1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include <crypto/hash.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
43 #include "block-group.h"
45 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
46 BTRFS_HEADER_FLAG_RELOC |\
47 BTRFS_SUPER_FLAG_ERROR |\
48 BTRFS_SUPER_FLAG_SEEDING |\
49 BTRFS_SUPER_FLAG_METADUMP |\
50 BTRFS_SUPER_FLAG_METADUMP_V2)
52 static const struct extent_io_ops btree_extent_io_ops;
53 static void end_workqueue_fn(struct btrfs_work *work);
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
68 * is complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
71 struct btrfs_end_io_wq {
75 struct btrfs_fs_info *info;
77 enum btrfs_wq_endio_type metadata;
78 struct btrfs_work work;
81 static struct kmem_cache *btrfs_end_io_wq_cache;
83 int __init btrfs_end_io_wq_init(void)
85 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 sizeof(struct btrfs_end_io_wq),
90 if (!btrfs_end_io_wq_cache)
95 void __cold btrfs_end_io_wq_exit(void)
97 kmem_cache_destroy(btrfs_end_io_wq_cache);
101 * async submit bios are used to offload expensive checksumming
102 * onto the worker threads. They checksum file and metadata bios
103 * just before they are sent down the IO stack.
105 struct async_submit_bio {
108 extent_submit_bio_start_t *submit_bio_start;
111 * bio_offset is optional, can be used if the pages in the bio
112 * can't tell us where in the file the bio should go
115 struct btrfs_work work;
120 * Lockdep class keys for extent_buffer->lock's in this root. For a given
121 * eb, the lockdep key is determined by the btrfs_root it belongs to and
122 * the level the eb occupies in the tree.
124 * Different roots are used for different purposes and may nest inside each
125 * other and they require separate keysets. As lockdep keys should be
126 * static, assign keysets according to the purpose of the root as indicated
127 * by btrfs_root->root_key.objectid. This ensures that all special purpose
128 * roots have separate keysets.
130 * Lock-nesting across peer nodes is always done with the immediate parent
131 * node locked thus preventing deadlock. As lockdep doesn't know this, use
132 * subclass to avoid triggering lockdep warning in such cases.
134 * The key is set by the readpage_end_io_hook after the buffer has passed
135 * csum validation but before the pages are unlocked. It is also set by
136 * btrfs_init_new_buffer on freshly allocated blocks.
138 * We also add a check to make sure the highest level of the tree is the
139 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
140 * needs update as well.
142 #ifdef CONFIG_DEBUG_LOCK_ALLOC
143 # if BTRFS_MAX_LEVEL != 8
147 static struct btrfs_lockdep_keyset {
148 u64 id; /* root objectid */
149 const char *name_stem; /* lock name stem */
150 char names[BTRFS_MAX_LEVEL + 1][20];
151 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
152 } btrfs_lockdep_keysets[] = {
153 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
154 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
155 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
156 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
157 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
158 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
159 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
160 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
161 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
162 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
163 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
164 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
165 { .id = 0, .name_stem = "tree" },
168 void __init btrfs_init_lockdep(void)
172 /* initialize lockdep class names */
173 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
174 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
176 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
177 snprintf(ks->names[j], sizeof(ks->names[j]),
178 "btrfs-%s-%02d", ks->name_stem, j);
182 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
185 struct btrfs_lockdep_keyset *ks;
187 BUG_ON(level >= ARRAY_SIZE(ks->keys));
189 /* find the matching keyset, id 0 is the default entry */
190 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
191 if (ks->id == objectid)
194 lockdep_set_class_and_name(&eb->lock,
195 &ks->keys[level], ks->names[level]);
201 * extents on the btree inode are pretty simple, there's one extent
202 * that covers the entire device
204 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
205 struct page *page, size_t pg_offset, u64 start, u64 len,
208 struct btrfs_fs_info *fs_info = inode->root->fs_info;
209 struct extent_map_tree *em_tree = &inode->extent_tree;
210 struct extent_map *em;
213 read_lock(&em_tree->lock);
214 em = lookup_extent_mapping(em_tree, start, len);
216 em->bdev = fs_info->fs_devices->latest_bdev;
217 read_unlock(&em_tree->lock);
220 read_unlock(&em_tree->lock);
222 em = alloc_extent_map();
224 em = ERR_PTR(-ENOMEM);
229 em->block_len = (u64)-1;
231 em->bdev = fs_info->fs_devices->latest_bdev;
233 write_lock(&em_tree->lock);
234 ret = add_extent_mapping(em_tree, em, 0);
235 if (ret == -EEXIST) {
237 em = lookup_extent_mapping(em_tree, start, len);
244 write_unlock(&em_tree->lock);
251 * Compute the csum of a btree block and store the result to provided buffer.
253 * Returns error if the extent buffer cannot be mapped.
255 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
257 struct btrfs_fs_info *fs_info = buf->fs_info;
258 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
260 unsigned long cur_len;
261 unsigned long offset = BTRFS_CSUM_SIZE;
263 unsigned long map_start;
264 unsigned long map_len;
267 shash->tfm = fs_info->csum_shash;
268 crypto_shash_init(shash);
270 len = buf->len - offset;
274 * Note: we don't need to check for the err == 1 case here, as
275 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
276 * and 'min_len = 32' and the currently implemented mapping
277 * algorithm we cannot cross a page boundary.
279 err = map_private_extent_buffer(buf, offset, 32,
280 &kaddr, &map_start, &map_len);
283 cur_len = min(len, map_len - (offset - map_start));
284 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
288 memset(result, 0, BTRFS_CSUM_SIZE);
290 crypto_shash_final(shash, result);
296 * we can't consider a given block up to date unless the transid of the
297 * block matches the transid in the parent node's pointer. This is how we
298 * detect blocks that either didn't get written at all or got written
299 * in the wrong place.
301 static int verify_parent_transid(struct extent_io_tree *io_tree,
302 struct extent_buffer *eb, u64 parent_transid,
305 struct extent_state *cached_state = NULL;
307 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
309 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
316 btrfs_tree_read_lock(eb);
317 btrfs_set_lock_blocking_read(eb);
320 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
322 if (extent_buffer_uptodate(eb) &&
323 btrfs_header_generation(eb) == parent_transid) {
327 btrfs_err_rl(eb->fs_info,
328 "parent transid verify failed on %llu wanted %llu found %llu",
330 parent_transid, btrfs_header_generation(eb));
334 * Things reading via commit roots that don't have normal protection,
335 * like send, can have a really old block in cache that may point at a
336 * block that has been freed and re-allocated. So don't clear uptodate
337 * if we find an eb that is under IO (dirty/writeback) because we could
338 * end up reading in the stale data and then writing it back out and
339 * making everybody very sad.
341 if (!extent_buffer_under_io(eb))
342 clear_extent_buffer_uptodate(eb);
344 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
347 btrfs_tree_read_unlock_blocking(eb);
351 static bool btrfs_supported_super_csum(u16 csum_type)
354 case BTRFS_CSUM_TYPE_CRC32:
355 case BTRFS_CSUM_TYPE_XXHASH:
356 case BTRFS_CSUM_TYPE_SHA256:
364 * Return 0 if the superblock checksum type matches the checksum value of that
365 * algorithm. Pass the raw disk superblock data.
367 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
370 struct btrfs_super_block *disk_sb =
371 (struct btrfs_super_block *)raw_disk_sb;
372 char result[BTRFS_CSUM_SIZE];
373 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
375 shash->tfm = fs_info->csum_shash;
376 crypto_shash_init(shash);
379 * The super_block structure does not span the whole
380 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
381 * filled with zeros and is included in the checksum.
383 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
384 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
385 crypto_shash_final(shash, result);
387 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
393 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
394 struct btrfs_key *first_key, u64 parent_transid)
396 struct btrfs_fs_info *fs_info = eb->fs_info;
398 struct btrfs_key found_key;
401 found_level = btrfs_header_level(eb);
402 if (found_level != level) {
403 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
404 KERN_ERR "BTRFS: tree level check failed\n");
406 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
407 eb->start, level, found_level);
415 * For live tree block (new tree blocks in current transaction),
416 * we need proper lock context to avoid race, which is impossible here.
417 * So we only checks tree blocks which is read from disk, whose
418 * generation <= fs_info->last_trans_committed.
420 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
423 /* We have @first_key, so this @eb must have at least one item */
424 if (btrfs_header_nritems(eb) == 0) {
426 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
428 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
433 btrfs_node_key_to_cpu(eb, &found_key, 0);
435 btrfs_item_key_to_cpu(eb, &found_key, 0);
436 ret = btrfs_comp_cpu_keys(first_key, &found_key);
439 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
440 KERN_ERR "BTRFS: tree first key check failed\n");
442 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
443 eb->start, parent_transid, first_key->objectid,
444 first_key->type, first_key->offset,
445 found_key.objectid, found_key.type,
452 * helper to read a given tree block, doing retries as required when
453 * the checksums don't match and we have alternate mirrors to try.
455 * @parent_transid: expected transid, skip check if 0
456 * @level: expected level, mandatory check
457 * @first_key: expected key of first slot, skip check if NULL
459 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
460 u64 parent_transid, int level,
461 struct btrfs_key *first_key)
463 struct btrfs_fs_info *fs_info = eb->fs_info;
464 struct extent_io_tree *io_tree;
469 int failed_mirror = 0;
471 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
473 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
474 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
476 if (verify_parent_transid(io_tree, eb,
479 else if (btrfs_verify_level_key(eb, level,
480 first_key, parent_transid))
486 num_copies = btrfs_num_copies(fs_info,
491 if (!failed_mirror) {
493 failed_mirror = eb->read_mirror;
497 if (mirror_num == failed_mirror)
500 if (mirror_num > num_copies)
504 if (failed && !ret && failed_mirror)
505 btrfs_repair_eb_io_failure(eb, failed_mirror);
511 * checksum a dirty tree block before IO. This has extra checks to make sure
512 * we only fill in the checksum field in the first page of a multi-page block
515 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
517 u64 start = page_offset(page);
519 u8 result[BTRFS_CSUM_SIZE];
520 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
521 struct extent_buffer *eb;
524 eb = (struct extent_buffer *)page->private;
525 if (page != eb->pages[0])
528 found_start = btrfs_header_bytenr(eb);
530 * Please do not consolidate these warnings into a single if.
531 * It is useful to know what went wrong.
533 if (WARN_ON(found_start != start))
535 if (WARN_ON(!PageUptodate(page)))
538 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
539 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
541 if (csum_tree_block(eb, result))
544 if (btrfs_header_level(eb))
545 ret = btrfs_check_node(eb);
547 ret = btrfs_check_leaf_full(eb);
550 btrfs_print_tree(eb, 0);
552 "block=%llu write time tree block corruption detected",
554 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
557 write_extent_buffer(eb, result, 0, csum_size);
562 static int check_tree_block_fsid(struct extent_buffer *eb)
564 struct btrfs_fs_info *fs_info = eb->fs_info;
565 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
566 u8 fsid[BTRFS_FSID_SIZE];
569 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
574 * Checking the incompat flag is only valid for the current
575 * fs. For seed devices it's forbidden to have their uuid
576 * changed so reading ->fsid in this case is fine
578 if (fs_devices == fs_info->fs_devices &&
579 btrfs_fs_incompat(fs_info, METADATA_UUID))
580 metadata_uuid = fs_devices->metadata_uuid;
582 metadata_uuid = fs_devices->fsid;
584 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
588 fs_devices = fs_devices->seed;
593 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
594 u64 phy_offset, struct page *page,
595 u64 start, u64 end, int mirror)
599 struct extent_buffer *eb;
600 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
601 struct btrfs_fs_info *fs_info = root->fs_info;
602 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
604 u8 result[BTRFS_CSUM_SIZE];
610 eb = (struct extent_buffer *)page->private;
612 /* the pending IO might have been the only thing that kept this buffer
613 * in memory. Make sure we have a ref for all this other checks
615 atomic_inc(&eb->refs);
617 reads_done = atomic_dec_and_test(&eb->io_pages);
621 eb->read_mirror = mirror;
622 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
627 found_start = btrfs_header_bytenr(eb);
628 if (found_start != eb->start) {
629 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
630 eb->start, found_start);
634 if (check_tree_block_fsid(eb)) {
635 btrfs_err_rl(fs_info, "bad fsid on block %llu",
640 found_level = btrfs_header_level(eb);
641 if (found_level >= BTRFS_MAX_LEVEL) {
642 btrfs_err(fs_info, "bad tree block level %d on %llu",
643 (int)btrfs_header_level(eb), eb->start);
648 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
651 ret = csum_tree_block(eb, result);
655 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
659 memcpy(&found, result, csum_size);
661 read_extent_buffer(eb, &val, 0, csum_size);
662 btrfs_warn_rl(fs_info,
663 "%s checksum verify failed on %llu wanted %x found %x level %d",
664 fs_info->sb->s_id, eb->start,
665 val, found, btrfs_header_level(eb));
671 * If this is a leaf block and it is corrupt, set the corrupt bit so
672 * that we don't try and read the other copies of this block, just
675 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
676 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
680 if (found_level > 0 && btrfs_check_node(eb))
684 set_extent_buffer_uptodate(eb);
687 "block=%llu read time tree block corruption detected",
691 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
692 btree_readahead_hook(eb, ret);
696 * our io error hook is going to dec the io pages
697 * again, we have to make sure it has something
700 atomic_inc(&eb->io_pages);
701 clear_extent_buffer_uptodate(eb);
703 free_extent_buffer(eb);
708 static void end_workqueue_bio(struct bio *bio)
710 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
711 struct btrfs_fs_info *fs_info;
712 struct btrfs_workqueue *wq;
714 fs_info = end_io_wq->info;
715 end_io_wq->status = bio->bi_status;
717 if (bio_op(bio) == REQ_OP_WRITE) {
718 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
719 wq = fs_info->endio_meta_write_workers;
720 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
721 wq = fs_info->endio_freespace_worker;
722 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
723 wq = fs_info->endio_raid56_workers;
725 wq = fs_info->endio_write_workers;
727 if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
728 wq = fs_info->endio_repair_workers;
729 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
730 wq = fs_info->endio_raid56_workers;
731 else if (end_io_wq->metadata)
732 wq = fs_info->endio_meta_workers;
734 wq = fs_info->endio_workers;
737 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
738 btrfs_queue_work(wq, &end_io_wq->work);
741 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
742 enum btrfs_wq_endio_type metadata)
744 struct btrfs_end_io_wq *end_io_wq;
746 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
748 return BLK_STS_RESOURCE;
750 end_io_wq->private = bio->bi_private;
751 end_io_wq->end_io = bio->bi_end_io;
752 end_io_wq->info = info;
753 end_io_wq->status = 0;
754 end_io_wq->bio = bio;
755 end_io_wq->metadata = metadata;
757 bio->bi_private = end_io_wq;
758 bio->bi_end_io = end_workqueue_bio;
762 static void run_one_async_start(struct btrfs_work *work)
764 struct async_submit_bio *async;
767 async = container_of(work, struct async_submit_bio, work);
768 ret = async->submit_bio_start(async->private_data, async->bio,
775 * In order to insert checksums into the metadata in large chunks, we wait
776 * until bio submission time. All the pages in the bio are checksummed and
777 * sums are attached onto the ordered extent record.
779 * At IO completion time the csums attached on the ordered extent record are
780 * inserted into the tree.
782 static void run_one_async_done(struct btrfs_work *work)
784 struct async_submit_bio *async;
788 async = container_of(work, struct async_submit_bio, work);
789 inode = async->private_data;
791 /* If an error occurred we just want to clean up the bio and move on */
793 async->bio->bi_status = async->status;
794 bio_endio(async->bio);
799 * All of the bios that pass through here are from async helpers.
800 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
801 * This changes nothing when cgroups aren't in use.
803 async->bio->bi_opf |= REQ_CGROUP_PUNT;
804 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
806 async->bio->bi_status = ret;
807 bio_endio(async->bio);
811 static void run_one_async_free(struct btrfs_work *work)
813 struct async_submit_bio *async;
815 async = container_of(work, struct async_submit_bio, work);
819 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
820 int mirror_num, unsigned long bio_flags,
821 u64 bio_offset, void *private_data,
822 extent_submit_bio_start_t *submit_bio_start)
824 struct async_submit_bio *async;
826 async = kmalloc(sizeof(*async), GFP_NOFS);
828 return BLK_STS_RESOURCE;
830 async->private_data = private_data;
832 async->mirror_num = mirror_num;
833 async->submit_bio_start = submit_bio_start;
835 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
838 async->bio_offset = bio_offset;
842 if (op_is_sync(bio->bi_opf))
843 btrfs_set_work_high_priority(&async->work);
845 btrfs_queue_work(fs_info->workers, &async->work);
849 static blk_status_t btree_csum_one_bio(struct bio *bio)
851 struct bio_vec *bvec;
852 struct btrfs_root *root;
854 struct bvec_iter_all iter_all;
856 ASSERT(!bio_flagged(bio, BIO_CLONED));
857 bio_for_each_segment_all(bvec, bio, iter_all) {
858 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
859 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
864 return errno_to_blk_status(ret);
867 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
871 * when we're called for a write, we're already in the async
872 * submission context. Just jump into btrfs_map_bio
874 return btree_csum_one_bio(bio);
877 static int check_async_write(struct btrfs_fs_info *fs_info,
878 struct btrfs_inode *bi)
880 if (atomic_read(&bi->sync_writers))
882 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
887 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
889 unsigned long bio_flags)
891 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
892 int async = check_async_write(fs_info, BTRFS_I(inode));
895 if (bio_op(bio) != REQ_OP_WRITE) {
897 * called for a read, do the setup so that checksum validation
898 * can happen in the async kernel threads
900 ret = btrfs_bio_wq_end_io(fs_info, bio,
901 BTRFS_WQ_ENDIO_METADATA);
904 ret = btrfs_map_bio(fs_info, bio, mirror_num);
906 ret = btree_csum_one_bio(bio);
909 ret = btrfs_map_bio(fs_info, bio, mirror_num);
912 * kthread helpers are used to submit writes so that
913 * checksumming can happen in parallel across all CPUs
915 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
916 0, inode, btree_submit_bio_start);
924 bio->bi_status = ret;
929 #ifdef CONFIG_MIGRATION
930 static int btree_migratepage(struct address_space *mapping,
931 struct page *newpage, struct page *page,
932 enum migrate_mode mode)
935 * we can't safely write a btree page from here,
936 * we haven't done the locking hook
941 * Buffers may be managed in a filesystem specific way.
942 * We must have no buffers or drop them.
944 if (page_has_private(page) &&
945 !try_to_release_page(page, GFP_KERNEL))
947 return migrate_page(mapping, newpage, page, mode);
952 static int btree_writepages(struct address_space *mapping,
953 struct writeback_control *wbc)
955 struct btrfs_fs_info *fs_info;
958 if (wbc->sync_mode == WB_SYNC_NONE) {
960 if (wbc->for_kupdate)
963 fs_info = BTRFS_I(mapping->host)->root->fs_info;
964 /* this is a bit racy, but that's ok */
965 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
966 BTRFS_DIRTY_METADATA_THRESH,
967 fs_info->dirty_metadata_batch);
971 return btree_write_cache_pages(mapping, wbc);
974 static int btree_readpage(struct file *file, struct page *page)
976 struct extent_io_tree *tree;
977 tree = &BTRFS_I(page->mapping->host)->io_tree;
978 return extent_read_full_page(tree, page, btree_get_extent, 0);
981 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
983 if (PageWriteback(page) || PageDirty(page))
986 return try_release_extent_buffer(page);
989 static void btree_invalidatepage(struct page *page, unsigned int offset,
992 struct extent_io_tree *tree;
993 tree = &BTRFS_I(page->mapping->host)->io_tree;
994 extent_invalidatepage(tree, page, offset);
995 btree_releasepage(page, GFP_NOFS);
996 if (PagePrivate(page)) {
997 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
998 "page private not zero on page %llu",
999 (unsigned long long)page_offset(page));
1000 ClearPagePrivate(page);
1001 set_page_private(page, 0);
1006 static int btree_set_page_dirty(struct page *page)
1009 struct extent_buffer *eb;
1011 BUG_ON(!PagePrivate(page));
1012 eb = (struct extent_buffer *)page->private;
1014 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1015 BUG_ON(!atomic_read(&eb->refs));
1016 btrfs_assert_tree_locked(eb);
1018 return __set_page_dirty_nobuffers(page);
1021 static const struct address_space_operations btree_aops = {
1022 .readpage = btree_readpage,
1023 .writepages = btree_writepages,
1024 .releasepage = btree_releasepage,
1025 .invalidatepage = btree_invalidatepage,
1026 #ifdef CONFIG_MIGRATION
1027 .migratepage = btree_migratepage,
1029 .set_page_dirty = btree_set_page_dirty,
1032 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1034 struct extent_buffer *buf = NULL;
1037 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1041 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1043 free_extent_buffer_stale(buf);
1045 free_extent_buffer(buf);
1048 struct extent_buffer *btrfs_find_create_tree_block(
1049 struct btrfs_fs_info *fs_info,
1052 if (btrfs_is_testing(fs_info))
1053 return alloc_test_extent_buffer(fs_info, bytenr);
1054 return alloc_extent_buffer(fs_info, bytenr);
1058 * Read tree block at logical address @bytenr and do variant basic but critical
1061 * @parent_transid: expected transid of this tree block, skip check if 0
1062 * @level: expected level, mandatory check
1063 * @first_key: expected key in slot 0, skip check if NULL
1065 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1066 u64 parent_transid, int level,
1067 struct btrfs_key *first_key)
1069 struct extent_buffer *buf = NULL;
1072 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1076 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1079 free_extent_buffer_stale(buf);
1080 return ERR_PTR(ret);
1086 void btrfs_clean_tree_block(struct extent_buffer *buf)
1088 struct btrfs_fs_info *fs_info = buf->fs_info;
1089 if (btrfs_header_generation(buf) ==
1090 fs_info->running_transaction->transid) {
1091 btrfs_assert_tree_locked(buf);
1093 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1094 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1096 fs_info->dirty_metadata_batch);
1097 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1098 btrfs_set_lock_blocking_write(buf);
1099 clear_extent_buffer_dirty(buf);
1104 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1106 struct btrfs_subvolume_writers *writers;
1109 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1111 return ERR_PTR(-ENOMEM);
1113 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1116 return ERR_PTR(ret);
1119 init_waitqueue_head(&writers->wait);
1124 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1126 percpu_counter_destroy(&writers->counter);
1130 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1133 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1135 root->commit_root = NULL;
1137 root->orphan_cleanup_state = 0;
1139 root->last_trans = 0;
1140 root->highest_objectid = 0;
1141 root->nr_delalloc_inodes = 0;
1142 root->nr_ordered_extents = 0;
1143 root->inode_tree = RB_ROOT;
1144 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1145 root->block_rsv = NULL;
1147 INIT_LIST_HEAD(&root->dirty_list);
1148 INIT_LIST_HEAD(&root->root_list);
1149 INIT_LIST_HEAD(&root->delalloc_inodes);
1150 INIT_LIST_HEAD(&root->delalloc_root);
1151 INIT_LIST_HEAD(&root->ordered_extents);
1152 INIT_LIST_HEAD(&root->ordered_root);
1153 INIT_LIST_HEAD(&root->reloc_dirty_list);
1154 INIT_LIST_HEAD(&root->logged_list[0]);
1155 INIT_LIST_HEAD(&root->logged_list[1]);
1156 spin_lock_init(&root->inode_lock);
1157 spin_lock_init(&root->delalloc_lock);
1158 spin_lock_init(&root->ordered_extent_lock);
1159 spin_lock_init(&root->accounting_lock);
1160 spin_lock_init(&root->log_extents_lock[0]);
1161 spin_lock_init(&root->log_extents_lock[1]);
1162 spin_lock_init(&root->qgroup_meta_rsv_lock);
1163 mutex_init(&root->objectid_mutex);
1164 mutex_init(&root->log_mutex);
1165 mutex_init(&root->ordered_extent_mutex);
1166 mutex_init(&root->delalloc_mutex);
1167 init_waitqueue_head(&root->log_writer_wait);
1168 init_waitqueue_head(&root->log_commit_wait[0]);
1169 init_waitqueue_head(&root->log_commit_wait[1]);
1170 INIT_LIST_HEAD(&root->log_ctxs[0]);
1171 INIT_LIST_HEAD(&root->log_ctxs[1]);
1172 atomic_set(&root->log_commit[0], 0);
1173 atomic_set(&root->log_commit[1], 0);
1174 atomic_set(&root->log_writers, 0);
1175 atomic_set(&root->log_batch, 0);
1176 refcount_set(&root->refs, 1);
1177 atomic_set(&root->will_be_snapshotted, 0);
1178 atomic_set(&root->snapshot_force_cow, 0);
1179 atomic_set(&root->nr_swapfiles, 0);
1180 root->log_transid = 0;
1181 root->log_transid_committed = -1;
1182 root->last_log_commit = 0;
1184 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1185 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1187 memset(&root->root_key, 0, sizeof(root->root_key));
1188 memset(&root->root_item, 0, sizeof(root->root_item));
1189 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1191 root->defrag_trans_start = fs_info->generation;
1193 root->defrag_trans_start = 0;
1194 root->root_key.objectid = objectid;
1197 spin_lock_init(&root->root_item_lock);
1198 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1201 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1204 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1206 root->fs_info = fs_info;
1210 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1211 /* Should only be used by the testing infrastructure */
1212 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1214 struct btrfs_root *root;
1217 return ERR_PTR(-EINVAL);
1219 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1221 return ERR_PTR(-ENOMEM);
1223 /* We don't use the stripesize in selftest, set it as sectorsize */
1224 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1225 root->alloc_bytenr = 0;
1231 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1234 struct btrfs_fs_info *fs_info = trans->fs_info;
1235 struct extent_buffer *leaf;
1236 struct btrfs_root *tree_root = fs_info->tree_root;
1237 struct btrfs_root *root;
1238 struct btrfs_key key;
1239 unsigned int nofs_flag;
1241 uuid_le uuid = NULL_UUID_LE;
1244 * We're holding a transaction handle, so use a NOFS memory allocation
1245 * context to avoid deadlock if reclaim happens.
1247 nofs_flag = memalloc_nofs_save();
1248 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1249 memalloc_nofs_restore(nofs_flag);
1251 return ERR_PTR(-ENOMEM);
1253 __setup_root(root, fs_info, objectid);
1254 root->root_key.objectid = objectid;
1255 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1256 root->root_key.offset = 0;
1258 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1260 ret = PTR_ERR(leaf);
1266 btrfs_mark_buffer_dirty(leaf);
1268 root->commit_root = btrfs_root_node(root);
1269 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1271 root->root_item.flags = 0;
1272 root->root_item.byte_limit = 0;
1273 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1274 btrfs_set_root_generation(&root->root_item, trans->transid);
1275 btrfs_set_root_level(&root->root_item, 0);
1276 btrfs_set_root_refs(&root->root_item, 1);
1277 btrfs_set_root_used(&root->root_item, leaf->len);
1278 btrfs_set_root_last_snapshot(&root->root_item, 0);
1279 btrfs_set_root_dirid(&root->root_item, 0);
1280 if (is_fstree(objectid))
1282 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1283 root->root_item.drop_level = 0;
1285 key.objectid = objectid;
1286 key.type = BTRFS_ROOT_ITEM_KEY;
1288 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1292 btrfs_tree_unlock(leaf);
1298 btrfs_tree_unlock(leaf);
1299 free_extent_buffer(root->commit_root);
1300 free_extent_buffer(leaf);
1304 return ERR_PTR(ret);
1307 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1308 struct btrfs_fs_info *fs_info)
1310 struct btrfs_root *root;
1311 struct extent_buffer *leaf;
1313 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1315 return ERR_PTR(-ENOMEM);
1317 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1319 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1320 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1321 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1324 * DON'T set REF_COWS for log trees
1326 * log trees do not get reference counted because they go away
1327 * before a real commit is actually done. They do store pointers
1328 * to file data extents, and those reference counts still get
1329 * updated (along with back refs to the log tree).
1332 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1336 return ERR_CAST(leaf);
1341 btrfs_mark_buffer_dirty(root->node);
1342 btrfs_tree_unlock(root->node);
1346 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1347 struct btrfs_fs_info *fs_info)
1349 struct btrfs_root *log_root;
1351 log_root = alloc_log_tree(trans, fs_info);
1352 if (IS_ERR(log_root))
1353 return PTR_ERR(log_root);
1354 WARN_ON(fs_info->log_root_tree);
1355 fs_info->log_root_tree = log_root;
1359 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1360 struct btrfs_root *root)
1362 struct btrfs_fs_info *fs_info = root->fs_info;
1363 struct btrfs_root *log_root;
1364 struct btrfs_inode_item *inode_item;
1366 log_root = alloc_log_tree(trans, fs_info);
1367 if (IS_ERR(log_root))
1368 return PTR_ERR(log_root);
1370 log_root->last_trans = trans->transid;
1371 log_root->root_key.offset = root->root_key.objectid;
1373 inode_item = &log_root->root_item.inode;
1374 btrfs_set_stack_inode_generation(inode_item, 1);
1375 btrfs_set_stack_inode_size(inode_item, 3);
1376 btrfs_set_stack_inode_nlink(inode_item, 1);
1377 btrfs_set_stack_inode_nbytes(inode_item,
1379 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1381 btrfs_set_root_node(&log_root->root_item, log_root->node);
1383 WARN_ON(root->log_root);
1384 root->log_root = log_root;
1385 root->log_transid = 0;
1386 root->log_transid_committed = -1;
1387 root->last_log_commit = 0;
1391 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1392 struct btrfs_key *key)
1394 struct btrfs_root *root;
1395 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1396 struct btrfs_path *path;
1401 path = btrfs_alloc_path();
1403 return ERR_PTR(-ENOMEM);
1405 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1411 __setup_root(root, fs_info, key->objectid);
1413 ret = btrfs_find_root(tree_root, key, path,
1414 &root->root_item, &root->root_key);
1421 generation = btrfs_root_generation(&root->root_item);
1422 level = btrfs_root_level(&root->root_item);
1423 root->node = read_tree_block(fs_info,
1424 btrfs_root_bytenr(&root->root_item),
1425 generation, level, NULL);
1426 if (IS_ERR(root->node)) {
1427 ret = PTR_ERR(root->node);
1429 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1431 free_extent_buffer(root->node);
1434 root->commit_root = btrfs_root_node(root);
1436 btrfs_free_path(path);
1442 root = ERR_PTR(ret);
1446 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1447 struct btrfs_key *location)
1449 struct btrfs_root *root;
1451 root = btrfs_read_tree_root(tree_root, location);
1455 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1456 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1457 btrfs_check_and_init_root_item(&root->root_item);
1463 int btrfs_init_fs_root(struct btrfs_root *root)
1466 struct btrfs_subvolume_writers *writers;
1468 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1469 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1471 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1476 writers = btrfs_alloc_subvolume_writers();
1477 if (IS_ERR(writers)) {
1478 ret = PTR_ERR(writers);
1481 root->subv_writers = writers;
1483 btrfs_init_free_ino_ctl(root);
1484 spin_lock_init(&root->ino_cache_lock);
1485 init_waitqueue_head(&root->ino_cache_wait);
1487 ret = get_anon_bdev(&root->anon_dev);
1491 mutex_lock(&root->objectid_mutex);
1492 ret = btrfs_find_highest_objectid(root,
1493 &root->highest_objectid);
1495 mutex_unlock(&root->objectid_mutex);
1499 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1501 mutex_unlock(&root->objectid_mutex);
1505 /* The caller is responsible to call btrfs_free_fs_root */
1509 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1512 struct btrfs_root *root;
1514 spin_lock(&fs_info->fs_roots_radix_lock);
1515 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1516 (unsigned long)root_id);
1517 spin_unlock(&fs_info->fs_roots_radix_lock);
1521 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1522 struct btrfs_root *root)
1526 ret = radix_tree_preload(GFP_NOFS);
1530 spin_lock(&fs_info->fs_roots_radix_lock);
1531 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1532 (unsigned long)root->root_key.objectid,
1535 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1536 spin_unlock(&fs_info->fs_roots_radix_lock);
1537 radix_tree_preload_end();
1542 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1543 struct btrfs_key *location,
1546 struct btrfs_root *root;
1547 struct btrfs_path *path;
1548 struct btrfs_key key;
1551 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1552 return fs_info->tree_root;
1553 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1554 return fs_info->extent_root;
1555 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1556 return fs_info->chunk_root;
1557 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1558 return fs_info->dev_root;
1559 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1560 return fs_info->csum_root;
1561 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1562 return fs_info->quota_root ? fs_info->quota_root :
1564 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1565 return fs_info->uuid_root ? fs_info->uuid_root :
1567 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1568 return fs_info->free_space_root ? fs_info->free_space_root :
1571 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1573 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1574 return ERR_PTR(-ENOENT);
1578 root = btrfs_read_fs_root(fs_info->tree_root, location);
1582 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1587 ret = btrfs_init_fs_root(root);
1591 path = btrfs_alloc_path();
1596 key.objectid = BTRFS_ORPHAN_OBJECTID;
1597 key.type = BTRFS_ORPHAN_ITEM_KEY;
1598 key.offset = location->objectid;
1600 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1601 btrfs_free_path(path);
1605 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1607 ret = btrfs_insert_fs_root(fs_info, root);
1609 if (ret == -EEXIST) {
1610 btrfs_free_fs_root(root);
1617 btrfs_free_fs_root(root);
1618 return ERR_PTR(ret);
1621 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1623 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1625 struct btrfs_device *device;
1626 struct backing_dev_info *bdi;
1629 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1632 bdi = device->bdev->bd_bdi;
1633 if (bdi_congested(bdi, bdi_bits)) {
1643 * called by the kthread helper functions to finally call the bio end_io
1644 * functions. This is where read checksum verification actually happens
1646 static void end_workqueue_fn(struct btrfs_work *work)
1649 struct btrfs_end_io_wq *end_io_wq;
1651 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1652 bio = end_io_wq->bio;
1654 bio->bi_status = end_io_wq->status;
1655 bio->bi_private = end_io_wq->private;
1656 bio->bi_end_io = end_io_wq->end_io;
1658 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1661 static int cleaner_kthread(void *arg)
1663 struct btrfs_root *root = arg;
1664 struct btrfs_fs_info *fs_info = root->fs_info;
1670 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1672 /* Make the cleaner go to sleep early. */
1673 if (btrfs_need_cleaner_sleep(fs_info))
1677 * Do not do anything if we might cause open_ctree() to block
1678 * before we have finished mounting the filesystem.
1680 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1683 if (!mutex_trylock(&fs_info->cleaner_mutex))
1687 * Avoid the problem that we change the status of the fs
1688 * during the above check and trylock.
1690 if (btrfs_need_cleaner_sleep(fs_info)) {
1691 mutex_unlock(&fs_info->cleaner_mutex);
1695 btrfs_run_delayed_iputs(fs_info);
1697 again = btrfs_clean_one_deleted_snapshot(root);
1698 mutex_unlock(&fs_info->cleaner_mutex);
1701 * The defragger has dealt with the R/O remount and umount,
1702 * needn't do anything special here.
1704 btrfs_run_defrag_inodes(fs_info);
1707 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1708 * with relocation (btrfs_relocate_chunk) and relocation
1709 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1710 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1711 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1712 * unused block groups.
1714 btrfs_delete_unused_bgs(fs_info);
1716 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1717 if (kthread_should_park())
1719 if (kthread_should_stop())
1722 set_current_state(TASK_INTERRUPTIBLE);
1724 __set_current_state(TASK_RUNNING);
1729 static int transaction_kthread(void *arg)
1731 struct btrfs_root *root = arg;
1732 struct btrfs_fs_info *fs_info = root->fs_info;
1733 struct btrfs_trans_handle *trans;
1734 struct btrfs_transaction *cur;
1737 unsigned long delay;
1741 cannot_commit = false;
1742 delay = HZ * fs_info->commit_interval;
1743 mutex_lock(&fs_info->transaction_kthread_mutex);
1745 spin_lock(&fs_info->trans_lock);
1746 cur = fs_info->running_transaction;
1748 spin_unlock(&fs_info->trans_lock);
1752 now = ktime_get_seconds();
1753 if (cur->state < TRANS_STATE_COMMIT_START &&
1754 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1755 (now < cur->start_time ||
1756 now - cur->start_time < fs_info->commit_interval)) {
1757 spin_unlock(&fs_info->trans_lock);
1761 transid = cur->transid;
1762 spin_unlock(&fs_info->trans_lock);
1764 /* If the file system is aborted, this will always fail. */
1765 trans = btrfs_attach_transaction(root);
1766 if (IS_ERR(trans)) {
1767 if (PTR_ERR(trans) != -ENOENT)
1768 cannot_commit = true;
1771 if (transid == trans->transid) {
1772 btrfs_commit_transaction(trans);
1774 btrfs_end_transaction(trans);
1777 wake_up_process(fs_info->cleaner_kthread);
1778 mutex_unlock(&fs_info->transaction_kthread_mutex);
1780 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1781 &fs_info->fs_state)))
1782 btrfs_cleanup_transaction(fs_info);
1783 if (!kthread_should_stop() &&
1784 (!btrfs_transaction_blocked(fs_info) ||
1786 schedule_timeout_interruptible(delay);
1787 } while (!kthread_should_stop());
1792 * this will find the highest generation in the array of
1793 * root backups. The index of the highest array is returned,
1794 * or -1 if we can't find anything.
1796 * We check to make sure the array is valid by comparing the
1797 * generation of the latest root in the array with the generation
1798 * in the super block. If they don't match we pitch it.
1800 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1803 int newest_index = -1;
1804 struct btrfs_root_backup *root_backup;
1807 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1808 root_backup = info->super_copy->super_roots + i;
1809 cur = btrfs_backup_tree_root_gen(root_backup);
1810 if (cur == newest_gen)
1814 /* check to see if we actually wrapped around */
1815 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1816 root_backup = info->super_copy->super_roots;
1817 cur = btrfs_backup_tree_root_gen(root_backup);
1818 if (cur == newest_gen)
1821 return newest_index;
1826 * find the oldest backup so we know where to store new entries
1827 * in the backup array. This will set the backup_root_index
1828 * field in the fs_info struct
1830 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1833 int newest_index = -1;
1835 newest_index = find_newest_super_backup(info, newest_gen);
1836 /* if there was garbage in there, just move along */
1837 if (newest_index == -1) {
1838 info->backup_root_index = 0;
1840 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1845 * copy all the root pointers into the super backup array.
1846 * this will bump the backup pointer by one when it is
1849 static void backup_super_roots(struct btrfs_fs_info *info)
1852 struct btrfs_root_backup *root_backup;
1855 next_backup = info->backup_root_index;
1856 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1857 BTRFS_NUM_BACKUP_ROOTS;
1860 * just overwrite the last backup if we're at the same generation
1861 * this happens only at umount
1863 root_backup = info->super_for_commit->super_roots + last_backup;
1864 if (btrfs_backup_tree_root_gen(root_backup) ==
1865 btrfs_header_generation(info->tree_root->node))
1866 next_backup = last_backup;
1868 root_backup = info->super_for_commit->super_roots + next_backup;
1871 * make sure all of our padding and empty slots get zero filled
1872 * regardless of which ones we use today
1874 memset(root_backup, 0, sizeof(*root_backup));
1876 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1878 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1879 btrfs_set_backup_tree_root_gen(root_backup,
1880 btrfs_header_generation(info->tree_root->node));
1882 btrfs_set_backup_tree_root_level(root_backup,
1883 btrfs_header_level(info->tree_root->node));
1885 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1886 btrfs_set_backup_chunk_root_gen(root_backup,
1887 btrfs_header_generation(info->chunk_root->node));
1888 btrfs_set_backup_chunk_root_level(root_backup,
1889 btrfs_header_level(info->chunk_root->node));
1891 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1892 btrfs_set_backup_extent_root_gen(root_backup,
1893 btrfs_header_generation(info->extent_root->node));
1894 btrfs_set_backup_extent_root_level(root_backup,
1895 btrfs_header_level(info->extent_root->node));
1898 * we might commit during log recovery, which happens before we set
1899 * the fs_root. Make sure it is valid before we fill it in.
1901 if (info->fs_root && info->fs_root->node) {
1902 btrfs_set_backup_fs_root(root_backup,
1903 info->fs_root->node->start);
1904 btrfs_set_backup_fs_root_gen(root_backup,
1905 btrfs_header_generation(info->fs_root->node));
1906 btrfs_set_backup_fs_root_level(root_backup,
1907 btrfs_header_level(info->fs_root->node));
1910 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1911 btrfs_set_backup_dev_root_gen(root_backup,
1912 btrfs_header_generation(info->dev_root->node));
1913 btrfs_set_backup_dev_root_level(root_backup,
1914 btrfs_header_level(info->dev_root->node));
1916 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1917 btrfs_set_backup_csum_root_gen(root_backup,
1918 btrfs_header_generation(info->csum_root->node));
1919 btrfs_set_backup_csum_root_level(root_backup,
1920 btrfs_header_level(info->csum_root->node));
1922 btrfs_set_backup_total_bytes(root_backup,
1923 btrfs_super_total_bytes(info->super_copy));
1924 btrfs_set_backup_bytes_used(root_backup,
1925 btrfs_super_bytes_used(info->super_copy));
1926 btrfs_set_backup_num_devices(root_backup,
1927 btrfs_super_num_devices(info->super_copy));
1930 * if we don't copy this out to the super_copy, it won't get remembered
1931 * for the next commit
1933 memcpy(&info->super_copy->super_roots,
1934 &info->super_for_commit->super_roots,
1935 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1939 * this copies info out of the root backup array and back into
1940 * the in-memory super block. It is meant to help iterate through
1941 * the array, so you send it the number of backups you've already
1942 * tried and the last backup index you used.
1944 * this returns -1 when it has tried all the backups
1946 static noinline int next_root_backup(struct btrfs_fs_info *info,
1947 struct btrfs_super_block *super,
1948 int *num_backups_tried, int *backup_index)
1950 struct btrfs_root_backup *root_backup;
1951 int newest = *backup_index;
1953 if (*num_backups_tried == 0) {
1954 u64 gen = btrfs_super_generation(super);
1956 newest = find_newest_super_backup(info, gen);
1960 *backup_index = newest;
1961 *num_backups_tried = 1;
1962 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1963 /* we've tried all the backups, all done */
1966 /* jump to the next oldest backup */
1967 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1968 BTRFS_NUM_BACKUP_ROOTS;
1969 *backup_index = newest;
1970 *num_backups_tried += 1;
1972 root_backup = super->super_roots + newest;
1974 btrfs_set_super_generation(super,
1975 btrfs_backup_tree_root_gen(root_backup));
1976 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1977 btrfs_set_super_root_level(super,
1978 btrfs_backup_tree_root_level(root_backup));
1979 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1982 * fixme: the total bytes and num_devices need to match or we should
1985 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1986 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1990 /* helper to cleanup workers */
1991 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1993 btrfs_destroy_workqueue(fs_info->fixup_workers);
1994 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1995 btrfs_destroy_workqueue(fs_info->workers);
1996 btrfs_destroy_workqueue(fs_info->endio_workers);
1997 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1998 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1999 btrfs_destroy_workqueue(fs_info->rmw_workers);
2000 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2001 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2002 btrfs_destroy_workqueue(fs_info->delayed_workers);
2003 btrfs_destroy_workqueue(fs_info->caching_workers);
2004 btrfs_destroy_workqueue(fs_info->readahead_workers);
2005 btrfs_destroy_workqueue(fs_info->flush_workers);
2006 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2008 * Now that all other work queues are destroyed, we can safely destroy
2009 * the queues used for metadata I/O, since tasks from those other work
2010 * queues can do metadata I/O operations.
2012 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2013 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2016 static void free_root_extent_buffers(struct btrfs_root *root)
2019 free_extent_buffer(root->node);
2020 free_extent_buffer(root->commit_root);
2022 root->commit_root = NULL;
2026 /* helper to cleanup tree roots */
2027 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2029 free_root_extent_buffers(info->tree_root);
2031 free_root_extent_buffers(info->dev_root);
2032 free_root_extent_buffers(info->extent_root);
2033 free_root_extent_buffers(info->csum_root);
2034 free_root_extent_buffers(info->quota_root);
2035 free_root_extent_buffers(info->uuid_root);
2036 if (free_chunk_root)
2037 free_root_extent_buffers(info->chunk_root);
2038 free_root_extent_buffers(info->free_space_root);
2041 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2044 struct btrfs_root *gang[8];
2047 while (!list_empty(&fs_info->dead_roots)) {
2048 gang[0] = list_entry(fs_info->dead_roots.next,
2049 struct btrfs_root, root_list);
2050 list_del(&gang[0]->root_list);
2052 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2053 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2055 free_extent_buffer(gang[0]->node);
2056 free_extent_buffer(gang[0]->commit_root);
2057 btrfs_put_fs_root(gang[0]);
2062 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2067 for (i = 0; i < ret; i++)
2068 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2071 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2072 btrfs_free_log_root_tree(NULL, fs_info);
2073 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2077 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2079 mutex_init(&fs_info->scrub_lock);
2080 atomic_set(&fs_info->scrubs_running, 0);
2081 atomic_set(&fs_info->scrub_pause_req, 0);
2082 atomic_set(&fs_info->scrubs_paused, 0);
2083 atomic_set(&fs_info->scrub_cancel_req, 0);
2084 init_waitqueue_head(&fs_info->scrub_pause_wait);
2085 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2088 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2090 spin_lock_init(&fs_info->balance_lock);
2091 mutex_init(&fs_info->balance_mutex);
2092 atomic_set(&fs_info->balance_pause_req, 0);
2093 atomic_set(&fs_info->balance_cancel_req, 0);
2094 fs_info->balance_ctl = NULL;
2095 init_waitqueue_head(&fs_info->balance_wait_q);
2098 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2100 struct inode *inode = fs_info->btree_inode;
2102 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2103 set_nlink(inode, 1);
2105 * we set the i_size on the btree inode to the max possible int.
2106 * the real end of the address space is determined by all of
2107 * the devices in the system
2109 inode->i_size = OFFSET_MAX;
2110 inode->i_mapping->a_ops = &btree_aops;
2112 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2113 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2114 IO_TREE_INODE_IO, inode);
2115 BTRFS_I(inode)->io_tree.track_uptodate = false;
2116 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2118 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2120 BTRFS_I(inode)->root = fs_info->tree_root;
2121 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2122 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2123 btrfs_insert_inode_hash(inode);
2126 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2128 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2129 init_rwsem(&fs_info->dev_replace.rwsem);
2130 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2133 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2135 spin_lock_init(&fs_info->qgroup_lock);
2136 mutex_init(&fs_info->qgroup_ioctl_lock);
2137 fs_info->qgroup_tree = RB_ROOT;
2138 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2139 fs_info->qgroup_seq = 1;
2140 fs_info->qgroup_ulist = NULL;
2141 fs_info->qgroup_rescan_running = false;
2142 mutex_init(&fs_info->qgroup_rescan_lock);
2145 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2146 struct btrfs_fs_devices *fs_devices)
2148 u32 max_active = fs_info->thread_pool_size;
2149 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2152 btrfs_alloc_workqueue(fs_info, "worker",
2153 flags | WQ_HIGHPRI, max_active, 16);
2155 fs_info->delalloc_workers =
2156 btrfs_alloc_workqueue(fs_info, "delalloc",
2157 flags, max_active, 2);
2159 fs_info->flush_workers =
2160 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2161 flags, max_active, 0);
2163 fs_info->caching_workers =
2164 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2166 fs_info->fixup_workers =
2167 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2170 * endios are largely parallel and should have a very
2173 fs_info->endio_workers =
2174 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2175 fs_info->endio_meta_workers =
2176 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2178 fs_info->endio_meta_write_workers =
2179 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2181 fs_info->endio_raid56_workers =
2182 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2184 fs_info->endio_repair_workers =
2185 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2186 fs_info->rmw_workers =
2187 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2188 fs_info->endio_write_workers =
2189 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2191 fs_info->endio_freespace_worker =
2192 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2194 fs_info->delayed_workers =
2195 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2197 fs_info->readahead_workers =
2198 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2200 fs_info->qgroup_rescan_workers =
2201 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2203 if (!(fs_info->workers && fs_info->delalloc_workers &&
2204 fs_info->flush_workers &&
2205 fs_info->endio_workers && fs_info->endio_meta_workers &&
2206 fs_info->endio_meta_write_workers &&
2207 fs_info->endio_repair_workers &&
2208 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2209 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2210 fs_info->caching_workers && fs_info->readahead_workers &&
2211 fs_info->fixup_workers && fs_info->delayed_workers &&
2212 fs_info->qgroup_rescan_workers)) {
2219 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2221 struct crypto_shash *csum_shash;
2222 const char *csum_name = btrfs_super_csum_name(csum_type);
2224 csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2226 if (IS_ERR(csum_shash)) {
2227 btrfs_err(fs_info, "error allocating %s hash for checksum",
2229 return PTR_ERR(csum_shash);
2232 fs_info->csum_shash = csum_shash;
2237 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2239 crypto_free_shash(fs_info->csum_shash);
2242 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2243 struct btrfs_fs_devices *fs_devices)
2246 struct btrfs_root *log_tree_root;
2247 struct btrfs_super_block *disk_super = fs_info->super_copy;
2248 u64 bytenr = btrfs_super_log_root(disk_super);
2249 int level = btrfs_super_log_root_level(disk_super);
2251 if (fs_devices->rw_devices == 0) {
2252 btrfs_warn(fs_info, "log replay required on RO media");
2256 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2260 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2262 log_tree_root->node = read_tree_block(fs_info, bytenr,
2263 fs_info->generation + 1,
2265 if (IS_ERR(log_tree_root->node)) {
2266 btrfs_warn(fs_info, "failed to read log tree");
2267 ret = PTR_ERR(log_tree_root->node);
2268 kfree(log_tree_root);
2270 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2271 btrfs_err(fs_info, "failed to read log tree");
2272 free_extent_buffer(log_tree_root->node);
2273 kfree(log_tree_root);
2276 /* returns with log_tree_root freed on success */
2277 ret = btrfs_recover_log_trees(log_tree_root);
2279 btrfs_handle_fs_error(fs_info, ret,
2280 "Failed to recover log tree");
2281 free_extent_buffer(log_tree_root->node);
2282 kfree(log_tree_root);
2286 if (sb_rdonly(fs_info->sb)) {
2287 ret = btrfs_commit_super(fs_info);
2295 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2297 struct btrfs_root *tree_root = fs_info->tree_root;
2298 struct btrfs_root *root;
2299 struct btrfs_key location;
2302 BUG_ON(!fs_info->tree_root);
2304 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2305 location.type = BTRFS_ROOT_ITEM_KEY;
2306 location.offset = 0;
2308 root = btrfs_read_tree_root(tree_root, &location);
2310 ret = PTR_ERR(root);
2313 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2314 fs_info->extent_root = root;
2316 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2317 root = btrfs_read_tree_root(tree_root, &location);
2319 ret = PTR_ERR(root);
2322 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2323 fs_info->dev_root = root;
2324 btrfs_init_devices_late(fs_info);
2326 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2327 root = btrfs_read_tree_root(tree_root, &location);
2329 ret = PTR_ERR(root);
2332 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2333 fs_info->csum_root = root;
2335 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2336 root = btrfs_read_tree_root(tree_root, &location);
2337 if (!IS_ERR(root)) {
2338 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2339 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2340 fs_info->quota_root = root;
2343 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2344 root = btrfs_read_tree_root(tree_root, &location);
2346 ret = PTR_ERR(root);
2350 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2351 fs_info->uuid_root = root;
2354 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2355 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2356 root = btrfs_read_tree_root(tree_root, &location);
2358 ret = PTR_ERR(root);
2361 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2362 fs_info->free_space_root = root;
2367 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2368 location.objectid, ret);
2373 * Real super block validation
2374 * NOTE: super csum type and incompat features will not be checked here.
2376 * @sb: super block to check
2377 * @mirror_num: the super block number to check its bytenr:
2378 * 0 the primary (1st) sb
2379 * 1, 2 2nd and 3rd backup copy
2380 * -1 skip bytenr check
2382 static int validate_super(struct btrfs_fs_info *fs_info,
2383 struct btrfs_super_block *sb, int mirror_num)
2385 u64 nodesize = btrfs_super_nodesize(sb);
2386 u64 sectorsize = btrfs_super_sectorsize(sb);
2389 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2390 btrfs_err(fs_info, "no valid FS found");
2393 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2394 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2395 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2398 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2399 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2400 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2403 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2404 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2405 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2408 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2409 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2410 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2415 * Check sectorsize and nodesize first, other check will need it.
2416 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2418 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2419 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2420 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2423 /* Only PAGE SIZE is supported yet */
2424 if (sectorsize != PAGE_SIZE) {
2426 "sectorsize %llu not supported yet, only support %lu",
2427 sectorsize, PAGE_SIZE);
2430 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2431 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2432 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2435 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2436 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2437 le32_to_cpu(sb->__unused_leafsize), nodesize);
2441 /* Root alignment check */
2442 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2443 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2444 btrfs_super_root(sb));
2447 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2448 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2449 btrfs_super_chunk_root(sb));
2452 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2453 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2454 btrfs_super_log_root(sb));
2458 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2459 BTRFS_FSID_SIZE) != 0) {
2461 "dev_item UUID does not match metadata fsid: %pU != %pU",
2462 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2467 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2470 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2471 btrfs_err(fs_info, "bytes_used is too small %llu",
2472 btrfs_super_bytes_used(sb));
2475 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2476 btrfs_err(fs_info, "invalid stripesize %u",
2477 btrfs_super_stripesize(sb));
2480 if (btrfs_super_num_devices(sb) > (1UL << 31))
2481 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2482 btrfs_super_num_devices(sb));
2483 if (btrfs_super_num_devices(sb) == 0) {
2484 btrfs_err(fs_info, "number of devices is 0");
2488 if (mirror_num >= 0 &&
2489 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2490 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2491 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2496 * Obvious sys_chunk_array corruptions, it must hold at least one key
2499 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2500 btrfs_err(fs_info, "system chunk array too big %u > %u",
2501 btrfs_super_sys_array_size(sb),
2502 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2505 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2506 + sizeof(struct btrfs_chunk)) {
2507 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2508 btrfs_super_sys_array_size(sb),
2509 sizeof(struct btrfs_disk_key)
2510 + sizeof(struct btrfs_chunk));
2515 * The generation is a global counter, we'll trust it more than the others
2516 * but it's still possible that it's the one that's wrong.
2518 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2520 "suspicious: generation < chunk_root_generation: %llu < %llu",
2521 btrfs_super_generation(sb),
2522 btrfs_super_chunk_root_generation(sb));
2523 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2524 && btrfs_super_cache_generation(sb) != (u64)-1)
2526 "suspicious: generation < cache_generation: %llu < %llu",
2527 btrfs_super_generation(sb),
2528 btrfs_super_cache_generation(sb));
2534 * Validation of super block at mount time.
2535 * Some checks already done early at mount time, like csum type and incompat
2536 * flags will be skipped.
2538 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2540 return validate_super(fs_info, fs_info->super_copy, 0);
2544 * Validation of super block at write time.
2545 * Some checks like bytenr check will be skipped as their values will be
2547 * Extra checks like csum type and incompat flags will be done here.
2549 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2550 struct btrfs_super_block *sb)
2554 ret = validate_super(fs_info, sb, -1);
2557 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2559 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2560 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2563 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2566 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2567 btrfs_super_incompat_flags(sb),
2568 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2574 "super block corruption detected before writing it to disk");
2578 int __cold open_ctree(struct super_block *sb,
2579 struct btrfs_fs_devices *fs_devices,
2588 struct btrfs_key location;
2589 struct buffer_head *bh;
2590 struct btrfs_super_block *disk_super;
2591 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2592 struct btrfs_root *tree_root;
2593 struct btrfs_root *chunk_root;
2596 int num_backups_tried = 0;
2597 int backup_index = 0;
2598 int clear_free_space_tree = 0;
2601 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2602 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2603 if (!tree_root || !chunk_root) {
2608 ret = init_srcu_struct(&fs_info->subvol_srcu);
2614 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2620 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2623 goto fail_dio_bytes;
2625 fs_info->dirty_metadata_batch = PAGE_SIZE *
2626 (1 + ilog2(nr_cpu_ids));
2628 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2631 goto fail_dirty_metadata_bytes;
2634 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2638 goto fail_delalloc_bytes;
2641 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2642 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2643 INIT_LIST_HEAD(&fs_info->trans_list);
2644 INIT_LIST_HEAD(&fs_info->dead_roots);
2645 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2646 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2647 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2648 spin_lock_init(&fs_info->delalloc_root_lock);
2649 spin_lock_init(&fs_info->trans_lock);
2650 spin_lock_init(&fs_info->fs_roots_radix_lock);
2651 spin_lock_init(&fs_info->delayed_iput_lock);
2652 spin_lock_init(&fs_info->defrag_inodes_lock);
2653 spin_lock_init(&fs_info->tree_mod_seq_lock);
2654 spin_lock_init(&fs_info->super_lock);
2655 spin_lock_init(&fs_info->buffer_lock);
2656 spin_lock_init(&fs_info->unused_bgs_lock);
2657 rwlock_init(&fs_info->tree_mod_log_lock);
2658 mutex_init(&fs_info->unused_bg_unpin_mutex);
2659 mutex_init(&fs_info->delete_unused_bgs_mutex);
2660 mutex_init(&fs_info->reloc_mutex);
2661 mutex_init(&fs_info->delalloc_root_mutex);
2662 seqlock_init(&fs_info->profiles_lock);
2664 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2665 INIT_LIST_HEAD(&fs_info->space_info);
2666 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2667 INIT_LIST_HEAD(&fs_info->unused_bgs);
2668 extent_map_tree_init(&fs_info->mapping_tree);
2669 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2670 BTRFS_BLOCK_RSV_GLOBAL);
2671 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2672 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2673 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2674 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2675 BTRFS_BLOCK_RSV_DELOPS);
2676 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2677 BTRFS_BLOCK_RSV_DELREFS);
2679 atomic_set(&fs_info->async_delalloc_pages, 0);
2680 atomic_set(&fs_info->defrag_running, 0);
2681 atomic_set(&fs_info->reada_works_cnt, 0);
2682 atomic_set(&fs_info->nr_delayed_iputs, 0);
2683 atomic64_set(&fs_info->tree_mod_seq, 0);
2685 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2686 fs_info->metadata_ratio = 0;
2687 fs_info->defrag_inodes = RB_ROOT;
2688 atomic64_set(&fs_info->free_chunk_space, 0);
2689 fs_info->tree_mod_log = RB_ROOT;
2690 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2691 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2692 /* readahead state */
2693 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2694 spin_lock_init(&fs_info->reada_lock);
2695 btrfs_init_ref_verify(fs_info);
2697 fs_info->thread_pool_size = min_t(unsigned long,
2698 num_online_cpus() + 2, 8);
2700 INIT_LIST_HEAD(&fs_info->ordered_roots);
2701 spin_lock_init(&fs_info->ordered_root_lock);
2703 fs_info->btree_inode = new_inode(sb);
2704 if (!fs_info->btree_inode) {
2706 goto fail_bio_counter;
2708 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2710 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2712 if (!fs_info->delayed_root) {
2716 btrfs_init_delayed_root(fs_info->delayed_root);
2718 btrfs_init_scrub(fs_info);
2719 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2720 fs_info->check_integrity_print_mask = 0;
2722 btrfs_init_balance(fs_info);
2723 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2725 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2726 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2728 btrfs_init_btree_inode(fs_info);
2730 spin_lock_init(&fs_info->block_group_cache_lock);
2731 fs_info->block_group_cache_tree = RB_ROOT;
2732 fs_info->first_logical_byte = (u64)-1;
2734 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2735 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2736 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2737 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2738 fs_info->pinned_extents = &fs_info->freed_extents[0];
2739 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2741 mutex_init(&fs_info->ordered_operations_mutex);
2742 mutex_init(&fs_info->tree_log_mutex);
2743 mutex_init(&fs_info->chunk_mutex);
2744 mutex_init(&fs_info->transaction_kthread_mutex);
2745 mutex_init(&fs_info->cleaner_mutex);
2746 mutex_init(&fs_info->ro_block_group_mutex);
2747 init_rwsem(&fs_info->commit_root_sem);
2748 init_rwsem(&fs_info->cleanup_work_sem);
2749 init_rwsem(&fs_info->subvol_sem);
2750 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2752 btrfs_init_dev_replace_locks(fs_info);
2753 btrfs_init_qgroup(fs_info);
2755 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2756 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2758 init_waitqueue_head(&fs_info->transaction_throttle);
2759 init_waitqueue_head(&fs_info->transaction_wait);
2760 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2761 init_waitqueue_head(&fs_info->async_submit_wait);
2762 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2764 /* Usable values until the real ones are cached from the superblock */
2765 fs_info->nodesize = 4096;
2766 fs_info->sectorsize = 4096;
2767 fs_info->stripesize = 4096;
2769 spin_lock_init(&fs_info->swapfile_pins_lock);
2770 fs_info->swapfile_pins = RB_ROOT;
2772 fs_info->send_in_progress = 0;
2774 ret = btrfs_alloc_stripe_hash_table(fs_info);
2780 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2782 invalidate_bdev(fs_devices->latest_bdev);
2785 * Read super block and check the signature bytes only
2787 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2794 * Verify the type first, if that or the the checksum value are
2795 * corrupted, we'll find out
2797 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2798 if (!btrfs_supported_super_csum(csum_type)) {
2799 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2806 ret = btrfs_init_csum_hash(fs_info, csum_type);
2813 * We want to check superblock checksum, the type is stored inside.
2814 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2816 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2817 btrfs_err(fs_info, "superblock checksum mismatch");
2824 * super_copy is zeroed at allocation time and we never touch the
2825 * following bytes up to INFO_SIZE, the checksum is calculated from
2826 * the whole block of INFO_SIZE
2828 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2831 disk_super = fs_info->super_copy;
2833 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2836 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2837 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2838 fs_info->super_copy->metadata_uuid,
2842 features = btrfs_super_flags(disk_super);
2843 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2844 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2845 btrfs_set_super_flags(disk_super, features);
2847 "found metadata UUID change in progress flag, clearing");
2850 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2851 sizeof(*fs_info->super_for_commit));
2853 ret = btrfs_validate_mount_super(fs_info);
2855 btrfs_err(fs_info, "superblock contains fatal errors");
2860 if (!btrfs_super_root(disk_super))
2863 /* check FS state, whether FS is broken. */
2864 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2865 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2868 * run through our array of backup supers and setup
2869 * our ring pointer to the oldest one
2871 generation = btrfs_super_generation(disk_super);
2872 find_oldest_super_backup(fs_info, generation);
2875 * In the long term, we'll store the compression type in the super
2876 * block, and it'll be used for per file compression control.
2878 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2880 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2886 features = btrfs_super_incompat_flags(disk_super) &
2887 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2890 "cannot mount because of unsupported optional features (%llx)",
2896 features = btrfs_super_incompat_flags(disk_super);
2897 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2898 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2899 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2900 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2901 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2903 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2904 btrfs_info(fs_info, "has skinny extents");
2907 * flag our filesystem as having big metadata blocks if
2908 * they are bigger than the page size
2910 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2911 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2913 "flagging fs with big metadata feature");
2914 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2917 nodesize = btrfs_super_nodesize(disk_super);
2918 sectorsize = btrfs_super_sectorsize(disk_super);
2919 stripesize = sectorsize;
2920 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2921 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2923 /* Cache block sizes */
2924 fs_info->nodesize = nodesize;
2925 fs_info->sectorsize = sectorsize;
2926 fs_info->stripesize = stripesize;
2929 * mixed block groups end up with duplicate but slightly offset
2930 * extent buffers for the same range. It leads to corruptions
2932 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2933 (sectorsize != nodesize)) {
2935 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2936 nodesize, sectorsize);
2941 * Needn't use the lock because there is no other task which will
2944 btrfs_set_super_incompat_flags(disk_super, features);
2946 features = btrfs_super_compat_ro_flags(disk_super) &
2947 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2948 if (!sb_rdonly(sb) && features) {
2950 "cannot mount read-write because of unsupported optional features (%llx)",
2956 ret = btrfs_init_workqueues(fs_info, fs_devices);
2959 goto fail_sb_buffer;
2962 sb->s_bdi->congested_fn = btrfs_congested_fn;
2963 sb->s_bdi->congested_data = fs_info;
2964 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2965 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2966 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2967 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2969 sb->s_blocksize = sectorsize;
2970 sb->s_blocksize_bits = blksize_bits(sectorsize);
2971 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2973 mutex_lock(&fs_info->chunk_mutex);
2974 ret = btrfs_read_sys_array(fs_info);
2975 mutex_unlock(&fs_info->chunk_mutex);
2977 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2978 goto fail_sb_buffer;
2981 generation = btrfs_super_chunk_root_generation(disk_super);
2982 level = btrfs_super_chunk_root_level(disk_super);
2984 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2986 chunk_root->node = read_tree_block(fs_info,
2987 btrfs_super_chunk_root(disk_super),
2988 generation, level, NULL);
2989 if (IS_ERR(chunk_root->node) ||
2990 !extent_buffer_uptodate(chunk_root->node)) {
2991 btrfs_err(fs_info, "failed to read chunk root");
2992 if (!IS_ERR(chunk_root->node))
2993 free_extent_buffer(chunk_root->node);
2994 chunk_root->node = NULL;
2995 goto fail_tree_roots;
2997 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2998 chunk_root->commit_root = btrfs_root_node(chunk_root);
3000 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3001 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3003 ret = btrfs_read_chunk_tree(fs_info);
3005 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3006 goto fail_tree_roots;
3010 * Keep the devid that is marked to be the target device for the
3011 * device replace procedure
3013 btrfs_free_extra_devids(fs_devices, 0);
3015 if (!fs_devices->latest_bdev) {
3016 btrfs_err(fs_info, "failed to read devices");
3017 goto fail_tree_roots;
3021 generation = btrfs_super_generation(disk_super);
3022 level = btrfs_super_root_level(disk_super);
3024 tree_root->node = read_tree_block(fs_info,
3025 btrfs_super_root(disk_super),
3026 generation, level, NULL);
3027 if (IS_ERR(tree_root->node) ||
3028 !extent_buffer_uptodate(tree_root->node)) {
3029 btrfs_warn(fs_info, "failed to read tree root");
3030 if (!IS_ERR(tree_root->node))
3031 free_extent_buffer(tree_root->node);
3032 tree_root->node = NULL;
3033 goto recovery_tree_root;
3036 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3037 tree_root->commit_root = btrfs_root_node(tree_root);
3038 btrfs_set_root_refs(&tree_root->root_item, 1);
3040 mutex_lock(&tree_root->objectid_mutex);
3041 ret = btrfs_find_highest_objectid(tree_root,
3042 &tree_root->highest_objectid);
3044 mutex_unlock(&tree_root->objectid_mutex);
3045 goto recovery_tree_root;
3048 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3050 mutex_unlock(&tree_root->objectid_mutex);
3052 ret = btrfs_read_roots(fs_info);
3054 goto recovery_tree_root;
3056 fs_info->generation = generation;
3057 fs_info->last_trans_committed = generation;
3059 ret = btrfs_verify_dev_extents(fs_info);
3062 "failed to verify dev extents against chunks: %d",
3064 goto fail_block_groups;
3066 ret = btrfs_recover_balance(fs_info);
3068 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3069 goto fail_block_groups;
3072 ret = btrfs_init_dev_stats(fs_info);
3074 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3075 goto fail_block_groups;
3078 ret = btrfs_init_dev_replace(fs_info);
3080 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3081 goto fail_block_groups;
3084 btrfs_free_extra_devids(fs_devices, 1);
3086 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3088 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3090 goto fail_block_groups;
3093 ret = btrfs_sysfs_add_device(fs_devices);
3095 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3097 goto fail_fsdev_sysfs;
3100 ret = btrfs_sysfs_add_mounted(fs_info);
3102 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3103 goto fail_fsdev_sysfs;
3106 ret = btrfs_init_space_info(fs_info);
3108 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3112 ret = btrfs_read_block_groups(fs_info);
3114 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3118 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3120 "writable mount is not allowed due to too many missing devices");
3124 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3126 if (IS_ERR(fs_info->cleaner_kthread))
3129 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3131 "btrfs-transaction");
3132 if (IS_ERR(fs_info->transaction_kthread))
3135 if (!btrfs_test_opt(fs_info, NOSSD) &&
3136 !fs_info->fs_devices->rotating) {
3137 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3141 * Mount does not set all options immediately, we can do it now and do
3142 * not have to wait for transaction commit
3144 btrfs_apply_pending_changes(fs_info);
3146 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3147 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3148 ret = btrfsic_mount(fs_info, fs_devices,
3149 btrfs_test_opt(fs_info,
3150 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3152 fs_info->check_integrity_print_mask);
3155 "failed to initialize integrity check module: %d",
3159 ret = btrfs_read_qgroup_config(fs_info);
3161 goto fail_trans_kthread;
3163 if (btrfs_build_ref_tree(fs_info))
3164 btrfs_err(fs_info, "couldn't build ref tree");
3166 /* do not make disk changes in broken FS or nologreplay is given */
3167 if (btrfs_super_log_root(disk_super) != 0 &&
3168 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3169 ret = btrfs_replay_log(fs_info, fs_devices);
3176 ret = btrfs_find_orphan_roots(fs_info);
3180 if (!sb_rdonly(sb)) {
3181 ret = btrfs_cleanup_fs_roots(fs_info);
3185 mutex_lock(&fs_info->cleaner_mutex);
3186 ret = btrfs_recover_relocation(tree_root);
3187 mutex_unlock(&fs_info->cleaner_mutex);
3189 btrfs_warn(fs_info, "failed to recover relocation: %d",
3196 location.objectid = BTRFS_FS_TREE_OBJECTID;
3197 location.type = BTRFS_ROOT_ITEM_KEY;
3198 location.offset = 0;
3200 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3201 if (IS_ERR(fs_info->fs_root)) {
3202 err = PTR_ERR(fs_info->fs_root);
3203 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3210 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3211 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3212 clear_free_space_tree = 1;
3213 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3214 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3215 btrfs_warn(fs_info, "free space tree is invalid");
3216 clear_free_space_tree = 1;
3219 if (clear_free_space_tree) {
3220 btrfs_info(fs_info, "clearing free space tree");
3221 ret = btrfs_clear_free_space_tree(fs_info);
3224 "failed to clear free space tree: %d", ret);
3225 close_ctree(fs_info);
3230 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3231 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3232 btrfs_info(fs_info, "creating free space tree");
3233 ret = btrfs_create_free_space_tree(fs_info);
3236 "failed to create free space tree: %d", ret);
3237 close_ctree(fs_info);
3242 down_read(&fs_info->cleanup_work_sem);
3243 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3244 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3245 up_read(&fs_info->cleanup_work_sem);
3246 close_ctree(fs_info);
3249 up_read(&fs_info->cleanup_work_sem);
3251 ret = btrfs_resume_balance_async(fs_info);
3253 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3254 close_ctree(fs_info);
3258 ret = btrfs_resume_dev_replace_async(fs_info);
3260 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3261 close_ctree(fs_info);
3265 btrfs_qgroup_rescan_resume(fs_info);
3267 if (!fs_info->uuid_root) {
3268 btrfs_info(fs_info, "creating UUID tree");
3269 ret = btrfs_create_uuid_tree(fs_info);
3272 "failed to create the UUID tree: %d", ret);
3273 close_ctree(fs_info);
3276 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3277 fs_info->generation !=
3278 btrfs_super_uuid_tree_generation(disk_super)) {
3279 btrfs_info(fs_info, "checking UUID tree");
3280 ret = btrfs_check_uuid_tree(fs_info);
3283 "failed to check the UUID tree: %d", ret);
3284 close_ctree(fs_info);
3288 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3290 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3293 * backuproot only affect mount behavior, and if open_ctree succeeded,
3294 * no need to keep the flag
3296 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3301 btrfs_free_qgroup_config(fs_info);
3303 kthread_stop(fs_info->transaction_kthread);
3304 btrfs_cleanup_transaction(fs_info);
3305 btrfs_free_fs_roots(fs_info);
3307 kthread_stop(fs_info->cleaner_kthread);
3310 * make sure we're done with the btree inode before we stop our
3313 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3316 btrfs_sysfs_remove_mounted(fs_info);
3319 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3322 btrfs_put_block_group_cache(fs_info);
3325 free_root_pointers(fs_info, true);
3326 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3329 btrfs_stop_all_workers(fs_info);
3330 btrfs_free_block_groups(fs_info);
3332 btrfs_free_csum_hash(fs_info);
3335 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3337 iput(fs_info->btree_inode);
3339 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3340 fail_delalloc_bytes:
3341 percpu_counter_destroy(&fs_info->delalloc_bytes);
3342 fail_dirty_metadata_bytes:
3343 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3345 percpu_counter_destroy(&fs_info->dio_bytes);
3347 cleanup_srcu_struct(&fs_info->subvol_srcu);
3349 btrfs_free_stripe_hash_table(fs_info);
3350 btrfs_close_devices(fs_info->fs_devices);
3354 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3355 goto fail_tree_roots;
3357 free_root_pointers(fs_info, false);
3359 /* don't use the log in recovery mode, it won't be valid */
3360 btrfs_set_super_log_root(disk_super, 0);
3362 /* we can't trust the free space cache either */
3363 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3365 ret = next_root_backup(fs_info, fs_info->super_copy,
3366 &num_backups_tried, &backup_index);
3368 goto fail_block_groups;
3369 goto retry_root_backup;
3371 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3373 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3376 set_buffer_uptodate(bh);
3378 struct btrfs_device *device = (struct btrfs_device *)
3381 btrfs_warn_rl_in_rcu(device->fs_info,
3382 "lost page write due to IO error on %s",
3383 rcu_str_deref(device->name));
3384 /* note, we don't set_buffer_write_io_error because we have
3385 * our own ways of dealing with the IO errors
3387 clear_buffer_uptodate(bh);
3388 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3394 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3395 struct buffer_head **bh_ret)
3397 struct buffer_head *bh;
3398 struct btrfs_super_block *super;
3401 bytenr = btrfs_sb_offset(copy_num);
3402 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3405 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3407 * If we fail to read from the underlying devices, as of now
3408 * the best option we have is to mark it EIO.
3413 super = (struct btrfs_super_block *)bh->b_data;
3414 if (btrfs_super_bytenr(super) != bytenr ||
3415 btrfs_super_magic(super) != BTRFS_MAGIC) {
3425 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3427 struct buffer_head *bh;
3428 struct buffer_head *latest = NULL;
3429 struct btrfs_super_block *super;
3434 /* we would like to check all the supers, but that would make
3435 * a btrfs mount succeed after a mkfs from a different FS.
3436 * So, we need to add a special mount option to scan for
3437 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3439 for (i = 0; i < 1; i++) {
3440 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3444 super = (struct btrfs_super_block *)bh->b_data;
3446 if (!latest || btrfs_super_generation(super) > transid) {
3449 transid = btrfs_super_generation(super);
3456 return ERR_PTR(ret);
3462 * Write superblock @sb to the @device. Do not wait for completion, all the
3463 * buffer heads we write are pinned.
3465 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3466 * the expected device size at commit time. Note that max_mirrors must be
3467 * same for write and wait phases.
3469 * Return number of errors when buffer head is not found or submission fails.
3471 static int write_dev_supers(struct btrfs_device *device,
3472 struct btrfs_super_block *sb, int max_mirrors)
3474 struct btrfs_fs_info *fs_info = device->fs_info;
3475 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3476 struct buffer_head *bh;
3483 if (max_mirrors == 0)
3484 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3486 shash->tfm = fs_info->csum_shash;
3488 for (i = 0; i < max_mirrors; i++) {
3489 bytenr = btrfs_sb_offset(i);
3490 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3491 device->commit_total_bytes)
3494 btrfs_set_super_bytenr(sb, bytenr);
3496 crypto_shash_init(shash);
3497 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3498 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3499 crypto_shash_final(shash, sb->csum);
3501 /* One reference for us, and we leave it for the caller */
3502 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3503 BTRFS_SUPER_INFO_SIZE);
3505 btrfs_err(device->fs_info,
3506 "couldn't get super buffer head for bytenr %llu",
3512 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3514 /* one reference for submit_bh */
3517 set_buffer_uptodate(bh);
3519 bh->b_end_io = btrfs_end_buffer_write_sync;
3520 bh->b_private = device;
3523 * we fua the first super. The others we allow
3526 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3527 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3528 op_flags |= REQ_FUA;
3529 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3533 return errors < i ? 0 : -1;
3537 * Wait for write completion of superblocks done by write_dev_supers,
3538 * @max_mirrors same for write and wait phases.
3540 * Return number of errors when buffer head is not found or not marked up to
3543 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3545 struct buffer_head *bh;
3548 bool primary_failed = false;
3551 if (max_mirrors == 0)
3552 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3554 for (i = 0; i < max_mirrors; i++) {
3555 bytenr = btrfs_sb_offset(i);
3556 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3557 device->commit_total_bytes)
3560 bh = __find_get_block(device->bdev,
3561 bytenr / BTRFS_BDEV_BLOCKSIZE,
3562 BTRFS_SUPER_INFO_SIZE);
3566 primary_failed = true;
3570 if (!buffer_uptodate(bh)) {
3573 primary_failed = true;
3576 /* drop our reference */
3579 /* drop the reference from the writing run */
3583 /* log error, force error return */
3584 if (primary_failed) {
3585 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3590 return errors < i ? 0 : -1;
3594 * endio for the write_dev_flush, this will wake anyone waiting
3595 * for the barrier when it is done
3597 static void btrfs_end_empty_barrier(struct bio *bio)
3599 complete(bio->bi_private);
3603 * Submit a flush request to the device if it supports it. Error handling is
3604 * done in the waiting counterpart.
3606 static void write_dev_flush(struct btrfs_device *device)
3608 struct request_queue *q = bdev_get_queue(device->bdev);
3609 struct bio *bio = device->flush_bio;
3611 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3615 bio->bi_end_io = btrfs_end_empty_barrier;
3616 bio_set_dev(bio, device->bdev);
3617 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3618 init_completion(&device->flush_wait);
3619 bio->bi_private = &device->flush_wait;
3621 btrfsic_submit_bio(bio);
3622 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3626 * If the flush bio has been submitted by write_dev_flush, wait for it.
3628 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3630 struct bio *bio = device->flush_bio;
3632 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3635 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3636 wait_for_completion_io(&device->flush_wait);
3638 return bio->bi_status;
3641 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3643 if (!btrfs_check_rw_degradable(fs_info, NULL))
3649 * send an empty flush down to each device in parallel,
3650 * then wait for them
3652 static int barrier_all_devices(struct btrfs_fs_info *info)
3654 struct list_head *head;
3655 struct btrfs_device *dev;
3656 int errors_wait = 0;
3659 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3660 /* send down all the barriers */
3661 head = &info->fs_devices->devices;
3662 list_for_each_entry(dev, head, dev_list) {
3663 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3667 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3668 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3671 write_dev_flush(dev);
3672 dev->last_flush_error = BLK_STS_OK;
3675 /* wait for all the barriers */
3676 list_for_each_entry(dev, head, dev_list) {
3677 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3683 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3684 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3687 ret = wait_dev_flush(dev);
3689 dev->last_flush_error = ret;
3690 btrfs_dev_stat_inc_and_print(dev,
3691 BTRFS_DEV_STAT_FLUSH_ERRS);
3698 * At some point we need the status of all disks
3699 * to arrive at the volume status. So error checking
3700 * is being pushed to a separate loop.
3702 return check_barrier_error(info);
3707 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3710 int min_tolerated = INT_MAX;
3712 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3713 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3714 min_tolerated = min_t(int, min_tolerated,
3715 btrfs_raid_array[BTRFS_RAID_SINGLE].
3716 tolerated_failures);
3718 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3719 if (raid_type == BTRFS_RAID_SINGLE)
3721 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3723 min_tolerated = min_t(int, min_tolerated,
3724 btrfs_raid_array[raid_type].
3725 tolerated_failures);
3728 if (min_tolerated == INT_MAX) {
3729 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3733 return min_tolerated;
3736 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3738 struct list_head *head;
3739 struct btrfs_device *dev;
3740 struct btrfs_super_block *sb;
3741 struct btrfs_dev_item *dev_item;
3745 int total_errors = 0;
3748 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3751 * max_mirrors == 0 indicates we're from commit_transaction,
3752 * not from fsync where the tree roots in fs_info have not
3753 * been consistent on disk.
3755 if (max_mirrors == 0)
3756 backup_super_roots(fs_info);
3758 sb = fs_info->super_for_commit;
3759 dev_item = &sb->dev_item;
3761 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3762 head = &fs_info->fs_devices->devices;
3763 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3766 ret = barrier_all_devices(fs_info);
3769 &fs_info->fs_devices->device_list_mutex);
3770 btrfs_handle_fs_error(fs_info, ret,
3771 "errors while submitting device barriers.");
3776 list_for_each_entry(dev, head, dev_list) {
3781 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3782 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3785 btrfs_set_stack_device_generation(dev_item, 0);
3786 btrfs_set_stack_device_type(dev_item, dev->type);
3787 btrfs_set_stack_device_id(dev_item, dev->devid);
3788 btrfs_set_stack_device_total_bytes(dev_item,
3789 dev->commit_total_bytes);
3790 btrfs_set_stack_device_bytes_used(dev_item,
3791 dev->commit_bytes_used);
3792 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3793 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3794 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3795 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3796 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3799 flags = btrfs_super_flags(sb);
3800 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3802 ret = btrfs_validate_write_super(fs_info, sb);
3804 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3805 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3806 "unexpected superblock corruption detected");
3810 ret = write_dev_supers(dev, sb, max_mirrors);
3814 if (total_errors > max_errors) {
3815 btrfs_err(fs_info, "%d errors while writing supers",
3817 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3819 /* FUA is masked off if unsupported and can't be the reason */
3820 btrfs_handle_fs_error(fs_info, -EIO,
3821 "%d errors while writing supers",
3827 list_for_each_entry(dev, head, dev_list) {
3830 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3831 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3834 ret = wait_dev_supers(dev, max_mirrors);
3838 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3839 if (total_errors > max_errors) {
3840 btrfs_handle_fs_error(fs_info, -EIO,
3841 "%d errors while writing supers",
3848 /* Drop a fs root from the radix tree and free it. */
3849 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3850 struct btrfs_root *root)
3852 spin_lock(&fs_info->fs_roots_radix_lock);
3853 radix_tree_delete(&fs_info->fs_roots_radix,
3854 (unsigned long)root->root_key.objectid);
3855 spin_unlock(&fs_info->fs_roots_radix_lock);
3857 if (btrfs_root_refs(&root->root_item) == 0)
3858 synchronize_srcu(&fs_info->subvol_srcu);
3860 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3861 btrfs_free_log(NULL, root);
3862 if (root->reloc_root) {
3863 free_extent_buffer(root->reloc_root->node);
3864 free_extent_buffer(root->reloc_root->commit_root);
3865 btrfs_put_fs_root(root->reloc_root);
3866 root->reloc_root = NULL;
3870 if (root->free_ino_pinned)
3871 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3872 if (root->free_ino_ctl)
3873 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3874 btrfs_free_fs_root(root);
3877 void btrfs_free_fs_root(struct btrfs_root *root)
3879 iput(root->ino_cache_inode);
3880 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3882 free_anon_bdev(root->anon_dev);
3883 if (root->subv_writers)
3884 btrfs_free_subvolume_writers(root->subv_writers);
3885 free_extent_buffer(root->node);
3886 free_extent_buffer(root->commit_root);
3887 kfree(root->free_ino_ctl);
3888 kfree(root->free_ino_pinned);
3889 btrfs_put_fs_root(root);
3892 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3894 u64 root_objectid = 0;
3895 struct btrfs_root *gang[8];
3898 unsigned int ret = 0;
3902 index = srcu_read_lock(&fs_info->subvol_srcu);
3903 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3904 (void **)gang, root_objectid,
3907 srcu_read_unlock(&fs_info->subvol_srcu, index);
3910 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3912 for (i = 0; i < ret; i++) {
3913 /* Avoid to grab roots in dead_roots */
3914 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3918 /* grab all the search result for later use */
3919 gang[i] = btrfs_grab_fs_root(gang[i]);
3921 srcu_read_unlock(&fs_info->subvol_srcu, index);
3923 for (i = 0; i < ret; i++) {
3926 root_objectid = gang[i]->root_key.objectid;
3927 err = btrfs_orphan_cleanup(gang[i]);
3930 btrfs_put_fs_root(gang[i]);
3935 /* release the uncleaned roots due to error */
3936 for (; i < ret; i++) {
3938 btrfs_put_fs_root(gang[i]);
3943 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3945 struct btrfs_root *root = fs_info->tree_root;
3946 struct btrfs_trans_handle *trans;
3948 mutex_lock(&fs_info->cleaner_mutex);
3949 btrfs_run_delayed_iputs(fs_info);
3950 mutex_unlock(&fs_info->cleaner_mutex);
3951 wake_up_process(fs_info->cleaner_kthread);
3953 /* wait until ongoing cleanup work done */
3954 down_write(&fs_info->cleanup_work_sem);
3955 up_write(&fs_info->cleanup_work_sem);
3957 trans = btrfs_join_transaction(root);
3959 return PTR_ERR(trans);
3960 return btrfs_commit_transaction(trans);
3963 void __cold close_ctree(struct btrfs_fs_info *fs_info)
3967 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3969 * We don't want the cleaner to start new transactions, add more delayed
3970 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3971 * because that frees the task_struct, and the transaction kthread might
3972 * still try to wake up the cleaner.
3974 kthread_park(fs_info->cleaner_kthread);
3976 /* wait for the qgroup rescan worker to stop */
3977 btrfs_qgroup_wait_for_completion(fs_info, false);
3979 /* wait for the uuid_scan task to finish */
3980 down(&fs_info->uuid_tree_rescan_sem);
3981 /* avoid complains from lockdep et al., set sem back to initial state */
3982 up(&fs_info->uuid_tree_rescan_sem);
3984 /* pause restriper - we want to resume on mount */
3985 btrfs_pause_balance(fs_info);
3987 btrfs_dev_replace_suspend_for_unmount(fs_info);
3989 btrfs_scrub_cancel(fs_info);
3991 /* wait for any defraggers to finish */
3992 wait_event(fs_info->transaction_wait,
3993 (atomic_read(&fs_info->defrag_running) == 0));
3995 /* clear out the rbtree of defraggable inodes */
3996 btrfs_cleanup_defrag_inodes(fs_info);
3998 cancel_work_sync(&fs_info->async_reclaim_work);
4000 if (!sb_rdonly(fs_info->sb)) {
4002 * The cleaner kthread is stopped, so do one final pass over
4003 * unused block groups.
4005 btrfs_delete_unused_bgs(fs_info);
4007 ret = btrfs_commit_super(fs_info);
4009 btrfs_err(fs_info, "commit super ret %d", ret);
4012 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4013 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4014 btrfs_error_commit_super(fs_info);
4016 kthread_stop(fs_info->transaction_kthread);
4017 kthread_stop(fs_info->cleaner_kthread);
4019 ASSERT(list_empty(&fs_info->delayed_iputs));
4020 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4022 btrfs_free_qgroup_config(fs_info);
4023 ASSERT(list_empty(&fs_info->delalloc_roots));
4025 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4026 btrfs_info(fs_info, "at unmount delalloc count %lld",
4027 percpu_counter_sum(&fs_info->delalloc_bytes));
4030 if (percpu_counter_sum(&fs_info->dio_bytes))
4031 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4032 percpu_counter_sum(&fs_info->dio_bytes));
4034 btrfs_sysfs_remove_mounted(fs_info);
4035 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4037 btrfs_free_fs_roots(fs_info);
4039 btrfs_put_block_group_cache(fs_info);
4042 * we must make sure there is not any read request to
4043 * submit after we stopping all workers.
4045 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4046 btrfs_stop_all_workers(fs_info);
4048 btrfs_free_block_groups(fs_info);
4050 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4051 free_root_pointers(fs_info, true);
4053 iput(fs_info->btree_inode);
4055 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4056 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4057 btrfsic_unmount(fs_info->fs_devices);
4060 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4061 btrfs_close_devices(fs_info->fs_devices);
4063 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4064 percpu_counter_destroy(&fs_info->delalloc_bytes);
4065 percpu_counter_destroy(&fs_info->dio_bytes);
4066 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4067 cleanup_srcu_struct(&fs_info->subvol_srcu);
4069 btrfs_free_csum_hash(fs_info);
4070 btrfs_free_stripe_hash_table(fs_info);
4071 btrfs_free_ref_cache(fs_info);
4074 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4078 struct inode *btree_inode = buf->pages[0]->mapping->host;
4080 ret = extent_buffer_uptodate(buf);
4084 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4085 parent_transid, atomic);
4091 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4093 struct btrfs_fs_info *fs_info;
4094 struct btrfs_root *root;
4095 u64 transid = btrfs_header_generation(buf);
4098 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4100 * This is a fast path so only do this check if we have sanity tests
4101 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4102 * outside of the sanity tests.
4104 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4107 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4108 fs_info = root->fs_info;
4109 btrfs_assert_tree_locked(buf);
4110 if (transid != fs_info->generation)
4111 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4112 buf->start, transid, fs_info->generation);
4113 was_dirty = set_extent_buffer_dirty(buf);
4115 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4117 fs_info->dirty_metadata_batch);
4118 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4120 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4121 * but item data not updated.
4122 * So here we should only check item pointers, not item data.
4124 if (btrfs_header_level(buf) == 0 &&
4125 btrfs_check_leaf_relaxed(buf)) {
4126 btrfs_print_leaf(buf);
4132 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4136 * looks as though older kernels can get into trouble with
4137 * this code, they end up stuck in balance_dirty_pages forever
4141 if (current->flags & PF_MEMALLOC)
4145 btrfs_balance_delayed_items(fs_info);
4147 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4148 BTRFS_DIRTY_METADATA_THRESH,
4149 fs_info->dirty_metadata_batch);
4151 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4155 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4157 __btrfs_btree_balance_dirty(fs_info, 1);
4160 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4162 __btrfs_btree_balance_dirty(fs_info, 0);
4165 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4166 struct btrfs_key *first_key)
4168 return btree_read_extent_buffer_pages(buf, parent_transid,
4172 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4174 /* cleanup FS via transaction */
4175 btrfs_cleanup_transaction(fs_info);
4177 mutex_lock(&fs_info->cleaner_mutex);
4178 btrfs_run_delayed_iputs(fs_info);
4179 mutex_unlock(&fs_info->cleaner_mutex);
4181 down_write(&fs_info->cleanup_work_sem);
4182 up_write(&fs_info->cleanup_work_sem);
4185 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4187 struct btrfs_ordered_extent *ordered;
4189 spin_lock(&root->ordered_extent_lock);
4191 * This will just short circuit the ordered completion stuff which will
4192 * make sure the ordered extent gets properly cleaned up.
4194 list_for_each_entry(ordered, &root->ordered_extents,
4196 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4197 spin_unlock(&root->ordered_extent_lock);
4200 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4202 struct btrfs_root *root;
4203 struct list_head splice;
4205 INIT_LIST_HEAD(&splice);
4207 spin_lock(&fs_info->ordered_root_lock);
4208 list_splice_init(&fs_info->ordered_roots, &splice);
4209 while (!list_empty(&splice)) {
4210 root = list_first_entry(&splice, struct btrfs_root,
4212 list_move_tail(&root->ordered_root,
4213 &fs_info->ordered_roots);
4215 spin_unlock(&fs_info->ordered_root_lock);
4216 btrfs_destroy_ordered_extents(root);
4219 spin_lock(&fs_info->ordered_root_lock);
4221 spin_unlock(&fs_info->ordered_root_lock);
4224 * We need this here because if we've been flipped read-only we won't
4225 * get sync() from the umount, so we need to make sure any ordered
4226 * extents that haven't had their dirty pages IO start writeout yet
4227 * actually get run and error out properly.
4229 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4232 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4233 struct btrfs_fs_info *fs_info)
4235 struct rb_node *node;
4236 struct btrfs_delayed_ref_root *delayed_refs;
4237 struct btrfs_delayed_ref_node *ref;
4240 delayed_refs = &trans->delayed_refs;
4242 spin_lock(&delayed_refs->lock);
4243 if (atomic_read(&delayed_refs->num_entries) == 0) {
4244 spin_unlock(&delayed_refs->lock);
4245 btrfs_info(fs_info, "delayed_refs has NO entry");
4249 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4250 struct btrfs_delayed_ref_head *head;
4252 bool pin_bytes = false;
4254 head = rb_entry(node, struct btrfs_delayed_ref_head,
4256 if (btrfs_delayed_ref_lock(delayed_refs, head))
4259 spin_lock(&head->lock);
4260 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4261 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4264 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4265 RB_CLEAR_NODE(&ref->ref_node);
4266 if (!list_empty(&ref->add_list))
4267 list_del(&ref->add_list);
4268 atomic_dec(&delayed_refs->num_entries);
4269 btrfs_put_delayed_ref(ref);
4271 if (head->must_insert_reserved)
4273 btrfs_free_delayed_extent_op(head->extent_op);
4274 btrfs_delete_ref_head(delayed_refs, head);
4275 spin_unlock(&head->lock);
4276 spin_unlock(&delayed_refs->lock);
4277 mutex_unlock(&head->mutex);
4280 btrfs_pin_extent(fs_info, head->bytenr,
4281 head->num_bytes, 1);
4282 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4283 btrfs_put_delayed_ref_head(head);
4285 spin_lock(&delayed_refs->lock);
4288 spin_unlock(&delayed_refs->lock);
4293 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4295 struct btrfs_inode *btrfs_inode;
4296 struct list_head splice;
4298 INIT_LIST_HEAD(&splice);
4300 spin_lock(&root->delalloc_lock);
4301 list_splice_init(&root->delalloc_inodes, &splice);
4303 while (!list_empty(&splice)) {
4304 struct inode *inode = NULL;
4305 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4307 __btrfs_del_delalloc_inode(root, btrfs_inode);
4308 spin_unlock(&root->delalloc_lock);
4311 * Make sure we get a live inode and that it'll not disappear
4314 inode = igrab(&btrfs_inode->vfs_inode);
4316 invalidate_inode_pages2(inode->i_mapping);
4319 spin_lock(&root->delalloc_lock);
4321 spin_unlock(&root->delalloc_lock);
4324 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4326 struct btrfs_root *root;
4327 struct list_head splice;
4329 INIT_LIST_HEAD(&splice);
4331 spin_lock(&fs_info->delalloc_root_lock);
4332 list_splice_init(&fs_info->delalloc_roots, &splice);
4333 while (!list_empty(&splice)) {
4334 root = list_first_entry(&splice, struct btrfs_root,
4336 root = btrfs_grab_fs_root(root);
4338 spin_unlock(&fs_info->delalloc_root_lock);
4340 btrfs_destroy_delalloc_inodes(root);
4341 btrfs_put_fs_root(root);
4343 spin_lock(&fs_info->delalloc_root_lock);
4345 spin_unlock(&fs_info->delalloc_root_lock);
4348 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4349 struct extent_io_tree *dirty_pages,
4353 struct extent_buffer *eb;
4358 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4363 clear_extent_bits(dirty_pages, start, end, mark);
4364 while (start <= end) {
4365 eb = find_extent_buffer(fs_info, start);
4366 start += fs_info->nodesize;
4369 wait_on_extent_buffer_writeback(eb);
4371 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4373 clear_extent_buffer_dirty(eb);
4374 free_extent_buffer_stale(eb);
4381 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4382 struct extent_io_tree *pinned_extents)
4384 struct extent_io_tree *unpin;
4390 unpin = pinned_extents;
4393 struct extent_state *cached_state = NULL;
4396 * The btrfs_finish_extent_commit() may get the same range as
4397 * ours between find_first_extent_bit and clear_extent_dirty.
4398 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4399 * the same extent range.
4401 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4402 ret = find_first_extent_bit(unpin, 0, &start, &end,
4403 EXTENT_DIRTY, &cached_state);
4405 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4409 clear_extent_dirty(unpin, start, end, &cached_state);
4410 free_extent_state(cached_state);
4411 btrfs_error_unpin_extent_range(fs_info, start, end);
4412 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4417 if (unpin == &fs_info->freed_extents[0])
4418 unpin = &fs_info->freed_extents[1];
4420 unpin = &fs_info->freed_extents[0];
4428 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4430 struct inode *inode;
4432 inode = cache->io_ctl.inode;
4434 invalidate_inode_pages2(inode->i_mapping);
4435 BTRFS_I(inode)->generation = 0;
4436 cache->io_ctl.inode = NULL;
4439 btrfs_put_block_group(cache);
4442 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4443 struct btrfs_fs_info *fs_info)
4445 struct btrfs_block_group_cache *cache;
4447 spin_lock(&cur_trans->dirty_bgs_lock);
4448 while (!list_empty(&cur_trans->dirty_bgs)) {
4449 cache = list_first_entry(&cur_trans->dirty_bgs,
4450 struct btrfs_block_group_cache,
4453 if (!list_empty(&cache->io_list)) {
4454 spin_unlock(&cur_trans->dirty_bgs_lock);
4455 list_del_init(&cache->io_list);
4456 btrfs_cleanup_bg_io(cache);
4457 spin_lock(&cur_trans->dirty_bgs_lock);
4460 list_del_init(&cache->dirty_list);
4461 spin_lock(&cache->lock);
4462 cache->disk_cache_state = BTRFS_DC_ERROR;
4463 spin_unlock(&cache->lock);
4465 spin_unlock(&cur_trans->dirty_bgs_lock);
4466 btrfs_put_block_group(cache);
4467 btrfs_delayed_refs_rsv_release(fs_info, 1);
4468 spin_lock(&cur_trans->dirty_bgs_lock);
4470 spin_unlock(&cur_trans->dirty_bgs_lock);
4473 * Refer to the definition of io_bgs member for details why it's safe
4474 * to use it without any locking
4476 while (!list_empty(&cur_trans->io_bgs)) {
4477 cache = list_first_entry(&cur_trans->io_bgs,
4478 struct btrfs_block_group_cache,
4481 list_del_init(&cache->io_list);
4482 spin_lock(&cache->lock);
4483 cache->disk_cache_state = BTRFS_DC_ERROR;
4484 spin_unlock(&cache->lock);
4485 btrfs_cleanup_bg_io(cache);
4489 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4490 struct btrfs_fs_info *fs_info)
4492 struct btrfs_device *dev, *tmp;
4494 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4495 ASSERT(list_empty(&cur_trans->dirty_bgs));
4496 ASSERT(list_empty(&cur_trans->io_bgs));
4498 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4500 list_del_init(&dev->post_commit_list);
4503 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4505 cur_trans->state = TRANS_STATE_COMMIT_START;
4506 wake_up(&fs_info->transaction_blocked_wait);
4508 cur_trans->state = TRANS_STATE_UNBLOCKED;
4509 wake_up(&fs_info->transaction_wait);
4511 btrfs_destroy_delayed_inodes(fs_info);
4512 btrfs_assert_delayed_root_empty(fs_info);
4514 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4516 btrfs_destroy_pinned_extent(fs_info,
4517 fs_info->pinned_extents);
4519 cur_trans->state =TRANS_STATE_COMPLETED;
4520 wake_up(&cur_trans->commit_wait);
4523 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4525 struct btrfs_transaction *t;
4527 mutex_lock(&fs_info->transaction_kthread_mutex);
4529 spin_lock(&fs_info->trans_lock);
4530 while (!list_empty(&fs_info->trans_list)) {
4531 t = list_first_entry(&fs_info->trans_list,
4532 struct btrfs_transaction, list);
4533 if (t->state >= TRANS_STATE_COMMIT_START) {
4534 refcount_inc(&t->use_count);
4535 spin_unlock(&fs_info->trans_lock);
4536 btrfs_wait_for_commit(fs_info, t->transid);
4537 btrfs_put_transaction(t);
4538 spin_lock(&fs_info->trans_lock);
4541 if (t == fs_info->running_transaction) {
4542 t->state = TRANS_STATE_COMMIT_DOING;
4543 spin_unlock(&fs_info->trans_lock);
4545 * We wait for 0 num_writers since we don't hold a trans
4546 * handle open currently for this transaction.
4548 wait_event(t->writer_wait,
4549 atomic_read(&t->num_writers) == 0);
4551 spin_unlock(&fs_info->trans_lock);
4553 btrfs_cleanup_one_transaction(t, fs_info);
4555 spin_lock(&fs_info->trans_lock);
4556 if (t == fs_info->running_transaction)
4557 fs_info->running_transaction = NULL;
4558 list_del_init(&t->list);
4559 spin_unlock(&fs_info->trans_lock);
4561 btrfs_put_transaction(t);
4562 trace_btrfs_transaction_commit(fs_info->tree_root);
4563 spin_lock(&fs_info->trans_lock);
4565 spin_unlock(&fs_info->trans_lock);
4566 btrfs_destroy_all_ordered_extents(fs_info);
4567 btrfs_destroy_delayed_inodes(fs_info);
4568 btrfs_assert_delayed_root_empty(fs_info);
4569 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4570 btrfs_destroy_all_delalloc_inodes(fs_info);
4571 mutex_unlock(&fs_info->transaction_kthread_mutex);
4576 static const struct extent_io_ops btree_extent_io_ops = {
4577 /* mandatory callbacks */
4578 .submit_bio_hook = btree_submit_bio_hook,
4579 .readpage_end_io_hook = btree_readpage_end_io_hook,