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 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
48 BTRFS_HEADER_FLAG_RELOC |\
49 BTRFS_SUPER_FLAG_ERROR |\
50 BTRFS_SUPER_FLAG_SEEDING |\
51 BTRFS_SUPER_FLAG_METADUMP |\
52 BTRFS_SUPER_FLAG_METADUMP_V2)
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
68 if (fs_info->csum_shash)
69 crypto_free_shash(fs_info->csum_shash);
73 * async submit bios are used to offload expensive checksumming
74 * onto the worker threads. They checksum file and metadata bios
75 * just before they are sent down the IO stack.
77 struct async_submit_bio {
80 extent_submit_bio_start_t *submit_bio_start;
83 /* Optional parameter for submit_bio_start used by direct io */
85 struct btrfs_work work;
90 * Compute the csum of a btree block and store the result to provided buffer.
92 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
94 struct btrfs_fs_info *fs_info = buf->fs_info;
95 const int num_pages = num_extent_pages(buf);
96 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
97 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
101 shash->tfm = fs_info->csum_shash;
102 crypto_shash_init(shash);
103 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
104 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
105 first_page_part - BTRFS_CSUM_SIZE);
107 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
108 kaddr = page_address(buf->pages[i]);
109 crypto_shash_update(shash, kaddr, PAGE_SIZE);
111 memset(result, 0, BTRFS_CSUM_SIZE);
112 crypto_shash_final(shash, result);
116 * we can't consider a given block up to date unless the transid of the
117 * block matches the transid in the parent node's pointer. This is how we
118 * detect blocks that either didn't get written at all or got written
119 * in the wrong place.
121 static int verify_parent_transid(struct extent_io_tree *io_tree,
122 struct extent_buffer *eb, u64 parent_transid,
125 struct extent_state *cached_state = NULL;
128 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
134 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state);
135 if (extent_buffer_uptodate(eb) &&
136 btrfs_header_generation(eb) == parent_transid) {
140 btrfs_err_rl(eb->fs_info,
141 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
142 eb->start, eb->read_mirror,
143 parent_transid, btrfs_header_generation(eb));
145 clear_extent_buffer_uptodate(eb);
147 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
152 static bool btrfs_supported_super_csum(u16 csum_type)
155 case BTRFS_CSUM_TYPE_CRC32:
156 case BTRFS_CSUM_TYPE_XXHASH:
157 case BTRFS_CSUM_TYPE_SHA256:
158 case BTRFS_CSUM_TYPE_BLAKE2:
166 * Return 0 if the superblock checksum type matches the checksum value of that
167 * algorithm. Pass the raw disk superblock data.
169 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
170 const struct btrfs_super_block *disk_sb)
172 char result[BTRFS_CSUM_SIZE];
173 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
175 shash->tfm = fs_info->csum_shash;
178 * The super_block structure does not span the whole
179 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
180 * filled with zeros and is included in the checksum.
182 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
183 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
185 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
191 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
192 struct btrfs_key *first_key, u64 parent_transid)
194 struct btrfs_fs_info *fs_info = eb->fs_info;
196 struct btrfs_key found_key;
199 found_level = btrfs_header_level(eb);
200 if (found_level != level) {
201 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
202 KERN_ERR "BTRFS: tree level check failed\n");
204 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
205 eb->start, level, found_level);
213 * For live tree block (new tree blocks in current transaction),
214 * we need proper lock context to avoid race, which is impossible here.
215 * So we only checks tree blocks which is read from disk, whose
216 * generation <= fs_info->last_trans_committed.
218 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
221 /* We have @first_key, so this @eb must have at least one item */
222 if (btrfs_header_nritems(eb) == 0) {
224 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
226 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
231 btrfs_node_key_to_cpu(eb, &found_key, 0);
233 btrfs_item_key_to_cpu(eb, &found_key, 0);
234 ret = btrfs_comp_cpu_keys(first_key, &found_key);
237 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
238 KERN_ERR "BTRFS: tree first key check failed\n");
240 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
241 eb->start, parent_transid, first_key->objectid,
242 first_key->type, first_key->offset,
243 found_key.objectid, found_key.type,
250 * helper to read a given tree block, doing retries as required when
251 * the checksums don't match and we have alternate mirrors to try.
253 * @parent_transid: expected transid, skip check if 0
254 * @level: expected level, mandatory check
255 * @first_key: expected key of first slot, skip check if NULL
257 int btrfs_read_extent_buffer(struct extent_buffer *eb,
258 u64 parent_transid, int level,
259 struct btrfs_key *first_key)
261 struct btrfs_fs_info *fs_info = eb->fs_info;
262 struct extent_io_tree *io_tree;
267 int failed_mirror = 0;
269 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
271 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
272 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
274 if (verify_parent_transid(io_tree, eb,
277 else if (btrfs_verify_level_key(eb, level,
278 first_key, parent_transid))
284 num_copies = btrfs_num_copies(fs_info,
289 if (!failed_mirror) {
291 failed_mirror = eb->read_mirror;
295 if (mirror_num == failed_mirror)
298 if (mirror_num > num_copies)
302 if (failed && !ret && failed_mirror)
303 btrfs_repair_eb_io_failure(eb, failed_mirror);
308 static int csum_one_extent_buffer(struct extent_buffer *eb)
310 struct btrfs_fs_info *fs_info = eb->fs_info;
311 u8 result[BTRFS_CSUM_SIZE];
314 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
315 offsetof(struct btrfs_header, fsid),
316 BTRFS_FSID_SIZE) == 0);
317 csum_tree_block(eb, result);
319 if (btrfs_header_level(eb))
320 ret = btrfs_check_node(eb);
322 ret = btrfs_check_leaf_full(eb);
328 * Also check the generation, the eb reached here must be newer than
329 * last committed. Or something seriously wrong happened.
331 if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
334 "block=%llu bad generation, have %llu expect > %llu",
335 eb->start, btrfs_header_generation(eb),
336 fs_info->last_trans_committed);
339 write_extent_buffer(eb, result, 0, fs_info->csum_size);
344 btrfs_print_tree(eb, 0);
345 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
348 * Be noisy if this is an extent buffer from a log tree. We don't abort
349 * a transaction in case there's a bad log tree extent buffer, we just
350 * fallback to a transaction commit. Still we want to know when there is
351 * a bad log tree extent buffer, as that may signal a bug somewhere.
353 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
354 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
358 /* Checksum all dirty extent buffers in one bio_vec */
359 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
360 struct bio_vec *bvec)
362 struct page *page = bvec->bv_page;
363 u64 bvec_start = page_offset(page) + bvec->bv_offset;
367 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
368 cur += fs_info->nodesize) {
369 struct extent_buffer *eb;
372 eb = find_extent_buffer(fs_info, cur);
373 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
376 /* A dirty eb shouldn't disappear from buffer_radix */
380 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
381 free_extent_buffer(eb);
384 if (WARN_ON(!uptodate)) {
385 free_extent_buffer(eb);
389 ret = csum_one_extent_buffer(eb);
390 free_extent_buffer(eb);
398 * Checksum a dirty tree block before IO. This has extra checks to make sure
399 * we only fill in the checksum field in the first page of a multi-page block.
400 * For subpage extent buffers we need bvec to also read the offset in the page.
402 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
404 struct page *page = bvec->bv_page;
405 u64 start = page_offset(page);
407 struct extent_buffer *eb;
409 if (fs_info->nodesize < PAGE_SIZE)
410 return csum_dirty_subpage_buffers(fs_info, bvec);
412 eb = (struct extent_buffer *)page->private;
413 if (page != eb->pages[0])
416 found_start = btrfs_header_bytenr(eb);
418 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
419 WARN_ON(found_start != 0);
424 * Please do not consolidate these warnings into a single if.
425 * It is useful to know what went wrong.
427 if (WARN_ON(found_start != start))
429 if (WARN_ON(!PageUptodate(page)))
432 return csum_one_extent_buffer(eb);
435 static int check_tree_block_fsid(struct extent_buffer *eb)
437 struct btrfs_fs_info *fs_info = eb->fs_info;
438 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
439 u8 fsid[BTRFS_FSID_SIZE];
442 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
445 * Checking the incompat flag is only valid for the current fs. For
446 * seed devices it's forbidden to have their uuid changed so reading
447 * ->fsid in this case is fine
449 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
450 metadata_uuid = fs_devices->metadata_uuid;
452 metadata_uuid = fs_devices->fsid;
454 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
457 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
458 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
464 /* Do basic extent buffer checks at read time */
465 static int validate_extent_buffer(struct extent_buffer *eb)
467 struct btrfs_fs_info *fs_info = eb->fs_info;
469 const u32 csum_size = fs_info->csum_size;
471 u8 result[BTRFS_CSUM_SIZE];
472 const u8 *header_csum;
475 found_start = btrfs_header_bytenr(eb);
476 if (found_start != eb->start) {
477 btrfs_err_rl(fs_info,
478 "bad tree block start, mirror %u want %llu have %llu",
479 eb->read_mirror, eb->start, found_start);
483 if (check_tree_block_fsid(eb)) {
484 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
485 eb->start, eb->read_mirror);
489 found_level = btrfs_header_level(eb);
490 if (found_level >= BTRFS_MAX_LEVEL) {
492 "bad tree block level, mirror %u level %d on logical %llu",
493 eb->read_mirror, btrfs_header_level(eb), eb->start);
498 csum_tree_block(eb, result);
499 header_csum = page_address(eb->pages[0]) +
500 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
502 if (memcmp(result, header_csum, csum_size) != 0) {
503 btrfs_warn_rl(fs_info,
504 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
505 eb->start, eb->read_mirror,
506 CSUM_FMT_VALUE(csum_size, header_csum),
507 CSUM_FMT_VALUE(csum_size, result),
508 btrfs_header_level(eb));
514 * If this is a leaf block and it is corrupt, set the corrupt bit so
515 * that we don't try and read the other copies of this block, just
518 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
519 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
523 if (found_level > 0 && btrfs_check_node(eb))
527 set_extent_buffer_uptodate(eb);
530 "read time tree block corruption detected on logical %llu mirror %u",
531 eb->start, eb->read_mirror);
536 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
539 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
540 struct extent_buffer *eb;
545 * We don't allow bio merge for subpage metadata read, so we should
546 * only get one eb for each endio hook.
548 ASSERT(end == start + fs_info->nodesize - 1);
549 ASSERT(PagePrivate(page));
551 eb = find_extent_buffer(fs_info, start);
553 * When we are reading one tree block, eb must have been inserted into
554 * the radix tree. If not, something is wrong.
558 reads_done = atomic_dec_and_test(&eb->io_pages);
559 /* Subpage read must finish in page read */
562 eb->read_mirror = mirror;
563 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
567 ret = validate_extent_buffer(eb);
571 set_extent_buffer_uptodate(eb);
573 free_extent_buffer(eb);
577 * end_bio_extent_readpage decrements io_pages in case of error,
578 * make sure it has something to decrement.
580 atomic_inc(&eb->io_pages);
581 clear_extent_buffer_uptodate(eb);
582 free_extent_buffer(eb);
586 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
587 struct page *page, u64 start, u64 end,
590 struct extent_buffer *eb;
594 ASSERT(page->private);
596 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
597 return validate_subpage_buffer(page, start, end, mirror);
599 eb = (struct extent_buffer *)page->private;
602 * The pending IO might have been the only thing that kept this buffer
603 * in memory. Make sure we have a ref for all this other checks
605 atomic_inc(&eb->refs);
607 reads_done = atomic_dec_and_test(&eb->io_pages);
611 eb->read_mirror = mirror;
612 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
616 ret = validate_extent_buffer(eb);
620 * our io error hook is going to dec the io pages
621 * again, we have to make sure it has something
624 atomic_inc(&eb->io_pages);
625 clear_extent_buffer_uptodate(eb);
627 free_extent_buffer(eb);
632 static void run_one_async_start(struct btrfs_work *work)
634 struct async_submit_bio *async;
637 async = container_of(work, struct async_submit_bio, work);
638 ret = async->submit_bio_start(async->inode, async->bio,
639 async->dio_file_offset);
645 * In order to insert checksums into the metadata in large chunks, we wait
646 * until bio submission time. All the pages in the bio are checksummed and
647 * sums are attached onto the ordered extent record.
649 * At IO completion time the csums attached on the ordered extent record are
650 * inserted into the tree.
652 static void run_one_async_done(struct btrfs_work *work)
654 struct async_submit_bio *async =
655 container_of(work, struct async_submit_bio, work);
656 struct inode *inode = async->inode;
657 struct btrfs_bio *bbio = btrfs_bio(async->bio);
659 /* If an error occurred we just want to clean up the bio and move on */
661 btrfs_bio_end_io(bbio, async->status);
666 * All of the bios that pass through here are from async helpers.
667 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
668 * This changes nothing when cgroups aren't in use.
670 async->bio->bi_opf |= REQ_CGROUP_PUNT;
671 btrfs_submit_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
674 static void run_one_async_free(struct btrfs_work *work)
676 struct async_submit_bio *async;
678 async = container_of(work, struct async_submit_bio, work);
683 * Submit bio to an async queue.
686 * - true if the work has been succesfuly submitted
687 * - false in case of error
689 bool btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, int mirror_num,
691 extent_submit_bio_start_t *submit_bio_start)
693 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
694 struct async_submit_bio *async;
696 async = kmalloc(sizeof(*async), GFP_NOFS);
700 async->inode = inode;
702 async->mirror_num = mirror_num;
703 async->submit_bio_start = submit_bio_start;
705 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
708 async->dio_file_offset = dio_file_offset;
712 if (op_is_sync(bio->bi_opf))
713 btrfs_queue_work(fs_info->hipri_workers, &async->work);
715 btrfs_queue_work(fs_info->workers, &async->work);
719 static blk_status_t btree_csum_one_bio(struct bio *bio)
721 struct bio_vec *bvec;
722 struct btrfs_root *root;
724 struct bvec_iter_all iter_all;
726 ASSERT(!bio_flagged(bio, BIO_CLONED));
727 bio_for_each_segment_all(bvec, bio, iter_all) {
728 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
729 ret = csum_dirty_buffer(root->fs_info, bvec);
734 return errno_to_blk_status(ret);
737 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
741 * when we're called for a write, we're already in the async
742 * submission context. Just jump into btrfs_submit_bio.
744 return btree_csum_one_bio(bio);
747 static bool should_async_write(struct btrfs_fs_info *fs_info,
748 struct btrfs_inode *bi)
750 if (btrfs_is_zoned(fs_info))
752 if (atomic_read(&bi->sync_writers))
754 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
759 void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_num)
761 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
762 struct btrfs_bio *bbio = btrfs_bio(bio);
765 bio->bi_opf |= REQ_META;
767 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
768 btrfs_submit_bio(fs_info, bio, mirror_num);
773 * Kthread helpers are used to submit writes so that checksumming can
774 * happen in parallel across all CPUs.
776 if (should_async_write(fs_info, BTRFS_I(inode)) &&
777 btrfs_wq_submit_bio(inode, bio, mirror_num, 0, btree_submit_bio_start))
780 ret = btree_csum_one_bio(bio);
782 btrfs_bio_end_io(bbio, ret);
786 btrfs_submit_bio(fs_info, bio, mirror_num);
789 #ifdef CONFIG_MIGRATION
790 static int btree_migrate_folio(struct address_space *mapping,
791 struct folio *dst, struct folio *src, enum migrate_mode mode)
794 * we can't safely write a btree page from here,
795 * we haven't done the locking hook
797 if (folio_test_dirty(src))
800 * Buffers may be managed in a filesystem specific way.
801 * We must have no buffers or drop them.
803 if (folio_get_private(src) &&
804 !filemap_release_folio(src, GFP_KERNEL))
806 return migrate_folio(mapping, dst, src, mode);
809 #define btree_migrate_folio NULL
812 static int btree_writepages(struct address_space *mapping,
813 struct writeback_control *wbc)
815 struct btrfs_fs_info *fs_info;
818 if (wbc->sync_mode == WB_SYNC_NONE) {
820 if (wbc->for_kupdate)
823 fs_info = BTRFS_I(mapping->host)->root->fs_info;
824 /* this is a bit racy, but that's ok */
825 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
826 BTRFS_DIRTY_METADATA_THRESH,
827 fs_info->dirty_metadata_batch);
831 return btree_write_cache_pages(mapping, wbc);
834 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
836 if (folio_test_writeback(folio) || folio_test_dirty(folio))
839 return try_release_extent_buffer(&folio->page);
842 static void btree_invalidate_folio(struct folio *folio, size_t offset,
845 struct extent_io_tree *tree;
846 tree = &BTRFS_I(folio->mapping->host)->io_tree;
847 extent_invalidate_folio(tree, folio, offset);
848 btree_release_folio(folio, GFP_NOFS);
849 if (folio_get_private(folio)) {
850 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
851 "folio private not zero on folio %llu",
852 (unsigned long long)folio_pos(folio));
853 folio_detach_private(folio);
858 static bool btree_dirty_folio(struct address_space *mapping,
861 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
862 struct btrfs_subpage_info *spi = fs_info->subpage_info;
863 struct btrfs_subpage *subpage;
864 struct extent_buffer *eb;
866 u64 page_start = folio_pos(folio);
868 if (fs_info->sectorsize == PAGE_SIZE) {
869 eb = folio_get_private(folio);
871 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
872 BUG_ON(!atomic_read(&eb->refs));
873 btrfs_assert_tree_write_locked(eb);
874 return filemap_dirty_folio(mapping, folio);
878 subpage = folio_get_private(folio);
880 for (cur_bit = spi->dirty_offset;
881 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
886 spin_lock_irqsave(&subpage->lock, flags);
887 if (!test_bit(cur_bit, subpage->bitmaps)) {
888 spin_unlock_irqrestore(&subpage->lock, flags);
891 spin_unlock_irqrestore(&subpage->lock, flags);
892 cur = page_start + cur_bit * fs_info->sectorsize;
894 eb = find_extent_buffer(fs_info, cur);
896 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
897 ASSERT(atomic_read(&eb->refs));
898 btrfs_assert_tree_write_locked(eb);
899 free_extent_buffer(eb);
901 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
903 return filemap_dirty_folio(mapping, folio);
906 #define btree_dirty_folio filemap_dirty_folio
909 static const struct address_space_operations btree_aops = {
910 .writepages = btree_writepages,
911 .release_folio = btree_release_folio,
912 .invalidate_folio = btree_invalidate_folio,
913 .migrate_folio = btree_migrate_folio,
914 .dirty_folio = btree_dirty_folio,
917 struct extent_buffer *btrfs_find_create_tree_block(
918 struct btrfs_fs_info *fs_info,
919 u64 bytenr, u64 owner_root,
922 if (btrfs_is_testing(fs_info))
923 return alloc_test_extent_buffer(fs_info, bytenr);
924 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
928 * Read tree block at logical address @bytenr and do variant basic but critical
931 * @owner_root: the objectid of the root owner for this block.
932 * @parent_transid: expected transid of this tree block, skip check if 0
933 * @level: expected level, mandatory check
934 * @first_key: expected key in slot 0, skip check if NULL
936 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
937 u64 owner_root, u64 parent_transid,
938 int level, struct btrfs_key *first_key)
940 struct extent_buffer *buf = NULL;
943 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
947 ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key);
949 free_extent_buffer_stale(buf);
952 if (btrfs_check_eb_owner(buf, owner_root)) {
953 free_extent_buffer_stale(buf);
954 return ERR_PTR(-EUCLEAN);
960 void btrfs_clean_tree_block(struct extent_buffer *buf)
962 struct btrfs_fs_info *fs_info = buf->fs_info;
963 if (btrfs_header_generation(buf) ==
964 fs_info->running_transaction->transid) {
965 btrfs_assert_tree_write_locked(buf);
967 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
968 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
970 fs_info->dirty_metadata_batch);
971 clear_extent_buffer_dirty(buf);
976 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
979 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
981 memset(&root->root_key, 0, sizeof(root->root_key));
982 memset(&root->root_item, 0, sizeof(root->root_item));
983 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
984 root->fs_info = fs_info;
985 root->root_key.objectid = objectid;
987 root->commit_root = NULL;
989 RB_CLEAR_NODE(&root->rb_node);
991 root->last_trans = 0;
992 root->free_objectid = 0;
993 root->nr_delalloc_inodes = 0;
994 root->nr_ordered_extents = 0;
995 root->inode_tree = RB_ROOT;
996 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
998 btrfs_init_root_block_rsv(root);
1000 INIT_LIST_HEAD(&root->dirty_list);
1001 INIT_LIST_HEAD(&root->root_list);
1002 INIT_LIST_HEAD(&root->delalloc_inodes);
1003 INIT_LIST_HEAD(&root->delalloc_root);
1004 INIT_LIST_HEAD(&root->ordered_extents);
1005 INIT_LIST_HEAD(&root->ordered_root);
1006 INIT_LIST_HEAD(&root->reloc_dirty_list);
1007 INIT_LIST_HEAD(&root->logged_list[0]);
1008 INIT_LIST_HEAD(&root->logged_list[1]);
1009 spin_lock_init(&root->inode_lock);
1010 spin_lock_init(&root->delalloc_lock);
1011 spin_lock_init(&root->ordered_extent_lock);
1012 spin_lock_init(&root->accounting_lock);
1013 spin_lock_init(&root->log_extents_lock[0]);
1014 spin_lock_init(&root->log_extents_lock[1]);
1015 spin_lock_init(&root->qgroup_meta_rsv_lock);
1016 mutex_init(&root->objectid_mutex);
1017 mutex_init(&root->log_mutex);
1018 mutex_init(&root->ordered_extent_mutex);
1019 mutex_init(&root->delalloc_mutex);
1020 init_waitqueue_head(&root->qgroup_flush_wait);
1021 init_waitqueue_head(&root->log_writer_wait);
1022 init_waitqueue_head(&root->log_commit_wait[0]);
1023 init_waitqueue_head(&root->log_commit_wait[1]);
1024 INIT_LIST_HEAD(&root->log_ctxs[0]);
1025 INIT_LIST_HEAD(&root->log_ctxs[1]);
1026 atomic_set(&root->log_commit[0], 0);
1027 atomic_set(&root->log_commit[1], 0);
1028 atomic_set(&root->log_writers, 0);
1029 atomic_set(&root->log_batch, 0);
1030 refcount_set(&root->refs, 1);
1031 atomic_set(&root->snapshot_force_cow, 0);
1032 atomic_set(&root->nr_swapfiles, 0);
1033 root->log_transid = 0;
1034 root->log_transid_committed = -1;
1035 root->last_log_commit = 0;
1038 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1039 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1040 extent_io_tree_init(fs_info, &root->log_csum_range,
1041 IO_TREE_LOG_CSUM_RANGE, NULL);
1044 spin_lock_init(&root->root_item_lock);
1045 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1046 #ifdef CONFIG_BTRFS_DEBUG
1047 INIT_LIST_HEAD(&root->leak_list);
1048 spin_lock(&fs_info->fs_roots_radix_lock);
1049 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1050 spin_unlock(&fs_info->fs_roots_radix_lock);
1054 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1055 u64 objectid, gfp_t flags)
1057 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1059 __setup_root(root, fs_info, objectid);
1063 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1064 /* Should only be used by the testing infrastructure */
1065 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1067 struct btrfs_root *root;
1070 return ERR_PTR(-EINVAL);
1072 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1074 return ERR_PTR(-ENOMEM);
1076 /* We don't use the stripesize in selftest, set it as sectorsize */
1077 root->alloc_bytenr = 0;
1083 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1085 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1086 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1088 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1091 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1093 const struct btrfs_key *key = k;
1094 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1096 return btrfs_comp_cpu_keys(key, &root->root_key);
1099 int btrfs_global_root_insert(struct btrfs_root *root)
1101 struct btrfs_fs_info *fs_info = root->fs_info;
1102 struct rb_node *tmp;
1105 write_lock(&fs_info->global_root_lock);
1106 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1107 write_unlock(&fs_info->global_root_lock);
1111 btrfs_warn(fs_info, "global root %llu %llu already exists",
1112 root->root_key.objectid, root->root_key.offset);
1117 void btrfs_global_root_delete(struct btrfs_root *root)
1119 struct btrfs_fs_info *fs_info = root->fs_info;
1121 write_lock(&fs_info->global_root_lock);
1122 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1123 write_unlock(&fs_info->global_root_lock);
1126 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1127 struct btrfs_key *key)
1129 struct rb_node *node;
1130 struct btrfs_root *root = NULL;
1132 read_lock(&fs_info->global_root_lock);
1133 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1135 root = container_of(node, struct btrfs_root, rb_node);
1136 read_unlock(&fs_info->global_root_lock);
1141 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1143 struct btrfs_block_group *block_group;
1146 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1150 block_group = btrfs_lookup_block_group(fs_info, bytenr);
1152 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1153 ASSERT(block_group);
1156 ret = block_group->global_root_id;
1157 btrfs_put_block_group(block_group);
1162 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1164 struct btrfs_key key = {
1165 .objectid = BTRFS_CSUM_TREE_OBJECTID,
1166 .type = BTRFS_ROOT_ITEM_KEY,
1167 .offset = btrfs_global_root_id(fs_info, bytenr),
1170 return btrfs_global_root(fs_info, &key);
1173 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1175 struct btrfs_key key = {
1176 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
1177 .type = BTRFS_ROOT_ITEM_KEY,
1178 .offset = btrfs_global_root_id(fs_info, bytenr),
1181 return btrfs_global_root(fs_info, &key);
1184 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1187 struct btrfs_fs_info *fs_info = trans->fs_info;
1188 struct extent_buffer *leaf;
1189 struct btrfs_root *tree_root = fs_info->tree_root;
1190 struct btrfs_root *root;
1191 struct btrfs_key key;
1192 unsigned int nofs_flag;
1196 * We're holding a transaction handle, so use a NOFS memory allocation
1197 * context to avoid deadlock if reclaim happens.
1199 nofs_flag = memalloc_nofs_save();
1200 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1201 memalloc_nofs_restore(nofs_flag);
1203 return ERR_PTR(-ENOMEM);
1205 root->root_key.objectid = objectid;
1206 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1207 root->root_key.offset = 0;
1209 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1210 BTRFS_NESTING_NORMAL);
1212 ret = PTR_ERR(leaf);
1218 btrfs_mark_buffer_dirty(leaf);
1220 root->commit_root = btrfs_root_node(root);
1221 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1223 btrfs_set_root_flags(&root->root_item, 0);
1224 btrfs_set_root_limit(&root->root_item, 0);
1225 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1226 btrfs_set_root_generation(&root->root_item, trans->transid);
1227 btrfs_set_root_level(&root->root_item, 0);
1228 btrfs_set_root_refs(&root->root_item, 1);
1229 btrfs_set_root_used(&root->root_item, leaf->len);
1230 btrfs_set_root_last_snapshot(&root->root_item, 0);
1231 btrfs_set_root_dirid(&root->root_item, 0);
1232 if (is_fstree(objectid))
1233 generate_random_guid(root->root_item.uuid);
1235 export_guid(root->root_item.uuid, &guid_null);
1236 btrfs_set_root_drop_level(&root->root_item, 0);
1238 btrfs_tree_unlock(leaf);
1240 key.objectid = objectid;
1241 key.type = BTRFS_ROOT_ITEM_KEY;
1243 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1251 btrfs_tree_unlock(leaf);
1253 btrfs_put_root(root);
1255 return ERR_PTR(ret);
1258 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1259 struct btrfs_fs_info *fs_info)
1261 struct btrfs_root *root;
1263 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1265 return ERR_PTR(-ENOMEM);
1267 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1268 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1269 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1274 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1275 struct btrfs_root *root)
1277 struct extent_buffer *leaf;
1280 * DON'T set SHAREABLE bit for log trees.
1282 * Log trees are not exposed to user space thus can't be snapshotted,
1283 * and they go away before a real commit is actually done.
1285 * They do store pointers to file data extents, and those reference
1286 * counts still get updated (along with back refs to the log tree).
1289 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1290 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1292 return PTR_ERR(leaf);
1296 btrfs_mark_buffer_dirty(root->node);
1297 btrfs_tree_unlock(root->node);
1302 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1303 struct btrfs_fs_info *fs_info)
1305 struct btrfs_root *log_root;
1307 log_root = alloc_log_tree(trans, fs_info);
1308 if (IS_ERR(log_root))
1309 return PTR_ERR(log_root);
1311 if (!btrfs_is_zoned(fs_info)) {
1312 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1315 btrfs_put_root(log_root);
1320 WARN_ON(fs_info->log_root_tree);
1321 fs_info->log_root_tree = log_root;
1325 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1326 struct btrfs_root *root)
1328 struct btrfs_fs_info *fs_info = root->fs_info;
1329 struct btrfs_root *log_root;
1330 struct btrfs_inode_item *inode_item;
1333 log_root = alloc_log_tree(trans, fs_info);
1334 if (IS_ERR(log_root))
1335 return PTR_ERR(log_root);
1337 ret = btrfs_alloc_log_tree_node(trans, log_root);
1339 btrfs_put_root(log_root);
1343 log_root->last_trans = trans->transid;
1344 log_root->root_key.offset = root->root_key.objectid;
1346 inode_item = &log_root->root_item.inode;
1347 btrfs_set_stack_inode_generation(inode_item, 1);
1348 btrfs_set_stack_inode_size(inode_item, 3);
1349 btrfs_set_stack_inode_nlink(inode_item, 1);
1350 btrfs_set_stack_inode_nbytes(inode_item,
1352 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1354 btrfs_set_root_node(&log_root->root_item, log_root->node);
1356 WARN_ON(root->log_root);
1357 root->log_root = log_root;
1358 root->log_transid = 0;
1359 root->log_transid_committed = -1;
1360 root->last_log_commit = 0;
1364 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1365 struct btrfs_path *path,
1366 struct btrfs_key *key)
1368 struct btrfs_root *root;
1369 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1374 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1376 return ERR_PTR(-ENOMEM);
1378 ret = btrfs_find_root(tree_root, key, path,
1379 &root->root_item, &root->root_key);
1386 generation = btrfs_root_generation(&root->root_item);
1387 level = btrfs_root_level(&root->root_item);
1388 root->node = read_tree_block(fs_info,
1389 btrfs_root_bytenr(&root->root_item),
1390 key->objectid, generation, level, NULL);
1391 if (IS_ERR(root->node)) {
1392 ret = PTR_ERR(root->node);
1396 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1402 * For real fs, and not log/reloc trees, root owner must
1403 * match its root node owner
1405 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1406 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1407 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1408 root->root_key.objectid != btrfs_header_owner(root->node)) {
1410 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1411 root->root_key.objectid, root->node->start,
1412 btrfs_header_owner(root->node),
1413 root->root_key.objectid);
1417 root->commit_root = btrfs_root_node(root);
1420 btrfs_put_root(root);
1421 return ERR_PTR(ret);
1424 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1425 struct btrfs_key *key)
1427 struct btrfs_root *root;
1428 struct btrfs_path *path;
1430 path = btrfs_alloc_path();
1432 return ERR_PTR(-ENOMEM);
1433 root = read_tree_root_path(tree_root, path, key);
1434 btrfs_free_path(path);
1440 * Initialize subvolume root in-memory structure
1442 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1444 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1447 unsigned int nofs_flag;
1450 * We might be called under a transaction (e.g. indirect backref
1451 * resolution) which could deadlock if it triggers memory reclaim
1453 nofs_flag = memalloc_nofs_save();
1454 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1455 memalloc_nofs_restore(nofs_flag);
1459 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1460 !btrfs_is_data_reloc_root(root) &&
1461 is_fstree(root->root_key.objectid)) {
1462 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1463 btrfs_check_and_init_root_item(&root->root_item);
1467 * Don't assign anonymous block device to roots that are not exposed to
1468 * userspace, the id pool is limited to 1M
1470 if (is_fstree(root->root_key.objectid) &&
1471 btrfs_root_refs(&root->root_item) > 0) {
1473 ret = get_anon_bdev(&root->anon_dev);
1477 root->anon_dev = anon_dev;
1481 mutex_lock(&root->objectid_mutex);
1482 ret = btrfs_init_root_free_objectid(root);
1484 mutex_unlock(&root->objectid_mutex);
1488 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1490 mutex_unlock(&root->objectid_mutex);
1494 /* The caller is responsible to call btrfs_free_fs_root */
1498 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1501 struct btrfs_root *root;
1503 spin_lock(&fs_info->fs_roots_radix_lock);
1504 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1505 (unsigned long)root_id);
1507 root = btrfs_grab_root(root);
1508 spin_unlock(&fs_info->fs_roots_radix_lock);
1512 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1515 struct btrfs_key key = {
1516 .objectid = objectid,
1517 .type = BTRFS_ROOT_ITEM_KEY,
1521 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1522 return btrfs_grab_root(fs_info->tree_root);
1523 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1524 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1525 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1526 return btrfs_grab_root(fs_info->chunk_root);
1527 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1528 return btrfs_grab_root(fs_info->dev_root);
1529 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1530 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1531 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1532 return btrfs_grab_root(fs_info->quota_root) ?
1533 fs_info->quota_root : ERR_PTR(-ENOENT);
1534 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1535 return btrfs_grab_root(fs_info->uuid_root) ?
1536 fs_info->uuid_root : ERR_PTR(-ENOENT);
1537 if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID)
1538 return btrfs_grab_root(fs_info->block_group_root) ?
1539 fs_info->block_group_root : ERR_PTR(-ENOENT);
1540 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1541 struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1543 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1548 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1549 struct btrfs_root *root)
1553 ret = radix_tree_preload(GFP_NOFS);
1557 spin_lock(&fs_info->fs_roots_radix_lock);
1558 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1559 (unsigned long)root->root_key.objectid,
1562 btrfs_grab_root(root);
1563 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1565 spin_unlock(&fs_info->fs_roots_radix_lock);
1566 radix_tree_preload_end();
1571 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1573 #ifdef CONFIG_BTRFS_DEBUG
1574 struct btrfs_root *root;
1576 while (!list_empty(&fs_info->allocated_roots)) {
1577 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1579 root = list_first_entry(&fs_info->allocated_roots,
1580 struct btrfs_root, leak_list);
1581 btrfs_err(fs_info, "leaked root %s refcount %d",
1582 btrfs_root_name(&root->root_key, buf),
1583 refcount_read(&root->refs));
1584 while (refcount_read(&root->refs) > 1)
1585 btrfs_put_root(root);
1586 btrfs_put_root(root);
1591 static void free_global_roots(struct btrfs_fs_info *fs_info)
1593 struct btrfs_root *root;
1594 struct rb_node *node;
1596 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1597 root = rb_entry(node, struct btrfs_root, rb_node);
1598 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1599 btrfs_put_root(root);
1603 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1605 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1606 percpu_counter_destroy(&fs_info->delalloc_bytes);
1607 percpu_counter_destroy(&fs_info->ordered_bytes);
1608 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1609 btrfs_free_csum_hash(fs_info);
1610 btrfs_free_stripe_hash_table(fs_info);
1611 btrfs_free_ref_cache(fs_info);
1612 kfree(fs_info->balance_ctl);
1613 kfree(fs_info->delayed_root);
1614 free_global_roots(fs_info);
1615 btrfs_put_root(fs_info->tree_root);
1616 btrfs_put_root(fs_info->chunk_root);
1617 btrfs_put_root(fs_info->dev_root);
1618 btrfs_put_root(fs_info->quota_root);
1619 btrfs_put_root(fs_info->uuid_root);
1620 btrfs_put_root(fs_info->fs_root);
1621 btrfs_put_root(fs_info->data_reloc_root);
1622 btrfs_put_root(fs_info->block_group_root);
1623 btrfs_check_leaked_roots(fs_info);
1624 btrfs_extent_buffer_leak_debug_check(fs_info);
1625 kfree(fs_info->super_copy);
1626 kfree(fs_info->super_for_commit);
1627 kfree(fs_info->subpage_info);
1633 * Get an in-memory reference of a root structure.
1635 * For essential trees like root/extent tree, we grab it from fs_info directly.
1636 * For subvolume trees, we check the cached filesystem roots first. If not
1637 * found, then read it from disk and add it to cached fs roots.
1639 * Caller should release the root by calling btrfs_put_root() after the usage.
1641 * NOTE: Reloc and log trees can't be read by this function as they share the
1642 * same root objectid.
1644 * @objectid: root id
1645 * @anon_dev: preallocated anonymous block device number for new roots,
1646 * pass NULL for a new allocation.
1647 * @check_ref: whether to check root item references, If true, return -ENOENT
1650 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1651 u64 objectid, dev_t *anon_dev,
1654 struct btrfs_root *root;
1655 struct btrfs_path *path;
1656 struct btrfs_key key;
1659 root = btrfs_get_global_root(fs_info, objectid);
1663 root = btrfs_lookup_fs_root(fs_info, objectid);
1666 * Some other caller may have read out the newly inserted
1667 * subvolume already (for things like backref walk etc). Not
1668 * that common but still possible. In that case, we just need
1669 * to free the anon_dev.
1671 if (unlikely(anon_dev && *anon_dev)) {
1672 free_anon_bdev(*anon_dev);
1676 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1677 btrfs_put_root(root);
1678 return ERR_PTR(-ENOENT);
1683 key.objectid = objectid;
1684 key.type = BTRFS_ROOT_ITEM_KEY;
1685 key.offset = (u64)-1;
1686 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1690 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1695 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1699 path = btrfs_alloc_path();
1704 key.objectid = BTRFS_ORPHAN_OBJECTID;
1705 key.type = BTRFS_ORPHAN_ITEM_KEY;
1706 key.offset = objectid;
1708 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1709 btrfs_free_path(path);
1713 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1715 ret = btrfs_insert_fs_root(fs_info, root);
1717 if (ret == -EEXIST) {
1718 btrfs_put_root(root);
1726 * If our caller provided us an anonymous device, then it's his
1727 * responsibility to free it in case we fail. So we have to set our
1728 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1729 * and once again by our caller.
1731 if (anon_dev && *anon_dev)
1733 btrfs_put_root(root);
1734 return ERR_PTR(ret);
1738 * Get in-memory reference of a root structure
1740 * @objectid: tree objectid
1741 * @check_ref: if set, verify that the tree exists and the item has at least
1744 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1745 u64 objectid, bool check_ref)
1747 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1751 * Get in-memory reference of a root structure, created as new, optionally pass
1752 * the anonymous block device id
1754 * @objectid: tree objectid
1755 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1756 * parameter value if not NULL
1758 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1759 u64 objectid, dev_t *anon_dev)
1761 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1765 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1766 * @fs_info: the fs_info
1767 * @objectid: the objectid we need to lookup
1769 * This is exclusively used for backref walking, and exists specifically because
1770 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1771 * creation time, which means we may have to read the tree_root in order to look
1772 * up a fs root that is not in memory. If the root is not in memory we will
1773 * read the tree root commit root and look up the fs root from there. This is a
1774 * temporary root, it will not be inserted into the radix tree as it doesn't
1775 * have the most uptodate information, it'll simply be discarded once the
1776 * backref code is finished using the root.
1778 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1779 struct btrfs_path *path,
1782 struct btrfs_root *root;
1783 struct btrfs_key key;
1785 ASSERT(path->search_commit_root && path->skip_locking);
1788 * This can return -ENOENT if we ask for a root that doesn't exist, but
1789 * since this is called via the backref walking code we won't be looking
1790 * up a root that doesn't exist, unless there's corruption. So if root
1791 * != NULL just return it.
1793 root = btrfs_get_global_root(fs_info, objectid);
1797 root = btrfs_lookup_fs_root(fs_info, objectid);
1801 key.objectid = objectid;
1802 key.type = BTRFS_ROOT_ITEM_KEY;
1803 key.offset = (u64)-1;
1804 root = read_tree_root_path(fs_info->tree_root, path, &key);
1805 btrfs_release_path(path);
1810 static int cleaner_kthread(void *arg)
1812 struct btrfs_fs_info *fs_info = arg;
1818 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1820 /* Make the cleaner go to sleep early. */
1821 if (btrfs_need_cleaner_sleep(fs_info))
1825 * Do not do anything if we might cause open_ctree() to block
1826 * before we have finished mounting the filesystem.
1828 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1831 if (!mutex_trylock(&fs_info->cleaner_mutex))
1835 * Avoid the problem that we change the status of the fs
1836 * during the above check and trylock.
1838 if (btrfs_need_cleaner_sleep(fs_info)) {
1839 mutex_unlock(&fs_info->cleaner_mutex);
1843 btrfs_run_delayed_iputs(fs_info);
1845 again = btrfs_clean_one_deleted_snapshot(fs_info);
1846 mutex_unlock(&fs_info->cleaner_mutex);
1849 * The defragger has dealt with the R/O remount and umount,
1850 * needn't do anything special here.
1852 btrfs_run_defrag_inodes(fs_info);
1855 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1856 * with relocation (btrfs_relocate_chunk) and relocation
1857 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1858 * after acquiring fs_info->reclaim_bgs_lock. So we
1859 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1860 * unused block groups.
1862 btrfs_delete_unused_bgs(fs_info);
1865 * Reclaim block groups in the reclaim_bgs list after we deleted
1866 * all unused block_groups. This possibly gives us some more free
1869 btrfs_reclaim_bgs(fs_info);
1871 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1872 if (kthread_should_park())
1874 if (kthread_should_stop())
1877 set_current_state(TASK_INTERRUPTIBLE);
1879 __set_current_state(TASK_RUNNING);
1884 static int transaction_kthread(void *arg)
1886 struct btrfs_root *root = arg;
1887 struct btrfs_fs_info *fs_info = root->fs_info;
1888 struct btrfs_trans_handle *trans;
1889 struct btrfs_transaction *cur;
1892 unsigned long delay;
1896 cannot_commit = false;
1897 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1898 mutex_lock(&fs_info->transaction_kthread_mutex);
1900 spin_lock(&fs_info->trans_lock);
1901 cur = fs_info->running_transaction;
1903 spin_unlock(&fs_info->trans_lock);
1907 delta = ktime_get_seconds() - cur->start_time;
1908 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1909 cur->state < TRANS_STATE_COMMIT_START &&
1910 delta < fs_info->commit_interval) {
1911 spin_unlock(&fs_info->trans_lock);
1912 delay -= msecs_to_jiffies((delta - 1) * 1000);
1914 msecs_to_jiffies(fs_info->commit_interval * 1000));
1917 transid = cur->transid;
1918 spin_unlock(&fs_info->trans_lock);
1920 /* If the file system is aborted, this will always fail. */
1921 trans = btrfs_attach_transaction(root);
1922 if (IS_ERR(trans)) {
1923 if (PTR_ERR(trans) != -ENOENT)
1924 cannot_commit = true;
1927 if (transid == trans->transid) {
1928 btrfs_commit_transaction(trans);
1930 btrfs_end_transaction(trans);
1933 wake_up_process(fs_info->cleaner_kthread);
1934 mutex_unlock(&fs_info->transaction_kthread_mutex);
1936 if (BTRFS_FS_ERROR(fs_info))
1937 btrfs_cleanup_transaction(fs_info);
1938 if (!kthread_should_stop() &&
1939 (!btrfs_transaction_blocked(fs_info) ||
1941 schedule_timeout_interruptible(delay);
1942 } while (!kthread_should_stop());
1947 * This will find the highest generation in the array of root backups. The
1948 * index of the highest array is returned, or -EINVAL if we can't find
1951 * We check to make sure the array is valid by comparing the
1952 * generation of the latest root in the array with the generation
1953 * in the super block. If they don't match we pitch it.
1955 static int find_newest_super_backup(struct btrfs_fs_info *info)
1957 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1959 struct btrfs_root_backup *root_backup;
1962 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1963 root_backup = info->super_copy->super_roots + i;
1964 cur = btrfs_backup_tree_root_gen(root_backup);
1965 if (cur == newest_gen)
1973 * copy all the root pointers into the super backup array.
1974 * this will bump the backup pointer by one when it is
1977 static void backup_super_roots(struct btrfs_fs_info *info)
1979 const int next_backup = info->backup_root_index;
1980 struct btrfs_root_backup *root_backup;
1982 root_backup = info->super_for_commit->super_roots + next_backup;
1985 * make sure all of our padding and empty slots get zero filled
1986 * regardless of which ones we use today
1988 memset(root_backup, 0, sizeof(*root_backup));
1990 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1992 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1993 btrfs_set_backup_tree_root_gen(root_backup,
1994 btrfs_header_generation(info->tree_root->node));
1996 btrfs_set_backup_tree_root_level(root_backup,
1997 btrfs_header_level(info->tree_root->node));
1999 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2000 btrfs_set_backup_chunk_root_gen(root_backup,
2001 btrfs_header_generation(info->chunk_root->node));
2002 btrfs_set_backup_chunk_root_level(root_backup,
2003 btrfs_header_level(info->chunk_root->node));
2005 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
2006 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2007 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2009 btrfs_set_backup_extent_root(root_backup,
2010 extent_root->node->start);
2011 btrfs_set_backup_extent_root_gen(root_backup,
2012 btrfs_header_generation(extent_root->node));
2013 btrfs_set_backup_extent_root_level(root_backup,
2014 btrfs_header_level(extent_root->node));
2016 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2017 btrfs_set_backup_csum_root_gen(root_backup,
2018 btrfs_header_generation(csum_root->node));
2019 btrfs_set_backup_csum_root_level(root_backup,
2020 btrfs_header_level(csum_root->node));
2024 * we might commit during log recovery, which happens before we set
2025 * the fs_root. Make sure it is valid before we fill it in.
2027 if (info->fs_root && info->fs_root->node) {
2028 btrfs_set_backup_fs_root(root_backup,
2029 info->fs_root->node->start);
2030 btrfs_set_backup_fs_root_gen(root_backup,
2031 btrfs_header_generation(info->fs_root->node));
2032 btrfs_set_backup_fs_root_level(root_backup,
2033 btrfs_header_level(info->fs_root->node));
2036 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2037 btrfs_set_backup_dev_root_gen(root_backup,
2038 btrfs_header_generation(info->dev_root->node));
2039 btrfs_set_backup_dev_root_level(root_backup,
2040 btrfs_header_level(info->dev_root->node));
2042 btrfs_set_backup_total_bytes(root_backup,
2043 btrfs_super_total_bytes(info->super_copy));
2044 btrfs_set_backup_bytes_used(root_backup,
2045 btrfs_super_bytes_used(info->super_copy));
2046 btrfs_set_backup_num_devices(root_backup,
2047 btrfs_super_num_devices(info->super_copy));
2050 * if we don't copy this out to the super_copy, it won't get remembered
2051 * for the next commit
2053 memcpy(&info->super_copy->super_roots,
2054 &info->super_for_commit->super_roots,
2055 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2059 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2060 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2062 * fs_info - filesystem whose backup roots need to be read
2063 * priority - priority of backup root required
2065 * Returns backup root index on success and -EINVAL otherwise.
2067 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2069 int backup_index = find_newest_super_backup(fs_info);
2070 struct btrfs_super_block *super = fs_info->super_copy;
2071 struct btrfs_root_backup *root_backup;
2073 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2075 return backup_index;
2077 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2078 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2083 root_backup = super->super_roots + backup_index;
2085 btrfs_set_super_generation(super,
2086 btrfs_backup_tree_root_gen(root_backup));
2087 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2088 btrfs_set_super_root_level(super,
2089 btrfs_backup_tree_root_level(root_backup));
2090 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2093 * Fixme: the total bytes and num_devices need to match or we should
2096 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2097 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2099 return backup_index;
2102 /* helper to cleanup workers */
2103 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2105 btrfs_destroy_workqueue(fs_info->fixup_workers);
2106 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2107 btrfs_destroy_workqueue(fs_info->hipri_workers);
2108 btrfs_destroy_workqueue(fs_info->workers);
2109 if (fs_info->endio_workers)
2110 destroy_workqueue(fs_info->endio_workers);
2111 if (fs_info->endio_raid56_workers)
2112 destroy_workqueue(fs_info->endio_raid56_workers);
2113 if (fs_info->rmw_workers)
2114 destroy_workqueue(fs_info->rmw_workers);
2115 if (fs_info->compressed_write_workers)
2116 destroy_workqueue(fs_info->compressed_write_workers);
2117 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2118 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2119 btrfs_destroy_workqueue(fs_info->delayed_workers);
2120 btrfs_destroy_workqueue(fs_info->caching_workers);
2121 btrfs_destroy_workqueue(fs_info->flush_workers);
2122 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2123 if (fs_info->discard_ctl.discard_workers)
2124 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2126 * Now that all other work queues are destroyed, we can safely destroy
2127 * the queues used for metadata I/O, since tasks from those other work
2128 * queues can do metadata I/O operations.
2130 if (fs_info->endio_meta_workers)
2131 destroy_workqueue(fs_info->endio_meta_workers);
2134 static void free_root_extent_buffers(struct btrfs_root *root)
2137 free_extent_buffer(root->node);
2138 free_extent_buffer(root->commit_root);
2140 root->commit_root = NULL;
2144 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2146 struct btrfs_root *root, *tmp;
2148 rbtree_postorder_for_each_entry_safe(root, tmp,
2149 &fs_info->global_root_tree,
2151 free_root_extent_buffers(root);
2154 /* helper to cleanup tree roots */
2155 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2157 free_root_extent_buffers(info->tree_root);
2159 free_global_root_pointers(info);
2160 free_root_extent_buffers(info->dev_root);
2161 free_root_extent_buffers(info->quota_root);
2162 free_root_extent_buffers(info->uuid_root);
2163 free_root_extent_buffers(info->fs_root);
2164 free_root_extent_buffers(info->data_reloc_root);
2165 free_root_extent_buffers(info->block_group_root);
2166 if (free_chunk_root)
2167 free_root_extent_buffers(info->chunk_root);
2170 void btrfs_put_root(struct btrfs_root *root)
2175 if (refcount_dec_and_test(&root->refs)) {
2176 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2177 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2179 free_anon_bdev(root->anon_dev);
2180 btrfs_drew_lock_destroy(&root->snapshot_lock);
2181 free_root_extent_buffers(root);
2182 #ifdef CONFIG_BTRFS_DEBUG
2183 spin_lock(&root->fs_info->fs_roots_radix_lock);
2184 list_del_init(&root->leak_list);
2185 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2191 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2194 struct btrfs_root *gang[8];
2197 while (!list_empty(&fs_info->dead_roots)) {
2198 gang[0] = list_entry(fs_info->dead_roots.next,
2199 struct btrfs_root, root_list);
2200 list_del(&gang[0]->root_list);
2202 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2203 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2204 btrfs_put_root(gang[0]);
2208 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2213 for (i = 0; i < ret; i++)
2214 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2218 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2220 mutex_init(&fs_info->scrub_lock);
2221 atomic_set(&fs_info->scrubs_running, 0);
2222 atomic_set(&fs_info->scrub_pause_req, 0);
2223 atomic_set(&fs_info->scrubs_paused, 0);
2224 atomic_set(&fs_info->scrub_cancel_req, 0);
2225 init_waitqueue_head(&fs_info->scrub_pause_wait);
2226 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2229 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2231 spin_lock_init(&fs_info->balance_lock);
2232 mutex_init(&fs_info->balance_mutex);
2233 atomic_set(&fs_info->balance_pause_req, 0);
2234 atomic_set(&fs_info->balance_cancel_req, 0);
2235 fs_info->balance_ctl = NULL;
2236 init_waitqueue_head(&fs_info->balance_wait_q);
2237 atomic_set(&fs_info->reloc_cancel_req, 0);
2240 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2242 struct inode *inode = fs_info->btree_inode;
2243 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
2244 fs_info->tree_root);
2246 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2247 set_nlink(inode, 1);
2249 * we set the i_size on the btree inode to the max possible int.
2250 * the real end of the address space is determined by all of
2251 * the devices in the system
2253 inode->i_size = OFFSET_MAX;
2254 inode->i_mapping->a_ops = &btree_aops;
2256 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2257 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2258 IO_TREE_BTREE_INODE_IO, NULL);
2259 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2261 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2262 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2263 BTRFS_I(inode)->location.type = 0;
2264 BTRFS_I(inode)->location.offset = 0;
2265 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2266 __insert_inode_hash(inode, hash);
2269 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2271 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2272 init_rwsem(&fs_info->dev_replace.rwsem);
2273 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2276 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2278 spin_lock_init(&fs_info->qgroup_lock);
2279 mutex_init(&fs_info->qgroup_ioctl_lock);
2280 fs_info->qgroup_tree = RB_ROOT;
2281 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2282 fs_info->qgroup_seq = 1;
2283 fs_info->qgroup_ulist = NULL;
2284 fs_info->qgroup_rescan_running = false;
2285 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
2286 mutex_init(&fs_info->qgroup_rescan_lock);
2289 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2291 u32 max_active = fs_info->thread_pool_size;
2292 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2295 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2296 fs_info->hipri_workers =
2297 btrfs_alloc_workqueue(fs_info, "worker-high",
2298 flags | WQ_HIGHPRI, max_active, 16);
2300 fs_info->delalloc_workers =
2301 btrfs_alloc_workqueue(fs_info, "delalloc",
2302 flags, max_active, 2);
2304 fs_info->flush_workers =
2305 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2306 flags, max_active, 0);
2308 fs_info->caching_workers =
2309 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2311 fs_info->fixup_workers =
2312 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2314 fs_info->endio_workers =
2315 alloc_workqueue("btrfs-endio", flags, max_active);
2316 fs_info->endio_meta_workers =
2317 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2318 fs_info->endio_raid56_workers =
2319 alloc_workqueue("btrfs-endio-raid56", flags, max_active);
2320 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2321 fs_info->endio_write_workers =
2322 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2324 fs_info->compressed_write_workers =
2325 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2326 fs_info->endio_freespace_worker =
2327 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2329 fs_info->delayed_workers =
2330 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2332 fs_info->qgroup_rescan_workers =
2333 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2334 fs_info->discard_ctl.discard_workers =
2335 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2337 if (!(fs_info->workers && fs_info->hipri_workers &&
2338 fs_info->delalloc_workers && fs_info->flush_workers &&
2339 fs_info->endio_workers && fs_info->endio_meta_workers &&
2340 fs_info->compressed_write_workers &&
2341 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2342 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2343 fs_info->caching_workers && fs_info->fixup_workers &&
2344 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2345 fs_info->discard_ctl.discard_workers)) {
2352 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2354 struct crypto_shash *csum_shash;
2355 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2357 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2359 if (IS_ERR(csum_shash)) {
2360 btrfs_err(fs_info, "error allocating %s hash for checksum",
2362 return PTR_ERR(csum_shash);
2365 fs_info->csum_shash = csum_shash;
2368 * Check if the checksum implementation is a fast accelerated one.
2369 * As-is this is a bit of a hack and should be replaced once the csum
2370 * implementations provide that information themselves.
2372 switch (csum_type) {
2373 case BTRFS_CSUM_TYPE_CRC32:
2374 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2375 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2381 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2382 btrfs_super_csum_name(csum_type),
2383 crypto_shash_driver_name(csum_shash));
2387 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2388 struct btrfs_fs_devices *fs_devices)
2391 struct btrfs_root *log_tree_root;
2392 struct btrfs_super_block *disk_super = fs_info->super_copy;
2393 u64 bytenr = btrfs_super_log_root(disk_super);
2394 int level = btrfs_super_log_root_level(disk_super);
2396 if (fs_devices->rw_devices == 0) {
2397 btrfs_warn(fs_info, "log replay required on RO media");
2401 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2406 log_tree_root->node = read_tree_block(fs_info, bytenr,
2407 BTRFS_TREE_LOG_OBJECTID,
2408 fs_info->generation + 1, level,
2410 if (IS_ERR(log_tree_root->node)) {
2411 btrfs_warn(fs_info, "failed to read log tree");
2412 ret = PTR_ERR(log_tree_root->node);
2413 log_tree_root->node = NULL;
2414 btrfs_put_root(log_tree_root);
2417 if (!extent_buffer_uptodate(log_tree_root->node)) {
2418 btrfs_err(fs_info, "failed to read log tree");
2419 btrfs_put_root(log_tree_root);
2423 /* returns with log_tree_root freed on success */
2424 ret = btrfs_recover_log_trees(log_tree_root);
2426 btrfs_handle_fs_error(fs_info, ret,
2427 "Failed to recover log tree");
2428 btrfs_put_root(log_tree_root);
2432 if (sb_rdonly(fs_info->sb)) {
2433 ret = btrfs_commit_super(fs_info);
2441 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2442 struct btrfs_path *path, u64 objectid,
2445 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2446 struct btrfs_root *root;
2447 u64 max_global_id = 0;
2449 struct btrfs_key key = {
2450 .objectid = objectid,
2451 .type = BTRFS_ROOT_ITEM_KEY,
2456 /* If we have IGNOREDATACSUMS skip loading these roots. */
2457 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2458 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2459 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2464 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2468 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2469 ret = btrfs_next_leaf(tree_root, path);
2478 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2479 if (key.objectid != objectid)
2481 btrfs_release_path(path);
2484 * Just worry about this for extent tree, it'll be the same for
2487 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2488 max_global_id = max(max_global_id, key.offset);
2491 root = read_tree_root_path(tree_root, path, &key);
2493 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2494 ret = PTR_ERR(root);
2497 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2498 ret = btrfs_global_root_insert(root);
2500 btrfs_put_root(root);
2505 btrfs_release_path(path);
2507 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2508 fs_info->nr_global_roots = max_global_id + 1;
2510 if (!found || ret) {
2511 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2512 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2514 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2515 ret = ret ? ret : -ENOENT;
2518 btrfs_err(fs_info, "failed to load root %s", name);
2523 static int load_global_roots(struct btrfs_root *tree_root)
2525 struct btrfs_path *path;
2528 path = btrfs_alloc_path();
2532 ret = load_global_roots_objectid(tree_root, path,
2533 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2536 ret = load_global_roots_objectid(tree_root, path,
2537 BTRFS_CSUM_TREE_OBJECTID, "csum");
2540 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2542 ret = load_global_roots_objectid(tree_root, path,
2543 BTRFS_FREE_SPACE_TREE_OBJECTID,
2546 btrfs_free_path(path);
2550 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2552 struct btrfs_root *tree_root = fs_info->tree_root;
2553 struct btrfs_root *root;
2554 struct btrfs_key location;
2557 BUG_ON(!fs_info->tree_root);
2559 ret = load_global_roots(tree_root);
2563 location.type = BTRFS_ROOT_ITEM_KEY;
2564 location.offset = 0;
2566 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2567 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2568 root = btrfs_read_tree_root(tree_root, &location);
2570 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2571 ret = PTR_ERR(root);
2575 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2576 fs_info->block_group_root = root;
2580 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2581 root = btrfs_read_tree_root(tree_root, &location);
2583 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2584 ret = PTR_ERR(root);
2588 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2589 fs_info->dev_root = root;
2591 /* Initialize fs_info for all devices in any case */
2592 ret = btrfs_init_devices_late(fs_info);
2597 * This tree can share blocks with some other fs tree during relocation
2598 * and we need a proper setup by btrfs_get_fs_root
2600 root = btrfs_get_fs_root(tree_root->fs_info,
2601 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2603 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2604 ret = PTR_ERR(root);
2608 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2609 fs_info->data_reloc_root = root;
2612 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2613 root = btrfs_read_tree_root(tree_root, &location);
2614 if (!IS_ERR(root)) {
2615 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2616 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2617 fs_info->quota_root = root;
2620 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2621 root = btrfs_read_tree_root(tree_root, &location);
2623 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2624 ret = PTR_ERR(root);
2629 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2630 fs_info->uuid_root = root;
2635 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2636 location.objectid, ret);
2641 * Real super block validation
2642 * NOTE: super csum type and incompat features will not be checked here.
2644 * @sb: super block to check
2645 * @mirror_num: the super block number to check its bytenr:
2646 * 0 the primary (1st) sb
2647 * 1, 2 2nd and 3rd backup copy
2648 * -1 skip bytenr check
2650 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2651 struct btrfs_super_block *sb, int mirror_num)
2653 u64 nodesize = btrfs_super_nodesize(sb);
2654 u64 sectorsize = btrfs_super_sectorsize(sb);
2657 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2658 btrfs_err(fs_info, "no valid FS found");
2661 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2662 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2663 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2666 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2667 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2668 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2671 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2672 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2673 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2676 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2677 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2678 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2683 * Check sectorsize and nodesize first, other check will need it.
2684 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2686 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2687 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2688 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2693 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2695 * We can support 16K sectorsize with 64K page size without problem,
2696 * but such sectorsize/pagesize combination doesn't make much sense.
2697 * 4K will be our future standard, PAGE_SIZE is supported from the very
2700 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2702 "sectorsize %llu not yet supported for page size %lu",
2703 sectorsize, PAGE_SIZE);
2707 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2708 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2709 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2712 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2713 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2714 le32_to_cpu(sb->__unused_leafsize), nodesize);
2718 /* Root alignment check */
2719 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2720 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2721 btrfs_super_root(sb));
2724 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2725 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2726 btrfs_super_chunk_root(sb));
2729 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2730 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2731 btrfs_super_log_root(sb));
2735 if (memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2737 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2738 sb->fsid, fs_info->fs_devices->fsid);
2742 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2743 BTRFS_FSID_SIZE) != 0) {
2745 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2746 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2751 * Artificial requirement for block-group-tree to force newer features
2752 * (free-space-tree, no-holes) so the test matrix is smaller.
2754 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2755 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2756 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2758 "block-group-tree feature requires fres-space-tree and no-holes");
2762 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2763 BTRFS_FSID_SIZE) != 0) {
2765 "dev_item UUID does not match metadata fsid: %pU != %pU",
2766 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2771 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2774 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2775 btrfs_err(fs_info, "bytes_used is too small %llu",
2776 btrfs_super_bytes_used(sb));
2779 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2780 btrfs_err(fs_info, "invalid stripesize %u",
2781 btrfs_super_stripesize(sb));
2784 if (btrfs_super_num_devices(sb) > (1UL << 31))
2785 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2786 btrfs_super_num_devices(sb));
2787 if (btrfs_super_num_devices(sb) == 0) {
2788 btrfs_err(fs_info, "number of devices is 0");
2792 if (mirror_num >= 0 &&
2793 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2794 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2795 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2800 * Obvious sys_chunk_array corruptions, it must hold at least one key
2803 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2804 btrfs_err(fs_info, "system chunk array too big %u > %u",
2805 btrfs_super_sys_array_size(sb),
2806 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2809 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2810 + sizeof(struct btrfs_chunk)) {
2811 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2812 btrfs_super_sys_array_size(sb),
2813 sizeof(struct btrfs_disk_key)
2814 + sizeof(struct btrfs_chunk));
2819 * The generation is a global counter, we'll trust it more than the others
2820 * but it's still possible that it's the one that's wrong.
2822 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2824 "suspicious: generation < chunk_root_generation: %llu < %llu",
2825 btrfs_super_generation(sb),
2826 btrfs_super_chunk_root_generation(sb));
2827 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2828 && btrfs_super_cache_generation(sb) != (u64)-1)
2830 "suspicious: generation < cache_generation: %llu < %llu",
2831 btrfs_super_generation(sb),
2832 btrfs_super_cache_generation(sb));
2838 * Validation of super block at mount time.
2839 * Some checks already done early at mount time, like csum type and incompat
2840 * flags will be skipped.
2842 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2844 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2848 * Validation of super block at write time.
2849 * Some checks like bytenr check will be skipped as their values will be
2851 * Extra checks like csum type and incompat flags will be done here.
2853 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2854 struct btrfs_super_block *sb)
2858 ret = btrfs_validate_super(fs_info, sb, -1);
2861 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2863 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2864 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2867 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2870 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2871 btrfs_super_incompat_flags(sb),
2872 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2878 "super block corruption detected before writing it to disk");
2882 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2886 root->node = read_tree_block(root->fs_info, bytenr,
2887 root->root_key.objectid, gen, level, NULL);
2888 if (IS_ERR(root->node)) {
2889 ret = PTR_ERR(root->node);
2893 if (!extent_buffer_uptodate(root->node)) {
2894 free_extent_buffer(root->node);
2899 btrfs_set_root_node(&root->root_item, root->node);
2900 root->commit_root = btrfs_root_node(root);
2901 btrfs_set_root_refs(&root->root_item, 1);
2905 static int load_important_roots(struct btrfs_fs_info *fs_info)
2907 struct btrfs_super_block *sb = fs_info->super_copy;
2911 bytenr = btrfs_super_root(sb);
2912 gen = btrfs_super_generation(sb);
2913 level = btrfs_super_root_level(sb);
2914 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2916 btrfs_warn(fs_info, "couldn't read tree root");
2922 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2924 int backup_index = find_newest_super_backup(fs_info);
2925 struct btrfs_super_block *sb = fs_info->super_copy;
2926 struct btrfs_root *tree_root = fs_info->tree_root;
2927 bool handle_error = false;
2931 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2933 if (!IS_ERR(tree_root->node))
2934 free_extent_buffer(tree_root->node);
2935 tree_root->node = NULL;
2937 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2940 free_root_pointers(fs_info, 0);
2943 * Don't use the log in recovery mode, it won't be
2946 btrfs_set_super_log_root(sb, 0);
2948 /* We can't trust the free space cache either */
2949 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2951 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2952 ret = read_backup_root(fs_info, i);
2958 ret = load_important_roots(fs_info);
2960 handle_error = true;
2965 * No need to hold btrfs_root::objectid_mutex since the fs
2966 * hasn't been fully initialised and we are the only user
2968 ret = btrfs_init_root_free_objectid(tree_root);
2970 handle_error = true;
2974 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2976 ret = btrfs_read_roots(fs_info);
2978 handle_error = true;
2982 /* All successful */
2983 fs_info->generation = btrfs_header_generation(tree_root->node);
2984 fs_info->last_trans_committed = fs_info->generation;
2985 fs_info->last_reloc_trans = 0;
2987 /* Always begin writing backup roots after the one being used */
2988 if (backup_index < 0) {
2989 fs_info->backup_root_index = 0;
2991 fs_info->backup_root_index = backup_index + 1;
2992 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3000 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3002 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
3003 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
3004 INIT_LIST_HEAD(&fs_info->trans_list);
3005 INIT_LIST_HEAD(&fs_info->dead_roots);
3006 INIT_LIST_HEAD(&fs_info->delayed_iputs);
3007 INIT_LIST_HEAD(&fs_info->delalloc_roots);
3008 INIT_LIST_HEAD(&fs_info->caching_block_groups);
3009 spin_lock_init(&fs_info->delalloc_root_lock);
3010 spin_lock_init(&fs_info->trans_lock);
3011 spin_lock_init(&fs_info->fs_roots_radix_lock);
3012 spin_lock_init(&fs_info->delayed_iput_lock);
3013 spin_lock_init(&fs_info->defrag_inodes_lock);
3014 spin_lock_init(&fs_info->super_lock);
3015 spin_lock_init(&fs_info->buffer_lock);
3016 spin_lock_init(&fs_info->unused_bgs_lock);
3017 spin_lock_init(&fs_info->treelog_bg_lock);
3018 spin_lock_init(&fs_info->zone_active_bgs_lock);
3019 spin_lock_init(&fs_info->relocation_bg_lock);
3020 rwlock_init(&fs_info->tree_mod_log_lock);
3021 rwlock_init(&fs_info->global_root_lock);
3022 mutex_init(&fs_info->unused_bg_unpin_mutex);
3023 mutex_init(&fs_info->reclaim_bgs_lock);
3024 mutex_init(&fs_info->reloc_mutex);
3025 mutex_init(&fs_info->delalloc_root_mutex);
3026 mutex_init(&fs_info->zoned_meta_io_lock);
3027 mutex_init(&fs_info->zoned_data_reloc_io_lock);
3028 seqlock_init(&fs_info->profiles_lock);
3030 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
3031 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
3032 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
3033 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
3034 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
3035 BTRFS_LOCKDEP_TRANS_COMMIT_START);
3036 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
3037 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
3038 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
3039 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
3040 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
3041 BTRFS_LOCKDEP_TRANS_COMPLETED);
3043 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3044 INIT_LIST_HEAD(&fs_info->space_info);
3045 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3046 INIT_LIST_HEAD(&fs_info->unused_bgs);
3047 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3048 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3049 #ifdef CONFIG_BTRFS_DEBUG
3050 INIT_LIST_HEAD(&fs_info->allocated_roots);
3051 INIT_LIST_HEAD(&fs_info->allocated_ebs);
3052 spin_lock_init(&fs_info->eb_leak_lock);
3054 extent_map_tree_init(&fs_info->mapping_tree);
3055 btrfs_init_block_rsv(&fs_info->global_block_rsv,
3056 BTRFS_BLOCK_RSV_GLOBAL);
3057 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3058 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3059 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3060 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3061 BTRFS_BLOCK_RSV_DELOPS);
3062 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3063 BTRFS_BLOCK_RSV_DELREFS);
3065 atomic_set(&fs_info->async_delalloc_pages, 0);
3066 atomic_set(&fs_info->defrag_running, 0);
3067 atomic_set(&fs_info->nr_delayed_iputs, 0);
3068 atomic64_set(&fs_info->tree_mod_seq, 0);
3069 fs_info->global_root_tree = RB_ROOT;
3070 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3071 fs_info->metadata_ratio = 0;
3072 fs_info->defrag_inodes = RB_ROOT;
3073 atomic64_set(&fs_info->free_chunk_space, 0);
3074 fs_info->tree_mod_log = RB_ROOT;
3075 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3076 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3077 btrfs_init_ref_verify(fs_info);
3079 fs_info->thread_pool_size = min_t(unsigned long,
3080 num_online_cpus() + 2, 8);
3082 INIT_LIST_HEAD(&fs_info->ordered_roots);
3083 spin_lock_init(&fs_info->ordered_root_lock);
3085 btrfs_init_scrub(fs_info);
3086 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3087 fs_info->check_integrity_print_mask = 0;
3089 btrfs_init_balance(fs_info);
3090 btrfs_init_async_reclaim_work(fs_info);
3092 rwlock_init(&fs_info->block_group_cache_lock);
3093 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3095 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3096 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
3098 mutex_init(&fs_info->ordered_operations_mutex);
3099 mutex_init(&fs_info->tree_log_mutex);
3100 mutex_init(&fs_info->chunk_mutex);
3101 mutex_init(&fs_info->transaction_kthread_mutex);
3102 mutex_init(&fs_info->cleaner_mutex);
3103 mutex_init(&fs_info->ro_block_group_mutex);
3104 init_rwsem(&fs_info->commit_root_sem);
3105 init_rwsem(&fs_info->cleanup_work_sem);
3106 init_rwsem(&fs_info->subvol_sem);
3107 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3109 btrfs_init_dev_replace_locks(fs_info);
3110 btrfs_init_qgroup(fs_info);
3111 btrfs_discard_init(fs_info);
3113 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3114 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3116 init_waitqueue_head(&fs_info->transaction_throttle);
3117 init_waitqueue_head(&fs_info->transaction_wait);
3118 init_waitqueue_head(&fs_info->transaction_blocked_wait);
3119 init_waitqueue_head(&fs_info->async_submit_wait);
3120 init_waitqueue_head(&fs_info->delayed_iputs_wait);
3122 /* Usable values until the real ones are cached from the superblock */
3123 fs_info->nodesize = 4096;
3124 fs_info->sectorsize = 4096;
3125 fs_info->sectorsize_bits = ilog2(4096);
3126 fs_info->stripesize = 4096;
3128 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3130 spin_lock_init(&fs_info->swapfile_pins_lock);
3131 fs_info->swapfile_pins = RB_ROOT;
3133 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3134 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3137 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3142 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3143 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3145 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3149 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3153 fs_info->dirty_metadata_batch = PAGE_SIZE *
3154 (1 + ilog2(nr_cpu_ids));
3156 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3160 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3165 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3167 if (!fs_info->delayed_root)
3169 btrfs_init_delayed_root(fs_info->delayed_root);
3172 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3174 return btrfs_alloc_stripe_hash_table(fs_info);
3177 static int btrfs_uuid_rescan_kthread(void *data)
3179 struct btrfs_fs_info *fs_info = data;
3183 * 1st step is to iterate through the existing UUID tree and
3184 * to delete all entries that contain outdated data.
3185 * 2nd step is to add all missing entries to the UUID tree.
3187 ret = btrfs_uuid_tree_iterate(fs_info);
3190 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3192 up(&fs_info->uuid_tree_rescan_sem);
3195 return btrfs_uuid_scan_kthread(data);
3198 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3200 struct task_struct *task;
3202 down(&fs_info->uuid_tree_rescan_sem);
3203 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3205 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3206 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3207 up(&fs_info->uuid_tree_rescan_sem);
3208 return PTR_ERR(task);
3215 * Some options only have meaning at mount time and shouldn't persist across
3216 * remounts, or be displayed. Clear these at the end of mount and remount
3219 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3221 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3222 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3226 * Mounting logic specific to read-write file systems. Shared by open_ctree
3227 * and btrfs_remount when remounting from read-only to read-write.
3229 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3232 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3233 bool rebuild_free_space_tree = false;
3235 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3236 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3237 rebuild_free_space_tree = true;
3238 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3239 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3240 btrfs_warn(fs_info, "free space tree is invalid");
3241 rebuild_free_space_tree = true;
3244 if (rebuild_free_space_tree) {
3245 btrfs_info(fs_info, "rebuilding free space tree");
3246 ret = btrfs_rebuild_free_space_tree(fs_info);
3249 "failed to rebuild free space tree: %d", ret);
3254 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3255 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3256 btrfs_info(fs_info, "disabling free space tree");
3257 ret = btrfs_delete_free_space_tree(fs_info);
3260 "failed to disable free space tree: %d", ret);
3266 * btrfs_find_orphan_roots() is responsible for finding all the dead
3267 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3268 * them into the fs_info->fs_roots_radix tree. This must be done before
3269 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3270 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3271 * item before the root's tree is deleted - this means that if we unmount
3272 * or crash before the deletion completes, on the next mount we will not
3273 * delete what remains of the tree because the orphan item does not
3274 * exists anymore, which is what tells us we have a pending deletion.
3276 ret = btrfs_find_orphan_roots(fs_info);
3280 ret = btrfs_cleanup_fs_roots(fs_info);
3284 down_read(&fs_info->cleanup_work_sem);
3285 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3286 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3287 up_read(&fs_info->cleanup_work_sem);
3290 up_read(&fs_info->cleanup_work_sem);
3292 mutex_lock(&fs_info->cleaner_mutex);
3293 ret = btrfs_recover_relocation(fs_info);
3294 mutex_unlock(&fs_info->cleaner_mutex);
3296 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3300 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3301 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3302 btrfs_info(fs_info, "creating free space tree");
3303 ret = btrfs_create_free_space_tree(fs_info);
3306 "failed to create free space tree: %d", ret);
3311 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3312 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3317 ret = btrfs_resume_balance_async(fs_info);
3321 ret = btrfs_resume_dev_replace_async(fs_info);
3323 btrfs_warn(fs_info, "failed to resume dev_replace");
3327 btrfs_qgroup_rescan_resume(fs_info);
3329 if (!fs_info->uuid_root) {
3330 btrfs_info(fs_info, "creating UUID tree");
3331 ret = btrfs_create_uuid_tree(fs_info);
3334 "failed to create the UUID tree %d", ret);
3344 * Do various sanity and dependency checks of different features.
3346 * @is_rw_mount: If the mount is read-write.
3348 * This is the place for less strict checks (like for subpage or artificial
3349 * feature dependencies).
3351 * For strict checks or possible corruption detection, see
3352 * btrfs_validate_super().
3354 * This should be called after btrfs_parse_options(), as some mount options
3355 * (space cache related) can modify on-disk format like free space tree and
3356 * screw up certain feature dependencies.
3358 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3360 struct btrfs_super_block *disk_super = fs_info->super_copy;
3361 u64 incompat = btrfs_super_incompat_flags(disk_super);
3362 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3363 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3365 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3367 "cannot mount because of unknown incompat features (0x%llx)",
3372 /* Runtime limitation for mixed block groups. */
3373 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3374 (fs_info->sectorsize != fs_info->nodesize)) {
3376 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3377 fs_info->nodesize, fs_info->sectorsize);
3381 /* Mixed backref is an always-enabled feature. */
3382 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3384 /* Set compression related flags just in case. */
3385 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3386 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3387 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3388 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3391 * An ancient flag, which should really be marked deprecated.
3392 * Such runtime limitation doesn't really need a incompat flag.
3394 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3395 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3397 if (compat_ro_unsupp && is_rw_mount) {
3399 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3405 * We have unsupported RO compat features, although RO mounted, we
3406 * should not cause any metadata writes, including log replay.
3407 * Or we could screw up whatever the new feature requires.
3409 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3410 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3412 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3418 * Artificial limitations for block group tree, to force
3419 * block-group-tree to rely on no-holes and free-space-tree.
3421 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3422 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3423 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3425 "block-group-tree feature requires no-holes and free-space-tree features");
3430 * Subpage runtime limitation on v1 cache.
3432 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3433 * we're already defaulting to v2 cache, no need to bother v1 as it's
3434 * going to be deprecated anyway.
3436 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3438 "v1 space cache is not supported for page size %lu with sectorsize %u",
3439 PAGE_SIZE, fs_info->sectorsize);
3443 /* This can be called by remount, we need to protect the super block. */
3444 spin_lock(&fs_info->super_lock);
3445 btrfs_set_super_incompat_flags(disk_super, incompat);
3446 spin_unlock(&fs_info->super_lock);
3451 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3460 struct btrfs_super_block *disk_super;
3461 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3462 struct btrfs_root *tree_root;
3463 struct btrfs_root *chunk_root;
3468 ret = init_mount_fs_info(fs_info, sb);
3474 /* These need to be init'ed before we start creating inodes and such. */
3475 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3477 fs_info->tree_root = tree_root;
3478 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3480 fs_info->chunk_root = chunk_root;
3481 if (!tree_root || !chunk_root) {
3486 fs_info->btree_inode = new_inode(sb);
3487 if (!fs_info->btree_inode) {
3491 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3492 btrfs_init_btree_inode(fs_info);
3494 invalidate_bdev(fs_devices->latest_dev->bdev);
3497 * Read super block and check the signature bytes only
3499 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3500 if (IS_ERR(disk_super)) {
3501 err = PTR_ERR(disk_super);
3505 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3507 * Verify the type first, if that or the checksum value are
3508 * corrupted, we'll find out
3510 csum_type = btrfs_super_csum_type(disk_super);
3511 if (!btrfs_supported_super_csum(csum_type)) {
3512 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3515 btrfs_release_disk_super(disk_super);
3519 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3521 ret = btrfs_init_csum_hash(fs_info, csum_type);
3524 btrfs_release_disk_super(disk_super);
3529 * We want to check superblock checksum, the type is stored inside.
3530 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3532 if (btrfs_check_super_csum(fs_info, disk_super)) {
3533 btrfs_err(fs_info, "superblock checksum mismatch");
3535 btrfs_release_disk_super(disk_super);
3540 * super_copy is zeroed at allocation time and we never touch the
3541 * following bytes up to INFO_SIZE, the checksum is calculated from
3542 * the whole block of INFO_SIZE
3544 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3545 btrfs_release_disk_super(disk_super);
3547 disk_super = fs_info->super_copy;
3550 features = btrfs_super_flags(disk_super);
3551 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3552 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3553 btrfs_set_super_flags(disk_super, features);
3555 "found metadata UUID change in progress flag, clearing");
3558 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3559 sizeof(*fs_info->super_for_commit));
3561 ret = btrfs_validate_mount_super(fs_info);
3563 btrfs_err(fs_info, "superblock contains fatal errors");
3568 if (!btrfs_super_root(disk_super))
3571 /* check FS state, whether FS is broken. */
3572 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3573 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3576 * In the long term, we'll store the compression type in the super
3577 * block, and it'll be used for per file compression control.
3579 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3582 /* Set up fs_info before parsing mount options */
3583 nodesize = btrfs_super_nodesize(disk_super);
3584 sectorsize = btrfs_super_sectorsize(disk_super);
3585 stripesize = sectorsize;
3586 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3587 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3589 fs_info->nodesize = nodesize;
3590 fs_info->sectorsize = sectorsize;
3591 fs_info->sectorsize_bits = ilog2(sectorsize);
3592 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3593 fs_info->stripesize = stripesize;
3595 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3601 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3607 if (sectorsize < PAGE_SIZE) {
3608 struct btrfs_subpage_info *subpage_info;
3611 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3612 * going to be deprecated.
3614 * Force to use v2 cache for subpage case.
3616 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3617 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3618 "forcing free space tree for sector size %u with page size %lu",
3619 sectorsize, PAGE_SIZE);
3622 "read-write for sector size %u with page size %lu is experimental",
3623 sectorsize, PAGE_SIZE);
3624 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3627 btrfs_init_subpage_info(subpage_info, sectorsize);
3628 fs_info->subpage_info = subpage_info;
3631 ret = btrfs_init_workqueues(fs_info);
3634 goto fail_sb_buffer;
3637 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3638 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3640 sb->s_blocksize = sectorsize;
3641 sb->s_blocksize_bits = blksize_bits(sectorsize);
3642 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3644 mutex_lock(&fs_info->chunk_mutex);
3645 ret = btrfs_read_sys_array(fs_info);
3646 mutex_unlock(&fs_info->chunk_mutex);
3648 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3649 goto fail_sb_buffer;
3652 generation = btrfs_super_chunk_root_generation(disk_super);
3653 level = btrfs_super_chunk_root_level(disk_super);
3654 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3657 btrfs_err(fs_info, "failed to read chunk root");
3658 goto fail_tree_roots;
3661 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3662 offsetof(struct btrfs_header, chunk_tree_uuid),
3665 ret = btrfs_read_chunk_tree(fs_info);
3667 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3668 goto fail_tree_roots;
3672 * At this point we know all the devices that make this filesystem,
3673 * including the seed devices but we don't know yet if the replace
3674 * target is required. So free devices that are not part of this
3675 * filesystem but skip the replace target device which is checked
3676 * below in btrfs_init_dev_replace().
3678 btrfs_free_extra_devids(fs_devices);
3679 if (!fs_devices->latest_dev->bdev) {
3680 btrfs_err(fs_info, "failed to read devices");
3681 goto fail_tree_roots;
3684 ret = init_tree_roots(fs_info);
3686 goto fail_tree_roots;
3689 * Get zone type information of zoned block devices. This will also
3690 * handle emulation of a zoned filesystem if a regular device has the
3691 * zoned incompat feature flag set.
3693 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3696 "zoned: failed to read device zone info: %d",
3698 goto fail_block_groups;
3702 * If we have a uuid root and we're not being told to rescan we need to
3703 * check the generation here so we can set the
3704 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3705 * transaction during a balance or the log replay without updating the
3706 * uuid generation, and then if we crash we would rescan the uuid tree,
3707 * even though it was perfectly fine.
3709 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3710 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3711 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3713 ret = btrfs_verify_dev_extents(fs_info);
3716 "failed to verify dev extents against chunks: %d",
3718 goto fail_block_groups;
3720 ret = btrfs_recover_balance(fs_info);
3722 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3723 goto fail_block_groups;
3726 ret = btrfs_init_dev_stats(fs_info);
3728 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3729 goto fail_block_groups;
3732 ret = btrfs_init_dev_replace(fs_info);
3734 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3735 goto fail_block_groups;
3738 ret = btrfs_check_zoned_mode(fs_info);
3740 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3742 goto fail_block_groups;
3745 ret = btrfs_sysfs_add_fsid(fs_devices);
3747 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3749 goto fail_block_groups;
3752 ret = btrfs_sysfs_add_mounted(fs_info);
3754 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3755 goto fail_fsdev_sysfs;
3758 ret = btrfs_init_space_info(fs_info);
3760 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3764 ret = btrfs_read_block_groups(fs_info);
3766 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3770 btrfs_free_zone_cache(fs_info);
3772 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3773 !btrfs_check_rw_degradable(fs_info, NULL)) {
3775 "writable mount is not allowed due to too many missing devices");
3779 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3781 if (IS_ERR(fs_info->cleaner_kthread))
3784 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3786 "btrfs-transaction");
3787 if (IS_ERR(fs_info->transaction_kthread))
3790 if (!btrfs_test_opt(fs_info, NOSSD) &&
3791 !fs_info->fs_devices->rotating) {
3792 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3796 * Mount does not set all options immediately, we can do it now and do
3797 * not have to wait for transaction commit
3799 btrfs_apply_pending_changes(fs_info);
3801 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3802 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3803 ret = btrfsic_mount(fs_info, fs_devices,
3804 btrfs_test_opt(fs_info,
3805 CHECK_INTEGRITY_DATA) ? 1 : 0,
3806 fs_info->check_integrity_print_mask);
3809 "failed to initialize integrity check module: %d",
3813 ret = btrfs_read_qgroup_config(fs_info);
3815 goto fail_trans_kthread;
3817 if (btrfs_build_ref_tree(fs_info))
3818 btrfs_err(fs_info, "couldn't build ref tree");
3820 /* do not make disk changes in broken FS or nologreplay is given */
3821 if (btrfs_super_log_root(disk_super) != 0 &&
3822 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3823 btrfs_info(fs_info, "start tree-log replay");
3824 ret = btrfs_replay_log(fs_info, fs_devices);
3831 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3832 if (IS_ERR(fs_info->fs_root)) {
3833 err = PTR_ERR(fs_info->fs_root);
3834 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3835 fs_info->fs_root = NULL;
3842 ret = btrfs_start_pre_rw_mount(fs_info);
3844 close_ctree(fs_info);
3847 btrfs_discard_resume(fs_info);
3849 if (fs_info->uuid_root &&
3850 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3851 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3852 btrfs_info(fs_info, "checking UUID tree");
3853 ret = btrfs_check_uuid_tree(fs_info);
3856 "failed to check the UUID tree: %d", ret);
3857 close_ctree(fs_info);
3862 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3864 /* Kick the cleaner thread so it'll start deleting snapshots. */
3865 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3866 wake_up_process(fs_info->cleaner_kthread);
3869 btrfs_clear_oneshot_options(fs_info);
3873 btrfs_free_qgroup_config(fs_info);
3875 kthread_stop(fs_info->transaction_kthread);
3876 btrfs_cleanup_transaction(fs_info);
3877 btrfs_free_fs_roots(fs_info);
3879 kthread_stop(fs_info->cleaner_kthread);
3882 * make sure we're done with the btree inode before we stop our
3885 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3888 btrfs_sysfs_remove_mounted(fs_info);
3891 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3894 btrfs_put_block_group_cache(fs_info);
3897 if (fs_info->data_reloc_root)
3898 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3899 free_root_pointers(fs_info, true);
3900 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3903 btrfs_stop_all_workers(fs_info);
3904 btrfs_free_block_groups(fs_info);
3906 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3908 iput(fs_info->btree_inode);
3910 btrfs_close_devices(fs_info->fs_devices);
3913 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3915 static void btrfs_end_super_write(struct bio *bio)
3917 struct btrfs_device *device = bio->bi_private;
3918 struct bio_vec *bvec;
3919 struct bvec_iter_all iter_all;
3922 bio_for_each_segment_all(bvec, bio, iter_all) {
3923 page = bvec->bv_page;
3925 if (bio->bi_status) {
3926 btrfs_warn_rl_in_rcu(device->fs_info,
3927 "lost page write due to IO error on %s (%d)",
3928 rcu_str_deref(device->name),
3929 blk_status_to_errno(bio->bi_status));
3930 ClearPageUptodate(page);
3932 btrfs_dev_stat_inc_and_print(device,
3933 BTRFS_DEV_STAT_WRITE_ERRS);
3935 SetPageUptodate(page);
3945 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3946 int copy_num, bool drop_cache)
3948 struct btrfs_super_block *super;
3950 u64 bytenr, bytenr_orig;
3951 struct address_space *mapping = bdev->bd_inode->i_mapping;
3954 bytenr_orig = btrfs_sb_offset(copy_num);
3955 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3957 return ERR_PTR(-EINVAL);
3959 return ERR_PTR(ret);
3961 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3962 return ERR_PTR(-EINVAL);
3965 /* This should only be called with the primary sb. */
3966 ASSERT(copy_num == 0);
3969 * Drop the page of the primary superblock, so later read will
3970 * always read from the device.
3972 invalidate_inode_pages2_range(mapping,
3973 bytenr >> PAGE_SHIFT,
3974 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3977 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3979 return ERR_CAST(page);
3981 super = page_address(page);
3982 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3983 btrfs_release_disk_super(super);
3984 return ERR_PTR(-ENODATA);
3987 if (btrfs_super_bytenr(super) != bytenr_orig) {
3988 btrfs_release_disk_super(super);
3989 return ERR_PTR(-EINVAL);
3996 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3998 struct btrfs_super_block *super, *latest = NULL;
4002 /* we would like to check all the supers, but that would make
4003 * a btrfs mount succeed after a mkfs from a different FS.
4004 * So, we need to add a special mount option to scan for
4005 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
4007 for (i = 0; i < 1; i++) {
4008 super = btrfs_read_dev_one_super(bdev, i, false);
4012 if (!latest || btrfs_super_generation(super) > transid) {
4014 btrfs_release_disk_super(super);
4017 transid = btrfs_super_generation(super);
4025 * Write superblock @sb to the @device. Do not wait for completion, all the
4026 * pages we use for writing are locked.
4028 * Write @max_mirrors copies of the superblock, where 0 means default that fit
4029 * the expected device size at commit time. Note that max_mirrors must be
4030 * same for write and wait phases.
4032 * Return number of errors when page is not found or submission fails.
4034 static int write_dev_supers(struct btrfs_device *device,
4035 struct btrfs_super_block *sb, int max_mirrors)
4037 struct btrfs_fs_info *fs_info = device->fs_info;
4038 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
4039 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4043 u64 bytenr, bytenr_orig;
4045 if (max_mirrors == 0)
4046 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4048 shash->tfm = fs_info->csum_shash;
4050 for (i = 0; i < max_mirrors; i++) {
4053 struct btrfs_super_block *disk_super;
4055 bytenr_orig = btrfs_sb_offset(i);
4056 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4057 if (ret == -ENOENT) {
4059 } else if (ret < 0) {
4060 btrfs_err(device->fs_info,
4061 "couldn't get super block location for mirror %d",
4066 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4067 device->commit_total_bytes)
4070 btrfs_set_super_bytenr(sb, bytenr_orig);
4072 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4073 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4076 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4079 btrfs_err(device->fs_info,
4080 "couldn't get super block page for bytenr %llu",
4086 /* Bump the refcount for wait_dev_supers() */
4089 disk_super = page_address(page);
4090 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4093 * Directly use bios here instead of relying on the page cache
4094 * to do I/O, so we don't lose the ability to do integrity
4097 bio = bio_alloc(device->bdev, 1,
4098 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4100 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4101 bio->bi_private = device;
4102 bio->bi_end_io = btrfs_end_super_write;
4103 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4104 offset_in_page(bytenr));
4107 * We FUA only the first super block. The others we allow to
4108 * go down lazy and there's a short window where the on-disk
4109 * copies might still contain the older version.
4111 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4112 bio->bi_opf |= REQ_FUA;
4114 btrfsic_check_bio(bio);
4117 if (btrfs_advance_sb_log(device, i))
4120 return errors < i ? 0 : -1;
4124 * Wait for write completion of superblocks done by write_dev_supers,
4125 * @max_mirrors same for write and wait phases.
4127 * Return number of errors when page is not found or not marked up to
4130 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4134 bool primary_failed = false;
4138 if (max_mirrors == 0)
4139 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4141 for (i = 0; i < max_mirrors; i++) {
4144 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4145 if (ret == -ENOENT) {
4147 } else if (ret < 0) {
4150 primary_failed = true;
4153 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4154 device->commit_total_bytes)
4157 page = find_get_page(device->bdev->bd_inode->i_mapping,
4158 bytenr >> PAGE_SHIFT);
4162 primary_failed = true;
4165 /* Page is submitted locked and unlocked once the IO completes */
4166 wait_on_page_locked(page);
4167 if (PageError(page)) {
4170 primary_failed = true;
4173 /* Drop our reference */
4176 /* Drop the reference from the writing run */
4180 /* log error, force error return */
4181 if (primary_failed) {
4182 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4187 return errors < i ? 0 : -1;
4191 * endio for the write_dev_flush, this will wake anyone waiting
4192 * for the barrier when it is done
4194 static void btrfs_end_empty_barrier(struct bio *bio)
4197 complete(bio->bi_private);
4201 * Submit a flush request to the device if it supports it. Error handling is
4202 * done in the waiting counterpart.
4204 static void write_dev_flush(struct btrfs_device *device)
4206 struct bio *bio = &device->flush_bio;
4208 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4210 * When a disk has write caching disabled, we skip submission of a bio
4211 * with flush and sync requests before writing the superblock, since
4212 * it's not needed. However when the integrity checker is enabled, this
4213 * results in reports that there are metadata blocks referred by a
4214 * superblock that were not properly flushed. So don't skip the bio
4215 * submission only when the integrity checker is enabled for the sake
4216 * of simplicity, since this is a debug tool and not meant for use in
4219 if (!bdev_write_cache(device->bdev))
4223 bio_init(bio, device->bdev, NULL, 0,
4224 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4225 bio->bi_end_io = btrfs_end_empty_barrier;
4226 init_completion(&device->flush_wait);
4227 bio->bi_private = &device->flush_wait;
4229 btrfsic_check_bio(bio);
4231 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4235 * If the flush bio has been submitted by write_dev_flush, wait for it.
4237 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4239 struct bio *bio = &device->flush_bio;
4241 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4244 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4245 wait_for_completion_io(&device->flush_wait);
4247 return bio->bi_status;
4250 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4252 if (!btrfs_check_rw_degradable(fs_info, NULL))
4258 * send an empty flush down to each device in parallel,
4259 * then wait for them
4261 static int barrier_all_devices(struct btrfs_fs_info *info)
4263 struct list_head *head;
4264 struct btrfs_device *dev;
4265 int errors_wait = 0;
4268 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4269 /* send down all the barriers */
4270 head = &info->fs_devices->devices;
4271 list_for_each_entry(dev, head, dev_list) {
4272 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4276 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4277 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4280 write_dev_flush(dev);
4281 dev->last_flush_error = BLK_STS_OK;
4284 /* wait for all the barriers */
4285 list_for_each_entry(dev, head, dev_list) {
4286 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4292 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4293 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4296 ret = wait_dev_flush(dev);
4298 dev->last_flush_error = ret;
4299 btrfs_dev_stat_inc_and_print(dev,
4300 BTRFS_DEV_STAT_FLUSH_ERRS);
4307 * At some point we need the status of all disks
4308 * to arrive at the volume status. So error checking
4309 * is being pushed to a separate loop.
4311 return check_barrier_error(info);
4316 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4319 int min_tolerated = INT_MAX;
4321 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4322 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4323 min_tolerated = min_t(int, min_tolerated,
4324 btrfs_raid_array[BTRFS_RAID_SINGLE].
4325 tolerated_failures);
4327 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4328 if (raid_type == BTRFS_RAID_SINGLE)
4330 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4332 min_tolerated = min_t(int, min_tolerated,
4333 btrfs_raid_array[raid_type].
4334 tolerated_failures);
4337 if (min_tolerated == INT_MAX) {
4338 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4342 return min_tolerated;
4345 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4347 struct list_head *head;
4348 struct btrfs_device *dev;
4349 struct btrfs_super_block *sb;
4350 struct btrfs_dev_item *dev_item;
4354 int total_errors = 0;
4357 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4360 * max_mirrors == 0 indicates we're from commit_transaction,
4361 * not from fsync where the tree roots in fs_info have not
4362 * been consistent on disk.
4364 if (max_mirrors == 0)
4365 backup_super_roots(fs_info);
4367 sb = fs_info->super_for_commit;
4368 dev_item = &sb->dev_item;
4370 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4371 head = &fs_info->fs_devices->devices;
4372 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4375 ret = barrier_all_devices(fs_info);
4378 &fs_info->fs_devices->device_list_mutex);
4379 btrfs_handle_fs_error(fs_info, ret,
4380 "errors while submitting device barriers.");
4385 list_for_each_entry(dev, head, dev_list) {
4390 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4391 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4394 btrfs_set_stack_device_generation(dev_item, 0);
4395 btrfs_set_stack_device_type(dev_item, dev->type);
4396 btrfs_set_stack_device_id(dev_item, dev->devid);
4397 btrfs_set_stack_device_total_bytes(dev_item,
4398 dev->commit_total_bytes);
4399 btrfs_set_stack_device_bytes_used(dev_item,
4400 dev->commit_bytes_used);
4401 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4402 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4403 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4404 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4405 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4408 flags = btrfs_super_flags(sb);
4409 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4411 ret = btrfs_validate_write_super(fs_info, sb);
4413 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4414 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4415 "unexpected superblock corruption detected");
4419 ret = write_dev_supers(dev, sb, max_mirrors);
4423 if (total_errors > max_errors) {
4424 btrfs_err(fs_info, "%d errors while writing supers",
4426 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4428 /* FUA is masked off if unsupported and can't be the reason */
4429 btrfs_handle_fs_error(fs_info, -EIO,
4430 "%d errors while writing supers",
4436 list_for_each_entry(dev, head, dev_list) {
4439 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4440 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4443 ret = wait_dev_supers(dev, max_mirrors);
4447 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4448 if (total_errors > max_errors) {
4449 btrfs_handle_fs_error(fs_info, -EIO,
4450 "%d errors while writing supers",
4457 /* Drop a fs root from the radix tree and free it. */
4458 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4459 struct btrfs_root *root)
4461 bool drop_ref = false;
4463 spin_lock(&fs_info->fs_roots_radix_lock);
4464 radix_tree_delete(&fs_info->fs_roots_radix,
4465 (unsigned long)root->root_key.objectid);
4466 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4468 spin_unlock(&fs_info->fs_roots_radix_lock);
4470 if (BTRFS_FS_ERROR(fs_info)) {
4471 ASSERT(root->log_root == NULL);
4472 if (root->reloc_root) {
4473 btrfs_put_root(root->reloc_root);
4474 root->reloc_root = NULL;
4479 btrfs_put_root(root);
4482 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4484 u64 root_objectid = 0;
4485 struct btrfs_root *gang[8];
4488 unsigned int ret = 0;
4491 spin_lock(&fs_info->fs_roots_radix_lock);
4492 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4493 (void **)gang, root_objectid,
4496 spin_unlock(&fs_info->fs_roots_radix_lock);
4499 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4501 for (i = 0; i < ret; i++) {
4502 /* Avoid to grab roots in dead_roots */
4503 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4507 /* grab all the search result for later use */
4508 gang[i] = btrfs_grab_root(gang[i]);
4510 spin_unlock(&fs_info->fs_roots_radix_lock);
4512 for (i = 0; i < ret; i++) {
4515 root_objectid = gang[i]->root_key.objectid;
4516 err = btrfs_orphan_cleanup(gang[i]);
4519 btrfs_put_root(gang[i]);
4524 /* release the uncleaned roots due to error */
4525 for (; i < ret; i++) {
4527 btrfs_put_root(gang[i]);
4532 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4534 struct btrfs_root *root = fs_info->tree_root;
4535 struct btrfs_trans_handle *trans;
4537 mutex_lock(&fs_info->cleaner_mutex);
4538 btrfs_run_delayed_iputs(fs_info);
4539 mutex_unlock(&fs_info->cleaner_mutex);
4540 wake_up_process(fs_info->cleaner_kthread);
4542 /* wait until ongoing cleanup work done */
4543 down_write(&fs_info->cleanup_work_sem);
4544 up_write(&fs_info->cleanup_work_sem);
4546 trans = btrfs_join_transaction(root);
4548 return PTR_ERR(trans);
4549 return btrfs_commit_transaction(trans);
4552 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4554 struct btrfs_transaction *trans;
4555 struct btrfs_transaction *tmp;
4558 if (list_empty(&fs_info->trans_list))
4562 * This function is only called at the very end of close_ctree(),
4563 * thus no other running transaction, no need to take trans_lock.
4565 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4566 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4567 struct extent_state *cached = NULL;
4568 u64 dirty_bytes = 0;
4574 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4575 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4576 dirty_bytes += found_end + 1 - found_start;
4577 cur = found_end + 1;
4580 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4581 trans->transid, dirty_bytes);
4582 btrfs_cleanup_one_transaction(trans, fs_info);
4584 if (trans == fs_info->running_transaction)
4585 fs_info->running_transaction = NULL;
4586 list_del_init(&trans->list);
4588 btrfs_put_transaction(trans);
4589 trace_btrfs_transaction_commit(fs_info);
4594 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4598 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4601 * If we had UNFINISHED_DROPS we could still be processing them, so
4602 * clear that bit and wake up relocation so it can stop.
4603 * We must do this before stopping the block group reclaim task, because
4604 * at btrfs_relocate_block_group() we wait for this bit, and after the
4605 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4606 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4609 btrfs_wake_unfinished_drop(fs_info);
4612 * We may have the reclaim task running and relocating a data block group,
4613 * in which case it may create delayed iputs. So stop it before we park
4614 * the cleaner kthread otherwise we can get new delayed iputs after
4615 * parking the cleaner, and that can make the async reclaim task to hang
4616 * if it's waiting for delayed iputs to complete, since the cleaner is
4617 * parked and can not run delayed iputs - this will make us hang when
4618 * trying to stop the async reclaim task.
4620 cancel_work_sync(&fs_info->reclaim_bgs_work);
4622 * We don't want the cleaner to start new transactions, add more delayed
4623 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4624 * because that frees the task_struct, and the transaction kthread might
4625 * still try to wake up the cleaner.
4627 kthread_park(fs_info->cleaner_kthread);
4629 /* wait for the qgroup rescan worker to stop */
4630 btrfs_qgroup_wait_for_completion(fs_info, false);
4632 /* wait for the uuid_scan task to finish */
4633 down(&fs_info->uuid_tree_rescan_sem);
4634 /* avoid complains from lockdep et al., set sem back to initial state */
4635 up(&fs_info->uuid_tree_rescan_sem);
4637 /* pause restriper - we want to resume on mount */
4638 btrfs_pause_balance(fs_info);
4640 btrfs_dev_replace_suspend_for_unmount(fs_info);
4642 btrfs_scrub_cancel(fs_info);
4644 /* wait for any defraggers to finish */
4645 wait_event(fs_info->transaction_wait,
4646 (atomic_read(&fs_info->defrag_running) == 0));
4648 /* clear out the rbtree of defraggable inodes */
4649 btrfs_cleanup_defrag_inodes(fs_info);
4652 * After we parked the cleaner kthread, ordered extents may have
4653 * completed and created new delayed iputs. If one of the async reclaim
4654 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4655 * can hang forever trying to stop it, because if a delayed iput is
4656 * added after it ran btrfs_run_delayed_iputs() and before it called
4657 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4658 * no one else to run iputs.
4660 * So wait for all ongoing ordered extents to complete and then run
4661 * delayed iputs. This works because once we reach this point no one
4662 * can either create new ordered extents nor create delayed iputs
4663 * through some other means.
4665 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4666 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4667 * but the delayed iput for the respective inode is made only when doing
4668 * the final btrfs_put_ordered_extent() (which must happen at
4669 * btrfs_finish_ordered_io() when we are unmounting).
4671 btrfs_flush_workqueue(fs_info->endio_write_workers);
4672 /* Ordered extents for free space inodes. */
4673 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4674 btrfs_run_delayed_iputs(fs_info);
4676 cancel_work_sync(&fs_info->async_reclaim_work);
4677 cancel_work_sync(&fs_info->async_data_reclaim_work);
4678 cancel_work_sync(&fs_info->preempt_reclaim_work);
4680 /* Cancel or finish ongoing discard work */
4681 btrfs_discard_cleanup(fs_info);
4683 if (!sb_rdonly(fs_info->sb)) {
4685 * The cleaner kthread is stopped, so do one final pass over
4686 * unused block groups.
4688 btrfs_delete_unused_bgs(fs_info);
4691 * There might be existing delayed inode workers still running
4692 * and holding an empty delayed inode item. We must wait for
4693 * them to complete first because they can create a transaction.
4694 * This happens when someone calls btrfs_balance_delayed_items()
4695 * and then a transaction commit runs the same delayed nodes
4696 * before any delayed worker has done something with the nodes.
4697 * We must wait for any worker here and not at transaction
4698 * commit time since that could cause a deadlock.
4699 * This is a very rare case.
4701 btrfs_flush_workqueue(fs_info->delayed_workers);
4703 ret = btrfs_commit_super(fs_info);
4705 btrfs_err(fs_info, "commit super ret %d", ret);
4708 if (BTRFS_FS_ERROR(fs_info))
4709 btrfs_error_commit_super(fs_info);
4711 kthread_stop(fs_info->transaction_kthread);
4712 kthread_stop(fs_info->cleaner_kthread);
4714 ASSERT(list_empty(&fs_info->delayed_iputs));
4715 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4717 if (btrfs_check_quota_leak(fs_info)) {
4718 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4719 btrfs_err(fs_info, "qgroup reserved space leaked");
4722 btrfs_free_qgroup_config(fs_info);
4723 ASSERT(list_empty(&fs_info->delalloc_roots));
4725 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4726 btrfs_info(fs_info, "at unmount delalloc count %lld",
4727 percpu_counter_sum(&fs_info->delalloc_bytes));
4730 if (percpu_counter_sum(&fs_info->ordered_bytes))
4731 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4732 percpu_counter_sum(&fs_info->ordered_bytes));
4734 btrfs_sysfs_remove_mounted(fs_info);
4735 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4737 btrfs_put_block_group_cache(fs_info);
4740 * we must make sure there is not any read request to
4741 * submit after we stopping all workers.
4743 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4744 btrfs_stop_all_workers(fs_info);
4746 /* We shouldn't have any transaction open at this point */
4747 warn_about_uncommitted_trans(fs_info);
4749 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4750 free_root_pointers(fs_info, true);
4751 btrfs_free_fs_roots(fs_info);
4754 * We must free the block groups after dropping the fs_roots as we could
4755 * have had an IO error and have left over tree log blocks that aren't
4756 * cleaned up until the fs roots are freed. This makes the block group
4757 * accounting appear to be wrong because there's pending reserved bytes,
4758 * so make sure we do the block group cleanup afterwards.
4760 btrfs_free_block_groups(fs_info);
4762 iput(fs_info->btree_inode);
4764 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4765 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4766 btrfsic_unmount(fs_info->fs_devices);
4769 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4770 btrfs_close_devices(fs_info->fs_devices);
4773 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4777 struct inode *btree_inode = buf->pages[0]->mapping->host;
4779 ret = extent_buffer_uptodate(buf);
4783 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4784 parent_transid, atomic);
4790 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4792 struct btrfs_fs_info *fs_info = buf->fs_info;
4793 u64 transid = btrfs_header_generation(buf);
4796 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4798 * This is a fast path so only do this check if we have sanity tests
4799 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4800 * outside of the sanity tests.
4802 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4805 btrfs_assert_tree_write_locked(buf);
4806 if (transid != fs_info->generation)
4807 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4808 buf->start, transid, fs_info->generation);
4809 was_dirty = set_extent_buffer_dirty(buf);
4811 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4813 fs_info->dirty_metadata_batch);
4814 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4816 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4817 * but item data not updated.
4818 * So here we should only check item pointers, not item data.
4820 if (btrfs_header_level(buf) == 0 &&
4821 btrfs_check_leaf_relaxed(buf)) {
4822 btrfs_print_leaf(buf);
4828 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4832 * looks as though older kernels can get into trouble with
4833 * this code, they end up stuck in balance_dirty_pages forever
4837 if (current->flags & PF_MEMALLOC)
4841 btrfs_balance_delayed_items(fs_info);
4843 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4844 BTRFS_DIRTY_METADATA_THRESH,
4845 fs_info->dirty_metadata_batch);
4847 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4851 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4853 __btrfs_btree_balance_dirty(fs_info, 1);
4856 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4858 __btrfs_btree_balance_dirty(fs_info, 0);
4861 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4863 /* cleanup FS via transaction */
4864 btrfs_cleanup_transaction(fs_info);
4866 mutex_lock(&fs_info->cleaner_mutex);
4867 btrfs_run_delayed_iputs(fs_info);
4868 mutex_unlock(&fs_info->cleaner_mutex);
4870 down_write(&fs_info->cleanup_work_sem);
4871 up_write(&fs_info->cleanup_work_sem);
4874 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4876 struct btrfs_root *gang[8];
4877 u64 root_objectid = 0;
4880 spin_lock(&fs_info->fs_roots_radix_lock);
4881 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4882 (void **)gang, root_objectid,
4883 ARRAY_SIZE(gang))) != 0) {
4886 for (i = 0; i < ret; i++)
4887 gang[i] = btrfs_grab_root(gang[i]);
4888 spin_unlock(&fs_info->fs_roots_radix_lock);
4890 for (i = 0; i < ret; i++) {
4893 root_objectid = gang[i]->root_key.objectid;
4894 btrfs_free_log(NULL, gang[i]);
4895 btrfs_put_root(gang[i]);
4898 spin_lock(&fs_info->fs_roots_radix_lock);
4900 spin_unlock(&fs_info->fs_roots_radix_lock);
4901 btrfs_free_log_root_tree(NULL, fs_info);
4904 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4906 struct btrfs_ordered_extent *ordered;
4908 spin_lock(&root->ordered_extent_lock);
4910 * This will just short circuit the ordered completion stuff which will
4911 * make sure the ordered extent gets properly cleaned up.
4913 list_for_each_entry(ordered, &root->ordered_extents,
4915 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4916 spin_unlock(&root->ordered_extent_lock);
4919 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4921 struct btrfs_root *root;
4922 struct list_head splice;
4924 INIT_LIST_HEAD(&splice);
4926 spin_lock(&fs_info->ordered_root_lock);
4927 list_splice_init(&fs_info->ordered_roots, &splice);
4928 while (!list_empty(&splice)) {
4929 root = list_first_entry(&splice, struct btrfs_root,
4931 list_move_tail(&root->ordered_root,
4932 &fs_info->ordered_roots);
4934 spin_unlock(&fs_info->ordered_root_lock);
4935 btrfs_destroy_ordered_extents(root);
4938 spin_lock(&fs_info->ordered_root_lock);
4940 spin_unlock(&fs_info->ordered_root_lock);
4943 * We need this here because if we've been flipped read-only we won't
4944 * get sync() from the umount, so we need to make sure any ordered
4945 * extents that haven't had their dirty pages IO start writeout yet
4946 * actually get run and error out properly.
4948 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4951 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4952 struct btrfs_fs_info *fs_info)
4954 struct rb_node *node;
4955 struct btrfs_delayed_ref_root *delayed_refs;
4956 struct btrfs_delayed_ref_node *ref;
4959 delayed_refs = &trans->delayed_refs;
4961 spin_lock(&delayed_refs->lock);
4962 if (atomic_read(&delayed_refs->num_entries) == 0) {
4963 spin_unlock(&delayed_refs->lock);
4964 btrfs_debug(fs_info, "delayed_refs has NO entry");
4968 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4969 struct btrfs_delayed_ref_head *head;
4971 bool pin_bytes = false;
4973 head = rb_entry(node, struct btrfs_delayed_ref_head,
4975 if (btrfs_delayed_ref_lock(delayed_refs, head))
4978 spin_lock(&head->lock);
4979 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4980 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4983 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4984 RB_CLEAR_NODE(&ref->ref_node);
4985 if (!list_empty(&ref->add_list))
4986 list_del(&ref->add_list);
4987 atomic_dec(&delayed_refs->num_entries);
4988 btrfs_put_delayed_ref(ref);
4990 if (head->must_insert_reserved)
4992 btrfs_free_delayed_extent_op(head->extent_op);
4993 btrfs_delete_ref_head(delayed_refs, head);
4994 spin_unlock(&head->lock);
4995 spin_unlock(&delayed_refs->lock);
4996 mutex_unlock(&head->mutex);
4999 struct btrfs_block_group *cache;
5001 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
5004 spin_lock(&cache->space_info->lock);
5005 spin_lock(&cache->lock);
5006 cache->pinned += head->num_bytes;
5007 btrfs_space_info_update_bytes_pinned(fs_info,
5008 cache->space_info, head->num_bytes);
5009 cache->reserved -= head->num_bytes;
5010 cache->space_info->bytes_reserved -= head->num_bytes;
5011 spin_unlock(&cache->lock);
5012 spin_unlock(&cache->space_info->lock);
5014 btrfs_put_block_group(cache);
5016 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
5017 head->bytenr + head->num_bytes - 1);
5019 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
5020 btrfs_put_delayed_ref_head(head);
5022 spin_lock(&delayed_refs->lock);
5024 btrfs_qgroup_destroy_extent_records(trans);
5026 spin_unlock(&delayed_refs->lock);
5031 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
5033 struct btrfs_inode *btrfs_inode;
5034 struct list_head splice;
5036 INIT_LIST_HEAD(&splice);
5038 spin_lock(&root->delalloc_lock);
5039 list_splice_init(&root->delalloc_inodes, &splice);
5041 while (!list_empty(&splice)) {
5042 struct inode *inode = NULL;
5043 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5045 __btrfs_del_delalloc_inode(root, btrfs_inode);
5046 spin_unlock(&root->delalloc_lock);
5049 * Make sure we get a live inode and that it'll not disappear
5052 inode = igrab(&btrfs_inode->vfs_inode);
5054 unsigned int nofs_flag;
5056 nofs_flag = memalloc_nofs_save();
5057 invalidate_inode_pages2(inode->i_mapping);
5058 memalloc_nofs_restore(nofs_flag);
5061 spin_lock(&root->delalloc_lock);
5063 spin_unlock(&root->delalloc_lock);
5066 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5068 struct btrfs_root *root;
5069 struct list_head splice;
5071 INIT_LIST_HEAD(&splice);
5073 spin_lock(&fs_info->delalloc_root_lock);
5074 list_splice_init(&fs_info->delalloc_roots, &splice);
5075 while (!list_empty(&splice)) {
5076 root = list_first_entry(&splice, struct btrfs_root,
5078 root = btrfs_grab_root(root);
5080 spin_unlock(&fs_info->delalloc_root_lock);
5082 btrfs_destroy_delalloc_inodes(root);
5083 btrfs_put_root(root);
5085 spin_lock(&fs_info->delalloc_root_lock);
5087 spin_unlock(&fs_info->delalloc_root_lock);
5090 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5091 struct extent_io_tree *dirty_pages,
5095 struct extent_buffer *eb;
5100 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5105 clear_extent_bits(dirty_pages, start, end, mark);
5106 while (start <= end) {
5107 eb = find_extent_buffer(fs_info, start);
5108 start += fs_info->nodesize;
5111 wait_on_extent_buffer_writeback(eb);
5113 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5115 clear_extent_buffer_dirty(eb);
5116 free_extent_buffer_stale(eb);
5123 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5124 struct extent_io_tree *unpin)
5131 struct extent_state *cached_state = NULL;
5134 * The btrfs_finish_extent_commit() may get the same range as
5135 * ours between find_first_extent_bit and clear_extent_dirty.
5136 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5137 * the same extent range.
5139 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5140 ret = find_first_extent_bit(unpin, 0, &start, &end,
5141 EXTENT_DIRTY, &cached_state);
5143 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5147 clear_extent_dirty(unpin, start, end, &cached_state);
5148 free_extent_state(cached_state);
5149 btrfs_error_unpin_extent_range(fs_info, start, end);
5150 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5157 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5159 struct inode *inode;
5161 inode = cache->io_ctl.inode;
5163 unsigned int nofs_flag;
5165 nofs_flag = memalloc_nofs_save();
5166 invalidate_inode_pages2(inode->i_mapping);
5167 memalloc_nofs_restore(nofs_flag);
5169 BTRFS_I(inode)->generation = 0;
5170 cache->io_ctl.inode = NULL;
5173 ASSERT(cache->io_ctl.pages == NULL);
5174 btrfs_put_block_group(cache);
5177 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5178 struct btrfs_fs_info *fs_info)
5180 struct btrfs_block_group *cache;
5182 spin_lock(&cur_trans->dirty_bgs_lock);
5183 while (!list_empty(&cur_trans->dirty_bgs)) {
5184 cache = list_first_entry(&cur_trans->dirty_bgs,
5185 struct btrfs_block_group,
5188 if (!list_empty(&cache->io_list)) {
5189 spin_unlock(&cur_trans->dirty_bgs_lock);
5190 list_del_init(&cache->io_list);
5191 btrfs_cleanup_bg_io(cache);
5192 spin_lock(&cur_trans->dirty_bgs_lock);
5195 list_del_init(&cache->dirty_list);
5196 spin_lock(&cache->lock);
5197 cache->disk_cache_state = BTRFS_DC_ERROR;
5198 spin_unlock(&cache->lock);
5200 spin_unlock(&cur_trans->dirty_bgs_lock);
5201 btrfs_put_block_group(cache);
5202 btrfs_delayed_refs_rsv_release(fs_info, 1);
5203 spin_lock(&cur_trans->dirty_bgs_lock);
5205 spin_unlock(&cur_trans->dirty_bgs_lock);
5208 * Refer to the definition of io_bgs member for details why it's safe
5209 * to use it without any locking
5211 while (!list_empty(&cur_trans->io_bgs)) {
5212 cache = list_first_entry(&cur_trans->io_bgs,
5213 struct btrfs_block_group,
5216 list_del_init(&cache->io_list);
5217 spin_lock(&cache->lock);
5218 cache->disk_cache_state = BTRFS_DC_ERROR;
5219 spin_unlock(&cache->lock);
5220 btrfs_cleanup_bg_io(cache);
5224 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5225 struct btrfs_fs_info *fs_info)
5227 struct btrfs_device *dev, *tmp;
5229 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5230 ASSERT(list_empty(&cur_trans->dirty_bgs));
5231 ASSERT(list_empty(&cur_trans->io_bgs));
5233 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5235 list_del_init(&dev->post_commit_list);
5238 btrfs_destroy_delayed_refs(cur_trans, fs_info);
5240 cur_trans->state = TRANS_STATE_COMMIT_START;
5241 wake_up(&fs_info->transaction_blocked_wait);
5243 cur_trans->state = TRANS_STATE_UNBLOCKED;
5244 wake_up(&fs_info->transaction_wait);
5246 btrfs_destroy_delayed_inodes(fs_info);
5248 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5250 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5252 btrfs_free_redirty_list(cur_trans);
5254 cur_trans->state =TRANS_STATE_COMPLETED;
5255 wake_up(&cur_trans->commit_wait);
5258 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5260 struct btrfs_transaction *t;
5262 mutex_lock(&fs_info->transaction_kthread_mutex);
5264 spin_lock(&fs_info->trans_lock);
5265 while (!list_empty(&fs_info->trans_list)) {
5266 t = list_first_entry(&fs_info->trans_list,
5267 struct btrfs_transaction, list);
5268 if (t->state >= TRANS_STATE_COMMIT_START) {
5269 refcount_inc(&t->use_count);
5270 spin_unlock(&fs_info->trans_lock);
5271 btrfs_wait_for_commit(fs_info, t->transid);
5272 btrfs_put_transaction(t);
5273 spin_lock(&fs_info->trans_lock);
5276 if (t == fs_info->running_transaction) {
5277 t->state = TRANS_STATE_COMMIT_DOING;
5278 spin_unlock(&fs_info->trans_lock);
5280 * We wait for 0 num_writers since we don't hold a trans
5281 * handle open currently for this transaction.
5283 wait_event(t->writer_wait,
5284 atomic_read(&t->num_writers) == 0);
5286 spin_unlock(&fs_info->trans_lock);
5288 btrfs_cleanup_one_transaction(t, fs_info);
5290 spin_lock(&fs_info->trans_lock);
5291 if (t == fs_info->running_transaction)
5292 fs_info->running_transaction = NULL;
5293 list_del_init(&t->list);
5294 spin_unlock(&fs_info->trans_lock);
5296 btrfs_put_transaction(t);
5297 trace_btrfs_transaction_commit(fs_info);
5298 spin_lock(&fs_info->trans_lock);
5300 spin_unlock(&fs_info->trans_lock);
5301 btrfs_destroy_all_ordered_extents(fs_info);
5302 btrfs_destroy_delayed_inodes(fs_info);
5303 btrfs_assert_delayed_root_empty(fs_info);
5304 btrfs_destroy_all_delalloc_inodes(fs_info);
5305 btrfs_drop_all_logs(fs_info);
5306 mutex_unlock(&fs_info->transaction_kthread_mutex);
5311 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5313 struct btrfs_path *path;
5315 struct extent_buffer *l;
5316 struct btrfs_key search_key;
5317 struct btrfs_key found_key;
5320 path = btrfs_alloc_path();
5324 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5325 search_key.type = -1;
5326 search_key.offset = (u64)-1;
5327 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5330 BUG_ON(ret == 0); /* Corruption */
5331 if (path->slots[0] > 0) {
5332 slot = path->slots[0] - 1;
5334 btrfs_item_key_to_cpu(l, &found_key, slot);
5335 root->free_objectid = max_t(u64, found_key.objectid + 1,
5336 BTRFS_FIRST_FREE_OBJECTID);
5338 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5342 btrfs_free_path(path);
5346 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5349 mutex_lock(&root->objectid_mutex);
5351 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5352 btrfs_warn(root->fs_info,
5353 "the objectid of root %llu reaches its highest value",
5354 root->root_key.objectid);
5359 *objectid = root->free_objectid++;
5362 mutex_unlock(&root->objectid_mutex);