1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include <crypto/hash.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
44 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
45 BTRFS_HEADER_FLAG_RELOC |\
46 BTRFS_SUPER_FLAG_ERROR |\
47 BTRFS_SUPER_FLAG_SEEDING |\
48 BTRFS_SUPER_FLAG_METADUMP |\
49 BTRFS_SUPER_FLAG_METADUMP_V2)
51 static const struct extent_io_ops btree_extent_io_ops;
52 static void end_workqueue_fn(struct btrfs_work *work);
53 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 struct btrfs_fs_info *fs_info);
56 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
57 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
58 struct extent_io_tree *dirty_pages,
60 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
61 struct extent_io_tree *pinned_extents);
62 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
63 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66 * btrfs_end_io_wq structs are used to do processing in task context when an IO
67 * is complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
70 struct btrfs_end_io_wq {
74 struct btrfs_fs_info *info;
76 enum btrfs_wq_endio_type metadata;
77 struct btrfs_work work;
80 static struct kmem_cache *btrfs_end_io_wq_cache;
82 int __init btrfs_end_io_wq_init(void)
84 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
85 sizeof(struct btrfs_end_io_wq),
89 if (!btrfs_end_io_wq_cache)
94 void __cold btrfs_end_io_wq_exit(void)
96 kmem_cache_destroy(btrfs_end_io_wq_cache);
100 * async submit bios are used to offload expensive checksumming
101 * onto the worker threads. They checksum file and metadata bios
102 * just before they are sent down the IO stack.
104 struct async_submit_bio {
107 extent_submit_bio_start_t *submit_bio_start;
110 * bio_offset is optional, can be used if the pages in the bio
111 * can't tell us where in the file the bio should go
114 struct btrfs_work work;
119 * Lockdep class keys for extent_buffer->lock's in this root. For a given
120 * eb, the lockdep key is determined by the btrfs_root it belongs to and
121 * the level the eb occupies in the tree.
123 * Different roots are used for different purposes and may nest inside each
124 * other and they require separate keysets. As lockdep keys should be
125 * static, assign keysets according to the purpose of the root as indicated
126 * by btrfs_root->root_key.objectid. This ensures that all special purpose
127 * roots have separate keysets.
129 * Lock-nesting across peer nodes is always done with the immediate parent
130 * node locked thus preventing deadlock. As lockdep doesn't know this, use
131 * subclass to avoid triggering lockdep warning in such cases.
133 * The key is set by the readpage_end_io_hook after the buffer has passed
134 * csum validation but before the pages are unlocked. It is also set by
135 * btrfs_init_new_buffer on freshly allocated blocks.
137 * We also add a check to make sure the highest level of the tree is the
138 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
139 * needs update as well.
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
142 # if BTRFS_MAX_LEVEL != 8
146 static struct btrfs_lockdep_keyset {
147 u64 id; /* root objectid */
148 const char *name_stem; /* lock name stem */
149 char names[BTRFS_MAX_LEVEL + 1][20];
150 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
151 } btrfs_lockdep_keysets[] = {
152 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
153 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
154 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
155 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
156 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
157 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
158 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
159 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
160 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
161 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
162 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
163 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
164 { .id = 0, .name_stem = "tree" },
167 void __init btrfs_init_lockdep(void)
171 /* initialize lockdep class names */
172 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
173 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
175 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
176 snprintf(ks->names[j], sizeof(ks->names[j]),
177 "btrfs-%s-%02d", ks->name_stem, j);
181 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
184 struct btrfs_lockdep_keyset *ks;
186 BUG_ON(level >= ARRAY_SIZE(ks->keys));
188 /* find the matching keyset, id 0 is the default entry */
189 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
190 if (ks->id == objectid)
193 lockdep_set_class_and_name(&eb->lock,
194 &ks->keys[level], ks->names[level]);
200 * extents on the btree inode are pretty simple, there's one extent
201 * that covers the entire device
203 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
204 struct page *page, size_t pg_offset, u64 start, u64 len,
207 struct btrfs_fs_info *fs_info = inode->root->fs_info;
208 struct extent_map_tree *em_tree = &inode->extent_tree;
209 struct extent_map *em;
212 read_lock(&em_tree->lock);
213 em = lookup_extent_mapping(em_tree, start, len);
215 em->bdev = fs_info->fs_devices->latest_bdev;
216 read_unlock(&em_tree->lock);
219 read_unlock(&em_tree->lock);
221 em = alloc_extent_map();
223 em = ERR_PTR(-ENOMEM);
228 em->block_len = (u64)-1;
230 em->bdev = fs_info->fs_devices->latest_bdev;
232 write_lock(&em_tree->lock);
233 ret = add_extent_mapping(em_tree, em, 0);
234 if (ret == -EEXIST) {
236 em = lookup_extent_mapping(em_tree, start, len);
243 write_unlock(&em_tree->lock);
249 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
251 return crc32c(seed, data, len);
254 void btrfs_csum_final(u32 crc, u8 *result)
256 put_unaligned_le32(~crc, result);
260 * Compute the csum of a btree block and store the result to provided buffer.
262 * Returns error if the extent buffer cannot be mapped.
264 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
267 unsigned long cur_len;
268 unsigned long offset = BTRFS_CSUM_SIZE;
270 unsigned long map_start;
271 unsigned long map_len;
275 len = buf->len - offset;
278 * Note: we don't need to check for the err == 1 case here, as
279 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
280 * and 'min_len = 32' and the currently implemented mapping
281 * algorithm we cannot cross a page boundary.
283 err = map_private_extent_buffer(buf, offset, 32,
284 &kaddr, &map_start, &map_len);
287 cur_len = min(len, map_len - (offset - map_start));
288 crc = btrfs_csum_data(kaddr + offset - map_start,
293 memset(result, 0, BTRFS_CSUM_SIZE);
295 btrfs_csum_final(crc, result);
301 * we can't consider a given block up to date unless the transid of the
302 * block matches the transid in the parent node's pointer. This is how we
303 * detect blocks that either didn't get written at all or got written
304 * in the wrong place.
306 static int verify_parent_transid(struct extent_io_tree *io_tree,
307 struct extent_buffer *eb, u64 parent_transid,
310 struct extent_state *cached_state = NULL;
312 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
314 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
321 btrfs_tree_read_lock(eb);
322 btrfs_set_lock_blocking_read(eb);
325 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
327 if (extent_buffer_uptodate(eb) &&
328 btrfs_header_generation(eb) == parent_transid) {
332 btrfs_err_rl(eb->fs_info,
333 "parent transid verify failed on %llu wanted %llu found %llu",
335 parent_transid, btrfs_header_generation(eb));
339 * Things reading via commit roots that don't have normal protection,
340 * like send, can have a really old block in cache that may point at a
341 * block that has been freed and re-allocated. So don't clear uptodate
342 * if we find an eb that is under IO (dirty/writeback) because we could
343 * end up reading in the stale data and then writing it back out and
344 * making everybody very sad.
346 if (!extent_buffer_under_io(eb))
347 clear_extent_buffer_uptodate(eb);
349 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
352 btrfs_tree_read_unlock_blocking(eb);
356 static bool btrfs_supported_super_csum(u16 csum_type)
359 case BTRFS_CSUM_TYPE_CRC32:
367 * Return 0 if the superblock checksum type matches the checksum value of that
368 * algorithm. Pass the raw disk superblock data.
370 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
373 struct btrfs_super_block *disk_sb =
374 (struct btrfs_super_block *)raw_disk_sb;
376 char result[BTRFS_CSUM_SIZE];
379 * The super_block structure does not span the whole
380 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
381 * filled with zeros and is included in the checksum.
383 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
384 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
385 btrfs_csum_final(crc, result);
387 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
393 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
394 struct btrfs_key *first_key, u64 parent_transid)
396 struct btrfs_fs_info *fs_info = eb->fs_info;
398 struct btrfs_key found_key;
401 found_level = btrfs_header_level(eb);
402 if (found_level != level) {
403 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
404 KERN_ERR "BTRFS: tree level check failed\n");
406 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
407 eb->start, level, found_level);
415 * For live tree block (new tree blocks in current transaction),
416 * we need proper lock context to avoid race, which is impossible here.
417 * So we only checks tree blocks which is read from disk, whose
418 * generation <= fs_info->last_trans_committed.
420 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
423 btrfs_node_key_to_cpu(eb, &found_key, 0);
425 btrfs_item_key_to_cpu(eb, &found_key, 0);
426 ret = btrfs_comp_cpu_keys(first_key, &found_key);
429 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
430 KERN_ERR "BTRFS: tree first key check failed\n");
432 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
433 eb->start, parent_transid, first_key->objectid,
434 first_key->type, first_key->offset,
435 found_key.objectid, found_key.type,
442 * helper to read a given tree block, doing retries as required when
443 * the checksums don't match and we have alternate mirrors to try.
445 * @parent_transid: expected transid, skip check if 0
446 * @level: expected level, mandatory check
447 * @first_key: expected key of first slot, skip check if NULL
449 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
450 u64 parent_transid, int level,
451 struct btrfs_key *first_key)
453 struct btrfs_fs_info *fs_info = eb->fs_info;
454 struct extent_io_tree *io_tree;
459 int failed_mirror = 0;
461 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
463 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
464 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
466 if (verify_parent_transid(io_tree, eb,
469 else if (btrfs_verify_level_key(eb, level,
470 first_key, parent_transid))
476 num_copies = btrfs_num_copies(fs_info,
481 if (!failed_mirror) {
483 failed_mirror = eb->read_mirror;
487 if (mirror_num == failed_mirror)
490 if (mirror_num > num_copies)
494 if (failed && !ret && failed_mirror)
495 btrfs_repair_eb_io_failure(eb, failed_mirror);
501 * checksum a dirty tree block before IO. This has extra checks to make sure
502 * we only fill in the checksum field in the first page of a multi-page block
505 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
507 u64 start = page_offset(page);
509 u8 result[BTRFS_CSUM_SIZE];
510 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
511 struct extent_buffer *eb;
514 eb = (struct extent_buffer *)page->private;
515 if (page != eb->pages[0])
518 found_start = btrfs_header_bytenr(eb);
520 * Please do not consolidate these warnings into a single if.
521 * It is useful to know what went wrong.
523 if (WARN_ON(found_start != start))
525 if (WARN_ON(!PageUptodate(page)))
528 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
529 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
531 if (csum_tree_block(eb, result))
534 if (btrfs_header_level(eb))
535 ret = btrfs_check_node(eb);
537 ret = btrfs_check_leaf_full(eb);
541 "block=%llu write time tree block corruption detected",
545 write_extent_buffer(eb, result, 0, csum_size);
550 static int check_tree_block_fsid(struct extent_buffer *eb)
552 struct btrfs_fs_info *fs_info = eb->fs_info;
553 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
554 u8 fsid[BTRFS_FSID_SIZE];
557 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
562 * Checking the incompat flag is only valid for the current
563 * fs. For seed devices it's forbidden to have their uuid
564 * changed so reading ->fsid in this case is fine
566 if (fs_devices == fs_info->fs_devices &&
567 btrfs_fs_incompat(fs_info, METADATA_UUID))
568 metadata_uuid = fs_devices->metadata_uuid;
570 metadata_uuid = fs_devices->fsid;
572 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
576 fs_devices = fs_devices->seed;
581 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
582 u64 phy_offset, struct page *page,
583 u64 start, u64 end, int mirror)
587 struct extent_buffer *eb;
588 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
589 struct btrfs_fs_info *fs_info = root->fs_info;
590 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
592 u8 result[BTRFS_CSUM_SIZE];
598 eb = (struct extent_buffer *)page->private;
600 /* the pending IO might have been the only thing that kept this buffer
601 * in memory. Make sure we have a ref for all this other checks
603 extent_buffer_get(eb);
605 reads_done = atomic_dec_and_test(&eb->io_pages);
609 eb->read_mirror = mirror;
610 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
615 found_start = btrfs_header_bytenr(eb);
616 if (found_start != eb->start) {
617 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
618 eb->start, found_start);
622 if (check_tree_block_fsid(eb)) {
623 btrfs_err_rl(fs_info, "bad fsid on block %llu",
628 found_level = btrfs_header_level(eb);
629 if (found_level >= BTRFS_MAX_LEVEL) {
630 btrfs_err(fs_info, "bad tree block level %d on %llu",
631 (int)btrfs_header_level(eb), eb->start);
636 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
639 ret = csum_tree_block(eb, result);
643 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
647 memcpy(&found, result, csum_size);
649 read_extent_buffer(eb, &val, 0, csum_size);
650 btrfs_warn_rl(fs_info,
651 "%s checksum verify failed on %llu wanted %x found %x level %d",
652 fs_info->sb->s_id, eb->start,
653 val, found, btrfs_header_level(eb));
659 * If this is a leaf block and it is corrupt, set the corrupt bit so
660 * that we don't try and read the other copies of this block, just
663 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
664 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
668 if (found_level > 0 && btrfs_check_node(eb))
672 set_extent_buffer_uptodate(eb);
675 "block=%llu read time tree block corruption detected",
679 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
680 btree_readahead_hook(eb, ret);
684 * our io error hook is going to dec the io pages
685 * again, we have to make sure it has something
688 atomic_inc(&eb->io_pages);
689 clear_extent_buffer_uptodate(eb);
691 free_extent_buffer(eb);
696 static void end_workqueue_bio(struct bio *bio)
698 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
699 struct btrfs_fs_info *fs_info;
700 struct btrfs_workqueue *wq;
701 btrfs_work_func_t func;
703 fs_info = end_io_wq->info;
704 end_io_wq->status = bio->bi_status;
706 if (bio_op(bio) == REQ_OP_WRITE) {
707 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
708 wq = fs_info->endio_meta_write_workers;
709 func = btrfs_endio_meta_write_helper;
710 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
711 wq = fs_info->endio_freespace_worker;
712 func = btrfs_freespace_write_helper;
713 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
714 wq = fs_info->endio_raid56_workers;
715 func = btrfs_endio_raid56_helper;
717 wq = fs_info->endio_write_workers;
718 func = btrfs_endio_write_helper;
721 if (unlikely(end_io_wq->metadata ==
722 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
723 wq = fs_info->endio_repair_workers;
724 func = btrfs_endio_repair_helper;
725 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
726 wq = fs_info->endio_raid56_workers;
727 func = btrfs_endio_raid56_helper;
728 } else if (end_io_wq->metadata) {
729 wq = fs_info->endio_meta_workers;
730 func = btrfs_endio_meta_helper;
732 wq = fs_info->endio_workers;
733 func = btrfs_endio_helper;
737 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
738 btrfs_queue_work(wq, &end_io_wq->work);
741 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
742 enum btrfs_wq_endio_type metadata)
744 struct btrfs_end_io_wq *end_io_wq;
746 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
748 return BLK_STS_RESOURCE;
750 end_io_wq->private = bio->bi_private;
751 end_io_wq->end_io = bio->bi_end_io;
752 end_io_wq->info = info;
753 end_io_wq->status = 0;
754 end_io_wq->bio = bio;
755 end_io_wq->metadata = metadata;
757 bio->bi_private = end_io_wq;
758 bio->bi_end_io = end_workqueue_bio;
762 static void run_one_async_start(struct btrfs_work *work)
764 struct async_submit_bio *async;
767 async = container_of(work, struct async_submit_bio, work);
768 ret = async->submit_bio_start(async->private_data, async->bio,
775 * In order to insert checksums into the metadata in large chunks, we wait
776 * until bio submission time. All the pages in the bio are checksummed and
777 * sums are attached onto the ordered extent record.
779 * At IO completion time the csums attached on the ordered extent record are
780 * inserted into the tree.
782 static void run_one_async_done(struct btrfs_work *work)
784 struct async_submit_bio *async;
788 async = container_of(work, struct async_submit_bio, work);
789 inode = async->private_data;
791 /* If an error occurred we just want to clean up the bio and move on */
793 async->bio->bi_status = async->status;
794 bio_endio(async->bio);
798 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
799 async->mirror_num, 1);
801 async->bio->bi_status = ret;
802 bio_endio(async->bio);
806 static void run_one_async_free(struct btrfs_work *work)
808 struct async_submit_bio *async;
810 async = container_of(work, struct async_submit_bio, work);
814 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
815 int mirror_num, unsigned long bio_flags,
816 u64 bio_offset, void *private_data,
817 extent_submit_bio_start_t *submit_bio_start)
819 struct async_submit_bio *async;
821 async = kmalloc(sizeof(*async), GFP_NOFS);
823 return BLK_STS_RESOURCE;
825 async->private_data = private_data;
827 async->mirror_num = mirror_num;
828 async->submit_bio_start = submit_bio_start;
830 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
831 run_one_async_done, run_one_async_free);
833 async->bio_offset = bio_offset;
837 if (op_is_sync(bio->bi_opf))
838 btrfs_set_work_high_priority(&async->work);
840 btrfs_queue_work(fs_info->workers, &async->work);
844 static blk_status_t btree_csum_one_bio(struct bio *bio)
846 struct bio_vec *bvec;
847 struct btrfs_root *root;
849 struct bvec_iter_all iter_all;
851 ASSERT(!bio_flagged(bio, BIO_CLONED));
852 bio_for_each_segment_all(bvec, bio, iter_all) {
853 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
854 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
859 return errno_to_blk_status(ret);
862 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
866 * when we're called for a write, we're already in the async
867 * submission context. Just jump into btrfs_map_bio
869 return btree_csum_one_bio(bio);
872 static int check_async_write(struct btrfs_fs_info *fs_info,
873 struct btrfs_inode *bi)
875 if (atomic_read(&bi->sync_writers))
877 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
882 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
884 unsigned long bio_flags)
886 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
887 int async = check_async_write(fs_info, BTRFS_I(inode));
890 if (bio_op(bio) != REQ_OP_WRITE) {
892 * called for a read, do the setup so that checksum validation
893 * can happen in the async kernel threads
895 ret = btrfs_bio_wq_end_io(fs_info, bio,
896 BTRFS_WQ_ENDIO_METADATA);
899 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
901 ret = btree_csum_one_bio(bio);
904 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
907 * kthread helpers are used to submit writes so that
908 * checksumming can happen in parallel across all CPUs
910 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
911 0, inode, btree_submit_bio_start);
919 bio->bi_status = ret;
924 #ifdef CONFIG_MIGRATION
925 static int btree_migratepage(struct address_space *mapping,
926 struct page *newpage, struct page *page,
927 enum migrate_mode mode)
930 * we can't safely write a btree page from here,
931 * we haven't done the locking hook
936 * Buffers may be managed in a filesystem specific way.
937 * We must have no buffers or drop them.
939 if (page_has_private(page) &&
940 !try_to_release_page(page, GFP_KERNEL))
942 return migrate_page(mapping, newpage, page, mode);
947 static int btree_writepages(struct address_space *mapping,
948 struct writeback_control *wbc)
950 struct btrfs_fs_info *fs_info;
953 if (wbc->sync_mode == WB_SYNC_NONE) {
955 if (wbc->for_kupdate)
958 fs_info = BTRFS_I(mapping->host)->root->fs_info;
959 /* this is a bit racy, but that's ok */
960 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
961 BTRFS_DIRTY_METADATA_THRESH,
962 fs_info->dirty_metadata_batch);
966 return btree_write_cache_pages(mapping, wbc);
969 static int btree_readpage(struct file *file, struct page *page)
971 struct extent_io_tree *tree;
972 tree = &BTRFS_I(page->mapping->host)->io_tree;
973 return extent_read_full_page(tree, page, btree_get_extent, 0);
976 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
978 if (PageWriteback(page) || PageDirty(page))
981 return try_release_extent_buffer(page);
984 static void btree_invalidatepage(struct page *page, unsigned int offset,
987 struct extent_io_tree *tree;
988 tree = &BTRFS_I(page->mapping->host)->io_tree;
989 extent_invalidatepage(tree, page, offset);
990 btree_releasepage(page, GFP_NOFS);
991 if (PagePrivate(page)) {
992 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
993 "page private not zero on page %llu",
994 (unsigned long long)page_offset(page));
995 ClearPagePrivate(page);
996 set_page_private(page, 0);
1001 static int btree_set_page_dirty(struct page *page)
1004 struct extent_buffer *eb;
1006 BUG_ON(!PagePrivate(page));
1007 eb = (struct extent_buffer *)page->private;
1009 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1010 BUG_ON(!atomic_read(&eb->refs));
1011 btrfs_assert_tree_locked(eb);
1013 return __set_page_dirty_nobuffers(page);
1016 static const struct address_space_operations btree_aops = {
1017 .readpage = btree_readpage,
1018 .writepages = btree_writepages,
1019 .releasepage = btree_releasepage,
1020 .invalidatepage = btree_invalidatepage,
1021 #ifdef CONFIG_MIGRATION
1022 .migratepage = btree_migratepage,
1024 .set_page_dirty = btree_set_page_dirty,
1027 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1029 struct extent_buffer *buf = NULL;
1032 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1036 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1038 free_extent_buffer_stale(buf);
1040 free_extent_buffer(buf);
1043 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1044 int mirror_num, struct extent_buffer **eb)
1046 struct extent_buffer *buf = NULL;
1049 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1053 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1055 ret = read_extent_buffer_pages(buf, WAIT_PAGE_LOCK, mirror_num);
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(buf, parent_transid,
1103 free_extent_buffer_stale(buf);
1104 return ERR_PTR(ret);
1110 void btrfs_clean_tree_block(struct extent_buffer *buf)
1112 struct btrfs_fs_info *fs_info = buf->fs_info;
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,
1258 struct btrfs_fs_info *fs_info = trans->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 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2165 fs_info->qgroup_seq = 1;
2166 fs_info->qgroup_ulist = NULL;
2167 fs_info->qgroup_rescan_running = false;
2168 mutex_init(&fs_info->qgroup_rescan_lock);
2171 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2172 struct btrfs_fs_devices *fs_devices)
2174 u32 max_active = fs_info->thread_pool_size;
2175 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2178 btrfs_alloc_workqueue(fs_info, "worker",
2179 flags | WQ_HIGHPRI, max_active, 16);
2181 fs_info->delalloc_workers =
2182 btrfs_alloc_workqueue(fs_info, "delalloc",
2183 flags, max_active, 2);
2185 fs_info->flush_workers =
2186 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2187 flags, max_active, 0);
2189 fs_info->caching_workers =
2190 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2193 * a higher idle thresh on the submit workers makes it much more
2194 * likely that bios will be send down in a sane order to the
2197 fs_info->submit_workers =
2198 btrfs_alloc_workqueue(fs_info, "submit", flags,
2199 min_t(u64, fs_devices->num_devices,
2202 fs_info->fixup_workers =
2203 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2206 * endios are largely parallel and should have a very
2209 fs_info->endio_workers =
2210 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2211 fs_info->endio_meta_workers =
2212 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2214 fs_info->endio_meta_write_workers =
2215 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2217 fs_info->endio_raid56_workers =
2218 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2220 fs_info->endio_repair_workers =
2221 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2222 fs_info->rmw_workers =
2223 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2224 fs_info->endio_write_workers =
2225 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2227 fs_info->endio_freespace_worker =
2228 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2230 fs_info->delayed_workers =
2231 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2233 fs_info->readahead_workers =
2234 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2236 fs_info->qgroup_rescan_workers =
2237 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2238 fs_info->extent_workers =
2239 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2240 min_t(u64, fs_devices->num_devices,
2243 if (!(fs_info->workers && fs_info->delalloc_workers &&
2244 fs_info->submit_workers && fs_info->flush_workers &&
2245 fs_info->endio_workers && fs_info->endio_meta_workers &&
2246 fs_info->endio_meta_write_workers &&
2247 fs_info->endio_repair_workers &&
2248 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2249 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2250 fs_info->caching_workers && fs_info->readahead_workers &&
2251 fs_info->fixup_workers && fs_info->delayed_workers &&
2252 fs_info->extent_workers &&
2253 fs_info->qgroup_rescan_workers)) {
2260 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2262 struct crypto_shash *csum_shash;
2263 const char *csum_name = btrfs_super_csum_name(csum_type);
2265 csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2267 if (IS_ERR(csum_shash)) {
2268 btrfs_err(fs_info, "error allocating %s hash for checksum",
2270 return PTR_ERR(csum_shash);
2273 fs_info->csum_shash = csum_shash;
2278 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2280 crypto_free_shash(fs_info->csum_shash);
2283 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2284 struct btrfs_fs_devices *fs_devices)
2287 struct btrfs_root *log_tree_root;
2288 struct btrfs_super_block *disk_super = fs_info->super_copy;
2289 u64 bytenr = btrfs_super_log_root(disk_super);
2290 int level = btrfs_super_log_root_level(disk_super);
2292 if (fs_devices->rw_devices == 0) {
2293 btrfs_warn(fs_info, "log replay required on RO media");
2297 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2301 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2303 log_tree_root->node = read_tree_block(fs_info, bytenr,
2304 fs_info->generation + 1,
2306 if (IS_ERR(log_tree_root->node)) {
2307 btrfs_warn(fs_info, "failed to read log tree");
2308 ret = PTR_ERR(log_tree_root->node);
2309 kfree(log_tree_root);
2311 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2312 btrfs_err(fs_info, "failed to read log tree");
2313 free_extent_buffer(log_tree_root->node);
2314 kfree(log_tree_root);
2317 /* returns with log_tree_root freed on success */
2318 ret = btrfs_recover_log_trees(log_tree_root);
2320 btrfs_handle_fs_error(fs_info, ret,
2321 "Failed to recover log tree");
2322 free_extent_buffer(log_tree_root->node);
2323 kfree(log_tree_root);
2327 if (sb_rdonly(fs_info->sb)) {
2328 ret = btrfs_commit_super(fs_info);
2336 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2338 struct btrfs_root *tree_root = fs_info->tree_root;
2339 struct btrfs_root *root;
2340 struct btrfs_key location;
2343 BUG_ON(!fs_info->tree_root);
2345 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2346 location.type = BTRFS_ROOT_ITEM_KEY;
2347 location.offset = 0;
2349 root = btrfs_read_tree_root(tree_root, &location);
2351 ret = PTR_ERR(root);
2354 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2355 fs_info->extent_root = root;
2357 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2358 root = btrfs_read_tree_root(tree_root, &location);
2360 ret = PTR_ERR(root);
2363 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2364 fs_info->dev_root = root;
2365 btrfs_init_devices_late(fs_info);
2367 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2368 root = btrfs_read_tree_root(tree_root, &location);
2370 ret = PTR_ERR(root);
2373 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2374 fs_info->csum_root = root;
2376 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2377 root = btrfs_read_tree_root(tree_root, &location);
2378 if (!IS_ERR(root)) {
2379 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2380 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2381 fs_info->quota_root = root;
2384 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2385 root = btrfs_read_tree_root(tree_root, &location);
2387 ret = PTR_ERR(root);
2391 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2392 fs_info->uuid_root = root;
2395 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2396 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2397 root = btrfs_read_tree_root(tree_root, &location);
2399 ret = PTR_ERR(root);
2402 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2403 fs_info->free_space_root = root;
2408 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2409 location.objectid, ret);
2414 * Real super block validation
2415 * NOTE: super csum type and incompat features will not be checked here.
2417 * @sb: super block to check
2418 * @mirror_num: the super block number to check its bytenr:
2419 * 0 the primary (1st) sb
2420 * 1, 2 2nd and 3rd backup copy
2421 * -1 skip bytenr check
2423 static int validate_super(struct btrfs_fs_info *fs_info,
2424 struct btrfs_super_block *sb, int mirror_num)
2426 u64 nodesize = btrfs_super_nodesize(sb);
2427 u64 sectorsize = btrfs_super_sectorsize(sb);
2430 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2431 btrfs_err(fs_info, "no valid FS found");
2434 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2435 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2436 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2439 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2440 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2441 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2444 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2445 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2446 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2449 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2450 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2451 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2456 * Check sectorsize and nodesize first, other check will need it.
2457 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2459 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2460 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2461 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2464 /* Only PAGE SIZE is supported yet */
2465 if (sectorsize != PAGE_SIZE) {
2467 "sectorsize %llu not supported yet, only support %lu",
2468 sectorsize, PAGE_SIZE);
2471 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2472 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2473 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2476 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2477 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2478 le32_to_cpu(sb->__unused_leafsize), nodesize);
2482 /* Root alignment check */
2483 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2484 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2485 btrfs_super_root(sb));
2488 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2489 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2490 btrfs_super_chunk_root(sb));
2493 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2494 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2495 btrfs_super_log_root(sb));
2499 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2500 BTRFS_FSID_SIZE) != 0) {
2502 "dev_item UUID does not match metadata fsid: %pU != %pU",
2503 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2508 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2511 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2512 btrfs_err(fs_info, "bytes_used is too small %llu",
2513 btrfs_super_bytes_used(sb));
2516 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2517 btrfs_err(fs_info, "invalid stripesize %u",
2518 btrfs_super_stripesize(sb));
2521 if (btrfs_super_num_devices(sb) > (1UL << 31))
2522 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2523 btrfs_super_num_devices(sb));
2524 if (btrfs_super_num_devices(sb) == 0) {
2525 btrfs_err(fs_info, "number of devices is 0");
2529 if (mirror_num >= 0 &&
2530 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2531 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2532 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2537 * Obvious sys_chunk_array corruptions, it must hold at least one key
2540 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2541 btrfs_err(fs_info, "system chunk array too big %u > %u",
2542 btrfs_super_sys_array_size(sb),
2543 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2546 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2547 + sizeof(struct btrfs_chunk)) {
2548 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2549 btrfs_super_sys_array_size(sb),
2550 sizeof(struct btrfs_disk_key)
2551 + sizeof(struct btrfs_chunk));
2556 * The generation is a global counter, we'll trust it more than the others
2557 * but it's still possible that it's the one that's wrong.
2559 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2561 "suspicious: generation < chunk_root_generation: %llu < %llu",
2562 btrfs_super_generation(sb),
2563 btrfs_super_chunk_root_generation(sb));
2564 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2565 && btrfs_super_cache_generation(sb) != (u64)-1)
2567 "suspicious: generation < cache_generation: %llu < %llu",
2568 btrfs_super_generation(sb),
2569 btrfs_super_cache_generation(sb));
2575 * Validation of super block at mount time.
2576 * Some checks already done early at mount time, like csum type and incompat
2577 * flags will be skipped.
2579 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2581 return validate_super(fs_info, fs_info->super_copy, 0);
2585 * Validation of super block at write time.
2586 * Some checks like bytenr check will be skipped as their values will be
2588 * Extra checks like csum type and incompat flags will be done here.
2590 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2591 struct btrfs_super_block *sb)
2595 ret = validate_super(fs_info, sb, -1);
2598 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2600 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2601 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2604 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2607 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2608 btrfs_super_incompat_flags(sb),
2609 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2615 "super block corruption detected before writing it to disk");
2619 int open_ctree(struct super_block *sb,
2620 struct btrfs_fs_devices *fs_devices,
2629 struct btrfs_key location;
2630 struct buffer_head *bh;
2631 struct btrfs_super_block *disk_super;
2632 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2633 struct btrfs_root *tree_root;
2634 struct btrfs_root *chunk_root;
2637 int num_backups_tried = 0;
2638 int backup_index = 0;
2639 int clear_free_space_tree = 0;
2642 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2643 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2644 if (!tree_root || !chunk_root) {
2649 ret = init_srcu_struct(&fs_info->subvol_srcu);
2655 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2661 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2664 goto fail_dio_bytes;
2666 fs_info->dirty_metadata_batch = PAGE_SIZE *
2667 (1 + ilog2(nr_cpu_ids));
2669 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2672 goto fail_dirty_metadata_bytes;
2675 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2679 goto fail_delalloc_bytes;
2682 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2683 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2684 INIT_LIST_HEAD(&fs_info->trans_list);
2685 INIT_LIST_HEAD(&fs_info->dead_roots);
2686 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2687 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2688 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2689 INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2690 spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2691 spin_lock_init(&fs_info->delalloc_root_lock);
2692 spin_lock_init(&fs_info->trans_lock);
2693 spin_lock_init(&fs_info->fs_roots_radix_lock);
2694 spin_lock_init(&fs_info->delayed_iput_lock);
2695 spin_lock_init(&fs_info->defrag_inodes_lock);
2696 spin_lock_init(&fs_info->tree_mod_seq_lock);
2697 spin_lock_init(&fs_info->super_lock);
2698 spin_lock_init(&fs_info->buffer_lock);
2699 spin_lock_init(&fs_info->unused_bgs_lock);
2700 rwlock_init(&fs_info->tree_mod_log_lock);
2701 mutex_init(&fs_info->unused_bg_unpin_mutex);
2702 mutex_init(&fs_info->delete_unused_bgs_mutex);
2703 mutex_init(&fs_info->reloc_mutex);
2704 mutex_init(&fs_info->delalloc_root_mutex);
2705 seqlock_init(&fs_info->profiles_lock);
2707 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2708 INIT_LIST_HEAD(&fs_info->space_info);
2709 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2710 INIT_LIST_HEAD(&fs_info->unused_bgs);
2711 extent_map_tree_init(&fs_info->mapping_tree);
2712 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2713 BTRFS_BLOCK_RSV_GLOBAL);
2714 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2715 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2716 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2717 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2718 BTRFS_BLOCK_RSV_DELOPS);
2719 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2720 BTRFS_BLOCK_RSV_DELREFS);
2722 atomic_set(&fs_info->async_delalloc_pages, 0);
2723 atomic_set(&fs_info->defrag_running, 0);
2724 atomic_set(&fs_info->reada_works_cnt, 0);
2725 atomic_set(&fs_info->nr_delayed_iputs, 0);
2726 atomic64_set(&fs_info->tree_mod_seq, 0);
2728 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2729 fs_info->metadata_ratio = 0;
2730 fs_info->defrag_inodes = RB_ROOT;
2731 atomic64_set(&fs_info->free_chunk_space, 0);
2732 fs_info->tree_mod_log = RB_ROOT;
2733 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2734 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2735 /* readahead state */
2736 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2737 spin_lock_init(&fs_info->reada_lock);
2738 btrfs_init_ref_verify(fs_info);
2740 fs_info->thread_pool_size = min_t(unsigned long,
2741 num_online_cpus() + 2, 8);
2743 INIT_LIST_HEAD(&fs_info->ordered_roots);
2744 spin_lock_init(&fs_info->ordered_root_lock);
2746 fs_info->btree_inode = new_inode(sb);
2747 if (!fs_info->btree_inode) {
2749 goto fail_bio_counter;
2751 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2753 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2755 if (!fs_info->delayed_root) {
2759 btrfs_init_delayed_root(fs_info->delayed_root);
2761 btrfs_init_scrub(fs_info);
2762 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2763 fs_info->check_integrity_print_mask = 0;
2765 btrfs_init_balance(fs_info);
2766 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2768 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2769 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2771 btrfs_init_btree_inode(fs_info);
2773 spin_lock_init(&fs_info->block_group_cache_lock);
2774 fs_info->block_group_cache_tree = RB_ROOT;
2775 fs_info->first_logical_byte = (u64)-1;
2777 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2778 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2779 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2780 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2781 fs_info->pinned_extents = &fs_info->freed_extents[0];
2782 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2784 mutex_init(&fs_info->ordered_operations_mutex);
2785 mutex_init(&fs_info->tree_log_mutex);
2786 mutex_init(&fs_info->chunk_mutex);
2787 mutex_init(&fs_info->transaction_kthread_mutex);
2788 mutex_init(&fs_info->cleaner_mutex);
2789 mutex_init(&fs_info->ro_block_group_mutex);
2790 init_rwsem(&fs_info->commit_root_sem);
2791 init_rwsem(&fs_info->cleanup_work_sem);
2792 init_rwsem(&fs_info->subvol_sem);
2793 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2795 btrfs_init_dev_replace_locks(fs_info);
2796 btrfs_init_qgroup(fs_info);
2798 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2799 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2801 init_waitqueue_head(&fs_info->transaction_throttle);
2802 init_waitqueue_head(&fs_info->transaction_wait);
2803 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2804 init_waitqueue_head(&fs_info->async_submit_wait);
2805 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2807 /* Usable values until the real ones are cached from the superblock */
2808 fs_info->nodesize = 4096;
2809 fs_info->sectorsize = 4096;
2810 fs_info->stripesize = 4096;
2812 spin_lock_init(&fs_info->swapfile_pins_lock);
2813 fs_info->swapfile_pins = RB_ROOT;
2815 ret = btrfs_alloc_stripe_hash_table(fs_info);
2821 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2823 invalidate_bdev(fs_devices->latest_bdev);
2826 * Read super block and check the signature bytes only
2828 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2835 * Verify the type first, if that or the the checksum value are
2836 * corrupted, we'll find out
2838 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2839 if (!btrfs_supported_super_csum(csum_type)) {
2840 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2847 ret = btrfs_init_csum_hash(fs_info, csum_type);
2854 * We want to check superblock checksum, the type is stored inside.
2855 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2857 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2858 btrfs_err(fs_info, "superblock checksum mismatch");
2865 * super_copy is zeroed at allocation time and we never touch the
2866 * following bytes up to INFO_SIZE, the checksum is calculated from
2867 * the whole block of INFO_SIZE
2869 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2872 disk_super = fs_info->super_copy;
2874 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2877 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2878 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2879 fs_info->super_copy->metadata_uuid,
2883 features = btrfs_super_flags(disk_super);
2884 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2885 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2886 btrfs_set_super_flags(disk_super, features);
2888 "found metadata UUID change in progress flag, clearing");
2891 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2892 sizeof(*fs_info->super_for_commit));
2894 ret = btrfs_validate_mount_super(fs_info);
2896 btrfs_err(fs_info, "superblock contains fatal errors");
2901 if (!btrfs_super_root(disk_super))
2904 /* check FS state, whether FS is broken. */
2905 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2906 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2909 * run through our array of backup supers and setup
2910 * our ring pointer to the oldest one
2912 generation = btrfs_super_generation(disk_super);
2913 find_oldest_super_backup(fs_info, generation);
2916 * In the long term, we'll store the compression type in the super
2917 * block, and it'll be used for per file compression control.
2919 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2921 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2927 features = btrfs_super_incompat_flags(disk_super) &
2928 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2931 "cannot mount because of unsupported optional features (%llx)",
2937 features = btrfs_super_incompat_flags(disk_super);
2938 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2939 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2940 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2941 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2942 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2944 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2945 btrfs_info(fs_info, "has skinny extents");
2948 * flag our filesystem as having big metadata blocks if
2949 * they are bigger than the page size
2951 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2952 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2954 "flagging fs with big metadata feature");
2955 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2958 nodesize = btrfs_super_nodesize(disk_super);
2959 sectorsize = btrfs_super_sectorsize(disk_super);
2960 stripesize = sectorsize;
2961 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2962 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2964 /* Cache block sizes */
2965 fs_info->nodesize = nodesize;
2966 fs_info->sectorsize = sectorsize;
2967 fs_info->stripesize = stripesize;
2970 * mixed block groups end up with duplicate but slightly offset
2971 * extent buffers for the same range. It leads to corruptions
2973 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2974 (sectorsize != nodesize)) {
2976 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2977 nodesize, sectorsize);
2982 * Needn't use the lock because there is no other task which will
2985 btrfs_set_super_incompat_flags(disk_super, features);
2987 features = btrfs_super_compat_ro_flags(disk_super) &
2988 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2989 if (!sb_rdonly(sb) && features) {
2991 "cannot mount read-write because of unsupported optional features (%llx)",
2997 ret = btrfs_init_workqueues(fs_info, fs_devices);
3000 goto fail_sb_buffer;
3003 sb->s_bdi->congested_fn = btrfs_congested_fn;
3004 sb->s_bdi->congested_data = fs_info;
3005 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3006 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3007 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3008 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3010 sb->s_blocksize = sectorsize;
3011 sb->s_blocksize_bits = blksize_bits(sectorsize);
3012 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3014 mutex_lock(&fs_info->chunk_mutex);
3015 ret = btrfs_read_sys_array(fs_info);
3016 mutex_unlock(&fs_info->chunk_mutex);
3018 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3019 goto fail_sb_buffer;
3022 generation = btrfs_super_chunk_root_generation(disk_super);
3023 level = btrfs_super_chunk_root_level(disk_super);
3025 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
3027 chunk_root->node = read_tree_block(fs_info,
3028 btrfs_super_chunk_root(disk_super),
3029 generation, level, NULL);
3030 if (IS_ERR(chunk_root->node) ||
3031 !extent_buffer_uptodate(chunk_root->node)) {
3032 btrfs_err(fs_info, "failed to read chunk root");
3033 if (!IS_ERR(chunk_root->node))
3034 free_extent_buffer(chunk_root->node);
3035 chunk_root->node = NULL;
3036 goto fail_tree_roots;
3038 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3039 chunk_root->commit_root = btrfs_root_node(chunk_root);
3041 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3042 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3044 ret = btrfs_read_chunk_tree(fs_info);
3046 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3047 goto fail_tree_roots;
3051 * Keep the devid that is marked to be the target device for the
3052 * device replace procedure
3054 btrfs_free_extra_devids(fs_devices, 0);
3056 if (!fs_devices->latest_bdev) {
3057 btrfs_err(fs_info, "failed to read devices");
3058 goto fail_tree_roots;
3062 generation = btrfs_super_generation(disk_super);
3063 level = btrfs_super_root_level(disk_super);
3065 tree_root->node = read_tree_block(fs_info,
3066 btrfs_super_root(disk_super),
3067 generation, level, NULL);
3068 if (IS_ERR(tree_root->node) ||
3069 !extent_buffer_uptodate(tree_root->node)) {
3070 btrfs_warn(fs_info, "failed to read tree root");
3071 if (!IS_ERR(tree_root->node))
3072 free_extent_buffer(tree_root->node);
3073 tree_root->node = NULL;
3074 goto recovery_tree_root;
3077 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3078 tree_root->commit_root = btrfs_root_node(tree_root);
3079 btrfs_set_root_refs(&tree_root->root_item, 1);
3081 mutex_lock(&tree_root->objectid_mutex);
3082 ret = btrfs_find_highest_objectid(tree_root,
3083 &tree_root->highest_objectid);
3085 mutex_unlock(&tree_root->objectid_mutex);
3086 goto recovery_tree_root;
3089 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3091 mutex_unlock(&tree_root->objectid_mutex);
3093 ret = btrfs_read_roots(fs_info);
3095 goto recovery_tree_root;
3097 fs_info->generation = generation;
3098 fs_info->last_trans_committed = generation;
3100 ret = btrfs_verify_dev_extents(fs_info);
3103 "failed to verify dev extents against chunks: %d",
3105 goto fail_block_groups;
3107 ret = btrfs_recover_balance(fs_info);
3109 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3110 goto fail_block_groups;
3113 ret = btrfs_init_dev_stats(fs_info);
3115 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3116 goto fail_block_groups;
3119 ret = btrfs_init_dev_replace(fs_info);
3121 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3122 goto fail_block_groups;
3125 btrfs_free_extra_devids(fs_devices, 1);
3127 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3129 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3131 goto fail_block_groups;
3134 ret = btrfs_sysfs_add_device(fs_devices);
3136 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3138 goto fail_fsdev_sysfs;
3141 ret = btrfs_sysfs_add_mounted(fs_info);
3143 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3144 goto fail_fsdev_sysfs;
3147 ret = btrfs_init_space_info(fs_info);
3149 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3153 ret = btrfs_read_block_groups(fs_info);
3155 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3159 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3161 "writable mount is not allowed due to too many missing devices");
3165 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3167 if (IS_ERR(fs_info->cleaner_kthread))
3170 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3172 "btrfs-transaction");
3173 if (IS_ERR(fs_info->transaction_kthread))
3176 if (!btrfs_test_opt(fs_info, NOSSD) &&
3177 !fs_info->fs_devices->rotating) {
3178 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3182 * Mount does not set all options immediately, we can do it now and do
3183 * not have to wait for transaction commit
3185 btrfs_apply_pending_changes(fs_info);
3187 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3188 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3189 ret = btrfsic_mount(fs_info, fs_devices,
3190 btrfs_test_opt(fs_info,
3191 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3193 fs_info->check_integrity_print_mask);
3196 "failed to initialize integrity check module: %d",
3200 ret = btrfs_read_qgroup_config(fs_info);
3202 goto fail_trans_kthread;
3204 if (btrfs_build_ref_tree(fs_info))
3205 btrfs_err(fs_info, "couldn't build ref tree");
3207 /* do not make disk changes in broken FS or nologreplay is given */
3208 if (btrfs_super_log_root(disk_super) != 0 &&
3209 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3210 ret = btrfs_replay_log(fs_info, fs_devices);
3217 ret = btrfs_find_orphan_roots(fs_info);
3221 if (!sb_rdonly(sb)) {
3222 ret = btrfs_cleanup_fs_roots(fs_info);
3226 mutex_lock(&fs_info->cleaner_mutex);
3227 ret = btrfs_recover_relocation(tree_root);
3228 mutex_unlock(&fs_info->cleaner_mutex);
3230 btrfs_warn(fs_info, "failed to recover relocation: %d",
3237 location.objectid = BTRFS_FS_TREE_OBJECTID;
3238 location.type = BTRFS_ROOT_ITEM_KEY;
3239 location.offset = 0;
3241 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3242 if (IS_ERR(fs_info->fs_root)) {
3243 err = PTR_ERR(fs_info->fs_root);
3244 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3251 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3252 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3253 clear_free_space_tree = 1;
3254 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3255 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3256 btrfs_warn(fs_info, "free space tree is invalid");
3257 clear_free_space_tree = 1;
3260 if (clear_free_space_tree) {
3261 btrfs_info(fs_info, "clearing free space tree");
3262 ret = btrfs_clear_free_space_tree(fs_info);
3265 "failed to clear free space tree: %d", ret);
3266 close_ctree(fs_info);
3271 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3272 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3273 btrfs_info(fs_info, "creating free space tree");
3274 ret = btrfs_create_free_space_tree(fs_info);
3277 "failed to create free space tree: %d", ret);
3278 close_ctree(fs_info);
3283 down_read(&fs_info->cleanup_work_sem);
3284 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3285 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3286 up_read(&fs_info->cleanup_work_sem);
3287 close_ctree(fs_info);
3290 up_read(&fs_info->cleanup_work_sem);
3292 ret = btrfs_resume_balance_async(fs_info);
3294 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3295 close_ctree(fs_info);
3299 ret = btrfs_resume_dev_replace_async(fs_info);
3301 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3302 close_ctree(fs_info);
3306 btrfs_qgroup_rescan_resume(fs_info);
3308 if (!fs_info->uuid_root) {
3309 btrfs_info(fs_info, "creating UUID tree");
3310 ret = btrfs_create_uuid_tree(fs_info);
3313 "failed to create the UUID tree: %d", ret);
3314 close_ctree(fs_info);
3317 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3318 fs_info->generation !=
3319 btrfs_super_uuid_tree_generation(disk_super)) {
3320 btrfs_info(fs_info, "checking UUID tree");
3321 ret = btrfs_check_uuid_tree(fs_info);
3324 "failed to check the UUID tree: %d", ret);
3325 close_ctree(fs_info);
3329 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3331 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3334 * backuproot only affect mount behavior, and if open_ctree succeeded,
3335 * no need to keep the flag
3337 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3342 btrfs_free_qgroup_config(fs_info);
3344 kthread_stop(fs_info->transaction_kthread);
3345 btrfs_cleanup_transaction(fs_info);
3346 btrfs_free_fs_roots(fs_info);
3348 kthread_stop(fs_info->cleaner_kthread);
3351 * make sure we're done with the btree inode before we stop our
3354 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3357 btrfs_sysfs_remove_mounted(fs_info);
3360 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3363 btrfs_put_block_group_cache(fs_info);
3366 free_root_pointers(fs_info, 1);
3367 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3370 btrfs_stop_all_workers(fs_info);
3371 btrfs_free_block_groups(fs_info);
3373 btrfs_free_csum_hash(fs_info);
3376 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3378 iput(fs_info->btree_inode);
3380 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3381 fail_delalloc_bytes:
3382 percpu_counter_destroy(&fs_info->delalloc_bytes);
3383 fail_dirty_metadata_bytes:
3384 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3386 percpu_counter_destroy(&fs_info->dio_bytes);
3388 cleanup_srcu_struct(&fs_info->subvol_srcu);
3390 btrfs_free_stripe_hash_table(fs_info);
3391 btrfs_close_devices(fs_info->fs_devices);
3395 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3396 goto fail_tree_roots;
3398 free_root_pointers(fs_info, 0);
3400 /* don't use the log in recovery mode, it won't be valid */
3401 btrfs_set_super_log_root(disk_super, 0);
3403 /* we can't trust the free space cache either */
3404 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3406 ret = next_root_backup(fs_info, fs_info->super_copy,
3407 &num_backups_tried, &backup_index);
3409 goto fail_block_groups;
3410 goto retry_root_backup;
3412 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3414 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3417 set_buffer_uptodate(bh);
3419 struct btrfs_device *device = (struct btrfs_device *)
3422 btrfs_warn_rl_in_rcu(device->fs_info,
3423 "lost page write due to IO error on %s",
3424 rcu_str_deref(device->name));
3425 /* note, we don't set_buffer_write_io_error because we have
3426 * our own ways of dealing with the IO errors
3428 clear_buffer_uptodate(bh);
3429 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3435 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3436 struct buffer_head **bh_ret)
3438 struct buffer_head *bh;
3439 struct btrfs_super_block *super;
3442 bytenr = btrfs_sb_offset(copy_num);
3443 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3446 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3448 * If we fail to read from the underlying devices, as of now
3449 * the best option we have is to mark it EIO.
3454 super = (struct btrfs_super_block *)bh->b_data;
3455 if (btrfs_super_bytenr(super) != bytenr ||
3456 btrfs_super_magic(super) != BTRFS_MAGIC) {
3466 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3468 struct buffer_head *bh;
3469 struct buffer_head *latest = NULL;
3470 struct btrfs_super_block *super;
3475 /* we would like to check all the supers, but that would make
3476 * a btrfs mount succeed after a mkfs from a different FS.
3477 * So, we need to add a special mount option to scan for
3478 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3480 for (i = 0; i < 1; i++) {
3481 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3485 super = (struct btrfs_super_block *)bh->b_data;
3487 if (!latest || btrfs_super_generation(super) > transid) {
3490 transid = btrfs_super_generation(super);
3497 return ERR_PTR(ret);
3503 * Write superblock @sb to the @device. Do not wait for completion, all the
3504 * buffer heads we write are pinned.
3506 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3507 * the expected device size at commit time. Note that max_mirrors must be
3508 * same for write and wait phases.
3510 * Return number of errors when buffer head is not found or submission fails.
3512 static int write_dev_supers(struct btrfs_device *device,
3513 struct btrfs_super_block *sb, int max_mirrors)
3515 struct buffer_head *bh;
3523 if (max_mirrors == 0)
3524 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3526 for (i = 0; i < max_mirrors; i++) {
3527 bytenr = btrfs_sb_offset(i);
3528 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3529 device->commit_total_bytes)
3532 btrfs_set_super_bytenr(sb, bytenr);
3535 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3536 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3537 btrfs_csum_final(crc, sb->csum);
3539 /* One reference for us, and we leave it for the caller */
3540 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3541 BTRFS_SUPER_INFO_SIZE);
3543 btrfs_err(device->fs_info,
3544 "couldn't get super buffer head for bytenr %llu",
3550 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3552 /* one reference for submit_bh */
3555 set_buffer_uptodate(bh);
3557 bh->b_end_io = btrfs_end_buffer_write_sync;
3558 bh->b_private = device;
3561 * we fua the first super. The others we allow
3564 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3565 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3566 op_flags |= REQ_FUA;
3567 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3571 return errors < i ? 0 : -1;
3575 * Wait for write completion of superblocks done by write_dev_supers,
3576 * @max_mirrors same for write and wait phases.
3578 * Return number of errors when buffer head is not found or not marked up to
3581 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3583 struct buffer_head *bh;
3586 bool primary_failed = false;
3589 if (max_mirrors == 0)
3590 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3592 for (i = 0; i < max_mirrors; i++) {
3593 bytenr = btrfs_sb_offset(i);
3594 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3595 device->commit_total_bytes)
3598 bh = __find_get_block(device->bdev,
3599 bytenr / BTRFS_BDEV_BLOCKSIZE,
3600 BTRFS_SUPER_INFO_SIZE);
3604 primary_failed = true;
3608 if (!buffer_uptodate(bh)) {
3611 primary_failed = true;
3614 /* drop our reference */
3617 /* drop the reference from the writing run */
3621 /* log error, force error return */
3622 if (primary_failed) {
3623 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3628 return errors < i ? 0 : -1;
3632 * endio for the write_dev_flush, this will wake anyone waiting
3633 * for the barrier when it is done
3635 static void btrfs_end_empty_barrier(struct bio *bio)
3637 complete(bio->bi_private);
3641 * Submit a flush request to the device if it supports it. Error handling is
3642 * done in the waiting counterpart.
3644 static void write_dev_flush(struct btrfs_device *device)
3646 struct request_queue *q = bdev_get_queue(device->bdev);
3647 struct bio *bio = device->flush_bio;
3649 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3653 bio->bi_end_io = btrfs_end_empty_barrier;
3654 bio_set_dev(bio, device->bdev);
3655 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3656 init_completion(&device->flush_wait);
3657 bio->bi_private = &device->flush_wait;
3659 btrfsic_submit_bio(bio);
3660 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3664 * If the flush bio has been submitted by write_dev_flush, wait for it.
3666 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3668 struct bio *bio = device->flush_bio;
3670 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3673 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3674 wait_for_completion_io(&device->flush_wait);
3676 return bio->bi_status;
3679 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3681 if (!btrfs_check_rw_degradable(fs_info, NULL))
3687 * send an empty flush down to each device in parallel,
3688 * then wait for them
3690 static int barrier_all_devices(struct btrfs_fs_info *info)
3692 struct list_head *head;
3693 struct btrfs_device *dev;
3694 int errors_wait = 0;
3697 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3698 /* send down all the barriers */
3699 head = &info->fs_devices->devices;
3700 list_for_each_entry(dev, head, dev_list) {
3701 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3705 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3706 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3709 write_dev_flush(dev);
3710 dev->last_flush_error = BLK_STS_OK;
3713 /* wait for all the barriers */
3714 list_for_each_entry(dev, head, dev_list) {
3715 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3721 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3722 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3725 ret = wait_dev_flush(dev);
3727 dev->last_flush_error = ret;
3728 btrfs_dev_stat_inc_and_print(dev,
3729 BTRFS_DEV_STAT_FLUSH_ERRS);
3736 * At some point we need the status of all disks
3737 * to arrive at the volume status. So error checking
3738 * is being pushed to a separate loop.
3740 return check_barrier_error(info);
3745 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3748 int min_tolerated = INT_MAX;
3750 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3751 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3752 min_tolerated = min_t(int, min_tolerated,
3753 btrfs_raid_array[BTRFS_RAID_SINGLE].
3754 tolerated_failures);
3756 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3757 if (raid_type == BTRFS_RAID_SINGLE)
3759 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3761 min_tolerated = min_t(int, min_tolerated,
3762 btrfs_raid_array[raid_type].
3763 tolerated_failures);
3766 if (min_tolerated == INT_MAX) {
3767 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3771 return min_tolerated;
3774 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3776 struct list_head *head;
3777 struct btrfs_device *dev;
3778 struct btrfs_super_block *sb;
3779 struct btrfs_dev_item *dev_item;
3783 int total_errors = 0;
3786 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3789 * max_mirrors == 0 indicates we're from commit_transaction,
3790 * not from fsync where the tree roots in fs_info have not
3791 * been consistent on disk.
3793 if (max_mirrors == 0)
3794 backup_super_roots(fs_info);
3796 sb = fs_info->super_for_commit;
3797 dev_item = &sb->dev_item;
3799 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3800 head = &fs_info->fs_devices->devices;
3801 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3804 ret = barrier_all_devices(fs_info);
3807 &fs_info->fs_devices->device_list_mutex);
3808 btrfs_handle_fs_error(fs_info, ret,
3809 "errors while submitting device barriers.");
3814 list_for_each_entry(dev, head, dev_list) {
3819 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3820 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3823 btrfs_set_stack_device_generation(dev_item, 0);
3824 btrfs_set_stack_device_type(dev_item, dev->type);
3825 btrfs_set_stack_device_id(dev_item, dev->devid);
3826 btrfs_set_stack_device_total_bytes(dev_item,
3827 dev->commit_total_bytes);
3828 btrfs_set_stack_device_bytes_used(dev_item,
3829 dev->commit_bytes_used);
3830 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3831 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3832 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3833 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3834 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3837 flags = btrfs_super_flags(sb);
3838 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3840 ret = btrfs_validate_write_super(fs_info, sb);
3842 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3843 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3844 "unexpected superblock corruption detected");
3848 ret = write_dev_supers(dev, sb, max_mirrors);
3852 if (total_errors > max_errors) {
3853 btrfs_err(fs_info, "%d errors while writing supers",
3855 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3857 /* FUA is masked off if unsupported and can't be the reason */
3858 btrfs_handle_fs_error(fs_info, -EIO,
3859 "%d errors while writing supers",
3865 list_for_each_entry(dev, head, dev_list) {
3868 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3869 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3872 ret = wait_dev_supers(dev, max_mirrors);
3876 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3877 if (total_errors > max_errors) {
3878 btrfs_handle_fs_error(fs_info, -EIO,
3879 "%d errors while writing supers",
3886 /* Drop a fs root from the radix tree and free it. */
3887 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3888 struct btrfs_root *root)
3890 spin_lock(&fs_info->fs_roots_radix_lock);
3891 radix_tree_delete(&fs_info->fs_roots_radix,
3892 (unsigned long)root->root_key.objectid);
3893 spin_unlock(&fs_info->fs_roots_radix_lock);
3895 if (btrfs_root_refs(&root->root_item) == 0)
3896 synchronize_srcu(&fs_info->subvol_srcu);
3898 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3899 btrfs_free_log(NULL, root);
3900 if (root->reloc_root) {
3901 free_extent_buffer(root->reloc_root->node);
3902 free_extent_buffer(root->reloc_root->commit_root);
3903 btrfs_put_fs_root(root->reloc_root);
3904 root->reloc_root = NULL;
3908 if (root->free_ino_pinned)
3909 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3910 if (root->free_ino_ctl)
3911 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3912 btrfs_free_fs_root(root);
3915 void btrfs_free_fs_root(struct btrfs_root *root)
3917 iput(root->ino_cache_inode);
3918 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3920 free_anon_bdev(root->anon_dev);
3921 if (root->subv_writers)
3922 btrfs_free_subvolume_writers(root->subv_writers);
3923 free_extent_buffer(root->node);
3924 free_extent_buffer(root->commit_root);
3925 kfree(root->free_ino_ctl);
3926 kfree(root->free_ino_pinned);
3927 btrfs_put_fs_root(root);
3930 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3932 u64 root_objectid = 0;
3933 struct btrfs_root *gang[8];
3936 unsigned int ret = 0;
3940 index = srcu_read_lock(&fs_info->subvol_srcu);
3941 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3942 (void **)gang, root_objectid,
3945 srcu_read_unlock(&fs_info->subvol_srcu, index);
3948 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3950 for (i = 0; i < ret; i++) {
3951 /* Avoid to grab roots in dead_roots */
3952 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3956 /* grab all the search result for later use */
3957 gang[i] = btrfs_grab_fs_root(gang[i]);
3959 srcu_read_unlock(&fs_info->subvol_srcu, index);
3961 for (i = 0; i < ret; i++) {
3964 root_objectid = gang[i]->root_key.objectid;
3965 err = btrfs_orphan_cleanup(gang[i]);
3968 btrfs_put_fs_root(gang[i]);
3973 /* release the uncleaned roots due to error */
3974 for (; i < ret; i++) {
3976 btrfs_put_fs_root(gang[i]);
3981 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3983 struct btrfs_root *root = fs_info->tree_root;
3984 struct btrfs_trans_handle *trans;
3986 mutex_lock(&fs_info->cleaner_mutex);
3987 btrfs_run_delayed_iputs(fs_info);
3988 mutex_unlock(&fs_info->cleaner_mutex);
3989 wake_up_process(fs_info->cleaner_kthread);
3991 /* wait until ongoing cleanup work done */
3992 down_write(&fs_info->cleanup_work_sem);
3993 up_write(&fs_info->cleanup_work_sem);
3995 trans = btrfs_join_transaction(root);
3997 return PTR_ERR(trans);
3998 return btrfs_commit_transaction(trans);
4001 void close_ctree(struct btrfs_fs_info *fs_info)
4005 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4007 * We don't want the cleaner to start new transactions, add more delayed
4008 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4009 * because that frees the task_struct, and the transaction kthread might
4010 * still try to wake up the cleaner.
4012 kthread_park(fs_info->cleaner_kthread);
4014 /* wait for the qgroup rescan worker to stop */
4015 btrfs_qgroup_wait_for_completion(fs_info, false);
4017 /* wait for the uuid_scan task to finish */
4018 down(&fs_info->uuid_tree_rescan_sem);
4019 /* avoid complains from lockdep et al., set sem back to initial state */
4020 up(&fs_info->uuid_tree_rescan_sem);
4022 /* pause restriper - we want to resume on mount */
4023 btrfs_pause_balance(fs_info);
4025 btrfs_dev_replace_suspend_for_unmount(fs_info);
4027 btrfs_scrub_cancel(fs_info);
4029 /* wait for any defraggers to finish */
4030 wait_event(fs_info->transaction_wait,
4031 (atomic_read(&fs_info->defrag_running) == 0));
4033 /* clear out the rbtree of defraggable inodes */
4034 btrfs_cleanup_defrag_inodes(fs_info);
4036 cancel_work_sync(&fs_info->async_reclaim_work);
4038 if (!sb_rdonly(fs_info->sb)) {
4040 * The cleaner kthread is stopped, so do one final pass over
4041 * unused block groups.
4043 btrfs_delete_unused_bgs(fs_info);
4045 ret = btrfs_commit_super(fs_info);
4047 btrfs_err(fs_info, "commit super ret %d", ret);
4050 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4051 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4052 btrfs_error_commit_super(fs_info);
4054 kthread_stop(fs_info->transaction_kthread);
4055 kthread_stop(fs_info->cleaner_kthread);
4057 ASSERT(list_empty(&fs_info->delayed_iputs));
4058 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4060 btrfs_free_qgroup_config(fs_info);
4061 ASSERT(list_empty(&fs_info->delalloc_roots));
4063 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4064 btrfs_info(fs_info, "at unmount delalloc count %lld",
4065 percpu_counter_sum(&fs_info->delalloc_bytes));
4068 if (percpu_counter_sum(&fs_info->dio_bytes))
4069 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4070 percpu_counter_sum(&fs_info->dio_bytes));
4072 btrfs_sysfs_remove_mounted(fs_info);
4073 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4075 btrfs_free_fs_roots(fs_info);
4077 btrfs_put_block_group_cache(fs_info);
4080 * we must make sure there is not any read request to
4081 * submit after we stopping all workers.
4083 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4084 btrfs_stop_all_workers(fs_info);
4086 btrfs_free_block_groups(fs_info);
4088 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4089 free_root_pointers(fs_info, 1);
4091 iput(fs_info->btree_inode);
4093 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4094 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4095 btrfsic_unmount(fs_info->fs_devices);
4098 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4099 btrfs_close_devices(fs_info->fs_devices);
4101 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4102 percpu_counter_destroy(&fs_info->delalloc_bytes);
4103 percpu_counter_destroy(&fs_info->dio_bytes);
4104 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4105 cleanup_srcu_struct(&fs_info->subvol_srcu);
4107 btrfs_free_stripe_hash_table(fs_info);
4108 btrfs_free_ref_cache(fs_info);
4111 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4115 struct inode *btree_inode = buf->pages[0]->mapping->host;
4117 ret = extent_buffer_uptodate(buf);
4121 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4122 parent_transid, atomic);
4128 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4130 struct btrfs_fs_info *fs_info;
4131 struct btrfs_root *root;
4132 u64 transid = btrfs_header_generation(buf);
4135 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4137 * This is a fast path so only do this check if we have sanity tests
4138 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4139 * outside of the sanity tests.
4141 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4144 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4145 fs_info = root->fs_info;
4146 btrfs_assert_tree_locked(buf);
4147 if (transid != fs_info->generation)
4148 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4149 buf->start, transid, fs_info->generation);
4150 was_dirty = set_extent_buffer_dirty(buf);
4152 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4154 fs_info->dirty_metadata_batch);
4155 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4157 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4158 * but item data not updated.
4159 * So here we should only check item pointers, not item data.
4161 if (btrfs_header_level(buf) == 0 &&
4162 btrfs_check_leaf_relaxed(buf)) {
4163 btrfs_print_leaf(buf);
4169 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4173 * looks as though older kernels can get into trouble with
4174 * this code, they end up stuck in balance_dirty_pages forever
4178 if (current->flags & PF_MEMALLOC)
4182 btrfs_balance_delayed_items(fs_info);
4184 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4185 BTRFS_DIRTY_METADATA_THRESH,
4186 fs_info->dirty_metadata_batch);
4188 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4192 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4194 __btrfs_btree_balance_dirty(fs_info, 1);
4197 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4199 __btrfs_btree_balance_dirty(fs_info, 0);
4202 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4203 struct btrfs_key *first_key)
4205 return btree_read_extent_buffer_pages(buf, parent_transid,
4209 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4211 /* cleanup FS via transaction */
4212 btrfs_cleanup_transaction(fs_info);
4214 mutex_lock(&fs_info->cleaner_mutex);
4215 btrfs_run_delayed_iputs(fs_info);
4216 mutex_unlock(&fs_info->cleaner_mutex);
4218 down_write(&fs_info->cleanup_work_sem);
4219 up_write(&fs_info->cleanup_work_sem);
4222 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4224 struct btrfs_ordered_extent *ordered;
4226 spin_lock(&root->ordered_extent_lock);
4228 * This will just short circuit the ordered completion stuff which will
4229 * make sure the ordered extent gets properly cleaned up.
4231 list_for_each_entry(ordered, &root->ordered_extents,
4233 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4234 spin_unlock(&root->ordered_extent_lock);
4237 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4239 struct btrfs_root *root;
4240 struct list_head splice;
4242 INIT_LIST_HEAD(&splice);
4244 spin_lock(&fs_info->ordered_root_lock);
4245 list_splice_init(&fs_info->ordered_roots, &splice);
4246 while (!list_empty(&splice)) {
4247 root = list_first_entry(&splice, struct btrfs_root,
4249 list_move_tail(&root->ordered_root,
4250 &fs_info->ordered_roots);
4252 spin_unlock(&fs_info->ordered_root_lock);
4253 btrfs_destroy_ordered_extents(root);
4256 spin_lock(&fs_info->ordered_root_lock);
4258 spin_unlock(&fs_info->ordered_root_lock);
4261 * We need this here because if we've been flipped read-only we won't
4262 * get sync() from the umount, so we need to make sure any ordered
4263 * extents that haven't had their dirty pages IO start writeout yet
4264 * actually get run and error out properly.
4266 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4269 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4270 struct btrfs_fs_info *fs_info)
4272 struct rb_node *node;
4273 struct btrfs_delayed_ref_root *delayed_refs;
4274 struct btrfs_delayed_ref_node *ref;
4277 delayed_refs = &trans->delayed_refs;
4279 spin_lock(&delayed_refs->lock);
4280 if (atomic_read(&delayed_refs->num_entries) == 0) {
4281 spin_unlock(&delayed_refs->lock);
4282 btrfs_info(fs_info, "delayed_refs has NO entry");
4286 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4287 struct btrfs_delayed_ref_head *head;
4289 bool pin_bytes = false;
4291 head = rb_entry(node, struct btrfs_delayed_ref_head,
4293 if (btrfs_delayed_ref_lock(delayed_refs, head))
4296 spin_lock(&head->lock);
4297 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4298 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4301 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4302 RB_CLEAR_NODE(&ref->ref_node);
4303 if (!list_empty(&ref->add_list))
4304 list_del(&ref->add_list);
4305 atomic_dec(&delayed_refs->num_entries);
4306 btrfs_put_delayed_ref(ref);
4308 if (head->must_insert_reserved)
4310 btrfs_free_delayed_extent_op(head->extent_op);
4311 btrfs_delete_ref_head(delayed_refs, head);
4312 spin_unlock(&head->lock);
4313 spin_unlock(&delayed_refs->lock);
4314 mutex_unlock(&head->mutex);
4317 btrfs_pin_extent(fs_info, head->bytenr,
4318 head->num_bytes, 1);
4319 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4320 btrfs_put_delayed_ref_head(head);
4322 spin_lock(&delayed_refs->lock);
4325 spin_unlock(&delayed_refs->lock);
4330 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4332 struct btrfs_inode *btrfs_inode;
4333 struct list_head splice;
4335 INIT_LIST_HEAD(&splice);
4337 spin_lock(&root->delalloc_lock);
4338 list_splice_init(&root->delalloc_inodes, &splice);
4340 while (!list_empty(&splice)) {
4341 struct inode *inode = NULL;
4342 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4344 __btrfs_del_delalloc_inode(root, btrfs_inode);
4345 spin_unlock(&root->delalloc_lock);
4348 * Make sure we get a live inode and that it'll not disappear
4351 inode = igrab(&btrfs_inode->vfs_inode);
4353 invalidate_inode_pages2(inode->i_mapping);
4356 spin_lock(&root->delalloc_lock);
4358 spin_unlock(&root->delalloc_lock);
4361 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4363 struct btrfs_root *root;
4364 struct list_head splice;
4366 INIT_LIST_HEAD(&splice);
4368 spin_lock(&fs_info->delalloc_root_lock);
4369 list_splice_init(&fs_info->delalloc_roots, &splice);
4370 while (!list_empty(&splice)) {
4371 root = list_first_entry(&splice, struct btrfs_root,
4373 root = btrfs_grab_fs_root(root);
4375 spin_unlock(&fs_info->delalloc_root_lock);
4377 btrfs_destroy_delalloc_inodes(root);
4378 btrfs_put_fs_root(root);
4380 spin_lock(&fs_info->delalloc_root_lock);
4382 spin_unlock(&fs_info->delalloc_root_lock);
4385 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4386 struct extent_io_tree *dirty_pages,
4390 struct extent_buffer *eb;
4395 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4400 clear_extent_bits(dirty_pages, start, end, mark);
4401 while (start <= end) {
4402 eb = find_extent_buffer(fs_info, start);
4403 start += fs_info->nodesize;
4406 wait_on_extent_buffer_writeback(eb);
4408 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4410 clear_extent_buffer_dirty(eb);
4411 free_extent_buffer_stale(eb);
4418 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4419 struct extent_io_tree *pinned_extents)
4421 struct extent_io_tree *unpin;
4427 unpin = pinned_extents;
4430 struct extent_state *cached_state = NULL;
4433 * The btrfs_finish_extent_commit() may get the same range as
4434 * ours between find_first_extent_bit and clear_extent_dirty.
4435 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4436 * the same extent range.
4438 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4439 ret = find_first_extent_bit(unpin, 0, &start, &end,
4440 EXTENT_DIRTY, &cached_state);
4442 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4446 clear_extent_dirty(unpin, start, end, &cached_state);
4447 free_extent_state(cached_state);
4448 btrfs_error_unpin_extent_range(fs_info, start, end);
4449 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4454 if (unpin == &fs_info->freed_extents[0])
4455 unpin = &fs_info->freed_extents[1];
4457 unpin = &fs_info->freed_extents[0];
4465 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4467 struct inode *inode;
4469 inode = cache->io_ctl.inode;
4471 invalidate_inode_pages2(inode->i_mapping);
4472 BTRFS_I(inode)->generation = 0;
4473 cache->io_ctl.inode = NULL;
4476 btrfs_put_block_group(cache);
4479 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4480 struct btrfs_fs_info *fs_info)
4482 struct btrfs_block_group_cache *cache;
4484 spin_lock(&cur_trans->dirty_bgs_lock);
4485 while (!list_empty(&cur_trans->dirty_bgs)) {
4486 cache = list_first_entry(&cur_trans->dirty_bgs,
4487 struct btrfs_block_group_cache,
4490 if (!list_empty(&cache->io_list)) {
4491 spin_unlock(&cur_trans->dirty_bgs_lock);
4492 list_del_init(&cache->io_list);
4493 btrfs_cleanup_bg_io(cache);
4494 spin_lock(&cur_trans->dirty_bgs_lock);
4497 list_del_init(&cache->dirty_list);
4498 spin_lock(&cache->lock);
4499 cache->disk_cache_state = BTRFS_DC_ERROR;
4500 spin_unlock(&cache->lock);
4502 spin_unlock(&cur_trans->dirty_bgs_lock);
4503 btrfs_put_block_group(cache);
4504 btrfs_delayed_refs_rsv_release(fs_info, 1);
4505 spin_lock(&cur_trans->dirty_bgs_lock);
4507 spin_unlock(&cur_trans->dirty_bgs_lock);
4510 * Refer to the definition of io_bgs member for details why it's safe
4511 * to use it without any locking
4513 while (!list_empty(&cur_trans->io_bgs)) {
4514 cache = list_first_entry(&cur_trans->io_bgs,
4515 struct btrfs_block_group_cache,
4518 list_del_init(&cache->io_list);
4519 spin_lock(&cache->lock);
4520 cache->disk_cache_state = BTRFS_DC_ERROR;
4521 spin_unlock(&cache->lock);
4522 btrfs_cleanup_bg_io(cache);
4526 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4527 struct btrfs_fs_info *fs_info)
4529 struct btrfs_device *dev, *tmp;
4531 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4532 ASSERT(list_empty(&cur_trans->dirty_bgs));
4533 ASSERT(list_empty(&cur_trans->io_bgs));
4535 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4537 list_del_init(&dev->post_commit_list);
4540 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4542 cur_trans->state = TRANS_STATE_COMMIT_START;
4543 wake_up(&fs_info->transaction_blocked_wait);
4545 cur_trans->state = TRANS_STATE_UNBLOCKED;
4546 wake_up(&fs_info->transaction_wait);
4548 btrfs_destroy_delayed_inodes(fs_info);
4549 btrfs_assert_delayed_root_empty(fs_info);
4551 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4553 btrfs_destroy_pinned_extent(fs_info,
4554 fs_info->pinned_extents);
4556 cur_trans->state =TRANS_STATE_COMPLETED;
4557 wake_up(&cur_trans->commit_wait);
4560 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4562 struct btrfs_transaction *t;
4564 mutex_lock(&fs_info->transaction_kthread_mutex);
4566 spin_lock(&fs_info->trans_lock);
4567 while (!list_empty(&fs_info->trans_list)) {
4568 t = list_first_entry(&fs_info->trans_list,
4569 struct btrfs_transaction, list);
4570 if (t->state >= TRANS_STATE_COMMIT_START) {
4571 refcount_inc(&t->use_count);
4572 spin_unlock(&fs_info->trans_lock);
4573 btrfs_wait_for_commit(fs_info, t->transid);
4574 btrfs_put_transaction(t);
4575 spin_lock(&fs_info->trans_lock);
4578 if (t == fs_info->running_transaction) {
4579 t->state = TRANS_STATE_COMMIT_DOING;
4580 spin_unlock(&fs_info->trans_lock);
4582 * We wait for 0 num_writers since we don't hold a trans
4583 * handle open currently for this transaction.
4585 wait_event(t->writer_wait,
4586 atomic_read(&t->num_writers) == 0);
4588 spin_unlock(&fs_info->trans_lock);
4590 btrfs_cleanup_one_transaction(t, fs_info);
4592 spin_lock(&fs_info->trans_lock);
4593 if (t == fs_info->running_transaction)
4594 fs_info->running_transaction = NULL;
4595 list_del_init(&t->list);
4596 spin_unlock(&fs_info->trans_lock);
4598 btrfs_put_transaction(t);
4599 trace_btrfs_transaction_commit(fs_info->tree_root);
4600 spin_lock(&fs_info->trans_lock);
4602 spin_unlock(&fs_info->trans_lock);
4603 btrfs_destroy_all_ordered_extents(fs_info);
4604 btrfs_destroy_delayed_inodes(fs_info);
4605 btrfs_assert_delayed_root_empty(fs_info);
4606 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4607 btrfs_destroy_all_delalloc_inodes(fs_info);
4608 mutex_unlock(&fs_info->transaction_kthread_mutex);
4613 static const struct extent_io_ops btree_extent_io_ops = {
4614 /* mandatory callbacks */
4615 .submit_bio_hook = btree_submit_bio_hook,
4616 .readpage_end_io_hook = btree_readpage_end_io_hook,