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/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include <crypto/hash.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
43 #include "block-group.h"
45 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
46 BTRFS_HEADER_FLAG_RELOC |\
47 BTRFS_SUPER_FLAG_ERROR |\
48 BTRFS_SUPER_FLAG_SEEDING |\
49 BTRFS_SUPER_FLAG_METADUMP |\
50 BTRFS_SUPER_FLAG_METADUMP_V2)
52 static const struct extent_io_ops btree_extent_io_ops;
53 static void end_workqueue_fn(struct btrfs_work *work);
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);
67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
68 * is complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
71 struct btrfs_end_io_wq {
75 struct btrfs_fs_info *info;
77 enum btrfs_wq_endio_type metadata;
78 struct btrfs_work work;
81 static struct kmem_cache *btrfs_end_io_wq_cache;
83 int __init btrfs_end_io_wq_init(void)
85 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 sizeof(struct btrfs_end_io_wq),
90 if (!btrfs_end_io_wq_cache)
95 void __cold btrfs_end_io_wq_exit(void)
97 kmem_cache_destroy(btrfs_end_io_wq_cache);
101 * async submit bios are used to offload expensive checksumming
102 * onto the worker threads. They checksum file and metadata bios
103 * just before they are sent down the IO stack.
105 struct async_submit_bio {
108 extent_submit_bio_start_t *submit_bio_start;
111 * bio_offset is optional, can be used if the pages in the bio
112 * can't tell us where in the file the bio should go
115 struct btrfs_work work;
120 * Lockdep class keys for extent_buffer->lock's in this root. For a given
121 * eb, the lockdep key is determined by the btrfs_root it belongs to and
122 * the level the eb occupies in the tree.
124 * Different roots are used for different purposes and may nest inside each
125 * other and they require separate keysets. As lockdep keys should be
126 * static, assign keysets according to the purpose of the root as indicated
127 * by btrfs_root->root_key.objectid. This ensures that all special purpose
128 * roots have separate keysets.
130 * Lock-nesting across peer nodes is always done with the immediate parent
131 * node locked thus preventing deadlock. As lockdep doesn't know this, use
132 * subclass to avoid triggering lockdep warning in such cases.
134 * The key is set by the readpage_end_io_hook after the buffer has passed
135 * csum validation but before the pages are unlocked. It is also set by
136 * btrfs_init_new_buffer on freshly allocated blocks.
138 * We also add a check to make sure the highest level of the tree is the
139 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
140 * needs update as well.
142 #ifdef CONFIG_DEBUG_LOCK_ALLOC
143 # if BTRFS_MAX_LEVEL != 8
147 static struct btrfs_lockdep_keyset {
148 u64 id; /* root objectid */
149 const char *name_stem; /* lock name stem */
150 char names[BTRFS_MAX_LEVEL + 1][20];
151 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
152 } btrfs_lockdep_keysets[] = {
153 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
154 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
155 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
156 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
157 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
158 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
159 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
160 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
161 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
162 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
163 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
164 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
165 { .id = 0, .name_stem = "tree" },
168 void __init btrfs_init_lockdep(void)
172 /* initialize lockdep class names */
173 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
174 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
176 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
177 snprintf(ks->names[j], sizeof(ks->names[j]),
178 "btrfs-%s-%02d", ks->name_stem, j);
182 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
185 struct btrfs_lockdep_keyset *ks;
187 BUG_ON(level >= ARRAY_SIZE(ks->keys));
189 /* find the matching keyset, id 0 is the default entry */
190 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
191 if (ks->id == objectid)
194 lockdep_set_class_and_name(&eb->lock,
195 &ks->keys[level], ks->names[level]);
201 * extents on the btree inode are pretty simple, there's one extent
202 * that covers the entire device
204 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
205 struct page *page, size_t pg_offset, u64 start, u64 len,
208 struct btrfs_fs_info *fs_info = inode->root->fs_info;
209 struct extent_map_tree *em_tree = &inode->extent_tree;
210 struct extent_map *em;
213 read_lock(&em_tree->lock);
214 em = lookup_extent_mapping(em_tree, start, len);
216 em->bdev = fs_info->fs_devices->latest_bdev;
217 read_unlock(&em_tree->lock);
220 read_unlock(&em_tree->lock);
222 em = alloc_extent_map();
224 em = ERR_PTR(-ENOMEM);
229 em->block_len = (u64)-1;
231 em->bdev = fs_info->fs_devices->latest_bdev;
233 write_lock(&em_tree->lock);
234 ret = add_extent_mapping(em_tree, em, 0);
235 if (ret == -EEXIST) {
237 em = lookup_extent_mapping(em_tree, start, len);
244 write_unlock(&em_tree->lock);
251 * Compute the csum of a btree block and store the result to provided buffer.
253 * Returns error if the extent buffer cannot be mapped.
255 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
257 struct btrfs_fs_info *fs_info = buf->fs_info;
258 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
260 unsigned long cur_len;
261 unsigned long offset = BTRFS_CSUM_SIZE;
263 unsigned long map_start;
264 unsigned long map_len;
267 shash->tfm = fs_info->csum_shash;
268 crypto_shash_init(shash);
270 len = buf->len - offset;
274 * Note: we don't need to check for the err == 1 case here, as
275 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
276 * and 'min_len = 32' and the currently implemented mapping
277 * algorithm we cannot cross a page boundary.
279 err = map_private_extent_buffer(buf, offset, 32,
280 &kaddr, &map_start, &map_len);
283 cur_len = min(len, map_len - (offset - map_start));
284 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
288 memset(result, 0, BTRFS_CSUM_SIZE);
290 crypto_shash_final(shash, result);
296 * we can't consider a given block up to date unless the transid of the
297 * block matches the transid in the parent node's pointer. This is how we
298 * detect blocks that either didn't get written at all or got written
299 * in the wrong place.
301 static int verify_parent_transid(struct extent_io_tree *io_tree,
302 struct extent_buffer *eb, u64 parent_transid,
305 struct extent_state *cached_state = NULL;
307 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
309 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
316 btrfs_tree_read_lock(eb);
317 btrfs_set_lock_blocking_read(eb);
320 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
322 if (extent_buffer_uptodate(eb) &&
323 btrfs_header_generation(eb) == parent_transid) {
327 btrfs_err_rl(eb->fs_info,
328 "parent transid verify failed on %llu wanted %llu found %llu",
330 parent_transid, btrfs_header_generation(eb));
334 * Things reading via commit roots that don't have normal protection,
335 * like send, can have a really old block in cache that may point at a
336 * block that has been freed and re-allocated. So don't clear uptodate
337 * if we find an eb that is under IO (dirty/writeback) because we could
338 * end up reading in the stale data and then writing it back out and
339 * making everybody very sad.
341 if (!extent_buffer_under_io(eb))
342 clear_extent_buffer_uptodate(eb);
344 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
347 btrfs_tree_read_unlock_blocking(eb);
351 static bool btrfs_supported_super_csum(u16 csum_type)
354 case BTRFS_CSUM_TYPE_CRC32:
362 * Return 0 if the superblock checksum type matches the checksum value of that
363 * algorithm. Pass the raw disk superblock data.
365 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
368 struct btrfs_super_block *disk_sb =
369 (struct btrfs_super_block *)raw_disk_sb;
370 char result[BTRFS_CSUM_SIZE];
371 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
373 shash->tfm = fs_info->csum_shash;
374 crypto_shash_init(shash);
377 * The super_block structure does not span the whole
378 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
379 * filled with zeros and is included in the checksum.
381 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
382 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
383 crypto_shash_final(shash, result);
385 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
391 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
392 struct btrfs_key *first_key, u64 parent_transid)
394 struct btrfs_fs_info *fs_info = eb->fs_info;
396 struct btrfs_key found_key;
399 found_level = btrfs_header_level(eb);
400 if (found_level != level) {
401 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
402 KERN_ERR "BTRFS: tree level check failed\n");
404 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
405 eb->start, level, found_level);
413 * For live tree block (new tree blocks in current transaction),
414 * we need proper lock context to avoid race, which is impossible here.
415 * So we only checks tree blocks which is read from disk, whose
416 * generation <= fs_info->last_trans_committed.
418 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
421 /* We have @first_key, so this @eb must have at least one item */
422 if (btrfs_header_nritems(eb) == 0) {
424 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
426 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
431 btrfs_node_key_to_cpu(eb, &found_key, 0);
433 btrfs_item_key_to_cpu(eb, &found_key, 0);
434 ret = btrfs_comp_cpu_keys(first_key, &found_key);
437 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
438 KERN_ERR "BTRFS: tree first key check failed\n");
440 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
441 eb->start, parent_transid, first_key->objectid,
442 first_key->type, first_key->offset,
443 found_key.objectid, found_key.type,
450 * helper to read a given tree block, doing retries as required when
451 * the checksums don't match and we have alternate mirrors to try.
453 * @parent_transid: expected transid, skip check if 0
454 * @level: expected level, mandatory check
455 * @first_key: expected key of first slot, skip check if NULL
457 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
458 u64 parent_transid, int level,
459 struct btrfs_key *first_key)
461 struct btrfs_fs_info *fs_info = eb->fs_info;
462 struct extent_io_tree *io_tree;
467 int failed_mirror = 0;
469 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
471 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
472 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
474 if (verify_parent_transid(io_tree, eb,
477 else if (btrfs_verify_level_key(eb, level,
478 first_key, parent_transid))
484 num_copies = btrfs_num_copies(fs_info,
489 if (!failed_mirror) {
491 failed_mirror = eb->read_mirror;
495 if (mirror_num == failed_mirror)
498 if (mirror_num > num_copies)
502 if (failed && !ret && failed_mirror)
503 btrfs_repair_eb_io_failure(eb, failed_mirror);
509 * checksum a dirty tree block before IO. This has extra checks to make sure
510 * we only fill in the checksum field in the first page of a multi-page block
513 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
515 u64 start = page_offset(page);
517 u8 result[BTRFS_CSUM_SIZE];
518 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
519 struct extent_buffer *eb;
522 eb = (struct extent_buffer *)page->private;
523 if (page != eb->pages[0])
526 found_start = btrfs_header_bytenr(eb);
528 * Please do not consolidate these warnings into a single if.
529 * It is useful to know what went wrong.
531 if (WARN_ON(found_start != start))
533 if (WARN_ON(!PageUptodate(page)))
536 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
537 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
539 if (csum_tree_block(eb, result))
542 if (btrfs_header_level(eb))
543 ret = btrfs_check_node(eb);
545 ret = btrfs_check_leaf_full(eb);
548 btrfs_print_tree(eb, 0);
550 "block=%llu write time tree block corruption detected",
552 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
555 write_extent_buffer(eb, result, 0, csum_size);
560 static int check_tree_block_fsid(struct extent_buffer *eb)
562 struct btrfs_fs_info *fs_info = eb->fs_info;
563 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
564 u8 fsid[BTRFS_FSID_SIZE];
567 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
572 * Checking the incompat flag is only valid for the current
573 * fs. For seed devices it's forbidden to have their uuid
574 * changed so reading ->fsid in this case is fine
576 if (fs_devices == fs_info->fs_devices &&
577 btrfs_fs_incompat(fs_info, METADATA_UUID))
578 metadata_uuid = fs_devices->metadata_uuid;
580 metadata_uuid = fs_devices->fsid;
582 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
586 fs_devices = fs_devices->seed;
591 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
592 u64 phy_offset, struct page *page,
593 u64 start, u64 end, int mirror)
597 struct extent_buffer *eb;
598 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
599 struct btrfs_fs_info *fs_info = root->fs_info;
600 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
602 u8 result[BTRFS_CSUM_SIZE];
608 eb = (struct extent_buffer *)page->private;
610 /* the pending IO might have been the only thing that kept this buffer
611 * in memory. Make sure we have a ref for all this other checks
613 atomic_inc(&eb->refs);
615 reads_done = atomic_dec_and_test(&eb->io_pages);
619 eb->read_mirror = mirror;
620 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
625 found_start = btrfs_header_bytenr(eb);
626 if (found_start != eb->start) {
627 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
628 eb->start, found_start);
632 if (check_tree_block_fsid(eb)) {
633 btrfs_err_rl(fs_info, "bad fsid on block %llu",
638 found_level = btrfs_header_level(eb);
639 if (found_level >= BTRFS_MAX_LEVEL) {
640 btrfs_err(fs_info, "bad tree block level %d on %llu",
641 (int)btrfs_header_level(eb), eb->start);
646 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
649 ret = csum_tree_block(eb, result);
653 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
657 memcpy(&found, result, csum_size);
659 read_extent_buffer(eb, &val, 0, csum_size);
660 btrfs_warn_rl(fs_info,
661 "%s checksum verify failed on %llu wanted %x found %x level %d",
662 fs_info->sb->s_id, eb->start,
663 val, found, btrfs_header_level(eb));
669 * If this is a leaf block and it is corrupt, set the corrupt bit so
670 * that we don't try and read the other copies of this block, just
673 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
674 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
678 if (found_level > 0 && btrfs_check_node(eb))
682 set_extent_buffer_uptodate(eb);
685 "block=%llu read time tree block corruption detected",
689 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
690 btree_readahead_hook(eb, ret);
694 * our io error hook is going to dec the io pages
695 * again, we have to make sure it has something
698 atomic_inc(&eb->io_pages);
699 clear_extent_buffer_uptodate(eb);
701 free_extent_buffer(eb);
706 static void end_workqueue_bio(struct bio *bio)
708 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
709 struct btrfs_fs_info *fs_info;
710 struct btrfs_workqueue *wq;
712 fs_info = end_io_wq->info;
713 end_io_wq->status = bio->bi_status;
715 if (bio_op(bio) == REQ_OP_WRITE) {
716 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
717 wq = fs_info->endio_meta_write_workers;
718 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
719 wq = fs_info->endio_freespace_worker;
720 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
721 wq = fs_info->endio_raid56_workers;
723 wq = fs_info->endio_write_workers;
725 if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
726 wq = fs_info->endio_repair_workers;
727 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
728 wq = fs_info->endio_raid56_workers;
729 else if (end_io_wq->metadata)
730 wq = fs_info->endio_meta_workers;
732 wq = fs_info->endio_workers;
735 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
736 btrfs_queue_work(wq, &end_io_wq->work);
739 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
740 enum btrfs_wq_endio_type metadata)
742 struct btrfs_end_io_wq *end_io_wq;
744 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
746 return BLK_STS_RESOURCE;
748 end_io_wq->private = bio->bi_private;
749 end_io_wq->end_io = bio->bi_end_io;
750 end_io_wq->info = info;
751 end_io_wq->status = 0;
752 end_io_wq->bio = bio;
753 end_io_wq->metadata = metadata;
755 bio->bi_private = end_io_wq;
756 bio->bi_end_io = end_workqueue_bio;
760 static void run_one_async_start(struct btrfs_work *work)
762 struct async_submit_bio *async;
765 async = container_of(work, struct async_submit_bio, work);
766 ret = async->submit_bio_start(async->private_data, async->bio,
773 * In order to insert checksums into the metadata in large chunks, we wait
774 * until bio submission time. All the pages in the bio are checksummed and
775 * sums are attached onto the ordered extent record.
777 * At IO completion time the csums attached on the ordered extent record are
778 * inserted into the tree.
780 static void run_one_async_done(struct btrfs_work *work)
782 struct async_submit_bio *async;
786 async = container_of(work, struct async_submit_bio, work);
787 inode = async->private_data;
789 /* If an error occurred we just want to clean up the bio and move on */
791 async->bio->bi_status = async->status;
792 bio_endio(async->bio);
797 * All of the bios that pass through here are from async helpers.
798 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
799 * This changes nothing when cgroups aren't in use.
801 async->bio->bi_opf |= REQ_CGROUP_PUNT;
802 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
804 async->bio->bi_status = ret;
805 bio_endio(async->bio);
809 static void run_one_async_free(struct btrfs_work *work)
811 struct async_submit_bio *async;
813 async = container_of(work, struct async_submit_bio, work);
817 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
818 int mirror_num, unsigned long bio_flags,
819 u64 bio_offset, void *private_data,
820 extent_submit_bio_start_t *submit_bio_start)
822 struct async_submit_bio *async;
824 async = kmalloc(sizeof(*async), GFP_NOFS);
826 return BLK_STS_RESOURCE;
828 async->private_data = private_data;
830 async->mirror_num = mirror_num;
831 async->submit_bio_start = submit_bio_start;
833 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
836 async->bio_offset = bio_offset;
840 if (op_is_sync(bio->bi_opf))
841 btrfs_set_work_high_priority(&async->work);
843 btrfs_queue_work(fs_info->workers, &async->work);
847 static blk_status_t btree_csum_one_bio(struct bio *bio)
849 struct bio_vec *bvec;
850 struct btrfs_root *root;
852 struct bvec_iter_all iter_all;
854 ASSERT(!bio_flagged(bio, BIO_CLONED));
855 bio_for_each_segment_all(bvec, bio, iter_all) {
856 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
857 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
862 return errno_to_blk_status(ret);
865 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
869 * when we're called for a write, we're already in the async
870 * submission context. Just jump into btrfs_map_bio
872 return btree_csum_one_bio(bio);
875 static int check_async_write(struct btrfs_fs_info *fs_info,
876 struct btrfs_inode *bi)
878 if (atomic_read(&bi->sync_writers))
880 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
885 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
887 unsigned long bio_flags)
889 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
890 int async = check_async_write(fs_info, BTRFS_I(inode));
893 if (bio_op(bio) != REQ_OP_WRITE) {
895 * called for a read, do the setup so that checksum validation
896 * can happen in the async kernel threads
898 ret = btrfs_bio_wq_end_io(fs_info, bio,
899 BTRFS_WQ_ENDIO_METADATA);
902 ret = btrfs_map_bio(fs_info, bio, mirror_num);
904 ret = btree_csum_one_bio(bio);
907 ret = btrfs_map_bio(fs_info, bio, mirror_num);
910 * kthread helpers are used to submit writes so that
911 * checksumming can happen in parallel across all CPUs
913 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
914 0, inode, btree_submit_bio_start);
922 bio->bi_status = ret;
927 #ifdef CONFIG_MIGRATION
928 static int btree_migratepage(struct address_space *mapping,
929 struct page *newpage, struct page *page,
930 enum migrate_mode mode)
933 * we can't safely write a btree page from here,
934 * we haven't done the locking hook
939 * Buffers may be managed in a filesystem specific way.
940 * We must have no buffers or drop them.
942 if (page_has_private(page) &&
943 !try_to_release_page(page, GFP_KERNEL))
945 return migrate_page(mapping, newpage, page, mode);
950 static int btree_writepages(struct address_space *mapping,
951 struct writeback_control *wbc)
953 struct btrfs_fs_info *fs_info;
956 if (wbc->sync_mode == WB_SYNC_NONE) {
958 if (wbc->for_kupdate)
961 fs_info = BTRFS_I(mapping->host)->root->fs_info;
962 /* this is a bit racy, but that's ok */
963 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
964 BTRFS_DIRTY_METADATA_THRESH,
965 fs_info->dirty_metadata_batch);
969 return btree_write_cache_pages(mapping, wbc);
972 static int btree_readpage(struct file *file, struct page *page)
974 struct extent_io_tree *tree;
975 tree = &BTRFS_I(page->mapping->host)->io_tree;
976 return extent_read_full_page(tree, page, btree_get_extent, 0);
979 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
981 if (PageWriteback(page) || PageDirty(page))
984 return try_release_extent_buffer(page);
987 static void btree_invalidatepage(struct page *page, unsigned int offset,
990 struct extent_io_tree *tree;
991 tree = &BTRFS_I(page->mapping->host)->io_tree;
992 extent_invalidatepage(tree, page, offset);
993 btree_releasepage(page, GFP_NOFS);
994 if (PagePrivate(page)) {
995 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
996 "page private not zero on page %llu",
997 (unsigned long long)page_offset(page));
998 ClearPagePrivate(page);
999 set_page_private(page, 0);
1004 static int btree_set_page_dirty(struct page *page)
1007 struct extent_buffer *eb;
1009 BUG_ON(!PagePrivate(page));
1010 eb = (struct extent_buffer *)page->private;
1012 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1013 BUG_ON(!atomic_read(&eb->refs));
1014 btrfs_assert_tree_locked(eb);
1016 return __set_page_dirty_nobuffers(page);
1019 static const struct address_space_operations btree_aops = {
1020 .readpage = btree_readpage,
1021 .writepages = btree_writepages,
1022 .releasepage = btree_releasepage,
1023 .invalidatepage = btree_invalidatepage,
1024 #ifdef CONFIG_MIGRATION
1025 .migratepage = btree_migratepage,
1027 .set_page_dirty = btree_set_page_dirty,
1030 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1032 struct extent_buffer *buf = NULL;
1035 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1039 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1041 free_extent_buffer_stale(buf);
1043 free_extent_buffer(buf);
1046 struct extent_buffer *btrfs_find_create_tree_block(
1047 struct btrfs_fs_info *fs_info,
1050 if (btrfs_is_testing(fs_info))
1051 return alloc_test_extent_buffer(fs_info, bytenr);
1052 return alloc_extent_buffer(fs_info, bytenr);
1056 * Read tree block at logical address @bytenr and do variant basic but critical
1059 * @parent_transid: expected transid of this tree block, skip check if 0
1060 * @level: expected level, mandatory check
1061 * @first_key: expected key in slot 0, skip check if NULL
1063 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1064 u64 parent_transid, int level,
1065 struct btrfs_key *first_key)
1067 struct extent_buffer *buf = NULL;
1070 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1074 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1077 free_extent_buffer_stale(buf);
1078 return ERR_PTR(ret);
1084 void btrfs_clean_tree_block(struct extent_buffer *buf)
1086 struct btrfs_fs_info *fs_info = buf->fs_info;
1087 if (btrfs_header_generation(buf) ==
1088 fs_info->running_transaction->transid) {
1089 btrfs_assert_tree_locked(buf);
1091 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1092 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1094 fs_info->dirty_metadata_batch);
1095 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1096 btrfs_set_lock_blocking_write(buf);
1097 clear_extent_buffer_dirty(buf);
1102 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1104 struct btrfs_subvolume_writers *writers;
1107 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1109 return ERR_PTR(-ENOMEM);
1111 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1114 return ERR_PTR(ret);
1117 init_waitqueue_head(&writers->wait);
1122 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1124 percpu_counter_destroy(&writers->counter);
1128 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1131 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1133 root->commit_root = NULL;
1135 root->orphan_cleanup_state = 0;
1137 root->last_trans = 0;
1138 root->highest_objectid = 0;
1139 root->nr_delalloc_inodes = 0;
1140 root->nr_ordered_extents = 0;
1141 root->inode_tree = RB_ROOT;
1142 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1143 root->block_rsv = NULL;
1145 INIT_LIST_HEAD(&root->dirty_list);
1146 INIT_LIST_HEAD(&root->root_list);
1147 INIT_LIST_HEAD(&root->delalloc_inodes);
1148 INIT_LIST_HEAD(&root->delalloc_root);
1149 INIT_LIST_HEAD(&root->ordered_extents);
1150 INIT_LIST_HEAD(&root->ordered_root);
1151 INIT_LIST_HEAD(&root->reloc_dirty_list);
1152 INIT_LIST_HEAD(&root->logged_list[0]);
1153 INIT_LIST_HEAD(&root->logged_list[1]);
1154 spin_lock_init(&root->inode_lock);
1155 spin_lock_init(&root->delalloc_lock);
1156 spin_lock_init(&root->ordered_extent_lock);
1157 spin_lock_init(&root->accounting_lock);
1158 spin_lock_init(&root->log_extents_lock[0]);
1159 spin_lock_init(&root->log_extents_lock[1]);
1160 spin_lock_init(&root->qgroup_meta_rsv_lock);
1161 mutex_init(&root->objectid_mutex);
1162 mutex_init(&root->log_mutex);
1163 mutex_init(&root->ordered_extent_mutex);
1164 mutex_init(&root->delalloc_mutex);
1165 init_waitqueue_head(&root->log_writer_wait);
1166 init_waitqueue_head(&root->log_commit_wait[0]);
1167 init_waitqueue_head(&root->log_commit_wait[1]);
1168 INIT_LIST_HEAD(&root->log_ctxs[0]);
1169 INIT_LIST_HEAD(&root->log_ctxs[1]);
1170 atomic_set(&root->log_commit[0], 0);
1171 atomic_set(&root->log_commit[1], 0);
1172 atomic_set(&root->log_writers, 0);
1173 atomic_set(&root->log_batch, 0);
1174 refcount_set(&root->refs, 1);
1175 atomic_set(&root->will_be_snapshotted, 0);
1176 atomic_set(&root->snapshot_force_cow, 0);
1177 atomic_set(&root->nr_swapfiles, 0);
1178 root->log_transid = 0;
1179 root->log_transid_committed = -1;
1180 root->last_log_commit = 0;
1182 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1183 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1185 memset(&root->root_key, 0, sizeof(root->root_key));
1186 memset(&root->root_item, 0, sizeof(root->root_item));
1187 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1189 root->defrag_trans_start = fs_info->generation;
1191 root->defrag_trans_start = 0;
1192 root->root_key.objectid = objectid;
1195 spin_lock_init(&root->root_item_lock);
1196 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1199 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1202 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1204 root->fs_info = fs_info;
1208 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1209 /* Should only be used by the testing infrastructure */
1210 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1212 struct btrfs_root *root;
1215 return ERR_PTR(-EINVAL);
1217 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1219 return ERR_PTR(-ENOMEM);
1221 /* We don't use the stripesize in selftest, set it as sectorsize */
1222 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1223 root->alloc_bytenr = 0;
1229 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1232 struct btrfs_fs_info *fs_info = trans->fs_info;
1233 struct extent_buffer *leaf;
1234 struct btrfs_root *tree_root = fs_info->tree_root;
1235 struct btrfs_root *root;
1236 struct btrfs_key key;
1237 unsigned int nofs_flag;
1239 uuid_le uuid = NULL_UUID_LE;
1242 * We're holding a transaction handle, so use a NOFS memory allocation
1243 * context to avoid deadlock if reclaim happens.
1245 nofs_flag = memalloc_nofs_save();
1246 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1247 memalloc_nofs_restore(nofs_flag);
1249 return ERR_PTR(-ENOMEM);
1251 __setup_root(root, fs_info, objectid);
1252 root->root_key.objectid = objectid;
1253 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1254 root->root_key.offset = 0;
1256 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1258 ret = PTR_ERR(leaf);
1264 btrfs_mark_buffer_dirty(leaf);
1266 root->commit_root = btrfs_root_node(root);
1267 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1269 root->root_item.flags = 0;
1270 root->root_item.byte_limit = 0;
1271 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1272 btrfs_set_root_generation(&root->root_item, trans->transid);
1273 btrfs_set_root_level(&root->root_item, 0);
1274 btrfs_set_root_refs(&root->root_item, 1);
1275 btrfs_set_root_used(&root->root_item, leaf->len);
1276 btrfs_set_root_last_snapshot(&root->root_item, 0);
1277 btrfs_set_root_dirid(&root->root_item, 0);
1278 if (is_fstree(objectid))
1280 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1281 root->root_item.drop_level = 0;
1283 key.objectid = objectid;
1284 key.type = BTRFS_ROOT_ITEM_KEY;
1286 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1290 btrfs_tree_unlock(leaf);
1296 btrfs_tree_unlock(leaf);
1297 free_extent_buffer(root->commit_root);
1298 free_extent_buffer(leaf);
1302 return ERR_PTR(ret);
1305 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1306 struct btrfs_fs_info *fs_info)
1308 struct btrfs_root *root;
1309 struct extent_buffer *leaf;
1311 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1313 return ERR_PTR(-ENOMEM);
1315 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1317 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1318 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1319 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1322 * DON'T set REF_COWS for log trees
1324 * log trees do not get reference counted because they go away
1325 * before a real commit is actually done. They do store pointers
1326 * to file data extents, and those reference counts still get
1327 * updated (along with back refs to the log tree).
1330 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1334 return ERR_CAST(leaf);
1339 btrfs_mark_buffer_dirty(root->node);
1340 btrfs_tree_unlock(root->node);
1344 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1345 struct btrfs_fs_info *fs_info)
1347 struct btrfs_root *log_root;
1349 log_root = alloc_log_tree(trans, fs_info);
1350 if (IS_ERR(log_root))
1351 return PTR_ERR(log_root);
1352 WARN_ON(fs_info->log_root_tree);
1353 fs_info->log_root_tree = log_root;
1357 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1358 struct btrfs_root *root)
1360 struct btrfs_fs_info *fs_info = root->fs_info;
1361 struct btrfs_root *log_root;
1362 struct btrfs_inode_item *inode_item;
1364 log_root = alloc_log_tree(trans, fs_info);
1365 if (IS_ERR(log_root))
1366 return PTR_ERR(log_root);
1368 log_root->last_trans = trans->transid;
1369 log_root->root_key.offset = root->root_key.objectid;
1371 inode_item = &log_root->root_item.inode;
1372 btrfs_set_stack_inode_generation(inode_item, 1);
1373 btrfs_set_stack_inode_size(inode_item, 3);
1374 btrfs_set_stack_inode_nlink(inode_item, 1);
1375 btrfs_set_stack_inode_nbytes(inode_item,
1377 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1379 btrfs_set_root_node(&log_root->root_item, log_root->node);
1381 WARN_ON(root->log_root);
1382 root->log_root = log_root;
1383 root->log_transid = 0;
1384 root->log_transid_committed = -1;
1385 root->last_log_commit = 0;
1389 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1390 struct btrfs_key *key)
1392 struct btrfs_root *root;
1393 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1394 struct btrfs_path *path;
1399 path = btrfs_alloc_path();
1401 return ERR_PTR(-ENOMEM);
1403 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1409 __setup_root(root, fs_info, key->objectid);
1411 ret = btrfs_find_root(tree_root, key, path,
1412 &root->root_item, &root->root_key);
1419 generation = btrfs_root_generation(&root->root_item);
1420 level = btrfs_root_level(&root->root_item);
1421 root->node = read_tree_block(fs_info,
1422 btrfs_root_bytenr(&root->root_item),
1423 generation, level, NULL);
1424 if (IS_ERR(root->node)) {
1425 ret = PTR_ERR(root->node);
1427 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1429 free_extent_buffer(root->node);
1432 root->commit_root = btrfs_root_node(root);
1434 btrfs_free_path(path);
1440 root = ERR_PTR(ret);
1444 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1445 struct btrfs_key *location)
1447 struct btrfs_root *root;
1449 root = btrfs_read_tree_root(tree_root, location);
1453 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1454 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1455 btrfs_check_and_init_root_item(&root->root_item);
1461 int btrfs_init_fs_root(struct btrfs_root *root)
1464 struct btrfs_subvolume_writers *writers;
1466 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1467 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1469 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1474 writers = btrfs_alloc_subvolume_writers();
1475 if (IS_ERR(writers)) {
1476 ret = PTR_ERR(writers);
1479 root->subv_writers = writers;
1481 btrfs_init_free_ino_ctl(root);
1482 spin_lock_init(&root->ino_cache_lock);
1483 init_waitqueue_head(&root->ino_cache_wait);
1485 ret = get_anon_bdev(&root->anon_dev);
1489 mutex_lock(&root->objectid_mutex);
1490 ret = btrfs_find_highest_objectid(root,
1491 &root->highest_objectid);
1493 mutex_unlock(&root->objectid_mutex);
1497 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1499 mutex_unlock(&root->objectid_mutex);
1503 /* The caller is responsible to call btrfs_free_fs_root */
1507 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1510 struct btrfs_root *root;
1512 spin_lock(&fs_info->fs_roots_radix_lock);
1513 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1514 (unsigned long)root_id);
1515 spin_unlock(&fs_info->fs_roots_radix_lock);
1519 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1520 struct btrfs_root *root)
1524 ret = radix_tree_preload(GFP_NOFS);
1528 spin_lock(&fs_info->fs_roots_radix_lock);
1529 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1530 (unsigned long)root->root_key.objectid,
1533 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1534 spin_unlock(&fs_info->fs_roots_radix_lock);
1535 radix_tree_preload_end();
1540 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1541 struct btrfs_key *location,
1544 struct btrfs_root *root;
1545 struct btrfs_path *path;
1546 struct btrfs_key key;
1549 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1550 return fs_info->tree_root;
1551 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1552 return fs_info->extent_root;
1553 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1554 return fs_info->chunk_root;
1555 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1556 return fs_info->dev_root;
1557 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1558 return fs_info->csum_root;
1559 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1560 return fs_info->quota_root ? fs_info->quota_root :
1562 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1563 return fs_info->uuid_root ? fs_info->uuid_root :
1565 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1566 return fs_info->free_space_root ? fs_info->free_space_root :
1569 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1571 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1572 return ERR_PTR(-ENOENT);
1576 root = btrfs_read_fs_root(fs_info->tree_root, location);
1580 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1585 ret = btrfs_init_fs_root(root);
1589 path = btrfs_alloc_path();
1594 key.objectid = BTRFS_ORPHAN_OBJECTID;
1595 key.type = BTRFS_ORPHAN_ITEM_KEY;
1596 key.offset = location->objectid;
1598 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1599 btrfs_free_path(path);
1603 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1605 ret = btrfs_insert_fs_root(fs_info, root);
1607 if (ret == -EEXIST) {
1608 btrfs_free_fs_root(root);
1615 btrfs_free_fs_root(root);
1616 return ERR_PTR(ret);
1619 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1621 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1623 struct btrfs_device *device;
1624 struct backing_dev_info *bdi;
1627 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1630 bdi = device->bdev->bd_bdi;
1631 if (bdi_congested(bdi, bdi_bits)) {
1641 * called by the kthread helper functions to finally call the bio end_io
1642 * functions. This is where read checksum verification actually happens
1644 static void end_workqueue_fn(struct btrfs_work *work)
1647 struct btrfs_end_io_wq *end_io_wq;
1649 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1650 bio = end_io_wq->bio;
1652 bio->bi_status = end_io_wq->status;
1653 bio->bi_private = end_io_wq->private;
1654 bio->bi_end_io = end_io_wq->end_io;
1656 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1659 static int cleaner_kthread(void *arg)
1661 struct btrfs_root *root = arg;
1662 struct btrfs_fs_info *fs_info = root->fs_info;
1668 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1670 /* Make the cleaner go to sleep early. */
1671 if (btrfs_need_cleaner_sleep(fs_info))
1675 * Do not do anything if we might cause open_ctree() to block
1676 * before we have finished mounting the filesystem.
1678 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1681 if (!mutex_trylock(&fs_info->cleaner_mutex))
1685 * Avoid the problem that we change the status of the fs
1686 * during the above check and trylock.
1688 if (btrfs_need_cleaner_sleep(fs_info)) {
1689 mutex_unlock(&fs_info->cleaner_mutex);
1693 btrfs_run_delayed_iputs(fs_info);
1695 again = btrfs_clean_one_deleted_snapshot(root);
1696 mutex_unlock(&fs_info->cleaner_mutex);
1699 * The defragger has dealt with the R/O remount and umount,
1700 * needn't do anything special here.
1702 btrfs_run_defrag_inodes(fs_info);
1705 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1706 * with relocation (btrfs_relocate_chunk) and relocation
1707 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1708 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1709 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1710 * unused block groups.
1712 btrfs_delete_unused_bgs(fs_info);
1714 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1715 if (kthread_should_park())
1717 if (kthread_should_stop())
1720 set_current_state(TASK_INTERRUPTIBLE);
1722 __set_current_state(TASK_RUNNING);
1727 static int transaction_kthread(void *arg)
1729 struct btrfs_root *root = arg;
1730 struct btrfs_fs_info *fs_info = root->fs_info;
1731 struct btrfs_trans_handle *trans;
1732 struct btrfs_transaction *cur;
1735 unsigned long delay;
1739 cannot_commit = false;
1740 delay = HZ * fs_info->commit_interval;
1741 mutex_lock(&fs_info->transaction_kthread_mutex);
1743 spin_lock(&fs_info->trans_lock);
1744 cur = fs_info->running_transaction;
1746 spin_unlock(&fs_info->trans_lock);
1750 now = ktime_get_seconds();
1751 if (cur->state < TRANS_STATE_COMMIT_START &&
1752 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1753 (now < cur->start_time ||
1754 now - cur->start_time < fs_info->commit_interval)) {
1755 spin_unlock(&fs_info->trans_lock);
1759 transid = cur->transid;
1760 spin_unlock(&fs_info->trans_lock);
1762 /* If the file system is aborted, this will always fail. */
1763 trans = btrfs_attach_transaction(root);
1764 if (IS_ERR(trans)) {
1765 if (PTR_ERR(trans) != -ENOENT)
1766 cannot_commit = true;
1769 if (transid == trans->transid) {
1770 btrfs_commit_transaction(trans);
1772 btrfs_end_transaction(trans);
1775 wake_up_process(fs_info->cleaner_kthread);
1776 mutex_unlock(&fs_info->transaction_kthread_mutex);
1778 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1779 &fs_info->fs_state)))
1780 btrfs_cleanup_transaction(fs_info);
1781 if (!kthread_should_stop() &&
1782 (!btrfs_transaction_blocked(fs_info) ||
1784 schedule_timeout_interruptible(delay);
1785 } while (!kthread_should_stop());
1790 * this will find the highest generation in the array of
1791 * root backups. The index of the highest array is returned,
1792 * or -1 if we can't find anything.
1794 * We check to make sure the array is valid by comparing the
1795 * generation of the latest root in the array with the generation
1796 * in the super block. If they don't match we pitch it.
1798 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1801 int newest_index = -1;
1802 struct btrfs_root_backup *root_backup;
1805 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1806 root_backup = info->super_copy->super_roots + i;
1807 cur = btrfs_backup_tree_root_gen(root_backup);
1808 if (cur == newest_gen)
1812 /* check to see if we actually wrapped around */
1813 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1814 root_backup = info->super_copy->super_roots;
1815 cur = btrfs_backup_tree_root_gen(root_backup);
1816 if (cur == newest_gen)
1819 return newest_index;
1824 * find the oldest backup so we know where to store new entries
1825 * in the backup array. This will set the backup_root_index
1826 * field in the fs_info struct
1828 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1831 int newest_index = -1;
1833 newest_index = find_newest_super_backup(info, newest_gen);
1834 /* if there was garbage in there, just move along */
1835 if (newest_index == -1) {
1836 info->backup_root_index = 0;
1838 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1843 * copy all the root pointers into the super backup array.
1844 * this will bump the backup pointer by one when it is
1847 static void backup_super_roots(struct btrfs_fs_info *info)
1850 struct btrfs_root_backup *root_backup;
1853 next_backup = info->backup_root_index;
1854 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1855 BTRFS_NUM_BACKUP_ROOTS;
1858 * just overwrite the last backup if we're at the same generation
1859 * this happens only at umount
1861 root_backup = info->super_for_commit->super_roots + last_backup;
1862 if (btrfs_backup_tree_root_gen(root_backup) ==
1863 btrfs_header_generation(info->tree_root->node))
1864 next_backup = last_backup;
1866 root_backup = info->super_for_commit->super_roots + next_backup;
1869 * make sure all of our padding and empty slots get zero filled
1870 * regardless of which ones we use today
1872 memset(root_backup, 0, sizeof(*root_backup));
1874 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1876 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1877 btrfs_set_backup_tree_root_gen(root_backup,
1878 btrfs_header_generation(info->tree_root->node));
1880 btrfs_set_backup_tree_root_level(root_backup,
1881 btrfs_header_level(info->tree_root->node));
1883 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1884 btrfs_set_backup_chunk_root_gen(root_backup,
1885 btrfs_header_generation(info->chunk_root->node));
1886 btrfs_set_backup_chunk_root_level(root_backup,
1887 btrfs_header_level(info->chunk_root->node));
1889 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1890 btrfs_set_backup_extent_root_gen(root_backup,
1891 btrfs_header_generation(info->extent_root->node));
1892 btrfs_set_backup_extent_root_level(root_backup,
1893 btrfs_header_level(info->extent_root->node));
1896 * we might commit during log recovery, which happens before we set
1897 * the fs_root. Make sure it is valid before we fill it in.
1899 if (info->fs_root && info->fs_root->node) {
1900 btrfs_set_backup_fs_root(root_backup,
1901 info->fs_root->node->start);
1902 btrfs_set_backup_fs_root_gen(root_backup,
1903 btrfs_header_generation(info->fs_root->node));
1904 btrfs_set_backup_fs_root_level(root_backup,
1905 btrfs_header_level(info->fs_root->node));
1908 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1909 btrfs_set_backup_dev_root_gen(root_backup,
1910 btrfs_header_generation(info->dev_root->node));
1911 btrfs_set_backup_dev_root_level(root_backup,
1912 btrfs_header_level(info->dev_root->node));
1914 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1915 btrfs_set_backup_csum_root_gen(root_backup,
1916 btrfs_header_generation(info->csum_root->node));
1917 btrfs_set_backup_csum_root_level(root_backup,
1918 btrfs_header_level(info->csum_root->node));
1920 btrfs_set_backup_total_bytes(root_backup,
1921 btrfs_super_total_bytes(info->super_copy));
1922 btrfs_set_backup_bytes_used(root_backup,
1923 btrfs_super_bytes_used(info->super_copy));
1924 btrfs_set_backup_num_devices(root_backup,
1925 btrfs_super_num_devices(info->super_copy));
1928 * if we don't copy this out to the super_copy, it won't get remembered
1929 * for the next commit
1931 memcpy(&info->super_copy->super_roots,
1932 &info->super_for_commit->super_roots,
1933 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1937 * this copies info out of the root backup array and back into
1938 * the in-memory super block. It is meant to help iterate through
1939 * the array, so you send it the number of backups you've already
1940 * tried and the last backup index you used.
1942 * this returns -1 when it has tried all the backups
1944 static noinline int next_root_backup(struct btrfs_fs_info *info,
1945 struct btrfs_super_block *super,
1946 int *num_backups_tried, int *backup_index)
1948 struct btrfs_root_backup *root_backup;
1949 int newest = *backup_index;
1951 if (*num_backups_tried == 0) {
1952 u64 gen = btrfs_super_generation(super);
1954 newest = find_newest_super_backup(info, gen);
1958 *backup_index = newest;
1959 *num_backups_tried = 1;
1960 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1961 /* we've tried all the backups, all done */
1964 /* jump to the next oldest backup */
1965 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1966 BTRFS_NUM_BACKUP_ROOTS;
1967 *backup_index = newest;
1968 *num_backups_tried += 1;
1970 root_backup = super->super_roots + newest;
1972 btrfs_set_super_generation(super,
1973 btrfs_backup_tree_root_gen(root_backup));
1974 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1975 btrfs_set_super_root_level(super,
1976 btrfs_backup_tree_root_level(root_backup));
1977 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1980 * fixme: the total bytes and num_devices need to match or we should
1983 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1984 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1988 /* helper to cleanup workers */
1989 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1991 btrfs_destroy_workqueue(fs_info->fixup_workers);
1992 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1993 btrfs_destroy_workqueue(fs_info->workers);
1994 btrfs_destroy_workqueue(fs_info->endio_workers);
1995 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1996 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1997 btrfs_destroy_workqueue(fs_info->rmw_workers);
1998 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1999 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2000 btrfs_destroy_workqueue(fs_info->delayed_workers);
2001 btrfs_destroy_workqueue(fs_info->caching_workers);
2002 btrfs_destroy_workqueue(fs_info->readahead_workers);
2003 btrfs_destroy_workqueue(fs_info->flush_workers);
2004 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2006 * Now that all other work queues are destroyed, we can safely destroy
2007 * the queues used for metadata I/O, since tasks from those other work
2008 * queues can do metadata I/O operations.
2010 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2011 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2014 static void free_root_extent_buffers(struct btrfs_root *root)
2017 free_extent_buffer(root->node);
2018 free_extent_buffer(root->commit_root);
2020 root->commit_root = NULL;
2024 /* helper to cleanup tree roots */
2025 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2027 free_root_extent_buffers(info->tree_root);
2029 free_root_extent_buffers(info->dev_root);
2030 free_root_extent_buffers(info->extent_root);
2031 free_root_extent_buffers(info->csum_root);
2032 free_root_extent_buffers(info->quota_root);
2033 free_root_extent_buffers(info->uuid_root);
2035 free_root_extent_buffers(info->chunk_root);
2036 free_root_extent_buffers(info->free_space_root);
2039 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2042 struct btrfs_root *gang[8];
2045 while (!list_empty(&fs_info->dead_roots)) {
2046 gang[0] = list_entry(fs_info->dead_roots.next,
2047 struct btrfs_root, root_list);
2048 list_del(&gang[0]->root_list);
2050 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2051 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2053 free_extent_buffer(gang[0]->node);
2054 free_extent_buffer(gang[0]->commit_root);
2055 btrfs_put_fs_root(gang[0]);
2060 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2065 for (i = 0; i < ret; i++)
2066 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2069 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2070 btrfs_free_log_root_tree(NULL, fs_info);
2071 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2075 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2077 mutex_init(&fs_info->scrub_lock);
2078 atomic_set(&fs_info->scrubs_running, 0);
2079 atomic_set(&fs_info->scrub_pause_req, 0);
2080 atomic_set(&fs_info->scrubs_paused, 0);
2081 atomic_set(&fs_info->scrub_cancel_req, 0);
2082 init_waitqueue_head(&fs_info->scrub_pause_wait);
2083 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2086 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2088 spin_lock_init(&fs_info->balance_lock);
2089 mutex_init(&fs_info->balance_mutex);
2090 atomic_set(&fs_info->balance_pause_req, 0);
2091 atomic_set(&fs_info->balance_cancel_req, 0);
2092 fs_info->balance_ctl = NULL;
2093 init_waitqueue_head(&fs_info->balance_wait_q);
2096 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2098 struct inode *inode = fs_info->btree_inode;
2100 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2101 set_nlink(inode, 1);
2103 * we set the i_size on the btree inode to the max possible int.
2104 * the real end of the address space is determined by all of
2105 * the devices in the system
2107 inode->i_size = OFFSET_MAX;
2108 inode->i_mapping->a_ops = &btree_aops;
2110 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2111 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2112 IO_TREE_INODE_IO, inode);
2113 BTRFS_I(inode)->io_tree.track_uptodate = false;
2114 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2116 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2118 BTRFS_I(inode)->root = fs_info->tree_root;
2119 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2120 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2121 btrfs_insert_inode_hash(inode);
2124 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2126 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2127 init_rwsem(&fs_info->dev_replace.rwsem);
2128 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2131 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2133 spin_lock_init(&fs_info->qgroup_lock);
2134 mutex_init(&fs_info->qgroup_ioctl_lock);
2135 fs_info->qgroup_tree = RB_ROOT;
2136 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2137 fs_info->qgroup_seq = 1;
2138 fs_info->qgroup_ulist = NULL;
2139 fs_info->qgroup_rescan_running = false;
2140 mutex_init(&fs_info->qgroup_rescan_lock);
2143 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2144 struct btrfs_fs_devices *fs_devices)
2146 u32 max_active = fs_info->thread_pool_size;
2147 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2150 btrfs_alloc_workqueue(fs_info, "worker",
2151 flags | WQ_HIGHPRI, max_active, 16);
2153 fs_info->delalloc_workers =
2154 btrfs_alloc_workqueue(fs_info, "delalloc",
2155 flags, max_active, 2);
2157 fs_info->flush_workers =
2158 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2159 flags, max_active, 0);
2161 fs_info->caching_workers =
2162 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2164 fs_info->fixup_workers =
2165 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2168 * endios are largely parallel and should have a very
2171 fs_info->endio_workers =
2172 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2173 fs_info->endio_meta_workers =
2174 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2176 fs_info->endio_meta_write_workers =
2177 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2179 fs_info->endio_raid56_workers =
2180 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2182 fs_info->endio_repair_workers =
2183 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2184 fs_info->rmw_workers =
2185 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2186 fs_info->endio_write_workers =
2187 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2189 fs_info->endio_freespace_worker =
2190 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2192 fs_info->delayed_workers =
2193 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2195 fs_info->readahead_workers =
2196 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2198 fs_info->qgroup_rescan_workers =
2199 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2201 if (!(fs_info->workers && fs_info->delalloc_workers &&
2202 fs_info->flush_workers &&
2203 fs_info->endio_workers && fs_info->endio_meta_workers &&
2204 fs_info->endio_meta_write_workers &&
2205 fs_info->endio_repair_workers &&
2206 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2207 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2208 fs_info->caching_workers && fs_info->readahead_workers &&
2209 fs_info->fixup_workers && fs_info->delayed_workers &&
2210 fs_info->qgroup_rescan_workers)) {
2217 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2219 struct crypto_shash *csum_shash;
2220 const char *csum_name = btrfs_super_csum_name(csum_type);
2222 csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2224 if (IS_ERR(csum_shash)) {
2225 btrfs_err(fs_info, "error allocating %s hash for checksum",
2227 return PTR_ERR(csum_shash);
2230 fs_info->csum_shash = csum_shash;
2235 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2237 crypto_free_shash(fs_info->csum_shash);
2240 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2241 struct btrfs_fs_devices *fs_devices)
2244 struct btrfs_root *log_tree_root;
2245 struct btrfs_super_block *disk_super = fs_info->super_copy;
2246 u64 bytenr = btrfs_super_log_root(disk_super);
2247 int level = btrfs_super_log_root_level(disk_super);
2249 if (fs_devices->rw_devices == 0) {
2250 btrfs_warn(fs_info, "log replay required on RO media");
2254 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2258 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2260 log_tree_root->node = read_tree_block(fs_info, bytenr,
2261 fs_info->generation + 1,
2263 if (IS_ERR(log_tree_root->node)) {
2264 btrfs_warn(fs_info, "failed to read log tree");
2265 ret = PTR_ERR(log_tree_root->node);
2266 kfree(log_tree_root);
2268 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2269 btrfs_err(fs_info, "failed to read log tree");
2270 free_extent_buffer(log_tree_root->node);
2271 kfree(log_tree_root);
2274 /* returns with log_tree_root freed on success */
2275 ret = btrfs_recover_log_trees(log_tree_root);
2277 btrfs_handle_fs_error(fs_info, ret,
2278 "Failed to recover log tree");
2279 free_extent_buffer(log_tree_root->node);
2280 kfree(log_tree_root);
2284 if (sb_rdonly(fs_info->sb)) {
2285 ret = btrfs_commit_super(fs_info);
2293 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2295 struct btrfs_root *tree_root = fs_info->tree_root;
2296 struct btrfs_root *root;
2297 struct btrfs_key location;
2300 BUG_ON(!fs_info->tree_root);
2302 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2303 location.type = BTRFS_ROOT_ITEM_KEY;
2304 location.offset = 0;
2306 root = btrfs_read_tree_root(tree_root, &location);
2308 ret = PTR_ERR(root);
2311 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2312 fs_info->extent_root = root;
2314 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2315 root = btrfs_read_tree_root(tree_root, &location);
2317 ret = PTR_ERR(root);
2320 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2321 fs_info->dev_root = root;
2322 btrfs_init_devices_late(fs_info);
2324 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2325 root = btrfs_read_tree_root(tree_root, &location);
2327 ret = PTR_ERR(root);
2330 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2331 fs_info->csum_root = root;
2333 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2334 root = btrfs_read_tree_root(tree_root, &location);
2335 if (!IS_ERR(root)) {
2336 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2337 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2338 fs_info->quota_root = root;
2341 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2342 root = btrfs_read_tree_root(tree_root, &location);
2344 ret = PTR_ERR(root);
2348 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2349 fs_info->uuid_root = root;
2352 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2353 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2354 root = btrfs_read_tree_root(tree_root, &location);
2356 ret = PTR_ERR(root);
2359 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2360 fs_info->free_space_root = root;
2365 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2366 location.objectid, ret);
2371 * Real super block validation
2372 * NOTE: super csum type and incompat features will not be checked here.
2374 * @sb: super block to check
2375 * @mirror_num: the super block number to check its bytenr:
2376 * 0 the primary (1st) sb
2377 * 1, 2 2nd and 3rd backup copy
2378 * -1 skip bytenr check
2380 static int validate_super(struct btrfs_fs_info *fs_info,
2381 struct btrfs_super_block *sb, int mirror_num)
2383 u64 nodesize = btrfs_super_nodesize(sb);
2384 u64 sectorsize = btrfs_super_sectorsize(sb);
2387 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2388 btrfs_err(fs_info, "no valid FS found");
2391 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2392 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2393 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2396 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2397 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2398 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2401 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2402 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2403 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2406 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2407 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2408 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2413 * Check sectorsize and nodesize first, other check will need it.
2414 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2416 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2417 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2418 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2421 /* Only PAGE SIZE is supported yet */
2422 if (sectorsize != PAGE_SIZE) {
2424 "sectorsize %llu not supported yet, only support %lu",
2425 sectorsize, PAGE_SIZE);
2428 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2429 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2430 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2433 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2434 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2435 le32_to_cpu(sb->__unused_leafsize), nodesize);
2439 /* Root alignment check */
2440 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2441 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2442 btrfs_super_root(sb));
2445 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2446 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2447 btrfs_super_chunk_root(sb));
2450 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2451 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2452 btrfs_super_log_root(sb));
2456 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2457 BTRFS_FSID_SIZE) != 0) {
2459 "dev_item UUID does not match metadata fsid: %pU != %pU",
2460 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2465 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2468 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2469 btrfs_err(fs_info, "bytes_used is too small %llu",
2470 btrfs_super_bytes_used(sb));
2473 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2474 btrfs_err(fs_info, "invalid stripesize %u",
2475 btrfs_super_stripesize(sb));
2478 if (btrfs_super_num_devices(sb) > (1UL << 31))
2479 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2480 btrfs_super_num_devices(sb));
2481 if (btrfs_super_num_devices(sb) == 0) {
2482 btrfs_err(fs_info, "number of devices is 0");
2486 if (mirror_num >= 0 &&
2487 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2488 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2489 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2494 * Obvious sys_chunk_array corruptions, it must hold at least one key
2497 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2498 btrfs_err(fs_info, "system chunk array too big %u > %u",
2499 btrfs_super_sys_array_size(sb),
2500 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2503 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2504 + sizeof(struct btrfs_chunk)) {
2505 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2506 btrfs_super_sys_array_size(sb),
2507 sizeof(struct btrfs_disk_key)
2508 + sizeof(struct btrfs_chunk));
2513 * The generation is a global counter, we'll trust it more than the others
2514 * but it's still possible that it's the one that's wrong.
2516 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2518 "suspicious: generation < chunk_root_generation: %llu < %llu",
2519 btrfs_super_generation(sb),
2520 btrfs_super_chunk_root_generation(sb));
2521 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2522 && btrfs_super_cache_generation(sb) != (u64)-1)
2524 "suspicious: generation < cache_generation: %llu < %llu",
2525 btrfs_super_generation(sb),
2526 btrfs_super_cache_generation(sb));
2532 * Validation of super block at mount time.
2533 * Some checks already done early at mount time, like csum type and incompat
2534 * flags will be skipped.
2536 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2538 return validate_super(fs_info, fs_info->super_copy, 0);
2542 * Validation of super block at write time.
2543 * Some checks like bytenr check will be skipped as their values will be
2545 * Extra checks like csum type and incompat flags will be done here.
2547 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2548 struct btrfs_super_block *sb)
2552 ret = validate_super(fs_info, sb, -1);
2555 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2557 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2558 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2561 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2564 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2565 btrfs_super_incompat_flags(sb),
2566 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2572 "super block corruption detected before writing it to disk");
2576 int __cold open_ctree(struct super_block *sb,
2577 struct btrfs_fs_devices *fs_devices,
2586 struct btrfs_key location;
2587 struct buffer_head *bh;
2588 struct btrfs_super_block *disk_super;
2589 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2590 struct btrfs_root *tree_root;
2591 struct btrfs_root *chunk_root;
2594 int num_backups_tried = 0;
2595 int backup_index = 0;
2596 int clear_free_space_tree = 0;
2599 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2600 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2601 if (!tree_root || !chunk_root) {
2606 ret = init_srcu_struct(&fs_info->subvol_srcu);
2612 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2618 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2621 goto fail_dio_bytes;
2623 fs_info->dirty_metadata_batch = PAGE_SIZE *
2624 (1 + ilog2(nr_cpu_ids));
2626 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2629 goto fail_dirty_metadata_bytes;
2632 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2636 goto fail_delalloc_bytes;
2639 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2640 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2641 INIT_LIST_HEAD(&fs_info->trans_list);
2642 INIT_LIST_HEAD(&fs_info->dead_roots);
2643 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2644 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2645 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2646 spin_lock_init(&fs_info->delalloc_root_lock);
2647 spin_lock_init(&fs_info->trans_lock);
2648 spin_lock_init(&fs_info->fs_roots_radix_lock);
2649 spin_lock_init(&fs_info->delayed_iput_lock);
2650 spin_lock_init(&fs_info->defrag_inodes_lock);
2651 spin_lock_init(&fs_info->tree_mod_seq_lock);
2652 spin_lock_init(&fs_info->super_lock);
2653 spin_lock_init(&fs_info->buffer_lock);
2654 spin_lock_init(&fs_info->unused_bgs_lock);
2655 rwlock_init(&fs_info->tree_mod_log_lock);
2656 mutex_init(&fs_info->unused_bg_unpin_mutex);
2657 mutex_init(&fs_info->delete_unused_bgs_mutex);
2658 mutex_init(&fs_info->reloc_mutex);
2659 mutex_init(&fs_info->delalloc_root_mutex);
2660 seqlock_init(&fs_info->profiles_lock);
2662 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2663 INIT_LIST_HEAD(&fs_info->space_info);
2664 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2665 INIT_LIST_HEAD(&fs_info->unused_bgs);
2666 extent_map_tree_init(&fs_info->mapping_tree);
2667 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2668 BTRFS_BLOCK_RSV_GLOBAL);
2669 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2670 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2671 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2672 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2673 BTRFS_BLOCK_RSV_DELOPS);
2674 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2675 BTRFS_BLOCK_RSV_DELREFS);
2677 atomic_set(&fs_info->async_delalloc_pages, 0);
2678 atomic_set(&fs_info->defrag_running, 0);
2679 atomic_set(&fs_info->reada_works_cnt, 0);
2680 atomic_set(&fs_info->nr_delayed_iputs, 0);
2681 atomic64_set(&fs_info->tree_mod_seq, 0);
2683 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2684 fs_info->metadata_ratio = 0;
2685 fs_info->defrag_inodes = RB_ROOT;
2686 atomic64_set(&fs_info->free_chunk_space, 0);
2687 fs_info->tree_mod_log = RB_ROOT;
2688 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2689 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2690 /* readahead state */
2691 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2692 spin_lock_init(&fs_info->reada_lock);
2693 btrfs_init_ref_verify(fs_info);
2695 fs_info->thread_pool_size = min_t(unsigned long,
2696 num_online_cpus() + 2, 8);
2698 INIT_LIST_HEAD(&fs_info->ordered_roots);
2699 spin_lock_init(&fs_info->ordered_root_lock);
2701 fs_info->btree_inode = new_inode(sb);
2702 if (!fs_info->btree_inode) {
2704 goto fail_bio_counter;
2706 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2708 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2710 if (!fs_info->delayed_root) {
2714 btrfs_init_delayed_root(fs_info->delayed_root);
2716 btrfs_init_scrub(fs_info);
2717 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2718 fs_info->check_integrity_print_mask = 0;
2720 btrfs_init_balance(fs_info);
2721 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2723 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2724 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2726 btrfs_init_btree_inode(fs_info);
2728 spin_lock_init(&fs_info->block_group_cache_lock);
2729 fs_info->block_group_cache_tree = RB_ROOT;
2730 fs_info->first_logical_byte = (u64)-1;
2732 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2733 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2734 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2735 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2736 fs_info->pinned_extents = &fs_info->freed_extents[0];
2737 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2739 mutex_init(&fs_info->ordered_operations_mutex);
2740 mutex_init(&fs_info->tree_log_mutex);
2741 mutex_init(&fs_info->chunk_mutex);
2742 mutex_init(&fs_info->transaction_kthread_mutex);
2743 mutex_init(&fs_info->cleaner_mutex);
2744 mutex_init(&fs_info->ro_block_group_mutex);
2745 init_rwsem(&fs_info->commit_root_sem);
2746 init_rwsem(&fs_info->cleanup_work_sem);
2747 init_rwsem(&fs_info->subvol_sem);
2748 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2750 btrfs_init_dev_replace_locks(fs_info);
2751 btrfs_init_qgroup(fs_info);
2753 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2754 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2756 init_waitqueue_head(&fs_info->transaction_throttle);
2757 init_waitqueue_head(&fs_info->transaction_wait);
2758 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2759 init_waitqueue_head(&fs_info->async_submit_wait);
2760 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2762 /* Usable values until the real ones are cached from the superblock */
2763 fs_info->nodesize = 4096;
2764 fs_info->sectorsize = 4096;
2765 fs_info->stripesize = 4096;
2767 spin_lock_init(&fs_info->swapfile_pins_lock);
2768 fs_info->swapfile_pins = RB_ROOT;
2770 fs_info->send_in_progress = 0;
2772 ret = btrfs_alloc_stripe_hash_table(fs_info);
2778 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2780 invalidate_bdev(fs_devices->latest_bdev);
2783 * Read super block and check the signature bytes only
2785 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2792 * Verify the type first, if that or the the checksum value are
2793 * corrupted, we'll find out
2795 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2796 if (!btrfs_supported_super_csum(csum_type)) {
2797 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2804 ret = btrfs_init_csum_hash(fs_info, csum_type);
2811 * We want to check superblock checksum, the type is stored inside.
2812 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2814 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2815 btrfs_err(fs_info, "superblock checksum mismatch");
2822 * super_copy is zeroed at allocation time and we never touch the
2823 * following bytes up to INFO_SIZE, the checksum is calculated from
2824 * the whole block of INFO_SIZE
2826 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2829 disk_super = fs_info->super_copy;
2831 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2834 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2835 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2836 fs_info->super_copy->metadata_uuid,
2840 features = btrfs_super_flags(disk_super);
2841 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2842 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2843 btrfs_set_super_flags(disk_super, features);
2845 "found metadata UUID change in progress flag, clearing");
2848 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2849 sizeof(*fs_info->super_for_commit));
2851 ret = btrfs_validate_mount_super(fs_info);
2853 btrfs_err(fs_info, "superblock contains fatal errors");
2858 if (!btrfs_super_root(disk_super))
2861 /* check FS state, whether FS is broken. */
2862 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2863 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2866 * run through our array of backup supers and setup
2867 * our ring pointer to the oldest one
2869 generation = btrfs_super_generation(disk_super);
2870 find_oldest_super_backup(fs_info, generation);
2873 * In the long term, we'll store the compression type in the super
2874 * block, and it'll be used for per file compression control.
2876 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2878 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2884 features = btrfs_super_incompat_flags(disk_super) &
2885 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2888 "cannot mount because of unsupported optional features (%llx)",
2894 features = btrfs_super_incompat_flags(disk_super);
2895 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2896 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2897 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2898 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2899 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2901 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2902 btrfs_info(fs_info, "has skinny extents");
2905 * flag our filesystem as having big metadata blocks if
2906 * they are bigger than the page size
2908 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2909 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2911 "flagging fs with big metadata feature");
2912 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2915 nodesize = btrfs_super_nodesize(disk_super);
2916 sectorsize = btrfs_super_sectorsize(disk_super);
2917 stripesize = sectorsize;
2918 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2919 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2921 /* Cache block sizes */
2922 fs_info->nodesize = nodesize;
2923 fs_info->sectorsize = sectorsize;
2924 fs_info->stripesize = stripesize;
2927 * mixed block groups end up with duplicate but slightly offset
2928 * extent buffers for the same range. It leads to corruptions
2930 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2931 (sectorsize != nodesize)) {
2933 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2934 nodesize, sectorsize);
2939 * Needn't use the lock because there is no other task which will
2942 btrfs_set_super_incompat_flags(disk_super, features);
2944 features = btrfs_super_compat_ro_flags(disk_super) &
2945 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2946 if (!sb_rdonly(sb) && features) {
2948 "cannot mount read-write because of unsupported optional features (%llx)",
2954 ret = btrfs_init_workqueues(fs_info, fs_devices);
2957 goto fail_sb_buffer;
2960 sb->s_bdi->congested_fn = btrfs_congested_fn;
2961 sb->s_bdi->congested_data = fs_info;
2962 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2963 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2964 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2965 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2967 sb->s_blocksize = sectorsize;
2968 sb->s_blocksize_bits = blksize_bits(sectorsize);
2969 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2971 mutex_lock(&fs_info->chunk_mutex);
2972 ret = btrfs_read_sys_array(fs_info);
2973 mutex_unlock(&fs_info->chunk_mutex);
2975 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2976 goto fail_sb_buffer;
2979 generation = btrfs_super_chunk_root_generation(disk_super);
2980 level = btrfs_super_chunk_root_level(disk_super);
2982 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2984 chunk_root->node = read_tree_block(fs_info,
2985 btrfs_super_chunk_root(disk_super),
2986 generation, level, NULL);
2987 if (IS_ERR(chunk_root->node) ||
2988 !extent_buffer_uptodate(chunk_root->node)) {
2989 btrfs_err(fs_info, "failed to read chunk root");
2990 if (!IS_ERR(chunk_root->node))
2991 free_extent_buffer(chunk_root->node);
2992 chunk_root->node = NULL;
2993 goto fail_tree_roots;
2995 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2996 chunk_root->commit_root = btrfs_root_node(chunk_root);
2998 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2999 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3001 ret = btrfs_read_chunk_tree(fs_info);
3003 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3004 goto fail_tree_roots;
3008 * Keep the devid that is marked to be the target device for the
3009 * device replace procedure
3011 btrfs_free_extra_devids(fs_devices, 0);
3013 if (!fs_devices->latest_bdev) {
3014 btrfs_err(fs_info, "failed to read devices");
3015 goto fail_tree_roots;
3019 generation = btrfs_super_generation(disk_super);
3020 level = btrfs_super_root_level(disk_super);
3022 tree_root->node = read_tree_block(fs_info,
3023 btrfs_super_root(disk_super),
3024 generation, level, NULL);
3025 if (IS_ERR(tree_root->node) ||
3026 !extent_buffer_uptodate(tree_root->node)) {
3027 btrfs_warn(fs_info, "failed to read tree root");
3028 if (!IS_ERR(tree_root->node))
3029 free_extent_buffer(tree_root->node);
3030 tree_root->node = NULL;
3031 goto recovery_tree_root;
3034 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3035 tree_root->commit_root = btrfs_root_node(tree_root);
3036 btrfs_set_root_refs(&tree_root->root_item, 1);
3038 mutex_lock(&tree_root->objectid_mutex);
3039 ret = btrfs_find_highest_objectid(tree_root,
3040 &tree_root->highest_objectid);
3042 mutex_unlock(&tree_root->objectid_mutex);
3043 goto recovery_tree_root;
3046 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3048 mutex_unlock(&tree_root->objectid_mutex);
3050 ret = btrfs_read_roots(fs_info);
3052 goto recovery_tree_root;
3054 fs_info->generation = generation;
3055 fs_info->last_trans_committed = generation;
3057 ret = btrfs_verify_dev_extents(fs_info);
3060 "failed to verify dev extents against chunks: %d",
3062 goto fail_block_groups;
3064 ret = btrfs_recover_balance(fs_info);
3066 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3067 goto fail_block_groups;
3070 ret = btrfs_init_dev_stats(fs_info);
3072 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3073 goto fail_block_groups;
3076 ret = btrfs_init_dev_replace(fs_info);
3078 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3079 goto fail_block_groups;
3082 btrfs_free_extra_devids(fs_devices, 1);
3084 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3086 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3088 goto fail_block_groups;
3091 ret = btrfs_sysfs_add_device(fs_devices);
3093 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3095 goto fail_fsdev_sysfs;
3098 ret = btrfs_sysfs_add_mounted(fs_info);
3100 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3101 goto fail_fsdev_sysfs;
3104 ret = btrfs_init_space_info(fs_info);
3106 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3110 ret = btrfs_read_block_groups(fs_info);
3112 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3116 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3118 "writable mount is not allowed due to too many missing devices");
3122 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3124 if (IS_ERR(fs_info->cleaner_kthread))
3127 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3129 "btrfs-transaction");
3130 if (IS_ERR(fs_info->transaction_kthread))
3133 if (!btrfs_test_opt(fs_info, NOSSD) &&
3134 !fs_info->fs_devices->rotating) {
3135 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3139 * Mount does not set all options immediately, we can do it now and do
3140 * not have to wait for transaction commit
3142 btrfs_apply_pending_changes(fs_info);
3144 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3145 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3146 ret = btrfsic_mount(fs_info, fs_devices,
3147 btrfs_test_opt(fs_info,
3148 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3150 fs_info->check_integrity_print_mask);
3153 "failed to initialize integrity check module: %d",
3157 ret = btrfs_read_qgroup_config(fs_info);
3159 goto fail_trans_kthread;
3161 if (btrfs_build_ref_tree(fs_info))
3162 btrfs_err(fs_info, "couldn't build ref tree");
3164 /* do not make disk changes in broken FS or nologreplay is given */
3165 if (btrfs_super_log_root(disk_super) != 0 &&
3166 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3167 ret = btrfs_replay_log(fs_info, fs_devices);
3174 ret = btrfs_find_orphan_roots(fs_info);
3178 if (!sb_rdonly(sb)) {
3179 ret = btrfs_cleanup_fs_roots(fs_info);
3183 mutex_lock(&fs_info->cleaner_mutex);
3184 ret = btrfs_recover_relocation(tree_root);
3185 mutex_unlock(&fs_info->cleaner_mutex);
3187 btrfs_warn(fs_info, "failed to recover relocation: %d",
3194 location.objectid = BTRFS_FS_TREE_OBJECTID;
3195 location.type = BTRFS_ROOT_ITEM_KEY;
3196 location.offset = 0;
3198 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3199 if (IS_ERR(fs_info->fs_root)) {
3200 err = PTR_ERR(fs_info->fs_root);
3201 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3208 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3209 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3210 clear_free_space_tree = 1;
3211 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3212 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3213 btrfs_warn(fs_info, "free space tree is invalid");
3214 clear_free_space_tree = 1;
3217 if (clear_free_space_tree) {
3218 btrfs_info(fs_info, "clearing free space tree");
3219 ret = btrfs_clear_free_space_tree(fs_info);
3222 "failed to clear free space tree: %d", ret);
3223 close_ctree(fs_info);
3228 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3229 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3230 btrfs_info(fs_info, "creating free space tree");
3231 ret = btrfs_create_free_space_tree(fs_info);
3234 "failed to create free space tree: %d", ret);
3235 close_ctree(fs_info);
3240 down_read(&fs_info->cleanup_work_sem);
3241 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3242 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3243 up_read(&fs_info->cleanup_work_sem);
3244 close_ctree(fs_info);
3247 up_read(&fs_info->cleanup_work_sem);
3249 ret = btrfs_resume_balance_async(fs_info);
3251 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3252 close_ctree(fs_info);
3256 ret = btrfs_resume_dev_replace_async(fs_info);
3258 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3259 close_ctree(fs_info);
3263 btrfs_qgroup_rescan_resume(fs_info);
3265 if (!fs_info->uuid_root) {
3266 btrfs_info(fs_info, "creating UUID tree");
3267 ret = btrfs_create_uuid_tree(fs_info);
3270 "failed to create the UUID tree: %d", ret);
3271 close_ctree(fs_info);
3274 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3275 fs_info->generation !=
3276 btrfs_super_uuid_tree_generation(disk_super)) {
3277 btrfs_info(fs_info, "checking UUID tree");
3278 ret = btrfs_check_uuid_tree(fs_info);
3281 "failed to check the UUID tree: %d", ret);
3282 close_ctree(fs_info);
3286 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3288 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3291 * backuproot only affect mount behavior, and if open_ctree succeeded,
3292 * no need to keep the flag
3294 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3299 btrfs_free_qgroup_config(fs_info);
3301 kthread_stop(fs_info->transaction_kthread);
3302 btrfs_cleanup_transaction(fs_info);
3303 btrfs_free_fs_roots(fs_info);
3305 kthread_stop(fs_info->cleaner_kthread);
3308 * make sure we're done with the btree inode before we stop our
3311 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3314 btrfs_sysfs_remove_mounted(fs_info);
3317 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3320 btrfs_put_block_group_cache(fs_info);
3323 free_root_pointers(fs_info, 1);
3324 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3327 btrfs_stop_all_workers(fs_info);
3328 btrfs_free_block_groups(fs_info);
3330 btrfs_free_csum_hash(fs_info);
3333 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3335 iput(fs_info->btree_inode);
3337 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3338 fail_delalloc_bytes:
3339 percpu_counter_destroy(&fs_info->delalloc_bytes);
3340 fail_dirty_metadata_bytes:
3341 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3343 percpu_counter_destroy(&fs_info->dio_bytes);
3345 cleanup_srcu_struct(&fs_info->subvol_srcu);
3347 btrfs_free_stripe_hash_table(fs_info);
3348 btrfs_close_devices(fs_info->fs_devices);
3352 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3353 goto fail_tree_roots;
3355 free_root_pointers(fs_info, 0);
3357 /* don't use the log in recovery mode, it won't be valid */
3358 btrfs_set_super_log_root(disk_super, 0);
3360 /* we can't trust the free space cache either */
3361 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3363 ret = next_root_backup(fs_info, fs_info->super_copy,
3364 &num_backups_tried, &backup_index);
3366 goto fail_block_groups;
3367 goto retry_root_backup;
3369 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3371 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3374 set_buffer_uptodate(bh);
3376 struct btrfs_device *device = (struct btrfs_device *)
3379 btrfs_warn_rl_in_rcu(device->fs_info,
3380 "lost page write due to IO error on %s",
3381 rcu_str_deref(device->name));
3382 /* note, we don't set_buffer_write_io_error because we have
3383 * our own ways of dealing with the IO errors
3385 clear_buffer_uptodate(bh);
3386 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3392 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3393 struct buffer_head **bh_ret)
3395 struct buffer_head *bh;
3396 struct btrfs_super_block *super;
3399 bytenr = btrfs_sb_offset(copy_num);
3400 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3403 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3405 * If we fail to read from the underlying devices, as of now
3406 * the best option we have is to mark it EIO.
3411 super = (struct btrfs_super_block *)bh->b_data;
3412 if (btrfs_super_bytenr(super) != bytenr ||
3413 btrfs_super_magic(super) != BTRFS_MAGIC) {
3423 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3425 struct buffer_head *bh;
3426 struct buffer_head *latest = NULL;
3427 struct btrfs_super_block *super;
3432 /* we would like to check all the supers, but that would make
3433 * a btrfs mount succeed after a mkfs from a different FS.
3434 * So, we need to add a special mount option to scan for
3435 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3437 for (i = 0; i < 1; i++) {
3438 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3442 super = (struct btrfs_super_block *)bh->b_data;
3444 if (!latest || btrfs_super_generation(super) > transid) {
3447 transid = btrfs_super_generation(super);
3454 return ERR_PTR(ret);
3460 * Write superblock @sb to the @device. Do not wait for completion, all the
3461 * buffer heads we write are pinned.
3463 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3464 * the expected device size at commit time. Note that max_mirrors must be
3465 * same for write and wait phases.
3467 * Return number of errors when buffer head is not found or submission fails.
3469 static int write_dev_supers(struct btrfs_device *device,
3470 struct btrfs_super_block *sb, int max_mirrors)
3472 struct btrfs_fs_info *fs_info = device->fs_info;
3473 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3474 struct buffer_head *bh;
3481 if (max_mirrors == 0)
3482 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3484 shash->tfm = fs_info->csum_shash;
3486 for (i = 0; i < max_mirrors; i++) {
3487 bytenr = btrfs_sb_offset(i);
3488 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3489 device->commit_total_bytes)
3492 btrfs_set_super_bytenr(sb, bytenr);
3494 crypto_shash_init(shash);
3495 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3496 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3497 crypto_shash_final(shash, sb->csum);
3499 /* One reference for us, and we leave it for the caller */
3500 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3501 BTRFS_SUPER_INFO_SIZE);
3503 btrfs_err(device->fs_info,
3504 "couldn't get super buffer head for bytenr %llu",
3510 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3512 /* one reference for submit_bh */
3515 set_buffer_uptodate(bh);
3517 bh->b_end_io = btrfs_end_buffer_write_sync;
3518 bh->b_private = device;
3521 * we fua the first super. The others we allow
3524 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3525 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3526 op_flags |= REQ_FUA;
3527 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3531 return errors < i ? 0 : -1;
3535 * Wait for write completion of superblocks done by write_dev_supers,
3536 * @max_mirrors same for write and wait phases.
3538 * Return number of errors when buffer head is not found or not marked up to
3541 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3543 struct buffer_head *bh;
3546 bool primary_failed = false;
3549 if (max_mirrors == 0)
3550 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3552 for (i = 0; i < max_mirrors; i++) {
3553 bytenr = btrfs_sb_offset(i);
3554 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3555 device->commit_total_bytes)
3558 bh = __find_get_block(device->bdev,
3559 bytenr / BTRFS_BDEV_BLOCKSIZE,
3560 BTRFS_SUPER_INFO_SIZE);
3564 primary_failed = true;
3568 if (!buffer_uptodate(bh)) {
3571 primary_failed = true;
3574 /* drop our reference */
3577 /* drop the reference from the writing run */
3581 /* log error, force error return */
3582 if (primary_failed) {
3583 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3588 return errors < i ? 0 : -1;
3592 * endio for the write_dev_flush, this will wake anyone waiting
3593 * for the barrier when it is done
3595 static void btrfs_end_empty_barrier(struct bio *bio)
3597 complete(bio->bi_private);
3601 * Submit a flush request to the device if it supports it. Error handling is
3602 * done in the waiting counterpart.
3604 static void write_dev_flush(struct btrfs_device *device)
3606 struct request_queue *q = bdev_get_queue(device->bdev);
3607 struct bio *bio = device->flush_bio;
3609 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3613 bio->bi_end_io = btrfs_end_empty_barrier;
3614 bio_set_dev(bio, device->bdev);
3615 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3616 init_completion(&device->flush_wait);
3617 bio->bi_private = &device->flush_wait;
3619 btrfsic_submit_bio(bio);
3620 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3624 * If the flush bio has been submitted by write_dev_flush, wait for it.
3626 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3628 struct bio *bio = device->flush_bio;
3630 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3633 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3634 wait_for_completion_io(&device->flush_wait);
3636 return bio->bi_status;
3639 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3641 if (!btrfs_check_rw_degradable(fs_info, NULL))
3647 * send an empty flush down to each device in parallel,
3648 * then wait for them
3650 static int barrier_all_devices(struct btrfs_fs_info *info)
3652 struct list_head *head;
3653 struct btrfs_device *dev;
3654 int errors_wait = 0;
3657 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3658 /* send down all the barriers */
3659 head = &info->fs_devices->devices;
3660 list_for_each_entry(dev, head, dev_list) {
3661 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3665 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3666 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3669 write_dev_flush(dev);
3670 dev->last_flush_error = BLK_STS_OK;
3673 /* wait for all the barriers */
3674 list_for_each_entry(dev, head, dev_list) {
3675 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3681 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3682 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3685 ret = wait_dev_flush(dev);
3687 dev->last_flush_error = ret;
3688 btrfs_dev_stat_inc_and_print(dev,
3689 BTRFS_DEV_STAT_FLUSH_ERRS);
3696 * At some point we need the status of all disks
3697 * to arrive at the volume status. So error checking
3698 * is being pushed to a separate loop.
3700 return check_barrier_error(info);
3705 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3708 int min_tolerated = INT_MAX;
3710 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3711 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3712 min_tolerated = min_t(int, min_tolerated,
3713 btrfs_raid_array[BTRFS_RAID_SINGLE].
3714 tolerated_failures);
3716 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3717 if (raid_type == BTRFS_RAID_SINGLE)
3719 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3721 min_tolerated = min_t(int, min_tolerated,
3722 btrfs_raid_array[raid_type].
3723 tolerated_failures);
3726 if (min_tolerated == INT_MAX) {
3727 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3731 return min_tolerated;
3734 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3736 struct list_head *head;
3737 struct btrfs_device *dev;
3738 struct btrfs_super_block *sb;
3739 struct btrfs_dev_item *dev_item;
3743 int total_errors = 0;
3746 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3749 * max_mirrors == 0 indicates we're from commit_transaction,
3750 * not from fsync where the tree roots in fs_info have not
3751 * been consistent on disk.
3753 if (max_mirrors == 0)
3754 backup_super_roots(fs_info);
3756 sb = fs_info->super_for_commit;
3757 dev_item = &sb->dev_item;
3759 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3760 head = &fs_info->fs_devices->devices;
3761 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3764 ret = barrier_all_devices(fs_info);
3767 &fs_info->fs_devices->device_list_mutex);
3768 btrfs_handle_fs_error(fs_info, ret,
3769 "errors while submitting device barriers.");
3774 list_for_each_entry(dev, head, dev_list) {
3779 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3780 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3783 btrfs_set_stack_device_generation(dev_item, 0);
3784 btrfs_set_stack_device_type(dev_item, dev->type);
3785 btrfs_set_stack_device_id(dev_item, dev->devid);
3786 btrfs_set_stack_device_total_bytes(dev_item,
3787 dev->commit_total_bytes);
3788 btrfs_set_stack_device_bytes_used(dev_item,
3789 dev->commit_bytes_used);
3790 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3791 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3792 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3793 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3794 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3797 flags = btrfs_super_flags(sb);
3798 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3800 ret = btrfs_validate_write_super(fs_info, sb);
3802 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3803 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3804 "unexpected superblock corruption detected");
3808 ret = write_dev_supers(dev, sb, max_mirrors);
3812 if (total_errors > max_errors) {
3813 btrfs_err(fs_info, "%d errors while writing supers",
3815 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3817 /* FUA is masked off if unsupported and can't be the reason */
3818 btrfs_handle_fs_error(fs_info, -EIO,
3819 "%d errors while writing supers",
3825 list_for_each_entry(dev, head, dev_list) {
3828 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3829 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3832 ret = wait_dev_supers(dev, max_mirrors);
3836 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3837 if (total_errors > max_errors) {
3838 btrfs_handle_fs_error(fs_info, -EIO,
3839 "%d errors while writing supers",
3846 /* Drop a fs root from the radix tree and free it. */
3847 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3848 struct btrfs_root *root)
3850 spin_lock(&fs_info->fs_roots_radix_lock);
3851 radix_tree_delete(&fs_info->fs_roots_radix,
3852 (unsigned long)root->root_key.objectid);
3853 spin_unlock(&fs_info->fs_roots_radix_lock);
3855 if (btrfs_root_refs(&root->root_item) == 0)
3856 synchronize_srcu(&fs_info->subvol_srcu);
3858 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3859 btrfs_free_log(NULL, root);
3860 if (root->reloc_root) {
3861 free_extent_buffer(root->reloc_root->node);
3862 free_extent_buffer(root->reloc_root->commit_root);
3863 btrfs_put_fs_root(root->reloc_root);
3864 root->reloc_root = NULL;
3868 if (root->free_ino_pinned)
3869 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3870 if (root->free_ino_ctl)
3871 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3872 btrfs_free_fs_root(root);
3875 void btrfs_free_fs_root(struct btrfs_root *root)
3877 iput(root->ino_cache_inode);
3878 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3880 free_anon_bdev(root->anon_dev);
3881 if (root->subv_writers)
3882 btrfs_free_subvolume_writers(root->subv_writers);
3883 free_extent_buffer(root->node);
3884 free_extent_buffer(root->commit_root);
3885 kfree(root->free_ino_ctl);
3886 kfree(root->free_ino_pinned);
3887 btrfs_put_fs_root(root);
3890 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3892 u64 root_objectid = 0;
3893 struct btrfs_root *gang[8];
3896 unsigned int ret = 0;
3900 index = srcu_read_lock(&fs_info->subvol_srcu);
3901 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3902 (void **)gang, root_objectid,
3905 srcu_read_unlock(&fs_info->subvol_srcu, index);
3908 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3910 for (i = 0; i < ret; i++) {
3911 /* Avoid to grab roots in dead_roots */
3912 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3916 /* grab all the search result for later use */
3917 gang[i] = btrfs_grab_fs_root(gang[i]);
3919 srcu_read_unlock(&fs_info->subvol_srcu, index);
3921 for (i = 0; i < ret; i++) {
3924 root_objectid = gang[i]->root_key.objectid;
3925 err = btrfs_orphan_cleanup(gang[i]);
3928 btrfs_put_fs_root(gang[i]);
3933 /* release the uncleaned roots due to error */
3934 for (; i < ret; i++) {
3936 btrfs_put_fs_root(gang[i]);
3941 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3943 struct btrfs_root *root = fs_info->tree_root;
3944 struct btrfs_trans_handle *trans;
3946 mutex_lock(&fs_info->cleaner_mutex);
3947 btrfs_run_delayed_iputs(fs_info);
3948 mutex_unlock(&fs_info->cleaner_mutex);
3949 wake_up_process(fs_info->cleaner_kthread);
3951 /* wait until ongoing cleanup work done */
3952 down_write(&fs_info->cleanup_work_sem);
3953 up_write(&fs_info->cleanup_work_sem);
3955 trans = btrfs_join_transaction(root);
3957 return PTR_ERR(trans);
3958 return btrfs_commit_transaction(trans);
3961 void __cold close_ctree(struct btrfs_fs_info *fs_info)
3965 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3967 * We don't want the cleaner to start new transactions, add more delayed
3968 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3969 * because that frees the task_struct, and the transaction kthread might
3970 * still try to wake up the cleaner.
3972 kthread_park(fs_info->cleaner_kthread);
3974 /* wait for the qgroup rescan worker to stop */
3975 btrfs_qgroup_wait_for_completion(fs_info, false);
3977 /* wait for the uuid_scan task to finish */
3978 down(&fs_info->uuid_tree_rescan_sem);
3979 /* avoid complains from lockdep et al., set sem back to initial state */
3980 up(&fs_info->uuid_tree_rescan_sem);
3982 /* pause restriper - we want to resume on mount */
3983 btrfs_pause_balance(fs_info);
3985 btrfs_dev_replace_suspend_for_unmount(fs_info);
3987 btrfs_scrub_cancel(fs_info);
3989 /* wait for any defraggers to finish */
3990 wait_event(fs_info->transaction_wait,
3991 (atomic_read(&fs_info->defrag_running) == 0));
3993 /* clear out the rbtree of defraggable inodes */
3994 btrfs_cleanup_defrag_inodes(fs_info);
3996 cancel_work_sync(&fs_info->async_reclaim_work);
3998 if (!sb_rdonly(fs_info->sb)) {
4000 * The cleaner kthread is stopped, so do one final pass over
4001 * unused block groups.
4003 btrfs_delete_unused_bgs(fs_info);
4005 ret = btrfs_commit_super(fs_info);
4007 btrfs_err(fs_info, "commit super ret %d", ret);
4010 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4011 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4012 btrfs_error_commit_super(fs_info);
4014 kthread_stop(fs_info->transaction_kthread);
4015 kthread_stop(fs_info->cleaner_kthread);
4017 ASSERT(list_empty(&fs_info->delayed_iputs));
4018 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4020 btrfs_free_qgroup_config(fs_info);
4021 ASSERT(list_empty(&fs_info->delalloc_roots));
4023 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4024 btrfs_info(fs_info, "at unmount delalloc count %lld",
4025 percpu_counter_sum(&fs_info->delalloc_bytes));
4028 if (percpu_counter_sum(&fs_info->dio_bytes))
4029 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4030 percpu_counter_sum(&fs_info->dio_bytes));
4032 btrfs_sysfs_remove_mounted(fs_info);
4033 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4035 btrfs_free_fs_roots(fs_info);
4037 btrfs_put_block_group_cache(fs_info);
4040 * we must make sure there is not any read request to
4041 * submit after we stopping all workers.
4043 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4044 btrfs_stop_all_workers(fs_info);
4046 btrfs_free_block_groups(fs_info);
4048 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4049 free_root_pointers(fs_info, 1);
4051 iput(fs_info->btree_inode);
4053 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4054 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4055 btrfsic_unmount(fs_info->fs_devices);
4058 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4059 btrfs_close_devices(fs_info->fs_devices);
4061 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4062 percpu_counter_destroy(&fs_info->delalloc_bytes);
4063 percpu_counter_destroy(&fs_info->dio_bytes);
4064 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4065 cleanup_srcu_struct(&fs_info->subvol_srcu);
4067 btrfs_free_csum_hash(fs_info);
4068 btrfs_free_stripe_hash_table(fs_info);
4069 btrfs_free_ref_cache(fs_info);
4072 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4076 struct inode *btree_inode = buf->pages[0]->mapping->host;
4078 ret = extent_buffer_uptodate(buf);
4082 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4083 parent_transid, atomic);
4089 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4091 struct btrfs_fs_info *fs_info;
4092 struct btrfs_root *root;
4093 u64 transid = btrfs_header_generation(buf);
4096 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4098 * This is a fast path so only do this check if we have sanity tests
4099 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4100 * outside of the sanity tests.
4102 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4105 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4106 fs_info = root->fs_info;
4107 btrfs_assert_tree_locked(buf);
4108 if (transid != fs_info->generation)
4109 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4110 buf->start, transid, fs_info->generation);
4111 was_dirty = set_extent_buffer_dirty(buf);
4113 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4115 fs_info->dirty_metadata_batch);
4116 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4118 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4119 * but item data not updated.
4120 * So here we should only check item pointers, not item data.
4122 if (btrfs_header_level(buf) == 0 &&
4123 btrfs_check_leaf_relaxed(buf)) {
4124 btrfs_print_leaf(buf);
4130 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4134 * looks as though older kernels can get into trouble with
4135 * this code, they end up stuck in balance_dirty_pages forever
4139 if (current->flags & PF_MEMALLOC)
4143 btrfs_balance_delayed_items(fs_info);
4145 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4146 BTRFS_DIRTY_METADATA_THRESH,
4147 fs_info->dirty_metadata_batch);
4149 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4153 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4155 __btrfs_btree_balance_dirty(fs_info, 1);
4158 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4160 __btrfs_btree_balance_dirty(fs_info, 0);
4163 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4164 struct btrfs_key *first_key)
4166 return btree_read_extent_buffer_pages(buf, parent_transid,
4170 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4172 /* cleanup FS via transaction */
4173 btrfs_cleanup_transaction(fs_info);
4175 mutex_lock(&fs_info->cleaner_mutex);
4176 btrfs_run_delayed_iputs(fs_info);
4177 mutex_unlock(&fs_info->cleaner_mutex);
4179 down_write(&fs_info->cleanup_work_sem);
4180 up_write(&fs_info->cleanup_work_sem);
4183 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4185 struct btrfs_ordered_extent *ordered;
4187 spin_lock(&root->ordered_extent_lock);
4189 * This will just short circuit the ordered completion stuff which will
4190 * make sure the ordered extent gets properly cleaned up.
4192 list_for_each_entry(ordered, &root->ordered_extents,
4194 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4195 spin_unlock(&root->ordered_extent_lock);
4198 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4200 struct btrfs_root *root;
4201 struct list_head splice;
4203 INIT_LIST_HEAD(&splice);
4205 spin_lock(&fs_info->ordered_root_lock);
4206 list_splice_init(&fs_info->ordered_roots, &splice);
4207 while (!list_empty(&splice)) {
4208 root = list_first_entry(&splice, struct btrfs_root,
4210 list_move_tail(&root->ordered_root,
4211 &fs_info->ordered_roots);
4213 spin_unlock(&fs_info->ordered_root_lock);
4214 btrfs_destroy_ordered_extents(root);
4217 spin_lock(&fs_info->ordered_root_lock);
4219 spin_unlock(&fs_info->ordered_root_lock);
4222 * We need this here because if we've been flipped read-only we won't
4223 * get sync() from the umount, so we need to make sure any ordered
4224 * extents that haven't had their dirty pages IO start writeout yet
4225 * actually get run and error out properly.
4227 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4230 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4231 struct btrfs_fs_info *fs_info)
4233 struct rb_node *node;
4234 struct btrfs_delayed_ref_root *delayed_refs;
4235 struct btrfs_delayed_ref_node *ref;
4238 delayed_refs = &trans->delayed_refs;
4240 spin_lock(&delayed_refs->lock);
4241 if (atomic_read(&delayed_refs->num_entries) == 0) {
4242 spin_unlock(&delayed_refs->lock);
4243 btrfs_info(fs_info, "delayed_refs has NO entry");
4247 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4248 struct btrfs_delayed_ref_head *head;
4250 bool pin_bytes = false;
4252 head = rb_entry(node, struct btrfs_delayed_ref_head,
4254 if (btrfs_delayed_ref_lock(delayed_refs, head))
4257 spin_lock(&head->lock);
4258 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4259 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4262 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4263 RB_CLEAR_NODE(&ref->ref_node);
4264 if (!list_empty(&ref->add_list))
4265 list_del(&ref->add_list);
4266 atomic_dec(&delayed_refs->num_entries);
4267 btrfs_put_delayed_ref(ref);
4269 if (head->must_insert_reserved)
4271 btrfs_free_delayed_extent_op(head->extent_op);
4272 btrfs_delete_ref_head(delayed_refs, head);
4273 spin_unlock(&head->lock);
4274 spin_unlock(&delayed_refs->lock);
4275 mutex_unlock(&head->mutex);
4278 btrfs_pin_extent(fs_info, head->bytenr,
4279 head->num_bytes, 1);
4280 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4281 btrfs_put_delayed_ref_head(head);
4283 spin_lock(&delayed_refs->lock);
4286 spin_unlock(&delayed_refs->lock);
4291 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4293 struct btrfs_inode *btrfs_inode;
4294 struct list_head splice;
4296 INIT_LIST_HEAD(&splice);
4298 spin_lock(&root->delalloc_lock);
4299 list_splice_init(&root->delalloc_inodes, &splice);
4301 while (!list_empty(&splice)) {
4302 struct inode *inode = NULL;
4303 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4305 __btrfs_del_delalloc_inode(root, btrfs_inode);
4306 spin_unlock(&root->delalloc_lock);
4309 * Make sure we get a live inode and that it'll not disappear
4312 inode = igrab(&btrfs_inode->vfs_inode);
4314 invalidate_inode_pages2(inode->i_mapping);
4317 spin_lock(&root->delalloc_lock);
4319 spin_unlock(&root->delalloc_lock);
4322 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4324 struct btrfs_root *root;
4325 struct list_head splice;
4327 INIT_LIST_HEAD(&splice);
4329 spin_lock(&fs_info->delalloc_root_lock);
4330 list_splice_init(&fs_info->delalloc_roots, &splice);
4331 while (!list_empty(&splice)) {
4332 root = list_first_entry(&splice, struct btrfs_root,
4334 root = btrfs_grab_fs_root(root);
4336 spin_unlock(&fs_info->delalloc_root_lock);
4338 btrfs_destroy_delalloc_inodes(root);
4339 btrfs_put_fs_root(root);
4341 spin_lock(&fs_info->delalloc_root_lock);
4343 spin_unlock(&fs_info->delalloc_root_lock);
4346 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4347 struct extent_io_tree *dirty_pages,
4351 struct extent_buffer *eb;
4356 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4361 clear_extent_bits(dirty_pages, start, end, mark);
4362 while (start <= end) {
4363 eb = find_extent_buffer(fs_info, start);
4364 start += fs_info->nodesize;
4367 wait_on_extent_buffer_writeback(eb);
4369 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4371 clear_extent_buffer_dirty(eb);
4372 free_extent_buffer_stale(eb);
4379 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4380 struct extent_io_tree *pinned_extents)
4382 struct extent_io_tree *unpin;
4388 unpin = pinned_extents;
4391 struct extent_state *cached_state = NULL;
4394 * The btrfs_finish_extent_commit() may get the same range as
4395 * ours between find_first_extent_bit and clear_extent_dirty.
4396 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4397 * the same extent range.
4399 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4400 ret = find_first_extent_bit(unpin, 0, &start, &end,
4401 EXTENT_DIRTY, &cached_state);
4403 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4407 clear_extent_dirty(unpin, start, end, &cached_state);
4408 free_extent_state(cached_state);
4409 btrfs_error_unpin_extent_range(fs_info, start, end);
4410 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4415 if (unpin == &fs_info->freed_extents[0])
4416 unpin = &fs_info->freed_extents[1];
4418 unpin = &fs_info->freed_extents[0];
4426 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4428 struct inode *inode;
4430 inode = cache->io_ctl.inode;
4432 invalidate_inode_pages2(inode->i_mapping);
4433 BTRFS_I(inode)->generation = 0;
4434 cache->io_ctl.inode = NULL;
4437 btrfs_put_block_group(cache);
4440 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4441 struct btrfs_fs_info *fs_info)
4443 struct btrfs_block_group_cache *cache;
4445 spin_lock(&cur_trans->dirty_bgs_lock);
4446 while (!list_empty(&cur_trans->dirty_bgs)) {
4447 cache = list_first_entry(&cur_trans->dirty_bgs,
4448 struct btrfs_block_group_cache,
4451 if (!list_empty(&cache->io_list)) {
4452 spin_unlock(&cur_trans->dirty_bgs_lock);
4453 list_del_init(&cache->io_list);
4454 btrfs_cleanup_bg_io(cache);
4455 spin_lock(&cur_trans->dirty_bgs_lock);
4458 list_del_init(&cache->dirty_list);
4459 spin_lock(&cache->lock);
4460 cache->disk_cache_state = BTRFS_DC_ERROR;
4461 spin_unlock(&cache->lock);
4463 spin_unlock(&cur_trans->dirty_bgs_lock);
4464 btrfs_put_block_group(cache);
4465 btrfs_delayed_refs_rsv_release(fs_info, 1);
4466 spin_lock(&cur_trans->dirty_bgs_lock);
4468 spin_unlock(&cur_trans->dirty_bgs_lock);
4471 * Refer to the definition of io_bgs member for details why it's safe
4472 * to use it without any locking
4474 while (!list_empty(&cur_trans->io_bgs)) {
4475 cache = list_first_entry(&cur_trans->io_bgs,
4476 struct btrfs_block_group_cache,
4479 list_del_init(&cache->io_list);
4480 spin_lock(&cache->lock);
4481 cache->disk_cache_state = BTRFS_DC_ERROR;
4482 spin_unlock(&cache->lock);
4483 btrfs_cleanup_bg_io(cache);
4487 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4488 struct btrfs_fs_info *fs_info)
4490 struct btrfs_device *dev, *tmp;
4492 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4493 ASSERT(list_empty(&cur_trans->dirty_bgs));
4494 ASSERT(list_empty(&cur_trans->io_bgs));
4496 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4498 list_del_init(&dev->post_commit_list);
4501 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4503 cur_trans->state = TRANS_STATE_COMMIT_START;
4504 wake_up(&fs_info->transaction_blocked_wait);
4506 cur_trans->state = TRANS_STATE_UNBLOCKED;
4507 wake_up(&fs_info->transaction_wait);
4509 btrfs_destroy_delayed_inodes(fs_info);
4510 btrfs_assert_delayed_root_empty(fs_info);
4512 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4514 btrfs_destroy_pinned_extent(fs_info,
4515 fs_info->pinned_extents);
4517 cur_trans->state =TRANS_STATE_COMPLETED;
4518 wake_up(&cur_trans->commit_wait);
4521 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4523 struct btrfs_transaction *t;
4525 mutex_lock(&fs_info->transaction_kthread_mutex);
4527 spin_lock(&fs_info->trans_lock);
4528 while (!list_empty(&fs_info->trans_list)) {
4529 t = list_first_entry(&fs_info->trans_list,
4530 struct btrfs_transaction, list);
4531 if (t->state >= TRANS_STATE_COMMIT_START) {
4532 refcount_inc(&t->use_count);
4533 spin_unlock(&fs_info->trans_lock);
4534 btrfs_wait_for_commit(fs_info, t->transid);
4535 btrfs_put_transaction(t);
4536 spin_lock(&fs_info->trans_lock);
4539 if (t == fs_info->running_transaction) {
4540 t->state = TRANS_STATE_COMMIT_DOING;
4541 spin_unlock(&fs_info->trans_lock);
4543 * We wait for 0 num_writers since we don't hold a trans
4544 * handle open currently for this transaction.
4546 wait_event(t->writer_wait,
4547 atomic_read(&t->num_writers) == 0);
4549 spin_unlock(&fs_info->trans_lock);
4551 btrfs_cleanup_one_transaction(t, fs_info);
4553 spin_lock(&fs_info->trans_lock);
4554 if (t == fs_info->running_transaction)
4555 fs_info->running_transaction = NULL;
4556 list_del_init(&t->list);
4557 spin_unlock(&fs_info->trans_lock);
4559 btrfs_put_transaction(t);
4560 trace_btrfs_transaction_commit(fs_info->tree_root);
4561 spin_lock(&fs_info->trans_lock);
4563 spin_unlock(&fs_info->trans_lock);
4564 btrfs_destroy_all_ordered_extents(fs_info);
4565 btrfs_destroy_delayed_inodes(fs_info);
4566 btrfs_assert_delayed_root_empty(fs_info);
4567 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4568 btrfs_destroy_all_delalloc_inodes(fs_info);
4569 mutex_unlock(&fs_info->transaction_kthread_mutex);
4574 static const struct extent_io_ops btree_extent_io_ops = {
4575 /* mandatory callbacks */
4576 .submit_bio_hook = btree_submit_bio_hook,
4577 .readpage_end_io_hook = btree_readpage_end_io_hook,