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:
355 case BTRFS_CSUM_TYPE_XXHASH:
363 * Return 0 if the superblock checksum type matches the checksum value of that
364 * algorithm. Pass the raw disk superblock data.
366 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
369 struct btrfs_super_block *disk_sb =
370 (struct btrfs_super_block *)raw_disk_sb;
371 char result[BTRFS_CSUM_SIZE];
372 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
374 shash->tfm = fs_info->csum_shash;
375 crypto_shash_init(shash);
378 * The super_block structure does not span the whole
379 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
380 * filled with zeros and is included in the checksum.
382 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
383 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
384 crypto_shash_final(shash, result);
386 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
392 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
393 struct btrfs_key *first_key, u64 parent_transid)
395 struct btrfs_fs_info *fs_info = eb->fs_info;
397 struct btrfs_key found_key;
400 found_level = btrfs_header_level(eb);
401 if (found_level != level) {
402 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
403 KERN_ERR "BTRFS: tree level check failed\n");
405 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
406 eb->start, level, found_level);
414 * For live tree block (new tree blocks in current transaction),
415 * we need proper lock context to avoid race, which is impossible here.
416 * So we only checks tree blocks which is read from disk, whose
417 * generation <= fs_info->last_trans_committed.
419 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
422 /* We have @first_key, so this @eb must have at least one item */
423 if (btrfs_header_nritems(eb) == 0) {
425 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
427 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
432 btrfs_node_key_to_cpu(eb, &found_key, 0);
434 btrfs_item_key_to_cpu(eb, &found_key, 0);
435 ret = btrfs_comp_cpu_keys(first_key, &found_key);
438 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
439 KERN_ERR "BTRFS: tree first key check failed\n");
441 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
442 eb->start, parent_transid, first_key->objectid,
443 first_key->type, first_key->offset,
444 found_key.objectid, found_key.type,
451 * helper to read a given tree block, doing retries as required when
452 * the checksums don't match and we have alternate mirrors to try.
454 * @parent_transid: expected transid, skip check if 0
455 * @level: expected level, mandatory check
456 * @first_key: expected key of first slot, skip check if NULL
458 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
459 u64 parent_transid, int level,
460 struct btrfs_key *first_key)
462 struct btrfs_fs_info *fs_info = eb->fs_info;
463 struct extent_io_tree *io_tree;
468 int failed_mirror = 0;
470 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
472 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
473 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
475 if (verify_parent_transid(io_tree, eb,
478 else if (btrfs_verify_level_key(eb, level,
479 first_key, parent_transid))
485 num_copies = btrfs_num_copies(fs_info,
490 if (!failed_mirror) {
492 failed_mirror = eb->read_mirror;
496 if (mirror_num == failed_mirror)
499 if (mirror_num > num_copies)
503 if (failed && !ret && failed_mirror)
504 btrfs_repair_eb_io_failure(eb, failed_mirror);
510 * checksum a dirty tree block before IO. This has extra checks to make sure
511 * we only fill in the checksum field in the first page of a multi-page block
514 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
516 u64 start = page_offset(page);
518 u8 result[BTRFS_CSUM_SIZE];
519 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
520 struct extent_buffer *eb;
523 eb = (struct extent_buffer *)page->private;
524 if (page != eb->pages[0])
527 found_start = btrfs_header_bytenr(eb);
529 * Please do not consolidate these warnings into a single if.
530 * It is useful to know what went wrong.
532 if (WARN_ON(found_start != start))
534 if (WARN_ON(!PageUptodate(page)))
537 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
538 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
540 if (csum_tree_block(eb, result))
543 if (btrfs_header_level(eb))
544 ret = btrfs_check_node(eb);
546 ret = btrfs_check_leaf_full(eb);
549 btrfs_print_tree(eb, 0);
551 "block=%llu write time tree block corruption detected",
553 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
556 write_extent_buffer(eb, result, 0, csum_size);
561 static int check_tree_block_fsid(struct extent_buffer *eb)
563 struct btrfs_fs_info *fs_info = eb->fs_info;
564 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
565 u8 fsid[BTRFS_FSID_SIZE];
568 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
573 * Checking the incompat flag is only valid for the current
574 * fs. For seed devices it's forbidden to have their uuid
575 * changed so reading ->fsid in this case is fine
577 if (fs_devices == fs_info->fs_devices &&
578 btrfs_fs_incompat(fs_info, METADATA_UUID))
579 metadata_uuid = fs_devices->metadata_uuid;
581 metadata_uuid = fs_devices->fsid;
583 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
587 fs_devices = fs_devices->seed;
592 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
593 u64 phy_offset, struct page *page,
594 u64 start, u64 end, int mirror)
598 struct extent_buffer *eb;
599 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
600 struct btrfs_fs_info *fs_info = root->fs_info;
601 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
603 u8 result[BTRFS_CSUM_SIZE];
609 eb = (struct extent_buffer *)page->private;
611 /* the pending IO might have been the only thing that kept this buffer
612 * in memory. Make sure we have a ref for all this other checks
614 atomic_inc(&eb->refs);
616 reads_done = atomic_dec_and_test(&eb->io_pages);
620 eb->read_mirror = mirror;
621 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
626 found_start = btrfs_header_bytenr(eb);
627 if (found_start != eb->start) {
628 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
629 eb->start, found_start);
633 if (check_tree_block_fsid(eb)) {
634 btrfs_err_rl(fs_info, "bad fsid on block %llu",
639 found_level = btrfs_header_level(eb);
640 if (found_level >= BTRFS_MAX_LEVEL) {
641 btrfs_err(fs_info, "bad tree block level %d on %llu",
642 (int)btrfs_header_level(eb), eb->start);
647 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
650 ret = csum_tree_block(eb, result);
654 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
658 memcpy(&found, result, csum_size);
660 read_extent_buffer(eb, &val, 0, csum_size);
661 btrfs_warn_rl(fs_info,
662 "%s checksum verify failed on %llu wanted %x found %x level %d",
663 fs_info->sb->s_id, eb->start,
664 val, found, btrfs_header_level(eb));
670 * If this is a leaf block and it is corrupt, set the corrupt bit so
671 * that we don't try and read the other copies of this block, just
674 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
675 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
679 if (found_level > 0 && btrfs_check_node(eb))
683 set_extent_buffer_uptodate(eb);
686 "block=%llu read time tree block corruption detected",
690 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
691 btree_readahead_hook(eb, ret);
695 * our io error hook is going to dec the io pages
696 * again, we have to make sure it has something
699 atomic_inc(&eb->io_pages);
700 clear_extent_buffer_uptodate(eb);
702 free_extent_buffer(eb);
707 static void end_workqueue_bio(struct bio *bio)
709 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
710 struct btrfs_fs_info *fs_info;
711 struct btrfs_workqueue *wq;
713 fs_info = end_io_wq->info;
714 end_io_wq->status = bio->bi_status;
716 if (bio_op(bio) == REQ_OP_WRITE) {
717 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
718 wq = fs_info->endio_meta_write_workers;
719 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
720 wq = fs_info->endio_freespace_worker;
721 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
722 wq = fs_info->endio_raid56_workers;
724 wq = fs_info->endio_write_workers;
726 if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
727 wq = fs_info->endio_repair_workers;
728 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
729 wq = fs_info->endio_raid56_workers;
730 else if (end_io_wq->metadata)
731 wq = fs_info->endio_meta_workers;
733 wq = fs_info->endio_workers;
736 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
737 btrfs_queue_work(wq, &end_io_wq->work);
740 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
741 enum btrfs_wq_endio_type metadata)
743 struct btrfs_end_io_wq *end_io_wq;
745 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
747 return BLK_STS_RESOURCE;
749 end_io_wq->private = bio->bi_private;
750 end_io_wq->end_io = bio->bi_end_io;
751 end_io_wq->info = info;
752 end_io_wq->status = 0;
753 end_io_wq->bio = bio;
754 end_io_wq->metadata = metadata;
756 bio->bi_private = end_io_wq;
757 bio->bi_end_io = end_workqueue_bio;
761 static void run_one_async_start(struct btrfs_work *work)
763 struct async_submit_bio *async;
766 async = container_of(work, struct async_submit_bio, work);
767 ret = async->submit_bio_start(async->private_data, async->bio,
774 * In order to insert checksums into the metadata in large chunks, we wait
775 * until bio submission time. All the pages in the bio are checksummed and
776 * sums are attached onto the ordered extent record.
778 * At IO completion time the csums attached on the ordered extent record are
779 * inserted into the tree.
781 static void run_one_async_done(struct btrfs_work *work)
783 struct async_submit_bio *async;
787 async = container_of(work, struct async_submit_bio, work);
788 inode = async->private_data;
790 /* If an error occurred we just want to clean up the bio and move on */
792 async->bio->bi_status = async->status;
793 bio_endio(async->bio);
798 * All of the bios that pass through here are from async helpers.
799 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
800 * This changes nothing when cgroups aren't in use.
802 async->bio->bi_opf |= REQ_CGROUP_PUNT;
803 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
805 async->bio->bi_status = ret;
806 bio_endio(async->bio);
810 static void run_one_async_free(struct btrfs_work *work)
812 struct async_submit_bio *async;
814 async = container_of(work, struct async_submit_bio, work);
818 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
819 int mirror_num, unsigned long bio_flags,
820 u64 bio_offset, void *private_data,
821 extent_submit_bio_start_t *submit_bio_start)
823 struct async_submit_bio *async;
825 async = kmalloc(sizeof(*async), GFP_NOFS);
827 return BLK_STS_RESOURCE;
829 async->private_data = private_data;
831 async->mirror_num = mirror_num;
832 async->submit_bio_start = submit_bio_start;
834 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
837 async->bio_offset = bio_offset;
841 if (op_is_sync(bio->bi_opf))
842 btrfs_set_work_high_priority(&async->work);
844 btrfs_queue_work(fs_info->workers, &async->work);
848 static blk_status_t btree_csum_one_bio(struct bio *bio)
850 struct bio_vec *bvec;
851 struct btrfs_root *root;
853 struct bvec_iter_all iter_all;
855 ASSERT(!bio_flagged(bio, BIO_CLONED));
856 bio_for_each_segment_all(bvec, bio, iter_all) {
857 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
858 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
863 return errno_to_blk_status(ret);
866 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
870 * when we're called for a write, we're already in the async
871 * submission context. Just jump into btrfs_map_bio
873 return btree_csum_one_bio(bio);
876 static int check_async_write(struct btrfs_fs_info *fs_info,
877 struct btrfs_inode *bi)
879 if (atomic_read(&bi->sync_writers))
881 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
886 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
888 unsigned long bio_flags)
890 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
891 int async = check_async_write(fs_info, BTRFS_I(inode));
894 if (bio_op(bio) != REQ_OP_WRITE) {
896 * called for a read, do the setup so that checksum validation
897 * can happen in the async kernel threads
899 ret = btrfs_bio_wq_end_io(fs_info, bio,
900 BTRFS_WQ_ENDIO_METADATA);
903 ret = btrfs_map_bio(fs_info, bio, mirror_num);
905 ret = btree_csum_one_bio(bio);
908 ret = btrfs_map_bio(fs_info, bio, mirror_num);
911 * kthread helpers are used to submit writes so that
912 * checksumming can happen in parallel across all CPUs
914 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
915 0, inode, btree_submit_bio_start);
923 bio->bi_status = ret;
928 #ifdef CONFIG_MIGRATION
929 static int btree_migratepage(struct address_space *mapping,
930 struct page *newpage, struct page *page,
931 enum migrate_mode mode)
934 * we can't safely write a btree page from here,
935 * we haven't done the locking hook
940 * Buffers may be managed in a filesystem specific way.
941 * We must have no buffers or drop them.
943 if (page_has_private(page) &&
944 !try_to_release_page(page, GFP_KERNEL))
946 return migrate_page(mapping, newpage, page, mode);
951 static int btree_writepages(struct address_space *mapping,
952 struct writeback_control *wbc)
954 struct btrfs_fs_info *fs_info;
957 if (wbc->sync_mode == WB_SYNC_NONE) {
959 if (wbc->for_kupdate)
962 fs_info = BTRFS_I(mapping->host)->root->fs_info;
963 /* this is a bit racy, but that's ok */
964 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
965 BTRFS_DIRTY_METADATA_THRESH,
966 fs_info->dirty_metadata_batch);
970 return btree_write_cache_pages(mapping, wbc);
973 static int btree_readpage(struct file *file, struct page *page)
975 struct extent_io_tree *tree;
976 tree = &BTRFS_I(page->mapping->host)->io_tree;
977 return extent_read_full_page(tree, page, btree_get_extent, 0);
980 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
982 if (PageWriteback(page) || PageDirty(page))
985 return try_release_extent_buffer(page);
988 static void btree_invalidatepage(struct page *page, unsigned int offset,
991 struct extent_io_tree *tree;
992 tree = &BTRFS_I(page->mapping->host)->io_tree;
993 extent_invalidatepage(tree, page, offset);
994 btree_releasepage(page, GFP_NOFS);
995 if (PagePrivate(page)) {
996 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
997 "page private not zero on page %llu",
998 (unsigned long long)page_offset(page));
999 ClearPagePrivate(page);
1000 set_page_private(page, 0);
1005 static int btree_set_page_dirty(struct page *page)
1008 struct extent_buffer *eb;
1010 BUG_ON(!PagePrivate(page));
1011 eb = (struct extent_buffer *)page->private;
1013 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1014 BUG_ON(!atomic_read(&eb->refs));
1015 btrfs_assert_tree_locked(eb);
1017 return __set_page_dirty_nobuffers(page);
1020 static const struct address_space_operations btree_aops = {
1021 .readpage = btree_readpage,
1022 .writepages = btree_writepages,
1023 .releasepage = btree_releasepage,
1024 .invalidatepage = btree_invalidatepage,
1025 #ifdef CONFIG_MIGRATION
1026 .migratepage = btree_migratepage,
1028 .set_page_dirty = btree_set_page_dirty,
1031 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1033 struct extent_buffer *buf = NULL;
1036 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1040 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1042 free_extent_buffer_stale(buf);
1044 free_extent_buffer(buf);
1047 struct extent_buffer *btrfs_find_create_tree_block(
1048 struct btrfs_fs_info *fs_info,
1051 if (btrfs_is_testing(fs_info))
1052 return alloc_test_extent_buffer(fs_info, bytenr);
1053 return alloc_extent_buffer(fs_info, bytenr);
1057 * Read tree block at logical address @bytenr and do variant basic but critical
1060 * @parent_transid: expected transid of this tree block, skip check if 0
1061 * @level: expected level, mandatory check
1062 * @first_key: expected key in slot 0, skip check if NULL
1064 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1065 u64 parent_transid, int level,
1066 struct btrfs_key *first_key)
1068 struct extent_buffer *buf = NULL;
1071 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1075 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1078 free_extent_buffer_stale(buf);
1079 return ERR_PTR(ret);
1085 void btrfs_clean_tree_block(struct extent_buffer *buf)
1087 struct btrfs_fs_info *fs_info = buf->fs_info;
1088 if (btrfs_header_generation(buf) ==
1089 fs_info->running_transaction->transid) {
1090 btrfs_assert_tree_locked(buf);
1092 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1093 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1095 fs_info->dirty_metadata_batch);
1096 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1097 btrfs_set_lock_blocking_write(buf);
1098 clear_extent_buffer_dirty(buf);
1103 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1105 struct btrfs_subvolume_writers *writers;
1108 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1110 return ERR_PTR(-ENOMEM);
1112 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1115 return ERR_PTR(ret);
1118 init_waitqueue_head(&writers->wait);
1123 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1125 percpu_counter_destroy(&writers->counter);
1129 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1132 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1134 root->commit_root = NULL;
1136 root->orphan_cleanup_state = 0;
1138 root->last_trans = 0;
1139 root->highest_objectid = 0;
1140 root->nr_delalloc_inodes = 0;
1141 root->nr_ordered_extents = 0;
1142 root->inode_tree = RB_ROOT;
1143 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1144 root->block_rsv = NULL;
1146 INIT_LIST_HEAD(&root->dirty_list);
1147 INIT_LIST_HEAD(&root->root_list);
1148 INIT_LIST_HEAD(&root->delalloc_inodes);
1149 INIT_LIST_HEAD(&root->delalloc_root);
1150 INIT_LIST_HEAD(&root->ordered_extents);
1151 INIT_LIST_HEAD(&root->ordered_root);
1152 INIT_LIST_HEAD(&root->reloc_dirty_list);
1153 INIT_LIST_HEAD(&root->logged_list[0]);
1154 INIT_LIST_HEAD(&root->logged_list[1]);
1155 spin_lock_init(&root->inode_lock);
1156 spin_lock_init(&root->delalloc_lock);
1157 spin_lock_init(&root->ordered_extent_lock);
1158 spin_lock_init(&root->accounting_lock);
1159 spin_lock_init(&root->log_extents_lock[0]);
1160 spin_lock_init(&root->log_extents_lock[1]);
1161 spin_lock_init(&root->qgroup_meta_rsv_lock);
1162 mutex_init(&root->objectid_mutex);
1163 mutex_init(&root->log_mutex);
1164 mutex_init(&root->ordered_extent_mutex);
1165 mutex_init(&root->delalloc_mutex);
1166 init_waitqueue_head(&root->log_writer_wait);
1167 init_waitqueue_head(&root->log_commit_wait[0]);
1168 init_waitqueue_head(&root->log_commit_wait[1]);
1169 INIT_LIST_HEAD(&root->log_ctxs[0]);
1170 INIT_LIST_HEAD(&root->log_ctxs[1]);
1171 atomic_set(&root->log_commit[0], 0);
1172 atomic_set(&root->log_commit[1], 0);
1173 atomic_set(&root->log_writers, 0);
1174 atomic_set(&root->log_batch, 0);
1175 refcount_set(&root->refs, 1);
1176 atomic_set(&root->will_be_snapshotted, 0);
1177 atomic_set(&root->snapshot_force_cow, 0);
1178 atomic_set(&root->nr_swapfiles, 0);
1179 root->log_transid = 0;
1180 root->log_transid_committed = -1;
1181 root->last_log_commit = 0;
1183 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1184 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1186 memset(&root->root_key, 0, sizeof(root->root_key));
1187 memset(&root->root_item, 0, sizeof(root->root_item));
1188 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1190 root->defrag_trans_start = fs_info->generation;
1192 root->defrag_trans_start = 0;
1193 root->root_key.objectid = objectid;
1196 spin_lock_init(&root->root_item_lock);
1197 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1200 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1203 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1205 root->fs_info = fs_info;
1209 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1210 /* Should only be used by the testing infrastructure */
1211 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1213 struct btrfs_root *root;
1216 return ERR_PTR(-EINVAL);
1218 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1220 return ERR_PTR(-ENOMEM);
1222 /* We don't use the stripesize in selftest, set it as sectorsize */
1223 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1224 root->alloc_bytenr = 0;
1230 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1233 struct btrfs_fs_info *fs_info = trans->fs_info;
1234 struct extent_buffer *leaf;
1235 struct btrfs_root *tree_root = fs_info->tree_root;
1236 struct btrfs_root *root;
1237 struct btrfs_key key;
1238 unsigned int nofs_flag;
1240 uuid_le uuid = NULL_UUID_LE;
1243 * We're holding a transaction handle, so use a NOFS memory allocation
1244 * context to avoid deadlock if reclaim happens.
1246 nofs_flag = memalloc_nofs_save();
1247 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1248 memalloc_nofs_restore(nofs_flag);
1250 return ERR_PTR(-ENOMEM);
1252 __setup_root(root, fs_info, objectid);
1253 root->root_key.objectid = objectid;
1254 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1255 root->root_key.offset = 0;
1257 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1259 ret = PTR_ERR(leaf);
1265 btrfs_mark_buffer_dirty(leaf);
1267 root->commit_root = btrfs_root_node(root);
1268 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1270 root->root_item.flags = 0;
1271 root->root_item.byte_limit = 0;
1272 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1273 btrfs_set_root_generation(&root->root_item, trans->transid);
1274 btrfs_set_root_level(&root->root_item, 0);
1275 btrfs_set_root_refs(&root->root_item, 1);
1276 btrfs_set_root_used(&root->root_item, leaf->len);
1277 btrfs_set_root_last_snapshot(&root->root_item, 0);
1278 btrfs_set_root_dirid(&root->root_item, 0);
1279 if (is_fstree(objectid))
1281 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1282 root->root_item.drop_level = 0;
1284 key.objectid = objectid;
1285 key.type = BTRFS_ROOT_ITEM_KEY;
1287 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1291 btrfs_tree_unlock(leaf);
1297 btrfs_tree_unlock(leaf);
1298 free_extent_buffer(root->commit_root);
1299 free_extent_buffer(leaf);
1303 return ERR_PTR(ret);
1306 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1307 struct btrfs_fs_info *fs_info)
1309 struct btrfs_root *root;
1310 struct extent_buffer *leaf;
1312 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1314 return ERR_PTR(-ENOMEM);
1316 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1318 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1319 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1320 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1323 * DON'T set REF_COWS for log trees
1325 * log trees do not get reference counted because they go away
1326 * before a real commit is actually done. They do store pointers
1327 * to file data extents, and those reference counts still get
1328 * updated (along with back refs to the log tree).
1331 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1335 return ERR_CAST(leaf);
1340 btrfs_mark_buffer_dirty(root->node);
1341 btrfs_tree_unlock(root->node);
1345 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1346 struct btrfs_fs_info *fs_info)
1348 struct btrfs_root *log_root;
1350 log_root = alloc_log_tree(trans, fs_info);
1351 if (IS_ERR(log_root))
1352 return PTR_ERR(log_root);
1353 WARN_ON(fs_info->log_root_tree);
1354 fs_info->log_root_tree = log_root;
1358 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1359 struct btrfs_root *root)
1361 struct btrfs_fs_info *fs_info = root->fs_info;
1362 struct btrfs_root *log_root;
1363 struct btrfs_inode_item *inode_item;
1365 log_root = alloc_log_tree(trans, fs_info);
1366 if (IS_ERR(log_root))
1367 return PTR_ERR(log_root);
1369 log_root->last_trans = trans->transid;
1370 log_root->root_key.offset = root->root_key.objectid;
1372 inode_item = &log_root->root_item.inode;
1373 btrfs_set_stack_inode_generation(inode_item, 1);
1374 btrfs_set_stack_inode_size(inode_item, 3);
1375 btrfs_set_stack_inode_nlink(inode_item, 1);
1376 btrfs_set_stack_inode_nbytes(inode_item,
1378 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1380 btrfs_set_root_node(&log_root->root_item, log_root->node);
1382 WARN_ON(root->log_root);
1383 root->log_root = log_root;
1384 root->log_transid = 0;
1385 root->log_transid_committed = -1;
1386 root->last_log_commit = 0;
1390 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1391 struct btrfs_key *key)
1393 struct btrfs_root *root;
1394 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1395 struct btrfs_path *path;
1400 path = btrfs_alloc_path();
1402 return ERR_PTR(-ENOMEM);
1404 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1410 __setup_root(root, fs_info, key->objectid);
1412 ret = btrfs_find_root(tree_root, key, path,
1413 &root->root_item, &root->root_key);
1420 generation = btrfs_root_generation(&root->root_item);
1421 level = btrfs_root_level(&root->root_item);
1422 root->node = read_tree_block(fs_info,
1423 btrfs_root_bytenr(&root->root_item),
1424 generation, level, NULL);
1425 if (IS_ERR(root->node)) {
1426 ret = PTR_ERR(root->node);
1428 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1430 free_extent_buffer(root->node);
1433 root->commit_root = btrfs_root_node(root);
1435 btrfs_free_path(path);
1441 root = ERR_PTR(ret);
1445 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1446 struct btrfs_key *location)
1448 struct btrfs_root *root;
1450 root = btrfs_read_tree_root(tree_root, location);
1454 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1455 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1456 btrfs_check_and_init_root_item(&root->root_item);
1462 int btrfs_init_fs_root(struct btrfs_root *root)
1465 struct btrfs_subvolume_writers *writers;
1467 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1468 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1470 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1475 writers = btrfs_alloc_subvolume_writers();
1476 if (IS_ERR(writers)) {
1477 ret = PTR_ERR(writers);
1480 root->subv_writers = writers;
1482 btrfs_init_free_ino_ctl(root);
1483 spin_lock_init(&root->ino_cache_lock);
1484 init_waitqueue_head(&root->ino_cache_wait);
1486 ret = get_anon_bdev(&root->anon_dev);
1490 mutex_lock(&root->objectid_mutex);
1491 ret = btrfs_find_highest_objectid(root,
1492 &root->highest_objectid);
1494 mutex_unlock(&root->objectid_mutex);
1498 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1500 mutex_unlock(&root->objectid_mutex);
1504 /* The caller is responsible to call btrfs_free_fs_root */
1508 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1511 struct btrfs_root *root;
1513 spin_lock(&fs_info->fs_roots_radix_lock);
1514 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1515 (unsigned long)root_id);
1516 spin_unlock(&fs_info->fs_roots_radix_lock);
1520 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1521 struct btrfs_root *root)
1525 ret = radix_tree_preload(GFP_NOFS);
1529 spin_lock(&fs_info->fs_roots_radix_lock);
1530 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1531 (unsigned long)root->root_key.objectid,
1534 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1535 spin_unlock(&fs_info->fs_roots_radix_lock);
1536 radix_tree_preload_end();
1541 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1542 struct btrfs_key *location,
1545 struct btrfs_root *root;
1546 struct btrfs_path *path;
1547 struct btrfs_key key;
1550 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1551 return fs_info->tree_root;
1552 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1553 return fs_info->extent_root;
1554 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1555 return fs_info->chunk_root;
1556 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1557 return fs_info->dev_root;
1558 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1559 return fs_info->csum_root;
1560 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1561 return fs_info->quota_root ? fs_info->quota_root :
1563 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1564 return fs_info->uuid_root ? fs_info->uuid_root :
1566 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1567 return fs_info->free_space_root ? fs_info->free_space_root :
1570 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1572 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1573 return ERR_PTR(-ENOENT);
1577 root = btrfs_read_fs_root(fs_info->tree_root, location);
1581 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1586 ret = btrfs_init_fs_root(root);
1590 path = btrfs_alloc_path();
1595 key.objectid = BTRFS_ORPHAN_OBJECTID;
1596 key.type = BTRFS_ORPHAN_ITEM_KEY;
1597 key.offset = location->objectid;
1599 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1600 btrfs_free_path(path);
1604 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1606 ret = btrfs_insert_fs_root(fs_info, root);
1608 if (ret == -EEXIST) {
1609 btrfs_free_fs_root(root);
1616 btrfs_free_fs_root(root);
1617 return ERR_PTR(ret);
1620 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1622 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1624 struct btrfs_device *device;
1625 struct backing_dev_info *bdi;
1628 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1631 bdi = device->bdev->bd_bdi;
1632 if (bdi_congested(bdi, bdi_bits)) {
1642 * called by the kthread helper functions to finally call the bio end_io
1643 * functions. This is where read checksum verification actually happens
1645 static void end_workqueue_fn(struct btrfs_work *work)
1648 struct btrfs_end_io_wq *end_io_wq;
1650 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1651 bio = end_io_wq->bio;
1653 bio->bi_status = end_io_wq->status;
1654 bio->bi_private = end_io_wq->private;
1655 bio->bi_end_io = end_io_wq->end_io;
1657 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1660 static int cleaner_kthread(void *arg)
1662 struct btrfs_root *root = arg;
1663 struct btrfs_fs_info *fs_info = root->fs_info;
1669 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1671 /* Make the cleaner go to sleep early. */
1672 if (btrfs_need_cleaner_sleep(fs_info))
1676 * Do not do anything if we might cause open_ctree() to block
1677 * before we have finished mounting the filesystem.
1679 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1682 if (!mutex_trylock(&fs_info->cleaner_mutex))
1686 * Avoid the problem that we change the status of the fs
1687 * during the above check and trylock.
1689 if (btrfs_need_cleaner_sleep(fs_info)) {
1690 mutex_unlock(&fs_info->cleaner_mutex);
1694 btrfs_run_delayed_iputs(fs_info);
1696 again = btrfs_clean_one_deleted_snapshot(root);
1697 mutex_unlock(&fs_info->cleaner_mutex);
1700 * The defragger has dealt with the R/O remount and umount,
1701 * needn't do anything special here.
1703 btrfs_run_defrag_inodes(fs_info);
1706 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1707 * with relocation (btrfs_relocate_chunk) and relocation
1708 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1709 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1710 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1711 * unused block groups.
1713 btrfs_delete_unused_bgs(fs_info);
1715 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1716 if (kthread_should_park())
1718 if (kthread_should_stop())
1721 set_current_state(TASK_INTERRUPTIBLE);
1723 __set_current_state(TASK_RUNNING);
1728 static int transaction_kthread(void *arg)
1730 struct btrfs_root *root = arg;
1731 struct btrfs_fs_info *fs_info = root->fs_info;
1732 struct btrfs_trans_handle *trans;
1733 struct btrfs_transaction *cur;
1736 unsigned long delay;
1740 cannot_commit = false;
1741 delay = HZ * fs_info->commit_interval;
1742 mutex_lock(&fs_info->transaction_kthread_mutex);
1744 spin_lock(&fs_info->trans_lock);
1745 cur = fs_info->running_transaction;
1747 spin_unlock(&fs_info->trans_lock);
1751 now = ktime_get_seconds();
1752 if (cur->state < TRANS_STATE_COMMIT_START &&
1753 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1754 (now < cur->start_time ||
1755 now - cur->start_time < fs_info->commit_interval)) {
1756 spin_unlock(&fs_info->trans_lock);
1760 transid = cur->transid;
1761 spin_unlock(&fs_info->trans_lock);
1763 /* If the file system is aborted, this will always fail. */
1764 trans = btrfs_attach_transaction(root);
1765 if (IS_ERR(trans)) {
1766 if (PTR_ERR(trans) != -ENOENT)
1767 cannot_commit = true;
1770 if (transid == trans->transid) {
1771 btrfs_commit_transaction(trans);
1773 btrfs_end_transaction(trans);
1776 wake_up_process(fs_info->cleaner_kthread);
1777 mutex_unlock(&fs_info->transaction_kthread_mutex);
1779 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1780 &fs_info->fs_state)))
1781 btrfs_cleanup_transaction(fs_info);
1782 if (!kthread_should_stop() &&
1783 (!btrfs_transaction_blocked(fs_info) ||
1785 schedule_timeout_interruptible(delay);
1786 } while (!kthread_should_stop());
1791 * this will find the highest generation in the array of
1792 * root backups. The index of the highest array is returned,
1793 * or -1 if we can't find anything.
1795 * We check to make sure the array is valid by comparing the
1796 * generation of the latest root in the array with the generation
1797 * in the super block. If they don't match we pitch it.
1799 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1802 int newest_index = -1;
1803 struct btrfs_root_backup *root_backup;
1806 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1807 root_backup = info->super_copy->super_roots + i;
1808 cur = btrfs_backup_tree_root_gen(root_backup);
1809 if (cur == newest_gen)
1813 /* check to see if we actually wrapped around */
1814 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1815 root_backup = info->super_copy->super_roots;
1816 cur = btrfs_backup_tree_root_gen(root_backup);
1817 if (cur == newest_gen)
1820 return newest_index;
1825 * find the oldest backup so we know where to store new entries
1826 * in the backup array. This will set the backup_root_index
1827 * field in the fs_info struct
1829 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1832 int newest_index = -1;
1834 newest_index = find_newest_super_backup(info, newest_gen);
1835 /* if there was garbage in there, just move along */
1836 if (newest_index == -1) {
1837 info->backup_root_index = 0;
1839 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1844 * copy all the root pointers into the super backup array.
1845 * this will bump the backup pointer by one when it is
1848 static void backup_super_roots(struct btrfs_fs_info *info)
1851 struct btrfs_root_backup *root_backup;
1854 next_backup = info->backup_root_index;
1855 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1856 BTRFS_NUM_BACKUP_ROOTS;
1859 * just overwrite the last backup if we're at the same generation
1860 * this happens only at umount
1862 root_backup = info->super_for_commit->super_roots + last_backup;
1863 if (btrfs_backup_tree_root_gen(root_backup) ==
1864 btrfs_header_generation(info->tree_root->node))
1865 next_backup = last_backup;
1867 root_backup = info->super_for_commit->super_roots + next_backup;
1870 * make sure all of our padding and empty slots get zero filled
1871 * regardless of which ones we use today
1873 memset(root_backup, 0, sizeof(*root_backup));
1875 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1877 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1878 btrfs_set_backup_tree_root_gen(root_backup,
1879 btrfs_header_generation(info->tree_root->node));
1881 btrfs_set_backup_tree_root_level(root_backup,
1882 btrfs_header_level(info->tree_root->node));
1884 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1885 btrfs_set_backup_chunk_root_gen(root_backup,
1886 btrfs_header_generation(info->chunk_root->node));
1887 btrfs_set_backup_chunk_root_level(root_backup,
1888 btrfs_header_level(info->chunk_root->node));
1890 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1891 btrfs_set_backup_extent_root_gen(root_backup,
1892 btrfs_header_generation(info->extent_root->node));
1893 btrfs_set_backup_extent_root_level(root_backup,
1894 btrfs_header_level(info->extent_root->node));
1897 * we might commit during log recovery, which happens before we set
1898 * the fs_root. Make sure it is valid before we fill it in.
1900 if (info->fs_root && info->fs_root->node) {
1901 btrfs_set_backup_fs_root(root_backup,
1902 info->fs_root->node->start);
1903 btrfs_set_backup_fs_root_gen(root_backup,
1904 btrfs_header_generation(info->fs_root->node));
1905 btrfs_set_backup_fs_root_level(root_backup,
1906 btrfs_header_level(info->fs_root->node));
1909 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1910 btrfs_set_backup_dev_root_gen(root_backup,
1911 btrfs_header_generation(info->dev_root->node));
1912 btrfs_set_backup_dev_root_level(root_backup,
1913 btrfs_header_level(info->dev_root->node));
1915 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1916 btrfs_set_backup_csum_root_gen(root_backup,
1917 btrfs_header_generation(info->csum_root->node));
1918 btrfs_set_backup_csum_root_level(root_backup,
1919 btrfs_header_level(info->csum_root->node));
1921 btrfs_set_backup_total_bytes(root_backup,
1922 btrfs_super_total_bytes(info->super_copy));
1923 btrfs_set_backup_bytes_used(root_backup,
1924 btrfs_super_bytes_used(info->super_copy));
1925 btrfs_set_backup_num_devices(root_backup,
1926 btrfs_super_num_devices(info->super_copy));
1929 * if we don't copy this out to the super_copy, it won't get remembered
1930 * for the next commit
1932 memcpy(&info->super_copy->super_roots,
1933 &info->super_for_commit->super_roots,
1934 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1938 * this copies info out of the root backup array and back into
1939 * the in-memory super block. It is meant to help iterate through
1940 * the array, so you send it the number of backups you've already
1941 * tried and the last backup index you used.
1943 * this returns -1 when it has tried all the backups
1945 static noinline int next_root_backup(struct btrfs_fs_info *info,
1946 struct btrfs_super_block *super,
1947 int *num_backups_tried, int *backup_index)
1949 struct btrfs_root_backup *root_backup;
1950 int newest = *backup_index;
1952 if (*num_backups_tried == 0) {
1953 u64 gen = btrfs_super_generation(super);
1955 newest = find_newest_super_backup(info, gen);
1959 *backup_index = newest;
1960 *num_backups_tried = 1;
1961 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1962 /* we've tried all the backups, all done */
1965 /* jump to the next oldest backup */
1966 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1967 BTRFS_NUM_BACKUP_ROOTS;
1968 *backup_index = newest;
1969 *num_backups_tried += 1;
1971 root_backup = super->super_roots + newest;
1973 btrfs_set_super_generation(super,
1974 btrfs_backup_tree_root_gen(root_backup));
1975 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1976 btrfs_set_super_root_level(super,
1977 btrfs_backup_tree_root_level(root_backup));
1978 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1981 * fixme: the total bytes and num_devices need to match or we should
1984 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1985 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1989 /* helper to cleanup workers */
1990 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1992 btrfs_destroy_workqueue(fs_info->fixup_workers);
1993 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1994 btrfs_destroy_workqueue(fs_info->workers);
1995 btrfs_destroy_workqueue(fs_info->endio_workers);
1996 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1997 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1998 btrfs_destroy_workqueue(fs_info->rmw_workers);
1999 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2000 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2001 btrfs_destroy_workqueue(fs_info->delayed_workers);
2002 btrfs_destroy_workqueue(fs_info->caching_workers);
2003 btrfs_destroy_workqueue(fs_info->readahead_workers);
2004 btrfs_destroy_workqueue(fs_info->flush_workers);
2005 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2007 * Now that all other work queues are destroyed, we can safely destroy
2008 * the queues used for metadata I/O, since tasks from those other work
2009 * queues can do metadata I/O operations.
2011 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2012 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2015 static void free_root_extent_buffers(struct btrfs_root *root)
2018 free_extent_buffer(root->node);
2019 free_extent_buffer(root->commit_root);
2021 root->commit_root = NULL;
2025 /* helper to cleanup tree roots */
2026 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2028 free_root_extent_buffers(info->tree_root);
2030 free_root_extent_buffers(info->dev_root);
2031 free_root_extent_buffers(info->extent_root);
2032 free_root_extent_buffers(info->csum_root);
2033 free_root_extent_buffers(info->quota_root);
2034 free_root_extent_buffers(info->uuid_root);
2035 if (free_chunk_root)
2036 free_root_extent_buffers(info->chunk_root);
2037 free_root_extent_buffers(info->free_space_root);
2040 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2043 struct btrfs_root *gang[8];
2046 while (!list_empty(&fs_info->dead_roots)) {
2047 gang[0] = list_entry(fs_info->dead_roots.next,
2048 struct btrfs_root, root_list);
2049 list_del(&gang[0]->root_list);
2051 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2052 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2054 free_extent_buffer(gang[0]->node);
2055 free_extent_buffer(gang[0]->commit_root);
2056 btrfs_put_fs_root(gang[0]);
2061 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2066 for (i = 0; i < ret; i++)
2067 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2070 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2071 btrfs_free_log_root_tree(NULL, fs_info);
2072 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2076 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2078 mutex_init(&fs_info->scrub_lock);
2079 atomic_set(&fs_info->scrubs_running, 0);
2080 atomic_set(&fs_info->scrub_pause_req, 0);
2081 atomic_set(&fs_info->scrubs_paused, 0);
2082 atomic_set(&fs_info->scrub_cancel_req, 0);
2083 init_waitqueue_head(&fs_info->scrub_pause_wait);
2084 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2087 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2089 spin_lock_init(&fs_info->balance_lock);
2090 mutex_init(&fs_info->balance_mutex);
2091 atomic_set(&fs_info->balance_pause_req, 0);
2092 atomic_set(&fs_info->balance_cancel_req, 0);
2093 fs_info->balance_ctl = NULL;
2094 init_waitqueue_head(&fs_info->balance_wait_q);
2097 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2099 struct inode *inode = fs_info->btree_inode;
2101 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2102 set_nlink(inode, 1);
2104 * we set the i_size on the btree inode to the max possible int.
2105 * the real end of the address space is determined by all of
2106 * the devices in the system
2108 inode->i_size = OFFSET_MAX;
2109 inode->i_mapping->a_ops = &btree_aops;
2111 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2112 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2113 IO_TREE_INODE_IO, inode);
2114 BTRFS_I(inode)->io_tree.track_uptodate = false;
2115 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2117 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2119 BTRFS_I(inode)->root = fs_info->tree_root;
2120 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2121 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2122 btrfs_insert_inode_hash(inode);
2125 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2127 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2128 init_rwsem(&fs_info->dev_replace.rwsem);
2129 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2132 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2134 spin_lock_init(&fs_info->qgroup_lock);
2135 mutex_init(&fs_info->qgroup_ioctl_lock);
2136 fs_info->qgroup_tree = RB_ROOT;
2137 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2138 fs_info->qgroup_seq = 1;
2139 fs_info->qgroup_ulist = NULL;
2140 fs_info->qgroup_rescan_running = false;
2141 mutex_init(&fs_info->qgroup_rescan_lock);
2144 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2145 struct btrfs_fs_devices *fs_devices)
2147 u32 max_active = fs_info->thread_pool_size;
2148 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2151 btrfs_alloc_workqueue(fs_info, "worker",
2152 flags | WQ_HIGHPRI, max_active, 16);
2154 fs_info->delalloc_workers =
2155 btrfs_alloc_workqueue(fs_info, "delalloc",
2156 flags, max_active, 2);
2158 fs_info->flush_workers =
2159 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2160 flags, max_active, 0);
2162 fs_info->caching_workers =
2163 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2165 fs_info->fixup_workers =
2166 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2169 * endios are largely parallel and should have a very
2172 fs_info->endio_workers =
2173 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2174 fs_info->endio_meta_workers =
2175 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2177 fs_info->endio_meta_write_workers =
2178 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2180 fs_info->endio_raid56_workers =
2181 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2183 fs_info->endio_repair_workers =
2184 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2185 fs_info->rmw_workers =
2186 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2187 fs_info->endio_write_workers =
2188 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2190 fs_info->endio_freespace_worker =
2191 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2193 fs_info->delayed_workers =
2194 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2196 fs_info->readahead_workers =
2197 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2199 fs_info->qgroup_rescan_workers =
2200 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2202 if (!(fs_info->workers && fs_info->delalloc_workers &&
2203 fs_info->flush_workers &&
2204 fs_info->endio_workers && fs_info->endio_meta_workers &&
2205 fs_info->endio_meta_write_workers &&
2206 fs_info->endio_repair_workers &&
2207 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2208 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2209 fs_info->caching_workers && fs_info->readahead_workers &&
2210 fs_info->fixup_workers && fs_info->delayed_workers &&
2211 fs_info->qgroup_rescan_workers)) {
2218 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2220 struct crypto_shash *csum_shash;
2221 const char *csum_name = btrfs_super_csum_name(csum_type);
2223 csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2225 if (IS_ERR(csum_shash)) {
2226 btrfs_err(fs_info, "error allocating %s hash for checksum",
2228 return PTR_ERR(csum_shash);
2231 fs_info->csum_shash = csum_shash;
2236 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2238 crypto_free_shash(fs_info->csum_shash);
2241 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2242 struct btrfs_fs_devices *fs_devices)
2245 struct btrfs_root *log_tree_root;
2246 struct btrfs_super_block *disk_super = fs_info->super_copy;
2247 u64 bytenr = btrfs_super_log_root(disk_super);
2248 int level = btrfs_super_log_root_level(disk_super);
2250 if (fs_devices->rw_devices == 0) {
2251 btrfs_warn(fs_info, "log replay required on RO media");
2255 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2259 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2261 log_tree_root->node = read_tree_block(fs_info, bytenr,
2262 fs_info->generation + 1,
2264 if (IS_ERR(log_tree_root->node)) {
2265 btrfs_warn(fs_info, "failed to read log tree");
2266 ret = PTR_ERR(log_tree_root->node);
2267 kfree(log_tree_root);
2269 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2270 btrfs_err(fs_info, "failed to read log tree");
2271 free_extent_buffer(log_tree_root->node);
2272 kfree(log_tree_root);
2275 /* returns with log_tree_root freed on success */
2276 ret = btrfs_recover_log_trees(log_tree_root);
2278 btrfs_handle_fs_error(fs_info, ret,
2279 "Failed to recover log tree");
2280 free_extent_buffer(log_tree_root->node);
2281 kfree(log_tree_root);
2285 if (sb_rdonly(fs_info->sb)) {
2286 ret = btrfs_commit_super(fs_info);
2294 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2296 struct btrfs_root *tree_root = fs_info->tree_root;
2297 struct btrfs_root *root;
2298 struct btrfs_key location;
2301 BUG_ON(!fs_info->tree_root);
2303 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2304 location.type = BTRFS_ROOT_ITEM_KEY;
2305 location.offset = 0;
2307 root = btrfs_read_tree_root(tree_root, &location);
2309 ret = PTR_ERR(root);
2312 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2313 fs_info->extent_root = root;
2315 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2316 root = btrfs_read_tree_root(tree_root, &location);
2318 ret = PTR_ERR(root);
2321 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2322 fs_info->dev_root = root;
2323 btrfs_init_devices_late(fs_info);
2325 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2326 root = btrfs_read_tree_root(tree_root, &location);
2328 ret = PTR_ERR(root);
2331 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2332 fs_info->csum_root = root;
2334 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2335 root = btrfs_read_tree_root(tree_root, &location);
2336 if (!IS_ERR(root)) {
2337 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2338 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2339 fs_info->quota_root = root;
2342 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2343 root = btrfs_read_tree_root(tree_root, &location);
2345 ret = PTR_ERR(root);
2349 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2350 fs_info->uuid_root = root;
2353 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2354 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2355 root = btrfs_read_tree_root(tree_root, &location);
2357 ret = PTR_ERR(root);
2360 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2361 fs_info->free_space_root = root;
2366 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2367 location.objectid, ret);
2372 * Real super block validation
2373 * NOTE: super csum type and incompat features will not be checked here.
2375 * @sb: super block to check
2376 * @mirror_num: the super block number to check its bytenr:
2377 * 0 the primary (1st) sb
2378 * 1, 2 2nd and 3rd backup copy
2379 * -1 skip bytenr check
2381 static int validate_super(struct btrfs_fs_info *fs_info,
2382 struct btrfs_super_block *sb, int mirror_num)
2384 u64 nodesize = btrfs_super_nodesize(sb);
2385 u64 sectorsize = btrfs_super_sectorsize(sb);
2388 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2389 btrfs_err(fs_info, "no valid FS found");
2392 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2393 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2394 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2397 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2398 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2399 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2402 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2403 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2404 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2407 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2408 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2409 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2414 * Check sectorsize and nodesize first, other check will need it.
2415 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2417 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2418 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2419 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2422 /* Only PAGE SIZE is supported yet */
2423 if (sectorsize != PAGE_SIZE) {
2425 "sectorsize %llu not supported yet, only support %lu",
2426 sectorsize, PAGE_SIZE);
2429 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2430 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2431 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2434 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2435 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2436 le32_to_cpu(sb->__unused_leafsize), nodesize);
2440 /* Root alignment check */
2441 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2442 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2443 btrfs_super_root(sb));
2446 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2447 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2448 btrfs_super_chunk_root(sb));
2451 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2452 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2453 btrfs_super_log_root(sb));
2457 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2458 BTRFS_FSID_SIZE) != 0) {
2460 "dev_item UUID does not match metadata fsid: %pU != %pU",
2461 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2466 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2469 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2470 btrfs_err(fs_info, "bytes_used is too small %llu",
2471 btrfs_super_bytes_used(sb));
2474 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2475 btrfs_err(fs_info, "invalid stripesize %u",
2476 btrfs_super_stripesize(sb));
2479 if (btrfs_super_num_devices(sb) > (1UL << 31))
2480 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2481 btrfs_super_num_devices(sb));
2482 if (btrfs_super_num_devices(sb) == 0) {
2483 btrfs_err(fs_info, "number of devices is 0");
2487 if (mirror_num >= 0 &&
2488 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2489 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2490 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2495 * Obvious sys_chunk_array corruptions, it must hold at least one key
2498 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2499 btrfs_err(fs_info, "system chunk array too big %u > %u",
2500 btrfs_super_sys_array_size(sb),
2501 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2504 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2505 + sizeof(struct btrfs_chunk)) {
2506 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2507 btrfs_super_sys_array_size(sb),
2508 sizeof(struct btrfs_disk_key)
2509 + sizeof(struct btrfs_chunk));
2514 * The generation is a global counter, we'll trust it more than the others
2515 * but it's still possible that it's the one that's wrong.
2517 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2519 "suspicious: generation < chunk_root_generation: %llu < %llu",
2520 btrfs_super_generation(sb),
2521 btrfs_super_chunk_root_generation(sb));
2522 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2523 && btrfs_super_cache_generation(sb) != (u64)-1)
2525 "suspicious: generation < cache_generation: %llu < %llu",
2526 btrfs_super_generation(sb),
2527 btrfs_super_cache_generation(sb));
2533 * Validation of super block at mount time.
2534 * Some checks already done early at mount time, like csum type and incompat
2535 * flags will be skipped.
2537 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2539 return validate_super(fs_info, fs_info->super_copy, 0);
2543 * Validation of super block at write time.
2544 * Some checks like bytenr check will be skipped as their values will be
2546 * Extra checks like csum type and incompat flags will be done here.
2548 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2549 struct btrfs_super_block *sb)
2553 ret = validate_super(fs_info, sb, -1);
2556 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2558 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2559 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2562 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2565 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2566 btrfs_super_incompat_flags(sb),
2567 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2573 "super block corruption detected before writing it to disk");
2577 int __cold open_ctree(struct super_block *sb,
2578 struct btrfs_fs_devices *fs_devices,
2587 struct btrfs_key location;
2588 struct buffer_head *bh;
2589 struct btrfs_super_block *disk_super;
2590 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2591 struct btrfs_root *tree_root;
2592 struct btrfs_root *chunk_root;
2595 int num_backups_tried = 0;
2596 int backup_index = 0;
2597 int clear_free_space_tree = 0;
2600 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2601 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2602 if (!tree_root || !chunk_root) {
2607 ret = init_srcu_struct(&fs_info->subvol_srcu);
2613 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2619 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2622 goto fail_dio_bytes;
2624 fs_info->dirty_metadata_batch = PAGE_SIZE *
2625 (1 + ilog2(nr_cpu_ids));
2627 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2630 goto fail_dirty_metadata_bytes;
2633 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2637 goto fail_delalloc_bytes;
2640 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2641 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2642 INIT_LIST_HEAD(&fs_info->trans_list);
2643 INIT_LIST_HEAD(&fs_info->dead_roots);
2644 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2645 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2646 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2647 spin_lock_init(&fs_info->delalloc_root_lock);
2648 spin_lock_init(&fs_info->trans_lock);
2649 spin_lock_init(&fs_info->fs_roots_radix_lock);
2650 spin_lock_init(&fs_info->delayed_iput_lock);
2651 spin_lock_init(&fs_info->defrag_inodes_lock);
2652 spin_lock_init(&fs_info->tree_mod_seq_lock);
2653 spin_lock_init(&fs_info->super_lock);
2654 spin_lock_init(&fs_info->buffer_lock);
2655 spin_lock_init(&fs_info->unused_bgs_lock);
2656 rwlock_init(&fs_info->tree_mod_log_lock);
2657 mutex_init(&fs_info->unused_bg_unpin_mutex);
2658 mutex_init(&fs_info->delete_unused_bgs_mutex);
2659 mutex_init(&fs_info->reloc_mutex);
2660 mutex_init(&fs_info->delalloc_root_mutex);
2661 seqlock_init(&fs_info->profiles_lock);
2663 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2664 INIT_LIST_HEAD(&fs_info->space_info);
2665 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2666 INIT_LIST_HEAD(&fs_info->unused_bgs);
2667 extent_map_tree_init(&fs_info->mapping_tree);
2668 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2669 BTRFS_BLOCK_RSV_GLOBAL);
2670 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2671 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2672 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2673 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2674 BTRFS_BLOCK_RSV_DELOPS);
2675 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2676 BTRFS_BLOCK_RSV_DELREFS);
2678 atomic_set(&fs_info->async_delalloc_pages, 0);
2679 atomic_set(&fs_info->defrag_running, 0);
2680 atomic_set(&fs_info->reada_works_cnt, 0);
2681 atomic_set(&fs_info->nr_delayed_iputs, 0);
2682 atomic64_set(&fs_info->tree_mod_seq, 0);
2684 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2685 fs_info->metadata_ratio = 0;
2686 fs_info->defrag_inodes = RB_ROOT;
2687 atomic64_set(&fs_info->free_chunk_space, 0);
2688 fs_info->tree_mod_log = RB_ROOT;
2689 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2690 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2691 /* readahead state */
2692 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2693 spin_lock_init(&fs_info->reada_lock);
2694 btrfs_init_ref_verify(fs_info);
2696 fs_info->thread_pool_size = min_t(unsigned long,
2697 num_online_cpus() + 2, 8);
2699 INIT_LIST_HEAD(&fs_info->ordered_roots);
2700 spin_lock_init(&fs_info->ordered_root_lock);
2702 fs_info->btree_inode = new_inode(sb);
2703 if (!fs_info->btree_inode) {
2705 goto fail_bio_counter;
2707 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2709 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2711 if (!fs_info->delayed_root) {
2715 btrfs_init_delayed_root(fs_info->delayed_root);
2717 btrfs_init_scrub(fs_info);
2718 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2719 fs_info->check_integrity_print_mask = 0;
2721 btrfs_init_balance(fs_info);
2722 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2724 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2725 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2727 btrfs_init_btree_inode(fs_info);
2729 spin_lock_init(&fs_info->block_group_cache_lock);
2730 fs_info->block_group_cache_tree = RB_ROOT;
2731 fs_info->first_logical_byte = (u64)-1;
2733 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2734 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2735 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2736 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2737 fs_info->pinned_extents = &fs_info->freed_extents[0];
2738 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2740 mutex_init(&fs_info->ordered_operations_mutex);
2741 mutex_init(&fs_info->tree_log_mutex);
2742 mutex_init(&fs_info->chunk_mutex);
2743 mutex_init(&fs_info->transaction_kthread_mutex);
2744 mutex_init(&fs_info->cleaner_mutex);
2745 mutex_init(&fs_info->ro_block_group_mutex);
2746 init_rwsem(&fs_info->commit_root_sem);
2747 init_rwsem(&fs_info->cleanup_work_sem);
2748 init_rwsem(&fs_info->subvol_sem);
2749 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2751 btrfs_init_dev_replace_locks(fs_info);
2752 btrfs_init_qgroup(fs_info);
2754 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2755 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2757 init_waitqueue_head(&fs_info->transaction_throttle);
2758 init_waitqueue_head(&fs_info->transaction_wait);
2759 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2760 init_waitqueue_head(&fs_info->async_submit_wait);
2761 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2763 /* Usable values until the real ones are cached from the superblock */
2764 fs_info->nodesize = 4096;
2765 fs_info->sectorsize = 4096;
2766 fs_info->stripesize = 4096;
2768 spin_lock_init(&fs_info->swapfile_pins_lock);
2769 fs_info->swapfile_pins = RB_ROOT;
2771 fs_info->send_in_progress = 0;
2773 ret = btrfs_alloc_stripe_hash_table(fs_info);
2779 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2781 invalidate_bdev(fs_devices->latest_bdev);
2784 * Read super block and check the signature bytes only
2786 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2793 * Verify the type first, if that or the the checksum value are
2794 * corrupted, we'll find out
2796 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2797 if (!btrfs_supported_super_csum(csum_type)) {
2798 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2805 ret = btrfs_init_csum_hash(fs_info, csum_type);
2812 * We want to check superblock checksum, the type is stored inside.
2813 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2815 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2816 btrfs_err(fs_info, "superblock checksum mismatch");
2823 * super_copy is zeroed at allocation time and we never touch the
2824 * following bytes up to INFO_SIZE, the checksum is calculated from
2825 * the whole block of INFO_SIZE
2827 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2830 disk_super = fs_info->super_copy;
2832 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2835 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2836 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2837 fs_info->super_copy->metadata_uuid,
2841 features = btrfs_super_flags(disk_super);
2842 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2843 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2844 btrfs_set_super_flags(disk_super, features);
2846 "found metadata UUID change in progress flag, clearing");
2849 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2850 sizeof(*fs_info->super_for_commit));
2852 ret = btrfs_validate_mount_super(fs_info);
2854 btrfs_err(fs_info, "superblock contains fatal errors");
2859 if (!btrfs_super_root(disk_super))
2862 /* check FS state, whether FS is broken. */
2863 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2864 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2867 * run through our array of backup supers and setup
2868 * our ring pointer to the oldest one
2870 generation = btrfs_super_generation(disk_super);
2871 find_oldest_super_backup(fs_info, generation);
2874 * In the long term, we'll store the compression type in the super
2875 * block, and it'll be used for per file compression control.
2877 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2879 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2885 features = btrfs_super_incompat_flags(disk_super) &
2886 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2889 "cannot mount because of unsupported optional features (%llx)",
2895 features = btrfs_super_incompat_flags(disk_super);
2896 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2897 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2898 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2899 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2900 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2902 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2903 btrfs_info(fs_info, "has skinny extents");
2906 * flag our filesystem as having big metadata blocks if
2907 * they are bigger than the page size
2909 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2910 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2912 "flagging fs with big metadata feature");
2913 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2916 nodesize = btrfs_super_nodesize(disk_super);
2917 sectorsize = btrfs_super_sectorsize(disk_super);
2918 stripesize = sectorsize;
2919 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2920 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2922 /* Cache block sizes */
2923 fs_info->nodesize = nodesize;
2924 fs_info->sectorsize = sectorsize;
2925 fs_info->stripesize = stripesize;
2928 * mixed block groups end up with duplicate but slightly offset
2929 * extent buffers for the same range. It leads to corruptions
2931 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2932 (sectorsize != nodesize)) {
2934 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2935 nodesize, sectorsize);
2940 * Needn't use the lock because there is no other task which will
2943 btrfs_set_super_incompat_flags(disk_super, features);
2945 features = btrfs_super_compat_ro_flags(disk_super) &
2946 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2947 if (!sb_rdonly(sb) && features) {
2949 "cannot mount read-write because of unsupported optional features (%llx)",
2955 ret = btrfs_init_workqueues(fs_info, fs_devices);
2958 goto fail_sb_buffer;
2961 sb->s_bdi->congested_fn = btrfs_congested_fn;
2962 sb->s_bdi->congested_data = fs_info;
2963 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2964 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2965 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2966 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2968 sb->s_blocksize = sectorsize;
2969 sb->s_blocksize_bits = blksize_bits(sectorsize);
2970 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2972 mutex_lock(&fs_info->chunk_mutex);
2973 ret = btrfs_read_sys_array(fs_info);
2974 mutex_unlock(&fs_info->chunk_mutex);
2976 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2977 goto fail_sb_buffer;
2980 generation = btrfs_super_chunk_root_generation(disk_super);
2981 level = btrfs_super_chunk_root_level(disk_super);
2983 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2985 chunk_root->node = read_tree_block(fs_info,
2986 btrfs_super_chunk_root(disk_super),
2987 generation, level, NULL);
2988 if (IS_ERR(chunk_root->node) ||
2989 !extent_buffer_uptodate(chunk_root->node)) {
2990 btrfs_err(fs_info, "failed to read chunk root");
2991 if (!IS_ERR(chunk_root->node))
2992 free_extent_buffer(chunk_root->node);
2993 chunk_root->node = NULL;
2994 goto fail_tree_roots;
2996 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2997 chunk_root->commit_root = btrfs_root_node(chunk_root);
2999 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3000 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3002 ret = btrfs_read_chunk_tree(fs_info);
3004 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3005 goto fail_tree_roots;
3009 * Keep the devid that is marked to be the target device for the
3010 * device replace procedure
3012 btrfs_free_extra_devids(fs_devices, 0);
3014 if (!fs_devices->latest_bdev) {
3015 btrfs_err(fs_info, "failed to read devices");
3016 goto fail_tree_roots;
3020 generation = btrfs_super_generation(disk_super);
3021 level = btrfs_super_root_level(disk_super);
3023 tree_root->node = read_tree_block(fs_info,
3024 btrfs_super_root(disk_super),
3025 generation, level, NULL);
3026 if (IS_ERR(tree_root->node) ||
3027 !extent_buffer_uptodate(tree_root->node)) {
3028 btrfs_warn(fs_info, "failed to read tree root");
3029 if (!IS_ERR(tree_root->node))
3030 free_extent_buffer(tree_root->node);
3031 tree_root->node = NULL;
3032 goto recovery_tree_root;
3035 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3036 tree_root->commit_root = btrfs_root_node(tree_root);
3037 btrfs_set_root_refs(&tree_root->root_item, 1);
3039 mutex_lock(&tree_root->objectid_mutex);
3040 ret = btrfs_find_highest_objectid(tree_root,
3041 &tree_root->highest_objectid);
3043 mutex_unlock(&tree_root->objectid_mutex);
3044 goto recovery_tree_root;
3047 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3049 mutex_unlock(&tree_root->objectid_mutex);
3051 ret = btrfs_read_roots(fs_info);
3053 goto recovery_tree_root;
3055 fs_info->generation = generation;
3056 fs_info->last_trans_committed = generation;
3058 ret = btrfs_verify_dev_extents(fs_info);
3061 "failed to verify dev extents against chunks: %d",
3063 goto fail_block_groups;
3065 ret = btrfs_recover_balance(fs_info);
3067 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3068 goto fail_block_groups;
3071 ret = btrfs_init_dev_stats(fs_info);
3073 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3074 goto fail_block_groups;
3077 ret = btrfs_init_dev_replace(fs_info);
3079 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3080 goto fail_block_groups;
3083 btrfs_free_extra_devids(fs_devices, 1);
3085 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3087 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3089 goto fail_block_groups;
3092 ret = btrfs_sysfs_add_device(fs_devices);
3094 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3096 goto fail_fsdev_sysfs;
3099 ret = btrfs_sysfs_add_mounted(fs_info);
3101 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3102 goto fail_fsdev_sysfs;
3105 ret = btrfs_init_space_info(fs_info);
3107 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3111 ret = btrfs_read_block_groups(fs_info);
3113 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3117 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3119 "writable mount is not allowed due to too many missing devices");
3123 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3125 if (IS_ERR(fs_info->cleaner_kthread))
3128 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3130 "btrfs-transaction");
3131 if (IS_ERR(fs_info->transaction_kthread))
3134 if (!btrfs_test_opt(fs_info, NOSSD) &&
3135 !fs_info->fs_devices->rotating) {
3136 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3140 * Mount does not set all options immediately, we can do it now and do
3141 * not have to wait for transaction commit
3143 btrfs_apply_pending_changes(fs_info);
3145 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3146 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3147 ret = btrfsic_mount(fs_info, fs_devices,
3148 btrfs_test_opt(fs_info,
3149 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3151 fs_info->check_integrity_print_mask);
3154 "failed to initialize integrity check module: %d",
3158 ret = btrfs_read_qgroup_config(fs_info);
3160 goto fail_trans_kthread;
3162 if (btrfs_build_ref_tree(fs_info))
3163 btrfs_err(fs_info, "couldn't build ref tree");
3165 /* do not make disk changes in broken FS or nologreplay is given */
3166 if (btrfs_super_log_root(disk_super) != 0 &&
3167 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3168 ret = btrfs_replay_log(fs_info, fs_devices);
3175 ret = btrfs_find_orphan_roots(fs_info);
3179 if (!sb_rdonly(sb)) {
3180 ret = btrfs_cleanup_fs_roots(fs_info);
3184 mutex_lock(&fs_info->cleaner_mutex);
3185 ret = btrfs_recover_relocation(tree_root);
3186 mutex_unlock(&fs_info->cleaner_mutex);
3188 btrfs_warn(fs_info, "failed to recover relocation: %d",
3195 location.objectid = BTRFS_FS_TREE_OBJECTID;
3196 location.type = BTRFS_ROOT_ITEM_KEY;
3197 location.offset = 0;
3199 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3200 if (IS_ERR(fs_info->fs_root)) {
3201 err = PTR_ERR(fs_info->fs_root);
3202 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3209 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3210 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3211 clear_free_space_tree = 1;
3212 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3213 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3214 btrfs_warn(fs_info, "free space tree is invalid");
3215 clear_free_space_tree = 1;
3218 if (clear_free_space_tree) {
3219 btrfs_info(fs_info, "clearing free space tree");
3220 ret = btrfs_clear_free_space_tree(fs_info);
3223 "failed to clear free space tree: %d", ret);
3224 close_ctree(fs_info);
3229 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3230 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3231 btrfs_info(fs_info, "creating free space tree");
3232 ret = btrfs_create_free_space_tree(fs_info);
3235 "failed to create free space tree: %d", ret);
3236 close_ctree(fs_info);
3241 down_read(&fs_info->cleanup_work_sem);
3242 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3243 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3244 up_read(&fs_info->cleanup_work_sem);
3245 close_ctree(fs_info);
3248 up_read(&fs_info->cleanup_work_sem);
3250 ret = btrfs_resume_balance_async(fs_info);
3252 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3253 close_ctree(fs_info);
3257 ret = btrfs_resume_dev_replace_async(fs_info);
3259 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3260 close_ctree(fs_info);
3264 btrfs_qgroup_rescan_resume(fs_info);
3266 if (!fs_info->uuid_root) {
3267 btrfs_info(fs_info, "creating UUID tree");
3268 ret = btrfs_create_uuid_tree(fs_info);
3271 "failed to create the UUID tree: %d", ret);
3272 close_ctree(fs_info);
3275 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3276 fs_info->generation !=
3277 btrfs_super_uuid_tree_generation(disk_super)) {
3278 btrfs_info(fs_info, "checking UUID tree");
3279 ret = btrfs_check_uuid_tree(fs_info);
3282 "failed to check the UUID tree: %d", ret);
3283 close_ctree(fs_info);
3287 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3289 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3292 * backuproot only affect mount behavior, and if open_ctree succeeded,
3293 * no need to keep the flag
3295 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3300 btrfs_free_qgroup_config(fs_info);
3302 kthread_stop(fs_info->transaction_kthread);
3303 btrfs_cleanup_transaction(fs_info);
3304 btrfs_free_fs_roots(fs_info);
3306 kthread_stop(fs_info->cleaner_kthread);
3309 * make sure we're done with the btree inode before we stop our
3312 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3315 btrfs_sysfs_remove_mounted(fs_info);
3318 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3321 btrfs_put_block_group_cache(fs_info);
3324 free_root_pointers(fs_info, true);
3325 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3328 btrfs_stop_all_workers(fs_info);
3329 btrfs_free_block_groups(fs_info);
3331 btrfs_free_csum_hash(fs_info);
3334 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3336 iput(fs_info->btree_inode);
3338 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3339 fail_delalloc_bytes:
3340 percpu_counter_destroy(&fs_info->delalloc_bytes);
3341 fail_dirty_metadata_bytes:
3342 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3344 percpu_counter_destroy(&fs_info->dio_bytes);
3346 cleanup_srcu_struct(&fs_info->subvol_srcu);
3348 btrfs_free_stripe_hash_table(fs_info);
3349 btrfs_close_devices(fs_info->fs_devices);
3353 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3354 goto fail_tree_roots;
3356 free_root_pointers(fs_info, false);
3358 /* don't use the log in recovery mode, it won't be valid */
3359 btrfs_set_super_log_root(disk_super, 0);
3361 /* we can't trust the free space cache either */
3362 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3364 ret = next_root_backup(fs_info, fs_info->super_copy,
3365 &num_backups_tried, &backup_index);
3367 goto fail_block_groups;
3368 goto retry_root_backup;
3370 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3372 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3375 set_buffer_uptodate(bh);
3377 struct btrfs_device *device = (struct btrfs_device *)
3380 btrfs_warn_rl_in_rcu(device->fs_info,
3381 "lost page write due to IO error on %s",
3382 rcu_str_deref(device->name));
3383 /* note, we don't set_buffer_write_io_error because we have
3384 * our own ways of dealing with the IO errors
3386 clear_buffer_uptodate(bh);
3387 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3393 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3394 struct buffer_head **bh_ret)
3396 struct buffer_head *bh;
3397 struct btrfs_super_block *super;
3400 bytenr = btrfs_sb_offset(copy_num);
3401 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3404 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3406 * If we fail to read from the underlying devices, as of now
3407 * the best option we have is to mark it EIO.
3412 super = (struct btrfs_super_block *)bh->b_data;
3413 if (btrfs_super_bytenr(super) != bytenr ||
3414 btrfs_super_magic(super) != BTRFS_MAGIC) {
3424 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3426 struct buffer_head *bh;
3427 struct buffer_head *latest = NULL;
3428 struct btrfs_super_block *super;
3433 /* we would like to check all the supers, but that would make
3434 * a btrfs mount succeed after a mkfs from a different FS.
3435 * So, we need to add a special mount option to scan for
3436 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3438 for (i = 0; i < 1; i++) {
3439 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3443 super = (struct btrfs_super_block *)bh->b_data;
3445 if (!latest || btrfs_super_generation(super) > transid) {
3448 transid = btrfs_super_generation(super);
3455 return ERR_PTR(ret);
3461 * Write superblock @sb to the @device. Do not wait for completion, all the
3462 * buffer heads we write are pinned.
3464 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3465 * the expected device size at commit time. Note that max_mirrors must be
3466 * same for write and wait phases.
3468 * Return number of errors when buffer head is not found or submission fails.
3470 static int write_dev_supers(struct btrfs_device *device,
3471 struct btrfs_super_block *sb, int max_mirrors)
3473 struct btrfs_fs_info *fs_info = device->fs_info;
3474 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3475 struct buffer_head *bh;
3482 if (max_mirrors == 0)
3483 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3485 shash->tfm = fs_info->csum_shash;
3487 for (i = 0; i < max_mirrors; i++) {
3488 bytenr = btrfs_sb_offset(i);
3489 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3490 device->commit_total_bytes)
3493 btrfs_set_super_bytenr(sb, bytenr);
3495 crypto_shash_init(shash);
3496 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3497 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3498 crypto_shash_final(shash, sb->csum);
3500 /* One reference for us, and we leave it for the caller */
3501 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3502 BTRFS_SUPER_INFO_SIZE);
3504 btrfs_err(device->fs_info,
3505 "couldn't get super buffer head for bytenr %llu",
3511 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3513 /* one reference for submit_bh */
3516 set_buffer_uptodate(bh);
3518 bh->b_end_io = btrfs_end_buffer_write_sync;
3519 bh->b_private = device;
3522 * we fua the first super. The others we allow
3525 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3526 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3527 op_flags |= REQ_FUA;
3528 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3532 return errors < i ? 0 : -1;
3536 * Wait for write completion of superblocks done by write_dev_supers,
3537 * @max_mirrors same for write and wait phases.
3539 * Return number of errors when buffer head is not found or not marked up to
3542 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3544 struct buffer_head *bh;
3547 bool primary_failed = false;
3550 if (max_mirrors == 0)
3551 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3553 for (i = 0; i < max_mirrors; i++) {
3554 bytenr = btrfs_sb_offset(i);
3555 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3556 device->commit_total_bytes)
3559 bh = __find_get_block(device->bdev,
3560 bytenr / BTRFS_BDEV_BLOCKSIZE,
3561 BTRFS_SUPER_INFO_SIZE);
3565 primary_failed = true;
3569 if (!buffer_uptodate(bh)) {
3572 primary_failed = true;
3575 /* drop our reference */
3578 /* drop the reference from the writing run */
3582 /* log error, force error return */
3583 if (primary_failed) {
3584 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3589 return errors < i ? 0 : -1;
3593 * endio for the write_dev_flush, this will wake anyone waiting
3594 * for the barrier when it is done
3596 static void btrfs_end_empty_barrier(struct bio *bio)
3598 complete(bio->bi_private);
3602 * Submit a flush request to the device if it supports it. Error handling is
3603 * done in the waiting counterpart.
3605 static void write_dev_flush(struct btrfs_device *device)
3607 struct request_queue *q = bdev_get_queue(device->bdev);
3608 struct bio *bio = device->flush_bio;
3610 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3614 bio->bi_end_io = btrfs_end_empty_barrier;
3615 bio_set_dev(bio, device->bdev);
3616 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3617 init_completion(&device->flush_wait);
3618 bio->bi_private = &device->flush_wait;
3620 btrfsic_submit_bio(bio);
3621 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3625 * If the flush bio has been submitted by write_dev_flush, wait for it.
3627 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3629 struct bio *bio = device->flush_bio;
3631 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3634 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3635 wait_for_completion_io(&device->flush_wait);
3637 return bio->bi_status;
3640 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3642 if (!btrfs_check_rw_degradable(fs_info, NULL))
3648 * send an empty flush down to each device in parallel,
3649 * then wait for them
3651 static int barrier_all_devices(struct btrfs_fs_info *info)
3653 struct list_head *head;
3654 struct btrfs_device *dev;
3655 int errors_wait = 0;
3658 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3659 /* send down all the barriers */
3660 head = &info->fs_devices->devices;
3661 list_for_each_entry(dev, head, dev_list) {
3662 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3666 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3667 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3670 write_dev_flush(dev);
3671 dev->last_flush_error = BLK_STS_OK;
3674 /* wait for all the barriers */
3675 list_for_each_entry(dev, head, dev_list) {
3676 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3682 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3683 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3686 ret = wait_dev_flush(dev);
3688 dev->last_flush_error = ret;
3689 btrfs_dev_stat_inc_and_print(dev,
3690 BTRFS_DEV_STAT_FLUSH_ERRS);
3697 * At some point we need the status of all disks
3698 * to arrive at the volume status. So error checking
3699 * is being pushed to a separate loop.
3701 return check_barrier_error(info);
3706 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3709 int min_tolerated = INT_MAX;
3711 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3712 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3713 min_tolerated = min_t(int, min_tolerated,
3714 btrfs_raid_array[BTRFS_RAID_SINGLE].
3715 tolerated_failures);
3717 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3718 if (raid_type == BTRFS_RAID_SINGLE)
3720 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3722 min_tolerated = min_t(int, min_tolerated,
3723 btrfs_raid_array[raid_type].
3724 tolerated_failures);
3727 if (min_tolerated == INT_MAX) {
3728 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3732 return min_tolerated;
3735 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3737 struct list_head *head;
3738 struct btrfs_device *dev;
3739 struct btrfs_super_block *sb;
3740 struct btrfs_dev_item *dev_item;
3744 int total_errors = 0;
3747 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3750 * max_mirrors == 0 indicates we're from commit_transaction,
3751 * not from fsync where the tree roots in fs_info have not
3752 * been consistent on disk.
3754 if (max_mirrors == 0)
3755 backup_super_roots(fs_info);
3757 sb = fs_info->super_for_commit;
3758 dev_item = &sb->dev_item;
3760 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3761 head = &fs_info->fs_devices->devices;
3762 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3765 ret = barrier_all_devices(fs_info);
3768 &fs_info->fs_devices->device_list_mutex);
3769 btrfs_handle_fs_error(fs_info, ret,
3770 "errors while submitting device barriers.");
3775 list_for_each_entry(dev, head, dev_list) {
3780 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3781 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3784 btrfs_set_stack_device_generation(dev_item, 0);
3785 btrfs_set_stack_device_type(dev_item, dev->type);
3786 btrfs_set_stack_device_id(dev_item, dev->devid);
3787 btrfs_set_stack_device_total_bytes(dev_item,
3788 dev->commit_total_bytes);
3789 btrfs_set_stack_device_bytes_used(dev_item,
3790 dev->commit_bytes_used);
3791 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3792 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3793 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3794 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3795 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3798 flags = btrfs_super_flags(sb);
3799 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3801 ret = btrfs_validate_write_super(fs_info, sb);
3803 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3804 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3805 "unexpected superblock corruption detected");
3809 ret = write_dev_supers(dev, sb, max_mirrors);
3813 if (total_errors > max_errors) {
3814 btrfs_err(fs_info, "%d errors while writing supers",
3816 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3818 /* FUA is masked off if unsupported and can't be the reason */
3819 btrfs_handle_fs_error(fs_info, -EIO,
3820 "%d errors while writing supers",
3826 list_for_each_entry(dev, head, dev_list) {
3829 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3830 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3833 ret = wait_dev_supers(dev, max_mirrors);
3837 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3838 if (total_errors > max_errors) {
3839 btrfs_handle_fs_error(fs_info, -EIO,
3840 "%d errors while writing supers",
3847 /* Drop a fs root from the radix tree and free it. */
3848 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3849 struct btrfs_root *root)
3851 spin_lock(&fs_info->fs_roots_radix_lock);
3852 radix_tree_delete(&fs_info->fs_roots_radix,
3853 (unsigned long)root->root_key.objectid);
3854 spin_unlock(&fs_info->fs_roots_radix_lock);
3856 if (btrfs_root_refs(&root->root_item) == 0)
3857 synchronize_srcu(&fs_info->subvol_srcu);
3859 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3860 btrfs_free_log(NULL, root);
3861 if (root->reloc_root) {
3862 free_extent_buffer(root->reloc_root->node);
3863 free_extent_buffer(root->reloc_root->commit_root);
3864 btrfs_put_fs_root(root->reloc_root);
3865 root->reloc_root = NULL;
3869 if (root->free_ino_pinned)
3870 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3871 if (root->free_ino_ctl)
3872 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3873 btrfs_free_fs_root(root);
3876 void btrfs_free_fs_root(struct btrfs_root *root)
3878 iput(root->ino_cache_inode);
3879 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3881 free_anon_bdev(root->anon_dev);
3882 if (root->subv_writers)
3883 btrfs_free_subvolume_writers(root->subv_writers);
3884 free_extent_buffer(root->node);
3885 free_extent_buffer(root->commit_root);
3886 kfree(root->free_ino_ctl);
3887 kfree(root->free_ino_pinned);
3888 btrfs_put_fs_root(root);
3891 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3893 u64 root_objectid = 0;
3894 struct btrfs_root *gang[8];
3897 unsigned int ret = 0;
3901 index = srcu_read_lock(&fs_info->subvol_srcu);
3902 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3903 (void **)gang, root_objectid,
3906 srcu_read_unlock(&fs_info->subvol_srcu, index);
3909 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3911 for (i = 0; i < ret; i++) {
3912 /* Avoid to grab roots in dead_roots */
3913 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3917 /* grab all the search result for later use */
3918 gang[i] = btrfs_grab_fs_root(gang[i]);
3920 srcu_read_unlock(&fs_info->subvol_srcu, index);
3922 for (i = 0; i < ret; i++) {
3925 root_objectid = gang[i]->root_key.objectid;
3926 err = btrfs_orphan_cleanup(gang[i]);
3929 btrfs_put_fs_root(gang[i]);
3934 /* release the uncleaned roots due to error */
3935 for (; i < ret; i++) {
3937 btrfs_put_fs_root(gang[i]);
3942 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3944 struct btrfs_root *root = fs_info->tree_root;
3945 struct btrfs_trans_handle *trans;
3947 mutex_lock(&fs_info->cleaner_mutex);
3948 btrfs_run_delayed_iputs(fs_info);
3949 mutex_unlock(&fs_info->cleaner_mutex);
3950 wake_up_process(fs_info->cleaner_kthread);
3952 /* wait until ongoing cleanup work done */
3953 down_write(&fs_info->cleanup_work_sem);
3954 up_write(&fs_info->cleanup_work_sem);
3956 trans = btrfs_join_transaction(root);
3958 return PTR_ERR(trans);
3959 return btrfs_commit_transaction(trans);
3962 void __cold close_ctree(struct btrfs_fs_info *fs_info)
3966 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3968 * We don't want the cleaner to start new transactions, add more delayed
3969 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3970 * because that frees the task_struct, and the transaction kthread might
3971 * still try to wake up the cleaner.
3973 kthread_park(fs_info->cleaner_kthread);
3975 /* wait for the qgroup rescan worker to stop */
3976 btrfs_qgroup_wait_for_completion(fs_info, false);
3978 /* wait for the uuid_scan task to finish */
3979 down(&fs_info->uuid_tree_rescan_sem);
3980 /* avoid complains from lockdep et al., set sem back to initial state */
3981 up(&fs_info->uuid_tree_rescan_sem);
3983 /* pause restriper - we want to resume on mount */
3984 btrfs_pause_balance(fs_info);
3986 btrfs_dev_replace_suspend_for_unmount(fs_info);
3988 btrfs_scrub_cancel(fs_info);
3990 /* wait for any defraggers to finish */
3991 wait_event(fs_info->transaction_wait,
3992 (atomic_read(&fs_info->defrag_running) == 0));
3994 /* clear out the rbtree of defraggable inodes */
3995 btrfs_cleanup_defrag_inodes(fs_info);
3997 cancel_work_sync(&fs_info->async_reclaim_work);
3999 if (!sb_rdonly(fs_info->sb)) {
4001 * The cleaner kthread is stopped, so do one final pass over
4002 * unused block groups.
4004 btrfs_delete_unused_bgs(fs_info);
4006 ret = btrfs_commit_super(fs_info);
4008 btrfs_err(fs_info, "commit super ret %d", ret);
4011 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4012 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4013 btrfs_error_commit_super(fs_info);
4015 kthread_stop(fs_info->transaction_kthread);
4016 kthread_stop(fs_info->cleaner_kthread);
4018 ASSERT(list_empty(&fs_info->delayed_iputs));
4019 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4021 btrfs_free_qgroup_config(fs_info);
4022 ASSERT(list_empty(&fs_info->delalloc_roots));
4024 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4025 btrfs_info(fs_info, "at unmount delalloc count %lld",
4026 percpu_counter_sum(&fs_info->delalloc_bytes));
4029 if (percpu_counter_sum(&fs_info->dio_bytes))
4030 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4031 percpu_counter_sum(&fs_info->dio_bytes));
4033 btrfs_sysfs_remove_mounted(fs_info);
4034 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4036 btrfs_free_fs_roots(fs_info);
4038 btrfs_put_block_group_cache(fs_info);
4041 * we must make sure there is not any read request to
4042 * submit after we stopping all workers.
4044 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4045 btrfs_stop_all_workers(fs_info);
4047 btrfs_free_block_groups(fs_info);
4049 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4050 free_root_pointers(fs_info, true);
4052 iput(fs_info->btree_inode);
4054 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4055 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4056 btrfsic_unmount(fs_info->fs_devices);
4059 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4060 btrfs_close_devices(fs_info->fs_devices);
4062 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4063 percpu_counter_destroy(&fs_info->delalloc_bytes);
4064 percpu_counter_destroy(&fs_info->dio_bytes);
4065 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4066 cleanup_srcu_struct(&fs_info->subvol_srcu);
4068 btrfs_free_csum_hash(fs_info);
4069 btrfs_free_stripe_hash_table(fs_info);
4070 btrfs_free_ref_cache(fs_info);
4073 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4077 struct inode *btree_inode = buf->pages[0]->mapping->host;
4079 ret = extent_buffer_uptodate(buf);
4083 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4084 parent_transid, atomic);
4090 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4092 struct btrfs_fs_info *fs_info;
4093 struct btrfs_root *root;
4094 u64 transid = btrfs_header_generation(buf);
4097 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4099 * This is a fast path so only do this check if we have sanity tests
4100 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4101 * outside of the sanity tests.
4103 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4106 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4107 fs_info = root->fs_info;
4108 btrfs_assert_tree_locked(buf);
4109 if (transid != fs_info->generation)
4110 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4111 buf->start, transid, fs_info->generation);
4112 was_dirty = set_extent_buffer_dirty(buf);
4114 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4116 fs_info->dirty_metadata_batch);
4117 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4119 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4120 * but item data not updated.
4121 * So here we should only check item pointers, not item data.
4123 if (btrfs_header_level(buf) == 0 &&
4124 btrfs_check_leaf_relaxed(buf)) {
4125 btrfs_print_leaf(buf);
4131 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4135 * looks as though older kernels can get into trouble with
4136 * this code, they end up stuck in balance_dirty_pages forever
4140 if (current->flags & PF_MEMALLOC)
4144 btrfs_balance_delayed_items(fs_info);
4146 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4147 BTRFS_DIRTY_METADATA_THRESH,
4148 fs_info->dirty_metadata_batch);
4150 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4154 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4156 __btrfs_btree_balance_dirty(fs_info, 1);
4159 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4161 __btrfs_btree_balance_dirty(fs_info, 0);
4164 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4165 struct btrfs_key *first_key)
4167 return btree_read_extent_buffer_pages(buf, parent_transid,
4171 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4173 /* cleanup FS via transaction */
4174 btrfs_cleanup_transaction(fs_info);
4176 mutex_lock(&fs_info->cleaner_mutex);
4177 btrfs_run_delayed_iputs(fs_info);
4178 mutex_unlock(&fs_info->cleaner_mutex);
4180 down_write(&fs_info->cleanup_work_sem);
4181 up_write(&fs_info->cleanup_work_sem);
4184 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4186 struct btrfs_ordered_extent *ordered;
4188 spin_lock(&root->ordered_extent_lock);
4190 * This will just short circuit the ordered completion stuff which will
4191 * make sure the ordered extent gets properly cleaned up.
4193 list_for_each_entry(ordered, &root->ordered_extents,
4195 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4196 spin_unlock(&root->ordered_extent_lock);
4199 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4201 struct btrfs_root *root;
4202 struct list_head splice;
4204 INIT_LIST_HEAD(&splice);
4206 spin_lock(&fs_info->ordered_root_lock);
4207 list_splice_init(&fs_info->ordered_roots, &splice);
4208 while (!list_empty(&splice)) {
4209 root = list_first_entry(&splice, struct btrfs_root,
4211 list_move_tail(&root->ordered_root,
4212 &fs_info->ordered_roots);
4214 spin_unlock(&fs_info->ordered_root_lock);
4215 btrfs_destroy_ordered_extents(root);
4218 spin_lock(&fs_info->ordered_root_lock);
4220 spin_unlock(&fs_info->ordered_root_lock);
4223 * We need this here because if we've been flipped read-only we won't
4224 * get sync() from the umount, so we need to make sure any ordered
4225 * extents that haven't had their dirty pages IO start writeout yet
4226 * actually get run and error out properly.
4228 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4231 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4232 struct btrfs_fs_info *fs_info)
4234 struct rb_node *node;
4235 struct btrfs_delayed_ref_root *delayed_refs;
4236 struct btrfs_delayed_ref_node *ref;
4239 delayed_refs = &trans->delayed_refs;
4241 spin_lock(&delayed_refs->lock);
4242 if (atomic_read(&delayed_refs->num_entries) == 0) {
4243 spin_unlock(&delayed_refs->lock);
4244 btrfs_info(fs_info, "delayed_refs has NO entry");
4248 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4249 struct btrfs_delayed_ref_head *head;
4251 bool pin_bytes = false;
4253 head = rb_entry(node, struct btrfs_delayed_ref_head,
4255 if (btrfs_delayed_ref_lock(delayed_refs, head))
4258 spin_lock(&head->lock);
4259 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4260 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4263 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4264 RB_CLEAR_NODE(&ref->ref_node);
4265 if (!list_empty(&ref->add_list))
4266 list_del(&ref->add_list);
4267 atomic_dec(&delayed_refs->num_entries);
4268 btrfs_put_delayed_ref(ref);
4270 if (head->must_insert_reserved)
4272 btrfs_free_delayed_extent_op(head->extent_op);
4273 btrfs_delete_ref_head(delayed_refs, head);
4274 spin_unlock(&head->lock);
4275 spin_unlock(&delayed_refs->lock);
4276 mutex_unlock(&head->mutex);
4279 btrfs_pin_extent(fs_info, head->bytenr,
4280 head->num_bytes, 1);
4281 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4282 btrfs_put_delayed_ref_head(head);
4284 spin_lock(&delayed_refs->lock);
4287 spin_unlock(&delayed_refs->lock);
4292 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4294 struct btrfs_inode *btrfs_inode;
4295 struct list_head splice;
4297 INIT_LIST_HEAD(&splice);
4299 spin_lock(&root->delalloc_lock);
4300 list_splice_init(&root->delalloc_inodes, &splice);
4302 while (!list_empty(&splice)) {
4303 struct inode *inode = NULL;
4304 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4306 __btrfs_del_delalloc_inode(root, btrfs_inode);
4307 spin_unlock(&root->delalloc_lock);
4310 * Make sure we get a live inode and that it'll not disappear
4313 inode = igrab(&btrfs_inode->vfs_inode);
4315 invalidate_inode_pages2(inode->i_mapping);
4318 spin_lock(&root->delalloc_lock);
4320 spin_unlock(&root->delalloc_lock);
4323 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4325 struct btrfs_root *root;
4326 struct list_head splice;
4328 INIT_LIST_HEAD(&splice);
4330 spin_lock(&fs_info->delalloc_root_lock);
4331 list_splice_init(&fs_info->delalloc_roots, &splice);
4332 while (!list_empty(&splice)) {
4333 root = list_first_entry(&splice, struct btrfs_root,
4335 root = btrfs_grab_fs_root(root);
4337 spin_unlock(&fs_info->delalloc_root_lock);
4339 btrfs_destroy_delalloc_inodes(root);
4340 btrfs_put_fs_root(root);
4342 spin_lock(&fs_info->delalloc_root_lock);
4344 spin_unlock(&fs_info->delalloc_root_lock);
4347 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4348 struct extent_io_tree *dirty_pages,
4352 struct extent_buffer *eb;
4357 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4362 clear_extent_bits(dirty_pages, start, end, mark);
4363 while (start <= end) {
4364 eb = find_extent_buffer(fs_info, start);
4365 start += fs_info->nodesize;
4368 wait_on_extent_buffer_writeback(eb);
4370 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4372 clear_extent_buffer_dirty(eb);
4373 free_extent_buffer_stale(eb);
4380 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4381 struct extent_io_tree *pinned_extents)
4383 struct extent_io_tree *unpin;
4389 unpin = pinned_extents;
4392 struct extent_state *cached_state = NULL;
4395 * The btrfs_finish_extent_commit() may get the same range as
4396 * ours between find_first_extent_bit and clear_extent_dirty.
4397 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4398 * the same extent range.
4400 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4401 ret = find_first_extent_bit(unpin, 0, &start, &end,
4402 EXTENT_DIRTY, &cached_state);
4404 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4408 clear_extent_dirty(unpin, start, end, &cached_state);
4409 free_extent_state(cached_state);
4410 btrfs_error_unpin_extent_range(fs_info, start, end);
4411 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4416 if (unpin == &fs_info->freed_extents[0])
4417 unpin = &fs_info->freed_extents[1];
4419 unpin = &fs_info->freed_extents[0];
4427 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4429 struct inode *inode;
4431 inode = cache->io_ctl.inode;
4433 invalidate_inode_pages2(inode->i_mapping);
4434 BTRFS_I(inode)->generation = 0;
4435 cache->io_ctl.inode = NULL;
4438 btrfs_put_block_group(cache);
4441 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4442 struct btrfs_fs_info *fs_info)
4444 struct btrfs_block_group_cache *cache;
4446 spin_lock(&cur_trans->dirty_bgs_lock);
4447 while (!list_empty(&cur_trans->dirty_bgs)) {
4448 cache = list_first_entry(&cur_trans->dirty_bgs,
4449 struct btrfs_block_group_cache,
4452 if (!list_empty(&cache->io_list)) {
4453 spin_unlock(&cur_trans->dirty_bgs_lock);
4454 list_del_init(&cache->io_list);
4455 btrfs_cleanup_bg_io(cache);
4456 spin_lock(&cur_trans->dirty_bgs_lock);
4459 list_del_init(&cache->dirty_list);
4460 spin_lock(&cache->lock);
4461 cache->disk_cache_state = BTRFS_DC_ERROR;
4462 spin_unlock(&cache->lock);
4464 spin_unlock(&cur_trans->dirty_bgs_lock);
4465 btrfs_put_block_group(cache);
4466 btrfs_delayed_refs_rsv_release(fs_info, 1);
4467 spin_lock(&cur_trans->dirty_bgs_lock);
4469 spin_unlock(&cur_trans->dirty_bgs_lock);
4472 * Refer to the definition of io_bgs member for details why it's safe
4473 * to use it without any locking
4475 while (!list_empty(&cur_trans->io_bgs)) {
4476 cache = list_first_entry(&cur_trans->io_bgs,
4477 struct btrfs_block_group_cache,
4480 list_del_init(&cache->io_list);
4481 spin_lock(&cache->lock);
4482 cache->disk_cache_state = BTRFS_DC_ERROR;
4483 spin_unlock(&cache->lock);
4484 btrfs_cleanup_bg_io(cache);
4488 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4489 struct btrfs_fs_info *fs_info)
4491 struct btrfs_device *dev, *tmp;
4493 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4494 ASSERT(list_empty(&cur_trans->dirty_bgs));
4495 ASSERT(list_empty(&cur_trans->io_bgs));
4497 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4499 list_del_init(&dev->post_commit_list);
4502 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4504 cur_trans->state = TRANS_STATE_COMMIT_START;
4505 wake_up(&fs_info->transaction_blocked_wait);
4507 cur_trans->state = TRANS_STATE_UNBLOCKED;
4508 wake_up(&fs_info->transaction_wait);
4510 btrfs_destroy_delayed_inodes(fs_info);
4511 btrfs_assert_delayed_root_empty(fs_info);
4513 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4515 btrfs_destroy_pinned_extent(fs_info,
4516 fs_info->pinned_extents);
4518 cur_trans->state =TRANS_STATE_COMPLETED;
4519 wake_up(&cur_trans->commit_wait);
4522 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4524 struct btrfs_transaction *t;
4526 mutex_lock(&fs_info->transaction_kthread_mutex);
4528 spin_lock(&fs_info->trans_lock);
4529 while (!list_empty(&fs_info->trans_list)) {
4530 t = list_first_entry(&fs_info->trans_list,
4531 struct btrfs_transaction, list);
4532 if (t->state >= TRANS_STATE_COMMIT_START) {
4533 refcount_inc(&t->use_count);
4534 spin_unlock(&fs_info->trans_lock);
4535 btrfs_wait_for_commit(fs_info, t->transid);
4536 btrfs_put_transaction(t);
4537 spin_lock(&fs_info->trans_lock);
4540 if (t == fs_info->running_transaction) {
4541 t->state = TRANS_STATE_COMMIT_DOING;
4542 spin_unlock(&fs_info->trans_lock);
4544 * We wait for 0 num_writers since we don't hold a trans
4545 * handle open currently for this transaction.
4547 wait_event(t->writer_wait,
4548 atomic_read(&t->num_writers) == 0);
4550 spin_unlock(&fs_info->trans_lock);
4552 btrfs_cleanup_one_transaction(t, fs_info);
4554 spin_lock(&fs_info->trans_lock);
4555 if (t == fs_info->running_transaction)
4556 fs_info->running_transaction = NULL;
4557 list_del_init(&t->list);
4558 spin_unlock(&fs_info->trans_lock);
4560 btrfs_put_transaction(t);
4561 trace_btrfs_transaction_commit(fs_info->tree_root);
4562 spin_lock(&fs_info->trans_lock);
4564 spin_unlock(&fs_info->trans_lock);
4565 btrfs_destroy_all_ordered_extents(fs_info);
4566 btrfs_destroy_delayed_inodes(fs_info);
4567 btrfs_assert_delayed_root_empty(fs_info);
4568 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4569 btrfs_destroy_all_delalloc_inodes(fs_info);
4570 mutex_unlock(&fs_info->transaction_kthread_mutex);
4575 static const struct extent_io_ops btree_extent_io_ops = {
4576 /* mandatory callbacks */
4577 .submit_bio_hook = btree_submit_bio_hook,
4578 .readpage_end_io_hook = btree_readpage_end_io_hook,