btrfs_err(fs_info, "%d errors while writing supers",
4067 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4069 /* FUA is masked off if unsupported and can't be the reason */
4070 btrfs_handle_fs_error(fs_info, -EIO,
4071 "%d errors while writing supers",
4077 list_for_each_entry(dev, head, dev_list) {
4080 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4081 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4084 ret = wait_dev_supers(dev, max_mirrors);
4088 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4089 if (total_errors > max_errors) {
4090 btrfs_handle_fs_error(fs_info, -EIO,
4091 "%d errors while writing supers",
4098 /* Drop a fs root from the radix tree and free it. */
4099 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4100 struct btrfs_root *root)
4102 bool drop_ref = false;
4104 spin_lock(&fs_info->fs_roots_radix_lock);
4105 radix_tree_delete(&fs_info->fs_roots_radix,
4106 (unsigned long)root->root_key.objectid);
4107 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4109 spin_unlock(&fs_info->fs_roots_radix_lock);
4111 if (BTRFS_FS_ERROR(fs_info)) {
4112 ASSERT(root->log_root == NULL);
4113 if (root->reloc_root) {
4114 btrfs_put_root(root->reloc_root);
4115 root->reloc_root = NULL;
4120 btrfs_put_root(root);
4123 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4125 u64 root_objectid = 0;
4126 struct btrfs_root *gang[8];
4129 unsigned int ret = 0;
4132 spin_lock(&fs_info->fs_roots_radix_lock);
4133 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4134 (void **)gang, root_objectid,
4137 spin_unlock(&fs_info->fs_roots_radix_lock);
4140 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4142 for (i = 0; i < ret; i++) {
4143 /* Avoid to grab roots in dead_roots */
4144 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4148 /* grab all the search result for later use */
4149 gang[i] = btrfs_grab_root(gang[i]);
4151 spin_unlock(&fs_info->fs_roots_radix_lock);
4153 for (i = 0; i < ret; i++) {
4156 root_objectid = gang[i]->root_key.objectid;
4157 err = btrfs_orphan_cleanup(gang[i]);
4160 btrfs_put_root(gang[i]);
4165 /* release the uncleaned roots due to error */
4166 for (; i < ret; i++) {
4168 btrfs_put_root(gang[i]);
4173 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4175 struct btrfs_root *root = fs_info->tree_root;
4176 struct btrfs_trans_handle *trans;
4178 mutex_lock(&fs_info->cleaner_mutex);
4179 btrfs_run_delayed_iputs(fs_info);
4180 mutex_unlock(&fs_info->cleaner_mutex);
4181 wake_up_process(fs_info->cleaner_kthread);
4183 /* wait until ongoing cleanup work done */
4184 down_write(&fs_info->cleanup_work_sem);
4185 up_write(&fs_info->cleanup_work_sem);
4187 trans = btrfs_join_transaction(root);
4189 return PTR_ERR(trans);
4190 return btrfs_commit_transaction(trans);
4193 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4195 struct btrfs_transaction *trans;
4196 struct btrfs_transaction *tmp;
4199 if (list_empty(&fs_info->trans_list))
4203 * This function is only called at the very end of close_ctree(),
4204 * thus no other running transaction, no need to take trans_lock.
4206 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4207 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4208 struct extent_state *cached = NULL;
4209 u64 dirty_bytes = 0;
4215 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4216 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4217 dirty_bytes += found_end + 1 - found_start;
4218 cur = found_end + 1;
4221 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4222 trans->transid, dirty_bytes);
4223 btrfs_cleanup_one_transaction(trans, fs_info);
4225 if (trans == fs_info->running_transaction)
4226 fs_info->running_transaction = NULL;
4227 list_del_init(&trans->list);
4229 btrfs_put_transaction(trans);
4230 trace_btrfs_transaction_commit(fs_info);
4235 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4239 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4242 * If we had UNFINISHED_DROPS we could still be processing them, so
4243 * clear that bit and wake up relocation so it can stop.
4244 * We must do this before stopping the block group reclaim task, because
4245 * at btrfs_relocate_block_group() we wait for this bit, and after the
4246 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4247 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4250 btrfs_wake_unfinished_drop(fs_info);
4253 * We may have the reclaim task running and relocating a data block group,
4254 * in which case it may create delayed iputs. So stop it before we park
4255 * the cleaner kthread otherwise we can get new delayed iputs after
4256 * parking the cleaner, and that can make the async reclaim task to hang
4257 * if it's waiting for delayed iputs to complete, since the cleaner is
4258 * parked and can not run delayed iputs - this will make us hang when
4259 * trying to stop the async reclaim task.
4261 cancel_work_sync(&fs_info->reclaim_bgs_work);
4263 * We don't want the cleaner to start new transactions, add more delayed
4264 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4265 * because that frees the task_struct, and the transaction kthread might
4266 * still try to wake up the cleaner.
4268 kthread_park(fs_info->cleaner_kthread);
4270 /* wait for the qgroup rescan worker to stop */
4271 btrfs_qgroup_wait_for_completion(fs_info, false);
4273 /* wait for the uuid_scan task to finish */
4274 down(&fs_info->uuid_tree_rescan_sem);
4275 /* avoid complains from lockdep et al., set sem back to initial state */
4276 up(&fs_info->uuid_tree_rescan_sem);
4278 /* pause restriper - we want to resume on mount */
4279 btrfs_pause_balance(fs_info);
4281 btrfs_dev_replace_suspend_for_unmount(fs_info);
4283 btrfs_scrub_cancel(fs_info);
4285 /* wait for any defraggers to finish */
4286 wait_event(fs_info->transaction_wait,
4287 (atomic_read(&fs_info->defrag_running) == 0));
4289 /* clear out the rbtree of defraggable inodes */
4290 btrfs_cleanup_defrag_inodes(fs_info);
4293 * After we parked the cleaner kthread, ordered extents may have
4294 * completed and created new delayed iputs. If one of the async reclaim
4295 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4296 * can hang forever trying to stop it, because if a delayed iput is
4297 * added after it ran btrfs_run_delayed_iputs() and before it called
4298 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4299 * no one else to run iputs.
4301 * So wait for all ongoing ordered extents to complete and then run
4302 * delayed iputs. This works because once we reach this point no one
4303 * can either create new ordered extents nor create delayed iputs
4304 * through some other means.
4306 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4307 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4308 * but the delayed iput for the respective inode is made only when doing
4309 * the final btrfs_put_ordered_extent() (which must happen at
4310 * btrfs_finish_ordered_io() when we are unmounting).
4312 btrfs_flush_workqueue(fs_info->endio_write_workers);
4313 /* Ordered extents for free space inodes. */
4314 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4315 btrfs_run_delayed_iputs(fs_info);
4317 cancel_work_sync(&fs_info->async_reclaim_work);
4318 cancel_work_sync(&fs_info->async_data_reclaim_work);
4319 cancel_work_sync(&fs_info->preempt_reclaim_work);
4321 /* Cancel or finish ongoing discard work */
4322 btrfs_discard_cleanup(fs_info);
4324 if (!sb_rdonly(fs_info->sb)) {
4326 * The cleaner kthread is stopped, so do one final pass over
4327 * unused block groups.
4329 btrfs_delete_unused_bgs(fs_info);
4332 * There might be existing delayed inode workers still running
4333 * and holding an empty delayed inode item. We must wait for
4334 * them to complete first because they can create a transaction.
4335 * This happens when someone calls btrfs_balance_delayed_items()
4336 * and then a transaction commit runs the same delayed nodes
4337 * before any delayed worker has done something with the nodes.
4338 * We must wait for any worker here and not at transaction
4339 * commit time since that could cause a deadlock.
4340 * This is a very rare case.
4342 btrfs_flush_workqueue(fs_info->delayed_workers);
4344 ret = btrfs_commit_super(fs_info);
4346 btrfs_err(fs_info, "commit super ret %d", ret);
4349 if (BTRFS_FS_ERROR(fs_info))
4350 btrfs_error_commit_super(fs_info);
4352 kthread_stop(fs_info->transaction_kthread);
4353 kthread_stop(fs_info->cleaner_kthread);
4355 ASSERT(list_empty(&fs_info->delayed_iputs));
4356 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4358 if (btrfs_check_quota_leak(fs_info)) {
4359 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4360 btrfs_err(fs_info, "qgroup reserved space leaked");
4363 btrfs_free_qgroup_config(fs_info);
4364 ASSERT(list_empty(&fs_info->delalloc_roots));
4366 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4367 btrfs_info(fs_info, "at unmount delalloc count %lld",
4368 percpu_counter_sum(&fs_info->delalloc_bytes));
4371 if (percpu_counter_sum(&fs_info->ordered_bytes))
4372 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4373 percpu_counter_sum(&fs_info->ordered_bytes));
4375 btrfs_sysfs_remove_mounted(fs_info);
4376 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4378 btrfs_put_block_group_cache(fs_info);
4381 * we must make sure there is not any read request to
4382 * submit after we stopping all workers.
4384 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4385 btrfs_stop_all_workers(fs_info);
4387 /* We shouldn't have any transaction open at this point */
4388 warn_about_uncommitted_trans(fs_info);
4390 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4391 free_root_pointers(fs_info, true);
4392 btrfs_free_fs_roots(fs_info);
4395 * We must free the block groups after dropping the fs_roots as we could
4396 * have had an IO error and have left over tree log blocks that aren't
4397 * cleaned up until the fs roots are freed. This makes the block group
4398 * accounting appear to be wrong because there's pending reserved bytes,
4399 * so make sure we do the block group cleanup afterwards.
4401 btrfs_free_block_groups(fs_info);
4403 iput(fs_info->btree_inode);
4405 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4406 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4407 btrfsic_unmount(fs_info->fs_devices);
4410 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4411 btrfs_close_devices(fs_info->fs_devices);
4414 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4416 struct btrfs_fs_info *fs_info = buf->fs_info;
4417 u64 transid = btrfs_header_generation(buf);
4419 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4421 * This is a fast path so only do this check if we have sanity tests
4422 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4423 * outside of the sanity tests.
4425 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4428 btrfs_assert_tree_write_locked(buf);
4429 if (transid != fs_info->generation)
4430 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4431 buf->start, transid, fs_info->generation);
4432 set_extent_buffer_dirty(buf);
4433 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4435 * btrfs_check_leaf() won't check item data if we don't have WRITTEN
4436 * set, so this will only validate the basic structure of the items.
4438 if (btrfs_header_level(buf) == 0 && btrfs_check_leaf(buf)) {
4439 btrfs_print_leaf(buf);
4445 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4449 * looks as though older kernels can get into trouble with
4450 * this code, they end up stuck in balance_dirty_pages forever
4454 if (current->flags & PF_MEMALLOC)
4458 btrfs_balance_delayed_items(fs_info);
4460 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4461 BTRFS_DIRTY_METADATA_THRESH,
4462 fs_info->dirty_metadata_batch);
4464 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4468 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4470 __btrfs_btree_balance_dirty(fs_info, 1);
4473 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4475 __btrfs_btree_balance_dirty(fs_info, 0);
4478 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4480 /* cleanup FS via transaction */
4481 btrfs_cleanup_transaction(fs_info);
4483 mutex_lock(&fs_info->cleaner_mutex);
4484 btrfs_run_delayed_iputs(fs_info);
4485 mutex_unlock(&fs_info->cleaner_mutex);
4487 down_write(&fs_info->cleanup_work_sem);
4488 up_write(&fs_info->cleanup_work_sem);
4491 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4493 struct btrfs_root *gang[8];
4494 u64 root_objectid = 0;
4497 spin_lock(&fs_info->fs_roots_radix_lock);
4498 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4499 (void **)gang, root_objectid,
4500 ARRAY_SIZE(gang))) != 0) {
4503 for (i = 0; i < ret; i++)
4504 gang[i] = btrfs_grab_root(gang[i]);
4505 spin_unlock(&fs_info->fs_roots_radix_lock);
4507 for (i = 0; i < ret; i++) {
4510 root_objectid = gang[i]->root_key.objectid;
4511 btrfs_free_log(NULL, gang[i]);
4512 btrfs_put_root(gang[i]);
4515 spin_lock(&fs_info->fs_roots_radix_lock);
4517 spin_unlock(&fs_info->fs_roots_radix_lock);
4518 btrfs_free_log_root_tree(NULL, fs_info);
4521 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4523 struct btrfs_ordered_extent *ordered;
4525 spin_lock(&root->ordered_extent_lock);
4527 * This will just short circuit the ordered completion stuff which will
4528 * make sure the ordered extent gets properly cleaned up.
4530 list_for_each_entry(ordered, &root->ordered_extents,
4532 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4533 spin_unlock(&root->ordered_extent_lock);
4536 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4538 struct btrfs_root *root;
4539 struct list_head splice;
4541 INIT_LIST_HEAD(&splice);
4543 spin_lock(&fs_info->ordered_root_lock);
4544 list_splice_init(&fs_info->ordered_roots, &splice);
4545 while (!list_empty(&splice)) {
4546 root = list_first_entry(&splice, struct btrfs_root,
4548 list_move_tail(&root->ordered_root,
4549 &fs_info->ordered_roots);
4551 spin_unlock(&fs_info->ordered_root_lock);
4552 btrfs_destroy_ordered_extents(root);
4555 spin_lock(&fs_info->ordered_root_lock);
4557 spin_unlock(&fs_info->ordered_root_lock);
4560 * We need this here because if we've been flipped read-only we won't
4561 * get sync() from the umount, so we need to make sure any ordered
4562 * extents that haven't had their dirty pages IO start writeout yet
4563 * actually get run and error out properly.
4565 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4568 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4569 struct btrfs_fs_info *fs_info)
4571 struct rb_node *node;
4572 struct btrfs_delayed_ref_root *delayed_refs;
4573 struct btrfs_delayed_ref_node *ref;
4575 delayed_refs = &trans->delayed_refs;
4577 spin_lock(&delayed_refs->lock);
4578 if (atomic_read(&delayed_refs->num_entries) == 0) {
4579 spin_unlock(&delayed_refs->lock);
4580 btrfs_debug(fs_info, "delayed_refs has NO entry");
4584 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4585 struct btrfs_delayed_ref_head *head;
4587 bool pin_bytes = false;
4589 head = rb_entry(node, struct btrfs_delayed_ref_head,
4591 if (btrfs_delayed_ref_lock(delayed_refs, head))
4594 spin_lock(&head->lock);
4595 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4596 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4598 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4599 RB_CLEAR_NODE(&ref->ref_node);
4600 if (!list_empty(&ref->add_list))
4601 list_del(&ref->add_list);
4602 atomic_dec(&delayed_refs->num_entries);
4603 btrfs_put_delayed_ref(ref);
4605 if (head->must_insert_reserved)
4607 btrfs_free_delayed_extent_op(head->extent_op);
4608 btrfs_delete_ref_head(delayed_refs, head);
4609 spin_unlock(&head->lock);
4610 spin_unlock(&delayed_refs->lock);
4611 mutex_unlock(&head->mutex);
4614 struct btrfs_block_group *cache;
4616 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4619 spin_lock(&cache->space_info->lock);
4620 spin_lock(&cache->lock);
4621 cache->pinned += head->num_bytes;
4622 btrfs_space_info_update_bytes_pinned(fs_info,
4623 cache->space_info, head->num_bytes);
4624 cache->reserved -= head->num_bytes;
4625 cache->space_info->bytes_reserved -= head->num_bytes;
4626 spin_unlock(&cache->lock);
4627 spin_unlock(&cache->space_info->lock);
4629 btrfs_put_block_group(cache);
4631 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4632 head->bytenr + head->num_bytes - 1);
4634 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4635 btrfs_put_delayed_ref_head(head);
4637 spin_lock(&delayed_refs->lock);
4639 btrfs_qgroup_destroy_extent_records(trans);
4641 spin_unlock(&delayed_refs->lock);
4644 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4646 struct btrfs_inode *btrfs_inode;
4647 struct list_head splice;
4649 INIT_LIST_HEAD(&splice);
4651 spin_lock(&root->delalloc_lock);
4652 list_splice_init(&root->delalloc_inodes, &splice);
4654 while (!list_empty(&splice)) {
4655 struct inode *inode = NULL;
4656 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4658 __btrfs_del_delalloc_inode(root, btrfs_inode);
4659 spin_unlock(&root->delalloc_lock);
4662 * Make sure we get a live inode and that it'll not disappear
4665 inode = igrab(&btrfs_inode->vfs_inode);
4667 unsigned int nofs_flag;
4669 nofs_flag = memalloc_nofs_save();
4670 invalidate_inode_pages2(inode->i_mapping);
4671 memalloc_nofs_restore(nofs_flag);
4674 spin_lock(&root->delalloc_lock);
4676 spin_unlock(&root->delalloc_lock);
4679 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4681 struct btrfs_root *root;
4682 struct list_head splice;
4684 INIT_LIST_HEAD(&splice);
4686 spin_lock(&fs_info->delalloc_root_lock);
4687 list_splice_init(&fs_info->delalloc_roots, &splice);
4688 while (!list_empty(&splice)) {
4689 root = list_first_entry(&splice, struct btrfs_root,
4691 root = btrfs_grab_root(root);
4693 spin_unlock(&fs_info->delalloc_root_lock);
4695 btrfs_destroy_delalloc_inodes(root);
4696 btrfs_put_root(root);
4698 spin_lock(&fs_info->delalloc_root_lock);
4700 spin_unlock(&fs_info->delalloc_root_lock);
4703 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4704 struct extent_io_tree *dirty_pages,
4708 struct extent_buffer *eb;
4713 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4718 clear_extent_bits(dirty_pages, start, end, mark);
4719 while (start <= end) {
4720 eb = find_extent_buffer(fs_info, start);
4721 start += fs_info->nodesize;
4725 btrfs_tree_lock(eb);
4726 wait_on_extent_buffer_writeback(eb);
4727 btrfs_clear_buffer_dirty(NULL, eb);
4728 btrfs_tree_unlock(eb);
4730 free_extent_buffer_stale(eb);
4737 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4738 struct extent_io_tree *unpin)
4745 struct extent_state *cached_state = NULL;
4748 * The btrfs_finish_extent_commit() may get the same range as
4749 * ours between find_first_extent_bit and clear_extent_dirty.
4750 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4751 * the same extent range.
4753 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4754 ret = find_first_extent_bit(unpin, 0, &start, &end,
4755 EXTENT_DIRTY, &cached_state);
4757 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4761 clear_extent_dirty(unpin, start, end, &cached_state);
4762 free_extent_state(cached_state);
4763 btrfs_error_unpin_extent_range(fs_info, start, end);
4764 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4771 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4773 struct inode *inode;
4775 inode = cache->io_ctl.inode;
4777 unsigned int nofs_flag;
4779 nofs_flag = memalloc_nofs_save();
4780 invalidate_inode_pages2(inode->i_mapping);
4781 memalloc_nofs_restore(nofs_flag);
4783 BTRFS_I(inode)->generation = 0;
4784 cache->io_ctl.inode = NULL;
4787 ASSERT(cache->io_ctl.pages == NULL);
4788 btrfs_put_block_group(cache);
4791 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4792 struct btrfs_fs_info *fs_info)
4794 struct btrfs_block_group *cache;
4796 spin_lock(&cur_trans->dirty_bgs_lock);
4797 while (!list_empty(&cur_trans->dirty_bgs)) {
4798 cache = list_first_entry(&cur_trans->dirty_bgs,
4799 struct btrfs_block_group,
4802 if (!list_empty(&cache->io_list)) {
4803 spin_unlock(&cur_trans->dirty_bgs_lock);
4804 list_del_init(&cache->io_list);
4805 btrfs_cleanup_bg_io(cache);
4806 spin_lock(&cur_trans->dirty_bgs_lock);
4809 list_del_init(&cache->dirty_list);
4810 spin_lock(&cache->lock);
4811 cache->disk_cache_state = BTRFS_DC_ERROR;
4812 spin_unlock(&cache->lock);
4814 spin_unlock(&cur_trans->dirty_bgs_lock);
4815 btrfs_put_block_group(cache);
4816 btrfs_delayed_refs_rsv_release(fs_info, 1);
4817 spin_lock(&cur_trans->dirty_bgs_lock);
4819 spin_unlock(&cur_trans->dirty_bgs_lock);
4822 * Refer to the definition of io_bgs member for details why it's safe
4823 * to use it without any locking
4825 while (!list_empty(&cur_trans->io_bgs)) {
4826 cache = list_first_entry(&cur_trans->io_bgs,
4827 struct btrfs_block_group,
4830 list_del_init(&cache->io_list);
4831 spin_lock(&cache->lock);
4832 cache->disk_cache_state = BTRFS_DC_ERROR;
4833 spin_unlock(&cache->lock);
4834 btrfs_cleanup_bg_io(cache);
4838 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4839 struct btrfs_fs_info *fs_info)
4841 struct btrfs_device *dev, *tmp;
4843 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4844 ASSERT(list_empty(&cur_trans->dirty_bgs));
4845 ASSERT(list_empty(&cur_trans->io_bgs));
4847 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4849 list_del_init(&dev->post_commit_list);
4852 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4854 cur_trans->state = TRANS_STATE_COMMIT_START;
4855 wake_up(&fs_info->transaction_blocked_wait);
4857 cur_trans->state = TRANS_STATE_UNBLOCKED;
4858 wake_up(&fs_info->transaction_wait);
4860 btrfs_destroy_delayed_inodes(fs_info);
4862 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4864 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4866 cur_trans->state =TRANS_STATE_COMPLETED;
4867 wake_up(&cur_trans->commit_wait);
4870 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4872 struct btrfs_transaction *t;
4874 mutex_lock(&fs_info->transaction_kthread_mutex);
4876 spin_lock(&fs_info->trans_lock);
4877 while (!list_empty(&fs_info->trans_list)) {
4878 t = list_first_entry(&fs_info->trans_list,
4879 struct btrfs_transaction, list);
4880 if (t->state >= TRANS_STATE_COMMIT_START) {
4881 refcount_inc(&t->use_count);
4882 spin_unlock(&fs_info->trans_lock);
4883 btrfs_wait_for_commit(fs_info, t->transid);
4884 btrfs_put_transaction(t);
4885 spin_lock(&fs_info->trans_lock);
4888 if (t == fs_info->running_transaction) {
4889 t->state = TRANS_STATE_COMMIT_DOING;
4890 spin_unlock(&fs_info->trans_lock);
4892 * We wait for 0 num_writers since we don't hold a trans
4893 * handle open currently for this transaction.
4895 wait_event(t->writer_wait,
4896 atomic_read(&t->num_writers) == 0);
4898 spin_unlock(&fs_info->trans_lock);
4900 btrfs_cleanup_one_transaction(t, fs_info);
4902 spin_lock(&fs_info->trans_lock);
4903 if (t == fs_info->running_transaction)
4904 fs_info->running_transaction = NULL;
4905 list_del_init(&t->list);
4906 spin_unlock(&fs_info->trans_lock);
4908 btrfs_put_transaction(t);
4909 trace_btrfs_transaction_commit(fs_info);
4910 spin_lock(&fs_info->trans_lock);
4912 spin_unlock(&fs_info->trans_lock);
4913 btrfs_destroy_all_ordered_extents(fs_info);
4914 btrfs_destroy_delayed_inodes(fs_info);
4915 btrfs_assert_delayed_root_empty(fs_info);
4916 btrfs_destroy_all_delalloc_inodes(fs_info);
4917 btrfs_drop_all_logs(fs_info);
4918 mutex_unlock(&fs_info->transaction_kthread_mutex);
4923 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4925 struct btrfs_path *path;
4927 struct extent_buffer *l;
4928 struct btrfs_key search_key;
4929 struct btrfs_key found_key;
4932 path = btrfs_alloc_path();
4936 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4937 search_key.type = -1;
4938 search_key.offset = (u64)-1;
4939 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4942 BUG_ON(ret == 0); /* Corruption */
4943 if (path->slots[0] > 0) {
4944 slot = path->slots[0] - 1;
4946 btrfs_item_key_to_cpu(l, &found_key, slot);
4947 root->free_objectid = max_t(u64, found_key.objectid + 1,
4948 BTRFS_FIRST_FREE_OBJECTID);
4950 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4954 btrfs_free_path(path);
4958 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4961 mutex_lock(&root->objectid_mutex);
4963 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4964 btrfs_warn(root->fs_info,
4965 "the objectid of root %llu reaches its highest value",
4966 root->root_key.objectid);
4971 *objectid = root->free_objectid++;
4974 mutex_unlock(&root->objectid_mutex);