1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include <crypto/hash.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
43 #include "block-group.h"
45 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
46 BTRFS_HEADER_FLAG_RELOC |\
47 BTRFS_SUPER_FLAG_ERROR |\
48 BTRFS_SUPER_FLAG_SEEDING |\
49 BTRFS_SUPER_FLAG_METADUMP |\
50 BTRFS_SUPER_FLAG_METADUMP_V2)
52 static const struct extent_io_ops btree_extent_io_ops;
53 static void end_workqueue_fn(struct btrfs_work *work);
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
68 * is complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
71 struct btrfs_end_io_wq {
75 struct btrfs_fs_info *info;
77 enum btrfs_wq_endio_type metadata;
78 struct btrfs_work work;
81 static struct kmem_cache *btrfs_end_io_wq_cache;
83 int __init btrfs_end_io_wq_init(void)
85 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 sizeof(struct btrfs_end_io_wq),
90 if (!btrfs_end_io_wq_cache)
95 void __cold btrfs_end_io_wq_exit(void)
97 kmem_cache_destroy(btrfs_end_io_wq_cache);
101 * async submit bios are used to offload expensive checksumming
102 * onto the worker threads. They checksum file and metadata bios
103 * just before they are sent down the IO stack.
105 struct async_submit_bio {
108 extent_submit_bio_start_t *submit_bio_start;
111 * bio_offset is optional, can be used if the pages in the bio
112 * can't tell us where in the file the bio should go
115 struct btrfs_work work;
120 * Lockdep class keys for extent_buffer->lock's in this root. For a given
121 * eb, the lockdep key is determined by the btrfs_root it belongs to and
122 * the level the eb occupies in the tree.
124 * Different roots are used for different purposes and may nest inside each
125 * other and they require separate keysets. As lockdep keys should be
126 * static, assign keysets according to the purpose of the root as indicated
127 * by btrfs_root->root_key.objectid. This ensures that all special purpose
128 * roots have separate keysets.
130 * Lock-nesting across peer nodes is always done with the immediate parent
131 * node locked thus preventing deadlock. As lockdep doesn't know this, use
132 * subclass to avoid triggering lockdep warning in such cases.
134 * The key is set by the readpage_end_io_hook after the buffer has passed
135 * csum validation but before the pages are unlocked. It is also set by
136 * btrfs_init_new_buffer on freshly allocated blocks.
138 * We also add a check to make sure the highest level of the tree is the
139 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
140 * needs update as well.
142 #ifdef CONFIG_DEBUG_LOCK_ALLOC
143 # if BTRFS_MAX_LEVEL != 8
147 static struct btrfs_lockdep_keyset {
148 u64 id; /* root objectid */
149 const char *name_stem; /* lock name stem */
150 char names[BTRFS_MAX_LEVEL + 1][20];
151 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
152 } btrfs_lockdep_keysets[] = {
153 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
154 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
155 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
156 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
157 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
158 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
159 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
160 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
161 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
162 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
163 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
164 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
165 { .id = 0, .name_stem = "tree" },
168 void __init btrfs_init_lockdep(void)
172 /* initialize lockdep class names */
173 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
174 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
176 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
177 snprintf(ks->names[j], sizeof(ks->names[j]),
178 "btrfs-%s-%02d", ks->name_stem, j);
182 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
185 struct btrfs_lockdep_keyset *ks;
187 BUG_ON(level >= ARRAY_SIZE(ks->keys));
189 /* find the matching keyset, id 0 is the default entry */
190 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
191 if (ks->id == objectid)
194 lockdep_set_class_and_name(&eb->lock,
195 &ks->keys[level], ks->names[level]);
201 * extents on the btree inode are pretty simple, there's one extent
202 * that covers the entire device
204 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
205 struct page *page, size_t pg_offset, u64 start, u64 len,
208 struct btrfs_fs_info *fs_info = inode->root->fs_info;
209 struct extent_map_tree *em_tree = &inode->extent_tree;
210 struct extent_map *em;
213 read_lock(&em_tree->lock);
214 em = lookup_extent_mapping(em_tree, start, len);
216 em->bdev = fs_info->fs_devices->latest_bdev;
217 read_unlock(&em_tree->lock);
220 read_unlock(&em_tree->lock);
222 em = alloc_extent_map();
224 em = ERR_PTR(-ENOMEM);
229 em->block_len = (u64)-1;
231 em->bdev = fs_info->fs_devices->latest_bdev;
233 write_lock(&em_tree->lock);
234 ret = add_extent_mapping(em_tree, em, 0);
235 if (ret == -EEXIST) {
237 em = lookup_extent_mapping(em_tree, start, len);
244 write_unlock(&em_tree->lock);
251 * Compute the csum of a btree block and store the result to provided buffer.
253 * Returns error if the extent buffer cannot be mapped.
255 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
257 struct btrfs_fs_info *fs_info = buf->fs_info;
258 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
260 unsigned long cur_len;
261 unsigned long offset = BTRFS_CSUM_SIZE;
263 unsigned long map_start;
264 unsigned long map_len;
267 shash->tfm = fs_info->csum_shash;
268 crypto_shash_init(shash);
270 len = buf->len - offset;
274 * Note: we don't need to check for the err == 1 case here, as
275 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
276 * and 'min_len = 32' and the currently implemented mapping
277 * algorithm we cannot cross a page boundary.
279 err = map_private_extent_buffer(buf, offset, 32,
280 &kaddr, &map_start, &map_len);
283 cur_len = min(len, map_len - (offset - map_start));
284 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
288 memset(result, 0, BTRFS_CSUM_SIZE);
290 crypto_shash_final(shash, result);
296 * we can't consider a given block up to date unless the transid of the
297 * block matches the transid in the parent node's pointer. This is how we
298 * detect blocks that either didn't get written at all or got written
299 * in the wrong place.
301 static int verify_parent_transid(struct extent_io_tree *io_tree,
302 struct extent_buffer *eb, u64 parent_transid,
305 struct extent_state *cached_state = NULL;
307 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
309 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
316 btrfs_tree_read_lock(eb);
317 btrfs_set_lock_blocking_read(eb);
320 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
322 if (extent_buffer_uptodate(eb) &&
323 btrfs_header_generation(eb) == parent_transid) {
327 btrfs_err_rl(eb->fs_info,
328 "parent transid verify failed on %llu wanted %llu found %llu",
330 parent_transid, btrfs_header_generation(eb));
334 * Things reading via commit roots that don't have normal protection,
335 * like send, can have a really old block in cache that may point at a
336 * block that has been freed and re-allocated. So don't clear uptodate
337 * if we find an eb that is under IO (dirty/writeback) because we could
338 * end up reading in the stale data and then writing it back out and
339 * making everybody very sad.
341 if (!extent_buffer_under_io(eb))
342 clear_extent_buffer_uptodate(eb);
344 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
347 btrfs_tree_read_unlock_blocking(eb);
351 static bool btrfs_supported_super_csum(u16 csum_type)
354 case BTRFS_CSUM_TYPE_CRC32:
362 * Return 0 if the superblock checksum type matches the checksum value of that
363 * algorithm. Pass the raw disk superblock data.
365 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
368 struct btrfs_super_block *disk_sb =
369 (struct btrfs_super_block *)raw_disk_sb;
370 char result[BTRFS_CSUM_SIZE];
371 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
373 shash->tfm = fs_info->csum_shash;
374 crypto_shash_init(shash);
377 * The super_block structure does not span the whole
378 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
379 * filled with zeros and is included in the checksum.
381 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
382 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
383 crypto_shash_final(shash, result);
385 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
391 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
392 struct btrfs_key *first_key, u64 parent_transid)
394 struct btrfs_fs_info *fs_info = eb->fs_info;
396 struct btrfs_key found_key;
399 found_level = btrfs_header_level(eb);
400 if (found_level != level) {
401 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
402 KERN_ERR "BTRFS: tree level check failed\n");
404 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
405 eb->start, level, found_level);
413 * For live tree block (new tree blocks in current transaction),
414 * we need proper lock context to avoid race, which is impossible here.
415 * So we only checks tree blocks which is read from disk, whose
416 * generation <= fs_info->last_trans_committed.
418 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
421 /* We have @first_key, so this @eb must have at least one item */
422 if (btrfs_header_nritems(eb) == 0) {
424 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
426 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
431 btrfs_node_key_to_cpu(eb, &found_key, 0);
433 btrfs_item_key_to_cpu(eb, &found_key, 0);
434 ret = btrfs_comp_cpu_keys(first_key, &found_key);
437 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
438 KERN_ERR "BTRFS: tree first key check failed\n");
440 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
441 eb->start, parent_transid, first_key->objectid,
442 first_key->type, first_key->offset,
443 found_key.objectid, found_key.type,
450 * helper to read a given tree block, doing retries as required when
451 * the checksums don't match and we have alternate mirrors to try.
453 * @parent_transid: expected transid, skip check if 0
454 * @level: expected level, mandatory check
455 * @first_key: expected key of first slot, skip check if NULL
457 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
458 u64 parent_transid, int level,
459 struct btrfs_key *first_key)
461 struct btrfs_fs_info *fs_info = eb->fs_info;
462 struct extent_io_tree *io_tree;
467 int failed_mirror = 0;
469 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
471 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
472 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
474 if (verify_parent_transid(io_tree, eb,
477 else if (btrfs_verify_level_key(eb, level,
478 first_key, parent_transid))
484 num_copies = btrfs_num_copies(fs_info,
489 if (!failed_mirror) {
491 failed_mirror = eb->read_mirror;
495 if (mirror_num == failed_mirror)
498 if (mirror_num > num_copies)
502 if (failed && !ret && failed_mirror)
503 btrfs_repair_eb_io_failure(eb, failed_mirror);
509 * checksum a dirty tree block before IO. This has extra checks to make sure
510 * we only fill in the checksum field in the first page of a multi-page block
513 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
515 u64 start = page_offset(page);
517 u8 result[BTRFS_CSUM_SIZE];
518 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
519 struct extent_buffer *eb;
522 eb = (struct extent_buffer *)page->private;
523 if (page != eb->pages[0])
526 found_start = btrfs_header_bytenr(eb);
528 * Please do not consolidate these warnings into a single if.
529 * It is useful to know what went wrong.
531 if (WARN_ON(found_start != start))
533 if (WARN_ON(!PageUptodate(page)))
536 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
537 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
539 if (csum_tree_block(eb, result))
542 if (btrfs_header_level(eb))
543 ret = btrfs_check_node(eb);
545 ret = btrfs_check_leaf_full(eb);
549 "block=%llu write time tree block corruption detected",
553 write_extent_buffer(eb, result, 0, csum_size);
558 static int check_tree_block_fsid(struct extent_buffer *eb)
560 struct btrfs_fs_info *fs_info = eb->fs_info;
561 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
562 u8 fsid[BTRFS_FSID_SIZE];
565 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
570 * Checking the incompat flag is only valid for the current
571 * fs. For seed devices it's forbidden to have their uuid
572 * changed so reading ->fsid in this case is fine
574 if (fs_devices == fs_info->fs_devices &&
575 btrfs_fs_incompat(fs_info, METADATA_UUID))
576 metadata_uuid = fs_devices->metadata_uuid;
578 metadata_uuid = fs_devices->fsid;
580 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
584 fs_devices = fs_devices->seed;
589 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
590 u64 phy_offset, struct page *page,
591 u64 start, u64 end, int mirror)
595 struct extent_buffer *eb;
596 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
597 struct btrfs_fs_info *fs_info = root->fs_info;
598 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
600 u8 result[BTRFS_CSUM_SIZE];
606 eb = (struct extent_buffer *)page->private;
608 /* the pending IO might have been the only thing that kept this buffer
609 * in memory. Make sure we have a ref for all this other checks
611 extent_buffer_get(eb);
613 reads_done = atomic_dec_and_test(&eb->io_pages);
617 eb->read_mirror = mirror;
618 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
623 found_start = btrfs_header_bytenr(eb);
624 if (found_start != eb->start) {
625 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
626 eb->start, found_start);
630 if (check_tree_block_fsid(eb)) {
631 btrfs_err_rl(fs_info, "bad fsid on block %llu",
636 found_level = btrfs_header_level(eb);
637 if (found_level >= BTRFS_MAX_LEVEL) {
638 btrfs_err(fs_info, "bad tree block level %d on %llu",
639 (int)btrfs_header_level(eb), eb->start);
644 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
647 ret = csum_tree_block(eb, result);
651 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
655 memcpy(&found, result, csum_size);
657 read_extent_buffer(eb, &val, 0, csum_size);
658 btrfs_warn_rl(fs_info,
659 "%s checksum verify failed on %llu wanted %x found %x level %d",
660 fs_info->sb->s_id, eb->start,
661 val, found, btrfs_header_level(eb));
667 * If this is a leaf block and it is corrupt, set the corrupt bit so
668 * that we don't try and read the other copies of this block, just
671 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
672 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
676 if (found_level > 0 && btrfs_check_node(eb))
680 set_extent_buffer_uptodate(eb);
683 "block=%llu read time tree block corruption detected",
687 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
688 btree_readahead_hook(eb, ret);
692 * our io error hook is going to dec the io pages
693 * again, we have to make sure it has something
696 atomic_inc(&eb->io_pages);
697 clear_extent_buffer_uptodate(eb);
699 free_extent_buffer(eb);
704 static void end_workqueue_bio(struct bio *bio)
706 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
707 struct btrfs_fs_info *fs_info;
708 struct btrfs_workqueue *wq;
709 btrfs_work_func_t func;
711 fs_info = end_io_wq->info;
712 end_io_wq->status = bio->bi_status;
714 if (bio_op(bio) == REQ_OP_WRITE) {
715 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
716 wq = fs_info->endio_meta_write_workers;
717 func = btrfs_endio_meta_write_helper;
718 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
719 wq = fs_info->endio_freespace_worker;
720 func = btrfs_freespace_write_helper;
721 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
722 wq = fs_info->endio_raid56_workers;
723 func = btrfs_endio_raid56_helper;
725 wq = fs_info->endio_write_workers;
726 func = btrfs_endio_write_helper;
729 if (unlikely(end_io_wq->metadata ==
730 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
731 wq = fs_info->endio_repair_workers;
732 func = btrfs_endio_repair_helper;
733 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
734 wq = fs_info->endio_raid56_workers;
735 func = btrfs_endio_raid56_helper;
736 } else if (end_io_wq->metadata) {
737 wq = fs_info->endio_meta_workers;
738 func = btrfs_endio_meta_helper;
740 wq = fs_info->endio_workers;
741 func = btrfs_endio_helper;
745 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
746 btrfs_queue_work(wq, &end_io_wq->work);
749 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
750 enum btrfs_wq_endio_type metadata)
752 struct btrfs_end_io_wq *end_io_wq;
754 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
756 return BLK_STS_RESOURCE;
758 end_io_wq->private = bio->bi_private;
759 end_io_wq->end_io = bio->bi_end_io;
760 end_io_wq->info = info;
761 end_io_wq->status = 0;
762 end_io_wq->bio = bio;
763 end_io_wq->metadata = metadata;
765 bio->bi_private = end_io_wq;
766 bio->bi_end_io = end_workqueue_bio;
770 static void run_one_async_start(struct btrfs_work *work)
772 struct async_submit_bio *async;
775 async = container_of(work, struct async_submit_bio, work);
776 ret = async->submit_bio_start(async->private_data, async->bio,
783 * In order to insert checksums into the metadata in large chunks, we wait
784 * until bio submission time. All the pages in the bio are checksummed and
785 * sums are attached onto the ordered extent record.
787 * At IO completion time the csums attached on the ordered extent record are
788 * inserted into the tree.
790 static void run_one_async_done(struct btrfs_work *work)
792 struct async_submit_bio *async;
796 async = container_of(work, struct async_submit_bio, work);
797 inode = async->private_data;
799 /* If an error occurred we just want to clean up the bio and move on */
801 async->bio->bi_status = async->status;
802 bio_endio(async->bio);
806 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
807 async->mirror_num, 1);
809 async->bio->bi_status = ret;
810 bio_endio(async->bio);
814 static void run_one_async_free(struct btrfs_work *work)
816 struct async_submit_bio *async;
818 async = container_of(work, struct async_submit_bio, work);
822 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
823 int mirror_num, unsigned long bio_flags,
824 u64 bio_offset, void *private_data,
825 extent_submit_bio_start_t *submit_bio_start)
827 struct async_submit_bio *async;
829 async = kmalloc(sizeof(*async), GFP_NOFS);
831 return BLK_STS_RESOURCE;
833 async->private_data = private_data;
835 async->mirror_num = mirror_num;
836 async->submit_bio_start = submit_bio_start;
838 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
839 run_one_async_done, run_one_async_free);
841 async->bio_offset = bio_offset;
845 if (op_is_sync(bio->bi_opf))
846 btrfs_set_work_high_priority(&async->work);
848 btrfs_queue_work(fs_info->workers, &async->work);
852 static blk_status_t btree_csum_one_bio(struct bio *bio)
854 struct bio_vec *bvec;
855 struct btrfs_root *root;
857 struct bvec_iter_all iter_all;
859 ASSERT(!bio_flagged(bio, BIO_CLONED));
860 bio_for_each_segment_all(bvec, bio, iter_all) {
861 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
862 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
867 return errno_to_blk_status(ret);
870 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
874 * when we're called for a write, we're already in the async
875 * submission context. Just jump into btrfs_map_bio
877 return btree_csum_one_bio(bio);
880 static int check_async_write(struct btrfs_fs_info *fs_info,
881 struct btrfs_inode *bi)
883 if (atomic_read(&bi->sync_writers))
885 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
890 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
892 unsigned long bio_flags)
894 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
895 int async = check_async_write(fs_info, BTRFS_I(inode));
898 if (bio_op(bio) != REQ_OP_WRITE) {
900 * called for a read, do the setup so that checksum validation
901 * can happen in the async kernel threads
903 ret = btrfs_bio_wq_end_io(fs_info, bio,
904 BTRFS_WQ_ENDIO_METADATA);
907 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
909 ret = btree_csum_one_bio(bio);
912 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
915 * kthread helpers are used to submit writes so that
916 * checksumming can happen in parallel across all CPUs
918 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
919 0, inode, btree_submit_bio_start);
927 bio->bi_status = ret;
932 #ifdef CONFIG_MIGRATION
933 static int btree_migratepage(struct address_space *mapping,
934 struct page *newpage, struct page *page,
935 enum migrate_mode mode)
938 * we can't safely write a btree page from here,
939 * we haven't done the locking hook
944 * Buffers may be managed in a filesystem specific way.
945 * We must have no buffers or drop them.
947 if (page_has_private(page) &&
948 !try_to_release_page(page, GFP_KERNEL))
950 return migrate_page(mapping, newpage, page, mode);
955 static int btree_writepages(struct address_space *mapping,
956 struct writeback_control *wbc)
958 struct btrfs_fs_info *fs_info;
961 if (wbc->sync_mode == WB_SYNC_NONE) {
963 if (wbc->for_kupdate)
966 fs_info = BTRFS_I(mapping->host)->root->fs_info;
967 /* this is a bit racy, but that's ok */
968 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
969 BTRFS_DIRTY_METADATA_THRESH,
970 fs_info->dirty_metadata_batch);
974 return btree_write_cache_pages(mapping, wbc);
977 static int btree_readpage(struct file *file, struct page *page)
979 struct extent_io_tree *tree;
980 tree = &BTRFS_I(page->mapping->host)->io_tree;
981 return extent_read_full_page(tree, page, btree_get_extent, 0);
984 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
986 if (PageWriteback(page) || PageDirty(page))
989 return try_release_extent_buffer(page);
992 static void btree_invalidatepage(struct page *page, unsigned int offset,
995 struct extent_io_tree *tree;
996 tree = &BTRFS_I(page->mapping->host)->io_tree;
997 extent_invalidatepage(tree, page, offset);
998 btree_releasepage(page, GFP_NOFS);
999 if (PagePrivate(page)) {
1000 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1001 "page private not zero on page %llu",
1002 (unsigned long long)page_offset(page));
1003 ClearPagePrivate(page);
1004 set_page_private(page, 0);
1009 static int btree_set_page_dirty(struct page *page)
1012 struct extent_buffer *eb;
1014 BUG_ON(!PagePrivate(page));
1015 eb = (struct extent_buffer *)page->private;
1017 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1018 BUG_ON(!atomic_read(&eb->refs));
1019 btrfs_assert_tree_locked(eb);
1021 return __set_page_dirty_nobuffers(page);
1024 static const struct address_space_operations btree_aops = {
1025 .readpage = btree_readpage,
1026 .writepages = btree_writepages,
1027 .releasepage = btree_releasepage,
1028 .invalidatepage = btree_invalidatepage,
1029 #ifdef CONFIG_MIGRATION
1030 .migratepage = btree_migratepage,
1032 .set_page_dirty = btree_set_page_dirty,
1035 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1037 struct extent_buffer *buf = NULL;
1040 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1044 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1046 free_extent_buffer_stale(buf);
1048 free_extent_buffer(buf);
1051 struct extent_buffer *btrfs_find_create_tree_block(
1052 struct btrfs_fs_info *fs_info,
1055 if (btrfs_is_testing(fs_info))
1056 return alloc_test_extent_buffer(fs_info, bytenr);
1057 return alloc_extent_buffer(fs_info, bytenr);
1061 * Read tree block at logical address @bytenr and do variant basic but critical
1064 * @parent_transid: expected transid of this tree block, skip check if 0
1065 * @level: expected level, mandatory check
1066 * @first_key: expected key in slot 0, skip check if NULL
1068 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1069 u64 parent_transid, int level,
1070 struct btrfs_key *first_key)
1072 struct extent_buffer *buf = NULL;
1075 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1079 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1082 free_extent_buffer_stale(buf);
1083 return ERR_PTR(ret);
1089 void btrfs_clean_tree_block(struct extent_buffer *buf)
1091 struct btrfs_fs_info *fs_info = buf->fs_info;
1092 if (btrfs_header_generation(buf) ==
1093 fs_info->running_transaction->transid) {
1094 btrfs_assert_tree_locked(buf);
1096 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1097 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1099 fs_info->dirty_metadata_batch);
1100 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1101 btrfs_set_lock_blocking_write(buf);
1102 clear_extent_buffer_dirty(buf);
1107 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1109 struct btrfs_subvolume_writers *writers;
1112 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1114 return ERR_PTR(-ENOMEM);
1116 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1119 return ERR_PTR(ret);
1122 init_waitqueue_head(&writers->wait);
1127 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1129 percpu_counter_destroy(&writers->counter);
1133 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1136 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1138 root->commit_root = NULL;
1140 root->orphan_cleanup_state = 0;
1142 root->last_trans = 0;
1143 root->highest_objectid = 0;
1144 root->nr_delalloc_inodes = 0;
1145 root->nr_ordered_extents = 0;
1146 root->inode_tree = RB_ROOT;
1147 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1148 root->block_rsv = NULL;
1150 INIT_LIST_HEAD(&root->dirty_list);
1151 INIT_LIST_HEAD(&root->root_list);
1152 INIT_LIST_HEAD(&root->delalloc_inodes);
1153 INIT_LIST_HEAD(&root->delalloc_root);
1154 INIT_LIST_HEAD(&root->ordered_extents);
1155 INIT_LIST_HEAD(&root->ordered_root);
1156 INIT_LIST_HEAD(&root->reloc_dirty_list);
1157 INIT_LIST_HEAD(&root->logged_list[0]);
1158 INIT_LIST_HEAD(&root->logged_list[1]);
1159 spin_lock_init(&root->inode_lock);
1160 spin_lock_init(&root->delalloc_lock);
1161 spin_lock_init(&root->ordered_extent_lock);
1162 spin_lock_init(&root->accounting_lock);
1163 spin_lock_init(&root->log_extents_lock[0]);
1164 spin_lock_init(&root->log_extents_lock[1]);
1165 spin_lock_init(&root->qgroup_meta_rsv_lock);
1166 mutex_init(&root->objectid_mutex);
1167 mutex_init(&root->log_mutex);
1168 mutex_init(&root->ordered_extent_mutex);
1169 mutex_init(&root->delalloc_mutex);
1170 init_waitqueue_head(&root->log_writer_wait);
1171 init_waitqueue_head(&root->log_commit_wait[0]);
1172 init_waitqueue_head(&root->log_commit_wait[1]);
1173 INIT_LIST_HEAD(&root->log_ctxs[0]);
1174 INIT_LIST_HEAD(&root->log_ctxs[1]);
1175 atomic_set(&root->log_commit[0], 0);
1176 atomic_set(&root->log_commit[1], 0);
1177 atomic_set(&root->log_writers, 0);
1178 atomic_set(&root->log_batch, 0);
1179 refcount_set(&root->refs, 1);
1180 atomic_set(&root->will_be_snapshotted, 0);
1181 atomic_set(&root->snapshot_force_cow, 0);
1182 atomic_set(&root->nr_swapfiles, 0);
1183 root->log_transid = 0;
1184 root->log_transid_committed = -1;
1185 root->last_log_commit = 0;
1187 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1188 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1190 memset(&root->root_key, 0, sizeof(root->root_key));
1191 memset(&root->root_item, 0, sizeof(root->root_item));
1192 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1194 root->defrag_trans_start = fs_info->generation;
1196 root->defrag_trans_start = 0;
1197 root->root_key.objectid = objectid;
1200 spin_lock_init(&root->root_item_lock);
1201 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1204 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1207 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1209 root->fs_info = fs_info;
1213 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1214 /* Should only be used by the testing infrastructure */
1215 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1217 struct btrfs_root *root;
1220 return ERR_PTR(-EINVAL);
1222 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1224 return ERR_PTR(-ENOMEM);
1226 /* We don't use the stripesize in selftest, set it as sectorsize */
1227 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1228 root->alloc_bytenr = 0;
1234 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1237 struct btrfs_fs_info *fs_info = trans->fs_info;
1238 struct extent_buffer *leaf;
1239 struct btrfs_root *tree_root = fs_info->tree_root;
1240 struct btrfs_root *root;
1241 struct btrfs_key key;
1242 unsigned int nofs_flag;
1244 uuid_le uuid = NULL_UUID_LE;
1247 * We're holding a transaction handle, so use a NOFS memory allocation
1248 * context to avoid deadlock if reclaim happens.
1250 nofs_flag = memalloc_nofs_save();
1251 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1252 memalloc_nofs_restore(nofs_flag);
1254 return ERR_PTR(-ENOMEM);
1256 __setup_root(root, fs_info, objectid);
1257 root->root_key.objectid = objectid;
1258 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1259 root->root_key.offset = 0;
1261 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1263 ret = PTR_ERR(leaf);
1269 btrfs_mark_buffer_dirty(leaf);
1271 root->commit_root = btrfs_root_node(root);
1272 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1274 root->root_item.flags = 0;
1275 root->root_item.byte_limit = 0;
1276 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1277 btrfs_set_root_generation(&root->root_item, trans->transid);
1278 btrfs_set_root_level(&root->root_item, 0);
1279 btrfs_set_root_refs(&root->root_item, 1);
1280 btrfs_set_root_used(&root->root_item, leaf->len);
1281 btrfs_set_root_last_snapshot(&root->root_item, 0);
1282 btrfs_set_root_dirid(&root->root_item, 0);
1283 if (is_fstree(objectid))
1285 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1286 root->root_item.drop_level = 0;
1288 key.objectid = objectid;
1289 key.type = BTRFS_ROOT_ITEM_KEY;
1291 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1295 btrfs_tree_unlock(leaf);
1301 btrfs_tree_unlock(leaf);
1302 free_extent_buffer(root->commit_root);
1303 free_extent_buffer(leaf);
1307 return ERR_PTR(ret);
1310 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1311 struct btrfs_fs_info *fs_info)
1313 struct btrfs_root *root;
1314 struct extent_buffer *leaf;
1316 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1318 return ERR_PTR(-ENOMEM);
1320 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1322 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1323 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1324 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1327 * DON'T set REF_COWS for log trees
1329 * log trees do not get reference counted because they go away
1330 * before a real commit is actually done. They do store pointers
1331 * to file data extents, and those reference counts still get
1332 * updated (along with back refs to the log tree).
1335 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1339 return ERR_CAST(leaf);
1344 btrfs_mark_buffer_dirty(root->node);
1345 btrfs_tree_unlock(root->node);
1349 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1350 struct btrfs_fs_info *fs_info)
1352 struct btrfs_root *log_root;
1354 log_root = alloc_log_tree(trans, fs_info);
1355 if (IS_ERR(log_root))
1356 return PTR_ERR(log_root);
1357 WARN_ON(fs_info->log_root_tree);
1358 fs_info->log_root_tree = log_root;
1362 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root)
1365 struct btrfs_fs_info *fs_info = root->fs_info;
1366 struct btrfs_root *log_root;
1367 struct btrfs_inode_item *inode_item;
1369 log_root = alloc_log_tree(trans, fs_info);
1370 if (IS_ERR(log_root))
1371 return PTR_ERR(log_root);
1373 log_root->last_trans = trans->transid;
1374 log_root->root_key.offset = root->root_key.objectid;
1376 inode_item = &log_root->root_item.inode;
1377 btrfs_set_stack_inode_generation(inode_item, 1);
1378 btrfs_set_stack_inode_size(inode_item, 3);
1379 btrfs_set_stack_inode_nlink(inode_item, 1);
1380 btrfs_set_stack_inode_nbytes(inode_item,
1382 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1384 btrfs_set_root_node(&log_root->root_item, log_root->node);
1386 WARN_ON(root->log_root);
1387 root->log_root = log_root;
1388 root->log_transid = 0;
1389 root->log_transid_committed = -1;
1390 root->last_log_commit = 0;
1394 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1395 struct btrfs_key *key)
1397 struct btrfs_root *root;
1398 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1399 struct btrfs_path *path;
1404 path = btrfs_alloc_path();
1406 return ERR_PTR(-ENOMEM);
1408 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1414 __setup_root(root, fs_info, key->objectid);
1416 ret = btrfs_find_root(tree_root, key, path,
1417 &root->root_item, &root->root_key);
1424 generation = btrfs_root_generation(&root->root_item);
1425 level = btrfs_root_level(&root->root_item);
1426 root->node = read_tree_block(fs_info,
1427 btrfs_root_bytenr(&root->root_item),
1428 generation, level, NULL);
1429 if (IS_ERR(root->node)) {
1430 ret = PTR_ERR(root->node);
1432 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1434 free_extent_buffer(root->node);
1437 root->commit_root = btrfs_root_node(root);
1439 btrfs_free_path(path);
1445 root = ERR_PTR(ret);
1449 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1450 struct btrfs_key *location)
1452 struct btrfs_root *root;
1454 root = btrfs_read_tree_root(tree_root, location);
1458 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1459 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1460 btrfs_check_and_init_root_item(&root->root_item);
1466 int btrfs_init_fs_root(struct btrfs_root *root)
1469 struct btrfs_subvolume_writers *writers;
1471 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1472 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1474 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1479 writers = btrfs_alloc_subvolume_writers();
1480 if (IS_ERR(writers)) {
1481 ret = PTR_ERR(writers);
1484 root->subv_writers = writers;
1486 btrfs_init_free_ino_ctl(root);
1487 spin_lock_init(&root->ino_cache_lock);
1488 init_waitqueue_head(&root->ino_cache_wait);
1490 ret = get_anon_bdev(&root->anon_dev);
1494 mutex_lock(&root->objectid_mutex);
1495 ret = btrfs_find_highest_objectid(root,
1496 &root->highest_objectid);
1498 mutex_unlock(&root->objectid_mutex);
1502 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1504 mutex_unlock(&root->objectid_mutex);
1508 /* The caller is responsible to call btrfs_free_fs_root */
1512 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1515 struct btrfs_root *root;
1517 spin_lock(&fs_info->fs_roots_radix_lock);
1518 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1519 (unsigned long)root_id);
1520 spin_unlock(&fs_info->fs_roots_radix_lock);
1524 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1525 struct btrfs_root *root)
1529 ret = radix_tree_preload(GFP_NOFS);
1533 spin_lock(&fs_info->fs_roots_radix_lock);
1534 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1535 (unsigned long)root->root_key.objectid,
1538 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1539 spin_unlock(&fs_info->fs_roots_radix_lock);
1540 radix_tree_preload_end();
1545 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1546 struct btrfs_key *location,
1549 struct btrfs_root *root;
1550 struct btrfs_path *path;
1551 struct btrfs_key key;
1554 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1555 return fs_info->tree_root;
1556 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1557 return fs_info->extent_root;
1558 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1559 return fs_info->chunk_root;
1560 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1561 return fs_info->dev_root;
1562 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1563 return fs_info->csum_root;
1564 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1565 return fs_info->quota_root ? fs_info->quota_root :
1567 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1568 return fs_info->uuid_root ? fs_info->uuid_root :
1570 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1571 return fs_info->free_space_root ? fs_info->free_space_root :
1574 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1576 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1577 return ERR_PTR(-ENOENT);
1581 root = btrfs_read_fs_root(fs_info->tree_root, location);
1585 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1590 ret = btrfs_init_fs_root(root);
1594 path = btrfs_alloc_path();
1599 key.objectid = BTRFS_ORPHAN_OBJECTID;
1600 key.type = BTRFS_ORPHAN_ITEM_KEY;
1601 key.offset = location->objectid;
1603 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1604 btrfs_free_path(path);
1608 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1610 ret = btrfs_insert_fs_root(fs_info, root);
1612 if (ret == -EEXIST) {
1613 btrfs_free_fs_root(root);
1620 btrfs_free_fs_root(root);
1621 return ERR_PTR(ret);
1624 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1626 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1628 struct btrfs_device *device;
1629 struct backing_dev_info *bdi;
1632 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1635 bdi = device->bdev->bd_bdi;
1636 if (bdi_congested(bdi, bdi_bits)) {
1646 * called by the kthread helper functions to finally call the bio end_io
1647 * functions. This is where read checksum verification actually happens
1649 static void end_workqueue_fn(struct btrfs_work *work)
1652 struct btrfs_end_io_wq *end_io_wq;
1654 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1655 bio = end_io_wq->bio;
1657 bio->bi_status = end_io_wq->status;
1658 bio->bi_private = end_io_wq->private;
1659 bio->bi_end_io = end_io_wq->end_io;
1660 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1664 static int cleaner_kthread(void *arg)
1666 struct btrfs_root *root = arg;
1667 struct btrfs_fs_info *fs_info = root->fs_info;
1673 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1675 /* Make the cleaner go to sleep early. */
1676 if (btrfs_need_cleaner_sleep(fs_info))
1680 * Do not do anything if we might cause open_ctree() to block
1681 * before we have finished mounting the filesystem.
1683 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1686 if (!mutex_trylock(&fs_info->cleaner_mutex))
1690 * Avoid the problem that we change the status of the fs
1691 * during the above check and trylock.
1693 if (btrfs_need_cleaner_sleep(fs_info)) {
1694 mutex_unlock(&fs_info->cleaner_mutex);
1698 btrfs_run_delayed_iputs(fs_info);
1700 again = btrfs_clean_one_deleted_snapshot(root);
1701 mutex_unlock(&fs_info->cleaner_mutex);
1704 * The defragger has dealt with the R/O remount and umount,
1705 * needn't do anything special here.
1707 btrfs_run_defrag_inodes(fs_info);
1710 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1711 * with relocation (btrfs_relocate_chunk) and relocation
1712 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1713 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1714 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1715 * unused block groups.
1717 btrfs_delete_unused_bgs(fs_info);
1719 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1720 if (kthread_should_park())
1722 if (kthread_should_stop())
1725 set_current_state(TASK_INTERRUPTIBLE);
1727 __set_current_state(TASK_RUNNING);
1732 static int transaction_kthread(void *arg)
1734 struct btrfs_root *root = arg;
1735 struct btrfs_fs_info *fs_info = root->fs_info;
1736 struct btrfs_trans_handle *trans;
1737 struct btrfs_transaction *cur;
1740 unsigned long delay;
1744 cannot_commit = false;
1745 delay = HZ * fs_info->commit_interval;
1746 mutex_lock(&fs_info->transaction_kthread_mutex);
1748 spin_lock(&fs_info->trans_lock);
1749 cur = fs_info->running_transaction;
1751 spin_unlock(&fs_info->trans_lock);
1755 now = ktime_get_seconds();
1756 if (cur->state < TRANS_STATE_BLOCKED &&
1757 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1758 (now < cur->start_time ||
1759 now - cur->start_time < fs_info->commit_interval)) {
1760 spin_unlock(&fs_info->trans_lock);
1764 transid = cur->transid;
1765 spin_unlock(&fs_info->trans_lock);
1767 /* If the file system is aborted, this will always fail. */
1768 trans = btrfs_attach_transaction(root);
1769 if (IS_ERR(trans)) {
1770 if (PTR_ERR(trans) != -ENOENT)
1771 cannot_commit = true;
1774 if (transid == trans->transid) {
1775 btrfs_commit_transaction(trans);
1777 btrfs_end_transaction(trans);
1780 wake_up_process(fs_info->cleaner_kthread);
1781 mutex_unlock(&fs_info->transaction_kthread_mutex);
1783 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1784 &fs_info->fs_state)))
1785 btrfs_cleanup_transaction(fs_info);
1786 if (!kthread_should_stop() &&
1787 (!btrfs_transaction_blocked(fs_info) ||
1789 schedule_timeout_interruptible(delay);
1790 } while (!kthread_should_stop());
1795 * this will find the highest generation in the array of
1796 * root backups. The index of the highest array is returned,
1797 * or -1 if we can't find anything.
1799 * We check to make sure the array is valid by comparing the
1800 * generation of the latest root in the array with the generation
1801 * in the super block. If they don't match we pitch it.
1803 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1806 int newest_index = -1;
1807 struct btrfs_root_backup *root_backup;
1810 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811 root_backup = info->super_copy->super_roots + i;
1812 cur = btrfs_backup_tree_root_gen(root_backup);
1813 if (cur == newest_gen)
1817 /* check to see if we actually wrapped around */
1818 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819 root_backup = info->super_copy->super_roots;
1820 cur = btrfs_backup_tree_root_gen(root_backup);
1821 if (cur == newest_gen)
1824 return newest_index;
1829 * find the oldest backup so we know where to store new entries
1830 * in the backup array. This will set the backup_root_index
1831 * field in the fs_info struct
1833 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1836 int newest_index = -1;
1838 newest_index = find_newest_super_backup(info, newest_gen);
1839 /* if there was garbage in there, just move along */
1840 if (newest_index == -1) {
1841 info->backup_root_index = 0;
1843 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1852 static void backup_super_roots(struct btrfs_fs_info *info)
1855 struct btrfs_root_backup *root_backup;
1858 next_backup = info->backup_root_index;
1859 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860 BTRFS_NUM_BACKUP_ROOTS;
1863 * just overwrite the last backup if we're at the same generation
1864 * this happens only at umount
1866 root_backup = info->super_for_commit->super_roots + last_backup;
1867 if (btrfs_backup_tree_root_gen(root_backup) ==
1868 btrfs_header_generation(info->tree_root->node))
1869 next_backup = last_backup;
1871 root_backup = info->super_for_commit->super_roots + next_backup;
1874 * make sure all of our padding and empty slots get zero filled
1875 * regardless of which ones we use today
1877 memset(root_backup, 0, sizeof(*root_backup));
1879 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1881 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882 btrfs_set_backup_tree_root_gen(root_backup,
1883 btrfs_header_generation(info->tree_root->node));
1885 btrfs_set_backup_tree_root_level(root_backup,
1886 btrfs_header_level(info->tree_root->node));
1888 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889 btrfs_set_backup_chunk_root_gen(root_backup,
1890 btrfs_header_generation(info->chunk_root->node));
1891 btrfs_set_backup_chunk_root_level(root_backup,
1892 btrfs_header_level(info->chunk_root->node));
1894 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895 btrfs_set_backup_extent_root_gen(root_backup,
1896 btrfs_header_generation(info->extent_root->node));
1897 btrfs_set_backup_extent_root_level(root_backup,
1898 btrfs_header_level(info->extent_root->node));
1901 * we might commit during log recovery, which happens before we set
1902 * the fs_root. Make sure it is valid before we fill it in.
1904 if (info->fs_root && info->fs_root->node) {
1905 btrfs_set_backup_fs_root(root_backup,
1906 info->fs_root->node->start);
1907 btrfs_set_backup_fs_root_gen(root_backup,
1908 btrfs_header_generation(info->fs_root->node));
1909 btrfs_set_backup_fs_root_level(root_backup,
1910 btrfs_header_level(info->fs_root->node));
1913 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914 btrfs_set_backup_dev_root_gen(root_backup,
1915 btrfs_header_generation(info->dev_root->node));
1916 btrfs_set_backup_dev_root_level(root_backup,
1917 btrfs_header_level(info->dev_root->node));
1919 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920 btrfs_set_backup_csum_root_gen(root_backup,
1921 btrfs_header_generation(info->csum_root->node));
1922 btrfs_set_backup_csum_root_level(root_backup,
1923 btrfs_header_level(info->csum_root->node));
1925 btrfs_set_backup_total_bytes(root_backup,
1926 btrfs_super_total_bytes(info->super_copy));
1927 btrfs_set_backup_bytes_used(root_backup,
1928 btrfs_super_bytes_used(info->super_copy));
1929 btrfs_set_backup_num_devices(root_backup,
1930 btrfs_super_num_devices(info->super_copy));
1933 * if we don't copy this out to the super_copy, it won't get remembered
1934 * for the next commit
1936 memcpy(&info->super_copy->super_roots,
1937 &info->super_for_commit->super_roots,
1938 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1942 * this copies info out of the root backup array and back into
1943 * the in-memory super block. It is meant to help iterate through
1944 * the array, so you send it the number of backups you've already
1945 * tried and the last backup index you used.
1947 * this returns -1 when it has tried all the backups
1949 static noinline int next_root_backup(struct btrfs_fs_info *info,
1950 struct btrfs_super_block *super,
1951 int *num_backups_tried, int *backup_index)
1953 struct btrfs_root_backup *root_backup;
1954 int newest = *backup_index;
1956 if (*num_backups_tried == 0) {
1957 u64 gen = btrfs_super_generation(super);
1959 newest = find_newest_super_backup(info, gen);
1963 *backup_index = newest;
1964 *num_backups_tried = 1;
1965 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966 /* we've tried all the backups, all done */
1969 /* jump to the next oldest backup */
1970 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 BTRFS_NUM_BACKUP_ROOTS;
1972 *backup_index = newest;
1973 *num_backups_tried += 1;
1975 root_backup = super->super_roots + newest;
1977 btrfs_set_super_generation(super,
1978 btrfs_backup_tree_root_gen(root_backup));
1979 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980 btrfs_set_super_root_level(super,
1981 btrfs_backup_tree_root_level(root_backup));
1982 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1985 * fixme: the total bytes and num_devices need to match or we should
1988 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1993 /* helper to cleanup workers */
1994 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1996 btrfs_destroy_workqueue(fs_info->fixup_workers);
1997 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1998 btrfs_destroy_workqueue(fs_info->workers);
1999 btrfs_destroy_workqueue(fs_info->endio_workers);
2000 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2001 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2002 btrfs_destroy_workqueue(fs_info->rmw_workers);
2003 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2004 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2005 btrfs_destroy_workqueue(fs_info->submit_workers);
2006 btrfs_destroy_workqueue(fs_info->delayed_workers);
2007 btrfs_destroy_workqueue(fs_info->caching_workers);
2008 btrfs_destroy_workqueue(fs_info->readahead_workers);
2009 btrfs_destroy_workqueue(fs_info->flush_workers);
2010 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2011 btrfs_destroy_workqueue(fs_info->extent_workers);
2013 * Now that all other work queues are destroyed, we can safely destroy
2014 * the queues used for metadata I/O, since tasks from those other work
2015 * queues can do metadata I/O operations.
2017 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2018 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2021 static void free_root_extent_buffers(struct btrfs_root *root)
2024 free_extent_buffer(root->node);
2025 free_extent_buffer(root->commit_root);
2027 root->commit_root = NULL;
2031 /* helper to cleanup tree roots */
2032 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2034 free_root_extent_buffers(info->tree_root);
2036 free_root_extent_buffers(info->dev_root);
2037 free_root_extent_buffers(info->extent_root);
2038 free_root_extent_buffers(info->csum_root);
2039 free_root_extent_buffers(info->quota_root);
2040 free_root_extent_buffers(info->uuid_root);
2042 free_root_extent_buffers(info->chunk_root);
2043 free_root_extent_buffers(info->free_space_root);
2046 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2049 struct btrfs_root *gang[8];
2052 while (!list_empty(&fs_info->dead_roots)) {
2053 gang[0] = list_entry(fs_info->dead_roots.next,
2054 struct btrfs_root, root_list);
2055 list_del(&gang[0]->root_list);
2057 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2058 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2060 free_extent_buffer(gang[0]->node);
2061 free_extent_buffer(gang[0]->commit_root);
2062 btrfs_put_fs_root(gang[0]);
2067 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2072 for (i = 0; i < ret; i++)
2073 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2076 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2077 btrfs_free_log_root_tree(NULL, fs_info);
2078 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2082 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2084 mutex_init(&fs_info->scrub_lock);
2085 atomic_set(&fs_info->scrubs_running, 0);
2086 atomic_set(&fs_info->scrub_pause_req, 0);
2087 atomic_set(&fs_info->scrubs_paused, 0);
2088 atomic_set(&fs_info->scrub_cancel_req, 0);
2089 init_waitqueue_head(&fs_info->scrub_pause_wait);
2090 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2093 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2095 spin_lock_init(&fs_info->balance_lock);
2096 mutex_init(&fs_info->balance_mutex);
2097 atomic_set(&fs_info->balance_pause_req, 0);
2098 atomic_set(&fs_info->balance_cancel_req, 0);
2099 fs_info->balance_ctl = NULL;
2100 init_waitqueue_head(&fs_info->balance_wait_q);
2103 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2105 struct inode *inode = fs_info->btree_inode;
2107 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2108 set_nlink(inode, 1);
2110 * we set the i_size on the btree inode to the max possible int.
2111 * the real end of the address space is determined by all of
2112 * the devices in the system
2114 inode->i_size = OFFSET_MAX;
2115 inode->i_mapping->a_ops = &btree_aops;
2117 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2118 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2119 IO_TREE_INODE_IO, inode);
2120 BTRFS_I(inode)->io_tree.track_uptodate = false;
2121 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2123 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2125 BTRFS_I(inode)->root = fs_info->tree_root;
2126 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2127 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2128 btrfs_insert_inode_hash(inode);
2131 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2133 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2134 init_rwsem(&fs_info->dev_replace.rwsem);
2135 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2138 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2140 spin_lock_init(&fs_info->qgroup_lock);
2141 mutex_init(&fs_info->qgroup_ioctl_lock);
2142 fs_info->qgroup_tree = RB_ROOT;
2143 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2144 fs_info->qgroup_seq = 1;
2145 fs_info->qgroup_ulist = NULL;
2146 fs_info->qgroup_rescan_running = false;
2147 mutex_init(&fs_info->qgroup_rescan_lock);
2150 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2151 struct btrfs_fs_devices *fs_devices)
2153 u32 max_active = fs_info->thread_pool_size;
2154 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2157 btrfs_alloc_workqueue(fs_info, "worker",
2158 flags | WQ_HIGHPRI, max_active, 16);
2160 fs_info->delalloc_workers =
2161 btrfs_alloc_workqueue(fs_info, "delalloc",
2162 flags, max_active, 2);
2164 fs_info->flush_workers =
2165 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2166 flags, max_active, 0);
2168 fs_info->caching_workers =
2169 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2172 * a higher idle thresh on the submit workers makes it much more
2173 * likely that bios will be send down in a sane order to the
2176 fs_info->submit_workers =
2177 btrfs_alloc_workqueue(fs_info, "submit", flags,
2178 min_t(u64, fs_devices->num_devices,
2181 fs_info->fixup_workers =
2182 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2185 * endios are largely parallel and should have a very
2188 fs_info->endio_workers =
2189 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2190 fs_info->endio_meta_workers =
2191 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2193 fs_info->endio_meta_write_workers =
2194 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2196 fs_info->endio_raid56_workers =
2197 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2199 fs_info->endio_repair_workers =
2200 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2201 fs_info->rmw_workers =
2202 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2203 fs_info->endio_write_workers =
2204 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2206 fs_info->endio_freespace_worker =
2207 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2209 fs_info->delayed_workers =
2210 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2212 fs_info->readahead_workers =
2213 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2215 fs_info->qgroup_rescan_workers =
2216 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2217 fs_info->extent_workers =
2218 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2219 min_t(u64, fs_devices->num_devices,
2222 if (!(fs_info->workers && fs_info->delalloc_workers &&
2223 fs_info->submit_workers && fs_info->flush_workers &&
2224 fs_info->endio_workers && fs_info->endio_meta_workers &&
2225 fs_info->endio_meta_write_workers &&
2226 fs_info->endio_repair_workers &&
2227 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2228 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2229 fs_info->caching_workers && fs_info->readahead_workers &&
2230 fs_info->fixup_workers && fs_info->delayed_workers &&
2231 fs_info->extent_workers &&
2232 fs_info->qgroup_rescan_workers)) {
2239 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2241 struct crypto_shash *csum_shash;
2242 const char *csum_name = btrfs_super_csum_name(csum_type);
2244 csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2246 if (IS_ERR(csum_shash)) {
2247 btrfs_err(fs_info, "error allocating %s hash for checksum",
2249 return PTR_ERR(csum_shash);
2252 fs_info->csum_shash = csum_shash;
2257 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2259 crypto_free_shash(fs_info->csum_shash);
2262 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2263 struct btrfs_fs_devices *fs_devices)
2266 struct btrfs_root *log_tree_root;
2267 struct btrfs_super_block *disk_super = fs_info->super_copy;
2268 u64 bytenr = btrfs_super_log_root(disk_super);
2269 int level = btrfs_super_log_root_level(disk_super);
2271 if (fs_devices->rw_devices == 0) {
2272 btrfs_warn(fs_info, "log replay required on RO media");
2276 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2280 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2282 log_tree_root->node = read_tree_block(fs_info, bytenr,
2283 fs_info->generation + 1,
2285 if (IS_ERR(log_tree_root->node)) {
2286 btrfs_warn(fs_info, "failed to read log tree");
2287 ret = PTR_ERR(log_tree_root->node);
2288 kfree(log_tree_root);
2290 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2291 btrfs_err(fs_info, "failed to read log tree");
2292 free_extent_buffer(log_tree_root->node);
2293 kfree(log_tree_root);
2296 /* returns with log_tree_root freed on success */
2297 ret = btrfs_recover_log_trees(log_tree_root);
2299 btrfs_handle_fs_error(fs_info, ret,
2300 "Failed to recover log tree");
2301 free_extent_buffer(log_tree_root->node);
2302 kfree(log_tree_root);
2306 if (sb_rdonly(fs_info->sb)) {
2307 ret = btrfs_commit_super(fs_info);
2315 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2317 struct btrfs_root *tree_root = fs_info->tree_root;
2318 struct btrfs_root *root;
2319 struct btrfs_key location;
2322 BUG_ON(!fs_info->tree_root);
2324 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2325 location.type = BTRFS_ROOT_ITEM_KEY;
2326 location.offset = 0;
2328 root = btrfs_read_tree_root(tree_root, &location);
2330 ret = PTR_ERR(root);
2333 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2334 fs_info->extent_root = root;
2336 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2337 root = btrfs_read_tree_root(tree_root, &location);
2339 ret = PTR_ERR(root);
2342 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2343 fs_info->dev_root = root;
2344 btrfs_init_devices_late(fs_info);
2346 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2347 root = btrfs_read_tree_root(tree_root, &location);
2349 ret = PTR_ERR(root);
2352 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2353 fs_info->csum_root = root;
2355 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2356 root = btrfs_read_tree_root(tree_root, &location);
2357 if (!IS_ERR(root)) {
2358 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2359 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2360 fs_info->quota_root = root;
2363 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2364 root = btrfs_read_tree_root(tree_root, &location);
2366 ret = PTR_ERR(root);
2370 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2371 fs_info->uuid_root = root;
2374 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2375 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2376 root = btrfs_read_tree_root(tree_root, &location);
2378 ret = PTR_ERR(root);
2381 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2382 fs_info->free_space_root = root;
2387 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2388 location.objectid, ret);
2393 * Real super block validation
2394 * NOTE: super csum type and incompat features will not be checked here.
2396 * @sb: super block to check
2397 * @mirror_num: the super block number to check its bytenr:
2398 * 0 the primary (1st) sb
2399 * 1, 2 2nd and 3rd backup copy
2400 * -1 skip bytenr check
2402 static int validate_super(struct btrfs_fs_info *fs_info,
2403 struct btrfs_super_block *sb, int mirror_num)
2405 u64 nodesize = btrfs_super_nodesize(sb);
2406 u64 sectorsize = btrfs_super_sectorsize(sb);
2409 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2410 btrfs_err(fs_info, "no valid FS found");
2413 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2414 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2415 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2418 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2419 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2420 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2423 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2424 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2425 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2428 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2429 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2430 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2435 * Check sectorsize and nodesize first, other check will need it.
2436 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2438 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2439 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2440 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2443 /* Only PAGE SIZE is supported yet */
2444 if (sectorsize != PAGE_SIZE) {
2446 "sectorsize %llu not supported yet, only support %lu",
2447 sectorsize, PAGE_SIZE);
2450 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2451 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2452 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2455 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2456 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2457 le32_to_cpu(sb->__unused_leafsize), nodesize);
2461 /* Root alignment check */
2462 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2463 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2464 btrfs_super_root(sb));
2467 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2468 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2469 btrfs_super_chunk_root(sb));
2472 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2473 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2474 btrfs_super_log_root(sb));
2478 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2479 BTRFS_FSID_SIZE) != 0) {
2481 "dev_item UUID does not match metadata fsid: %pU != %pU",
2482 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2487 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2490 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2491 btrfs_err(fs_info, "bytes_used is too small %llu",
2492 btrfs_super_bytes_used(sb));
2495 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2496 btrfs_err(fs_info, "invalid stripesize %u",
2497 btrfs_super_stripesize(sb));
2500 if (btrfs_super_num_devices(sb) > (1UL << 31))
2501 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2502 btrfs_super_num_devices(sb));
2503 if (btrfs_super_num_devices(sb) == 0) {
2504 btrfs_err(fs_info, "number of devices is 0");
2508 if (mirror_num >= 0 &&
2509 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2510 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2511 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2516 * Obvious sys_chunk_array corruptions, it must hold at least one key
2519 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2520 btrfs_err(fs_info, "system chunk array too big %u > %u",
2521 btrfs_super_sys_array_size(sb),
2522 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2525 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2526 + sizeof(struct btrfs_chunk)) {
2527 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2528 btrfs_super_sys_array_size(sb),
2529 sizeof(struct btrfs_disk_key)
2530 + sizeof(struct btrfs_chunk));
2535 * The generation is a global counter, we'll trust it more than the others
2536 * but it's still possible that it's the one that's wrong.
2538 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2540 "suspicious: generation < chunk_root_generation: %llu < %llu",
2541 btrfs_super_generation(sb),
2542 btrfs_super_chunk_root_generation(sb));
2543 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2544 && btrfs_super_cache_generation(sb) != (u64)-1)
2546 "suspicious: generation < cache_generation: %llu < %llu",
2547 btrfs_super_generation(sb),
2548 btrfs_super_cache_generation(sb));
2554 * Validation of super block at mount time.
2555 * Some checks already done early at mount time, like csum type and incompat
2556 * flags will be skipped.
2558 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2560 return validate_super(fs_info, fs_info->super_copy, 0);
2564 * Validation of super block at write time.
2565 * Some checks like bytenr check will be skipped as their values will be
2567 * Extra checks like csum type and incompat flags will be done here.
2569 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2570 struct btrfs_super_block *sb)
2574 ret = validate_super(fs_info, sb, -1);
2577 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2579 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2580 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2583 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2586 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2587 btrfs_super_incompat_flags(sb),
2588 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2594 "super block corruption detected before writing it to disk");
2598 int open_ctree(struct super_block *sb,
2599 struct btrfs_fs_devices *fs_devices,
2608 struct btrfs_key location;
2609 struct buffer_head *bh;
2610 struct btrfs_super_block *disk_super;
2611 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2612 struct btrfs_root *tree_root;
2613 struct btrfs_root *chunk_root;
2616 int num_backups_tried = 0;
2617 int backup_index = 0;
2618 int clear_free_space_tree = 0;
2621 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2622 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2623 if (!tree_root || !chunk_root) {
2628 ret = init_srcu_struct(&fs_info->subvol_srcu);
2634 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2640 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2643 goto fail_dio_bytes;
2645 fs_info->dirty_metadata_batch = PAGE_SIZE *
2646 (1 + ilog2(nr_cpu_ids));
2648 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2651 goto fail_dirty_metadata_bytes;
2654 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2658 goto fail_delalloc_bytes;
2661 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2662 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2663 INIT_LIST_HEAD(&fs_info->trans_list);
2664 INIT_LIST_HEAD(&fs_info->dead_roots);
2665 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2666 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2667 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2668 spin_lock_init(&fs_info->delalloc_root_lock);
2669 spin_lock_init(&fs_info->trans_lock);
2670 spin_lock_init(&fs_info->fs_roots_radix_lock);
2671 spin_lock_init(&fs_info->delayed_iput_lock);
2672 spin_lock_init(&fs_info->defrag_inodes_lock);
2673 spin_lock_init(&fs_info->tree_mod_seq_lock);
2674 spin_lock_init(&fs_info->super_lock);
2675 spin_lock_init(&fs_info->buffer_lock);
2676 spin_lock_init(&fs_info->unused_bgs_lock);
2677 rwlock_init(&fs_info->tree_mod_log_lock);
2678 mutex_init(&fs_info->unused_bg_unpin_mutex);
2679 mutex_init(&fs_info->delete_unused_bgs_mutex);
2680 mutex_init(&fs_info->reloc_mutex);
2681 mutex_init(&fs_info->delalloc_root_mutex);
2682 seqlock_init(&fs_info->profiles_lock);
2684 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2685 INIT_LIST_HEAD(&fs_info->space_info);
2686 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2687 INIT_LIST_HEAD(&fs_info->unused_bgs);
2688 extent_map_tree_init(&fs_info->mapping_tree);
2689 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2690 BTRFS_BLOCK_RSV_GLOBAL);
2691 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2692 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2693 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2694 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2695 BTRFS_BLOCK_RSV_DELOPS);
2696 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2697 BTRFS_BLOCK_RSV_DELREFS);
2699 atomic_set(&fs_info->async_delalloc_pages, 0);
2700 atomic_set(&fs_info->defrag_running, 0);
2701 atomic_set(&fs_info->reada_works_cnt, 0);
2702 atomic_set(&fs_info->nr_delayed_iputs, 0);
2703 atomic64_set(&fs_info->tree_mod_seq, 0);
2705 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2706 fs_info->metadata_ratio = 0;
2707 fs_info->defrag_inodes = RB_ROOT;
2708 atomic64_set(&fs_info->free_chunk_space, 0);
2709 fs_info->tree_mod_log = RB_ROOT;
2710 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2711 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2712 /* readahead state */
2713 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2714 spin_lock_init(&fs_info->reada_lock);
2715 btrfs_init_ref_verify(fs_info);
2717 fs_info->thread_pool_size = min_t(unsigned long,
2718 num_online_cpus() + 2, 8);
2720 INIT_LIST_HEAD(&fs_info->ordered_roots);
2721 spin_lock_init(&fs_info->ordered_root_lock);
2723 fs_info->btree_inode = new_inode(sb);
2724 if (!fs_info->btree_inode) {
2726 goto fail_bio_counter;
2728 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2730 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2732 if (!fs_info->delayed_root) {
2736 btrfs_init_delayed_root(fs_info->delayed_root);
2738 btrfs_init_scrub(fs_info);
2739 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2740 fs_info->check_integrity_print_mask = 0;
2742 btrfs_init_balance(fs_info);
2743 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2745 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2746 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2748 btrfs_init_btree_inode(fs_info);
2750 spin_lock_init(&fs_info->block_group_cache_lock);
2751 fs_info->block_group_cache_tree = RB_ROOT;
2752 fs_info->first_logical_byte = (u64)-1;
2754 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2755 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2756 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2757 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2758 fs_info->pinned_extents = &fs_info->freed_extents[0];
2759 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2761 mutex_init(&fs_info->ordered_operations_mutex);
2762 mutex_init(&fs_info->tree_log_mutex);
2763 mutex_init(&fs_info->chunk_mutex);
2764 mutex_init(&fs_info->transaction_kthread_mutex);
2765 mutex_init(&fs_info->cleaner_mutex);
2766 mutex_init(&fs_info->ro_block_group_mutex);
2767 init_rwsem(&fs_info->commit_root_sem);
2768 init_rwsem(&fs_info->cleanup_work_sem);
2769 init_rwsem(&fs_info->subvol_sem);
2770 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2772 btrfs_init_dev_replace_locks(fs_info);
2773 btrfs_init_qgroup(fs_info);
2775 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2776 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2778 init_waitqueue_head(&fs_info->transaction_throttle);
2779 init_waitqueue_head(&fs_info->transaction_wait);
2780 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2781 init_waitqueue_head(&fs_info->async_submit_wait);
2782 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2784 /* Usable values until the real ones are cached from the superblock */
2785 fs_info->nodesize = 4096;
2786 fs_info->sectorsize = 4096;
2787 fs_info->stripesize = 4096;
2789 spin_lock_init(&fs_info->swapfile_pins_lock);
2790 fs_info->swapfile_pins = RB_ROOT;
2792 fs_info->send_in_progress = 0;
2794 ret = btrfs_alloc_stripe_hash_table(fs_info);
2800 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2802 invalidate_bdev(fs_devices->latest_bdev);
2805 * Read super block and check the signature bytes only
2807 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2814 * Verify the type first, if that or the the checksum value are
2815 * corrupted, we'll find out
2817 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2818 if (!btrfs_supported_super_csum(csum_type)) {
2819 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2826 ret = btrfs_init_csum_hash(fs_info, csum_type);
2833 * We want to check superblock checksum, the type is stored inside.
2834 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2836 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2837 btrfs_err(fs_info, "superblock checksum mismatch");
2844 * super_copy is zeroed at allocation time and we never touch the
2845 * following bytes up to INFO_SIZE, the checksum is calculated from
2846 * the whole block of INFO_SIZE
2848 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2851 disk_super = fs_info->super_copy;
2853 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2856 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2857 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2858 fs_info->super_copy->metadata_uuid,
2862 features = btrfs_super_flags(disk_super);
2863 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2864 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2865 btrfs_set_super_flags(disk_super, features);
2867 "found metadata UUID change in progress flag, clearing");
2870 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2871 sizeof(*fs_info->super_for_commit));
2873 ret = btrfs_validate_mount_super(fs_info);
2875 btrfs_err(fs_info, "superblock contains fatal errors");
2880 if (!btrfs_super_root(disk_super))
2883 /* check FS state, whether FS is broken. */
2884 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2885 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2888 * run through our array of backup supers and setup
2889 * our ring pointer to the oldest one
2891 generation = btrfs_super_generation(disk_super);
2892 find_oldest_super_backup(fs_info, generation);
2895 * In the long term, we'll store the compression type in the super
2896 * block, and it'll be used for per file compression control.
2898 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2900 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2906 features = btrfs_super_incompat_flags(disk_super) &
2907 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2910 "cannot mount because of unsupported optional features (%llx)",
2916 features = btrfs_super_incompat_flags(disk_super);
2917 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2918 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2919 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2920 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2921 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2923 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2924 btrfs_info(fs_info, "has skinny extents");
2927 * flag our filesystem as having big metadata blocks if
2928 * they are bigger than the page size
2930 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2931 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2933 "flagging fs with big metadata feature");
2934 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2937 nodesize = btrfs_super_nodesize(disk_super);
2938 sectorsize = btrfs_super_sectorsize(disk_super);
2939 stripesize = sectorsize;
2940 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2941 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2943 /* Cache block sizes */
2944 fs_info->nodesize = nodesize;
2945 fs_info->sectorsize = sectorsize;
2946 fs_info->stripesize = stripesize;
2949 * mixed block groups end up with duplicate but slightly offset
2950 * extent buffers for the same range. It leads to corruptions
2952 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2953 (sectorsize != nodesize)) {
2955 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2956 nodesize, sectorsize);
2961 * Needn't use the lock because there is no other task which will
2964 btrfs_set_super_incompat_flags(disk_super, features);
2966 features = btrfs_super_compat_ro_flags(disk_super) &
2967 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2968 if (!sb_rdonly(sb) && features) {
2970 "cannot mount read-write because of unsupported optional features (%llx)",
2976 ret = btrfs_init_workqueues(fs_info, fs_devices);
2979 goto fail_sb_buffer;
2982 sb->s_bdi->congested_fn = btrfs_congested_fn;
2983 sb->s_bdi->congested_data = fs_info;
2984 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2985 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2986 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2987 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2989 sb->s_blocksize = sectorsize;
2990 sb->s_blocksize_bits = blksize_bits(sectorsize);
2991 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2993 mutex_lock(&fs_info->chunk_mutex);
2994 ret = btrfs_read_sys_array(fs_info);
2995 mutex_unlock(&fs_info->chunk_mutex);
2997 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2998 goto fail_sb_buffer;
3001 generation = btrfs_super_chunk_root_generation(disk_super);
3002 level = btrfs_super_chunk_root_level(disk_super);
3004 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
3006 chunk_root->node = read_tree_block(fs_info,
3007 btrfs_super_chunk_root(disk_super),
3008 generation, level, NULL);
3009 if (IS_ERR(chunk_root->node) ||
3010 !extent_buffer_uptodate(chunk_root->node)) {
3011 btrfs_err(fs_info, "failed to read chunk root");
3012 if (!IS_ERR(chunk_root->node))
3013 free_extent_buffer(chunk_root->node);
3014 chunk_root->node = NULL;
3015 goto fail_tree_roots;
3017 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3018 chunk_root->commit_root = btrfs_root_node(chunk_root);
3020 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3021 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3023 ret = btrfs_read_chunk_tree(fs_info);
3025 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3026 goto fail_tree_roots;
3030 * Keep the devid that is marked to be the target device for the
3031 * device replace procedure
3033 btrfs_free_extra_devids(fs_devices, 0);
3035 if (!fs_devices->latest_bdev) {
3036 btrfs_err(fs_info, "failed to read devices");
3037 goto fail_tree_roots;
3041 generation = btrfs_super_generation(disk_super);
3042 level = btrfs_super_root_level(disk_super);
3044 tree_root->node = read_tree_block(fs_info,
3045 btrfs_super_root(disk_super),
3046 generation, level, NULL);
3047 if (IS_ERR(tree_root->node) ||
3048 !extent_buffer_uptodate(tree_root->node)) {
3049 btrfs_warn(fs_info, "failed to read tree root");
3050 if (!IS_ERR(tree_root->node))
3051 free_extent_buffer(tree_root->node);
3052 tree_root->node = NULL;
3053 goto recovery_tree_root;
3056 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3057 tree_root->commit_root = btrfs_root_node(tree_root);
3058 btrfs_set_root_refs(&tree_root->root_item, 1);
3060 mutex_lock(&tree_root->objectid_mutex);
3061 ret = btrfs_find_highest_objectid(tree_root,
3062 &tree_root->highest_objectid);
3064 mutex_unlock(&tree_root->objectid_mutex);
3065 goto recovery_tree_root;
3068 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3070 mutex_unlock(&tree_root->objectid_mutex);
3072 ret = btrfs_read_roots(fs_info);
3074 goto recovery_tree_root;
3076 fs_info->generation = generation;
3077 fs_info->last_trans_committed = generation;
3079 ret = btrfs_verify_dev_extents(fs_info);
3082 "failed to verify dev extents against chunks: %d",
3084 goto fail_block_groups;
3086 ret = btrfs_recover_balance(fs_info);
3088 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3089 goto fail_block_groups;
3092 ret = btrfs_init_dev_stats(fs_info);
3094 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3095 goto fail_block_groups;
3098 ret = btrfs_init_dev_replace(fs_info);
3100 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3101 goto fail_block_groups;
3104 btrfs_free_extra_devids(fs_devices, 1);
3106 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3108 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3110 goto fail_block_groups;
3113 ret = btrfs_sysfs_add_device(fs_devices);
3115 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3117 goto fail_fsdev_sysfs;
3120 ret = btrfs_sysfs_add_mounted(fs_info);
3122 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3123 goto fail_fsdev_sysfs;
3126 ret = btrfs_init_space_info(fs_info);
3128 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3132 ret = btrfs_read_block_groups(fs_info);
3134 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3138 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3140 "writable mount is not allowed due to too many missing devices");
3144 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3146 if (IS_ERR(fs_info->cleaner_kthread))
3149 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3151 "btrfs-transaction");
3152 if (IS_ERR(fs_info->transaction_kthread))
3155 if (!btrfs_test_opt(fs_info, NOSSD) &&
3156 !fs_info->fs_devices->rotating) {
3157 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3161 * Mount does not set all options immediately, we can do it now and do
3162 * not have to wait for transaction commit
3164 btrfs_apply_pending_changes(fs_info);
3166 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3167 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3168 ret = btrfsic_mount(fs_info, fs_devices,
3169 btrfs_test_opt(fs_info,
3170 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3172 fs_info->check_integrity_print_mask);
3175 "failed to initialize integrity check module: %d",
3179 ret = btrfs_read_qgroup_config(fs_info);
3181 goto fail_trans_kthread;
3183 if (btrfs_build_ref_tree(fs_info))
3184 btrfs_err(fs_info, "couldn't build ref tree");
3186 /* do not make disk changes in broken FS or nologreplay is given */
3187 if (btrfs_super_log_root(disk_super) != 0 &&
3188 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3189 ret = btrfs_replay_log(fs_info, fs_devices);
3196 ret = btrfs_find_orphan_roots(fs_info);
3200 if (!sb_rdonly(sb)) {
3201 ret = btrfs_cleanup_fs_roots(fs_info);
3205 mutex_lock(&fs_info->cleaner_mutex);
3206 ret = btrfs_recover_relocation(tree_root);
3207 mutex_unlock(&fs_info->cleaner_mutex);
3209 btrfs_warn(fs_info, "failed to recover relocation: %d",
3216 location.objectid = BTRFS_FS_TREE_OBJECTID;
3217 location.type = BTRFS_ROOT_ITEM_KEY;
3218 location.offset = 0;
3220 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3221 if (IS_ERR(fs_info->fs_root)) {
3222 err = PTR_ERR(fs_info->fs_root);
3223 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3230 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3231 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3232 clear_free_space_tree = 1;
3233 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3234 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3235 btrfs_warn(fs_info, "free space tree is invalid");
3236 clear_free_space_tree = 1;
3239 if (clear_free_space_tree) {
3240 btrfs_info(fs_info, "clearing free space tree");
3241 ret = btrfs_clear_free_space_tree(fs_info);
3244 "failed to clear free space tree: %d", ret);
3245 close_ctree(fs_info);
3250 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3251 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3252 btrfs_info(fs_info, "creating free space tree");
3253 ret = btrfs_create_free_space_tree(fs_info);
3256 "failed to create free space tree: %d", ret);
3257 close_ctree(fs_info);
3262 down_read(&fs_info->cleanup_work_sem);
3263 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3264 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3265 up_read(&fs_info->cleanup_work_sem);
3266 close_ctree(fs_info);
3269 up_read(&fs_info->cleanup_work_sem);
3271 ret = btrfs_resume_balance_async(fs_info);
3273 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3274 close_ctree(fs_info);
3278 ret = btrfs_resume_dev_replace_async(fs_info);
3280 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3281 close_ctree(fs_info);
3285 btrfs_qgroup_rescan_resume(fs_info);
3287 if (!fs_info->uuid_root) {
3288 btrfs_info(fs_info, "creating UUID tree");
3289 ret = btrfs_create_uuid_tree(fs_info);
3292 "failed to create the UUID tree: %d", ret);
3293 close_ctree(fs_info);
3296 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3297 fs_info->generation !=
3298 btrfs_super_uuid_tree_generation(disk_super)) {
3299 btrfs_info(fs_info, "checking UUID tree");
3300 ret = btrfs_check_uuid_tree(fs_info);
3303 "failed to check the UUID tree: %d", ret);
3304 close_ctree(fs_info);
3308 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3310 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3313 * backuproot only affect mount behavior, and if open_ctree succeeded,
3314 * no need to keep the flag
3316 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3321 btrfs_free_qgroup_config(fs_info);
3323 kthread_stop(fs_info->transaction_kthread);
3324 btrfs_cleanup_transaction(fs_info);
3325 btrfs_free_fs_roots(fs_info);
3327 kthread_stop(fs_info->cleaner_kthread);
3330 * make sure we're done with the btree inode before we stop our
3333 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3336 btrfs_sysfs_remove_mounted(fs_info);
3339 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3342 btrfs_put_block_group_cache(fs_info);
3345 free_root_pointers(fs_info, 1);
3346 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3349 btrfs_stop_all_workers(fs_info);
3350 btrfs_free_block_groups(fs_info);
3352 btrfs_free_csum_hash(fs_info);
3355 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3357 iput(fs_info->btree_inode);
3359 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3360 fail_delalloc_bytes:
3361 percpu_counter_destroy(&fs_info->delalloc_bytes);
3362 fail_dirty_metadata_bytes:
3363 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3365 percpu_counter_destroy(&fs_info->dio_bytes);
3367 cleanup_srcu_struct(&fs_info->subvol_srcu);
3369 btrfs_free_stripe_hash_table(fs_info);
3370 btrfs_close_devices(fs_info->fs_devices);
3374 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3375 goto fail_tree_roots;
3377 free_root_pointers(fs_info, 0);
3379 /* don't use the log in recovery mode, it won't be valid */
3380 btrfs_set_super_log_root(disk_super, 0);
3382 /* we can't trust the free space cache either */
3383 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3385 ret = next_root_backup(fs_info, fs_info->super_copy,
3386 &num_backups_tried, &backup_index);
3388 goto fail_block_groups;
3389 goto retry_root_backup;
3391 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3393 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3396 set_buffer_uptodate(bh);
3398 struct btrfs_device *device = (struct btrfs_device *)
3401 btrfs_warn_rl_in_rcu(device->fs_info,
3402 "lost page write due to IO error on %s",
3403 rcu_str_deref(device->name));
3404 /* note, we don't set_buffer_write_io_error because we have
3405 * our own ways of dealing with the IO errors
3407 clear_buffer_uptodate(bh);
3408 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3414 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3415 struct buffer_head **bh_ret)
3417 struct buffer_head *bh;
3418 struct btrfs_super_block *super;
3421 bytenr = btrfs_sb_offset(copy_num);
3422 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3425 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3427 * If we fail to read from the underlying devices, as of now
3428 * the best option we have is to mark it EIO.
3433 super = (struct btrfs_super_block *)bh->b_data;
3434 if (btrfs_super_bytenr(super) != bytenr ||
3435 btrfs_super_magic(super) != BTRFS_MAGIC) {
3445 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3447 struct buffer_head *bh;
3448 struct buffer_head *latest = NULL;
3449 struct btrfs_super_block *super;
3454 /* we would like to check all the supers, but that would make
3455 * a btrfs mount succeed after a mkfs from a different FS.
3456 * So, we need to add a special mount option to scan for
3457 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3459 for (i = 0; i < 1; i++) {
3460 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3464 super = (struct btrfs_super_block *)bh->b_data;
3466 if (!latest || btrfs_super_generation(super) > transid) {
3469 transid = btrfs_super_generation(super);
3476 return ERR_PTR(ret);
3482 * Write superblock @sb to the @device. Do not wait for completion, all the
3483 * buffer heads we write are pinned.
3485 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3486 * the expected device size at commit time. Note that max_mirrors must be
3487 * same for write and wait phases.
3489 * Return number of errors when buffer head is not found or submission fails.
3491 static int write_dev_supers(struct btrfs_device *device,
3492 struct btrfs_super_block *sb, int max_mirrors)
3494 struct btrfs_fs_info *fs_info = device->fs_info;
3495 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3496 struct buffer_head *bh;
3503 if (max_mirrors == 0)
3504 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3506 shash->tfm = fs_info->csum_shash;
3508 for (i = 0; i < max_mirrors; i++) {
3509 bytenr = btrfs_sb_offset(i);
3510 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3511 device->commit_total_bytes)
3514 btrfs_set_super_bytenr(sb, bytenr);
3516 crypto_shash_init(shash);
3517 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3518 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3519 crypto_shash_final(shash, sb->csum);
3521 /* One reference for us, and we leave it for the caller */
3522 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3523 BTRFS_SUPER_INFO_SIZE);
3525 btrfs_err(device->fs_info,
3526 "couldn't get super buffer head for bytenr %llu",
3532 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3534 /* one reference for submit_bh */
3537 set_buffer_uptodate(bh);
3539 bh->b_end_io = btrfs_end_buffer_write_sync;
3540 bh->b_private = device;
3543 * we fua the first super. The others we allow
3546 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3547 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3548 op_flags |= REQ_FUA;
3549 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3553 return errors < i ? 0 : -1;
3557 * Wait for write completion of superblocks done by write_dev_supers,
3558 * @max_mirrors same for write and wait phases.
3560 * Return number of errors when buffer head is not found or not marked up to
3563 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3565 struct buffer_head *bh;
3568 bool primary_failed = false;
3571 if (max_mirrors == 0)
3572 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3574 for (i = 0; i < max_mirrors; i++) {
3575 bytenr = btrfs_sb_offset(i);
3576 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3577 device->commit_total_bytes)
3580 bh = __find_get_block(device->bdev,
3581 bytenr / BTRFS_BDEV_BLOCKSIZE,
3582 BTRFS_SUPER_INFO_SIZE);
3586 primary_failed = true;
3590 if (!buffer_uptodate(bh)) {
3593 primary_failed = true;
3596 /* drop our reference */
3599 /* drop the reference from the writing run */
3603 /* log error, force error return */
3604 if (primary_failed) {
3605 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3610 return errors < i ? 0 : -1;
3614 * endio for the write_dev_flush, this will wake anyone waiting
3615 * for the barrier when it is done
3617 static void btrfs_end_empty_barrier(struct bio *bio)
3619 complete(bio->bi_private);
3623 * Submit a flush request to the device if it supports it. Error handling is
3624 * done in the waiting counterpart.
3626 static void write_dev_flush(struct btrfs_device *device)
3628 struct request_queue *q = bdev_get_queue(device->bdev);
3629 struct bio *bio = device->flush_bio;
3631 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3635 bio->bi_end_io = btrfs_end_empty_barrier;
3636 bio_set_dev(bio, device->bdev);
3637 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3638 init_completion(&device->flush_wait);
3639 bio->bi_private = &device->flush_wait;
3641 btrfsic_submit_bio(bio);
3642 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3646 * If the flush bio has been submitted by write_dev_flush, wait for it.
3648 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3650 struct bio *bio = device->flush_bio;
3652 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3655 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3656 wait_for_completion_io(&device->flush_wait);
3658 return bio->bi_status;
3661 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3663 if (!btrfs_check_rw_degradable(fs_info, NULL))
3669 * send an empty flush down to each device in parallel,
3670 * then wait for them
3672 static int barrier_all_devices(struct btrfs_fs_info *info)
3674 struct list_head *head;
3675 struct btrfs_device *dev;
3676 int errors_wait = 0;
3679 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3680 /* send down all the barriers */
3681 head = &info->fs_devices->devices;
3682 list_for_each_entry(dev, head, dev_list) {
3683 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3687 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3688 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3691 write_dev_flush(dev);
3692 dev->last_flush_error = BLK_STS_OK;
3695 /* wait for all the barriers */
3696 list_for_each_entry(dev, head, dev_list) {
3697 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3703 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3704 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3707 ret = wait_dev_flush(dev);
3709 dev->last_flush_error = ret;
3710 btrfs_dev_stat_inc_and_print(dev,
3711 BTRFS_DEV_STAT_FLUSH_ERRS);
3718 * At some point we need the status of all disks
3719 * to arrive at the volume status. So error checking
3720 * is being pushed to a separate loop.
3722 return check_barrier_error(info);
3727 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3730 int min_tolerated = INT_MAX;
3732 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3733 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3734 min_tolerated = min_t(int, min_tolerated,
3735 btrfs_raid_array[BTRFS_RAID_SINGLE].
3736 tolerated_failures);
3738 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3739 if (raid_type == BTRFS_RAID_SINGLE)
3741 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3743 min_tolerated = min_t(int, min_tolerated,
3744 btrfs_raid_array[raid_type].
3745 tolerated_failures);
3748 if (min_tolerated == INT_MAX) {
3749 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3753 return min_tolerated;
3756 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3758 struct list_head *head;
3759 struct btrfs_device *dev;
3760 struct btrfs_super_block *sb;
3761 struct btrfs_dev_item *dev_item;
3765 int total_errors = 0;
3768 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3771 * max_mirrors == 0 indicates we're from commit_transaction,
3772 * not from fsync where the tree roots in fs_info have not
3773 * been consistent on disk.
3775 if (max_mirrors == 0)
3776 backup_super_roots(fs_info);
3778 sb = fs_info->super_for_commit;
3779 dev_item = &sb->dev_item;
3781 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3782 head = &fs_info->fs_devices->devices;
3783 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3786 ret = barrier_all_devices(fs_info);
3789 &fs_info->fs_devices->device_list_mutex);
3790 btrfs_handle_fs_error(fs_info, ret,
3791 "errors while submitting device barriers.");
3796 list_for_each_entry(dev, head, dev_list) {
3801 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3802 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3805 btrfs_set_stack_device_generation(dev_item, 0);
3806 btrfs_set_stack_device_type(dev_item, dev->type);
3807 btrfs_set_stack_device_id(dev_item, dev->devid);
3808 btrfs_set_stack_device_total_bytes(dev_item,
3809 dev->commit_total_bytes);
3810 btrfs_set_stack_device_bytes_used(dev_item,
3811 dev->commit_bytes_used);
3812 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3813 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3814 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3815 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3816 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3819 flags = btrfs_super_flags(sb);
3820 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3822 ret = btrfs_validate_write_super(fs_info, sb);
3824 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3825 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3826 "unexpected superblock corruption detected");
3830 ret = write_dev_supers(dev, sb, max_mirrors);
3834 if (total_errors > max_errors) {
3835 btrfs_err(fs_info, "%d errors while writing supers",
3837 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3839 /* FUA is masked off if unsupported and can't be the reason */
3840 btrfs_handle_fs_error(fs_info, -EIO,
3841 "%d errors while writing supers",
3847 list_for_each_entry(dev, head, dev_list) {
3850 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3851 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3854 ret = wait_dev_supers(dev, max_mirrors);
3858 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3859 if (total_errors > max_errors) {
3860 btrfs_handle_fs_error(fs_info, -EIO,
3861 "%d errors while writing supers",
3868 /* Drop a fs root from the radix tree and free it. */
3869 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3870 struct btrfs_root *root)
3872 spin_lock(&fs_info->fs_roots_radix_lock);
3873 radix_tree_delete(&fs_info->fs_roots_radix,
3874 (unsigned long)root->root_key.objectid);
3875 spin_unlock(&fs_info->fs_roots_radix_lock);
3877 if (btrfs_root_refs(&root->root_item) == 0)
3878 synchronize_srcu(&fs_info->subvol_srcu);
3880 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3881 btrfs_free_log(NULL, root);
3882 if (root->reloc_root) {
3883 free_extent_buffer(root->reloc_root->node);
3884 free_extent_buffer(root->reloc_root->commit_root);
3885 btrfs_put_fs_root(root->reloc_root);
3886 root->reloc_root = NULL;
3890 if (root->free_ino_pinned)
3891 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3892 if (root->free_ino_ctl)
3893 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3894 btrfs_free_fs_root(root);
3897 void btrfs_free_fs_root(struct btrfs_root *root)
3899 iput(root->ino_cache_inode);
3900 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3902 free_anon_bdev(root->anon_dev);
3903 if (root->subv_writers)
3904 btrfs_free_subvolume_writers(root->subv_writers);
3905 free_extent_buffer(root->node);
3906 free_extent_buffer(root->commit_root);
3907 kfree(root->free_ino_ctl);
3908 kfree(root->free_ino_pinned);
3909 btrfs_put_fs_root(root);
3912 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3914 u64 root_objectid = 0;
3915 struct btrfs_root *gang[8];
3918 unsigned int ret = 0;
3922 index = srcu_read_lock(&fs_info->subvol_srcu);
3923 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3924 (void **)gang, root_objectid,
3927 srcu_read_unlock(&fs_info->subvol_srcu, index);
3930 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3932 for (i = 0; i < ret; i++) {
3933 /* Avoid to grab roots in dead_roots */
3934 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3938 /* grab all the search result for later use */
3939 gang[i] = btrfs_grab_fs_root(gang[i]);
3941 srcu_read_unlock(&fs_info->subvol_srcu, index);
3943 for (i = 0; i < ret; i++) {
3946 root_objectid = gang[i]->root_key.objectid;
3947 err = btrfs_orphan_cleanup(gang[i]);
3950 btrfs_put_fs_root(gang[i]);
3955 /* release the uncleaned roots due to error */
3956 for (; i < ret; i++) {
3958 btrfs_put_fs_root(gang[i]);
3963 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3965 struct btrfs_root *root = fs_info->tree_root;
3966 struct btrfs_trans_handle *trans;
3968 mutex_lock(&fs_info->cleaner_mutex);
3969 btrfs_run_delayed_iputs(fs_info);
3970 mutex_unlock(&fs_info->cleaner_mutex);
3971 wake_up_process(fs_info->cleaner_kthread);
3973 /* wait until ongoing cleanup work done */
3974 down_write(&fs_info->cleanup_work_sem);
3975 up_write(&fs_info->cleanup_work_sem);
3977 trans = btrfs_join_transaction(root);
3979 return PTR_ERR(trans);
3980 return btrfs_commit_transaction(trans);
3983 void close_ctree(struct btrfs_fs_info *fs_info)
3987 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3989 * We don't want the cleaner to start new transactions, add more delayed
3990 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3991 * because that frees the task_struct, and the transaction kthread might
3992 * still try to wake up the cleaner.
3994 kthread_park(fs_info->cleaner_kthread);
3996 /* wait for the qgroup rescan worker to stop */
3997 btrfs_qgroup_wait_for_completion(fs_info, false);
3999 /* wait for the uuid_scan task to finish */
4000 down(&fs_info->uuid_tree_rescan_sem);
4001 /* avoid complains from lockdep et al., set sem back to initial state */
4002 up(&fs_info->uuid_tree_rescan_sem);
4004 /* pause restriper - we want to resume on mount */
4005 btrfs_pause_balance(fs_info);
4007 btrfs_dev_replace_suspend_for_unmount(fs_info);
4009 btrfs_scrub_cancel(fs_info);
4011 /* wait for any defraggers to finish */
4012 wait_event(fs_info->transaction_wait,
4013 (atomic_read(&fs_info->defrag_running) == 0));
4015 /* clear out the rbtree of defraggable inodes */
4016 btrfs_cleanup_defrag_inodes(fs_info);
4018 cancel_work_sync(&fs_info->async_reclaim_work);
4020 if (!sb_rdonly(fs_info->sb)) {
4022 * The cleaner kthread is stopped, so do one final pass over
4023 * unused block groups.
4025 btrfs_delete_unused_bgs(fs_info);
4027 ret = btrfs_commit_super(fs_info);
4029 btrfs_err(fs_info, "commit super ret %d", ret);
4032 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4033 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4034 btrfs_error_commit_super(fs_info);
4036 kthread_stop(fs_info->transaction_kthread);
4037 kthread_stop(fs_info->cleaner_kthread);
4039 ASSERT(list_empty(&fs_info->delayed_iputs));
4040 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4042 btrfs_free_qgroup_config(fs_info);
4043 ASSERT(list_empty(&fs_info->delalloc_roots));
4045 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4046 btrfs_info(fs_info, "at unmount delalloc count %lld",
4047 percpu_counter_sum(&fs_info->delalloc_bytes));
4050 if (percpu_counter_sum(&fs_info->dio_bytes))
4051 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4052 percpu_counter_sum(&fs_info->dio_bytes));
4054 btrfs_sysfs_remove_mounted(fs_info);
4055 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4057 btrfs_free_fs_roots(fs_info);
4059 btrfs_put_block_group_cache(fs_info);
4062 * we must make sure there is not any read request to
4063 * submit after we stopping all workers.
4065 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4066 btrfs_stop_all_workers(fs_info);
4068 btrfs_free_block_groups(fs_info);
4070 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4071 free_root_pointers(fs_info, 1);
4073 iput(fs_info->btree_inode);
4075 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4076 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4077 btrfsic_unmount(fs_info->fs_devices);
4080 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4081 btrfs_close_devices(fs_info->fs_devices);
4083 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4084 percpu_counter_destroy(&fs_info->delalloc_bytes);
4085 percpu_counter_destroy(&fs_info->dio_bytes);
4086 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4087 cleanup_srcu_struct(&fs_info->subvol_srcu);
4089 btrfs_free_csum_hash(fs_info);
4090 btrfs_free_stripe_hash_table(fs_info);
4091 btrfs_free_ref_cache(fs_info);
4094 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4098 struct inode *btree_inode = buf->pages[0]->mapping->host;
4100 ret = extent_buffer_uptodate(buf);
4104 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4105 parent_transid, atomic);
4111 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4113 struct btrfs_fs_info *fs_info;
4114 struct btrfs_root *root;
4115 u64 transid = btrfs_header_generation(buf);
4118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4120 * This is a fast path so only do this check if we have sanity tests
4121 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4122 * outside of the sanity tests.
4124 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4127 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4128 fs_info = root->fs_info;
4129 btrfs_assert_tree_locked(buf);
4130 if (transid != fs_info->generation)
4131 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4132 buf->start, transid, fs_info->generation);
4133 was_dirty = set_extent_buffer_dirty(buf);
4135 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4137 fs_info->dirty_metadata_batch);
4138 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4140 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4141 * but item data not updated.
4142 * So here we should only check item pointers, not item data.
4144 if (btrfs_header_level(buf) == 0 &&
4145 btrfs_check_leaf_relaxed(buf)) {
4146 btrfs_print_leaf(buf);
4152 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4156 * looks as though older kernels can get into trouble with
4157 * this code, they end up stuck in balance_dirty_pages forever
4161 if (current->flags & PF_MEMALLOC)
4165 btrfs_balance_delayed_items(fs_info);
4167 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4168 BTRFS_DIRTY_METADATA_THRESH,
4169 fs_info->dirty_metadata_batch);
4171 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4175 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4177 __btrfs_btree_balance_dirty(fs_info, 1);
4180 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4182 __btrfs_btree_balance_dirty(fs_info, 0);
4185 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4186 struct btrfs_key *first_key)
4188 return btree_read_extent_buffer_pages(buf, parent_transid,
4192 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4194 /* cleanup FS via transaction */
4195 btrfs_cleanup_transaction(fs_info);
4197 mutex_lock(&fs_info->cleaner_mutex);
4198 btrfs_run_delayed_iputs(fs_info);
4199 mutex_unlock(&fs_info->cleaner_mutex);
4201 down_write(&fs_info->cleanup_work_sem);
4202 up_write(&fs_info->cleanup_work_sem);
4205 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4207 struct btrfs_ordered_extent *ordered;
4209 spin_lock(&root->ordered_extent_lock);
4211 * This will just short circuit the ordered completion stuff which will
4212 * make sure the ordered extent gets properly cleaned up.
4214 list_for_each_entry(ordered, &root->ordered_extents,
4216 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4217 spin_unlock(&root->ordered_extent_lock);
4220 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4222 struct btrfs_root *root;
4223 struct list_head splice;
4225 INIT_LIST_HEAD(&splice);
4227 spin_lock(&fs_info->ordered_root_lock);
4228 list_splice_init(&fs_info->ordered_roots, &splice);
4229 while (!list_empty(&splice)) {
4230 root = list_first_entry(&splice, struct btrfs_root,
4232 list_move_tail(&root->ordered_root,
4233 &fs_info->ordered_roots);
4235 spin_unlock(&fs_info->ordered_root_lock);
4236 btrfs_destroy_ordered_extents(root);
4239 spin_lock(&fs_info->ordered_root_lock);
4241 spin_unlock(&fs_info->ordered_root_lock);
4244 * We need this here because if we've been flipped read-only we won't
4245 * get sync() from the umount, so we need to make sure any ordered
4246 * extents that haven't had their dirty pages IO start writeout yet
4247 * actually get run and error out properly.
4249 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4252 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4253 struct btrfs_fs_info *fs_info)
4255 struct rb_node *node;
4256 struct btrfs_delayed_ref_root *delayed_refs;
4257 struct btrfs_delayed_ref_node *ref;
4260 delayed_refs = &trans->delayed_refs;
4262 spin_lock(&delayed_refs->lock);
4263 if (atomic_read(&delayed_refs->num_entries) == 0) {
4264 spin_unlock(&delayed_refs->lock);
4265 btrfs_info(fs_info, "delayed_refs has NO entry");
4269 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4270 struct btrfs_delayed_ref_head *head;
4272 bool pin_bytes = false;
4274 head = rb_entry(node, struct btrfs_delayed_ref_head,
4276 if (btrfs_delayed_ref_lock(delayed_refs, head))
4279 spin_lock(&head->lock);
4280 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4281 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4284 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4285 RB_CLEAR_NODE(&ref->ref_node);
4286 if (!list_empty(&ref->add_list))
4287 list_del(&ref->add_list);
4288 atomic_dec(&delayed_refs->num_entries);
4289 btrfs_put_delayed_ref(ref);
4291 if (head->must_insert_reserved)
4293 btrfs_free_delayed_extent_op(head->extent_op);
4294 btrfs_delete_ref_head(delayed_refs, head);
4295 spin_unlock(&head->lock);
4296 spin_unlock(&delayed_refs->lock);
4297 mutex_unlock(&head->mutex);
4300 btrfs_pin_extent(fs_info, head->bytenr,
4301 head->num_bytes, 1);
4302 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4303 btrfs_put_delayed_ref_head(head);
4305 spin_lock(&delayed_refs->lock);
4308 spin_unlock(&delayed_refs->lock);
4313 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4315 struct btrfs_inode *btrfs_inode;
4316 struct list_head splice;
4318 INIT_LIST_HEAD(&splice);
4320 spin_lock(&root->delalloc_lock);
4321 list_splice_init(&root->delalloc_inodes, &splice);
4323 while (!list_empty(&splice)) {
4324 struct inode *inode = NULL;
4325 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4327 __btrfs_del_delalloc_inode(root, btrfs_inode);
4328 spin_unlock(&root->delalloc_lock);
4331 * Make sure we get a live inode and that it'll not disappear
4334 inode = igrab(&btrfs_inode->vfs_inode);
4336 invalidate_inode_pages2(inode->i_mapping);
4339 spin_lock(&root->delalloc_lock);
4341 spin_unlock(&root->delalloc_lock);
4344 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4346 struct btrfs_root *root;
4347 struct list_head splice;
4349 INIT_LIST_HEAD(&splice);
4351 spin_lock(&fs_info->delalloc_root_lock);
4352 list_splice_init(&fs_info->delalloc_roots, &splice);
4353 while (!list_empty(&splice)) {
4354 root = list_first_entry(&splice, struct btrfs_root,
4356 root = btrfs_grab_fs_root(root);
4358 spin_unlock(&fs_info->delalloc_root_lock);
4360 btrfs_destroy_delalloc_inodes(root);
4361 btrfs_put_fs_root(root);
4363 spin_lock(&fs_info->delalloc_root_lock);
4365 spin_unlock(&fs_info->delalloc_root_lock);
4368 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4369 struct extent_io_tree *dirty_pages,
4373 struct extent_buffer *eb;
4378 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4383 clear_extent_bits(dirty_pages, start, end, mark);
4384 while (start <= end) {
4385 eb = find_extent_buffer(fs_info, start);
4386 start += fs_info->nodesize;
4389 wait_on_extent_buffer_writeback(eb);
4391 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4393 clear_extent_buffer_dirty(eb);
4394 free_extent_buffer_stale(eb);
4401 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4402 struct extent_io_tree *pinned_extents)
4404 struct extent_io_tree *unpin;
4410 unpin = pinned_extents;
4413 struct extent_state *cached_state = NULL;
4416 * The btrfs_finish_extent_commit() may get the same range as
4417 * ours between find_first_extent_bit and clear_extent_dirty.
4418 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4419 * the same extent range.
4421 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4422 ret = find_first_extent_bit(unpin, 0, &start, &end,
4423 EXTENT_DIRTY, &cached_state);
4425 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4429 clear_extent_dirty(unpin, start, end, &cached_state);
4430 free_extent_state(cached_state);
4431 btrfs_error_unpin_extent_range(fs_info, start, end);
4432 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4437 if (unpin == &fs_info->freed_extents[0])
4438 unpin = &fs_info->freed_extents[1];
4440 unpin = &fs_info->freed_extents[0];
4448 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4450 struct inode *inode;
4452 inode = cache->io_ctl.inode;
4454 invalidate_inode_pages2(inode->i_mapping);
4455 BTRFS_I(inode)->generation = 0;
4456 cache->io_ctl.inode = NULL;
4459 btrfs_put_block_group(cache);
4462 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4463 struct btrfs_fs_info *fs_info)
4465 struct btrfs_block_group_cache *cache;
4467 spin_lock(&cur_trans->dirty_bgs_lock);
4468 while (!list_empty(&cur_trans->dirty_bgs)) {
4469 cache = list_first_entry(&cur_trans->dirty_bgs,
4470 struct btrfs_block_group_cache,
4473 if (!list_empty(&cache->io_list)) {
4474 spin_unlock(&cur_trans->dirty_bgs_lock);
4475 list_del_init(&cache->io_list);
4476 btrfs_cleanup_bg_io(cache);
4477 spin_lock(&cur_trans->dirty_bgs_lock);
4480 list_del_init(&cache->dirty_list);
4481 spin_lock(&cache->lock);
4482 cache->disk_cache_state = BTRFS_DC_ERROR;
4483 spin_unlock(&cache->lock);
4485 spin_unlock(&cur_trans->dirty_bgs_lock);
4486 btrfs_put_block_group(cache);
4487 btrfs_delayed_refs_rsv_release(fs_info, 1);
4488 spin_lock(&cur_trans->dirty_bgs_lock);
4490 spin_unlock(&cur_trans->dirty_bgs_lock);
4493 * Refer to the definition of io_bgs member for details why it's safe
4494 * to use it without any locking
4496 while (!list_empty(&cur_trans->io_bgs)) {
4497 cache = list_first_entry(&cur_trans->io_bgs,
4498 struct btrfs_block_group_cache,
4501 list_del_init(&cache->io_list);
4502 spin_lock(&cache->lock);
4503 cache->disk_cache_state = BTRFS_DC_ERROR;
4504 spin_unlock(&cache->lock);
4505 btrfs_cleanup_bg_io(cache);
4509 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4510 struct btrfs_fs_info *fs_info)
4512 struct btrfs_device *dev, *tmp;
4514 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4515 ASSERT(list_empty(&cur_trans->dirty_bgs));
4516 ASSERT(list_empty(&cur_trans->io_bgs));
4518 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4520 list_del_init(&dev->post_commit_list);
4523 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4525 cur_trans->state = TRANS_STATE_COMMIT_START;
4526 wake_up(&fs_info->transaction_blocked_wait);
4528 cur_trans->state = TRANS_STATE_UNBLOCKED;
4529 wake_up(&fs_info->transaction_wait);
4531 btrfs_destroy_delayed_inodes(fs_info);
4532 btrfs_assert_delayed_root_empty(fs_info);
4534 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4536 btrfs_destroy_pinned_extent(fs_info,
4537 fs_info->pinned_extents);
4539 cur_trans->state =TRANS_STATE_COMPLETED;
4540 wake_up(&cur_trans->commit_wait);
4543 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4545 struct btrfs_transaction *t;
4547 mutex_lock(&fs_info->transaction_kthread_mutex);
4549 spin_lock(&fs_info->trans_lock);
4550 while (!list_empty(&fs_info->trans_list)) {
4551 t = list_first_entry(&fs_info->trans_list,
4552 struct btrfs_transaction, list);
4553 if (t->state >= TRANS_STATE_COMMIT_START) {
4554 refcount_inc(&t->use_count);
4555 spin_unlock(&fs_info->trans_lock);
4556 btrfs_wait_for_commit(fs_info, t->transid);
4557 btrfs_put_transaction(t);
4558 spin_lock(&fs_info->trans_lock);
4561 if (t == fs_info->running_transaction) {
4562 t->state = TRANS_STATE_COMMIT_DOING;
4563 spin_unlock(&fs_info->trans_lock);
4565 * We wait for 0 num_writers since we don't hold a trans
4566 * handle open currently for this transaction.
4568 wait_event(t->writer_wait,
4569 atomic_read(&t->num_writers) == 0);
4571 spin_unlock(&fs_info->trans_lock);
4573 btrfs_cleanup_one_transaction(t, fs_info);
4575 spin_lock(&fs_info->trans_lock);
4576 if (t == fs_info->running_transaction)
4577 fs_info->running_transaction = NULL;
4578 list_del_init(&t->list);
4579 spin_unlock(&fs_info->trans_lock);
4581 btrfs_put_transaction(t);
4582 trace_btrfs_transaction_commit(fs_info->tree_root);
4583 spin_lock(&fs_info->trans_lock);
4585 spin_unlock(&fs_info->trans_lock);
4586 btrfs_destroy_all_ordered_extents(fs_info);
4587 btrfs_destroy_delayed_inodes(fs_info);
4588 btrfs_assert_delayed_root_empty(fs_info);
4589 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4590 btrfs_destroy_all_delalloc_inodes(fs_info);
4591 mutex_unlock(&fs_info->transaction_kthread_mutex);
4596 static const struct extent_io_ops btree_extent_io_ops = {
4597 /* mandatory callbacks */
4598 .submit_bio_hook = btree_submit_bio_hook,
4599 .readpage_end_io_hook = btree_readpage_end_io_hook,