1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/scatterlist.h>
9 #include <linux/swap.h>
10 #include <linux/radix-tree.h>
11 #include <linux/writeback.h>
12 #include <linux/buffer_head.h>
13 #include <linux/workqueue.h>
14 #include <linux/kthread.h>
15 #include <linux/slab.h>
16 #include <linux/migrate.h>
17 #include <linux/ratelimit.h>
18 #include <linux/uuid.h>
19 #include <linux/semaphore.h>
20 #include <linux/error-injection.h>
21 #include <linux/crc32c.h>
22 #include <asm/unaligned.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
45 #include <asm/cpufeature.h>
48 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
49 BTRFS_HEADER_FLAG_RELOC |\
50 BTRFS_SUPER_FLAG_ERROR |\
51 BTRFS_SUPER_FLAG_SEEDING |\
52 BTRFS_SUPER_FLAG_METADUMP |\
53 BTRFS_SUPER_FLAG_METADUMP_V2)
55 static const struct extent_io_ops btree_extent_io_ops;
56 static void end_workqueue_fn(struct btrfs_work *work);
57 static void free_fs_root(struct btrfs_root *root);
58 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
59 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
60 struct btrfs_fs_info *fs_info);
61 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
62 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *dirty_pages,
65 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
66 struct extent_io_tree *pinned_extents);
67 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
68 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
71 * btrfs_end_io_wq structs are used to do processing in task context when an IO
72 * is complete. This is used during reads to verify checksums, and it is used
73 * by writes to insert metadata for new file extents after IO is complete.
75 struct btrfs_end_io_wq {
79 struct btrfs_fs_info *info;
81 enum btrfs_wq_endio_type metadata;
82 struct btrfs_work work;
85 static struct kmem_cache *btrfs_end_io_wq_cache;
87 int __init btrfs_end_io_wq_init(void)
89 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
90 sizeof(struct btrfs_end_io_wq),
94 if (!btrfs_end_io_wq_cache)
99 void __cold btrfs_end_io_wq_exit(void)
101 kmem_cache_destroy(btrfs_end_io_wq_cache);
105 * async submit bios are used to offload expensive checksumming
106 * onto the worker threads. They checksum file and metadata bios
107 * just before they are sent down the IO stack.
109 struct async_submit_bio {
111 struct btrfs_fs_info *fs_info;
113 extent_submit_bio_start_t *submit_bio_start;
114 extent_submit_bio_done_t *submit_bio_done;
116 unsigned long bio_flags;
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->objectid. This ensures that all special purpose roots
135 * 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, u64 start, u64 len,
215 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
216 struct extent_map_tree *em_tree = &inode->extent_tree;
217 struct extent_map *em;
220 read_lock(&em_tree->lock);
221 em = lookup_extent_mapping(em_tree, start, len);
223 em->bdev = fs_info->fs_devices->latest_bdev;
224 read_unlock(&em_tree->lock);
227 read_unlock(&em_tree->lock);
229 em = alloc_extent_map();
231 em = ERR_PTR(-ENOMEM);
236 em->block_len = (u64)-1;
238 em->bdev = fs_info->fs_devices->latest_bdev;
240 write_lock(&em_tree->lock);
241 ret = add_extent_mapping(em_tree, em, 0);
242 if (ret == -EEXIST) {
244 em = lookup_extent_mapping(em_tree, start, len);
251 write_unlock(&em_tree->lock);
257 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
259 return crc32c(seed, data, len);
262 void btrfs_csum_final(u32 crc, u8 *result)
264 put_unaligned_le32(~crc, result);
268 * compute the csum for a btree block, and either verify it or write it
269 * into the csum field of the block.
271 static int csum_tree_block(struct btrfs_fs_info *fs_info,
272 struct extent_buffer *buf,
275 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
276 char result[BTRFS_CSUM_SIZE];
278 unsigned long cur_len;
279 unsigned long offset = BTRFS_CSUM_SIZE;
281 unsigned long map_start;
282 unsigned long map_len;
286 len = buf->len - offset;
288 err = map_private_extent_buffer(buf, offset, 32,
289 &kaddr, &map_start, &map_len);
292 cur_len = min(len, map_len - (offset - map_start));
293 crc = btrfs_csum_data(kaddr + offset - map_start,
298 memset(result, 0, BTRFS_CSUM_SIZE);
300 btrfs_csum_final(crc, result);
303 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
306 memcpy(&found, result, csum_size);
308 read_extent_buffer(buf, &val, 0, csum_size);
309 btrfs_warn_rl(fs_info,
310 "%s checksum verify failed on %llu wanted %X found %X level %d",
311 fs_info->sb->s_id, buf->start,
312 val, found, btrfs_header_level(buf));
316 write_extent_buffer(buf, result, 0, csum_size);
323 * we can't consider a given block up to date unless the transid of the
324 * block matches the transid in the parent node's pointer. This is how we
325 * detect blocks that either didn't get written at all or got written
326 * in the wrong place.
328 static int verify_parent_transid(struct extent_io_tree *io_tree,
329 struct extent_buffer *eb, u64 parent_transid,
332 struct extent_state *cached_state = NULL;
334 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
336 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
343 btrfs_tree_read_lock(eb);
344 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
347 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
349 if (extent_buffer_uptodate(eb) &&
350 btrfs_header_generation(eb) == parent_transid) {
354 btrfs_err_rl(eb->fs_info,
355 "parent transid verify failed on %llu wanted %llu found %llu",
357 parent_transid, btrfs_header_generation(eb));
361 * Things reading via commit roots that don't have normal protection,
362 * like send, can have a really old block in cache that may point at a
363 * block that has been freed and re-allocated. So don't clear uptodate
364 * if we find an eb that is under IO (dirty/writeback) because we could
365 * end up reading in the stale data and then writing it back out and
366 * making everybody very sad.
368 if (!extent_buffer_under_io(eb))
369 clear_extent_buffer_uptodate(eb);
371 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
374 btrfs_tree_read_unlock_blocking(eb);
379 * Return 0 if the superblock checksum type matches the checksum value of that
380 * algorithm. Pass the raw disk superblock data.
382 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
385 struct btrfs_super_block *disk_sb =
386 (struct btrfs_super_block *)raw_disk_sb;
387 u16 csum_type = btrfs_super_csum_type(disk_sb);
390 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
392 char result[sizeof(crc)];
395 * The super_block structure does not span the whole
396 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
397 * is filled with zeros and is included in the checksum.
399 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
400 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
401 btrfs_csum_final(crc, result);
403 if (memcmp(raw_disk_sb, result, sizeof(result)))
407 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
408 btrfs_err(fs_info, "unsupported checksum algorithm %u",
416 static int verify_level_key(struct btrfs_fs_info *fs_info,
417 struct extent_buffer *eb, int level,
418 struct btrfs_key *first_key, u64 parent_transid)
421 struct btrfs_key found_key;
424 found_level = btrfs_header_level(eb);
425 if (found_level != level) {
426 #ifdef CONFIG_BTRFS_DEBUG
429 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
430 eb->start, level, found_level);
439 * For live tree block (new tree blocks in current transaction),
440 * we need proper lock context to avoid race, which is impossible here.
441 * So we only checks tree blocks which is read from disk, whose
442 * generation <= fs_info->last_trans_committed.
444 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
447 btrfs_node_key_to_cpu(eb, &found_key, 0);
449 btrfs_item_key_to_cpu(eb, &found_key, 0);
450 ret = btrfs_comp_cpu_keys(first_key, &found_key);
452 #ifdef CONFIG_BTRFS_DEBUG
456 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
457 eb->start, parent_transid, first_key->objectid,
458 first_key->type, first_key->offset,
459 found_key.objectid, found_key.type,
467 * helper to read a given tree block, doing retries as required when
468 * the checksums don't match and we have alternate mirrors to try.
470 * @parent_transid: expected transid, skip check if 0
471 * @level: expected level, mandatory check
472 * @first_key: expected key of first slot, skip check if NULL
474 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
475 struct extent_buffer *eb,
476 u64 parent_transid, int level,
477 struct btrfs_key *first_key)
479 struct extent_io_tree *io_tree;
484 int failed_mirror = 0;
486 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
487 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
489 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
492 if (verify_parent_transid(io_tree, eb,
495 else if (verify_level_key(fs_info, eb, level,
496 first_key, parent_transid))
503 * This buffer's crc is fine, but its contents are corrupted, so
504 * there is no reason to read the other copies, they won't be
507 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
511 num_copies = btrfs_num_copies(fs_info,
516 if (!failed_mirror) {
518 failed_mirror = eb->read_mirror;
522 if (mirror_num == failed_mirror)
525 if (mirror_num > num_copies)
529 if (failed && !ret && failed_mirror)
530 repair_eb_io_failure(fs_info, eb, failed_mirror);
536 * checksum a dirty tree block before IO. This has extra checks to make sure
537 * we only fill in the checksum field in the first page of a multi-page block
540 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
542 u64 start = page_offset(page);
544 struct extent_buffer *eb;
546 eb = (struct extent_buffer *)page->private;
547 if (page != eb->pages[0])
550 found_start = btrfs_header_bytenr(eb);
552 * Please do not consolidate these warnings into a single if.
553 * It is useful to know what went wrong.
555 if (WARN_ON(found_start != start))
557 if (WARN_ON(!PageUptodate(page)))
560 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
561 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
563 return csum_tree_block(fs_info, eb, 0);
566 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
567 struct extent_buffer *eb)
569 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
570 u8 fsid[BTRFS_FSID_SIZE];
573 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
575 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
579 fs_devices = fs_devices->seed;
584 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
585 u64 phy_offset, struct page *page,
586 u64 start, u64 end, int mirror)
590 struct extent_buffer *eb;
591 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
592 struct btrfs_fs_info *fs_info = root->fs_info;
599 eb = (struct extent_buffer *)page->private;
601 /* the pending IO might have been the only thing that kept this buffer
602 * in memory. Make sure we have a ref for all this other checks
604 extent_buffer_get(eb);
606 reads_done = atomic_dec_and_test(&eb->io_pages);
610 eb->read_mirror = mirror;
611 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
616 found_start = btrfs_header_bytenr(eb);
617 if (found_start != eb->start) {
618 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
619 eb->start, found_start);
623 if (check_tree_block_fsid(fs_info, eb)) {
624 btrfs_err_rl(fs_info, "bad fsid on block %llu",
629 found_level = btrfs_header_level(eb);
630 if (found_level >= BTRFS_MAX_LEVEL) {
631 btrfs_err(fs_info, "bad tree block level %d on %llu",
632 (int)btrfs_header_level(eb), eb->start);
637 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
640 ret = csum_tree_block(fs_info, eb, 1);
645 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 * that we don't try and read the other copies of this block, just
649 if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
650 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
654 if (found_level > 0 && btrfs_check_node(fs_info, eb))
658 set_extent_buffer_uptodate(eb);
661 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 btree_readahead_hook(eb, ret);
666 * our io error hook is going to dec the io pages
667 * again, we have to make sure it has something
670 atomic_inc(&eb->io_pages);
671 clear_extent_buffer_uptodate(eb);
673 free_extent_buffer(eb);
678 static int btree_io_failed_hook(struct page *page, int failed_mirror)
680 struct extent_buffer *eb;
682 eb = (struct extent_buffer *)page->private;
683 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
684 eb->read_mirror = failed_mirror;
685 atomic_dec(&eb->io_pages);
686 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
687 btree_readahead_hook(eb, -EIO);
688 return -EIO; /* we fixed nothing */
691 static void end_workqueue_bio(struct bio *bio)
693 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
694 struct btrfs_fs_info *fs_info;
695 struct btrfs_workqueue *wq;
696 btrfs_work_func_t func;
698 fs_info = end_io_wq->info;
699 end_io_wq->status = bio->bi_status;
701 if (bio_op(bio) == REQ_OP_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
703 wq = fs_info->endio_meta_write_workers;
704 func = btrfs_endio_meta_write_helper;
705 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
706 wq = fs_info->endio_freespace_worker;
707 func = btrfs_freespace_write_helper;
708 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
709 wq = fs_info->endio_raid56_workers;
710 func = btrfs_endio_raid56_helper;
712 wq = fs_info->endio_write_workers;
713 func = btrfs_endio_write_helper;
716 if (unlikely(end_io_wq->metadata ==
717 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
718 wq = fs_info->endio_repair_workers;
719 func = btrfs_endio_repair_helper;
720 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
721 wq = fs_info->endio_raid56_workers;
722 func = btrfs_endio_raid56_helper;
723 } else if (end_io_wq->metadata) {
724 wq = fs_info->endio_meta_workers;
725 func = btrfs_endio_meta_helper;
727 wq = fs_info->endio_workers;
728 func = btrfs_endio_helper;
732 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
733 btrfs_queue_work(wq, &end_io_wq->work);
736 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
737 enum btrfs_wq_endio_type metadata)
739 struct btrfs_end_io_wq *end_io_wq;
741 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
743 return BLK_STS_RESOURCE;
745 end_io_wq->private = bio->bi_private;
746 end_io_wq->end_io = bio->bi_end_io;
747 end_io_wq->info = info;
748 end_io_wq->status = 0;
749 end_io_wq->bio = bio;
750 end_io_wq->metadata = metadata;
752 bio->bi_private = end_io_wq;
753 bio->bi_end_io = end_workqueue_bio;
757 static void run_one_async_start(struct btrfs_work *work)
759 struct async_submit_bio *async;
762 async = container_of(work, struct async_submit_bio, work);
763 ret = async->submit_bio_start(async->private_data, async->bio,
769 static void run_one_async_done(struct btrfs_work *work)
771 struct async_submit_bio *async;
773 async = container_of(work, struct async_submit_bio, work);
775 /* If an error occurred we just want to clean up the bio and move on */
777 async->bio->bi_status = async->status;
778 bio_endio(async->bio);
782 async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
785 static void run_one_async_free(struct btrfs_work *work)
787 struct async_submit_bio *async;
789 async = container_of(work, struct async_submit_bio, work);
793 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
794 int mirror_num, unsigned long bio_flags,
795 u64 bio_offset, void *private_data,
796 extent_submit_bio_start_t *submit_bio_start,
797 extent_submit_bio_done_t *submit_bio_done)
799 struct async_submit_bio *async;
801 async = kmalloc(sizeof(*async), GFP_NOFS);
803 return BLK_STS_RESOURCE;
805 async->private_data = private_data;
806 async->fs_info = fs_info;
808 async->mirror_num = mirror_num;
809 async->submit_bio_start = submit_bio_start;
810 async->submit_bio_done = submit_bio_done;
812 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
813 run_one_async_done, run_one_async_free);
815 async->bio_flags = bio_flags;
816 async->bio_offset = bio_offset;
820 if (op_is_sync(bio->bi_opf))
821 btrfs_set_work_high_priority(&async->work);
823 btrfs_queue_work(fs_info->workers, &async->work);
827 static blk_status_t btree_csum_one_bio(struct bio *bio)
829 struct bio_vec *bvec;
830 struct btrfs_root *root;
833 ASSERT(!bio_flagged(bio, BIO_CLONED));
834 bio_for_each_segment_all(bvec, bio, i) {
835 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
836 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
841 return errno_to_blk_status(ret);
844 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
848 * when we're called for a write, we're already in the async
849 * submission context. Just jump into btrfs_map_bio
851 return btree_csum_one_bio(bio);
854 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
857 struct inode *inode = private_data;
861 * when we're called for a write, we're already in the async
862 * submission context. Just jump into btrfs_map_bio
864 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
866 bio->bi_status = ret;
872 static int check_async_write(struct btrfs_inode *bi)
874 if (atomic_read(&bi->sync_writers))
877 if (static_cpu_has(X86_FEATURE_XMM4_2))
883 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
884 int mirror_num, unsigned long bio_flags,
887 struct inode *inode = private_data;
888 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
889 int async = check_async_write(BTRFS_I(inode));
892 if (bio_op(bio) != REQ_OP_WRITE) {
894 * called for a read, do the setup so that checksum validation
895 * can happen in the async kernel threads
897 ret = btrfs_bio_wq_end_io(fs_info, bio,
898 BTRFS_WQ_ENDIO_METADATA);
901 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
903 ret = btree_csum_one_bio(bio);
906 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
909 * kthread helpers are used to submit writes so that
910 * checksumming can happen in parallel across all CPUs
912 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
913 bio_offset, private_data,
914 btree_submit_bio_start,
915 btree_submit_bio_done);
923 bio->bi_status = ret;
928 #ifdef CONFIG_MIGRATION
929 static int btree_migratepage(struct address_space *mapping,
930 struct page *newpage, struct page *page,
931 enum migrate_mode mode)
934 * we can't safely write a btree page from here,
935 * we haven't done the locking hook
940 * Buffers may be managed in a filesystem specific way.
941 * We must have no buffers or drop them.
943 if (page_has_private(page) &&
944 !try_to_release_page(page, GFP_KERNEL))
946 return migrate_page(mapping, newpage, page, mode);
951 static int btree_writepages(struct address_space *mapping,
952 struct writeback_control *wbc)
954 struct btrfs_fs_info *fs_info;
957 if (wbc->sync_mode == WB_SYNC_NONE) {
959 if (wbc->for_kupdate)
962 fs_info = BTRFS_I(mapping->host)->root->fs_info;
963 /* this is a bit racy, but that's ok */
964 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
965 BTRFS_DIRTY_METADATA_THRESH);
969 return btree_write_cache_pages(mapping, wbc);
972 static int btree_readpage(struct file *file, struct page *page)
974 struct extent_io_tree *tree;
975 tree = &BTRFS_I(page->mapping->host)->io_tree;
976 return extent_read_full_page(tree, page, btree_get_extent, 0);
979 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
981 if (PageWriteback(page) || PageDirty(page))
984 return try_release_extent_buffer(page);
987 static void btree_invalidatepage(struct page *page, unsigned int offset,
990 struct extent_io_tree *tree;
991 tree = &BTRFS_I(page->mapping->host)->io_tree;
992 extent_invalidatepage(tree, page, offset);
993 btree_releasepage(page, GFP_NOFS);
994 if (PagePrivate(page)) {
995 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
996 "page private not zero on page %llu",
997 (unsigned long long)page_offset(page));
998 ClearPagePrivate(page);
999 set_page_private(page, 0);
1004 static int btree_set_page_dirty(struct page *page)
1007 struct extent_buffer *eb;
1009 BUG_ON(!PagePrivate(page));
1010 eb = (struct extent_buffer *)page->private;
1012 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1013 BUG_ON(!atomic_read(&eb->refs));
1014 btrfs_assert_tree_locked(eb);
1016 return __set_page_dirty_nobuffers(page);
1019 static const struct address_space_operations btree_aops = {
1020 .readpage = btree_readpage,
1021 .writepages = btree_writepages,
1022 .releasepage = btree_releasepage,
1023 .invalidatepage = btree_invalidatepage,
1024 #ifdef CONFIG_MIGRATION
1025 .migratepage = btree_migratepage,
1027 .set_page_dirty = btree_set_page_dirty,
1030 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1032 struct extent_buffer *buf = NULL;
1033 struct inode *btree_inode = fs_info->btree_inode;
1035 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1038 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1040 free_extent_buffer(buf);
1043 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1044 int mirror_num, struct extent_buffer **eb)
1046 struct extent_buffer *buf = NULL;
1047 struct inode *btree_inode = fs_info->btree_inode;
1048 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1051 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1055 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1057 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1060 free_extent_buffer(buf);
1064 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1065 free_extent_buffer(buf);
1067 } else if (extent_buffer_uptodate(buf)) {
1070 free_extent_buffer(buf);
1075 struct extent_buffer *btrfs_find_create_tree_block(
1076 struct btrfs_fs_info *fs_info,
1079 if (btrfs_is_testing(fs_info))
1080 return alloc_test_extent_buffer(fs_info, bytenr);
1081 return alloc_extent_buffer(fs_info, bytenr);
1085 int btrfs_write_tree_block(struct extent_buffer *buf)
1087 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1088 buf->start + buf->len - 1);
1091 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1093 filemap_fdatawait_range(buf->pages[0]->mapping,
1094 buf->start, buf->start + buf->len - 1);
1098 * Read tree block at logical address @bytenr and do variant basic but critical
1101 * @parent_transid: expected transid of this tree block, skip check if 0
1102 * @level: expected level, mandatory check
1103 * @first_key: expected key in slot 0, skip check if NULL
1105 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1106 u64 parent_transid, int level,
1107 struct btrfs_key *first_key)
1109 struct extent_buffer *buf = NULL;
1112 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1116 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1119 free_extent_buffer(buf);
1120 return ERR_PTR(ret);
1126 void clean_tree_block(struct btrfs_fs_info *fs_info,
1127 struct extent_buffer *buf)
1129 if (btrfs_header_generation(buf) ==
1130 fs_info->running_transaction->transid) {
1131 btrfs_assert_tree_locked(buf);
1133 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1134 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1136 fs_info->dirty_metadata_batch);
1137 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1138 btrfs_set_lock_blocking(buf);
1139 clear_extent_buffer_dirty(buf);
1144 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1146 struct btrfs_subvolume_writers *writers;
1149 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1151 return ERR_PTR(-ENOMEM);
1153 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1156 return ERR_PTR(ret);
1159 init_waitqueue_head(&writers->wait);
1164 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1166 percpu_counter_destroy(&writers->counter);
1170 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1173 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1175 root->commit_root = NULL;
1177 root->orphan_cleanup_state = 0;
1179 root->objectid = objectid;
1180 root->last_trans = 0;
1181 root->highest_objectid = 0;
1182 root->nr_delalloc_inodes = 0;
1183 root->nr_ordered_extents = 0;
1185 root->inode_tree = RB_ROOT;
1186 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1187 root->block_rsv = NULL;
1189 INIT_LIST_HEAD(&root->dirty_list);
1190 INIT_LIST_HEAD(&root->root_list);
1191 INIT_LIST_HEAD(&root->delalloc_inodes);
1192 INIT_LIST_HEAD(&root->delalloc_root);
1193 INIT_LIST_HEAD(&root->ordered_extents);
1194 INIT_LIST_HEAD(&root->ordered_root);
1195 INIT_LIST_HEAD(&root->logged_list[0]);
1196 INIT_LIST_HEAD(&root->logged_list[1]);
1197 spin_lock_init(&root->inode_lock);
1198 spin_lock_init(&root->delalloc_lock);
1199 spin_lock_init(&root->ordered_extent_lock);
1200 spin_lock_init(&root->accounting_lock);
1201 spin_lock_init(&root->log_extents_lock[0]);
1202 spin_lock_init(&root->log_extents_lock[1]);
1203 spin_lock_init(&root->qgroup_meta_rsv_lock);
1204 mutex_init(&root->objectid_mutex);
1205 mutex_init(&root->log_mutex);
1206 mutex_init(&root->ordered_extent_mutex);
1207 mutex_init(&root->delalloc_mutex);
1208 init_waitqueue_head(&root->log_writer_wait);
1209 init_waitqueue_head(&root->log_commit_wait[0]);
1210 init_waitqueue_head(&root->log_commit_wait[1]);
1211 INIT_LIST_HEAD(&root->log_ctxs[0]);
1212 INIT_LIST_HEAD(&root->log_ctxs[1]);
1213 atomic_set(&root->log_commit[0], 0);
1214 atomic_set(&root->log_commit[1], 0);
1215 atomic_set(&root->log_writers, 0);
1216 atomic_set(&root->log_batch, 0);
1217 refcount_set(&root->refs, 1);
1218 atomic_set(&root->will_be_snapshotted, 0);
1219 root->log_transid = 0;
1220 root->log_transid_committed = -1;
1221 root->last_log_commit = 0;
1223 extent_io_tree_init(&root->dirty_log_pages, NULL);
1225 memset(&root->root_key, 0, sizeof(root->root_key));
1226 memset(&root->root_item, 0, sizeof(root->root_item));
1227 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1229 root->defrag_trans_start = fs_info->generation;
1231 root->defrag_trans_start = 0;
1232 root->root_key.objectid = objectid;
1235 spin_lock_init(&root->root_item_lock);
1238 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1241 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1243 root->fs_info = fs_info;
1247 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1248 /* Should only be used by the testing infrastructure */
1249 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1251 struct btrfs_root *root;
1254 return ERR_PTR(-EINVAL);
1256 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1258 return ERR_PTR(-ENOMEM);
1260 /* We don't use the stripesize in selftest, set it as sectorsize */
1261 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1262 root->alloc_bytenr = 0;
1268 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1269 struct btrfs_fs_info *fs_info,
1272 struct extent_buffer *leaf;
1273 struct btrfs_root *tree_root = fs_info->tree_root;
1274 struct btrfs_root *root;
1275 struct btrfs_key key;
1277 uuid_le uuid = NULL_UUID_LE;
1279 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1281 return ERR_PTR(-ENOMEM);
1283 __setup_root(root, fs_info, objectid);
1284 root->root_key.objectid = objectid;
1285 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1286 root->root_key.offset = 0;
1288 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1290 ret = PTR_ERR(leaf);
1296 btrfs_mark_buffer_dirty(leaf);
1298 root->commit_root = btrfs_root_node(root);
1299 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1301 root->root_item.flags = 0;
1302 root->root_item.byte_limit = 0;
1303 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1304 btrfs_set_root_generation(&root->root_item, trans->transid);
1305 btrfs_set_root_level(&root->root_item, 0);
1306 btrfs_set_root_refs(&root->root_item, 1);
1307 btrfs_set_root_used(&root->root_item, leaf->len);
1308 btrfs_set_root_last_snapshot(&root->root_item, 0);
1309 btrfs_set_root_dirid(&root->root_item, 0);
1310 if (is_fstree(objectid))
1312 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1313 root->root_item.drop_level = 0;
1315 key.objectid = objectid;
1316 key.type = BTRFS_ROOT_ITEM_KEY;
1318 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1322 btrfs_tree_unlock(leaf);
1328 btrfs_tree_unlock(leaf);
1329 free_extent_buffer(root->commit_root);
1330 free_extent_buffer(leaf);
1334 return ERR_PTR(ret);
1337 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1338 struct btrfs_fs_info *fs_info)
1340 struct btrfs_root *root;
1341 struct extent_buffer *leaf;
1343 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1345 return ERR_PTR(-ENOMEM);
1347 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1349 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1350 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1351 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1354 * DON'T set REF_COWS for log trees
1356 * log trees do not get reference counted because they go away
1357 * before a real commit is actually done. They do store pointers
1358 * to file data extents, and those reference counts still get
1359 * updated (along with back refs to the log tree).
1362 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1366 return ERR_CAST(leaf);
1371 btrfs_mark_buffer_dirty(root->node);
1372 btrfs_tree_unlock(root->node);
1376 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1377 struct btrfs_fs_info *fs_info)
1379 struct btrfs_root *log_root;
1381 log_root = alloc_log_tree(trans, fs_info);
1382 if (IS_ERR(log_root))
1383 return PTR_ERR(log_root);
1384 WARN_ON(fs_info->log_root_tree);
1385 fs_info->log_root_tree = log_root;
1389 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1390 struct btrfs_root *root)
1392 struct btrfs_fs_info *fs_info = root->fs_info;
1393 struct btrfs_root *log_root;
1394 struct btrfs_inode_item *inode_item;
1396 log_root = alloc_log_tree(trans, fs_info);
1397 if (IS_ERR(log_root))
1398 return PTR_ERR(log_root);
1400 log_root->last_trans = trans->transid;
1401 log_root->root_key.offset = root->root_key.objectid;
1403 inode_item = &log_root->root_item.inode;
1404 btrfs_set_stack_inode_generation(inode_item, 1);
1405 btrfs_set_stack_inode_size(inode_item, 3);
1406 btrfs_set_stack_inode_nlink(inode_item, 1);
1407 btrfs_set_stack_inode_nbytes(inode_item,
1409 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1411 btrfs_set_root_node(&log_root->root_item, log_root->node);
1413 WARN_ON(root->log_root);
1414 root->log_root = log_root;
1415 root->log_transid = 0;
1416 root->log_transid_committed = -1;
1417 root->last_log_commit = 0;
1421 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1422 struct btrfs_key *key)
1424 struct btrfs_root *root;
1425 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1426 struct btrfs_path *path;
1431 path = btrfs_alloc_path();
1433 return ERR_PTR(-ENOMEM);
1435 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1441 __setup_root(root, fs_info, key->objectid);
1443 ret = btrfs_find_root(tree_root, key, path,
1444 &root->root_item, &root->root_key);
1451 generation = btrfs_root_generation(&root->root_item);
1452 level = btrfs_root_level(&root->root_item);
1453 root->node = read_tree_block(fs_info,
1454 btrfs_root_bytenr(&root->root_item),
1455 generation, level, NULL);
1456 if (IS_ERR(root->node)) {
1457 ret = PTR_ERR(root->node);
1459 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1461 free_extent_buffer(root->node);
1464 root->commit_root = btrfs_root_node(root);
1466 btrfs_free_path(path);
1472 root = ERR_PTR(ret);
1476 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1477 struct btrfs_key *location)
1479 struct btrfs_root *root;
1481 root = btrfs_read_tree_root(tree_root, location);
1485 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1486 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1487 btrfs_check_and_init_root_item(&root->root_item);
1493 int btrfs_init_fs_root(struct btrfs_root *root)
1496 struct btrfs_subvolume_writers *writers;
1498 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1499 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1501 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1506 writers = btrfs_alloc_subvolume_writers();
1507 if (IS_ERR(writers)) {
1508 ret = PTR_ERR(writers);
1511 root->subv_writers = writers;
1513 btrfs_init_free_ino_ctl(root);
1514 spin_lock_init(&root->ino_cache_lock);
1515 init_waitqueue_head(&root->ino_cache_wait);
1517 ret = get_anon_bdev(&root->anon_dev);
1521 mutex_lock(&root->objectid_mutex);
1522 ret = btrfs_find_highest_objectid(root,
1523 &root->highest_objectid);
1525 mutex_unlock(&root->objectid_mutex);
1529 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1531 mutex_unlock(&root->objectid_mutex);
1535 /* the caller is responsible to call free_fs_root */
1539 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1542 struct btrfs_root *root;
1544 spin_lock(&fs_info->fs_roots_radix_lock);
1545 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1546 (unsigned long)root_id);
1547 spin_unlock(&fs_info->fs_roots_radix_lock);
1551 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1552 struct btrfs_root *root)
1556 ret = radix_tree_preload(GFP_NOFS);
1560 spin_lock(&fs_info->fs_roots_radix_lock);
1561 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1562 (unsigned long)root->root_key.objectid,
1565 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1566 spin_unlock(&fs_info->fs_roots_radix_lock);
1567 radix_tree_preload_end();
1572 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1573 struct btrfs_key *location,
1576 struct btrfs_root *root;
1577 struct btrfs_path *path;
1578 struct btrfs_key key;
1581 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1582 return fs_info->tree_root;
1583 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1584 return fs_info->extent_root;
1585 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1586 return fs_info->chunk_root;
1587 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1588 return fs_info->dev_root;
1589 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1590 return fs_info->csum_root;
1591 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1592 return fs_info->quota_root ? fs_info->quota_root :
1594 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1595 return fs_info->uuid_root ? fs_info->uuid_root :
1597 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1598 return fs_info->free_space_root ? fs_info->free_space_root :
1601 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1603 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1604 return ERR_PTR(-ENOENT);
1608 root = btrfs_read_fs_root(fs_info->tree_root, location);
1612 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1617 ret = btrfs_init_fs_root(root);
1621 path = btrfs_alloc_path();
1626 key.objectid = BTRFS_ORPHAN_OBJECTID;
1627 key.type = BTRFS_ORPHAN_ITEM_KEY;
1628 key.offset = location->objectid;
1630 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1631 btrfs_free_path(path);
1635 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1637 ret = btrfs_insert_fs_root(fs_info, root);
1639 if (ret == -EEXIST) {
1648 return ERR_PTR(ret);
1651 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1653 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1655 struct btrfs_device *device;
1656 struct backing_dev_info *bdi;
1659 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1662 bdi = device->bdev->bd_bdi;
1663 if (bdi_congested(bdi, bdi_bits)) {
1673 * called by the kthread helper functions to finally call the bio end_io
1674 * functions. This is where read checksum verification actually happens
1676 static void end_workqueue_fn(struct btrfs_work *work)
1679 struct btrfs_end_io_wq *end_io_wq;
1681 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1682 bio = end_io_wq->bio;
1684 bio->bi_status = end_io_wq->status;
1685 bio->bi_private = end_io_wq->private;
1686 bio->bi_end_io = end_io_wq->end_io;
1687 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1691 static int cleaner_kthread(void *arg)
1693 struct btrfs_root *root = arg;
1694 struct btrfs_fs_info *fs_info = root->fs_info;
1696 struct btrfs_trans_handle *trans;
1701 /* Make the cleaner go to sleep early. */
1702 if (btrfs_need_cleaner_sleep(fs_info))
1706 * Do not do anything if we might cause open_ctree() to block
1707 * before we have finished mounting the filesystem.
1709 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1712 if (!mutex_trylock(&fs_info->cleaner_mutex))
1716 * Avoid the problem that we change the status of the fs
1717 * during the above check and trylock.
1719 if (btrfs_need_cleaner_sleep(fs_info)) {
1720 mutex_unlock(&fs_info->cleaner_mutex);
1724 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1725 btrfs_run_delayed_iputs(fs_info);
1726 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1728 again = btrfs_clean_one_deleted_snapshot(root);
1729 mutex_unlock(&fs_info->cleaner_mutex);
1732 * The defragger has dealt with the R/O remount and umount,
1733 * needn't do anything special here.
1735 btrfs_run_defrag_inodes(fs_info);
1738 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1739 * with relocation (btrfs_relocate_chunk) and relocation
1740 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1741 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1742 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1743 * unused block groups.
1745 btrfs_delete_unused_bgs(fs_info);
1748 set_current_state(TASK_INTERRUPTIBLE);
1749 if (!kthread_should_stop())
1751 __set_current_state(TASK_RUNNING);
1753 } while (!kthread_should_stop());
1756 * Transaction kthread is stopped before us and wakes us up.
1757 * However we might have started a new transaction and COWed some
1758 * tree blocks when deleting unused block groups for example. So
1759 * make sure we commit the transaction we started to have a clean
1760 * shutdown when evicting the btree inode - if it has dirty pages
1761 * when we do the final iput() on it, eviction will trigger a
1762 * writeback for it which will fail with null pointer dereferences
1763 * since work queues and other resources were already released and
1764 * destroyed by the time the iput/eviction/writeback is made.
1766 trans = btrfs_attach_transaction(root);
1767 if (IS_ERR(trans)) {
1768 if (PTR_ERR(trans) != -ENOENT)
1770 "cleaner transaction attach returned %ld",
1775 ret = btrfs_commit_transaction(trans);
1778 "cleaner open transaction commit returned %d",
1785 static int transaction_kthread(void *arg)
1787 struct btrfs_root *root = arg;
1788 struct btrfs_fs_info *fs_info = root->fs_info;
1789 struct btrfs_trans_handle *trans;
1790 struct btrfs_transaction *cur;
1793 unsigned long delay;
1797 cannot_commit = false;
1798 delay = HZ * fs_info->commit_interval;
1799 mutex_lock(&fs_info->transaction_kthread_mutex);
1801 spin_lock(&fs_info->trans_lock);
1802 cur = fs_info->running_transaction;
1804 spin_unlock(&fs_info->trans_lock);
1808 now = ktime_get_seconds();
1809 if (cur->state < TRANS_STATE_BLOCKED &&
1810 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1811 (now < cur->start_time ||
1812 now - cur->start_time < fs_info->commit_interval)) {
1813 spin_unlock(&fs_info->trans_lock);
1817 transid = cur->transid;
1818 spin_unlock(&fs_info->trans_lock);
1820 /* If the file system is aborted, this will always fail. */
1821 trans = btrfs_attach_transaction(root);
1822 if (IS_ERR(trans)) {
1823 if (PTR_ERR(trans) != -ENOENT)
1824 cannot_commit = true;
1827 if (transid == trans->transid) {
1828 btrfs_commit_transaction(trans);
1830 btrfs_end_transaction(trans);
1833 wake_up_process(fs_info->cleaner_kthread);
1834 mutex_unlock(&fs_info->transaction_kthread_mutex);
1836 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1837 &fs_info->fs_state)))
1838 btrfs_cleanup_transaction(fs_info);
1839 if (!kthread_should_stop() &&
1840 (!btrfs_transaction_blocked(fs_info) ||
1842 schedule_timeout_interruptible(delay);
1843 } while (!kthread_should_stop());
1848 * this will find the highest generation in the array of
1849 * root backups. The index of the highest array is returned,
1850 * or -1 if we can't find anything.
1852 * We check to make sure the array is valid by comparing the
1853 * generation of the latest root in the array with the generation
1854 * in the super block. If they don't match we pitch it.
1856 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1859 int newest_index = -1;
1860 struct btrfs_root_backup *root_backup;
1863 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1864 root_backup = info->super_copy->super_roots + i;
1865 cur = btrfs_backup_tree_root_gen(root_backup);
1866 if (cur == newest_gen)
1870 /* check to see if we actually wrapped around */
1871 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1872 root_backup = info->super_copy->super_roots;
1873 cur = btrfs_backup_tree_root_gen(root_backup);
1874 if (cur == newest_gen)
1877 return newest_index;
1882 * find the oldest backup so we know where to store new entries
1883 * in the backup array. This will set the backup_root_index
1884 * field in the fs_info struct
1886 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1889 int newest_index = -1;
1891 newest_index = find_newest_super_backup(info, newest_gen);
1892 /* if there was garbage in there, just move along */
1893 if (newest_index == -1) {
1894 info->backup_root_index = 0;
1896 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1901 * copy all the root pointers into the super backup array.
1902 * this will bump the backup pointer by one when it is
1905 static void backup_super_roots(struct btrfs_fs_info *info)
1908 struct btrfs_root_backup *root_backup;
1911 next_backup = info->backup_root_index;
1912 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1913 BTRFS_NUM_BACKUP_ROOTS;
1916 * just overwrite the last backup if we're at the same generation
1917 * this happens only at umount
1919 root_backup = info->super_for_commit->super_roots + last_backup;
1920 if (btrfs_backup_tree_root_gen(root_backup) ==
1921 btrfs_header_generation(info->tree_root->node))
1922 next_backup = last_backup;
1924 root_backup = info->super_for_commit->super_roots + next_backup;
1927 * make sure all of our padding and empty slots get zero filled
1928 * regardless of which ones we use today
1930 memset(root_backup, 0, sizeof(*root_backup));
1932 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1934 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1935 btrfs_set_backup_tree_root_gen(root_backup,
1936 btrfs_header_generation(info->tree_root->node));
1938 btrfs_set_backup_tree_root_level(root_backup,
1939 btrfs_header_level(info->tree_root->node));
1941 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1942 btrfs_set_backup_chunk_root_gen(root_backup,
1943 btrfs_header_generation(info->chunk_root->node));
1944 btrfs_set_backup_chunk_root_level(root_backup,
1945 btrfs_header_level(info->chunk_root->node));
1947 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1948 btrfs_set_backup_extent_root_gen(root_backup,
1949 btrfs_header_generation(info->extent_root->node));
1950 btrfs_set_backup_extent_root_level(root_backup,
1951 btrfs_header_level(info->extent_root->node));
1954 * we might commit during log recovery, which happens before we set
1955 * the fs_root. Make sure it is valid before we fill it in.
1957 if (info->fs_root && info->fs_root->node) {
1958 btrfs_set_backup_fs_root(root_backup,
1959 info->fs_root->node->start);
1960 btrfs_set_backup_fs_root_gen(root_backup,
1961 btrfs_header_generation(info->fs_root->node));
1962 btrfs_set_backup_fs_root_level(root_backup,
1963 btrfs_header_level(info->fs_root->node));
1966 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1967 btrfs_set_backup_dev_root_gen(root_backup,
1968 btrfs_header_generation(info->dev_root->node));
1969 btrfs_set_backup_dev_root_level(root_backup,
1970 btrfs_header_level(info->dev_root->node));
1972 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1973 btrfs_set_backup_csum_root_gen(root_backup,
1974 btrfs_header_generation(info->csum_root->node));
1975 btrfs_set_backup_csum_root_level(root_backup,
1976 btrfs_header_level(info->csum_root->node));
1978 btrfs_set_backup_total_bytes(root_backup,
1979 btrfs_super_total_bytes(info->super_copy));
1980 btrfs_set_backup_bytes_used(root_backup,
1981 btrfs_super_bytes_used(info->super_copy));
1982 btrfs_set_backup_num_devices(root_backup,
1983 btrfs_super_num_devices(info->super_copy));
1986 * if we don't copy this out to the super_copy, it won't get remembered
1987 * for the next commit
1989 memcpy(&info->super_copy->super_roots,
1990 &info->super_for_commit->super_roots,
1991 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1995 * this copies info out of the root backup array and back into
1996 * the in-memory super block. It is meant to help iterate through
1997 * the array, so you send it the number of backups you've already
1998 * tried and the last backup index you used.
2000 * this returns -1 when it has tried all the backups
2002 static noinline int next_root_backup(struct btrfs_fs_info *info,
2003 struct btrfs_super_block *super,
2004 int *num_backups_tried, int *backup_index)
2006 struct btrfs_root_backup *root_backup;
2007 int newest = *backup_index;
2009 if (*num_backups_tried == 0) {
2010 u64 gen = btrfs_super_generation(super);
2012 newest = find_newest_super_backup(info, gen);
2016 *backup_index = newest;
2017 *num_backups_tried = 1;
2018 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2019 /* we've tried all the backups, all done */
2022 /* jump to the next oldest backup */
2023 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2024 BTRFS_NUM_BACKUP_ROOTS;
2025 *backup_index = newest;
2026 *num_backups_tried += 1;
2028 root_backup = super->super_roots + newest;
2030 btrfs_set_super_generation(super,
2031 btrfs_backup_tree_root_gen(root_backup));
2032 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2033 btrfs_set_super_root_level(super,
2034 btrfs_backup_tree_root_level(root_backup));
2035 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2038 * fixme: the total bytes and num_devices need to match or we should
2041 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2042 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2046 /* helper to cleanup workers */
2047 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2049 btrfs_destroy_workqueue(fs_info->fixup_workers);
2050 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2051 btrfs_destroy_workqueue(fs_info->workers);
2052 btrfs_destroy_workqueue(fs_info->endio_workers);
2053 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2054 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2055 btrfs_destroy_workqueue(fs_info->rmw_workers);
2056 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2057 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2058 btrfs_destroy_workqueue(fs_info->submit_workers);
2059 btrfs_destroy_workqueue(fs_info->delayed_workers);
2060 btrfs_destroy_workqueue(fs_info->caching_workers);
2061 btrfs_destroy_workqueue(fs_info->readahead_workers);
2062 btrfs_destroy_workqueue(fs_info->flush_workers);
2063 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2064 btrfs_destroy_workqueue(fs_info->extent_workers);
2066 * Now that all other work queues are destroyed, we can safely destroy
2067 * the queues used for metadata I/O, since tasks from those other work
2068 * queues can do metadata I/O operations.
2070 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2071 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2074 static void free_root_extent_buffers(struct btrfs_root *root)
2077 free_extent_buffer(root->node);
2078 free_extent_buffer(root->commit_root);
2080 root->commit_root = NULL;
2084 /* helper to cleanup tree roots */
2085 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2087 free_root_extent_buffers(info->tree_root);
2089 free_root_extent_buffers(info->dev_root);
2090 free_root_extent_buffers(info->extent_root);
2091 free_root_extent_buffers(info->csum_root);
2092 free_root_extent_buffers(info->quota_root);
2093 free_root_extent_buffers(info->uuid_root);
2095 free_root_extent_buffers(info->chunk_root);
2096 free_root_extent_buffers(info->free_space_root);
2099 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2102 struct btrfs_root *gang[8];
2105 while (!list_empty(&fs_info->dead_roots)) {
2106 gang[0] = list_entry(fs_info->dead_roots.next,
2107 struct btrfs_root, root_list);
2108 list_del(&gang[0]->root_list);
2110 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2111 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2113 free_extent_buffer(gang[0]->node);
2114 free_extent_buffer(gang[0]->commit_root);
2115 btrfs_put_fs_root(gang[0]);
2120 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2125 for (i = 0; i < ret; i++)
2126 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2129 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2130 btrfs_free_log_root_tree(NULL, fs_info);
2131 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2135 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2137 mutex_init(&fs_info->scrub_lock);
2138 atomic_set(&fs_info->scrubs_running, 0);
2139 atomic_set(&fs_info->scrub_pause_req, 0);
2140 atomic_set(&fs_info->scrubs_paused, 0);
2141 atomic_set(&fs_info->scrub_cancel_req, 0);
2142 init_waitqueue_head(&fs_info->scrub_pause_wait);
2143 fs_info->scrub_workers_refcnt = 0;
2146 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2148 spin_lock_init(&fs_info->balance_lock);
2149 mutex_init(&fs_info->balance_mutex);
2150 atomic_set(&fs_info->balance_pause_req, 0);
2151 atomic_set(&fs_info->balance_cancel_req, 0);
2152 fs_info->balance_ctl = NULL;
2153 init_waitqueue_head(&fs_info->balance_wait_q);
2156 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2158 struct inode *inode = fs_info->btree_inode;
2160 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2161 set_nlink(inode, 1);
2163 * we set the i_size on the btree inode to the max possible int.
2164 * the real end of the address space is determined by all of
2165 * the devices in the system
2167 inode->i_size = OFFSET_MAX;
2168 inode->i_mapping->a_ops = &btree_aops;
2170 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2171 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2172 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2173 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2175 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2177 BTRFS_I(inode)->root = fs_info->tree_root;
2178 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2179 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2180 btrfs_insert_inode_hash(inode);
2183 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2185 fs_info->dev_replace.lock_owner = 0;
2186 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2187 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2188 rwlock_init(&fs_info->dev_replace.lock);
2189 atomic_set(&fs_info->dev_replace.read_locks, 0);
2190 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2191 init_waitqueue_head(&fs_info->replace_wait);
2192 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2195 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2197 spin_lock_init(&fs_info->qgroup_lock);
2198 mutex_init(&fs_info->qgroup_ioctl_lock);
2199 fs_info->qgroup_tree = RB_ROOT;
2200 fs_info->qgroup_op_tree = RB_ROOT;
2201 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2202 fs_info->qgroup_seq = 1;
2203 fs_info->qgroup_ulist = NULL;
2204 fs_info->qgroup_rescan_running = false;
2205 mutex_init(&fs_info->qgroup_rescan_lock);
2208 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2209 struct btrfs_fs_devices *fs_devices)
2211 u32 max_active = fs_info->thread_pool_size;
2212 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2215 btrfs_alloc_workqueue(fs_info, "worker",
2216 flags | WQ_HIGHPRI, max_active, 16);
2218 fs_info->delalloc_workers =
2219 btrfs_alloc_workqueue(fs_info, "delalloc",
2220 flags, max_active, 2);
2222 fs_info->flush_workers =
2223 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2224 flags, max_active, 0);
2226 fs_info->caching_workers =
2227 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2230 * a higher idle thresh on the submit workers makes it much more
2231 * likely that bios will be send down in a sane order to the
2234 fs_info->submit_workers =
2235 btrfs_alloc_workqueue(fs_info, "submit", flags,
2236 min_t(u64, fs_devices->num_devices,
2239 fs_info->fixup_workers =
2240 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2243 * endios are largely parallel and should have a very
2246 fs_info->endio_workers =
2247 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2248 fs_info->endio_meta_workers =
2249 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2251 fs_info->endio_meta_write_workers =
2252 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2254 fs_info->endio_raid56_workers =
2255 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2257 fs_info->endio_repair_workers =
2258 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2259 fs_info->rmw_workers =
2260 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2261 fs_info->endio_write_workers =
2262 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2264 fs_info->endio_freespace_worker =
2265 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2267 fs_info->delayed_workers =
2268 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2270 fs_info->readahead_workers =
2271 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2273 fs_info->qgroup_rescan_workers =
2274 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2275 fs_info->extent_workers =
2276 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2277 min_t(u64, fs_devices->num_devices,
2280 if (!(fs_info->workers && fs_info->delalloc_workers &&
2281 fs_info->submit_workers && fs_info->flush_workers &&
2282 fs_info->endio_workers && fs_info->endio_meta_workers &&
2283 fs_info->endio_meta_write_workers &&
2284 fs_info->endio_repair_workers &&
2285 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2286 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2287 fs_info->caching_workers && fs_info->readahead_workers &&
2288 fs_info->fixup_workers && fs_info->delayed_workers &&
2289 fs_info->extent_workers &&
2290 fs_info->qgroup_rescan_workers)) {
2297 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2298 struct btrfs_fs_devices *fs_devices)
2301 struct btrfs_root *log_tree_root;
2302 struct btrfs_super_block *disk_super = fs_info->super_copy;
2303 u64 bytenr = btrfs_super_log_root(disk_super);
2304 int level = btrfs_super_log_root_level(disk_super);
2306 if (fs_devices->rw_devices == 0) {
2307 btrfs_warn(fs_info, "log replay required on RO media");
2311 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2315 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2317 log_tree_root->node = read_tree_block(fs_info, bytenr,
2318 fs_info->generation + 1,
2320 if (IS_ERR(log_tree_root->node)) {
2321 btrfs_warn(fs_info, "failed to read log tree");
2322 ret = PTR_ERR(log_tree_root->node);
2323 kfree(log_tree_root);
2325 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2326 btrfs_err(fs_info, "failed to read log tree");
2327 free_extent_buffer(log_tree_root->node);
2328 kfree(log_tree_root);
2331 /* returns with log_tree_root freed on success */
2332 ret = btrfs_recover_log_trees(log_tree_root);
2334 btrfs_handle_fs_error(fs_info, ret,
2335 "Failed to recover log tree");
2336 free_extent_buffer(log_tree_root->node);
2337 kfree(log_tree_root);
2341 if (sb_rdonly(fs_info->sb)) {
2342 ret = btrfs_commit_super(fs_info);
2350 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2352 struct btrfs_root *tree_root = fs_info->tree_root;
2353 struct btrfs_root *root;
2354 struct btrfs_key location;
2357 BUG_ON(!fs_info->tree_root);
2359 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2360 location.type = BTRFS_ROOT_ITEM_KEY;
2361 location.offset = 0;
2363 root = btrfs_read_tree_root(tree_root, &location);
2365 ret = PTR_ERR(root);
2368 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2369 fs_info->extent_root = root;
2371 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2372 root = btrfs_read_tree_root(tree_root, &location);
2374 ret = PTR_ERR(root);
2377 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2378 fs_info->dev_root = root;
2379 btrfs_init_devices_late(fs_info);
2381 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2382 root = btrfs_read_tree_root(tree_root, &location);
2384 ret = PTR_ERR(root);
2387 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2388 fs_info->csum_root = root;
2390 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2391 root = btrfs_read_tree_root(tree_root, &location);
2392 if (!IS_ERR(root)) {
2393 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2394 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2395 fs_info->quota_root = root;
2398 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2399 root = btrfs_read_tree_root(tree_root, &location);
2401 ret = PTR_ERR(root);
2405 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2406 fs_info->uuid_root = root;
2409 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2410 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2411 root = btrfs_read_tree_root(tree_root, &location);
2413 ret = PTR_ERR(root);
2416 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2417 fs_info->free_space_root = root;
2422 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2423 location.objectid, ret);
2428 * Real super block validation
2429 * NOTE: super csum type and incompat features will not be checked here.
2431 * @sb: super block to check
2432 * @mirror_num: the super block number to check its bytenr:
2433 * 0 the primary (1st) sb
2434 * 1, 2 2nd and 3rd backup copy
2435 * -1 skip bytenr check
2437 static int validate_super(struct btrfs_fs_info *fs_info,
2438 struct btrfs_super_block *sb, int mirror_num)
2440 u64 nodesize = btrfs_super_nodesize(sb);
2441 u64 sectorsize = btrfs_super_sectorsize(sb);
2444 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2445 btrfs_err(fs_info, "no valid FS found");
2448 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2449 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2450 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2453 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2454 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2455 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2458 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2459 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2460 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2463 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2464 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2465 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2470 * Check sectorsize and nodesize first, other check will need it.
2471 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2473 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2474 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2475 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2478 /* Only PAGE SIZE is supported yet */
2479 if (sectorsize != PAGE_SIZE) {
2481 "sectorsize %llu not supported yet, only support %lu",
2482 sectorsize, PAGE_SIZE);
2485 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2486 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2487 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2490 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2491 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2492 le32_to_cpu(sb->__unused_leafsize), nodesize);
2496 /* Root alignment check */
2497 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2498 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2499 btrfs_super_root(sb));
2502 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2503 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2504 btrfs_super_chunk_root(sb));
2507 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2508 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2509 btrfs_super_log_root(sb));
2513 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
2515 "dev_item UUID does not match fsid: %pU != %pU",
2516 fs_info->fsid, sb->dev_item.fsid);
2521 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2524 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2525 btrfs_err(fs_info, "bytes_used is too small %llu",
2526 btrfs_super_bytes_used(sb));
2529 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2530 btrfs_err(fs_info, "invalid stripesize %u",
2531 btrfs_super_stripesize(sb));
2534 if (btrfs_super_num_devices(sb) > (1UL << 31))
2535 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2536 btrfs_super_num_devices(sb));
2537 if (btrfs_super_num_devices(sb) == 0) {
2538 btrfs_err(fs_info, "number of devices is 0");
2542 if (mirror_num >= 0 &&
2543 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2544 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2545 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2550 * Obvious sys_chunk_array corruptions, it must hold at least one key
2553 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2554 btrfs_err(fs_info, "system chunk array too big %u > %u",
2555 btrfs_super_sys_array_size(sb),
2556 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2559 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2560 + sizeof(struct btrfs_chunk)) {
2561 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2562 btrfs_super_sys_array_size(sb),
2563 sizeof(struct btrfs_disk_key)
2564 + sizeof(struct btrfs_chunk));
2569 * The generation is a global counter, we'll trust it more than the others
2570 * but it's still possible that it's the one that's wrong.
2572 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2574 "suspicious: generation < chunk_root_generation: %llu < %llu",
2575 btrfs_super_generation(sb),
2576 btrfs_super_chunk_root_generation(sb));
2577 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2578 && btrfs_super_cache_generation(sb) != (u64)-1)
2580 "suspicious: generation < cache_generation: %llu < %llu",
2581 btrfs_super_generation(sb),
2582 btrfs_super_cache_generation(sb));
2588 * Validation of super block at mount time.
2589 * Some checks already done early at mount time, like csum type and incompat
2590 * flags will be skipped.
2592 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2594 return validate_super(fs_info, fs_info->super_copy, 0);
2598 * Validation of super block at write time.
2599 * Some checks like bytenr check will be skipped as their values will be
2601 * Extra checks like csum type and incompat flags will be done here.
2603 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2604 struct btrfs_super_block *sb)
2608 ret = validate_super(fs_info, sb, -1);
2611 if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2613 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2614 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2617 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2620 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2621 btrfs_super_incompat_flags(sb),
2622 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2628 "super block corruption detected before writing it to disk");
2632 int open_ctree(struct super_block *sb,
2633 struct btrfs_fs_devices *fs_devices,
2641 struct btrfs_key location;
2642 struct buffer_head *bh;
2643 struct btrfs_super_block *disk_super;
2644 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2645 struct btrfs_root *tree_root;
2646 struct btrfs_root *chunk_root;
2649 int num_backups_tried = 0;
2650 int backup_index = 0;
2651 int clear_free_space_tree = 0;
2654 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2655 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2656 if (!tree_root || !chunk_root) {
2661 ret = init_srcu_struct(&fs_info->subvol_srcu);
2667 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2672 fs_info->dirty_metadata_batch = PAGE_SIZE *
2673 (1 + ilog2(nr_cpu_ids));
2675 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2678 goto fail_dirty_metadata_bytes;
2681 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2684 goto fail_delalloc_bytes;
2687 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2688 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2689 INIT_LIST_HEAD(&fs_info->trans_list);
2690 INIT_LIST_HEAD(&fs_info->dead_roots);
2691 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2692 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2693 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2694 INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2695 spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2696 spin_lock_init(&fs_info->delalloc_root_lock);
2697 spin_lock_init(&fs_info->trans_lock);
2698 spin_lock_init(&fs_info->fs_roots_radix_lock);
2699 spin_lock_init(&fs_info->delayed_iput_lock);
2700 spin_lock_init(&fs_info->defrag_inodes_lock);
2701 spin_lock_init(&fs_info->tree_mod_seq_lock);
2702 spin_lock_init(&fs_info->super_lock);
2703 spin_lock_init(&fs_info->qgroup_op_lock);
2704 spin_lock_init(&fs_info->buffer_lock);
2705 spin_lock_init(&fs_info->unused_bgs_lock);
2706 rwlock_init(&fs_info->tree_mod_log_lock);
2707 mutex_init(&fs_info->unused_bg_unpin_mutex);
2708 mutex_init(&fs_info->delete_unused_bgs_mutex);
2709 mutex_init(&fs_info->reloc_mutex);
2710 mutex_init(&fs_info->delalloc_root_mutex);
2711 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2712 seqlock_init(&fs_info->profiles_lock);
2714 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2715 INIT_LIST_HEAD(&fs_info->space_info);
2716 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2717 INIT_LIST_HEAD(&fs_info->unused_bgs);
2718 btrfs_mapping_init(&fs_info->mapping_tree);
2719 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2720 BTRFS_BLOCK_RSV_GLOBAL);
2721 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2722 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2723 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2724 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2725 BTRFS_BLOCK_RSV_DELOPS);
2726 atomic_set(&fs_info->async_delalloc_pages, 0);
2727 atomic_set(&fs_info->defrag_running, 0);
2728 atomic_set(&fs_info->qgroup_op_seq, 0);
2729 atomic_set(&fs_info->reada_works_cnt, 0);
2730 atomic64_set(&fs_info->tree_mod_seq, 0);
2732 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2733 fs_info->metadata_ratio = 0;
2734 fs_info->defrag_inodes = RB_ROOT;
2735 atomic64_set(&fs_info->free_chunk_space, 0);
2736 fs_info->tree_mod_log = RB_ROOT;
2737 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2738 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2739 /* readahead state */
2740 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2741 spin_lock_init(&fs_info->reada_lock);
2742 btrfs_init_ref_verify(fs_info);
2744 fs_info->thread_pool_size = min_t(unsigned long,
2745 num_online_cpus() + 2, 8);
2747 INIT_LIST_HEAD(&fs_info->ordered_roots);
2748 spin_lock_init(&fs_info->ordered_root_lock);
2750 fs_info->btree_inode = new_inode(sb);
2751 if (!fs_info->btree_inode) {
2753 goto fail_bio_counter;
2755 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2757 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2759 if (!fs_info->delayed_root) {
2763 btrfs_init_delayed_root(fs_info->delayed_root);
2765 btrfs_init_scrub(fs_info);
2766 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2767 fs_info->check_integrity_print_mask = 0;
2769 btrfs_init_balance(fs_info);
2770 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2772 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2773 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2775 btrfs_init_btree_inode(fs_info);
2777 spin_lock_init(&fs_info->block_group_cache_lock);
2778 fs_info->block_group_cache_tree = RB_ROOT;
2779 fs_info->first_logical_byte = (u64)-1;
2781 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2782 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2783 fs_info->pinned_extents = &fs_info->freed_extents[0];
2784 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2786 mutex_init(&fs_info->ordered_operations_mutex);
2787 mutex_init(&fs_info->tree_log_mutex);
2788 mutex_init(&fs_info->chunk_mutex);
2789 mutex_init(&fs_info->transaction_kthread_mutex);
2790 mutex_init(&fs_info->cleaner_mutex);
2791 mutex_init(&fs_info->ro_block_group_mutex);
2792 init_rwsem(&fs_info->commit_root_sem);
2793 init_rwsem(&fs_info->cleanup_work_sem);
2794 init_rwsem(&fs_info->subvol_sem);
2795 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2797 btrfs_init_dev_replace_locks(fs_info);
2798 btrfs_init_qgroup(fs_info);
2800 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2801 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2803 init_waitqueue_head(&fs_info->transaction_throttle);
2804 init_waitqueue_head(&fs_info->transaction_wait);
2805 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2806 init_waitqueue_head(&fs_info->async_submit_wait);
2808 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2810 /* Usable values until the real ones are cached from the superblock */
2811 fs_info->nodesize = 4096;
2812 fs_info->sectorsize = 4096;
2813 fs_info->stripesize = 4096;
2815 ret = btrfs_alloc_stripe_hash_table(fs_info);
2821 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2823 invalidate_bdev(fs_devices->latest_bdev);
2826 * Read super block and check the signature bytes only
2828 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2835 * We want to check superblock checksum, the type is stored inside.
2836 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2838 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2839 btrfs_err(fs_info, "superblock checksum mismatch");
2846 * super_copy is zeroed at allocation time and we never touch the
2847 * following bytes up to INFO_SIZE, the checksum is calculated from
2848 * the whole block of INFO_SIZE
2850 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2851 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2852 sizeof(*fs_info->super_for_commit));
2855 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2857 ret = btrfs_validate_mount_super(fs_info);
2859 btrfs_err(fs_info, "superblock contains fatal errors");
2864 disk_super = fs_info->super_copy;
2865 if (!btrfs_super_root(disk_super))
2868 /* check FS state, whether FS is broken. */
2869 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2870 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2873 * run through our array of backup supers and setup
2874 * our ring pointer to the oldest one
2876 generation = btrfs_super_generation(disk_super);
2877 find_oldest_super_backup(fs_info, generation);
2880 * In the long term, we'll store the compression type in the super
2881 * block, and it'll be used for per file compression control.
2883 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2885 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2891 features = btrfs_super_incompat_flags(disk_super) &
2892 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2895 "cannot mount because of unsupported optional features (%llx)",
2901 features = btrfs_super_incompat_flags(disk_super);
2902 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2903 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2904 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2905 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2906 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2908 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2909 btrfs_info(fs_info, "has skinny extents");
2912 * flag our filesystem as having big metadata blocks if
2913 * they are bigger than the page size
2915 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2916 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2918 "flagging fs with big metadata feature");
2919 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2922 nodesize = btrfs_super_nodesize(disk_super);
2923 sectorsize = btrfs_super_sectorsize(disk_super);
2924 stripesize = sectorsize;
2925 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2926 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2928 /* Cache block sizes */
2929 fs_info->nodesize = nodesize;
2930 fs_info->sectorsize = sectorsize;
2931 fs_info->stripesize = stripesize;
2934 * mixed block groups end up with duplicate but slightly offset
2935 * extent buffers for the same range. It leads to corruptions
2937 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2938 (sectorsize != nodesize)) {
2940 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2941 nodesize, sectorsize);
2946 * Needn't use the lock because there is no other task which will
2949 btrfs_set_super_incompat_flags(disk_super, features);
2951 features = btrfs_super_compat_ro_flags(disk_super) &
2952 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2953 if (!sb_rdonly(sb) && features) {
2955 "cannot mount read-write because of unsupported optional features (%llx)",
2961 ret = btrfs_init_workqueues(fs_info, fs_devices);
2964 goto fail_sb_buffer;
2967 sb->s_bdi->congested_fn = btrfs_congested_fn;
2968 sb->s_bdi->congested_data = fs_info;
2969 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2970 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2971 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2972 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2974 sb->s_blocksize = sectorsize;
2975 sb->s_blocksize_bits = blksize_bits(sectorsize);
2976 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2978 mutex_lock(&fs_info->chunk_mutex);
2979 ret = btrfs_read_sys_array(fs_info);
2980 mutex_unlock(&fs_info->chunk_mutex);
2982 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2983 goto fail_sb_buffer;
2986 generation = btrfs_super_chunk_root_generation(disk_super);
2987 level = btrfs_super_chunk_root_level(disk_super);
2989 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2991 chunk_root->node = read_tree_block(fs_info,
2992 btrfs_super_chunk_root(disk_super),
2993 generation, level, NULL);
2994 if (IS_ERR(chunk_root->node) ||
2995 !extent_buffer_uptodate(chunk_root->node)) {
2996 btrfs_err(fs_info, "failed to read chunk root");
2997 if (!IS_ERR(chunk_root->node))
2998 free_extent_buffer(chunk_root->node);
2999 chunk_root->node = NULL;
3000 goto fail_tree_roots;
3002 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3003 chunk_root->commit_root = btrfs_root_node(chunk_root);
3005 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3006 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3008 ret = btrfs_read_chunk_tree(fs_info);
3010 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3011 goto fail_tree_roots;
3015 * Keep the devid that is marked to be the target device for the
3016 * device replace procedure
3018 btrfs_free_extra_devids(fs_devices, 0);
3020 if (!fs_devices->latest_bdev) {
3021 btrfs_err(fs_info, "failed to read devices");
3022 goto fail_tree_roots;
3026 generation = btrfs_super_generation(disk_super);
3027 level = btrfs_super_root_level(disk_super);
3029 tree_root->node = read_tree_block(fs_info,
3030 btrfs_super_root(disk_super),
3031 generation, level, NULL);
3032 if (IS_ERR(tree_root->node) ||
3033 !extent_buffer_uptodate(tree_root->node)) {
3034 btrfs_warn(fs_info, "failed to read tree root");
3035 if (!IS_ERR(tree_root->node))
3036 free_extent_buffer(tree_root->node);
3037 tree_root->node = NULL;
3038 goto recovery_tree_root;
3041 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3042 tree_root->commit_root = btrfs_root_node(tree_root);
3043 btrfs_set_root_refs(&tree_root->root_item, 1);
3045 mutex_lock(&tree_root->objectid_mutex);
3046 ret = btrfs_find_highest_objectid(tree_root,
3047 &tree_root->highest_objectid);
3049 mutex_unlock(&tree_root->objectid_mutex);
3050 goto recovery_tree_root;
3053 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3055 mutex_unlock(&tree_root->objectid_mutex);
3057 ret = btrfs_read_roots(fs_info);
3059 goto recovery_tree_root;
3061 fs_info->generation = generation;
3062 fs_info->last_trans_committed = generation;
3064 ret = btrfs_recover_balance(fs_info);
3066 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3067 goto fail_block_groups;
3070 ret = btrfs_init_dev_stats(fs_info);
3072 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3073 goto fail_block_groups;
3076 ret = btrfs_init_dev_replace(fs_info);
3078 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3079 goto fail_block_groups;
3082 btrfs_free_extra_devids(fs_devices, 1);
3084 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3086 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3088 goto fail_block_groups;
3091 ret = btrfs_sysfs_add_device(fs_devices);
3093 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3095 goto fail_fsdev_sysfs;
3098 ret = btrfs_sysfs_add_mounted(fs_info);
3100 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3101 goto fail_fsdev_sysfs;
3104 ret = btrfs_init_space_info(fs_info);
3106 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3110 ret = btrfs_read_block_groups(fs_info);
3112 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3116 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3118 "writeable mount is not allowed due to too many missing devices");
3122 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3124 if (IS_ERR(fs_info->cleaner_kthread))
3127 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3129 "btrfs-transaction");
3130 if (IS_ERR(fs_info->transaction_kthread))
3133 if (!btrfs_test_opt(fs_info, NOSSD) &&
3134 !fs_info->fs_devices->rotating) {
3135 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3139 * Mount does not set all options immediately, we can do it now and do
3140 * not have to wait for transaction commit
3142 btrfs_apply_pending_changes(fs_info);
3144 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3145 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3146 ret = btrfsic_mount(fs_info, fs_devices,
3147 btrfs_test_opt(fs_info,
3148 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3150 fs_info->check_integrity_print_mask);
3153 "failed to initialize integrity check module: %d",
3157 ret = btrfs_read_qgroup_config(fs_info);
3159 goto fail_trans_kthread;
3161 if (btrfs_build_ref_tree(fs_info))
3162 btrfs_err(fs_info, "couldn't build ref tree");
3164 /* do not make disk changes in broken FS or nologreplay is given */
3165 if (btrfs_super_log_root(disk_super) != 0 &&
3166 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3167 ret = btrfs_replay_log(fs_info, fs_devices);
3174 ret = btrfs_find_orphan_roots(fs_info);
3178 if (!sb_rdonly(sb)) {
3179 ret = btrfs_cleanup_fs_roots(fs_info);
3183 mutex_lock(&fs_info->cleaner_mutex);
3184 ret = btrfs_recover_relocation(tree_root);
3185 mutex_unlock(&fs_info->cleaner_mutex);
3187 btrfs_warn(fs_info, "failed to recover relocation: %d",
3194 location.objectid = BTRFS_FS_TREE_OBJECTID;
3195 location.type = BTRFS_ROOT_ITEM_KEY;
3196 location.offset = 0;
3198 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3199 if (IS_ERR(fs_info->fs_root)) {
3200 err = PTR_ERR(fs_info->fs_root);
3201 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3208 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3209 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3210 clear_free_space_tree = 1;
3211 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3212 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3213 btrfs_warn(fs_info, "free space tree is invalid");
3214 clear_free_space_tree = 1;
3217 if (clear_free_space_tree) {
3218 btrfs_info(fs_info, "clearing free space tree");
3219 ret = btrfs_clear_free_space_tree(fs_info);
3222 "failed to clear free space tree: %d", ret);
3223 close_ctree(fs_info);
3228 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3229 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3230 btrfs_info(fs_info, "creating free space tree");
3231 ret = btrfs_create_free_space_tree(fs_info);
3234 "failed to create free space tree: %d", ret);
3235 close_ctree(fs_info);
3240 down_read(&fs_info->cleanup_work_sem);
3241 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3242 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3243 up_read(&fs_info->cleanup_work_sem);
3244 close_ctree(fs_info);
3247 up_read(&fs_info->cleanup_work_sem);
3249 ret = btrfs_resume_balance_async(fs_info);
3251 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3252 close_ctree(fs_info);
3256 ret = btrfs_resume_dev_replace_async(fs_info);
3258 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);