1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41 #include "block-group.h"
43 #include "space-info.h"
47 #include "accessors.h"
48 #include "extent-tree.h"
49 #include "root-tree.h"
51 #include "uuid-tree.h"
52 #include "relocation.h"
56 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
57 BTRFS_HEADER_FLAG_RELOC |\
58 BTRFS_SUPER_FLAG_ERROR |\
59 BTRFS_SUPER_FLAG_SEEDING |\
60 BTRFS_SUPER_FLAG_METADUMP |\
61 BTRFS_SUPER_FLAG_METADUMP_V2)
63 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_fs_info *fs_info);
66 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
67 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
68 struct extent_io_tree *dirty_pages,
70 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
71 struct extent_io_tree *pinned_extents);
72 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
73 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
75 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
77 if (fs_info->csum_shash)
78 crypto_free_shash(fs_info->csum_shash);
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
86 struct async_submit_bio {
87 struct btrfs_inode *inode;
89 enum btrfs_wq_submit_cmd submit_cmd;
92 /* Optional parameter for used by direct io */
94 struct btrfs_work work;
99 * Compute the csum of a btree block and store the result to provided buffer.
101 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
103 struct btrfs_fs_info *fs_info = buf->fs_info;
104 const int num_pages = num_extent_pages(buf);
105 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
106 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
110 shash->tfm = fs_info->csum_shash;
111 crypto_shash_init(shash);
112 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
113 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
114 first_page_part - BTRFS_CSUM_SIZE);
116 for (i = 1; i < num_pages; i++) {
117 kaddr = page_address(buf->pages[i]);
118 crypto_shash_update(shash, kaddr, PAGE_SIZE);
120 memset(result, 0, BTRFS_CSUM_SIZE);
121 crypto_shash_final(shash, result);
125 * we can't consider a given block up to date unless the transid of the
126 * block matches the transid in the parent node's pointer. This is how we
127 * detect blocks that either didn't get written at all or got written
128 * in the wrong place.
130 static int verify_parent_transid(struct extent_io_tree *io_tree,
131 struct extent_buffer *eb, u64 parent_transid,
134 struct extent_state *cached_state = NULL;
137 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
143 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state);
144 if (extent_buffer_uptodate(eb) &&
145 btrfs_header_generation(eb) == parent_transid) {
149 btrfs_err_rl(eb->fs_info,
150 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
151 eb->start, eb->read_mirror,
152 parent_transid, btrfs_header_generation(eb));
154 clear_extent_buffer_uptodate(eb);
156 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
161 static bool btrfs_supported_super_csum(u16 csum_type)
164 case BTRFS_CSUM_TYPE_CRC32:
165 case BTRFS_CSUM_TYPE_XXHASH:
166 case BTRFS_CSUM_TYPE_SHA256:
167 case BTRFS_CSUM_TYPE_BLAKE2:
175 * Return 0 if the superblock checksum type matches the checksum value of that
176 * algorithm. Pass the raw disk superblock data.
178 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
179 const struct btrfs_super_block *disk_sb)
181 char result[BTRFS_CSUM_SIZE];
182 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
184 shash->tfm = fs_info->csum_shash;
187 * The super_block structure does not span the whole
188 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
189 * filled with zeros and is included in the checksum.
191 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
192 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
194 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
200 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
201 struct btrfs_key *first_key, u64 parent_transid)
203 struct btrfs_fs_info *fs_info = eb->fs_info;
205 struct btrfs_key found_key;
208 found_level = btrfs_header_level(eb);
209 if (found_level != level) {
210 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
211 KERN_ERR "BTRFS: tree level check failed\n");
213 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
214 eb->start, level, found_level);
222 * For live tree block (new tree blocks in current transaction),
223 * we need proper lock context to avoid race, which is impossible here.
224 * So we only checks tree blocks which is read from disk, whose
225 * generation <= fs_info->last_trans_committed.
227 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
230 /* We have @first_key, so this @eb must have at least one item */
231 if (btrfs_header_nritems(eb) == 0) {
233 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
235 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
240 btrfs_node_key_to_cpu(eb, &found_key, 0);
242 btrfs_item_key_to_cpu(eb, &found_key, 0);
243 ret = btrfs_comp_cpu_keys(first_key, &found_key);
246 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
247 KERN_ERR "BTRFS: tree first key check failed\n");
249 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
250 eb->start, parent_transid, first_key->objectid,
251 first_key->type, first_key->offset,
252 found_key.objectid, found_key.type,
258 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
261 struct btrfs_fs_info *fs_info = eb->fs_info;
262 u64 start = eb->start;
263 int i, num_pages = num_extent_pages(eb);
266 if (sb_rdonly(fs_info->sb))
269 for (i = 0; i < num_pages; i++) {
270 struct page *p = eb->pages[i];
272 ret = btrfs_repair_io_failure(fs_info, 0, start, PAGE_SIZE,
273 start, p, start - page_offset(p), mirror_num);
283 * helper to read a given tree block, doing retries as required when
284 * the checksums don't match and we have alternate mirrors to try.
286 * @check: expected tree parentness check, see the comments of the
287 * structure for details.
289 int btrfs_read_extent_buffer(struct extent_buffer *eb,
290 struct btrfs_tree_parent_check *check)
292 struct btrfs_fs_info *fs_info = eb->fs_info;
297 int failed_mirror = 0;
302 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
303 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
307 num_copies = btrfs_num_copies(fs_info,
312 if (!failed_mirror) {
314 failed_mirror = eb->read_mirror;
318 if (mirror_num == failed_mirror)
321 if (mirror_num > num_copies)
325 if (failed && !ret && failed_mirror)
326 btrfs_repair_eb_io_failure(eb, failed_mirror);
331 static int csum_one_extent_buffer(struct extent_buffer *eb)
333 struct btrfs_fs_info *fs_info = eb->fs_info;
334 u8 result[BTRFS_CSUM_SIZE];
337 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
338 offsetof(struct btrfs_header, fsid),
339 BTRFS_FSID_SIZE) == 0);
340 csum_tree_block(eb, result);
342 if (btrfs_header_level(eb))
343 ret = btrfs_check_node(eb);
345 ret = btrfs_check_leaf_full(eb);
351 * Also check the generation, the eb reached here must be newer than
352 * last committed. Or something seriously wrong happened.
354 if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
357 "block=%llu bad generation, have %llu expect > %llu",
358 eb->start, btrfs_header_generation(eb),
359 fs_info->last_trans_committed);
362 write_extent_buffer(eb, result, 0, fs_info->csum_size);
367 btrfs_print_tree(eb, 0);
368 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
370 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
374 /* Checksum all dirty extent buffers in one bio_vec */
375 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
376 struct bio_vec *bvec)
378 struct page *page = bvec->bv_page;
379 u64 bvec_start = page_offset(page) + bvec->bv_offset;
383 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
384 cur += fs_info->nodesize) {
385 struct extent_buffer *eb;
388 eb = find_extent_buffer(fs_info, cur);
389 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
392 /* A dirty eb shouldn't disappear from buffer_radix */
396 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
397 free_extent_buffer(eb);
400 if (WARN_ON(!uptodate)) {
401 free_extent_buffer(eb);
405 ret = csum_one_extent_buffer(eb);
406 free_extent_buffer(eb);
414 * Checksum a dirty tree block before IO. This has extra checks to make sure
415 * we only fill in the checksum field in the first page of a multi-page block.
416 * For subpage extent buffers we need bvec to also read the offset in the page.
418 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
420 struct page *page = bvec->bv_page;
421 u64 start = page_offset(page);
423 struct extent_buffer *eb;
425 if (fs_info->nodesize < PAGE_SIZE)
426 return csum_dirty_subpage_buffers(fs_info, bvec);
428 eb = (struct extent_buffer *)page->private;
429 if (page != eb->pages[0])
432 found_start = btrfs_header_bytenr(eb);
434 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
435 WARN_ON(found_start != 0);
440 * Please do not consolidate these warnings into a single if.
441 * It is useful to know what went wrong.
443 if (WARN_ON(found_start != start))
445 if (WARN_ON(!PageUptodate(page)))
448 return csum_one_extent_buffer(eb);
451 static int check_tree_block_fsid(struct extent_buffer *eb)
453 struct btrfs_fs_info *fs_info = eb->fs_info;
454 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
455 u8 fsid[BTRFS_FSID_SIZE];
458 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
461 * Checking the incompat flag is only valid for the current fs. For
462 * seed devices it's forbidden to have their uuid changed so reading
463 * ->fsid in this case is fine
465 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
466 metadata_uuid = fs_devices->metadata_uuid;
468 metadata_uuid = fs_devices->fsid;
470 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
473 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
474 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
480 /* Do basic extent buffer checks at read time */
481 static int validate_extent_buffer(struct extent_buffer *eb,
482 struct btrfs_tree_parent_check *check)
484 struct btrfs_fs_info *fs_info = eb->fs_info;
486 const u32 csum_size = fs_info->csum_size;
488 u8 result[BTRFS_CSUM_SIZE];
489 const u8 *header_csum;
494 found_start = btrfs_header_bytenr(eb);
495 if (found_start != eb->start) {
496 btrfs_err_rl(fs_info,
497 "bad tree block start, mirror %u want %llu have %llu",
498 eb->read_mirror, eb->start, found_start);
502 if (check_tree_block_fsid(eb)) {
503 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
504 eb->start, eb->read_mirror);
508 found_level = btrfs_header_level(eb);
509 if (found_level >= BTRFS_MAX_LEVEL) {
511 "bad tree block level, mirror %u level %d on logical %llu",
512 eb->read_mirror, btrfs_header_level(eb), eb->start);
517 csum_tree_block(eb, result);
518 header_csum = page_address(eb->pages[0]) +
519 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
521 if (memcmp(result, header_csum, csum_size) != 0) {
522 btrfs_warn_rl(fs_info,
523 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
524 eb->start, eb->read_mirror,
525 CSUM_FMT_VALUE(csum_size, header_csum),
526 CSUM_FMT_VALUE(csum_size, result),
527 btrfs_header_level(eb));
532 if (found_level != check->level) {
534 "level verify failed on logical %llu mirror %u wanted %u found %u",
535 eb->start, eb->read_mirror, check->level, found_level);
539 if (unlikely(check->transid &&
540 btrfs_header_generation(eb) != check->transid)) {
541 btrfs_err_rl(eb->fs_info,
542 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
543 eb->start, eb->read_mirror, check->transid,
544 btrfs_header_generation(eb));
548 if (check->has_first_key) {
549 struct btrfs_key *expect_key = &check->first_key;
550 struct btrfs_key found_key;
553 btrfs_node_key_to_cpu(eb, &found_key, 0);
555 btrfs_item_key_to_cpu(eb, &found_key, 0);
556 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
558 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
559 eb->start, check->transid,
560 expect_key->objectid,
561 expect_key->type, expect_key->offset,
562 found_key.objectid, found_key.type,
568 if (check->owner_root) {
569 ret = btrfs_check_eb_owner(eb, check->owner_root);
575 * If this is a leaf block and it is corrupt, set the corrupt bit so
576 * that we don't try and read the other copies of this block, just
579 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
580 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
584 if (found_level > 0 && btrfs_check_node(eb))
588 set_extent_buffer_uptodate(eb);
591 "read time tree block corruption detected on logical %llu mirror %u",
592 eb->start, eb->read_mirror);
597 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
598 int mirror, struct btrfs_tree_parent_check *check)
600 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
601 struct extent_buffer *eb;
608 * We don't allow bio merge for subpage metadata read, so we should
609 * only get one eb for each endio hook.
611 ASSERT(end == start + fs_info->nodesize - 1);
612 ASSERT(PagePrivate(page));
614 eb = find_extent_buffer(fs_info, start);
616 * When we are reading one tree block, eb must have been inserted into
617 * the radix tree. If not, something is wrong.
621 reads_done = atomic_dec_and_test(&eb->io_pages);
622 /* Subpage read must finish in page read */
625 eb->read_mirror = mirror;
626 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
630 ret = validate_extent_buffer(eb, check);
634 set_extent_buffer_uptodate(eb);
636 free_extent_buffer(eb);
640 * end_bio_extent_readpage decrements io_pages in case of error,
641 * make sure it has something to decrement.
643 atomic_inc(&eb->io_pages);
644 clear_extent_buffer_uptodate(eb);
645 free_extent_buffer(eb);
649 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
650 struct page *page, u64 start, u64 end,
653 struct extent_buffer *eb;
657 ASSERT(page->private);
659 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
660 return validate_subpage_buffer(page, start, end, mirror,
661 &bbio->parent_check);
663 eb = (struct extent_buffer *)page->private;
666 * The pending IO might have been the only thing that kept this buffer
667 * in memory. Make sure we have a ref for all this other checks
669 atomic_inc(&eb->refs);
671 reads_done = atomic_dec_and_test(&eb->io_pages);
675 eb->read_mirror = mirror;
676 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
680 ret = validate_extent_buffer(eb, &bbio->parent_check);
684 * our io error hook is going to dec the io pages
685 * again, we have to make sure it has something
688 atomic_inc(&eb->io_pages);
689 clear_extent_buffer_uptodate(eb);
691 free_extent_buffer(eb);
696 static void run_one_async_start(struct btrfs_work *work)
698 struct async_submit_bio *async;
701 async = container_of(work, struct async_submit_bio, work);
702 switch (async->submit_cmd) {
703 case WQ_SUBMIT_METADATA:
704 ret = btree_submit_bio_start(async->bio);
707 ret = btrfs_submit_bio_start(async->inode, async->bio);
709 case WQ_SUBMIT_DATA_DIO:
710 ret = btrfs_submit_bio_start_direct_io(async->inode,
711 async->bio, async->dio_file_offset);
719 * In order to insert checksums into the metadata in large chunks, we wait
720 * until bio submission time. All the pages in the bio are checksummed and
721 * sums are attached onto the ordered extent record.
723 * At IO completion time the csums attached on the ordered extent record are
724 * inserted into the tree.
726 static void run_one_async_done(struct btrfs_work *work)
728 struct async_submit_bio *async =
729 container_of(work, struct async_submit_bio, work);
730 struct btrfs_inode *inode = async->inode;
731 struct btrfs_bio *bbio = btrfs_bio(async->bio);
733 /* If an error occurred we just want to clean up the bio and move on */
735 btrfs_bio_end_io(bbio, async->status);
740 * All of the bios that pass through here are from async helpers.
741 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
742 * This changes nothing when cgroups aren't in use.
744 async->bio->bi_opf |= REQ_CGROUP_PUNT;
745 btrfs_submit_bio(inode->root->fs_info, async->bio, async->mirror_num);
748 static void run_one_async_free(struct btrfs_work *work)
750 struct async_submit_bio *async;
752 async = container_of(work, struct async_submit_bio, work);
757 * Submit bio to an async queue.
760 * - true if the work has been succesfuly submitted
761 * - false in case of error
763 bool btrfs_wq_submit_bio(struct btrfs_inode *inode, struct bio *bio, int mirror_num,
764 u64 dio_file_offset, enum btrfs_wq_submit_cmd cmd)
766 struct btrfs_fs_info *fs_info = inode->root->fs_info;
767 struct async_submit_bio *async;
769 async = kmalloc(sizeof(*async), GFP_NOFS);
773 async->inode = inode;
775 async->mirror_num = mirror_num;
776 async->submit_cmd = cmd;
778 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
781 async->dio_file_offset = dio_file_offset;
785 if (op_is_sync(bio->bi_opf))
786 btrfs_queue_work(fs_info->hipri_workers, &async->work);
788 btrfs_queue_work(fs_info->workers, &async->work);
792 static blk_status_t btree_csum_one_bio(struct bio *bio)
794 struct bio_vec *bvec;
795 struct btrfs_root *root;
797 struct bvec_iter_all iter_all;
799 ASSERT(!bio_flagged(bio, BIO_CLONED));
800 bio_for_each_segment_all(bvec, bio, iter_all) {
801 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
802 ret = csum_dirty_buffer(root->fs_info, bvec);
807 return errno_to_blk_status(ret);
810 blk_status_t btree_submit_bio_start(struct bio *bio)
813 * when we're called for a write, we're already in the async
814 * submission context. Just jump into btrfs_submit_bio.
816 return btree_csum_one_bio(bio);
819 static bool should_async_write(struct btrfs_fs_info *fs_info,
820 struct btrfs_inode *bi)
822 if (btrfs_is_zoned(fs_info))
824 if (atomic_read(&bi->sync_writers))
826 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
831 void btrfs_submit_metadata_bio(struct btrfs_inode *inode, struct bio *bio, int mirror_num)
833 struct btrfs_fs_info *fs_info = inode->root->fs_info;
834 struct btrfs_bio *bbio = btrfs_bio(bio);
837 bio->bi_opf |= REQ_META;
838 bbio->is_metadata = 1;
840 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
841 btrfs_submit_bio(fs_info, bio, mirror_num);
846 * Kthread helpers are used to submit writes so that checksumming can
847 * happen in parallel across all CPUs.
849 if (should_async_write(fs_info, inode) &&
850 btrfs_wq_submit_bio(inode, bio, mirror_num, 0, WQ_SUBMIT_METADATA))
853 ret = btree_csum_one_bio(bio);
855 btrfs_bio_end_io(bbio, ret);
859 btrfs_submit_bio(fs_info, bio, mirror_num);
862 #ifdef CONFIG_MIGRATION
863 static int btree_migrate_folio(struct address_space *mapping,
864 struct folio *dst, struct folio *src, enum migrate_mode mode)
867 * we can't safely write a btree page from here,
868 * we haven't done the locking hook
870 if (folio_test_dirty(src))
873 * Buffers may be managed in a filesystem specific way.
874 * We must have no buffers or drop them.
876 if (folio_get_private(src) &&
877 !filemap_release_folio(src, GFP_KERNEL))
879 return migrate_folio(mapping, dst, src, mode);
882 #define btree_migrate_folio NULL
885 static int btree_writepages(struct address_space *mapping,
886 struct writeback_control *wbc)
888 struct btrfs_fs_info *fs_info;
891 if (wbc->sync_mode == WB_SYNC_NONE) {
893 if (wbc->for_kupdate)
896 fs_info = BTRFS_I(mapping->host)->root->fs_info;
897 /* this is a bit racy, but that's ok */
898 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
899 BTRFS_DIRTY_METADATA_THRESH,
900 fs_info->dirty_metadata_batch);
904 return btree_write_cache_pages(mapping, wbc);
907 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
909 if (folio_test_writeback(folio) || folio_test_dirty(folio))
912 return try_release_extent_buffer(&folio->page);
915 static void btree_invalidate_folio(struct folio *folio, size_t offset,
918 struct extent_io_tree *tree;
919 tree = &BTRFS_I(folio->mapping->host)->io_tree;
920 extent_invalidate_folio(tree, folio, offset);
921 btree_release_folio(folio, GFP_NOFS);
922 if (folio_get_private(folio)) {
923 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
924 "folio private not zero on folio %llu",
925 (unsigned long long)folio_pos(folio));
926 folio_detach_private(folio);
931 static bool btree_dirty_folio(struct address_space *mapping,
934 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
935 struct btrfs_subpage *subpage;
936 struct extent_buffer *eb;
938 u64 page_start = folio_pos(folio);
940 if (fs_info->sectorsize == PAGE_SIZE) {
941 eb = folio_get_private(folio);
943 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
944 BUG_ON(!atomic_read(&eb->refs));
945 btrfs_assert_tree_write_locked(eb);
946 return filemap_dirty_folio(mapping, folio);
948 subpage = folio_get_private(folio);
950 ASSERT(subpage->dirty_bitmap);
951 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
954 u16 tmp = (1 << cur_bit);
956 spin_lock_irqsave(&subpage->lock, flags);
957 if (!(tmp & subpage->dirty_bitmap)) {
958 spin_unlock_irqrestore(&subpage->lock, flags);
962 spin_unlock_irqrestore(&subpage->lock, flags);
963 cur = page_start + cur_bit * fs_info->sectorsize;
965 eb = find_extent_buffer(fs_info, cur);
967 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
968 ASSERT(atomic_read(&eb->refs));
969 btrfs_assert_tree_write_locked(eb);
970 free_extent_buffer(eb);
972 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
974 return filemap_dirty_folio(mapping, folio);
977 #define btree_dirty_folio filemap_dirty_folio
980 static const struct address_space_operations btree_aops = {
981 .writepages = btree_writepages,
982 .release_folio = btree_release_folio,
983 .invalidate_folio = btree_invalidate_folio,
984 .migrate_folio = btree_migrate_folio,
985 .dirty_folio = btree_dirty_folio,
988 struct extent_buffer *btrfs_find_create_tree_block(
989 struct btrfs_fs_info *fs_info,
990 u64 bytenr, u64 owner_root,
993 if (btrfs_is_testing(fs_info))
994 return alloc_test_extent_buffer(fs_info, bytenr);
995 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
999 * Read tree block at logical address @bytenr and do variant basic but critical
1002 * @check: expected tree parentness check, see comments of the
1003 * structure for details.
1005 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1006 struct btrfs_tree_parent_check *check)
1008 struct extent_buffer *buf = NULL;
1013 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
1018 ret = btrfs_read_extent_buffer(buf, check);
1020 free_extent_buffer_stale(buf);
1021 return ERR_PTR(ret);
1023 if (btrfs_check_eb_owner(buf, check->owner_root)) {
1024 free_extent_buffer_stale(buf);
1025 return ERR_PTR(-EUCLEAN);
1031 void btrfs_clean_tree_block(struct extent_buffer *buf)
1033 struct btrfs_fs_info *fs_info = buf->fs_info;
1034 if (btrfs_header_generation(buf) ==
1035 fs_info->running_transaction->transid) {
1036 btrfs_assert_tree_write_locked(buf);
1038 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1039 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1041 fs_info->dirty_metadata_batch);
1042 clear_extent_buffer_dirty(buf);
1047 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1050 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1052 memset(&root->root_key, 0, sizeof(root->root_key));
1053 memset(&root->root_item, 0, sizeof(root->root_item));
1054 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1055 root->fs_info = fs_info;
1056 root->root_key.objectid = objectid;
1058 root->commit_root = NULL;
1060 RB_CLEAR_NODE(&root->rb_node);
1062 root->last_trans = 0;
1063 root->free_objectid = 0;
1064 root->nr_delalloc_inodes = 0;
1065 root->nr_ordered_extents = 0;
1066 root->inode_tree = RB_ROOT;
1067 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1069 btrfs_init_root_block_rsv(root);
1071 INIT_LIST_HEAD(&root->dirty_list);
1072 INIT_LIST_HEAD(&root->root_list);
1073 INIT_LIST_HEAD(&root->delalloc_inodes);
1074 INIT_LIST_HEAD(&root->delalloc_root);
1075 INIT_LIST_HEAD(&root->ordered_extents);
1076 INIT_LIST_HEAD(&root->ordered_root);
1077 INIT_LIST_HEAD(&root->reloc_dirty_list);
1078 INIT_LIST_HEAD(&root->logged_list[0]);
1079 INIT_LIST_HEAD(&root->logged_list[1]);
1080 spin_lock_init(&root->inode_lock);
1081 spin_lock_init(&root->delalloc_lock);
1082 spin_lock_init(&root->ordered_extent_lock);
1083 spin_lock_init(&root->accounting_lock);
1084 spin_lock_init(&root->log_extents_lock[0]);
1085 spin_lock_init(&root->log_extents_lock[1]);
1086 spin_lock_init(&root->qgroup_meta_rsv_lock);
1087 mutex_init(&root->objectid_mutex);
1088 mutex_init(&root->log_mutex);
1089 mutex_init(&root->ordered_extent_mutex);
1090 mutex_init(&root->delalloc_mutex);
1091 init_waitqueue_head(&root->qgroup_flush_wait);
1092 init_waitqueue_head(&root->log_writer_wait);
1093 init_waitqueue_head(&root->log_commit_wait[0]);
1094 init_waitqueue_head(&root->log_commit_wait[1]);
1095 INIT_LIST_HEAD(&root->log_ctxs[0]);
1096 INIT_LIST_HEAD(&root->log_ctxs[1]);
1097 atomic_set(&root->log_commit[0], 0);
1098 atomic_set(&root->log_commit[1], 0);
1099 atomic_set(&root->log_writers, 0);
1100 atomic_set(&root->log_batch, 0);
1101 refcount_set(&root->refs, 1);
1102 atomic_set(&root->snapshot_force_cow, 0);
1103 atomic_set(&root->nr_swapfiles, 0);
1104 root->log_transid = 0;
1105 root->log_transid_committed = -1;
1106 root->last_log_commit = 0;
1109 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1110 IO_TREE_ROOT_DIRTY_LOG_PAGES);
1111 extent_io_tree_init(fs_info, &root->log_csum_range,
1112 IO_TREE_LOG_CSUM_RANGE);
1115 spin_lock_init(&root->root_item_lock);
1116 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1117 #ifdef CONFIG_BTRFS_DEBUG
1118 INIT_LIST_HEAD(&root->leak_list);
1119 spin_lock(&fs_info->fs_roots_radix_lock);
1120 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1121 spin_unlock(&fs_info->fs_roots_radix_lock);
1125 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1126 u64 objectid, gfp_t flags)
1128 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1130 __setup_root(root, fs_info, objectid);
1134 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1135 /* Should only be used by the testing infrastructure */
1136 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1138 struct btrfs_root *root;
1141 return ERR_PTR(-EINVAL);
1143 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1145 return ERR_PTR(-ENOMEM);
1147 /* We don't use the stripesize in selftest, set it as sectorsize */
1148 root->alloc_bytenr = 0;
1154 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1156 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1157 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1159 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1162 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1164 const struct btrfs_key *key = k;
1165 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1167 return btrfs_comp_cpu_keys(key, &root->root_key);
1170 int btrfs_global_root_insert(struct btrfs_root *root)
1172 struct btrfs_fs_info *fs_info = root->fs_info;
1173 struct rb_node *tmp;
1175 write_lock(&fs_info->global_root_lock);
1176 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1177 write_unlock(&fs_info->global_root_lock);
1180 return tmp ? -EEXIST : 0;
1183 void btrfs_global_root_delete(struct btrfs_root *root)
1185 struct btrfs_fs_info *fs_info = root->fs_info;
1187 write_lock(&fs_info->global_root_lock);
1188 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1189 write_unlock(&fs_info->global_root_lock);
1192 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1193 struct btrfs_key *key)
1195 struct rb_node *node;
1196 struct btrfs_root *root = NULL;
1198 read_lock(&fs_info->global_root_lock);
1199 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1201 root = container_of(node, struct btrfs_root, rb_node);
1202 read_unlock(&fs_info->global_root_lock);
1207 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1209 struct btrfs_block_group *block_group;
1212 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1216 block_group = btrfs_lookup_block_group(fs_info, bytenr);
1218 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1219 ASSERT(block_group);
1222 ret = block_group->global_root_id;
1223 btrfs_put_block_group(block_group);
1228 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1230 struct btrfs_key key = {
1231 .objectid = BTRFS_CSUM_TREE_OBJECTID,
1232 .type = BTRFS_ROOT_ITEM_KEY,
1233 .offset = btrfs_global_root_id(fs_info, bytenr),
1236 return btrfs_global_root(fs_info, &key);
1239 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1241 struct btrfs_key key = {
1242 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
1243 .type = BTRFS_ROOT_ITEM_KEY,
1244 .offset = btrfs_global_root_id(fs_info, bytenr),
1247 return btrfs_global_root(fs_info, &key);
1250 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
1252 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
1253 return fs_info->block_group_root;
1254 return btrfs_extent_root(fs_info, 0);
1257 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1260 struct btrfs_fs_info *fs_info = trans->fs_info;
1261 struct extent_buffer *leaf;
1262 struct btrfs_root *tree_root = fs_info->tree_root;
1263 struct btrfs_root *root;
1264 struct btrfs_key key;
1265 unsigned int nofs_flag;
1269 * We're holding a transaction handle, so use a NOFS memory allocation
1270 * context to avoid deadlock if reclaim happens.
1272 nofs_flag = memalloc_nofs_save();
1273 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1274 memalloc_nofs_restore(nofs_flag);
1276 return ERR_PTR(-ENOMEM);
1278 root->root_key.objectid = objectid;
1279 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1280 root->root_key.offset = 0;
1282 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1283 BTRFS_NESTING_NORMAL);
1285 ret = PTR_ERR(leaf);
1291 btrfs_mark_buffer_dirty(leaf);
1293 root->commit_root = btrfs_root_node(root);
1294 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1296 btrfs_set_root_flags(&root->root_item, 0);
1297 btrfs_set_root_limit(&root->root_item, 0);
1298 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1299 btrfs_set_root_generation(&root->root_item, trans->transid);
1300 btrfs_set_root_level(&root->root_item, 0);
1301 btrfs_set_root_refs(&root->root_item, 1);
1302 btrfs_set_root_used(&root->root_item, leaf->len);
1303 btrfs_set_root_last_snapshot(&root->root_item, 0);
1304 btrfs_set_root_dirid(&root->root_item, 0);
1305 if (is_fstree(objectid))
1306 generate_random_guid(root->root_item.uuid);
1308 export_guid(root->root_item.uuid, &guid_null);
1309 btrfs_set_root_drop_level(&root->root_item, 0);
1311 btrfs_tree_unlock(leaf);
1313 key.objectid = objectid;
1314 key.type = BTRFS_ROOT_ITEM_KEY;
1316 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1323 btrfs_put_root(root);
1325 return ERR_PTR(ret);
1328 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1329 struct btrfs_fs_info *fs_info)
1331 struct btrfs_root *root;
1333 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1335 return ERR_PTR(-ENOMEM);
1337 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1338 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1339 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1344 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1345 struct btrfs_root *root)
1347 struct extent_buffer *leaf;
1350 * DON'T set SHAREABLE bit for log trees.
1352 * Log trees are not exposed to user space thus can't be snapshotted,
1353 * and they go away before a real commit is actually done.
1355 * They do store pointers to file data extents, and those reference
1356 * counts still get updated (along with back refs to the log tree).
1359 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1360 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1362 return PTR_ERR(leaf);
1366 btrfs_mark_buffer_dirty(root->node);
1367 btrfs_tree_unlock(root->node);
1372 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1373 struct btrfs_fs_info *fs_info)
1375 struct btrfs_root *log_root;
1377 log_root = alloc_log_tree(trans, fs_info);
1378 if (IS_ERR(log_root))
1379 return PTR_ERR(log_root);
1381 if (!btrfs_is_zoned(fs_info)) {
1382 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1385 btrfs_put_root(log_root);
1390 WARN_ON(fs_info->log_root_tree);
1391 fs_info->log_root_tree = log_root;
1395 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1396 struct btrfs_root *root)
1398 struct btrfs_fs_info *fs_info = root->fs_info;
1399 struct btrfs_root *log_root;
1400 struct btrfs_inode_item *inode_item;
1403 log_root = alloc_log_tree(trans, fs_info);
1404 if (IS_ERR(log_root))
1405 return PTR_ERR(log_root);
1407 ret = btrfs_alloc_log_tree_node(trans, log_root);
1409 btrfs_put_root(log_root);
1413 log_root->last_trans = trans->transid;
1414 log_root->root_key.offset = root->root_key.objectid;
1416 inode_item = &log_root->root_item.inode;
1417 btrfs_set_stack_inode_generation(inode_item, 1);
1418 btrfs_set_stack_inode_size(inode_item, 3);
1419 btrfs_set_stack_inode_nlink(inode_item, 1);
1420 btrfs_set_stack_inode_nbytes(inode_item,
1422 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1424 btrfs_set_root_node(&log_root->root_item, log_root->node);
1426 WARN_ON(root->log_root);
1427 root->log_root = log_root;
1428 root->log_transid = 0;
1429 root->log_transid_committed = -1;
1430 root->last_log_commit = 0;
1434 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1435 struct btrfs_path *path,
1436 struct btrfs_key *key)
1438 struct btrfs_root *root;
1439 struct btrfs_tree_parent_check check = { 0 };
1440 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1445 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1447 return ERR_PTR(-ENOMEM);
1449 ret = btrfs_find_root(tree_root, key, path,
1450 &root->root_item, &root->root_key);
1457 generation = btrfs_root_generation(&root->root_item);
1458 level = btrfs_root_level(&root->root_item);
1459 check.level = level;
1460 check.transid = generation;
1461 check.owner_root = key->objectid;
1462 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1464 if (IS_ERR(root->node)) {
1465 ret = PTR_ERR(root->node);
1469 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1475 * For real fs, and not log/reloc trees, root owner must
1476 * match its root node owner
1478 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1479 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1480 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1481 root->root_key.objectid != btrfs_header_owner(root->node)) {
1483 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1484 root->root_key.objectid, root->node->start,
1485 btrfs_header_owner(root->node),
1486 root->root_key.objectid);
1490 root->commit_root = btrfs_root_node(root);
1493 btrfs_put_root(root);
1494 return ERR_PTR(ret);
1497 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1498 struct btrfs_key *key)
1500 struct btrfs_root *root;
1501 struct btrfs_path *path;
1503 path = btrfs_alloc_path();
1505 return ERR_PTR(-ENOMEM);
1506 root = read_tree_root_path(tree_root, path, key);
1507 btrfs_free_path(path);
1513 * Initialize subvolume root in-memory structure
1515 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1517 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1520 unsigned int nofs_flag;
1523 * We might be called under a transaction (e.g. indirect backref
1524 * resolution) which could deadlock if it triggers memory reclaim
1526 nofs_flag = memalloc_nofs_save();
1527 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1528 memalloc_nofs_restore(nofs_flag);
1532 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1533 !btrfs_is_data_reloc_root(root)) {
1534 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1535 btrfs_check_and_init_root_item(&root->root_item);
1539 * Don't assign anonymous block device to roots that are not exposed to
1540 * userspace, the id pool is limited to 1M
1542 if (is_fstree(root->root_key.objectid) &&
1543 btrfs_root_refs(&root->root_item) > 0) {
1545 ret = get_anon_bdev(&root->anon_dev);
1549 root->anon_dev = anon_dev;
1553 mutex_lock(&root->objectid_mutex);
1554 ret = btrfs_init_root_free_objectid(root);
1556 mutex_unlock(&root->objectid_mutex);
1560 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1562 mutex_unlock(&root->objectid_mutex);
1566 /* The caller is responsible to call btrfs_free_fs_root */
1570 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1573 struct btrfs_root *root;
1575 spin_lock(&fs_info->fs_roots_radix_lock);
1576 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1577 (unsigned long)root_id);
1579 root = btrfs_grab_root(root);
1580 spin_unlock(&fs_info->fs_roots_radix_lock);
1584 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1587 struct btrfs_key key = {
1588 .objectid = objectid,
1589 .type = BTRFS_ROOT_ITEM_KEY,
1593 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1594 return btrfs_grab_root(fs_info->tree_root);
1595 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1596 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1597 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1598 return btrfs_grab_root(fs_info->chunk_root);
1599 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1600 return btrfs_grab_root(fs_info->dev_root);
1601 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1602 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1603 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1604 return btrfs_grab_root(fs_info->quota_root) ?
1605 fs_info->quota_root : ERR_PTR(-ENOENT);
1606 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1607 return btrfs_grab_root(fs_info->uuid_root) ?
1608 fs_info->uuid_root : ERR_PTR(-ENOENT);
1609 if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID)
1610 return btrfs_grab_root(fs_info->block_group_root) ?
1611 fs_info->block_group_root : ERR_PTR(-ENOENT);
1612 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1613 struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1615 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1620 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1621 struct btrfs_root *root)
1625 ret = radix_tree_preload(GFP_NOFS);
1629 spin_lock(&fs_info->fs_roots_radix_lock);
1630 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1631 (unsigned long)root->root_key.objectid,
1634 btrfs_grab_root(root);
1635 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1637 spin_unlock(&fs_info->fs_roots_radix_lock);
1638 radix_tree_preload_end();
1643 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1645 #ifdef CONFIG_BTRFS_DEBUG
1646 struct btrfs_root *root;
1648 while (!list_empty(&fs_info->allocated_roots)) {
1649 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1651 root = list_first_entry(&fs_info->allocated_roots,
1652 struct btrfs_root, leak_list);
1653 btrfs_err(fs_info, "leaked root %s refcount %d",
1654 btrfs_root_name(&root->root_key, buf),
1655 refcount_read(&root->refs));
1656 while (refcount_read(&root->refs) > 1)
1657 btrfs_put_root(root);
1658 btrfs_put_root(root);
1663 static void free_global_roots(struct btrfs_fs_info *fs_info)
1665 struct btrfs_root *root;
1666 struct rb_node *node;
1668 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1669 root = rb_entry(node, struct btrfs_root, rb_node);
1670 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1671 btrfs_put_root(root);
1675 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1677 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1678 percpu_counter_destroy(&fs_info->delalloc_bytes);
1679 percpu_counter_destroy(&fs_info->ordered_bytes);
1680 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1681 btrfs_free_csum_hash(fs_info);
1682 btrfs_free_stripe_hash_table(fs_info);
1683 btrfs_free_ref_cache(fs_info);
1684 kfree(fs_info->balance_ctl);
1685 kfree(fs_info->delayed_root);
1686 free_global_roots(fs_info);
1687 btrfs_put_root(fs_info->tree_root);
1688 btrfs_put_root(fs_info->chunk_root);
1689 btrfs_put_root(fs_info->dev_root);
1690 btrfs_put_root(fs_info->quota_root);
1691 btrfs_put_root(fs_info->uuid_root);
1692 btrfs_put_root(fs_info->fs_root);
1693 btrfs_put_root(fs_info->data_reloc_root);
1694 btrfs_put_root(fs_info->block_group_root);
1695 btrfs_check_leaked_roots(fs_info);
1696 btrfs_extent_buffer_leak_debug_check(fs_info);
1697 kfree(fs_info->super_copy);
1698 kfree(fs_info->super_for_commit);
1699 kfree(fs_info->subpage_info);
1705 * Get an in-memory reference of a root structure.
1707 * For essential trees like root/extent tree, we grab it from fs_info directly.
1708 * For subvolume trees, we check the cached filesystem roots first. If not
1709 * found, then read it from disk and add it to cached fs roots.
1711 * Caller should release the root by calling btrfs_put_root() after the usage.
1713 * NOTE: Reloc and log trees can't be read by this function as they share the
1714 * same root objectid.
1716 * @objectid: root id
1717 * @anon_dev: preallocated anonymous block device number for new roots,
1718 * pass 0 for new allocation.
1719 * @check_ref: whether to check root item references, If true, return -ENOENT
1722 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1723 u64 objectid, dev_t anon_dev,
1726 struct btrfs_root *root;
1727 struct btrfs_path *path;
1728 struct btrfs_key key;
1731 root = btrfs_get_global_root(fs_info, objectid);
1735 root = btrfs_lookup_fs_root(fs_info, objectid);
1737 /* Shouldn't get preallocated anon_dev for cached roots */
1739 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1740 btrfs_put_root(root);
1741 return ERR_PTR(-ENOENT);
1746 key.objectid = objectid;
1747 key.type = BTRFS_ROOT_ITEM_KEY;
1748 key.offset = (u64)-1;
1749 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1753 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1758 ret = btrfs_init_fs_root(root, anon_dev);
1762 path = btrfs_alloc_path();
1767 key.objectid = BTRFS_ORPHAN_OBJECTID;
1768 key.type = BTRFS_ORPHAN_ITEM_KEY;
1769 key.offset = objectid;
1771 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1772 btrfs_free_path(path);
1776 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1778 ret = btrfs_insert_fs_root(fs_info, root);
1780 if (ret == -EEXIST) {
1781 btrfs_put_root(root);
1789 * If our caller provided us an anonymous device, then it's his
1790 * responsibility to free it in case we fail. So we have to set our
1791 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1792 * and once again by our caller.
1796 btrfs_put_root(root);
1797 return ERR_PTR(ret);
1801 * Get in-memory reference of a root structure
1803 * @objectid: tree objectid
1804 * @check_ref: if set, verify that the tree exists and the item has at least
1807 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1808 u64 objectid, bool check_ref)
1810 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1814 * Get in-memory reference of a root structure, created as new, optionally pass
1815 * the anonymous block device id
1817 * @objectid: tree objectid
1818 * @anon_dev: if zero, allocate a new anonymous block device or use the
1821 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1822 u64 objectid, dev_t anon_dev)
1824 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1828 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1829 * @fs_info: the fs_info
1830 * @objectid: the objectid we need to lookup
1832 * This is exclusively used for backref walking, and exists specifically because
1833 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1834 * creation time, which means we may have to read the tree_root in order to look
1835 * up a fs root that is not in memory. If the root is not in memory we will
1836 * read the tree root commit root and look up the fs root from there. This is a
1837 * temporary root, it will not be inserted into the radix tree as it doesn't
1838 * have the most uptodate information, it'll simply be discarded once the
1839 * backref code is finished using the root.
1841 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1842 struct btrfs_path *path,
1845 struct btrfs_root *root;
1846 struct btrfs_key key;
1848 ASSERT(path->search_commit_root && path->skip_locking);
1851 * This can return -ENOENT if we ask for a root that doesn't exist, but
1852 * since this is called via the backref walking code we won't be looking
1853 * up a root that doesn't exist, unless there's corruption. So if root
1854 * != NULL just return it.
1856 root = btrfs_get_global_root(fs_info, objectid);
1860 root = btrfs_lookup_fs_root(fs_info, objectid);
1864 key.objectid = objectid;
1865 key.type = BTRFS_ROOT_ITEM_KEY;
1866 key.offset = (u64)-1;
1867 root = read_tree_root_path(fs_info->tree_root, path, &key);
1868 btrfs_release_path(path);
1873 static int cleaner_kthread(void *arg)
1875 struct btrfs_fs_info *fs_info = arg;
1881 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1883 /* Make the cleaner go to sleep early. */
1884 if (btrfs_need_cleaner_sleep(fs_info))
1888 * Do not do anything if we might cause open_ctree() to block
1889 * before we have finished mounting the filesystem.
1891 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1894 if (!mutex_trylock(&fs_info->cleaner_mutex))
1898 * Avoid the problem that we change the status of the fs
1899 * during the above check and trylock.
1901 if (btrfs_need_cleaner_sleep(fs_info)) {
1902 mutex_unlock(&fs_info->cleaner_mutex);
1906 btrfs_run_delayed_iputs(fs_info);
1908 again = btrfs_clean_one_deleted_snapshot(fs_info);
1909 mutex_unlock(&fs_info->cleaner_mutex);
1912 * The defragger has dealt with the R/O remount and umount,
1913 * needn't do anything special here.
1915 btrfs_run_defrag_inodes(fs_info);
1918 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1919 * with relocation (btrfs_relocate_chunk) and relocation
1920 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1921 * after acquiring fs_info->reclaim_bgs_lock. So we
1922 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1923 * unused block groups.
1925 btrfs_delete_unused_bgs(fs_info);
1928 * Reclaim block groups in the reclaim_bgs list after we deleted
1929 * all unused block_groups. This possibly gives us some more free
1932 btrfs_reclaim_bgs(fs_info);
1934 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1935 if (kthread_should_park())
1937 if (kthread_should_stop())
1940 set_current_state(TASK_INTERRUPTIBLE);
1942 __set_current_state(TASK_RUNNING);
1947 static int transaction_kthread(void *arg)
1949 struct btrfs_root *root = arg;
1950 struct btrfs_fs_info *fs_info = root->fs_info;
1951 struct btrfs_trans_handle *trans;
1952 struct btrfs_transaction *cur;
1955 unsigned long delay;
1959 cannot_commit = false;
1960 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1961 mutex_lock(&fs_info->transaction_kthread_mutex);
1963 spin_lock(&fs_info->trans_lock);
1964 cur = fs_info->running_transaction;
1966 spin_unlock(&fs_info->trans_lock);
1970 delta = ktime_get_seconds() - cur->start_time;
1971 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1972 cur->state < TRANS_STATE_COMMIT_START &&
1973 delta < fs_info->commit_interval) {
1974 spin_unlock(&fs_info->trans_lock);
1975 delay -= msecs_to_jiffies((delta - 1) * 1000);
1977 msecs_to_jiffies(fs_info->commit_interval * 1000));
1980 transid = cur->transid;
1981 spin_unlock(&fs_info->trans_lock);
1983 /* If the file system is aborted, this will always fail. */
1984 trans = btrfs_attach_transaction(root);
1985 if (IS_ERR(trans)) {
1986 if (PTR_ERR(trans) != -ENOENT)
1987 cannot_commit = true;
1990 if (transid == trans->transid) {
1991 btrfs_commit_transaction(trans);
1993 btrfs_end_transaction(trans);
1996 wake_up_process(fs_info->cleaner_kthread);
1997 mutex_unlock(&fs_info->transaction_kthread_mutex);
1999 if (BTRFS_FS_ERROR(fs_info))
2000 btrfs_cleanup_transaction(fs_info);
2001 if (!kthread_should_stop() &&
2002 (!btrfs_transaction_blocked(fs_info) ||
2004 schedule_timeout_interruptible(delay);
2005 } while (!kthread_should_stop());
2010 * This will find the highest generation in the array of root backups. The
2011 * index of the highest array is returned, or -EINVAL if we can't find
2014 * We check to make sure the array is valid by comparing the
2015 * generation of the latest root in the array with the generation
2016 * in the super block. If they don't match we pitch it.
2018 static int find_newest_super_backup(struct btrfs_fs_info *info)
2020 const u64 newest_gen = btrfs_super_generation(info->super_copy);
2022 struct btrfs_root_backup *root_backup;
2025 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2026 root_backup = info->super_copy->super_roots + i;
2027 cur = btrfs_backup_tree_root_gen(root_backup);
2028 if (cur == newest_gen)
2036 * copy all the root pointers into the super backup array.
2037 * this will bump the backup pointer by one when it is
2040 static void backup_super_roots(struct btrfs_fs_info *info)
2042 const int next_backup = info->backup_root_index;
2043 struct btrfs_root_backup *root_backup;
2045 root_backup = info->super_for_commit->super_roots + next_backup;
2048 * make sure all of our padding and empty slots get zero filled
2049 * regardless of which ones we use today
2051 memset(root_backup, 0, sizeof(*root_backup));
2053 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2055 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2056 btrfs_set_backup_tree_root_gen(root_backup,
2057 btrfs_header_generation(info->tree_root->node));
2059 btrfs_set_backup_tree_root_level(root_backup,
2060 btrfs_header_level(info->tree_root->node));
2062 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2063 btrfs_set_backup_chunk_root_gen(root_backup,
2064 btrfs_header_generation(info->chunk_root->node));
2065 btrfs_set_backup_chunk_root_level(root_backup,
2066 btrfs_header_level(info->chunk_root->node));
2068 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
2069 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2070 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2072 btrfs_set_backup_extent_root(root_backup,
2073 extent_root->node->start);
2074 btrfs_set_backup_extent_root_gen(root_backup,
2075 btrfs_header_generation(extent_root->node));
2076 btrfs_set_backup_extent_root_level(root_backup,
2077 btrfs_header_level(extent_root->node));
2079 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2080 btrfs_set_backup_csum_root_gen(root_backup,
2081 btrfs_header_generation(csum_root->node));
2082 btrfs_set_backup_csum_root_level(root_backup,
2083 btrfs_header_level(csum_root->node));
2087 * we might commit during log recovery, which happens before we set
2088 * the fs_root. Make sure it is valid before we fill it in.
2090 if (info->fs_root && info->fs_root->node) {
2091 btrfs_set_backup_fs_root(root_backup,
2092 info->fs_root->node->start);
2093 btrfs_set_backup_fs_root_gen(root_backup,
2094 btrfs_header_generation(info->fs_root->node));
2095 btrfs_set_backup_fs_root_level(root_backup,
2096 btrfs_header_level(info->fs_root->node));
2099 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2100 btrfs_set_backup_dev_root_gen(root_backup,
2101 btrfs_header_generation(info->dev_root->node));
2102 btrfs_set_backup_dev_root_level(root_backup,
2103 btrfs_header_level(info->dev_root->node));
2105 btrfs_set_backup_total_bytes(root_backup,
2106 btrfs_super_total_bytes(info->super_copy));
2107 btrfs_set_backup_bytes_used(root_backup,
2108 btrfs_super_bytes_used(info->super_copy));
2109 btrfs_set_backup_num_devices(root_backup,
2110 btrfs_super_num_devices(info->super_copy));
2113 * if we don't copy this out to the super_copy, it won't get remembered
2114 * for the next commit
2116 memcpy(&info->super_copy->super_roots,
2117 &info->super_for_commit->super_roots,
2118 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2122 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2123 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2125 * fs_info - filesystem whose backup roots need to be read
2126 * priority - priority of backup root required
2128 * Returns backup root index on success and -EINVAL otherwise.
2130 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2132 int backup_index = find_newest_super_backup(fs_info);
2133 struct btrfs_super_block *super = fs_info->super_copy;
2134 struct btrfs_root_backup *root_backup;
2136 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2138 return backup_index;
2140 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2141 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2146 root_backup = super->super_roots + backup_index;
2148 btrfs_set_super_generation(super,
2149 btrfs_backup_tree_root_gen(root_backup));
2150 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2151 btrfs_set_super_root_level(super,
2152 btrfs_backup_tree_root_level(root_backup));
2153 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2156 * Fixme: the total bytes and num_devices need to match or we should
2159 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2160 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2162 return backup_index;
2165 /* helper to cleanup workers */
2166 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2168 btrfs_destroy_workqueue(fs_info->fixup_workers);
2169 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2170 btrfs_destroy_workqueue(fs_info->hipri_workers);
2171 btrfs_destroy_workqueue(fs_info->workers);
2172 if (fs_info->endio_workers)
2173 destroy_workqueue(fs_info->endio_workers);
2174 if (fs_info->rmw_workers)
2175 destroy_workqueue(fs_info->rmw_workers);
2176 if (fs_info->compressed_write_workers)
2177 destroy_workqueue(fs_info->compressed_write_workers);
2178 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2179 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2180 btrfs_destroy_workqueue(fs_info->delayed_workers);
2181 btrfs_destroy_workqueue(fs_info->caching_workers);
2182 btrfs_destroy_workqueue(fs_info->flush_workers);
2183 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2184 if (fs_info->discard_ctl.discard_workers)
2185 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2187 * Now that all other work queues are destroyed, we can safely destroy
2188 * the queues used for metadata I/O, since tasks from those other work
2189 * queues can do metadata I/O operations.
2191 if (fs_info->endio_meta_workers)
2192 destroy_workqueue(fs_info->endio_meta_workers);
2195 static void free_root_extent_buffers(struct btrfs_root *root)
2198 free_extent_buffer(root->node);
2199 free_extent_buffer(root->commit_root);
2201 root->commit_root = NULL;
2205 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2207 struct btrfs_root *root, *tmp;
2209 rbtree_postorder_for_each_entry_safe(root, tmp,
2210 &fs_info->global_root_tree,
2212 free_root_extent_buffers(root);
2215 /* helper to cleanup tree roots */
2216 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2218 free_root_extent_buffers(info->tree_root);
2220 free_global_root_pointers(info);
2221 free_root_extent_buffers(info->dev_root);
2222 free_root_extent_buffers(info->quota_root);
2223 free_root_extent_buffers(info->uuid_root);
2224 free_root_extent_buffers(info->fs_root);
2225 free_root_extent_buffers(info->data_reloc_root);
2226 free_root_extent_buffers(info->block_group_root);
2227 if (free_chunk_root)
2228 free_root_extent_buffers(info->chunk_root);
2231 void btrfs_put_root(struct btrfs_root *root)
2236 if (refcount_dec_and_test(&root->refs)) {
2237 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2238 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2240 free_anon_bdev(root->anon_dev);
2241 btrfs_drew_lock_destroy(&root->snapshot_lock);
2242 free_root_extent_buffers(root);
2243 #ifdef CONFIG_BTRFS_DEBUG
2244 spin_lock(&root->fs_info->fs_roots_radix_lock);
2245 list_del_init(&root->leak_list);
2246 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2252 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2255 struct btrfs_root *gang[8];
2258 while (!list_empty(&fs_info->dead_roots)) {
2259 gang[0] = list_entry(fs_info->dead_roots.next,
2260 struct btrfs_root, root_list);
2261 list_del(&gang[0]->root_list);
2263 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2264 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2265 btrfs_put_root(gang[0]);
2269 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2274 for (i = 0; i < ret; i++)
2275 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2279 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2281 mutex_init(&fs_info->scrub_lock);
2282 atomic_set(&fs_info->scrubs_running, 0);
2283 atomic_set(&fs_info->scrub_pause_req, 0);
2284 atomic_set(&fs_info->scrubs_paused, 0);
2285 atomic_set(&fs_info->scrub_cancel_req, 0);
2286 init_waitqueue_head(&fs_info->scrub_pause_wait);
2287 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2290 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2292 spin_lock_init(&fs_info->balance_lock);
2293 mutex_init(&fs_info->balance_mutex);
2294 atomic_set(&fs_info->balance_pause_req, 0);
2295 atomic_set(&fs_info->balance_cancel_req, 0);
2296 fs_info->balance_ctl = NULL;
2297 init_waitqueue_head(&fs_info->balance_wait_q);
2298 atomic_set(&fs_info->reloc_cancel_req, 0);
2301 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2303 struct inode *inode = fs_info->btree_inode;
2304 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
2305 fs_info->tree_root);
2307 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2308 set_nlink(inode, 1);
2310 * we set the i_size on the btree inode to the max possible int.
2311 * the real end of the address space is determined by all of
2312 * the devices in the system
2314 inode->i_size = OFFSET_MAX;
2315 inode->i_mapping->a_ops = &btree_aops;
2317 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2318 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2319 IO_TREE_BTREE_INODE_IO);
2320 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2322 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2323 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2324 BTRFS_I(inode)->location.type = 0;
2325 BTRFS_I(inode)->location.offset = 0;
2326 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2327 __insert_inode_hash(inode, hash);
2330 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2332 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2333 init_rwsem(&fs_info->dev_replace.rwsem);
2334 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2337 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2339 spin_lock_init(&fs_info->qgroup_lock);
2340 mutex_init(&fs_info->qgroup_ioctl_lock);
2341 fs_info->qgroup_tree = RB_ROOT;
2342 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2343 fs_info->qgroup_seq = 1;
2344 fs_info->qgroup_ulist = NULL;
2345 fs_info->qgroup_rescan_running = false;
2346 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
2347 mutex_init(&fs_info->qgroup_rescan_lock);
2350 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2352 u32 max_active = fs_info->thread_pool_size;
2353 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2356 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2357 fs_info->hipri_workers =
2358 btrfs_alloc_workqueue(fs_info, "worker-high",
2359 flags | WQ_HIGHPRI, max_active, 16);
2361 fs_info->delalloc_workers =
2362 btrfs_alloc_workqueue(fs_info, "delalloc",
2363 flags, max_active, 2);
2365 fs_info->flush_workers =
2366 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2367 flags, max_active, 0);
2369 fs_info->caching_workers =
2370 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2372 fs_info->fixup_workers =
2373 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2375 fs_info->endio_workers =
2376 alloc_workqueue("btrfs-endio", flags, max_active);
2377 fs_info->endio_meta_workers =
2378 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2379 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2380 fs_info->endio_write_workers =
2381 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2383 fs_info->compressed_write_workers =
2384 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2385 fs_info->endio_freespace_worker =
2386 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2388 fs_info->delayed_workers =
2389 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2391 fs_info->qgroup_rescan_workers =
2392 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2393 fs_info->discard_ctl.discard_workers =
2394 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2396 if (!(fs_info->workers && fs_info->hipri_workers &&
2397 fs_info->delalloc_workers && fs_info->flush_workers &&
2398 fs_info->endio_workers && fs_info->endio_meta_workers &&
2399 fs_info->compressed_write_workers &&
2400 fs_info->endio_write_workers &&
2401 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2402 fs_info->caching_workers && fs_info->fixup_workers &&
2403 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2404 fs_info->discard_ctl.discard_workers)) {
2411 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2413 struct crypto_shash *csum_shash;
2414 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2416 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2418 if (IS_ERR(csum_shash)) {
2419 btrfs_err(fs_info, "error allocating %s hash for checksum",
2421 return PTR_ERR(csum_shash);
2424 fs_info->csum_shash = csum_shash;
2426 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2427 btrfs_super_csum_name(csum_type),
2428 crypto_shash_driver_name(csum_shash));
2432 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2433 struct btrfs_fs_devices *fs_devices)
2436 struct btrfs_tree_parent_check check = { 0 };
2437 struct btrfs_root *log_tree_root;
2438 struct btrfs_super_block *disk_super = fs_info->super_copy;
2439 u64 bytenr = btrfs_super_log_root(disk_super);
2440 int level = btrfs_super_log_root_level(disk_super);
2442 if (fs_devices->rw_devices == 0) {
2443 btrfs_warn(fs_info, "log replay required on RO media");
2447 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2452 check.level = level;
2453 check.transid = fs_info->generation + 1;
2454 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2455 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2456 if (IS_ERR(log_tree_root->node)) {
2457 btrfs_warn(fs_info, "failed to read log tree");
2458 ret = PTR_ERR(log_tree_root->node);
2459 log_tree_root->node = NULL;
2460 btrfs_put_root(log_tree_root);
2463 if (!extent_buffer_uptodate(log_tree_root->node)) {
2464 btrfs_err(fs_info, "failed to read log tree");
2465 btrfs_put_root(log_tree_root);
2469 /* returns with log_tree_root freed on success */
2470 ret = btrfs_recover_log_trees(log_tree_root);
2472 btrfs_handle_fs_error(fs_info, ret,
2473 "Failed to recover log tree");
2474 btrfs_put_root(log_tree_root);
2478 if (sb_rdonly(fs_info->sb)) {
2479 ret = btrfs_commit_super(fs_info);
2487 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2488 struct btrfs_path *path, u64 objectid,
2491 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2492 struct btrfs_root *root;
2493 u64 max_global_id = 0;
2495 struct btrfs_key key = {
2496 .objectid = objectid,
2497 .type = BTRFS_ROOT_ITEM_KEY,
2502 /* If we have IGNOREDATACSUMS skip loading these roots. */
2503 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2504 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2505 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2510 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2514 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2515 ret = btrfs_next_leaf(tree_root, path);
2524 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2525 if (key.objectid != objectid)
2527 btrfs_release_path(path);
2530 * Just worry about this for extent tree, it'll be the same for
2533 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2534 max_global_id = max(max_global_id, key.offset);
2537 root = read_tree_root_path(tree_root, path, &key);
2539 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2540 ret = PTR_ERR(root);
2543 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2544 ret = btrfs_global_root_insert(root);
2546 btrfs_put_root(root);
2551 btrfs_release_path(path);
2553 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2554 fs_info->nr_global_roots = max_global_id + 1;
2556 if (!found || ret) {
2557 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2558 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2560 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2561 ret = ret ? ret : -ENOENT;
2564 btrfs_err(fs_info, "failed to load root %s", name);
2569 static int load_global_roots(struct btrfs_root *tree_root)
2571 struct btrfs_path *path;
2574 path = btrfs_alloc_path();
2578 ret = load_global_roots_objectid(tree_root, path,
2579 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2582 ret = load_global_roots_objectid(tree_root, path,
2583 BTRFS_CSUM_TREE_OBJECTID, "csum");
2586 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2588 ret = load_global_roots_objectid(tree_root, path,
2589 BTRFS_FREE_SPACE_TREE_OBJECTID,
2592 btrfs_free_path(path);
2596 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2598 struct btrfs_root *tree_root = fs_info->tree_root;
2599 struct btrfs_root *root;
2600 struct btrfs_key location;
2603 BUG_ON(!fs_info->tree_root);
2605 ret = load_global_roots(tree_root);
2609 location.type = BTRFS_ROOT_ITEM_KEY;
2610 location.offset = 0;
2612 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2613 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2614 root = btrfs_read_tree_root(tree_root, &location);
2616 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2617 ret = PTR_ERR(root);
2621 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2622 fs_info->block_group_root = root;
2626 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2627 root = btrfs_read_tree_root(tree_root, &location);
2629 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2630 ret = PTR_ERR(root);
2634 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2635 fs_info->dev_root = root;
2637 /* Initialize fs_info for all devices in any case */
2638 ret = btrfs_init_devices_late(fs_info);
2643 * This tree can share blocks with some other fs tree during relocation
2644 * and we need a proper setup by btrfs_get_fs_root
2646 root = btrfs_get_fs_root(tree_root->fs_info,
2647 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2649 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2650 ret = PTR_ERR(root);
2654 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2655 fs_info->data_reloc_root = root;
2658 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2659 root = btrfs_read_tree_root(tree_root, &location);
2660 if (!IS_ERR(root)) {
2661 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2662 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2663 fs_info->quota_root = root;
2666 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2667 root = btrfs_read_tree_root(tree_root, &location);
2669 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2670 ret = PTR_ERR(root);
2675 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2676 fs_info->uuid_root = root;
2681 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2682 location.objectid, ret);
2687 * Real super block validation
2688 * NOTE: super csum type and incompat features will not be checked here.
2690 * @sb: super block to check
2691 * @mirror_num: the super block number to check its bytenr:
2692 * 0 the primary (1st) sb
2693 * 1, 2 2nd and 3rd backup copy
2694 * -1 skip bytenr check
2696 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2697 struct btrfs_super_block *sb, int mirror_num)
2699 u64 nodesize = btrfs_super_nodesize(sb);
2700 u64 sectorsize = btrfs_super_sectorsize(sb);
2703 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2704 btrfs_err(fs_info, "no valid FS found");
2707 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2708 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2709 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2712 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2713 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2714 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2717 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2718 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2719 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2722 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2723 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2724 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2729 * Check sectorsize and nodesize first, other check will need it.
2730 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2732 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2733 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2734 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2739 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2741 * We can support 16K sectorsize with 64K page size without problem,
2742 * but such sectorsize/pagesize combination doesn't make much sense.
2743 * 4K will be our future standard, PAGE_SIZE is supported from the very
2746 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2748 "sectorsize %llu not yet supported for page size %lu",
2749 sectorsize, PAGE_SIZE);
2753 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2754 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2755 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2758 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2759 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2760 le32_to_cpu(sb->__unused_leafsize), nodesize);
2764 /* Root alignment check */
2765 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2766 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2767 btrfs_super_root(sb));
2770 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2771 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2772 btrfs_super_chunk_root(sb));
2775 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2776 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2777 btrfs_super_log_root(sb));
2781 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2784 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2785 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2789 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2790 memcmp(fs_info->fs_devices->metadata_uuid,
2791 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2793 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2794 fs_info->super_copy->metadata_uuid,
2795 fs_info->fs_devices->metadata_uuid);
2800 * Artificial requirement for block-group-tree to force newer features
2801 * (free-space-tree, no-holes) so the test matrix is smaller.
2803 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2804 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2805 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2807 "block-group-tree feature requires fres-space-tree and no-holes");
2811 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2812 BTRFS_FSID_SIZE) != 0) {
2814 "dev_item UUID does not match metadata fsid: %pU != %pU",
2815 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2820 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2823 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2824 btrfs_err(fs_info, "bytes_used is too small %llu",
2825 btrfs_super_bytes_used(sb));
2828 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2829 btrfs_err(fs_info, "invalid stripesize %u",
2830 btrfs_super_stripesize(sb));
2833 if (btrfs_super_num_devices(sb) > (1UL << 31))
2834 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2835 btrfs_super_num_devices(sb));
2836 if (btrfs_super_num_devices(sb) == 0) {
2837 btrfs_err(fs_info, "number of devices is 0");
2841 if (mirror_num >= 0 &&
2842 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2843 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2844 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2849 * Obvious sys_chunk_array corruptions, it must hold at least one key
2852 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2853 btrfs_err(fs_info, "system chunk array too big %u > %u",
2854 btrfs_super_sys_array_size(sb),
2855 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2858 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2859 + sizeof(struct btrfs_chunk)) {
2860 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2861 btrfs_super_sys_array_size(sb),
2862 sizeof(struct btrfs_disk_key)
2863 + sizeof(struct btrfs_chunk));
2868 * The generation is a global counter, we'll trust it more than the others
2869 * but it's still possible that it's the one that's wrong.
2871 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2873 "suspicious: generation < chunk_root_generation: %llu < %llu",
2874 btrfs_super_generation(sb),
2875 btrfs_super_chunk_root_generation(sb));
2876 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2877 && btrfs_super_cache_generation(sb) != (u64)-1)
2879 "suspicious: generation < cache_generation: %llu < %llu",
2880 btrfs_super_generation(sb),
2881 btrfs_super_cache_generation(sb));
2887 * Validation of super block at mount time.
2888 * Some checks already done early at mount time, like csum type and incompat
2889 * flags will be skipped.
2891 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2893 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2897 * Validation of super block at write time.
2898 * Some checks like bytenr check will be skipped as their values will be
2900 * Extra checks like csum type and incompat flags will be done here.
2902 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2903 struct btrfs_super_block *sb)
2907 ret = btrfs_validate_super(fs_info, sb, -1);
2910 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2912 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2913 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2916 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2919 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2920 btrfs_super_incompat_flags(sb),
2921 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2927 "super block corruption detected before writing it to disk");
2931 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2933 struct btrfs_tree_parent_check check = {
2936 .owner_root = root->root_key.objectid
2940 root->node = read_tree_block(root->fs_info, bytenr, &check);
2941 if (IS_ERR(root->node)) {
2942 ret = PTR_ERR(root->node);
2946 if (!extent_buffer_uptodate(root->node)) {
2947 free_extent_buffer(root->node);
2952 btrfs_set_root_node(&root->root_item, root->node);
2953 root->commit_root = btrfs_root_node(root);
2954 btrfs_set_root_refs(&root->root_item, 1);
2958 static int load_important_roots(struct btrfs_fs_info *fs_info)
2960 struct btrfs_super_block *sb = fs_info->super_copy;
2964 bytenr = btrfs_super_root(sb);
2965 gen = btrfs_super_generation(sb);
2966 level = btrfs_super_root_level(sb);
2967 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2969 btrfs_warn(fs_info, "couldn't read tree root");
2975 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2977 int backup_index = find_newest_super_backup(fs_info);
2978 struct btrfs_super_block *sb = fs_info->super_copy;
2979 struct btrfs_root *tree_root = fs_info->tree_root;
2980 bool handle_error = false;
2984 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2986 if (!IS_ERR(tree_root->node))
2987 free_extent_buffer(tree_root->node);
2988 tree_root->node = NULL;
2990 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2993 free_root_pointers(fs_info, 0);
2996 * Don't use the log in recovery mode, it won't be
2999 btrfs_set_super_log_root(sb, 0);
3001 /* We can't trust the free space cache either */
3002 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3004 ret = read_backup_root(fs_info, i);
3010 ret = load_important_roots(fs_info);
3012 handle_error = true;
3017 * No need to hold btrfs_root::objectid_mutex since the fs
3018 * hasn't been fully initialised and we are the only user
3020 ret = btrfs_init_root_free_objectid(tree_root);
3022 handle_error = true;
3026 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
3028 ret = btrfs_read_roots(fs_info);
3030 handle_error = true;
3034 /* All successful */
3035 fs_info->generation = btrfs_header_generation(tree_root->node);
3036 fs_info->last_trans_committed = fs_info->generation;
3037 fs_info->last_reloc_trans = 0;
3039 /* Always begin writing backup roots after the one being used */
3040 if (backup_index < 0) {
3041 fs_info->backup_root_index = 0;
3043 fs_info->backup_root_index = backup_index + 1;
3044 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3052 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3054 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
3055 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
3056 INIT_LIST_HEAD(&fs_info->trans_list);
3057 INIT_LIST_HEAD(&fs_info->dead_roots);
3058 INIT_LIST_HEAD(&fs_info->delayed_iputs);
3059 INIT_LIST_HEAD(&fs_info->delalloc_roots);
3060 INIT_LIST_HEAD(&fs_info->caching_block_groups);
3061 spin_lock_init(&fs_info->delalloc_root_lock);
3062 spin_lock_init(&fs_info->trans_lock);
3063 spin_lock_init(&fs_info->fs_roots_radix_lock);
3064 spin_lock_init(&fs_info->delayed_iput_lock);
3065 spin_lock_init(&fs_info->defrag_inodes_lock);
3066 spin_lock_init(&fs_info->super_lock);
3067 spin_lock_init(&fs_info->buffer_lock);
3068 spin_lock_init(&fs_info->unused_bgs_lock);
3069 spin_lock_init(&fs_info->treelog_bg_lock);
3070 spin_lock_init(&fs_info->zone_active_bgs_lock);
3071 spin_lock_init(&fs_info->relocation_bg_lock);
3072 rwlock_init(&fs_info->tree_mod_log_lock);
3073 rwlock_init(&fs_info->global_root_lock);
3074 mutex_init(&fs_info->unused_bg_unpin_mutex);
3075 mutex_init(&fs_info->reclaim_bgs_lock);
3076 mutex_init(&fs_info->reloc_mutex);
3077 mutex_init(&fs_info->delalloc_root_mutex);
3078 mutex_init(&fs_info->zoned_meta_io_lock);
3079 mutex_init(&fs_info->zoned_data_reloc_io_lock);
3080 seqlock_init(&fs_info->profiles_lock);
3082 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
3083 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
3084 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
3085 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
3086 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
3087 BTRFS_LOCKDEP_TRANS_COMMIT_START);
3088 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
3089 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
3090 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
3091 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
3092 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
3093 BTRFS_LOCKDEP_TRANS_COMPLETED);
3095 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3096 INIT_LIST_HEAD(&fs_info->space_info);
3097 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3098 INIT_LIST_HEAD(&fs_info->unused_bgs);
3099 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3100 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3101 #ifdef CONFIG_BTRFS_DEBUG
3102 INIT_LIST_HEAD(&fs_info->allocated_roots);
3103 INIT_LIST_HEAD(&fs_info->allocated_ebs);
3104 spin_lock_init(&fs_info->eb_leak_lock);
3106 extent_map_tree_init(&fs_info->mapping_tree);
3107 btrfs_init_block_rsv(&fs_info->global_block_rsv,
3108 BTRFS_BLOCK_RSV_GLOBAL);
3109 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3110 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3111 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3112 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3113 BTRFS_BLOCK_RSV_DELOPS);
3114 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3115 BTRFS_BLOCK_RSV_DELREFS);
3117 atomic_set(&fs_info->async_delalloc_pages, 0);
3118 atomic_set(&fs_info->defrag_running, 0);
3119 atomic_set(&fs_info->nr_delayed_iputs, 0);
3120 atomic64_set(&fs_info->tree_mod_seq, 0);
3121 fs_info->global_root_tree = RB_ROOT;
3122 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3123 fs_info->metadata_ratio = 0;
3124 fs_info->defrag_inodes = RB_ROOT;
3125 atomic64_set(&fs_info->free_chunk_space, 0);
3126 fs_info->tree_mod_log = RB_ROOT;
3127 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3128 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3129 btrfs_init_ref_verify(fs_info);
3131 fs_info->thread_pool_size = min_t(unsigned long,
3132 num_online_cpus() + 2, 8);
3134 INIT_LIST_HEAD(&fs_info->ordered_roots);
3135 spin_lock_init(&fs_info->ordered_root_lock);
3137 btrfs_init_scrub(fs_info);
3138 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3139 fs_info->check_integrity_print_mask = 0;
3141 btrfs_init_balance(fs_info);
3142 btrfs_init_async_reclaim_work(fs_info);
3144 rwlock_init(&fs_info->block_group_cache_lock);
3145 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3147 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3148 IO_TREE_FS_EXCLUDED_EXTENTS);
3150 mutex_init(&fs_info->ordered_operations_mutex);
3151 mutex_init(&fs_info->tree_log_mutex);
3152 mutex_init(&fs_info->chunk_mutex);
3153 mutex_init(&fs_info->transaction_kthread_mutex);
3154 mutex_init(&fs_info->cleaner_mutex);
3155 mutex_init(&fs_info->ro_block_group_mutex);
3156 init_rwsem(&fs_info->commit_root_sem);
3157 init_rwsem(&fs_info->cleanup_work_sem);
3158 init_rwsem(&fs_info->subvol_sem);
3159 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3161 btrfs_init_dev_replace_locks(fs_info);
3162 btrfs_init_qgroup(fs_info);
3163 btrfs_discard_init(fs_info);
3165 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3166 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3168 init_waitqueue_head(&fs_info->transaction_throttle);
3169 init_waitqueue_head(&fs_info->transaction_wait);
3170 init_waitqueue_head(&fs_info->transaction_blocked_wait);
3171 init_waitqueue_head(&fs_info->async_submit_wait);
3172 init_waitqueue_head(&fs_info->delayed_iputs_wait);
3174 /* Usable values until the real ones are cached from the superblock */
3175 fs_info->nodesize = 4096;
3176 fs_info->sectorsize = 4096;
3177 fs_info->sectorsize_bits = ilog2(4096);
3178 fs_info->stripesize = 4096;
3180 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3182 spin_lock_init(&fs_info->swapfile_pins_lock);
3183 fs_info->swapfile_pins = RB_ROOT;
3185 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3186 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3189 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3194 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3195 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3197 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3201 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3205 fs_info->dirty_metadata_batch = PAGE_SIZE *
3206 (1 + ilog2(nr_cpu_ids));
3208 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3212 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3217 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3219 if (!fs_info->delayed_root)
3221 btrfs_init_delayed_root(fs_info->delayed_root);
3224 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3226 return btrfs_alloc_stripe_hash_table(fs_info);
3229 static int btrfs_uuid_rescan_kthread(void *data)
3231 struct btrfs_fs_info *fs_info = data;
3235 * 1st step is to iterate through the existing UUID tree and
3236 * to delete all entries that contain outdated data.
3237 * 2nd step is to add all missing entries to the UUID tree.
3239 ret = btrfs_uuid_tree_iterate(fs_info);
3242 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3244 up(&fs_info->uuid_tree_rescan_sem);
3247 return btrfs_uuid_scan_kthread(data);
3250 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3252 struct task_struct *task;
3254 down(&fs_info->uuid_tree_rescan_sem);
3255 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3257 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3258 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3259 up(&fs_info->uuid_tree_rescan_sem);
3260 return PTR_ERR(task);
3267 * Some options only have meaning at mount time and shouldn't persist across
3268 * remounts, or be displayed. Clear these at the end of mount and remount
3271 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3273 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3274 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3278 * Mounting logic specific to read-write file systems. Shared by open_ctree
3279 * and btrfs_remount when remounting from read-only to read-write.
3281 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3284 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3285 bool clear_free_space_tree = false;
3287 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3288 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3289 clear_free_space_tree = true;
3290 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3291 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3292 btrfs_warn(fs_info, "free space tree is invalid");
3293 clear_free_space_tree = true;
3296 if (clear_free_space_tree) {
3297 btrfs_info(fs_info, "clearing free space tree");
3298 ret = btrfs_clear_free_space_tree(fs_info);
3301 "failed to clear free space tree: %d", ret);
3307 * btrfs_find_orphan_roots() is responsible for finding all the dead
3308 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3309 * them into the fs_info->fs_roots_radix tree. This must be done before
3310 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3311 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3312 * item before the root's tree is deleted - this means that if we unmount
3313 * or crash before the deletion completes, on the next mount we will not
3314 * delete what remains of the tree because the orphan item does not
3315 * exists anymore, which is what tells us we have a pending deletion.
3317 ret = btrfs_find_orphan_roots(fs_info);
3321 ret = btrfs_cleanup_fs_roots(fs_info);
3325 down_read(&fs_info->cleanup_work_sem);
3326 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3327 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3328 up_read(&fs_info->cleanup_work_sem);
3331 up_read(&fs_info->cleanup_work_sem);
3333 mutex_lock(&fs_info->cleaner_mutex);
3334 ret = btrfs_recover_relocation(fs_info);
3335 mutex_unlock(&fs_info->cleaner_mutex);
3337 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3341 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3342 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3343 btrfs_info(fs_info, "creating free space tree");
3344 ret = btrfs_create_free_space_tree(fs_info);
3347 "failed to create free space tree: %d", ret);
3352 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3353 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3358 ret = btrfs_resume_balance_async(fs_info);
3362 ret = btrfs_resume_dev_replace_async(fs_info);
3364 btrfs_warn(fs_info, "failed to resume dev_replace");
3368 btrfs_qgroup_rescan_resume(fs_info);
3370 if (!fs_info->uuid_root) {
3371 btrfs_info(fs_info, "creating UUID tree");
3372 ret = btrfs_create_uuid_tree(fs_info);
3375 "failed to create the UUID tree %d", ret);
3385 * Do various sanity and dependency checks of different features.
3387 * @is_rw_mount: If the mount is read-write.
3389 * This is the place for less strict checks (like for subpage or artificial
3390 * feature dependencies).
3392 * For strict checks or possible corruption detection, see
3393 * btrfs_validate_super().
3395 * This should be called after btrfs_parse_options(), as some mount options
3396 * (space cache related) can modify on-disk format like free space tree and
3397 * screw up certain feature dependencies.
3399 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3401 struct btrfs_super_block *disk_super = fs_info->super_copy;
3402 u64 incompat = btrfs_super_incompat_flags(disk_super);
3403 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3404 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3406 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3408 "cannot mount because of unknown incompat features (0x%llx)",
3413 /* Runtime limitation for mixed block groups. */
3414 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3415 (fs_info->sectorsize != fs_info->nodesize)) {
3417 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3418 fs_info->nodesize, fs_info->sectorsize);
3422 /* Mixed backref is an always-enabled feature. */
3423 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3425 /* Set compression related flags just in case. */
3426 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3427 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3428 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3429 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3432 * An ancient flag, which should really be marked deprecated.
3433 * Such runtime limitation doesn't really need a incompat flag.
3435 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3436 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3438 if (compat_ro_unsupp && is_rw_mount) {
3440 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3446 * We have unsupported RO compat features, although RO mounted, we
3447 * should not cause any metadata writes, including log replay.
3448 * Or we could screw up whatever the new feature requires.
3450 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3451 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3453 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3459 * Artificial limitations for block group tree, to force
3460 * block-group-tree to rely on no-holes and free-space-tree.
3462 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3463 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3464 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3466 "block-group-tree feature requires no-holes and free-space-tree features");
3471 * Subpage runtime limitation on v1 cache.
3473 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3474 * we're already defaulting to v2 cache, no need to bother v1 as it's
3475 * going to be deprecated anyway.
3477 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3479 "v1 space cache is not supported for page size %lu with sectorsize %u",
3480 PAGE_SIZE, fs_info->sectorsize);
3484 /* This can be called by remount, we need to protect the super block. */
3485 spin_lock(&fs_info->super_lock);
3486 btrfs_set_super_incompat_flags(disk_super, incompat);
3487 spin_unlock(&fs_info->super_lock);
3492 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3501 struct btrfs_super_block *disk_super;
3502 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3503 struct btrfs_root *tree_root;
3504 struct btrfs_root *chunk_root;
3509 ret = init_mount_fs_info(fs_info, sb);
3515 /* These need to be init'ed before we start creating inodes and such. */
3516 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3518 fs_info->tree_root = tree_root;
3519 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3521 fs_info->chunk_root = chunk_root;
3522 if (!tree_root || !chunk_root) {
3527 fs_info->btree_inode = new_inode(sb);
3528 if (!fs_info->btree_inode) {
3532 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3533 btrfs_init_btree_inode(fs_info);
3535 invalidate_bdev(fs_devices->latest_dev->bdev);
3538 * Read super block and check the signature bytes only
3540 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3541 if (IS_ERR(disk_super)) {
3542 err = PTR_ERR(disk_super);
3547 * Verify the type first, if that or the checksum value are
3548 * corrupted, we'll find out
3550 csum_type = btrfs_super_csum_type(disk_super);
3551 if (!btrfs_supported_super_csum(csum_type)) {
3552 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3555 btrfs_release_disk_super(disk_super);
3559 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3561 ret = btrfs_init_csum_hash(fs_info, csum_type);
3564 btrfs_release_disk_super(disk_super);
3569 * We want to check superblock checksum, the type is stored inside.
3570 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3572 if (btrfs_check_super_csum(fs_info, disk_super)) {
3573 btrfs_err(fs_info, "superblock checksum mismatch");
3575 btrfs_release_disk_super(disk_super);
3580 * super_copy is zeroed at allocation time and we never touch the
3581 * following bytes up to INFO_SIZE, the checksum is calculated from
3582 * the whole block of INFO_SIZE
3584 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3585 btrfs_release_disk_super(disk_super);
3587 disk_super = fs_info->super_copy;
3590 features = btrfs_super_flags(disk_super);
3591 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3592 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3593 btrfs_set_super_flags(disk_super, features);
3595 "found metadata UUID change in progress flag, clearing");
3598 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3599 sizeof(*fs_info->super_for_commit));
3601 ret = btrfs_validate_mount_super(fs_info);
3603 btrfs_err(fs_info, "superblock contains fatal errors");
3608 if (!btrfs_super_root(disk_super))
3611 /* check FS state, whether FS is broken. */
3612 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3613 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3616 * In the long term, we'll store the compression type in the super
3617 * block, and it'll be used for per file compression control.
3619 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3622 /* Set up fs_info before parsing mount options */
3623 nodesize = btrfs_super_nodesize(disk_super);
3624 sectorsize = btrfs_super_sectorsize(disk_super);
3625 stripesize = sectorsize;
3626 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3627 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3629 fs_info->nodesize = nodesize;
3630 fs_info->sectorsize = sectorsize;
3631 fs_info->sectorsize_bits = ilog2(sectorsize);
3632 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3633 fs_info->stripesize = stripesize;
3635 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3641 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3647 if (sectorsize < PAGE_SIZE) {
3648 struct btrfs_subpage_info *subpage_info;
3651 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3652 * going to be deprecated.
3654 * Force to use v2 cache for subpage case.
3656 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3657 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3658 "forcing free space tree for sector size %u with page size %lu",
3659 sectorsize, PAGE_SIZE);
3662 "read-write for sector size %u with page size %lu is experimental",
3663 sectorsize, PAGE_SIZE);
3664 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3667 btrfs_init_subpage_info(subpage_info, sectorsize);
3668 fs_info->subpage_info = subpage_info;
3671 ret = btrfs_init_workqueues(fs_info);
3674 goto fail_sb_buffer;
3677 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3678 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3680 sb->s_blocksize = sectorsize;
3681 sb->s_blocksize_bits = blksize_bits(sectorsize);
3682 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3684 mutex_lock(&fs_info->chunk_mutex);
3685 ret = btrfs_read_sys_array(fs_info);
3686 mutex_unlock(&fs_info->chunk_mutex);
3688 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3689 goto fail_sb_buffer;
3692 generation = btrfs_super_chunk_root_generation(disk_super);
3693 level = btrfs_super_chunk_root_level(disk_super);
3694 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3697 btrfs_err(fs_info, "failed to read chunk root");
3698 goto fail_tree_roots;
3701 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3702 offsetof(struct btrfs_header, chunk_tree_uuid),
3705 ret = btrfs_read_chunk_tree(fs_info);
3707 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3708 goto fail_tree_roots;
3712 * At this point we know all the devices that make this filesystem,
3713 * including the seed devices but we don't know yet if the replace
3714 * target is required. So free devices that are not part of this
3715 * filesystem but skip the replace target device which is checked
3716 * below in btrfs_init_dev_replace().
3718 btrfs_free_extra_devids(fs_devices);
3719 if (!fs_devices->latest_dev->bdev) {
3720 btrfs_err(fs_info, "failed to read devices");
3721 goto fail_tree_roots;
3724 ret = init_tree_roots(fs_info);
3726 goto fail_tree_roots;
3729 * Get zone type information of zoned block devices. This will also
3730 * handle emulation of a zoned filesystem if a regular device has the
3731 * zoned incompat feature flag set.
3733 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3736 "zoned: failed to read device zone info: %d",
3738 goto fail_block_groups;
3742 * If we have a uuid root and we're not being told to rescan we need to
3743 * check the generation here so we can set the
3744 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3745 * transaction during a balance or the log replay without updating the
3746 * uuid generation, and then if we crash we would rescan the uuid tree,
3747 * even though it was perfectly fine.
3749 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3750 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3751 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3753 ret = btrfs_verify_dev_extents(fs_info);
3756 "failed to verify dev extents against chunks: %d",
3758 goto fail_block_groups;
3760 ret = btrfs_recover_balance(fs_info);
3762 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3763 goto fail_block_groups;
3766 ret = btrfs_init_dev_stats(fs_info);
3768 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3769 goto fail_block_groups;
3772 ret = btrfs_init_dev_replace(fs_info);
3774 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3775 goto fail_block_groups;
3778 ret = btrfs_check_zoned_mode(fs_info);
3780 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3782 goto fail_block_groups;
3785 ret = btrfs_sysfs_add_fsid(fs_devices);
3787 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3789 goto fail_block_groups;
3792 ret = btrfs_sysfs_add_mounted(fs_info);
3794 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3795 goto fail_fsdev_sysfs;
3798 ret = btrfs_init_space_info(fs_info);
3800 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3804 ret = btrfs_read_block_groups(fs_info);
3806 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3810 btrfs_free_zone_cache(fs_info);
3812 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3813 !btrfs_check_rw_degradable(fs_info, NULL)) {
3815 "writable mount is not allowed due to too many missing devices");
3819 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3821 if (IS_ERR(fs_info->cleaner_kthread))
3824 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3826 "btrfs-transaction");
3827 if (IS_ERR(fs_info->transaction_kthread))
3830 if (!btrfs_test_opt(fs_info, NOSSD) &&
3831 !fs_info->fs_devices->rotating) {
3832 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3836 * For devices supporting discard turn on discard=async automatically,
3837 * unless it's already set or disabled. This could be turned off by
3838 * nodiscard for the same mount.
3840 if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3841 btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3842 btrfs_test_opt(fs_info, NODISCARD)) &&
3843 fs_info->fs_devices->discardable) {
3844 btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3845 "auto enabling async discard");
3846 btrfs_clear_opt(fs_info->mount_opt, NODISCARD);
3849 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3850 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3851 ret = btrfsic_mount(fs_info, fs_devices,
3852 btrfs_test_opt(fs_info,
3853 CHECK_INTEGRITY_DATA) ? 1 : 0,
3854 fs_info->check_integrity_print_mask);
3857 "failed to initialize integrity check module: %d",
3861 ret = btrfs_read_qgroup_config(fs_info);
3863 goto fail_trans_kthread;
3865 if (btrfs_build_ref_tree(fs_info))
3866 btrfs_err(fs_info, "couldn't build ref tree");
3868 /* do not make disk changes in broken FS or nologreplay is given */
3869 if (btrfs_super_log_root(disk_super) != 0 &&
3870 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3871 btrfs_info(fs_info, "start tree-log replay");
3872 ret = btrfs_replay_log(fs_info, fs_devices);
3879 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3880 if (IS_ERR(fs_info->fs_root)) {
3881 err = PTR_ERR(fs_info->fs_root);
3882 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3883 fs_info->fs_root = NULL;
3890 ret = btrfs_start_pre_rw_mount(fs_info);
3892 close_ctree(fs_info);
3895 btrfs_discard_resume(fs_info);
3897 if (fs_info->uuid_root &&
3898 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3899 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3900 btrfs_info(fs_info, "checking UUID tree");
3901 ret = btrfs_check_uuid_tree(fs_info);
3904 "failed to check the UUID tree: %d", ret);
3905 close_ctree(fs_info);
3910 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3912 /* Kick the cleaner thread so it'll start deleting snapshots. */
3913 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3914 wake_up_process(fs_info->cleaner_kthread);
3917 btrfs_clear_oneshot_options(fs_info);
3921 btrfs_free_qgroup_config(fs_info);
3923 kthread_stop(fs_info->transaction_kthread);
3924 btrfs_cleanup_transaction(fs_info);
3925 btrfs_free_fs_roots(fs_info);
3927 kthread_stop(fs_info->cleaner_kthread);
3930 * make sure we're done with the btree inode before we stop our
3933 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3936 btrfs_sysfs_remove_mounted(fs_info);
3939 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3942 btrfs_put_block_group_cache(fs_info);
3945 if (fs_info->data_reloc_root)
3946 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3947 free_root_pointers(fs_info, true);
3948 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3951 btrfs_stop_all_workers(fs_info);
3952 btrfs_free_block_groups(fs_info);
3954 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3956 iput(fs_info->btree_inode);
3958 btrfs_close_devices(fs_info->fs_devices);
3961 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3963 static void btrfs_end_super_write(struct bio *bio)
3965 struct btrfs_device *device = bio->bi_private;
3966 struct bio_vec *bvec;
3967 struct bvec_iter_all iter_all;
3970 bio_for_each_segment_all(bvec, bio, iter_all) {
3971 page = bvec->bv_page;
3973 if (bio->bi_status) {
3974 btrfs_warn_rl_in_rcu(device->fs_info,
3975 "lost page write due to IO error on %s (%d)",
3976 btrfs_dev_name(device),
3977 blk_status_to_errno(bio->bi_status));
3978 ClearPageUptodate(page);
3980 btrfs_dev_stat_inc_and_print(device,
3981 BTRFS_DEV_STAT_WRITE_ERRS);
3983 SetPageUptodate(page);
3993 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3994 int copy_num, bool drop_cache)
3996 struct btrfs_super_block *super;
3998 u64 bytenr, bytenr_orig;
3999 struct address_space *mapping = bdev->bd_inode->i_mapping;
4002 bytenr_orig = btrfs_sb_offset(copy_num);
4003 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
4005 return ERR_PTR(-EINVAL);
4007 return ERR_PTR(ret);
4009 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
4010 return ERR_PTR(-EINVAL);
4013 /* This should only be called with the primary sb. */
4014 ASSERT(copy_num == 0);
4017 * Drop the page of the primary superblock, so later read will
4018 * always read from the device.
4020 invalidate_inode_pages2_range(mapping,
4021 bytenr >> PAGE_SHIFT,
4022 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
4025 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
4027 return ERR_CAST(page);
4029 super = page_address(page);
4030 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
4031 btrfs_release_disk_super(super);
4032 return ERR_PTR(-ENODATA);
4035 if (btrfs_super_bytenr(super) != bytenr_orig) {
4036 btrfs_release_disk_super(super);
4037 return ERR_PTR(-EINVAL);
4044 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
4046 struct btrfs_super_block *super, *latest = NULL;
4050 /* we would like to check all the supers, but that would make
4051 * a btrfs mount succeed after a mkfs from a different FS.
4052 * So, we need to add a special mount option to scan for
4053 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
4055 for (i = 0; i < 1; i++) {
4056 super = btrfs_read_dev_one_super(bdev, i, false);
4060 if (!latest || btrfs_super_generation(super) > transid) {
4062 btrfs_release_disk_super(super);
4065 transid = btrfs_super_generation(super);
4073 * Write superblock @sb to the @device. Do not wait for completion, all the
4074 * pages we use for writing are locked.
4076 * Write @max_mirrors copies of the superblock, where 0 means default that fit
4077 * the expected device size at commit time. Note that max_mirrors must be
4078 * same for write and wait phases.
4080 * Return number of errors when page is not found or submission fails.
4082 static int write_dev_supers(struct btrfs_device *device,
4083 struct btrfs_super_block *sb, int max_mirrors)
4085 struct btrfs_fs_info *fs_info = device->fs_info;
4086 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
4087 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4091 u64 bytenr, bytenr_orig;
4093 if (max_mirrors == 0)
4094 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4096 shash->tfm = fs_info->csum_shash;
4098 for (i = 0; i < max_mirrors; i++) {
4101 struct btrfs_super_block *disk_super;
4103 bytenr_orig = btrfs_sb_offset(i);
4104 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4105 if (ret == -ENOENT) {
4107 } else if (ret < 0) {
4108 btrfs_err(device->fs_info,
4109 "couldn't get super block location for mirror %d",
4114 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4115 device->commit_total_bytes)
4118 btrfs_set_super_bytenr(sb, bytenr_orig);
4120 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4121 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4124 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4127 btrfs_err(device->fs_info,
4128 "couldn't get super block page for bytenr %llu",
4134 /* Bump the refcount for wait_dev_supers() */
4137 disk_super = page_address(page);
4138 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4141 * Directly use bios here instead of relying on the page cache
4142 * to do I/O, so we don't lose the ability to do integrity
4145 bio = bio_alloc(device->bdev, 1,
4146 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4148 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4149 bio->bi_private = device;
4150 bio->bi_end_io = btrfs_end_super_write;
4151 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4152 offset_in_page(bytenr));
4155 * We FUA only the first super block. The others we allow to
4156 * go down lazy and there's a short window where the on-disk
4157 * copies might still contain the older version.
4159 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4160 bio->bi_opf |= REQ_FUA;
4162 btrfsic_check_bio(bio);
4165 if (btrfs_advance_sb_log(device, i))
4168 return errors < i ? 0 : -1;
4172 * Wait for write completion of superblocks done by write_dev_supers,
4173 * @max_mirrors same for write and wait phases.
4175 * Return number of errors when page is not found or not marked up to
4178 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4182 bool primary_failed = false;
4186 if (max_mirrors == 0)
4187 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4189 for (i = 0; i < max_mirrors; i++) {
4192 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4193 if (ret == -ENOENT) {
4195 } else if (ret < 0) {
4198 primary_failed = true;
4201 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4202 device->commit_total_bytes)
4205 page = find_get_page(device->bdev->bd_inode->i_mapping,
4206 bytenr >> PAGE_SHIFT);
4210 primary_failed = true;
4213 /* Page is submitted locked and unlocked once the IO completes */
4214 wait_on_page_locked(page);
4215 if (PageError(page)) {
4218 primary_failed = true;
4221 /* Drop our reference */
4224 /* Drop the reference from the writing run */
4228 /* log error, force error return */
4229 if (primary_failed) {
4230 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4235 return errors < i ? 0 : -1;
4239 * endio for the write_dev_flush, this will wake anyone waiting
4240 * for the barrier when it is done
4242 static void btrfs_end_empty_barrier(struct bio *bio)
4245 complete(bio->bi_private);
4249 * Submit a flush request to the device if it supports it. Error handling is
4250 * done in the waiting counterpart.
4252 static void write_dev_flush(struct btrfs_device *device)
4254 struct bio *bio = &device->flush_bio;
4256 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4258 * When a disk has write caching disabled, we skip submission of a bio
4259 * with flush and sync requests before writing the superblock, since
4260 * it's not needed. However when the integrity checker is enabled, this
4261 * results in reports that there are metadata blocks referred by a
4262 * superblock that were not properly flushed. So don't skip the bio
4263 * submission only when the integrity checker is enabled for the sake
4264 * of simplicity, since this is a debug tool and not meant for use in
4267 if (!bdev_write_cache(device->bdev))
4271 bio_init(bio, device->bdev, NULL, 0,
4272 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4273 bio->bi_end_io = btrfs_end_empty_barrier;
4274 init_completion(&device->flush_wait);
4275 bio->bi_private = &device->flush_wait;
4277 btrfsic_check_bio(bio);
4279 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4283 * If the flush bio has been submitted by write_dev_flush, wait for it.
4285 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4287 struct bio *bio = &device->flush_bio;
4289 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4292 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4293 wait_for_completion_io(&device->flush_wait);
4295 return bio->bi_status;
4298 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4300 if (!btrfs_check_rw_degradable(fs_info, NULL))
4306 * send an empty flush down to each device in parallel,
4307 * then wait for them
4309 static int barrier_all_devices(struct btrfs_fs_info *info)
4311 struct list_head *head;
4312 struct btrfs_device *dev;
4313 int errors_wait = 0;
4316 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4317 /* send down all the barriers */
4318 head = &info->fs_devices->devices;
4319 list_for_each_entry(dev, head, dev_list) {
4320 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4324 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4325 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4328 write_dev_flush(dev);
4329 dev->last_flush_error = BLK_STS_OK;
4332 /* wait for all the barriers */
4333 list_for_each_entry(dev, head, dev_list) {
4334 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4340 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4341 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4344 ret = wait_dev_flush(dev);
4346 dev->last_flush_error = ret;
4347 btrfs_dev_stat_inc_and_print(dev,
4348 BTRFS_DEV_STAT_FLUSH_ERRS);
4355 * At some point we need the status of all disks
4356 * to arrive at the volume status. So error checking
4357 * is being pushed to a separate loop.
4359 return check_barrier_error(info);
4364 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4367 int min_tolerated = INT_MAX;
4369 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4370 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4371 min_tolerated = min_t(int, min_tolerated,
4372 btrfs_raid_array[BTRFS_RAID_SINGLE].
4373 tolerated_failures);
4375 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4376 if (raid_type == BTRFS_RAID_SINGLE)
4378 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4380 min_tolerated = min_t(int, min_tolerated,
4381 btrfs_raid_array[raid_type].
4382 tolerated_failures);
4385 if (min_tolerated == INT_MAX) {
4386 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4390 return min_tolerated;
4393 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4395 struct list_head *head;
4396 struct btrfs_device *dev;
4397 struct btrfs_super_block *sb;
4398 struct btrfs_dev_item *dev_item;
4402 int total_errors = 0;
4405 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4408 * max_mirrors == 0 indicates we're from commit_transaction,
4409 * not from fsync where the tree roots in fs_info have not
4410 * been consistent on disk.
4412 if (max_mirrors == 0)
4413 backup_super_roots(fs_info);
4415 sb = fs_info->super_for_commit;
4416 dev_item = &sb->dev_item;
4418 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4419 head = &fs_info->fs_devices->devices;
4420 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4423 ret = barrier_all_devices(fs_info);
4426 &fs_info->fs_devices->device_list_mutex);
4427 btrfs_handle_fs_error(fs_info, ret,
4428 "errors while submitting device barriers.");
4433 list_for_each_entry(dev, head, dev_list) {
4438 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4439 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4442 btrfs_set_stack_device_generation(dev_item, 0);
4443 btrfs_set_stack_device_type(dev_item, dev->type);
4444 btrfs_set_stack_device_id(dev_item, dev->devid);
4445 btrfs_set_stack_device_total_bytes(dev_item,
4446 dev->commit_total_bytes);
4447 btrfs_set_stack_device_bytes_used(dev_item,
4448 dev->commit_bytes_used);
4449 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4450 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4451 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4452 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4453 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4456 flags = btrfs_super_flags(sb);
4457 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4459 ret = btrfs_validate_write_super(fs_info, sb);
4461 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4462 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4463 "unexpected superblock corruption detected");
4467 ret = write_dev_supers(dev, sb, max_mirrors);
4471 if (total_errors > max_errors) {
4472 btrfs_err(fs_info, "%d errors while writing supers",
4474 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4476 /* FUA is masked off if unsupported and can't be the reason */
4477 btrfs_handle_fs_error(fs_info, -EIO,
4478 "%d errors while writing supers",
4484 list_for_each_entry(dev, head, dev_list) {
4487 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4488 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4491 ret = wait_dev_supers(dev, max_mirrors);
4495 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4496 if (total_errors > max_errors) {
4497 btrfs_handle_fs_error(fs_info, -EIO,
4498 "%d errors while writing supers",
4505 /* Drop a fs root from the radix tree and free it. */
4506 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4507 struct btrfs_root *root)
4509 bool drop_ref = false;
4511 spin_lock(&fs_info->fs_roots_radix_lock);
4512 radix_tree_delete(&fs_info->fs_roots_radix,
4513 (unsigned long)root->root_key.objectid);
4514 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4516 spin_unlock(&fs_info->fs_roots_radix_lock);
4518 if (BTRFS_FS_ERROR(fs_info)) {
4519 ASSERT(root->log_root == NULL);
4520 if (root->reloc_root) {
4521 btrfs_put_root(root->reloc_root);
4522 root->reloc_root = NULL;
4527 btrfs_put_root(root);
4530 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4532 u64 root_objectid = 0;
4533 struct btrfs_root *gang[8];
4536 unsigned int ret = 0;
4539 spin_lock(&fs_info->fs_roots_radix_lock);
4540 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4541 (void **)gang, root_objectid,
4544 spin_unlock(&fs_info->fs_roots_radix_lock);
4547 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4549 for (i = 0; i < ret; i++) {
4550 /* Avoid to grab roots in dead_roots */
4551 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4555 /* grab all the search result for later use */
4556 gang[i] = btrfs_grab_root(gang[i]);
4558 spin_unlock(&fs_info->fs_roots_radix_lock);
4560 for (i = 0; i < ret; i++) {
4563 root_objectid = gang[i]->root_key.objectid;
4564 err = btrfs_orphan_cleanup(gang[i]);
4567 btrfs_put_root(gang[i]);
4572 /* release the uncleaned roots due to error */
4573 for (; i < ret; i++) {
4575 btrfs_put_root(gang[i]);
4580 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4582 struct btrfs_root *root = fs_info->tree_root;
4583 struct btrfs_trans_handle *trans;
4585 mutex_lock(&fs_info->cleaner_mutex);
4586 btrfs_run_delayed_iputs(fs_info);
4587 mutex_unlock(&fs_info->cleaner_mutex);
4588 wake_up_process(fs_info->cleaner_kthread);
4590 /* wait until ongoing cleanup work done */
4591 down_write(&fs_info->cleanup_work_sem);
4592 up_write(&fs_info->cleanup_work_sem);
4594 trans = btrfs_join_transaction(root);
4596 return PTR_ERR(trans);
4597 return btrfs_commit_transaction(trans);
4600 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4602 struct btrfs_transaction *trans;
4603 struct btrfs_transaction *tmp;
4606 if (list_empty(&fs_info->trans_list))
4610 * This function is only called at the very end of close_ctree(),
4611 * thus no other running transaction, no need to take trans_lock.
4613 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4614 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4615 struct extent_state *cached = NULL;
4616 u64 dirty_bytes = 0;
4622 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4623 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4624 dirty_bytes += found_end + 1 - found_start;
4625 cur = found_end + 1;
4628 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4629 trans->transid, dirty_bytes);
4630 btrfs_cleanup_one_transaction(trans, fs_info);
4632 if (trans == fs_info->running_transaction)
4633 fs_info->running_transaction = NULL;
4634 list_del_init(&trans->list);
4636 btrfs_put_transaction(trans);
4637 trace_btrfs_transaction_commit(fs_info);
4642 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4646 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4649 * If we had UNFINISHED_DROPS we could still be processing them, so
4650 * clear that bit and wake up relocation so it can stop.
4651 * We must do this before stopping the block group reclaim task, because
4652 * at btrfs_relocate_block_group() we wait for this bit, and after the
4653 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4654 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4657 btrfs_wake_unfinished_drop(fs_info);
4660 * We may have the reclaim task running and relocating a data block group,
4661 * in which case it may create delayed iputs. So stop it before we park
4662 * the cleaner kthread otherwise we can get new delayed iputs after
4663 * parking the cleaner, and that can make the async reclaim task to hang
4664 * if it's waiting for delayed iputs to complete, since the cleaner is
4665 * parked and can not run delayed iputs - this will make us hang when
4666 * trying to stop the async reclaim task.
4668 cancel_work_sync(&fs_info->reclaim_bgs_work);
4670 * We don't want the cleaner to start new transactions, add more delayed
4671 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4672 * because that frees the task_struct, and the transaction kthread might
4673 * still try to wake up the cleaner.
4675 kthread_park(fs_info->cleaner_kthread);
4677 /* wait for the qgroup rescan worker to stop */
4678 btrfs_qgroup_wait_for_completion(fs_info, false);
4680 /* wait for the uuid_scan task to finish */
4681 down(&fs_info->uuid_tree_rescan_sem);
4682 /* avoid complains from lockdep et al., set sem back to initial state */
4683 up(&fs_info->uuid_tree_rescan_sem);
4685 /* pause restriper - we want to resume on mount */
4686 btrfs_pause_balance(fs_info);
4688 btrfs_dev_replace_suspend_for_unmount(fs_info);
4690 btrfs_scrub_cancel(fs_info);
4692 /* wait for any defraggers to finish */
4693 wait_event(fs_info->transaction_wait,
4694 (atomic_read(&fs_info->defrag_running) == 0));
4696 /* clear out the rbtree of defraggable inodes */
4697 btrfs_cleanup_defrag_inodes(fs_info);
4700 * After we parked the cleaner kthread, ordered extents may have
4701 * completed and created new delayed iputs. If one of the async reclaim
4702 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4703 * can hang forever trying to stop it, because if a delayed iput is
4704 * added after it ran btrfs_run_delayed_iputs() and before it called
4705 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4706 * no one else to run iputs.
4708 * So wait for all ongoing ordered extents to complete and then run
4709 * delayed iputs. This works because once we reach this point no one
4710 * can either create new ordered extents nor create delayed iputs
4711 * through some other means.
4713 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4714 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4715 * but the delayed iput for the respective inode is made only when doing
4716 * the final btrfs_put_ordered_extent() (which must happen at
4717 * btrfs_finish_ordered_io() when we are unmounting).
4719 btrfs_flush_workqueue(fs_info->endio_write_workers);
4720 /* Ordered extents for free space inodes. */
4721 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4722 btrfs_run_delayed_iputs(fs_info);
4724 cancel_work_sync(&fs_info->async_reclaim_work);
4725 cancel_work_sync(&fs_info->async_data_reclaim_work);
4726 cancel_work_sync(&fs_info->preempt_reclaim_work);
4728 /* Cancel or finish ongoing discard work */
4729 btrfs_discard_cleanup(fs_info);
4731 if (!sb_rdonly(fs_info->sb)) {
4733 * The cleaner kthread is stopped, so do one final pass over
4734 * unused block groups.
4736 btrfs_delete_unused_bgs(fs_info);
4739 * There might be existing delayed inode workers still running
4740 * and holding an empty delayed inode item. We must wait for
4741 * them to complete first because they can create a transaction.
4742 * This happens when someone calls btrfs_balance_delayed_items()
4743 * and then a transaction commit runs the same delayed nodes
4744 * before any delayed worker has done something with the nodes.
4745 * We must wait for any worker here and not at transaction
4746 * commit time since that could cause a deadlock.
4747 * This is a very rare case.
4749 btrfs_flush_workqueue(fs_info->delayed_workers);
4751 ret = btrfs_commit_super(fs_info);
4753 btrfs_err(fs_info, "commit super ret %d", ret);
4756 if (BTRFS_FS_ERROR(fs_info))
4757 btrfs_error_commit_super(fs_info);
4759 kthread_stop(fs_info->transaction_kthread);
4760 kthread_stop(fs_info->cleaner_kthread);
4762 ASSERT(list_empty(&fs_info->delayed_iputs));
4763 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4765 if (btrfs_check_quota_leak(fs_info)) {
4766 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4767 btrfs_err(fs_info, "qgroup reserved space leaked");
4770 btrfs_free_qgroup_config(fs_info);
4771 ASSERT(list_empty(&fs_info->delalloc_roots));
4773 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4774 btrfs_info(fs_info, "at unmount delalloc count %lld",
4775 percpu_counter_sum(&fs_info->delalloc_bytes));
4778 if (percpu_counter_sum(&fs_info->ordered_bytes))
4779 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4780 percpu_counter_sum(&fs_info->ordered_bytes));
4782 btrfs_sysfs_remove_mounted(fs_info);
4783 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4785 btrfs_put_block_group_cache(fs_info);
4788 * we must make sure there is not any read request to
4789 * submit after we stopping all workers.
4791 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4792 btrfs_stop_all_workers(fs_info);
4794 /* We shouldn't have any transaction open at this point */
4795 warn_about_uncommitted_trans(fs_info);
4797 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4798 free_root_pointers(fs_info, true);
4799 btrfs_free_fs_roots(fs_info);
4802 * We must free the block groups after dropping the fs_roots as we could
4803 * have had an IO error and have left over tree log blocks that aren't
4804 * cleaned up until the fs roots are freed. This makes the block group
4805 * accounting appear to be wrong because there's pending reserved bytes,
4806 * so make sure we do the block group cleanup afterwards.
4808 btrfs_free_block_groups(fs_info);
4810 iput(fs_info->btree_inode);
4812 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4813 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4814 btrfsic_unmount(fs_info->fs_devices);
4817 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4818 btrfs_close_devices(fs_info->fs_devices);
4821 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4825 struct inode *btree_inode = buf->pages[0]->mapping->host;
4827 ret = extent_buffer_uptodate(buf);
4831 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4832 parent_transid, atomic);
4838 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4840 struct btrfs_fs_info *fs_info = buf->fs_info;
4841 u64 transid = btrfs_header_generation(buf);
4844 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4846 * This is a fast path so only do this check if we have sanity tests
4847 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4848 * outside of the sanity tests.
4850 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4853 btrfs_assert_tree_write_locked(buf);
4854 if (transid != fs_info->generation)
4855 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4856 buf->start, transid, fs_info->generation);
4857 was_dirty = set_extent_buffer_dirty(buf);
4859 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4861 fs_info->dirty_metadata_batch);
4862 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4864 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4865 * but item data not updated.
4866 * So here we should only check item pointers, not item data.
4868 if (btrfs_header_level(buf) == 0 &&
4869 btrfs_check_leaf_relaxed(buf)) {
4870 btrfs_print_leaf(buf);
4876 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4880 * looks as though older kernels can get into trouble with
4881 * this code, they end up stuck in balance_dirty_pages forever
4885 if (current->flags & PF_MEMALLOC)
4889 btrfs_balance_delayed_items(fs_info);
4891 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4892 BTRFS_DIRTY_METADATA_THRESH,
4893 fs_info->dirty_metadata_batch);
4895 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4899 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4901 __btrfs_btree_balance_dirty(fs_info, 1);
4904 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4906 __btrfs_btree_balance_dirty(fs_info, 0);
4909 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4911 /* cleanup FS via transaction */
4912 btrfs_cleanup_transaction(fs_info);
4914 mutex_lock(&fs_info->cleaner_mutex);
4915 btrfs_run_delayed_iputs(fs_info);
4916 mutex_unlock(&fs_info->cleaner_mutex);
4918 down_write(&fs_info->cleanup_work_sem);
4919 up_write(&fs_info->cleanup_work_sem);
4922 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4924 struct btrfs_root *gang[8];
4925 u64 root_objectid = 0;
4928 spin_lock(&fs_info->fs_roots_radix_lock);
4929 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4930 (void **)gang, root_objectid,
4931 ARRAY_SIZE(gang))) != 0) {
4934 for (i = 0; i < ret; i++)
4935 gang[i] = btrfs_grab_root(gang[i]);
4936 spin_unlock(&fs_info->fs_roots_radix_lock);
4938 for (i = 0; i < ret; i++) {
4941 root_objectid = gang[i]->root_key.objectid;
4942 btrfs_free_log(NULL, gang[i]);
4943 btrfs_put_root(gang[i]);
4946 spin_lock(&fs_info->fs_roots_radix_lock);
4948 spin_unlock(&fs_info->fs_roots_radix_lock);
4949 btrfs_free_log_root_tree(NULL, fs_info);
4952 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4954 struct btrfs_ordered_extent *ordered;
4956 spin_lock(&root->ordered_extent_lock);
4958 * This will just short circuit the ordered completion stuff which will
4959 * make sure the ordered extent gets properly cleaned up.
4961 list_for_each_entry(ordered, &root->ordered_extents,
4963 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4964 spin_unlock(&root->ordered_extent_lock);
4967 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4969 struct btrfs_root *root;
4970 struct list_head splice;
4972 INIT_LIST_HEAD(&splice);
4974 spin_lock(&fs_info->ordered_root_lock);
4975 list_splice_init(&fs_info->ordered_roots, &splice);
4976 while (!list_empty(&splice)) {
4977 root = list_first_entry(&splice, struct btrfs_root,
4979 list_move_tail(&root->ordered_root,
4980 &fs_info->ordered_roots);
4982 spin_unlock(&fs_info->ordered_root_lock);
4983 btrfs_destroy_ordered_extents(root);
4986 spin_lock(&fs_info->ordered_root_lock);
4988 spin_unlock(&fs_info->ordered_root_lock);
4991 * We need this here because if we've been flipped read-only we won't
4992 * get sync() from the umount, so we need to make sure any ordered
4993 * extents that haven't had their dirty pages IO start writeout yet
4994 * actually get run and error out properly.
4996 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4999 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
5000 struct btrfs_fs_info *fs_info)
5002 struct rb_node *node;
5003 struct btrfs_delayed_ref_root *delayed_refs;
5004 struct btrfs_delayed_ref_node *ref;
5007 delayed_refs = &trans->delayed_refs;
5009 spin_lock(&delayed_refs->lock);
5010 if (atomic_read(&delayed_refs->num_entries) == 0) {
5011 spin_unlock(&delayed_refs->lock);
5012 btrfs_debug(fs_info, "delayed_refs has NO entry");
5016 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
5017 struct btrfs_delayed_ref_head *head;
5019 bool pin_bytes = false;
5021 head = rb_entry(node, struct btrfs_delayed_ref_head,
5023 if (btrfs_delayed_ref_lock(delayed_refs, head))
5026 spin_lock(&head->lock);
5027 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
5028 ref = rb_entry(n, struct btrfs_delayed_ref_node,
5031 rb_erase_cached(&ref->ref_node, &head->ref_tree);
5032 RB_CLEAR_NODE(&ref->ref_node);
5033 if (!list_empty(&ref->add_list))
5034 list_del(&ref->add_list);
5035 atomic_dec(&delayed_refs->num_entries);
5036 btrfs_put_delayed_ref(ref);
5038 if (head->must_insert_reserved)
5040 btrfs_free_delayed_extent_op(head->extent_op);
5041 btrfs_delete_ref_head(delayed_refs, head);
5042 spin_unlock(&head->lock);
5043 spin_unlock(&delayed_refs->lock);
5044 mutex_unlock(&head->mutex);
5047 struct btrfs_block_group *cache;
5049 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
5052 spin_lock(&cache->space_info->lock);
5053 spin_lock(&cache->lock);
5054 cache->pinned += head->num_bytes;
5055 btrfs_space_info_update_bytes_pinned(fs_info,
5056 cache->space_info, head->num_bytes);
5057 cache->reserved -= head->num_bytes;
5058 cache->space_info->bytes_reserved -= head->num_bytes;
5059 spin_unlock(&cache->lock);
5060 spin_unlock(&cache->space_info->lock);
5062 btrfs_put_block_group(cache);
5064 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
5065 head->bytenr + head->num_bytes - 1);
5067 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
5068 btrfs_put_delayed_ref_head(head);
5070 spin_lock(&delayed_refs->lock);
5072 btrfs_qgroup_destroy_extent_records(trans);
5074 spin_unlock(&delayed_refs->lock);
5079 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
5081 struct btrfs_inode *btrfs_inode;
5082 struct list_head splice;
5084 INIT_LIST_HEAD(&splice);
5086 spin_lock(&root->delalloc_lock);
5087 list_splice_init(&root->delalloc_inodes, &splice);
5089 while (!list_empty(&splice)) {
5090 struct inode *inode = NULL;
5091 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5093 __btrfs_del_delalloc_inode(root, btrfs_inode);
5094 spin_unlock(&root->delalloc_lock);
5097 * Make sure we get a live inode and that it'll not disappear
5100 inode = igrab(&btrfs_inode->vfs_inode);
5102 invalidate_inode_pages2(inode->i_mapping);
5105 spin_lock(&root->delalloc_lock);
5107 spin_unlock(&root->delalloc_lock);
5110 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5112 struct btrfs_root *root;
5113 struct list_head splice;
5115 INIT_LIST_HEAD(&splice);
5117 spin_lock(&fs_info->delalloc_root_lock);
5118 list_splice_init(&fs_info->delalloc_roots, &splice);
5119 while (!list_empty(&splice)) {
5120 root = list_first_entry(&splice, struct btrfs_root,
5122 root = btrfs_grab_root(root);
5124 spin_unlock(&fs_info->delalloc_root_lock);
5126 btrfs_destroy_delalloc_inodes(root);
5127 btrfs_put_root(root);
5129 spin_lock(&fs_info->delalloc_root_lock);
5131 spin_unlock(&fs_info->delalloc_root_lock);
5134 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5135 struct extent_io_tree *dirty_pages,
5139 struct extent_buffer *eb;
5144 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5149 clear_extent_bits(dirty_pages, start, end, mark);
5150 while (start <= end) {
5151 eb = find_extent_buffer(fs_info, start);
5152 start += fs_info->nodesize;
5155 wait_on_extent_buffer_writeback(eb);
5157 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5159 clear_extent_buffer_dirty(eb);
5160 free_extent_buffer_stale(eb);
5167 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5168 struct extent_io_tree *unpin)
5175 struct extent_state *cached_state = NULL;
5178 * The btrfs_finish_extent_commit() may get the same range as
5179 * ours between find_first_extent_bit and clear_extent_dirty.
5180 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5181 * the same extent range.
5183 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5184 ret = find_first_extent_bit(unpin, 0, &start, &end,
5185 EXTENT_DIRTY, &cached_state);
5187 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5191 clear_extent_dirty(unpin, start, end, &cached_state);
5192 free_extent_state(cached_state);
5193 btrfs_error_unpin_extent_range(fs_info, start, end);
5194 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5201 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5203 struct inode *inode;
5205 inode = cache->io_ctl.inode;
5207 invalidate_inode_pages2(inode->i_mapping);
5208 BTRFS_I(inode)->generation = 0;
5209 cache->io_ctl.inode = NULL;
5212 ASSERT(cache->io_ctl.pages == NULL);
5213 btrfs_put_block_group(cache);
5216 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5217 struct btrfs_fs_info *fs_info)
5219 struct btrfs_block_group *cache;
5221 spin_lock(&cur_trans->dirty_bgs_lock);
5222 while (!list_empty(&cur_trans->dirty_bgs)) {
5223 cache = list_first_entry(&cur_trans->dirty_bgs,
5224 struct btrfs_block_group,
5227 if (!list_empty(&cache->io_list)) {
5228 spin_unlock(&cur_trans->dirty_bgs_lock);
5229 list_del_init(&cache->io_list);
5230 btrfs_cleanup_bg_io(cache);
5231 spin_lock(&cur_trans->dirty_bgs_lock);
5234 list_del_init(&cache->dirty_list);
5235 spin_lock(&cache->lock);
5236 cache->disk_cache_state = BTRFS_DC_ERROR;
5237 spin_unlock(&cache->lock);
5239 spin_unlock(&cur_trans->dirty_bgs_lock);
5240 btrfs_put_block_group(cache);
5241 btrfs_delayed_refs_rsv_release(fs_info, 1);
5242 spin_lock(&cur_trans->dirty_bgs_lock);
5244 spin_unlock(&cur_trans->dirty_bgs_lock);
5247 * Refer to the definition of io_bgs member for details why it's safe
5248 * to use it without any locking
5250 while (!list_empty(&cur_trans->io_bgs)) {
5251 cache = list_first_entry(&cur_trans->io_bgs,
5252 struct btrfs_block_group,
5255 list_del_init(&cache->io_list);
5256 spin_lock(&cache->lock);
5257 cache->disk_cache_state = BTRFS_DC_ERROR;
5258 spin_unlock(&cache->lock);
5259 btrfs_cleanup_bg_io(cache);
5263 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5264 struct btrfs_fs_info *fs_info)
5266 struct btrfs_device *dev, *tmp;
5268 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5269 ASSERT(list_empty(&cur_trans->dirty_bgs));
5270 ASSERT(list_empty(&cur_trans->io_bgs));
5272 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5274 list_del_init(&dev->post_commit_list);
5277 btrfs_destroy_delayed_refs(cur_trans, fs_info);
5279 cur_trans->state = TRANS_STATE_COMMIT_START;
5280 wake_up(&fs_info->transaction_blocked_wait);
5282 cur_trans->state = TRANS_STATE_UNBLOCKED;
5283 wake_up(&fs_info->transaction_wait);
5285 btrfs_destroy_delayed_inodes(fs_info);
5287 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5289 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5291 btrfs_free_redirty_list(cur_trans);
5293 cur_trans->state =TRANS_STATE_COMPLETED;
5294 wake_up(&cur_trans->commit_wait);
5297 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5299 struct btrfs_transaction *t;
5301 mutex_lock(&fs_info->transaction_kthread_mutex);
5303 spin_lock(&fs_info->trans_lock);
5304 while (!list_empty(&fs_info->trans_list)) {
5305 t = list_first_entry(&fs_info->trans_list,
5306 struct btrfs_transaction, list);
5307 if (t->state >= TRANS_STATE_COMMIT_START) {
5308 refcount_inc(&t->use_count);
5309 spin_unlock(&fs_info->trans_lock);
5310 btrfs_wait_for_commit(fs_info, t->transid);
5311 btrfs_put_transaction(t);
5312 spin_lock(&fs_info->trans_lock);
5315 if (t == fs_info->running_transaction) {
5316 t->state = TRANS_STATE_COMMIT_DOING;
5317 spin_unlock(&fs_info->trans_lock);
5319 * We wait for 0 num_writers since we don't hold a trans
5320 * handle open currently for this transaction.
5322 wait_event(t->writer_wait,
5323 atomic_read(&t->num_writers) == 0);
5325 spin_unlock(&fs_info->trans_lock);
5327 btrfs_cleanup_one_transaction(t, fs_info);
5329 spin_lock(&fs_info->trans_lock);
5330 if (t == fs_info->running_transaction)
5331 fs_info->running_transaction = NULL;
5332 list_del_init(&t->list);
5333 spin_unlock(&fs_info->trans_lock);
5335 btrfs_put_transaction(t);
5336 trace_btrfs_transaction_commit(fs_info);
5337 spin_lock(&fs_info->trans_lock);
5339 spin_unlock(&fs_info->trans_lock);
5340 btrfs_destroy_all_ordered_extents(fs_info);
5341 btrfs_destroy_delayed_inodes(fs_info);
5342 btrfs_assert_delayed_root_empty(fs_info);
5343 btrfs_destroy_all_delalloc_inodes(fs_info);
5344 btrfs_drop_all_logs(fs_info);
5345 mutex_unlock(&fs_info->transaction_kthread_mutex);
5350 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5352 struct btrfs_path *path;
5354 struct extent_buffer *l;
5355 struct btrfs_key search_key;
5356 struct btrfs_key found_key;
5359 path = btrfs_alloc_path();
5363 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5364 search_key.type = -1;
5365 search_key.offset = (u64)-1;
5366 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5369 BUG_ON(ret == 0); /* Corruption */
5370 if (path->slots[0] > 0) {
5371 slot = path->slots[0] - 1;
5373 btrfs_item_key_to_cpu(l, &found_key, slot);
5374 root->free_objectid = max_t(u64, found_key.objectid + 1,
5375 BTRFS_FIRST_FREE_OBJECTID);
5377 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5381 btrfs_free_path(path);
5385 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5388 mutex_lock(&root->objectid_mutex);
5390 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5391 btrfs_warn(root->fs_info,
5392 "the objectid of root %llu reaches its highest value",
5393 root->root_key.objectid);
5398 *objectid = root->free_objectid++;
5401 mutex_unlock(&root->objectid_mutex);