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 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
259 struct btrfs_fs_info *fs_info = buf->fs_info;
260 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
261 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
265 shash->tfm = fs_info->csum_shash;
266 crypto_shash_init(shash);
267 kaddr = page_address(buf->pages[0]);
268 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
269 PAGE_SIZE - BTRFS_CSUM_SIZE);
271 for (i = 1; i < num_pages; i++) {
272 kaddr = page_address(buf->pages[i]);
273 crypto_shash_update(shash, kaddr, PAGE_SIZE);
275 memset(result, 0, BTRFS_CSUM_SIZE);
276 crypto_shash_final(shash, result);
280 * we can't consider a given block up to date unless the transid of the
281 * block matches the transid in the parent node's pointer. This is how we
282 * detect blocks that either didn't get written at all or got written
283 * in the wrong place.
285 static int verify_parent_transid(struct extent_io_tree *io_tree,
286 struct extent_buffer *eb, u64 parent_transid,
289 struct extent_state *cached_state = NULL;
291 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
293 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
300 btrfs_tree_read_lock(eb);
301 btrfs_set_lock_blocking_read(eb);
304 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
306 if (extent_buffer_uptodate(eb) &&
307 btrfs_header_generation(eb) == parent_transid) {
311 btrfs_err_rl(eb->fs_info,
312 "parent transid verify failed on %llu wanted %llu found %llu",
314 parent_transid, btrfs_header_generation(eb));
318 * Things reading via commit roots that don't have normal protection,
319 * like send, can have a really old block in cache that may point at a
320 * block that has been freed and re-allocated. So don't clear uptodate
321 * if we find an eb that is under IO (dirty/writeback) because we could
322 * end up reading in the stale data and then writing it back out and
323 * making everybody very sad.
325 if (!extent_buffer_under_io(eb))
326 clear_extent_buffer_uptodate(eb);
328 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
331 btrfs_tree_read_unlock_blocking(eb);
335 static bool btrfs_supported_super_csum(u16 csum_type)
338 case BTRFS_CSUM_TYPE_CRC32:
339 case BTRFS_CSUM_TYPE_XXHASH:
340 case BTRFS_CSUM_TYPE_SHA256:
341 case BTRFS_CSUM_TYPE_BLAKE2:
349 * Return 0 if the superblock checksum type matches the checksum value of that
350 * algorithm. Pass the raw disk superblock data.
352 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
355 struct btrfs_super_block *disk_sb =
356 (struct btrfs_super_block *)raw_disk_sb;
357 char result[BTRFS_CSUM_SIZE];
358 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
360 shash->tfm = fs_info->csum_shash;
363 * The super_block structure does not span the whole
364 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
365 * filled with zeros and is included in the checksum.
367 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
368 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
370 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
376 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
377 struct btrfs_key *first_key, u64 parent_transid)
379 struct btrfs_fs_info *fs_info = eb->fs_info;
381 struct btrfs_key found_key;
384 found_level = btrfs_header_level(eb);
385 if (found_level != level) {
386 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
387 KERN_ERR "BTRFS: tree level check failed\n");
389 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
390 eb->start, level, found_level);
398 * For live tree block (new tree blocks in current transaction),
399 * we need proper lock context to avoid race, which is impossible here.
400 * So we only checks tree blocks which is read from disk, whose
401 * generation <= fs_info->last_trans_committed.
403 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
406 /* We have @first_key, so this @eb must have at least one item */
407 if (btrfs_header_nritems(eb) == 0) {
409 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
411 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
416 btrfs_node_key_to_cpu(eb, &found_key, 0);
418 btrfs_item_key_to_cpu(eb, &found_key, 0);
419 ret = btrfs_comp_cpu_keys(first_key, &found_key);
422 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
423 KERN_ERR "BTRFS: tree first key check failed\n");
425 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
426 eb->start, parent_transid, first_key->objectid,
427 first_key->type, first_key->offset,
428 found_key.objectid, found_key.type,
435 * helper to read a given tree block, doing retries as required when
436 * the checksums don't match and we have alternate mirrors to try.
438 * @parent_transid: expected transid, skip check if 0
439 * @level: expected level, mandatory check
440 * @first_key: expected key of first slot, skip check if NULL
442 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
443 u64 parent_transid, int level,
444 struct btrfs_key *first_key)
446 struct btrfs_fs_info *fs_info = eb->fs_info;
447 struct extent_io_tree *io_tree;
452 int failed_mirror = 0;
454 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
456 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
457 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
459 if (verify_parent_transid(io_tree, eb,
462 else if (btrfs_verify_level_key(eb, level,
463 first_key, parent_transid))
469 num_copies = btrfs_num_copies(fs_info,
474 if (!failed_mirror) {
476 failed_mirror = eb->read_mirror;
480 if (mirror_num == failed_mirror)
483 if (mirror_num > num_copies)
487 if (failed && !ret && failed_mirror)
488 btrfs_repair_eb_io_failure(eb, failed_mirror);
494 * checksum a dirty tree block before IO. This has extra checks to make sure
495 * we only fill in the checksum field in the first page of a multi-page block
498 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
500 u64 start = page_offset(page);
502 u8 result[BTRFS_CSUM_SIZE];
503 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
504 struct extent_buffer *eb;
507 eb = (struct extent_buffer *)page->private;
508 if (page != eb->pages[0])
511 found_start = btrfs_header_bytenr(eb);
513 * Please do not consolidate these warnings into a single if.
514 * It is useful to know what went wrong.
516 if (WARN_ON(found_start != start))
518 if (WARN_ON(!PageUptodate(page)))
521 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
522 offsetof(struct btrfs_header, fsid),
523 BTRFS_FSID_SIZE) == 0);
525 csum_tree_block(eb, result);
527 if (btrfs_header_level(eb))
528 ret = btrfs_check_node(eb);
530 ret = btrfs_check_leaf_full(eb);
533 btrfs_print_tree(eb, 0);
535 "block=%llu write time tree block corruption detected",
537 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
540 write_extent_buffer(eb, result, 0, csum_size);
545 static int check_tree_block_fsid(struct extent_buffer *eb)
547 struct btrfs_fs_info *fs_info = eb->fs_info;
548 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
549 u8 fsid[BTRFS_FSID_SIZE];
552 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
558 * Checking the incompat flag is only valid for the current
559 * fs. For seed devices it's forbidden to have their uuid
560 * changed so reading ->fsid in this case is fine
562 if (fs_devices == fs_info->fs_devices &&
563 btrfs_fs_incompat(fs_info, METADATA_UUID))
564 metadata_uuid = fs_devices->metadata_uuid;
566 metadata_uuid = fs_devices->fsid;
568 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
572 fs_devices = fs_devices->seed;
577 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
578 u64 phy_offset, struct page *page,
579 u64 start, u64 end, int mirror)
583 struct extent_buffer *eb;
584 struct btrfs_fs_info *fs_info;
587 u8 result[BTRFS_CSUM_SIZE];
593 eb = (struct extent_buffer *)page->private;
594 fs_info = eb->fs_info;
595 csum_size = btrfs_super_csum_size(fs_info->super_copy);
597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
600 atomic_inc(&eb->refs);
602 reads_done = atomic_dec_and_test(&eb->io_pages);
606 eb->read_mirror = mirror;
607 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
612 found_start = btrfs_header_bytenr(eb);
613 if (found_start != eb->start) {
614 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
615 eb->start, found_start);
619 if (check_tree_block_fsid(eb)) {
620 btrfs_err_rl(fs_info, "bad fsid on block %llu",
625 found_level = btrfs_header_level(eb);
626 if (found_level >= BTRFS_MAX_LEVEL) {
627 btrfs_err(fs_info, "bad tree block level %d on %llu",
628 (int)btrfs_header_level(eb), eb->start);
633 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
636 csum_tree_block(eb, result);
638 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
639 u8 val[BTRFS_CSUM_SIZE] = { 0 };
641 read_extent_buffer(eb, &val, 0, csum_size);
642 btrfs_warn_rl(fs_info,
643 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
644 fs_info->sb->s_id, eb->start,
645 CSUM_FMT_VALUE(csum_size, val),
646 CSUM_FMT_VALUE(csum_size, result),
647 btrfs_header_level(eb));
653 * If this is a leaf block and it is corrupt, set the corrupt bit so
654 * that we don't try and read the other copies of this block, just
657 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
658 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
662 if (found_level > 0 && btrfs_check_node(eb))
666 set_extent_buffer_uptodate(eb);
669 "block=%llu read time tree block corruption detected",
673 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
674 btree_readahead_hook(eb, ret);
678 * our io error hook is going to dec the io pages
679 * again, we have to make sure it has something
682 atomic_inc(&eb->io_pages);
683 clear_extent_buffer_uptodate(eb);
685 free_extent_buffer(eb);
690 static void end_workqueue_bio(struct bio *bio)
692 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
693 struct btrfs_fs_info *fs_info;
694 struct btrfs_workqueue *wq;
696 fs_info = end_io_wq->info;
697 end_io_wq->status = bio->bi_status;
699 if (bio_op(bio) == REQ_OP_WRITE) {
700 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
701 wq = fs_info->endio_meta_write_workers;
702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
703 wq = fs_info->endio_freespace_worker;
704 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
705 wq = fs_info->endio_raid56_workers;
707 wq = fs_info->endio_write_workers;
709 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
710 wq = fs_info->endio_raid56_workers;
711 else if (end_io_wq->metadata)
712 wq = fs_info->endio_meta_workers;
714 wq = fs_info->endio_workers;
717 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
718 btrfs_queue_work(wq, &end_io_wq->work);
721 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
722 enum btrfs_wq_endio_type metadata)
724 struct btrfs_end_io_wq *end_io_wq;
726 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
728 return BLK_STS_RESOURCE;
730 end_io_wq->private = bio->bi_private;
731 end_io_wq->end_io = bio->bi_end_io;
732 end_io_wq->info = info;
733 end_io_wq->status = 0;
734 end_io_wq->bio = bio;
735 end_io_wq->metadata = metadata;
737 bio->bi_private = end_io_wq;
738 bio->bi_end_io = end_workqueue_bio;
742 static void run_one_async_start(struct btrfs_work *work)
744 struct async_submit_bio *async;
747 async = container_of(work, struct async_submit_bio, work);
748 ret = async->submit_bio_start(async->private_data, async->bio,
755 * In order to insert checksums into the metadata in large chunks, we wait
756 * until bio submission time. All the pages in the bio are checksummed and
757 * sums are attached onto the ordered extent record.
759 * At IO completion time the csums attached on the ordered extent record are
760 * inserted into the tree.
762 static void run_one_async_done(struct btrfs_work *work)
764 struct async_submit_bio *async;
768 async = container_of(work, struct async_submit_bio, work);
769 inode = async->private_data;
771 /* If an error occurred we just want to clean up the bio and move on */
773 async->bio->bi_status = async->status;
774 bio_endio(async->bio);
779 * All of the bios that pass through here are from async helpers.
780 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
781 * This changes nothing when cgroups aren't in use.
783 async->bio->bi_opf |= REQ_CGROUP_PUNT;
784 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
786 async->bio->bi_status = ret;
787 bio_endio(async->bio);
791 static void run_one_async_free(struct btrfs_work *work)
793 struct async_submit_bio *async;
795 async = container_of(work, struct async_submit_bio, work);
799 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
800 int mirror_num, unsigned long bio_flags,
801 u64 bio_offset, void *private_data,
802 extent_submit_bio_start_t *submit_bio_start)
804 struct async_submit_bio *async;
806 async = kmalloc(sizeof(*async), GFP_NOFS);
808 return BLK_STS_RESOURCE;
810 async->private_data = private_data;
812 async->mirror_num = mirror_num;
813 async->submit_bio_start = submit_bio_start;
815 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
818 async->bio_offset = bio_offset;
822 if (op_is_sync(bio->bi_opf))
823 btrfs_set_work_high_priority(&async->work);
825 btrfs_queue_work(fs_info->workers, &async->work);
829 static blk_status_t btree_csum_one_bio(struct bio *bio)
831 struct bio_vec *bvec;
832 struct btrfs_root *root;
834 struct bvec_iter_all iter_all;
836 ASSERT(!bio_flagged(bio, BIO_CLONED));
837 bio_for_each_segment_all(bvec, bio, iter_all) {
838 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
839 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
844 return errno_to_blk_status(ret);
847 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
851 * when we're called for a write, we're already in the async
852 * submission context. Just jump into btrfs_map_bio
854 return btree_csum_one_bio(bio);
857 static int check_async_write(struct btrfs_fs_info *fs_info,
858 struct btrfs_inode *bi)
860 if (atomic_read(&bi->sync_writers))
862 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
867 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
869 unsigned long bio_flags)
871 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
872 int async = check_async_write(fs_info, BTRFS_I(inode));
875 if (bio_op(bio) != REQ_OP_WRITE) {
877 * called for a read, do the setup so that checksum validation
878 * can happen in the async kernel threads
880 ret = btrfs_bio_wq_end_io(fs_info, bio,
881 BTRFS_WQ_ENDIO_METADATA);
884 ret = btrfs_map_bio(fs_info, bio, mirror_num);
886 ret = btree_csum_one_bio(bio);
889 ret = btrfs_map_bio(fs_info, bio, mirror_num);
892 * kthread helpers are used to submit writes so that
893 * checksumming can happen in parallel across all CPUs
895 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
896 0, inode, btree_submit_bio_start);
904 bio->bi_status = ret;
909 #ifdef CONFIG_MIGRATION
910 static int btree_migratepage(struct address_space *mapping,
911 struct page *newpage, struct page *page,
912 enum migrate_mode mode)
915 * we can't safely write a btree page from here,
916 * we haven't done the locking hook
921 * Buffers may be managed in a filesystem specific way.
922 * We must have no buffers or drop them.
924 if (page_has_private(page) &&
925 !try_to_release_page(page, GFP_KERNEL))
927 return migrate_page(mapping, newpage, page, mode);
932 static int btree_writepages(struct address_space *mapping,
933 struct writeback_control *wbc)
935 struct btrfs_fs_info *fs_info;
938 if (wbc->sync_mode == WB_SYNC_NONE) {
940 if (wbc->for_kupdate)
943 fs_info = BTRFS_I(mapping->host)->root->fs_info;
944 /* this is a bit racy, but that's ok */
945 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
946 BTRFS_DIRTY_METADATA_THRESH,
947 fs_info->dirty_metadata_batch);
951 return btree_write_cache_pages(mapping, wbc);
954 static int btree_readpage(struct file *file, struct page *page)
956 return extent_read_full_page(page, btree_get_extent, 0);
959 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
961 if (PageWriteback(page) || PageDirty(page))
964 return try_release_extent_buffer(page);
967 static void btree_invalidatepage(struct page *page, unsigned int offset,
970 struct extent_io_tree *tree;
971 tree = &BTRFS_I(page->mapping->host)->io_tree;
972 extent_invalidatepage(tree, page, offset);
973 btree_releasepage(page, GFP_NOFS);
974 if (PagePrivate(page)) {
975 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
976 "page private not zero on page %llu",
977 (unsigned long long)page_offset(page));
978 detach_page_private(page);
982 static int btree_set_page_dirty(struct page *page)
985 struct extent_buffer *eb;
987 BUG_ON(!PagePrivate(page));
988 eb = (struct extent_buffer *)page->private;
990 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
991 BUG_ON(!atomic_read(&eb->refs));
992 btrfs_assert_tree_locked(eb);
994 return __set_page_dirty_nobuffers(page);
997 static const struct address_space_operations btree_aops = {
998 .readpage = btree_readpage,
999 .writepages = btree_writepages,
1000 .releasepage = btree_releasepage,
1001 .invalidatepage = btree_invalidatepage,
1002 #ifdef CONFIG_MIGRATION
1003 .migratepage = btree_migratepage,
1005 .set_page_dirty = btree_set_page_dirty,
1008 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1010 struct extent_buffer *buf = NULL;
1013 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1017 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1019 free_extent_buffer_stale(buf);
1021 free_extent_buffer(buf);
1024 struct extent_buffer *btrfs_find_create_tree_block(
1025 struct btrfs_fs_info *fs_info,
1028 if (btrfs_is_testing(fs_info))
1029 return alloc_test_extent_buffer(fs_info, bytenr);
1030 return alloc_extent_buffer(fs_info, bytenr);
1034 * Read tree block at logical address @bytenr and do variant basic but critical
1037 * @parent_transid: expected transid of this tree block, skip check if 0
1038 * @level: expected level, mandatory check
1039 * @first_key: expected key in slot 0, skip check if NULL
1041 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1042 u64 parent_transid, int level,
1043 struct btrfs_key *first_key)
1045 struct extent_buffer *buf = NULL;
1048 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1052 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1055 free_extent_buffer_stale(buf);
1056 return ERR_PTR(ret);
1062 void btrfs_clean_tree_block(struct extent_buffer *buf)
1064 struct btrfs_fs_info *fs_info = buf->fs_info;
1065 if (btrfs_header_generation(buf) ==
1066 fs_info->running_transaction->transid) {
1067 btrfs_assert_tree_locked(buf);
1069 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1070 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1072 fs_info->dirty_metadata_batch);
1073 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1074 btrfs_set_lock_blocking_write(buf);
1075 clear_extent_buffer_dirty(buf);
1080 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1083 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1084 root->fs_info = fs_info;
1086 root->commit_root = NULL;
1088 root->orphan_cleanup_state = 0;
1090 root->last_trans = 0;
1091 root->highest_objectid = 0;
1092 root->nr_delalloc_inodes = 0;
1093 root->nr_ordered_extents = 0;
1094 root->inode_tree = RB_ROOT;
1095 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1096 root->block_rsv = NULL;
1098 INIT_LIST_HEAD(&root->dirty_list);
1099 INIT_LIST_HEAD(&root->root_list);
1100 INIT_LIST_HEAD(&root->delalloc_inodes);
1101 INIT_LIST_HEAD(&root->delalloc_root);
1102 INIT_LIST_HEAD(&root->ordered_extents);
1103 INIT_LIST_HEAD(&root->ordered_root);
1104 INIT_LIST_HEAD(&root->reloc_dirty_list);
1105 INIT_LIST_HEAD(&root->logged_list[0]);
1106 INIT_LIST_HEAD(&root->logged_list[1]);
1107 spin_lock_init(&root->inode_lock);
1108 spin_lock_init(&root->delalloc_lock);
1109 spin_lock_init(&root->ordered_extent_lock);
1110 spin_lock_init(&root->accounting_lock);
1111 spin_lock_init(&root->log_extents_lock[0]);
1112 spin_lock_init(&root->log_extents_lock[1]);
1113 spin_lock_init(&root->qgroup_meta_rsv_lock);
1114 mutex_init(&root->objectid_mutex);
1115 mutex_init(&root->log_mutex);
1116 mutex_init(&root->ordered_extent_mutex);
1117 mutex_init(&root->delalloc_mutex);
1118 init_waitqueue_head(&root->qgroup_flush_wait);
1119 init_waitqueue_head(&root->log_writer_wait);
1120 init_waitqueue_head(&root->log_commit_wait[0]);
1121 init_waitqueue_head(&root->log_commit_wait[1]);
1122 INIT_LIST_HEAD(&root->log_ctxs[0]);
1123 INIT_LIST_HEAD(&root->log_ctxs[1]);
1124 atomic_set(&root->log_commit[0], 0);
1125 atomic_set(&root->log_commit[1], 0);
1126 atomic_set(&root->log_writers, 0);
1127 atomic_set(&root->log_batch, 0);
1128 refcount_set(&root->refs, 1);
1129 atomic_set(&root->snapshot_force_cow, 0);
1130 atomic_set(&root->nr_swapfiles, 0);
1131 root->log_transid = 0;
1132 root->log_transid_committed = -1;
1133 root->last_log_commit = 0;
1135 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1136 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1137 extent_io_tree_init(fs_info, &root->log_csum_range,
1138 IO_TREE_LOG_CSUM_RANGE, NULL);
1141 memset(&root->root_key, 0, sizeof(root->root_key));
1142 memset(&root->root_item, 0, sizeof(root->root_item));
1143 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1144 root->root_key.objectid = objectid;
1147 spin_lock_init(&root->root_item_lock);
1148 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1149 #ifdef CONFIG_BTRFS_DEBUG
1150 INIT_LIST_HEAD(&root->leak_list);
1151 spin_lock(&fs_info->fs_roots_radix_lock);
1152 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1153 spin_unlock(&fs_info->fs_roots_radix_lock);
1157 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1158 u64 objectid, gfp_t flags)
1160 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1162 __setup_root(root, fs_info, objectid);
1166 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1167 /* Should only be used by the testing infrastructure */
1168 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1170 struct btrfs_root *root;
1173 return ERR_PTR(-EINVAL);
1175 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1177 return ERR_PTR(-ENOMEM);
1179 /* We don't use the stripesize in selftest, set it as sectorsize */
1180 root->alloc_bytenr = 0;
1186 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1189 struct btrfs_fs_info *fs_info = trans->fs_info;
1190 struct extent_buffer *leaf;
1191 struct btrfs_root *tree_root = fs_info->tree_root;
1192 struct btrfs_root *root;
1193 struct btrfs_key key;
1194 unsigned int nofs_flag;
1198 * We're holding a transaction handle, so use a NOFS memory allocation
1199 * context to avoid deadlock if reclaim happens.
1201 nofs_flag = memalloc_nofs_save();
1202 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1203 memalloc_nofs_restore(nofs_flag);
1205 return ERR_PTR(-ENOMEM);
1207 root->root_key.objectid = objectid;
1208 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1209 root->root_key.offset = 0;
1211 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1213 ret = PTR_ERR(leaf);
1219 btrfs_mark_buffer_dirty(leaf);
1221 root->commit_root = btrfs_root_node(root);
1222 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1224 root->root_item.flags = 0;
1225 root->root_item.byte_limit = 0;
1226 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1227 btrfs_set_root_generation(&root->root_item, trans->transid);
1228 btrfs_set_root_level(&root->root_item, 0);
1229 btrfs_set_root_refs(&root->root_item, 1);
1230 btrfs_set_root_used(&root->root_item, leaf->len);
1231 btrfs_set_root_last_snapshot(&root->root_item, 0);
1232 btrfs_set_root_dirid(&root->root_item, 0);
1233 if (is_fstree(objectid))
1234 generate_random_guid(root->root_item.uuid);
1236 export_guid(root->root_item.uuid, &guid_null);
1237 root->root_item.drop_level = 0;
1239 key.objectid = objectid;
1240 key.type = BTRFS_ROOT_ITEM_KEY;
1242 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1246 btrfs_tree_unlock(leaf);
1252 btrfs_tree_unlock(leaf);
1253 btrfs_put_root(root);
1255 return ERR_PTR(ret);
1258 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1259 struct btrfs_fs_info *fs_info)
1261 struct btrfs_root *root;
1262 struct extent_buffer *leaf;
1264 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1266 return ERR_PTR(-ENOMEM);
1268 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1269 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1270 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1273 * DON'T set SHAREABLE bit for log trees.
1275 * Log trees are not exposed to user space thus can't be snapshotted,
1276 * and they go away before a real commit is actually done.
1278 * They do store pointers to file data extents, and those reference
1279 * counts still get updated (along with back refs to the log tree).
1282 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1285 btrfs_put_root(root);
1286 return ERR_CAST(leaf);
1291 btrfs_mark_buffer_dirty(root->node);
1292 btrfs_tree_unlock(root->node);
1296 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1297 struct btrfs_fs_info *fs_info)
1299 struct btrfs_root *log_root;
1301 log_root = alloc_log_tree(trans, fs_info);
1302 if (IS_ERR(log_root))
1303 return PTR_ERR(log_root);
1304 WARN_ON(fs_info->log_root_tree);
1305 fs_info->log_root_tree = log_root;
1309 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1310 struct btrfs_root *root)
1312 struct btrfs_fs_info *fs_info = root->fs_info;
1313 struct btrfs_root *log_root;
1314 struct btrfs_inode_item *inode_item;
1316 log_root = alloc_log_tree(trans, fs_info);
1317 if (IS_ERR(log_root))
1318 return PTR_ERR(log_root);
1320 log_root->last_trans = trans->transid;
1321 log_root->root_key.offset = root->root_key.objectid;
1323 inode_item = &log_root->root_item.inode;
1324 btrfs_set_stack_inode_generation(inode_item, 1);
1325 btrfs_set_stack_inode_size(inode_item, 3);
1326 btrfs_set_stack_inode_nlink(inode_item, 1);
1327 btrfs_set_stack_inode_nbytes(inode_item,
1329 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1331 btrfs_set_root_node(&log_root->root_item, log_root->node);
1333 WARN_ON(root->log_root);
1334 root->log_root = log_root;
1335 root->log_transid = 0;
1336 root->log_transid_committed = -1;
1337 root->last_log_commit = 0;
1341 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1342 struct btrfs_key *key)
1344 struct btrfs_root *root;
1345 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1346 struct btrfs_path *path;
1351 path = btrfs_alloc_path();
1353 return ERR_PTR(-ENOMEM);
1355 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1361 ret = btrfs_find_root(tree_root, key, path,
1362 &root->root_item, &root->root_key);
1369 generation = btrfs_root_generation(&root->root_item);
1370 level = btrfs_root_level(&root->root_item);
1371 root->node = read_tree_block(fs_info,
1372 btrfs_root_bytenr(&root->root_item),
1373 generation, level, NULL);
1374 if (IS_ERR(root->node)) {
1375 ret = PTR_ERR(root->node);
1378 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1382 root->commit_root = btrfs_root_node(root);
1384 btrfs_free_path(path);
1388 btrfs_put_root(root);
1390 root = ERR_PTR(ret);
1395 * Initialize subvolume root in-memory structure
1397 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1399 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1402 unsigned int nofs_flag;
1404 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1405 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1407 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1413 * We might be called under a transaction (e.g. indirect backref
1414 * resolution) which could deadlock if it triggers memory reclaim
1416 nofs_flag = memalloc_nofs_save();
1417 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1418 memalloc_nofs_restore(nofs_flag);
1422 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1423 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1424 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1425 btrfs_check_and_init_root_item(&root->root_item);
1428 btrfs_init_free_ino_ctl(root);
1429 spin_lock_init(&root->ino_cache_lock);
1430 init_waitqueue_head(&root->ino_cache_wait);
1433 * Don't assign anonymous block device to roots that are not exposed to
1434 * userspace, the id pool is limited to 1M
1436 if (is_fstree(root->root_key.objectid) &&
1437 btrfs_root_refs(&root->root_item) > 0) {
1439 ret = get_anon_bdev(&root->anon_dev);
1443 root->anon_dev = anon_dev;
1447 mutex_lock(&root->objectid_mutex);
1448 ret = btrfs_find_highest_objectid(root,
1449 &root->highest_objectid);
1451 mutex_unlock(&root->objectid_mutex);
1455 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1457 mutex_unlock(&root->objectid_mutex);
1461 /* The caller is responsible to call btrfs_free_fs_root */
1465 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1468 struct btrfs_root *root;
1470 spin_lock(&fs_info->fs_roots_radix_lock);
1471 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1472 (unsigned long)root_id);
1474 root = btrfs_grab_root(root);
1475 spin_unlock(&fs_info->fs_roots_radix_lock);
1479 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1480 struct btrfs_root *root)
1484 ret = radix_tree_preload(GFP_NOFS);
1488 spin_lock(&fs_info->fs_roots_radix_lock);
1489 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1490 (unsigned long)root->root_key.objectid,
1493 btrfs_grab_root(root);
1494 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1496 spin_unlock(&fs_info->fs_roots_radix_lock);
1497 radix_tree_preload_end();
1502 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1504 #ifdef CONFIG_BTRFS_DEBUG
1505 struct btrfs_root *root;
1507 while (!list_empty(&fs_info->allocated_roots)) {
1508 root = list_first_entry(&fs_info->allocated_roots,
1509 struct btrfs_root, leak_list);
1510 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1511 root->root_key.objectid, root->root_key.offset,
1512 refcount_read(&root->refs));
1513 while (refcount_read(&root->refs) > 1)
1514 btrfs_put_root(root);
1515 btrfs_put_root(root);
1520 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1522 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1523 percpu_counter_destroy(&fs_info->delalloc_bytes);
1524 percpu_counter_destroy(&fs_info->dio_bytes);
1525 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1526 btrfs_free_csum_hash(fs_info);
1527 btrfs_free_stripe_hash_table(fs_info);
1528 btrfs_free_ref_cache(fs_info);
1529 kfree(fs_info->balance_ctl);
1530 kfree(fs_info->delayed_root);
1531 btrfs_put_root(fs_info->extent_root);
1532 btrfs_put_root(fs_info->tree_root);
1533 btrfs_put_root(fs_info->chunk_root);
1534 btrfs_put_root(fs_info->dev_root);
1535 btrfs_put_root(fs_info->csum_root);
1536 btrfs_put_root(fs_info->quota_root);
1537 btrfs_put_root(fs_info->uuid_root);
1538 btrfs_put_root(fs_info->free_space_root);
1539 btrfs_put_root(fs_info->fs_root);
1540 btrfs_put_root(fs_info->data_reloc_root);
1541 btrfs_check_leaked_roots(fs_info);
1542 btrfs_extent_buffer_leak_debug_check(fs_info);
1543 kfree(fs_info->super_copy);
1544 kfree(fs_info->super_for_commit);
1550 * Get an in-memory reference of a root structure.
1552 * For essential trees like root/extent tree, we grab it from fs_info directly.
1553 * For subvolume trees, we check the cached filesystem roots first. If not
1554 * found, then read it from disk and add it to cached fs roots.
1556 * Caller should release the root by calling btrfs_put_root() after the usage.
1558 * NOTE: Reloc and log trees can't be read by this function as they share the
1559 * same root objectid.
1561 * @objectid: root id
1562 * @anon_dev: preallocated anonymous block device number for new roots,
1563 * pass 0 for new allocation.
1564 * @check_ref: whether to check root item references, If true, return -ENOENT
1567 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1568 u64 objectid, dev_t anon_dev,
1571 struct btrfs_root *root;
1572 struct btrfs_path *path;
1573 struct btrfs_key key;
1576 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1577 return btrfs_grab_root(fs_info->tree_root);
1578 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1579 return btrfs_grab_root(fs_info->extent_root);
1580 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1581 return btrfs_grab_root(fs_info->chunk_root);
1582 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1583 return btrfs_grab_root(fs_info->dev_root);
1584 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1585 return btrfs_grab_root(fs_info->csum_root);
1586 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1587 return btrfs_grab_root(fs_info->quota_root) ?
1588 fs_info->quota_root : ERR_PTR(-ENOENT);
1589 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1590 return btrfs_grab_root(fs_info->uuid_root) ?
1591 fs_info->uuid_root : ERR_PTR(-ENOENT);
1592 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1593 return btrfs_grab_root(fs_info->free_space_root) ?
1594 fs_info->free_space_root : ERR_PTR(-ENOENT);
1596 root = btrfs_lookup_fs_root(fs_info, objectid);
1598 /* Shouldn't get preallocated anon_dev for cached roots */
1600 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1601 btrfs_put_root(root);
1602 return ERR_PTR(-ENOENT);
1607 key.objectid = objectid;
1608 key.type = BTRFS_ROOT_ITEM_KEY;
1609 key.offset = (u64)-1;
1610 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1614 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1619 ret = btrfs_init_fs_root(root, anon_dev);
1623 path = btrfs_alloc_path();
1628 key.objectid = BTRFS_ORPHAN_OBJECTID;
1629 key.type = BTRFS_ORPHAN_ITEM_KEY;
1630 key.offset = objectid;
1632 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1633 btrfs_free_path(path);
1637 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1639 ret = btrfs_insert_fs_root(fs_info, root);
1641 btrfs_put_root(root);
1648 btrfs_put_root(root);
1649 return ERR_PTR(ret);
1653 * Get in-memory reference of a root structure
1655 * @objectid: tree objectid
1656 * @check_ref: if set, verify that the tree exists and the item has at least
1659 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1660 u64 objectid, bool check_ref)
1662 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1666 * Get in-memory reference of a root structure, created as new, optionally pass
1667 * the anonymous block device id
1669 * @objectid: tree objectid
1670 * @anon_dev: if zero, allocate a new anonymous block device or use the
1673 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1674 u64 objectid, dev_t anon_dev)
1676 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1680 * called by the kthread helper functions to finally call the bio end_io
1681 * functions. This is where read checksum verification actually happens
1683 static void end_workqueue_fn(struct btrfs_work *work)
1686 struct btrfs_end_io_wq *end_io_wq;
1688 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1689 bio = end_io_wq->bio;
1691 bio->bi_status = end_io_wq->status;
1692 bio->bi_private = end_io_wq->private;
1693 bio->bi_end_io = end_io_wq->end_io;
1695 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1698 static int cleaner_kthread(void *arg)
1700 struct btrfs_root *root = arg;
1701 struct btrfs_fs_info *fs_info = root->fs_info;
1707 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1709 /* Make the cleaner go to sleep early. */
1710 if (btrfs_need_cleaner_sleep(fs_info))
1714 * Do not do anything if we might cause open_ctree() to block
1715 * before we have finished mounting the filesystem.
1717 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1720 if (!mutex_trylock(&fs_info->cleaner_mutex))
1724 * Avoid the problem that we change the status of the fs
1725 * during the above check and trylock.
1727 if (btrfs_need_cleaner_sleep(fs_info)) {
1728 mutex_unlock(&fs_info->cleaner_mutex);
1732 btrfs_run_delayed_iputs(fs_info);
1734 again = btrfs_clean_one_deleted_snapshot(root);
1735 mutex_unlock(&fs_info->cleaner_mutex);
1738 * The defragger has dealt with the R/O remount and umount,
1739 * needn't do anything special here.
1741 btrfs_run_defrag_inodes(fs_info);
1744 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1745 * with relocation (btrfs_relocate_chunk) and relocation
1746 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1747 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1748 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1749 * unused block groups.
1751 btrfs_delete_unused_bgs(fs_info);
1753 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1754 if (kthread_should_park())
1756 if (kthread_should_stop())
1759 set_current_state(TASK_INTERRUPTIBLE);
1761 __set_current_state(TASK_RUNNING);
1766 static int transaction_kthread(void *arg)
1768 struct btrfs_root *root = arg;
1769 struct btrfs_fs_info *fs_info = root->fs_info;
1770 struct btrfs_trans_handle *trans;
1771 struct btrfs_transaction *cur;
1774 unsigned long delay;
1778 cannot_commit = false;
1779 delay = HZ * fs_info->commit_interval;
1780 mutex_lock(&fs_info->transaction_kthread_mutex);
1782 spin_lock(&fs_info->trans_lock);
1783 cur = fs_info->running_transaction;
1785 spin_unlock(&fs_info->trans_lock);
1789 now = ktime_get_seconds();
1790 if (cur->state < TRANS_STATE_COMMIT_START &&
1791 (now < cur->start_time ||
1792 now - cur->start_time < fs_info->commit_interval)) {
1793 spin_unlock(&fs_info->trans_lock);
1797 transid = cur->transid;
1798 spin_unlock(&fs_info->trans_lock);
1800 /* If the file system is aborted, this will always fail. */
1801 trans = btrfs_attach_transaction(root);
1802 if (IS_ERR(trans)) {
1803 if (PTR_ERR(trans) != -ENOENT)
1804 cannot_commit = true;
1807 if (transid == trans->transid) {
1808 btrfs_commit_transaction(trans);
1810 btrfs_end_transaction(trans);
1813 wake_up_process(fs_info->cleaner_kthread);
1814 mutex_unlock(&fs_info->transaction_kthread_mutex);
1816 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1817 &fs_info->fs_state)))
1818 btrfs_cleanup_transaction(fs_info);
1819 if (!kthread_should_stop() &&
1820 (!btrfs_transaction_blocked(fs_info) ||
1822 schedule_timeout_interruptible(delay);
1823 } while (!kthread_should_stop());
1828 * This will find the highest generation in the array of root backups. The
1829 * index of the highest array is returned, or -EINVAL if we can't find
1832 * We check to make sure the array is valid by comparing the
1833 * generation of the latest root in the array with the generation
1834 * in the super block. If they don't match we pitch it.
1836 static int find_newest_super_backup(struct btrfs_fs_info *info)
1838 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1840 struct btrfs_root_backup *root_backup;
1843 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1844 root_backup = info->super_copy->super_roots + i;
1845 cur = btrfs_backup_tree_root_gen(root_backup);
1846 if (cur == newest_gen)
1854 * copy all the root pointers into the super backup array.
1855 * this will bump the backup pointer by one when it is
1858 static void backup_super_roots(struct btrfs_fs_info *info)
1860 const int next_backup = info->backup_root_index;
1861 struct btrfs_root_backup *root_backup;
1863 root_backup = info->super_for_commit->super_roots + next_backup;
1866 * make sure all of our padding and empty slots get zero filled
1867 * regardless of which ones we use today
1869 memset(root_backup, 0, sizeof(*root_backup));
1871 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1873 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1874 btrfs_set_backup_tree_root_gen(root_backup,
1875 btrfs_header_generation(info->tree_root->node));
1877 btrfs_set_backup_tree_root_level(root_backup,
1878 btrfs_header_level(info->tree_root->node));
1880 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1881 btrfs_set_backup_chunk_root_gen(root_backup,
1882 btrfs_header_generation(info->chunk_root->node));
1883 btrfs_set_backup_chunk_root_level(root_backup,
1884 btrfs_header_level(info->chunk_root->node));
1886 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1887 btrfs_set_backup_extent_root_gen(root_backup,
1888 btrfs_header_generation(info->extent_root->node));
1889 btrfs_set_backup_extent_root_level(root_backup,
1890 btrfs_header_level(info->extent_root->node));
1893 * we might commit during log recovery, which happens before we set
1894 * the fs_root. Make sure it is valid before we fill it in.
1896 if (info->fs_root && info->fs_root->node) {
1897 btrfs_set_backup_fs_root(root_backup,
1898 info->fs_root->node->start);
1899 btrfs_set_backup_fs_root_gen(root_backup,
1900 btrfs_header_generation(info->fs_root->node));
1901 btrfs_set_backup_fs_root_level(root_backup,
1902 btrfs_header_level(info->fs_root->node));
1905 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1906 btrfs_set_backup_dev_root_gen(root_backup,
1907 btrfs_header_generation(info->dev_root->node));
1908 btrfs_set_backup_dev_root_level(root_backup,
1909 btrfs_header_level(info->dev_root->node));
1911 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1912 btrfs_set_backup_csum_root_gen(root_backup,
1913 btrfs_header_generation(info->csum_root->node));
1914 btrfs_set_backup_csum_root_level(root_backup,
1915 btrfs_header_level(info->csum_root->node));
1917 btrfs_set_backup_total_bytes(root_backup,
1918 btrfs_super_total_bytes(info->super_copy));
1919 btrfs_set_backup_bytes_used(root_backup,
1920 btrfs_super_bytes_used(info->super_copy));
1921 btrfs_set_backup_num_devices(root_backup,
1922 btrfs_super_num_devices(info->super_copy));
1925 * if we don't copy this out to the super_copy, it won't get remembered
1926 * for the next commit
1928 memcpy(&info->super_copy->super_roots,
1929 &info->super_for_commit->super_roots,
1930 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1934 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1935 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1937 * fs_info - filesystem whose backup roots need to be read
1938 * priority - priority of backup root required
1940 * Returns backup root index on success and -EINVAL otherwise.
1942 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1944 int backup_index = find_newest_super_backup(fs_info);
1945 struct btrfs_super_block *super = fs_info->super_copy;
1946 struct btrfs_root_backup *root_backup;
1948 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1950 return backup_index;
1952 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1953 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1958 root_backup = super->super_roots + backup_index;
1960 btrfs_set_super_generation(super,
1961 btrfs_backup_tree_root_gen(root_backup));
1962 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1963 btrfs_set_super_root_level(super,
1964 btrfs_backup_tree_root_level(root_backup));
1965 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1968 * Fixme: the total bytes and num_devices need to match or we should
1971 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1972 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1974 return backup_index;
1977 /* helper to cleanup workers */
1978 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1980 btrfs_destroy_workqueue(fs_info->fixup_workers);
1981 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1982 btrfs_destroy_workqueue(fs_info->workers);
1983 btrfs_destroy_workqueue(fs_info->endio_workers);
1984 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1985 btrfs_destroy_workqueue(fs_info->rmw_workers);
1986 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1987 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1988 btrfs_destroy_workqueue(fs_info->delayed_workers);
1989 btrfs_destroy_workqueue(fs_info->caching_workers);
1990 btrfs_destroy_workqueue(fs_info->readahead_workers);
1991 btrfs_destroy_workqueue(fs_info->flush_workers);
1992 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1993 if (fs_info->discard_ctl.discard_workers)
1994 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1996 * Now that all other work queues are destroyed, we can safely destroy
1997 * the queues used for metadata I/O, since tasks from those other work
1998 * queues can do metadata I/O operations.
2000 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2001 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2004 static void free_root_extent_buffers(struct btrfs_root *root)
2007 free_extent_buffer(root->node);
2008 free_extent_buffer(root->commit_root);
2010 root->commit_root = NULL;
2014 /* helper to cleanup tree roots */
2015 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2017 free_root_extent_buffers(info->tree_root);
2019 free_root_extent_buffers(info->dev_root);
2020 free_root_extent_buffers(info->extent_root);
2021 free_root_extent_buffers(info->csum_root);
2022 free_root_extent_buffers(info->quota_root);
2023 free_root_extent_buffers(info->uuid_root);
2024 free_root_extent_buffers(info->fs_root);
2025 free_root_extent_buffers(info->data_reloc_root);
2026 if (free_chunk_root)
2027 free_root_extent_buffers(info->chunk_root);
2028 free_root_extent_buffers(info->free_space_root);
2031 void btrfs_put_root(struct btrfs_root *root)
2036 if (refcount_dec_and_test(&root->refs)) {
2037 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2038 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2040 free_anon_bdev(root->anon_dev);
2041 btrfs_drew_lock_destroy(&root->snapshot_lock);
2042 free_root_extent_buffers(root);
2043 kfree(root->free_ino_ctl);
2044 kfree(root->free_ino_pinned);
2045 #ifdef CONFIG_BTRFS_DEBUG
2046 spin_lock(&root->fs_info->fs_roots_radix_lock);
2047 list_del_init(&root->leak_list);
2048 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2054 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2057 struct btrfs_root *gang[8];
2060 while (!list_empty(&fs_info->dead_roots)) {
2061 gang[0] = list_entry(fs_info->dead_roots.next,
2062 struct btrfs_root, root_list);
2063 list_del(&gang[0]->root_list);
2065 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2066 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2067 btrfs_put_root(gang[0]);
2071 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2076 for (i = 0; i < ret; i++)
2077 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2081 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2083 mutex_init(&fs_info->scrub_lock);
2084 atomic_set(&fs_info->scrubs_running, 0);
2085 atomic_set(&fs_info->scrub_pause_req, 0);
2086 atomic_set(&fs_info->scrubs_paused, 0);
2087 atomic_set(&fs_info->scrub_cancel_req, 0);
2088 init_waitqueue_head(&fs_info->scrub_pause_wait);
2089 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2092 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2094 spin_lock_init(&fs_info->balance_lock);
2095 mutex_init(&fs_info->balance_mutex);
2096 atomic_set(&fs_info->balance_pause_req, 0);
2097 atomic_set(&fs_info->balance_cancel_req, 0);
2098 fs_info->balance_ctl = NULL;
2099 init_waitqueue_head(&fs_info->balance_wait_q);
2102 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2104 struct inode *inode = fs_info->btree_inode;
2106 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2107 set_nlink(inode, 1);
2109 * we set the i_size on the btree inode to the max possible int.
2110 * the real end of the address space is determined by all of
2111 * the devices in the system
2113 inode->i_size = OFFSET_MAX;
2114 inode->i_mapping->a_ops = &btree_aops;
2116 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2117 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2118 IO_TREE_INODE_IO, inode);
2119 BTRFS_I(inode)->io_tree.track_uptodate = false;
2120 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2122 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2124 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2125 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2126 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2127 btrfs_insert_inode_hash(inode);
2130 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2132 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2133 init_rwsem(&fs_info->dev_replace.rwsem);
2134 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2137 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2139 spin_lock_init(&fs_info->qgroup_lock);
2140 mutex_init(&fs_info->qgroup_ioctl_lock);
2141 fs_info->qgroup_tree = RB_ROOT;
2142 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2143 fs_info->qgroup_seq = 1;
2144 fs_info->qgroup_ulist = NULL;
2145 fs_info->qgroup_rescan_running = false;
2146 mutex_init(&fs_info->qgroup_rescan_lock);
2149 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2150 struct btrfs_fs_devices *fs_devices)
2152 u32 max_active = fs_info->thread_pool_size;
2153 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2156 btrfs_alloc_workqueue(fs_info, "worker",
2157 flags | WQ_HIGHPRI, max_active, 16);
2159 fs_info->delalloc_workers =
2160 btrfs_alloc_workqueue(fs_info, "delalloc",
2161 flags, max_active, 2);
2163 fs_info->flush_workers =
2164 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2165 flags, max_active, 0);
2167 fs_info->caching_workers =
2168 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2170 fs_info->fixup_workers =
2171 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2174 * endios are largely parallel and should have a very
2177 fs_info->endio_workers =
2178 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2179 fs_info->endio_meta_workers =
2180 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2182 fs_info->endio_meta_write_workers =
2183 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2185 fs_info->endio_raid56_workers =
2186 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2188 fs_info->rmw_workers =
2189 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2190 fs_info->endio_write_workers =
2191 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2193 fs_info->endio_freespace_worker =
2194 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2196 fs_info->delayed_workers =
2197 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2199 fs_info->readahead_workers =
2200 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2202 fs_info->qgroup_rescan_workers =
2203 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2204 fs_info->discard_ctl.discard_workers =
2205 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2207 if (!(fs_info->workers && fs_info->delalloc_workers &&
2208 fs_info->flush_workers &&
2209 fs_info->endio_workers && fs_info->endio_meta_workers &&
2210 fs_info->endio_meta_write_workers &&
2211 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2212 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2213 fs_info->caching_workers && fs_info->readahead_workers &&
2214 fs_info->fixup_workers && fs_info->delayed_workers &&
2215 fs_info->qgroup_rescan_workers &&
2216 fs_info->discard_ctl.discard_workers)) {
2223 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2225 struct crypto_shash *csum_shash;
2226 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2228 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2230 if (IS_ERR(csum_shash)) {
2231 btrfs_err(fs_info, "error allocating %s hash for checksum",
2233 return PTR_ERR(csum_shash);
2236 fs_info->csum_shash = csum_shash;
2241 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2242 struct btrfs_fs_devices *fs_devices)
2245 struct btrfs_root *log_tree_root;
2246 struct btrfs_super_block *disk_super = fs_info->super_copy;
2247 u64 bytenr = btrfs_super_log_root(disk_super);
2248 int level = btrfs_super_log_root_level(disk_super);
2250 if (fs_devices->rw_devices == 0) {
2251 btrfs_warn(fs_info, "log replay required on RO media");
2255 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2260 log_tree_root->node = read_tree_block(fs_info, bytenr,
2261 fs_info->generation + 1,
2263 if (IS_ERR(log_tree_root->node)) {
2264 btrfs_warn(fs_info, "failed to read log tree");
2265 ret = PTR_ERR(log_tree_root->node);
2266 log_tree_root->node = NULL;
2267 btrfs_put_root(log_tree_root);
2269 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2270 btrfs_err(fs_info, "failed to read log tree");
2271 btrfs_put_root(log_tree_root);
2274 /* returns with log_tree_root freed on success */
2275 ret = btrfs_recover_log_trees(log_tree_root);
2277 btrfs_handle_fs_error(fs_info, ret,
2278 "Failed to recover log tree");
2279 btrfs_put_root(log_tree_root);
2283 if (sb_rdonly(fs_info->sb)) {
2284 ret = btrfs_commit_super(fs_info);
2292 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2294 struct btrfs_root *tree_root = fs_info->tree_root;
2295 struct btrfs_root *root;
2296 struct btrfs_key location;
2299 BUG_ON(!fs_info->tree_root);
2301 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2302 location.type = BTRFS_ROOT_ITEM_KEY;
2303 location.offset = 0;
2305 root = btrfs_read_tree_root(tree_root, &location);
2307 ret = PTR_ERR(root);
2310 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2311 fs_info->extent_root = root;
2313 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2314 root = btrfs_read_tree_root(tree_root, &location);
2316 ret = PTR_ERR(root);
2319 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2320 fs_info->dev_root = root;
2321 btrfs_init_devices_late(fs_info);
2323 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2324 root = btrfs_read_tree_root(tree_root, &location);
2326 ret = PTR_ERR(root);
2329 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2330 fs_info->csum_root = root;
2333 * This tree can share blocks with some other fs tree during relocation
2334 * and we need a proper setup by btrfs_get_fs_root
2336 root = btrfs_get_fs_root(tree_root->fs_info,
2337 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2339 ret = PTR_ERR(root);
2342 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2343 fs_info->data_reloc_root = root;
2345 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2346 root = btrfs_read_tree_root(tree_root, &location);
2347 if (!IS_ERR(root)) {
2348 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2349 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2350 fs_info->quota_root = root;
2353 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2354 root = btrfs_read_tree_root(tree_root, &location);
2356 ret = PTR_ERR(root);
2360 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2361 fs_info->uuid_root = root;
2364 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2365 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2366 root = btrfs_read_tree_root(tree_root, &location);
2368 ret = PTR_ERR(root);
2371 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2372 fs_info->free_space_root = root;
2377 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2378 location.objectid, ret);
2383 * Real super block validation
2384 * NOTE: super csum type and incompat features will not be checked here.
2386 * @sb: super block to check
2387 * @mirror_num: the super block number to check its bytenr:
2388 * 0 the primary (1st) sb
2389 * 1, 2 2nd and 3rd backup copy
2390 * -1 skip bytenr check
2392 static int validate_super(struct btrfs_fs_info *fs_info,
2393 struct btrfs_super_block *sb, int mirror_num)
2395 u64 nodesize = btrfs_super_nodesize(sb);
2396 u64 sectorsize = btrfs_super_sectorsize(sb);
2399 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2400 btrfs_err(fs_info, "no valid FS found");
2403 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2404 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2405 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2408 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2409 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2410 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2413 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2414 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2415 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2418 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2419 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2420 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2425 * Check sectorsize and nodesize first, other check will need it.
2426 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2428 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2429 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2430 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2433 /* Only PAGE SIZE is supported yet */
2434 if (sectorsize != PAGE_SIZE) {
2436 "sectorsize %llu not supported yet, only support %lu",
2437 sectorsize, PAGE_SIZE);
2440 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2441 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2442 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2445 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2446 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2447 le32_to_cpu(sb->__unused_leafsize), nodesize);
2451 /* Root alignment check */
2452 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2453 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2454 btrfs_super_root(sb));
2457 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2458 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2459 btrfs_super_chunk_root(sb));
2462 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2463 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2464 btrfs_super_log_root(sb));
2468 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2469 BTRFS_FSID_SIZE) != 0) {
2471 "dev_item UUID does not match metadata fsid: %pU != %pU",
2472 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2477 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2480 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2481 btrfs_err(fs_info, "bytes_used is too small %llu",
2482 btrfs_super_bytes_used(sb));
2485 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2486 btrfs_err(fs_info, "invalid stripesize %u",
2487 btrfs_super_stripesize(sb));
2490 if (btrfs_super_num_devices(sb) > (1UL << 31))
2491 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2492 btrfs_super_num_devices(sb));
2493 if (btrfs_super_num_devices(sb) == 0) {
2494 btrfs_err(fs_info, "number of devices is 0");
2498 if (mirror_num >= 0 &&
2499 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2500 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2501 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2506 * Obvious sys_chunk_array corruptions, it must hold at least one key
2509 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2510 btrfs_err(fs_info, "system chunk array too big %u > %u",
2511 btrfs_super_sys_array_size(sb),
2512 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2515 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2516 + sizeof(struct btrfs_chunk)) {
2517 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2518 btrfs_super_sys_array_size(sb),
2519 sizeof(struct btrfs_disk_key)
2520 + sizeof(struct btrfs_chunk));
2525 * The generation is a global counter, we'll trust it more than the others
2526 * but it's still possible that it's the one that's wrong.
2528 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2530 "suspicious: generation < chunk_root_generation: %llu < %llu",
2531 btrfs_super_generation(sb),
2532 btrfs_super_chunk_root_generation(sb));
2533 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2534 && btrfs_super_cache_generation(sb) != (u64)-1)
2536 "suspicious: generation < cache_generation: %llu < %llu",
2537 btrfs_super_generation(sb),
2538 btrfs_super_cache_generation(sb));
2544 * Validation of super block at mount time.
2545 * Some checks already done early at mount time, like csum type and incompat
2546 * flags will be skipped.
2548 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2550 return validate_super(fs_info, fs_info->super_copy, 0);
2554 * Validation of super block at write time.
2555 * Some checks like bytenr check will be skipped as their values will be
2557 * Extra checks like csum type and incompat flags will be done here.
2559 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2560 struct btrfs_super_block *sb)
2564 ret = validate_super(fs_info, sb, -1);
2567 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2569 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2570 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2573 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2576 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2577 btrfs_super_incompat_flags(sb),
2578 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2584 "super block corruption detected before writing it to disk");
2588 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2590 int backup_index = find_newest_super_backup(fs_info);
2591 struct btrfs_super_block *sb = fs_info->super_copy;
2592 struct btrfs_root *tree_root = fs_info->tree_root;
2593 bool handle_error = false;
2597 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2602 if (!IS_ERR(tree_root->node))
2603 free_extent_buffer(tree_root->node);
2604 tree_root->node = NULL;
2606 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2609 free_root_pointers(fs_info, 0);
2612 * Don't use the log in recovery mode, it won't be
2615 btrfs_set_super_log_root(sb, 0);
2617 /* We can't trust the free space cache either */
2618 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2620 ret = read_backup_root(fs_info, i);
2625 generation = btrfs_super_generation(sb);
2626 level = btrfs_super_root_level(sb);
2627 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2628 generation, level, NULL);
2629 if (IS_ERR(tree_root->node) ||
2630 !extent_buffer_uptodate(tree_root->node)) {
2631 handle_error = true;
2633 if (IS_ERR(tree_root->node)) {
2634 ret = PTR_ERR(tree_root->node);
2635 tree_root->node = NULL;
2636 } else if (!extent_buffer_uptodate(tree_root->node)) {
2640 btrfs_warn(fs_info, "failed to read tree root");
2644 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2645 tree_root->commit_root = btrfs_root_node(tree_root);
2646 btrfs_set_root_refs(&tree_root->root_item, 1);
2649 * No need to hold btrfs_root::objectid_mutex since the fs
2650 * hasn't been fully initialised and we are the only user
2652 ret = btrfs_find_highest_objectid(tree_root,
2653 &tree_root->highest_objectid);
2655 handle_error = true;
2659 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2661 ret = btrfs_read_roots(fs_info);
2663 handle_error = true;
2667 /* All successful */
2668 fs_info->generation = generation;
2669 fs_info->last_trans_committed = generation;
2671 /* Always begin writing backup roots after the one being used */
2672 if (backup_index < 0) {
2673 fs_info->backup_root_index = 0;
2675 fs_info->backup_root_index = backup_index + 1;
2676 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2684 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2686 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2687 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2688 INIT_LIST_HEAD(&fs_info->trans_list);
2689 INIT_LIST_HEAD(&fs_info->dead_roots);
2690 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2691 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2692 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2693 spin_lock_init(&fs_info->delalloc_root_lock);
2694 spin_lock_init(&fs_info->trans_lock);
2695 spin_lock_init(&fs_info->fs_roots_radix_lock);
2696 spin_lock_init(&fs_info->delayed_iput_lock);
2697 spin_lock_init(&fs_info->defrag_inodes_lock);
2698 spin_lock_init(&fs_info->super_lock);
2699 spin_lock_init(&fs_info->buffer_lock);
2700 spin_lock_init(&fs_info->unused_bgs_lock);
2701 rwlock_init(&fs_info->tree_mod_log_lock);
2702 mutex_init(&fs_info->unused_bg_unpin_mutex);
2703 mutex_init(&fs_info->delete_unused_bgs_mutex);
2704 mutex_init(&fs_info->reloc_mutex);
2705 mutex_init(&fs_info->delalloc_root_mutex);
2706 seqlock_init(&fs_info->profiles_lock);
2708 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2709 INIT_LIST_HEAD(&fs_info->space_info);
2710 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2711 INIT_LIST_HEAD(&fs_info->unused_bgs);
2712 #ifdef CONFIG_BTRFS_DEBUG
2713 INIT_LIST_HEAD(&fs_info->allocated_roots);
2714 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2715 spin_lock_init(&fs_info->eb_leak_lock);
2717 extent_map_tree_init(&fs_info->mapping_tree);
2718 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2719 BTRFS_BLOCK_RSV_GLOBAL);
2720 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2721 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2722 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2723 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2724 BTRFS_BLOCK_RSV_DELOPS);
2725 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2726 BTRFS_BLOCK_RSV_DELREFS);
2728 atomic_set(&fs_info->async_delalloc_pages, 0);
2729 atomic_set(&fs_info->defrag_running, 0);
2730 atomic_set(&fs_info->reada_works_cnt, 0);
2731 atomic_set(&fs_info->nr_delayed_iputs, 0);
2732 atomic64_set(&fs_info->tree_mod_seq, 0);
2733 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2734 fs_info->metadata_ratio = 0;
2735 fs_info->defrag_inodes = RB_ROOT;
2736 atomic64_set(&fs_info->free_chunk_space, 0);
2737 fs_info->tree_mod_log = RB_ROOT;
2738 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2739 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2740 /* readahead state */
2741 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2742 spin_lock_init(&fs_info->reada_lock);
2743 btrfs_init_ref_verify(fs_info);
2745 fs_info->thread_pool_size = min_t(unsigned long,
2746 num_online_cpus() + 2, 8);
2748 INIT_LIST_HEAD(&fs_info->ordered_roots);
2749 spin_lock_init(&fs_info->ordered_root_lock);
2751 btrfs_init_scrub(fs_info);
2752 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2753 fs_info->check_integrity_print_mask = 0;
2755 btrfs_init_balance(fs_info);
2756 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2758 spin_lock_init(&fs_info->block_group_cache_lock);
2759 fs_info->block_group_cache_tree = RB_ROOT;
2760 fs_info->first_logical_byte = (u64)-1;
2762 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2763 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2764 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2766 mutex_init(&fs_info->ordered_operations_mutex);
2767 mutex_init(&fs_info->tree_log_mutex);
2768 mutex_init(&fs_info->chunk_mutex);
2769 mutex_init(&fs_info->transaction_kthread_mutex);
2770 mutex_init(&fs_info->cleaner_mutex);
2771 mutex_init(&fs_info->ro_block_group_mutex);
2772 init_rwsem(&fs_info->commit_root_sem);
2773 init_rwsem(&fs_info->cleanup_work_sem);
2774 init_rwsem(&fs_info->subvol_sem);
2775 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2777 btrfs_init_dev_replace_locks(fs_info);
2778 btrfs_init_qgroup(fs_info);
2779 btrfs_discard_init(fs_info);
2781 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2782 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2784 init_waitqueue_head(&fs_info->transaction_throttle);
2785 init_waitqueue_head(&fs_info->transaction_wait);
2786 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2787 init_waitqueue_head(&fs_info->async_submit_wait);
2788 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2790 /* Usable values until the real ones are cached from the superblock */
2791 fs_info->nodesize = 4096;
2792 fs_info->sectorsize = 4096;
2793 fs_info->stripesize = 4096;
2795 spin_lock_init(&fs_info->swapfile_pins_lock);
2796 fs_info->swapfile_pins = RB_ROOT;
2798 fs_info->send_in_progress = 0;
2801 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2806 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2807 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2809 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2813 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2817 fs_info->dirty_metadata_batch = PAGE_SIZE *
2818 (1 + ilog2(nr_cpu_ids));
2820 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2824 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2829 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2831 if (!fs_info->delayed_root)
2833 btrfs_init_delayed_root(fs_info->delayed_root);
2835 return btrfs_alloc_stripe_hash_table(fs_info);
2838 static int btrfs_uuid_rescan_kthread(void *data)
2840 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2844 * 1st step is to iterate through the existing UUID tree and
2845 * to delete all entries that contain outdated data.
2846 * 2nd step is to add all missing entries to the UUID tree.
2848 ret = btrfs_uuid_tree_iterate(fs_info);
2851 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2853 up(&fs_info->uuid_tree_rescan_sem);
2856 return btrfs_uuid_scan_kthread(data);
2859 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2861 struct task_struct *task;
2863 down(&fs_info->uuid_tree_rescan_sem);
2864 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2866 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2867 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2868 up(&fs_info->uuid_tree_rescan_sem);
2869 return PTR_ERR(task);
2875 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2884 struct btrfs_super_block *disk_super;
2885 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2886 struct btrfs_root *tree_root;
2887 struct btrfs_root *chunk_root;
2890 int clear_free_space_tree = 0;
2893 ret = init_mount_fs_info(fs_info, sb);
2899 /* These need to be init'ed before we start creating inodes and such. */
2900 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2902 fs_info->tree_root = tree_root;
2903 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2905 fs_info->chunk_root = chunk_root;
2906 if (!tree_root || !chunk_root) {
2911 fs_info->btree_inode = new_inode(sb);
2912 if (!fs_info->btree_inode) {
2916 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2917 btrfs_init_btree_inode(fs_info);
2919 invalidate_bdev(fs_devices->latest_bdev);
2922 * Read super block and check the signature bytes only
2924 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2925 if (IS_ERR(disk_super)) {
2926 err = PTR_ERR(disk_super);
2931 * Verify the type first, if that or the the checksum value are
2932 * corrupted, we'll find out
2934 csum_type = btrfs_super_csum_type(disk_super);
2935 if (!btrfs_supported_super_csum(csum_type)) {
2936 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2939 btrfs_release_disk_super(disk_super);
2943 ret = btrfs_init_csum_hash(fs_info, csum_type);
2946 btrfs_release_disk_super(disk_super);
2951 * We want to check superblock checksum, the type is stored inside.
2952 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2954 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2955 btrfs_err(fs_info, "superblock checksum mismatch");
2957 btrfs_release_disk_super(disk_super);
2962 * super_copy is zeroed at allocation time and we never touch the
2963 * following bytes up to INFO_SIZE, the checksum is calculated from
2964 * the whole block of INFO_SIZE
2966 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2967 btrfs_release_disk_super(disk_super);
2969 disk_super = fs_info->super_copy;
2971 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2974 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2975 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2976 fs_info->super_copy->metadata_uuid,
2980 features = btrfs_super_flags(disk_super);
2981 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2982 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2983 btrfs_set_super_flags(disk_super, features);
2985 "found metadata UUID change in progress flag, clearing");
2988 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2989 sizeof(*fs_info->super_for_commit));
2991 ret = btrfs_validate_mount_super(fs_info);
2993 btrfs_err(fs_info, "superblock contains fatal errors");
2998 if (!btrfs_super_root(disk_super))
3001 /* check FS state, whether FS is broken. */
3002 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3003 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3006 * In the long term, we'll store the compression type in the super
3007 * block, and it'll be used for per file compression control.
3009 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3011 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3017 features = btrfs_super_incompat_flags(disk_super) &
3018 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3021 "cannot mount because of unsupported optional features (%llx)",
3027 features = btrfs_super_incompat_flags(disk_super);
3028 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3029 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3030 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3031 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3032 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3034 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3035 btrfs_info(fs_info, "has skinny extents");
3038 * flag our filesystem as having big metadata blocks if
3039 * they are bigger than the page size
3041 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3042 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3044 "flagging fs with big metadata feature");
3045 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3048 nodesize = btrfs_super_nodesize(disk_super);
3049 sectorsize = btrfs_super_sectorsize(disk_super);
3050 stripesize = sectorsize;
3051 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3052 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3054 /* Cache block sizes */
3055 fs_info->nodesize = nodesize;
3056 fs_info->sectorsize = sectorsize;
3057 fs_info->stripesize = stripesize;
3060 * mixed block groups end up with duplicate but slightly offset
3061 * extent buffers for the same range. It leads to corruptions
3063 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3064 (sectorsize != nodesize)) {
3066 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3067 nodesize, sectorsize);
3072 * Needn't use the lock because there is no other task which will
3075 btrfs_set_super_incompat_flags(disk_super, features);
3077 features = btrfs_super_compat_ro_flags(disk_super) &
3078 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3079 if (!sb_rdonly(sb) && features) {
3081 "cannot mount read-write because of unsupported optional features (%llx)",
3087 ret = btrfs_init_workqueues(fs_info, fs_devices);
3090 goto fail_sb_buffer;
3093 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3094 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3095 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3096 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3098 sb->s_blocksize = sectorsize;
3099 sb->s_blocksize_bits = blksize_bits(sectorsize);
3100 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3102 mutex_lock(&fs_info->chunk_mutex);
3103 ret = btrfs_read_sys_array(fs_info);
3104 mutex_unlock(&fs_info->chunk_mutex);
3106 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3107 goto fail_sb_buffer;
3110 generation = btrfs_super_chunk_root_generation(disk_super);
3111 level = btrfs_super_chunk_root_level(disk_super);
3113 chunk_root->node = read_tree_block(fs_info,
3114 btrfs_super_chunk_root(disk_super),
3115 generation, level, NULL);
3116 if (IS_ERR(chunk_root->node) ||
3117 !extent_buffer_uptodate(chunk_root->node)) {
3118 btrfs_err(fs_info, "failed to read chunk root");
3119 if (!IS_ERR(chunk_root->node))
3120 free_extent_buffer(chunk_root->node);
3121 chunk_root->node = NULL;
3122 goto fail_tree_roots;
3124 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3125 chunk_root->commit_root = btrfs_root_node(chunk_root);
3127 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3128 offsetof(struct btrfs_header, chunk_tree_uuid),
3131 ret = btrfs_read_chunk_tree(fs_info);
3133 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3134 goto fail_tree_roots;
3138 * Keep the devid that is marked to be the target device for the
3139 * device replace procedure
3141 btrfs_free_extra_devids(fs_devices, 0);
3143 if (!fs_devices->latest_bdev) {
3144 btrfs_err(fs_info, "failed to read devices");
3145 goto fail_tree_roots;
3148 ret = init_tree_roots(fs_info);
3150 goto fail_tree_roots;
3153 * If we have a uuid root and we're not being told to rescan we need to
3154 * check the generation here so we can set the
3155 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3156 * transaction during a balance or the log replay without updating the
3157 * uuid generation, and then if we crash we would rescan the uuid tree,
3158 * even though it was perfectly fine.
3160 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3161 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3162 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3164 ret = btrfs_verify_dev_extents(fs_info);
3167 "failed to verify dev extents against chunks: %d",
3169 goto fail_block_groups;
3171 ret = btrfs_recover_balance(fs_info);
3173 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3174 goto fail_block_groups;
3177 ret = btrfs_init_dev_stats(fs_info);
3179 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3180 goto fail_block_groups;
3183 ret = btrfs_init_dev_replace(fs_info);
3185 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3186 goto fail_block_groups;
3189 btrfs_free_extra_devids(fs_devices, 1);
3191 ret = btrfs_sysfs_add_fsid(fs_devices);
3193 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3195 goto fail_block_groups;
3198 ret = btrfs_sysfs_add_mounted(fs_info);
3200 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3201 goto fail_fsdev_sysfs;
3204 ret = btrfs_init_space_info(fs_info);
3206 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3210 ret = btrfs_read_block_groups(fs_info);
3212 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3216 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3218 "writable mount is not allowed due to too many missing devices");
3222 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3224 if (IS_ERR(fs_info->cleaner_kthread))
3227 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3229 "btrfs-transaction");
3230 if (IS_ERR(fs_info->transaction_kthread))
3233 if (!btrfs_test_opt(fs_info, NOSSD) &&
3234 !fs_info->fs_devices->rotating) {
3235 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3239 * Mount does not set all options immediately, we can do it now and do
3240 * not have to wait for transaction commit
3242 btrfs_apply_pending_changes(fs_info);
3244 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3245 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3246 ret = btrfsic_mount(fs_info, fs_devices,
3247 btrfs_test_opt(fs_info,
3248 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3250 fs_info->check_integrity_print_mask);
3253 "failed to initialize integrity check module: %d",
3257 ret = btrfs_read_qgroup_config(fs_info);
3259 goto fail_trans_kthread;
3261 if (btrfs_build_ref_tree(fs_info))
3262 btrfs_err(fs_info, "couldn't build ref tree");
3264 /* do not make disk changes in broken FS or nologreplay is given */
3265 if (btrfs_super_log_root(disk_super) != 0 &&
3266 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3267 btrfs_info(fs_info, "start tree-log replay");
3268 ret = btrfs_replay_log(fs_info, fs_devices);
3275 ret = btrfs_find_orphan_roots(fs_info);
3279 if (!sb_rdonly(sb)) {
3280 ret = btrfs_cleanup_fs_roots(fs_info);
3284 mutex_lock(&fs_info->cleaner_mutex);
3285 ret = btrfs_recover_relocation(tree_root);
3286 mutex_unlock(&fs_info->cleaner_mutex);
3288 btrfs_warn(fs_info, "failed to recover relocation: %d",
3295 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3296 if (IS_ERR(fs_info->fs_root)) {
3297 err = PTR_ERR(fs_info->fs_root);
3298 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3299 fs_info->fs_root = NULL;
3306 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3307 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3308 clear_free_space_tree = 1;
3309 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3310 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3311 btrfs_warn(fs_info, "free space tree is invalid");
3312 clear_free_space_tree = 1;
3315 if (clear_free_space_tree) {
3316 btrfs_info(fs_info, "clearing free space tree");
3317 ret = btrfs_clear_free_space_tree(fs_info);
3320 "failed to clear free space tree: %d", ret);
3321 close_ctree(fs_info);
3326 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3327 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3328 btrfs_info(fs_info, "creating free space tree");
3329 ret = btrfs_create_free_space_tree(fs_info);
3332 "failed to create free space tree: %d", ret);
3333 close_ctree(fs_info);
3338 down_read(&fs_info->cleanup_work_sem);
3339 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3340 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3341 up_read(&fs_info->cleanup_work_sem);
3342 close_ctree(fs_info);
3345 up_read(&fs_info->cleanup_work_sem);
3347 ret = btrfs_resume_balance_async(fs_info);
3349 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3350 close_ctree(fs_info);
3354 ret = btrfs_resume_dev_replace_async(fs_info);
3356 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3357 close_ctree(fs_info);
3361 btrfs_qgroup_rescan_resume(fs_info);
3362 btrfs_discard_resume(fs_info);
3364 if (!fs_info->uuid_root) {
3365 btrfs_info(fs_info, "creating UUID tree");
3366 ret = btrfs_create_uuid_tree(fs_info);
3369 "failed to create the UUID tree: %d", ret);
3370 close_ctree(fs_info);
3373 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3374 fs_info->generation !=
3375 btrfs_super_uuid_tree_generation(disk_super)) {
3376 btrfs_info(fs_info, "checking UUID tree");
3377 ret = btrfs_check_uuid_tree(fs_info);
3380 "failed to check the UUID tree: %d", ret);
3381 close_ctree(fs_info);
3385 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3388 * backuproot only affect mount behavior, and if open_ctree succeeded,
3389 * no need to keep the flag
3391 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3396 btrfs_free_qgroup_config(fs_info);
3398 kthread_stop(fs_info->transaction_kthread);
3399 btrfs_cleanup_transaction(fs_info);
3400 btrfs_free_fs_roots(fs_info);
3402 kthread_stop(fs_info->cleaner_kthread);
3405 * make sure we're done with the btree inode before we stop our
3408 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3411 btrfs_sysfs_remove_mounted(fs_info);
3414 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3417 btrfs_put_block_group_cache(fs_info);
3420 if (fs_info->data_reloc_root)
3421 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3422 free_root_pointers(fs_info, true);
3423 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3426 btrfs_stop_all_workers(fs_info);
3427 btrfs_free_block_groups(fs_info);
3429 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3431 iput(fs_info->btree_inode);
3433 btrfs_close_devices(fs_info->fs_devices);
3436 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3438 static void btrfs_end_super_write(struct bio *bio)
3440 struct btrfs_device *device = bio->bi_private;
3441 struct bio_vec *bvec;
3442 struct bvec_iter_all iter_all;
3445 bio_for_each_segment_all(bvec, bio, iter_all) {
3446 page = bvec->bv_page;
3448 if (bio->bi_status) {
3449 btrfs_warn_rl_in_rcu(device->fs_info,
3450 "lost page write due to IO error on %s (%d)",
3451 rcu_str_deref(device->name),
3452 blk_status_to_errno(bio->bi_status));
3453 ClearPageUptodate(page);
3455 btrfs_dev_stat_inc_and_print(device,
3456 BTRFS_DEV_STAT_WRITE_ERRS);
3458 SetPageUptodate(page);
3468 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3471 struct btrfs_super_block *super;
3474 struct address_space *mapping = bdev->bd_inode->i_mapping;
3476 bytenr = btrfs_sb_offset(copy_num);
3477 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3478 return ERR_PTR(-EINVAL);
3480 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3482 return ERR_CAST(page);
3484 super = page_address(page);
3485 if (btrfs_super_bytenr(super) != bytenr ||
3486 btrfs_super_magic(super) != BTRFS_MAGIC) {
3487 btrfs_release_disk_super(super);
3488 return ERR_PTR(-EINVAL);
3495 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3497 struct btrfs_super_block *super, *latest = NULL;
3501 /* we would like to check all the supers, but that would make
3502 * a btrfs mount succeed after a mkfs from a different FS.
3503 * So, we need to add a special mount option to scan for
3504 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3506 for (i = 0; i < 1; i++) {
3507 super = btrfs_read_dev_one_super(bdev, i);
3511 if (!latest || btrfs_super_generation(super) > transid) {
3513 btrfs_release_disk_super(super);
3516 transid = btrfs_super_generation(super);
3524 * Write superblock @sb to the @device. Do not wait for completion, all the
3525 * pages we use for writing are locked.
3527 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3528 * the expected device size at commit time. Note that max_mirrors must be
3529 * same for write and wait phases.
3531 * Return number of errors when page is not found or submission fails.
3533 static int write_dev_supers(struct btrfs_device *device,
3534 struct btrfs_super_block *sb, int max_mirrors)
3536 struct btrfs_fs_info *fs_info = device->fs_info;
3537 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3538 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3543 if (max_mirrors == 0)
3544 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3546 shash->tfm = fs_info->csum_shash;
3548 for (i = 0; i < max_mirrors; i++) {
3551 struct btrfs_super_block *disk_super;
3553 bytenr = btrfs_sb_offset(i);
3554 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3555 device->commit_total_bytes)
3558 btrfs_set_super_bytenr(sb, bytenr);
3560 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3561 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3564 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3567 btrfs_err(device->fs_info,
3568 "couldn't get super block page for bytenr %llu",
3574 /* Bump the refcount for wait_dev_supers() */
3577 disk_super = page_address(page);
3578 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3581 * Directly use bios here instead of relying on the page cache
3582 * to do I/O, so we don't lose the ability to do integrity
3585 bio = bio_alloc(GFP_NOFS, 1);
3586 bio_set_dev(bio, device->bdev);
3587 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3588 bio->bi_private = device;
3589 bio->bi_end_io = btrfs_end_super_write;
3590 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3591 offset_in_page(bytenr));
3594 * We FUA only the first super block. The others we allow to
3595 * go down lazy and there's a short window where the on-disk
3596 * copies might still contain the older version.
3598 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3599 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3600 bio->bi_opf |= REQ_FUA;
3602 btrfsic_submit_bio(bio);
3604 return errors < i ? 0 : -1;
3608 * Wait for write completion of superblocks done by write_dev_supers,
3609 * @max_mirrors same for write and wait phases.
3611 * Return number of errors when page is not found or not marked up to
3614 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3618 bool primary_failed = false;
3621 if (max_mirrors == 0)
3622 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3624 for (i = 0; i < max_mirrors; i++) {
3627 bytenr = btrfs_sb_offset(i);
3628 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3629 device->commit_total_bytes)
3632 page = find_get_page(device->bdev->bd_inode->i_mapping,
3633 bytenr >> PAGE_SHIFT);
3637 primary_failed = true;
3640 /* Page is submitted locked and unlocked once the IO completes */
3641 wait_on_page_locked(page);
3642 if (PageError(page)) {
3645 primary_failed = true;
3648 /* Drop our reference */
3651 /* Drop the reference from the writing run */
3655 /* log error, force error return */
3656 if (primary_failed) {
3657 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3662 return errors < i ? 0 : -1;
3666 * endio for the write_dev_flush, this will wake anyone waiting
3667 * for the barrier when it is done
3669 static void btrfs_end_empty_barrier(struct bio *bio)
3671 complete(bio->bi_private);
3675 * Submit a flush request to the device if it supports it. Error handling is
3676 * done in the waiting counterpart.
3678 static void write_dev_flush(struct btrfs_device *device)
3680 struct request_queue *q = bdev_get_queue(device->bdev);
3681 struct bio *bio = device->flush_bio;
3683 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3687 bio->bi_end_io = btrfs_end_empty_barrier;
3688 bio_set_dev(bio, device->bdev);
3689 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3690 init_completion(&device->flush_wait);
3691 bio->bi_private = &device->flush_wait;
3693 btrfsic_submit_bio(bio);
3694 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3698 * If the flush bio has been submitted by write_dev_flush, wait for it.
3700 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3702 struct bio *bio = device->flush_bio;
3704 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3707 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3708 wait_for_completion_io(&device->flush_wait);
3710 return bio->bi_status;
3713 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3715 if (!btrfs_check_rw_degradable(fs_info, NULL))
3721 * send an empty flush down to each device in parallel,
3722 * then wait for them
3724 static int barrier_all_devices(struct btrfs_fs_info *info)
3726 struct list_head *head;
3727 struct btrfs_device *dev;
3728 int errors_wait = 0;
3731 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3732 /* send down all the barriers */
3733 head = &info->fs_devices->devices;
3734 list_for_each_entry(dev, head, dev_list) {
3735 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3739 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3740 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3743 write_dev_flush(dev);
3744 dev->last_flush_error = BLK_STS_OK;
3747 /* wait for all the barriers */
3748 list_for_each_entry(dev, head, dev_list) {
3749 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3755 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3756 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3759 ret = wait_dev_flush(dev);
3761 dev->last_flush_error = ret;
3762 btrfs_dev_stat_inc_and_print(dev,
3763 BTRFS_DEV_STAT_FLUSH_ERRS);
3770 * At some point we need the status of all disks
3771 * to arrive at the volume status. So error checking
3772 * is being pushed to a separate loop.
3774 return check_barrier_error(info);
3779 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3782 int min_tolerated = INT_MAX;
3784 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3785 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3786 min_tolerated = min_t(int, min_tolerated,
3787 btrfs_raid_array[BTRFS_RAID_SINGLE].
3788 tolerated_failures);
3790 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3791 if (raid_type == BTRFS_RAID_SINGLE)
3793 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3795 min_tolerated = min_t(int, min_tolerated,
3796 btrfs_raid_array[raid_type].
3797 tolerated_failures);
3800 if (min_tolerated == INT_MAX) {
3801 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3805 return min_tolerated;
3808 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3810 struct list_head *head;
3811 struct btrfs_device *dev;
3812 struct btrfs_super_block *sb;
3813 struct btrfs_dev_item *dev_item;
3817 int total_errors = 0;
3820 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3823 * max_mirrors == 0 indicates we're from commit_transaction,
3824 * not from fsync where the tree roots in fs_info have not
3825 * been consistent on disk.
3827 if (max_mirrors == 0)
3828 backup_super_roots(fs_info);
3830 sb = fs_info->super_for_commit;
3831 dev_item = &sb->dev_item;
3833 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3834 head = &fs_info->fs_devices->devices;
3835 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3838 ret = barrier_all_devices(fs_info);
3841 &fs_info->fs_devices->device_list_mutex);
3842 btrfs_handle_fs_error(fs_info, ret,
3843 "errors while submitting device barriers.");
3848 list_for_each_entry(dev, head, dev_list) {
3853 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3854 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3857 btrfs_set_stack_device_generation(dev_item, 0);
3858 btrfs_set_stack_device_type(dev_item, dev->type);
3859 btrfs_set_stack_device_id(dev_item, dev->devid);
3860 btrfs_set_stack_device_total_bytes(dev_item,
3861 dev->commit_total_bytes);
3862 btrfs_set_stack_device_bytes_used(dev_item,
3863 dev->commit_bytes_used);
3864 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3865 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3866 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3867 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3868 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3871 flags = btrfs_super_flags(sb);
3872 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3874 ret = btrfs_validate_write_super(fs_info, sb);
3876 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3877 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3878 "unexpected superblock corruption detected");
3882 ret = write_dev_supers(dev, sb, max_mirrors);
3886 if (total_errors > max_errors) {
3887 btrfs_err(fs_info, "%d errors while writing supers",
3889 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3891 /* FUA is masked off if unsupported and can't be the reason */
3892 btrfs_handle_fs_error(fs_info, -EIO,
3893 "%d errors while writing supers",
3899 list_for_each_entry(dev, head, dev_list) {
3902 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3903 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3906 ret = wait_dev_supers(dev, max_mirrors);
3910 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3911 if (total_errors > max_errors) {
3912 btrfs_handle_fs_error(fs_info, -EIO,
3913 "%d errors while writing supers",
3920 /* Drop a fs root from the radix tree and free it. */
3921 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3922 struct btrfs_root *root)
3924 bool drop_ref = false;
3926 spin_lock(&fs_info->fs_roots_radix_lock);
3927 radix_tree_delete(&fs_info->fs_roots_radix,
3928 (unsigned long)root->root_key.objectid);
3929 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3931 spin_unlock(&fs_info->fs_roots_radix_lock);
3933 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3934 ASSERT(root->log_root == NULL);
3935 if (root->reloc_root) {
3936 btrfs_put_root(root->reloc_root);
3937 root->reloc_root = NULL;
3941 if (root->free_ino_pinned)
3942 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3943 if (root->free_ino_ctl)
3944 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3945 if (root->ino_cache_inode) {
3946 iput(root->ino_cache_inode);
3947 root->ino_cache_inode = NULL;
3950 btrfs_put_root(root);
3953 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3955 u64 root_objectid = 0;
3956 struct btrfs_root *gang[8];
3959 unsigned int ret = 0;
3962 spin_lock(&fs_info->fs_roots_radix_lock);
3963 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3964 (void **)gang, root_objectid,
3967 spin_unlock(&fs_info->fs_roots_radix_lock);
3970 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3972 for (i = 0; i < ret; i++) {
3973 /* Avoid to grab roots in dead_roots */
3974 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3978 /* grab all the search result for later use */
3979 gang[i] = btrfs_grab_root(gang[i]);
3981 spin_unlock(&fs_info->fs_roots_radix_lock);
3983 for (i = 0; i < ret; i++) {
3986 root_objectid = gang[i]->root_key.objectid;
3987 err = btrfs_orphan_cleanup(gang[i]);
3990 btrfs_put_root(gang[i]);
3995 /* release the uncleaned roots due to error */
3996 for (; i < ret; i++) {
3998 btrfs_put_root(gang[i]);
4003 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4005 struct btrfs_root *root = fs_info->tree_root;
4006 struct btrfs_trans_handle *trans;
4008 mutex_lock(&fs_info->cleaner_mutex);
4009 btrfs_run_delayed_iputs(fs_info);
4010 mutex_unlock(&fs_info->cleaner_mutex);
4011 wake_up_process(fs_info->cleaner_kthread);
4013 /* wait until ongoing cleanup work done */
4014 down_write(&fs_info->cleanup_work_sem);
4015 up_write(&fs_info->cleanup_work_sem);
4017 trans = btrfs_join_transaction(root);
4019 return PTR_ERR(trans);
4020 return btrfs_commit_transaction(trans);
4023 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4027 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4029 * We don't want the cleaner to start new transactions, add more delayed
4030 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4031 * because that frees the task_struct, and the transaction kthread might
4032 * still try to wake up the cleaner.
4034 kthread_park(fs_info->cleaner_kthread);
4036 /* wait for the qgroup rescan worker to stop */
4037 btrfs_qgroup_wait_for_completion(fs_info, false);
4039 /* wait for the uuid_scan task to finish */
4040 down(&fs_info->uuid_tree_rescan_sem);
4041 /* avoid complains from lockdep et al., set sem back to initial state */
4042 up(&fs_info->uuid_tree_rescan_sem);
4044 /* pause restriper - we want to resume on mount */
4045 btrfs_pause_balance(fs_info);
4047 btrfs_dev_replace_suspend_for_unmount(fs_info);
4049 btrfs_scrub_cancel(fs_info);
4051 /* wait for any defraggers to finish */
4052 wait_event(fs_info->transaction_wait,
4053 (atomic_read(&fs_info->defrag_running) == 0));
4055 /* clear out the rbtree of defraggable inodes */
4056 btrfs_cleanup_defrag_inodes(fs_info);
4058 cancel_work_sync(&fs_info->async_reclaim_work);
4060 /* Cancel or finish ongoing discard work */
4061 btrfs_discard_cleanup(fs_info);
4063 if (!sb_rdonly(fs_info->sb)) {
4065 * The cleaner kthread is stopped, so do one final pass over
4066 * unused block groups.
4068 btrfs_delete_unused_bgs(fs_info);
4071 * There might be existing delayed inode workers still running
4072 * and holding an empty delayed inode item. We must wait for
4073 * them to complete first because they can create a transaction.
4074 * This happens when someone calls btrfs_balance_delayed_items()
4075 * and then a transaction commit runs the same delayed nodes
4076 * before any delayed worker has done something with the nodes.
4077 * We must wait for any worker here and not at transaction
4078 * commit time since that could cause a deadlock.
4079 * This is a very rare case.
4081 btrfs_flush_workqueue(fs_info->delayed_workers);
4083 ret = btrfs_commit_super(fs_info);
4085 btrfs_err(fs_info, "commit super ret %d", ret);
4088 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4089 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4090 btrfs_error_commit_super(fs_info);
4092 kthread_stop(fs_info->transaction_kthread);
4093 kthread_stop(fs_info->cleaner_kthread);
4095 ASSERT(list_empty(&fs_info->delayed_iputs));
4096 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4098 if (btrfs_check_quota_leak(fs_info)) {
4099 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4100 btrfs_err(fs_info, "qgroup reserved space leaked");
4103 btrfs_free_qgroup_config(fs_info);
4104 ASSERT(list_empty(&fs_info->delalloc_roots));
4106 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4107 btrfs_info(fs_info, "at unmount delalloc count %lld",
4108 percpu_counter_sum(&fs_info->delalloc_bytes));
4111 if (percpu_counter_sum(&fs_info->dio_bytes))
4112 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4113 percpu_counter_sum(&fs_info->dio_bytes));
4115 btrfs_sysfs_remove_mounted(fs_info);
4116 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4118 btrfs_put_block_group_cache(fs_info);
4121 * we must make sure there is not any read request to
4122 * submit after we stopping all workers.
4124 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4125 btrfs_stop_all_workers(fs_info);
4127 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4128 free_root_pointers(fs_info, true);
4129 btrfs_free_fs_roots(fs_info);
4132 * We must free the block groups after dropping the fs_roots as we could
4133 * have had an IO error and have left over tree log blocks that aren't
4134 * cleaned up until the fs roots are freed. This makes the block group
4135 * accounting appear to be wrong because there's pending reserved bytes,
4136 * so make sure we do the block group cleanup afterwards.
4138 btrfs_free_block_groups(fs_info);
4140 iput(fs_info->btree_inode);
4142 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4143 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4144 btrfsic_unmount(fs_info->fs_devices);
4147 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4148 btrfs_close_devices(fs_info->fs_devices);
4151 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4155 struct inode *btree_inode = buf->pages[0]->mapping->host;
4157 ret = extent_buffer_uptodate(buf);
4161 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4162 parent_transid, atomic);
4168 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4170 struct btrfs_fs_info *fs_info;
4171 struct btrfs_root *root;
4172 u64 transid = btrfs_header_generation(buf);
4175 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4177 * This is a fast path so only do this check if we have sanity tests
4178 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4179 * outside of the sanity tests.
4181 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4184 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4185 fs_info = root->fs_info;
4186 btrfs_assert_tree_locked(buf);
4187 if (transid != fs_info->generation)
4188 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4189 buf->start, transid, fs_info->generation);
4190 was_dirty = set_extent_buffer_dirty(buf);
4192 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4194 fs_info->dirty_metadata_batch);
4195 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4197 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4198 * but item data not updated.
4199 * So here we should only check item pointers, not item data.
4201 if (btrfs_header_level(buf) == 0 &&
4202 btrfs_check_leaf_relaxed(buf)) {
4203 btrfs_print_leaf(buf);
4209 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4213 * looks as though older kernels can get into trouble with
4214 * this code, they end up stuck in balance_dirty_pages forever
4218 if (current->flags & PF_MEMALLOC)
4222 btrfs_balance_delayed_items(fs_info);
4224 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4225 BTRFS_DIRTY_METADATA_THRESH,
4226 fs_info->dirty_metadata_batch);
4228 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4232 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4234 __btrfs_btree_balance_dirty(fs_info, 1);
4237 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4239 __btrfs_btree_balance_dirty(fs_info, 0);
4242 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4243 struct btrfs_key *first_key)
4245 return btree_read_extent_buffer_pages(buf, parent_transid,
4249 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4251 /* cleanup FS via transaction */
4252 btrfs_cleanup_transaction(fs_info);
4254 mutex_lock(&fs_info->cleaner_mutex);
4255 btrfs_run_delayed_iputs(fs_info);
4256 mutex_unlock(&fs_info->cleaner_mutex);
4258 down_write(&fs_info->cleanup_work_sem);
4259 up_write(&fs_info->cleanup_work_sem);
4262 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4264 struct btrfs_root *gang[8];
4265 u64 root_objectid = 0;
4268 spin_lock(&fs_info->fs_roots_radix_lock);
4269 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4270 (void **)gang, root_objectid,
4271 ARRAY_SIZE(gang))) != 0) {
4274 for (i = 0; i < ret; i++)
4275 gang[i] = btrfs_grab_root(gang[i]);
4276 spin_unlock(&fs_info->fs_roots_radix_lock);
4278 for (i = 0; i < ret; i++) {
4281 root_objectid = gang[i]->root_key.objectid;
4282 btrfs_free_log(NULL, gang[i]);
4283 btrfs_put_root(gang[i]);
4286 spin_lock(&fs_info->fs_roots_radix_lock);
4288 spin_unlock(&fs_info->fs_roots_radix_lock);
4289 btrfs_free_log_root_tree(NULL, fs_info);
4292 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4294 struct btrfs_ordered_extent *ordered;
4296 spin_lock(&root->ordered_extent_lock);
4298 * This will just short circuit the ordered completion stuff which will
4299 * make sure the ordered extent gets properly cleaned up.
4301 list_for_each_entry(ordered, &root->ordered_extents,
4303 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4304 spin_unlock(&root->ordered_extent_lock);
4307 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4309 struct btrfs_root *root;
4310 struct list_head splice;
4312 INIT_LIST_HEAD(&splice);
4314 spin_lock(&fs_info->ordered_root_lock);
4315 list_splice_init(&fs_info->ordered_roots, &splice);
4316 while (!list_empty(&splice)) {
4317 root = list_first_entry(&splice, struct btrfs_root,
4319 list_move_tail(&root->ordered_root,
4320 &fs_info->ordered_roots);
4322 spin_unlock(&fs_info->ordered_root_lock);
4323 btrfs_destroy_ordered_extents(root);
4326 spin_lock(&fs_info->ordered_root_lock);
4328 spin_unlock(&fs_info->ordered_root_lock);
4331 * We need this here because if we've been flipped read-only we won't
4332 * get sync() from the umount, so we need to make sure any ordered
4333 * extents that haven't had their dirty pages IO start writeout yet
4334 * actually get run and error out properly.
4336 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4339 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4340 struct btrfs_fs_info *fs_info)
4342 struct rb_node *node;
4343 struct btrfs_delayed_ref_root *delayed_refs;
4344 struct btrfs_delayed_ref_node *ref;
4347 delayed_refs = &trans->delayed_refs;
4349 spin_lock(&delayed_refs->lock);
4350 if (atomic_read(&delayed_refs->num_entries) == 0) {
4351 spin_unlock(&delayed_refs->lock);
4352 btrfs_debug(fs_info, "delayed_refs has NO entry");
4356 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4357 struct btrfs_delayed_ref_head *head;
4359 bool pin_bytes = false;
4361 head = rb_entry(node, struct btrfs_delayed_ref_head,
4363 if (btrfs_delayed_ref_lock(delayed_refs, head))
4366 spin_lock(&head->lock);
4367 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4368 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4371 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4372 RB_CLEAR_NODE(&ref->ref_node);
4373 if (!list_empty(&ref->add_list))
4374 list_del(&ref->add_list);
4375 atomic_dec(&delayed_refs->num_entries);
4376 btrfs_put_delayed_ref(ref);
4378 if (head->must_insert_reserved)
4380 btrfs_free_delayed_extent_op(head->extent_op);
4381 btrfs_delete_ref_head(delayed_refs, head);
4382 spin_unlock(&head->lock);
4383 spin_unlock(&delayed_refs->lock);
4384 mutex_unlock(&head->mutex);
4387 struct btrfs_block_group *cache;
4389 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4392 spin_lock(&cache->space_info->lock);
4393 spin_lock(&cache->lock);
4394 cache->pinned += head->num_bytes;
4395 btrfs_space_info_update_bytes_pinned(fs_info,
4396 cache->space_info, head->num_bytes);
4397 cache->reserved -= head->num_bytes;
4398 cache->space_info->bytes_reserved -= head->num_bytes;
4399 spin_unlock(&cache->lock);
4400 spin_unlock(&cache->space_info->lock);
4401 percpu_counter_add_batch(
4402 &cache->space_info->total_bytes_pinned,
4403 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4405 btrfs_put_block_group(cache);
4407 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4408 head->bytenr + head->num_bytes - 1);
4410 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4411 btrfs_put_delayed_ref_head(head);
4413 spin_lock(&delayed_refs->lock);
4415 btrfs_qgroup_destroy_extent_records(trans);
4417 spin_unlock(&delayed_refs->lock);
4422 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4424 struct btrfs_inode *btrfs_inode;
4425 struct list_head splice;
4427 INIT_LIST_HEAD(&splice);
4429 spin_lock(&root->delalloc_lock);
4430 list_splice_init(&root->delalloc_inodes, &splice);
4432 while (!list_empty(&splice)) {
4433 struct inode *inode = NULL;
4434 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4436 __btrfs_del_delalloc_inode(root, btrfs_inode);
4437 spin_unlock(&root->delalloc_lock);
4440 * Make sure we get a live inode and that it'll not disappear
4443 inode = igrab(&btrfs_inode->vfs_inode);
4445 invalidate_inode_pages2(inode->i_mapping);
4448 spin_lock(&root->delalloc_lock);
4450 spin_unlock(&root->delalloc_lock);
4453 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4455 struct btrfs_root *root;
4456 struct list_head splice;
4458 INIT_LIST_HEAD(&splice);
4460 spin_lock(&fs_info->delalloc_root_lock);
4461 list_splice_init(&fs_info->delalloc_roots, &splice);
4462 while (!list_empty(&splice)) {
4463 root = list_first_entry(&splice, struct btrfs_root,
4465 root = btrfs_grab_root(root);
4467 spin_unlock(&fs_info->delalloc_root_lock);
4469 btrfs_destroy_delalloc_inodes(root);
4470 btrfs_put_root(root);
4472 spin_lock(&fs_info->delalloc_root_lock);
4474 spin_unlock(&fs_info->delalloc_root_lock);
4477 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4478 struct extent_io_tree *dirty_pages,
4482 struct extent_buffer *eb;
4487 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4492 clear_extent_bits(dirty_pages, start, end, mark);
4493 while (start <= end) {
4494 eb = find_extent_buffer(fs_info, start);
4495 start += fs_info->nodesize;
4498 wait_on_extent_buffer_writeback(eb);
4500 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4502 clear_extent_buffer_dirty(eb);
4503 free_extent_buffer_stale(eb);
4510 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4511 struct extent_io_tree *unpin)
4518 struct extent_state *cached_state = NULL;
4521 * The btrfs_finish_extent_commit() may get the same range as
4522 * ours between find_first_extent_bit and clear_extent_dirty.
4523 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4524 * the same extent range.
4526 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4527 ret = find_first_extent_bit(unpin, 0, &start, &end,
4528 EXTENT_DIRTY, &cached_state);
4530 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4534 clear_extent_dirty(unpin, start, end, &cached_state);
4535 free_extent_state(cached_state);
4536 btrfs_error_unpin_extent_range(fs_info, start, end);
4537 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4544 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4546 struct inode *inode;
4548 inode = cache->io_ctl.inode;
4550 invalidate_inode_pages2(inode->i_mapping);
4551 BTRFS_I(inode)->generation = 0;
4552 cache->io_ctl.inode = NULL;
4555 ASSERT(cache->io_ctl.pages == NULL);
4556 btrfs_put_block_group(cache);
4559 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4560 struct btrfs_fs_info *fs_info)
4562 struct btrfs_block_group *cache;
4564 spin_lock(&cur_trans->dirty_bgs_lock);
4565 while (!list_empty(&cur_trans->dirty_bgs)) {
4566 cache = list_first_entry(&cur_trans->dirty_bgs,
4567 struct btrfs_block_group,
4570 if (!list_empty(&cache->io_list)) {
4571 spin_unlock(&cur_trans->dirty_bgs_lock);
4572 list_del_init(&cache->io_list);
4573 btrfs_cleanup_bg_io(cache);
4574 spin_lock(&cur_trans->dirty_bgs_lock);
4577 list_del_init(&cache->dirty_list);
4578 spin_lock(&cache->lock);
4579 cache->disk_cache_state = BTRFS_DC_ERROR;
4580 spin_unlock(&cache->lock);
4582 spin_unlock(&cur_trans->dirty_bgs_lock);
4583 btrfs_put_block_group(cache);
4584 btrfs_delayed_refs_rsv_release(fs_info, 1);
4585 spin_lock(&cur_trans->dirty_bgs_lock);
4587 spin_unlock(&cur_trans->dirty_bgs_lock);
4590 * Refer to the definition of io_bgs member for details why it's safe
4591 * to use it without any locking
4593 while (!list_empty(&cur_trans->io_bgs)) {
4594 cache = list_first_entry(&cur_trans->io_bgs,
4595 struct btrfs_block_group,
4598 list_del_init(&cache->io_list);
4599 spin_lock(&cache->lock);
4600 cache->disk_cache_state = BTRFS_DC_ERROR;
4601 spin_unlock(&cache->lock);
4602 btrfs_cleanup_bg_io(cache);
4606 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4607 struct btrfs_fs_info *fs_info)
4609 struct btrfs_device *dev, *tmp;
4611 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4612 ASSERT(list_empty(&cur_trans->dirty_bgs));
4613 ASSERT(list_empty(&cur_trans->io_bgs));
4615 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4617 list_del_init(&dev->post_commit_list);
4620 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4622 cur_trans->state = TRANS_STATE_COMMIT_START;
4623 wake_up(&fs_info->transaction_blocked_wait);
4625 cur_trans->state = TRANS_STATE_UNBLOCKED;
4626 wake_up(&fs_info->transaction_wait);
4628 btrfs_destroy_delayed_inodes(fs_info);
4630 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4632 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4634 cur_trans->state =TRANS_STATE_COMPLETED;
4635 wake_up(&cur_trans->commit_wait);
4638 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4640 struct btrfs_transaction *t;
4642 mutex_lock(&fs_info->transaction_kthread_mutex);
4644 spin_lock(&fs_info->trans_lock);
4645 while (!list_empty(&fs_info->trans_list)) {
4646 t = list_first_entry(&fs_info->trans_list,
4647 struct btrfs_transaction, list);
4648 if (t->state >= TRANS_STATE_COMMIT_START) {
4649 refcount_inc(&t->use_count);
4650 spin_unlock(&fs_info->trans_lock);
4651 btrfs_wait_for_commit(fs_info, t->transid);
4652 btrfs_put_transaction(t);
4653 spin_lock(&fs_info->trans_lock);
4656 if (t == fs_info->running_transaction) {
4657 t->state = TRANS_STATE_COMMIT_DOING;
4658 spin_unlock(&fs_info->trans_lock);
4660 * We wait for 0 num_writers since we don't hold a trans
4661 * handle open currently for this transaction.
4663 wait_event(t->writer_wait,
4664 atomic_read(&t->num_writers) == 0);
4666 spin_unlock(&fs_info->trans_lock);
4668 btrfs_cleanup_one_transaction(t, fs_info);
4670 spin_lock(&fs_info->trans_lock);
4671 if (t == fs_info->running_transaction)
4672 fs_info->running_transaction = NULL;
4673 list_del_init(&t->list);
4674 spin_unlock(&fs_info->trans_lock);
4676 btrfs_put_transaction(t);
4677 trace_btrfs_transaction_commit(fs_info->tree_root);
4678 spin_lock(&fs_info->trans_lock);
4680 spin_unlock(&fs_info->trans_lock);
4681 btrfs_destroy_all_ordered_extents(fs_info);
4682 btrfs_destroy_delayed_inodes(fs_info);
4683 btrfs_assert_delayed_root_empty(fs_info);
4684 btrfs_destroy_all_delalloc_inodes(fs_info);
4685 btrfs_drop_all_logs(fs_info);
4686 mutex_unlock(&fs_info->transaction_kthread_mutex);
4691 static const struct extent_io_ops btree_extent_io_ops = {
4692 /* mandatory callbacks */
4693 .submit_bio_hook = btree_submit_bio_hook,
4694 .readpage_end_io_hook = btree_readpage_end_io_hook,