1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include <crypto/hash.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
44 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
45 BTRFS_HEADER_FLAG_RELOC |\
46 BTRFS_SUPER_FLAG_ERROR |\
47 BTRFS_SUPER_FLAG_SEEDING |\
48 BTRFS_SUPER_FLAG_METADUMP |\
49 BTRFS_SUPER_FLAG_METADUMP_V2)
51 static const struct extent_io_ops btree_extent_io_ops;
52 static void end_workqueue_fn(struct btrfs_work *work);
53 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 struct btrfs_fs_info *fs_info);
56 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
57 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
58 struct extent_io_tree *dirty_pages,
60 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
61 struct extent_io_tree *pinned_extents);
62 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
63 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66 * btrfs_end_io_wq structs are used to do processing in task context when an IO
67 * is complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
70 struct btrfs_end_io_wq {
74 struct btrfs_fs_info *info;
76 enum btrfs_wq_endio_type metadata;
77 struct btrfs_work work;
80 static struct kmem_cache *btrfs_end_io_wq_cache;
82 int __init btrfs_end_io_wq_init(void)
84 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
85 sizeof(struct btrfs_end_io_wq),
89 if (!btrfs_end_io_wq_cache)
94 void __cold btrfs_end_io_wq_exit(void)
96 kmem_cache_destroy(btrfs_end_io_wq_cache);
100 * async submit bios are used to offload expensive checksumming
101 * onto the worker threads. They checksum file and metadata bios
102 * just before they are sent down the IO stack.
104 struct async_submit_bio {
107 extent_submit_bio_start_t *submit_bio_start;
110 * bio_offset is optional, can be used if the pages in the bio
111 * can't tell us where in the file the bio should go
114 struct btrfs_work work;
119 * Lockdep class keys for extent_buffer->lock's in this root. For a given
120 * eb, the lockdep key is determined by the btrfs_root it belongs to and
121 * the level the eb occupies in the tree.
123 * Different roots are used for different purposes and may nest inside each
124 * other and they require separate keysets. As lockdep keys should be
125 * static, assign keysets according to the purpose of the root as indicated
126 * by btrfs_root->root_key.objectid. This ensures that all special purpose
127 * roots have separate keysets.
129 * Lock-nesting across peer nodes is always done with the immediate parent
130 * node locked thus preventing deadlock. As lockdep doesn't know this, use
131 * subclass to avoid triggering lockdep warning in such cases.
133 * The key is set by the readpage_end_io_hook after the buffer has passed
134 * csum validation but before the pages are unlocked. It is also set by
135 * btrfs_init_new_buffer on freshly allocated blocks.
137 * We also add a check to make sure the highest level of the tree is the
138 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
139 * needs update as well.
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
142 # if BTRFS_MAX_LEVEL != 8
146 static struct btrfs_lockdep_keyset {
147 u64 id; /* root objectid */
148 const char *name_stem; /* lock name stem */
149 char names[BTRFS_MAX_LEVEL + 1][20];
150 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
151 } btrfs_lockdep_keysets[] = {
152 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
153 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
154 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
155 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
156 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
157 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
158 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
159 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
160 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
161 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
162 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
163 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
164 { .id = 0, .name_stem = "tree" },
167 void __init btrfs_init_lockdep(void)
171 /* initialize lockdep class names */
172 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
173 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
175 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
176 snprintf(ks->names[j], sizeof(ks->names[j]),
177 "btrfs-%s-%02d", ks->name_stem, j);
181 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
184 struct btrfs_lockdep_keyset *ks;
186 BUG_ON(level >= ARRAY_SIZE(ks->keys));
188 /* find the matching keyset, id 0 is the default entry */
189 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
190 if (ks->id == objectid)
193 lockdep_set_class_and_name(&eb->lock,
194 &ks->keys[level], ks->names[level]);
200 * extents on the btree inode are pretty simple, there's one extent
201 * that covers the entire device
203 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
204 struct page *page, size_t pg_offset, u64 start, u64 len,
207 struct btrfs_fs_info *fs_info = inode->root->fs_info;
208 struct extent_map_tree *em_tree = &inode->extent_tree;
209 struct extent_map *em;
212 read_lock(&em_tree->lock);
213 em = lookup_extent_mapping(em_tree, start, len);
215 em->bdev = fs_info->fs_devices->latest_bdev;
216 read_unlock(&em_tree->lock);
219 read_unlock(&em_tree->lock);
221 em = alloc_extent_map();
223 em = ERR_PTR(-ENOMEM);
228 em->block_len = (u64)-1;
230 em->bdev = fs_info->fs_devices->latest_bdev;
232 write_lock(&em_tree->lock);
233 ret = add_extent_mapping(em_tree, em, 0);
234 if (ret == -EEXIST) {
236 em = lookup_extent_mapping(em_tree, start, len);
243 write_unlock(&em_tree->lock);
250 * Compute the csum of a btree block and store the result to provided buffer.
252 * Returns error if the extent buffer cannot be mapped.
254 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
256 struct btrfs_fs_info *fs_info = buf->fs_info;
257 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
259 unsigned long cur_len;
260 unsigned long offset = BTRFS_CSUM_SIZE;
262 unsigned long map_start;
263 unsigned long map_len;
266 shash->tfm = fs_info->csum_shash;
267 crypto_shash_init(shash);
269 len = buf->len - offset;
273 * Note: we don't need to check for the err == 1 case here, as
274 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
275 * and 'min_len = 32' and the currently implemented mapping
276 * algorithm we cannot cross a page boundary.
278 err = map_private_extent_buffer(buf, offset, 32,
279 &kaddr, &map_start, &map_len);
282 cur_len = min(len, map_len - (offset - map_start));
283 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
287 memset(result, 0, BTRFS_CSUM_SIZE);
289 crypto_shash_final(shash, result);
295 * we can't consider a given block up to date unless the transid of the
296 * block matches the transid in the parent node's pointer. This is how we
297 * detect blocks that either didn't get written at all or got written
298 * in the wrong place.
300 static int verify_parent_transid(struct extent_io_tree *io_tree,
301 struct extent_buffer *eb, u64 parent_transid,
304 struct extent_state *cached_state = NULL;
306 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
308 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
315 btrfs_tree_read_lock(eb);
316 btrfs_set_lock_blocking_read(eb);
319 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
321 if (extent_buffer_uptodate(eb) &&
322 btrfs_header_generation(eb) == parent_transid) {
326 btrfs_err_rl(eb->fs_info,
327 "parent transid verify failed on %llu wanted %llu found %llu",
329 parent_transid, btrfs_header_generation(eb));
333 * Things reading via commit roots that don't have normal protection,
334 * like send, can have a really old block in cache that may point at a
335 * block that has been freed and re-allocated. So don't clear uptodate
336 * if we find an eb that is under IO (dirty/writeback) because we could
337 * end up reading in the stale data and then writing it back out and
338 * making everybody very sad.
340 if (!extent_buffer_under_io(eb))
341 clear_extent_buffer_uptodate(eb);
343 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
346 btrfs_tree_read_unlock_blocking(eb);
350 static bool btrfs_supported_super_csum(u16 csum_type)
353 case BTRFS_CSUM_TYPE_CRC32:
361 * Return 0 if the superblock checksum type matches the checksum value of that
362 * algorithm. Pass the raw disk superblock data.
364 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
367 struct btrfs_super_block *disk_sb =
368 (struct btrfs_super_block *)raw_disk_sb;
369 char result[BTRFS_CSUM_SIZE];
370 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
372 shash->tfm = fs_info->csum_shash;
373 crypto_shash_init(shash);
376 * The super_block structure does not span the whole
377 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
378 * filled with zeros and is included in the checksum.
380 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
381 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382 crypto_shash_final(shash, result);
384 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
390 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
391 struct btrfs_key *first_key, u64 parent_transid)
393 struct btrfs_fs_info *fs_info = eb->fs_info;
395 struct btrfs_key found_key;
398 found_level = btrfs_header_level(eb);
399 if (found_level != level) {
400 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
401 KERN_ERR "BTRFS: tree level check failed\n");
403 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
404 eb->start, level, found_level);
412 * For live tree block (new tree blocks in current transaction),
413 * we need proper lock context to avoid race, which is impossible here.
414 * So we only checks tree blocks which is read from disk, whose
415 * generation <= fs_info->last_trans_committed.
417 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
420 btrfs_node_key_to_cpu(eb, &found_key, 0);
422 btrfs_item_key_to_cpu(eb, &found_key, 0);
423 ret = btrfs_comp_cpu_keys(first_key, &found_key);
426 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
427 KERN_ERR "BTRFS: tree first key check failed\n");
429 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
430 eb->start, parent_transid, first_key->objectid,
431 first_key->type, first_key->offset,
432 found_key.objectid, found_key.type,
439 * helper to read a given tree block, doing retries as required when
440 * the checksums don't match and we have alternate mirrors to try.
442 * @parent_transid: expected transid, skip check if 0
443 * @level: expected level, mandatory check
444 * @first_key: expected key of first slot, skip check if NULL
446 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
447 u64 parent_transid, int level,
448 struct btrfs_key *first_key)
450 struct btrfs_fs_info *fs_info = eb->fs_info;
451 struct extent_io_tree *io_tree;
456 int failed_mirror = 0;
458 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
460 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
461 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
463 if (verify_parent_transid(io_tree, eb,
466 else if (btrfs_verify_level_key(eb, level,
467 first_key, parent_transid))
473 num_copies = btrfs_num_copies(fs_info,
478 if (!failed_mirror) {
480 failed_mirror = eb->read_mirror;
484 if (mirror_num == failed_mirror)
487 if (mirror_num > num_copies)
491 if (failed && !ret && failed_mirror)
492 btrfs_repair_eb_io_failure(eb, failed_mirror);
498 * checksum a dirty tree block before IO. This has extra checks to make sure
499 * we only fill in the checksum field in the first page of a multi-page block
502 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
504 u64 start = page_offset(page);
506 u8 result[BTRFS_CSUM_SIZE];
507 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
508 struct extent_buffer *eb;
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
515 found_start = btrfs_header_bytenr(eb);
517 * Please do not consolidate these warnings into a single if.
518 * It is useful to know what went wrong.
520 if (WARN_ON(found_start != start))
522 if (WARN_ON(!PageUptodate(page)))
525 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
526 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
528 if (csum_tree_block(eb, result))
531 if (btrfs_header_level(eb))
532 ret = btrfs_check_node(eb);
534 ret = btrfs_check_leaf_full(eb);
538 "block=%llu write time tree block corruption detected",
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, btrfs_header_fsid(), BTRFS_FSID_SIZE);
559 * Checking the incompat flag is only valid for the current
560 * fs. For seed devices it's forbidden to have their uuid
561 * changed so reading ->fsid in this case is fine
563 if (fs_devices == fs_info->fs_devices &&
564 btrfs_fs_incompat(fs_info, METADATA_UUID))
565 metadata_uuid = fs_devices->metadata_uuid;
567 metadata_uuid = fs_devices->fsid;
569 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
573 fs_devices = fs_devices->seed;
578 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
579 u64 phy_offset, struct page *page,
580 u64 start, u64 end, int mirror)
584 struct extent_buffer *eb;
585 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
586 struct btrfs_fs_info *fs_info = root->fs_info;
587 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
589 u8 result[BTRFS_CSUM_SIZE];
595 eb = (struct extent_buffer *)page->private;
597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
600 extent_buffer_get(eb);
602 reads_done = atomic_dec_and_test(&eb->io_pages);
606 eb->read_mirror = mirror;
607 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
612 found_start = btrfs_header_bytenr(eb);
613 if (found_start != eb->start) {
614 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
615 eb->start, found_start);
619 if (check_tree_block_fsid(eb)) {
620 btrfs_err_rl(fs_info, "bad fsid on block %llu",
625 found_level = btrfs_header_level(eb);
626 if (found_level >= BTRFS_MAX_LEVEL) {
627 btrfs_err(fs_info, "bad tree block level %d on %llu",
628 (int)btrfs_header_level(eb), eb->start);
633 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
636 ret = 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;
698 btrfs_work_func_t func;
700 fs_info = end_io_wq->info;
701 end_io_wq->status = bio->bi_status;
703 if (bio_op(bio) == REQ_OP_WRITE) {
704 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
705 wq = fs_info->endio_meta_write_workers;
706 func = btrfs_endio_meta_write_helper;
707 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
708 wq = fs_info->endio_freespace_worker;
709 func = btrfs_freespace_write_helper;
710 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
711 wq = fs_info->endio_raid56_workers;
712 func = btrfs_endio_raid56_helper;
714 wq = fs_info->endio_write_workers;
715 func = btrfs_endio_write_helper;
718 if (unlikely(end_io_wq->metadata ==
719 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
720 wq = fs_info->endio_repair_workers;
721 func = btrfs_endio_repair_helper;
722 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
723 wq = fs_info->endio_raid56_workers;
724 func = btrfs_endio_raid56_helper;
725 } else if (end_io_wq->metadata) {
726 wq = fs_info->endio_meta_workers;
727 func = btrfs_endio_meta_helper;
729 wq = fs_info->endio_workers;
730 func = btrfs_endio_helper;
734 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
735 btrfs_queue_work(wq, &end_io_wq->work);
738 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
739 enum btrfs_wq_endio_type metadata)
741 struct btrfs_end_io_wq *end_io_wq;
743 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
745 return BLK_STS_RESOURCE;
747 end_io_wq->private = bio->bi_private;
748 end_io_wq->end_io = bio->bi_end_io;
749 end_io_wq->info = info;
750 end_io_wq->status = 0;
751 end_io_wq->bio = bio;
752 end_io_wq->metadata = metadata;
754 bio->bi_private = end_io_wq;
755 bio->bi_end_io = end_workqueue_bio;
759 static void run_one_async_start(struct btrfs_work *work)
761 struct async_submit_bio *async;
764 async = container_of(work, struct async_submit_bio, work);
765 ret = async->submit_bio_start(async->private_data, async->bio,
772 * In order to insert checksums into the metadata in large chunks, we wait
773 * until bio submission time. All the pages in the bio are checksummed and
774 * sums are attached onto the ordered extent record.
776 * At IO completion time the csums attached on the ordered extent record are
777 * inserted into the tree.
779 static void run_one_async_done(struct btrfs_work *work)
781 struct async_submit_bio *async;
785 async = container_of(work, struct async_submit_bio, work);
786 inode = async->private_data;
788 /* If an error occurred we just want to clean up the bio and move on */
790 async->bio->bi_status = async->status;
791 bio_endio(async->bio);
795 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
796 async->mirror_num, 1);
798 async->bio->bi_status = ret;
799 bio_endio(async->bio);
803 static void run_one_async_free(struct btrfs_work *work)
805 struct async_submit_bio *async;
807 async = container_of(work, struct async_submit_bio, work);
811 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
812 int mirror_num, unsigned long bio_flags,
813 u64 bio_offset, void *private_data,
814 extent_submit_bio_start_t *submit_bio_start)
816 struct async_submit_bio *async;
818 async = kmalloc(sizeof(*async), GFP_NOFS);
820 return BLK_STS_RESOURCE;
822 async->private_data = private_data;
824 async->mirror_num = mirror_num;
825 async->submit_bio_start = submit_bio_start;
827 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
828 run_one_async_done, run_one_async_free);
830 async->bio_offset = bio_offset;
834 if (op_is_sync(bio->bi_opf))
835 btrfs_set_work_high_priority(&async->work);
837 btrfs_queue_work(fs_info->workers, &async->work);
841 static blk_status_t btree_csum_one_bio(struct bio *bio)
843 struct bio_vec *bvec;
844 struct btrfs_root *root;
846 struct bvec_iter_all iter_all;
848 ASSERT(!bio_flagged(bio, BIO_CLONED));
849 bio_for_each_segment_all(bvec, bio, iter_all) {
850 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
851 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
856 return errno_to_blk_status(ret);
859 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
863 * when we're called for a write, we're already in the async
864 * submission context. Just jump into btrfs_map_bio
866 return btree_csum_one_bio(bio);
869 static int check_async_write(struct btrfs_fs_info *fs_info,
870 struct btrfs_inode *bi)
872 if (atomic_read(&bi->sync_writers))
874 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
879 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
881 unsigned long bio_flags)
883 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
884 int async = check_async_write(fs_info, BTRFS_I(inode));
887 if (bio_op(bio) != REQ_OP_WRITE) {
889 * called for a read, do the setup so that checksum validation
890 * can happen in the async kernel threads
892 ret = btrfs_bio_wq_end_io(fs_info, bio,
893 BTRFS_WQ_ENDIO_METADATA);
896 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
898 ret = btree_csum_one_bio(bio);
901 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
904 * kthread helpers are used to submit writes so that
905 * checksumming can happen in parallel across all CPUs
907 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
908 0, inode, btree_submit_bio_start);
916 bio->bi_status = ret;
921 #ifdef CONFIG_MIGRATION
922 static int btree_migratepage(struct address_space *mapping,
923 struct page *newpage, struct page *page,
924 enum migrate_mode mode)
927 * we can't safely write a btree page from here,
928 * we haven't done the locking hook
933 * Buffers may be managed in a filesystem specific way.
934 * We must have no buffers or drop them.
936 if (page_has_private(page) &&
937 !try_to_release_page(page, GFP_KERNEL))
939 return migrate_page(mapping, newpage, page, mode);
944 static int btree_writepages(struct address_space *mapping,
945 struct writeback_control *wbc)
947 struct btrfs_fs_info *fs_info;
950 if (wbc->sync_mode == WB_SYNC_NONE) {
952 if (wbc->for_kupdate)
955 fs_info = BTRFS_I(mapping->host)->root->fs_info;
956 /* this is a bit racy, but that's ok */
957 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
958 BTRFS_DIRTY_METADATA_THRESH,
959 fs_info->dirty_metadata_batch);
963 return btree_write_cache_pages(mapping, wbc);
966 static int btree_readpage(struct file *file, struct page *page)
968 struct extent_io_tree *tree;
969 tree = &BTRFS_I(page->mapping->host)->io_tree;
970 return extent_read_full_page(tree, page, btree_get_extent, 0);
973 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
975 if (PageWriteback(page) || PageDirty(page))
978 return try_release_extent_buffer(page);
981 static void btree_invalidatepage(struct page *page, unsigned int offset,
984 struct extent_io_tree *tree;
985 tree = &BTRFS_I(page->mapping->host)->io_tree;
986 extent_invalidatepage(tree, page, offset);
987 btree_releasepage(page, GFP_NOFS);
988 if (PagePrivate(page)) {
989 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
990 "page private not zero on page %llu",
991 (unsigned long long)page_offset(page));
992 ClearPagePrivate(page);
993 set_page_private(page, 0);
998 static int btree_set_page_dirty(struct page *page)
1001 struct extent_buffer *eb;
1003 BUG_ON(!PagePrivate(page));
1004 eb = (struct extent_buffer *)page->private;
1006 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1007 BUG_ON(!atomic_read(&eb->refs));
1008 btrfs_assert_tree_locked(eb);
1010 return __set_page_dirty_nobuffers(page);
1013 static const struct address_space_operations btree_aops = {
1014 .readpage = btree_readpage,
1015 .writepages = btree_writepages,
1016 .releasepage = btree_releasepage,
1017 .invalidatepage = btree_invalidatepage,
1018 #ifdef CONFIG_MIGRATION
1019 .migratepage = btree_migratepage,
1021 .set_page_dirty = btree_set_page_dirty,
1024 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1026 struct extent_buffer *buf = NULL;
1029 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1033 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1035 free_extent_buffer_stale(buf);
1037 free_extent_buffer(buf);
1040 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1041 int mirror_num, struct extent_buffer **eb)
1043 struct extent_buffer *buf = NULL;
1046 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1050 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1052 ret = read_extent_buffer_pages(buf, WAIT_PAGE_LOCK, mirror_num);
1054 free_extent_buffer_stale(buf);
1058 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1059 free_extent_buffer_stale(buf);
1061 } else if (extent_buffer_uptodate(buf)) {
1064 free_extent_buffer(buf);
1069 struct extent_buffer *btrfs_find_create_tree_block(
1070 struct btrfs_fs_info *fs_info,
1073 if (btrfs_is_testing(fs_info))
1074 return alloc_test_extent_buffer(fs_info, bytenr);
1075 return alloc_extent_buffer(fs_info, bytenr);
1079 * Read tree block at logical address @bytenr and do variant basic but critical
1082 * @parent_transid: expected transid of this tree block, skip check if 0
1083 * @level: expected level, mandatory check
1084 * @first_key: expected key in slot 0, skip check if NULL
1086 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1087 u64 parent_transid, int level,
1088 struct btrfs_key *first_key)
1090 struct extent_buffer *buf = NULL;
1093 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1097 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1100 free_extent_buffer_stale(buf);
1101 return ERR_PTR(ret);
1107 void btrfs_clean_tree_block(struct extent_buffer *buf)
1109 struct btrfs_fs_info *fs_info = buf->fs_info;
1110 if (btrfs_header_generation(buf) ==
1111 fs_info->running_transaction->transid) {
1112 btrfs_assert_tree_locked(buf);
1114 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1115 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1117 fs_info->dirty_metadata_batch);
1118 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1119 btrfs_set_lock_blocking_write(buf);
1120 clear_extent_buffer_dirty(buf);
1125 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1127 struct btrfs_subvolume_writers *writers;
1130 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1132 return ERR_PTR(-ENOMEM);
1134 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1137 return ERR_PTR(ret);
1140 init_waitqueue_head(&writers->wait);
1145 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1147 percpu_counter_destroy(&writers->counter);
1151 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1154 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1156 root->commit_root = NULL;
1158 root->orphan_cleanup_state = 0;
1160 root->last_trans = 0;
1161 root->highest_objectid = 0;
1162 root->nr_delalloc_inodes = 0;
1163 root->nr_ordered_extents = 0;
1164 root->inode_tree = RB_ROOT;
1165 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1166 root->block_rsv = NULL;
1168 INIT_LIST_HEAD(&root->dirty_list);
1169 INIT_LIST_HEAD(&root->root_list);
1170 INIT_LIST_HEAD(&root->delalloc_inodes);
1171 INIT_LIST_HEAD(&root->delalloc_root);
1172 INIT_LIST_HEAD(&root->ordered_extents);
1173 INIT_LIST_HEAD(&root->ordered_root);
1174 INIT_LIST_HEAD(&root->reloc_dirty_list);
1175 INIT_LIST_HEAD(&root->logged_list[0]);
1176 INIT_LIST_HEAD(&root->logged_list[1]);
1177 spin_lock_init(&root->inode_lock);
1178 spin_lock_init(&root->delalloc_lock);
1179 spin_lock_init(&root->ordered_extent_lock);
1180 spin_lock_init(&root->accounting_lock);
1181 spin_lock_init(&root->log_extents_lock[0]);
1182 spin_lock_init(&root->log_extents_lock[1]);
1183 spin_lock_init(&root->qgroup_meta_rsv_lock);
1184 mutex_init(&root->objectid_mutex);
1185 mutex_init(&root->log_mutex);
1186 mutex_init(&root->ordered_extent_mutex);
1187 mutex_init(&root->delalloc_mutex);
1188 init_waitqueue_head(&root->log_writer_wait);
1189 init_waitqueue_head(&root->log_commit_wait[0]);
1190 init_waitqueue_head(&root->log_commit_wait[1]);
1191 INIT_LIST_HEAD(&root->log_ctxs[0]);
1192 INIT_LIST_HEAD(&root->log_ctxs[1]);
1193 atomic_set(&root->log_commit[0], 0);
1194 atomic_set(&root->log_commit[1], 0);
1195 atomic_set(&root->log_writers, 0);
1196 atomic_set(&root->log_batch, 0);
1197 refcount_set(&root->refs, 1);
1198 atomic_set(&root->will_be_snapshotted, 0);
1199 atomic_set(&root->snapshot_force_cow, 0);
1200 atomic_set(&root->nr_swapfiles, 0);
1201 root->log_transid = 0;
1202 root->log_transid_committed = -1;
1203 root->last_log_commit = 0;
1205 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1206 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1208 memset(&root->root_key, 0, sizeof(root->root_key));
1209 memset(&root->root_item, 0, sizeof(root->root_item));
1210 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1212 root->defrag_trans_start = fs_info->generation;
1214 root->defrag_trans_start = 0;
1215 root->root_key.objectid = objectid;
1218 spin_lock_init(&root->root_item_lock);
1219 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1222 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1225 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1227 root->fs_info = fs_info;
1231 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1232 /* Should only be used by the testing infrastructure */
1233 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1235 struct btrfs_root *root;
1238 return ERR_PTR(-EINVAL);
1240 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1242 return ERR_PTR(-ENOMEM);
1244 /* We don't use the stripesize in selftest, set it as sectorsize */
1245 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1246 root->alloc_bytenr = 0;
1252 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1255 struct btrfs_fs_info *fs_info = trans->fs_info;
1256 struct extent_buffer *leaf;
1257 struct btrfs_root *tree_root = fs_info->tree_root;
1258 struct btrfs_root *root;
1259 struct btrfs_key key;
1260 unsigned int nofs_flag;
1262 uuid_le uuid = NULL_UUID_LE;
1265 * We're holding a transaction handle, so use a NOFS memory allocation
1266 * context to avoid deadlock if reclaim happens.
1268 nofs_flag = memalloc_nofs_save();
1269 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1270 memalloc_nofs_restore(nofs_flag);
1272 return ERR_PTR(-ENOMEM);
1274 __setup_root(root, fs_info, objectid);
1275 root->root_key.objectid = objectid;
1276 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1277 root->root_key.offset = 0;
1279 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1281 ret = PTR_ERR(leaf);
1287 btrfs_mark_buffer_dirty(leaf);
1289 root->commit_root = btrfs_root_node(root);
1290 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1292 root->root_item.flags = 0;
1293 root->root_item.byte_limit = 0;
1294 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1295 btrfs_set_root_generation(&root->root_item, trans->transid);
1296 btrfs_set_root_level(&root->root_item, 0);
1297 btrfs_set_root_refs(&root->root_item, 1);
1298 btrfs_set_root_used(&root->root_item, leaf->len);
1299 btrfs_set_root_last_snapshot(&root->root_item, 0);
1300 btrfs_set_root_dirid(&root->root_item, 0);
1301 if (is_fstree(objectid))
1303 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1304 root->root_item.drop_level = 0;
1306 key.objectid = objectid;
1307 key.type = BTRFS_ROOT_ITEM_KEY;
1309 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1313 btrfs_tree_unlock(leaf);
1319 btrfs_tree_unlock(leaf);
1320 free_extent_buffer(root->commit_root);
1321 free_extent_buffer(leaf);
1325 return ERR_PTR(ret);
1328 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1329 struct btrfs_fs_info *fs_info)
1331 struct btrfs_root *root;
1332 struct extent_buffer *leaf;
1334 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1336 return ERR_PTR(-ENOMEM);
1338 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1340 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1341 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1342 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1345 * DON'T set REF_COWS for log trees
1347 * log trees do not get reference counted because they go away
1348 * before a real commit is actually done. They do store pointers
1349 * to file data extents, and those reference counts still get
1350 * updated (along with back refs to the log tree).
1353 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1357 return ERR_CAST(leaf);
1362 btrfs_mark_buffer_dirty(root->node);
1363 btrfs_tree_unlock(root->node);
1367 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1368 struct btrfs_fs_info *fs_info)
1370 struct btrfs_root *log_root;
1372 log_root = alloc_log_tree(trans, fs_info);
1373 if (IS_ERR(log_root))
1374 return PTR_ERR(log_root);
1375 WARN_ON(fs_info->log_root_tree);
1376 fs_info->log_root_tree = log_root;
1380 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1381 struct btrfs_root *root)
1383 struct btrfs_fs_info *fs_info = root->fs_info;
1384 struct btrfs_root *log_root;
1385 struct btrfs_inode_item *inode_item;
1387 log_root = alloc_log_tree(trans, fs_info);
1388 if (IS_ERR(log_root))
1389 return PTR_ERR(log_root);
1391 log_root->last_trans = trans->transid;
1392 log_root->root_key.offset = root->root_key.objectid;
1394 inode_item = &log_root->root_item.inode;
1395 btrfs_set_stack_inode_generation(inode_item, 1);
1396 btrfs_set_stack_inode_size(inode_item, 3);
1397 btrfs_set_stack_inode_nlink(inode_item, 1);
1398 btrfs_set_stack_inode_nbytes(inode_item,
1400 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1402 btrfs_set_root_node(&log_root->root_item, log_root->node);
1404 WARN_ON(root->log_root);
1405 root->log_root = log_root;
1406 root->log_transid = 0;
1407 root->log_transid_committed = -1;
1408 root->last_log_commit = 0;
1412 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1413 struct btrfs_key *key)
1415 struct btrfs_root *root;
1416 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1417 struct btrfs_path *path;
1422 path = btrfs_alloc_path();
1424 return ERR_PTR(-ENOMEM);
1426 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1432 __setup_root(root, fs_info, key->objectid);
1434 ret = btrfs_find_root(tree_root, key, path,
1435 &root->root_item, &root->root_key);
1442 generation = btrfs_root_generation(&root->root_item);
1443 level = btrfs_root_level(&root->root_item);
1444 root->node = read_tree_block(fs_info,
1445 btrfs_root_bytenr(&root->root_item),
1446 generation, level, NULL);
1447 if (IS_ERR(root->node)) {
1448 ret = PTR_ERR(root->node);
1450 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1452 free_extent_buffer(root->node);
1455 root->commit_root = btrfs_root_node(root);
1457 btrfs_free_path(path);
1463 root = ERR_PTR(ret);
1467 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1468 struct btrfs_key *location)
1470 struct btrfs_root *root;
1472 root = btrfs_read_tree_root(tree_root, location);
1476 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1477 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1478 btrfs_check_and_init_root_item(&root->root_item);
1484 int btrfs_init_fs_root(struct btrfs_root *root)
1487 struct btrfs_subvolume_writers *writers;
1489 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1490 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1492 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1497 writers = btrfs_alloc_subvolume_writers();
1498 if (IS_ERR(writers)) {
1499 ret = PTR_ERR(writers);
1502 root->subv_writers = writers;
1504 btrfs_init_free_ino_ctl(root);
1505 spin_lock_init(&root->ino_cache_lock);
1506 init_waitqueue_head(&root->ino_cache_wait);
1508 ret = get_anon_bdev(&root->anon_dev);
1512 mutex_lock(&root->objectid_mutex);
1513 ret = btrfs_find_highest_objectid(root,
1514 &root->highest_objectid);
1516 mutex_unlock(&root->objectid_mutex);
1520 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1522 mutex_unlock(&root->objectid_mutex);
1526 /* The caller is responsible to call btrfs_free_fs_root */
1530 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1533 struct btrfs_root *root;
1535 spin_lock(&fs_info->fs_roots_radix_lock);
1536 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1537 (unsigned long)root_id);
1538 spin_unlock(&fs_info->fs_roots_radix_lock);
1542 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1543 struct btrfs_root *root)
1547 ret = radix_tree_preload(GFP_NOFS);
1551 spin_lock(&fs_info->fs_roots_radix_lock);
1552 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1553 (unsigned long)root->root_key.objectid,
1556 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1557 spin_unlock(&fs_info->fs_roots_radix_lock);
1558 radix_tree_preload_end();
1563 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1564 struct btrfs_key *location,
1567 struct btrfs_root *root;
1568 struct btrfs_path *path;
1569 struct btrfs_key key;
1572 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1573 return fs_info->tree_root;
1574 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1575 return fs_info->extent_root;
1576 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1577 return fs_info->chunk_root;
1578 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1579 return fs_info->dev_root;
1580 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1581 return fs_info->csum_root;
1582 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1583 return fs_info->quota_root ? fs_info->quota_root :
1585 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1586 return fs_info->uuid_root ? fs_info->uuid_root :
1588 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1589 return fs_info->free_space_root ? fs_info->free_space_root :
1592 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1594 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1595 return ERR_PTR(-ENOENT);
1599 root = btrfs_read_fs_root(fs_info->tree_root, location);
1603 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1608 ret = btrfs_init_fs_root(root);
1612 path = btrfs_alloc_path();
1617 key.objectid = BTRFS_ORPHAN_OBJECTID;
1618 key.type = BTRFS_ORPHAN_ITEM_KEY;
1619 key.offset = location->objectid;
1621 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1622 btrfs_free_path(path);
1626 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1628 ret = btrfs_insert_fs_root(fs_info, root);
1630 if (ret == -EEXIST) {
1631 btrfs_free_fs_root(root);
1638 btrfs_free_fs_root(root);
1639 return ERR_PTR(ret);
1642 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1644 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1646 struct btrfs_device *device;
1647 struct backing_dev_info *bdi;
1650 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1653 bdi = device->bdev->bd_bdi;
1654 if (bdi_congested(bdi, bdi_bits)) {
1664 * called by the kthread helper functions to finally call the bio end_io
1665 * functions. This is where read checksum verification actually happens
1667 static void end_workqueue_fn(struct btrfs_work *work)
1670 struct btrfs_end_io_wq *end_io_wq;
1672 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1673 bio = end_io_wq->bio;
1675 bio->bi_status = end_io_wq->status;
1676 bio->bi_private = end_io_wq->private;
1677 bio->bi_end_io = end_io_wq->end_io;
1678 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1682 static int cleaner_kthread(void *arg)
1684 struct btrfs_root *root = arg;
1685 struct btrfs_fs_info *fs_info = root->fs_info;
1691 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1693 /* Make the cleaner go to sleep early. */
1694 if (btrfs_need_cleaner_sleep(fs_info))
1698 * Do not do anything if we might cause open_ctree() to block
1699 * before we have finished mounting the filesystem.
1701 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1704 if (!mutex_trylock(&fs_info->cleaner_mutex))
1708 * Avoid the problem that we change the status of the fs
1709 * during the above check and trylock.
1711 if (btrfs_need_cleaner_sleep(fs_info)) {
1712 mutex_unlock(&fs_info->cleaner_mutex);
1716 btrfs_run_delayed_iputs(fs_info);
1718 again = btrfs_clean_one_deleted_snapshot(root);
1719 mutex_unlock(&fs_info->cleaner_mutex);
1722 * The defragger has dealt with the R/O remount and umount,
1723 * needn't do anything special here.
1725 btrfs_run_defrag_inodes(fs_info);
1728 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1729 * with relocation (btrfs_relocate_chunk) and relocation
1730 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1731 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1732 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1733 * unused block groups.
1735 btrfs_delete_unused_bgs(fs_info);
1737 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1738 if (kthread_should_park())
1740 if (kthread_should_stop())
1743 set_current_state(TASK_INTERRUPTIBLE);
1745 __set_current_state(TASK_RUNNING);
1750 static int transaction_kthread(void *arg)
1752 struct btrfs_root *root = arg;
1753 struct btrfs_fs_info *fs_info = root->fs_info;
1754 struct btrfs_trans_handle *trans;
1755 struct btrfs_transaction *cur;
1758 unsigned long delay;
1762 cannot_commit = false;
1763 delay = HZ * fs_info->commit_interval;
1764 mutex_lock(&fs_info->transaction_kthread_mutex);
1766 spin_lock(&fs_info->trans_lock);
1767 cur = fs_info->running_transaction;
1769 spin_unlock(&fs_info->trans_lock);
1773 now = ktime_get_seconds();
1774 if (cur->state < TRANS_STATE_BLOCKED &&
1775 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1776 (now < cur->start_time ||
1777 now - cur->start_time < fs_info->commit_interval)) {
1778 spin_unlock(&fs_info->trans_lock);
1782 transid = cur->transid;
1783 spin_unlock(&fs_info->trans_lock);
1785 /* If the file system is aborted, this will always fail. */
1786 trans = btrfs_attach_transaction(root);
1787 if (IS_ERR(trans)) {
1788 if (PTR_ERR(trans) != -ENOENT)
1789 cannot_commit = true;
1792 if (transid == trans->transid) {
1793 btrfs_commit_transaction(trans);
1795 btrfs_end_transaction(trans);
1798 wake_up_process(fs_info->cleaner_kthread);
1799 mutex_unlock(&fs_info->transaction_kthread_mutex);
1801 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1802 &fs_info->fs_state)))
1803 btrfs_cleanup_transaction(fs_info);
1804 if (!kthread_should_stop() &&
1805 (!btrfs_transaction_blocked(fs_info) ||
1807 schedule_timeout_interruptible(delay);
1808 } while (!kthread_should_stop());
1813 * this will find the highest generation in the array of
1814 * root backups. The index of the highest array is returned,
1815 * or -1 if we can't find anything.
1817 * We check to make sure the array is valid by comparing the
1818 * generation of the latest root in the array with the generation
1819 * in the super block. If they don't match we pitch it.
1821 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1824 int newest_index = -1;
1825 struct btrfs_root_backup *root_backup;
1828 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1829 root_backup = info->super_copy->super_roots + i;
1830 cur = btrfs_backup_tree_root_gen(root_backup);
1831 if (cur == newest_gen)
1835 /* check to see if we actually wrapped around */
1836 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1837 root_backup = info->super_copy->super_roots;
1838 cur = btrfs_backup_tree_root_gen(root_backup);
1839 if (cur == newest_gen)
1842 return newest_index;
1847 * find the oldest backup so we know where to store new entries
1848 * in the backup array. This will set the backup_root_index
1849 * field in the fs_info struct
1851 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1854 int newest_index = -1;
1856 newest_index = find_newest_super_backup(info, newest_gen);
1857 /* if there was garbage in there, just move along */
1858 if (newest_index == -1) {
1859 info->backup_root_index = 0;
1861 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1866 * copy all the root pointers into the super backup array.
1867 * this will bump the backup pointer by one when it is
1870 static void backup_super_roots(struct btrfs_fs_info *info)
1873 struct btrfs_root_backup *root_backup;
1876 next_backup = info->backup_root_index;
1877 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1878 BTRFS_NUM_BACKUP_ROOTS;
1881 * just overwrite the last backup if we're at the same generation
1882 * this happens only at umount
1884 root_backup = info->super_for_commit->super_roots + last_backup;
1885 if (btrfs_backup_tree_root_gen(root_backup) ==
1886 btrfs_header_generation(info->tree_root->node))
1887 next_backup = last_backup;
1889 root_backup = info->super_for_commit->super_roots + next_backup;
1892 * make sure all of our padding and empty slots get zero filled
1893 * regardless of which ones we use today
1895 memset(root_backup, 0, sizeof(*root_backup));
1897 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1899 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1900 btrfs_set_backup_tree_root_gen(root_backup,
1901 btrfs_header_generation(info->tree_root->node));
1903 btrfs_set_backup_tree_root_level(root_backup,
1904 btrfs_header_level(info->tree_root->node));
1906 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1907 btrfs_set_backup_chunk_root_gen(root_backup,
1908 btrfs_header_generation(info->chunk_root->node));
1909 btrfs_set_backup_chunk_root_level(root_backup,
1910 btrfs_header_level(info->chunk_root->node));
1912 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1913 btrfs_set_backup_extent_root_gen(root_backup,
1914 btrfs_header_generation(info->extent_root->node));
1915 btrfs_set_backup_extent_root_level(root_backup,
1916 btrfs_header_level(info->extent_root->node));
1919 * we might commit during log recovery, which happens before we set
1920 * the fs_root. Make sure it is valid before we fill it in.
1922 if (info->fs_root && info->fs_root->node) {
1923 btrfs_set_backup_fs_root(root_backup,
1924 info->fs_root->node->start);
1925 btrfs_set_backup_fs_root_gen(root_backup,
1926 btrfs_header_generation(info->fs_root->node));
1927 btrfs_set_backup_fs_root_level(root_backup,
1928 btrfs_header_level(info->fs_root->node));
1931 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1932 btrfs_set_backup_dev_root_gen(root_backup,
1933 btrfs_header_generation(info->dev_root->node));
1934 btrfs_set_backup_dev_root_level(root_backup,
1935 btrfs_header_level(info->dev_root->node));
1937 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1938 btrfs_set_backup_csum_root_gen(root_backup,
1939 btrfs_header_generation(info->csum_root->node));
1940 btrfs_set_backup_csum_root_level(root_backup,
1941 btrfs_header_level(info->csum_root->node));
1943 btrfs_set_backup_total_bytes(root_backup,
1944 btrfs_super_total_bytes(info->super_copy));
1945 btrfs_set_backup_bytes_used(root_backup,
1946 btrfs_super_bytes_used(info->super_copy));
1947 btrfs_set_backup_num_devices(root_backup,
1948 btrfs_super_num_devices(info->super_copy));
1951 * if we don't copy this out to the super_copy, it won't get remembered
1952 * for the next commit
1954 memcpy(&info->super_copy->super_roots,
1955 &info->super_for_commit->super_roots,
1956 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1960 * this copies info out of the root backup array and back into
1961 * the in-memory super block. It is meant to help iterate through
1962 * the array, so you send it the number of backups you've already
1963 * tried and the last backup index you used.
1965 * this returns -1 when it has tried all the backups
1967 static noinline int next_root_backup(struct btrfs_fs_info *info,
1968 struct btrfs_super_block *super,
1969 int *num_backups_tried, int *backup_index)
1971 struct btrfs_root_backup *root_backup;
1972 int newest = *backup_index;
1974 if (*num_backups_tried == 0) {
1975 u64 gen = btrfs_super_generation(super);
1977 newest = find_newest_super_backup(info, gen);
1981 *backup_index = newest;
1982 *num_backups_tried = 1;
1983 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1984 /* we've tried all the backups, all done */
1987 /* jump to the next oldest backup */
1988 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1989 BTRFS_NUM_BACKUP_ROOTS;
1990 *backup_index = newest;
1991 *num_backups_tried += 1;
1993 root_backup = super->super_roots + newest;
1995 btrfs_set_super_generation(super,
1996 btrfs_backup_tree_root_gen(root_backup));
1997 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1998 btrfs_set_super_root_level(super,
1999 btrfs_backup_tree_root_level(root_backup));
2000 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2003 * fixme: the total bytes and num_devices need to match or we should
2006 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2007 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2011 /* helper to cleanup workers */
2012 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2014 btrfs_destroy_workqueue(fs_info->fixup_workers);
2015 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2016 btrfs_destroy_workqueue(fs_info->workers);
2017 btrfs_destroy_workqueue(fs_info->endio_workers);
2018 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2019 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2020 btrfs_destroy_workqueue(fs_info->rmw_workers);
2021 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2022 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2023 btrfs_destroy_workqueue(fs_info->submit_workers);
2024 btrfs_destroy_workqueue(fs_info->delayed_workers);
2025 btrfs_destroy_workqueue(fs_info->caching_workers);
2026 btrfs_destroy_workqueue(fs_info->readahead_workers);
2027 btrfs_destroy_workqueue(fs_info->flush_workers);
2028 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2029 btrfs_destroy_workqueue(fs_info->extent_workers);
2031 * Now that all other work queues are destroyed, we can safely destroy
2032 * the queues used for metadata I/O, since tasks from those other work
2033 * queues can do metadata I/O operations.
2035 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2036 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2039 static void free_root_extent_buffers(struct btrfs_root *root)
2042 free_extent_buffer(root->node);
2043 free_extent_buffer(root->commit_root);
2045 root->commit_root = NULL;
2049 /* helper to cleanup tree roots */
2050 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2052 free_root_extent_buffers(info->tree_root);
2054 free_root_extent_buffers(info->dev_root);
2055 free_root_extent_buffers(info->extent_root);
2056 free_root_extent_buffers(info->csum_root);
2057 free_root_extent_buffers(info->quota_root);
2058 free_root_extent_buffers(info->uuid_root);
2060 free_root_extent_buffers(info->chunk_root);
2061 free_root_extent_buffers(info->free_space_root);
2064 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2067 struct btrfs_root *gang[8];
2070 while (!list_empty(&fs_info->dead_roots)) {
2071 gang[0] = list_entry(fs_info->dead_roots.next,
2072 struct btrfs_root, root_list);
2073 list_del(&gang[0]->root_list);
2075 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2076 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2078 free_extent_buffer(gang[0]->node);
2079 free_extent_buffer(gang[0]->commit_root);
2080 btrfs_put_fs_root(gang[0]);
2085 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2090 for (i = 0; i < ret; i++)
2091 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2094 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2095 btrfs_free_log_root_tree(NULL, fs_info);
2096 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2100 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2102 mutex_init(&fs_info->scrub_lock);
2103 atomic_set(&fs_info->scrubs_running, 0);
2104 atomic_set(&fs_info->scrub_pause_req, 0);
2105 atomic_set(&fs_info->scrubs_paused, 0);
2106 atomic_set(&fs_info->scrub_cancel_req, 0);
2107 init_waitqueue_head(&fs_info->scrub_pause_wait);
2108 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2111 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2113 spin_lock_init(&fs_info->balance_lock);
2114 mutex_init(&fs_info->balance_mutex);
2115 atomic_set(&fs_info->balance_pause_req, 0);
2116 atomic_set(&fs_info->balance_cancel_req, 0);
2117 fs_info->balance_ctl = NULL;
2118 init_waitqueue_head(&fs_info->balance_wait_q);
2121 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2123 struct inode *inode = fs_info->btree_inode;
2125 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2126 set_nlink(inode, 1);
2128 * we set the i_size on the btree inode to the max possible int.
2129 * the real end of the address space is determined by all of
2130 * the devices in the system
2132 inode->i_size = OFFSET_MAX;
2133 inode->i_mapping->a_ops = &btree_aops;
2135 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2136 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2137 IO_TREE_INODE_IO, inode);
2138 BTRFS_I(inode)->io_tree.track_uptodate = false;
2139 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2141 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2143 BTRFS_I(inode)->root = fs_info->tree_root;
2144 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2145 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2146 btrfs_insert_inode_hash(inode);
2149 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2151 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2152 init_rwsem(&fs_info->dev_replace.rwsem);
2153 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2156 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2158 spin_lock_init(&fs_info->qgroup_lock);
2159 mutex_init(&fs_info->qgroup_ioctl_lock);
2160 fs_info->qgroup_tree = RB_ROOT;
2161 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2162 fs_info->qgroup_seq = 1;
2163 fs_info->qgroup_ulist = NULL;
2164 fs_info->qgroup_rescan_running = false;
2165 mutex_init(&fs_info->qgroup_rescan_lock);
2168 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2169 struct btrfs_fs_devices *fs_devices)
2171 u32 max_active = fs_info->thread_pool_size;
2172 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2175 btrfs_alloc_workqueue(fs_info, "worker",
2176 flags | WQ_HIGHPRI, max_active, 16);
2178 fs_info->delalloc_workers =
2179 btrfs_alloc_workqueue(fs_info, "delalloc",
2180 flags, max_active, 2);
2182 fs_info->flush_workers =
2183 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2184 flags, max_active, 0);
2186 fs_info->caching_workers =
2187 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2190 * a higher idle thresh on the submit workers makes it much more
2191 * likely that bios will be send down in a sane order to the
2194 fs_info->submit_workers =
2195 btrfs_alloc_workqueue(fs_info, "submit", flags,
2196 min_t(u64, fs_devices->num_devices,
2199 fs_info->fixup_workers =
2200 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2203 * endios are largely parallel and should have a very
2206 fs_info->endio_workers =
2207 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2208 fs_info->endio_meta_workers =
2209 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2211 fs_info->endio_meta_write_workers =
2212 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2214 fs_info->endio_raid56_workers =
2215 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2217 fs_info->endio_repair_workers =
2218 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2219 fs_info->rmw_workers =
2220 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2221 fs_info->endio_write_workers =
2222 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2224 fs_info->endio_freespace_worker =
2225 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2227 fs_info->delayed_workers =
2228 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2230 fs_info->readahead_workers =
2231 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2233 fs_info->qgroup_rescan_workers =
2234 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2235 fs_info->extent_workers =
2236 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2237 min_t(u64, fs_devices->num_devices,
2240 if (!(fs_info->workers && fs_info->delalloc_workers &&
2241 fs_info->submit_workers && fs_info->flush_workers &&
2242 fs_info->endio_workers && fs_info->endio_meta_workers &&
2243 fs_info->endio_meta_write_workers &&
2244 fs_info->endio_repair_workers &&
2245 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2246 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2247 fs_info->caching_workers && fs_info->readahead_workers &&
2248 fs_info->fixup_workers && fs_info->delayed_workers &&
2249 fs_info->extent_workers &&
2250 fs_info->qgroup_rescan_workers)) {
2257 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2259 struct crypto_shash *csum_shash;
2260 const char *csum_name = btrfs_super_csum_name(csum_type);
2262 csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2264 if (IS_ERR(csum_shash)) {
2265 btrfs_err(fs_info, "error allocating %s hash for checksum",
2267 return PTR_ERR(csum_shash);
2270 fs_info->csum_shash = csum_shash;
2275 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2277 crypto_free_shash(fs_info->csum_shash);
2280 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2281 struct btrfs_fs_devices *fs_devices)
2284 struct btrfs_root *log_tree_root;
2285 struct btrfs_super_block *disk_super = fs_info->super_copy;
2286 u64 bytenr = btrfs_super_log_root(disk_super);
2287 int level = btrfs_super_log_root_level(disk_super);
2289 if (fs_devices->rw_devices == 0) {
2290 btrfs_warn(fs_info, "log replay required on RO media");
2294 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2298 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2300 log_tree_root->node = read_tree_block(fs_info, bytenr,
2301 fs_info->generation + 1,
2303 if (IS_ERR(log_tree_root->node)) {
2304 btrfs_warn(fs_info, "failed to read log tree");
2305 ret = PTR_ERR(log_tree_root->node);
2306 kfree(log_tree_root);
2308 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2309 btrfs_err(fs_info, "failed to read log tree");
2310 free_extent_buffer(log_tree_root->node);
2311 kfree(log_tree_root);
2314 /* returns with log_tree_root freed on success */
2315 ret = btrfs_recover_log_trees(log_tree_root);
2317 btrfs_handle_fs_error(fs_info, ret,
2318 "Failed to recover log tree");
2319 free_extent_buffer(log_tree_root->node);
2320 kfree(log_tree_root);
2324 if (sb_rdonly(fs_info->sb)) {
2325 ret = btrfs_commit_super(fs_info);
2333 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2335 struct btrfs_root *tree_root = fs_info->tree_root;
2336 struct btrfs_root *root;
2337 struct btrfs_key location;
2340 BUG_ON(!fs_info->tree_root);
2342 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2343 location.type = BTRFS_ROOT_ITEM_KEY;
2344 location.offset = 0;
2346 root = btrfs_read_tree_root(tree_root, &location);
2348 ret = PTR_ERR(root);
2351 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2352 fs_info->extent_root = root;
2354 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2355 root = btrfs_read_tree_root(tree_root, &location);
2357 ret = PTR_ERR(root);
2360 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2361 fs_info->dev_root = root;
2362 btrfs_init_devices_late(fs_info);
2364 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2365 root = btrfs_read_tree_root(tree_root, &location);
2367 ret = PTR_ERR(root);
2370 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2371 fs_info->csum_root = root;
2373 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2374 root = btrfs_read_tree_root(tree_root, &location);
2375 if (!IS_ERR(root)) {
2376 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2377 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2378 fs_info->quota_root = root;
2381 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2382 root = btrfs_read_tree_root(tree_root, &location);
2384 ret = PTR_ERR(root);
2388 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2389 fs_info->uuid_root = root;
2392 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2393 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2394 root = btrfs_read_tree_root(tree_root, &location);
2396 ret = PTR_ERR(root);
2399 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2400 fs_info->free_space_root = root;
2405 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2406 location.objectid, ret);
2411 * Real super block validation
2412 * NOTE: super csum type and incompat features will not be checked here.
2414 * @sb: super block to check
2415 * @mirror_num: the super block number to check its bytenr:
2416 * 0 the primary (1st) sb
2417 * 1, 2 2nd and 3rd backup copy
2418 * -1 skip bytenr check
2420 static int validate_super(struct btrfs_fs_info *fs_info,
2421 struct btrfs_super_block *sb, int mirror_num)
2423 u64 nodesize = btrfs_super_nodesize(sb);
2424 u64 sectorsize = btrfs_super_sectorsize(sb);
2427 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2428 btrfs_err(fs_info, "no valid FS found");
2431 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2432 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2433 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2436 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2437 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2438 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2441 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2442 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2443 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2446 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2447 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2448 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2453 * Check sectorsize and nodesize first, other check will need it.
2454 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2456 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2457 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2458 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2461 /* Only PAGE SIZE is supported yet */
2462 if (sectorsize != PAGE_SIZE) {
2464 "sectorsize %llu not supported yet, only support %lu",
2465 sectorsize, PAGE_SIZE);
2468 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2469 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2470 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2473 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2474 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2475 le32_to_cpu(sb->__unused_leafsize), nodesize);
2479 /* Root alignment check */
2480 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2481 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2482 btrfs_super_root(sb));
2485 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2486 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2487 btrfs_super_chunk_root(sb));
2490 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2491 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2492 btrfs_super_log_root(sb));
2496 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2497 BTRFS_FSID_SIZE) != 0) {
2499 "dev_item UUID does not match metadata fsid: %pU != %pU",
2500 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2505 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2508 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2509 btrfs_err(fs_info, "bytes_used is too small %llu",
2510 btrfs_super_bytes_used(sb));
2513 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2514 btrfs_err(fs_info, "invalid stripesize %u",
2515 btrfs_super_stripesize(sb));
2518 if (btrfs_super_num_devices(sb) > (1UL << 31))
2519 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2520 btrfs_super_num_devices(sb));
2521 if (btrfs_super_num_devices(sb) == 0) {
2522 btrfs_err(fs_info, "number of devices is 0");
2526 if (mirror_num >= 0 &&
2527 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2528 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2529 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2534 * Obvious sys_chunk_array corruptions, it must hold at least one key
2537 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2538 btrfs_err(fs_info, "system chunk array too big %u > %u",
2539 btrfs_super_sys_array_size(sb),
2540 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2543 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2544 + sizeof(struct btrfs_chunk)) {
2545 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2546 btrfs_super_sys_array_size(sb),
2547 sizeof(struct btrfs_disk_key)
2548 + sizeof(struct btrfs_chunk));
2553 * The generation is a global counter, we'll trust it more than the others
2554 * but it's still possible that it's the one that's wrong.
2556 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2558 "suspicious: generation < chunk_root_generation: %llu < %llu",
2559 btrfs_super_generation(sb),
2560 btrfs_super_chunk_root_generation(sb));
2561 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2562 && btrfs_super_cache_generation(sb) != (u64)-1)
2564 "suspicious: generation < cache_generation: %llu < %llu",
2565 btrfs_super_generation(sb),
2566 btrfs_super_cache_generation(sb));
2572 * Validation of super block at mount time.
2573 * Some checks already done early at mount time, like csum type and incompat
2574 * flags will be skipped.
2576 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2578 return validate_super(fs_info, fs_info->super_copy, 0);
2582 * Validation of super block at write time.
2583 * Some checks like bytenr check will be skipped as their values will be
2585 * Extra checks like csum type and incompat flags will be done here.
2587 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2588 struct btrfs_super_block *sb)
2592 ret = validate_super(fs_info, sb, -1);
2595 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2597 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2598 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2601 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2604 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2605 btrfs_super_incompat_flags(sb),
2606 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2612 "super block corruption detected before writing it to disk");
2616 int open_ctree(struct super_block *sb,
2617 struct btrfs_fs_devices *fs_devices,
2626 struct btrfs_key location;
2627 struct buffer_head *bh;
2628 struct btrfs_super_block *disk_super;
2629 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2630 struct btrfs_root *tree_root;
2631 struct btrfs_root *chunk_root;
2634 int num_backups_tried = 0;
2635 int backup_index = 0;
2636 int clear_free_space_tree = 0;
2639 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2640 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2641 if (!tree_root || !chunk_root) {
2646 ret = init_srcu_struct(&fs_info->subvol_srcu);
2652 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2658 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2661 goto fail_dio_bytes;
2663 fs_info->dirty_metadata_batch = PAGE_SIZE *
2664 (1 + ilog2(nr_cpu_ids));
2666 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2669 goto fail_dirty_metadata_bytes;
2672 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2676 goto fail_delalloc_bytes;
2679 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2680 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2681 INIT_LIST_HEAD(&fs_info->trans_list);
2682 INIT_LIST_HEAD(&fs_info->dead_roots);
2683 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2684 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2685 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2686 INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2687 spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2688 spin_lock_init(&fs_info->delalloc_root_lock);
2689 spin_lock_init(&fs_info->trans_lock);
2690 spin_lock_init(&fs_info->fs_roots_radix_lock);
2691 spin_lock_init(&fs_info->delayed_iput_lock);
2692 spin_lock_init(&fs_info->defrag_inodes_lock);
2693 spin_lock_init(&fs_info->tree_mod_seq_lock);
2694 spin_lock_init(&fs_info->super_lock);
2695 spin_lock_init(&fs_info->buffer_lock);
2696 spin_lock_init(&fs_info->unused_bgs_lock);
2697 rwlock_init(&fs_info->tree_mod_log_lock);
2698 mutex_init(&fs_info->unused_bg_unpin_mutex);
2699 mutex_init(&fs_info->delete_unused_bgs_mutex);
2700 mutex_init(&fs_info->reloc_mutex);
2701 mutex_init(&fs_info->delalloc_root_mutex);
2702 seqlock_init(&fs_info->profiles_lock);
2704 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2705 INIT_LIST_HEAD(&fs_info->space_info);
2706 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2707 INIT_LIST_HEAD(&fs_info->unused_bgs);
2708 extent_map_tree_init(&fs_info->mapping_tree);
2709 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2710 BTRFS_BLOCK_RSV_GLOBAL);
2711 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2712 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2713 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2714 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2715 BTRFS_BLOCK_RSV_DELOPS);
2716 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2717 BTRFS_BLOCK_RSV_DELREFS);
2719 atomic_set(&fs_info->async_delalloc_pages, 0);
2720 atomic_set(&fs_info->defrag_running, 0);
2721 atomic_set(&fs_info->reada_works_cnt, 0);
2722 atomic_set(&fs_info->nr_delayed_iputs, 0);
2723 atomic64_set(&fs_info->tree_mod_seq, 0);
2725 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2726 fs_info->metadata_ratio = 0;
2727 fs_info->defrag_inodes = RB_ROOT;
2728 atomic64_set(&fs_info->free_chunk_space, 0);
2729 fs_info->tree_mod_log = RB_ROOT;
2730 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2731 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2732 /* readahead state */
2733 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2734 spin_lock_init(&fs_info->reada_lock);
2735 btrfs_init_ref_verify(fs_info);
2737 fs_info->thread_pool_size = min_t(unsigned long,
2738 num_online_cpus() + 2, 8);
2740 INIT_LIST_HEAD(&fs_info->ordered_roots);
2741 spin_lock_init(&fs_info->ordered_root_lock);
2743 fs_info->btree_inode = new_inode(sb);
2744 if (!fs_info->btree_inode) {
2746 goto fail_bio_counter;
2748 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2750 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2752 if (!fs_info->delayed_root) {
2756 btrfs_init_delayed_root(fs_info->delayed_root);
2758 btrfs_init_scrub(fs_info);
2759 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2760 fs_info->check_integrity_print_mask = 0;
2762 btrfs_init_balance(fs_info);
2763 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2765 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2766 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2768 btrfs_init_btree_inode(fs_info);
2770 spin_lock_init(&fs_info->block_group_cache_lock);
2771 fs_info->block_group_cache_tree = RB_ROOT;
2772 fs_info->first_logical_byte = (u64)-1;
2774 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2775 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2776 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2777 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2778 fs_info->pinned_extents = &fs_info->freed_extents[0];
2779 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2781 mutex_init(&fs_info->ordered_operations_mutex);
2782 mutex_init(&fs_info->tree_log_mutex);
2783 mutex_init(&fs_info->chunk_mutex);
2784 mutex_init(&fs_info->transaction_kthread_mutex);
2785 mutex_init(&fs_info->cleaner_mutex);
2786 mutex_init(&fs_info->ro_block_group_mutex);
2787 init_rwsem(&fs_info->commit_root_sem);
2788 init_rwsem(&fs_info->cleanup_work_sem);
2789 init_rwsem(&fs_info->subvol_sem);
2790 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2792 btrfs_init_dev_replace_locks(fs_info);
2793 btrfs_init_qgroup(fs_info);
2795 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2796 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2798 init_waitqueue_head(&fs_info->transaction_throttle);
2799 init_waitqueue_head(&fs_info->transaction_wait);
2800 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2801 init_waitqueue_head(&fs_info->async_submit_wait);
2802 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2804 /* Usable values until the real ones are cached from the superblock */
2805 fs_info->nodesize = 4096;
2806 fs_info->sectorsize = 4096;
2807 fs_info->stripesize = 4096;
2809 spin_lock_init(&fs_info->swapfile_pins_lock);
2810 fs_info->swapfile_pins = RB_ROOT;
2812 fs_info->send_in_progress = 0;
2814 ret = btrfs_alloc_stripe_hash_table(fs_info);
2820 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2822 invalidate_bdev(fs_devices->latest_bdev);
2825 * Read super block and check the signature bytes only
2827 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2834 * Verify the type first, if that or the the checksum value are
2835 * corrupted, we'll find out
2837 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2838 if (!btrfs_supported_super_csum(csum_type)) {
2839 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2846 ret = btrfs_init_csum_hash(fs_info, csum_type);
2853 * We want to check superblock checksum, the type is stored inside.
2854 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2856 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2857 btrfs_err(fs_info, "superblock checksum mismatch");
2864 * super_copy is zeroed at allocation time and we never touch the
2865 * following bytes up to INFO_SIZE, the checksum is calculated from
2866 * the whole block of INFO_SIZE
2868 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2871 disk_super = fs_info->super_copy;
2873 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2876 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2877 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2878 fs_info->super_copy->metadata_uuid,
2882 features = btrfs_super_flags(disk_super);
2883 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2884 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2885 btrfs_set_super_flags(disk_super, features);
2887 "found metadata UUID change in progress flag, clearing");
2890 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2891 sizeof(*fs_info->super_for_commit));
2893 ret = btrfs_validate_mount_super(fs_info);
2895 btrfs_err(fs_info, "superblock contains fatal errors");
2900 if (!btrfs_super_root(disk_super))
2903 /* check FS state, whether FS is broken. */
2904 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2905 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2908 * run through our array of backup supers and setup
2909 * our ring pointer to the oldest one
2911 generation = btrfs_super_generation(disk_super);
2912 find_oldest_super_backup(fs_info, generation);
2915 * In the long term, we'll store the compression type in the super
2916 * block, and it'll be used for per file compression control.
2918 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2920 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2926 features = btrfs_super_incompat_flags(disk_super) &
2927 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2930 "cannot mount because of unsupported optional features (%llx)",
2936 features = btrfs_super_incompat_flags(disk_super);
2937 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2938 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2939 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2940 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2941 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2943 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2944 btrfs_info(fs_info, "has skinny extents");
2947 * flag our filesystem as having big metadata blocks if
2948 * they are bigger than the page size
2950 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2951 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2953 "flagging fs with big metadata feature");
2954 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2957 nodesize = btrfs_super_nodesize(disk_super);
2958 sectorsize = btrfs_super_sectorsize(disk_super);
2959 stripesize = sectorsize;
2960 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2961 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2963 /* Cache block sizes */
2964 fs_info->nodesize = nodesize;
2965 fs_info->sectorsize = sectorsize;
2966 fs_info->stripesize = stripesize;
2969 * mixed block groups end up with duplicate but slightly offset
2970 * extent buffers for the same range. It leads to corruptions
2972 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2973 (sectorsize != nodesize)) {
2975 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2976 nodesize, sectorsize);
2981 * Needn't use the lock because there is no other task which will
2984 btrfs_set_super_incompat_flags(disk_super, features);
2986 features = btrfs_super_compat_ro_flags(disk_super) &
2987 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2988 if (!sb_rdonly(sb) && features) {
2990 "cannot mount read-write because of unsupported optional features (%llx)",
2996 ret = btrfs_init_workqueues(fs_info, fs_devices);
2999 goto fail_sb_buffer;
3002 sb->s_bdi->congested_fn = btrfs_congested_fn;
3003 sb->s_bdi->congested_data = fs_info;
3004 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3005 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3006 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3007 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3009 sb->s_blocksize = sectorsize;
3010 sb->s_blocksize_bits = blksize_bits(sectorsize);
3011 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3013 mutex_lock(&fs_info->chunk_mutex);
3014 ret = btrfs_read_sys_array(fs_info);
3015 mutex_unlock(&fs_info->chunk_mutex);
3017 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3018 goto fail_sb_buffer;
3021 generation = btrfs_super_chunk_root_generation(disk_super);
3022 level = btrfs_super_chunk_root_level(disk_super);
3024 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
3026 chunk_root->node = read_tree_block(fs_info,
3027 btrfs_super_chunk_root(disk_super),
3028 generation, level, NULL);
3029 if (IS_ERR(chunk_root->node) ||
3030 !extent_buffer_uptodate(chunk_root->node)) {
3031 btrfs_err(fs_info, "failed to read chunk root");
3032 if (!IS_ERR(chunk_root->node))
3033 free_extent_buffer(chunk_root->node);
3034 chunk_root->node = NULL;
3035 goto fail_tree_roots;
3037 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3038 chunk_root->commit_root = btrfs_root_node(chunk_root);
3040 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3041 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3043 ret = btrfs_read_chunk_tree(fs_info);
3045 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3046 goto fail_tree_roots;
3050 * Keep the devid that is marked to be the target device for the
3051 * device replace procedure
3053 btrfs_free_extra_devids(fs_devices, 0);
3055 if (!fs_devices->latest_bdev) {
3056 btrfs_err(fs_info, "failed to read devices");
3057 goto fail_tree_roots;
3061 generation = btrfs_super_generation(disk_super);
3062 level = btrfs_super_root_level(disk_super);
3064 tree_root->node = read_tree_block(fs_info,
3065 btrfs_super_root(disk_super),
3066 generation, level, NULL);
3067 if (IS_ERR(tree_root->node) ||
3068 !extent_buffer_uptodate(tree_root->node)) {
3069 btrfs_warn(fs_info, "failed to read tree root");
3070 if (!IS_ERR(tree_root->node))
3071 free_extent_buffer(tree_root->node);
3072 tree_root->node = NULL;
3073 goto recovery_tree_root;
3076 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3077 tree_root->commit_root = btrfs_root_node(tree_root);
3078 btrfs_set_root_refs(&tree_root->root_item, 1);
3080 mutex_lock(&tree_root->objectid_mutex);
3081 ret = btrfs_find_highest_objectid(tree_root,
3082 &tree_root->highest_objectid);
3084 mutex_unlock(&tree_root->objectid_mutex);
3085 goto recovery_tree_root;
3088 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3090 mutex_unlock(&tree_root->objectid_mutex);
3092 ret = btrfs_read_roots(fs_info);
3094 goto recovery_tree_root;
3096 fs_info->generation = generation;
3097 fs_info->last_trans_committed = generation;
3099 ret = btrfs_verify_dev_extents(fs_info);
3102 "failed to verify dev extents against chunks: %d",
3104 goto fail_block_groups;
3106 ret = btrfs_recover_balance(fs_info);
3108 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3109 goto fail_block_groups;
3112 ret = btrfs_init_dev_stats(fs_info);
3114 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3115 goto fail_block_groups;
3118 ret = btrfs_init_dev_replace(fs_info);
3120 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3121 goto fail_block_groups;
3124 btrfs_free_extra_devids(fs_devices, 1);
3126 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3128 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3130 goto fail_block_groups;
3133 ret = btrfs_sysfs_add_device(fs_devices);
3135 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3137 goto fail_fsdev_sysfs;
3140 ret = btrfs_sysfs_add_mounted(fs_info);
3142 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3143 goto fail_fsdev_sysfs;
3146 ret = btrfs_init_space_info(fs_info);
3148 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3152 ret = btrfs_read_block_groups(fs_info);
3154 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3158 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3160 "writable mount is not allowed due to too many missing devices");
3164 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3166 if (IS_ERR(fs_info->cleaner_kthread))
3169 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3171 "btrfs-transaction");
3172 if (IS_ERR(fs_info->transaction_kthread))
3175 if (!btrfs_test_opt(fs_info, NOSSD) &&
3176 !fs_info->fs_devices->rotating) {
3177 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3181 * Mount does not set all options immediately, we can do it now and do
3182 * not have to wait for transaction commit
3184 btrfs_apply_pending_changes(fs_info);
3186 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3187 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3188 ret = btrfsic_mount(fs_info, fs_devices,
3189 btrfs_test_opt(fs_info,
3190 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3192 fs_info->check_integrity_print_mask);
3195 "failed to initialize integrity check module: %d",
3199 ret = btrfs_read_qgroup_config(fs_info);
3201 goto fail_trans_kthread;
3203 if (btrfs_build_ref_tree(fs_info))
3204 btrfs_err(fs_info, "couldn't build ref tree");
3206 /* do not make disk changes in broken FS or nologreplay is given */
3207 if (btrfs_super_log_root(disk_super) != 0 &&
3208 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3209 ret = btrfs_replay_log(fs_info, fs_devices);
3216 ret = btrfs_find_orphan_roots(fs_info);
3220 if (!sb_rdonly(sb)) {
3221 ret = btrfs_cleanup_fs_roots(fs_info);
3225 mutex_lock(&fs_info->cleaner_mutex);
3226 ret = btrfs_recover_relocation(tree_root);
3227 mutex_unlock(&fs_info->cleaner_mutex);
3229 btrfs_warn(fs_info, "failed to recover relocation: %d",
3236 location.objectid = BTRFS_FS_TREE_OBJECTID;
3237 location.type = BTRFS_ROOT_ITEM_KEY;
3238 location.offset = 0;
3240 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3241 if (IS_ERR(fs_info->fs_root)) {
3242 err = PTR_ERR(fs_info->fs_root);
3243 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3250 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3251 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3252 clear_free_space_tree = 1;
3253 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3254 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3255 btrfs_warn(fs_info, "free space tree is invalid");
3256 clear_free_space_tree = 1;
3259 if (clear_free_space_tree) {
3260 btrfs_info(fs_info, "clearing free space tree");
3261 ret = btrfs_clear_free_space_tree(fs_info);
3264 "failed to clear free space tree: %d", ret);
3265 close_ctree(fs_info);
3270 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3271 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3272 btrfs_info(fs_info, "creating free space tree");
3273 ret = btrfs_create_free_space_tree(fs_info);
3276 "failed to create free space tree: %d", ret);
3277 close_ctree(fs_info);
3282 down_read(&fs_info->cleanup_work_sem);
3283 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3284 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3285 up_read(&fs_info->cleanup_work_sem);
3286 close_ctree(fs_info);
3289 up_read(&fs_info->cleanup_work_sem);
3291 ret = btrfs_resume_balance_async(fs_info);
3293 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3294 close_ctree(fs_info);
3298 ret = btrfs_resume_dev_replace_async(fs_info);
3300 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3301 close_ctree(fs_info);
3305 btrfs_qgroup_rescan_resume(fs_info);
3307 if (!fs_info->uuid_root) {
3308 btrfs_info(fs_info, "creating UUID tree");
3309 ret = btrfs_create_uuid_tree(fs_info);
3312 "failed to create the UUID tree: %d", ret);
3313 close_ctree(fs_info);
3316 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3317 fs_info->generation !=
3318 btrfs_super_uuid_tree_generation(disk_super)) {
3319 btrfs_info(fs_info, "checking UUID tree");
3320 ret = btrfs_check_uuid_tree(fs_info);
3323 "failed to check the UUID tree: %d", ret);
3324 close_ctree(fs_info);
3328 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3330 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3333 * backuproot only affect mount behavior, and if open_ctree succeeded,
3334 * no need to keep the flag
3336 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3341 btrfs_free_qgroup_config(fs_info);
3343 kthread_stop(fs_info->transaction_kthread);
3344 btrfs_cleanup_transaction(fs_info);
3345 btrfs_free_fs_roots(fs_info);
3347 kthread_stop(fs_info->cleaner_kthread);
3350 * make sure we're done with the btree inode before we stop our
3353 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3356 btrfs_sysfs_remove_mounted(fs_info);
3359 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3362 btrfs_put_block_group_cache(fs_info);
3365 free_root_pointers(fs_info, 1);
3366 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3369 btrfs_stop_all_workers(fs_info);
3370 btrfs_free_block_groups(fs_info);
3372 btrfs_free_csum_hash(fs_info);
3375 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3377 iput(fs_info->btree_inode);
3379 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3380 fail_delalloc_bytes:
3381 percpu_counter_destroy(&fs_info->delalloc_bytes);
3382 fail_dirty_metadata_bytes:
3383 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3385 percpu_counter_destroy(&fs_info->dio_bytes);
3387 cleanup_srcu_struct(&fs_info->subvol_srcu);
3389 btrfs_free_stripe_hash_table(fs_info);
3390 btrfs_close_devices(fs_info->fs_devices);
3394 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3395 goto fail_tree_roots;
3397 free_root_pointers(fs_info, 0);
3399 /* don't use the log in recovery mode, it won't be valid */
3400 btrfs_set_super_log_root(disk_super, 0);
3402 /* we can't trust the free space cache either */
3403 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3405 ret = next_root_backup(fs_info, fs_info->super_copy,
3406 &num_backups_tried, &backup_index);
3408 goto fail_block_groups;
3409 goto retry_root_backup;
3411 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3413 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3416 set_buffer_uptodate(bh);
3418 struct btrfs_device *device = (struct btrfs_device *)
3421 btrfs_warn_rl_in_rcu(device->fs_info,
3422 "lost page write due to IO error on %s",
3423 rcu_str_deref(device->name));
3424 /* note, we don't set_buffer_write_io_error because we have
3425 * our own ways of dealing with the IO errors
3427 clear_buffer_uptodate(bh);
3428 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3434 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3435 struct buffer_head **bh_ret)
3437 struct buffer_head *bh;
3438 struct btrfs_super_block *super;
3441 bytenr = btrfs_sb_offset(copy_num);
3442 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3445 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3447 * If we fail to read from the underlying devices, as of now
3448 * the best option we have is to mark it EIO.
3453 super = (struct btrfs_super_block *)bh->b_data;
3454 if (btrfs_super_bytenr(super) != bytenr ||
3455 btrfs_super_magic(super) != BTRFS_MAGIC) {
3465 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3467 struct buffer_head *bh;
3468 struct buffer_head *latest = NULL;
3469 struct btrfs_super_block *super;
3474 /* we would like to check all the supers, but that would make
3475 * a btrfs mount succeed after a mkfs from a different FS.
3476 * So, we need to add a special mount option to scan for
3477 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3479 for (i = 0; i < 1; i++) {
3480 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3484 super = (struct btrfs_super_block *)bh->b_data;
3486 if (!latest || btrfs_super_generation(super) > transid) {
3489 transid = btrfs_super_generation(super);
3496 return ERR_PTR(ret);
3502 * Write superblock @sb to the @device. Do not wait for completion, all the
3503 * buffer heads we write are pinned.
3505 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3506 * the expected device size at commit time. Note that max_mirrors must be
3507 * same for write and wait phases.
3509 * Return number of errors when buffer head is not found or submission fails.
3511 static int write_dev_supers(struct btrfs_device *device,
3512 struct btrfs_super_block *sb, int max_mirrors)
3514 struct btrfs_fs_info *fs_info = device->fs_info;
3515 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3516 struct buffer_head *bh;
3523 if (max_mirrors == 0)
3524 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3526 shash->tfm = fs_info->csum_shash;
3528 for (i = 0; i < max_mirrors; i++) {
3529 bytenr = btrfs_sb_offset(i);
3530 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3531 device->commit_total_bytes)
3534 btrfs_set_super_bytenr(sb, bytenr);
3536 crypto_shash_init(shash);
3537 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3538 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3539 crypto_shash_final(shash, sb->csum);
3541 /* One reference for us, and we leave it for the caller */
3542 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3543 BTRFS_SUPER_INFO_SIZE);
3545 btrfs_err(device->fs_info,
3546 "couldn't get super buffer head for bytenr %llu",
3552 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3554 /* one reference for submit_bh */
3557 set_buffer_uptodate(bh);
3559 bh->b_end_io = btrfs_end_buffer_write_sync;
3560 bh->b_private = device;
3563 * we fua the first super. The others we allow
3566 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3567 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3568 op_flags |= REQ_FUA;
3569 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3573 return errors < i ? 0 : -1;
3577 * Wait for write completion of superblocks done by write_dev_supers,
3578 * @max_mirrors same for write and wait phases.
3580 * Return number of errors when buffer head is not found or not marked up to
3583 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3585 struct buffer_head *bh;
3588 bool primary_failed = false;
3591 if (max_mirrors == 0)
3592 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3594 for (i = 0; i < max_mirrors; i++) {
3595 bytenr = btrfs_sb_offset(i);
3596 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3597 device->commit_total_bytes)
3600 bh = __find_get_block(device->bdev,
3601 bytenr / BTRFS_BDEV_BLOCKSIZE,
3602 BTRFS_SUPER_INFO_SIZE);
3606 primary_failed = true;
3610 if (!buffer_uptodate(bh)) {
3613 primary_failed = true;
3616 /* drop our reference */
3619 /* drop the reference from the writing run */
3623 /* log error, force error return */
3624 if (primary_failed) {
3625 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3630 return errors < i ? 0 : -1;
3634 * endio for the write_dev_flush, this will wake anyone waiting
3635 * for the barrier when it is done
3637 static void btrfs_end_empty_barrier(struct bio *bio)
3639 complete(bio->bi_private);
3643 * Submit a flush request to the device if it supports it. Error handling is
3644 * done in the waiting counterpart.
3646 static void write_dev_flush(struct btrfs_device *device)
3648 struct request_queue *q = bdev_get_queue(device->bdev);
3649 struct bio *bio = device->flush_bio;
3651 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3655 bio->bi_end_io = btrfs_end_empty_barrier;
3656 bio_set_dev(bio, device->bdev);
3657 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3658 init_completion(&device->flush_wait);
3659 bio->bi_private = &device->flush_wait;
3661 btrfsic_submit_bio(bio);
3662 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3666 * If the flush bio has been submitted by write_dev_flush, wait for it.
3668 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3670 struct bio *bio = device->flush_bio;
3672 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3675 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3676 wait_for_completion_io(&device->flush_wait);
3678 return bio->bi_status;
3681 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3683 if (!btrfs_check_rw_degradable(fs_info, NULL))
3689 * send an empty flush down to each device in parallel,
3690 * then wait for them
3692 static int barrier_all_devices(struct btrfs_fs_info *info)
3694 struct list_head *head;
3695 struct btrfs_device *dev;
3696 int errors_wait = 0;
3699 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3700 /* send down all the barriers */
3701 head = &info->fs_devices->devices;
3702 list_for_each_entry(dev, head, dev_list) {
3703 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3707 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3708 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3711 write_dev_flush(dev);
3712 dev->last_flush_error = BLK_STS_OK;
3715 /* wait for all the barriers */
3716 list_for_each_entry(dev, head, dev_list) {
3717 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3723 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3724 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3727 ret = wait_dev_flush(dev);
3729 dev->last_flush_error = ret;
3730 btrfs_dev_stat_inc_and_print(dev,
3731 BTRFS_DEV_STAT_FLUSH_ERRS);
3738 * At some point we need the status of all disks
3739 * to arrive at the volume status. So error checking
3740 * is being pushed to a separate loop.
3742 return check_barrier_error(info);
3747 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3750 int min_tolerated = INT_MAX;
3752 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3753 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3754 min_tolerated = min_t(int, min_tolerated,
3755 btrfs_raid_array[BTRFS_RAID_SINGLE].
3756 tolerated_failures);
3758 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3759 if (raid_type == BTRFS_RAID_SINGLE)
3761 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3763 min_tolerated = min_t(int, min_tolerated,
3764 btrfs_raid_array[raid_type].
3765 tolerated_failures);
3768 if (min_tolerated == INT_MAX) {
3769 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3773 return min_tolerated;
3776 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3778 struct list_head *head;
3779 struct btrfs_device *dev;
3780 struct btrfs_super_block *sb;
3781 struct btrfs_dev_item *dev_item;
3785 int total_errors = 0;
3788 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3791 * max_mirrors == 0 indicates we're from commit_transaction,
3792 * not from fsync where the tree roots in fs_info have not
3793 * been consistent on disk.
3795 if (max_mirrors == 0)
3796 backup_super_roots(fs_info);
3798 sb = fs_info->super_for_commit;
3799 dev_item = &sb->dev_item;
3801 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3802 head = &fs_info->fs_devices->devices;
3803 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3806 ret = barrier_all_devices(fs_info);
3809 &fs_info->fs_devices->device_list_mutex);
3810 btrfs_handle_fs_error(fs_info, ret,
3811 "errors while submitting device barriers.");
3816 list_for_each_entry(dev, head, dev_list) {
3821 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3822 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3825 btrfs_set_stack_device_generation(dev_item, 0);
3826 btrfs_set_stack_device_type(dev_item, dev->type);
3827 btrfs_set_stack_device_id(dev_item, dev->devid);
3828 btrfs_set_stack_device_total_bytes(dev_item,
3829 dev->commit_total_bytes);
3830 btrfs_set_stack_device_bytes_used(dev_item,
3831 dev->commit_bytes_used);
3832 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3833 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3834 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3835 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3836 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3839 flags = btrfs_super_flags(sb);
3840 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3842 ret = btrfs_validate_write_super(fs_info, sb);
3844 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3845 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3846 "unexpected superblock corruption detected");
3850 ret = write_dev_supers(dev, sb, max_mirrors);
3854 if (total_errors > max_errors) {
3855 btrfs_err(fs_info, "%d errors while writing supers",
3857 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3859 /* FUA is masked off if unsupported and can't be the reason */
3860 btrfs_handle_fs_error(fs_info, -EIO,
3861 "%d errors while writing supers",
3867 list_for_each_entry(dev, head, dev_list) {
3870 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3871 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3874 ret = wait_dev_supers(dev, max_mirrors);
3878 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3879 if (total_errors > max_errors) {
3880 btrfs_handle_fs_error(fs_info, -EIO,
3881 "%d errors while writing supers",
3888 /* Drop a fs root from the radix tree and free it. */
3889 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3890 struct btrfs_root *root)
3892 spin_lock(&fs_info->fs_roots_radix_lock);
3893 radix_tree_delete(&fs_info->fs_roots_radix,
3894 (unsigned long)root->root_key.objectid);
3895 spin_unlock(&fs_info->fs_roots_radix_lock);
3897 if (btrfs_root_refs(&root->root_item) == 0)
3898 synchronize_srcu(&fs_info->subvol_srcu);
3900 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3901 btrfs_free_log(NULL, root);
3902 if (root->reloc_root) {
3903 free_extent_buffer(root->reloc_root->node);
3904 free_extent_buffer(root->reloc_root->commit_root);
3905 btrfs_put_fs_root(root->reloc_root);
3906 root->reloc_root = NULL;
3910 if (root->free_ino_pinned)
3911 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3912 if (root->free_ino_ctl)
3913 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3914 btrfs_free_fs_root(root);
3917 void btrfs_free_fs_root(struct btrfs_root *root)
3919 iput(root->ino_cache_inode);
3920 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3922 free_anon_bdev(root->anon_dev);
3923 if (root->subv_writers)
3924 btrfs_free_subvolume_writers(root->subv_writers);
3925 free_extent_buffer(root->node);
3926 free_extent_buffer(root->commit_root);
3927 kfree(root->free_ino_ctl);
3928 kfree(root->free_ino_pinned);
3929 btrfs_put_fs_root(root);
3932 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3934 u64 root_objectid = 0;
3935 struct btrfs_root *gang[8];
3938 unsigned int ret = 0;
3942 index = srcu_read_lock(&fs_info->subvol_srcu);
3943 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3944 (void **)gang, root_objectid,
3947 srcu_read_unlock(&fs_info->subvol_srcu, index);
3950 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3952 for (i = 0; i < ret; i++) {
3953 /* Avoid to grab roots in dead_roots */
3954 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3958 /* grab all the search result for later use */
3959 gang[i] = btrfs_grab_fs_root(gang[i]);
3961 srcu_read_unlock(&fs_info->subvol_srcu, index);
3963 for (i = 0; i < ret; i++) {
3966 root_objectid = gang[i]->root_key.objectid;
3967 err = btrfs_orphan_cleanup(gang[i]);
3970 btrfs_put_fs_root(gang[i]);
3975 /* release the uncleaned roots due to error */
3976 for (; i < ret; i++) {
3978 btrfs_put_fs_root(gang[i]);
3983 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3985 struct btrfs_root *root = fs_info->tree_root;
3986 struct btrfs_trans_handle *trans;
3988 mutex_lock(&fs_info->cleaner_mutex);
3989 btrfs_run_delayed_iputs(fs_info);
3990 mutex_unlock(&fs_info->cleaner_mutex);
3991 wake_up_process(fs_info->cleaner_kthread);
3993 /* wait until ongoing cleanup work done */
3994 down_write(&fs_info->cleanup_work_sem);
3995 up_write(&fs_info->cleanup_work_sem);
3997 trans = btrfs_join_transaction(root);
3999 return PTR_ERR(trans);
4000 return btrfs_commit_transaction(trans);
4003 void close_ctree(struct btrfs_fs_info *fs_info)
4007 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4009 * We don't want the cleaner to start new transactions, add more delayed
4010 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4011 * because that frees the task_struct, and the transaction kthread might
4012 * still try to wake up the cleaner.
4014 kthread_park(fs_info->cleaner_kthread);
4016 /* wait for the qgroup rescan worker to stop */
4017 btrfs_qgroup_wait_for_completion(fs_info, false);
4019 /* wait for the uuid_scan task to finish */
4020 down(&fs_info->uuid_tree_rescan_sem);
4021 /* avoid complains from lockdep et al., set sem back to initial state */
4022 up(&fs_info->uuid_tree_rescan_sem);
4024 /* pause restriper - we want to resume on mount */
4025 btrfs_pause_balance(fs_info);
4027 btrfs_dev_replace_suspend_for_unmount(fs_info);
4029 btrfs_scrub_cancel(fs_info);
4031 /* wait for any defraggers to finish */
4032 wait_event(fs_info->transaction_wait,
4033 (atomic_read(&fs_info->defrag_running) == 0));
4035 /* clear out the rbtree of defraggable inodes */
4036 btrfs_cleanup_defrag_inodes(fs_info);
4038 cancel_work_sync(&fs_info->async_reclaim_work);
4040 if (!sb_rdonly(fs_info->sb)) {
4042 * The cleaner kthread is stopped, so do one final pass over
4043 * unused block groups.
4045 btrfs_delete_unused_bgs(fs_info);
4047 ret = btrfs_commit_super(fs_info);
4049 btrfs_err(fs_info, "commit super ret %d", ret);
4052 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4053 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4054 btrfs_error_commit_super(fs_info);
4056 kthread_stop(fs_info->transaction_kthread);
4057 kthread_stop(fs_info->cleaner_kthread);
4059 ASSERT(list_empty(&fs_info->delayed_iputs));
4060 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4062 btrfs_free_qgroup_config(fs_info);
4063 ASSERT(list_empty(&fs_info->delalloc_roots));
4065 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4066 btrfs_info(fs_info, "at unmount delalloc count %lld",
4067 percpu_counter_sum(&fs_info->delalloc_bytes));
4070 if (percpu_counter_sum(&fs_info->dio_bytes))
4071 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4072 percpu_counter_sum(&fs_info->dio_bytes));
4074 btrfs_sysfs_remove_mounted(fs_info);
4075 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4077 btrfs_free_fs_roots(fs_info);
4079 btrfs_put_block_group_cache(fs_info);
4082 * we must make sure there is not any read request to
4083 * submit after we stopping all workers.
4085 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4086 btrfs_stop_all_workers(fs_info);
4088 btrfs_free_block_groups(fs_info);
4090 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4091 free_root_pointers(fs_info, 1);
4093 iput(fs_info->btree_inode);
4095 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4096 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4097 btrfsic_unmount(fs_info->fs_devices);
4100 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4101 btrfs_close_devices(fs_info->fs_devices);
4103 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4104 percpu_counter_destroy(&fs_info->delalloc_bytes);
4105 percpu_counter_destroy(&fs_info->dio_bytes);
4106 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4107 cleanup_srcu_struct(&fs_info->subvol_srcu);
4109 btrfs_free_stripe_hash_table(fs_info);
4110 btrfs_free_ref_cache(fs_info);
4113 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4117 struct inode *btree_inode = buf->pages[0]->mapping->host;
4119 ret = extent_buffer_uptodate(buf);
4123 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4124 parent_transid, atomic);
4130 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4132 struct btrfs_fs_info *fs_info;
4133 struct btrfs_root *root;
4134 u64 transid = btrfs_header_generation(buf);
4137 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4139 * This is a fast path so only do this check if we have sanity tests
4140 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4141 * outside of the sanity tests.
4143 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4146 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4147 fs_info = root->fs_info;
4148 btrfs_assert_tree_locked(buf);
4149 if (transid != fs_info->generation)
4150 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4151 buf->start, transid, fs_info->generation);
4152 was_dirty = set_extent_buffer_dirty(buf);
4154 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4156 fs_info->dirty_metadata_batch);
4157 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4159 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4160 * but item data not updated.
4161 * So here we should only check item pointers, not item data.
4163 if (btrfs_header_level(buf) == 0 &&
4164 btrfs_check_leaf_relaxed(buf)) {
4165 btrfs_print_leaf(buf);
4171 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4175 * looks as though older kernels can get into trouble with
4176 * this code, they end up stuck in balance_dirty_pages forever
4180 if (current->flags & PF_MEMALLOC)
4184 btrfs_balance_delayed_items(fs_info);
4186 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4187 BTRFS_DIRTY_METADATA_THRESH,
4188 fs_info->dirty_metadata_batch);
4190 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4194 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4196 __btrfs_btree_balance_dirty(fs_info, 1);
4199 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4201 __btrfs_btree_balance_dirty(fs_info, 0);
4204 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4205 struct btrfs_key *first_key)
4207 return btree_read_extent_buffer_pages(buf, parent_transid,
4211 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4213 /* cleanup FS via transaction */
4214 btrfs_cleanup_transaction(fs_info);
4216 mutex_lock(&fs_info->cleaner_mutex);
4217 btrfs_run_delayed_iputs(fs_info);
4218 mutex_unlock(&fs_info->cleaner_mutex);
4220 down_write(&fs_info->cleanup_work_sem);
4221 up_write(&fs_info->cleanup_work_sem);
4224 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4226 struct btrfs_ordered_extent *ordered;
4228 spin_lock(&root->ordered_extent_lock);
4230 * This will just short circuit the ordered completion stuff which will
4231 * make sure the ordered extent gets properly cleaned up.
4233 list_for_each_entry(ordered, &root->ordered_extents,
4235 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4236 spin_unlock(&root->ordered_extent_lock);
4239 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4241 struct btrfs_root *root;
4242 struct list_head splice;
4244 INIT_LIST_HEAD(&splice);
4246 spin_lock(&fs_info->ordered_root_lock);
4247 list_splice_init(&fs_info->ordered_roots, &splice);
4248 while (!list_empty(&splice)) {
4249 root = list_first_entry(&splice, struct btrfs_root,
4251 list_move_tail(&root->ordered_root,
4252 &fs_info->ordered_roots);
4254 spin_unlock(&fs_info->ordered_root_lock);
4255 btrfs_destroy_ordered_extents(root);
4258 spin_lock(&fs_info->ordered_root_lock);
4260 spin_unlock(&fs_info->ordered_root_lock);
4263 * We need this here because if we've been flipped read-only we won't
4264 * get sync() from the umount, so we need to make sure any ordered
4265 * extents that haven't had their dirty pages IO start writeout yet
4266 * actually get run and error out properly.
4268 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4271 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4272 struct btrfs_fs_info *fs_info)
4274 struct rb_node *node;
4275 struct btrfs_delayed_ref_root *delayed_refs;
4276 struct btrfs_delayed_ref_node *ref;
4279 delayed_refs = &trans->delayed_refs;
4281 spin_lock(&delayed_refs->lock);
4282 if (atomic_read(&delayed_refs->num_entries) == 0) {
4283 spin_unlock(&delayed_refs->lock);
4284 btrfs_info(fs_info, "delayed_refs has NO entry");
4288 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4289 struct btrfs_delayed_ref_head *head;
4291 bool pin_bytes = false;
4293 head = rb_entry(node, struct btrfs_delayed_ref_head,
4295 if (btrfs_delayed_ref_lock(delayed_refs, head))
4298 spin_lock(&head->lock);
4299 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4300 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4303 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4304 RB_CLEAR_NODE(&ref->ref_node);
4305 if (!list_empty(&ref->add_list))
4306 list_del(&ref->add_list);
4307 atomic_dec(&delayed_refs->num_entries);
4308 btrfs_put_delayed_ref(ref);
4310 if (head->must_insert_reserved)
4312 btrfs_free_delayed_extent_op(head->extent_op);
4313 btrfs_delete_ref_head(delayed_refs, head);
4314 spin_unlock(&head->lock);
4315 spin_unlock(&delayed_refs->lock);
4316 mutex_unlock(&head->mutex);
4319 btrfs_pin_extent(fs_info, head->bytenr,
4320 head->num_bytes, 1);
4321 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4322 btrfs_put_delayed_ref_head(head);
4324 spin_lock(&delayed_refs->lock);
4327 spin_unlock(&delayed_refs->lock);
4332 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4334 struct btrfs_inode *btrfs_inode;
4335 struct list_head splice;
4337 INIT_LIST_HEAD(&splice);
4339 spin_lock(&root->delalloc_lock);
4340 list_splice_init(&root->delalloc_inodes, &splice);
4342 while (!list_empty(&splice)) {
4343 struct inode *inode = NULL;
4344 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4346 __btrfs_del_delalloc_inode(root, btrfs_inode);
4347 spin_unlock(&root->delalloc_lock);
4350 * Make sure we get a live inode and that it'll not disappear
4353 inode = igrab(&btrfs_inode->vfs_inode);
4355 invalidate_inode_pages2(inode->i_mapping);
4358 spin_lock(&root->delalloc_lock);
4360 spin_unlock(&root->delalloc_lock);
4363 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4365 struct btrfs_root *root;
4366 struct list_head splice;
4368 INIT_LIST_HEAD(&splice);
4370 spin_lock(&fs_info->delalloc_root_lock);
4371 list_splice_init(&fs_info->delalloc_roots, &splice);
4372 while (!list_empty(&splice)) {
4373 root = list_first_entry(&splice, struct btrfs_root,
4375 root = btrfs_grab_fs_root(root);
4377 spin_unlock(&fs_info->delalloc_root_lock);
4379 btrfs_destroy_delalloc_inodes(root);
4380 btrfs_put_fs_root(root);
4382 spin_lock(&fs_info->delalloc_root_lock);
4384 spin_unlock(&fs_info->delalloc_root_lock);
4387 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4388 struct extent_io_tree *dirty_pages,
4392 struct extent_buffer *eb;
4397 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4402 clear_extent_bits(dirty_pages, start, end, mark);
4403 while (start <= end) {
4404 eb = find_extent_buffer(fs_info, start);
4405 start += fs_info->nodesize;
4408 wait_on_extent_buffer_writeback(eb);
4410 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4412 clear_extent_buffer_dirty(eb);
4413 free_extent_buffer_stale(eb);
4420 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4421 struct extent_io_tree *pinned_extents)
4423 struct extent_io_tree *unpin;
4429 unpin = pinned_extents;
4432 struct extent_state *cached_state = NULL;
4435 * The btrfs_finish_extent_commit() may get the same range as
4436 * ours between find_first_extent_bit and clear_extent_dirty.
4437 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4438 * the same extent range.
4440 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4441 ret = find_first_extent_bit(unpin, 0, &start, &end,
4442 EXTENT_DIRTY, &cached_state);
4444 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4448 clear_extent_dirty(unpin, start, end, &cached_state);
4449 free_extent_state(cached_state);
4450 btrfs_error_unpin_extent_range(fs_info, start, end);
4451 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4456 if (unpin == &fs_info->freed_extents[0])
4457 unpin = &fs_info->freed_extents[1];
4459 unpin = &fs_info->freed_extents[0];
4467 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4469 struct inode *inode;
4471 inode = cache->io_ctl.inode;
4473 invalidate_inode_pages2(inode->i_mapping);
4474 BTRFS_I(inode)->generation = 0;
4475 cache->io_ctl.inode = NULL;
4478 btrfs_put_block_group(cache);
4481 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4482 struct btrfs_fs_info *fs_info)
4484 struct btrfs_block_group_cache *cache;
4486 spin_lock(&cur_trans->dirty_bgs_lock);
4487 while (!list_empty(&cur_trans->dirty_bgs)) {
4488 cache = list_first_entry(&cur_trans->dirty_bgs,
4489 struct btrfs_block_group_cache,
4492 if (!list_empty(&cache->io_list)) {
4493 spin_unlock(&cur_trans->dirty_bgs_lock);
4494 list_del_init(&cache->io_list);
4495 btrfs_cleanup_bg_io(cache);
4496 spin_lock(&cur_trans->dirty_bgs_lock);
4499 list_del_init(&cache->dirty_list);
4500 spin_lock(&cache->lock);
4501 cache->disk_cache_state = BTRFS_DC_ERROR;
4502 spin_unlock(&cache->lock);
4504 spin_unlock(&cur_trans->dirty_bgs_lock);
4505 btrfs_put_block_group(cache);
4506 btrfs_delayed_refs_rsv_release(fs_info, 1);
4507 spin_lock(&cur_trans->dirty_bgs_lock);
4509 spin_unlock(&cur_trans->dirty_bgs_lock);
4512 * Refer to the definition of io_bgs member for details why it's safe
4513 * to use it without any locking
4515 while (!list_empty(&cur_trans->io_bgs)) {
4516 cache = list_first_entry(&cur_trans->io_bgs,
4517 struct btrfs_block_group_cache,
4520 list_del_init(&cache->io_list);
4521 spin_lock(&cache->lock);
4522 cache->disk_cache_state = BTRFS_DC_ERROR;
4523 spin_unlock(&cache->lock);
4524 btrfs_cleanup_bg_io(cache);
4528 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4529 struct btrfs_fs_info *fs_info)
4531 struct btrfs_device *dev, *tmp;
4533 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4534 ASSERT(list_empty(&cur_trans->dirty_bgs));
4535 ASSERT(list_empty(&cur_trans->io_bgs));
4537 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4539 list_del_init(&dev->post_commit_list);
4542 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4544 cur_trans->state = TRANS_STATE_COMMIT_START;
4545 wake_up(&fs_info->transaction_blocked_wait);
4547 cur_trans->state = TRANS_STATE_UNBLOCKED;
4548 wake_up(&fs_info->transaction_wait);
4550 btrfs_destroy_delayed_inodes(fs_info);
4551 btrfs_assert_delayed_root_empty(fs_info);
4553 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4555 btrfs_destroy_pinned_extent(fs_info,
4556 fs_info->pinned_extents);
4558 cur_trans->state =TRANS_STATE_COMPLETED;
4559 wake_up(&cur_trans->commit_wait);
4562 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4564 struct btrfs_transaction *t;
4566 mutex_lock(&fs_info->transaction_kthread_mutex);
4568 spin_lock(&fs_info->trans_lock);
4569 while (!list_empty(&fs_info->trans_list)) {
4570 t = list_first_entry(&fs_info->trans_list,
4571 struct btrfs_transaction, list);
4572 if (t->state >= TRANS_STATE_COMMIT_START) {
4573 refcount_inc(&t->use_count);
4574 spin_unlock(&fs_info->trans_lock);
4575 btrfs_wait_for_commit(fs_info, t->transid);
4576 btrfs_put_transaction(t);
4577 spin_lock(&fs_info->trans_lock);
4580 if (t == fs_info->running_transaction) {
4581 t->state = TRANS_STATE_COMMIT_DOING;
4582 spin_unlock(&fs_info->trans_lock);
4584 * We wait for 0 num_writers since we don't hold a trans
4585 * handle open currently for this transaction.
4587 wait_event(t->writer_wait,
4588 atomic_read(&t->num_writers) == 0);
4590 spin_unlock(&fs_info->trans_lock);
4592 btrfs_cleanup_one_transaction(t, fs_info);
4594 spin_lock(&fs_info->trans_lock);
4595 if (t == fs_info->running_transaction)
4596 fs_info->running_transaction = NULL;
4597 list_del_init(&t->list);
4598 spin_unlock(&fs_info->trans_lock);
4600 btrfs_put_transaction(t);
4601 trace_btrfs_transaction_commit(fs_info->tree_root);
4602 spin_lock(&fs_info->trans_lock);
4604 spin_unlock(&fs_info->trans_lock);
4605 btrfs_destroy_all_ordered_extents(fs_info);
4606 btrfs_destroy_delayed_inodes(fs_info);
4607 btrfs_assert_delayed_root_empty(fs_info);
4608 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4609 btrfs_destroy_all_delalloc_inodes(fs_info);
4610 mutex_unlock(&fs_info->transaction_kthread_mutex);
4615 static const struct extent_io_ops btree_extent_io_ops = {
4616 /* mandatory callbacks */
4617 .submit_bio_hook = btree_submit_bio_hook,
4618 .readpage_end_io_hook = btree_readpage_end_io_hook,