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>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "inode-map.h"
33 #include "check-integrity.h"
34 #include "rcu-string.h"
35 #include "dev-replace.h"
39 #include "compression.h"
40 #include "tree-checker.h"
41 #include "ref-verify.h"
44 #include <asm/cpufeature.h>
47 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
48 BTRFS_HEADER_FLAG_RELOC |\
49 BTRFS_SUPER_FLAG_ERROR |\
50 BTRFS_SUPER_FLAG_SEEDING |\
51 BTRFS_SUPER_FLAG_METADUMP |\
52 BTRFS_SUPER_FLAG_METADUMP_V2)
54 static const struct extent_io_ops btree_extent_io_ops;
55 static void end_workqueue_fn(struct btrfs_work *work);
56 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
57 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
58 struct btrfs_fs_info *fs_info);
59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
61 struct extent_io_tree *dirty_pages,
63 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
64 struct extent_io_tree *pinned_extents);
65 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
66 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
69 * btrfs_end_io_wq structs are used to do processing in task context when an IO
70 * is complete. This is used during reads to verify checksums, and it is used
71 * by writes to insert metadata for new file extents after IO is complete.
73 struct btrfs_end_io_wq {
77 struct btrfs_fs_info *info;
79 enum btrfs_wq_endio_type metadata;
80 struct btrfs_work work;
83 static struct kmem_cache *btrfs_end_io_wq_cache;
85 int __init btrfs_end_io_wq_init(void)
87 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
88 sizeof(struct btrfs_end_io_wq),
92 if (!btrfs_end_io_wq_cache)
97 void __cold btrfs_end_io_wq_exit(void)
99 kmem_cache_destroy(btrfs_end_io_wq_cache);
103 * async submit bios are used to offload expensive checksumming
104 * onto the worker threads. They checksum file and metadata bios
105 * just before they are sent down the IO stack.
107 struct async_submit_bio {
110 extent_submit_bio_start_t *submit_bio_start;
113 * bio_offset is optional, can be used if the pages in the bio
114 * can't tell us where in the file the bio should go
117 struct btrfs_work work;
122 * Lockdep class keys for extent_buffer->lock's in this root. For a given
123 * eb, the lockdep key is determined by the btrfs_root it belongs to and
124 * the level the eb occupies in the tree.
126 * Different roots are used for different purposes and may nest inside each
127 * other and they require separate keysets. As lockdep keys should be
128 * static, assign keysets according to the purpose of the root as indicated
129 * by btrfs_root->root_key.objectid. This ensures that all special purpose
130 * roots have separate keysets.
132 * Lock-nesting across peer nodes is always done with the immediate parent
133 * node locked thus preventing deadlock. As lockdep doesn't know this, use
134 * subclass to avoid triggering lockdep warning in such cases.
136 * The key is set by the readpage_end_io_hook after the buffer has passed
137 * csum validation but before the pages are unlocked. It is also set by
138 * btrfs_init_new_buffer on freshly allocated blocks.
140 * We also add a check to make sure the highest level of the tree is the
141 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
142 * needs update as well.
144 #ifdef CONFIG_DEBUG_LOCK_ALLOC
145 # if BTRFS_MAX_LEVEL != 8
149 static struct btrfs_lockdep_keyset {
150 u64 id; /* root objectid */
151 const char *name_stem; /* lock name stem */
152 char names[BTRFS_MAX_LEVEL + 1][20];
153 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
154 } btrfs_lockdep_keysets[] = {
155 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
156 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
157 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
158 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
159 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
160 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
161 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
162 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
163 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
164 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
165 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
166 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
167 { .id = 0, .name_stem = "tree" },
170 void __init btrfs_init_lockdep(void)
174 /* initialize lockdep class names */
175 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
176 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
178 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
179 snprintf(ks->names[j], sizeof(ks->names[j]),
180 "btrfs-%s-%02d", ks->name_stem, j);
184 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
187 struct btrfs_lockdep_keyset *ks;
189 BUG_ON(level >= ARRAY_SIZE(ks->keys));
191 /* find the matching keyset, id 0 is the default entry */
192 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
193 if (ks->id == objectid)
196 lockdep_set_class_and_name(&eb->lock,
197 &ks->keys[level], ks->names[level]);
203 * extents on the btree inode are pretty simple, there's one extent
204 * that covers the entire device
206 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
207 struct page *page, size_t pg_offset, u64 start, u64 len,
210 struct btrfs_fs_info *fs_info = inode->root->fs_info;
211 struct extent_map_tree *em_tree = &inode->extent_tree;
212 struct extent_map *em;
215 read_lock(&em_tree->lock);
216 em = lookup_extent_mapping(em_tree, start, len);
218 em->bdev = fs_info->fs_devices->latest_bdev;
219 read_unlock(&em_tree->lock);
222 read_unlock(&em_tree->lock);
224 em = alloc_extent_map();
226 em = ERR_PTR(-ENOMEM);
231 em->block_len = (u64)-1;
233 em->bdev = fs_info->fs_devices->latest_bdev;
235 write_lock(&em_tree->lock);
236 ret = add_extent_mapping(em_tree, em, 0);
237 if (ret == -EEXIST) {
239 em = lookup_extent_mapping(em_tree, start, len);
246 write_unlock(&em_tree->lock);
252 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
254 return crc32c(seed, data, len);
257 void btrfs_csum_final(u32 crc, u8 *result)
259 put_unaligned_le32(~crc, result);
263 * Compute the csum of a btree block and store the result to provided buffer.
265 * Returns error if the extent buffer cannot be mapped.
267 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
270 unsigned long cur_len;
271 unsigned long offset = BTRFS_CSUM_SIZE;
273 unsigned long map_start;
274 unsigned long map_len;
278 len = buf->len - offset;
281 * Note: we don't need to check for the err == 1 case here, as
282 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
283 * and 'min_len = 32' and the currently implemented mapping
284 * algorithm we cannot cross a page boundary.
286 err = map_private_extent_buffer(buf, offset, 32,
287 &kaddr, &map_start, &map_len);
290 cur_len = min(len, map_len - (offset - map_start));
291 crc = btrfs_csum_data(kaddr + offset - map_start,
296 memset(result, 0, BTRFS_CSUM_SIZE);
298 btrfs_csum_final(crc, result);
304 * we can't consider a given block up to date unless the transid of the
305 * block matches the transid in the parent node's pointer. This is how we
306 * detect blocks that either didn't get written at all or got written
307 * in the wrong place.
309 static int verify_parent_transid(struct extent_io_tree *io_tree,
310 struct extent_buffer *eb, u64 parent_transid,
313 struct extent_state *cached_state = NULL;
315 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
317 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
324 btrfs_tree_read_lock(eb);
325 btrfs_set_lock_blocking_read(eb);
328 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
330 if (extent_buffer_uptodate(eb) &&
331 btrfs_header_generation(eb) == parent_transid) {
335 btrfs_err_rl(eb->fs_info,
336 "parent transid verify failed on %llu wanted %llu found %llu",
338 parent_transid, btrfs_header_generation(eb));
342 * Things reading via commit roots that don't have normal protection,
343 * like send, can have a really old block in cache that may point at a
344 * block that has been freed and re-allocated. So don't clear uptodate
345 * if we find an eb that is under IO (dirty/writeback) because we could
346 * end up reading in the stale data and then writing it back out and
347 * making everybody very sad.
349 if (!extent_buffer_under_io(eb))
350 clear_extent_buffer_uptodate(eb);
352 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 btrfs_tree_read_unlock_blocking(eb);
360 * Return 0 if the superblock checksum type matches the checksum value of that
361 * algorithm. Pass the raw disk superblock data.
363 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
366 struct btrfs_super_block *disk_sb =
367 (struct btrfs_super_block *)raw_disk_sb;
368 u16 csum_type = btrfs_super_csum_type(disk_sb);
371 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
373 char result[sizeof(crc)];
376 * The super_block structure does not span the whole
377 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378 * is filled with zeros and is included in the checksum.
380 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
381 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382 btrfs_csum_final(crc, result);
384 if (memcmp(raw_disk_sb, result, sizeof(result)))
388 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
389 btrfs_err(fs_info, "unsupported checksum algorithm %u",
397 int btrfs_verify_level_key(struct btrfs_fs_info *fs_info,
398 struct extent_buffer *eb, int level,
399 struct btrfs_key *first_key, u64 parent_transid)
402 struct btrfs_key found_key;
405 found_level = btrfs_header_level(eb);
406 if (found_level != level) {
407 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
408 KERN_ERR "BTRFS: tree level check failed\n");
410 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
411 eb->start, level, found_level);
419 * For live tree block (new tree blocks in current transaction),
420 * we need proper lock context to avoid race, which is impossible here.
421 * So we only checks tree blocks which is read from disk, whose
422 * generation <= fs_info->last_trans_committed.
424 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
427 btrfs_node_key_to_cpu(eb, &found_key, 0);
429 btrfs_item_key_to_cpu(eb, &found_key, 0);
430 ret = btrfs_comp_cpu_keys(first_key, &found_key);
433 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
434 KERN_ERR "BTRFS: tree first key check failed\n");
436 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
437 eb->start, parent_transid, first_key->objectid,
438 first_key->type, first_key->offset,
439 found_key.objectid, found_key.type,
446 * helper to read a given tree block, doing retries as required when
447 * the checksums don't match and we have alternate mirrors to try.
449 * @parent_transid: expected transid, skip check if 0
450 * @level: expected level, mandatory check
451 * @first_key: expected key of first slot, skip check if NULL
453 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
454 struct extent_buffer *eb,
455 u64 parent_transid, int level,
456 struct btrfs_key *first_key)
458 struct extent_io_tree *io_tree;
463 int failed_mirror = 0;
465 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
467 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
468 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
471 if (verify_parent_transid(io_tree, eb,
474 else if (btrfs_verify_level_key(fs_info, eb, level,
475 first_key, parent_transid))
481 num_copies = btrfs_num_copies(fs_info,
486 if (!failed_mirror) {
488 failed_mirror = eb->read_mirror;
492 if (mirror_num == failed_mirror)
495 if (mirror_num > num_copies)
499 if (failed && !ret && failed_mirror)
500 btrfs_repair_eb_io_failure(eb, failed_mirror);
506 * checksum a dirty tree block before IO. This has extra checks to make sure
507 * we only fill in the checksum field in the first page of a multi-page block
510 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
512 u64 start = page_offset(page);
514 u8 result[BTRFS_CSUM_SIZE];
515 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
516 struct extent_buffer *eb;
518 eb = (struct extent_buffer *)page->private;
519 if (page != eb->pages[0])
522 found_start = btrfs_header_bytenr(eb);
524 * Please do not consolidate these warnings into a single if.
525 * It is useful to know what went wrong.
527 if (WARN_ON(found_start != start))
529 if (WARN_ON(!PageUptodate(page)))
532 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
533 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
535 if (csum_tree_block(eb, result))
538 write_extent_buffer(eb, result, 0, csum_size);
542 static int check_tree_block_fsid(struct extent_buffer *eb)
544 struct btrfs_fs_info *fs_info = eb->fs_info;
545 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
546 u8 fsid[BTRFS_FSID_SIZE];
549 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
554 * Checking the incompat flag is only valid for the current
555 * fs. For seed devices it's forbidden to have their uuid
556 * changed so reading ->fsid in this case is fine
558 if (fs_devices == fs_info->fs_devices &&
559 btrfs_fs_incompat(fs_info, METADATA_UUID))
560 metadata_uuid = fs_devices->metadata_uuid;
562 metadata_uuid = fs_devices->fsid;
564 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
568 fs_devices = fs_devices->seed;
573 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
574 u64 phy_offset, struct page *page,
575 u64 start, u64 end, int mirror)
579 struct extent_buffer *eb;
580 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
581 struct btrfs_fs_info *fs_info = root->fs_info;
582 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
584 u8 result[BTRFS_CSUM_SIZE];
590 eb = (struct extent_buffer *)page->private;
592 /* the pending IO might have been the only thing that kept this buffer
593 * in memory. Make sure we have a ref for all this other checks
595 extent_buffer_get(eb);
597 reads_done = atomic_dec_and_test(&eb->io_pages);
601 eb->read_mirror = mirror;
602 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
607 found_start = btrfs_header_bytenr(eb);
608 if (found_start != eb->start) {
609 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
610 eb->start, found_start);
614 if (check_tree_block_fsid(eb)) {
615 btrfs_err_rl(fs_info, "bad fsid on block %llu",
620 found_level = btrfs_header_level(eb);
621 if (found_level >= BTRFS_MAX_LEVEL) {
622 btrfs_err(fs_info, "bad tree block level %d on %llu",
623 (int)btrfs_header_level(eb), eb->start);
628 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
631 ret = csum_tree_block(eb, result);
635 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
639 memcpy(&found, result, csum_size);
641 read_extent_buffer(eb, &val, 0, csum_size);
642 btrfs_warn_rl(fs_info,
643 "%s checksum verify failed on %llu wanted %x found %x level %d",
644 fs_info->sb->s_id, eb->start,
645 val, found, btrfs_header_level(eb));
651 * If this is a leaf block and it is corrupt, set the corrupt bit so
652 * that we don't try and read the other copies of this block, just
655 if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
656 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
660 if (found_level > 0 && btrfs_check_node(fs_info, eb))
664 set_extent_buffer_uptodate(eb);
667 "block=%llu read time tree block corruption detected",
671 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
672 btree_readahead_hook(eb, ret);
676 * our io error hook is going to dec the io pages
677 * again, we have to make sure it has something
680 atomic_inc(&eb->io_pages);
681 clear_extent_buffer_uptodate(eb);
683 free_extent_buffer(eb);
688 static void end_workqueue_bio(struct bio *bio)
690 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
691 struct btrfs_fs_info *fs_info;
692 struct btrfs_workqueue *wq;
693 btrfs_work_func_t func;
695 fs_info = end_io_wq->info;
696 end_io_wq->status = bio->bi_status;
698 if (bio_op(bio) == REQ_OP_WRITE) {
699 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
700 wq = fs_info->endio_meta_write_workers;
701 func = btrfs_endio_meta_write_helper;
702 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
703 wq = fs_info->endio_freespace_worker;
704 func = btrfs_freespace_write_helper;
705 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
706 wq = fs_info->endio_raid56_workers;
707 func = btrfs_endio_raid56_helper;
709 wq = fs_info->endio_write_workers;
710 func = btrfs_endio_write_helper;
713 if (unlikely(end_io_wq->metadata ==
714 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
715 wq = fs_info->endio_repair_workers;
716 func = btrfs_endio_repair_helper;
717 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
718 wq = fs_info->endio_raid56_workers;
719 func = btrfs_endio_raid56_helper;
720 } else if (end_io_wq->metadata) {
721 wq = fs_info->endio_meta_workers;
722 func = btrfs_endio_meta_helper;
724 wq = fs_info->endio_workers;
725 func = btrfs_endio_helper;
729 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
730 btrfs_queue_work(wq, &end_io_wq->work);
733 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
734 enum btrfs_wq_endio_type metadata)
736 struct btrfs_end_io_wq *end_io_wq;
738 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
740 return BLK_STS_RESOURCE;
742 end_io_wq->private = bio->bi_private;
743 end_io_wq->end_io = bio->bi_end_io;
744 end_io_wq->info = info;
745 end_io_wq->status = 0;
746 end_io_wq->bio = bio;
747 end_io_wq->metadata = metadata;
749 bio->bi_private = end_io_wq;
750 bio->bi_end_io = end_workqueue_bio;
754 static void run_one_async_start(struct btrfs_work *work)
756 struct async_submit_bio *async;
759 async = container_of(work, struct async_submit_bio, work);
760 ret = async->submit_bio_start(async->private_data, async->bio,
767 * In order to insert checksums into the metadata in large chunks, we wait
768 * until bio submission time. All the pages in the bio are checksummed and
769 * sums are attached onto the ordered extent record.
771 * At IO completion time the csums attached on the ordered extent record are
772 * inserted into the tree.
774 static void run_one_async_done(struct btrfs_work *work)
776 struct async_submit_bio *async;
780 async = container_of(work, struct async_submit_bio, work);
781 inode = async->private_data;
783 /* If an error occurred we just want to clean up the bio and move on */
785 async->bio->bi_status = async->status;
786 bio_endio(async->bio);
790 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
791 async->mirror_num, 1);
793 async->bio->bi_status = ret;
794 bio_endio(async->bio);
798 static void run_one_async_free(struct btrfs_work *work)
800 struct async_submit_bio *async;
802 async = container_of(work, struct async_submit_bio, work);
806 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
807 int mirror_num, unsigned long bio_flags,
808 u64 bio_offset, void *private_data,
809 extent_submit_bio_start_t *submit_bio_start)
811 struct async_submit_bio *async;
813 async = kmalloc(sizeof(*async), GFP_NOFS);
815 return BLK_STS_RESOURCE;
817 async->private_data = private_data;
819 async->mirror_num = mirror_num;
820 async->submit_bio_start = submit_bio_start;
822 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
823 run_one_async_done, run_one_async_free);
825 async->bio_offset = bio_offset;
829 if (op_is_sync(bio->bi_opf))
830 btrfs_set_work_high_priority(&async->work);
832 btrfs_queue_work(fs_info->workers, &async->work);
836 static blk_status_t btree_csum_one_bio(struct bio *bio)
838 struct bio_vec *bvec;
839 struct btrfs_root *root;
841 struct bvec_iter_all iter_all;
843 ASSERT(!bio_flagged(bio, BIO_CLONED));
844 bio_for_each_segment_all(bvec, bio, i, iter_all) {
845 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
846 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
851 return errno_to_blk_status(ret);
854 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
858 * when we're called for a write, we're already in the async
859 * submission context. Just jump into btrfs_map_bio
861 return btree_csum_one_bio(bio);
864 static int check_async_write(struct btrfs_inode *bi)
866 if (atomic_read(&bi->sync_writers))
869 if (static_cpu_has(X86_FEATURE_XMM4_2))
875 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
876 int mirror_num, unsigned long bio_flags,
879 struct inode *inode = private_data;
880 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
881 int async = check_async_write(BTRFS_I(inode));
884 if (bio_op(bio) != REQ_OP_WRITE) {
886 * called for a read, do the setup so that checksum validation
887 * can happen in the async kernel threads
889 ret = btrfs_bio_wq_end_io(fs_info, bio,
890 BTRFS_WQ_ENDIO_METADATA);
893 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
895 ret = btree_csum_one_bio(bio);
898 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
901 * kthread helpers are used to submit writes so that
902 * checksumming can happen in parallel across all CPUs
904 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
905 bio_offset, private_data,
906 btree_submit_bio_start);
914 bio->bi_status = ret;
919 #ifdef CONFIG_MIGRATION
920 static int btree_migratepage(struct address_space *mapping,
921 struct page *newpage, struct page *page,
922 enum migrate_mode mode)
925 * we can't safely write a btree page from here,
926 * we haven't done the locking hook
931 * Buffers may be managed in a filesystem specific way.
932 * We must have no buffers or drop them.
934 if (page_has_private(page) &&
935 !try_to_release_page(page, GFP_KERNEL))
937 return migrate_page(mapping, newpage, page, mode);
942 static int btree_writepages(struct address_space *mapping,
943 struct writeback_control *wbc)
945 struct btrfs_fs_info *fs_info;
948 if (wbc->sync_mode == WB_SYNC_NONE) {
950 if (wbc->for_kupdate)
953 fs_info = BTRFS_I(mapping->host)->root->fs_info;
954 /* this is a bit racy, but that's ok */
955 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
956 BTRFS_DIRTY_METADATA_THRESH,
957 fs_info->dirty_metadata_batch);
961 return btree_write_cache_pages(mapping, wbc);
964 static int btree_readpage(struct file *file, struct page *page)
966 struct extent_io_tree *tree;
967 tree = &BTRFS_I(page->mapping->host)->io_tree;
968 return extent_read_full_page(tree, page, btree_get_extent, 0);
971 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
973 if (PageWriteback(page) || PageDirty(page))
976 return try_release_extent_buffer(page);
979 static void btree_invalidatepage(struct page *page, unsigned int offset,
982 struct extent_io_tree *tree;
983 tree = &BTRFS_I(page->mapping->host)->io_tree;
984 extent_invalidatepage(tree, page, offset);
985 btree_releasepage(page, GFP_NOFS);
986 if (PagePrivate(page)) {
987 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
988 "page private not zero on page %llu",
989 (unsigned long long)page_offset(page));
990 ClearPagePrivate(page);
991 set_page_private(page, 0);
996 static int btree_set_page_dirty(struct page *page)
999 struct extent_buffer *eb;
1001 BUG_ON(!PagePrivate(page));
1002 eb = (struct extent_buffer *)page->private;
1004 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1005 BUG_ON(!atomic_read(&eb->refs));
1006 btrfs_assert_tree_locked(eb);
1008 return __set_page_dirty_nobuffers(page);
1011 static const struct address_space_operations btree_aops = {
1012 .readpage = btree_readpage,
1013 .writepages = btree_writepages,
1014 .releasepage = btree_releasepage,
1015 .invalidatepage = btree_invalidatepage,
1016 #ifdef CONFIG_MIGRATION
1017 .migratepage = btree_migratepage,
1019 .set_page_dirty = btree_set_page_dirty,
1022 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1024 struct extent_buffer *buf = NULL;
1025 struct inode *btree_inode = fs_info->btree_inode;
1028 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1032 ret = read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf,
1035 free_extent_buffer_stale(buf);
1037 free_extent_buffer(buf);
1040 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1041 int mirror_num, struct extent_buffer **eb)
1043 struct extent_buffer *buf = NULL;
1044 struct inode *btree_inode = fs_info->btree_inode;
1045 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1048 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1052 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1054 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1057 free_extent_buffer_stale(buf);
1061 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1062 free_extent_buffer_stale(buf);
1064 } else if (extent_buffer_uptodate(buf)) {
1067 free_extent_buffer(buf);
1072 struct extent_buffer *btrfs_find_create_tree_block(
1073 struct btrfs_fs_info *fs_info,
1076 if (btrfs_is_testing(fs_info))
1077 return alloc_test_extent_buffer(fs_info, bytenr);
1078 return alloc_extent_buffer(fs_info, bytenr);
1082 * Read tree block at logical address @bytenr and do variant basic but critical
1085 * @parent_transid: expected transid of this tree block, skip check if 0
1086 * @level: expected level, mandatory check
1087 * @first_key: expected key in slot 0, skip check if NULL
1089 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1090 u64 parent_transid, int level,
1091 struct btrfs_key *first_key)
1093 struct extent_buffer *buf = NULL;
1096 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1100 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1103 free_extent_buffer_stale(buf);
1104 return ERR_PTR(ret);
1110 void clean_tree_block(struct btrfs_fs_info *fs_info,
1111 struct extent_buffer *buf)
1113 if (btrfs_header_generation(buf) ==
1114 fs_info->running_transaction->transid) {
1115 btrfs_assert_tree_locked(buf);
1117 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1118 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1120 fs_info->dirty_metadata_batch);
1121 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1122 btrfs_set_lock_blocking_write(buf);
1123 clear_extent_buffer_dirty(buf);
1128 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1130 struct btrfs_subvolume_writers *writers;
1133 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1135 return ERR_PTR(-ENOMEM);
1137 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1140 return ERR_PTR(ret);
1143 init_waitqueue_head(&writers->wait);
1148 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1150 percpu_counter_destroy(&writers->counter);
1154 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1157 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1159 root->commit_root = NULL;
1161 root->orphan_cleanup_state = 0;
1163 root->last_trans = 0;
1164 root->highest_objectid = 0;
1165 root->nr_delalloc_inodes = 0;
1166 root->nr_ordered_extents = 0;
1167 root->inode_tree = RB_ROOT;
1168 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1169 root->block_rsv = NULL;
1171 INIT_LIST_HEAD(&root->dirty_list);
1172 INIT_LIST_HEAD(&root->root_list);
1173 INIT_LIST_HEAD(&root->delalloc_inodes);
1174 INIT_LIST_HEAD(&root->delalloc_root);
1175 INIT_LIST_HEAD(&root->ordered_extents);
1176 INIT_LIST_HEAD(&root->ordered_root);
1177 INIT_LIST_HEAD(&root->reloc_dirty_list);
1178 INIT_LIST_HEAD(&root->logged_list[0]);
1179 INIT_LIST_HEAD(&root->logged_list[1]);
1180 spin_lock_init(&root->inode_lock);
1181 spin_lock_init(&root->delalloc_lock);
1182 spin_lock_init(&root->ordered_extent_lock);
1183 spin_lock_init(&root->accounting_lock);
1184 spin_lock_init(&root->log_extents_lock[0]);
1185 spin_lock_init(&root->log_extents_lock[1]);
1186 spin_lock_init(&root->qgroup_meta_rsv_lock);
1187 mutex_init(&root->objectid_mutex);
1188 mutex_init(&root->log_mutex);
1189 mutex_init(&root->ordered_extent_mutex);
1190 mutex_init(&root->delalloc_mutex);
1191 init_waitqueue_head(&root->log_writer_wait);
1192 init_waitqueue_head(&root->log_commit_wait[0]);
1193 init_waitqueue_head(&root->log_commit_wait[1]);
1194 INIT_LIST_HEAD(&root->log_ctxs[0]);
1195 INIT_LIST_HEAD(&root->log_ctxs[1]);
1196 atomic_set(&root->log_commit[0], 0);
1197 atomic_set(&root->log_commit[1], 0);
1198 atomic_set(&root->log_writers, 0);
1199 atomic_set(&root->log_batch, 0);
1200 refcount_set(&root->refs, 1);
1201 atomic_set(&root->will_be_snapshotted, 0);
1202 atomic_set(&root->snapshot_force_cow, 0);
1203 atomic_set(&root->nr_swapfiles, 0);
1204 root->log_transid = 0;
1205 root->log_transid_committed = -1;
1206 root->last_log_commit = 0;
1208 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1209 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1211 memset(&root->root_key, 0, sizeof(root->root_key));
1212 memset(&root->root_item, 0, sizeof(root->root_item));
1213 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1215 root->defrag_trans_start = fs_info->generation;
1217 root->defrag_trans_start = 0;
1218 root->root_key.objectid = objectid;
1221 spin_lock_init(&root->root_item_lock);
1222 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1225 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1228 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1230 root->fs_info = fs_info;
1234 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1235 /* Should only be used by the testing infrastructure */
1236 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1238 struct btrfs_root *root;
1241 return ERR_PTR(-EINVAL);
1243 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1245 return ERR_PTR(-ENOMEM);
1247 /* We don't use the stripesize in selftest, set it as sectorsize */
1248 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1249 root->alloc_bytenr = 0;
1255 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1256 struct btrfs_fs_info *fs_info,
1259 struct extent_buffer *leaf;
1260 struct btrfs_root *tree_root = fs_info->tree_root;
1261 struct btrfs_root *root;
1262 struct btrfs_key key;
1263 unsigned int nofs_flag;
1265 uuid_le uuid = NULL_UUID_LE;
1268 * We're holding a transaction handle, so use a NOFS memory allocation
1269 * context to avoid deadlock if reclaim happens.
1271 nofs_flag = memalloc_nofs_save();
1272 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1273 memalloc_nofs_restore(nofs_flag);
1275 return ERR_PTR(-ENOMEM);
1277 __setup_root(root, fs_info, objectid);
1278 root->root_key.objectid = objectid;
1279 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1280 root->root_key.offset = 0;
1282 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1284 ret = PTR_ERR(leaf);
1290 btrfs_mark_buffer_dirty(leaf);
1292 root->commit_root = btrfs_root_node(root);
1293 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1295 root->root_item.flags = 0;
1296 root->root_item.byte_limit = 0;
1297 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1298 btrfs_set_root_generation(&root->root_item, trans->transid);
1299 btrfs_set_root_level(&root->root_item, 0);
1300 btrfs_set_root_refs(&root->root_item, 1);
1301 btrfs_set_root_used(&root->root_item, leaf->len);
1302 btrfs_set_root_last_snapshot(&root->root_item, 0);
1303 btrfs_set_root_dirid(&root->root_item, 0);
1304 if (is_fstree(objectid))
1306 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1307 root->root_item.drop_level = 0;
1309 key.objectid = objectid;
1310 key.type = BTRFS_ROOT_ITEM_KEY;
1312 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1316 btrfs_tree_unlock(leaf);
1322 btrfs_tree_unlock(leaf);
1323 free_extent_buffer(root->commit_root);
1324 free_extent_buffer(leaf);
1328 return ERR_PTR(ret);
1331 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1332 struct btrfs_fs_info *fs_info)
1334 struct btrfs_root *root;
1335 struct extent_buffer *leaf;
1337 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1339 return ERR_PTR(-ENOMEM);
1341 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1343 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1344 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1345 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1348 * DON'T set REF_COWS for log trees
1350 * log trees do not get reference counted because they go away
1351 * before a real commit is actually done. They do store pointers
1352 * to file data extents, and those reference counts still get
1353 * updated (along with back refs to the log tree).
1356 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1360 return ERR_CAST(leaf);
1365 btrfs_mark_buffer_dirty(root->node);
1366 btrfs_tree_unlock(root->node);
1370 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1371 struct btrfs_fs_info *fs_info)
1373 struct btrfs_root *log_root;
1375 log_root = alloc_log_tree(trans, fs_info);
1376 if (IS_ERR(log_root))
1377 return PTR_ERR(log_root);
1378 WARN_ON(fs_info->log_root_tree);
1379 fs_info->log_root_tree = log_root;
1383 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1384 struct btrfs_root *root)
1386 struct btrfs_fs_info *fs_info = root->fs_info;
1387 struct btrfs_root *log_root;
1388 struct btrfs_inode_item *inode_item;
1390 log_root = alloc_log_tree(trans, fs_info);
1391 if (IS_ERR(log_root))
1392 return PTR_ERR(log_root);
1394 log_root->last_trans = trans->transid;
1395 log_root->root_key.offset = root->root_key.objectid;
1397 inode_item = &log_root->root_item.inode;
1398 btrfs_set_stack_inode_generation(inode_item, 1);
1399 btrfs_set_stack_inode_size(inode_item, 3);
1400 btrfs_set_stack_inode_nlink(inode_item, 1);
1401 btrfs_set_stack_inode_nbytes(inode_item,
1403 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1405 btrfs_set_root_node(&log_root->root_item, log_root->node);
1407 WARN_ON(root->log_root);
1408 root->log_root = log_root;
1409 root->log_transid = 0;
1410 root->log_transid_committed = -1;
1411 root->last_log_commit = 0;
1415 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1416 struct btrfs_key *key)
1418 struct btrfs_root *root;
1419 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1420 struct btrfs_path *path;
1425 path = btrfs_alloc_path();
1427 return ERR_PTR(-ENOMEM);
1429 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1435 __setup_root(root, fs_info, key->objectid);
1437 ret = btrfs_find_root(tree_root, key, path,
1438 &root->root_item, &root->root_key);
1445 generation = btrfs_root_generation(&root->root_item);
1446 level = btrfs_root_level(&root->root_item);
1447 root->node = read_tree_block(fs_info,
1448 btrfs_root_bytenr(&root->root_item),
1449 generation, level, NULL);
1450 if (IS_ERR(root->node)) {
1451 ret = PTR_ERR(root->node);
1453 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1455 free_extent_buffer(root->node);
1458 root->commit_root = btrfs_root_node(root);
1460 btrfs_free_path(path);
1466 root = ERR_PTR(ret);
1470 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1471 struct btrfs_key *location)
1473 struct btrfs_root *root;
1475 root = btrfs_read_tree_root(tree_root, location);
1479 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1480 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1481 btrfs_check_and_init_root_item(&root->root_item);
1487 int btrfs_init_fs_root(struct btrfs_root *root)
1490 struct btrfs_subvolume_writers *writers;
1492 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1493 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1495 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1500 writers = btrfs_alloc_subvolume_writers();
1501 if (IS_ERR(writers)) {
1502 ret = PTR_ERR(writers);
1505 root->subv_writers = writers;
1507 btrfs_init_free_ino_ctl(root);
1508 spin_lock_init(&root->ino_cache_lock);
1509 init_waitqueue_head(&root->ino_cache_wait);
1511 ret = get_anon_bdev(&root->anon_dev);
1515 mutex_lock(&root->objectid_mutex);
1516 ret = btrfs_find_highest_objectid(root,
1517 &root->highest_objectid);
1519 mutex_unlock(&root->objectid_mutex);
1523 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1525 mutex_unlock(&root->objectid_mutex);
1529 /* The caller is responsible to call btrfs_free_fs_root */
1533 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1536 struct btrfs_root *root;
1538 spin_lock(&fs_info->fs_roots_radix_lock);
1539 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1540 (unsigned long)root_id);
1541 spin_unlock(&fs_info->fs_roots_radix_lock);
1545 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1546 struct btrfs_root *root)
1550 ret = radix_tree_preload(GFP_NOFS);
1554 spin_lock(&fs_info->fs_roots_radix_lock);
1555 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1556 (unsigned long)root->root_key.objectid,
1559 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1560 spin_unlock(&fs_info->fs_roots_radix_lock);
1561 radix_tree_preload_end();
1566 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1567 struct btrfs_key *location,
1570 struct btrfs_root *root;
1571 struct btrfs_path *path;
1572 struct btrfs_key key;
1575 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1576 return fs_info->tree_root;
1577 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1578 return fs_info->extent_root;
1579 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1580 return fs_info->chunk_root;
1581 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1582 return fs_info->dev_root;
1583 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1584 return fs_info->csum_root;
1585 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1586 return fs_info->quota_root ? fs_info->quota_root :
1588 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1589 return fs_info->uuid_root ? fs_info->uuid_root :
1591 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1592 return fs_info->free_space_root ? fs_info->free_space_root :
1595 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1597 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1598 return ERR_PTR(-ENOENT);
1602 root = btrfs_read_fs_root(fs_info->tree_root, location);
1606 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1611 ret = btrfs_init_fs_root(root);
1615 path = btrfs_alloc_path();
1620 key.objectid = BTRFS_ORPHAN_OBJECTID;
1621 key.type = BTRFS_ORPHAN_ITEM_KEY;
1622 key.offset = location->objectid;
1624 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1625 btrfs_free_path(path);
1629 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1631 ret = btrfs_insert_fs_root(fs_info, root);
1633 if (ret == -EEXIST) {
1634 btrfs_free_fs_root(root);
1641 btrfs_free_fs_root(root);
1642 return ERR_PTR(ret);
1645 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1647 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1649 struct btrfs_device *device;
1650 struct backing_dev_info *bdi;
1653 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1656 bdi = device->bdev->bd_bdi;
1657 if (bdi_congested(bdi, bdi_bits)) {
1667 * called by the kthread helper functions to finally call the bio end_io
1668 * functions. This is where read checksum verification actually happens
1670 static void end_workqueue_fn(struct btrfs_work *work)
1673 struct btrfs_end_io_wq *end_io_wq;
1675 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1676 bio = end_io_wq->bio;
1678 bio->bi_status = end_io_wq->status;
1679 bio->bi_private = end_io_wq->private;
1680 bio->bi_end_io = end_io_wq->end_io;
1681 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1685 static int cleaner_kthread(void *arg)
1687 struct btrfs_root *root = arg;
1688 struct btrfs_fs_info *fs_info = root->fs_info;
1694 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1696 /* Make the cleaner go to sleep early. */
1697 if (btrfs_need_cleaner_sleep(fs_info))
1701 * Do not do anything if we might cause open_ctree() to block
1702 * before we have finished mounting the filesystem.
1704 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1707 if (!mutex_trylock(&fs_info->cleaner_mutex))
1711 * Avoid the problem that we change the status of the fs
1712 * during the above check and trylock.
1714 if (btrfs_need_cleaner_sleep(fs_info)) {
1715 mutex_unlock(&fs_info->cleaner_mutex);
1719 btrfs_run_delayed_iputs(fs_info);
1721 again = btrfs_clean_one_deleted_snapshot(root);
1722 mutex_unlock(&fs_info->cleaner_mutex);
1725 * The defragger has dealt with the R/O remount and umount,
1726 * needn't do anything special here.
1728 btrfs_run_defrag_inodes(fs_info);
1731 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1732 * with relocation (btrfs_relocate_chunk) and relocation
1733 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1734 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1735 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1736 * unused block groups.
1738 btrfs_delete_unused_bgs(fs_info);
1740 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1741 if (kthread_should_park())
1743 if (kthread_should_stop())
1746 set_current_state(TASK_INTERRUPTIBLE);
1748 __set_current_state(TASK_RUNNING);
1753 static int transaction_kthread(void *arg)
1755 struct btrfs_root *root = arg;
1756 struct btrfs_fs_info *fs_info = root->fs_info;
1757 struct btrfs_trans_handle *trans;
1758 struct btrfs_transaction *cur;
1761 unsigned long delay;
1765 cannot_commit = false;
1766 delay = HZ * fs_info->commit_interval;
1767 mutex_lock(&fs_info->transaction_kthread_mutex);
1769 spin_lock(&fs_info->trans_lock);
1770 cur = fs_info->running_transaction;
1772 spin_unlock(&fs_info->trans_lock);
1776 now = ktime_get_seconds();
1777 if (cur->state < TRANS_STATE_BLOCKED &&
1778 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1779 (now < cur->start_time ||
1780 now - cur->start_time < fs_info->commit_interval)) {
1781 spin_unlock(&fs_info->trans_lock);
1785 transid = cur->transid;
1786 spin_unlock(&fs_info->trans_lock);
1788 /* If the file system is aborted, this will always fail. */
1789 trans = btrfs_attach_transaction(root);
1790 if (IS_ERR(trans)) {
1791 if (PTR_ERR(trans) != -ENOENT)
1792 cannot_commit = true;
1795 if (transid == trans->transid) {
1796 btrfs_commit_transaction(trans);
1798 btrfs_end_transaction(trans);
1801 wake_up_process(fs_info->cleaner_kthread);
1802 mutex_unlock(&fs_info->transaction_kthread_mutex);
1804 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1805 &fs_info->fs_state)))
1806 btrfs_cleanup_transaction(fs_info);
1807 if (!kthread_should_stop() &&
1808 (!btrfs_transaction_blocked(fs_info) ||
1810 schedule_timeout_interruptible(delay);
1811 } while (!kthread_should_stop());
1816 * this will find the highest generation in the array of
1817 * root backups. The index of the highest array is returned,
1818 * or -1 if we can't find anything.
1820 * We check to make sure the array is valid by comparing the
1821 * generation of the latest root in the array with the generation
1822 * in the super block. If they don't match we pitch it.
1824 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1827 int newest_index = -1;
1828 struct btrfs_root_backup *root_backup;
1831 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1832 root_backup = info->super_copy->super_roots + i;
1833 cur = btrfs_backup_tree_root_gen(root_backup);
1834 if (cur == newest_gen)
1838 /* check to see if we actually wrapped around */
1839 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1840 root_backup = info->super_copy->super_roots;
1841 cur = btrfs_backup_tree_root_gen(root_backup);
1842 if (cur == newest_gen)
1845 return newest_index;
1850 * find the oldest backup so we know where to store new entries
1851 * in the backup array. This will set the backup_root_index
1852 * field in the fs_info struct
1854 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1857 int newest_index = -1;
1859 newest_index = find_newest_super_backup(info, newest_gen);
1860 /* if there was garbage in there, just move along */
1861 if (newest_index == -1) {
1862 info->backup_root_index = 0;
1864 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1869 * copy all the root pointers into the super backup array.
1870 * this will bump the backup pointer by one when it is
1873 static void backup_super_roots(struct btrfs_fs_info *info)
1876 struct btrfs_root_backup *root_backup;
1879 next_backup = info->backup_root_index;
1880 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1881 BTRFS_NUM_BACKUP_ROOTS;
1884 * just overwrite the last backup if we're at the same generation
1885 * this happens only at umount
1887 root_backup = info->super_for_commit->super_roots + last_backup;
1888 if (btrfs_backup_tree_root_gen(root_backup) ==
1889 btrfs_header_generation(info->tree_root->node))
1890 next_backup = last_backup;
1892 root_backup = info->super_for_commit->super_roots + next_backup;
1895 * make sure all of our padding and empty slots get zero filled
1896 * regardless of which ones we use today
1898 memset(root_backup, 0, sizeof(*root_backup));
1900 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1902 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1903 btrfs_set_backup_tree_root_gen(root_backup,
1904 btrfs_header_generation(info->tree_root->node));
1906 btrfs_set_backup_tree_root_level(root_backup,
1907 btrfs_header_level(info->tree_root->node));
1909 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1910 btrfs_set_backup_chunk_root_gen(root_backup,
1911 btrfs_header_generation(info->chunk_root->node));
1912 btrfs_set_backup_chunk_root_level(root_backup,
1913 btrfs_header_level(info->chunk_root->node));
1915 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1916 btrfs_set_backup_extent_root_gen(root_backup,
1917 btrfs_header_generation(info->extent_root->node));
1918 btrfs_set_backup_extent_root_level(root_backup,
1919 btrfs_header_level(info->extent_root->node));
1922 * we might commit during log recovery, which happens before we set
1923 * the fs_root. Make sure it is valid before we fill it in.
1925 if (info->fs_root && info->fs_root->node) {
1926 btrfs_set_backup_fs_root(root_backup,
1927 info->fs_root->node->start);
1928 btrfs_set_backup_fs_root_gen(root_backup,
1929 btrfs_header_generation(info->fs_root->node));
1930 btrfs_set_backup_fs_root_level(root_backup,
1931 btrfs_header_level(info->fs_root->node));
1934 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1935 btrfs_set_backup_dev_root_gen(root_backup,
1936 btrfs_header_generation(info->dev_root->node));
1937 btrfs_set_backup_dev_root_level(root_backup,
1938 btrfs_header_level(info->dev_root->node));
1940 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1941 btrfs_set_backup_csum_root_gen(root_backup,
1942 btrfs_header_generation(info->csum_root->node));
1943 btrfs_set_backup_csum_root_level(root_backup,
1944 btrfs_header_level(info->csum_root->node));
1946 btrfs_set_backup_total_bytes(root_backup,
1947 btrfs_super_total_bytes(info->super_copy));
1948 btrfs_set_backup_bytes_used(root_backup,
1949 btrfs_super_bytes_used(info->super_copy));
1950 btrfs_set_backup_num_devices(root_backup,
1951 btrfs_super_num_devices(info->super_copy));
1954 * if we don't copy this out to the super_copy, it won't get remembered
1955 * for the next commit
1957 memcpy(&info->super_copy->super_roots,
1958 &info->super_for_commit->super_roots,
1959 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1963 * this copies info out of the root backup array and back into
1964 * the in-memory super block. It is meant to help iterate through
1965 * the array, so you send it the number of backups you've already
1966 * tried and the last backup index you used.
1968 * this returns -1 when it has tried all the backups
1970 static noinline int next_root_backup(struct btrfs_fs_info *info,
1971 struct btrfs_super_block *super,
1972 int *num_backups_tried, int *backup_index)
1974 struct btrfs_root_backup *root_backup;
1975 int newest = *backup_index;
1977 if (*num_backups_tried == 0) {
1978 u64 gen = btrfs_super_generation(super);
1980 newest = find_newest_super_backup(info, gen);
1984 *backup_index = newest;
1985 *num_backups_tried = 1;
1986 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1987 /* we've tried all the backups, all done */
1990 /* jump to the next oldest backup */
1991 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1992 BTRFS_NUM_BACKUP_ROOTS;
1993 *backup_index = newest;
1994 *num_backups_tried += 1;
1996 root_backup = super->super_roots + newest;
1998 btrfs_set_super_generation(super,
1999 btrfs_backup_tree_root_gen(root_backup));
2000 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2001 btrfs_set_super_root_level(super,
2002 btrfs_backup_tree_root_level(root_backup));
2003 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2006 * fixme: the total bytes and num_devices need to match or we should
2009 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2010 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2014 /* helper to cleanup workers */
2015 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2017 btrfs_destroy_workqueue(fs_info->fixup_workers);
2018 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2019 btrfs_destroy_workqueue(fs_info->workers);
2020 btrfs_destroy_workqueue(fs_info->endio_workers);
2021 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2022 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2023 btrfs_destroy_workqueue(fs_info->rmw_workers);
2024 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2025 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2026 btrfs_destroy_workqueue(fs_info->submit_workers);
2027 btrfs_destroy_workqueue(fs_info->delayed_workers);
2028 btrfs_destroy_workqueue(fs_info->caching_workers);
2029 btrfs_destroy_workqueue(fs_info->readahead_workers);
2030 btrfs_destroy_workqueue(fs_info->flush_workers);
2031 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2032 btrfs_destroy_workqueue(fs_info->extent_workers);
2034 * Now that all other work queues are destroyed, we can safely destroy
2035 * the queues used for metadata I/O, since tasks from those other work
2036 * queues can do metadata I/O operations.
2038 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2039 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2042 static void free_root_extent_buffers(struct btrfs_root *root)
2045 free_extent_buffer(root->node);
2046 free_extent_buffer(root->commit_root);
2048 root->commit_root = NULL;
2052 /* helper to cleanup tree roots */
2053 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2055 free_root_extent_buffers(info->tree_root);
2057 free_root_extent_buffers(info->dev_root);
2058 free_root_extent_buffers(info->extent_root);
2059 free_root_extent_buffers(info->csum_root);
2060 free_root_extent_buffers(info->quota_root);
2061 free_root_extent_buffers(info->uuid_root);
2063 free_root_extent_buffers(info->chunk_root);
2064 free_root_extent_buffers(info->free_space_root);
2067 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2070 struct btrfs_root *gang[8];
2073 while (!list_empty(&fs_info->dead_roots)) {
2074 gang[0] = list_entry(fs_info->dead_roots.next,
2075 struct btrfs_root, root_list);
2076 list_del(&gang[0]->root_list);
2078 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2079 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2081 free_extent_buffer(gang[0]->node);
2082 free_extent_buffer(gang[0]->commit_root);
2083 btrfs_put_fs_root(gang[0]);
2088 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2093 for (i = 0; i < ret; i++)
2094 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2097 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2098 btrfs_free_log_root_tree(NULL, fs_info);
2099 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2103 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2105 mutex_init(&fs_info->scrub_lock);
2106 atomic_set(&fs_info->scrubs_running, 0);
2107 atomic_set(&fs_info->scrub_pause_req, 0);
2108 atomic_set(&fs_info->scrubs_paused, 0);
2109 atomic_set(&fs_info->scrub_cancel_req, 0);
2110 init_waitqueue_head(&fs_info->scrub_pause_wait);
2111 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2114 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2116 spin_lock_init(&fs_info->balance_lock);
2117 mutex_init(&fs_info->balance_mutex);
2118 atomic_set(&fs_info->balance_pause_req, 0);
2119 atomic_set(&fs_info->balance_cancel_req, 0);
2120 fs_info->balance_ctl = NULL;
2121 init_waitqueue_head(&fs_info->balance_wait_q);
2124 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2126 struct inode *inode = fs_info->btree_inode;
2128 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2129 set_nlink(inode, 1);
2131 * we set the i_size on the btree inode to the max possible int.
2132 * the real end of the address space is determined by all of
2133 * the devices in the system
2135 inode->i_size = OFFSET_MAX;
2136 inode->i_mapping->a_ops = &btree_aops;
2138 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2139 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2140 IO_TREE_INODE_IO, inode);
2141 BTRFS_I(inode)->io_tree.track_uptodate = false;
2142 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2144 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2146 BTRFS_I(inode)->root = fs_info->tree_root;
2147 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2148 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2149 btrfs_insert_inode_hash(inode);
2152 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2154 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2155 init_rwsem(&fs_info->dev_replace.rwsem);
2156 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2159 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2161 spin_lock_init(&fs_info->qgroup_lock);
2162 mutex_init(&fs_info->qgroup_ioctl_lock);
2163 fs_info->qgroup_tree = RB_ROOT;
2164 fs_info->qgroup_op_tree = RB_ROOT;
2165 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2166 fs_info->qgroup_seq = 1;
2167 fs_info->qgroup_ulist = NULL;
2168 fs_info->qgroup_rescan_running = false;
2169 mutex_init(&fs_info->qgroup_rescan_lock);
2172 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2173 struct btrfs_fs_devices *fs_devices)
2175 u32 max_active = fs_info->thread_pool_size;
2176 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2179 btrfs_alloc_workqueue(fs_info, "worker",
2180 flags | WQ_HIGHPRI, max_active, 16);
2182 fs_info->delalloc_workers =
2183 btrfs_alloc_workqueue(fs_info, "delalloc",
2184 flags, max_active, 2);
2186 fs_info->flush_workers =
2187 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2188 flags, max_active, 0);
2190 fs_info->caching_workers =
2191 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2194 * a higher idle thresh on the submit workers makes it much more
2195 * likely that bios will be send down in a sane order to the
2198 fs_info->submit_workers =
2199 btrfs_alloc_workqueue(fs_info, "submit", flags,
2200 min_t(u64, fs_devices->num_devices,
2203 fs_info->fixup_workers =
2204 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2207 * endios are largely parallel and should have a very
2210 fs_info->endio_workers =
2211 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2212 fs_info->endio_meta_workers =
2213 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2215 fs_info->endio_meta_write_workers =
2216 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2218 fs_info->endio_raid56_workers =
2219 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2221 fs_info->endio_repair_workers =
2222 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2223 fs_info->rmw_workers =
2224 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2225 fs_info->endio_write_workers =
2226 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2228 fs_info->endio_freespace_worker =
2229 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2231 fs_info->delayed_workers =
2232 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2234 fs_info->readahead_workers =
2235 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2237 fs_info->qgroup_rescan_workers =
2238 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2239 fs_info->extent_workers =
2240 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2241 min_t(u64, fs_devices->num_devices,
2244 if (!(fs_info->workers && fs_info->delalloc_workers &&
2245 fs_info->submit_workers && fs_info->flush_workers &&
2246 fs_info->endio_workers && fs_info->endio_meta_workers &&
2247 fs_info->endio_meta_write_workers &&
2248 fs_info->endio_repair_workers &&
2249 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2250 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2251 fs_info->caching_workers && fs_info->readahead_workers &&
2252 fs_info->fixup_workers && fs_info->delayed_workers &&
2253 fs_info->extent_workers &&
2254 fs_info->qgroup_rescan_workers)) {
2261 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2262 struct btrfs_fs_devices *fs_devices)
2265 struct btrfs_root *log_tree_root;
2266 struct btrfs_super_block *disk_super = fs_info->super_copy;
2267 u64 bytenr = btrfs_super_log_root(disk_super);
2268 int level = btrfs_super_log_root_level(disk_super);
2270 if (fs_devices->rw_devices == 0) {
2271 btrfs_warn(fs_info, "log replay required on RO media");
2275 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2279 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2281 log_tree_root->node = read_tree_block(fs_info, bytenr,
2282 fs_info->generation + 1,
2284 if (IS_ERR(log_tree_root->node)) {
2285 btrfs_warn(fs_info, "failed to read log tree");
2286 ret = PTR_ERR(log_tree_root->node);
2287 kfree(log_tree_root);
2289 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2290 btrfs_err(fs_info, "failed to read log tree");
2291 free_extent_buffer(log_tree_root->node);
2292 kfree(log_tree_root);
2295 /* returns with log_tree_root freed on success */
2296 ret = btrfs_recover_log_trees(log_tree_root);
2298 btrfs_handle_fs_error(fs_info, ret,
2299 "Failed to recover log tree");
2300 free_extent_buffer(log_tree_root->node);
2301 kfree(log_tree_root);
2305 if (sb_rdonly(fs_info->sb)) {
2306 ret = btrfs_commit_super(fs_info);
2314 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2316 struct btrfs_root *tree_root = fs_info->tree_root;
2317 struct btrfs_root *root;
2318 struct btrfs_key location;
2321 BUG_ON(!fs_info->tree_root);
2323 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2324 location.type = BTRFS_ROOT_ITEM_KEY;
2325 location.offset = 0;
2327 root = btrfs_read_tree_root(tree_root, &location);
2329 ret = PTR_ERR(root);
2332 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2333 fs_info->extent_root = root;
2335 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2336 root = btrfs_read_tree_root(tree_root, &location);
2338 ret = PTR_ERR(root);
2341 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2342 fs_info->dev_root = root;
2343 btrfs_init_devices_late(fs_info);
2345 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2346 root = btrfs_read_tree_root(tree_root, &location);
2348 ret = PTR_ERR(root);
2351 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2352 fs_info->csum_root = root;
2354 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2355 root = btrfs_read_tree_root(tree_root, &location);
2356 if (!IS_ERR(root)) {
2357 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2358 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2359 fs_info->quota_root = root;
2362 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2363 root = btrfs_read_tree_root(tree_root, &location);
2365 ret = PTR_ERR(root);
2369 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2370 fs_info->uuid_root = root;
2373 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2374 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2375 root = btrfs_read_tree_root(tree_root, &location);
2377 ret = PTR_ERR(root);
2380 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2381 fs_info->free_space_root = root;
2386 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2387 location.objectid, ret);
2392 * Real super block validation
2393 * NOTE: super csum type and incompat features will not be checked here.
2395 * @sb: super block to check
2396 * @mirror_num: the super block number to check its bytenr:
2397 * 0 the primary (1st) sb
2398 * 1, 2 2nd and 3rd backup copy
2399 * -1 skip bytenr check
2401 static int validate_super(struct btrfs_fs_info *fs_info,
2402 struct btrfs_super_block *sb, int mirror_num)
2404 u64 nodesize = btrfs_super_nodesize(sb);
2405 u64 sectorsize = btrfs_super_sectorsize(sb);
2408 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2409 btrfs_err(fs_info, "no valid FS found");
2412 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2413 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2414 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2417 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2418 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2419 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2422 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2423 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2424 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2427 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2428 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2429 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2434 * Check sectorsize and nodesize first, other check will need it.
2435 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2437 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2438 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2439 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2442 /* Only PAGE SIZE is supported yet */
2443 if (sectorsize != PAGE_SIZE) {
2445 "sectorsize %llu not supported yet, only support %lu",
2446 sectorsize, PAGE_SIZE);
2449 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2450 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2451 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2454 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2455 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2456 le32_to_cpu(sb->__unused_leafsize), nodesize);
2460 /* Root alignment check */
2461 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2462 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2463 btrfs_super_root(sb));
2466 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2467 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2468 btrfs_super_chunk_root(sb));
2471 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2472 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2473 btrfs_super_log_root(sb));
2477 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2478 BTRFS_FSID_SIZE) != 0) {
2480 "dev_item UUID does not match metadata fsid: %pU != %pU",
2481 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2486 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2489 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2490 btrfs_err(fs_info, "bytes_used is too small %llu",
2491 btrfs_super_bytes_used(sb));
2494 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2495 btrfs_err(fs_info, "invalid stripesize %u",
2496 btrfs_super_stripesize(sb));
2499 if (btrfs_super_num_devices(sb) > (1UL << 31))
2500 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2501 btrfs_super_num_devices(sb));
2502 if (btrfs_super_num_devices(sb) == 0) {
2503 btrfs_err(fs_info, "number of devices is 0");
2507 if (mirror_num >= 0 &&
2508 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2509 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2510 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2515 * Obvious sys_chunk_array corruptions, it must hold at least one key
2518 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2519 btrfs_err(fs_info, "system chunk array too big %u > %u",
2520 btrfs_super_sys_array_size(sb),
2521 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2524 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2525 + sizeof(struct btrfs_chunk)) {
2526 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2527 btrfs_super_sys_array_size(sb),
2528 sizeof(struct btrfs_disk_key)
2529 + sizeof(struct btrfs_chunk));
2534 * The generation is a global counter, we'll trust it more than the others
2535 * but it's still possible that it's the one that's wrong.
2537 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2539 "suspicious: generation < chunk_root_generation: %llu < %llu",
2540 btrfs_super_generation(sb),
2541 btrfs_super_chunk_root_generation(sb));
2542 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2543 && btrfs_super_cache_generation(sb) != (u64)-1)
2545 "suspicious: generation < cache_generation: %llu < %llu",
2546 btrfs_super_generation(sb),
2547 btrfs_super_cache_generation(sb));
2553 * Validation of super block at mount time.
2554 * Some checks already done early at mount time, like csum type and incompat
2555 * flags will be skipped.
2557 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2559 return validate_super(fs_info, fs_info->super_copy, 0);
2563 * Validation of super block at write time.
2564 * Some checks like bytenr check will be skipped as their values will be
2566 * Extra checks like csum type and incompat flags will be done here.
2568 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2569 struct btrfs_super_block *sb)
2573 ret = validate_super(fs_info, sb, -1);
2576 if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2578 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2579 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2582 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2585 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2586 btrfs_super_incompat_flags(sb),
2587 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2593 "super block corruption detected before writing it to disk");
2597 int open_ctree(struct super_block *sb,
2598 struct btrfs_fs_devices *fs_devices,
2606 struct btrfs_key location;
2607 struct buffer_head *bh;
2608 struct btrfs_super_block *disk_super;
2609 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2610 struct btrfs_root *tree_root;
2611 struct btrfs_root *chunk_root;
2614 int num_backups_tried = 0;
2615 int backup_index = 0;
2616 int clear_free_space_tree = 0;
2619 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2620 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2621 if (!tree_root || !chunk_root) {
2626 ret = init_srcu_struct(&fs_info->subvol_srcu);
2632 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2637 fs_info->dirty_metadata_batch = PAGE_SIZE *
2638 (1 + ilog2(nr_cpu_ids));
2640 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2643 goto fail_dirty_metadata_bytes;
2646 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2650 goto fail_delalloc_bytes;
2653 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2654 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2655 INIT_LIST_HEAD(&fs_info->trans_list);
2656 INIT_LIST_HEAD(&fs_info->dead_roots);
2657 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2658 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2659 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2660 INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2661 spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2662 spin_lock_init(&fs_info->delalloc_root_lock);
2663 spin_lock_init(&fs_info->trans_lock);
2664 spin_lock_init(&fs_info->fs_roots_radix_lock);
2665 spin_lock_init(&fs_info->delayed_iput_lock);
2666 spin_lock_init(&fs_info->defrag_inodes_lock);
2667 spin_lock_init(&fs_info->tree_mod_seq_lock);
2668 spin_lock_init(&fs_info->super_lock);
2669 spin_lock_init(&fs_info->qgroup_op_lock);
2670 spin_lock_init(&fs_info->buffer_lock);
2671 spin_lock_init(&fs_info->unused_bgs_lock);
2672 rwlock_init(&fs_info->tree_mod_log_lock);
2673 mutex_init(&fs_info->unused_bg_unpin_mutex);
2674 mutex_init(&fs_info->delete_unused_bgs_mutex);
2675 mutex_init(&fs_info->reloc_mutex);
2676 mutex_init(&fs_info->delalloc_root_mutex);
2677 seqlock_init(&fs_info->profiles_lock);
2679 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2680 INIT_LIST_HEAD(&fs_info->space_info);
2681 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2682 INIT_LIST_HEAD(&fs_info->unused_bgs);
2683 btrfs_mapping_init(&fs_info->mapping_tree);
2684 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2685 BTRFS_BLOCK_RSV_GLOBAL);
2686 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2687 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2688 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2689 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2690 BTRFS_BLOCK_RSV_DELOPS);
2691 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2692 BTRFS_BLOCK_RSV_DELREFS);
2694 atomic_set(&fs_info->async_delalloc_pages, 0);
2695 atomic_set(&fs_info->defrag_running, 0);
2696 atomic_set(&fs_info->qgroup_op_seq, 0);
2697 atomic_set(&fs_info->reada_works_cnt, 0);
2698 atomic_set(&fs_info->nr_delayed_iputs, 0);
2699 atomic64_set(&fs_info->tree_mod_seq, 0);
2701 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2702 fs_info->metadata_ratio = 0;
2703 fs_info->defrag_inodes = RB_ROOT;
2704 atomic64_set(&fs_info->free_chunk_space, 0);
2705 fs_info->tree_mod_log = RB_ROOT;
2706 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2707 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2708 /* readahead state */
2709 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2710 spin_lock_init(&fs_info->reada_lock);
2711 btrfs_init_ref_verify(fs_info);
2713 fs_info->thread_pool_size = min_t(unsigned long,
2714 num_online_cpus() + 2, 8);
2716 INIT_LIST_HEAD(&fs_info->ordered_roots);
2717 spin_lock_init(&fs_info->ordered_root_lock);
2719 fs_info->btree_inode = new_inode(sb);
2720 if (!fs_info->btree_inode) {
2722 goto fail_bio_counter;
2724 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2726 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2728 if (!fs_info->delayed_root) {
2732 btrfs_init_delayed_root(fs_info->delayed_root);
2734 btrfs_init_scrub(fs_info);
2735 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2736 fs_info->check_integrity_print_mask = 0;
2738 btrfs_init_balance(fs_info);
2739 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2741 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2742 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2744 btrfs_init_btree_inode(fs_info);
2746 spin_lock_init(&fs_info->block_group_cache_lock);
2747 fs_info->block_group_cache_tree = RB_ROOT;
2748 fs_info->first_logical_byte = (u64)-1;
2750 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2751 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2752 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2753 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2754 fs_info->pinned_extents = &fs_info->freed_extents[0];
2755 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2757 mutex_init(&fs_info->ordered_operations_mutex);
2758 mutex_init(&fs_info->tree_log_mutex);
2759 mutex_init(&fs_info->chunk_mutex);
2760 mutex_init(&fs_info->transaction_kthread_mutex);
2761 mutex_init(&fs_info->cleaner_mutex);
2762 mutex_init(&fs_info->ro_block_group_mutex);
2763 init_rwsem(&fs_info->commit_root_sem);
2764 init_rwsem(&fs_info->cleanup_work_sem);
2765 init_rwsem(&fs_info->subvol_sem);
2766 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2768 btrfs_init_dev_replace_locks(fs_info);
2769 btrfs_init_qgroup(fs_info);
2771 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2772 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2774 init_waitqueue_head(&fs_info->transaction_throttle);
2775 init_waitqueue_head(&fs_info->transaction_wait);
2776 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2777 init_waitqueue_head(&fs_info->async_submit_wait);
2778 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2780 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2782 /* Usable values until the real ones are cached from the superblock */
2783 fs_info->nodesize = 4096;
2784 fs_info->sectorsize = 4096;
2785 fs_info->stripesize = 4096;
2787 spin_lock_init(&fs_info->swapfile_pins_lock);
2788 fs_info->swapfile_pins = RB_ROOT;
2790 ret = btrfs_alloc_stripe_hash_table(fs_info);
2796 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2798 invalidate_bdev(fs_devices->latest_bdev);
2801 * Read super block and check the signature bytes only
2803 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2810 * We want to check superblock checksum, the type is stored inside.
2811 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2813 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2814 btrfs_err(fs_info, "superblock checksum mismatch");
2821 * super_copy is zeroed at allocation time and we never touch the
2822 * following bytes up to INFO_SIZE, the checksum is calculated from
2823 * the whole block of INFO_SIZE
2825 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2828 disk_super = fs_info->super_copy;
2830 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2833 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2834 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2835 fs_info->super_copy->metadata_uuid,
2839 features = btrfs_super_flags(disk_super);
2840 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2841 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2842 btrfs_set_super_flags(disk_super, features);
2844 "found metadata UUID change in progress flag, clearing");
2847 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2848 sizeof(*fs_info->super_for_commit));
2850 ret = btrfs_validate_mount_super(fs_info);
2852 btrfs_err(fs_info, "superblock contains fatal errors");
2857 if (!btrfs_super_root(disk_super))
2860 /* check FS state, whether FS is broken. */
2861 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2862 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2865 * run through our array of backup supers and setup
2866 * our ring pointer to the oldest one
2868 generation = btrfs_super_generation(disk_super);
2869 find_oldest_super_backup(fs_info, generation);
2872 * In the long term, we'll store the compression type in the super
2873 * block, and it'll be used for per file compression control.
2875 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2877 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2883 features = btrfs_super_incompat_flags(disk_super) &
2884 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2887 "cannot mount because of unsupported optional features (%llx)",
2893 features = btrfs_super_incompat_flags(disk_super);
2894 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2895 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2896 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2897 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2898 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2900 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2901 btrfs_info(fs_info, "has skinny extents");
2904 * flag our filesystem as having big metadata blocks if
2905 * they are bigger than the page size
2907 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2908 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2910 "flagging fs with big metadata feature");
2911 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2914 nodesize = btrfs_super_nodesize(disk_super);
2915 sectorsize = btrfs_super_sectorsize(disk_super);
2916 stripesize = sectorsize;
2917 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2918 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2920 /* Cache block sizes */
2921 fs_info->nodesize = nodesize;
2922 fs_info->sectorsize = sectorsize;
2923 fs_info->stripesize = stripesize;
2926 * mixed block groups end up with duplicate but slightly offset
2927 * extent buffers for the same range. It leads to corruptions
2929 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2930 (sectorsize != nodesize)) {
2932 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2933 nodesize, sectorsize);
2938 * Needn't use the lock because there is no other task which will
2941 btrfs_set_super_incompat_flags(disk_super, features);
2943 features = btrfs_super_compat_ro_flags(disk_super) &
2944 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2945 if (!sb_rdonly(sb) && features) {
2947 "cannot mount read-write because of unsupported optional features (%llx)",
2953 ret = btrfs_init_workqueues(fs_info, fs_devices);
2956 goto fail_sb_buffer;
2959 sb->s_bdi->congested_fn = btrfs_congested_fn;
2960 sb->s_bdi->congested_data = fs_info;
2961 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2962 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2963 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2964 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2966 sb->s_blocksize = sectorsize;
2967 sb->s_blocksize_bits = blksize_bits(sectorsize);
2968 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2970 mutex_lock(&fs_info->chunk_mutex);
2971 ret = btrfs_read_sys_array(fs_info);
2972 mutex_unlock(&fs_info->chunk_mutex);
2974 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2975 goto fail_sb_buffer;
2978 generation = btrfs_super_chunk_root_generation(disk_super);
2979 level = btrfs_super_chunk_root_level(disk_super);
2981 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2983 chunk_root->node = read_tree_block(fs_info,
2984 btrfs_super_chunk_root(disk_super),
2985 generation, level, NULL);
2986 if (IS_ERR(chunk_root->node) ||
2987 !extent_buffer_uptodate(chunk_root->node)) {
2988 btrfs_err(fs_info, "failed to read chunk root");
2989 if (!IS_ERR(chunk_root->node))
2990 free_extent_buffer(chunk_root->node);
2991 chunk_root->node = NULL;
2992 goto fail_tree_roots;
2994 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2995 chunk_root->commit_root = btrfs_root_node(chunk_root);
2997 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2998 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3000 ret = btrfs_read_chunk_tree(fs_info);
3002 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3003 goto fail_tree_roots;
3007 * Keep the devid that is marked to be the target device for the
3008 * device replace procedure
3010 btrfs_free_extra_devids(fs_devices, 0);
3012 if (!fs_devices->latest_bdev) {
3013 btrfs_err(fs_info, "failed to read devices");
3014 goto fail_tree_roots;
3018 generation = btrfs_super_generation(disk_super);
3019 level = btrfs_super_root_level(disk_super);
3021 tree_root->node = read_tree_block(fs_info,
3022 btrfs_super_root(disk_super),
3023 generation, level, NULL);
3024 if (IS_ERR(tree_root->node) ||
3025 !extent_buffer_uptodate(tree_root->node)) {
3026 btrfs_warn(fs_info, "failed to read tree root");
3027 if (!IS_ERR(tree_root->node))
3028 free_extent_buffer(tree_root->node);
3029 tree_root->node = NULL;
3030 goto recovery_tree_root;
3033 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3034 tree_root->commit_root = btrfs_root_node(tree_root);
3035 btrfs_set_root_refs(&tree_root->root_item, 1);
3037 mutex_lock(&tree_root->objectid_mutex);
3038 ret = btrfs_find_highest_objectid(tree_root,
3039 &tree_root->highest_objectid);
3041 mutex_unlock(&tree_root->objectid_mutex);
3042 goto recovery_tree_root;
3045 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3047 mutex_unlock(&tree_root->objectid_mutex);
3049 ret = btrfs_read_roots(fs_info);
3051 goto recovery_tree_root;
3053 fs_info->generation = generation;
3054 fs_info->last_trans_committed = generation;
3056 ret = btrfs_verify_dev_extents(fs_info);
3059 "failed to verify dev extents against chunks: %d",
3061 goto fail_block_groups;
3063 ret = btrfs_recover_balance(fs_info);
3065 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3066 goto fail_block_groups;
3069 ret = btrfs_init_dev_stats(fs_info);
3071 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3072 goto fail_block_groups;
3075 ret = btrfs_init_dev_replace(fs_info);
3077 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3078 goto fail_block_groups;
3081 btrfs_free_extra_devids(fs_devices, 1);
3083 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3085 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3087 goto fail_block_groups;
3090 ret = btrfs_sysfs_add_device(fs_devices);
3092 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3094 goto fail_fsdev_sysfs;
3097 ret = btrfs_sysfs_add_mounted(fs_info);
3099 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3100 goto fail_fsdev_sysfs;
3103 ret = btrfs_init_space_info(fs_info);
3105 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3109 ret = btrfs_read_block_groups(fs_info);
3111 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3115 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3117 "writable mount is not allowed due to too many missing devices");
3121 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3123 if (IS_ERR(fs_info->cleaner_kthread))
3126 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3128 "btrfs-transaction");
3129 if (IS_ERR(fs_info->transaction_kthread))
3132 if (!btrfs_test_opt(fs_info, NOSSD) &&
3133 !fs_info->fs_devices->rotating) {
3134 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3138 * Mount does not set all options immediately, we can do it now and do
3139 * not have to wait for transaction commit
3141 btrfs_apply_pending_changes(fs_info);
3143 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3144 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3145 ret = btrfsic_mount(fs_info, fs_devices,
3146 btrfs_test_opt(fs_info,
3147 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3149 fs_info->check_integrity_print_mask);
3152 "failed to initialize integrity check module: %d",
3156 ret = btrfs_read_qgroup_config(fs_info);
3158 goto fail_trans_kthread;
3160 if (btrfs_build_ref_tree(fs_info))
3161 btrfs_err(fs_info, "couldn't build ref tree");
3163 /* do not make disk changes in broken FS or nologreplay is given */
3164 if (btrfs_super_log_root(disk_super) != 0 &&
3165 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3166 ret = btrfs_replay_log(fs_info, fs_devices);
3173 ret = btrfs_find_orphan_roots(fs_info);
3177 if (!sb_rdonly(sb)) {
3178 ret = btrfs_cleanup_fs_roots(fs_info);
3182 mutex_lock(&fs_info->cleaner_mutex);
3183 ret = btrfs_recover_relocation(tree_root);
3184 mutex_unlock(&fs_info->cleaner_mutex);
3186 btrfs_warn(fs_info, "failed to recover relocation: %d",
3193 location.objectid = BTRFS_FS_TREE_OBJECTID;
3194 location.type = BTRFS_ROOT_ITEM_KEY;
3195 location.offset = 0;
3197 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3198 if (IS_ERR(fs_info->fs_root)) {
3199 err = PTR_ERR(fs_info->fs_root);
3200 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3207 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3208 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3209 clear_free_space_tree = 1;
3210 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3211 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3212 btrfs_warn(fs_info, "free space tree is invalid");
3213 clear_free_space_tree = 1;
3216 if (clear_free_space_tree) {
3217 btrfs_info(fs_info, "clearing free space tree");
3218 ret = btrfs_clear_free_space_tree(fs_info);
3221 "failed to clear free space tree: %d", ret);
3222 close_ctree(fs_info);
3227 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3228 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3229 btrfs_info(fs_info, "creating free space tree");
3230 ret = btrfs_create_free_space_tree(fs_info);
3233 "failed to create free space tree: %d", ret);
3234 close_ctree(fs_info);
3239 down_read(&fs_info->cleanup_work_sem);
3240 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3241 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3242 up_read(&fs_info->cleanup_work_sem);
3243 close_ctree(fs_info);
3246 up_read(&fs_info->cleanup_work_sem);
3248 ret = btrfs_resume_balance_async(fs_info);
3250 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3251 close_ctree(fs_info);
3255 ret = btrfs_resume_dev_replace_async(fs_info);
3257 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3258 close_ctree(fs_info);
3262 btrfs_qgroup_rescan_resume(fs_info);
3264 if (!fs_info->uuid_root) {
3265 btrfs_info(fs_info, "creating UUID tree");
3266 ret = btrfs_create_uuid_tree(fs_info);
3269 "failed to create the UUID tree: %d", ret);
3270 close_ctree(fs_info);
3273 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3274 fs_info->generation !=
3275 btrfs_super_uuid_tree_generation(disk_super)) {
3276 btrfs_info(fs_info, "checking UUID tree");
3277 ret = btrfs_check_uuid_tree(fs_info);
3280 "failed to check the UUID tree: %d", ret);
3281 close_ctree(fs_info);
3285 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3287 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3290 * backuproot only affect mount behavior, and if open_ctree succeeded,
3291 * no need to keep the flag
3293 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3298 btrfs_free_qgroup_config(fs_info);
3300 kthread_stop(fs_info->transaction_kthread);
3301 btrfs_cleanup_transaction(fs_info);
3302 btrfs_free_fs_roots(fs_info);
3304 kthread_stop(fs_info->cleaner_kthread);
3307 * make sure we're done with the btree inode before we stop our
3310 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3313 btrfs_sysfs_remove_mounted(fs_info);
3316 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3319 btrfs_put_block_group_cache(fs_info);
3322 free_root_pointers(fs_info, 1);
3323 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3326 btrfs_stop_all_workers(fs_info);
3327 btrfs_free_block_groups(fs_info);
3330 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3332 iput(fs_info->btree_inode);
3334 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3335 fail_delalloc_bytes:
3336 percpu_counter_destroy(&fs_info->delalloc_bytes);
3337 fail_dirty_metadata_bytes:
3338 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3340 cleanup_srcu_struct(&fs_info->subvol_srcu);
3342 btrfs_free_stripe_hash_table(fs_info);
3343 btrfs_close_devices(fs_info->fs_devices);
3347 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3348 goto fail_tree_roots;
3350 free_root_pointers(fs_info, 0);
3352 /* don't use the log in recovery mode, it won't be valid */
3353 btrfs_set_super_log_root(disk_super, 0);
3355 /* we can't trust the free space cache either */
3356 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3358 ret = next_root_backup(fs_info, fs_info->super_copy,
3359 &num_backups_tried, &backup_index);
3361 goto fail_block_groups;
3362 goto retry_root_backup;
3364 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3366 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3369 set_buffer_uptodate(bh);
3371 struct btrfs_device *device = (struct btrfs_device *)
3374 btrfs_warn_rl_in_rcu(device->fs_info,
3375 "lost page write due to IO error on %s",
3376 rcu_str_deref(device->name));
3377 /* note, we don't set_buffer_write_io_error because we have
3378 * our own ways of dealing with the IO errors
3380 clear_buffer_uptodate(bh);
3381 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3387 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3388 struct buffer_head **bh_ret)
3390 struct buffer_head *bh;
3391 struct btrfs_super_block *super;
3394 bytenr = btrfs_sb_offset(copy_num);
3395 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3398 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3400 * If we fail to read from the underlying devices, as of now
3401 * the best option we have is to mark it EIO.
3406 super = (struct btrfs_super_block *)bh->b_data;
3407 if (btrfs_super_bytenr(super) != bytenr ||
3408 btrfs_super_magic(super) != BTRFS_MAGIC) {
3418 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3420 struct buffer_head *bh;
3421 struct buffer_head *latest = NULL;
3422 struct btrfs_super_block *super;
3427 /* we would like to check all the supers, but that would make
3428 * a btrfs mount succeed after a mkfs from a different FS.
3429 * So, we need to add a special mount option to scan for
3430 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3432 for (i = 0; i < 1; i++) {
3433 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3437 super = (struct btrfs_super_block *)bh->b_data;
3439 if (!latest || btrfs_super_generation(super) > transid) {
3442 transid = btrfs_super_generation(super);
3449 return ERR_PTR(ret);
3455 * Write superblock @sb to the @device. Do not wait for completion, all the
3456 * buffer heads we write are pinned.
3458 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3459 * the expected device size at commit time. Note that max_mirrors must be
3460 * same for write and wait phases.
3462 * Return number of errors when buffer head is not found or submission fails.
3464 static int write_dev_supers(struct btrfs_device *device,
3465 struct btrfs_super_block *sb, int max_mirrors)
3467 struct buffer_head *bh;
3475 if (max_mirrors == 0)
3476 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3478 for (i = 0; i < max_mirrors; i++) {
3479 bytenr = btrfs_sb_offset(i);
3480 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3481 device->commit_total_bytes)
3484 btrfs_set_super_bytenr(sb, bytenr);
3487 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3488 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3489 btrfs_csum_final(crc, sb->csum);
3491 /* One reference for us, and we leave it for the caller */
3492 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3493 BTRFS_SUPER_INFO_SIZE);
3495 btrfs_err(device->fs_info,
3496 "couldn't get super buffer head for bytenr %llu",
3502 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3504 /* one reference for submit_bh */
3507 set_buffer_uptodate(bh);
3509 bh->b_end_io = btrfs_end_buffer_write_sync;
3510 bh->b_private = device;
3513 * we fua the first super. The others we allow
3516 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3517 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3518 op_flags |= REQ_FUA;
3519 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3523 return errors < i ? 0 : -1;
3527 * Wait for write completion of superblocks done by write_dev_supers,
3528 * @max_mirrors same for write and wait phases.
3530 * Return number of errors when buffer head is not found or not marked up to
3533 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3535 struct buffer_head *bh;
3538 bool primary_failed = false;
3541 if (max_mirrors == 0)
3542 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3544 for (i = 0; i < max_mirrors; i++) {
3545 bytenr = btrfs_sb_offset(i);
3546 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3547 device->commit_total_bytes)
3550 bh = __find_get_block(device->bdev,
3551 bytenr / BTRFS_BDEV_BLOCKSIZE,
3552 BTRFS_SUPER_INFO_SIZE);
3556 primary_failed = true;
3560 if (!buffer_uptodate(bh)) {
3563 primary_failed = true;
3566 /* drop our reference */
3569 /* drop the reference from the writing run */
3573 /* log error, force error return */
3574 if (primary_failed) {
3575 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3580 return errors < i ? 0 : -1;
3584 * endio for the write_dev_flush, this will wake anyone waiting
3585 * for the barrier when it is done
3587 static void btrfs_end_empty_barrier(struct bio *bio)
3589 complete(bio->bi_private);
3593 * Submit a flush request to the device if it supports it. Error handling is
3594 * done in the waiting counterpart.
3596 static void write_dev_flush(struct btrfs_device *device)
3598 struct request_queue *q = bdev_get_queue(device->bdev);
3599 struct bio *bio = device->flush_bio;
3601 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3605 bio->bi_end_io = btrfs_end_empty_barrier;
3606 bio_set_dev(bio, device->bdev);
3607 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3608 init_completion(&device->flush_wait);
3609 bio->bi_private = &device->flush_wait;
3611 btrfsic_submit_bio(bio);
3612 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3616 * If the flush bio has been submitted by write_dev_flush, wait for it.
3618 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3620 struct bio *bio = device->flush_bio;
3622 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3625 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3626 wait_for_completion_io(&device->flush_wait);
3628 return bio->bi_status;
3631 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3633 if (!btrfs_check_rw_degradable(fs_info, NULL))
3639 * send an empty flush down to each device in parallel,
3640 * then wait for them
3642 static int barrier_all_devices(struct btrfs_fs_info *info)
3644 struct list_head *head;
3645 struct btrfs_device *dev;
3646 int errors_wait = 0;
3649 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3650 /* send down all the barriers */
3651 head = &info->fs_devices->devices;
3652 list_for_each_entry(dev, head, dev_list) {
3653 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3657 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3658 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3661 write_dev_flush(dev);
3662 dev->last_flush_error = BLK_STS_OK;
3665 /* wait for all the barriers */
3666 list_for_each_entry(dev, head, dev_list) {
3667 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3673 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3674 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3677 ret = wait_dev_flush(dev);
3679 dev->last_flush_error = ret;
3680 btrfs_dev_stat_inc_and_print(dev,
3681 BTRFS_DEV_STAT_FLUSH_ERRS);
3688 * At some point we need the status of all disks
3689 * to arrive at the volume status. So error checking
3690 * is being pushed to a separate loop.
3692 return check_barrier_error(info);
3697 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3700 int min_tolerated = INT_MAX;
3702 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3703 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3704 min_tolerated = min(min_tolerated,
3705 btrfs_raid_array[BTRFS_RAID_SINGLE].
3706 tolerated_failures);
3708 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3709 if (raid_type == BTRFS_RAID_SINGLE)
3711 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3713 min_tolerated = min(min_tolerated,
3714 btrfs_raid_array[raid_type].
3715 tolerated_failures);
3718 if (min_tolerated == INT_MAX) {
3719 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3723 return min_tolerated;
3726 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3728 struct list_head *head;
3729 struct btrfs_device *dev;
3730 struct btrfs_super_block *sb;
3731 struct btrfs_dev_item *dev_item;
3735 int total_errors = 0;
3738 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3741 * max_mirrors == 0 indicates we're from commit_transaction,
3742 * not from fsync where the tree roots in fs_info have not
3743 * been consistent on disk.
3745 if (max_mirrors == 0)
3746 backup_super_roots(fs_info);
3748 sb = fs_info->super_for_commit;
3749 dev_item = &sb->dev_item;
3751 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3752 head = &fs_info->fs_devices->devices;
3753 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3756 ret = barrier_all_devices(fs_info);
3759 &fs_info->fs_devices->device_list_mutex);
3760 btrfs_handle_fs_error(fs_info, ret,
3761 "errors while submitting device barriers.");
3766 list_for_each_entry(dev, head, dev_list) {
3771 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3772 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3775 btrfs_set_stack_device_generation(dev_item, 0);
3776 btrfs_set_stack_device_type(dev_item, dev->type);
3777 btrfs_set_stack_device_id(dev_item, dev->devid);
3778 btrfs_set_stack_device_total_bytes(dev_item,
3779 dev->commit_total_bytes);
3780 btrfs_set_stack_device_bytes_used(dev_item,
3781 dev->commit_bytes_used);
3782 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3783 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3784 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3785 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3786 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3789 flags = btrfs_super_flags(sb);
3790 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3792 ret = btrfs_validate_write_super(fs_info, sb);
3794 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3795 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3796 "unexpected superblock corruption detected");
3800 ret = write_dev_supers(dev, sb, max_mirrors);
3804 if (total_errors > max_errors) {
3805 btrfs_err(fs_info, "%d errors while writing supers",
3807 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3809 /* FUA is masked off if unsupported and can't be the reason */
3810 btrfs_handle_fs_error(fs_info, -EIO,
3811 "%d errors while writing supers",
3817 list_for_each_entry(dev, head, dev_list) {
3820 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3821 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3824 ret = wait_dev_supers(dev, max_mirrors);
3828 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3829 if (total_errors > max_errors) {
3830 btrfs_handle_fs_error(fs_info, -EIO,
3831 "%d errors while writing supers",
3838 /* Drop a fs root from the radix tree and free it. */
3839 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3840 struct btrfs_root *root)
3842 spin_lock(&fs_info->fs_roots_radix_lock);
3843 radix_tree_delete(&fs_info->fs_roots_radix,
3844 (unsigned long)root->root_key.objectid);
3845 spin_unlock(&fs_info->fs_roots_radix_lock);
3847 if (btrfs_root_refs(&root->root_item) == 0)
3848 synchronize_srcu(&fs_info->subvol_srcu);
3850 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3851 btrfs_free_log(NULL, root);
3852 if (root->reloc_root) {
3853 free_extent_buffer(root->reloc_root->node);
3854 free_extent_buffer(root->reloc_root->commit_root);
3855 btrfs_put_fs_root(root->reloc_root);
3856 root->reloc_root = NULL;
3860 if (root->free_ino_pinned)
3861 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3862 if (root->free_ino_ctl)
3863 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3864 btrfs_free_fs_root(root);
3867 void btrfs_free_fs_root(struct btrfs_root *root)
3869 iput(root->ino_cache_inode);
3870 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3872 free_anon_bdev(root->anon_dev);
3873 if (root->subv_writers)
3874 btrfs_free_subvolume_writers(root->subv_writers);
3875 free_extent_buffer(root->node);
3876 free_extent_buffer(root->commit_root);
3877 kfree(root->free_ino_ctl);
3878 kfree(root->free_ino_pinned);
3879 btrfs_put_fs_root(root);
3882 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3884 u64 root_objectid = 0;
3885 struct btrfs_root *gang[8];
3888 unsigned int ret = 0;
3892 index = srcu_read_lock(&fs_info->subvol_srcu);
3893 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3894 (void **)gang, root_objectid,
3897 srcu_read_unlock(&fs_info->subvol_srcu, index);
3900 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3902 for (i = 0; i < ret; i++) {
3903 /* Avoid to grab roots in dead_roots */
3904 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3908 /* grab all the search result for later use */
3909 gang[i] = btrfs_grab_fs_root(gang[i]);
3911 srcu_read_unlock(&fs_info->subvol_srcu, index);
3913 for (i = 0; i < ret; i++) {
3916 root_objectid = gang[i]->root_key.objectid;
3917 err = btrfs_orphan_cleanup(gang[i]);
3920 btrfs_put_fs_root(gang[i]);
3925 /* release the uncleaned roots due to error */
3926 for (; i < ret; i++) {
3928 btrfs_put_fs_root(gang[i]);
3933 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3935 struct btrfs_root *root = fs_info->tree_root;
3936 struct btrfs_trans_handle *trans;
3938 mutex_lock(&fs_info->cleaner_mutex);
3939 btrfs_run_delayed_iputs(fs_info);
3940 mutex_unlock(&fs_info->cleaner_mutex);
3941 wake_up_process(fs_info->cleaner_kthread);
3943 /* wait until ongoing cleanup work done */
3944 down_write(&fs_info->cleanup_work_sem);
3945 up_write(&fs_info->cleanup_work_sem);
3947 trans = btrfs_join_transaction(root);
3949 return PTR_ERR(trans);
3950 return btrfs_commit_transaction(trans);
3953 void close_ctree(struct btrfs_fs_info *fs_info)
3957 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3959 * We don't want the cleaner to start new transactions, add more delayed
3960 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3961 * because that frees the task_struct, and the transaction kthread might
3962 * still try to wake up the cleaner.
3964 kthread_park(fs_info->cleaner_kthread);
3966 /* wait for the qgroup rescan worker to stop */
3967 btrfs_qgroup_wait_for_completion(fs_info, false);
3969 /* wait for the uuid_scan task to finish */
3970 down(&fs_info->uuid_tree_rescan_sem);
3971 /* avoid complains from lockdep et al., set sem back to initial state */
3972 up(&fs_info->uuid_tree_rescan_sem);
3974 /* pause restriper - we want to resume on mount */
3975 btrfs_pause_balance(fs_info);
3977 btrfs_dev_replace_suspend_for_unmount(fs_info);
3979 btrfs_scrub_cancel(fs_info);
3981 /* wait for any defraggers to finish */
3982 wait_event(fs_info->transaction_wait,
3983 (atomic_read(&fs_info->defrag_running) == 0));
3985 /* clear out the rbtree of defraggable inodes */
3986 btrfs_cleanup_defrag_inodes(fs_info);
3988 cancel_work_sync(&fs_info->async_reclaim_work);
3990 if (!sb_rdonly(fs_info->sb)) {
3992 * The cleaner kthread is stopped, so do one final pass over
3993 * unused block groups.
3995 btrfs_delete_unused_bgs(fs_info);
3997 ret = btrfs_commit_super(fs_info);
3999 btrfs_err(fs_info, "commit super ret %d", ret);
4002 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4003 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4004 btrfs_error_commit_super(fs_info);
4006 kthread_stop(fs_info->transaction_kthread);
4007 kthread_stop(fs_info->cleaner_kthread);
4009 ASSERT(list_empty(&fs_info->delayed_iputs));
4010 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4012 btrfs_free_qgroup_config(fs_info);
4013 ASSERT(list_empty(&fs_info->delalloc_roots));
4015 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4016 btrfs_info(fs_info, "at unmount delalloc count %lld",
4017 percpu_counter_sum(&fs_info->delalloc_bytes));
4020 btrfs_sysfs_remove_mounted(fs_info);
4021 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4023 btrfs_free_fs_roots(fs_info);
4025 btrfs_put_block_group_cache(fs_info);
4028 * we must make sure there is not any read request to
4029 * submit after we stopping all workers.
4031 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4032 btrfs_stop_all_workers(fs_info);
4034 btrfs_free_block_groups(fs_info);
4036 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4037 free_root_pointers(fs_info, 1);
4039 iput(fs_info->btree_inode);
4041 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4042 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4043 btrfsic_unmount(fs_info->fs_devices);
4046 btrfs_close_devices(fs_info->fs_devices);
4047 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4049 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4050 percpu_counter_destroy(&fs_info->delalloc_bytes);
4051 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4052 cleanup_srcu_struct(&fs_info->subvol_srcu);
4054 btrfs_free_stripe_hash_table(fs_info);
4055 btrfs_free_ref_cache(fs_info);
4057 while (!list_empty(&fs_info->pinned_chunks)) {
4058 struct extent_map *em;
4060 em = list_first_entry(&fs_info->pinned_chunks,
4061 struct extent_map, list);
4062 list_del_init(&em->list);
4063 free_extent_map(em);
4067 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4071 struct inode *btree_inode = buf->pages[0]->mapping->host;
4073 ret = extent_buffer_uptodate(buf);
4077 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4078 parent_transid, atomic);
4084 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4086 struct btrfs_fs_info *fs_info;
4087 struct btrfs_root *root;
4088 u64 transid = btrfs_header_generation(buf);
4091 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4093 * This is a fast path so only do this check if we have sanity tests
4094 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4095 * outside of the sanity tests.
4097 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4100 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4101 fs_info = root->fs_info;
4102 btrfs_assert_tree_locked(buf);
4103 if (transid != fs_info->generation)
4104 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4105 buf->start, transid, fs_info->generation);
4106 was_dirty = set_extent_buffer_dirty(buf);
4108 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4110 fs_info->dirty_metadata_batch);
4111 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4113 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4114 * but item data not updated.
4115 * So here we should only check item pointers, not item data.
4117 if (btrfs_header_level(buf) == 0 &&
4118 btrfs_check_leaf_relaxed(fs_info, buf)) {
4119 btrfs_print_leaf(buf);
4125 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4129 * looks as though older kernels can get into trouble with
4130 * this code, they end up stuck in balance_dirty_pages forever
4134 if (current->flags & PF_MEMALLOC)
4138 btrfs_balance_delayed_items(fs_info);
4140 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4141 BTRFS_DIRTY_METADATA_THRESH,
4142 fs_info->dirty_metadata_batch);
4144 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4148 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4150 __btrfs_btree_balance_dirty(fs_info, 1);
4153 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4155 __btrfs_btree_balance_dirty(fs_info, 0);
4158 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4159 struct btrfs_key *first_key)
4161 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4162 struct btrfs_fs_info *fs_info = root->fs_info;
4164 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4168 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4170 /* cleanup FS via transaction */
4171 btrfs_cleanup_transaction(fs_info);
4173 mutex_lock(&fs_info->cleaner_mutex);
4174 btrfs_run_delayed_iputs(fs_info);
4175 mutex_unlock(&fs_info->cleaner_mutex);
4177 down_write(&fs_info->cleanup_work_sem);
4178 up_write(&fs_info->cleanup_work_sem);
4181 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4183 struct btrfs_ordered_extent *ordered;
4185 spin_lock(&root->ordered_extent_lock);
4187 * This will just short circuit the ordered completion stuff which will
4188 * make sure the ordered extent gets properly cleaned up.
4190 list_for_each_entry(ordered, &root->ordered_extents,
4192 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4193 spin_unlock(&root->ordered_extent_lock);
4196 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4198 struct btrfs_root *root;
4199 struct list_head splice;
4201 INIT_LIST_HEAD(&splice);
4203 spin_lock(&fs_info->ordered_root_lock);
4204 list_splice_init(&fs_info->ordered_roots, &splice);
4205 while (!list_empty(&splice)) {
4206 root = list_first_entry(&splice, struct btrfs_root,
4208 list_move_tail(&root->ordered_root,
4209 &fs_info->ordered_roots);
4211 spin_unlock(&fs_info->ordered_root_lock);
4212 btrfs_destroy_ordered_extents(root);
4215 spin_lock(&fs_info->ordered_root_lock);
4217 spin_unlock(&fs_info->ordered_root_lock);
4220 * We need this here because if we've been flipped read-only we won't
4221 * get sync() from the umount, so we need to make sure any ordered
4222 * extents that haven't had their dirty pages IO start writeout yet
4223 * actually get run and error out properly.
4225 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4228 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4229 struct btrfs_fs_info *fs_info)
4231 struct rb_node *node;
4232 struct btrfs_delayed_ref_root *delayed_refs;
4233 struct btrfs_delayed_ref_node *ref;
4236 delayed_refs = &trans->delayed_refs;
4238 spin_lock(&delayed_refs->lock);
4239 if (atomic_read(&delayed_refs->num_entries) == 0) {
4240 spin_unlock(&delayed_refs->lock);
4241 btrfs_info(fs_info, "delayed_refs has NO entry");
4245 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4246 struct btrfs_delayed_ref_head *head;
4248 bool pin_bytes = false;
4250 head = rb_entry(node, struct btrfs_delayed_ref_head,
4252 if (btrfs_delayed_ref_lock(delayed_refs, head))
4255 spin_lock(&head->lock);
4256 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4257 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4260 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4261 RB_CLEAR_NODE(&ref->ref_node);
4262 if (!list_empty(&ref->add_list))
4263 list_del(&ref->add_list);
4264 atomic_dec(&delayed_refs->num_entries);
4265 btrfs_put_delayed_ref(ref);
4267 if (head->must_insert_reserved)
4269 btrfs_free_delayed_extent_op(head->extent_op);
4270 btrfs_delete_ref_head(delayed_refs, head);
4271 spin_unlock(&head->lock);
4272 spin_unlock(&delayed_refs->lock);
4273 mutex_unlock(&head->mutex);
4276 btrfs_pin_extent(fs_info, head->bytenr,
4277 head->num_bytes, 1);
4278 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4279 btrfs_put_delayed_ref_head(head);
4281 spin_lock(&delayed_refs->lock);
4284 spin_unlock(&delayed_refs->lock);
4289 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4291 struct btrfs_inode *btrfs_inode;
4292 struct list_head splice;
4294 INIT_LIST_HEAD(&splice);
4296 spin_lock(&root->delalloc_lock);
4297 list_splice_init(&root->delalloc_inodes, &splice);
4299 while (!list_empty(&splice)) {
4300 struct inode *inode = NULL;
4301 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4303 __btrfs_del_delalloc_inode(root, btrfs_inode);
4304 spin_unlock(&root->delalloc_lock);
4307 * Make sure we get a live inode and that it'll not disappear
4310 inode = igrab(&btrfs_inode->vfs_inode);
4312 invalidate_inode_pages2(inode->i_mapping);
4315 spin_lock(&root->delalloc_lock);
4317 spin_unlock(&root->delalloc_lock);
4320 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4322 struct btrfs_root *root;
4323 struct list_head splice;
4325 INIT_LIST_HEAD(&splice);
4327 spin_lock(&fs_info->delalloc_root_lock);
4328 list_splice_init(&fs_info->delalloc_roots, &splice);
4329 while (!list_empty(&splice)) {
4330 root = list_first_entry(&splice, struct btrfs_root,
4332 root = btrfs_grab_fs_root(root);
4334 spin_unlock(&fs_info->delalloc_root_lock);
4336 btrfs_destroy_delalloc_inodes(root);
4337 btrfs_put_fs_root(root);
4339 spin_lock(&fs_info->delalloc_root_lock);
4341 spin_unlock(&fs_info->delalloc_root_lock);
4344 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4345 struct extent_io_tree *dirty_pages,
4349 struct extent_buffer *eb;
4354 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4359 clear_extent_bits(dirty_pages, start, end, mark);
4360 while (start <= end) {
4361 eb = find_extent_buffer(fs_info, start);
4362 start += fs_info->nodesize;
4365 wait_on_extent_buffer_writeback(eb);
4367 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4369 clear_extent_buffer_dirty(eb);
4370 free_extent_buffer_stale(eb);
4377 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4378 struct extent_io_tree *pinned_extents)
4380 struct extent_io_tree *unpin;
4386 unpin = pinned_extents;
4389 struct extent_state *cached_state = NULL;
4392 * The btrfs_finish_extent_commit() may get the same range as
4393 * ours between find_first_extent_bit and clear_extent_dirty.
4394 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4395 * the same extent range.
4397 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4398 ret = find_first_extent_bit(unpin, 0, &start, &end,
4399 EXTENT_DIRTY, &cached_state);
4401 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4405 clear_extent_dirty(unpin, start, end, &cached_state);
4406 free_extent_state(cached_state);
4407 btrfs_error_unpin_extent_range(fs_info, start, end);
4408 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4413 if (unpin == &fs_info->freed_extents[0])
4414 unpin = &fs_info->freed_extents[1];
4416 unpin = &fs_info->freed_extents[0];
4424 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4426 struct inode *inode;
4428 inode = cache->io_ctl.inode;
4430 invalidate_inode_pages2(inode->i_mapping);
4431 BTRFS_I(inode)->generation = 0;
4432 cache->io_ctl.inode = NULL;
4435 btrfs_put_block_group(cache);
4438 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4439 struct btrfs_fs_info *fs_info)
4441 struct btrfs_block_group_cache *cache;
4443 spin_lock(&cur_trans->dirty_bgs_lock);
4444 while (!list_empty(&cur_trans->dirty_bgs)) {
4445 cache = list_first_entry(&cur_trans->dirty_bgs,
4446 struct btrfs_block_group_cache,
4449 if (!list_empty(&cache->io_list)) {
4450 spin_unlock(&cur_trans->dirty_bgs_lock);
4451 list_del_init(&cache->io_list);
4452 btrfs_cleanup_bg_io(cache);
4453 spin_lock(&cur_trans->dirty_bgs_lock);
4456 list_del_init(&cache->dirty_list);
4457 spin_lock(&cache->lock);
4458 cache->disk_cache_state = BTRFS_DC_ERROR;
4459 spin_unlock(&cache->lock);
4461 spin_unlock(&cur_trans->dirty_bgs_lock);
4462 btrfs_put_block_group(cache);
4463 btrfs_delayed_refs_rsv_release(fs_info, 1);
4464 spin_lock(&cur_trans->dirty_bgs_lock);
4466 spin_unlock(&cur_trans->dirty_bgs_lock);
4469 * Refer to the definition of io_bgs member for details why it's safe
4470 * to use it without any locking
4472 while (!list_empty(&cur_trans->io_bgs)) {
4473 cache = list_first_entry(&cur_trans->io_bgs,
4474 struct btrfs_block_group_cache,
4477 list_del_init(&cache->io_list);
4478 spin_lock(&cache->lock);
4479 cache->disk_cache_state = BTRFS_DC_ERROR;
4480 spin_unlock(&cache->lock);
4481 btrfs_cleanup_bg_io(cache);
4485 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4486 struct btrfs_fs_info *fs_info)
4488 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4489 ASSERT(list_empty(&cur_trans->dirty_bgs));
4490 ASSERT(list_empty(&cur_trans->io_bgs));
4492 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4494 cur_trans->state = TRANS_STATE_COMMIT_START;
4495 wake_up(&fs_info->transaction_blocked_wait);
4497 cur_trans->state = TRANS_STATE_UNBLOCKED;
4498 wake_up(&fs_info->transaction_wait);
4500 btrfs_destroy_delayed_inodes(fs_info);
4501 btrfs_assert_delayed_root_empty(fs_info);
4503 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4505 btrfs_destroy_pinned_extent(fs_info,
4506 fs_info->pinned_extents);
4508 cur_trans->state =TRANS_STATE_COMPLETED;
4509 wake_up(&cur_trans->commit_wait);
4512 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4514 struct btrfs_transaction *t;
4516 mutex_lock(&fs_info->transaction_kthread_mutex);
4518 spin_lock(&fs_info->trans_lock);
4519 while (!list_empty(&fs_info->trans_list)) {
4520 t = list_first_entry(&fs_info->trans_list,
4521 struct btrfs_transaction, list);
4522 if (t->state >= TRANS_STATE_COMMIT_START) {
4523 refcount_inc(&t->use_count);
4524 spin_unlock(&fs_info->trans_lock);
4525 btrfs_wait_for_commit(fs_info, t->transid);
4526 btrfs_put_transaction(t);
4527 spin_lock(&fs_info->trans_lock);
4530 if (t == fs_info->running_transaction) {
4531 t->state = TRANS_STATE_COMMIT_DOING;
4532 spin_unlock(&fs_info->trans_lock);
4534 * We wait for 0 num_writers since we don't hold a trans
4535 * handle open currently for this transaction.
4537 wait_event(t->writer_wait,
4538 atomic_read(&t->num_writers) == 0);
4540 spin_unlock(&fs_info->trans_lock);
4542 btrfs_cleanup_one_transaction(t, fs_info);
4544 spin_lock(&fs_info->trans_lock);
4545 if (t == fs_info->running_transaction)
4546 fs_info->running_transaction = NULL;
4547 list_del_init(&t->list);
4548 spin_unlock(&fs_info->trans_lock);
4550 btrfs_put_transaction(t);
4551 trace_btrfs_transaction_commit(fs_info->tree_root);
4552 spin_lock(&fs_info->trans_lock);
4554 spin_unlock(&fs_info->trans_lock);
4555 btrfs_destroy_all_ordered_extents(fs_info);
4556 btrfs_destroy_delayed_inodes(fs_info);
4557 btrfs_assert_delayed_root_empty(fs_info);
4558 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4559 btrfs_destroy_all_delalloc_inodes(fs_info);
4560 mutex_unlock(&fs_info->transaction_kthread_mutex);
4565 static const struct extent_io_ops btree_extent_io_ops = {
4566 /* mandatory callbacks */
4567 .submit_bio_hook = btree_submit_bio_hook,
4568 .readpage_end_io_hook = btree_readpage_end_io_hook,