1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "rcu-string.h"
33 #include "dev-replace.h"
37 #include "compression.h"
38 #include "tree-checker.h"
39 #include "ref-verify.h"
40 #include "block-group.h"
42 #include "space-info.h"
46 #include "accessors.h"
47 #include "extent-tree.h"
48 #include "root-tree.h"
50 #include "uuid-tree.h"
51 #include "relocation.h"
55 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
56 BTRFS_HEADER_FLAG_RELOC |\
57 BTRFS_SUPER_FLAG_ERROR |\
58 BTRFS_SUPER_FLAG_SEEDING |\
59 BTRFS_SUPER_FLAG_METADUMP |\
60 BTRFS_SUPER_FLAG_METADUMP_V2)
62 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
63 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
65 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
67 if (fs_info->csum_shash)
68 crypto_free_shash(fs_info->csum_shash);
72 * Compute the csum of a btree block and store the result to provided buffer.
74 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
76 struct btrfs_fs_info *fs_info = buf->fs_info;
77 const int num_pages = num_extent_pages(buf);
78 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
79 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
83 shash->tfm = fs_info->csum_shash;
84 crypto_shash_init(shash);
85 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
86 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
87 first_page_part - BTRFS_CSUM_SIZE);
89 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
90 kaddr = page_address(buf->pages[i]);
91 crypto_shash_update(shash, kaddr, PAGE_SIZE);
93 memset(result, 0, BTRFS_CSUM_SIZE);
94 crypto_shash_final(shash, result);
98 * we can't consider a given block up to date unless the transid of the
99 * block matches the transid in the parent node's pointer. This is how we
100 * detect blocks that either didn't get written at all or got written
101 * in the wrong place.
103 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
105 if (!extent_buffer_uptodate(eb))
108 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
114 if (!extent_buffer_uptodate(eb) ||
115 btrfs_header_generation(eb) != parent_transid) {
116 btrfs_err_rl(eb->fs_info,
117 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
118 eb->start, eb->read_mirror,
119 parent_transid, btrfs_header_generation(eb));
120 clear_extent_buffer_uptodate(eb);
126 static bool btrfs_supported_super_csum(u16 csum_type)
129 case BTRFS_CSUM_TYPE_CRC32:
130 case BTRFS_CSUM_TYPE_XXHASH:
131 case BTRFS_CSUM_TYPE_SHA256:
132 case BTRFS_CSUM_TYPE_BLAKE2:
140 * Return 0 if the superblock checksum type matches the checksum value of that
141 * algorithm. Pass the raw disk superblock data.
143 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
144 const struct btrfs_super_block *disk_sb)
146 char result[BTRFS_CSUM_SIZE];
147 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
149 shash->tfm = fs_info->csum_shash;
152 * The super_block structure does not span the whole
153 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
154 * filled with zeros and is included in the checksum.
156 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
157 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
159 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
165 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
168 struct btrfs_fs_info *fs_info = eb->fs_info;
169 int i, num_pages = num_extent_pages(eb);
172 if (sb_rdonly(fs_info->sb))
175 for (i = 0; i < num_pages; i++) {
176 struct page *p = eb->pages[i];
177 u64 start = max_t(u64, eb->start, page_offset(p));
178 u64 end = min_t(u64, eb->start + eb->len, page_offset(p) + PAGE_SIZE);
179 u32 len = end - start;
181 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
182 start, p, offset_in_page(start), mirror_num);
191 * helper to read a given tree block, doing retries as required when
192 * the checksums don't match and we have alternate mirrors to try.
194 * @check: expected tree parentness check, see the comments of the
195 * structure for details.
197 int btrfs_read_extent_buffer(struct extent_buffer *eb,
198 struct btrfs_tree_parent_check *check)
200 struct btrfs_fs_info *fs_info = eb->fs_info;
205 int failed_mirror = 0;
210 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
211 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
215 num_copies = btrfs_num_copies(fs_info,
220 if (!failed_mirror) {
222 failed_mirror = eb->read_mirror;
226 if (mirror_num == failed_mirror)
229 if (mirror_num > num_copies)
233 if (failed && !ret && failed_mirror)
234 btrfs_repair_eb_io_failure(eb, failed_mirror);
240 * Checksum a dirty tree block before IO.
242 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
244 struct extent_buffer *eb = bbio->private;
245 struct btrfs_fs_info *fs_info = eb->fs_info;
246 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 last_trans = btrfs_get_last_trans_committed(fs_info);
286 if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
289 "block=%llu bad generation, have %llu expect > %llu",
290 eb->start, btrfs_header_generation(eb), last_trans);
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];
317 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
321 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
322 * This is then overwritten by metadata_uuid if it is present in the
323 * device_list_add(). The same true for a seed device as well. So use of
324 * fs_devices::metadata_uuid is appropriate here.
326 if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
329 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
330 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
336 /* Do basic extent buffer checks at read time */
337 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
338 struct btrfs_tree_parent_check *check)
340 struct btrfs_fs_info *fs_info = eb->fs_info;
342 const u32 csum_size = fs_info->csum_size;
344 u8 result[BTRFS_CSUM_SIZE];
345 const u8 *header_csum;
350 found_start = btrfs_header_bytenr(eb);
351 if (found_start != eb->start) {
352 btrfs_err_rl(fs_info,
353 "bad tree block start, mirror %u want %llu have %llu",
354 eb->read_mirror, eb->start, found_start);
358 if (check_tree_block_fsid(eb)) {
359 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
360 eb->start, eb->read_mirror);
364 found_level = btrfs_header_level(eb);
365 if (found_level >= BTRFS_MAX_LEVEL) {
367 "bad tree block level, mirror %u level %d on logical %llu",
368 eb->read_mirror, btrfs_header_level(eb), eb->start);
373 csum_tree_block(eb, result);
374 header_csum = page_address(eb->pages[0]) +
375 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
377 if (memcmp(result, header_csum, csum_size) != 0) {
378 btrfs_warn_rl(fs_info,
379 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
380 eb->start, eb->read_mirror,
381 CSUM_FMT_VALUE(csum_size, header_csum),
382 CSUM_FMT_VALUE(csum_size, result),
383 btrfs_header_level(eb));
388 if (found_level != check->level) {
390 "level verify failed on logical %llu mirror %u wanted %u found %u",
391 eb->start, eb->read_mirror, check->level, found_level);
395 if (unlikely(check->transid &&
396 btrfs_header_generation(eb) != check->transid)) {
397 btrfs_err_rl(eb->fs_info,
398 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
399 eb->start, eb->read_mirror, check->transid,
400 btrfs_header_generation(eb));
404 if (check->has_first_key) {
405 struct btrfs_key *expect_key = &check->first_key;
406 struct btrfs_key found_key;
409 btrfs_node_key_to_cpu(eb, &found_key, 0);
411 btrfs_item_key_to_cpu(eb, &found_key, 0);
412 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
414 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
415 eb->start, check->transid,
416 expect_key->objectid,
417 expect_key->type, expect_key->offset,
418 found_key.objectid, found_key.type,
424 if (check->owner_root) {
425 ret = btrfs_check_eb_owner(eb, check->owner_root);
431 * If this is a leaf block and it is corrupt, set the corrupt bit so
432 * that we don't try and read the other copies of this block, just
435 if (found_level == 0 && btrfs_check_leaf(eb)) {
436 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
440 if (found_level > 0 && btrfs_check_node(eb))
445 "read time tree block corruption detected on logical %llu mirror %u",
446 eb->start, eb->read_mirror);
451 #ifdef CONFIG_MIGRATION
452 static int btree_migrate_folio(struct address_space *mapping,
453 struct folio *dst, struct folio *src, enum migrate_mode mode)
456 * we can't safely write a btree page from here,
457 * we haven't done the locking hook
459 if (folio_test_dirty(src))
462 * Buffers may be managed in a filesystem specific way.
463 * We must have no buffers or drop them.
465 if (folio_get_private(src) &&
466 !filemap_release_folio(src, GFP_KERNEL))
468 return migrate_folio(mapping, dst, src, mode);
471 #define btree_migrate_folio NULL
474 static int btree_writepages(struct address_space *mapping,
475 struct writeback_control *wbc)
477 struct btrfs_fs_info *fs_info;
480 if (wbc->sync_mode == WB_SYNC_NONE) {
482 if (wbc->for_kupdate)
485 fs_info = BTRFS_I(mapping->host)->root->fs_info;
486 /* this is a bit racy, but that's ok */
487 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
488 BTRFS_DIRTY_METADATA_THRESH,
489 fs_info->dirty_metadata_batch);
493 return btree_write_cache_pages(mapping, wbc);
496 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
498 if (folio_test_writeback(folio) || folio_test_dirty(folio))
501 return try_release_extent_buffer(&folio->page);
504 static void btree_invalidate_folio(struct folio *folio, size_t offset,
507 struct extent_io_tree *tree;
508 tree = &BTRFS_I(folio->mapping->host)->io_tree;
509 extent_invalidate_folio(tree, folio, offset);
510 btree_release_folio(folio, GFP_NOFS);
511 if (folio_get_private(folio)) {
512 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
513 "folio private not zero on folio %llu",
514 (unsigned long long)folio_pos(folio));
515 folio_detach_private(folio);
520 static bool btree_dirty_folio(struct address_space *mapping,
523 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
524 struct btrfs_subpage_info *spi = fs_info->subpage_info;
525 struct btrfs_subpage *subpage;
526 struct extent_buffer *eb;
528 u64 page_start = folio_pos(folio);
530 if (fs_info->sectorsize == PAGE_SIZE) {
531 eb = folio_get_private(folio);
533 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
534 BUG_ON(!atomic_read(&eb->refs));
535 btrfs_assert_tree_write_locked(eb);
536 return filemap_dirty_folio(mapping, folio);
540 subpage = folio_get_private(folio);
542 for (cur_bit = spi->dirty_offset;
543 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
548 spin_lock_irqsave(&subpage->lock, flags);
549 if (!test_bit(cur_bit, subpage->bitmaps)) {
550 spin_unlock_irqrestore(&subpage->lock, flags);
553 spin_unlock_irqrestore(&subpage->lock, flags);
554 cur = page_start + cur_bit * fs_info->sectorsize;
556 eb = find_extent_buffer(fs_info, cur);
558 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
559 ASSERT(atomic_read(&eb->refs));
560 btrfs_assert_tree_write_locked(eb);
561 free_extent_buffer(eb);
563 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
565 return filemap_dirty_folio(mapping, folio);
568 #define btree_dirty_folio filemap_dirty_folio
571 static const struct address_space_operations btree_aops = {
572 .writepages = btree_writepages,
573 .release_folio = btree_release_folio,
574 .invalidate_folio = btree_invalidate_folio,
575 .migrate_folio = btree_migrate_folio,
576 .dirty_folio = btree_dirty_folio,
579 struct extent_buffer *btrfs_find_create_tree_block(
580 struct btrfs_fs_info *fs_info,
581 u64 bytenr, u64 owner_root,
584 if (btrfs_is_testing(fs_info))
585 return alloc_test_extent_buffer(fs_info, bytenr);
586 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
590 * Read tree block at logical address @bytenr and do variant basic but critical
593 * @check: expected tree parentness check, see comments of the
594 * structure for details.
596 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
597 struct btrfs_tree_parent_check *check)
599 struct extent_buffer *buf = NULL;
604 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
609 ret = btrfs_read_extent_buffer(buf, check);
611 free_extent_buffer_stale(buf);
614 if (btrfs_check_eb_owner(buf, check->owner_root)) {
615 free_extent_buffer_stale(buf);
616 return ERR_PTR(-EUCLEAN);
622 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
625 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
627 memset(&root->root_key, 0, sizeof(root->root_key));
628 memset(&root->root_item, 0, sizeof(root->root_item));
629 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
630 root->fs_info = fs_info;
631 root->root_key.objectid = objectid;
633 root->commit_root = NULL;
635 RB_CLEAR_NODE(&root->rb_node);
637 root->last_trans = 0;
638 root->free_objectid = 0;
639 root->nr_delalloc_inodes = 0;
640 root->nr_ordered_extents = 0;
641 root->inode_tree = RB_ROOT;
642 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
644 btrfs_init_root_block_rsv(root);
646 INIT_LIST_HEAD(&root->dirty_list);
647 INIT_LIST_HEAD(&root->root_list);
648 INIT_LIST_HEAD(&root->delalloc_inodes);
649 INIT_LIST_HEAD(&root->delalloc_root);
650 INIT_LIST_HEAD(&root->ordered_extents);
651 INIT_LIST_HEAD(&root->ordered_root);
652 INIT_LIST_HEAD(&root->reloc_dirty_list);
653 INIT_LIST_HEAD(&root->logged_list[0]);
654 INIT_LIST_HEAD(&root->logged_list[1]);
655 spin_lock_init(&root->inode_lock);
656 spin_lock_init(&root->delalloc_lock);
657 spin_lock_init(&root->ordered_extent_lock);
658 spin_lock_init(&root->accounting_lock);
659 spin_lock_init(&root->log_extents_lock[0]);
660 spin_lock_init(&root->log_extents_lock[1]);
661 spin_lock_init(&root->qgroup_meta_rsv_lock);
662 mutex_init(&root->objectid_mutex);
663 mutex_init(&root->log_mutex);
664 mutex_init(&root->ordered_extent_mutex);
665 mutex_init(&root->delalloc_mutex);
666 init_waitqueue_head(&root->qgroup_flush_wait);
667 init_waitqueue_head(&root->log_writer_wait);
668 init_waitqueue_head(&root->log_commit_wait[0]);
669 init_waitqueue_head(&root->log_commit_wait[1]);
670 INIT_LIST_HEAD(&root->log_ctxs[0]);
671 INIT_LIST_HEAD(&root->log_ctxs[1]);
672 atomic_set(&root->log_commit[0], 0);
673 atomic_set(&root->log_commit[1], 0);
674 atomic_set(&root->log_writers, 0);
675 atomic_set(&root->log_batch, 0);
676 refcount_set(&root->refs, 1);
677 atomic_set(&root->snapshot_force_cow, 0);
678 atomic_set(&root->nr_swapfiles, 0);
679 btrfs_set_root_log_transid(root, 0);
680 root->log_transid_committed = -1;
681 btrfs_set_root_last_log_commit(root, 0);
684 extent_io_tree_init(fs_info, &root->dirty_log_pages,
685 IO_TREE_ROOT_DIRTY_LOG_PAGES);
686 extent_io_tree_init(fs_info, &root->log_csum_range,
687 IO_TREE_LOG_CSUM_RANGE);
690 spin_lock_init(&root->root_item_lock);
691 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
692 #ifdef CONFIG_BTRFS_DEBUG
693 INIT_LIST_HEAD(&root->leak_list);
694 spin_lock(&fs_info->fs_roots_radix_lock);
695 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
696 spin_unlock(&fs_info->fs_roots_radix_lock);
700 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
701 u64 objectid, gfp_t flags)
703 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
705 __setup_root(root, fs_info, objectid);
709 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
710 /* Should only be used by the testing infrastructure */
711 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
713 struct btrfs_root *root;
716 return ERR_PTR(-EINVAL);
718 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
720 return ERR_PTR(-ENOMEM);
722 /* We don't use the stripesize in selftest, set it as sectorsize */
723 root->alloc_bytenr = 0;
729 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
731 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
732 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
734 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
737 static int global_root_key_cmp(const void *k, const struct rb_node *node)
739 const struct btrfs_key *key = k;
740 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
742 return btrfs_comp_cpu_keys(key, &root->root_key);
745 int btrfs_global_root_insert(struct btrfs_root *root)
747 struct btrfs_fs_info *fs_info = root->fs_info;
751 write_lock(&fs_info->global_root_lock);
752 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
753 write_unlock(&fs_info->global_root_lock);
757 btrfs_warn(fs_info, "global root %llu %llu already exists",
758 root->root_key.objectid, root->root_key.offset);
763 void btrfs_global_root_delete(struct btrfs_root *root)
765 struct btrfs_fs_info *fs_info = root->fs_info;
767 write_lock(&fs_info->global_root_lock);
768 rb_erase(&root->rb_node, &fs_info->global_root_tree);
769 write_unlock(&fs_info->global_root_lock);
772 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
773 struct btrfs_key *key)
775 struct rb_node *node;
776 struct btrfs_root *root = NULL;
778 read_lock(&fs_info->global_root_lock);
779 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
781 root = container_of(node, struct btrfs_root, rb_node);
782 read_unlock(&fs_info->global_root_lock);
787 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
789 struct btrfs_block_group *block_group;
792 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
796 block_group = btrfs_lookup_block_group(fs_info, bytenr);
798 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
802 ret = block_group->global_root_id;
803 btrfs_put_block_group(block_group);
808 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
810 struct btrfs_key key = {
811 .objectid = BTRFS_CSUM_TREE_OBJECTID,
812 .type = BTRFS_ROOT_ITEM_KEY,
813 .offset = btrfs_global_root_id(fs_info, bytenr),
816 return btrfs_global_root(fs_info, &key);
819 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
821 struct btrfs_key key = {
822 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
823 .type = BTRFS_ROOT_ITEM_KEY,
824 .offset = btrfs_global_root_id(fs_info, bytenr),
827 return btrfs_global_root(fs_info, &key);
830 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
832 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
833 return fs_info->block_group_root;
834 return btrfs_extent_root(fs_info, 0);
837 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
840 struct btrfs_fs_info *fs_info = trans->fs_info;
841 struct extent_buffer *leaf;
842 struct btrfs_root *tree_root = fs_info->tree_root;
843 struct btrfs_root *root;
844 struct btrfs_key key;
845 unsigned int nofs_flag;
849 * We're holding a transaction handle, so use a NOFS memory allocation
850 * context to avoid deadlock if reclaim happens.
852 nofs_flag = memalloc_nofs_save();
853 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
854 memalloc_nofs_restore(nofs_flag);
856 return ERR_PTR(-ENOMEM);
858 root->root_key.objectid = objectid;
859 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
860 root->root_key.offset = 0;
862 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
863 0, BTRFS_NESTING_NORMAL);
871 btrfs_mark_buffer_dirty(trans, leaf);
873 root->commit_root = btrfs_root_node(root);
874 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
876 btrfs_set_root_flags(&root->root_item, 0);
877 btrfs_set_root_limit(&root->root_item, 0);
878 btrfs_set_root_bytenr(&root->root_item, leaf->start);
879 btrfs_set_root_generation(&root->root_item, trans->transid);
880 btrfs_set_root_level(&root->root_item, 0);
881 btrfs_set_root_refs(&root->root_item, 1);
882 btrfs_set_root_used(&root->root_item, leaf->len);
883 btrfs_set_root_last_snapshot(&root->root_item, 0);
884 btrfs_set_root_dirid(&root->root_item, 0);
885 if (is_fstree(objectid))
886 generate_random_guid(root->root_item.uuid);
888 export_guid(root->root_item.uuid, &guid_null);
889 btrfs_set_root_drop_level(&root->root_item, 0);
891 btrfs_tree_unlock(leaf);
893 key.objectid = objectid;
894 key.type = BTRFS_ROOT_ITEM_KEY;
896 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
903 btrfs_put_root(root);
908 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
909 struct btrfs_fs_info *fs_info)
911 struct btrfs_root *root;
913 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
915 return ERR_PTR(-ENOMEM);
917 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
918 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
919 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
924 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
925 struct btrfs_root *root)
927 struct extent_buffer *leaf;
930 * DON'T set SHAREABLE bit for log trees.
932 * Log trees are not exposed to user space thus can't be snapshotted,
933 * and they go away before a real commit is actually done.
935 * They do store pointers to file data extents, and those reference
936 * counts still get updated (along with back refs to the log tree).
939 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
940 NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
942 return PTR_ERR(leaf);
946 btrfs_mark_buffer_dirty(trans, root->node);
947 btrfs_tree_unlock(root->node);
952 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
953 struct btrfs_fs_info *fs_info)
955 struct btrfs_root *log_root;
957 log_root = alloc_log_tree(trans, fs_info);
958 if (IS_ERR(log_root))
959 return PTR_ERR(log_root);
961 if (!btrfs_is_zoned(fs_info)) {
962 int ret = btrfs_alloc_log_tree_node(trans, log_root);
965 btrfs_put_root(log_root);
970 WARN_ON(fs_info->log_root_tree);
971 fs_info->log_root_tree = log_root;
975 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
976 struct btrfs_root *root)
978 struct btrfs_fs_info *fs_info = root->fs_info;
979 struct btrfs_root *log_root;
980 struct btrfs_inode_item *inode_item;
983 log_root = alloc_log_tree(trans, fs_info);
984 if (IS_ERR(log_root))
985 return PTR_ERR(log_root);
987 ret = btrfs_alloc_log_tree_node(trans, log_root);
989 btrfs_put_root(log_root);
993 log_root->last_trans = trans->transid;
994 log_root->root_key.offset = root->root_key.objectid;
996 inode_item = &log_root->root_item.inode;
997 btrfs_set_stack_inode_generation(inode_item, 1);
998 btrfs_set_stack_inode_size(inode_item, 3);
999 btrfs_set_stack_inode_nlink(inode_item, 1);
1000 btrfs_set_stack_inode_nbytes(inode_item,
1002 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1004 btrfs_set_root_node(&log_root->root_item, log_root->node);
1006 WARN_ON(root->log_root);
1007 root->log_root = log_root;
1008 btrfs_set_root_log_transid(root, 0);
1009 root->log_transid_committed = -1;
1010 btrfs_set_root_last_log_commit(root, 0);
1014 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1015 struct btrfs_path *path,
1016 struct btrfs_key *key)
1018 struct btrfs_root *root;
1019 struct btrfs_tree_parent_check check = { 0 };
1020 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1025 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1027 return ERR_PTR(-ENOMEM);
1029 ret = btrfs_find_root(tree_root, key, path,
1030 &root->root_item, &root->root_key);
1037 generation = btrfs_root_generation(&root->root_item);
1038 level = btrfs_root_level(&root->root_item);
1039 check.level = level;
1040 check.transid = generation;
1041 check.owner_root = key->objectid;
1042 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1044 if (IS_ERR(root->node)) {
1045 ret = PTR_ERR(root->node);
1049 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1055 * For real fs, and not log/reloc trees, root owner must
1056 * match its root node owner
1058 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1059 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1060 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1061 root->root_key.objectid != btrfs_header_owner(root->node)) {
1063 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1064 root->root_key.objectid, root->node->start,
1065 btrfs_header_owner(root->node),
1066 root->root_key.objectid);
1070 root->commit_root = btrfs_root_node(root);
1073 btrfs_put_root(root);
1074 return ERR_PTR(ret);
1077 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1078 struct btrfs_key *key)
1080 struct btrfs_root *root;
1081 struct btrfs_path *path;
1083 path = btrfs_alloc_path();
1085 return ERR_PTR(-ENOMEM);
1086 root = read_tree_root_path(tree_root, path, key);
1087 btrfs_free_path(path);
1093 * Initialize subvolume root in-memory structure
1095 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1097 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1101 btrfs_drew_lock_init(&root->snapshot_lock);
1103 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1104 !btrfs_is_data_reloc_root(root) &&
1105 is_fstree(root->root_key.objectid)) {
1106 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1107 btrfs_check_and_init_root_item(&root->root_item);
1111 * Don't assign anonymous block device to roots that are not exposed to
1112 * userspace, the id pool is limited to 1M
1114 if (is_fstree(root->root_key.objectid) &&
1115 btrfs_root_refs(&root->root_item) > 0) {
1117 ret = get_anon_bdev(&root->anon_dev);
1121 root->anon_dev = anon_dev;
1125 mutex_lock(&root->objectid_mutex);
1126 ret = btrfs_init_root_free_objectid(root);
1128 mutex_unlock(&root->objectid_mutex);
1132 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1134 mutex_unlock(&root->objectid_mutex);
1138 /* The caller is responsible to call btrfs_free_fs_root */
1142 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1145 struct btrfs_root *root;
1147 spin_lock(&fs_info->fs_roots_radix_lock);
1148 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1149 (unsigned long)root_id);
1150 root = btrfs_grab_root(root);
1151 spin_unlock(&fs_info->fs_roots_radix_lock);
1155 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1158 struct btrfs_key key = {
1159 .objectid = objectid,
1160 .type = BTRFS_ROOT_ITEM_KEY,
1165 case BTRFS_ROOT_TREE_OBJECTID:
1166 return btrfs_grab_root(fs_info->tree_root);
1167 case BTRFS_EXTENT_TREE_OBJECTID:
1168 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1169 case BTRFS_CHUNK_TREE_OBJECTID:
1170 return btrfs_grab_root(fs_info->chunk_root);
1171 case BTRFS_DEV_TREE_OBJECTID:
1172 return btrfs_grab_root(fs_info->dev_root);
1173 case BTRFS_CSUM_TREE_OBJECTID:
1174 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1175 case BTRFS_QUOTA_TREE_OBJECTID:
1176 return btrfs_grab_root(fs_info->quota_root);
1177 case BTRFS_UUID_TREE_OBJECTID:
1178 return btrfs_grab_root(fs_info->uuid_root);
1179 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1180 return btrfs_grab_root(fs_info->block_group_root);
1181 case BTRFS_FREE_SPACE_TREE_OBJECTID:
1182 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1183 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1184 return btrfs_grab_root(fs_info->stripe_root);
1190 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1191 struct btrfs_root *root)
1195 ret = radix_tree_preload(GFP_NOFS);
1199 spin_lock(&fs_info->fs_roots_radix_lock);
1200 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1201 (unsigned long)root->root_key.objectid,
1204 btrfs_grab_root(root);
1205 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1207 spin_unlock(&fs_info->fs_roots_radix_lock);
1208 radix_tree_preload_end();
1213 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1215 #ifdef CONFIG_BTRFS_DEBUG
1216 struct btrfs_root *root;
1218 while (!list_empty(&fs_info->allocated_roots)) {
1219 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1221 root = list_first_entry(&fs_info->allocated_roots,
1222 struct btrfs_root, leak_list);
1223 btrfs_err(fs_info, "leaked root %s refcount %d",
1224 btrfs_root_name(&root->root_key, buf),
1225 refcount_read(&root->refs));
1226 while (refcount_read(&root->refs) > 1)
1227 btrfs_put_root(root);
1228 btrfs_put_root(root);
1233 static void free_global_roots(struct btrfs_fs_info *fs_info)
1235 struct btrfs_root *root;
1236 struct rb_node *node;
1238 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1239 root = rb_entry(node, struct btrfs_root, rb_node);
1240 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1241 btrfs_put_root(root);
1245 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1247 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1248 percpu_counter_destroy(&fs_info->delalloc_bytes);
1249 percpu_counter_destroy(&fs_info->ordered_bytes);
1250 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1251 btrfs_free_csum_hash(fs_info);
1252 btrfs_free_stripe_hash_table(fs_info);
1253 btrfs_free_ref_cache(fs_info);
1254 kfree(fs_info->balance_ctl);
1255 kfree(fs_info->delayed_root);
1256 free_global_roots(fs_info);
1257 btrfs_put_root(fs_info->tree_root);
1258 btrfs_put_root(fs_info->chunk_root);
1259 btrfs_put_root(fs_info->dev_root);
1260 btrfs_put_root(fs_info->quota_root);
1261 btrfs_put_root(fs_info->uuid_root);
1262 btrfs_put_root(fs_info->fs_root);
1263 btrfs_put_root(fs_info->data_reloc_root);
1264 btrfs_put_root(fs_info->block_group_root);
1265 btrfs_put_root(fs_info->stripe_root);
1266 btrfs_check_leaked_roots(fs_info);
1267 btrfs_extent_buffer_leak_debug_check(fs_info);
1268 kfree(fs_info->super_copy);
1269 kfree(fs_info->super_for_commit);
1270 kfree(fs_info->subpage_info);
1276 * Get an in-memory reference of a root structure.
1278 * For essential trees like root/extent tree, we grab it from fs_info directly.
1279 * For subvolume trees, we check the cached filesystem roots first. If not
1280 * found, then read it from disk and add it to cached fs roots.
1282 * Caller should release the root by calling btrfs_put_root() after the usage.
1284 * NOTE: Reloc and log trees can't be read by this function as they share the
1285 * same root objectid.
1287 * @objectid: root id
1288 * @anon_dev: preallocated anonymous block device number for new roots,
1289 * pass 0 for new allocation.
1290 * @check_ref: whether to check root item references, If true, return -ENOENT
1293 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1294 u64 objectid, dev_t anon_dev,
1297 struct btrfs_root *root;
1298 struct btrfs_path *path;
1299 struct btrfs_key key;
1302 root = btrfs_get_global_root(fs_info, objectid);
1307 * If we're called for non-subvolume trees, and above function didn't
1308 * find one, do not try to read it from disk.
1310 * This is namely for free-space-tree and quota tree, which can change
1311 * at runtime and should only be grabbed from fs_info.
1313 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1314 return ERR_PTR(-ENOENT);
1316 root = btrfs_lookup_fs_root(fs_info, objectid);
1318 /* Shouldn't get preallocated anon_dev for cached roots */
1320 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1321 btrfs_put_root(root);
1322 return ERR_PTR(-ENOENT);
1327 key.objectid = objectid;
1328 key.type = BTRFS_ROOT_ITEM_KEY;
1329 key.offset = (u64)-1;
1330 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1334 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1339 ret = btrfs_init_fs_root(root, anon_dev);
1343 path = btrfs_alloc_path();
1348 key.objectid = BTRFS_ORPHAN_OBJECTID;
1349 key.type = BTRFS_ORPHAN_ITEM_KEY;
1350 key.offset = objectid;
1352 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1353 btrfs_free_path(path);
1357 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1359 ret = btrfs_insert_fs_root(fs_info, root);
1361 if (ret == -EEXIST) {
1362 btrfs_put_root(root);
1370 * If our caller provided us an anonymous device, then it's his
1371 * responsibility to free it in case we fail. So we have to set our
1372 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1373 * and once again by our caller.
1377 btrfs_put_root(root);
1378 return ERR_PTR(ret);
1382 * Get in-memory reference of a root structure
1384 * @objectid: tree objectid
1385 * @check_ref: if set, verify that the tree exists and the item has at least
1388 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1389 u64 objectid, bool check_ref)
1391 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1395 * Get in-memory reference of a root structure, created as new, optionally pass
1396 * the anonymous block device id
1398 * @objectid: tree objectid
1399 * @anon_dev: if zero, allocate a new anonymous block device or use the
1402 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1403 u64 objectid, dev_t anon_dev)
1405 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1409 * Return a root for the given objectid.
1411 * @fs_info: the fs_info
1412 * @objectid: the objectid we need to lookup
1414 * This is exclusively used for backref walking, and exists specifically because
1415 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1416 * creation time, which means we may have to read the tree_root in order to look
1417 * up a fs root that is not in memory. If the root is not in memory we will
1418 * read the tree root commit root and look up the fs root from there. This is a
1419 * temporary root, it will not be inserted into the radix tree as it doesn't
1420 * have the most uptodate information, it'll simply be discarded once the
1421 * backref code is finished using the root.
1423 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1424 struct btrfs_path *path,
1427 struct btrfs_root *root;
1428 struct btrfs_key key;
1430 ASSERT(path->search_commit_root && path->skip_locking);
1433 * This can return -ENOENT if we ask for a root that doesn't exist, but
1434 * since this is called via the backref walking code we won't be looking
1435 * up a root that doesn't exist, unless there's corruption. So if root
1436 * != NULL just return it.
1438 root = btrfs_get_global_root(fs_info, objectid);
1442 root = btrfs_lookup_fs_root(fs_info, objectid);
1446 key.objectid = objectid;
1447 key.type = BTRFS_ROOT_ITEM_KEY;
1448 key.offset = (u64)-1;
1449 root = read_tree_root_path(fs_info->tree_root, path, &key);
1450 btrfs_release_path(path);
1455 static int cleaner_kthread(void *arg)
1457 struct btrfs_fs_info *fs_info = arg;
1463 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1465 /* Make the cleaner go to sleep early. */
1466 if (btrfs_need_cleaner_sleep(fs_info))
1470 * Do not do anything if we might cause open_ctree() to block
1471 * before we have finished mounting the filesystem.
1473 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1476 if (!mutex_trylock(&fs_info->cleaner_mutex))
1480 * Avoid the problem that we change the status of the fs
1481 * during the above check and trylock.
1483 if (btrfs_need_cleaner_sleep(fs_info)) {
1484 mutex_unlock(&fs_info->cleaner_mutex);
1488 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1489 btrfs_sysfs_feature_update(fs_info);
1491 btrfs_run_delayed_iputs(fs_info);
1493 again = btrfs_clean_one_deleted_snapshot(fs_info);
1494 mutex_unlock(&fs_info->cleaner_mutex);
1497 * The defragger has dealt with the R/O remount and umount,
1498 * needn't do anything special here.
1500 btrfs_run_defrag_inodes(fs_info);
1503 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1504 * with relocation (btrfs_relocate_chunk) and relocation
1505 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1506 * after acquiring fs_info->reclaim_bgs_lock. So we
1507 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1508 * unused block groups.
1510 btrfs_delete_unused_bgs(fs_info);
1513 * Reclaim block groups in the reclaim_bgs list after we deleted
1514 * all unused block_groups. This possibly gives us some more free
1517 btrfs_reclaim_bgs(fs_info);
1519 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1520 if (kthread_should_park())
1522 if (kthread_should_stop())
1525 set_current_state(TASK_INTERRUPTIBLE);
1527 __set_current_state(TASK_RUNNING);
1532 static int transaction_kthread(void *arg)
1534 struct btrfs_root *root = arg;
1535 struct btrfs_fs_info *fs_info = root->fs_info;
1536 struct btrfs_trans_handle *trans;
1537 struct btrfs_transaction *cur;
1540 unsigned long delay;
1544 cannot_commit = false;
1545 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1546 mutex_lock(&fs_info->transaction_kthread_mutex);
1548 spin_lock(&fs_info->trans_lock);
1549 cur = fs_info->running_transaction;
1551 spin_unlock(&fs_info->trans_lock);
1555 delta = ktime_get_seconds() - cur->start_time;
1556 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1557 cur->state < TRANS_STATE_COMMIT_PREP &&
1558 delta < fs_info->commit_interval) {
1559 spin_unlock(&fs_info->trans_lock);
1560 delay -= msecs_to_jiffies((delta - 1) * 1000);
1562 msecs_to_jiffies(fs_info->commit_interval * 1000));
1565 transid = cur->transid;
1566 spin_unlock(&fs_info->trans_lock);
1568 /* If the file system is aborted, this will always fail. */
1569 trans = btrfs_attach_transaction(root);
1570 if (IS_ERR(trans)) {
1571 if (PTR_ERR(trans) != -ENOENT)
1572 cannot_commit = true;
1575 if (transid == trans->transid) {
1576 btrfs_commit_transaction(trans);
1578 btrfs_end_transaction(trans);
1581 wake_up_process(fs_info->cleaner_kthread);
1582 mutex_unlock(&fs_info->transaction_kthread_mutex);
1584 if (BTRFS_FS_ERROR(fs_info))
1585 btrfs_cleanup_transaction(fs_info);
1586 if (!kthread_should_stop() &&
1587 (!btrfs_transaction_blocked(fs_info) ||
1589 schedule_timeout_interruptible(delay);
1590 } while (!kthread_should_stop());
1595 * This will find the highest generation in the array of root backups. The
1596 * index of the highest array is returned, or -EINVAL if we can't find
1599 * We check to make sure the array is valid by comparing the
1600 * generation of the latest root in the array with the generation
1601 * in the super block. If they don't match we pitch it.
1603 static int find_newest_super_backup(struct btrfs_fs_info *info)
1605 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1607 struct btrfs_root_backup *root_backup;
1610 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1611 root_backup = info->super_copy->super_roots + i;
1612 cur = btrfs_backup_tree_root_gen(root_backup);
1613 if (cur == newest_gen)
1621 * copy all the root pointers into the super backup array.
1622 * this will bump the backup pointer by one when it is
1625 static void backup_super_roots(struct btrfs_fs_info *info)
1627 const int next_backup = info->backup_root_index;
1628 struct btrfs_root_backup *root_backup;
1630 root_backup = info->super_for_commit->super_roots + next_backup;
1633 * make sure all of our padding and empty slots get zero filled
1634 * regardless of which ones we use today
1636 memset(root_backup, 0, sizeof(*root_backup));
1638 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1640 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1641 btrfs_set_backup_tree_root_gen(root_backup,
1642 btrfs_header_generation(info->tree_root->node));
1644 btrfs_set_backup_tree_root_level(root_backup,
1645 btrfs_header_level(info->tree_root->node));
1647 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1648 btrfs_set_backup_chunk_root_gen(root_backup,
1649 btrfs_header_generation(info->chunk_root->node));
1650 btrfs_set_backup_chunk_root_level(root_backup,
1651 btrfs_header_level(info->chunk_root->node));
1653 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1654 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1655 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1657 btrfs_set_backup_extent_root(root_backup,
1658 extent_root->node->start);
1659 btrfs_set_backup_extent_root_gen(root_backup,
1660 btrfs_header_generation(extent_root->node));
1661 btrfs_set_backup_extent_root_level(root_backup,
1662 btrfs_header_level(extent_root->node));
1664 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1665 btrfs_set_backup_csum_root_gen(root_backup,
1666 btrfs_header_generation(csum_root->node));
1667 btrfs_set_backup_csum_root_level(root_backup,
1668 btrfs_header_level(csum_root->node));
1672 * we might commit during log recovery, which happens before we set
1673 * the fs_root. Make sure it is valid before we fill it in.
1675 if (info->fs_root && info->fs_root->node) {
1676 btrfs_set_backup_fs_root(root_backup,
1677 info->fs_root->node->start);
1678 btrfs_set_backup_fs_root_gen(root_backup,
1679 btrfs_header_generation(info->fs_root->node));
1680 btrfs_set_backup_fs_root_level(root_backup,
1681 btrfs_header_level(info->fs_root->node));
1684 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1685 btrfs_set_backup_dev_root_gen(root_backup,
1686 btrfs_header_generation(info->dev_root->node));
1687 btrfs_set_backup_dev_root_level(root_backup,
1688 btrfs_header_level(info->dev_root->node));
1690 btrfs_set_backup_total_bytes(root_backup,
1691 btrfs_super_total_bytes(info->super_copy));
1692 btrfs_set_backup_bytes_used(root_backup,
1693 btrfs_super_bytes_used(info->super_copy));
1694 btrfs_set_backup_num_devices(root_backup,
1695 btrfs_super_num_devices(info->super_copy));
1698 * if we don't copy this out to the super_copy, it won't get remembered
1699 * for the next commit
1701 memcpy(&info->super_copy->super_roots,
1702 &info->super_for_commit->super_roots,
1703 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1707 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1708 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1710 * @fs_info: filesystem whose backup roots need to be read
1711 * @priority: priority of backup root required
1713 * Returns backup root index on success and -EINVAL otherwise.
1715 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1717 int backup_index = find_newest_super_backup(fs_info);
1718 struct btrfs_super_block *super = fs_info->super_copy;
1719 struct btrfs_root_backup *root_backup;
1721 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1723 return backup_index;
1725 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1726 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1731 root_backup = super->super_roots + backup_index;
1733 btrfs_set_super_generation(super,
1734 btrfs_backup_tree_root_gen(root_backup));
1735 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1736 btrfs_set_super_root_level(super,
1737 btrfs_backup_tree_root_level(root_backup));
1738 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1741 * Fixme: the total bytes and num_devices need to match or we should
1744 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1745 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1747 return backup_index;
1750 /* helper to cleanup workers */
1751 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1753 btrfs_destroy_workqueue(fs_info->fixup_workers);
1754 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1755 btrfs_destroy_workqueue(fs_info->workers);
1756 if (fs_info->endio_workers)
1757 destroy_workqueue(fs_info->endio_workers);
1758 if (fs_info->rmw_workers)
1759 destroy_workqueue(fs_info->rmw_workers);
1760 if (fs_info->compressed_write_workers)
1761 destroy_workqueue(fs_info->compressed_write_workers);
1762 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1763 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1764 btrfs_destroy_workqueue(fs_info->delayed_workers);
1765 btrfs_destroy_workqueue(fs_info->caching_workers);
1766 btrfs_destroy_workqueue(fs_info->flush_workers);
1767 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1768 if (fs_info->discard_ctl.discard_workers)
1769 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1771 * Now that all other work queues are destroyed, we can safely destroy
1772 * the queues used for metadata I/O, since tasks from those other work
1773 * queues can do metadata I/O operations.
1775 if (fs_info->endio_meta_workers)
1776 destroy_workqueue(fs_info->endio_meta_workers);
1779 static void free_root_extent_buffers(struct btrfs_root *root)
1782 free_extent_buffer(root->node);
1783 free_extent_buffer(root->commit_root);
1785 root->commit_root = NULL;
1789 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1791 struct btrfs_root *root, *tmp;
1793 rbtree_postorder_for_each_entry_safe(root, tmp,
1794 &fs_info->global_root_tree,
1796 free_root_extent_buffers(root);
1799 /* helper to cleanup tree roots */
1800 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1802 free_root_extent_buffers(info->tree_root);
1804 free_global_root_pointers(info);
1805 free_root_extent_buffers(info->dev_root);
1806 free_root_extent_buffers(info->quota_root);
1807 free_root_extent_buffers(info->uuid_root);
1808 free_root_extent_buffers(info->fs_root);
1809 free_root_extent_buffers(info->data_reloc_root);
1810 free_root_extent_buffers(info->block_group_root);
1811 free_root_extent_buffers(info->stripe_root);
1812 if (free_chunk_root)
1813 free_root_extent_buffers(info->chunk_root);
1816 void btrfs_put_root(struct btrfs_root *root)
1821 if (refcount_dec_and_test(&root->refs)) {
1822 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1823 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1825 free_anon_bdev(root->anon_dev);
1826 free_root_extent_buffers(root);
1827 #ifdef CONFIG_BTRFS_DEBUG
1828 spin_lock(&root->fs_info->fs_roots_radix_lock);
1829 list_del_init(&root->leak_list);
1830 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1836 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1839 struct btrfs_root *gang[8];
1842 while (!list_empty(&fs_info->dead_roots)) {
1843 gang[0] = list_entry(fs_info->dead_roots.next,
1844 struct btrfs_root, root_list);
1845 list_del(&gang[0]->root_list);
1847 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1848 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1849 btrfs_put_root(gang[0]);
1853 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1858 for (i = 0; i < ret; i++)
1859 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1863 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1865 mutex_init(&fs_info->scrub_lock);
1866 atomic_set(&fs_info->scrubs_running, 0);
1867 atomic_set(&fs_info->scrub_pause_req, 0);
1868 atomic_set(&fs_info->scrubs_paused, 0);
1869 atomic_set(&fs_info->scrub_cancel_req, 0);
1870 init_waitqueue_head(&fs_info->scrub_pause_wait);
1871 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1874 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1876 spin_lock_init(&fs_info->balance_lock);
1877 mutex_init(&fs_info->balance_mutex);
1878 atomic_set(&fs_info->balance_pause_req, 0);
1879 atomic_set(&fs_info->balance_cancel_req, 0);
1880 fs_info->balance_ctl = NULL;
1881 init_waitqueue_head(&fs_info->balance_wait_q);
1882 atomic_set(&fs_info->reloc_cancel_req, 0);
1885 static int btrfs_init_btree_inode(struct super_block *sb)
1887 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1888 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1889 fs_info->tree_root);
1890 struct inode *inode;
1892 inode = new_inode(sb);
1896 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1897 set_nlink(inode, 1);
1899 * we set the i_size on the btree inode to the max possible int.
1900 * the real end of the address space is determined by all of
1901 * the devices in the system
1903 inode->i_size = OFFSET_MAX;
1904 inode->i_mapping->a_ops = &btree_aops;
1905 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1907 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1908 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1909 IO_TREE_BTREE_INODE_IO);
1910 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1912 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1913 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1914 BTRFS_I(inode)->location.type = 0;
1915 BTRFS_I(inode)->location.offset = 0;
1916 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1917 __insert_inode_hash(inode, hash);
1918 fs_info->btree_inode = inode;
1923 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1925 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1926 init_rwsem(&fs_info->dev_replace.rwsem);
1927 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1930 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1932 spin_lock_init(&fs_info->qgroup_lock);
1933 mutex_init(&fs_info->qgroup_ioctl_lock);
1934 fs_info->qgroup_tree = RB_ROOT;
1935 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1936 fs_info->qgroup_seq = 1;
1937 fs_info->qgroup_ulist = NULL;
1938 fs_info->qgroup_rescan_running = false;
1939 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1940 mutex_init(&fs_info->qgroup_rescan_lock);
1943 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1945 u32 max_active = fs_info->thread_pool_size;
1946 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1947 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1950 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1952 fs_info->delalloc_workers =
1953 btrfs_alloc_workqueue(fs_info, "delalloc",
1954 flags, max_active, 2);
1956 fs_info->flush_workers =
1957 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1958 flags, max_active, 0);
1960 fs_info->caching_workers =
1961 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1963 fs_info->fixup_workers =
1964 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1966 fs_info->endio_workers =
1967 alloc_workqueue("btrfs-endio", flags, max_active);
1968 fs_info->endio_meta_workers =
1969 alloc_workqueue("btrfs-endio-meta", flags, max_active);
1970 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1971 fs_info->endio_write_workers =
1972 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1974 fs_info->compressed_write_workers =
1975 alloc_workqueue("btrfs-compressed-write", flags, max_active);
1976 fs_info->endio_freespace_worker =
1977 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
1979 fs_info->delayed_workers =
1980 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
1982 fs_info->qgroup_rescan_workers =
1983 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
1985 fs_info->discard_ctl.discard_workers =
1986 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
1988 if (!(fs_info->workers &&
1989 fs_info->delalloc_workers && fs_info->flush_workers &&
1990 fs_info->endio_workers && fs_info->endio_meta_workers &&
1991 fs_info->compressed_write_workers &&
1992 fs_info->endio_write_workers &&
1993 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
1994 fs_info->caching_workers && fs_info->fixup_workers &&
1995 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
1996 fs_info->discard_ctl.discard_workers)) {
2003 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2005 struct crypto_shash *csum_shash;
2006 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2008 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2010 if (IS_ERR(csum_shash)) {
2011 btrfs_err(fs_info, "error allocating %s hash for checksum",
2013 return PTR_ERR(csum_shash);
2016 fs_info->csum_shash = csum_shash;
2019 * Check if the checksum implementation is a fast accelerated one.
2020 * As-is this is a bit of a hack and should be replaced once the csum
2021 * implementations provide that information themselves.
2023 switch (csum_type) {
2024 case BTRFS_CSUM_TYPE_CRC32:
2025 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2026 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2028 case BTRFS_CSUM_TYPE_XXHASH:
2029 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2035 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2036 btrfs_super_csum_name(csum_type),
2037 crypto_shash_driver_name(csum_shash));
2041 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2042 struct btrfs_fs_devices *fs_devices)
2045 struct btrfs_tree_parent_check check = { 0 };
2046 struct btrfs_root *log_tree_root;
2047 struct btrfs_super_block *disk_super = fs_info->super_copy;
2048 u64 bytenr = btrfs_super_log_root(disk_super);
2049 int level = btrfs_super_log_root_level(disk_super);
2051 if (fs_devices->rw_devices == 0) {
2052 btrfs_warn(fs_info, "log replay required on RO media");
2056 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2061 check.level = level;
2062 check.transid = fs_info->generation + 1;
2063 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2064 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2065 if (IS_ERR(log_tree_root->node)) {
2066 btrfs_warn(fs_info, "failed to read log tree");
2067 ret = PTR_ERR(log_tree_root->node);
2068 log_tree_root->node = NULL;
2069 btrfs_put_root(log_tree_root);
2072 if (!extent_buffer_uptodate(log_tree_root->node)) {
2073 btrfs_err(fs_info, "failed to read log tree");
2074 btrfs_put_root(log_tree_root);
2078 /* returns with log_tree_root freed on success */
2079 ret = btrfs_recover_log_trees(log_tree_root);
2081 btrfs_handle_fs_error(fs_info, ret,
2082 "Failed to recover log tree");
2083 btrfs_put_root(log_tree_root);
2087 if (sb_rdonly(fs_info->sb)) {
2088 ret = btrfs_commit_super(fs_info);
2096 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2097 struct btrfs_path *path, u64 objectid,
2100 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2101 struct btrfs_root *root;
2102 u64 max_global_id = 0;
2104 struct btrfs_key key = {
2105 .objectid = objectid,
2106 .type = BTRFS_ROOT_ITEM_KEY,
2111 /* If we have IGNOREDATACSUMS skip loading these roots. */
2112 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2113 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2114 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2119 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2123 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2124 ret = btrfs_next_leaf(tree_root, path);
2133 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2134 if (key.objectid != objectid)
2136 btrfs_release_path(path);
2139 * Just worry about this for extent tree, it'll be the same for
2142 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2143 max_global_id = max(max_global_id, key.offset);
2146 root = read_tree_root_path(tree_root, path, &key);
2148 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2149 ret = PTR_ERR(root);
2152 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2153 ret = btrfs_global_root_insert(root);
2155 btrfs_put_root(root);
2160 btrfs_release_path(path);
2162 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2163 fs_info->nr_global_roots = max_global_id + 1;
2165 if (!found || ret) {
2166 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2167 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2169 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2170 ret = ret ? ret : -ENOENT;
2173 btrfs_err(fs_info, "failed to load root %s", name);
2178 static int load_global_roots(struct btrfs_root *tree_root)
2180 struct btrfs_path *path;
2183 path = btrfs_alloc_path();
2187 ret = load_global_roots_objectid(tree_root, path,
2188 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2191 ret = load_global_roots_objectid(tree_root, path,
2192 BTRFS_CSUM_TREE_OBJECTID, "csum");
2195 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2197 ret = load_global_roots_objectid(tree_root, path,
2198 BTRFS_FREE_SPACE_TREE_OBJECTID,
2201 btrfs_free_path(path);
2205 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2207 struct btrfs_root *tree_root = fs_info->tree_root;
2208 struct btrfs_root *root;
2209 struct btrfs_key location;
2212 BUG_ON(!fs_info->tree_root);
2214 ret = load_global_roots(tree_root);
2218 location.type = BTRFS_ROOT_ITEM_KEY;
2219 location.offset = 0;
2221 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2222 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2223 root = btrfs_read_tree_root(tree_root, &location);
2225 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2226 ret = PTR_ERR(root);
2230 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2231 fs_info->block_group_root = root;
2235 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2236 root = btrfs_read_tree_root(tree_root, &location);
2238 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2239 ret = PTR_ERR(root);
2243 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2244 fs_info->dev_root = root;
2246 /* Initialize fs_info for all devices in any case */
2247 ret = btrfs_init_devices_late(fs_info);
2252 * This tree can share blocks with some other fs tree during relocation
2253 * and we need a proper setup by btrfs_get_fs_root
2255 root = btrfs_get_fs_root(tree_root->fs_info,
2256 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2258 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2259 ret = PTR_ERR(root);
2263 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2264 fs_info->data_reloc_root = root;
2267 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2268 root = btrfs_read_tree_root(tree_root, &location);
2269 if (!IS_ERR(root)) {
2270 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2271 fs_info->quota_root = root;
2274 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2275 root = btrfs_read_tree_root(tree_root, &location);
2277 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2278 ret = PTR_ERR(root);
2283 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2284 fs_info->uuid_root = root;
2287 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2288 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2289 root = btrfs_read_tree_root(tree_root, &location);
2291 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2292 ret = PTR_ERR(root);
2296 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2297 fs_info->stripe_root = root;
2303 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2304 location.objectid, ret);
2309 * Real super block validation
2310 * NOTE: super csum type and incompat features will not be checked here.
2312 * @sb: super block to check
2313 * @mirror_num: the super block number to check its bytenr:
2314 * 0 the primary (1st) sb
2315 * 1, 2 2nd and 3rd backup copy
2316 * -1 skip bytenr check
2318 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2319 struct btrfs_super_block *sb, int mirror_num)
2321 u64 nodesize = btrfs_super_nodesize(sb);
2322 u64 sectorsize = btrfs_super_sectorsize(sb);
2325 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2326 btrfs_err(fs_info, "no valid FS found");
2329 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2330 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2331 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2334 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2335 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2336 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2339 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2340 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2341 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2344 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2345 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2346 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2351 * Check sectorsize and nodesize first, other check will need it.
2352 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2354 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2355 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2356 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2361 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2363 * We can support 16K sectorsize with 64K page size without problem,
2364 * but such sectorsize/pagesize combination doesn't make much sense.
2365 * 4K will be our future standard, PAGE_SIZE is supported from the very
2368 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2370 "sectorsize %llu not yet supported for page size %lu",
2371 sectorsize, PAGE_SIZE);
2375 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2376 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2377 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2380 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2381 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2382 le32_to_cpu(sb->__unused_leafsize), nodesize);
2386 /* Root alignment check */
2387 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2388 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2389 btrfs_super_root(sb));
2392 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2393 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2394 btrfs_super_chunk_root(sb));
2397 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2398 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2399 btrfs_super_log_root(sb));
2403 if (!fs_info->fs_devices->temp_fsid &&
2404 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2406 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2407 sb->fsid, fs_info->fs_devices->fsid);
2411 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2412 BTRFS_FSID_SIZE) != 0) {
2414 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2415 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2419 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2420 BTRFS_FSID_SIZE) != 0) {
2422 "dev_item UUID does not match metadata fsid: %pU != %pU",
2423 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2428 * Artificial requirement for block-group-tree to force newer features
2429 * (free-space-tree, no-holes) so the test matrix is smaller.
2431 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2432 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2433 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2435 "block-group-tree feature requires fres-space-tree and no-holes");
2440 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2443 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2444 btrfs_err(fs_info, "bytes_used is too small %llu",
2445 btrfs_super_bytes_used(sb));
2448 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2449 btrfs_err(fs_info, "invalid stripesize %u",
2450 btrfs_super_stripesize(sb));
2453 if (btrfs_super_num_devices(sb) > (1UL << 31))
2454 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2455 btrfs_super_num_devices(sb));
2456 if (btrfs_super_num_devices(sb) == 0) {
2457 btrfs_err(fs_info, "number of devices is 0");
2461 if (mirror_num >= 0 &&
2462 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2463 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2464 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2469 * Obvious sys_chunk_array corruptions, it must hold at least one key
2472 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2473 btrfs_err(fs_info, "system chunk array too big %u > %u",
2474 btrfs_super_sys_array_size(sb),
2475 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2478 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2479 + sizeof(struct btrfs_chunk)) {
2480 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2481 btrfs_super_sys_array_size(sb),
2482 sizeof(struct btrfs_disk_key)
2483 + sizeof(struct btrfs_chunk));
2488 * The generation is a global counter, we'll trust it more than the others
2489 * but it's still possible that it's the one that's wrong.
2491 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2493 "suspicious: generation < chunk_root_generation: %llu < %llu",
2494 btrfs_super_generation(sb),
2495 btrfs_super_chunk_root_generation(sb));
2496 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2497 && btrfs_super_cache_generation(sb) != (u64)-1)
2499 "suspicious: generation < cache_generation: %llu < %llu",
2500 btrfs_super_generation(sb),
2501 btrfs_super_cache_generation(sb));
2507 * Validation of super block at mount time.
2508 * Some checks already done early at mount time, like csum type and incompat
2509 * flags will be skipped.
2511 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2513 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2517 * Validation of super block at write time.
2518 * Some checks like bytenr check will be skipped as their values will be
2520 * Extra checks like csum type and incompat flags will be done here.
2522 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2523 struct btrfs_super_block *sb)
2527 ret = btrfs_validate_super(fs_info, sb, -1);
2530 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2532 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2533 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2536 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2539 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2540 btrfs_super_incompat_flags(sb),
2541 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2547 "super block corruption detected before writing it to disk");
2551 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2553 struct btrfs_tree_parent_check check = {
2556 .owner_root = root->root_key.objectid
2560 root->node = read_tree_block(root->fs_info, bytenr, &check);
2561 if (IS_ERR(root->node)) {
2562 ret = PTR_ERR(root->node);
2566 if (!extent_buffer_uptodate(root->node)) {
2567 free_extent_buffer(root->node);
2572 btrfs_set_root_node(&root->root_item, root->node);
2573 root->commit_root = btrfs_root_node(root);
2574 btrfs_set_root_refs(&root->root_item, 1);
2578 static int load_important_roots(struct btrfs_fs_info *fs_info)
2580 struct btrfs_super_block *sb = fs_info->super_copy;
2584 bytenr = btrfs_super_root(sb);
2585 gen = btrfs_super_generation(sb);
2586 level = btrfs_super_root_level(sb);
2587 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2589 btrfs_warn(fs_info, "couldn't read tree root");
2595 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2597 int backup_index = find_newest_super_backup(fs_info);
2598 struct btrfs_super_block *sb = fs_info->super_copy;
2599 struct btrfs_root *tree_root = fs_info->tree_root;
2600 bool handle_error = false;
2604 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2606 if (!IS_ERR(tree_root->node))
2607 free_extent_buffer(tree_root->node);
2608 tree_root->node = NULL;
2610 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2613 free_root_pointers(fs_info, 0);
2616 * Don't use the log in recovery mode, it won't be
2619 btrfs_set_super_log_root(sb, 0);
2621 /* We can't trust the free space cache either */
2622 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2624 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2625 ret = read_backup_root(fs_info, i);
2631 ret = load_important_roots(fs_info);
2633 handle_error = true;
2638 * No need to hold btrfs_root::objectid_mutex since the fs
2639 * hasn't been fully initialised and we are the only user
2641 ret = btrfs_init_root_free_objectid(tree_root);
2643 handle_error = true;
2647 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2649 ret = btrfs_read_roots(fs_info);
2651 handle_error = true;
2655 /* All successful */
2656 fs_info->generation = btrfs_header_generation(tree_root->node);
2657 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2658 fs_info->last_reloc_trans = 0;
2660 /* Always begin writing backup roots after the one being used */
2661 if (backup_index < 0) {
2662 fs_info->backup_root_index = 0;
2664 fs_info->backup_root_index = backup_index + 1;
2665 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2673 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2675 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2676 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2677 INIT_LIST_HEAD(&fs_info->trans_list);
2678 INIT_LIST_HEAD(&fs_info->dead_roots);
2679 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2680 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2681 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2682 spin_lock_init(&fs_info->delalloc_root_lock);
2683 spin_lock_init(&fs_info->trans_lock);
2684 spin_lock_init(&fs_info->fs_roots_radix_lock);
2685 spin_lock_init(&fs_info->delayed_iput_lock);
2686 spin_lock_init(&fs_info->defrag_inodes_lock);
2687 spin_lock_init(&fs_info->super_lock);
2688 spin_lock_init(&fs_info->buffer_lock);
2689 spin_lock_init(&fs_info->unused_bgs_lock);
2690 spin_lock_init(&fs_info->treelog_bg_lock);
2691 spin_lock_init(&fs_info->zone_active_bgs_lock);
2692 spin_lock_init(&fs_info->relocation_bg_lock);
2693 rwlock_init(&fs_info->tree_mod_log_lock);
2694 rwlock_init(&fs_info->global_root_lock);
2695 mutex_init(&fs_info->unused_bg_unpin_mutex);
2696 mutex_init(&fs_info->reclaim_bgs_lock);
2697 mutex_init(&fs_info->reloc_mutex);
2698 mutex_init(&fs_info->delalloc_root_mutex);
2699 mutex_init(&fs_info->zoned_meta_io_lock);
2700 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2701 seqlock_init(&fs_info->profiles_lock);
2703 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2704 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2705 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2706 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2707 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2708 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2709 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2710 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2711 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2712 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2713 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2714 BTRFS_LOCKDEP_TRANS_COMPLETED);
2716 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2717 INIT_LIST_HEAD(&fs_info->space_info);
2718 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2719 INIT_LIST_HEAD(&fs_info->unused_bgs);
2720 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2721 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2722 #ifdef CONFIG_BTRFS_DEBUG
2723 INIT_LIST_HEAD(&fs_info->allocated_roots);
2724 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2725 spin_lock_init(&fs_info->eb_leak_lock);
2727 extent_map_tree_init(&fs_info->mapping_tree);
2728 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2729 BTRFS_BLOCK_RSV_GLOBAL);
2730 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2731 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2732 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2733 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2734 BTRFS_BLOCK_RSV_DELOPS);
2735 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2736 BTRFS_BLOCK_RSV_DELREFS);
2738 atomic_set(&fs_info->async_delalloc_pages, 0);
2739 atomic_set(&fs_info->defrag_running, 0);
2740 atomic_set(&fs_info->nr_delayed_iputs, 0);
2741 atomic64_set(&fs_info->tree_mod_seq, 0);
2742 fs_info->global_root_tree = RB_ROOT;
2743 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2744 fs_info->metadata_ratio = 0;
2745 fs_info->defrag_inodes = RB_ROOT;
2746 atomic64_set(&fs_info->free_chunk_space, 0);
2747 fs_info->tree_mod_log = RB_ROOT;
2748 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2749 btrfs_init_ref_verify(fs_info);
2751 fs_info->thread_pool_size = min_t(unsigned long,
2752 num_online_cpus() + 2, 8);
2754 INIT_LIST_HEAD(&fs_info->ordered_roots);
2755 spin_lock_init(&fs_info->ordered_root_lock);
2757 btrfs_init_scrub(fs_info);
2758 btrfs_init_balance(fs_info);
2759 btrfs_init_async_reclaim_work(fs_info);
2761 rwlock_init(&fs_info->block_group_cache_lock);
2762 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2764 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2765 IO_TREE_FS_EXCLUDED_EXTENTS);
2767 mutex_init(&fs_info->ordered_operations_mutex);
2768 mutex_init(&fs_info->tree_log_mutex);
2769 mutex_init(&fs_info->chunk_mutex);
2770 mutex_init(&fs_info->transaction_kthread_mutex);
2771 mutex_init(&fs_info->cleaner_mutex);
2772 mutex_init(&fs_info->ro_block_group_mutex);
2773 init_rwsem(&fs_info->commit_root_sem);
2774 init_rwsem(&fs_info->cleanup_work_sem);
2775 init_rwsem(&fs_info->subvol_sem);
2776 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2778 btrfs_init_dev_replace_locks(fs_info);
2779 btrfs_init_qgroup(fs_info);
2780 btrfs_discard_init(fs_info);
2782 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2783 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2785 init_waitqueue_head(&fs_info->transaction_throttle);
2786 init_waitqueue_head(&fs_info->transaction_wait);
2787 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2788 init_waitqueue_head(&fs_info->async_submit_wait);
2789 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2791 /* Usable values until the real ones are cached from the superblock */
2792 fs_info->nodesize = 4096;
2793 fs_info->sectorsize = 4096;
2794 fs_info->sectorsize_bits = ilog2(4096);
2795 fs_info->stripesize = 4096;
2797 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2799 spin_lock_init(&fs_info->swapfile_pins_lock);
2800 fs_info->swapfile_pins = RB_ROOT;
2802 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2803 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2806 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2811 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2812 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2814 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2818 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2822 fs_info->dirty_metadata_batch = PAGE_SIZE *
2823 (1 + ilog2(nr_cpu_ids));
2825 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2829 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2834 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2836 if (!fs_info->delayed_root)
2838 btrfs_init_delayed_root(fs_info->delayed_root);
2841 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2843 return btrfs_alloc_stripe_hash_table(fs_info);
2846 static int btrfs_uuid_rescan_kthread(void *data)
2848 struct btrfs_fs_info *fs_info = data;
2852 * 1st step is to iterate through the existing UUID tree and
2853 * to delete all entries that contain outdated data.
2854 * 2nd step is to add all missing entries to the UUID tree.
2856 ret = btrfs_uuid_tree_iterate(fs_info);
2859 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2861 up(&fs_info->uuid_tree_rescan_sem);
2864 return btrfs_uuid_scan_kthread(data);
2867 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2869 struct task_struct *task;
2871 down(&fs_info->uuid_tree_rescan_sem);
2872 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2874 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2875 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2876 up(&fs_info->uuid_tree_rescan_sem);
2877 return PTR_ERR(task);
2883 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2885 u64 root_objectid = 0;
2886 struct btrfs_root *gang[8];
2889 unsigned int ret = 0;
2892 spin_lock(&fs_info->fs_roots_radix_lock);
2893 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2894 (void **)gang, root_objectid,
2897 spin_unlock(&fs_info->fs_roots_radix_lock);
2900 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2902 for (i = 0; i < ret; i++) {
2903 /* Avoid to grab roots in dead_roots. */
2904 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2908 /* Grab all the search result for later use. */
2909 gang[i] = btrfs_grab_root(gang[i]);
2911 spin_unlock(&fs_info->fs_roots_radix_lock);
2913 for (i = 0; i < ret; i++) {
2916 root_objectid = gang[i]->root_key.objectid;
2917 err = btrfs_orphan_cleanup(gang[i]);
2920 btrfs_put_root(gang[i]);
2925 /* Release the uncleaned roots due to error. */
2926 for (; i < ret; i++) {
2928 btrfs_put_root(gang[i]);
2934 * Some options only have meaning at mount time and shouldn't persist across
2935 * remounts, or be displayed. Clear these at the end of mount and remount
2938 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
2940 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
2941 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
2945 * Mounting logic specific to read-write file systems. Shared by open_ctree
2946 * and btrfs_remount when remounting from read-only to read-write.
2948 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2951 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2952 bool rebuild_free_space_tree = false;
2954 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2955 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2956 rebuild_free_space_tree = true;
2957 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2958 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2959 btrfs_warn(fs_info, "free space tree is invalid");
2960 rebuild_free_space_tree = true;
2963 if (rebuild_free_space_tree) {
2964 btrfs_info(fs_info, "rebuilding free space tree");
2965 ret = btrfs_rebuild_free_space_tree(fs_info);
2968 "failed to rebuild free space tree: %d", ret);
2973 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2974 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
2975 btrfs_info(fs_info, "disabling free space tree");
2976 ret = btrfs_delete_free_space_tree(fs_info);
2979 "failed to disable free space tree: %d", ret);
2985 * btrfs_find_orphan_roots() is responsible for finding all the dead
2986 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
2987 * them into the fs_info->fs_roots_radix tree. This must be done before
2988 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
2989 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
2990 * item before the root's tree is deleted - this means that if we unmount
2991 * or crash before the deletion completes, on the next mount we will not
2992 * delete what remains of the tree because the orphan item does not
2993 * exists anymore, which is what tells us we have a pending deletion.
2995 ret = btrfs_find_orphan_roots(fs_info);
2999 ret = btrfs_cleanup_fs_roots(fs_info);
3003 down_read(&fs_info->cleanup_work_sem);
3004 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3005 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3006 up_read(&fs_info->cleanup_work_sem);
3009 up_read(&fs_info->cleanup_work_sem);
3011 mutex_lock(&fs_info->cleaner_mutex);
3012 ret = btrfs_recover_relocation(fs_info);
3013 mutex_unlock(&fs_info->cleaner_mutex);
3015 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3019 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3020 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3021 btrfs_info(fs_info, "creating free space tree");
3022 ret = btrfs_create_free_space_tree(fs_info);
3025 "failed to create free space tree: %d", ret);
3030 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3031 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3036 ret = btrfs_resume_balance_async(fs_info);
3040 ret = btrfs_resume_dev_replace_async(fs_info);
3042 btrfs_warn(fs_info, "failed to resume dev_replace");
3046 btrfs_qgroup_rescan_resume(fs_info);
3048 if (!fs_info->uuid_root) {
3049 btrfs_info(fs_info, "creating UUID tree");
3050 ret = btrfs_create_uuid_tree(fs_info);
3053 "failed to create the UUID tree %d", ret);
3063 * Do various sanity and dependency checks of different features.
3065 * @is_rw_mount: If the mount is read-write.
3067 * This is the place for less strict checks (like for subpage or artificial
3068 * feature dependencies).
3070 * For strict checks or possible corruption detection, see
3071 * btrfs_validate_super().
3073 * This should be called after btrfs_parse_options(), as some mount options
3074 * (space cache related) can modify on-disk format like free space tree and
3075 * screw up certain feature dependencies.
3077 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3079 struct btrfs_super_block *disk_super = fs_info->super_copy;
3080 u64 incompat = btrfs_super_incompat_flags(disk_super);
3081 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3082 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3084 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3086 "cannot mount because of unknown incompat features (0x%llx)",
3091 /* Runtime limitation for mixed block groups. */
3092 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3093 (fs_info->sectorsize != fs_info->nodesize)) {
3095 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3096 fs_info->nodesize, fs_info->sectorsize);
3100 /* Mixed backref is an always-enabled feature. */
3101 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3103 /* Set compression related flags just in case. */
3104 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3105 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3106 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3107 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3110 * An ancient flag, which should really be marked deprecated.
3111 * Such runtime limitation doesn't really need a incompat flag.
3113 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3114 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3116 if (compat_ro_unsupp && is_rw_mount) {
3118 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3124 * We have unsupported RO compat features, although RO mounted, we
3125 * should not cause any metadata writes, including log replay.
3126 * Or we could screw up whatever the new feature requires.
3128 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3129 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3131 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3137 * Artificial limitations for block group tree, to force
3138 * block-group-tree to rely on no-holes and free-space-tree.
3140 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3141 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3142 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3144 "block-group-tree feature requires no-holes and free-space-tree features");
3149 * Subpage runtime limitation on v1 cache.
3151 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3152 * we're already defaulting to v2 cache, no need to bother v1 as it's
3153 * going to be deprecated anyway.
3155 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3157 "v1 space cache is not supported for page size %lu with sectorsize %u",
3158 PAGE_SIZE, fs_info->sectorsize);
3162 /* This can be called by remount, we need to protect the super block. */
3163 spin_lock(&fs_info->super_lock);
3164 btrfs_set_super_incompat_flags(disk_super, incompat);
3165 spin_unlock(&fs_info->super_lock);
3170 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3178 struct btrfs_super_block *disk_super;
3179 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3180 struct btrfs_root *tree_root;
3181 struct btrfs_root *chunk_root;
3185 ret = init_mount_fs_info(fs_info, sb);
3189 /* These need to be init'ed before we start creating inodes and such. */
3190 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3192 fs_info->tree_root = tree_root;
3193 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3195 fs_info->chunk_root = chunk_root;
3196 if (!tree_root || !chunk_root) {
3201 ret = btrfs_init_btree_inode(sb);
3205 invalidate_bdev(fs_devices->latest_dev->bdev);
3208 * Read super block and check the signature bytes only
3210 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3211 if (IS_ERR(disk_super)) {
3212 ret = PTR_ERR(disk_super);
3216 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3218 * Verify the type first, if that or the checksum value are
3219 * corrupted, we'll find out
3221 csum_type = btrfs_super_csum_type(disk_super);
3222 if (!btrfs_supported_super_csum(csum_type)) {
3223 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3226 btrfs_release_disk_super(disk_super);
3230 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3232 ret = btrfs_init_csum_hash(fs_info, csum_type);
3234 btrfs_release_disk_super(disk_super);
3239 * We want to check superblock checksum, the type is stored inside.
3240 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3242 if (btrfs_check_super_csum(fs_info, disk_super)) {
3243 btrfs_err(fs_info, "superblock checksum mismatch");
3245 btrfs_release_disk_super(disk_super);
3250 * super_copy is zeroed at allocation time and we never touch the
3251 * following bytes up to INFO_SIZE, the checksum is calculated from
3252 * the whole block of INFO_SIZE
3254 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3255 btrfs_release_disk_super(disk_super);
3257 disk_super = fs_info->super_copy;
3259 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3260 sizeof(*fs_info->super_for_commit));
3262 ret = btrfs_validate_mount_super(fs_info);
3264 btrfs_err(fs_info, "superblock contains fatal errors");
3269 if (!btrfs_super_root(disk_super)) {
3270 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3275 /* check FS state, whether FS is broken. */
3276 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3277 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3280 * In the long term, we'll store the compression type in the super
3281 * block, and it'll be used for per file compression control.
3283 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3286 /* Set up fs_info before parsing mount options */
3287 nodesize = btrfs_super_nodesize(disk_super);
3288 sectorsize = btrfs_super_sectorsize(disk_super);
3289 stripesize = sectorsize;
3290 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3291 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3293 fs_info->nodesize = nodesize;
3294 fs_info->sectorsize = sectorsize;
3295 fs_info->sectorsize_bits = ilog2(sectorsize);
3296 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3297 fs_info->stripesize = stripesize;
3299 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3303 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3307 if (sectorsize < PAGE_SIZE) {
3308 struct btrfs_subpage_info *subpage_info;
3311 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3312 * going to be deprecated.
3314 * Force to use v2 cache for subpage case.
3316 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3317 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3318 "forcing free space tree for sector size %u with page size %lu",
3319 sectorsize, PAGE_SIZE);
3322 "read-write for sector size %u with page size %lu is experimental",
3323 sectorsize, PAGE_SIZE);
3324 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3325 if (!subpage_info) {
3329 btrfs_init_subpage_info(subpage_info, sectorsize);
3330 fs_info->subpage_info = subpage_info;
3333 ret = btrfs_init_workqueues(fs_info);
3335 goto fail_sb_buffer;
3337 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3338 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3340 sb->s_blocksize = sectorsize;
3341 sb->s_blocksize_bits = blksize_bits(sectorsize);
3342 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3344 mutex_lock(&fs_info->chunk_mutex);
3345 ret = btrfs_read_sys_array(fs_info);
3346 mutex_unlock(&fs_info->chunk_mutex);
3348 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3349 goto fail_sb_buffer;
3352 generation = btrfs_super_chunk_root_generation(disk_super);
3353 level = btrfs_super_chunk_root_level(disk_super);
3354 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3357 btrfs_err(fs_info, "failed to read chunk root");
3358 goto fail_tree_roots;
3361 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3362 offsetof(struct btrfs_header, chunk_tree_uuid),
3365 ret = btrfs_read_chunk_tree(fs_info);
3367 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3368 goto fail_tree_roots;
3372 * At this point we know all the devices that make this filesystem,
3373 * including the seed devices but we don't know yet if the replace
3374 * target is required. So free devices that are not part of this
3375 * filesystem but skip the replace target device which is checked
3376 * below in btrfs_init_dev_replace().
3378 btrfs_free_extra_devids(fs_devices);
3379 if (!fs_devices->latest_dev->bdev) {
3380 btrfs_err(fs_info, "failed to read devices");
3382 goto fail_tree_roots;
3385 ret = init_tree_roots(fs_info);
3387 goto fail_tree_roots;
3390 * Get zone type information of zoned block devices. This will also
3391 * handle emulation of a zoned filesystem if a regular device has the
3392 * zoned incompat feature flag set.
3394 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3397 "zoned: failed to read device zone info: %d", ret);
3398 goto fail_block_groups;
3402 * If we have a uuid root and we're not being told to rescan we need to
3403 * check the generation here so we can set the
3404 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3405 * transaction during a balance or the log replay without updating the
3406 * uuid generation, and then if we crash we would rescan the uuid tree,
3407 * even though it was perfectly fine.
3409 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3410 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3411 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3413 ret = btrfs_verify_dev_extents(fs_info);
3416 "failed to verify dev extents against chunks: %d",
3418 goto fail_block_groups;
3420 ret = btrfs_recover_balance(fs_info);
3422 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3423 goto fail_block_groups;
3426 ret = btrfs_init_dev_stats(fs_info);
3428 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3429 goto fail_block_groups;
3432 ret = btrfs_init_dev_replace(fs_info);
3434 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3435 goto fail_block_groups;
3438 ret = btrfs_check_zoned_mode(fs_info);
3440 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3442 goto fail_block_groups;
3445 ret = btrfs_sysfs_add_fsid(fs_devices);
3447 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3449 goto fail_block_groups;
3452 ret = btrfs_sysfs_add_mounted(fs_info);
3454 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3455 goto fail_fsdev_sysfs;
3458 ret = btrfs_init_space_info(fs_info);
3460 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3464 ret = btrfs_read_block_groups(fs_info);
3466 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3470 btrfs_free_zone_cache(fs_info);
3472 btrfs_check_active_zone_reservation(fs_info);
3474 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3475 !btrfs_check_rw_degradable(fs_info, NULL)) {
3477 "writable mount is not allowed due to too many missing devices");
3482 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3484 if (IS_ERR(fs_info->cleaner_kthread)) {
3485 ret = PTR_ERR(fs_info->cleaner_kthread);
3489 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3491 "btrfs-transaction");
3492 if (IS_ERR(fs_info->transaction_kthread)) {
3493 ret = PTR_ERR(fs_info->transaction_kthread);
3497 if (!btrfs_test_opt(fs_info, NOSSD) &&
3498 !fs_info->fs_devices->rotating) {
3499 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3503 * For devices supporting discard turn on discard=async automatically,
3504 * unless it's already set or disabled. This could be turned off by
3505 * nodiscard for the same mount.
3507 * The zoned mode piggy backs on the discard functionality for
3508 * resetting a zone. There is no reason to delay the zone reset as it is
3509 * fast enough. So, do not enable async discard for zoned mode.
3511 if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3512 btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3513 btrfs_test_opt(fs_info, NODISCARD)) &&
3514 fs_info->fs_devices->discardable &&
3515 !btrfs_is_zoned(fs_info)) {
3516 btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3517 "auto enabling async discard");
3520 ret = btrfs_read_qgroup_config(fs_info);
3522 goto fail_trans_kthread;
3524 if (btrfs_build_ref_tree(fs_info))
3525 btrfs_err(fs_info, "couldn't build ref tree");
3527 /* do not make disk changes in broken FS or nologreplay is given */
3528 if (btrfs_super_log_root(disk_super) != 0 &&
3529 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3530 btrfs_info(fs_info, "start tree-log replay");
3531 ret = btrfs_replay_log(fs_info, fs_devices);
3536 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3537 if (IS_ERR(fs_info->fs_root)) {
3538 ret = PTR_ERR(fs_info->fs_root);
3539 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3540 fs_info->fs_root = NULL;
3547 ret = btrfs_start_pre_rw_mount(fs_info);
3549 close_ctree(fs_info);
3552 btrfs_discard_resume(fs_info);
3554 if (fs_info->uuid_root &&
3555 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3556 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3557 btrfs_info(fs_info, "checking UUID tree");
3558 ret = btrfs_check_uuid_tree(fs_info);
3561 "failed to check the UUID tree: %d", ret);
3562 close_ctree(fs_info);
3567 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3569 /* Kick the cleaner thread so it'll start deleting snapshots. */
3570 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3571 wake_up_process(fs_info->cleaner_kthread);
3574 btrfs_clear_oneshot_options(fs_info);
3578 btrfs_free_qgroup_config(fs_info);
3580 kthread_stop(fs_info->transaction_kthread);
3581 btrfs_cleanup_transaction(fs_info);
3582 btrfs_free_fs_roots(fs_info);
3584 kthread_stop(fs_info->cleaner_kthread);
3587 * make sure we're done with the btree inode before we stop our
3590 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3593 btrfs_sysfs_remove_mounted(fs_info);
3596 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3599 btrfs_put_block_group_cache(fs_info);
3602 if (fs_info->data_reloc_root)
3603 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3604 free_root_pointers(fs_info, true);
3605 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3608 btrfs_stop_all_workers(fs_info);
3609 btrfs_free_block_groups(fs_info);
3611 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3613 iput(fs_info->btree_inode);
3615 btrfs_close_devices(fs_info->fs_devices);
3619 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3621 static void btrfs_end_super_write(struct bio *bio)
3623 struct btrfs_device *device = bio->bi_private;
3624 struct bio_vec *bvec;
3625 struct bvec_iter_all iter_all;
3628 bio_for_each_segment_all(bvec, bio, iter_all) {
3629 page = bvec->bv_page;
3631 if (bio->bi_status) {
3632 btrfs_warn_rl_in_rcu(device->fs_info,
3633 "lost page write due to IO error on %s (%d)",
3634 btrfs_dev_name(device),
3635 blk_status_to_errno(bio->bi_status));
3636 ClearPageUptodate(page);
3638 btrfs_dev_stat_inc_and_print(device,
3639 BTRFS_DEV_STAT_WRITE_ERRS);
3641 SetPageUptodate(page);
3651 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3652 int copy_num, bool drop_cache)
3654 struct btrfs_super_block *super;
3656 u64 bytenr, bytenr_orig;
3657 struct address_space *mapping = bdev->bd_inode->i_mapping;
3660 bytenr_orig = btrfs_sb_offset(copy_num);
3661 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3663 return ERR_PTR(-EINVAL);
3665 return ERR_PTR(ret);
3667 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3668 return ERR_PTR(-EINVAL);
3671 /* This should only be called with the primary sb. */
3672 ASSERT(copy_num == 0);
3675 * Drop the page of the primary superblock, so later read will
3676 * always read from the device.
3678 invalidate_inode_pages2_range(mapping,
3679 bytenr >> PAGE_SHIFT,
3680 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3683 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3685 return ERR_CAST(page);
3687 super = page_address(page);
3688 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3689 btrfs_release_disk_super(super);
3690 return ERR_PTR(-ENODATA);
3693 if (btrfs_super_bytenr(super) != bytenr_orig) {
3694 btrfs_release_disk_super(super);
3695 return ERR_PTR(-EINVAL);
3702 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3704 struct btrfs_super_block *super, *latest = NULL;
3708 /* we would like to check all the supers, but that would make
3709 * a btrfs mount succeed after a mkfs from a different FS.
3710 * So, we need to add a special mount option to scan for
3711 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3713 for (i = 0; i < 1; i++) {
3714 super = btrfs_read_dev_one_super(bdev, i, false);
3718 if (!latest || btrfs_super_generation(super) > transid) {
3720 btrfs_release_disk_super(super);
3723 transid = btrfs_super_generation(super);
3731 * Write superblock @sb to the @device. Do not wait for completion, all the
3732 * pages we use for writing are locked.
3734 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3735 * the expected device size at commit time. Note that max_mirrors must be
3736 * same for write and wait phases.
3738 * Return number of errors when page is not found or submission fails.
3740 static int write_dev_supers(struct btrfs_device *device,
3741 struct btrfs_super_block *sb, int max_mirrors)
3743 struct btrfs_fs_info *fs_info = device->fs_info;
3744 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3745 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3749 u64 bytenr, bytenr_orig;
3751 if (max_mirrors == 0)
3752 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3754 shash->tfm = fs_info->csum_shash;
3756 for (i = 0; i < max_mirrors; i++) {
3759 struct btrfs_super_block *disk_super;
3761 bytenr_orig = btrfs_sb_offset(i);
3762 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3763 if (ret == -ENOENT) {
3765 } else if (ret < 0) {
3766 btrfs_err(device->fs_info,
3767 "couldn't get super block location for mirror %d",
3772 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3773 device->commit_total_bytes)
3776 btrfs_set_super_bytenr(sb, bytenr_orig);
3778 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3779 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3782 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3785 btrfs_err(device->fs_info,
3786 "couldn't get super block page for bytenr %llu",
3792 /* Bump the refcount for wait_dev_supers() */
3795 disk_super = page_address(page);
3796 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3799 * Directly use bios here instead of relying on the page cache
3800 * to do I/O, so we don't lose the ability to do integrity
3803 bio = bio_alloc(device->bdev, 1,
3804 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3806 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3807 bio->bi_private = device;
3808 bio->bi_end_io = btrfs_end_super_write;
3809 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3810 offset_in_page(bytenr));
3813 * We FUA only the first super block. The others we allow to
3814 * go down lazy and there's a short window where the on-disk
3815 * copies might still contain the older version.
3817 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3818 bio->bi_opf |= REQ_FUA;
3821 if (btrfs_advance_sb_log(device, i))
3824 return errors < i ? 0 : -1;
3828 * Wait for write completion of superblocks done by write_dev_supers,
3829 * @max_mirrors same for write and wait phases.
3831 * Return number of errors when page is not found or not marked up to
3834 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3838 bool primary_failed = false;
3842 if (max_mirrors == 0)
3843 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3845 for (i = 0; i < max_mirrors; i++) {
3848 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3849 if (ret == -ENOENT) {
3851 } else if (ret < 0) {
3854 primary_failed = true;
3857 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3858 device->commit_total_bytes)
3861 page = find_get_page(device->bdev->bd_inode->i_mapping,
3862 bytenr >> PAGE_SHIFT);
3866 primary_failed = true;
3869 /* Page is submitted locked and unlocked once the IO completes */
3870 wait_on_page_locked(page);
3871 if (PageError(page)) {
3874 primary_failed = true;
3877 /* Drop our reference */
3880 /* Drop the reference from the writing run */
3884 /* log error, force error return */
3885 if (primary_failed) {
3886 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3891 return errors < i ? 0 : -1;
3895 * endio for the write_dev_flush, this will wake anyone waiting
3896 * for the barrier when it is done
3898 static void btrfs_end_empty_barrier(struct bio *bio)
3901 complete(bio->bi_private);
3905 * Submit a flush request to the device if it supports it. Error handling is
3906 * done in the waiting counterpart.
3908 static void write_dev_flush(struct btrfs_device *device)
3910 struct bio *bio = &device->flush_bio;
3912 device->last_flush_error = BLK_STS_OK;
3914 bio_init(bio, device->bdev, NULL, 0,
3915 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3916 bio->bi_end_io = btrfs_end_empty_barrier;
3917 init_completion(&device->flush_wait);
3918 bio->bi_private = &device->flush_wait;
3920 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3924 * If the flush bio has been submitted by write_dev_flush, wait for it.
3925 * Return true for any error, and false otherwise.
3927 static bool wait_dev_flush(struct btrfs_device *device)
3929 struct bio *bio = &device->flush_bio;
3931 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3934 wait_for_completion_io(&device->flush_wait);
3936 if (bio->bi_status) {
3937 device->last_flush_error = bio->bi_status;
3938 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3946 * send an empty flush down to each device in parallel,
3947 * then wait for them
3949 static int barrier_all_devices(struct btrfs_fs_info *info)
3951 struct list_head *head;
3952 struct btrfs_device *dev;
3953 int errors_wait = 0;
3955 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3956 /* send down all the barriers */
3957 head = &info->fs_devices->devices;
3958 list_for_each_entry(dev, head, dev_list) {
3959 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3963 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3964 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3967 write_dev_flush(dev);
3970 /* wait for all the barriers */
3971 list_for_each_entry(dev, head, dev_list) {
3972 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3978 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3979 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3982 if (wait_dev_flush(dev))
3987 * Checks last_flush_error of disks in order to determine the device
3990 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3996 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3999 int min_tolerated = INT_MAX;
4001 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4002 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4003 min_tolerated = min_t(int, min_tolerated,
4004 btrfs_raid_array[BTRFS_RAID_SINGLE].
4005 tolerated_failures);
4007 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4008 if (raid_type == BTRFS_RAID_SINGLE)
4010 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4012 min_tolerated = min_t(int, min_tolerated,
4013 btrfs_raid_array[raid_type].
4014 tolerated_failures);
4017 if (min_tolerated == INT_MAX) {
4018 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4022 return min_tolerated;
4025 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4027 struct list_head *head;
4028 struct btrfs_device *dev;
4029 struct btrfs_super_block *sb;
4030 struct btrfs_dev_item *dev_item;
4034 int total_errors = 0;
4037 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4040 * max_mirrors == 0 indicates we're from commit_transaction,
4041 * not from fsync where the tree roots in fs_info have not
4042 * been consistent on disk.
4044 if (max_mirrors == 0)
4045 backup_super_roots(fs_info);
4047 sb = fs_info->super_for_commit;
4048 dev_item = &sb->dev_item;
4050 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4051 head = &fs_info->fs_devices->devices;
4052 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4055 ret = barrier_all_devices(fs_info);
4058 &fs_info->fs_devices->device_list_mutex);
4059 btrfs_handle_fs_error(fs_info, ret,
4060 "errors while submitting device barriers.");
4065 list_for_each_entry(dev, head, dev_list) {
4070 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4071 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4074 btrfs_set_stack_device_generation(dev_item, 0);
4075 btrfs_set_stack_device_type(dev_item, dev->type);
4076 btrfs_set_stack_device_id(dev_item, dev->devid);
4077 btrfs_set_stack_device_total_bytes(dev_item,
4078 dev->commit_total_bytes);
4079 btrfs_set_stack_device_bytes_used(dev_item,
4080 dev->commit_bytes_used);
4081 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4082 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4083 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4084 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4085 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4088 flags = btrfs_super_flags(sb);
4089 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4091 ret = btrfs_validate_write_super(fs_info, sb);
4093 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4094 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4095 "unexpected superblock corruption detected");
4099 ret = write_dev_supers(dev, sb, max_mirrors);
4103 if (total_errors > max_errors) {
4104 btrfs_err(fs_info, "%d errors while writing supers",
4106 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4108 /* FUA is masked off if unsupported and can't be the reason */
4109 btrfs_handle_fs_error(fs_info, -EIO,
4110 "%d errors while writing supers",
4116 list_for_each_entry(dev, head, dev_list) {
4119 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4120 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4123 ret = wait_dev_supers(dev, max_mirrors);
4127 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4128 if (total_errors > max_errors) {
4129 btrfs_handle_fs_error(fs_info, -EIO,
4130 "%d errors while writing supers",
4137 /* Drop a fs root from the radix tree and free it. */
4138 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4139 struct btrfs_root *root)
4141 bool drop_ref = false;
4143 spin_lock(&fs_info->fs_roots_radix_lock);
4144 radix_tree_delete(&fs_info->fs_roots_radix,
4145 (unsigned long)root->root_key.objectid);
4146 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4148 spin_unlock(&fs_info->fs_roots_radix_lock);
4150 if (BTRFS_FS_ERROR(fs_info)) {
4151 ASSERT(root->log_root == NULL);
4152 if (root->reloc_root) {
4153 btrfs_put_root(root->reloc_root);
4154 root->reloc_root = NULL;
4159 btrfs_put_root(root);
4162 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4164 struct btrfs_root *root = fs_info->tree_root;
4165 struct btrfs_trans_handle *trans;
4167 mutex_lock(&fs_info->cleaner_mutex);
4168 btrfs_run_delayed_iputs(fs_info);
4169 mutex_unlock(&fs_info->cleaner_mutex);
4170 wake_up_process(fs_info->cleaner_kthread);
4172 /* wait until ongoing cleanup work done */
4173 down_write(&fs_info->cleanup_work_sem);
4174 up_write(&fs_info->cleanup_work_sem);
4176 trans = btrfs_join_transaction(root);
4178 return PTR_ERR(trans);
4179 return btrfs_commit_transaction(trans);
4182 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4184 struct btrfs_transaction *trans;
4185 struct btrfs_transaction *tmp;
4188 if (list_empty(&fs_info->trans_list))
4192 * This function is only called at the very end of close_ctree(),
4193 * thus no other running transaction, no need to take trans_lock.
4195 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4196 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4197 struct extent_state *cached = NULL;
4198 u64 dirty_bytes = 0;
4204 while (find_first_extent_bit(&trans->dirty_pages, cur,
4205 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4206 dirty_bytes += found_end + 1 - found_start;
4207 cur = found_end + 1;
4210 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4211 trans->transid, dirty_bytes);
4212 btrfs_cleanup_one_transaction(trans, fs_info);
4214 if (trans == fs_info->running_transaction)
4215 fs_info->running_transaction = NULL;
4216 list_del_init(&trans->list);
4218 btrfs_put_transaction(trans);
4219 trace_btrfs_transaction_commit(fs_info);
4224 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4228 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4231 * If we had UNFINISHED_DROPS we could still be processing them, so
4232 * clear that bit and wake up relocation so it can stop.
4233 * We must do this before stopping the block group reclaim task, because
4234 * at btrfs_relocate_block_group() we wait for this bit, and after the
4235 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4236 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4239 btrfs_wake_unfinished_drop(fs_info);
4242 * We may have the reclaim task running and relocating a data block group,
4243 * in which case it may create delayed iputs. So stop it before we park
4244 * the cleaner kthread otherwise we can get new delayed iputs after
4245 * parking the cleaner, and that can make the async reclaim task to hang
4246 * if it's waiting for delayed iputs to complete, since the cleaner is
4247 * parked and can not run delayed iputs - this will make us hang when
4248 * trying to stop the async reclaim task.
4250 cancel_work_sync(&fs_info->reclaim_bgs_work);
4252 * We don't want the cleaner to start new transactions, add more delayed
4253 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4254 * because that frees the task_struct, and the transaction kthread might
4255 * still try to wake up the cleaner.
4257 kthread_park(fs_info->cleaner_kthread);
4259 /* wait for the qgroup rescan worker to stop */
4260 btrfs_qgroup_wait_for_completion(fs_info, false);
4262 /* wait for the uuid_scan task to finish */
4263 down(&fs_info->uuid_tree_rescan_sem);
4264 /* avoid complains from lockdep et al., set sem back to initial state */
4265 up(&fs_info->uuid_tree_rescan_sem);
4267 /* pause restriper - we want to resume on mount */
4268 btrfs_pause_balance(fs_info);
4270 btrfs_dev_replace_suspend_for_unmount(fs_info);
4272 btrfs_scrub_cancel(fs_info);
4274 /* wait for any defraggers to finish */
4275 wait_event(fs_info->transaction_wait,
4276 (atomic_read(&fs_info->defrag_running) == 0));
4278 /* clear out the rbtree of defraggable inodes */
4279 btrfs_cleanup_defrag_inodes(fs_info);
4282 * After we parked the cleaner kthread, ordered extents may have
4283 * completed and created new delayed iputs. If one of the async reclaim
4284 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4285 * can hang forever trying to stop it, because if a delayed iput is
4286 * added after it ran btrfs_run_delayed_iputs() and before it called
4287 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4288 * no one else to run iputs.
4290 * So wait for all ongoing ordered extents to complete and then run
4291 * delayed iputs. This works because once we reach this point no one
4292 * can either create new ordered extents nor create delayed iputs
4293 * through some other means.
4295 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4296 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4297 * but the delayed iput for the respective inode is made only when doing
4298 * the final btrfs_put_ordered_extent() (which must happen at
4299 * btrfs_finish_ordered_io() when we are unmounting).
4301 btrfs_flush_workqueue(fs_info->endio_write_workers);
4302 /* Ordered extents for free space inodes. */
4303 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4304 btrfs_run_delayed_iputs(fs_info);
4306 cancel_work_sync(&fs_info->async_reclaim_work);
4307 cancel_work_sync(&fs_info->async_data_reclaim_work);
4308 cancel_work_sync(&fs_info->preempt_reclaim_work);
4310 /* Cancel or finish ongoing discard work */
4311 btrfs_discard_cleanup(fs_info);
4313 if (!sb_rdonly(fs_info->sb)) {
4315 * The cleaner kthread is stopped, so do one final pass over
4316 * unused block groups.
4318 btrfs_delete_unused_bgs(fs_info);
4321 * There might be existing delayed inode workers still running
4322 * and holding an empty delayed inode item. We must wait for
4323 * them to complete first because they can create a transaction.
4324 * This happens when someone calls btrfs_balance_delayed_items()
4325 * and then a transaction commit runs the same delayed nodes
4326 * before any delayed worker has done something with the nodes.
4327 * We must wait for any worker here and not at transaction
4328 * commit time since that could cause a deadlock.
4329 * This is a very rare case.
4331 btrfs_flush_workqueue(fs_info->delayed_workers);
4333 ret = btrfs_commit_super(fs_info);
4335 btrfs_err(fs_info, "commit super ret %d", ret);
4338 if (BTRFS_FS_ERROR(fs_info))
4339 btrfs_error_commit_super(fs_info);
4341 kthread_stop(fs_info->transaction_kthread);
4342 kthread_stop(fs_info->cleaner_kthread);
4344 ASSERT(list_empty(&fs_info->delayed_iputs));
4345 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4347 if (btrfs_check_quota_leak(fs_info)) {
4348 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4349 btrfs_err(fs_info, "qgroup reserved space leaked");
4352 btrfs_free_qgroup_config(fs_info);
4353 ASSERT(list_empty(&fs_info->delalloc_roots));
4355 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4356 btrfs_info(fs_info, "at unmount delalloc count %lld",
4357 percpu_counter_sum(&fs_info->delalloc_bytes));
4360 if (percpu_counter_sum(&fs_info->ordered_bytes))
4361 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4362 percpu_counter_sum(&fs_info->ordered_bytes));
4364 btrfs_sysfs_remove_mounted(fs_info);
4365 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4367 btrfs_put_block_group_cache(fs_info);
4370 * we must make sure there is not any read request to
4371 * submit after we stopping all workers.
4373 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4374 btrfs_stop_all_workers(fs_info);
4376 /* We shouldn't have any transaction open at this point */
4377 warn_about_uncommitted_trans(fs_info);
4379 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4380 free_root_pointers(fs_info, true);
4381 btrfs_free_fs_roots(fs_info);
4384 * We must free the block groups after dropping the fs_roots as we could
4385 * have had an IO error and have left over tree log blocks that aren't
4386 * cleaned up until the fs roots are freed. This makes the block group
4387 * accounting appear to be wrong because there's pending reserved bytes,
4388 * so make sure we do the block group cleanup afterwards.
4390 btrfs_free_block_groups(fs_info);
4392 iput(fs_info->btree_inode);
4394 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4395 btrfs_close_devices(fs_info->fs_devices);
4398 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4399 struct extent_buffer *buf)
4401 struct btrfs_fs_info *fs_info = buf->fs_info;
4402 u64 transid = btrfs_header_generation(buf);
4404 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4406 * This is a fast path so only do this check if we have sanity tests
4407 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4408 * outside of the sanity tests.
4410 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4413 /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4414 ASSERT(trans->transid == fs_info->generation);
4415 btrfs_assert_tree_write_locked(buf);
4416 if (unlikely(transid != fs_info->generation)) {
4417 btrfs_abort_transaction(trans, -EUCLEAN);
4419 "dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4420 buf->start, transid, fs_info->generation);
4422 set_extent_buffer_dirty(buf);
4425 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4429 * looks as though older kernels can get into trouble with
4430 * this code, they end up stuck in balance_dirty_pages forever
4434 if (current->flags & PF_MEMALLOC)
4438 btrfs_balance_delayed_items(fs_info);
4440 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4441 BTRFS_DIRTY_METADATA_THRESH,
4442 fs_info->dirty_metadata_batch);
4444 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4448 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4450 __btrfs_btree_balance_dirty(fs_info, 1);
4453 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4455 __btrfs_btree_balance_dirty(fs_info, 0);
4458 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4460 /* cleanup FS via transaction */
4461 btrfs_cleanup_transaction(fs_info);
4463 mutex_lock(&fs_info->cleaner_mutex);
4464 btrfs_run_delayed_iputs(fs_info);
4465 mutex_unlock(&fs_info->cleaner_mutex);
4467 down_write(&fs_info->cleanup_work_sem);
4468 up_write(&fs_info->cleanup_work_sem);
4471 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4473 struct btrfs_root *gang[8];
4474 u64 root_objectid = 0;
4477 spin_lock(&fs_info->fs_roots_radix_lock);
4478 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4479 (void **)gang, root_objectid,
4480 ARRAY_SIZE(gang))) != 0) {
4483 for (i = 0; i < ret; i++)
4484 gang[i] = btrfs_grab_root(gang[i]);
4485 spin_unlock(&fs_info->fs_roots_radix_lock);
4487 for (i = 0; i < ret; i++) {
4490 root_objectid = gang[i]->root_key.objectid;
4491 btrfs_free_log(NULL, gang[i]);
4492 btrfs_put_root(gang[i]);
4495 spin_lock(&fs_info->fs_roots_radix_lock);
4497 spin_unlock(&fs_info->fs_roots_radix_lock);
4498 btrfs_free_log_root_tree(NULL, fs_info);
4501 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4503 struct btrfs_ordered_extent *ordered;
4505 spin_lock(&root->ordered_extent_lock);
4507 * This will just short circuit the ordered completion stuff which will
4508 * make sure the ordered extent gets properly cleaned up.
4510 list_for_each_entry(ordered, &root->ordered_extents,
4512 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4513 spin_unlock(&root->ordered_extent_lock);
4516 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4518 struct btrfs_root *root;
4521 spin_lock(&fs_info->ordered_root_lock);
4522 list_splice_init(&fs_info->ordered_roots, &splice);
4523 while (!list_empty(&splice)) {
4524 root = list_first_entry(&splice, struct btrfs_root,
4526 list_move_tail(&root->ordered_root,
4527 &fs_info->ordered_roots);
4529 spin_unlock(&fs_info->ordered_root_lock);
4530 btrfs_destroy_ordered_extents(root);
4533 spin_lock(&fs_info->ordered_root_lock);
4535 spin_unlock(&fs_info->ordered_root_lock);
4538 * We need this here because if we've been flipped read-only we won't
4539 * get sync() from the umount, so we need to make sure any ordered
4540 * extents that haven't had their dirty pages IO start writeout yet
4541 * actually get run and error out properly.
4543 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4546 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4547 struct btrfs_fs_info *fs_info)
4549 struct rb_node *node;
4550 struct btrfs_delayed_ref_root *delayed_refs;
4551 struct btrfs_delayed_ref_node *ref;
4553 delayed_refs = &trans->delayed_refs;
4555 spin_lock(&delayed_refs->lock);
4556 if (atomic_read(&delayed_refs->num_entries) == 0) {
4557 spin_unlock(&delayed_refs->lock);
4558 btrfs_debug(fs_info, "delayed_refs has NO entry");
4562 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4563 struct btrfs_delayed_ref_head *head;
4565 bool pin_bytes = false;
4567 head = rb_entry(node, struct btrfs_delayed_ref_head,
4569 if (btrfs_delayed_ref_lock(delayed_refs, head))
4572 spin_lock(&head->lock);
4573 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4574 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4576 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4577 RB_CLEAR_NODE(&ref->ref_node);
4578 if (!list_empty(&ref->add_list))
4579 list_del(&ref->add_list);
4580 atomic_dec(&delayed_refs->num_entries);
4581 btrfs_put_delayed_ref(ref);
4582 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4584 if (head->must_insert_reserved)
4586 btrfs_free_delayed_extent_op(head->extent_op);
4587 btrfs_delete_ref_head(delayed_refs, head);
4588 spin_unlock(&head->lock);
4589 spin_unlock(&delayed_refs->lock);
4590 mutex_unlock(&head->mutex);
4593 struct btrfs_block_group *cache;
4595 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4598 spin_lock(&cache->space_info->lock);
4599 spin_lock(&cache->lock);
4600 cache->pinned += head->num_bytes;
4601 btrfs_space_info_update_bytes_pinned(fs_info,
4602 cache->space_info, head->num_bytes);
4603 cache->reserved -= head->num_bytes;
4604 cache->space_info->bytes_reserved -= head->num_bytes;
4605 spin_unlock(&cache->lock);
4606 spin_unlock(&cache->space_info->lock);
4608 btrfs_put_block_group(cache);
4610 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4611 head->bytenr + head->num_bytes - 1);
4613 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4614 btrfs_put_delayed_ref_head(head);
4616 spin_lock(&delayed_refs->lock);
4618 btrfs_qgroup_destroy_extent_records(trans);
4620 spin_unlock(&delayed_refs->lock);
4623 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4625 struct btrfs_inode *btrfs_inode;
4628 spin_lock(&root->delalloc_lock);
4629 list_splice_init(&root->delalloc_inodes, &splice);
4631 while (!list_empty(&splice)) {
4632 struct inode *inode = NULL;
4633 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4635 __btrfs_del_delalloc_inode(root, btrfs_inode);
4636 spin_unlock(&root->delalloc_lock);
4639 * Make sure we get a live inode and that it'll not disappear
4642 inode = igrab(&btrfs_inode->vfs_inode);
4644 unsigned int nofs_flag;
4646 nofs_flag = memalloc_nofs_save();
4647 invalidate_inode_pages2(inode->i_mapping);
4648 memalloc_nofs_restore(nofs_flag);
4651 spin_lock(&root->delalloc_lock);
4653 spin_unlock(&root->delalloc_lock);
4656 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4658 struct btrfs_root *root;
4661 spin_lock(&fs_info->delalloc_root_lock);
4662 list_splice_init(&fs_info->delalloc_roots, &splice);
4663 while (!list_empty(&splice)) {
4664 root = list_first_entry(&splice, struct btrfs_root,
4666 root = btrfs_grab_root(root);
4668 spin_unlock(&fs_info->delalloc_root_lock);
4670 btrfs_destroy_delalloc_inodes(root);
4671 btrfs_put_root(root);
4673 spin_lock(&fs_info->delalloc_root_lock);
4675 spin_unlock(&fs_info->delalloc_root_lock);
4678 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4679 struct extent_io_tree *dirty_pages,
4682 struct extent_buffer *eb;
4686 while (find_first_extent_bit(dirty_pages, start, &start, &end,
4688 clear_extent_bits(dirty_pages, start, end, mark);
4689 while (start <= end) {
4690 eb = find_extent_buffer(fs_info, start);
4691 start += fs_info->nodesize;
4695 btrfs_tree_lock(eb);
4696 wait_on_extent_buffer_writeback(eb);
4697 btrfs_clear_buffer_dirty(NULL, eb);
4698 btrfs_tree_unlock(eb);
4700 free_extent_buffer_stale(eb);
4705 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4706 struct extent_io_tree *unpin)
4712 struct extent_state *cached_state = NULL;
4715 * The btrfs_finish_extent_commit() may get the same range as
4716 * ours between find_first_extent_bit and clear_extent_dirty.
4717 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4718 * the same extent range.
4720 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4721 if (!find_first_extent_bit(unpin, 0, &start, &end,
4722 EXTENT_DIRTY, &cached_state)) {
4723 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4727 clear_extent_dirty(unpin, start, end, &cached_state);
4728 free_extent_state(cached_state);
4729 btrfs_error_unpin_extent_range(fs_info, start, end);
4730 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4735 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4737 struct inode *inode;
4739 inode = cache->io_ctl.inode;
4741 unsigned int nofs_flag;
4743 nofs_flag = memalloc_nofs_save();
4744 invalidate_inode_pages2(inode->i_mapping);
4745 memalloc_nofs_restore(nofs_flag);
4747 BTRFS_I(inode)->generation = 0;
4748 cache->io_ctl.inode = NULL;
4751 ASSERT(cache->io_ctl.pages == NULL);
4752 btrfs_put_block_group(cache);
4755 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4756 struct btrfs_fs_info *fs_info)
4758 struct btrfs_block_group *cache;
4760 spin_lock(&cur_trans->dirty_bgs_lock);
4761 while (!list_empty(&cur_trans->dirty_bgs)) {
4762 cache = list_first_entry(&cur_trans->dirty_bgs,
4763 struct btrfs_block_group,
4766 if (!list_empty(&cache->io_list)) {
4767 spin_unlock(&cur_trans->dirty_bgs_lock);
4768 list_del_init(&cache->io_list);
4769 btrfs_cleanup_bg_io(cache);
4770 spin_lock(&cur_trans->dirty_bgs_lock);
4773 list_del_init(&cache->dirty_list);
4774 spin_lock(&cache->lock);
4775 cache->disk_cache_state = BTRFS_DC_ERROR;
4776 spin_unlock(&cache->lock);
4778 spin_unlock(&cur_trans->dirty_bgs_lock);
4779 btrfs_put_block_group(cache);
4780 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4781 spin_lock(&cur_trans->dirty_bgs_lock);
4783 spin_unlock(&cur_trans->dirty_bgs_lock);
4786 * Refer to the definition of io_bgs member for details why it's safe
4787 * to use it without any locking
4789 while (!list_empty(&cur_trans->io_bgs)) {
4790 cache = list_first_entry(&cur_trans->io_bgs,
4791 struct btrfs_block_group,
4794 list_del_init(&cache->io_list);
4795 spin_lock(&cache->lock);
4796 cache->disk_cache_state = BTRFS_DC_ERROR;
4797 spin_unlock(&cache->lock);
4798 btrfs_cleanup_bg_io(cache);
4802 static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4804 struct btrfs_root *gang[8];
4808 spin_lock(&fs_info->fs_roots_radix_lock);
4810 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4813 BTRFS_ROOT_TRANS_TAG);
4816 for (i = 0; i < ret; i++) {
4817 struct btrfs_root *root = gang[i];
4819 btrfs_qgroup_free_meta_all_pertrans(root);
4820 radix_tree_tag_clear(&fs_info->fs_roots_radix,
4821 (unsigned long)root->root_key.objectid,
4822 BTRFS_ROOT_TRANS_TAG);
4825 spin_unlock(&fs_info->fs_roots_radix_lock);
4828 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4829 struct btrfs_fs_info *fs_info)
4831 struct btrfs_device *dev, *tmp;
4833 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4834 ASSERT(list_empty(&cur_trans->dirty_bgs));
4835 ASSERT(list_empty(&cur_trans->io_bgs));
4837 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4839 list_del_init(&dev->post_commit_list);
4842 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4844 cur_trans->state = TRANS_STATE_COMMIT_START;
4845 wake_up(&fs_info->transaction_blocked_wait);
4847 cur_trans->state = TRANS_STATE_UNBLOCKED;
4848 wake_up(&fs_info->transaction_wait);
4850 btrfs_destroy_delayed_inodes(fs_info);
4852 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4854 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4856 btrfs_free_all_qgroup_pertrans(fs_info);
4858 cur_trans->state =TRANS_STATE_COMPLETED;
4859 wake_up(&cur_trans->commit_wait);
4862 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4864 struct btrfs_transaction *t;
4866 mutex_lock(&fs_info->transaction_kthread_mutex);
4868 spin_lock(&fs_info->trans_lock);
4869 while (!list_empty(&fs_info->trans_list)) {
4870 t = list_first_entry(&fs_info->trans_list,
4871 struct btrfs_transaction, list);
4872 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4873 refcount_inc(&t->use_count);
4874 spin_unlock(&fs_info->trans_lock);
4875 btrfs_wait_for_commit(fs_info, t->transid);
4876 btrfs_put_transaction(t);
4877 spin_lock(&fs_info->trans_lock);
4880 if (t == fs_info->running_transaction) {
4881 t->state = TRANS_STATE_COMMIT_DOING;
4882 spin_unlock(&fs_info->trans_lock);
4884 * We wait for 0 num_writers since we don't hold a trans
4885 * handle open currently for this transaction.
4887 wait_event(t->writer_wait,
4888 atomic_read(&t->num_writers) == 0);
4890 spin_unlock(&fs_info->trans_lock);
4892 btrfs_cleanup_one_transaction(t, fs_info);
4894 spin_lock(&fs_info->trans_lock);
4895 if (t == fs_info->running_transaction)
4896 fs_info->running_transaction = NULL;
4897 list_del_init(&t->list);
4898 spin_unlock(&fs_info->trans_lock);
4900 btrfs_put_transaction(t);
4901 trace_btrfs_transaction_commit(fs_info);
4902 spin_lock(&fs_info->trans_lock);
4904 spin_unlock(&fs_info->trans_lock);
4905 btrfs_destroy_all_ordered_extents(fs_info);
4906 btrfs_destroy_delayed_inodes(fs_info);
4907 btrfs_assert_delayed_root_empty(fs_info);
4908 btrfs_destroy_all_delalloc_inodes(fs_info);
4909 btrfs_drop_all_logs(fs_info);
4910 mutex_unlock(&fs_info->transaction_kthread_mutex);
4915 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4917 struct btrfs_path *path;
4919 struct extent_buffer *l;
4920 struct btrfs_key search_key;
4921 struct btrfs_key found_key;
4924 path = btrfs_alloc_path();
4928 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4929 search_key.type = -1;
4930 search_key.offset = (u64)-1;
4931 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4934 BUG_ON(ret == 0); /* Corruption */
4935 if (path->slots[0] > 0) {
4936 slot = path->slots[0] - 1;
4938 btrfs_item_key_to_cpu(l, &found_key, slot);
4939 root->free_objectid = max_t(u64, found_key.objectid + 1,
4940 BTRFS_FIRST_FREE_OBJECTID);
4942 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4946 btrfs_free_path(path);
4950 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4953 mutex_lock(&root->objectid_mutex);
4955 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4956 btrfs_warn(root->fs_info,
4957 "the objectid of root %llu reaches its highest value",
4958 root->root_key.objectid);
4963 *objectid = root->free_objectid++;
4966 mutex_unlock(&root->objectid_mutex);