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/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "inode-map.h"
33 #include "check-integrity.h"
34 #include "rcu-string.h"
35 #include "dev-replace.h"
39 #include "compression.h"
40 #include "tree-checker.h"
41 #include "ref-verify.h"
42 #include "block-group.h"
44 #include "space-info.h"
46 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
53 static const struct extent_io_ops btree_extent_io_ops;
54 static void end_workqueue_fn(struct btrfs_work *work);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_fs_info *fs_info);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 struct extent_io_tree *dirty_pages,
62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
72 struct btrfs_end_io_wq {
76 struct btrfs_fs_info *info;
78 enum btrfs_wq_endio_type metadata;
79 struct btrfs_work work;
82 static struct kmem_cache *btrfs_end_io_wq_cache;
84 int __init btrfs_end_io_wq_init(void)
86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq),
91 if (!btrfs_end_io_wq_cache)
96 void __cold btrfs_end_io_wq_exit(void)
98 kmem_cache_destroy(btrfs_end_io_wq_cache);
101 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
103 if (fs_info->csum_shash)
104 crypto_free_shash(fs_info->csum_shash);
108 * async submit bios are used to offload expensive checksumming
109 * onto the worker threads. They checksum file and metadata bios
110 * just before they are sent down the IO stack.
112 struct async_submit_bio {
115 extent_submit_bio_start_t *submit_bio_start;
118 * bio_offset is optional, can be used if the pages in the bio
119 * can't tell us where in the file the bio should go
122 struct btrfs_work work;
127 * Lockdep class keys for extent_buffer->lock's in this root. For a given
128 * eb, the lockdep key is determined by the btrfs_root it belongs to and
129 * the level the eb occupies in the tree.
131 * Different roots are used for different purposes and may nest inside each
132 * other and they require separate keysets. As lockdep keys should be
133 * static, assign keysets according to the purpose of the root as indicated
134 * by btrfs_root->root_key.objectid. This ensures that all special purpose
135 * roots have separate keysets.
137 * Lock-nesting across peer nodes is always done with the immediate parent
138 * node locked thus preventing deadlock. As lockdep doesn't know this, use
139 * subclass to avoid triggering lockdep warning in such cases.
141 * The key is set by the readpage_end_io_hook after the buffer has passed
142 * csum validation but before the pages are unlocked. It is also set by
143 * btrfs_init_new_buffer on freshly allocated blocks.
145 * We also add a check to make sure the highest level of the tree is the
146 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
147 * needs update as well.
149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
150 # if BTRFS_MAX_LEVEL != 8
154 static struct btrfs_lockdep_keyset {
155 u64 id; /* root objectid */
156 const char *name_stem; /* lock name stem */
157 char names[BTRFS_MAX_LEVEL + 1][20];
158 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
159 } btrfs_lockdep_keysets[] = {
160 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
161 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
162 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
163 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
164 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
165 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
166 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
167 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
168 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
169 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
170 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
171 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
172 { .id = 0, .name_stem = "tree" },
175 void __init btrfs_init_lockdep(void)
179 /* initialize lockdep class names */
180 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
181 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
183 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
184 snprintf(ks->names[j], sizeof(ks->names[j]),
185 "btrfs-%s-%02d", ks->name_stem, j);
189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
192 struct btrfs_lockdep_keyset *ks;
194 BUG_ON(level >= ARRAY_SIZE(ks->keys));
196 /* find the matching keyset, id 0 is the default entry */
197 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
198 if (ks->id == objectid)
201 lockdep_set_class_and_name(&eb->lock,
202 &ks->keys[level], ks->names[level]);
208 * extents on the btree inode are pretty simple, there's one extent
209 * that covers the entire device
211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
212 struct page *page, size_t pg_offset,
215 struct extent_map_tree *em_tree = &inode->extent_tree;
216 struct extent_map *em;
219 read_lock(&em_tree->lock);
220 em = lookup_extent_mapping(em_tree, start, len);
222 read_unlock(&em_tree->lock);
225 read_unlock(&em_tree->lock);
227 em = alloc_extent_map();
229 em = ERR_PTR(-ENOMEM);
234 em->block_len = (u64)-1;
237 write_lock(&em_tree->lock);
238 ret = add_extent_mapping(em_tree, em, 0);
239 if (ret == -EEXIST) {
241 em = lookup_extent_mapping(em_tree, start, len);
248 write_unlock(&em_tree->lock);
255 * Compute the csum of a btree block and store the result to provided buffer.
257 * Returns error if the extent buffer cannot be mapped.
259 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
261 struct btrfs_fs_info *fs_info = buf->fs_info;
262 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
267 unsigned long map_start;
268 unsigned long map_len;
271 shash->tfm = fs_info->csum_shash;
272 crypto_shash_init(shash);
274 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 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
292 memset(result, 0, BTRFS_CSUM_SIZE);
294 crypto_shash_final(shash, result);
300 * we can't consider a given block up to date unless the transid of the
301 * block matches the transid in the parent node's pointer. This is how we
302 * detect blocks that either didn't get written at all or got written
303 * in the wrong place.
305 static int verify_parent_transid(struct extent_io_tree *io_tree,
306 struct extent_buffer *eb, u64 parent_transid,
309 struct extent_state *cached_state = NULL;
311 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
313 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
320 btrfs_tree_read_lock(eb);
321 btrfs_set_lock_blocking_read(eb);
324 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
326 if (extent_buffer_uptodate(eb) &&
327 btrfs_header_generation(eb) == parent_transid) {
331 btrfs_err_rl(eb->fs_info,
332 "parent transid verify failed on %llu wanted %llu found %llu",
334 parent_transid, btrfs_header_generation(eb));
338 * Things reading via commit roots that don't have normal protection,
339 * like send, can have a really old block in cache that may point at a
340 * block that has been freed and re-allocated. So don't clear uptodate
341 * if we find an eb that is under IO (dirty/writeback) because we could
342 * end up reading in the stale data and then writing it back out and
343 * making everybody very sad.
345 if (!extent_buffer_under_io(eb))
346 clear_extent_buffer_uptodate(eb);
348 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
351 btrfs_tree_read_unlock_blocking(eb);
355 static bool btrfs_supported_super_csum(u16 csum_type)
358 case BTRFS_CSUM_TYPE_CRC32:
359 case BTRFS_CSUM_TYPE_XXHASH:
360 case BTRFS_CSUM_TYPE_SHA256:
361 case BTRFS_CSUM_TYPE_BLAKE2:
369 * Return 0 if the superblock checksum type matches the checksum value of that
370 * algorithm. Pass the raw disk superblock data.
372 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
375 struct btrfs_super_block *disk_sb =
376 (struct btrfs_super_block *)raw_disk_sb;
377 char result[BTRFS_CSUM_SIZE];
378 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
380 shash->tfm = fs_info->csum_shash;
381 crypto_shash_init(shash);
384 * The super_block structure does not span the whole
385 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
386 * filled with zeros and is included in the checksum.
388 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
389 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
390 crypto_shash_final(shash, result);
392 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
398 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
399 struct btrfs_key *first_key, u64 parent_transid)
401 struct btrfs_fs_info *fs_info = eb->fs_info;
403 struct btrfs_key found_key;
406 found_level = btrfs_header_level(eb);
407 if (found_level != level) {
408 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
409 KERN_ERR "BTRFS: tree level check failed\n");
411 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
412 eb->start, level, found_level);
420 * For live tree block (new tree blocks in current transaction),
421 * we need proper lock context to avoid race, which is impossible here.
422 * So we only checks tree blocks which is read from disk, whose
423 * generation <= fs_info->last_trans_committed.
425 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
428 /* We have @first_key, so this @eb must have at least one item */
429 if (btrfs_header_nritems(eb) == 0) {
431 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
433 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
438 btrfs_node_key_to_cpu(eb, &found_key, 0);
440 btrfs_item_key_to_cpu(eb, &found_key, 0);
441 ret = btrfs_comp_cpu_keys(first_key, &found_key);
444 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
445 KERN_ERR "BTRFS: tree first key check failed\n");
447 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
448 eb->start, parent_transid, first_key->objectid,
449 first_key->type, first_key->offset,
450 found_key.objectid, found_key.type,
457 * helper to read a given tree block, doing retries as required when
458 * the checksums don't match and we have alternate mirrors to try.
460 * @parent_transid: expected transid, skip check if 0
461 * @level: expected level, mandatory check
462 * @first_key: expected key of first slot, skip check if NULL
464 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
465 u64 parent_transid, int level,
466 struct btrfs_key *first_key)
468 struct btrfs_fs_info *fs_info = eb->fs_info;
469 struct extent_io_tree *io_tree;
474 int failed_mirror = 0;
476 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
478 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
479 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
481 if (verify_parent_transid(io_tree, eb,
484 else if (btrfs_verify_level_key(eb, level,
485 first_key, parent_transid))
491 num_copies = btrfs_num_copies(fs_info,
496 if (!failed_mirror) {
498 failed_mirror = eb->read_mirror;
502 if (mirror_num == failed_mirror)
505 if (mirror_num > num_copies)
509 if (failed && !ret && failed_mirror)
510 btrfs_repair_eb_io_failure(eb, failed_mirror);
516 * checksum a dirty tree block before IO. This has extra checks to make sure
517 * we only fill in the checksum field in the first page of a multi-page block
520 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
522 u64 start = page_offset(page);
524 u8 result[BTRFS_CSUM_SIZE];
525 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
526 struct extent_buffer *eb;
529 eb = (struct extent_buffer *)page->private;
530 if (page != eb->pages[0])
533 found_start = btrfs_header_bytenr(eb);
535 * Please do not consolidate these warnings into a single if.
536 * It is useful to know what went wrong.
538 if (WARN_ON(found_start != start))
540 if (WARN_ON(!PageUptodate(page)))
543 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
544 offsetof(struct btrfs_header, fsid),
545 BTRFS_FSID_SIZE) == 0);
547 if (csum_tree_block(eb, result))
550 if (btrfs_header_level(eb))
551 ret = btrfs_check_node(eb);
553 ret = btrfs_check_leaf_full(eb);
556 btrfs_print_tree(eb, 0);
558 "block=%llu write time tree block corruption detected",
560 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
563 write_extent_buffer(eb, result, 0, csum_size);
568 static int check_tree_block_fsid(struct extent_buffer *eb)
570 struct btrfs_fs_info *fs_info = eb->fs_info;
571 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
572 u8 fsid[BTRFS_FSID_SIZE];
575 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
581 * Checking the incompat flag is only valid for the current
582 * fs. For seed devices it's forbidden to have their uuid
583 * changed so reading ->fsid in this case is fine
585 if (fs_devices == fs_info->fs_devices &&
586 btrfs_fs_incompat(fs_info, METADATA_UUID))
587 metadata_uuid = fs_devices->metadata_uuid;
589 metadata_uuid = fs_devices->fsid;
591 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
595 fs_devices = fs_devices->seed;
600 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
601 u64 phy_offset, struct page *page,
602 u64 start, u64 end, int mirror)
606 struct extent_buffer *eb;
607 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
608 struct btrfs_fs_info *fs_info = root->fs_info;
609 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
611 u8 result[BTRFS_CSUM_SIZE];
617 eb = (struct extent_buffer *)page->private;
619 /* the pending IO might have been the only thing that kept this buffer
620 * in memory. Make sure we have a ref for all this other checks
622 atomic_inc(&eb->refs);
624 reads_done = atomic_dec_and_test(&eb->io_pages);
628 eb->read_mirror = mirror;
629 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
634 found_start = btrfs_header_bytenr(eb);
635 if (found_start != eb->start) {
636 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
637 eb->start, found_start);
641 if (check_tree_block_fsid(eb)) {
642 btrfs_err_rl(fs_info, "bad fsid on block %llu",
647 found_level = btrfs_header_level(eb);
648 if (found_level >= BTRFS_MAX_LEVEL) {
649 btrfs_err(fs_info, "bad tree block level %d on %llu",
650 (int)btrfs_header_level(eb), eb->start);
655 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
658 ret = csum_tree_block(eb, result);
662 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
666 memcpy(&found, result, csum_size);
668 read_extent_buffer(eb, &val, 0, csum_size);
669 btrfs_warn_rl(fs_info,
670 "%s checksum verify failed on %llu wanted %x found %x level %d",
671 fs_info->sb->s_id, eb->start,
672 val, found, btrfs_header_level(eb));
678 * If this is a leaf block and it is corrupt, set the corrupt bit so
679 * that we don't try and read the other copies of this block, just
682 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
683 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
687 if (found_level > 0 && btrfs_check_node(eb))
691 set_extent_buffer_uptodate(eb);
694 "block=%llu read time tree block corruption detected",
698 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
699 btree_readahead_hook(eb, ret);
703 * our io error hook is going to dec the io pages
704 * again, we have to make sure it has something
707 atomic_inc(&eb->io_pages);
708 clear_extent_buffer_uptodate(eb);
710 free_extent_buffer(eb);
715 static void end_workqueue_bio(struct bio *bio)
717 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
718 struct btrfs_fs_info *fs_info;
719 struct btrfs_workqueue *wq;
721 fs_info = end_io_wq->info;
722 end_io_wq->status = bio->bi_status;
724 if (bio_op(bio) == REQ_OP_WRITE) {
725 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
726 wq = fs_info->endio_meta_write_workers;
727 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
728 wq = fs_info->endio_freespace_worker;
729 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
730 wq = fs_info->endio_raid56_workers;
732 wq = fs_info->endio_write_workers;
734 if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
735 wq = fs_info->endio_repair_workers;
736 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
737 wq = fs_info->endio_raid56_workers;
738 else if (end_io_wq->metadata)
739 wq = fs_info->endio_meta_workers;
741 wq = fs_info->endio_workers;
744 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
745 btrfs_queue_work(wq, &end_io_wq->work);
748 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
749 enum btrfs_wq_endio_type metadata)
751 struct btrfs_end_io_wq *end_io_wq;
753 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
755 return BLK_STS_RESOURCE;
757 end_io_wq->private = bio->bi_private;
758 end_io_wq->end_io = bio->bi_end_io;
759 end_io_wq->info = info;
760 end_io_wq->status = 0;
761 end_io_wq->bio = bio;
762 end_io_wq->metadata = metadata;
764 bio->bi_private = end_io_wq;
765 bio->bi_end_io = end_workqueue_bio;
769 static void run_one_async_start(struct btrfs_work *work)
771 struct async_submit_bio *async;
774 async = container_of(work, struct async_submit_bio, work);
775 ret = async->submit_bio_start(async->private_data, async->bio,
782 * In order to insert checksums into the metadata in large chunks, we wait
783 * until bio submission time. All the pages in the bio are checksummed and
784 * sums are attached onto the ordered extent record.
786 * At IO completion time the csums attached on the ordered extent record are
787 * inserted into the tree.
789 static void run_one_async_done(struct btrfs_work *work)
791 struct async_submit_bio *async;
795 async = container_of(work, struct async_submit_bio, work);
796 inode = async->private_data;
798 /* If an error occurred we just want to clean up the bio and move on */
800 async->bio->bi_status = async->status;
801 bio_endio(async->bio);
806 * All of the bios that pass through here are from async helpers.
807 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
808 * This changes nothing when cgroups aren't in use.
810 async->bio->bi_opf |= REQ_CGROUP_PUNT;
811 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
813 async->bio->bi_status = ret;
814 bio_endio(async->bio);
818 static void run_one_async_free(struct btrfs_work *work)
820 struct async_submit_bio *async;
822 async = container_of(work, struct async_submit_bio, work);
826 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
827 int mirror_num, unsigned long bio_flags,
828 u64 bio_offset, void *private_data,
829 extent_submit_bio_start_t *submit_bio_start)
831 struct async_submit_bio *async;
833 async = kmalloc(sizeof(*async), GFP_NOFS);
835 return BLK_STS_RESOURCE;
837 async->private_data = private_data;
839 async->mirror_num = mirror_num;
840 async->submit_bio_start = submit_bio_start;
842 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
845 async->bio_offset = bio_offset;
849 if (op_is_sync(bio->bi_opf))
850 btrfs_set_work_high_priority(&async->work);
852 btrfs_queue_work(fs_info->workers, &async->work);
856 static blk_status_t btree_csum_one_bio(struct bio *bio)
858 struct bio_vec *bvec;
859 struct btrfs_root *root;
861 struct bvec_iter_all iter_all;
863 ASSERT(!bio_flagged(bio, BIO_CLONED));
864 bio_for_each_segment_all(bvec, bio, iter_all) {
865 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
866 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
871 return errno_to_blk_status(ret);
874 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
878 * when we're called for a write, we're already in the async
879 * submission context. Just jump into btrfs_map_bio
881 return btree_csum_one_bio(bio);
884 static int check_async_write(struct btrfs_fs_info *fs_info,
885 struct btrfs_inode *bi)
887 if (atomic_read(&bi->sync_writers))
889 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
894 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
896 unsigned long bio_flags)
898 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
899 int async = check_async_write(fs_info, BTRFS_I(inode));
902 if (bio_op(bio) != REQ_OP_WRITE) {
904 * called for a read, do the setup so that checksum validation
905 * can happen in the async kernel threads
907 ret = btrfs_bio_wq_end_io(fs_info, bio,
908 BTRFS_WQ_ENDIO_METADATA);
911 ret = btrfs_map_bio(fs_info, bio, mirror_num);
913 ret = btree_csum_one_bio(bio);
916 ret = btrfs_map_bio(fs_info, bio, mirror_num);
919 * kthread helpers are used to submit writes so that
920 * checksumming can happen in parallel across all CPUs
922 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
923 0, inode, btree_submit_bio_start);
931 bio->bi_status = ret;
936 #ifdef CONFIG_MIGRATION
937 static int btree_migratepage(struct address_space *mapping,
938 struct page *newpage, struct page *page,
939 enum migrate_mode mode)
942 * we can't safely write a btree page from here,
943 * we haven't done the locking hook
948 * Buffers may be managed in a filesystem specific way.
949 * We must have no buffers or drop them.
951 if (page_has_private(page) &&
952 !try_to_release_page(page, GFP_KERNEL))
954 return migrate_page(mapping, newpage, page, mode);
959 static int btree_writepages(struct address_space *mapping,
960 struct writeback_control *wbc)
962 struct btrfs_fs_info *fs_info;
965 if (wbc->sync_mode == WB_SYNC_NONE) {
967 if (wbc->for_kupdate)
970 fs_info = BTRFS_I(mapping->host)->root->fs_info;
971 /* this is a bit racy, but that's ok */
972 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
973 BTRFS_DIRTY_METADATA_THRESH,
974 fs_info->dirty_metadata_batch);
978 return btree_write_cache_pages(mapping, wbc);
981 static int btree_readpage(struct file *file, struct page *page)
983 return extent_read_full_page(page, btree_get_extent, 0);
986 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
988 if (PageWriteback(page) || PageDirty(page))
991 return try_release_extent_buffer(page);
994 static void btree_invalidatepage(struct page *page, unsigned int offset,
997 struct extent_io_tree *tree;
998 tree = &BTRFS_I(page->mapping->host)->io_tree;
999 extent_invalidatepage(tree, page, offset);
1000 btree_releasepage(page, GFP_NOFS);
1001 if (PagePrivate(page)) {
1002 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1003 "page private not zero on page %llu",
1004 (unsigned long long)page_offset(page));
1005 ClearPagePrivate(page);
1006 set_page_private(page, 0);
1011 static int btree_set_page_dirty(struct page *page)
1014 struct extent_buffer *eb;
1016 BUG_ON(!PagePrivate(page));
1017 eb = (struct extent_buffer *)page->private;
1019 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1020 BUG_ON(!atomic_read(&eb->refs));
1021 btrfs_assert_tree_locked(eb);
1023 return __set_page_dirty_nobuffers(page);
1026 static const struct address_space_operations btree_aops = {
1027 .readpage = btree_readpage,
1028 .writepages = btree_writepages,
1029 .releasepage = btree_releasepage,
1030 .invalidatepage = btree_invalidatepage,
1031 #ifdef CONFIG_MIGRATION
1032 .migratepage = btree_migratepage,
1034 .set_page_dirty = btree_set_page_dirty,
1037 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1039 struct extent_buffer *buf = NULL;
1042 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1046 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1048 free_extent_buffer_stale(buf);
1050 free_extent_buffer(buf);
1053 struct extent_buffer *btrfs_find_create_tree_block(
1054 struct btrfs_fs_info *fs_info,
1057 if (btrfs_is_testing(fs_info))
1058 return alloc_test_extent_buffer(fs_info, bytenr);
1059 return alloc_extent_buffer(fs_info, bytenr);
1063 * Read tree block at logical address @bytenr and do variant basic but critical
1066 * @parent_transid: expected transid of this tree block, skip check if 0
1067 * @level: expected level, mandatory check
1068 * @first_key: expected key in slot 0, skip check if NULL
1070 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1071 u64 parent_transid, int level,
1072 struct btrfs_key *first_key)
1074 struct extent_buffer *buf = NULL;
1077 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1081 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1084 free_extent_buffer_stale(buf);
1085 return ERR_PTR(ret);
1091 void btrfs_clean_tree_block(struct extent_buffer *buf)
1093 struct btrfs_fs_info *fs_info = buf->fs_info;
1094 if (btrfs_header_generation(buf) ==
1095 fs_info->running_transaction->transid) {
1096 btrfs_assert_tree_locked(buf);
1098 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1099 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1101 fs_info->dirty_metadata_batch);
1102 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1103 btrfs_set_lock_blocking_write(buf);
1104 clear_extent_buffer_dirty(buf);
1109 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1112 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1113 root->fs_info = fs_info;
1115 root->commit_root = NULL;
1117 root->orphan_cleanup_state = 0;
1119 root->last_trans = 0;
1120 root->highest_objectid = 0;
1121 root->nr_delalloc_inodes = 0;
1122 root->nr_ordered_extents = 0;
1123 root->inode_tree = RB_ROOT;
1124 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1125 root->block_rsv = NULL;
1127 INIT_LIST_HEAD(&root->dirty_list);
1128 INIT_LIST_HEAD(&root->root_list);
1129 INIT_LIST_HEAD(&root->delalloc_inodes);
1130 INIT_LIST_HEAD(&root->delalloc_root);
1131 INIT_LIST_HEAD(&root->ordered_extents);
1132 INIT_LIST_HEAD(&root->ordered_root);
1133 INIT_LIST_HEAD(&root->reloc_dirty_list);
1134 INIT_LIST_HEAD(&root->logged_list[0]);
1135 INIT_LIST_HEAD(&root->logged_list[1]);
1136 spin_lock_init(&root->inode_lock);
1137 spin_lock_init(&root->delalloc_lock);
1138 spin_lock_init(&root->ordered_extent_lock);
1139 spin_lock_init(&root->accounting_lock);
1140 spin_lock_init(&root->log_extents_lock[0]);
1141 spin_lock_init(&root->log_extents_lock[1]);
1142 spin_lock_init(&root->qgroup_meta_rsv_lock);
1143 mutex_init(&root->objectid_mutex);
1144 mutex_init(&root->log_mutex);
1145 mutex_init(&root->ordered_extent_mutex);
1146 mutex_init(&root->delalloc_mutex);
1147 init_waitqueue_head(&root->log_writer_wait);
1148 init_waitqueue_head(&root->log_commit_wait[0]);
1149 init_waitqueue_head(&root->log_commit_wait[1]);
1150 INIT_LIST_HEAD(&root->log_ctxs[0]);
1151 INIT_LIST_HEAD(&root->log_ctxs[1]);
1152 atomic_set(&root->log_commit[0], 0);
1153 atomic_set(&root->log_commit[1], 0);
1154 atomic_set(&root->log_writers, 0);
1155 atomic_set(&root->log_batch, 0);
1156 refcount_set(&root->refs, 1);
1157 atomic_set(&root->snapshot_force_cow, 0);
1158 atomic_set(&root->nr_swapfiles, 0);
1159 root->log_transid = 0;
1160 root->log_transid_committed = -1;
1161 root->last_log_commit = 0;
1163 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1164 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1166 memset(&root->root_key, 0, sizeof(root->root_key));
1167 memset(&root->root_item, 0, sizeof(root->root_item));
1168 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1170 root->defrag_trans_start = fs_info->generation;
1172 root->defrag_trans_start = 0;
1173 root->root_key.objectid = objectid;
1176 spin_lock_init(&root->root_item_lock);
1177 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1178 #ifdef CONFIG_BTRFS_DEBUG
1179 INIT_LIST_HEAD(&root->leak_list);
1180 spin_lock(&fs_info->fs_roots_radix_lock);
1181 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1182 spin_unlock(&fs_info->fs_roots_radix_lock);
1186 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1187 u64 objectid, gfp_t flags)
1189 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1191 __setup_root(root, fs_info, objectid);
1195 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1196 /* Should only be used by the testing infrastructure */
1197 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1199 struct btrfs_root *root;
1202 return ERR_PTR(-EINVAL);
1204 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1206 return ERR_PTR(-ENOMEM);
1208 /* We don't use the stripesize in selftest, set it as sectorsize */
1209 root->alloc_bytenr = 0;
1215 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1218 struct btrfs_fs_info *fs_info = trans->fs_info;
1219 struct extent_buffer *leaf;
1220 struct btrfs_root *tree_root = fs_info->tree_root;
1221 struct btrfs_root *root;
1222 struct btrfs_key key;
1223 unsigned int nofs_flag;
1225 uuid_le uuid = NULL_UUID_LE;
1228 * We're holding a transaction handle, so use a NOFS memory allocation
1229 * context to avoid deadlock if reclaim happens.
1231 nofs_flag = memalloc_nofs_save();
1232 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1233 memalloc_nofs_restore(nofs_flag);
1235 return ERR_PTR(-ENOMEM);
1237 root->root_key.objectid = objectid;
1238 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1239 root->root_key.offset = 0;
1241 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1243 ret = PTR_ERR(leaf);
1249 btrfs_mark_buffer_dirty(leaf);
1251 root->commit_root = btrfs_root_node(root);
1252 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1254 root->root_item.flags = 0;
1255 root->root_item.byte_limit = 0;
1256 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1257 btrfs_set_root_generation(&root->root_item, trans->transid);
1258 btrfs_set_root_level(&root->root_item, 0);
1259 btrfs_set_root_refs(&root->root_item, 1);
1260 btrfs_set_root_used(&root->root_item, leaf->len);
1261 btrfs_set_root_last_snapshot(&root->root_item, 0);
1262 btrfs_set_root_dirid(&root->root_item, 0);
1263 if (is_fstree(objectid))
1265 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1266 root->root_item.drop_level = 0;
1268 key.objectid = objectid;
1269 key.type = BTRFS_ROOT_ITEM_KEY;
1271 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1275 btrfs_tree_unlock(leaf);
1281 btrfs_tree_unlock(leaf);
1282 free_extent_buffer(root->commit_root);
1283 free_extent_buffer(leaf);
1285 btrfs_put_root(root);
1287 return ERR_PTR(ret);
1290 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1291 struct btrfs_fs_info *fs_info)
1293 struct btrfs_root *root;
1294 struct extent_buffer *leaf;
1296 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1298 return ERR_PTR(-ENOMEM);
1300 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1301 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1302 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1305 * DON'T set REF_COWS for log trees
1307 * log trees do not get reference counted because they go away
1308 * before a real commit is actually done. They do store pointers
1309 * to file data extents, and those reference counts still get
1310 * updated (along with back refs to the log tree).
1313 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1316 btrfs_put_root(root);
1317 return ERR_CAST(leaf);
1322 btrfs_mark_buffer_dirty(root->node);
1323 btrfs_tree_unlock(root->node);
1327 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1328 struct btrfs_fs_info *fs_info)
1330 struct btrfs_root *log_root;
1332 log_root = alloc_log_tree(trans, fs_info);
1333 if (IS_ERR(log_root))
1334 return PTR_ERR(log_root);
1335 WARN_ON(fs_info->log_root_tree);
1336 fs_info->log_root_tree = log_root;
1340 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1341 struct btrfs_root *root)
1343 struct btrfs_fs_info *fs_info = root->fs_info;
1344 struct btrfs_root *log_root;
1345 struct btrfs_inode_item *inode_item;
1347 log_root = alloc_log_tree(trans, fs_info);
1348 if (IS_ERR(log_root))
1349 return PTR_ERR(log_root);
1351 log_root->last_trans = trans->transid;
1352 log_root->root_key.offset = root->root_key.objectid;
1354 inode_item = &log_root->root_item.inode;
1355 btrfs_set_stack_inode_generation(inode_item, 1);
1356 btrfs_set_stack_inode_size(inode_item, 3);
1357 btrfs_set_stack_inode_nlink(inode_item, 1);
1358 btrfs_set_stack_inode_nbytes(inode_item,
1360 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1362 btrfs_set_root_node(&log_root->root_item, log_root->node);
1364 WARN_ON(root->log_root);
1365 root->log_root = log_root;
1366 root->log_transid = 0;
1367 root->log_transid_committed = -1;
1368 root->last_log_commit = 0;
1372 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1373 struct btrfs_key *key)
1375 struct btrfs_root *root;
1376 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1377 struct btrfs_path *path;
1382 path = btrfs_alloc_path();
1384 return ERR_PTR(-ENOMEM);
1386 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1392 ret = btrfs_find_root(tree_root, key, path,
1393 &root->root_item, &root->root_key);
1400 generation = btrfs_root_generation(&root->root_item);
1401 level = btrfs_root_level(&root->root_item);
1402 root->node = read_tree_block(fs_info,
1403 btrfs_root_bytenr(&root->root_item),
1404 generation, level, NULL);
1405 if (IS_ERR(root->node)) {
1406 ret = PTR_ERR(root->node);
1408 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1410 free_extent_buffer(root->node);
1413 root->commit_root = btrfs_root_node(root);
1415 btrfs_free_path(path);
1419 btrfs_put_root(root);
1421 root = ERR_PTR(ret);
1425 static int btrfs_init_fs_root(struct btrfs_root *root)
1428 unsigned int nofs_flag;
1430 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1431 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1433 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1439 * We might be called under a transaction (e.g. indirect backref
1440 * resolution) which could deadlock if it triggers memory reclaim
1442 nofs_flag = memalloc_nofs_save();
1443 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1444 memalloc_nofs_restore(nofs_flag);
1448 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1449 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1450 btrfs_check_and_init_root_item(&root->root_item);
1453 btrfs_init_free_ino_ctl(root);
1454 spin_lock_init(&root->ino_cache_lock);
1455 init_waitqueue_head(&root->ino_cache_wait);
1457 ret = get_anon_bdev(&root->anon_dev);
1461 mutex_lock(&root->objectid_mutex);
1462 ret = btrfs_find_highest_objectid(root,
1463 &root->highest_objectid);
1465 mutex_unlock(&root->objectid_mutex);
1469 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1471 mutex_unlock(&root->objectid_mutex);
1475 /* The caller is responsible to call btrfs_free_fs_root */
1479 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1482 struct btrfs_root *root;
1484 spin_lock(&fs_info->fs_roots_radix_lock);
1485 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1486 (unsigned long)root_id);
1488 root = btrfs_grab_root(root);
1489 spin_unlock(&fs_info->fs_roots_radix_lock);
1493 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1494 struct btrfs_root *root)
1498 ret = radix_tree_preload(GFP_NOFS);
1502 spin_lock(&fs_info->fs_roots_radix_lock);
1503 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1504 (unsigned long)root->root_key.objectid,
1507 btrfs_grab_root(root);
1508 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1510 spin_unlock(&fs_info->fs_roots_radix_lock);
1511 radix_tree_preload_end();
1516 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1518 #ifdef CONFIG_BTRFS_DEBUG
1519 struct btrfs_root *root;
1521 while (!list_empty(&fs_info->allocated_roots)) {
1522 root = list_first_entry(&fs_info->allocated_roots,
1523 struct btrfs_root, leak_list);
1524 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1525 root->root_key.objectid, root->root_key.offset,
1526 refcount_read(&root->refs));
1527 while (refcount_read(&root->refs) > 1)
1528 btrfs_put_root(root);
1529 btrfs_put_root(root);
1534 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1536 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1537 percpu_counter_destroy(&fs_info->delalloc_bytes);
1538 percpu_counter_destroy(&fs_info->dio_bytes);
1539 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1540 btrfs_free_csum_hash(fs_info);
1541 btrfs_free_stripe_hash_table(fs_info);
1542 btrfs_free_ref_cache(fs_info);
1543 kfree(fs_info->balance_ctl);
1544 kfree(fs_info->delayed_root);
1545 btrfs_put_root(fs_info->extent_root);
1546 btrfs_put_root(fs_info->tree_root);
1547 btrfs_put_root(fs_info->chunk_root);
1548 btrfs_put_root(fs_info->dev_root);
1549 btrfs_put_root(fs_info->csum_root);
1550 btrfs_put_root(fs_info->quota_root);
1551 btrfs_put_root(fs_info->uuid_root);
1552 btrfs_put_root(fs_info->free_space_root);
1553 btrfs_put_root(fs_info->fs_root);
1554 btrfs_check_leaked_roots(fs_info);
1555 kfree(fs_info->super_copy);
1556 kfree(fs_info->super_for_commit);
1561 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1562 struct btrfs_key *location,
1565 struct btrfs_root *root;
1566 struct btrfs_path *path;
1567 struct btrfs_key key;
1570 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1571 return btrfs_grab_root(fs_info->tree_root);
1572 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1573 return btrfs_grab_root(fs_info->extent_root);
1574 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1575 return btrfs_grab_root(fs_info->chunk_root);
1576 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1577 return btrfs_grab_root(fs_info->dev_root);
1578 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1579 return btrfs_grab_root(fs_info->csum_root);
1580 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1581 return btrfs_grab_root(fs_info->quota_root) ?
1582 fs_info->quota_root : ERR_PTR(-ENOENT);
1583 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1584 return btrfs_grab_root(fs_info->uuid_root) ?
1585 fs_info->uuid_root : ERR_PTR(-ENOENT);
1586 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1587 return btrfs_grab_root(fs_info->free_space_root) ?
1588 fs_info->free_space_root : ERR_PTR(-ENOENT);
1590 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1592 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1593 btrfs_put_root(root);
1594 return ERR_PTR(-ENOENT);
1599 root = btrfs_read_tree_root(fs_info->tree_root, location);
1603 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1608 ret = btrfs_init_fs_root(root);
1612 path = btrfs_alloc_path();
1617 key.objectid = BTRFS_ORPHAN_OBJECTID;
1618 key.type = BTRFS_ORPHAN_ITEM_KEY;
1619 key.offset = location->objectid;
1621 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1622 btrfs_free_path(path);
1626 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1629 * All roots have two refs on them at all times, one for the mounted fs,
1630 * and one for being in the radix tree. This way we only free the root
1631 * when we are unmounting or deleting the subvolume. We get one ref
1632 * from __setup_root, one for inserting it into the radix tree, and then
1633 * we have the third for returning it, and the caller will put it when
1634 * it's done with the root.
1636 btrfs_grab_root(root);
1637 ret = btrfs_insert_fs_root(fs_info, root);
1639 btrfs_put_root(root);
1640 if (ret == -EEXIST) {
1641 btrfs_free_fs_root(root);
1648 btrfs_free_fs_root(root);
1649 return ERR_PTR(ret);
1652 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1654 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1656 struct btrfs_device *device;
1657 struct backing_dev_info *bdi;
1660 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1663 bdi = device->bdev->bd_bdi;
1664 if (bdi_congested(bdi, bdi_bits)) {
1674 * called by the kthread helper functions to finally call the bio end_io
1675 * functions. This is where read checksum verification actually happens
1677 static void end_workqueue_fn(struct btrfs_work *work)
1680 struct btrfs_end_io_wq *end_io_wq;
1682 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1683 bio = end_io_wq->bio;
1685 bio->bi_status = end_io_wq->status;
1686 bio->bi_private = end_io_wq->private;
1687 bio->bi_end_io = end_io_wq->end_io;
1689 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1692 static int cleaner_kthread(void *arg)
1694 struct btrfs_root *root = arg;
1695 struct btrfs_fs_info *fs_info = root->fs_info;
1701 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1703 /* Make the cleaner go to sleep early. */
1704 if (btrfs_need_cleaner_sleep(fs_info))
1708 * Do not do anything if we might cause open_ctree() to block
1709 * before we have finished mounting the filesystem.
1711 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1714 if (!mutex_trylock(&fs_info->cleaner_mutex))
1718 * Avoid the problem that we change the status of the fs
1719 * during the above check and trylock.
1721 if (btrfs_need_cleaner_sleep(fs_info)) {
1722 mutex_unlock(&fs_info->cleaner_mutex);
1726 btrfs_run_delayed_iputs(fs_info);
1728 again = btrfs_clean_one_deleted_snapshot(root);
1729 mutex_unlock(&fs_info->cleaner_mutex);
1732 * The defragger has dealt with the R/O remount and umount,
1733 * needn't do anything special here.
1735 btrfs_run_defrag_inodes(fs_info);
1738 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1739 * with relocation (btrfs_relocate_chunk) and relocation
1740 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1741 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1742 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1743 * unused block groups.
1745 btrfs_delete_unused_bgs(fs_info);
1747 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1748 if (kthread_should_park())
1750 if (kthread_should_stop())
1753 set_current_state(TASK_INTERRUPTIBLE);
1755 __set_current_state(TASK_RUNNING);
1760 static int transaction_kthread(void *arg)
1762 struct btrfs_root *root = arg;
1763 struct btrfs_fs_info *fs_info = root->fs_info;
1764 struct btrfs_trans_handle *trans;
1765 struct btrfs_transaction *cur;
1768 unsigned long delay;
1772 cannot_commit = false;
1773 delay = HZ * fs_info->commit_interval;
1774 mutex_lock(&fs_info->transaction_kthread_mutex);
1776 spin_lock(&fs_info->trans_lock);
1777 cur = fs_info->running_transaction;
1779 spin_unlock(&fs_info->trans_lock);
1783 now = ktime_get_seconds();
1784 if (cur->state < TRANS_STATE_COMMIT_START &&
1785 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1786 (now < cur->start_time ||
1787 now - cur->start_time < fs_info->commit_interval)) {
1788 spin_unlock(&fs_info->trans_lock);
1792 transid = cur->transid;
1793 spin_unlock(&fs_info->trans_lock);
1795 /* If the file system is aborted, this will always fail. */
1796 trans = btrfs_attach_transaction(root);
1797 if (IS_ERR(trans)) {
1798 if (PTR_ERR(trans) != -ENOENT)
1799 cannot_commit = true;
1802 if (transid == trans->transid) {
1803 btrfs_commit_transaction(trans);
1805 btrfs_end_transaction(trans);
1808 wake_up_process(fs_info->cleaner_kthread);
1809 mutex_unlock(&fs_info->transaction_kthread_mutex);
1811 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1812 &fs_info->fs_state)))
1813 btrfs_cleanup_transaction(fs_info);
1814 if (!kthread_should_stop() &&
1815 (!btrfs_transaction_blocked(fs_info) ||
1817 schedule_timeout_interruptible(delay);
1818 } while (!kthread_should_stop());
1823 * This will find the highest generation in the array of root backups. The
1824 * index of the highest array is returned, or -EINVAL if we can't find
1827 * We check to make sure the array is valid by comparing the
1828 * generation of the latest root in the array with the generation
1829 * in the super block. If they don't match we pitch it.
1831 static int find_newest_super_backup(struct btrfs_fs_info *info)
1833 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1835 struct btrfs_root_backup *root_backup;
1838 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1839 root_backup = info->super_copy->super_roots + i;
1840 cur = btrfs_backup_tree_root_gen(root_backup);
1841 if (cur == newest_gen)
1849 * copy all the root pointers into the super backup array.
1850 * this will bump the backup pointer by one when it is
1853 static void backup_super_roots(struct btrfs_fs_info *info)
1855 const int next_backup = info->backup_root_index;
1856 struct btrfs_root_backup *root_backup;
1858 root_backup = info->super_for_commit->super_roots + next_backup;
1861 * make sure all of our padding and empty slots get zero filled
1862 * regardless of which ones we use today
1864 memset(root_backup, 0, sizeof(*root_backup));
1866 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1868 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1869 btrfs_set_backup_tree_root_gen(root_backup,
1870 btrfs_header_generation(info->tree_root->node));
1872 btrfs_set_backup_tree_root_level(root_backup,
1873 btrfs_header_level(info->tree_root->node));
1875 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1876 btrfs_set_backup_chunk_root_gen(root_backup,
1877 btrfs_header_generation(info->chunk_root->node));
1878 btrfs_set_backup_chunk_root_level(root_backup,
1879 btrfs_header_level(info->chunk_root->node));
1881 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1882 btrfs_set_backup_extent_root_gen(root_backup,
1883 btrfs_header_generation(info->extent_root->node));
1884 btrfs_set_backup_extent_root_level(root_backup,
1885 btrfs_header_level(info->extent_root->node));
1888 * we might commit during log recovery, which happens before we set
1889 * the fs_root. Make sure it is valid before we fill it in.
1891 if (info->fs_root && info->fs_root->node) {
1892 btrfs_set_backup_fs_root(root_backup,
1893 info->fs_root->node->start);
1894 btrfs_set_backup_fs_root_gen(root_backup,
1895 btrfs_header_generation(info->fs_root->node));
1896 btrfs_set_backup_fs_root_level(root_backup,
1897 btrfs_header_level(info->fs_root->node));
1900 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1901 btrfs_set_backup_dev_root_gen(root_backup,
1902 btrfs_header_generation(info->dev_root->node));
1903 btrfs_set_backup_dev_root_level(root_backup,
1904 btrfs_header_level(info->dev_root->node));
1906 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1907 btrfs_set_backup_csum_root_gen(root_backup,
1908 btrfs_header_generation(info->csum_root->node));
1909 btrfs_set_backup_csum_root_level(root_backup,
1910 btrfs_header_level(info->csum_root->node));
1912 btrfs_set_backup_total_bytes(root_backup,
1913 btrfs_super_total_bytes(info->super_copy));
1914 btrfs_set_backup_bytes_used(root_backup,
1915 btrfs_super_bytes_used(info->super_copy));
1916 btrfs_set_backup_num_devices(root_backup,
1917 btrfs_super_num_devices(info->super_copy));
1920 * if we don't copy this out to the super_copy, it won't get remembered
1921 * for the next commit
1923 memcpy(&info->super_copy->super_roots,
1924 &info->super_for_commit->super_roots,
1925 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1929 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1930 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1932 * fs_info - filesystem whose backup roots need to be read
1933 * priority - priority of backup root required
1935 * Returns backup root index on success and -EINVAL otherwise.
1937 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1939 int backup_index = find_newest_super_backup(fs_info);
1940 struct btrfs_super_block *super = fs_info->super_copy;
1941 struct btrfs_root_backup *root_backup;
1943 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1945 return backup_index;
1947 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1948 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1953 root_backup = super->super_roots + backup_index;
1955 btrfs_set_super_generation(super,
1956 btrfs_backup_tree_root_gen(root_backup));
1957 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1958 btrfs_set_super_root_level(super,
1959 btrfs_backup_tree_root_level(root_backup));
1960 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1963 * Fixme: the total bytes and num_devices need to match or we should
1966 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1967 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1969 return backup_index;
1972 /* helper to cleanup workers */
1973 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1975 btrfs_destroy_workqueue(fs_info->fixup_workers);
1976 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1977 btrfs_destroy_workqueue(fs_info->workers);
1978 btrfs_destroy_workqueue(fs_info->endio_workers);
1979 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1980 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1981 btrfs_destroy_workqueue(fs_info->rmw_workers);
1982 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1983 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1984 btrfs_destroy_workqueue(fs_info->delayed_workers);
1985 btrfs_destroy_workqueue(fs_info->caching_workers);
1986 btrfs_destroy_workqueue(fs_info->readahead_workers);
1987 btrfs_destroy_workqueue(fs_info->flush_workers);
1988 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1989 if (fs_info->discard_ctl.discard_workers)
1990 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1992 * Now that all other work queues are destroyed, we can safely destroy
1993 * the queues used for metadata I/O, since tasks from those other work
1994 * queues can do metadata I/O operations.
1996 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
1997 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2000 static void free_root_extent_buffers(struct btrfs_root *root)
2003 free_extent_buffer(root->node);
2004 free_extent_buffer(root->commit_root);
2006 root->commit_root = NULL;
2010 /* helper to cleanup tree roots */
2011 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2013 free_root_extent_buffers(info->tree_root);
2015 free_root_extent_buffers(info->dev_root);
2016 free_root_extent_buffers(info->extent_root);
2017 free_root_extent_buffers(info->csum_root);
2018 free_root_extent_buffers(info->quota_root);
2019 free_root_extent_buffers(info->uuid_root);
2020 if (free_chunk_root)
2021 free_root_extent_buffers(info->chunk_root);
2022 free_root_extent_buffers(info->free_space_root);
2025 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2028 struct btrfs_root *gang[8];
2031 while (!list_empty(&fs_info->dead_roots)) {
2032 gang[0] = list_entry(fs_info->dead_roots.next,
2033 struct btrfs_root, root_list);
2034 list_del(&gang[0]->root_list);
2036 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2037 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2039 free_extent_buffer(gang[0]->node);
2040 free_extent_buffer(gang[0]->commit_root);
2041 btrfs_put_root(gang[0]);
2046 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2051 for (i = 0; i < ret; i++)
2052 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2055 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
2056 btrfs_free_log_root_tree(NULL, fs_info);
2059 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2061 mutex_init(&fs_info->scrub_lock);
2062 atomic_set(&fs_info->scrubs_running, 0);
2063 atomic_set(&fs_info->scrub_pause_req, 0);
2064 atomic_set(&fs_info->scrubs_paused, 0);
2065 atomic_set(&fs_info->scrub_cancel_req, 0);
2066 init_waitqueue_head(&fs_info->scrub_pause_wait);
2067 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2070 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2072 spin_lock_init(&fs_info->balance_lock);
2073 mutex_init(&fs_info->balance_mutex);
2074 atomic_set(&fs_info->balance_pause_req, 0);
2075 atomic_set(&fs_info->balance_cancel_req, 0);
2076 fs_info->balance_ctl = NULL;
2077 init_waitqueue_head(&fs_info->balance_wait_q);
2080 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2082 struct inode *inode = fs_info->btree_inode;
2084 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2085 set_nlink(inode, 1);
2087 * we set the i_size on the btree inode to the max possible int.
2088 * the real end of the address space is determined by all of
2089 * the devices in the system
2091 inode->i_size = OFFSET_MAX;
2092 inode->i_mapping->a_ops = &btree_aops;
2094 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2095 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2096 IO_TREE_INODE_IO, inode);
2097 BTRFS_I(inode)->io_tree.track_uptodate = false;
2098 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2100 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2102 BTRFS_I(inode)->root = fs_info->tree_root;
2103 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2104 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2105 btrfs_insert_inode_hash(inode);
2108 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2110 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2111 init_rwsem(&fs_info->dev_replace.rwsem);
2112 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2115 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2117 spin_lock_init(&fs_info->qgroup_lock);
2118 mutex_init(&fs_info->qgroup_ioctl_lock);
2119 fs_info->qgroup_tree = RB_ROOT;
2120 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2121 fs_info->qgroup_seq = 1;
2122 fs_info->qgroup_ulist = NULL;
2123 fs_info->qgroup_rescan_running = false;
2124 mutex_init(&fs_info->qgroup_rescan_lock);
2127 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2128 struct btrfs_fs_devices *fs_devices)
2130 u32 max_active = fs_info->thread_pool_size;
2131 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2134 btrfs_alloc_workqueue(fs_info, "worker",
2135 flags | WQ_HIGHPRI, max_active, 16);
2137 fs_info->delalloc_workers =
2138 btrfs_alloc_workqueue(fs_info, "delalloc",
2139 flags, max_active, 2);
2141 fs_info->flush_workers =
2142 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2143 flags, max_active, 0);
2145 fs_info->caching_workers =
2146 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2148 fs_info->fixup_workers =
2149 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2152 * endios are largely parallel and should have a very
2155 fs_info->endio_workers =
2156 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2157 fs_info->endio_meta_workers =
2158 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2160 fs_info->endio_meta_write_workers =
2161 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2163 fs_info->endio_raid56_workers =
2164 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2166 fs_info->endio_repair_workers =
2167 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2168 fs_info->rmw_workers =
2169 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2170 fs_info->endio_write_workers =
2171 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2173 fs_info->endio_freespace_worker =
2174 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2176 fs_info->delayed_workers =
2177 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2179 fs_info->readahead_workers =
2180 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2182 fs_info->qgroup_rescan_workers =
2183 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2184 fs_info->discard_ctl.discard_workers =
2185 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2187 if (!(fs_info->workers && fs_info->delalloc_workers &&
2188 fs_info->flush_workers &&
2189 fs_info->endio_workers && fs_info->endio_meta_workers &&
2190 fs_info->endio_meta_write_workers &&
2191 fs_info->endio_repair_workers &&
2192 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2193 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2194 fs_info->caching_workers && fs_info->readahead_workers &&
2195 fs_info->fixup_workers && fs_info->delayed_workers &&
2196 fs_info->qgroup_rescan_workers &&
2197 fs_info->discard_ctl.discard_workers)) {
2204 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2206 struct crypto_shash *csum_shash;
2207 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2209 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2211 if (IS_ERR(csum_shash)) {
2212 btrfs_err(fs_info, "error allocating %s hash for checksum",
2214 return PTR_ERR(csum_shash);
2217 fs_info->csum_shash = csum_shash;
2222 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2223 struct btrfs_fs_devices *fs_devices)
2226 struct btrfs_root *log_tree_root;
2227 struct btrfs_super_block *disk_super = fs_info->super_copy;
2228 u64 bytenr = btrfs_super_log_root(disk_super);
2229 int level = btrfs_super_log_root_level(disk_super);
2231 if (fs_devices->rw_devices == 0) {
2232 btrfs_warn(fs_info, "log replay required on RO media");
2236 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2241 log_tree_root->node = read_tree_block(fs_info, bytenr,
2242 fs_info->generation + 1,
2244 if (IS_ERR(log_tree_root->node)) {
2245 btrfs_warn(fs_info, "failed to read log tree");
2246 ret = PTR_ERR(log_tree_root->node);
2247 btrfs_put_root(log_tree_root);
2249 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2250 btrfs_err(fs_info, "failed to read log tree");
2251 free_extent_buffer(log_tree_root->node);
2252 btrfs_put_root(log_tree_root);
2255 /* returns with log_tree_root freed on success */
2256 ret = btrfs_recover_log_trees(log_tree_root);
2258 btrfs_handle_fs_error(fs_info, ret,
2259 "Failed to recover log tree");
2260 free_extent_buffer(log_tree_root->node);
2261 btrfs_put_root(log_tree_root);
2265 if (sb_rdonly(fs_info->sb)) {
2266 ret = btrfs_commit_super(fs_info);
2274 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2276 struct btrfs_root *tree_root = fs_info->tree_root;
2277 struct btrfs_root *root;
2278 struct btrfs_key location;
2281 BUG_ON(!fs_info->tree_root);
2283 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2284 location.type = BTRFS_ROOT_ITEM_KEY;
2285 location.offset = 0;
2287 root = btrfs_read_tree_root(tree_root, &location);
2289 ret = PTR_ERR(root);
2292 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2293 fs_info->extent_root = root;
2295 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2296 root = btrfs_read_tree_root(tree_root, &location);
2298 ret = PTR_ERR(root);
2301 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2302 fs_info->dev_root = root;
2303 btrfs_init_devices_late(fs_info);
2305 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2306 root = btrfs_read_tree_root(tree_root, &location);
2308 ret = PTR_ERR(root);
2311 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2312 fs_info->csum_root = root;
2314 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2315 root = btrfs_read_tree_root(tree_root, &location);
2316 if (!IS_ERR(root)) {
2317 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2318 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2319 fs_info->quota_root = root;
2322 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2323 root = btrfs_read_tree_root(tree_root, &location);
2325 ret = PTR_ERR(root);
2329 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2330 fs_info->uuid_root = root;
2333 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2334 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2335 root = btrfs_read_tree_root(tree_root, &location);
2337 ret = PTR_ERR(root);
2340 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2341 fs_info->free_space_root = root;
2346 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2347 location.objectid, ret);
2352 * Real super block validation
2353 * NOTE: super csum type and incompat features will not be checked here.
2355 * @sb: super block to check
2356 * @mirror_num: the super block number to check its bytenr:
2357 * 0 the primary (1st) sb
2358 * 1, 2 2nd and 3rd backup copy
2359 * -1 skip bytenr check
2361 static int validate_super(struct btrfs_fs_info *fs_info,
2362 struct btrfs_super_block *sb, int mirror_num)
2364 u64 nodesize = btrfs_super_nodesize(sb);
2365 u64 sectorsize = btrfs_super_sectorsize(sb);
2368 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2369 btrfs_err(fs_info, "no valid FS found");
2372 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2373 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2374 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2377 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2378 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2379 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2382 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2383 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2384 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2387 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2388 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2389 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2394 * Check sectorsize and nodesize first, other check will need it.
2395 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2397 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2398 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2399 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2402 /* Only PAGE SIZE is supported yet */
2403 if (sectorsize != PAGE_SIZE) {
2405 "sectorsize %llu not supported yet, only support %lu",
2406 sectorsize, PAGE_SIZE);
2409 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2410 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2411 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2414 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2415 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2416 le32_to_cpu(sb->__unused_leafsize), nodesize);
2420 /* Root alignment check */
2421 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2422 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2423 btrfs_super_root(sb));
2426 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2427 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2428 btrfs_super_chunk_root(sb));
2431 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2432 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2433 btrfs_super_log_root(sb));
2437 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2438 BTRFS_FSID_SIZE) != 0) {
2440 "dev_item UUID does not match metadata fsid: %pU != %pU",
2441 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2446 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2449 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2450 btrfs_err(fs_info, "bytes_used is too small %llu",
2451 btrfs_super_bytes_used(sb));
2454 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2455 btrfs_err(fs_info, "invalid stripesize %u",
2456 btrfs_super_stripesize(sb));
2459 if (btrfs_super_num_devices(sb) > (1UL << 31))
2460 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2461 btrfs_super_num_devices(sb));
2462 if (btrfs_super_num_devices(sb) == 0) {
2463 btrfs_err(fs_info, "number of devices is 0");
2467 if (mirror_num >= 0 &&
2468 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2469 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2470 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2475 * Obvious sys_chunk_array corruptions, it must hold at least one key
2478 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2479 btrfs_err(fs_info, "system chunk array too big %u > %u",
2480 btrfs_super_sys_array_size(sb),
2481 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2484 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2485 + sizeof(struct btrfs_chunk)) {
2486 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2487 btrfs_super_sys_array_size(sb),
2488 sizeof(struct btrfs_disk_key)
2489 + sizeof(struct btrfs_chunk));
2494 * The generation is a global counter, we'll trust it more than the others
2495 * but it's still possible that it's the one that's wrong.
2497 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2499 "suspicious: generation < chunk_root_generation: %llu < %llu",
2500 btrfs_super_generation(sb),
2501 btrfs_super_chunk_root_generation(sb));
2502 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2503 && btrfs_super_cache_generation(sb) != (u64)-1)
2505 "suspicious: generation < cache_generation: %llu < %llu",
2506 btrfs_super_generation(sb),
2507 btrfs_super_cache_generation(sb));
2513 * Validation of super block at mount time.
2514 * Some checks already done early at mount time, like csum type and incompat
2515 * flags will be skipped.
2517 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2519 return validate_super(fs_info, fs_info->super_copy, 0);
2523 * Validation of super block at write time.
2524 * Some checks like bytenr check will be skipped as their values will be
2526 * Extra checks like csum type and incompat flags will be done here.
2528 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2529 struct btrfs_super_block *sb)
2533 ret = validate_super(fs_info, sb, -1);
2536 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2538 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2539 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2542 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2545 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2546 btrfs_super_incompat_flags(sb),
2547 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2553 "super block corruption detected before writing it to disk");
2557 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2559 int backup_index = find_newest_super_backup(fs_info);
2560 struct btrfs_super_block *sb = fs_info->super_copy;
2561 struct btrfs_root *tree_root = fs_info->tree_root;
2562 bool handle_error = false;
2566 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2571 if (!IS_ERR(tree_root->node))
2572 free_extent_buffer(tree_root->node);
2573 tree_root->node = NULL;
2575 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2578 free_root_pointers(fs_info, 0);
2581 * Don't use the log in recovery mode, it won't be
2584 btrfs_set_super_log_root(sb, 0);
2586 /* We can't trust the free space cache either */
2587 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2589 ret = read_backup_root(fs_info, i);
2594 generation = btrfs_super_generation(sb);
2595 level = btrfs_super_root_level(sb);
2596 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2597 generation, level, NULL);
2598 if (IS_ERR(tree_root->node) ||
2599 !extent_buffer_uptodate(tree_root->node)) {
2600 handle_error = true;
2602 if (IS_ERR(tree_root->node))
2603 ret = PTR_ERR(tree_root->node);
2604 else if (!extent_buffer_uptodate(tree_root->node))
2607 btrfs_warn(fs_info, "failed to read tree root");
2611 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2612 tree_root->commit_root = btrfs_root_node(tree_root);
2613 btrfs_set_root_refs(&tree_root->root_item, 1);
2616 * No need to hold btrfs_root::objectid_mutex since the fs
2617 * hasn't been fully initialised and we are the only user
2619 ret = btrfs_find_highest_objectid(tree_root,
2620 &tree_root->highest_objectid);
2622 handle_error = true;
2626 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2628 ret = btrfs_read_roots(fs_info);
2630 handle_error = true;
2634 /* All successful */
2635 fs_info->generation = generation;
2636 fs_info->last_trans_committed = generation;
2638 /* Always begin writing backup roots after the one being used */
2639 if (backup_index < 0) {
2640 fs_info->backup_root_index = 0;
2642 fs_info->backup_root_index = backup_index + 1;
2643 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2651 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2653 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2654 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2655 INIT_LIST_HEAD(&fs_info->trans_list);
2656 INIT_LIST_HEAD(&fs_info->dead_roots);
2657 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2658 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2659 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2660 spin_lock_init(&fs_info->delalloc_root_lock);
2661 spin_lock_init(&fs_info->trans_lock);
2662 spin_lock_init(&fs_info->fs_roots_radix_lock);
2663 spin_lock_init(&fs_info->delayed_iput_lock);
2664 spin_lock_init(&fs_info->defrag_inodes_lock);
2665 spin_lock_init(&fs_info->super_lock);
2666 spin_lock_init(&fs_info->buffer_lock);
2667 spin_lock_init(&fs_info->unused_bgs_lock);
2668 rwlock_init(&fs_info->tree_mod_log_lock);
2669 mutex_init(&fs_info->unused_bg_unpin_mutex);
2670 mutex_init(&fs_info->delete_unused_bgs_mutex);
2671 mutex_init(&fs_info->reloc_mutex);
2672 mutex_init(&fs_info->delalloc_root_mutex);
2673 seqlock_init(&fs_info->profiles_lock);
2675 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2676 INIT_LIST_HEAD(&fs_info->space_info);
2677 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2678 INIT_LIST_HEAD(&fs_info->unused_bgs);
2679 #ifdef CONFIG_BTRFS_DEBUG
2680 INIT_LIST_HEAD(&fs_info->allocated_roots);
2682 extent_map_tree_init(&fs_info->mapping_tree);
2683 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2684 BTRFS_BLOCK_RSV_GLOBAL);
2685 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2686 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2687 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2688 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2689 BTRFS_BLOCK_RSV_DELOPS);
2690 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2691 BTRFS_BLOCK_RSV_DELREFS);
2693 atomic_set(&fs_info->async_delalloc_pages, 0);
2694 atomic_set(&fs_info->defrag_running, 0);
2695 atomic_set(&fs_info->reada_works_cnt, 0);
2696 atomic_set(&fs_info->nr_delayed_iputs, 0);
2697 atomic64_set(&fs_info->tree_mod_seq, 0);
2698 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2699 fs_info->metadata_ratio = 0;
2700 fs_info->defrag_inodes = RB_ROOT;
2701 atomic64_set(&fs_info->free_chunk_space, 0);
2702 fs_info->tree_mod_log = RB_ROOT;
2703 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2704 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2705 /* readahead state */
2706 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2707 spin_lock_init(&fs_info->reada_lock);
2708 btrfs_init_ref_verify(fs_info);
2710 fs_info->thread_pool_size = min_t(unsigned long,
2711 num_online_cpus() + 2, 8);
2713 INIT_LIST_HEAD(&fs_info->ordered_roots);
2714 spin_lock_init(&fs_info->ordered_root_lock);
2716 btrfs_init_scrub(fs_info);
2717 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2718 fs_info->check_integrity_print_mask = 0;
2720 btrfs_init_balance(fs_info);
2721 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2723 spin_lock_init(&fs_info->block_group_cache_lock);
2724 fs_info->block_group_cache_tree = RB_ROOT;
2725 fs_info->first_logical_byte = (u64)-1;
2727 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2728 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2729 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2731 mutex_init(&fs_info->ordered_operations_mutex);
2732 mutex_init(&fs_info->tree_log_mutex);
2733 mutex_init(&fs_info->chunk_mutex);
2734 mutex_init(&fs_info->transaction_kthread_mutex);
2735 mutex_init(&fs_info->cleaner_mutex);
2736 mutex_init(&fs_info->ro_block_group_mutex);
2737 init_rwsem(&fs_info->commit_root_sem);
2738 init_rwsem(&fs_info->cleanup_work_sem);
2739 init_rwsem(&fs_info->subvol_sem);
2740 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2742 btrfs_init_dev_replace_locks(fs_info);
2743 btrfs_init_qgroup(fs_info);
2744 btrfs_discard_init(fs_info);
2746 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2747 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2749 init_waitqueue_head(&fs_info->transaction_throttle);
2750 init_waitqueue_head(&fs_info->transaction_wait);
2751 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2752 init_waitqueue_head(&fs_info->async_submit_wait);
2753 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2755 /* Usable values until the real ones are cached from the superblock */
2756 fs_info->nodesize = 4096;
2757 fs_info->sectorsize = 4096;
2758 fs_info->stripesize = 4096;
2760 spin_lock_init(&fs_info->swapfile_pins_lock);
2761 fs_info->swapfile_pins = RB_ROOT;
2763 fs_info->send_in_progress = 0;
2766 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2771 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2772 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2774 ret = init_srcu_struct(&fs_info->subvol_srcu);
2778 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2782 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2786 fs_info->dirty_metadata_batch = PAGE_SIZE *
2787 (1 + ilog2(nr_cpu_ids));
2789 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2793 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2798 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2800 if (!fs_info->delayed_root) {
2804 btrfs_init_delayed_root(fs_info->delayed_root);
2806 ret = btrfs_alloc_stripe_hash_table(fs_info);
2812 cleanup_srcu_struct(&fs_info->subvol_srcu);
2816 static int btrfs_uuid_rescan_kthread(void *data)
2818 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2822 * 1st step is to iterate through the existing UUID tree and
2823 * to delete all entries that contain outdated data.
2824 * 2nd step is to add all missing entries to the UUID tree.
2826 ret = btrfs_uuid_tree_iterate(fs_info);
2829 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2831 up(&fs_info->uuid_tree_rescan_sem);
2834 return btrfs_uuid_scan_kthread(data);
2837 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2839 struct task_struct *task;
2841 down(&fs_info->uuid_tree_rescan_sem);
2842 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2844 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2845 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2846 up(&fs_info->uuid_tree_rescan_sem);
2847 return PTR_ERR(task);
2853 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2862 struct btrfs_key location;
2863 struct btrfs_super_block *disk_super;
2864 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2865 struct btrfs_root *tree_root;
2866 struct btrfs_root *chunk_root;
2869 int clear_free_space_tree = 0;
2872 ret = init_mount_fs_info(fs_info, sb);
2878 /* These need to be init'ed before we start creating inodes and such. */
2879 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2881 fs_info->tree_root = tree_root;
2882 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2884 fs_info->chunk_root = chunk_root;
2885 if (!tree_root || !chunk_root) {
2890 fs_info->btree_inode = new_inode(sb);
2891 if (!fs_info->btree_inode) {
2895 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2896 btrfs_init_btree_inode(fs_info);
2898 invalidate_bdev(fs_devices->latest_bdev);
2901 * Read super block and check the signature bytes only
2903 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2904 if (IS_ERR(disk_super)) {
2905 err = PTR_ERR(disk_super);
2910 * Verify the type first, if that or the the checksum value are
2911 * corrupted, we'll find out
2913 csum_type = btrfs_super_csum_type(disk_super);
2914 if (!btrfs_supported_super_csum(csum_type)) {
2915 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2918 btrfs_release_disk_super(disk_super);
2922 ret = btrfs_init_csum_hash(fs_info, csum_type);
2925 btrfs_release_disk_super(disk_super);
2930 * We want to check superblock checksum, the type is stored inside.
2931 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2933 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2934 btrfs_err(fs_info, "superblock checksum mismatch");
2936 btrfs_release_disk_super(disk_super);
2941 * super_copy is zeroed at allocation time and we never touch the
2942 * following bytes up to INFO_SIZE, the checksum is calculated from
2943 * the whole block of INFO_SIZE
2945 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2946 btrfs_release_disk_super(disk_super);
2948 disk_super = fs_info->super_copy;
2950 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2953 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2954 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2955 fs_info->super_copy->metadata_uuid,
2959 features = btrfs_super_flags(disk_super);
2960 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2961 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2962 btrfs_set_super_flags(disk_super, features);
2964 "found metadata UUID change in progress flag, clearing");
2967 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2968 sizeof(*fs_info->super_for_commit));
2970 ret = btrfs_validate_mount_super(fs_info);
2972 btrfs_err(fs_info, "superblock contains fatal errors");
2977 if (!btrfs_super_root(disk_super))
2980 /* check FS state, whether FS is broken. */
2981 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2982 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2985 * In the long term, we'll store the compression type in the super
2986 * block, and it'll be used for per file compression control.
2988 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2990 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2996 features = btrfs_super_incompat_flags(disk_super) &
2997 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3000 "cannot mount because of unsupported optional features (%llx)",
3006 features = btrfs_super_incompat_flags(disk_super);
3007 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3008 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3009 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3010 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3011 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3013 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3014 btrfs_info(fs_info, "has skinny extents");
3017 * flag our filesystem as having big metadata blocks if
3018 * they are bigger than the page size
3020 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3021 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3023 "flagging fs with big metadata feature");
3024 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3027 nodesize = btrfs_super_nodesize(disk_super);
3028 sectorsize = btrfs_super_sectorsize(disk_super);
3029 stripesize = sectorsize;
3030 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3031 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3033 /* Cache block sizes */
3034 fs_info->nodesize = nodesize;
3035 fs_info->sectorsize = sectorsize;
3036 fs_info->stripesize = stripesize;
3039 * mixed block groups end up with duplicate but slightly offset
3040 * extent buffers for the same range. It leads to corruptions
3042 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3043 (sectorsize != nodesize)) {
3045 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3046 nodesize, sectorsize);
3051 * Needn't use the lock because there is no other task which will
3054 btrfs_set_super_incompat_flags(disk_super, features);
3056 features = btrfs_super_compat_ro_flags(disk_super) &
3057 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3058 if (!sb_rdonly(sb) && features) {
3060 "cannot mount read-write because of unsupported optional features (%llx)",
3066 ret = btrfs_init_workqueues(fs_info, fs_devices);
3069 goto fail_sb_buffer;
3072 sb->s_bdi->congested_fn = btrfs_congested_fn;
3073 sb->s_bdi->congested_data = fs_info;
3074 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3075 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3076 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3077 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3079 sb->s_blocksize = sectorsize;
3080 sb->s_blocksize_bits = blksize_bits(sectorsize);
3081 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3083 mutex_lock(&fs_info->chunk_mutex);
3084 ret = btrfs_read_sys_array(fs_info);
3085 mutex_unlock(&fs_info->chunk_mutex);
3087 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3088 goto fail_sb_buffer;
3091 generation = btrfs_super_chunk_root_generation(disk_super);
3092 level = btrfs_super_chunk_root_level(disk_super);
3094 chunk_root->node = read_tree_block(fs_info,
3095 btrfs_super_chunk_root(disk_super),
3096 generation, level, NULL);
3097 if (IS_ERR(chunk_root->node) ||
3098 !extent_buffer_uptodate(chunk_root->node)) {
3099 btrfs_err(fs_info, "failed to read chunk root");
3100 if (!IS_ERR(chunk_root->node))
3101 free_extent_buffer(chunk_root->node);
3102 chunk_root->node = NULL;
3103 goto fail_tree_roots;
3105 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3106 chunk_root->commit_root = btrfs_root_node(chunk_root);
3108 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3109 offsetof(struct btrfs_header, chunk_tree_uuid),
3112 ret = btrfs_read_chunk_tree(fs_info);
3114 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3115 goto fail_tree_roots;
3119 * Keep the devid that is marked to be the target device for the
3120 * device replace procedure
3122 btrfs_free_extra_devids(fs_devices, 0);
3124 if (!fs_devices->latest_bdev) {
3125 btrfs_err(fs_info, "failed to read devices");
3126 goto fail_tree_roots;
3129 ret = init_tree_roots(fs_info);
3131 goto fail_tree_roots;
3134 * If we have a uuid root and we're not being told to rescan we need to
3135 * check the generation here so we can set the
3136 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3137 * transaction during a balance or the log replay without updating the
3138 * uuid generation, and then if we crash we would rescan the uuid tree,
3139 * even though it was perfectly fine.
3141 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3142 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3143 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3145 ret = btrfs_verify_dev_extents(fs_info);
3148 "failed to verify dev extents against chunks: %d",
3150 goto fail_block_groups;
3152 ret = btrfs_recover_balance(fs_info);
3154 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3155 goto fail_block_groups;
3158 ret = btrfs_init_dev_stats(fs_info);
3160 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3161 goto fail_block_groups;
3164 ret = btrfs_init_dev_replace(fs_info);
3166 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3167 goto fail_block_groups;
3170 btrfs_free_extra_devids(fs_devices, 1);
3172 ret = btrfs_sysfs_add_fsid(fs_devices);
3174 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3176 goto fail_block_groups;
3179 ret = btrfs_sysfs_add_mounted(fs_info);
3181 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3182 goto fail_fsdev_sysfs;
3185 ret = btrfs_init_space_info(fs_info);
3187 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3191 ret = btrfs_read_block_groups(fs_info);
3193 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3197 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3199 "writable mount is not allowed due to too many missing devices");
3203 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3205 if (IS_ERR(fs_info->cleaner_kthread))
3208 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3210 "btrfs-transaction");
3211 if (IS_ERR(fs_info->transaction_kthread))
3214 if (!btrfs_test_opt(fs_info, NOSSD) &&
3215 !fs_info->fs_devices->rotating) {
3216 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3220 * Mount does not set all options immediately, we can do it now and do
3221 * not have to wait for transaction commit
3223 btrfs_apply_pending_changes(fs_info);
3225 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3226 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3227 ret = btrfsic_mount(fs_info, fs_devices,
3228 btrfs_test_opt(fs_info,
3229 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3231 fs_info->check_integrity_print_mask);
3234 "failed to initialize integrity check module: %d",
3238 ret = btrfs_read_qgroup_config(fs_info);
3240 goto fail_trans_kthread;
3242 if (btrfs_build_ref_tree(fs_info))
3243 btrfs_err(fs_info, "couldn't build ref tree");
3245 /* do not make disk changes in broken FS or nologreplay is given */
3246 if (btrfs_super_log_root(disk_super) != 0 &&
3247 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3248 btrfs_info(fs_info, "start tree-log replay");
3249 ret = btrfs_replay_log(fs_info, fs_devices);
3256 ret = btrfs_find_orphan_roots(fs_info);
3260 if (!sb_rdonly(sb)) {
3261 ret = btrfs_cleanup_fs_roots(fs_info);
3265 mutex_lock(&fs_info->cleaner_mutex);
3266 ret = btrfs_recover_relocation(tree_root);
3267 mutex_unlock(&fs_info->cleaner_mutex);
3269 btrfs_warn(fs_info, "failed to recover relocation: %d",
3276 location.objectid = BTRFS_FS_TREE_OBJECTID;
3277 location.type = BTRFS_ROOT_ITEM_KEY;
3278 location.offset = 0;
3280 fs_info->fs_root = btrfs_get_fs_root(fs_info, &location, true);
3281 if (IS_ERR(fs_info->fs_root)) {
3282 err = PTR_ERR(fs_info->fs_root);
3283 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3284 fs_info->fs_root = NULL;
3291 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3292 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3293 clear_free_space_tree = 1;
3294 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3295 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3296 btrfs_warn(fs_info, "free space tree is invalid");
3297 clear_free_space_tree = 1;
3300 if (clear_free_space_tree) {
3301 btrfs_info(fs_info, "clearing free space tree");
3302 ret = btrfs_clear_free_space_tree(fs_info);
3305 "failed to clear free space tree: %d", ret);
3306 close_ctree(fs_info);
3311 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3312 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3313 btrfs_info(fs_info, "creating free space tree");
3314 ret = btrfs_create_free_space_tree(fs_info);
3317 "failed to create free space tree: %d", ret);
3318 close_ctree(fs_info);
3323 down_read(&fs_info->cleanup_work_sem);
3324 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3325 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3326 up_read(&fs_info->cleanup_work_sem);
3327 close_ctree(fs_info);
3330 up_read(&fs_info->cleanup_work_sem);
3332 ret = btrfs_resume_balance_async(fs_info);
3334 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3335 close_ctree(fs_info);
3339 ret = btrfs_resume_dev_replace_async(fs_info);
3341 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3342 close_ctree(fs_info);
3346 btrfs_qgroup_rescan_resume(fs_info);
3347 btrfs_discard_resume(fs_info);
3349 if (!fs_info->uuid_root) {
3350 btrfs_info(fs_info, "creating UUID tree");
3351 ret = btrfs_create_uuid_tree(fs_info);
3354 "failed to create the UUID tree: %d", ret);
3355 close_ctree(fs_info);
3358 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3359 fs_info->generation !=
3360 btrfs_super_uuid_tree_generation(disk_super)) {
3361 btrfs_info(fs_info, "checking UUID tree");
3362 ret = btrfs_check_uuid_tree(fs_info);
3365 "failed to check the UUID tree: %d", ret);
3366 close_ctree(fs_info);
3370 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3373 * backuproot only affect mount behavior, and if open_ctree succeeded,
3374 * no need to keep the flag
3376 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3381 btrfs_free_qgroup_config(fs_info);
3383 kthread_stop(fs_info->transaction_kthread);
3384 btrfs_cleanup_transaction(fs_info);
3385 btrfs_free_fs_roots(fs_info);
3387 kthread_stop(fs_info->cleaner_kthread);
3390 * make sure we're done with the btree inode before we stop our
3393 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3396 btrfs_sysfs_remove_mounted(fs_info);
3399 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3402 btrfs_put_block_group_cache(fs_info);
3405 free_root_pointers(fs_info, true);
3406 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3409 btrfs_stop_all_workers(fs_info);
3410 btrfs_free_block_groups(fs_info);
3412 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3414 iput(fs_info->btree_inode);
3416 cleanup_srcu_struct(&fs_info->subvol_srcu);
3418 btrfs_close_devices(fs_info->fs_devices);
3421 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3423 static void btrfs_end_super_write(struct bio *bio)
3425 struct btrfs_device *device = bio->bi_private;
3426 struct bio_vec *bvec;
3427 struct bvec_iter_all iter_all;
3430 bio_for_each_segment_all(bvec, bio, iter_all) {
3431 page = bvec->bv_page;
3433 if (bio->bi_status) {
3434 btrfs_warn_rl_in_rcu(device->fs_info,
3435 "lost page write due to IO error on %s (%d)",
3436 rcu_str_deref(device->name),
3437 blk_status_to_errno(bio->bi_status));
3438 ClearPageUptodate(page);
3440 btrfs_dev_stat_inc_and_print(device,
3441 BTRFS_DEV_STAT_WRITE_ERRS);
3443 SetPageUptodate(page);
3453 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3456 struct btrfs_super_block *super;
3459 struct address_space *mapping = bdev->bd_inode->i_mapping;
3461 bytenr = btrfs_sb_offset(copy_num);
3462 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3463 return ERR_PTR(-EINVAL);
3465 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3467 return ERR_CAST(page);
3469 super = page_address(page);
3470 if (btrfs_super_bytenr(super) != bytenr ||
3471 btrfs_super_magic(super) != BTRFS_MAGIC) {
3472 btrfs_release_disk_super(super);
3473 return ERR_PTR(-EINVAL);
3480 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3482 struct btrfs_super_block *super, *latest = NULL;
3486 /* we would like to check all the supers, but that would make
3487 * a btrfs mount succeed after a mkfs from a different FS.
3488 * So, we need to add a special mount option to scan for
3489 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3491 for (i = 0; i < 1; i++) {
3492 super = btrfs_read_dev_one_super(bdev, i);
3496 if (!latest || btrfs_super_generation(super) > transid) {
3498 btrfs_release_disk_super(super);
3501 transid = btrfs_super_generation(super);
3509 * Write superblock @sb to the @device. Do not wait for completion, all the
3510 * pages we use for writing are locked.
3512 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3513 * the expected device size at commit time. Note that max_mirrors must be
3514 * same for write and wait phases.
3516 * Return number of errors when page is not found or submission fails.
3518 static int write_dev_supers(struct btrfs_device *device,
3519 struct btrfs_super_block *sb, int max_mirrors)
3521 struct btrfs_fs_info *fs_info = device->fs_info;
3522 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3523 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3528 if (max_mirrors == 0)
3529 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3531 shash->tfm = fs_info->csum_shash;
3533 for (i = 0; i < max_mirrors; i++) {
3536 struct btrfs_super_block *disk_super;
3538 bytenr = btrfs_sb_offset(i);
3539 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3540 device->commit_total_bytes)
3543 btrfs_set_super_bytenr(sb, bytenr);
3545 crypto_shash_init(shash);
3546 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3547 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3548 crypto_shash_final(shash, sb->csum);
3550 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3553 btrfs_err(device->fs_info,
3554 "couldn't get super block page for bytenr %llu",
3560 /* Bump the refcount for wait_dev_supers() */
3563 disk_super = page_address(page);
3564 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3567 * Directly use bios here instead of relying on the page cache
3568 * to do I/O, so we don't lose the ability to do integrity
3571 bio = bio_alloc(GFP_NOFS, 1);
3572 bio_set_dev(bio, device->bdev);
3573 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3574 bio->bi_private = device;
3575 bio->bi_end_io = btrfs_end_super_write;
3576 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3577 offset_in_page(bytenr));
3580 * We FUA only the first super block. The others we allow to
3581 * go down lazy and there's a short window where the on-disk
3582 * copies might still contain the older version.
3584 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3585 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3586 bio->bi_opf |= REQ_FUA;
3588 btrfsic_submit_bio(bio);
3590 return errors < i ? 0 : -1;
3594 * Wait for write completion of superblocks done by write_dev_supers,
3595 * @max_mirrors same for write and wait phases.
3597 * Return number of errors when page is not found or not marked up to
3600 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3604 bool primary_failed = false;
3607 if (max_mirrors == 0)
3608 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3610 for (i = 0; i < max_mirrors; i++) {
3613 bytenr = btrfs_sb_offset(i);
3614 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3615 device->commit_total_bytes)
3618 page = find_get_page(device->bdev->bd_inode->i_mapping,
3619 bytenr >> PAGE_SHIFT);
3623 primary_failed = true;
3626 /* Page is submitted locked and unlocked once the IO completes */
3627 wait_on_page_locked(page);
3628 if (PageError(page)) {
3631 primary_failed = true;
3634 /* Drop our reference */
3637 /* Drop the reference from the writing run */
3641 /* log error, force error return */
3642 if (primary_failed) {
3643 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3648 return errors < i ? 0 : -1;
3652 * endio for the write_dev_flush, this will wake anyone waiting
3653 * for the barrier when it is done
3655 static void btrfs_end_empty_barrier(struct bio *bio)
3657 complete(bio->bi_private);
3661 * Submit a flush request to the device if it supports it. Error handling is
3662 * done in the waiting counterpart.
3664 static void write_dev_flush(struct btrfs_device *device)
3666 struct request_queue *q = bdev_get_queue(device->bdev);
3667 struct bio *bio = device->flush_bio;
3669 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3673 bio->bi_end_io = btrfs_end_empty_barrier;
3674 bio_set_dev(bio, device->bdev);
3675 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3676 init_completion(&device->flush_wait);
3677 bio->bi_private = &device->flush_wait;
3679 btrfsic_submit_bio(bio);
3680 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3684 * If the flush bio has been submitted by write_dev_flush, wait for it.
3686 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3688 struct bio *bio = device->flush_bio;
3690 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3693 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3694 wait_for_completion_io(&device->flush_wait);
3696 return bio->bi_status;
3699 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3701 if (!btrfs_check_rw_degradable(fs_info, NULL))
3707 * send an empty flush down to each device in parallel,
3708 * then wait for them
3710 static int barrier_all_devices(struct btrfs_fs_info *info)
3712 struct list_head *head;
3713 struct btrfs_device *dev;
3714 int errors_wait = 0;
3717 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3718 /* send down all the barriers */
3719 head = &info->fs_devices->devices;
3720 list_for_each_entry(dev, head, dev_list) {
3721 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3725 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3726 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3729 write_dev_flush(dev);
3730 dev->last_flush_error = BLK_STS_OK;
3733 /* wait for all the barriers */
3734 list_for_each_entry(dev, head, dev_list) {
3735 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3741 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3742 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3745 ret = wait_dev_flush(dev);
3747 dev->last_flush_error = ret;
3748 btrfs_dev_stat_inc_and_print(dev,
3749 BTRFS_DEV_STAT_FLUSH_ERRS);
3756 * At some point we need the status of all disks
3757 * to arrive at the volume status. So error checking
3758 * is being pushed to a separate loop.
3760 return check_barrier_error(info);
3765 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3768 int min_tolerated = INT_MAX;
3770 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3771 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3772 min_tolerated = min_t(int, min_tolerated,
3773 btrfs_raid_array[BTRFS_RAID_SINGLE].
3774 tolerated_failures);
3776 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3777 if (raid_type == BTRFS_RAID_SINGLE)
3779 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3781 min_tolerated = min_t(int, min_tolerated,
3782 btrfs_raid_array[raid_type].
3783 tolerated_failures);
3786 if (min_tolerated == INT_MAX) {
3787 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3791 return min_tolerated;
3794 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3796 struct list_head *head;
3797 struct btrfs_device *dev;
3798 struct btrfs_super_block *sb;
3799 struct btrfs_dev_item *dev_item;
3803 int total_errors = 0;
3806 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3809 * max_mirrors == 0 indicates we're from commit_transaction,
3810 * not from fsync where the tree roots in fs_info have not
3811 * been consistent on disk.
3813 if (max_mirrors == 0)
3814 backup_super_roots(fs_info);
3816 sb = fs_info->super_for_commit;
3817 dev_item = &sb->dev_item;
3819 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3820 head = &fs_info->fs_devices->devices;
3821 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3824 ret = barrier_all_devices(fs_info);
3827 &fs_info->fs_devices->device_list_mutex);
3828 btrfs_handle_fs_error(fs_info, ret,
3829 "errors while submitting device barriers.");
3834 list_for_each_entry(dev, head, dev_list) {
3839 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3840 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3843 btrfs_set_stack_device_generation(dev_item, 0);
3844 btrfs_set_stack_device_type(dev_item, dev->type);
3845 btrfs_set_stack_device_id(dev_item, dev->devid);
3846 btrfs_set_stack_device_total_bytes(dev_item,
3847 dev->commit_total_bytes);
3848 btrfs_set_stack_device_bytes_used(dev_item,
3849 dev->commit_bytes_used);
3850 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3851 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3852 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3853 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3854 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3857 flags = btrfs_super_flags(sb);
3858 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3860 ret = btrfs_validate_write_super(fs_info, sb);
3862 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3863 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3864 "unexpected superblock corruption detected");
3868 ret = write_dev_supers(dev, sb, max_mirrors);
3872 if (total_errors > max_errors) {
3873 btrfs_err(fs_info, "%d errors while writing supers",
3875 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3877 /* FUA is masked off if unsupported and can't be the reason */
3878 btrfs_handle_fs_error(fs_info, -EIO,
3879 "%d errors while writing supers",
3885 list_for_each_entry(dev, head, dev_list) {
3888 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3889 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3892 ret = wait_dev_supers(dev, max_mirrors);
3896 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3897 if (total_errors > max_errors) {
3898 btrfs_handle_fs_error(fs_info, -EIO,
3899 "%d errors while writing supers",
3906 /* Drop a fs root from the radix tree and free it. */
3907 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3908 struct btrfs_root *root)
3910 spin_lock(&fs_info->fs_roots_radix_lock);
3911 radix_tree_delete(&fs_info->fs_roots_radix,
3912 (unsigned long)root->root_key.objectid);
3913 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3914 btrfs_put_root(root);
3915 spin_unlock(&fs_info->fs_roots_radix_lock);
3917 if (btrfs_root_refs(&root->root_item) == 0)
3918 synchronize_srcu(&fs_info->subvol_srcu);
3920 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3921 btrfs_free_log(NULL, root);
3922 if (root->reloc_root) {
3923 free_extent_buffer(root->reloc_root->node);
3924 free_extent_buffer(root->reloc_root->commit_root);
3925 btrfs_put_root(root->reloc_root);
3926 root->reloc_root = NULL;
3930 if (root->free_ino_pinned)
3931 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3932 if (root->free_ino_ctl)
3933 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3934 btrfs_free_fs_root(root);
3937 void btrfs_free_fs_root(struct btrfs_root *root)
3939 iput(root->ino_cache_inode);
3940 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3942 free_anon_bdev(root->anon_dev);
3943 btrfs_drew_lock_destroy(&root->snapshot_lock);
3944 free_extent_buffer(root->node);
3945 free_extent_buffer(root->commit_root);
3946 kfree(root->free_ino_ctl);
3947 kfree(root->free_ino_pinned);
3948 btrfs_put_root(root);
3951 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3953 u64 root_objectid = 0;
3954 struct btrfs_root *gang[8];
3957 unsigned int ret = 0;
3961 index = srcu_read_lock(&fs_info->subvol_srcu);
3962 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3963 (void **)gang, root_objectid,
3966 srcu_read_unlock(&fs_info->subvol_srcu, index);
3969 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3971 for (i = 0; i < ret; i++) {
3972 /* Avoid to grab roots in dead_roots */
3973 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3977 /* grab all the search result for later use */
3978 gang[i] = btrfs_grab_root(gang[i]);
3980 srcu_read_unlock(&fs_info->subvol_srcu, index);
3982 for (i = 0; i < ret; i++) {
3985 root_objectid = gang[i]->root_key.objectid;
3986 err = btrfs_orphan_cleanup(gang[i]);
3989 btrfs_put_root(gang[i]);
3994 /* release the uncleaned roots due to error */
3995 for (; i < ret; i++) {
3997 btrfs_put_root(gang[i]);
4002 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4004 struct btrfs_root *root = fs_info->tree_root;
4005 struct btrfs_trans_handle *trans;
4007 mutex_lock(&fs_info->cleaner_mutex);
4008 btrfs_run_delayed_iputs(fs_info);
4009 mutex_unlock(&fs_info->cleaner_mutex);
4010 wake_up_process(fs_info->cleaner_kthread);
4012 /* wait until ongoing cleanup work done */
4013 down_write(&fs_info->cleanup_work_sem);
4014 up_write(&fs_info->cleanup_work_sem);
4016 trans = btrfs_join_transaction(root);
4018 return PTR_ERR(trans);
4019 return btrfs_commit_transaction(trans);
4022 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4026 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4028 * We don't want the cleaner to start new transactions, add more delayed
4029 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4030 * because that frees the task_struct, and the transaction kthread might
4031 * still try to wake up the cleaner.
4033 kthread_park(fs_info->cleaner_kthread);
4035 /* wait for the qgroup rescan worker to stop */
4036 btrfs_qgroup_wait_for_completion(fs_info, false);
4038 /* wait for the uuid_scan task to finish */
4039 down(&fs_info->uuid_tree_rescan_sem);
4040 /* avoid complains from lockdep et al., set sem back to initial state */
4041 up(&fs_info->uuid_tree_rescan_sem);
4043 /* pause restriper - we want to resume on mount */
4044 btrfs_pause_balance(fs_info);
4046 btrfs_dev_replace_suspend_for_unmount(fs_info);
4048 btrfs_scrub_cancel(fs_info);
4050 /* wait for any defraggers to finish */
4051 wait_event(fs_info->transaction_wait,
4052 (atomic_read(&fs_info->defrag_running) == 0));
4054 /* clear out the rbtree of defraggable inodes */
4055 btrfs_cleanup_defrag_inodes(fs_info);
4057 cancel_work_sync(&fs_info->async_reclaim_work);
4059 /* Cancel or finish ongoing discard work */
4060 btrfs_discard_cleanup(fs_info);
4062 if (!sb_rdonly(fs_info->sb)) {
4064 * The cleaner kthread is stopped, so do one final pass over
4065 * unused block groups.
4067 btrfs_delete_unused_bgs(fs_info);
4069 ret = btrfs_commit_super(fs_info);
4071 btrfs_err(fs_info, "commit super ret %d", ret);
4074 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4075 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4076 btrfs_error_commit_super(fs_info);
4078 kthread_stop(fs_info->transaction_kthread);
4079 kthread_stop(fs_info->cleaner_kthread);
4081 ASSERT(list_empty(&fs_info->delayed_iputs));
4082 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4084 btrfs_free_qgroup_config(fs_info);
4085 ASSERT(list_empty(&fs_info->delalloc_roots));
4087 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4088 btrfs_info(fs_info, "at unmount delalloc count %lld",
4089 percpu_counter_sum(&fs_info->delalloc_bytes));
4092 if (percpu_counter_sum(&fs_info->dio_bytes))
4093 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4094 percpu_counter_sum(&fs_info->dio_bytes));
4096 btrfs_sysfs_remove_mounted(fs_info);
4097 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4099 btrfs_free_fs_roots(fs_info);
4101 btrfs_put_block_group_cache(fs_info);
4104 * we must make sure there is not any read request to
4105 * submit after we stopping all workers.
4107 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4108 btrfs_stop_all_workers(fs_info);
4110 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4111 free_root_pointers(fs_info, true);
4114 * We must free the block groups after dropping the fs_roots as we could
4115 * have had an IO error and have left over tree log blocks that aren't
4116 * cleaned up until the fs roots are freed. This makes the block group
4117 * accounting appear to be wrong because there's pending reserved bytes,
4118 * so make sure we do the block group cleanup afterwards.
4120 btrfs_free_block_groups(fs_info);
4122 iput(fs_info->btree_inode);
4124 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4125 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4126 btrfsic_unmount(fs_info->fs_devices);
4129 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4130 btrfs_close_devices(fs_info->fs_devices);
4131 cleanup_srcu_struct(&fs_info->subvol_srcu);
4134 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4138 struct inode *btree_inode = buf->pages[0]->mapping->host;
4140 ret = extent_buffer_uptodate(buf);
4144 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4145 parent_transid, atomic);
4151 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4153 struct btrfs_fs_info *fs_info;
4154 struct btrfs_root *root;
4155 u64 transid = btrfs_header_generation(buf);
4158 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4160 * This is a fast path so only do this check if we have sanity tests
4161 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4162 * outside of the sanity tests.
4164 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4167 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4168 fs_info = root->fs_info;
4169 btrfs_assert_tree_locked(buf);
4170 if (transid != fs_info->generation)
4171 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4172 buf->start, transid, fs_info->generation);
4173 was_dirty = set_extent_buffer_dirty(buf);
4175 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4177 fs_info->dirty_metadata_batch);
4178 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4180 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4181 * but item data not updated.
4182 * So here we should only check item pointers, not item data.
4184 if (btrfs_header_level(buf) == 0 &&
4185 btrfs_check_leaf_relaxed(buf)) {
4186 btrfs_print_leaf(buf);
4192 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4196 * looks as though older kernels can get into trouble with
4197 * this code, they end up stuck in balance_dirty_pages forever
4201 if (current->flags & PF_MEMALLOC)
4205 btrfs_balance_delayed_items(fs_info);
4207 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4208 BTRFS_DIRTY_METADATA_THRESH,
4209 fs_info->dirty_metadata_batch);
4211 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4215 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4217 __btrfs_btree_balance_dirty(fs_info, 1);
4220 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4222 __btrfs_btree_balance_dirty(fs_info, 0);
4225 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4226 struct btrfs_key *first_key)
4228 return btree_read_extent_buffer_pages(buf, parent_transid,
4232 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4234 /* cleanup FS via transaction */
4235 btrfs_cleanup_transaction(fs_info);
4237 mutex_lock(&fs_info->cleaner_mutex);
4238 btrfs_run_delayed_iputs(fs_info);
4239 mutex_unlock(&fs_info->cleaner_mutex);
4241 down_write(&fs_info->cleanup_work_sem);
4242 up_write(&fs_info->cleanup_work_sem);
4245 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4247 struct btrfs_ordered_extent *ordered;
4249 spin_lock(&root->ordered_extent_lock);
4251 * This will just short circuit the ordered completion stuff which will
4252 * make sure the ordered extent gets properly cleaned up.
4254 list_for_each_entry(ordered, &root->ordered_extents,
4256 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4257 spin_unlock(&root->ordered_extent_lock);
4260 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4262 struct btrfs_root *root;
4263 struct list_head splice;
4265 INIT_LIST_HEAD(&splice);
4267 spin_lock(&fs_info->ordered_root_lock);
4268 list_splice_init(&fs_info->ordered_roots, &splice);
4269 while (!list_empty(&splice)) {
4270 root = list_first_entry(&splice, struct btrfs_root,
4272 list_move_tail(&root->ordered_root,
4273 &fs_info->ordered_roots);
4275 spin_unlock(&fs_info->ordered_root_lock);
4276 btrfs_destroy_ordered_extents(root);
4279 spin_lock(&fs_info->ordered_root_lock);
4281 spin_unlock(&fs_info->ordered_root_lock);
4284 * We need this here because if we've been flipped read-only we won't
4285 * get sync() from the umount, so we need to make sure any ordered
4286 * extents that haven't had their dirty pages IO start writeout yet
4287 * actually get run and error out properly.
4289 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4292 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4293 struct btrfs_fs_info *fs_info)
4295 struct rb_node *node;
4296 struct btrfs_delayed_ref_root *delayed_refs;
4297 struct btrfs_delayed_ref_node *ref;
4300 delayed_refs = &trans->delayed_refs;
4302 spin_lock(&delayed_refs->lock);
4303 if (atomic_read(&delayed_refs->num_entries) == 0) {
4304 spin_unlock(&delayed_refs->lock);
4305 btrfs_debug(fs_info, "delayed_refs has NO entry");
4309 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4310 struct btrfs_delayed_ref_head *head;
4312 bool pin_bytes = false;
4314 head = rb_entry(node, struct btrfs_delayed_ref_head,
4316 if (btrfs_delayed_ref_lock(delayed_refs, head))
4319 spin_lock(&head->lock);
4320 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4321 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4324 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4325 RB_CLEAR_NODE(&ref->ref_node);
4326 if (!list_empty(&ref->add_list))
4327 list_del(&ref->add_list);
4328 atomic_dec(&delayed_refs->num_entries);
4329 btrfs_put_delayed_ref(ref);
4331 if (head->must_insert_reserved)
4333 btrfs_free_delayed_extent_op(head->extent_op);
4334 btrfs_delete_ref_head(delayed_refs, head);
4335 spin_unlock(&head->lock);
4336 spin_unlock(&delayed_refs->lock);
4337 mutex_unlock(&head->mutex);
4340 struct btrfs_block_group *cache;
4342 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4345 spin_lock(&cache->space_info->lock);
4346 spin_lock(&cache->lock);
4347 cache->pinned += head->num_bytes;
4348 btrfs_space_info_update_bytes_pinned(fs_info,
4349 cache->space_info, head->num_bytes);
4350 cache->reserved -= head->num_bytes;
4351 cache->space_info->bytes_reserved -= head->num_bytes;
4352 spin_unlock(&cache->lock);
4353 spin_unlock(&cache->space_info->lock);
4354 percpu_counter_add_batch(
4355 &cache->space_info->total_bytes_pinned,
4356 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4358 btrfs_put_block_group(cache);
4360 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4361 head->bytenr + head->num_bytes - 1);
4363 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4364 btrfs_put_delayed_ref_head(head);
4366 spin_lock(&delayed_refs->lock);
4368 btrfs_qgroup_destroy_extent_records(trans);
4370 spin_unlock(&delayed_refs->lock);
4375 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4377 struct btrfs_inode *btrfs_inode;
4378 struct list_head splice;
4380 INIT_LIST_HEAD(&splice);
4382 spin_lock(&root->delalloc_lock);
4383 list_splice_init(&root->delalloc_inodes, &splice);
4385 while (!list_empty(&splice)) {
4386 struct inode *inode = NULL;
4387 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4389 __btrfs_del_delalloc_inode(root, btrfs_inode);
4390 spin_unlock(&root->delalloc_lock);
4393 * Make sure we get a live inode and that it'll not disappear
4396 inode = igrab(&btrfs_inode->vfs_inode);
4398 invalidate_inode_pages2(inode->i_mapping);
4401 spin_lock(&root->delalloc_lock);
4403 spin_unlock(&root->delalloc_lock);
4406 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4408 struct btrfs_root *root;
4409 struct list_head splice;
4411 INIT_LIST_HEAD(&splice);
4413 spin_lock(&fs_info->delalloc_root_lock);
4414 list_splice_init(&fs_info->delalloc_roots, &splice);
4415 while (!list_empty(&splice)) {
4416 root = list_first_entry(&splice, struct btrfs_root,
4418 root = btrfs_grab_root(root);
4420 spin_unlock(&fs_info->delalloc_root_lock);
4422 btrfs_destroy_delalloc_inodes(root);
4423 btrfs_put_root(root);
4425 spin_lock(&fs_info->delalloc_root_lock);
4427 spin_unlock(&fs_info->delalloc_root_lock);
4430 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4431 struct extent_io_tree *dirty_pages,
4435 struct extent_buffer *eb;
4440 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4445 clear_extent_bits(dirty_pages, start, end, mark);
4446 while (start <= end) {
4447 eb = find_extent_buffer(fs_info, start);
4448 start += fs_info->nodesize;
4451 wait_on_extent_buffer_writeback(eb);
4453 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4455 clear_extent_buffer_dirty(eb);
4456 free_extent_buffer_stale(eb);
4463 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4464 struct extent_io_tree *unpin)
4471 struct extent_state *cached_state = NULL;
4474 * The btrfs_finish_extent_commit() may get the same range as
4475 * ours between find_first_extent_bit and clear_extent_dirty.
4476 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4477 * the same extent range.
4479 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4480 ret = find_first_extent_bit(unpin, 0, &start, &end,
4481 EXTENT_DIRTY, &cached_state);
4483 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4487 clear_extent_dirty(unpin, start, end, &cached_state);
4488 free_extent_state(cached_state);
4489 btrfs_error_unpin_extent_range(fs_info, start, end);
4490 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4497 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4499 struct inode *inode;
4501 inode = cache->io_ctl.inode;
4503 invalidate_inode_pages2(inode->i_mapping);
4504 BTRFS_I(inode)->generation = 0;
4505 cache->io_ctl.inode = NULL;
4508 btrfs_put_block_group(cache);
4511 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4512 struct btrfs_fs_info *fs_info)
4514 struct btrfs_block_group *cache;
4516 spin_lock(&cur_trans->dirty_bgs_lock);
4517 while (!list_empty(&cur_trans->dirty_bgs)) {
4518 cache = list_first_entry(&cur_trans->dirty_bgs,
4519 struct btrfs_block_group,
4522 if (!list_empty(&cache->io_list)) {
4523 spin_unlock(&cur_trans->dirty_bgs_lock);
4524 list_del_init(&cache->io_list);
4525 btrfs_cleanup_bg_io(cache);
4526 spin_lock(&cur_trans->dirty_bgs_lock);
4529 list_del_init(&cache->dirty_list);
4530 spin_lock(&cache->lock);
4531 cache->disk_cache_state = BTRFS_DC_ERROR;
4532 spin_unlock(&cache->lock);
4534 spin_unlock(&cur_trans->dirty_bgs_lock);
4535 btrfs_put_block_group(cache);
4536 btrfs_delayed_refs_rsv_release(fs_info, 1);
4537 spin_lock(&cur_trans->dirty_bgs_lock);
4539 spin_unlock(&cur_trans->dirty_bgs_lock);
4542 * Refer to the definition of io_bgs member for details why it's safe
4543 * to use it without any locking
4545 while (!list_empty(&cur_trans->io_bgs)) {
4546 cache = list_first_entry(&cur_trans->io_bgs,
4547 struct btrfs_block_group,
4550 list_del_init(&cache->io_list);
4551 spin_lock(&cache->lock);
4552 cache->disk_cache_state = BTRFS_DC_ERROR;
4553 spin_unlock(&cache->lock);
4554 btrfs_cleanup_bg_io(cache);
4558 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4559 struct btrfs_fs_info *fs_info)
4561 struct btrfs_device *dev, *tmp;
4563 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4564 ASSERT(list_empty(&cur_trans->dirty_bgs));
4565 ASSERT(list_empty(&cur_trans->io_bgs));
4567 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4569 list_del_init(&dev->post_commit_list);
4572 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4574 cur_trans->state = TRANS_STATE_COMMIT_START;
4575 wake_up(&fs_info->transaction_blocked_wait);
4577 cur_trans->state = TRANS_STATE_UNBLOCKED;
4578 wake_up(&fs_info->transaction_wait);
4580 btrfs_destroy_delayed_inodes(fs_info);
4582 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4584 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4586 cur_trans->state =TRANS_STATE_COMPLETED;
4587 wake_up(&cur_trans->commit_wait);
4590 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4592 struct btrfs_transaction *t;
4594 mutex_lock(&fs_info->transaction_kthread_mutex);
4596 spin_lock(&fs_info->trans_lock);
4597 while (!list_empty(&fs_info->trans_list)) {
4598 t = list_first_entry(&fs_info->trans_list,
4599 struct btrfs_transaction, list);
4600 if (t->state >= TRANS_STATE_COMMIT_START) {
4601 refcount_inc(&t->use_count);
4602 spin_unlock(&fs_info->trans_lock);
4603 btrfs_wait_for_commit(fs_info, t->transid);
4604 btrfs_put_transaction(t);
4605 spin_lock(&fs_info->trans_lock);
4608 if (t == fs_info->running_transaction) {
4609 t->state = TRANS_STATE_COMMIT_DOING;
4610 spin_unlock(&fs_info->trans_lock);
4612 * We wait for 0 num_writers since we don't hold a trans
4613 * handle open currently for this transaction.
4615 wait_event(t->writer_wait,
4616 atomic_read(&t->num_writers) == 0);
4618 spin_unlock(&fs_info->trans_lock);
4620 btrfs_cleanup_one_transaction(t, fs_info);
4622 spin_lock(&fs_info->trans_lock);
4623 if (t == fs_info->running_transaction)
4624 fs_info->running_transaction = NULL;
4625 list_del_init(&t->list);
4626 spin_unlock(&fs_info->trans_lock);
4628 btrfs_put_transaction(t);
4629 trace_btrfs_transaction_commit(fs_info->tree_root);
4630 spin_lock(&fs_info->trans_lock);
4632 spin_unlock(&fs_info->trans_lock);
4633 btrfs_destroy_all_ordered_extents(fs_info);
4634 btrfs_destroy_delayed_inodes(fs_info);
4635 btrfs_assert_delayed_root_empty(fs_info);
4636 btrfs_destroy_all_delalloc_inodes(fs_info);
4637 mutex_unlock(&fs_info->transaction_kthread_mutex);
4642 static const struct extent_io_ops btree_extent_io_ops = {
4643 /* mandatory callbacks */
4644 .submit_bio_hook = btree_submit_bio_hook,
4645 .readpage_end_io_hook = btree_readpage_end_io_hook,