1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "inode-map.h"
33 #include "check-integrity.h"
34 #include "rcu-string.h"
35 #include "dev-replace.h"
39 #include "compression.h"
40 #include "tree-checker.h"
41 #include "ref-verify.h"
42 #include "block-group.h"
44 #include "space-info.h"
46 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
53 static const struct extent_io_ops btree_extent_io_ops;
54 static void end_workqueue_fn(struct btrfs_work *work);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_fs_info *fs_info);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 struct extent_io_tree *dirty_pages,
62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
72 struct btrfs_end_io_wq {
76 struct btrfs_fs_info *info;
78 enum btrfs_wq_endio_type metadata;
79 struct btrfs_work work;
82 static struct kmem_cache *btrfs_end_io_wq_cache;
84 int __init btrfs_end_io_wq_init(void)
86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq),
91 if (!btrfs_end_io_wq_cache)
96 void __cold btrfs_end_io_wq_exit(void)
98 kmem_cache_destroy(btrfs_end_io_wq_cache);
101 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
103 if (fs_info->csum_shash)
104 crypto_free_shash(fs_info->csum_shash);
108 * async submit bios are used to offload expensive checksumming
109 * onto the worker threads. They checksum file and metadata bios
110 * just before they are sent down the IO stack.
112 struct async_submit_bio {
115 extent_submit_bio_start_t *submit_bio_start;
118 * bio_offset is optional, can be used if the pages in the bio
119 * can't tell us where in the file the bio should go
122 struct btrfs_work work;
127 * Lockdep class keys for extent_buffer->lock's in this root. For a given
128 * eb, the lockdep key is determined by the btrfs_root it belongs to and
129 * the level the eb occupies in the tree.
131 * Different roots are used for different purposes and may nest inside each
132 * other and they require separate keysets. As lockdep keys should be
133 * static, assign keysets according to the purpose of the root as indicated
134 * by btrfs_root->root_key.objectid. This ensures that all special purpose
135 * roots have separate keysets.
137 * Lock-nesting across peer nodes is always done with the immediate parent
138 * node locked thus preventing deadlock. As lockdep doesn't know this, use
139 * subclass to avoid triggering lockdep warning in such cases.
141 * The key is set by the readpage_end_io_hook after the buffer has passed
142 * csum validation but before the pages are unlocked. It is also set by
143 * btrfs_init_new_buffer on freshly allocated blocks.
145 * We also add a check to make sure the highest level of the tree is the
146 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
147 * needs update as well.
149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
150 # if BTRFS_MAX_LEVEL != 8
154 static struct btrfs_lockdep_keyset {
155 u64 id; /* root objectid */
156 const char *name_stem; /* lock name stem */
157 char names[BTRFS_MAX_LEVEL + 1][20];
158 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
159 } btrfs_lockdep_keysets[] = {
160 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
161 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
162 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
163 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
164 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
165 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
166 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
167 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
168 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
169 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
170 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
171 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
172 { .id = 0, .name_stem = "tree" },
175 void __init btrfs_init_lockdep(void)
179 /* initialize lockdep class names */
180 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
181 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
183 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
184 snprintf(ks->names[j], sizeof(ks->names[j]),
185 "btrfs-%s-%02d", ks->name_stem, j);
189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
192 struct btrfs_lockdep_keyset *ks;
194 BUG_ON(level >= ARRAY_SIZE(ks->keys));
196 /* find the matching keyset, id 0 is the default entry */
197 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
198 if (ks->id == objectid)
201 lockdep_set_class_and_name(&eb->lock,
202 &ks->keys[level], ks->names[level]);
208 * extents on the btree inode are pretty simple, there's one extent
209 * that covers the entire device
211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
212 struct page *page, size_t pg_offset,
215 struct extent_map_tree *em_tree = &inode->extent_tree;
216 struct extent_map *em;
219 read_lock(&em_tree->lock);
220 em = lookup_extent_mapping(em_tree, start, len);
222 read_unlock(&em_tree->lock);
225 read_unlock(&em_tree->lock);
227 em = alloc_extent_map();
229 em = ERR_PTR(-ENOMEM);
234 em->block_len = (u64)-1;
237 write_lock(&em_tree->lock);
238 ret = add_extent_mapping(em_tree, em, 0);
239 if (ret == -EEXIST) {
241 em = lookup_extent_mapping(em_tree, start, len);
248 write_unlock(&em_tree->lock);
255 * Compute the csum of a btree block and store the result to provided buffer.
257 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
259 struct btrfs_fs_info *fs_info = buf->fs_info;
260 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
261 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
265 shash->tfm = fs_info->csum_shash;
266 crypto_shash_init(shash);
267 kaddr = page_address(buf->pages[0]);
268 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
269 PAGE_SIZE - BTRFS_CSUM_SIZE);
271 for (i = 1; i < num_pages; i++) {
272 kaddr = page_address(buf->pages[i]);
273 crypto_shash_update(shash, kaddr, PAGE_SIZE);
275 memset(result, 0, BTRFS_CSUM_SIZE);
276 crypto_shash_final(shash, result);
280 * we can't consider a given block up to date unless the transid of the
281 * block matches the transid in the parent node's pointer. This is how we
282 * detect blocks that either didn't get written at all or got written
283 * in the wrong place.
285 static int verify_parent_transid(struct extent_io_tree *io_tree,
286 struct extent_buffer *eb, u64 parent_transid,
289 struct extent_state *cached_state = NULL;
291 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
293 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
300 btrfs_tree_read_lock(eb);
301 btrfs_set_lock_blocking_read(eb);
304 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
306 if (extent_buffer_uptodate(eb) &&
307 btrfs_header_generation(eb) == parent_transid) {
311 btrfs_err_rl(eb->fs_info,
312 "parent transid verify failed on %llu wanted %llu found %llu",
314 parent_transid, btrfs_header_generation(eb));
318 * Things reading via commit roots that don't have normal protection,
319 * like send, can have a really old block in cache that may point at a
320 * block that has been freed and re-allocated. So don't clear uptodate
321 * if we find an eb that is under IO (dirty/writeback) because we could
322 * end up reading in the stale data and then writing it back out and
323 * making everybody very sad.
325 if (!extent_buffer_under_io(eb))
326 clear_extent_buffer_uptodate(eb);
328 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
331 btrfs_tree_read_unlock_blocking(eb);
335 static bool btrfs_supported_super_csum(u16 csum_type)
338 case BTRFS_CSUM_TYPE_CRC32:
339 case BTRFS_CSUM_TYPE_XXHASH:
340 case BTRFS_CSUM_TYPE_SHA256:
341 case BTRFS_CSUM_TYPE_BLAKE2:
349 * Return 0 if the superblock checksum type matches the checksum value of that
350 * algorithm. Pass the raw disk superblock data.
352 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
355 struct btrfs_super_block *disk_sb =
356 (struct btrfs_super_block *)raw_disk_sb;
357 char result[BTRFS_CSUM_SIZE];
358 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
360 shash->tfm = fs_info->csum_shash;
361 crypto_shash_init(shash);
364 * The super_block structure does not span the whole
365 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
366 * filled with zeros and is included in the checksum.
368 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
369 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
370 crypto_shash_final(shash, result);
372 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
378 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
379 struct btrfs_key *first_key, u64 parent_transid)
381 struct btrfs_fs_info *fs_info = eb->fs_info;
383 struct btrfs_key found_key;
386 found_level = btrfs_header_level(eb);
387 if (found_level != level) {
388 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
389 KERN_ERR "BTRFS: tree level check failed\n");
391 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
392 eb->start, level, found_level);
400 * For live tree block (new tree blocks in current transaction),
401 * we need proper lock context to avoid race, which is impossible here.
402 * So we only checks tree blocks which is read from disk, whose
403 * generation <= fs_info->last_trans_committed.
405 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
408 /* We have @first_key, so this @eb must have at least one item */
409 if (btrfs_header_nritems(eb) == 0) {
411 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
413 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
418 btrfs_node_key_to_cpu(eb, &found_key, 0);
420 btrfs_item_key_to_cpu(eb, &found_key, 0);
421 ret = btrfs_comp_cpu_keys(first_key, &found_key);
424 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
425 KERN_ERR "BTRFS: tree first key check failed\n");
427 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
428 eb->start, parent_transid, first_key->objectid,
429 first_key->type, first_key->offset,
430 found_key.objectid, found_key.type,
437 * helper to read a given tree block, doing retries as required when
438 * the checksums don't match and we have alternate mirrors to try.
440 * @parent_transid: expected transid, skip check if 0
441 * @level: expected level, mandatory check
442 * @first_key: expected key of first slot, skip check if NULL
444 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
445 u64 parent_transid, int level,
446 struct btrfs_key *first_key)
448 struct btrfs_fs_info *fs_info = eb->fs_info;
449 struct extent_io_tree *io_tree;
454 int failed_mirror = 0;
456 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
458 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
459 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
461 if (verify_parent_transid(io_tree, eb,
464 else if (btrfs_verify_level_key(eb, level,
465 first_key, parent_transid))
471 num_copies = btrfs_num_copies(fs_info,
476 if (!failed_mirror) {
478 failed_mirror = eb->read_mirror;
482 if (mirror_num == failed_mirror)
485 if (mirror_num > num_copies)
489 if (failed && !ret && failed_mirror)
490 btrfs_repair_eb_io_failure(eb, failed_mirror);
496 * checksum a dirty tree block before IO. This has extra checks to make sure
497 * we only fill in the checksum field in the first page of a multi-page block
500 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
502 u64 start = page_offset(page);
504 u8 result[BTRFS_CSUM_SIZE];
505 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
506 struct extent_buffer *eb;
509 eb = (struct extent_buffer *)page->private;
510 if (page != eb->pages[0])
513 found_start = btrfs_header_bytenr(eb);
515 * Please do not consolidate these warnings into a single if.
516 * It is useful to know what went wrong.
518 if (WARN_ON(found_start != start))
520 if (WARN_ON(!PageUptodate(page)))
523 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
524 offsetof(struct btrfs_header, fsid),
525 BTRFS_FSID_SIZE) == 0);
527 csum_tree_block(eb, result);
529 if (btrfs_header_level(eb))
530 ret = btrfs_check_node(eb);
532 ret = btrfs_check_leaf_full(eb);
535 btrfs_print_tree(eb, 0);
537 "block=%llu write time tree block corruption detected",
539 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
542 write_extent_buffer(eb, result, 0, csum_size);
547 static int check_tree_block_fsid(struct extent_buffer *eb)
549 struct btrfs_fs_info *fs_info = eb->fs_info;
550 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
551 u8 fsid[BTRFS_FSID_SIZE];
554 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
560 * Checking the incompat flag is only valid for the current
561 * fs. For seed devices it's forbidden to have their uuid
562 * changed so reading ->fsid in this case is fine
564 if (fs_devices == fs_info->fs_devices &&
565 btrfs_fs_incompat(fs_info, METADATA_UUID))
566 metadata_uuid = fs_devices->metadata_uuid;
568 metadata_uuid = fs_devices->fsid;
570 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
574 fs_devices = fs_devices->seed;
579 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
580 u64 phy_offset, struct page *page,
581 u64 start, u64 end, int mirror)
585 struct extent_buffer *eb;
586 struct btrfs_fs_info *fs_info;
589 u8 result[BTRFS_CSUM_SIZE];
595 eb = (struct extent_buffer *)page->private;
596 fs_info = eb->fs_info;
597 csum_size = btrfs_super_csum_size(fs_info->super_copy);
599 /* the pending IO might have been the only thing that kept this buffer
600 * in memory. Make sure we have a ref for all this other checks
602 atomic_inc(&eb->refs);
604 reads_done = atomic_dec_and_test(&eb->io_pages);
608 eb->read_mirror = mirror;
609 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
614 found_start = btrfs_header_bytenr(eb);
615 if (found_start != eb->start) {
616 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
617 eb->start, found_start);
621 if (check_tree_block_fsid(eb)) {
622 btrfs_err_rl(fs_info, "bad fsid on block %llu",
627 found_level = btrfs_header_level(eb);
628 if (found_level >= BTRFS_MAX_LEVEL) {
629 btrfs_err(fs_info, "bad tree block level %d on %llu",
630 (int)btrfs_header_level(eb), eb->start);
635 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
638 csum_tree_block(eb, result);
640 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
644 memcpy(&found, result, csum_size);
646 read_extent_buffer(eb, &val, 0, csum_size);
647 btrfs_warn_rl(fs_info,
648 "%s checksum verify failed on %llu wanted %x found %x level %d",
649 fs_info->sb->s_id, eb->start,
650 val, found, btrfs_header_level(eb));
656 * If this is a leaf block and it is corrupt, set the corrupt bit so
657 * that we don't try and read the other copies of this block, just
660 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
661 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
665 if (found_level > 0 && btrfs_check_node(eb))
669 set_extent_buffer_uptodate(eb);
672 "block=%llu read time tree block corruption detected",
676 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
677 btree_readahead_hook(eb, ret);
681 * our io error hook is going to dec the io pages
682 * again, we have to make sure it has something
685 atomic_inc(&eb->io_pages);
686 clear_extent_buffer_uptodate(eb);
688 free_extent_buffer(eb);
693 static void end_workqueue_bio(struct bio *bio)
695 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
696 struct btrfs_fs_info *fs_info;
697 struct btrfs_workqueue *wq;
699 fs_info = end_io_wq->info;
700 end_io_wq->status = bio->bi_status;
702 if (bio_op(bio) == REQ_OP_WRITE) {
703 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
704 wq = fs_info->endio_meta_write_workers;
705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
706 wq = fs_info->endio_freespace_worker;
707 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
708 wq = fs_info->endio_raid56_workers;
710 wq = fs_info->endio_write_workers;
712 if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
713 wq = fs_info->endio_repair_workers;
714 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
715 wq = fs_info->endio_raid56_workers;
716 else if (end_io_wq->metadata)
717 wq = fs_info->endio_meta_workers;
719 wq = fs_info->endio_workers;
722 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
723 btrfs_queue_work(wq, &end_io_wq->work);
726 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
727 enum btrfs_wq_endio_type metadata)
729 struct btrfs_end_io_wq *end_io_wq;
731 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
733 return BLK_STS_RESOURCE;
735 end_io_wq->private = bio->bi_private;
736 end_io_wq->end_io = bio->bi_end_io;
737 end_io_wq->info = info;
738 end_io_wq->status = 0;
739 end_io_wq->bio = bio;
740 end_io_wq->metadata = metadata;
742 bio->bi_private = end_io_wq;
743 bio->bi_end_io = end_workqueue_bio;
747 static void run_one_async_start(struct btrfs_work *work)
749 struct async_submit_bio *async;
752 async = container_of(work, struct async_submit_bio, work);
753 ret = async->submit_bio_start(async->private_data, async->bio,
760 * In order to insert checksums into the metadata in large chunks, we wait
761 * until bio submission time. All the pages in the bio are checksummed and
762 * sums are attached onto the ordered extent record.
764 * At IO completion time the csums attached on the ordered extent record are
765 * inserted into the tree.
767 static void run_one_async_done(struct btrfs_work *work)
769 struct async_submit_bio *async;
773 async = container_of(work, struct async_submit_bio, work);
774 inode = async->private_data;
776 /* If an error occurred we just want to clean up the bio and move on */
778 async->bio->bi_status = async->status;
779 bio_endio(async->bio);
784 * All of the bios that pass through here are from async helpers.
785 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
786 * This changes nothing when cgroups aren't in use.
788 async->bio->bi_opf |= REQ_CGROUP_PUNT;
789 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
791 async->bio->bi_status = ret;
792 bio_endio(async->bio);
796 static void run_one_async_free(struct btrfs_work *work)
798 struct async_submit_bio *async;
800 async = container_of(work, struct async_submit_bio, work);
804 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
805 int mirror_num, unsigned long bio_flags,
806 u64 bio_offset, void *private_data,
807 extent_submit_bio_start_t *submit_bio_start)
809 struct async_submit_bio *async;
811 async = kmalloc(sizeof(*async), GFP_NOFS);
813 return BLK_STS_RESOURCE;
815 async->private_data = private_data;
817 async->mirror_num = mirror_num;
818 async->submit_bio_start = submit_bio_start;
820 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
823 async->bio_offset = bio_offset;
827 if (op_is_sync(bio->bi_opf))
828 btrfs_set_work_high_priority(&async->work);
830 btrfs_queue_work(fs_info->workers, &async->work);
834 static blk_status_t btree_csum_one_bio(struct bio *bio)
836 struct bio_vec *bvec;
837 struct btrfs_root *root;
839 struct bvec_iter_all iter_all;
841 ASSERT(!bio_flagged(bio, BIO_CLONED));
842 bio_for_each_segment_all(bvec, bio, iter_all) {
843 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
844 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
849 return errno_to_blk_status(ret);
852 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
856 * when we're called for a write, we're already in the async
857 * submission context. Just jump into btrfs_map_bio
859 return btree_csum_one_bio(bio);
862 static int check_async_write(struct btrfs_fs_info *fs_info,
863 struct btrfs_inode *bi)
865 if (atomic_read(&bi->sync_writers))
867 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
872 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
874 unsigned long bio_flags)
876 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
877 int async = check_async_write(fs_info, BTRFS_I(inode));
880 if (bio_op(bio) != REQ_OP_WRITE) {
882 * called for a read, do the setup so that checksum validation
883 * can happen in the async kernel threads
885 ret = btrfs_bio_wq_end_io(fs_info, bio,
886 BTRFS_WQ_ENDIO_METADATA);
889 ret = btrfs_map_bio(fs_info, bio, mirror_num);
891 ret = btree_csum_one_bio(bio);
894 ret = btrfs_map_bio(fs_info, bio, mirror_num);
897 * kthread helpers are used to submit writes so that
898 * checksumming can happen in parallel across all CPUs
900 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
901 0, inode, btree_submit_bio_start);
909 bio->bi_status = ret;
914 #ifdef CONFIG_MIGRATION
915 static int btree_migratepage(struct address_space *mapping,
916 struct page *newpage, struct page *page,
917 enum migrate_mode mode)
920 * we can't safely write a btree page from here,
921 * we haven't done the locking hook
926 * Buffers may be managed in a filesystem specific way.
927 * We must have no buffers or drop them.
929 if (page_has_private(page) &&
930 !try_to_release_page(page, GFP_KERNEL))
932 return migrate_page(mapping, newpage, page, mode);
937 static int btree_writepages(struct address_space *mapping,
938 struct writeback_control *wbc)
940 struct btrfs_fs_info *fs_info;
943 if (wbc->sync_mode == WB_SYNC_NONE) {
945 if (wbc->for_kupdate)
948 fs_info = BTRFS_I(mapping->host)->root->fs_info;
949 /* this is a bit racy, but that's ok */
950 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
951 BTRFS_DIRTY_METADATA_THRESH,
952 fs_info->dirty_metadata_batch);
956 return btree_write_cache_pages(mapping, wbc);
959 static int btree_readpage(struct file *file, struct page *page)
961 return extent_read_full_page(page, btree_get_extent, 0);
964 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
966 if (PageWriteback(page) || PageDirty(page))
969 return try_release_extent_buffer(page);
972 static void btree_invalidatepage(struct page *page, unsigned int offset,
975 struct extent_io_tree *tree;
976 tree = &BTRFS_I(page->mapping->host)->io_tree;
977 extent_invalidatepage(tree, page, offset);
978 btree_releasepage(page, GFP_NOFS);
979 if (PagePrivate(page)) {
980 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
981 "page private not zero on page %llu",
982 (unsigned long long)page_offset(page));
983 ClearPagePrivate(page);
984 set_page_private(page, 0);
989 static int btree_set_page_dirty(struct page *page)
992 struct extent_buffer *eb;
994 BUG_ON(!PagePrivate(page));
995 eb = (struct extent_buffer *)page->private;
997 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
998 BUG_ON(!atomic_read(&eb->refs));
999 btrfs_assert_tree_locked(eb);
1001 return __set_page_dirty_nobuffers(page);
1004 static const struct address_space_operations btree_aops = {
1005 .readpage = btree_readpage,
1006 .writepages = btree_writepages,
1007 .releasepage = btree_releasepage,
1008 .invalidatepage = btree_invalidatepage,
1009 #ifdef CONFIG_MIGRATION
1010 .migratepage = btree_migratepage,
1012 .set_page_dirty = btree_set_page_dirty,
1015 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1017 struct extent_buffer *buf = NULL;
1020 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1024 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1026 free_extent_buffer_stale(buf);
1028 free_extent_buffer(buf);
1031 struct extent_buffer *btrfs_find_create_tree_block(
1032 struct btrfs_fs_info *fs_info,
1035 if (btrfs_is_testing(fs_info))
1036 return alloc_test_extent_buffer(fs_info, bytenr);
1037 return alloc_extent_buffer(fs_info, bytenr);
1041 * Read tree block at logical address @bytenr and do variant basic but critical
1044 * @parent_transid: expected transid of this tree block, skip check if 0
1045 * @level: expected level, mandatory check
1046 * @first_key: expected key in slot 0, skip check if NULL
1048 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1049 u64 parent_transid, int level,
1050 struct btrfs_key *first_key)
1052 struct extent_buffer *buf = NULL;
1055 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1059 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1062 free_extent_buffer_stale(buf);
1063 return ERR_PTR(ret);
1069 void btrfs_clean_tree_block(struct extent_buffer *buf)
1071 struct btrfs_fs_info *fs_info = buf->fs_info;
1072 if (btrfs_header_generation(buf) ==
1073 fs_info->running_transaction->transid) {
1074 btrfs_assert_tree_locked(buf);
1076 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1077 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1079 fs_info->dirty_metadata_batch);
1080 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1081 btrfs_set_lock_blocking_write(buf);
1082 clear_extent_buffer_dirty(buf);
1087 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1090 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1091 root->fs_info = fs_info;
1093 root->commit_root = NULL;
1095 root->orphan_cleanup_state = 0;
1097 root->last_trans = 0;
1098 root->highest_objectid = 0;
1099 root->nr_delalloc_inodes = 0;
1100 root->nr_ordered_extents = 0;
1101 root->inode_tree = RB_ROOT;
1102 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1103 root->block_rsv = NULL;
1105 INIT_LIST_HEAD(&root->dirty_list);
1106 INIT_LIST_HEAD(&root->root_list);
1107 INIT_LIST_HEAD(&root->delalloc_inodes);
1108 INIT_LIST_HEAD(&root->delalloc_root);
1109 INIT_LIST_HEAD(&root->ordered_extents);
1110 INIT_LIST_HEAD(&root->ordered_root);
1111 INIT_LIST_HEAD(&root->reloc_dirty_list);
1112 INIT_LIST_HEAD(&root->logged_list[0]);
1113 INIT_LIST_HEAD(&root->logged_list[1]);
1114 spin_lock_init(&root->inode_lock);
1115 spin_lock_init(&root->delalloc_lock);
1116 spin_lock_init(&root->ordered_extent_lock);
1117 spin_lock_init(&root->accounting_lock);
1118 spin_lock_init(&root->log_extents_lock[0]);
1119 spin_lock_init(&root->log_extents_lock[1]);
1120 spin_lock_init(&root->qgroup_meta_rsv_lock);
1121 mutex_init(&root->objectid_mutex);
1122 mutex_init(&root->log_mutex);
1123 mutex_init(&root->ordered_extent_mutex);
1124 mutex_init(&root->delalloc_mutex);
1125 init_waitqueue_head(&root->log_writer_wait);
1126 init_waitqueue_head(&root->log_commit_wait[0]);
1127 init_waitqueue_head(&root->log_commit_wait[1]);
1128 INIT_LIST_HEAD(&root->log_ctxs[0]);
1129 INIT_LIST_HEAD(&root->log_ctxs[1]);
1130 atomic_set(&root->log_commit[0], 0);
1131 atomic_set(&root->log_commit[1], 0);
1132 atomic_set(&root->log_writers, 0);
1133 atomic_set(&root->log_batch, 0);
1134 refcount_set(&root->refs, 1);
1135 atomic_set(&root->snapshot_force_cow, 0);
1136 atomic_set(&root->nr_swapfiles, 0);
1137 root->log_transid = 0;
1138 root->log_transid_committed = -1;
1139 root->last_log_commit = 0;
1141 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1142 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1144 memset(&root->root_key, 0, sizeof(root->root_key));
1145 memset(&root->root_item, 0, sizeof(root->root_item));
1146 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1148 root->defrag_trans_start = fs_info->generation;
1150 root->defrag_trans_start = 0;
1151 root->root_key.objectid = objectid;
1154 spin_lock_init(&root->root_item_lock);
1155 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1156 #ifdef CONFIG_BTRFS_DEBUG
1157 INIT_LIST_HEAD(&root->leak_list);
1158 spin_lock(&fs_info->fs_roots_radix_lock);
1159 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1160 spin_unlock(&fs_info->fs_roots_radix_lock);
1164 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1165 u64 objectid, gfp_t flags)
1167 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1169 __setup_root(root, fs_info, objectid);
1173 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1174 /* Should only be used by the testing infrastructure */
1175 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1177 struct btrfs_root *root;
1180 return ERR_PTR(-EINVAL);
1182 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1184 return ERR_PTR(-ENOMEM);
1186 /* We don't use the stripesize in selftest, set it as sectorsize */
1187 root->alloc_bytenr = 0;
1193 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1196 struct btrfs_fs_info *fs_info = trans->fs_info;
1197 struct extent_buffer *leaf;
1198 struct btrfs_root *tree_root = fs_info->tree_root;
1199 struct btrfs_root *root;
1200 struct btrfs_key key;
1201 unsigned int nofs_flag;
1205 * We're holding a transaction handle, so use a NOFS memory allocation
1206 * context to avoid deadlock if reclaim happens.
1208 nofs_flag = memalloc_nofs_save();
1209 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1210 memalloc_nofs_restore(nofs_flag);
1212 return ERR_PTR(-ENOMEM);
1214 root->root_key.objectid = objectid;
1215 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1216 root->root_key.offset = 0;
1218 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1220 ret = PTR_ERR(leaf);
1226 btrfs_mark_buffer_dirty(leaf);
1228 root->commit_root = btrfs_root_node(root);
1229 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1231 root->root_item.flags = 0;
1232 root->root_item.byte_limit = 0;
1233 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1234 btrfs_set_root_generation(&root->root_item, trans->transid);
1235 btrfs_set_root_level(&root->root_item, 0);
1236 btrfs_set_root_refs(&root->root_item, 1);
1237 btrfs_set_root_used(&root->root_item, leaf->len);
1238 btrfs_set_root_last_snapshot(&root->root_item, 0);
1239 btrfs_set_root_dirid(&root->root_item, 0);
1240 if (is_fstree(objectid))
1241 generate_random_guid(root->root_item.uuid);
1243 export_guid(root->root_item.uuid, &guid_null);
1244 root->root_item.drop_level = 0;
1246 key.objectid = objectid;
1247 key.type = BTRFS_ROOT_ITEM_KEY;
1249 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1253 btrfs_tree_unlock(leaf);
1259 btrfs_tree_unlock(leaf);
1260 btrfs_put_root(root);
1262 return ERR_PTR(ret);
1265 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1266 struct btrfs_fs_info *fs_info)
1268 struct btrfs_root *root;
1269 struct extent_buffer *leaf;
1271 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1273 return ERR_PTR(-ENOMEM);
1275 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1276 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1277 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1280 * DON'T set REF_COWS for log trees
1282 * log trees do not get reference counted because they go away
1283 * before a real commit is actually done. They do store pointers
1284 * to file data extents, and those reference counts still get
1285 * updated (along with back refs to the log tree).
1288 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1291 btrfs_put_root(root);
1292 return ERR_CAST(leaf);
1297 btrfs_mark_buffer_dirty(root->node);
1298 btrfs_tree_unlock(root->node);
1302 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1303 struct btrfs_fs_info *fs_info)
1305 struct btrfs_root *log_root;
1307 log_root = alloc_log_tree(trans, fs_info);
1308 if (IS_ERR(log_root))
1309 return PTR_ERR(log_root);
1310 WARN_ON(fs_info->log_root_tree);
1311 fs_info->log_root_tree = log_root;
1315 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1316 struct btrfs_root *root)
1318 struct btrfs_fs_info *fs_info = root->fs_info;
1319 struct btrfs_root *log_root;
1320 struct btrfs_inode_item *inode_item;
1322 log_root = alloc_log_tree(trans, fs_info);
1323 if (IS_ERR(log_root))
1324 return PTR_ERR(log_root);
1326 log_root->last_trans = trans->transid;
1327 log_root->root_key.offset = root->root_key.objectid;
1329 inode_item = &log_root->root_item.inode;
1330 btrfs_set_stack_inode_generation(inode_item, 1);
1331 btrfs_set_stack_inode_size(inode_item, 3);
1332 btrfs_set_stack_inode_nlink(inode_item, 1);
1333 btrfs_set_stack_inode_nbytes(inode_item,
1335 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1337 btrfs_set_root_node(&log_root->root_item, log_root->node);
1339 WARN_ON(root->log_root);
1340 root->log_root = log_root;
1341 root->log_transid = 0;
1342 root->log_transid_committed = -1;
1343 root->last_log_commit = 0;
1347 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1348 struct btrfs_key *key)
1350 struct btrfs_root *root;
1351 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1352 struct btrfs_path *path;
1357 path = btrfs_alloc_path();
1359 return ERR_PTR(-ENOMEM);
1361 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1367 ret = btrfs_find_root(tree_root, key, path,
1368 &root->root_item, &root->root_key);
1375 generation = btrfs_root_generation(&root->root_item);
1376 level = btrfs_root_level(&root->root_item);
1377 root->node = read_tree_block(fs_info,
1378 btrfs_root_bytenr(&root->root_item),
1379 generation, level, NULL);
1380 if (IS_ERR(root->node)) {
1381 ret = PTR_ERR(root->node);
1384 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1388 root->commit_root = btrfs_root_node(root);
1390 btrfs_free_path(path);
1394 btrfs_put_root(root);
1396 root = ERR_PTR(ret);
1400 static int btrfs_init_fs_root(struct btrfs_root *root)
1403 unsigned int nofs_flag;
1405 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1406 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1408 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1414 * We might be called under a transaction (e.g. indirect backref
1415 * resolution) which could deadlock if it triggers memory reclaim
1417 nofs_flag = memalloc_nofs_save();
1418 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1419 memalloc_nofs_restore(nofs_flag);
1423 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1424 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1425 btrfs_check_and_init_root_item(&root->root_item);
1428 btrfs_init_free_ino_ctl(root);
1429 spin_lock_init(&root->ino_cache_lock);
1430 init_waitqueue_head(&root->ino_cache_wait);
1432 ret = get_anon_bdev(&root->anon_dev);
1436 mutex_lock(&root->objectid_mutex);
1437 ret = btrfs_find_highest_objectid(root,
1438 &root->highest_objectid);
1440 mutex_unlock(&root->objectid_mutex);
1444 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1446 mutex_unlock(&root->objectid_mutex);
1450 /* The caller is responsible to call btrfs_free_fs_root */
1454 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1457 struct btrfs_root *root;
1459 spin_lock(&fs_info->fs_roots_radix_lock);
1460 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1461 (unsigned long)root_id);
1463 root = btrfs_grab_root(root);
1464 spin_unlock(&fs_info->fs_roots_radix_lock);
1468 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1469 struct btrfs_root *root)
1473 ret = radix_tree_preload(GFP_NOFS);
1477 spin_lock(&fs_info->fs_roots_radix_lock);
1478 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1479 (unsigned long)root->root_key.objectid,
1482 btrfs_grab_root(root);
1483 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1485 spin_unlock(&fs_info->fs_roots_radix_lock);
1486 radix_tree_preload_end();
1491 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1493 #ifdef CONFIG_BTRFS_DEBUG
1494 struct btrfs_root *root;
1496 while (!list_empty(&fs_info->allocated_roots)) {
1497 root = list_first_entry(&fs_info->allocated_roots,
1498 struct btrfs_root, leak_list);
1499 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1500 root->root_key.objectid, root->root_key.offset,
1501 refcount_read(&root->refs));
1502 while (refcount_read(&root->refs) > 1)
1503 btrfs_put_root(root);
1504 btrfs_put_root(root);
1509 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1511 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1512 percpu_counter_destroy(&fs_info->delalloc_bytes);
1513 percpu_counter_destroy(&fs_info->dio_bytes);
1514 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1515 btrfs_free_csum_hash(fs_info);
1516 btrfs_free_stripe_hash_table(fs_info);
1517 btrfs_free_ref_cache(fs_info);
1518 kfree(fs_info->balance_ctl);
1519 kfree(fs_info->delayed_root);
1520 btrfs_put_root(fs_info->extent_root);
1521 btrfs_put_root(fs_info->tree_root);
1522 btrfs_put_root(fs_info->chunk_root);
1523 btrfs_put_root(fs_info->dev_root);
1524 btrfs_put_root(fs_info->csum_root);
1525 btrfs_put_root(fs_info->quota_root);
1526 btrfs_put_root(fs_info->uuid_root);
1527 btrfs_put_root(fs_info->free_space_root);
1528 btrfs_put_root(fs_info->fs_root);
1529 btrfs_check_leaked_roots(fs_info);
1530 btrfs_extent_buffer_leak_debug_check(fs_info);
1531 kfree(fs_info->super_copy);
1532 kfree(fs_info->super_for_commit);
1537 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1538 struct btrfs_key *location,
1541 struct btrfs_root *root;
1542 struct btrfs_path *path;
1543 struct btrfs_key key;
1546 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1547 return btrfs_grab_root(fs_info->tree_root);
1548 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1549 return btrfs_grab_root(fs_info->extent_root);
1550 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1551 return btrfs_grab_root(fs_info->chunk_root);
1552 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1553 return btrfs_grab_root(fs_info->dev_root);
1554 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1555 return btrfs_grab_root(fs_info->csum_root);
1556 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1557 return btrfs_grab_root(fs_info->quota_root) ?
1558 fs_info->quota_root : ERR_PTR(-ENOENT);
1559 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1560 return btrfs_grab_root(fs_info->uuid_root) ?
1561 fs_info->uuid_root : ERR_PTR(-ENOENT);
1562 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1563 return btrfs_grab_root(fs_info->free_space_root) ?
1564 fs_info->free_space_root : ERR_PTR(-ENOENT);
1566 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1568 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1569 btrfs_put_root(root);
1570 return ERR_PTR(-ENOENT);
1575 root = btrfs_read_tree_root(fs_info->tree_root, location);
1579 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1584 ret = btrfs_init_fs_root(root);
1588 path = btrfs_alloc_path();
1593 key.objectid = BTRFS_ORPHAN_OBJECTID;
1594 key.type = BTRFS_ORPHAN_ITEM_KEY;
1595 key.offset = location->objectid;
1597 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1598 btrfs_free_path(path);
1602 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1604 ret = btrfs_insert_fs_root(fs_info, root);
1606 btrfs_put_root(root);
1613 btrfs_put_root(root);
1614 return ERR_PTR(ret);
1617 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1619 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1621 struct btrfs_device *device;
1622 struct backing_dev_info *bdi;
1625 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1628 bdi = device->bdev->bd_bdi;
1629 if (bdi_congested(bdi, bdi_bits)) {
1639 * called by the kthread helper functions to finally call the bio end_io
1640 * functions. This is where read checksum verification actually happens
1642 static void end_workqueue_fn(struct btrfs_work *work)
1645 struct btrfs_end_io_wq *end_io_wq;
1647 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1648 bio = end_io_wq->bio;
1650 bio->bi_status = end_io_wq->status;
1651 bio->bi_private = end_io_wq->private;
1652 bio->bi_end_io = end_io_wq->end_io;
1654 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1657 static int cleaner_kthread(void *arg)
1659 struct btrfs_root *root = arg;
1660 struct btrfs_fs_info *fs_info = root->fs_info;
1666 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1668 /* Make the cleaner go to sleep early. */
1669 if (btrfs_need_cleaner_sleep(fs_info))
1673 * Do not do anything if we might cause open_ctree() to block
1674 * before we have finished mounting the filesystem.
1676 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1679 if (!mutex_trylock(&fs_info->cleaner_mutex))
1683 * Avoid the problem that we change the status of the fs
1684 * during the above check and trylock.
1686 if (btrfs_need_cleaner_sleep(fs_info)) {
1687 mutex_unlock(&fs_info->cleaner_mutex);
1691 btrfs_run_delayed_iputs(fs_info);
1693 again = btrfs_clean_one_deleted_snapshot(root);
1694 mutex_unlock(&fs_info->cleaner_mutex);
1697 * The defragger has dealt with the R/O remount and umount,
1698 * needn't do anything special here.
1700 btrfs_run_defrag_inodes(fs_info);
1703 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1704 * with relocation (btrfs_relocate_chunk) and relocation
1705 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1706 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1707 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1708 * unused block groups.
1710 btrfs_delete_unused_bgs(fs_info);
1712 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1713 if (kthread_should_park())
1715 if (kthread_should_stop())
1718 set_current_state(TASK_INTERRUPTIBLE);
1720 __set_current_state(TASK_RUNNING);
1725 static int transaction_kthread(void *arg)
1727 struct btrfs_root *root = arg;
1728 struct btrfs_fs_info *fs_info = root->fs_info;
1729 struct btrfs_trans_handle *trans;
1730 struct btrfs_transaction *cur;
1733 unsigned long delay;
1737 cannot_commit = false;
1738 delay = HZ * fs_info->commit_interval;
1739 mutex_lock(&fs_info->transaction_kthread_mutex);
1741 spin_lock(&fs_info->trans_lock);
1742 cur = fs_info->running_transaction;
1744 spin_unlock(&fs_info->trans_lock);
1748 now = ktime_get_seconds();
1749 if (cur->state < TRANS_STATE_COMMIT_START &&
1750 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1751 (now < cur->start_time ||
1752 now - cur->start_time < fs_info->commit_interval)) {
1753 spin_unlock(&fs_info->trans_lock);
1757 transid = cur->transid;
1758 spin_unlock(&fs_info->trans_lock);
1760 /* If the file system is aborted, this will always fail. */
1761 trans = btrfs_attach_transaction(root);
1762 if (IS_ERR(trans)) {
1763 if (PTR_ERR(trans) != -ENOENT)
1764 cannot_commit = true;
1767 if (transid == trans->transid) {
1768 btrfs_commit_transaction(trans);
1770 btrfs_end_transaction(trans);
1773 wake_up_process(fs_info->cleaner_kthread);
1774 mutex_unlock(&fs_info->transaction_kthread_mutex);
1776 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1777 &fs_info->fs_state)))
1778 btrfs_cleanup_transaction(fs_info);
1779 if (!kthread_should_stop() &&
1780 (!btrfs_transaction_blocked(fs_info) ||
1782 schedule_timeout_interruptible(delay);
1783 } while (!kthread_should_stop());
1788 * This will find the highest generation in the array of root backups. The
1789 * index of the highest array is returned, or -EINVAL if we can't find
1792 * We check to make sure the array is valid by comparing the
1793 * generation of the latest root in the array with the generation
1794 * in the super block. If they don't match we pitch it.
1796 static int find_newest_super_backup(struct btrfs_fs_info *info)
1798 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1800 struct btrfs_root_backup *root_backup;
1803 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1804 root_backup = info->super_copy->super_roots + i;
1805 cur = btrfs_backup_tree_root_gen(root_backup);
1806 if (cur == newest_gen)
1814 * copy all the root pointers into the super backup array.
1815 * this will bump the backup pointer by one when it is
1818 static void backup_super_roots(struct btrfs_fs_info *info)
1820 const int next_backup = info->backup_root_index;
1821 struct btrfs_root_backup *root_backup;
1823 root_backup = info->super_for_commit->super_roots + next_backup;
1826 * make sure all of our padding and empty slots get zero filled
1827 * regardless of which ones we use today
1829 memset(root_backup, 0, sizeof(*root_backup));
1831 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1833 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1834 btrfs_set_backup_tree_root_gen(root_backup,
1835 btrfs_header_generation(info->tree_root->node));
1837 btrfs_set_backup_tree_root_level(root_backup,
1838 btrfs_header_level(info->tree_root->node));
1840 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1841 btrfs_set_backup_chunk_root_gen(root_backup,
1842 btrfs_header_generation(info->chunk_root->node));
1843 btrfs_set_backup_chunk_root_level(root_backup,
1844 btrfs_header_level(info->chunk_root->node));
1846 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1847 btrfs_set_backup_extent_root_gen(root_backup,
1848 btrfs_header_generation(info->extent_root->node));
1849 btrfs_set_backup_extent_root_level(root_backup,
1850 btrfs_header_level(info->extent_root->node));
1853 * we might commit during log recovery, which happens before we set
1854 * the fs_root. Make sure it is valid before we fill it in.
1856 if (info->fs_root && info->fs_root->node) {
1857 btrfs_set_backup_fs_root(root_backup,
1858 info->fs_root->node->start);
1859 btrfs_set_backup_fs_root_gen(root_backup,
1860 btrfs_header_generation(info->fs_root->node));
1861 btrfs_set_backup_fs_root_level(root_backup,
1862 btrfs_header_level(info->fs_root->node));
1865 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1866 btrfs_set_backup_dev_root_gen(root_backup,
1867 btrfs_header_generation(info->dev_root->node));
1868 btrfs_set_backup_dev_root_level(root_backup,
1869 btrfs_header_level(info->dev_root->node));
1871 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1872 btrfs_set_backup_csum_root_gen(root_backup,
1873 btrfs_header_generation(info->csum_root->node));
1874 btrfs_set_backup_csum_root_level(root_backup,
1875 btrfs_header_level(info->csum_root->node));
1877 btrfs_set_backup_total_bytes(root_backup,
1878 btrfs_super_total_bytes(info->super_copy));
1879 btrfs_set_backup_bytes_used(root_backup,
1880 btrfs_super_bytes_used(info->super_copy));
1881 btrfs_set_backup_num_devices(root_backup,
1882 btrfs_super_num_devices(info->super_copy));
1885 * if we don't copy this out to the super_copy, it won't get remembered
1886 * for the next commit
1888 memcpy(&info->super_copy->super_roots,
1889 &info->super_for_commit->super_roots,
1890 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1894 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1895 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1897 * fs_info - filesystem whose backup roots need to be read
1898 * priority - priority of backup root required
1900 * Returns backup root index on success and -EINVAL otherwise.
1902 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1904 int backup_index = find_newest_super_backup(fs_info);
1905 struct btrfs_super_block *super = fs_info->super_copy;
1906 struct btrfs_root_backup *root_backup;
1908 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1910 return backup_index;
1912 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1913 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1918 root_backup = super->super_roots + backup_index;
1920 btrfs_set_super_generation(super,
1921 btrfs_backup_tree_root_gen(root_backup));
1922 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1923 btrfs_set_super_root_level(super,
1924 btrfs_backup_tree_root_level(root_backup));
1925 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1928 * Fixme: the total bytes and num_devices need to match or we should
1931 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1932 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1934 return backup_index;
1937 /* helper to cleanup workers */
1938 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1940 btrfs_destroy_workqueue(fs_info->fixup_workers);
1941 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1942 btrfs_destroy_workqueue(fs_info->workers);
1943 btrfs_destroy_workqueue(fs_info->endio_workers);
1944 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1945 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1946 btrfs_destroy_workqueue(fs_info->rmw_workers);
1947 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1948 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1949 btrfs_destroy_workqueue(fs_info->delayed_workers);
1950 btrfs_destroy_workqueue(fs_info->caching_workers);
1951 btrfs_destroy_workqueue(fs_info->readahead_workers);
1952 btrfs_destroy_workqueue(fs_info->flush_workers);
1953 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1954 if (fs_info->discard_ctl.discard_workers)
1955 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1957 * Now that all other work queues are destroyed, we can safely destroy
1958 * the queues used for metadata I/O, since tasks from those other work
1959 * queues can do metadata I/O operations.
1961 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
1962 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
1965 static void free_root_extent_buffers(struct btrfs_root *root)
1968 free_extent_buffer(root->node);
1969 free_extent_buffer(root->commit_root);
1971 root->commit_root = NULL;
1975 /* helper to cleanup tree roots */
1976 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1978 free_root_extent_buffers(info->tree_root);
1980 free_root_extent_buffers(info->dev_root);
1981 free_root_extent_buffers(info->extent_root);
1982 free_root_extent_buffers(info->csum_root);
1983 free_root_extent_buffers(info->quota_root);
1984 free_root_extent_buffers(info->uuid_root);
1985 free_root_extent_buffers(info->fs_root);
1986 if (free_chunk_root)
1987 free_root_extent_buffers(info->chunk_root);
1988 free_root_extent_buffers(info->free_space_root);
1991 void btrfs_put_root(struct btrfs_root *root)
1996 if (refcount_dec_and_test(&root->refs)) {
1997 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1999 free_anon_bdev(root->anon_dev);
2000 btrfs_drew_lock_destroy(&root->snapshot_lock);
2001 free_extent_buffer(root->node);
2002 free_extent_buffer(root->commit_root);
2003 kfree(root->free_ino_ctl);
2004 kfree(root->free_ino_pinned);
2005 #ifdef CONFIG_BTRFS_DEBUG
2006 spin_lock(&root->fs_info->fs_roots_radix_lock);
2007 list_del_init(&root->leak_list);
2008 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2014 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2017 struct btrfs_root *gang[8];
2020 while (!list_empty(&fs_info->dead_roots)) {
2021 gang[0] = list_entry(fs_info->dead_roots.next,
2022 struct btrfs_root, root_list);
2023 list_del(&gang[0]->root_list);
2025 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2026 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2027 btrfs_put_root(gang[0]);
2031 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2036 for (i = 0; i < ret; i++)
2037 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2041 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2043 mutex_init(&fs_info->scrub_lock);
2044 atomic_set(&fs_info->scrubs_running, 0);
2045 atomic_set(&fs_info->scrub_pause_req, 0);
2046 atomic_set(&fs_info->scrubs_paused, 0);
2047 atomic_set(&fs_info->scrub_cancel_req, 0);
2048 init_waitqueue_head(&fs_info->scrub_pause_wait);
2049 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2052 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2054 spin_lock_init(&fs_info->balance_lock);
2055 mutex_init(&fs_info->balance_mutex);
2056 atomic_set(&fs_info->balance_pause_req, 0);
2057 atomic_set(&fs_info->balance_cancel_req, 0);
2058 fs_info->balance_ctl = NULL;
2059 init_waitqueue_head(&fs_info->balance_wait_q);
2062 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2064 struct inode *inode = fs_info->btree_inode;
2066 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2067 set_nlink(inode, 1);
2069 * we set the i_size on the btree inode to the max possible int.
2070 * the real end of the address space is determined by all of
2071 * the devices in the system
2073 inode->i_size = OFFSET_MAX;
2074 inode->i_mapping->a_ops = &btree_aops;
2076 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2077 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2078 IO_TREE_INODE_IO, inode);
2079 BTRFS_I(inode)->io_tree.track_uptodate = false;
2080 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2082 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2084 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2085 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2086 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2087 btrfs_insert_inode_hash(inode);
2090 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2092 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2093 init_rwsem(&fs_info->dev_replace.rwsem);
2094 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2097 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2099 spin_lock_init(&fs_info->qgroup_lock);
2100 mutex_init(&fs_info->qgroup_ioctl_lock);
2101 fs_info->qgroup_tree = RB_ROOT;
2102 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2103 fs_info->qgroup_seq = 1;
2104 fs_info->qgroup_ulist = NULL;
2105 fs_info->qgroup_rescan_running = false;
2106 mutex_init(&fs_info->qgroup_rescan_lock);
2109 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2110 struct btrfs_fs_devices *fs_devices)
2112 u32 max_active = fs_info->thread_pool_size;
2113 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2116 btrfs_alloc_workqueue(fs_info, "worker",
2117 flags | WQ_HIGHPRI, max_active, 16);
2119 fs_info->delalloc_workers =
2120 btrfs_alloc_workqueue(fs_info, "delalloc",
2121 flags, max_active, 2);
2123 fs_info->flush_workers =
2124 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2125 flags, max_active, 0);
2127 fs_info->caching_workers =
2128 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2130 fs_info->fixup_workers =
2131 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2134 * endios are largely parallel and should have a very
2137 fs_info->endio_workers =
2138 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2139 fs_info->endio_meta_workers =
2140 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2142 fs_info->endio_meta_write_workers =
2143 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2145 fs_info->endio_raid56_workers =
2146 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2148 fs_info->endio_repair_workers =
2149 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2150 fs_info->rmw_workers =
2151 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2152 fs_info->endio_write_workers =
2153 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2155 fs_info->endio_freespace_worker =
2156 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2158 fs_info->delayed_workers =
2159 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2161 fs_info->readahead_workers =
2162 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2164 fs_info->qgroup_rescan_workers =
2165 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2166 fs_info->discard_ctl.discard_workers =
2167 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2169 if (!(fs_info->workers && fs_info->delalloc_workers &&
2170 fs_info->flush_workers &&
2171 fs_info->endio_workers && fs_info->endio_meta_workers &&
2172 fs_info->endio_meta_write_workers &&
2173 fs_info->endio_repair_workers &&
2174 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2175 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2176 fs_info->caching_workers && fs_info->readahead_workers &&
2177 fs_info->fixup_workers && fs_info->delayed_workers &&
2178 fs_info->qgroup_rescan_workers &&
2179 fs_info->discard_ctl.discard_workers)) {
2186 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2188 struct crypto_shash *csum_shash;
2189 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2191 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2193 if (IS_ERR(csum_shash)) {
2194 btrfs_err(fs_info, "error allocating %s hash for checksum",
2196 return PTR_ERR(csum_shash);
2199 fs_info->csum_shash = csum_shash;
2204 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2205 struct btrfs_fs_devices *fs_devices)
2208 struct btrfs_root *log_tree_root;
2209 struct btrfs_super_block *disk_super = fs_info->super_copy;
2210 u64 bytenr = btrfs_super_log_root(disk_super);
2211 int level = btrfs_super_log_root_level(disk_super);
2213 if (fs_devices->rw_devices == 0) {
2214 btrfs_warn(fs_info, "log replay required on RO media");
2218 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2223 log_tree_root->node = read_tree_block(fs_info, bytenr,
2224 fs_info->generation + 1,
2226 if (IS_ERR(log_tree_root->node)) {
2227 btrfs_warn(fs_info, "failed to read log tree");
2228 ret = PTR_ERR(log_tree_root->node);
2229 log_tree_root->node = NULL;
2230 btrfs_put_root(log_tree_root);
2232 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2233 btrfs_err(fs_info, "failed to read log tree");
2234 btrfs_put_root(log_tree_root);
2237 /* returns with log_tree_root freed on success */
2238 ret = btrfs_recover_log_trees(log_tree_root);
2240 btrfs_handle_fs_error(fs_info, ret,
2241 "Failed to recover log tree");
2242 btrfs_put_root(log_tree_root);
2246 if (sb_rdonly(fs_info->sb)) {
2247 ret = btrfs_commit_super(fs_info);
2255 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2257 struct btrfs_root *tree_root = fs_info->tree_root;
2258 struct btrfs_root *root;
2259 struct btrfs_key location;
2262 BUG_ON(!fs_info->tree_root);
2264 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2265 location.type = BTRFS_ROOT_ITEM_KEY;
2266 location.offset = 0;
2268 root = btrfs_read_tree_root(tree_root, &location);
2270 ret = PTR_ERR(root);
2273 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2274 fs_info->extent_root = root;
2276 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2277 root = btrfs_read_tree_root(tree_root, &location);
2279 ret = PTR_ERR(root);
2282 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2283 fs_info->dev_root = root;
2284 btrfs_init_devices_late(fs_info);
2286 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2287 root = btrfs_read_tree_root(tree_root, &location);
2289 ret = PTR_ERR(root);
2292 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2293 fs_info->csum_root = root;
2295 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2296 root = btrfs_read_tree_root(tree_root, &location);
2297 if (!IS_ERR(root)) {
2298 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2299 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2300 fs_info->quota_root = root;
2303 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2304 root = btrfs_read_tree_root(tree_root, &location);
2306 ret = PTR_ERR(root);
2310 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2311 fs_info->uuid_root = root;
2314 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2315 location.objectid = BTRFS_FREE_SPACE_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->free_space_root = root;
2327 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2328 location.objectid, ret);
2333 * Real super block validation
2334 * NOTE: super csum type and incompat features will not be checked here.
2336 * @sb: super block to check
2337 * @mirror_num: the super block number to check its bytenr:
2338 * 0 the primary (1st) sb
2339 * 1, 2 2nd and 3rd backup copy
2340 * -1 skip bytenr check
2342 static int validate_super(struct btrfs_fs_info *fs_info,
2343 struct btrfs_super_block *sb, int mirror_num)
2345 u64 nodesize = btrfs_super_nodesize(sb);
2346 u64 sectorsize = btrfs_super_sectorsize(sb);
2349 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2350 btrfs_err(fs_info, "no valid FS found");
2353 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2354 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2355 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2358 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2359 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2360 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2363 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2364 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2365 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2368 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2369 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2370 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2375 * Check sectorsize and nodesize first, other check will need it.
2376 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2378 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2379 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2380 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2383 /* Only PAGE SIZE is supported yet */
2384 if (sectorsize != PAGE_SIZE) {
2386 "sectorsize %llu not supported yet, only support %lu",
2387 sectorsize, PAGE_SIZE);
2390 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2391 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2392 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2395 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2396 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2397 le32_to_cpu(sb->__unused_leafsize), nodesize);
2401 /* Root alignment check */
2402 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2403 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2404 btrfs_super_root(sb));
2407 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2408 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2409 btrfs_super_chunk_root(sb));
2412 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2413 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2414 btrfs_super_log_root(sb));
2418 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2419 BTRFS_FSID_SIZE) != 0) {
2421 "dev_item UUID does not match metadata fsid: %pU != %pU",
2422 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2427 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2430 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2431 btrfs_err(fs_info, "bytes_used is too small %llu",
2432 btrfs_super_bytes_used(sb));
2435 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2436 btrfs_err(fs_info, "invalid stripesize %u",
2437 btrfs_super_stripesize(sb));
2440 if (btrfs_super_num_devices(sb) > (1UL << 31))
2441 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2442 btrfs_super_num_devices(sb));
2443 if (btrfs_super_num_devices(sb) == 0) {
2444 btrfs_err(fs_info, "number of devices is 0");
2448 if (mirror_num >= 0 &&
2449 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2450 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2451 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2456 * Obvious sys_chunk_array corruptions, it must hold at least one key
2459 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2460 btrfs_err(fs_info, "system chunk array too big %u > %u",
2461 btrfs_super_sys_array_size(sb),
2462 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2465 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2466 + sizeof(struct btrfs_chunk)) {
2467 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2468 btrfs_super_sys_array_size(sb),
2469 sizeof(struct btrfs_disk_key)
2470 + sizeof(struct btrfs_chunk));
2475 * The generation is a global counter, we'll trust it more than the others
2476 * but it's still possible that it's the one that's wrong.
2478 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2480 "suspicious: generation < chunk_root_generation: %llu < %llu",
2481 btrfs_super_generation(sb),
2482 btrfs_super_chunk_root_generation(sb));
2483 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2484 && btrfs_super_cache_generation(sb) != (u64)-1)
2486 "suspicious: generation < cache_generation: %llu < %llu",
2487 btrfs_super_generation(sb),
2488 btrfs_super_cache_generation(sb));
2494 * Validation of super block at mount time.
2495 * Some checks already done early at mount time, like csum type and incompat
2496 * flags will be skipped.
2498 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2500 return validate_super(fs_info, fs_info->super_copy, 0);
2504 * Validation of super block at write time.
2505 * Some checks like bytenr check will be skipped as their values will be
2507 * Extra checks like csum type and incompat flags will be done here.
2509 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2510 struct btrfs_super_block *sb)
2514 ret = validate_super(fs_info, sb, -1);
2517 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2519 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2520 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2523 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2526 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2527 btrfs_super_incompat_flags(sb),
2528 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2534 "super block corruption detected before writing it to disk");
2538 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2540 int backup_index = find_newest_super_backup(fs_info);
2541 struct btrfs_super_block *sb = fs_info->super_copy;
2542 struct btrfs_root *tree_root = fs_info->tree_root;
2543 bool handle_error = false;
2547 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2552 if (!IS_ERR(tree_root->node))
2553 free_extent_buffer(tree_root->node);
2554 tree_root->node = NULL;
2556 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2559 free_root_pointers(fs_info, 0);
2562 * Don't use the log in recovery mode, it won't be
2565 btrfs_set_super_log_root(sb, 0);
2567 /* We can't trust the free space cache either */
2568 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2570 ret = read_backup_root(fs_info, i);
2575 generation = btrfs_super_generation(sb);
2576 level = btrfs_super_root_level(sb);
2577 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2578 generation, level, NULL);
2579 if (IS_ERR(tree_root->node) ||
2580 !extent_buffer_uptodate(tree_root->node)) {
2581 handle_error = true;
2583 if (IS_ERR(tree_root->node))
2584 ret = PTR_ERR(tree_root->node);
2585 else if (!extent_buffer_uptodate(tree_root->node))
2588 btrfs_warn(fs_info, "failed to read tree root");
2592 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2593 tree_root->commit_root = btrfs_root_node(tree_root);
2594 btrfs_set_root_refs(&tree_root->root_item, 1);
2597 * No need to hold btrfs_root::objectid_mutex since the fs
2598 * hasn't been fully initialised and we are the only user
2600 ret = btrfs_find_highest_objectid(tree_root,
2601 &tree_root->highest_objectid);
2603 handle_error = true;
2607 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2609 ret = btrfs_read_roots(fs_info);
2611 handle_error = true;
2615 /* All successful */
2616 fs_info->generation = generation;
2617 fs_info->last_trans_committed = generation;
2619 /* Always begin writing backup roots after the one being used */
2620 if (backup_index < 0) {
2621 fs_info->backup_root_index = 0;
2623 fs_info->backup_root_index = backup_index + 1;
2624 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2632 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2634 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2635 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2636 INIT_LIST_HEAD(&fs_info->trans_list);
2637 INIT_LIST_HEAD(&fs_info->dead_roots);
2638 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2639 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2640 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2641 spin_lock_init(&fs_info->delalloc_root_lock);
2642 spin_lock_init(&fs_info->trans_lock);
2643 spin_lock_init(&fs_info->fs_roots_radix_lock);
2644 spin_lock_init(&fs_info->delayed_iput_lock);
2645 spin_lock_init(&fs_info->defrag_inodes_lock);
2646 spin_lock_init(&fs_info->super_lock);
2647 spin_lock_init(&fs_info->buffer_lock);
2648 spin_lock_init(&fs_info->unused_bgs_lock);
2649 rwlock_init(&fs_info->tree_mod_log_lock);
2650 mutex_init(&fs_info->unused_bg_unpin_mutex);
2651 mutex_init(&fs_info->delete_unused_bgs_mutex);
2652 mutex_init(&fs_info->reloc_mutex);
2653 mutex_init(&fs_info->delalloc_root_mutex);
2654 seqlock_init(&fs_info->profiles_lock);
2656 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2657 INIT_LIST_HEAD(&fs_info->space_info);
2658 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2659 INIT_LIST_HEAD(&fs_info->unused_bgs);
2660 #ifdef CONFIG_BTRFS_DEBUG
2661 INIT_LIST_HEAD(&fs_info->allocated_roots);
2662 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2663 spin_lock_init(&fs_info->eb_leak_lock);
2665 extent_map_tree_init(&fs_info->mapping_tree);
2666 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2667 BTRFS_BLOCK_RSV_GLOBAL);
2668 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2669 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2670 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2671 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2672 BTRFS_BLOCK_RSV_DELOPS);
2673 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2674 BTRFS_BLOCK_RSV_DELREFS);
2676 atomic_set(&fs_info->async_delalloc_pages, 0);
2677 atomic_set(&fs_info->defrag_running, 0);
2678 atomic_set(&fs_info->reada_works_cnt, 0);
2679 atomic_set(&fs_info->nr_delayed_iputs, 0);
2680 atomic64_set(&fs_info->tree_mod_seq, 0);
2681 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2682 fs_info->metadata_ratio = 0;
2683 fs_info->defrag_inodes = RB_ROOT;
2684 atomic64_set(&fs_info->free_chunk_space, 0);
2685 fs_info->tree_mod_log = RB_ROOT;
2686 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2687 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2688 /* readahead state */
2689 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2690 spin_lock_init(&fs_info->reada_lock);
2691 btrfs_init_ref_verify(fs_info);
2693 fs_info->thread_pool_size = min_t(unsigned long,
2694 num_online_cpus() + 2, 8);
2696 INIT_LIST_HEAD(&fs_info->ordered_roots);
2697 spin_lock_init(&fs_info->ordered_root_lock);
2699 btrfs_init_scrub(fs_info);
2700 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2701 fs_info->check_integrity_print_mask = 0;
2703 btrfs_init_balance(fs_info);
2704 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2706 spin_lock_init(&fs_info->block_group_cache_lock);
2707 fs_info->block_group_cache_tree = RB_ROOT;
2708 fs_info->first_logical_byte = (u64)-1;
2710 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2711 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2712 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2714 mutex_init(&fs_info->ordered_operations_mutex);
2715 mutex_init(&fs_info->tree_log_mutex);
2716 mutex_init(&fs_info->chunk_mutex);
2717 mutex_init(&fs_info->transaction_kthread_mutex);
2718 mutex_init(&fs_info->cleaner_mutex);
2719 mutex_init(&fs_info->ro_block_group_mutex);
2720 init_rwsem(&fs_info->commit_root_sem);
2721 init_rwsem(&fs_info->cleanup_work_sem);
2722 init_rwsem(&fs_info->subvol_sem);
2723 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2725 btrfs_init_dev_replace_locks(fs_info);
2726 btrfs_init_qgroup(fs_info);
2727 btrfs_discard_init(fs_info);
2729 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2730 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2732 init_waitqueue_head(&fs_info->transaction_throttle);
2733 init_waitqueue_head(&fs_info->transaction_wait);
2734 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2735 init_waitqueue_head(&fs_info->async_submit_wait);
2736 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2738 /* Usable values until the real ones are cached from the superblock */
2739 fs_info->nodesize = 4096;
2740 fs_info->sectorsize = 4096;
2741 fs_info->stripesize = 4096;
2743 spin_lock_init(&fs_info->swapfile_pins_lock);
2744 fs_info->swapfile_pins = RB_ROOT;
2746 fs_info->send_in_progress = 0;
2749 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2754 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2755 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2757 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2761 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2765 fs_info->dirty_metadata_batch = PAGE_SIZE *
2766 (1 + ilog2(nr_cpu_ids));
2768 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2772 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2777 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2779 if (!fs_info->delayed_root)
2781 btrfs_init_delayed_root(fs_info->delayed_root);
2783 return btrfs_alloc_stripe_hash_table(fs_info);
2786 static int btrfs_uuid_rescan_kthread(void *data)
2788 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2792 * 1st step is to iterate through the existing UUID tree and
2793 * to delete all entries that contain outdated data.
2794 * 2nd step is to add all missing entries to the UUID tree.
2796 ret = btrfs_uuid_tree_iterate(fs_info);
2799 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2801 up(&fs_info->uuid_tree_rescan_sem);
2804 return btrfs_uuid_scan_kthread(data);
2807 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2809 struct task_struct *task;
2811 down(&fs_info->uuid_tree_rescan_sem);
2812 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2814 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2815 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2816 up(&fs_info->uuid_tree_rescan_sem);
2817 return PTR_ERR(task);
2823 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2832 struct btrfs_key location;
2833 struct btrfs_super_block *disk_super;
2834 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2835 struct btrfs_root *tree_root;
2836 struct btrfs_root *chunk_root;
2839 int clear_free_space_tree = 0;
2842 ret = init_mount_fs_info(fs_info, sb);
2848 /* These need to be init'ed before we start creating inodes and such. */
2849 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2851 fs_info->tree_root = tree_root;
2852 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2854 fs_info->chunk_root = chunk_root;
2855 if (!tree_root || !chunk_root) {
2860 fs_info->btree_inode = new_inode(sb);
2861 if (!fs_info->btree_inode) {
2865 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2866 btrfs_init_btree_inode(fs_info);
2868 invalidate_bdev(fs_devices->latest_bdev);
2871 * Read super block and check the signature bytes only
2873 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2874 if (IS_ERR(disk_super)) {
2875 err = PTR_ERR(disk_super);
2880 * Verify the type first, if that or the the checksum value are
2881 * corrupted, we'll find out
2883 csum_type = btrfs_super_csum_type(disk_super);
2884 if (!btrfs_supported_super_csum(csum_type)) {
2885 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2888 btrfs_release_disk_super(disk_super);
2892 ret = btrfs_init_csum_hash(fs_info, csum_type);
2895 btrfs_release_disk_super(disk_super);
2900 * We want to check superblock checksum, the type is stored inside.
2901 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2903 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2904 btrfs_err(fs_info, "superblock checksum mismatch");
2906 btrfs_release_disk_super(disk_super);
2911 * super_copy is zeroed at allocation time and we never touch the
2912 * following bytes up to INFO_SIZE, the checksum is calculated from
2913 * the whole block of INFO_SIZE
2915 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2916 btrfs_release_disk_super(disk_super);
2918 disk_super = fs_info->super_copy;
2920 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2923 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2924 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2925 fs_info->super_copy->metadata_uuid,
2929 features = btrfs_super_flags(disk_super);
2930 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2931 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2932 btrfs_set_super_flags(disk_super, features);
2934 "found metadata UUID change in progress flag, clearing");
2937 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2938 sizeof(*fs_info->super_for_commit));
2940 ret = btrfs_validate_mount_super(fs_info);
2942 btrfs_err(fs_info, "superblock contains fatal errors");
2947 if (!btrfs_super_root(disk_super))
2950 /* check FS state, whether FS is broken. */
2951 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2952 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2955 * In the long term, we'll store the compression type in the super
2956 * block, and it'll be used for per file compression control.
2958 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2960 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2966 features = btrfs_super_incompat_flags(disk_super) &
2967 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2970 "cannot mount because of unsupported optional features (%llx)",
2976 features = btrfs_super_incompat_flags(disk_super);
2977 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2978 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2979 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2980 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2981 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2983 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2984 btrfs_info(fs_info, "has skinny extents");
2987 * flag our filesystem as having big metadata blocks if
2988 * they are bigger than the page size
2990 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2991 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2993 "flagging fs with big metadata feature");
2994 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2997 nodesize = btrfs_super_nodesize(disk_super);
2998 sectorsize = btrfs_super_sectorsize(disk_super);
2999 stripesize = sectorsize;
3000 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3001 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3003 /* Cache block sizes */
3004 fs_info->nodesize = nodesize;
3005 fs_info->sectorsize = sectorsize;
3006 fs_info->stripesize = stripesize;
3009 * mixed block groups end up with duplicate but slightly offset
3010 * extent buffers for the same range. It leads to corruptions
3012 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3013 (sectorsize != nodesize)) {
3015 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3016 nodesize, sectorsize);
3021 * Needn't use the lock because there is no other task which will
3024 btrfs_set_super_incompat_flags(disk_super, features);
3026 features = btrfs_super_compat_ro_flags(disk_super) &
3027 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3028 if (!sb_rdonly(sb) && features) {
3030 "cannot mount read-write because of unsupported optional features (%llx)",
3036 ret = btrfs_init_workqueues(fs_info, fs_devices);
3039 goto fail_sb_buffer;
3042 sb->s_bdi->congested_fn = btrfs_congested_fn;
3043 sb->s_bdi->congested_data = fs_info;
3044 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3045 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3046 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3047 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3049 sb->s_blocksize = sectorsize;
3050 sb->s_blocksize_bits = blksize_bits(sectorsize);
3051 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3053 mutex_lock(&fs_info->chunk_mutex);
3054 ret = btrfs_read_sys_array(fs_info);
3055 mutex_unlock(&fs_info->chunk_mutex);
3057 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3058 goto fail_sb_buffer;
3061 generation = btrfs_super_chunk_root_generation(disk_super);
3062 level = btrfs_super_chunk_root_level(disk_super);
3064 chunk_root->node = read_tree_block(fs_info,
3065 btrfs_super_chunk_root(disk_super),
3066 generation, level, NULL);
3067 if (IS_ERR(chunk_root->node) ||
3068 !extent_buffer_uptodate(chunk_root->node)) {
3069 btrfs_err(fs_info, "failed to read chunk root");
3070 if (!IS_ERR(chunk_root->node))
3071 free_extent_buffer(chunk_root->node);
3072 chunk_root->node = NULL;
3073 goto fail_tree_roots;
3075 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3076 chunk_root->commit_root = btrfs_root_node(chunk_root);
3078 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3079 offsetof(struct btrfs_header, chunk_tree_uuid),
3082 ret = btrfs_read_chunk_tree(fs_info);
3084 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3085 goto fail_tree_roots;
3089 * Keep the devid that is marked to be the target device for the
3090 * device replace procedure
3092 btrfs_free_extra_devids(fs_devices, 0);
3094 if (!fs_devices->latest_bdev) {
3095 btrfs_err(fs_info, "failed to read devices");
3096 goto fail_tree_roots;
3099 ret = init_tree_roots(fs_info);
3101 goto fail_tree_roots;
3104 * If we have a uuid root and we're not being told to rescan we need to
3105 * check the generation here so we can set the
3106 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3107 * transaction during a balance or the log replay without updating the
3108 * uuid generation, and then if we crash we would rescan the uuid tree,
3109 * even though it was perfectly fine.
3111 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3112 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3113 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3115 ret = btrfs_verify_dev_extents(fs_info);
3118 "failed to verify dev extents against chunks: %d",
3120 goto fail_block_groups;
3122 ret = btrfs_recover_balance(fs_info);
3124 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3125 goto fail_block_groups;
3128 ret = btrfs_init_dev_stats(fs_info);
3130 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3131 goto fail_block_groups;
3134 ret = btrfs_init_dev_replace(fs_info);
3136 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3137 goto fail_block_groups;
3140 btrfs_free_extra_devids(fs_devices, 1);
3142 ret = btrfs_sysfs_add_fsid(fs_devices);
3144 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3146 goto fail_block_groups;
3149 ret = btrfs_sysfs_add_mounted(fs_info);
3151 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3152 goto fail_fsdev_sysfs;
3155 ret = btrfs_init_space_info(fs_info);
3157 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3161 ret = btrfs_read_block_groups(fs_info);
3163 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3167 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3169 "writable mount is not allowed due to too many missing devices");
3173 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3175 if (IS_ERR(fs_info->cleaner_kthread))
3178 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3180 "btrfs-transaction");
3181 if (IS_ERR(fs_info->transaction_kthread))
3184 if (!btrfs_test_opt(fs_info, NOSSD) &&
3185 !fs_info->fs_devices->rotating) {
3186 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3190 * Mount does not set all options immediately, we can do it now and do
3191 * not have to wait for transaction commit
3193 btrfs_apply_pending_changes(fs_info);
3195 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3196 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3197 ret = btrfsic_mount(fs_info, fs_devices,
3198 btrfs_test_opt(fs_info,
3199 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3201 fs_info->check_integrity_print_mask);
3204 "failed to initialize integrity check module: %d",
3208 ret = btrfs_read_qgroup_config(fs_info);
3210 goto fail_trans_kthread;
3212 if (btrfs_build_ref_tree(fs_info))
3213 btrfs_err(fs_info, "couldn't build ref tree");
3215 /* do not make disk changes in broken FS or nologreplay is given */
3216 if (btrfs_super_log_root(disk_super) != 0 &&
3217 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3218 btrfs_info(fs_info, "start tree-log replay");
3219 ret = btrfs_replay_log(fs_info, fs_devices);
3226 ret = btrfs_find_orphan_roots(fs_info);
3230 if (!sb_rdonly(sb)) {
3231 ret = btrfs_cleanup_fs_roots(fs_info);
3235 mutex_lock(&fs_info->cleaner_mutex);
3236 ret = btrfs_recover_relocation(tree_root);
3237 mutex_unlock(&fs_info->cleaner_mutex);
3239 btrfs_warn(fs_info, "failed to recover relocation: %d",
3246 location.objectid = BTRFS_FS_TREE_OBJECTID;
3247 location.type = BTRFS_ROOT_ITEM_KEY;
3248 location.offset = 0;
3250 fs_info->fs_root = btrfs_get_fs_root(fs_info, &location, true);
3251 if (IS_ERR(fs_info->fs_root)) {
3252 err = PTR_ERR(fs_info->fs_root);
3253 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3254 fs_info->fs_root = NULL;
3261 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3262 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3263 clear_free_space_tree = 1;
3264 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3265 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3266 btrfs_warn(fs_info, "free space tree is invalid");
3267 clear_free_space_tree = 1;
3270 if (clear_free_space_tree) {
3271 btrfs_info(fs_info, "clearing free space tree");
3272 ret = btrfs_clear_free_space_tree(fs_info);
3275 "failed to clear free space tree: %d", ret);
3276 close_ctree(fs_info);
3281 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3282 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3283 btrfs_info(fs_info, "creating free space tree");
3284 ret = btrfs_create_free_space_tree(fs_info);
3287 "failed to create free space tree: %d", ret);
3288 close_ctree(fs_info);
3293 down_read(&fs_info->cleanup_work_sem);
3294 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3295 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3296 up_read(&fs_info->cleanup_work_sem);
3297 close_ctree(fs_info);
3300 up_read(&fs_info->cleanup_work_sem);
3302 ret = btrfs_resume_balance_async(fs_info);
3304 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3305 close_ctree(fs_info);
3309 ret = btrfs_resume_dev_replace_async(fs_info);
3311 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3312 close_ctree(fs_info);
3316 btrfs_qgroup_rescan_resume(fs_info);
3317 btrfs_discard_resume(fs_info);
3319 if (!fs_info->uuid_root) {
3320 btrfs_info(fs_info, "creating UUID tree");
3321 ret = btrfs_create_uuid_tree(fs_info);
3324 "failed to create the UUID tree: %d", ret);
3325 close_ctree(fs_info);
3328 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3329 fs_info->generation !=
3330 btrfs_super_uuid_tree_generation(disk_super)) {
3331 btrfs_info(fs_info, "checking UUID tree");
3332 ret = btrfs_check_uuid_tree(fs_info);
3335 "failed to check the UUID tree: %d", ret);
3336 close_ctree(fs_info);
3340 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3343 * backuproot only affect mount behavior, and if open_ctree succeeded,
3344 * no need to keep the flag
3346 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3351 btrfs_free_qgroup_config(fs_info);
3353 kthread_stop(fs_info->transaction_kthread);
3354 btrfs_cleanup_transaction(fs_info);
3355 btrfs_free_fs_roots(fs_info);
3357 kthread_stop(fs_info->cleaner_kthread);
3360 * make sure we're done with the btree inode before we stop our
3363 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3366 btrfs_sysfs_remove_mounted(fs_info);
3369 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3372 btrfs_put_block_group_cache(fs_info);
3375 free_root_pointers(fs_info, true);
3376 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3379 btrfs_stop_all_workers(fs_info);
3380 btrfs_free_block_groups(fs_info);
3382 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3384 iput(fs_info->btree_inode);
3386 btrfs_close_devices(fs_info->fs_devices);
3389 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3391 static void btrfs_end_super_write(struct bio *bio)
3393 struct btrfs_device *device = bio->bi_private;
3394 struct bio_vec *bvec;
3395 struct bvec_iter_all iter_all;
3398 bio_for_each_segment_all(bvec, bio, iter_all) {
3399 page = bvec->bv_page;
3401 if (bio->bi_status) {
3402 btrfs_warn_rl_in_rcu(device->fs_info,
3403 "lost page write due to IO error on %s (%d)",
3404 rcu_str_deref(device->name),
3405 blk_status_to_errno(bio->bi_status));
3406 ClearPageUptodate(page);
3408 btrfs_dev_stat_inc_and_print(device,
3409 BTRFS_DEV_STAT_WRITE_ERRS);
3411 SetPageUptodate(page);
3421 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3424 struct btrfs_super_block *super;
3427 struct address_space *mapping = bdev->bd_inode->i_mapping;
3429 bytenr = btrfs_sb_offset(copy_num);
3430 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3431 return ERR_PTR(-EINVAL);
3433 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3435 return ERR_CAST(page);
3437 super = page_address(page);
3438 if (btrfs_super_bytenr(super) != bytenr ||
3439 btrfs_super_magic(super) != BTRFS_MAGIC) {
3440 btrfs_release_disk_super(super);
3441 return ERR_PTR(-EINVAL);
3448 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3450 struct btrfs_super_block *super, *latest = NULL;
3454 /* we would like to check all the supers, but that would make
3455 * a btrfs mount succeed after a mkfs from a different FS.
3456 * So, we need to add a special mount option to scan for
3457 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3459 for (i = 0; i < 1; i++) {
3460 super = btrfs_read_dev_one_super(bdev, i);
3464 if (!latest || btrfs_super_generation(super) > transid) {
3466 btrfs_release_disk_super(super);
3469 transid = btrfs_super_generation(super);
3477 * Write superblock @sb to the @device. Do not wait for completion, all the
3478 * pages we use for writing are locked.
3480 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3481 * the expected device size at commit time. Note that max_mirrors must be
3482 * same for write and wait phases.
3484 * Return number of errors when page is not found or submission fails.
3486 static int write_dev_supers(struct btrfs_device *device,
3487 struct btrfs_super_block *sb, int max_mirrors)
3489 struct btrfs_fs_info *fs_info = device->fs_info;
3490 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3491 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3496 if (max_mirrors == 0)
3497 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3499 shash->tfm = fs_info->csum_shash;
3501 for (i = 0; i < max_mirrors; i++) {
3504 struct btrfs_super_block *disk_super;
3506 bytenr = btrfs_sb_offset(i);
3507 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3508 device->commit_total_bytes)
3511 btrfs_set_super_bytenr(sb, bytenr);
3513 crypto_shash_init(shash);
3514 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3515 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3516 crypto_shash_final(shash, sb->csum);
3518 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3521 btrfs_err(device->fs_info,
3522 "couldn't get super block page for bytenr %llu",
3528 /* Bump the refcount for wait_dev_supers() */
3531 disk_super = page_address(page);
3532 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3535 * Directly use bios here instead of relying on the page cache
3536 * to do I/O, so we don't lose the ability to do integrity
3539 bio = bio_alloc(GFP_NOFS, 1);
3540 bio_set_dev(bio, device->bdev);
3541 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3542 bio->bi_private = device;
3543 bio->bi_end_io = btrfs_end_super_write;
3544 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3545 offset_in_page(bytenr));
3548 * We FUA only the first super block. The others we allow to
3549 * go down lazy and there's a short window where the on-disk
3550 * copies might still contain the older version.
3552 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3553 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3554 bio->bi_opf |= REQ_FUA;
3556 btrfsic_submit_bio(bio);
3558 return errors < i ? 0 : -1;
3562 * Wait for write completion of superblocks done by write_dev_supers,
3563 * @max_mirrors same for write and wait phases.
3565 * Return number of errors when page is not found or not marked up to
3568 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3572 bool primary_failed = false;
3575 if (max_mirrors == 0)
3576 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3578 for (i = 0; i < max_mirrors; i++) {
3581 bytenr = btrfs_sb_offset(i);
3582 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3583 device->commit_total_bytes)
3586 page = find_get_page(device->bdev->bd_inode->i_mapping,
3587 bytenr >> PAGE_SHIFT);
3591 primary_failed = true;
3594 /* Page is submitted locked and unlocked once the IO completes */
3595 wait_on_page_locked(page);
3596 if (PageError(page)) {
3599 primary_failed = true;
3602 /* Drop our reference */
3605 /* Drop the reference from the writing run */
3609 /* log error, force error return */
3610 if (primary_failed) {
3611 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3616 return errors < i ? 0 : -1;
3620 * endio for the write_dev_flush, this will wake anyone waiting
3621 * for the barrier when it is done
3623 static void btrfs_end_empty_barrier(struct bio *bio)
3625 complete(bio->bi_private);
3629 * Submit a flush request to the device if it supports it. Error handling is
3630 * done in the waiting counterpart.
3632 static void write_dev_flush(struct btrfs_device *device)
3634 struct request_queue *q = bdev_get_queue(device->bdev);
3635 struct bio *bio = device->flush_bio;
3637 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3641 bio->bi_end_io = btrfs_end_empty_barrier;
3642 bio_set_dev(bio, device->bdev);
3643 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3644 init_completion(&device->flush_wait);
3645 bio->bi_private = &device->flush_wait;
3647 btrfsic_submit_bio(bio);
3648 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3652 * If the flush bio has been submitted by write_dev_flush, wait for it.
3654 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3656 struct bio *bio = device->flush_bio;
3658 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3661 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3662 wait_for_completion_io(&device->flush_wait);
3664 return bio->bi_status;
3667 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3669 if (!btrfs_check_rw_degradable(fs_info, NULL))
3675 * send an empty flush down to each device in parallel,
3676 * then wait for them
3678 static int barrier_all_devices(struct btrfs_fs_info *info)
3680 struct list_head *head;
3681 struct btrfs_device *dev;
3682 int errors_wait = 0;
3685 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3686 /* send down all the barriers */
3687 head = &info->fs_devices->devices;
3688 list_for_each_entry(dev, head, dev_list) {
3689 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3693 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3694 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3697 write_dev_flush(dev);
3698 dev->last_flush_error = BLK_STS_OK;
3701 /* wait for all the barriers */
3702 list_for_each_entry(dev, head, dev_list) {
3703 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3709 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3710 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3713 ret = wait_dev_flush(dev);
3715 dev->last_flush_error = ret;
3716 btrfs_dev_stat_inc_and_print(dev,
3717 BTRFS_DEV_STAT_FLUSH_ERRS);
3724 * At some point we need the status of all disks
3725 * to arrive at the volume status. So error checking
3726 * is being pushed to a separate loop.
3728 return check_barrier_error(info);
3733 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3736 int min_tolerated = INT_MAX;
3738 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3739 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3740 min_tolerated = min_t(int, min_tolerated,
3741 btrfs_raid_array[BTRFS_RAID_SINGLE].
3742 tolerated_failures);
3744 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3745 if (raid_type == BTRFS_RAID_SINGLE)
3747 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3749 min_tolerated = min_t(int, min_tolerated,
3750 btrfs_raid_array[raid_type].
3751 tolerated_failures);
3754 if (min_tolerated == INT_MAX) {
3755 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3759 return min_tolerated;
3762 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3764 struct list_head *head;
3765 struct btrfs_device *dev;
3766 struct btrfs_super_block *sb;
3767 struct btrfs_dev_item *dev_item;
3771 int total_errors = 0;
3774 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3777 * max_mirrors == 0 indicates we're from commit_transaction,
3778 * not from fsync where the tree roots in fs_info have not
3779 * been consistent on disk.
3781 if (max_mirrors == 0)
3782 backup_super_roots(fs_info);
3784 sb = fs_info->super_for_commit;
3785 dev_item = &sb->dev_item;
3787 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3788 head = &fs_info->fs_devices->devices;
3789 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3792 ret = barrier_all_devices(fs_info);
3795 &fs_info->fs_devices->device_list_mutex);
3796 btrfs_handle_fs_error(fs_info, ret,
3797 "errors while submitting device barriers.");
3802 list_for_each_entry(dev, head, dev_list) {
3807 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3808 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3811 btrfs_set_stack_device_generation(dev_item, 0);
3812 btrfs_set_stack_device_type(dev_item, dev->type);
3813 btrfs_set_stack_device_id(dev_item, dev->devid);
3814 btrfs_set_stack_device_total_bytes(dev_item,
3815 dev->commit_total_bytes);
3816 btrfs_set_stack_device_bytes_used(dev_item,
3817 dev->commit_bytes_used);
3818 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3819 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3820 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3821 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3822 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3825 flags = btrfs_super_flags(sb);
3826 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3828 ret = btrfs_validate_write_super(fs_info, sb);
3830 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3831 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3832 "unexpected superblock corruption detected");
3836 ret = write_dev_supers(dev, sb, max_mirrors);
3840 if (total_errors > max_errors) {
3841 btrfs_err(fs_info, "%d errors while writing supers",
3843 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3845 /* FUA is masked off if unsupported and can't be the reason */
3846 btrfs_handle_fs_error(fs_info, -EIO,
3847 "%d errors while writing supers",
3853 list_for_each_entry(dev, head, dev_list) {
3856 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3857 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3860 ret = wait_dev_supers(dev, max_mirrors);
3864 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3865 if (total_errors > max_errors) {
3866 btrfs_handle_fs_error(fs_info, -EIO,
3867 "%d errors while writing supers",
3874 /* Drop a fs root from the radix tree and free it. */
3875 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3876 struct btrfs_root *root)
3878 bool drop_ref = false;
3880 spin_lock(&fs_info->fs_roots_radix_lock);
3881 radix_tree_delete(&fs_info->fs_roots_radix,
3882 (unsigned long)root->root_key.objectid);
3883 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3885 spin_unlock(&fs_info->fs_roots_radix_lock);
3887 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3888 ASSERT(root->log_root == NULL);
3889 if (root->reloc_root) {
3890 btrfs_put_root(root->reloc_root);
3891 root->reloc_root = NULL;
3895 if (root->free_ino_pinned)
3896 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3897 if (root->free_ino_ctl)
3898 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3899 if (root->ino_cache_inode) {
3900 iput(root->ino_cache_inode);
3901 root->ino_cache_inode = NULL;
3904 btrfs_put_root(root);
3907 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3909 u64 root_objectid = 0;
3910 struct btrfs_root *gang[8];
3913 unsigned int ret = 0;
3916 spin_lock(&fs_info->fs_roots_radix_lock);
3917 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3918 (void **)gang, root_objectid,
3921 spin_unlock(&fs_info->fs_roots_radix_lock);
3924 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3926 for (i = 0; i < ret; i++) {
3927 /* Avoid to grab roots in dead_roots */
3928 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3932 /* grab all the search result for later use */
3933 gang[i] = btrfs_grab_root(gang[i]);
3935 spin_unlock(&fs_info->fs_roots_radix_lock);
3937 for (i = 0; i < ret; i++) {
3940 root_objectid = gang[i]->root_key.objectid;
3941 err = btrfs_orphan_cleanup(gang[i]);
3944 btrfs_put_root(gang[i]);
3949 /* release the uncleaned roots due to error */
3950 for (; i < ret; i++) {
3952 btrfs_put_root(gang[i]);
3957 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3959 struct btrfs_root *root = fs_info->tree_root;
3960 struct btrfs_trans_handle *trans;
3962 mutex_lock(&fs_info->cleaner_mutex);
3963 btrfs_run_delayed_iputs(fs_info);
3964 mutex_unlock(&fs_info->cleaner_mutex);
3965 wake_up_process(fs_info->cleaner_kthread);
3967 /* wait until ongoing cleanup work done */
3968 down_write(&fs_info->cleanup_work_sem);
3969 up_write(&fs_info->cleanup_work_sem);
3971 trans = btrfs_join_transaction(root);
3973 return PTR_ERR(trans);
3974 return btrfs_commit_transaction(trans);
3977 void __cold close_ctree(struct btrfs_fs_info *fs_info)
3981 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3983 * We don't want the cleaner to start new transactions, add more delayed
3984 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3985 * because that frees the task_struct, and the transaction kthread might
3986 * still try to wake up the cleaner.
3988 kthread_park(fs_info->cleaner_kthread);
3990 /* wait for the qgroup rescan worker to stop */
3991 btrfs_qgroup_wait_for_completion(fs_info, false);
3993 /* wait for the uuid_scan task to finish */
3994 down(&fs_info->uuid_tree_rescan_sem);
3995 /* avoid complains from lockdep et al., set sem back to initial state */
3996 up(&fs_info->uuid_tree_rescan_sem);
3998 /* pause restriper - we want to resume on mount */
3999 btrfs_pause_balance(fs_info);
4001 btrfs_dev_replace_suspend_for_unmount(fs_info);
4003 btrfs_scrub_cancel(fs_info);
4005 /* wait for any defraggers to finish */
4006 wait_event(fs_info->transaction_wait,
4007 (atomic_read(&fs_info->defrag_running) == 0));
4009 /* clear out the rbtree of defraggable inodes */
4010 btrfs_cleanup_defrag_inodes(fs_info);
4012 cancel_work_sync(&fs_info->async_reclaim_work);
4014 /* Cancel or finish ongoing discard work */
4015 btrfs_discard_cleanup(fs_info);
4017 if (!sb_rdonly(fs_info->sb)) {
4019 * The cleaner kthread is stopped, so do one final pass over
4020 * unused block groups.
4022 btrfs_delete_unused_bgs(fs_info);
4025 * There might be existing delayed inode workers still running
4026 * and holding an empty delayed inode item. We must wait for
4027 * them to complete first because they can create a transaction.
4028 * This happens when someone calls btrfs_balance_delayed_items()
4029 * and then a transaction commit runs the same delayed nodes
4030 * before any delayed worker has done something with the nodes.
4031 * We must wait for any worker here and not at transaction
4032 * commit time since that could cause a deadlock.
4033 * This is a very rare case.
4035 btrfs_flush_workqueue(fs_info->delayed_workers);
4037 ret = btrfs_commit_super(fs_info);
4039 btrfs_err(fs_info, "commit super ret %d", ret);
4042 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4043 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4044 btrfs_error_commit_super(fs_info);
4046 kthread_stop(fs_info->transaction_kthread);
4047 kthread_stop(fs_info->cleaner_kthread);
4049 ASSERT(list_empty(&fs_info->delayed_iputs));
4050 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4052 btrfs_free_qgroup_config(fs_info);
4053 ASSERT(list_empty(&fs_info->delalloc_roots));
4055 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4056 btrfs_info(fs_info, "at unmount delalloc count %lld",
4057 percpu_counter_sum(&fs_info->delalloc_bytes));
4060 if (percpu_counter_sum(&fs_info->dio_bytes))
4061 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4062 percpu_counter_sum(&fs_info->dio_bytes));
4064 btrfs_sysfs_remove_mounted(fs_info);
4065 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4067 btrfs_put_block_group_cache(fs_info);
4070 * we must make sure there is not any read request to
4071 * submit after we stopping all workers.
4073 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4074 btrfs_stop_all_workers(fs_info);
4076 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4077 free_root_pointers(fs_info, true);
4078 btrfs_free_fs_roots(fs_info);
4081 * We must free the block groups after dropping the fs_roots as we could
4082 * have had an IO error and have left over tree log blocks that aren't
4083 * cleaned up until the fs roots are freed. This makes the block group
4084 * accounting appear to be wrong because there's pending reserved bytes,
4085 * so make sure we do the block group cleanup afterwards.
4087 btrfs_free_block_groups(fs_info);
4089 iput(fs_info->btree_inode);
4091 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4092 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4093 btrfsic_unmount(fs_info->fs_devices);
4096 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4097 btrfs_close_devices(fs_info->fs_devices);
4100 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4104 struct inode *btree_inode = buf->pages[0]->mapping->host;
4106 ret = extent_buffer_uptodate(buf);
4110 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4111 parent_transid, atomic);
4117 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4119 struct btrfs_fs_info *fs_info;
4120 struct btrfs_root *root;
4121 u64 transid = btrfs_header_generation(buf);
4124 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4126 * This is a fast path so only do this check if we have sanity tests
4127 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4128 * outside of the sanity tests.
4130 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4133 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4134 fs_info = root->fs_info;
4135 btrfs_assert_tree_locked(buf);
4136 if (transid != fs_info->generation)
4137 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4138 buf->start, transid, fs_info->generation);
4139 was_dirty = set_extent_buffer_dirty(buf);
4141 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4143 fs_info->dirty_metadata_batch);
4144 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4146 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4147 * but item data not updated.
4148 * So here we should only check item pointers, not item data.
4150 if (btrfs_header_level(buf) == 0 &&
4151 btrfs_check_leaf_relaxed(buf)) {
4152 btrfs_print_leaf(buf);
4158 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4162 * looks as though older kernels can get into trouble with
4163 * this code, they end up stuck in balance_dirty_pages forever
4167 if (current->flags & PF_MEMALLOC)
4171 btrfs_balance_delayed_items(fs_info);
4173 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4174 BTRFS_DIRTY_METADATA_THRESH,
4175 fs_info->dirty_metadata_batch);
4177 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4181 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4183 __btrfs_btree_balance_dirty(fs_info, 1);
4186 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4188 __btrfs_btree_balance_dirty(fs_info, 0);
4191 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4192 struct btrfs_key *first_key)
4194 return btree_read_extent_buffer_pages(buf, parent_transid,
4198 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4200 /* cleanup FS via transaction */
4201 btrfs_cleanup_transaction(fs_info);
4203 mutex_lock(&fs_info->cleaner_mutex);
4204 btrfs_run_delayed_iputs(fs_info);
4205 mutex_unlock(&fs_info->cleaner_mutex);
4207 down_write(&fs_info->cleanup_work_sem);
4208 up_write(&fs_info->cleanup_work_sem);
4211 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4213 struct btrfs_root *gang[8];
4214 u64 root_objectid = 0;
4217 spin_lock(&fs_info->fs_roots_radix_lock);
4218 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4219 (void **)gang, root_objectid,
4220 ARRAY_SIZE(gang))) != 0) {
4223 for (i = 0; i < ret; i++)
4224 gang[i] = btrfs_grab_root(gang[i]);
4225 spin_unlock(&fs_info->fs_roots_radix_lock);
4227 for (i = 0; i < ret; i++) {
4230 root_objectid = gang[i]->root_key.objectid;
4231 btrfs_free_log(NULL, gang[i]);
4232 btrfs_put_root(gang[i]);
4235 spin_lock(&fs_info->fs_roots_radix_lock);
4237 spin_unlock(&fs_info->fs_roots_radix_lock);
4238 btrfs_free_log_root_tree(NULL, fs_info);
4241 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4243 struct btrfs_ordered_extent *ordered;
4245 spin_lock(&root->ordered_extent_lock);
4247 * This will just short circuit the ordered completion stuff which will
4248 * make sure the ordered extent gets properly cleaned up.
4250 list_for_each_entry(ordered, &root->ordered_extents,
4252 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4253 spin_unlock(&root->ordered_extent_lock);
4256 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4258 struct btrfs_root *root;
4259 struct list_head splice;
4261 INIT_LIST_HEAD(&splice);
4263 spin_lock(&fs_info->ordered_root_lock);
4264 list_splice_init(&fs_info->ordered_roots, &splice);
4265 while (!list_empty(&splice)) {
4266 root = list_first_entry(&splice, struct btrfs_root,
4268 list_move_tail(&root->ordered_root,
4269 &fs_info->ordered_roots);
4271 spin_unlock(&fs_info->ordered_root_lock);
4272 btrfs_destroy_ordered_extents(root);
4275 spin_lock(&fs_info->ordered_root_lock);
4277 spin_unlock(&fs_info->ordered_root_lock);
4280 * We need this here because if we've been flipped read-only we won't
4281 * get sync() from the umount, so we need to make sure any ordered
4282 * extents that haven't had their dirty pages IO start writeout yet
4283 * actually get run and error out properly.
4285 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4288 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4289 struct btrfs_fs_info *fs_info)
4291 struct rb_node *node;
4292 struct btrfs_delayed_ref_root *delayed_refs;
4293 struct btrfs_delayed_ref_node *ref;
4296 delayed_refs = &trans->delayed_refs;
4298 spin_lock(&delayed_refs->lock);
4299 if (atomic_read(&delayed_refs->num_entries) == 0) {
4300 spin_unlock(&delayed_refs->lock);
4301 btrfs_debug(fs_info, "delayed_refs has NO entry");
4305 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4306 struct btrfs_delayed_ref_head *head;
4308 bool pin_bytes = false;
4310 head = rb_entry(node, struct btrfs_delayed_ref_head,
4312 if (btrfs_delayed_ref_lock(delayed_refs, head))
4315 spin_lock(&head->lock);
4316 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4317 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4320 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4321 RB_CLEAR_NODE(&ref->ref_node);
4322 if (!list_empty(&ref->add_list))
4323 list_del(&ref->add_list);
4324 atomic_dec(&delayed_refs->num_entries);
4325 btrfs_put_delayed_ref(ref);
4327 if (head->must_insert_reserved)
4329 btrfs_free_delayed_extent_op(head->extent_op);
4330 btrfs_delete_ref_head(delayed_refs, head);
4331 spin_unlock(&head->lock);
4332 spin_unlock(&delayed_refs->lock);
4333 mutex_unlock(&head->mutex);
4336 struct btrfs_block_group *cache;
4338 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4341 spin_lock(&cache->space_info->lock);
4342 spin_lock(&cache->lock);
4343 cache->pinned += head->num_bytes;
4344 btrfs_space_info_update_bytes_pinned(fs_info,
4345 cache->space_info, head->num_bytes);
4346 cache->reserved -= head->num_bytes;
4347 cache->space_info->bytes_reserved -= head->num_bytes;
4348 spin_unlock(&cache->lock);
4349 spin_unlock(&cache->space_info->lock);
4350 percpu_counter_add_batch(
4351 &cache->space_info->total_bytes_pinned,
4352 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4354 btrfs_put_block_group(cache);
4356 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4357 head->bytenr + head->num_bytes - 1);
4359 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4360 btrfs_put_delayed_ref_head(head);
4362 spin_lock(&delayed_refs->lock);
4364 btrfs_qgroup_destroy_extent_records(trans);
4366 spin_unlock(&delayed_refs->lock);
4371 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4373 struct btrfs_inode *btrfs_inode;
4374 struct list_head splice;
4376 INIT_LIST_HEAD(&splice);
4378 spin_lock(&root->delalloc_lock);
4379 list_splice_init(&root->delalloc_inodes, &splice);
4381 while (!list_empty(&splice)) {
4382 struct inode *inode = NULL;
4383 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4385 __btrfs_del_delalloc_inode(root, btrfs_inode);
4386 spin_unlock(&root->delalloc_lock);
4389 * Make sure we get a live inode and that it'll not disappear
4392 inode = igrab(&btrfs_inode->vfs_inode);
4394 invalidate_inode_pages2(inode->i_mapping);
4397 spin_lock(&root->delalloc_lock);
4399 spin_unlock(&root->delalloc_lock);
4402 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4404 struct btrfs_root *root;
4405 struct list_head splice;
4407 INIT_LIST_HEAD(&splice);
4409 spin_lock(&fs_info->delalloc_root_lock);
4410 list_splice_init(&fs_info->delalloc_roots, &splice);
4411 while (!list_empty(&splice)) {
4412 root = list_first_entry(&splice, struct btrfs_root,
4414 root = btrfs_grab_root(root);
4416 spin_unlock(&fs_info->delalloc_root_lock);
4418 btrfs_destroy_delalloc_inodes(root);
4419 btrfs_put_root(root);
4421 spin_lock(&fs_info->delalloc_root_lock);
4423 spin_unlock(&fs_info->delalloc_root_lock);
4426 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4427 struct extent_io_tree *dirty_pages,
4431 struct extent_buffer *eb;
4436 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4441 clear_extent_bits(dirty_pages, start, end, mark);
4442 while (start <= end) {
4443 eb = find_extent_buffer(fs_info, start);
4444 start += fs_info->nodesize;
4447 wait_on_extent_buffer_writeback(eb);
4449 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4451 clear_extent_buffer_dirty(eb);
4452 free_extent_buffer_stale(eb);
4459 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4460 struct extent_io_tree *unpin)
4467 struct extent_state *cached_state = NULL;
4470 * The btrfs_finish_extent_commit() may get the same range as
4471 * ours between find_first_extent_bit and clear_extent_dirty.
4472 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4473 * the same extent range.
4475 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4476 ret = find_first_extent_bit(unpin, 0, &start, &end,
4477 EXTENT_DIRTY, &cached_state);
4479 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4483 clear_extent_dirty(unpin, start, end, &cached_state);
4484 free_extent_state(cached_state);
4485 btrfs_error_unpin_extent_range(fs_info, start, end);
4486 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4493 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4495 struct inode *inode;
4497 inode = cache->io_ctl.inode;
4499 invalidate_inode_pages2(inode->i_mapping);
4500 BTRFS_I(inode)->generation = 0;
4501 cache->io_ctl.inode = NULL;
4504 btrfs_put_block_group(cache);
4507 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4508 struct btrfs_fs_info *fs_info)
4510 struct btrfs_block_group *cache;
4512 spin_lock(&cur_trans->dirty_bgs_lock);
4513 while (!list_empty(&cur_trans->dirty_bgs)) {
4514 cache = list_first_entry(&cur_trans->dirty_bgs,
4515 struct btrfs_block_group,
4518 if (!list_empty(&cache->io_list)) {
4519 spin_unlock(&cur_trans->dirty_bgs_lock);
4520 list_del_init(&cache->io_list);
4521 btrfs_cleanup_bg_io(cache);
4522 spin_lock(&cur_trans->dirty_bgs_lock);
4525 list_del_init(&cache->dirty_list);
4526 spin_lock(&cache->lock);
4527 cache->disk_cache_state = BTRFS_DC_ERROR;
4528 spin_unlock(&cache->lock);
4530 spin_unlock(&cur_trans->dirty_bgs_lock);
4531 btrfs_put_block_group(cache);
4532 btrfs_delayed_refs_rsv_release(fs_info, 1);
4533 spin_lock(&cur_trans->dirty_bgs_lock);
4535 spin_unlock(&cur_trans->dirty_bgs_lock);
4538 * Refer to the definition of io_bgs member for details why it's safe
4539 * to use it without any locking
4541 while (!list_empty(&cur_trans->io_bgs)) {
4542 cache = list_first_entry(&cur_trans->io_bgs,
4543 struct btrfs_block_group,
4546 list_del_init(&cache->io_list);
4547 spin_lock(&cache->lock);
4548 cache->disk_cache_state = BTRFS_DC_ERROR;
4549 spin_unlock(&cache->lock);
4550 btrfs_cleanup_bg_io(cache);
4554 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4555 struct btrfs_fs_info *fs_info)
4557 struct btrfs_device *dev, *tmp;
4559 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4560 ASSERT(list_empty(&cur_trans->dirty_bgs));
4561 ASSERT(list_empty(&cur_trans->io_bgs));
4563 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4565 list_del_init(&dev->post_commit_list);
4568 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4570 cur_trans->state = TRANS_STATE_COMMIT_START;
4571 wake_up(&fs_info->transaction_blocked_wait);
4573 cur_trans->state = TRANS_STATE_UNBLOCKED;
4574 wake_up(&fs_info->transaction_wait);
4576 btrfs_destroy_delayed_inodes(fs_info);
4578 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4580 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4582 cur_trans->state =TRANS_STATE_COMPLETED;
4583 wake_up(&cur_trans->commit_wait);
4586 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4588 struct btrfs_transaction *t;
4590 mutex_lock(&fs_info->transaction_kthread_mutex);
4592 spin_lock(&fs_info->trans_lock);
4593 while (!list_empty(&fs_info->trans_list)) {
4594 t = list_first_entry(&fs_info->trans_list,
4595 struct btrfs_transaction, list);
4596 if (t->state >= TRANS_STATE_COMMIT_START) {
4597 refcount_inc(&t->use_count);
4598 spin_unlock(&fs_info->trans_lock);
4599 btrfs_wait_for_commit(fs_info, t->transid);
4600 btrfs_put_transaction(t);
4601 spin_lock(&fs_info->trans_lock);
4604 if (t == fs_info->running_transaction) {
4605 t->state = TRANS_STATE_COMMIT_DOING;
4606 spin_unlock(&fs_info->trans_lock);
4608 * We wait for 0 num_writers since we don't hold a trans
4609 * handle open currently for this transaction.
4611 wait_event(t->writer_wait,
4612 atomic_read(&t->num_writers) == 0);
4614 spin_unlock(&fs_info->trans_lock);
4616 btrfs_cleanup_one_transaction(t, fs_info);
4618 spin_lock(&fs_info->trans_lock);
4619 if (t == fs_info->running_transaction)
4620 fs_info->running_transaction = NULL;
4621 list_del_init(&t->list);
4622 spin_unlock(&fs_info->trans_lock);
4624 btrfs_put_transaction(t);
4625 trace_btrfs_transaction_commit(fs_info->tree_root);
4626 spin_lock(&fs_info->trans_lock);
4628 spin_unlock(&fs_info->trans_lock);
4629 btrfs_destroy_all_ordered_extents(fs_info);
4630 btrfs_destroy_delayed_inodes(fs_info);
4631 btrfs_assert_delayed_root_empty(fs_info);
4632 btrfs_destroy_all_delalloc_inodes(fs_info);
4633 btrfs_drop_all_logs(fs_info);
4634 mutex_unlock(&fs_info->transaction_kthread_mutex);
4639 static const struct extent_io_ops btree_extent_io_ops = {
4640 /* mandatory callbacks */
4641 .submit_bio_hook = btree_submit_bio_hook,
4642 .readpage_end_io_hook = btree_readpage_end_io_hook,