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 void end_workqueue_fn(struct btrfs_work *work);
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
68 * is complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
71 struct btrfs_end_io_wq {
75 struct btrfs_fs_info *info;
77 enum btrfs_wq_endio_type metadata;
78 struct btrfs_work work;
81 static struct kmem_cache *btrfs_end_io_wq_cache;
83 int __init btrfs_end_io_wq_init(void)
85 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 sizeof(struct btrfs_end_io_wq),
90 if (!btrfs_end_io_wq_cache)
95 void __cold btrfs_end_io_wq_exit(void)
97 kmem_cache_destroy(btrfs_end_io_wq_cache);
100 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102 if (fs_info->csum_shash)
103 crypto_free_shash(fs_info->csum_shash);
107 * async submit bios are used to offload expensive checksumming
108 * onto the worker threads. They checksum file and metadata bios
109 * just before they are sent down the IO stack.
111 struct async_submit_bio {
114 extent_submit_bio_start_t *submit_bio_start;
117 * bio_offset is optional, can be used if the pages in the bio
118 * can't tell us where in the file the bio should go
121 struct btrfs_work work;
126 * Lockdep class keys for extent_buffer->lock's in this root. For a given
127 * eb, the lockdep key is determined by the btrfs_root it belongs to and
128 * the level the eb occupies in the tree.
130 * Different roots are used for different purposes and may nest inside each
131 * other and they require separate keysets. As lockdep keys should be
132 * static, assign keysets according to the purpose of the root as indicated
133 * by btrfs_root->root_key.objectid. This ensures that all special purpose
134 * roots have separate keysets.
136 * Lock-nesting across peer nodes is always done with the immediate parent
137 * node locked thus preventing deadlock. As lockdep doesn't know this, use
138 * subclass to avoid triggering lockdep warning in such cases.
140 * The key is set by the readpage_end_io_hook after the buffer has passed
141 * csum validation but before the pages are unlocked. It is also set by
142 * btrfs_init_new_buffer on freshly allocated blocks.
144 * We also add a check to make sure the highest level of the tree is the
145 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
146 * needs update as well.
148 #ifdef CONFIG_DEBUG_LOCK_ALLOC
149 # if BTRFS_MAX_LEVEL != 8
153 static struct btrfs_lockdep_keyset {
154 u64 id; /* root objectid */
155 const char *name_stem; /* lock name stem */
156 char names[BTRFS_MAX_LEVEL + 1][20];
157 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
158 } btrfs_lockdep_keysets[] = {
159 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
160 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
161 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
162 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
163 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
164 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
165 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
166 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
167 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
168 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
169 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
170 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
171 { .id = 0, .name_stem = "tree" },
174 void __init btrfs_init_lockdep(void)
178 /* initialize lockdep class names */
179 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
180 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
182 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
183 snprintf(ks->names[j], sizeof(ks->names[j]),
184 "btrfs-%s-%02d", ks->name_stem, j);
188 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
191 struct btrfs_lockdep_keyset *ks;
193 BUG_ON(level >= ARRAY_SIZE(ks->keys));
195 /* find the matching keyset, id 0 is the default entry */
196 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
197 if (ks->id == objectid)
200 lockdep_set_class_and_name(&eb->lock,
201 &ks->keys[level], ks->names[level]);
207 * Compute the csum of a btree block and store the result to provided buffer.
209 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
211 struct btrfs_fs_info *fs_info = buf->fs_info;
212 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
213 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
217 shash->tfm = fs_info->csum_shash;
218 crypto_shash_init(shash);
219 kaddr = page_address(buf->pages[0]);
220 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
221 PAGE_SIZE - BTRFS_CSUM_SIZE);
223 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
224 kaddr = page_address(buf->pages[i]);
225 crypto_shash_update(shash, kaddr, PAGE_SIZE);
227 memset(result, 0, BTRFS_CSUM_SIZE);
228 crypto_shash_final(shash, result);
232 * we can't consider a given block up to date unless the transid of the
233 * block matches the transid in the parent node's pointer. This is how we
234 * detect blocks that either didn't get written at all or got written
235 * in the wrong place.
237 static int verify_parent_transid(struct extent_io_tree *io_tree,
238 struct extent_buffer *eb, u64 parent_transid,
241 struct extent_state *cached_state = NULL;
243 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
245 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
252 btrfs_tree_read_lock(eb);
253 btrfs_set_lock_blocking_read(eb);
256 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
258 if (extent_buffer_uptodate(eb) &&
259 btrfs_header_generation(eb) == parent_transid) {
263 btrfs_err_rl(eb->fs_info,
264 "parent transid verify failed on %llu wanted %llu found %llu",
266 parent_transid, btrfs_header_generation(eb));
270 * Things reading via commit roots that don't have normal protection,
271 * like send, can have a really old block in cache that may point at a
272 * block that has been freed and re-allocated. So don't clear uptodate
273 * if we find an eb that is under IO (dirty/writeback) because we could
274 * end up reading in the stale data and then writing it back out and
275 * making everybody very sad.
277 if (!extent_buffer_under_io(eb))
278 clear_extent_buffer_uptodate(eb);
280 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
283 btrfs_tree_read_unlock_blocking(eb);
287 static bool btrfs_supported_super_csum(u16 csum_type)
290 case BTRFS_CSUM_TYPE_CRC32:
291 case BTRFS_CSUM_TYPE_XXHASH:
292 case BTRFS_CSUM_TYPE_SHA256:
293 case BTRFS_CSUM_TYPE_BLAKE2:
301 * Return 0 if the superblock checksum type matches the checksum value of that
302 * algorithm. Pass the raw disk superblock data.
304 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
307 struct btrfs_super_block *disk_sb =
308 (struct btrfs_super_block *)raw_disk_sb;
309 char result[BTRFS_CSUM_SIZE];
310 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
312 shash->tfm = fs_info->csum_shash;
315 * The super_block structure does not span the whole
316 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
317 * filled with zeros and is included in the checksum.
319 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
320 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
322 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
328 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
329 struct btrfs_key *first_key, u64 parent_transid)
331 struct btrfs_fs_info *fs_info = eb->fs_info;
333 struct btrfs_key found_key;
336 found_level = btrfs_header_level(eb);
337 if (found_level != level) {
338 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
339 KERN_ERR "BTRFS: tree level check failed\n");
341 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
342 eb->start, level, found_level);
350 * For live tree block (new tree blocks in current transaction),
351 * we need proper lock context to avoid race, which is impossible here.
352 * So we only checks tree blocks which is read from disk, whose
353 * generation <= fs_info->last_trans_committed.
355 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
358 /* We have @first_key, so this @eb must have at least one item */
359 if (btrfs_header_nritems(eb) == 0) {
361 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
363 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
368 btrfs_node_key_to_cpu(eb, &found_key, 0);
370 btrfs_item_key_to_cpu(eb, &found_key, 0);
371 ret = btrfs_comp_cpu_keys(first_key, &found_key);
374 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
375 KERN_ERR "BTRFS: tree first key check failed\n");
377 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
378 eb->start, parent_transid, first_key->objectid,
379 first_key->type, first_key->offset,
380 found_key.objectid, found_key.type,
387 * helper to read a given tree block, doing retries as required when
388 * the checksums don't match and we have alternate mirrors to try.
390 * @parent_transid: expected transid, skip check if 0
391 * @level: expected level, mandatory check
392 * @first_key: expected key of first slot, skip check if NULL
394 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
395 u64 parent_transid, int level,
396 struct btrfs_key *first_key)
398 struct btrfs_fs_info *fs_info = eb->fs_info;
399 struct extent_io_tree *io_tree;
404 int failed_mirror = 0;
406 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
408 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
409 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
411 if (verify_parent_transid(io_tree, eb,
414 else if (btrfs_verify_level_key(eb, level,
415 first_key, parent_transid))
421 num_copies = btrfs_num_copies(fs_info,
426 if (!failed_mirror) {
428 failed_mirror = eb->read_mirror;
432 if (mirror_num == failed_mirror)
435 if (mirror_num > num_copies)
439 if (failed && !ret && failed_mirror)
440 btrfs_repair_eb_io_failure(eb, failed_mirror);
446 * checksum a dirty tree block before IO. This has extra checks to make sure
447 * we only fill in the checksum field in the first page of a multi-page block
450 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
452 u64 start = page_offset(page);
454 u8 result[BTRFS_CSUM_SIZE];
455 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
456 struct extent_buffer *eb;
459 eb = (struct extent_buffer *)page->private;
460 if (page != eb->pages[0])
463 found_start = btrfs_header_bytenr(eb);
465 * Please do not consolidate these warnings into a single if.
466 * It is useful to know what went wrong.
468 if (WARN_ON(found_start != start))
470 if (WARN_ON(!PageUptodate(page)))
473 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
474 offsetof(struct btrfs_header, fsid),
475 BTRFS_FSID_SIZE) == 0);
477 csum_tree_block(eb, result);
479 if (btrfs_header_level(eb))
480 ret = btrfs_check_node(eb);
482 ret = btrfs_check_leaf_full(eb);
485 btrfs_print_tree(eb, 0);
487 "block=%llu write time tree block corruption detected",
489 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
492 write_extent_buffer(eb, result, 0, csum_size);
497 static int check_tree_block_fsid(struct extent_buffer *eb)
499 struct btrfs_fs_info *fs_info = eb->fs_info;
500 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
501 u8 fsid[BTRFS_FSID_SIZE];
504 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
507 * Checking the incompat flag is only valid for the current fs. For
508 * seed devices it's forbidden to have their uuid changed so reading
509 * ->fsid in this case is fine
511 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
512 metadata_uuid = fs_devices->metadata_uuid;
514 metadata_uuid = fs_devices->fsid;
516 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
519 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
520 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
526 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, u64 phy_offset,
527 struct page *page, u64 start, u64 end,
532 struct extent_buffer *eb;
533 struct btrfs_fs_info *fs_info;
536 u8 result[BTRFS_CSUM_SIZE];
542 eb = (struct extent_buffer *)page->private;
543 fs_info = eb->fs_info;
544 csum_size = btrfs_super_csum_size(fs_info->super_copy);
546 /* the pending IO might have been the only thing that kept this buffer
547 * in memory. Make sure we have a ref for all this other checks
549 atomic_inc(&eb->refs);
551 reads_done = atomic_dec_and_test(&eb->io_pages);
555 eb->read_mirror = mirror;
556 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
561 found_start = btrfs_header_bytenr(eb);
562 if (found_start != eb->start) {
563 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
564 eb->start, found_start);
568 if (check_tree_block_fsid(eb)) {
569 btrfs_err_rl(fs_info, "bad fsid on block %llu",
574 found_level = btrfs_header_level(eb);
575 if (found_level >= BTRFS_MAX_LEVEL) {
576 btrfs_err(fs_info, "bad tree block level %d on %llu",
577 (int)btrfs_header_level(eb), eb->start);
582 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
585 csum_tree_block(eb, result);
587 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
588 u8 val[BTRFS_CSUM_SIZE] = { 0 };
590 read_extent_buffer(eb, &val, 0, csum_size);
591 btrfs_warn_rl(fs_info,
592 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
593 fs_info->sb->s_id, eb->start,
594 CSUM_FMT_VALUE(csum_size, val),
595 CSUM_FMT_VALUE(csum_size, result),
596 btrfs_header_level(eb));
602 * If this is a leaf block and it is corrupt, set the corrupt bit so
603 * that we don't try and read the other copies of this block, just
606 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
607 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
611 if (found_level > 0 && btrfs_check_node(eb))
615 set_extent_buffer_uptodate(eb);
618 "block=%llu read time tree block corruption detected",
622 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
623 btree_readahead_hook(eb, ret);
627 * our io error hook is going to dec the io pages
628 * again, we have to make sure it has something
631 atomic_inc(&eb->io_pages);
632 clear_extent_buffer_uptodate(eb);
634 free_extent_buffer(eb);
639 static void end_workqueue_bio(struct bio *bio)
641 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
642 struct btrfs_fs_info *fs_info;
643 struct btrfs_workqueue *wq;
645 fs_info = end_io_wq->info;
646 end_io_wq->status = bio->bi_status;
648 if (bio_op(bio) == REQ_OP_WRITE) {
649 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
650 wq = fs_info->endio_meta_write_workers;
651 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
652 wq = fs_info->endio_freespace_worker;
653 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
654 wq = fs_info->endio_raid56_workers;
656 wq = fs_info->endio_write_workers;
658 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
659 wq = fs_info->endio_raid56_workers;
660 else if (end_io_wq->metadata)
661 wq = fs_info->endio_meta_workers;
663 wq = fs_info->endio_workers;
666 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
667 btrfs_queue_work(wq, &end_io_wq->work);
670 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
671 enum btrfs_wq_endio_type metadata)
673 struct btrfs_end_io_wq *end_io_wq;
675 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
677 return BLK_STS_RESOURCE;
679 end_io_wq->private = bio->bi_private;
680 end_io_wq->end_io = bio->bi_end_io;
681 end_io_wq->info = info;
682 end_io_wq->status = 0;
683 end_io_wq->bio = bio;
684 end_io_wq->metadata = metadata;
686 bio->bi_private = end_io_wq;
687 bio->bi_end_io = end_workqueue_bio;
691 static void run_one_async_start(struct btrfs_work *work)
693 struct async_submit_bio *async;
696 async = container_of(work, struct async_submit_bio, work);
697 ret = async->submit_bio_start(async->private_data, async->bio,
704 * In order to insert checksums into the metadata in large chunks, we wait
705 * until bio submission time. All the pages in the bio are checksummed and
706 * sums are attached onto the ordered extent record.
708 * At IO completion time the csums attached on the ordered extent record are
709 * inserted into the tree.
711 static void run_one_async_done(struct btrfs_work *work)
713 struct async_submit_bio *async;
717 async = container_of(work, struct async_submit_bio, work);
718 inode = async->private_data;
720 /* If an error occurred we just want to clean up the bio and move on */
722 async->bio->bi_status = async->status;
723 bio_endio(async->bio);
728 * All of the bios that pass through here are from async helpers.
729 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
730 * This changes nothing when cgroups aren't in use.
732 async->bio->bi_opf |= REQ_CGROUP_PUNT;
733 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
735 async->bio->bi_status = ret;
736 bio_endio(async->bio);
740 static void run_one_async_free(struct btrfs_work *work)
742 struct async_submit_bio *async;
744 async = container_of(work, struct async_submit_bio, work);
748 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
749 int mirror_num, unsigned long bio_flags,
750 u64 bio_offset, void *private_data,
751 extent_submit_bio_start_t *submit_bio_start)
753 struct async_submit_bio *async;
755 async = kmalloc(sizeof(*async), GFP_NOFS);
757 return BLK_STS_RESOURCE;
759 async->private_data = private_data;
761 async->mirror_num = mirror_num;
762 async->submit_bio_start = submit_bio_start;
764 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
767 async->bio_offset = bio_offset;
771 if (op_is_sync(bio->bi_opf))
772 btrfs_set_work_high_priority(&async->work);
774 btrfs_queue_work(fs_info->workers, &async->work);
778 static blk_status_t btree_csum_one_bio(struct bio *bio)
780 struct bio_vec *bvec;
781 struct btrfs_root *root;
783 struct bvec_iter_all iter_all;
785 ASSERT(!bio_flagged(bio, BIO_CLONED));
786 bio_for_each_segment_all(bvec, bio, iter_all) {
787 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
788 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
793 return errno_to_blk_status(ret);
796 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
800 * when we're called for a write, we're already in the async
801 * submission context. Just jump into btrfs_map_bio
803 return btree_csum_one_bio(bio);
806 static int check_async_write(struct btrfs_fs_info *fs_info,
807 struct btrfs_inode *bi)
809 if (atomic_read(&bi->sync_writers))
811 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
816 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
817 int mirror_num, unsigned long bio_flags)
819 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
820 int async = check_async_write(fs_info, BTRFS_I(inode));
823 if (bio_op(bio) != REQ_OP_WRITE) {
825 * called for a read, do the setup so that checksum validation
826 * can happen in the async kernel threads
828 ret = btrfs_bio_wq_end_io(fs_info, bio,
829 BTRFS_WQ_ENDIO_METADATA);
832 ret = btrfs_map_bio(fs_info, bio, mirror_num);
834 ret = btree_csum_one_bio(bio);
837 ret = btrfs_map_bio(fs_info, bio, mirror_num);
840 * kthread helpers are used to submit writes so that
841 * checksumming can happen in parallel across all CPUs
843 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
844 0, inode, btree_submit_bio_start);
852 bio->bi_status = ret;
857 #ifdef CONFIG_MIGRATION
858 static int btree_migratepage(struct address_space *mapping,
859 struct page *newpage, struct page *page,
860 enum migrate_mode mode)
863 * we can't safely write a btree page from here,
864 * we haven't done the locking hook
869 * Buffers may be managed in a filesystem specific way.
870 * We must have no buffers or drop them.
872 if (page_has_private(page) &&
873 !try_to_release_page(page, GFP_KERNEL))
875 return migrate_page(mapping, newpage, page, mode);
880 static int btree_writepages(struct address_space *mapping,
881 struct writeback_control *wbc)
883 struct btrfs_fs_info *fs_info;
886 if (wbc->sync_mode == WB_SYNC_NONE) {
888 if (wbc->for_kupdate)
891 fs_info = BTRFS_I(mapping->host)->root->fs_info;
892 /* this is a bit racy, but that's ok */
893 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
894 BTRFS_DIRTY_METADATA_THRESH,
895 fs_info->dirty_metadata_batch);
899 return btree_write_cache_pages(mapping, wbc);
902 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
904 if (PageWriteback(page) || PageDirty(page))
907 return try_release_extent_buffer(page);
910 static void btree_invalidatepage(struct page *page, unsigned int offset,
913 struct extent_io_tree *tree;
914 tree = &BTRFS_I(page->mapping->host)->io_tree;
915 extent_invalidatepage(tree, page, offset);
916 btree_releasepage(page, GFP_NOFS);
917 if (PagePrivate(page)) {
918 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
919 "page private not zero on page %llu",
920 (unsigned long long)page_offset(page));
921 detach_page_private(page);
925 static int btree_set_page_dirty(struct page *page)
928 struct extent_buffer *eb;
930 BUG_ON(!PagePrivate(page));
931 eb = (struct extent_buffer *)page->private;
933 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
934 BUG_ON(!atomic_read(&eb->refs));
935 btrfs_assert_tree_locked(eb);
937 return __set_page_dirty_nobuffers(page);
940 static const struct address_space_operations btree_aops = {
941 .writepages = btree_writepages,
942 .releasepage = btree_releasepage,
943 .invalidatepage = btree_invalidatepage,
944 #ifdef CONFIG_MIGRATION
945 .migratepage = btree_migratepage,
947 .set_page_dirty = btree_set_page_dirty,
950 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
952 struct extent_buffer *buf = NULL;
955 buf = btrfs_find_create_tree_block(fs_info, bytenr);
959 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
961 free_extent_buffer_stale(buf);
963 free_extent_buffer(buf);
966 struct extent_buffer *btrfs_find_create_tree_block(
967 struct btrfs_fs_info *fs_info,
970 if (btrfs_is_testing(fs_info))
971 return alloc_test_extent_buffer(fs_info, bytenr);
972 return alloc_extent_buffer(fs_info, bytenr);
976 * Read tree block at logical address @bytenr and do variant basic but critical
979 * @parent_transid: expected transid of this tree block, skip check if 0
980 * @level: expected level, mandatory check
981 * @first_key: expected key in slot 0, skip check if NULL
983 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
984 u64 parent_transid, int level,
985 struct btrfs_key *first_key)
987 struct extent_buffer *buf = NULL;
990 buf = btrfs_find_create_tree_block(fs_info, bytenr);
994 ret = btree_read_extent_buffer_pages(buf, parent_transid,
997 free_extent_buffer_stale(buf);
1004 void btrfs_clean_tree_block(struct extent_buffer *buf)
1006 struct btrfs_fs_info *fs_info = buf->fs_info;
1007 if (btrfs_header_generation(buf) ==
1008 fs_info->running_transaction->transid) {
1009 btrfs_assert_tree_locked(buf);
1011 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1012 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1014 fs_info->dirty_metadata_batch);
1015 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1016 btrfs_set_lock_blocking_write(buf);
1017 clear_extent_buffer_dirty(buf);
1022 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1025 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1026 root->fs_info = fs_info;
1028 root->commit_root = NULL;
1030 root->orphan_cleanup_state = 0;
1032 root->last_trans = 0;
1033 root->highest_objectid = 0;
1034 root->nr_delalloc_inodes = 0;
1035 root->nr_ordered_extents = 0;
1036 root->inode_tree = RB_ROOT;
1037 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1038 root->block_rsv = NULL;
1040 INIT_LIST_HEAD(&root->dirty_list);
1041 INIT_LIST_HEAD(&root->root_list);
1042 INIT_LIST_HEAD(&root->delalloc_inodes);
1043 INIT_LIST_HEAD(&root->delalloc_root);
1044 INIT_LIST_HEAD(&root->ordered_extents);
1045 INIT_LIST_HEAD(&root->ordered_root);
1046 INIT_LIST_HEAD(&root->reloc_dirty_list);
1047 INIT_LIST_HEAD(&root->logged_list[0]);
1048 INIT_LIST_HEAD(&root->logged_list[1]);
1049 spin_lock_init(&root->inode_lock);
1050 spin_lock_init(&root->delalloc_lock);
1051 spin_lock_init(&root->ordered_extent_lock);
1052 spin_lock_init(&root->accounting_lock);
1053 spin_lock_init(&root->log_extents_lock[0]);
1054 spin_lock_init(&root->log_extents_lock[1]);
1055 spin_lock_init(&root->qgroup_meta_rsv_lock);
1056 mutex_init(&root->objectid_mutex);
1057 mutex_init(&root->log_mutex);
1058 mutex_init(&root->ordered_extent_mutex);
1059 mutex_init(&root->delalloc_mutex);
1060 init_waitqueue_head(&root->qgroup_flush_wait);
1061 init_waitqueue_head(&root->log_writer_wait);
1062 init_waitqueue_head(&root->log_commit_wait[0]);
1063 init_waitqueue_head(&root->log_commit_wait[1]);
1064 INIT_LIST_HEAD(&root->log_ctxs[0]);
1065 INIT_LIST_HEAD(&root->log_ctxs[1]);
1066 atomic_set(&root->log_commit[0], 0);
1067 atomic_set(&root->log_commit[1], 0);
1068 atomic_set(&root->log_writers, 0);
1069 atomic_set(&root->log_batch, 0);
1070 refcount_set(&root->refs, 1);
1071 atomic_set(&root->snapshot_force_cow, 0);
1072 atomic_set(&root->nr_swapfiles, 0);
1073 root->log_transid = 0;
1074 root->log_transid_committed = -1;
1075 root->last_log_commit = 0;
1077 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1078 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1079 extent_io_tree_init(fs_info, &root->log_csum_range,
1080 IO_TREE_LOG_CSUM_RANGE, NULL);
1083 memset(&root->root_key, 0, sizeof(root->root_key));
1084 memset(&root->root_item, 0, sizeof(root->root_item));
1085 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1086 root->root_key.objectid = objectid;
1089 spin_lock_init(&root->root_item_lock);
1090 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1091 #ifdef CONFIG_BTRFS_DEBUG
1092 INIT_LIST_HEAD(&root->leak_list);
1093 spin_lock(&fs_info->fs_roots_radix_lock);
1094 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1095 spin_unlock(&fs_info->fs_roots_radix_lock);
1099 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1100 u64 objectid, gfp_t flags)
1102 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1104 __setup_root(root, fs_info, objectid);
1108 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1109 /* Should only be used by the testing infrastructure */
1110 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1112 struct btrfs_root *root;
1115 return ERR_PTR(-EINVAL);
1117 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1119 return ERR_PTR(-ENOMEM);
1121 /* We don't use the stripesize in selftest, set it as sectorsize */
1122 root->alloc_bytenr = 0;
1128 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1131 struct btrfs_fs_info *fs_info = trans->fs_info;
1132 struct extent_buffer *leaf;
1133 struct btrfs_root *tree_root = fs_info->tree_root;
1134 struct btrfs_root *root;
1135 struct btrfs_key key;
1136 unsigned int nofs_flag;
1140 * We're holding a transaction handle, so use a NOFS memory allocation
1141 * context to avoid deadlock if reclaim happens.
1143 nofs_flag = memalloc_nofs_save();
1144 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1145 memalloc_nofs_restore(nofs_flag);
1147 return ERR_PTR(-ENOMEM);
1149 root->root_key.objectid = objectid;
1150 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1151 root->root_key.offset = 0;
1153 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1154 BTRFS_NESTING_NORMAL);
1156 ret = PTR_ERR(leaf);
1162 btrfs_mark_buffer_dirty(leaf);
1164 root->commit_root = btrfs_root_node(root);
1165 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1167 root->root_item.flags = 0;
1168 root->root_item.byte_limit = 0;
1169 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1170 btrfs_set_root_generation(&root->root_item, trans->transid);
1171 btrfs_set_root_level(&root->root_item, 0);
1172 btrfs_set_root_refs(&root->root_item, 1);
1173 btrfs_set_root_used(&root->root_item, leaf->len);
1174 btrfs_set_root_last_snapshot(&root->root_item, 0);
1175 btrfs_set_root_dirid(&root->root_item, 0);
1176 if (is_fstree(objectid))
1177 generate_random_guid(root->root_item.uuid);
1179 export_guid(root->root_item.uuid, &guid_null);
1180 root->root_item.drop_level = 0;
1182 key.objectid = objectid;
1183 key.type = BTRFS_ROOT_ITEM_KEY;
1185 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1189 btrfs_tree_unlock(leaf);
1195 btrfs_tree_unlock(leaf);
1196 btrfs_put_root(root);
1198 return ERR_PTR(ret);
1201 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1202 struct btrfs_fs_info *fs_info)
1204 struct btrfs_root *root;
1205 struct extent_buffer *leaf;
1207 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1209 return ERR_PTR(-ENOMEM);
1211 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1212 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1213 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1216 * DON'T set SHAREABLE bit for log trees.
1218 * Log trees are not exposed to user space thus can't be snapshotted,
1219 * and they go away before a real commit is actually done.
1221 * They do store pointers to file data extents, and those reference
1222 * counts still get updated (along with back refs to the log tree).
1225 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1226 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1228 btrfs_put_root(root);
1229 return ERR_CAST(leaf);
1234 btrfs_mark_buffer_dirty(root->node);
1235 btrfs_tree_unlock(root->node);
1239 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1240 struct btrfs_fs_info *fs_info)
1242 struct btrfs_root *log_root;
1244 log_root = alloc_log_tree(trans, fs_info);
1245 if (IS_ERR(log_root))
1246 return PTR_ERR(log_root);
1247 WARN_ON(fs_info->log_root_tree);
1248 fs_info->log_root_tree = log_root;
1252 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1253 struct btrfs_root *root)
1255 struct btrfs_fs_info *fs_info = root->fs_info;
1256 struct btrfs_root *log_root;
1257 struct btrfs_inode_item *inode_item;
1259 log_root = alloc_log_tree(trans, fs_info);
1260 if (IS_ERR(log_root))
1261 return PTR_ERR(log_root);
1263 log_root->last_trans = trans->transid;
1264 log_root->root_key.offset = root->root_key.objectid;
1266 inode_item = &log_root->root_item.inode;
1267 btrfs_set_stack_inode_generation(inode_item, 1);
1268 btrfs_set_stack_inode_size(inode_item, 3);
1269 btrfs_set_stack_inode_nlink(inode_item, 1);
1270 btrfs_set_stack_inode_nbytes(inode_item,
1272 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1274 btrfs_set_root_node(&log_root->root_item, log_root->node);
1276 WARN_ON(root->log_root);
1277 root->log_root = log_root;
1278 root->log_transid = 0;
1279 root->log_transid_committed = -1;
1280 root->last_log_commit = 0;
1284 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1285 struct btrfs_path *path,
1286 struct btrfs_key *key)
1288 struct btrfs_root *root;
1289 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1294 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1296 return ERR_PTR(-ENOMEM);
1298 ret = btrfs_find_root(tree_root, key, path,
1299 &root->root_item, &root->root_key);
1306 generation = btrfs_root_generation(&root->root_item);
1307 level = btrfs_root_level(&root->root_item);
1308 root->node = read_tree_block(fs_info,
1309 btrfs_root_bytenr(&root->root_item),
1310 generation, level, NULL);
1311 if (IS_ERR(root->node)) {
1312 ret = PTR_ERR(root->node);
1315 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1319 root->commit_root = btrfs_root_node(root);
1322 btrfs_put_root(root);
1323 return ERR_PTR(ret);
1326 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1327 struct btrfs_key *key)
1329 struct btrfs_root *root;
1330 struct btrfs_path *path;
1332 path = btrfs_alloc_path();
1334 return ERR_PTR(-ENOMEM);
1335 root = read_tree_root_path(tree_root, path, key);
1336 btrfs_free_path(path);
1342 * Initialize subvolume root in-memory structure
1344 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1346 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1349 unsigned int nofs_flag;
1351 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1352 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1354 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1360 * We might be called under a transaction (e.g. indirect backref
1361 * resolution) which could deadlock if it triggers memory reclaim
1363 nofs_flag = memalloc_nofs_save();
1364 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1365 memalloc_nofs_restore(nofs_flag);
1369 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1370 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1371 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1372 btrfs_check_and_init_root_item(&root->root_item);
1375 btrfs_init_free_ino_ctl(root);
1376 spin_lock_init(&root->ino_cache_lock);
1377 init_waitqueue_head(&root->ino_cache_wait);
1380 * Don't assign anonymous block device to roots that are not exposed to
1381 * userspace, the id pool is limited to 1M
1383 if (is_fstree(root->root_key.objectid) &&
1384 btrfs_root_refs(&root->root_item) > 0) {
1386 ret = get_anon_bdev(&root->anon_dev);
1390 root->anon_dev = anon_dev;
1394 mutex_lock(&root->objectid_mutex);
1395 ret = btrfs_find_highest_objectid(root,
1396 &root->highest_objectid);
1398 mutex_unlock(&root->objectid_mutex);
1402 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1404 mutex_unlock(&root->objectid_mutex);
1408 /* The caller is responsible to call btrfs_free_fs_root */
1412 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1415 struct btrfs_root *root;
1417 spin_lock(&fs_info->fs_roots_radix_lock);
1418 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1419 (unsigned long)root_id);
1421 root = btrfs_grab_root(root);
1422 spin_unlock(&fs_info->fs_roots_radix_lock);
1426 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1429 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1430 return btrfs_grab_root(fs_info->tree_root);
1431 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1432 return btrfs_grab_root(fs_info->extent_root);
1433 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1434 return btrfs_grab_root(fs_info->chunk_root);
1435 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1436 return btrfs_grab_root(fs_info->dev_root);
1437 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1438 return btrfs_grab_root(fs_info->csum_root);
1439 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1440 return btrfs_grab_root(fs_info->quota_root) ?
1441 fs_info->quota_root : ERR_PTR(-ENOENT);
1442 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1443 return btrfs_grab_root(fs_info->uuid_root) ?
1444 fs_info->uuid_root : ERR_PTR(-ENOENT);
1445 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1446 return btrfs_grab_root(fs_info->free_space_root) ?
1447 fs_info->free_space_root : ERR_PTR(-ENOENT);
1451 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1452 struct btrfs_root *root)
1456 ret = radix_tree_preload(GFP_NOFS);
1460 spin_lock(&fs_info->fs_roots_radix_lock);
1461 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1462 (unsigned long)root->root_key.objectid,
1465 btrfs_grab_root(root);
1466 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1468 spin_unlock(&fs_info->fs_roots_radix_lock);
1469 radix_tree_preload_end();
1474 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1476 #ifdef CONFIG_BTRFS_DEBUG
1477 struct btrfs_root *root;
1479 while (!list_empty(&fs_info->allocated_roots)) {
1480 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1482 root = list_first_entry(&fs_info->allocated_roots,
1483 struct btrfs_root, leak_list);
1484 btrfs_err(fs_info, "leaked root %s refcount %d",
1485 btrfs_root_name(&root->root_key, buf),
1486 refcount_read(&root->refs));
1487 while (refcount_read(&root->refs) > 1)
1488 btrfs_put_root(root);
1489 btrfs_put_root(root);
1494 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1496 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1497 percpu_counter_destroy(&fs_info->delalloc_bytes);
1498 percpu_counter_destroy(&fs_info->dio_bytes);
1499 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1500 btrfs_free_csum_hash(fs_info);
1501 btrfs_free_stripe_hash_table(fs_info);
1502 btrfs_free_ref_cache(fs_info);
1503 kfree(fs_info->balance_ctl);
1504 kfree(fs_info->delayed_root);
1505 btrfs_put_root(fs_info->extent_root);
1506 btrfs_put_root(fs_info->tree_root);
1507 btrfs_put_root(fs_info->chunk_root);
1508 btrfs_put_root(fs_info->dev_root);
1509 btrfs_put_root(fs_info->csum_root);
1510 btrfs_put_root(fs_info->quota_root);
1511 btrfs_put_root(fs_info->uuid_root);
1512 btrfs_put_root(fs_info->free_space_root);
1513 btrfs_put_root(fs_info->fs_root);
1514 btrfs_put_root(fs_info->data_reloc_root);
1515 btrfs_check_leaked_roots(fs_info);
1516 btrfs_extent_buffer_leak_debug_check(fs_info);
1517 kfree(fs_info->super_copy);
1518 kfree(fs_info->super_for_commit);
1524 * Get an in-memory reference of a root structure.
1526 * For essential trees like root/extent tree, we grab it from fs_info directly.
1527 * For subvolume trees, we check the cached filesystem roots first. If not
1528 * found, then read it from disk and add it to cached fs roots.
1530 * Caller should release the root by calling btrfs_put_root() after the usage.
1532 * NOTE: Reloc and log trees can't be read by this function as they share the
1533 * same root objectid.
1535 * @objectid: root id
1536 * @anon_dev: preallocated anonymous block device number for new roots,
1537 * pass 0 for new allocation.
1538 * @check_ref: whether to check root item references, If true, return -ENOENT
1541 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1542 u64 objectid, dev_t anon_dev,
1545 struct btrfs_root *root;
1546 struct btrfs_path *path;
1547 struct btrfs_key key;
1550 root = btrfs_get_global_root(fs_info, objectid);
1554 root = btrfs_lookup_fs_root(fs_info, objectid);
1557 * Some other caller may have read out the newly inserted
1558 * subvolume already (for things like backref walk etc). Not
1559 * that common but still possible. In that case, we just need
1560 * to free the anon_dev.
1562 if (unlikely(anon_dev)) {
1563 free_anon_bdev(anon_dev);
1567 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1568 btrfs_put_root(root);
1569 return ERR_PTR(-ENOENT);
1574 key.objectid = objectid;
1575 key.type = BTRFS_ROOT_ITEM_KEY;
1576 key.offset = (u64)-1;
1577 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1581 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1586 ret = btrfs_init_fs_root(root, anon_dev);
1590 path = btrfs_alloc_path();
1595 key.objectid = BTRFS_ORPHAN_OBJECTID;
1596 key.type = BTRFS_ORPHAN_ITEM_KEY;
1597 key.offset = objectid;
1599 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1600 btrfs_free_path(path);
1604 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1606 ret = btrfs_insert_fs_root(fs_info, root);
1608 if (ret == -EEXIST) {
1609 btrfs_put_root(root);
1617 * If our caller provided us an anonymous device, then it's his
1618 * responsability to free it in case we fail. So we have to set our
1619 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1620 * and once again by our caller.
1624 btrfs_put_root(root);
1625 return ERR_PTR(ret);
1629 * Get in-memory reference of a root structure
1631 * @objectid: tree objectid
1632 * @check_ref: if set, verify that the tree exists and the item has at least
1635 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1636 u64 objectid, bool check_ref)
1638 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1642 * Get in-memory reference of a root structure, created as new, optionally pass
1643 * the anonymous block device id
1645 * @objectid: tree objectid
1646 * @anon_dev: if zero, allocate a new anonymous block device or use the
1649 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1650 u64 objectid, dev_t anon_dev)
1652 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1656 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1657 * @fs_info: the fs_info
1658 * @objectid: the objectid we need to lookup
1660 * This is exclusively used for backref walking, and exists specifically because
1661 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1662 * creation time, which means we may have to read the tree_root in order to look
1663 * up a fs root that is not in memory. If the root is not in memory we will
1664 * read the tree root commit root and look up the fs root from there. This is a
1665 * temporary root, it will not be inserted into the radix tree as it doesn't
1666 * have the most uptodate information, it'll simply be discarded once the
1667 * backref code is finished using the root.
1669 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1670 struct btrfs_path *path,
1673 struct btrfs_root *root;
1674 struct btrfs_key key;
1676 ASSERT(path->search_commit_root && path->skip_locking);
1679 * This can return -ENOENT if we ask for a root that doesn't exist, but
1680 * since this is called via the backref walking code we won't be looking
1681 * up a root that doesn't exist, unless there's corruption. So if root
1682 * != NULL just return it.
1684 root = btrfs_get_global_root(fs_info, objectid);
1688 root = btrfs_lookup_fs_root(fs_info, objectid);
1692 key.objectid = objectid;
1693 key.type = BTRFS_ROOT_ITEM_KEY;
1694 key.offset = (u64)-1;
1695 root = read_tree_root_path(fs_info->tree_root, path, &key);
1696 btrfs_release_path(path);
1702 * called by the kthread helper functions to finally call the bio end_io
1703 * functions. This is where read checksum verification actually happens
1705 static void end_workqueue_fn(struct btrfs_work *work)
1708 struct btrfs_end_io_wq *end_io_wq;
1710 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1711 bio = end_io_wq->bio;
1713 bio->bi_status = end_io_wq->status;
1714 bio->bi_private = end_io_wq->private;
1715 bio->bi_end_io = end_io_wq->end_io;
1717 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1720 static int cleaner_kthread(void *arg)
1722 struct btrfs_root *root = arg;
1723 struct btrfs_fs_info *fs_info = root->fs_info;
1729 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1731 /* Make the cleaner go to sleep early. */
1732 if (btrfs_need_cleaner_sleep(fs_info))
1736 * Do not do anything if we might cause open_ctree() to block
1737 * before we have finished mounting the filesystem.
1739 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1742 if (!mutex_trylock(&fs_info->cleaner_mutex))
1746 * Avoid the problem that we change the status of the fs
1747 * during the above check and trylock.
1749 if (btrfs_need_cleaner_sleep(fs_info)) {
1750 mutex_unlock(&fs_info->cleaner_mutex);
1754 btrfs_run_delayed_iputs(fs_info);
1756 again = btrfs_clean_one_deleted_snapshot(root);
1757 mutex_unlock(&fs_info->cleaner_mutex);
1760 * The defragger has dealt with the R/O remount and umount,
1761 * needn't do anything special here.
1763 btrfs_run_defrag_inodes(fs_info);
1766 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1767 * with relocation (btrfs_relocate_chunk) and relocation
1768 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1769 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1770 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1771 * unused block groups.
1773 btrfs_delete_unused_bgs(fs_info);
1775 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1776 if (kthread_should_park())
1778 if (kthread_should_stop())
1781 set_current_state(TASK_INTERRUPTIBLE);
1783 __set_current_state(TASK_RUNNING);
1788 static int transaction_kthread(void *arg)
1790 struct btrfs_root *root = arg;
1791 struct btrfs_fs_info *fs_info = root->fs_info;
1792 struct btrfs_trans_handle *trans;
1793 struct btrfs_transaction *cur;
1796 unsigned long delay;
1800 cannot_commit = false;
1801 delay = HZ * fs_info->commit_interval;
1802 mutex_lock(&fs_info->transaction_kthread_mutex);
1804 spin_lock(&fs_info->trans_lock);
1805 cur = fs_info->running_transaction;
1807 spin_unlock(&fs_info->trans_lock);
1811 now = ktime_get_seconds();
1812 if (cur->state < TRANS_STATE_COMMIT_START &&
1813 (now < cur->start_time ||
1814 now - cur->start_time < fs_info->commit_interval)) {
1815 spin_unlock(&fs_info->trans_lock);
1819 transid = cur->transid;
1820 spin_unlock(&fs_info->trans_lock);
1822 /* If the file system is aborted, this will always fail. */
1823 trans = btrfs_attach_transaction(root);
1824 if (IS_ERR(trans)) {
1825 if (PTR_ERR(trans) != -ENOENT)
1826 cannot_commit = true;
1829 if (transid == trans->transid) {
1830 btrfs_commit_transaction(trans);
1832 btrfs_end_transaction(trans);
1835 wake_up_process(fs_info->cleaner_kthread);
1836 mutex_unlock(&fs_info->transaction_kthread_mutex);
1838 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1839 &fs_info->fs_state)))
1840 btrfs_cleanup_transaction(fs_info);
1841 if (!kthread_should_stop() &&
1842 (!btrfs_transaction_blocked(fs_info) ||
1844 schedule_timeout_interruptible(delay);
1845 } while (!kthread_should_stop());
1850 * This will find the highest generation in the array of root backups. The
1851 * index of the highest array is returned, or -EINVAL if we can't find
1854 * We check to make sure the array is valid by comparing the
1855 * generation of the latest root in the array with the generation
1856 * in the super block. If they don't match we pitch it.
1858 static int find_newest_super_backup(struct btrfs_fs_info *info)
1860 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1862 struct btrfs_root_backup *root_backup;
1865 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1866 root_backup = info->super_copy->super_roots + i;
1867 cur = btrfs_backup_tree_root_gen(root_backup);
1868 if (cur == newest_gen)
1876 * copy all the root pointers into the super backup array.
1877 * this will bump the backup pointer by one when it is
1880 static void backup_super_roots(struct btrfs_fs_info *info)
1882 const int next_backup = info->backup_root_index;
1883 struct btrfs_root_backup *root_backup;
1885 root_backup = info->super_for_commit->super_roots + next_backup;
1888 * make sure all of our padding and empty slots get zero filled
1889 * regardless of which ones we use today
1891 memset(root_backup, 0, sizeof(*root_backup));
1893 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1895 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1896 btrfs_set_backup_tree_root_gen(root_backup,
1897 btrfs_header_generation(info->tree_root->node));
1899 btrfs_set_backup_tree_root_level(root_backup,
1900 btrfs_header_level(info->tree_root->node));
1902 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1903 btrfs_set_backup_chunk_root_gen(root_backup,
1904 btrfs_header_generation(info->chunk_root->node));
1905 btrfs_set_backup_chunk_root_level(root_backup,
1906 btrfs_header_level(info->chunk_root->node));
1908 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1909 btrfs_set_backup_extent_root_gen(root_backup,
1910 btrfs_header_generation(info->extent_root->node));
1911 btrfs_set_backup_extent_root_level(root_backup,
1912 btrfs_header_level(info->extent_root->node));
1915 * we might commit during log recovery, which happens before we set
1916 * the fs_root. Make sure it is valid before we fill it in.
1918 if (info->fs_root && info->fs_root->node) {
1919 btrfs_set_backup_fs_root(root_backup,
1920 info->fs_root->node->start);
1921 btrfs_set_backup_fs_root_gen(root_backup,
1922 btrfs_header_generation(info->fs_root->node));
1923 btrfs_set_backup_fs_root_level(root_backup,
1924 btrfs_header_level(info->fs_root->node));
1927 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1928 btrfs_set_backup_dev_root_gen(root_backup,
1929 btrfs_header_generation(info->dev_root->node));
1930 btrfs_set_backup_dev_root_level(root_backup,
1931 btrfs_header_level(info->dev_root->node));
1933 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1934 btrfs_set_backup_csum_root_gen(root_backup,
1935 btrfs_header_generation(info->csum_root->node));
1936 btrfs_set_backup_csum_root_level(root_backup,
1937 btrfs_header_level(info->csum_root->node));
1939 btrfs_set_backup_total_bytes(root_backup,
1940 btrfs_super_total_bytes(info->super_copy));
1941 btrfs_set_backup_bytes_used(root_backup,
1942 btrfs_super_bytes_used(info->super_copy));
1943 btrfs_set_backup_num_devices(root_backup,
1944 btrfs_super_num_devices(info->super_copy));
1947 * if we don't copy this out to the super_copy, it won't get remembered
1948 * for the next commit
1950 memcpy(&info->super_copy->super_roots,
1951 &info->super_for_commit->super_roots,
1952 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1956 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1957 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1959 * fs_info - filesystem whose backup roots need to be read
1960 * priority - priority of backup root required
1962 * Returns backup root index on success and -EINVAL otherwise.
1964 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1966 int backup_index = find_newest_super_backup(fs_info);
1967 struct btrfs_super_block *super = fs_info->super_copy;
1968 struct btrfs_root_backup *root_backup;
1970 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1972 return backup_index;
1974 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1975 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1980 root_backup = super->super_roots + backup_index;
1982 btrfs_set_super_generation(super,
1983 btrfs_backup_tree_root_gen(root_backup));
1984 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1985 btrfs_set_super_root_level(super,
1986 btrfs_backup_tree_root_level(root_backup));
1987 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1990 * Fixme: the total bytes and num_devices need to match or we should
1993 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1994 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1996 return backup_index;
1999 /* helper to cleanup workers */
2000 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2002 btrfs_destroy_workqueue(fs_info->fixup_workers);
2003 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2004 btrfs_destroy_workqueue(fs_info->workers);
2005 btrfs_destroy_workqueue(fs_info->endio_workers);
2006 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2007 btrfs_destroy_workqueue(fs_info->rmw_workers);
2008 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2009 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2010 btrfs_destroy_workqueue(fs_info->delayed_workers);
2011 btrfs_destroy_workqueue(fs_info->caching_workers);
2012 btrfs_destroy_workqueue(fs_info->readahead_workers);
2013 btrfs_destroy_workqueue(fs_info->flush_workers);
2014 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2015 if (fs_info->discard_ctl.discard_workers)
2016 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2018 * Now that all other work queues are destroyed, we can safely destroy
2019 * the queues used for metadata I/O, since tasks from those other work
2020 * queues can do metadata I/O operations.
2022 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2023 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2026 static void free_root_extent_buffers(struct btrfs_root *root)
2029 free_extent_buffer(root->node);
2030 free_extent_buffer(root->commit_root);
2032 root->commit_root = NULL;
2036 /* helper to cleanup tree roots */
2037 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2039 free_root_extent_buffers(info->tree_root);
2041 free_root_extent_buffers(info->dev_root);
2042 free_root_extent_buffers(info->extent_root);
2043 free_root_extent_buffers(info->csum_root);
2044 free_root_extent_buffers(info->quota_root);
2045 free_root_extent_buffers(info->uuid_root);
2046 free_root_extent_buffers(info->fs_root);
2047 free_root_extent_buffers(info->data_reloc_root);
2048 if (free_chunk_root)
2049 free_root_extent_buffers(info->chunk_root);
2050 free_root_extent_buffers(info->free_space_root);
2053 void btrfs_put_root(struct btrfs_root *root)
2058 if (refcount_dec_and_test(&root->refs)) {
2059 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2060 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2062 free_anon_bdev(root->anon_dev);
2063 btrfs_drew_lock_destroy(&root->snapshot_lock);
2064 free_root_extent_buffers(root);
2065 kfree(root->free_ino_ctl);
2066 kfree(root->free_ino_pinned);
2067 #ifdef CONFIG_BTRFS_DEBUG
2068 spin_lock(&root->fs_info->fs_roots_radix_lock);
2069 list_del_init(&root->leak_list);
2070 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2076 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2079 struct btrfs_root *gang[8];
2082 while (!list_empty(&fs_info->dead_roots)) {
2083 gang[0] = list_entry(fs_info->dead_roots.next,
2084 struct btrfs_root, root_list);
2085 list_del(&gang[0]->root_list);
2087 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2088 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2089 btrfs_put_root(gang[0]);
2093 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2098 for (i = 0; i < ret; i++)
2099 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2103 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2105 mutex_init(&fs_info->scrub_lock);
2106 atomic_set(&fs_info->scrubs_running, 0);
2107 atomic_set(&fs_info->scrub_pause_req, 0);
2108 atomic_set(&fs_info->scrubs_paused, 0);
2109 atomic_set(&fs_info->scrub_cancel_req, 0);
2110 init_waitqueue_head(&fs_info->scrub_pause_wait);
2111 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2114 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2116 spin_lock_init(&fs_info->balance_lock);
2117 mutex_init(&fs_info->balance_mutex);
2118 atomic_set(&fs_info->balance_pause_req, 0);
2119 atomic_set(&fs_info->balance_cancel_req, 0);
2120 fs_info->balance_ctl = NULL;
2121 init_waitqueue_head(&fs_info->balance_wait_q);
2124 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2126 struct inode *inode = fs_info->btree_inode;
2128 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2129 set_nlink(inode, 1);
2131 * we set the i_size on the btree inode to the max possible int.
2132 * the real end of the address space is determined by all of
2133 * the devices in the system
2135 inode->i_size = OFFSET_MAX;
2136 inode->i_mapping->a_ops = &btree_aops;
2138 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2139 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2140 IO_TREE_BTREE_INODE_IO, inode);
2141 BTRFS_I(inode)->io_tree.track_uptodate = false;
2142 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2144 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2145 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2146 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2147 btrfs_insert_inode_hash(inode);
2150 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2152 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2153 init_rwsem(&fs_info->dev_replace.rwsem);
2154 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2157 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2159 spin_lock_init(&fs_info->qgroup_lock);
2160 mutex_init(&fs_info->qgroup_ioctl_lock);
2161 fs_info->qgroup_tree = RB_ROOT;
2162 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2163 fs_info->qgroup_seq = 1;
2164 fs_info->qgroup_ulist = NULL;
2165 fs_info->qgroup_rescan_running = false;
2166 mutex_init(&fs_info->qgroup_rescan_lock);
2169 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2170 struct btrfs_fs_devices *fs_devices)
2172 u32 max_active = fs_info->thread_pool_size;
2173 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2176 btrfs_alloc_workqueue(fs_info, "worker",
2177 flags | WQ_HIGHPRI, max_active, 16);
2179 fs_info->delalloc_workers =
2180 btrfs_alloc_workqueue(fs_info, "delalloc",
2181 flags, max_active, 2);
2183 fs_info->flush_workers =
2184 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2185 flags, max_active, 0);
2187 fs_info->caching_workers =
2188 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2190 fs_info->fixup_workers =
2191 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2194 * endios are largely parallel and should have a very
2197 fs_info->endio_workers =
2198 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2199 fs_info->endio_meta_workers =
2200 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2202 fs_info->endio_meta_write_workers =
2203 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2205 fs_info->endio_raid56_workers =
2206 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2208 fs_info->rmw_workers =
2209 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2210 fs_info->endio_write_workers =
2211 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2213 fs_info->endio_freespace_worker =
2214 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2216 fs_info->delayed_workers =
2217 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2219 fs_info->readahead_workers =
2220 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2222 fs_info->qgroup_rescan_workers =
2223 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2224 fs_info->discard_ctl.discard_workers =
2225 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2227 if (!(fs_info->workers && fs_info->delalloc_workers &&
2228 fs_info->flush_workers &&
2229 fs_info->endio_workers && fs_info->endio_meta_workers &&
2230 fs_info->endio_meta_write_workers &&
2231 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2232 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2233 fs_info->caching_workers && fs_info->readahead_workers &&
2234 fs_info->fixup_workers && fs_info->delayed_workers &&
2235 fs_info->qgroup_rescan_workers &&
2236 fs_info->discard_ctl.discard_workers)) {
2243 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2245 struct crypto_shash *csum_shash;
2246 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2248 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2250 if (IS_ERR(csum_shash)) {
2251 btrfs_err(fs_info, "error allocating %s hash for checksum",
2253 return PTR_ERR(csum_shash);
2256 fs_info->csum_shash = csum_shash;
2259 * Check if the checksum implementation is a fast accelerated one.
2260 * As-is this is a bit of a hack and should be replaced once the csum
2261 * implementations provide that information themselves.
2263 switch (csum_type) {
2264 case BTRFS_CSUM_TYPE_CRC32:
2265 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2266 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2268 case BTRFS_CSUM_TYPE_XXHASH:
2269 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2275 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2276 btrfs_super_csum_name(csum_type),
2277 crypto_shash_driver_name(csum_shash));
2281 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2282 struct btrfs_fs_devices *fs_devices)
2285 struct btrfs_root *log_tree_root;
2286 struct btrfs_super_block *disk_super = fs_info->super_copy;
2287 u64 bytenr = btrfs_super_log_root(disk_super);
2288 int level = btrfs_super_log_root_level(disk_super);
2290 if (fs_devices->rw_devices == 0) {
2291 btrfs_warn(fs_info, "log replay required on RO media");
2295 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2300 log_tree_root->node = read_tree_block(fs_info, bytenr,
2301 fs_info->generation + 1,
2303 if (IS_ERR(log_tree_root->node)) {
2304 btrfs_warn(fs_info, "failed to read log tree");
2305 ret = PTR_ERR(log_tree_root->node);
2306 log_tree_root->node = NULL;
2307 btrfs_put_root(log_tree_root);
2309 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2310 btrfs_err(fs_info, "failed to read log tree");
2311 btrfs_put_root(log_tree_root);
2314 /* returns with log_tree_root freed on success */
2315 ret = btrfs_recover_log_trees(log_tree_root);
2317 btrfs_handle_fs_error(fs_info, ret,
2318 "Failed to recover log tree");
2319 btrfs_put_root(log_tree_root);
2323 if (sb_rdonly(fs_info->sb)) {
2324 ret = btrfs_commit_super(fs_info);
2332 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2334 struct btrfs_root *tree_root = fs_info->tree_root;
2335 struct btrfs_root *root;
2336 struct btrfs_key location;
2339 BUG_ON(!fs_info->tree_root);
2341 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2342 location.type = BTRFS_ROOT_ITEM_KEY;
2343 location.offset = 0;
2345 root = btrfs_read_tree_root(tree_root, &location);
2347 ret = PTR_ERR(root);
2350 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2351 fs_info->extent_root = root;
2353 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2354 root = btrfs_read_tree_root(tree_root, &location);
2356 ret = PTR_ERR(root);
2359 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2360 fs_info->dev_root = root;
2361 btrfs_init_devices_late(fs_info);
2363 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2364 root = btrfs_read_tree_root(tree_root, &location);
2366 ret = PTR_ERR(root);
2369 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2370 fs_info->csum_root = root;
2373 * This tree can share blocks with some other fs tree during relocation
2374 * and we need a proper setup by btrfs_get_fs_root
2376 root = btrfs_get_fs_root(tree_root->fs_info,
2377 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2379 ret = PTR_ERR(root);
2382 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2383 fs_info->data_reloc_root = root;
2385 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2386 root = btrfs_read_tree_root(tree_root, &location);
2387 if (!IS_ERR(root)) {
2388 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2389 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2390 fs_info->quota_root = root;
2393 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2394 root = btrfs_read_tree_root(tree_root, &location);
2396 ret = PTR_ERR(root);
2400 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2401 fs_info->uuid_root = root;
2404 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2405 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2406 root = btrfs_read_tree_root(tree_root, &location);
2408 ret = PTR_ERR(root);
2411 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2412 fs_info->free_space_root = root;
2417 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2418 location.objectid, ret);
2423 * Real super block validation
2424 * NOTE: super csum type and incompat features will not be checked here.
2426 * @sb: super block to check
2427 * @mirror_num: the super block number to check its bytenr:
2428 * 0 the primary (1st) sb
2429 * 1, 2 2nd and 3rd backup copy
2430 * -1 skip bytenr check
2432 static int validate_super(struct btrfs_fs_info *fs_info,
2433 struct btrfs_super_block *sb, int mirror_num)
2435 u64 nodesize = btrfs_super_nodesize(sb);
2436 u64 sectorsize = btrfs_super_sectorsize(sb);
2439 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2440 btrfs_err(fs_info, "no valid FS found");
2443 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2444 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2445 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2448 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2449 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2450 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2453 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2454 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2455 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2458 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2459 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2460 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2465 * Check sectorsize and nodesize first, other check will need it.
2466 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2468 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2469 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2470 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2473 /* Only PAGE SIZE is supported yet */
2474 if (sectorsize != PAGE_SIZE) {
2476 "sectorsize %llu not supported yet, only support %lu",
2477 sectorsize, PAGE_SIZE);
2480 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2481 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2482 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2485 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2486 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2487 le32_to_cpu(sb->__unused_leafsize), nodesize);
2491 /* Root alignment check */
2492 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2493 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2494 btrfs_super_root(sb));
2497 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2498 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2499 btrfs_super_chunk_root(sb));
2502 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2503 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2504 btrfs_super_log_root(sb));
2508 if (memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2510 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2511 sb->fsid, fs_info->fs_devices->fsid);
2515 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2516 BTRFS_FSID_SIZE) != 0) {
2518 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2519 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2523 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2524 BTRFS_FSID_SIZE) != 0) {
2526 "dev_item UUID does not match metadata fsid: %pU != %pU",
2527 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2532 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2535 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2536 btrfs_err(fs_info, "bytes_used is too small %llu",
2537 btrfs_super_bytes_used(sb));
2540 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2541 btrfs_err(fs_info, "invalid stripesize %u",
2542 btrfs_super_stripesize(sb));
2545 if (btrfs_super_num_devices(sb) > (1UL << 31))
2546 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2547 btrfs_super_num_devices(sb));
2548 if (btrfs_super_num_devices(sb) == 0) {
2549 btrfs_err(fs_info, "number of devices is 0");
2553 if (mirror_num >= 0 &&
2554 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2555 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2556 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2561 * Obvious sys_chunk_array corruptions, it must hold at least one key
2564 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2565 btrfs_err(fs_info, "system chunk array too big %u > %u",
2566 btrfs_super_sys_array_size(sb),
2567 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2570 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2571 + sizeof(struct btrfs_chunk)) {
2572 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2573 btrfs_super_sys_array_size(sb),
2574 sizeof(struct btrfs_disk_key)
2575 + sizeof(struct btrfs_chunk));
2580 * The generation is a global counter, we'll trust it more than the others
2581 * but it's still possible that it's the one that's wrong.
2583 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2585 "suspicious: generation < chunk_root_generation: %llu < %llu",
2586 btrfs_super_generation(sb),
2587 btrfs_super_chunk_root_generation(sb));
2588 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2589 && btrfs_super_cache_generation(sb) != (u64)-1)
2591 "suspicious: generation < cache_generation: %llu < %llu",
2592 btrfs_super_generation(sb),
2593 btrfs_super_cache_generation(sb));
2599 * Validation of super block at mount time.
2600 * Some checks already done early at mount time, like csum type and incompat
2601 * flags will be skipped.
2603 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2605 return validate_super(fs_info, fs_info->super_copy, 0);
2609 * Validation of super block at write time.
2610 * Some checks like bytenr check will be skipped as their values will be
2612 * Extra checks like csum type and incompat flags will be done here.
2614 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2615 struct btrfs_super_block *sb)
2619 ret = validate_super(fs_info, sb, -1);
2622 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2624 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2625 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2628 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2631 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2632 btrfs_super_incompat_flags(sb),
2633 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2639 "super block corruption detected before writing it to disk");
2643 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2645 int backup_index = find_newest_super_backup(fs_info);
2646 struct btrfs_super_block *sb = fs_info->super_copy;
2647 struct btrfs_root *tree_root = fs_info->tree_root;
2648 bool handle_error = false;
2652 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2657 if (!IS_ERR(tree_root->node))
2658 free_extent_buffer(tree_root->node);
2659 tree_root->node = NULL;
2661 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2664 free_root_pointers(fs_info, 0);
2667 * Don't use the log in recovery mode, it won't be
2670 btrfs_set_super_log_root(sb, 0);
2672 /* We can't trust the free space cache either */
2673 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2675 ret = read_backup_root(fs_info, i);
2680 generation = btrfs_super_generation(sb);
2681 level = btrfs_super_root_level(sb);
2682 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2683 generation, level, NULL);
2684 if (IS_ERR(tree_root->node)) {
2685 handle_error = true;
2686 ret = PTR_ERR(tree_root->node);
2687 tree_root->node = NULL;
2688 btrfs_warn(fs_info, "couldn't read tree root");
2691 } else if (!extent_buffer_uptodate(tree_root->node)) {
2692 handle_error = true;
2694 btrfs_warn(fs_info, "error while reading tree root");
2698 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2699 tree_root->commit_root = btrfs_root_node(tree_root);
2700 btrfs_set_root_refs(&tree_root->root_item, 1);
2703 * No need to hold btrfs_root::objectid_mutex since the fs
2704 * hasn't been fully initialised and we are the only user
2706 ret = btrfs_find_highest_objectid(tree_root,
2707 &tree_root->highest_objectid);
2709 handle_error = true;
2713 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2715 ret = btrfs_read_roots(fs_info);
2717 handle_error = true;
2721 /* All successful */
2722 fs_info->generation = generation;
2723 fs_info->last_trans_committed = generation;
2725 /* Always begin writing backup roots after the one being used */
2726 if (backup_index < 0) {
2727 fs_info->backup_root_index = 0;
2729 fs_info->backup_root_index = backup_index + 1;
2730 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2738 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2740 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2741 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2742 INIT_LIST_HEAD(&fs_info->trans_list);
2743 INIT_LIST_HEAD(&fs_info->dead_roots);
2744 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2745 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2746 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2747 spin_lock_init(&fs_info->delalloc_root_lock);
2748 spin_lock_init(&fs_info->trans_lock);
2749 spin_lock_init(&fs_info->fs_roots_radix_lock);
2750 spin_lock_init(&fs_info->delayed_iput_lock);
2751 spin_lock_init(&fs_info->defrag_inodes_lock);
2752 spin_lock_init(&fs_info->super_lock);
2753 spin_lock_init(&fs_info->buffer_lock);
2754 spin_lock_init(&fs_info->unused_bgs_lock);
2755 rwlock_init(&fs_info->tree_mod_log_lock);
2756 mutex_init(&fs_info->unused_bg_unpin_mutex);
2757 mutex_init(&fs_info->delete_unused_bgs_mutex);
2758 mutex_init(&fs_info->reloc_mutex);
2759 mutex_init(&fs_info->delalloc_root_mutex);
2760 seqlock_init(&fs_info->profiles_lock);
2762 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2763 INIT_LIST_HEAD(&fs_info->space_info);
2764 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2765 INIT_LIST_HEAD(&fs_info->unused_bgs);
2766 #ifdef CONFIG_BTRFS_DEBUG
2767 INIT_LIST_HEAD(&fs_info->allocated_roots);
2768 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2769 spin_lock_init(&fs_info->eb_leak_lock);
2771 extent_map_tree_init(&fs_info->mapping_tree);
2772 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2773 BTRFS_BLOCK_RSV_GLOBAL);
2774 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2775 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2776 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2777 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2778 BTRFS_BLOCK_RSV_DELOPS);
2779 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2780 BTRFS_BLOCK_RSV_DELREFS);
2782 atomic_set(&fs_info->async_delalloc_pages, 0);
2783 atomic_set(&fs_info->defrag_running, 0);
2784 atomic_set(&fs_info->reada_works_cnt, 0);
2785 atomic_set(&fs_info->nr_delayed_iputs, 0);
2786 atomic64_set(&fs_info->tree_mod_seq, 0);
2787 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2788 fs_info->metadata_ratio = 0;
2789 fs_info->defrag_inodes = RB_ROOT;
2790 atomic64_set(&fs_info->free_chunk_space, 0);
2791 fs_info->tree_mod_log = RB_ROOT;
2792 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2793 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2794 /* readahead state */
2795 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2796 spin_lock_init(&fs_info->reada_lock);
2797 btrfs_init_ref_verify(fs_info);
2799 fs_info->thread_pool_size = min_t(unsigned long,
2800 num_online_cpus() + 2, 8);
2802 INIT_LIST_HEAD(&fs_info->ordered_roots);
2803 spin_lock_init(&fs_info->ordered_root_lock);
2805 btrfs_init_scrub(fs_info);
2806 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2807 fs_info->check_integrity_print_mask = 0;
2809 btrfs_init_balance(fs_info);
2810 btrfs_init_async_reclaim_work(fs_info);
2812 spin_lock_init(&fs_info->block_group_cache_lock);
2813 fs_info->block_group_cache_tree = RB_ROOT;
2814 fs_info->first_logical_byte = (u64)-1;
2816 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2817 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2818 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2820 mutex_init(&fs_info->ordered_operations_mutex);
2821 mutex_init(&fs_info->tree_log_mutex);
2822 mutex_init(&fs_info->chunk_mutex);
2823 mutex_init(&fs_info->transaction_kthread_mutex);
2824 mutex_init(&fs_info->cleaner_mutex);
2825 mutex_init(&fs_info->ro_block_group_mutex);
2826 init_rwsem(&fs_info->commit_root_sem);
2827 init_rwsem(&fs_info->cleanup_work_sem);
2828 init_rwsem(&fs_info->subvol_sem);
2829 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2831 btrfs_init_dev_replace_locks(fs_info);
2832 btrfs_init_qgroup(fs_info);
2833 btrfs_discard_init(fs_info);
2835 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2836 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2838 init_waitqueue_head(&fs_info->transaction_throttle);
2839 init_waitqueue_head(&fs_info->transaction_wait);
2840 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2841 init_waitqueue_head(&fs_info->async_submit_wait);
2842 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2844 /* Usable values until the real ones are cached from the superblock */
2845 fs_info->nodesize = 4096;
2846 fs_info->sectorsize = 4096;
2847 fs_info->stripesize = 4096;
2849 spin_lock_init(&fs_info->swapfile_pins_lock);
2850 fs_info->swapfile_pins = RB_ROOT;
2852 fs_info->send_in_progress = 0;
2855 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2860 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2861 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2863 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2867 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2871 fs_info->dirty_metadata_batch = PAGE_SIZE *
2872 (1 + ilog2(nr_cpu_ids));
2874 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2878 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2883 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2885 if (!fs_info->delayed_root)
2887 btrfs_init_delayed_root(fs_info->delayed_root);
2889 return btrfs_alloc_stripe_hash_table(fs_info);
2892 static int btrfs_uuid_rescan_kthread(void *data)
2894 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2898 * 1st step is to iterate through the existing UUID tree and
2899 * to delete all entries that contain outdated data.
2900 * 2nd step is to add all missing entries to the UUID tree.
2902 ret = btrfs_uuid_tree_iterate(fs_info);
2905 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2907 up(&fs_info->uuid_tree_rescan_sem);
2910 return btrfs_uuid_scan_kthread(data);
2913 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2915 struct task_struct *task;
2917 down(&fs_info->uuid_tree_rescan_sem);
2918 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2920 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2921 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2922 up(&fs_info->uuid_tree_rescan_sem);
2923 return PTR_ERR(task);
2929 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2938 struct btrfs_super_block *disk_super;
2939 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2940 struct btrfs_root *tree_root;
2941 struct btrfs_root *chunk_root;
2944 int clear_free_space_tree = 0;
2947 ret = init_mount_fs_info(fs_info, sb);
2953 /* These need to be init'ed before we start creating inodes and such. */
2954 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2956 fs_info->tree_root = tree_root;
2957 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2959 fs_info->chunk_root = chunk_root;
2960 if (!tree_root || !chunk_root) {
2965 fs_info->btree_inode = new_inode(sb);
2966 if (!fs_info->btree_inode) {
2970 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2971 btrfs_init_btree_inode(fs_info);
2973 invalidate_bdev(fs_devices->latest_bdev);
2976 * Read super block and check the signature bytes only
2978 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2979 if (IS_ERR(disk_super)) {
2980 err = PTR_ERR(disk_super);
2985 * Verify the type first, if that or the checksum value are
2986 * corrupted, we'll find out
2988 csum_type = btrfs_super_csum_type(disk_super);
2989 if (!btrfs_supported_super_csum(csum_type)) {
2990 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2993 btrfs_release_disk_super(disk_super);
2997 ret = btrfs_init_csum_hash(fs_info, csum_type);
3000 btrfs_release_disk_super(disk_super);
3005 * We want to check superblock checksum, the type is stored inside.
3006 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3008 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3009 btrfs_err(fs_info, "superblock checksum mismatch");
3011 btrfs_release_disk_super(disk_super);
3016 * super_copy is zeroed at allocation time and we never touch the
3017 * following bytes up to INFO_SIZE, the checksum is calculated from
3018 * the whole block of INFO_SIZE
3020 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3021 btrfs_release_disk_super(disk_super);
3023 disk_super = fs_info->super_copy;
3026 features = btrfs_super_flags(disk_super);
3027 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3028 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3029 btrfs_set_super_flags(disk_super, features);
3031 "found metadata UUID change in progress flag, clearing");
3034 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3035 sizeof(*fs_info->super_for_commit));
3037 ret = btrfs_validate_mount_super(fs_info);
3039 btrfs_err(fs_info, "superblock contains fatal errors");
3044 if (!btrfs_super_root(disk_super))
3047 /* check FS state, whether FS is broken. */
3048 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3049 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3052 * In the long term, we'll store the compression type in the super
3053 * block, and it'll be used for per file compression control.
3055 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3058 * Flag our filesystem as having big metadata blocks if they are bigger
3059 * than the page size
3061 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3062 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3064 "flagging fs with big metadata feature");
3065 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3068 /* Set up fs_info before parsing mount options */
3069 nodesize = btrfs_super_nodesize(disk_super);
3070 sectorsize = btrfs_super_sectorsize(disk_super);
3071 stripesize = sectorsize;
3072 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3073 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3075 /* Cache block sizes */
3076 fs_info->nodesize = nodesize;
3077 fs_info->sectorsize = sectorsize;
3078 fs_info->stripesize = stripesize;
3080 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3086 features = btrfs_super_incompat_flags(disk_super) &
3087 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3090 "cannot mount because of unsupported optional features (0x%llx)",
3096 features = btrfs_super_incompat_flags(disk_super);
3097 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3098 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3099 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3100 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3101 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3103 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3104 btrfs_info(fs_info, "has skinny extents");
3107 * mixed block groups end up with duplicate but slightly offset
3108 * extent buffers for the same range. It leads to corruptions
3110 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3111 (sectorsize != nodesize)) {
3113 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3114 nodesize, sectorsize);
3119 * Needn't use the lock because there is no other task which will
3122 btrfs_set_super_incompat_flags(disk_super, features);
3124 features = btrfs_super_compat_ro_flags(disk_super) &
3125 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3126 if (!sb_rdonly(sb) && features) {
3128 "cannot mount read-write because of unsupported optional features (0x%llx)",
3134 * We have unsupported RO compat features, although RO mounted, we
3135 * should not cause any metadata write, including log replay.
3136 * Or we could screw up whatever the new feature requires.
3138 if (unlikely(features && btrfs_super_log_root(disk_super) &&
3139 !btrfs_test_opt(fs_info, NOLOGREPLAY))) {
3141 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3148 ret = btrfs_init_workqueues(fs_info, fs_devices);
3151 goto fail_sb_buffer;
3154 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3155 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3157 sb->s_blocksize = sectorsize;
3158 sb->s_blocksize_bits = blksize_bits(sectorsize);
3159 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3161 mutex_lock(&fs_info->chunk_mutex);
3162 ret = btrfs_read_sys_array(fs_info);
3163 mutex_unlock(&fs_info->chunk_mutex);
3165 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3166 goto fail_sb_buffer;
3169 generation = btrfs_super_chunk_root_generation(disk_super);
3170 level = btrfs_super_chunk_root_level(disk_super);
3172 chunk_root->node = read_tree_block(fs_info,
3173 btrfs_super_chunk_root(disk_super),
3174 generation, level, NULL);
3175 if (IS_ERR(chunk_root->node) ||
3176 !extent_buffer_uptodate(chunk_root->node)) {
3177 btrfs_err(fs_info, "failed to read chunk root");
3178 if (!IS_ERR(chunk_root->node))
3179 free_extent_buffer(chunk_root->node);
3180 chunk_root->node = NULL;
3181 goto fail_tree_roots;
3183 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3184 chunk_root->commit_root = btrfs_root_node(chunk_root);
3186 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3187 offsetof(struct btrfs_header, chunk_tree_uuid),
3190 ret = btrfs_read_chunk_tree(fs_info);
3192 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3193 goto fail_tree_roots;
3197 * Keep the devid that is marked to be the target device for the
3198 * device replace procedure
3200 btrfs_free_extra_devids(fs_devices, 0);
3202 if (!fs_devices->latest_bdev) {
3203 btrfs_err(fs_info, "failed to read devices");
3204 goto fail_tree_roots;
3207 ret = init_tree_roots(fs_info);
3209 goto fail_tree_roots;
3212 * If we have a uuid root and we're not being told to rescan we need to
3213 * check the generation here so we can set the
3214 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3215 * transaction during a balance or the log replay without updating the
3216 * uuid generation, and then if we crash we would rescan the uuid tree,
3217 * even though it was perfectly fine.
3219 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3220 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3221 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3223 ret = btrfs_verify_dev_extents(fs_info);
3226 "failed to verify dev extents against chunks: %d",
3228 goto fail_block_groups;
3230 ret = btrfs_recover_balance(fs_info);
3232 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3233 goto fail_block_groups;
3236 ret = btrfs_init_dev_stats(fs_info);
3238 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3239 goto fail_block_groups;
3242 ret = btrfs_init_dev_replace(fs_info);
3244 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3245 goto fail_block_groups;
3248 btrfs_free_extra_devids(fs_devices, 1);
3250 ret = btrfs_sysfs_add_fsid(fs_devices);
3252 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3254 goto fail_block_groups;
3257 ret = btrfs_sysfs_add_mounted(fs_info);
3259 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3260 goto fail_fsdev_sysfs;
3263 ret = btrfs_init_space_info(fs_info);
3265 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3269 ret = btrfs_read_block_groups(fs_info);
3271 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3275 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3276 !btrfs_check_rw_degradable(fs_info, NULL)) {
3278 "writable mount is not allowed due to too many missing devices");
3282 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3284 if (IS_ERR(fs_info->cleaner_kthread))
3287 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3289 "btrfs-transaction");
3290 if (IS_ERR(fs_info->transaction_kthread))
3293 if (!btrfs_test_opt(fs_info, NOSSD) &&
3294 !fs_info->fs_devices->rotating) {
3295 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3299 * Mount does not set all options immediately, we can do it now and do
3300 * not have to wait for transaction commit
3302 btrfs_apply_pending_changes(fs_info);
3304 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3305 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3306 ret = btrfsic_mount(fs_info, fs_devices,
3307 btrfs_test_opt(fs_info,
3308 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3310 fs_info->check_integrity_print_mask);
3313 "failed to initialize integrity check module: %d",
3317 ret = btrfs_read_qgroup_config(fs_info);
3319 goto fail_trans_kthread;
3321 if (btrfs_build_ref_tree(fs_info))
3322 btrfs_err(fs_info, "couldn't build ref tree");
3324 /* do not make disk changes in broken FS or nologreplay is given */
3325 if (btrfs_super_log_root(disk_super) != 0 &&
3326 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3327 btrfs_info(fs_info, "start tree-log replay");
3328 ret = btrfs_replay_log(fs_info, fs_devices);
3335 ret = btrfs_find_orphan_roots(fs_info);
3339 if (!sb_rdonly(sb)) {
3340 ret = btrfs_cleanup_fs_roots(fs_info);
3344 mutex_lock(&fs_info->cleaner_mutex);
3345 ret = btrfs_recover_relocation(tree_root);
3346 mutex_unlock(&fs_info->cleaner_mutex);
3348 btrfs_warn(fs_info, "failed to recover relocation: %d",
3355 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3356 if (IS_ERR(fs_info->fs_root)) {
3357 err = PTR_ERR(fs_info->fs_root);
3358 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3359 fs_info->fs_root = NULL;
3366 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3367 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3368 clear_free_space_tree = 1;
3369 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3370 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3371 btrfs_warn(fs_info, "free space tree is invalid");
3372 clear_free_space_tree = 1;
3375 if (clear_free_space_tree) {
3376 btrfs_info(fs_info, "clearing free space tree");
3377 ret = btrfs_clear_free_space_tree(fs_info);
3380 "failed to clear free space tree: %d", ret);
3381 close_ctree(fs_info);
3386 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3387 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3388 btrfs_info(fs_info, "creating free space tree");
3389 ret = btrfs_create_free_space_tree(fs_info);
3392 "failed to create free space tree: %d", ret);
3393 close_ctree(fs_info);
3398 down_read(&fs_info->cleanup_work_sem);
3399 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3400 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3401 up_read(&fs_info->cleanup_work_sem);
3402 close_ctree(fs_info);
3405 up_read(&fs_info->cleanup_work_sem);
3407 ret = btrfs_resume_balance_async(fs_info);
3409 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3410 close_ctree(fs_info);
3414 ret = btrfs_resume_dev_replace_async(fs_info);
3416 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3417 close_ctree(fs_info);
3421 btrfs_qgroup_rescan_resume(fs_info);
3422 btrfs_discard_resume(fs_info);
3424 if (!fs_info->uuid_root) {
3425 btrfs_info(fs_info, "creating UUID tree");
3426 ret = btrfs_create_uuid_tree(fs_info);
3429 "failed to create the UUID tree: %d", ret);
3430 close_ctree(fs_info);
3433 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3434 fs_info->generation !=
3435 btrfs_super_uuid_tree_generation(disk_super)) {
3436 btrfs_info(fs_info, "checking UUID tree");
3437 ret = btrfs_check_uuid_tree(fs_info);
3440 "failed to check the UUID tree: %d", ret);
3441 close_ctree(fs_info);
3445 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3448 * backuproot only affect mount behavior, and if open_ctree succeeded,
3449 * no need to keep the flag
3451 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3456 btrfs_free_qgroup_config(fs_info);
3458 kthread_stop(fs_info->transaction_kthread);
3459 btrfs_cleanup_transaction(fs_info);
3460 btrfs_free_fs_roots(fs_info);
3462 kthread_stop(fs_info->cleaner_kthread);
3465 * make sure we're done with the btree inode before we stop our
3468 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3471 btrfs_sysfs_remove_mounted(fs_info);
3474 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3477 btrfs_put_block_group_cache(fs_info);
3480 if (fs_info->data_reloc_root)
3481 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3482 free_root_pointers(fs_info, true);
3483 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3486 btrfs_stop_all_workers(fs_info);
3487 btrfs_free_block_groups(fs_info);
3489 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3491 iput(fs_info->btree_inode);
3493 btrfs_close_devices(fs_info->fs_devices);
3496 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3498 static void btrfs_end_super_write(struct bio *bio)
3500 struct btrfs_device *device = bio->bi_private;
3501 struct bio_vec *bvec;
3502 struct bvec_iter_all iter_all;
3505 bio_for_each_segment_all(bvec, bio, iter_all) {
3506 page = bvec->bv_page;
3508 if (bio->bi_status) {
3509 btrfs_warn_rl_in_rcu(device->fs_info,
3510 "lost page write due to IO error on %s (%d)",
3511 rcu_str_deref(device->name),
3512 blk_status_to_errno(bio->bi_status));
3513 ClearPageUptodate(page);
3515 btrfs_dev_stat_inc_and_print(device,
3516 BTRFS_DEV_STAT_WRITE_ERRS);
3518 SetPageUptodate(page);
3528 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3531 struct btrfs_super_block *super;
3534 struct address_space *mapping = bdev->bd_inode->i_mapping;
3536 bytenr = btrfs_sb_offset(copy_num);
3537 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3538 return ERR_PTR(-EINVAL);
3540 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3542 return ERR_CAST(page);
3544 super = page_address(page);
3545 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3546 btrfs_release_disk_super(super);
3547 return ERR_PTR(-ENODATA);
3550 if (btrfs_super_bytenr(super) != bytenr) {
3551 btrfs_release_disk_super(super);
3552 return ERR_PTR(-EINVAL);
3559 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3561 struct btrfs_super_block *super, *latest = NULL;
3565 /* we would like to check all the supers, but that would make
3566 * a btrfs mount succeed after a mkfs from a different FS.
3567 * So, we need to add a special mount option to scan for
3568 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3570 for (i = 0; i < 1; i++) {
3571 super = btrfs_read_dev_one_super(bdev, i);
3575 if (!latest || btrfs_super_generation(super) > transid) {
3577 btrfs_release_disk_super(super);
3580 transid = btrfs_super_generation(super);
3588 * Write superblock @sb to the @device. Do not wait for completion, all the
3589 * pages we use for writing are locked.
3591 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3592 * the expected device size at commit time. Note that max_mirrors must be
3593 * same for write and wait phases.
3595 * Return number of errors when page is not found or submission fails.
3597 static int write_dev_supers(struct btrfs_device *device,
3598 struct btrfs_super_block *sb, int max_mirrors)
3600 struct btrfs_fs_info *fs_info = device->fs_info;
3601 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3602 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3607 if (max_mirrors == 0)
3608 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3610 shash->tfm = fs_info->csum_shash;
3612 for (i = 0; i < max_mirrors; i++) {
3615 struct btrfs_super_block *disk_super;
3617 bytenr = btrfs_sb_offset(i);
3618 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3619 device->commit_total_bytes)
3622 btrfs_set_super_bytenr(sb, bytenr);
3624 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3625 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3628 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3631 btrfs_err(device->fs_info,
3632 "couldn't get super block page for bytenr %llu",
3638 /* Bump the refcount for wait_dev_supers() */
3641 disk_super = page_address(page);
3642 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3645 * Directly use bios here instead of relying on the page cache
3646 * to do I/O, so we don't lose the ability to do integrity
3649 bio = bio_alloc(GFP_NOFS, 1);
3650 bio_set_dev(bio, device->bdev);
3651 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3652 bio->bi_private = device;
3653 bio->bi_end_io = btrfs_end_super_write;
3654 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3655 offset_in_page(bytenr));
3658 * We FUA only the first super block. The others we allow to
3659 * go down lazy and there's a short window where the on-disk
3660 * copies might still contain the older version.
3662 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3663 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3664 bio->bi_opf |= REQ_FUA;
3666 btrfsic_submit_bio(bio);
3668 return errors < i ? 0 : -1;
3672 * Wait for write completion of superblocks done by write_dev_supers,
3673 * @max_mirrors same for write and wait phases.
3675 * Return number of errors when page is not found or not marked up to
3678 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3682 bool primary_failed = false;
3685 if (max_mirrors == 0)
3686 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3688 for (i = 0; i < max_mirrors; i++) {
3691 bytenr = btrfs_sb_offset(i);
3692 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3693 device->commit_total_bytes)
3696 page = find_get_page(device->bdev->bd_inode->i_mapping,
3697 bytenr >> PAGE_SHIFT);
3701 primary_failed = true;
3704 /* Page is submitted locked and unlocked once the IO completes */
3705 wait_on_page_locked(page);
3706 if (PageError(page)) {
3709 primary_failed = true;
3712 /* Drop our reference */
3715 /* Drop the reference from the writing run */
3719 /* log error, force error return */
3720 if (primary_failed) {
3721 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3726 return errors < i ? 0 : -1;
3730 * endio for the write_dev_flush, this will wake anyone waiting
3731 * for the barrier when it is done
3733 static void btrfs_end_empty_barrier(struct bio *bio)
3735 complete(bio->bi_private);
3739 * Submit a flush request to the device if it supports it. Error handling is
3740 * done in the waiting counterpart.
3742 static void write_dev_flush(struct btrfs_device *device)
3744 struct bio *bio = device->flush_bio;
3746 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3748 * When a disk has write caching disabled, we skip submission of a bio
3749 * with flush and sync requests before writing the superblock, since
3750 * it's not needed. However when the integrity checker is enabled, this
3751 * results in reports that there are metadata blocks referred by a
3752 * superblock that were not properly flushed. So don't skip the bio
3753 * submission only when the integrity checker is enabled for the sake
3754 * of simplicity, since this is a debug tool and not meant for use in
3757 struct request_queue *q = bdev_get_queue(device->bdev);
3758 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3763 bio->bi_end_io = btrfs_end_empty_barrier;
3764 bio_set_dev(bio, device->bdev);
3765 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3766 init_completion(&device->flush_wait);
3767 bio->bi_private = &device->flush_wait;
3769 btrfsic_submit_bio(bio);
3770 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3774 * If the flush bio has been submitted by write_dev_flush, wait for it.
3776 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3778 struct bio *bio = device->flush_bio;
3780 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3783 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3784 wait_for_completion_io(&device->flush_wait);
3786 return bio->bi_status;
3789 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3791 if (!btrfs_check_rw_degradable(fs_info, NULL))
3797 * send an empty flush down to each device in parallel,
3798 * then wait for them
3800 static int barrier_all_devices(struct btrfs_fs_info *info)
3802 struct list_head *head;
3803 struct btrfs_device *dev;
3804 int errors_wait = 0;
3807 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3808 /* send down all the barriers */
3809 head = &info->fs_devices->devices;
3810 list_for_each_entry(dev, head, dev_list) {
3811 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3815 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3816 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3819 write_dev_flush(dev);
3820 dev->last_flush_error = BLK_STS_OK;
3823 /* wait for all the barriers */
3824 list_for_each_entry(dev, head, dev_list) {
3825 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3831 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3832 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3835 ret = wait_dev_flush(dev);
3837 dev->last_flush_error = ret;
3838 btrfs_dev_stat_inc_and_print(dev,
3839 BTRFS_DEV_STAT_FLUSH_ERRS);
3846 * At some point we need the status of all disks
3847 * to arrive at the volume status. So error checking
3848 * is being pushed to a separate loop.
3850 return check_barrier_error(info);
3855 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3858 int min_tolerated = INT_MAX;
3860 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3861 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3862 min_tolerated = min_t(int, min_tolerated,
3863 btrfs_raid_array[BTRFS_RAID_SINGLE].
3864 tolerated_failures);
3866 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3867 if (raid_type == BTRFS_RAID_SINGLE)
3869 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3871 min_tolerated = min_t(int, min_tolerated,
3872 btrfs_raid_array[raid_type].
3873 tolerated_failures);
3876 if (min_tolerated == INT_MAX) {
3877 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3881 return min_tolerated;
3884 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3886 struct list_head *head;
3887 struct btrfs_device *dev;
3888 struct btrfs_super_block *sb;
3889 struct btrfs_dev_item *dev_item;
3893 int total_errors = 0;
3896 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3899 * max_mirrors == 0 indicates we're from commit_transaction,
3900 * not from fsync where the tree roots in fs_info have not
3901 * been consistent on disk.
3903 if (max_mirrors == 0)
3904 backup_super_roots(fs_info);
3906 sb = fs_info->super_for_commit;
3907 dev_item = &sb->dev_item;
3909 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3910 head = &fs_info->fs_devices->devices;
3911 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3914 ret = barrier_all_devices(fs_info);
3917 &fs_info->fs_devices->device_list_mutex);
3918 btrfs_handle_fs_error(fs_info, ret,
3919 "errors while submitting device barriers.");
3924 list_for_each_entry(dev, head, dev_list) {
3929 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3930 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3933 btrfs_set_stack_device_generation(dev_item, 0);
3934 btrfs_set_stack_device_type(dev_item, dev->type);
3935 btrfs_set_stack_device_id(dev_item, dev->devid);
3936 btrfs_set_stack_device_total_bytes(dev_item,
3937 dev->commit_total_bytes);
3938 btrfs_set_stack_device_bytes_used(dev_item,
3939 dev->commit_bytes_used);
3940 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3941 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3942 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3943 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3944 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3947 flags = btrfs_super_flags(sb);
3948 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3950 ret = btrfs_validate_write_super(fs_info, sb);
3952 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3953 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3954 "unexpected superblock corruption detected");
3958 ret = write_dev_supers(dev, sb, max_mirrors);
3962 if (total_errors > max_errors) {
3963 btrfs_err(fs_info, "%d errors while writing supers",
3965 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3967 /* FUA is masked off if unsupported and can't be the reason */
3968 btrfs_handle_fs_error(fs_info, -EIO,
3969 "%d errors while writing supers",
3975 list_for_each_entry(dev, head, dev_list) {
3978 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3979 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3982 ret = wait_dev_supers(dev, max_mirrors);
3986 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3987 if (total_errors > max_errors) {
3988 btrfs_handle_fs_error(fs_info, -EIO,
3989 "%d errors while writing supers",
3996 /* Drop a fs root from the radix tree and free it. */
3997 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3998 struct btrfs_root *root)
4000 bool drop_ref = false;
4002 spin_lock(&fs_info->fs_roots_radix_lock);
4003 radix_tree_delete(&fs_info->fs_roots_radix,
4004 (unsigned long)root->root_key.objectid);
4005 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4007 spin_unlock(&fs_info->fs_roots_radix_lock);
4009 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4010 ASSERT(root->log_root == NULL);
4011 if (root->reloc_root) {
4012 btrfs_put_root(root->reloc_root);
4013 root->reloc_root = NULL;
4017 if (root->free_ino_pinned)
4018 __btrfs_remove_free_space_cache(root->free_ino_pinned);
4019 if (root->free_ino_ctl)
4020 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4021 if (root->ino_cache_inode) {
4022 iput(root->ino_cache_inode);
4023 root->ino_cache_inode = NULL;
4026 btrfs_put_root(root);
4029 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4031 u64 root_objectid = 0;
4032 struct btrfs_root *gang[8];
4035 unsigned int ret = 0;
4038 spin_lock(&fs_info->fs_roots_radix_lock);
4039 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4040 (void **)gang, root_objectid,
4043 spin_unlock(&fs_info->fs_roots_radix_lock);
4046 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4048 for (i = 0; i < ret; i++) {
4049 /* Avoid to grab roots in dead_roots */
4050 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4054 /* grab all the search result for later use */
4055 gang[i] = btrfs_grab_root(gang[i]);
4057 spin_unlock(&fs_info->fs_roots_radix_lock);
4059 for (i = 0; i < ret; i++) {
4062 root_objectid = gang[i]->root_key.objectid;
4063 err = btrfs_orphan_cleanup(gang[i]);
4066 btrfs_put_root(gang[i]);
4071 /* release the uncleaned roots due to error */
4072 for (; i < ret; i++) {
4074 btrfs_put_root(gang[i]);
4079 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4081 struct btrfs_root *root = fs_info->tree_root;
4082 struct btrfs_trans_handle *trans;
4084 mutex_lock(&fs_info->cleaner_mutex);
4085 btrfs_run_delayed_iputs(fs_info);
4086 mutex_unlock(&fs_info->cleaner_mutex);
4087 wake_up_process(fs_info->cleaner_kthread);
4089 /* wait until ongoing cleanup work done */
4090 down_write(&fs_info->cleanup_work_sem);
4091 up_write(&fs_info->cleanup_work_sem);
4093 trans = btrfs_join_transaction(root);
4095 return PTR_ERR(trans);
4096 return btrfs_commit_transaction(trans);
4099 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4103 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4105 * We don't want the cleaner to start new transactions, add more delayed
4106 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4107 * because that frees the task_struct, and the transaction kthread might
4108 * still try to wake up the cleaner.
4110 kthread_park(fs_info->cleaner_kthread);
4112 /* wait for the qgroup rescan worker to stop */
4113 btrfs_qgroup_wait_for_completion(fs_info, false);
4115 /* wait for the uuid_scan task to finish */
4116 down(&fs_info->uuid_tree_rescan_sem);
4117 /* avoid complains from lockdep et al., set sem back to initial state */
4118 up(&fs_info->uuid_tree_rescan_sem);
4120 /* pause restriper - we want to resume on mount */
4121 btrfs_pause_balance(fs_info);
4123 btrfs_dev_replace_suspend_for_unmount(fs_info);
4125 btrfs_scrub_cancel(fs_info);
4127 /* wait for any defraggers to finish */
4128 wait_event(fs_info->transaction_wait,
4129 (atomic_read(&fs_info->defrag_running) == 0));
4131 /* clear out the rbtree of defraggable inodes */
4132 btrfs_cleanup_defrag_inodes(fs_info);
4135 * After we parked the cleaner kthread, ordered extents may have
4136 * completed and created new delayed iputs. If one of the async reclaim
4137 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4138 * can hang forever trying to stop it, because if a delayed iput is
4139 * added after it ran btrfs_run_delayed_iputs() and before it called
4140 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4141 * no one else to run iputs.
4143 * So wait for all ongoing ordered extents to complete and then run
4144 * delayed iputs. This works because once we reach this point no one
4145 * can either create new ordered extents nor create delayed iputs
4146 * through some other means.
4148 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4149 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4150 * but the delayed iput for the respective inode is made only when doing
4151 * the final btrfs_put_ordered_extent() (which must happen at
4152 * btrfs_finish_ordered_io() when we are unmounting).
4154 btrfs_flush_workqueue(fs_info->endio_write_workers);
4155 /* Ordered extents for free space inodes. */
4156 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4157 btrfs_run_delayed_iputs(fs_info);
4159 cancel_work_sync(&fs_info->async_reclaim_work);
4160 cancel_work_sync(&fs_info->async_data_reclaim_work);
4162 /* Cancel or finish ongoing discard work */
4163 btrfs_discard_cleanup(fs_info);
4165 if (!sb_rdonly(fs_info->sb)) {
4167 * The cleaner kthread is stopped, so do one final pass over
4168 * unused block groups.
4170 btrfs_delete_unused_bgs(fs_info);
4173 * There might be existing delayed inode workers still running
4174 * and holding an empty delayed inode item. We must wait for
4175 * them to complete first because they can create a transaction.
4176 * This happens when someone calls btrfs_balance_delayed_items()
4177 * and then a transaction commit runs the same delayed nodes
4178 * before any delayed worker has done something with the nodes.
4179 * We must wait for any worker here and not at transaction
4180 * commit time since that could cause a deadlock.
4181 * This is a very rare case.
4183 btrfs_flush_workqueue(fs_info->delayed_workers);
4185 ret = btrfs_commit_super(fs_info);
4187 btrfs_err(fs_info, "commit super ret %d", ret);
4190 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4191 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4192 btrfs_error_commit_super(fs_info);
4194 kthread_stop(fs_info->transaction_kthread);
4195 kthread_stop(fs_info->cleaner_kthread);
4197 ASSERT(list_empty(&fs_info->delayed_iputs));
4198 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4200 if (btrfs_check_quota_leak(fs_info)) {
4201 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4202 btrfs_err(fs_info, "qgroup reserved space leaked");
4205 btrfs_free_qgroup_config(fs_info);
4206 ASSERT(list_empty(&fs_info->delalloc_roots));
4208 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4209 btrfs_info(fs_info, "at unmount delalloc count %lld",
4210 percpu_counter_sum(&fs_info->delalloc_bytes));
4213 if (percpu_counter_sum(&fs_info->dio_bytes))
4214 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4215 percpu_counter_sum(&fs_info->dio_bytes));
4217 btrfs_sysfs_remove_mounted(fs_info);
4218 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4220 btrfs_put_block_group_cache(fs_info);
4223 * we must make sure there is not any read request to
4224 * submit after we stopping all workers.
4226 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4227 btrfs_stop_all_workers(fs_info);
4229 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4230 free_root_pointers(fs_info, true);
4231 btrfs_free_fs_roots(fs_info);
4234 * We must free the block groups after dropping the fs_roots as we could
4235 * have had an IO error and have left over tree log blocks that aren't
4236 * cleaned up until the fs roots are freed. This makes the block group
4237 * accounting appear to be wrong because there's pending reserved bytes,
4238 * so make sure we do the block group cleanup afterwards.
4240 btrfs_free_block_groups(fs_info);
4242 iput(fs_info->btree_inode);
4244 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4245 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4246 btrfsic_unmount(fs_info->fs_devices);
4249 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4250 btrfs_close_devices(fs_info->fs_devices);
4253 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4257 struct inode *btree_inode = buf->pages[0]->mapping->host;
4259 ret = extent_buffer_uptodate(buf);
4263 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4264 parent_transid, atomic);
4270 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4272 struct btrfs_fs_info *fs_info;
4273 struct btrfs_root *root;
4274 u64 transid = btrfs_header_generation(buf);
4277 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4279 * This is a fast path so only do this check if we have sanity tests
4280 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4281 * outside of the sanity tests.
4283 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4286 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4287 fs_info = root->fs_info;
4288 btrfs_assert_tree_locked(buf);
4289 if (transid != fs_info->generation)
4290 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4291 buf->start, transid, fs_info->generation);
4292 was_dirty = set_extent_buffer_dirty(buf);
4294 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4296 fs_info->dirty_metadata_batch);
4297 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4299 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4300 * but item data not updated.
4301 * So here we should only check item pointers, not item data.
4303 if (btrfs_header_level(buf) == 0 &&
4304 btrfs_check_leaf_relaxed(buf)) {
4305 btrfs_print_leaf(buf);
4311 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4315 * looks as though older kernels can get into trouble with
4316 * this code, they end up stuck in balance_dirty_pages forever
4320 if (current->flags & PF_MEMALLOC)
4324 btrfs_balance_delayed_items(fs_info);
4326 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4327 BTRFS_DIRTY_METADATA_THRESH,
4328 fs_info->dirty_metadata_batch);
4330 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4334 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4336 __btrfs_btree_balance_dirty(fs_info, 1);
4339 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4341 __btrfs_btree_balance_dirty(fs_info, 0);
4344 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4345 struct btrfs_key *first_key)
4347 return btree_read_extent_buffer_pages(buf, parent_transid,
4351 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4353 /* cleanup FS via transaction */
4354 btrfs_cleanup_transaction(fs_info);
4356 mutex_lock(&fs_info->cleaner_mutex);
4357 btrfs_run_delayed_iputs(fs_info);
4358 mutex_unlock(&fs_info->cleaner_mutex);
4360 down_write(&fs_info->cleanup_work_sem);
4361 up_write(&fs_info->cleanup_work_sem);
4364 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4366 struct btrfs_root *gang[8];
4367 u64 root_objectid = 0;
4370 spin_lock(&fs_info->fs_roots_radix_lock);
4371 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4372 (void **)gang, root_objectid,
4373 ARRAY_SIZE(gang))) != 0) {
4376 for (i = 0; i < ret; i++)
4377 gang[i] = btrfs_grab_root(gang[i]);
4378 spin_unlock(&fs_info->fs_roots_radix_lock);
4380 for (i = 0; i < ret; i++) {
4383 root_objectid = gang[i]->root_key.objectid;
4384 btrfs_free_log(NULL, gang[i]);
4385 btrfs_put_root(gang[i]);
4388 spin_lock(&fs_info->fs_roots_radix_lock);
4390 spin_unlock(&fs_info->fs_roots_radix_lock);
4391 btrfs_free_log_root_tree(NULL, fs_info);
4394 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4396 struct btrfs_ordered_extent *ordered;
4398 spin_lock(&root->ordered_extent_lock);
4400 * This will just short circuit the ordered completion stuff which will
4401 * make sure the ordered extent gets properly cleaned up.
4403 list_for_each_entry(ordered, &root->ordered_extents,
4405 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4406 spin_unlock(&root->ordered_extent_lock);
4409 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4411 struct btrfs_root *root;
4412 struct list_head splice;
4414 INIT_LIST_HEAD(&splice);
4416 spin_lock(&fs_info->ordered_root_lock);
4417 list_splice_init(&fs_info->ordered_roots, &splice);
4418 while (!list_empty(&splice)) {
4419 root = list_first_entry(&splice, struct btrfs_root,
4421 list_move_tail(&root->ordered_root,
4422 &fs_info->ordered_roots);
4424 spin_unlock(&fs_info->ordered_root_lock);
4425 btrfs_destroy_ordered_extents(root);
4428 spin_lock(&fs_info->ordered_root_lock);
4430 spin_unlock(&fs_info->ordered_root_lock);
4433 * We need this here because if we've been flipped read-only we won't
4434 * get sync() from the umount, so we need to make sure any ordered
4435 * extents that haven't had their dirty pages IO start writeout yet
4436 * actually get run and error out properly.
4438 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4441 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4442 struct btrfs_fs_info *fs_info)
4444 struct rb_node *node;
4445 struct btrfs_delayed_ref_root *delayed_refs;
4446 struct btrfs_delayed_ref_node *ref;
4449 delayed_refs = &trans->delayed_refs;
4451 spin_lock(&delayed_refs->lock);
4452 if (atomic_read(&delayed_refs->num_entries) == 0) {
4453 spin_unlock(&delayed_refs->lock);
4454 btrfs_debug(fs_info, "delayed_refs has NO entry");
4458 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4459 struct btrfs_delayed_ref_head *head;
4461 bool pin_bytes = false;
4463 head = rb_entry(node, struct btrfs_delayed_ref_head,
4465 if (btrfs_delayed_ref_lock(delayed_refs, head))
4468 spin_lock(&head->lock);
4469 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4470 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4473 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4474 RB_CLEAR_NODE(&ref->ref_node);
4475 if (!list_empty(&ref->add_list))
4476 list_del(&ref->add_list);
4477 atomic_dec(&delayed_refs->num_entries);
4478 btrfs_put_delayed_ref(ref);
4480 if (head->must_insert_reserved)
4482 btrfs_free_delayed_extent_op(head->extent_op);
4483 btrfs_delete_ref_head(delayed_refs, head);
4484 spin_unlock(&head->lock);
4485 spin_unlock(&delayed_refs->lock);
4486 mutex_unlock(&head->mutex);
4489 struct btrfs_block_group *cache;
4491 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4494 spin_lock(&cache->space_info->lock);
4495 spin_lock(&cache->lock);
4496 cache->pinned += head->num_bytes;
4497 btrfs_space_info_update_bytes_pinned(fs_info,
4498 cache->space_info, head->num_bytes);
4499 cache->reserved -= head->num_bytes;
4500 cache->space_info->bytes_reserved -= head->num_bytes;
4501 spin_unlock(&cache->lock);
4502 spin_unlock(&cache->space_info->lock);
4503 percpu_counter_add_batch(
4504 &cache->space_info->total_bytes_pinned,
4505 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4507 btrfs_put_block_group(cache);
4509 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4510 head->bytenr + head->num_bytes - 1);
4512 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4513 btrfs_put_delayed_ref_head(head);
4515 spin_lock(&delayed_refs->lock);
4517 btrfs_qgroup_destroy_extent_records(trans);
4519 spin_unlock(&delayed_refs->lock);
4524 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4526 struct btrfs_inode *btrfs_inode;
4527 struct list_head splice;
4529 INIT_LIST_HEAD(&splice);
4531 spin_lock(&root->delalloc_lock);
4532 list_splice_init(&root->delalloc_inodes, &splice);
4534 while (!list_empty(&splice)) {
4535 struct inode *inode = NULL;
4536 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4538 __btrfs_del_delalloc_inode(root, btrfs_inode);
4539 spin_unlock(&root->delalloc_lock);
4542 * Make sure we get a live inode and that it'll not disappear
4545 inode = igrab(&btrfs_inode->vfs_inode);
4547 unsigned int nofs_flag;
4549 nofs_flag = memalloc_nofs_save();
4550 invalidate_inode_pages2(inode->i_mapping);
4551 memalloc_nofs_restore(nofs_flag);
4554 spin_lock(&root->delalloc_lock);
4556 spin_unlock(&root->delalloc_lock);
4559 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4561 struct btrfs_root *root;
4562 struct list_head splice;
4564 INIT_LIST_HEAD(&splice);
4566 spin_lock(&fs_info->delalloc_root_lock);
4567 list_splice_init(&fs_info->delalloc_roots, &splice);
4568 while (!list_empty(&splice)) {
4569 root = list_first_entry(&splice, struct btrfs_root,
4571 root = btrfs_grab_root(root);
4573 spin_unlock(&fs_info->delalloc_root_lock);
4575 btrfs_destroy_delalloc_inodes(root);
4576 btrfs_put_root(root);
4578 spin_lock(&fs_info->delalloc_root_lock);
4580 spin_unlock(&fs_info->delalloc_root_lock);
4583 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4584 struct extent_io_tree *dirty_pages,
4588 struct extent_buffer *eb;
4593 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4598 clear_extent_bits(dirty_pages, start, end, mark);
4599 while (start <= end) {
4600 eb = find_extent_buffer(fs_info, start);
4601 start += fs_info->nodesize;
4604 wait_on_extent_buffer_writeback(eb);
4606 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4608 clear_extent_buffer_dirty(eb);
4609 free_extent_buffer_stale(eb);
4616 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4617 struct extent_io_tree *unpin)
4624 struct extent_state *cached_state = NULL;
4627 * The btrfs_finish_extent_commit() may get the same range as
4628 * ours between find_first_extent_bit and clear_extent_dirty.
4629 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4630 * the same extent range.
4632 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4633 ret = find_first_extent_bit(unpin, 0, &start, &end,
4634 EXTENT_DIRTY, &cached_state);
4636 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4640 clear_extent_dirty(unpin, start, end, &cached_state);
4641 free_extent_state(cached_state);
4642 btrfs_error_unpin_extent_range(fs_info, start, end);
4643 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4650 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4652 struct inode *inode;
4654 inode = cache->io_ctl.inode;
4656 unsigned int nofs_flag;
4658 nofs_flag = memalloc_nofs_save();
4659 invalidate_inode_pages2(inode->i_mapping);
4660 memalloc_nofs_restore(nofs_flag);
4662 BTRFS_I(inode)->generation = 0;
4663 cache->io_ctl.inode = NULL;
4666 ASSERT(cache->io_ctl.pages == NULL);
4667 btrfs_put_block_group(cache);
4670 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4671 struct btrfs_fs_info *fs_info)
4673 struct btrfs_block_group *cache;
4675 spin_lock(&cur_trans->dirty_bgs_lock);
4676 while (!list_empty(&cur_trans->dirty_bgs)) {
4677 cache = list_first_entry(&cur_trans->dirty_bgs,
4678 struct btrfs_block_group,
4681 if (!list_empty(&cache->io_list)) {
4682 spin_unlock(&cur_trans->dirty_bgs_lock);
4683 list_del_init(&cache->io_list);
4684 btrfs_cleanup_bg_io(cache);
4685 spin_lock(&cur_trans->dirty_bgs_lock);
4688 list_del_init(&cache->dirty_list);
4689 spin_lock(&cache->lock);
4690 cache->disk_cache_state = BTRFS_DC_ERROR;
4691 spin_unlock(&cache->lock);
4693 spin_unlock(&cur_trans->dirty_bgs_lock);
4694 btrfs_put_block_group(cache);
4695 btrfs_delayed_refs_rsv_release(fs_info, 1);
4696 spin_lock(&cur_trans->dirty_bgs_lock);
4698 spin_unlock(&cur_trans->dirty_bgs_lock);
4701 * Refer to the definition of io_bgs member for details why it's safe
4702 * to use it without any locking
4704 while (!list_empty(&cur_trans->io_bgs)) {
4705 cache = list_first_entry(&cur_trans->io_bgs,
4706 struct btrfs_block_group,
4709 list_del_init(&cache->io_list);
4710 spin_lock(&cache->lock);
4711 cache->disk_cache_state = BTRFS_DC_ERROR;
4712 spin_unlock(&cache->lock);
4713 btrfs_cleanup_bg_io(cache);
4717 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4718 struct btrfs_fs_info *fs_info)
4720 struct btrfs_device *dev, *tmp;
4722 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4723 ASSERT(list_empty(&cur_trans->dirty_bgs));
4724 ASSERT(list_empty(&cur_trans->io_bgs));
4726 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4728 list_del_init(&dev->post_commit_list);
4731 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4733 cur_trans->state = TRANS_STATE_COMMIT_START;
4734 wake_up(&fs_info->transaction_blocked_wait);
4736 cur_trans->state = TRANS_STATE_UNBLOCKED;
4737 wake_up(&fs_info->transaction_wait);
4739 btrfs_destroy_delayed_inodes(fs_info);
4741 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4743 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4745 cur_trans->state =TRANS_STATE_COMPLETED;
4746 wake_up(&cur_trans->commit_wait);
4749 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4751 struct btrfs_transaction *t;
4753 mutex_lock(&fs_info->transaction_kthread_mutex);
4755 spin_lock(&fs_info->trans_lock);
4756 while (!list_empty(&fs_info->trans_list)) {
4757 t = list_first_entry(&fs_info->trans_list,
4758 struct btrfs_transaction, list);
4759 if (t->state >= TRANS_STATE_COMMIT_START) {
4760 refcount_inc(&t->use_count);
4761 spin_unlock(&fs_info->trans_lock);
4762 btrfs_wait_for_commit(fs_info, t->transid);
4763 btrfs_put_transaction(t);
4764 spin_lock(&fs_info->trans_lock);
4767 if (t == fs_info->running_transaction) {
4768 t->state = TRANS_STATE_COMMIT_DOING;
4769 spin_unlock(&fs_info->trans_lock);
4771 * We wait for 0 num_writers since we don't hold a trans
4772 * handle open currently for this transaction.
4774 wait_event(t->writer_wait,
4775 atomic_read(&t->num_writers) == 0);
4777 spin_unlock(&fs_info->trans_lock);
4779 btrfs_cleanup_one_transaction(t, fs_info);
4781 spin_lock(&fs_info->trans_lock);
4782 if (t == fs_info->running_transaction)
4783 fs_info->running_transaction = NULL;
4784 list_del_init(&t->list);
4785 spin_unlock(&fs_info->trans_lock);
4787 btrfs_put_transaction(t);
4788 trace_btrfs_transaction_commit(fs_info->tree_root);
4789 spin_lock(&fs_info->trans_lock);
4791 spin_unlock(&fs_info->trans_lock);
4792 btrfs_destroy_all_ordered_extents(fs_info);
4793 btrfs_destroy_delayed_inodes(fs_info);
4794 btrfs_assert_delayed_root_empty(fs_info);
4795 btrfs_destroy_all_delalloc_inodes(fs_info);
4796 btrfs_drop_all_logs(fs_info);
4797 mutex_unlock(&fs_info->transaction_kthread_mutex);
4802 int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
4804 struct btrfs_path *path;
4806 struct extent_buffer *l;
4807 struct btrfs_key search_key;
4808 struct btrfs_key found_key;
4811 path = btrfs_alloc_path();
4815 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4816 search_key.type = -1;
4817 search_key.offset = (u64)-1;
4818 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4821 BUG_ON(ret == 0); /* Corruption */
4822 if (path->slots[0] > 0) {
4823 slot = path->slots[0] - 1;
4825 btrfs_item_key_to_cpu(l, &found_key, slot);
4826 *objectid = max_t(u64, found_key.objectid,
4827 BTRFS_FIRST_FREE_OBJECTID - 1);
4829 *objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
4833 btrfs_free_path(path);
4837 int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
4840 mutex_lock(&root->objectid_mutex);
4842 if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4843 btrfs_warn(root->fs_info,
4844 "the objectid of root %llu reaches its highest value",
4845 root->root_key.objectid);
4850 *objectid = ++root->highest_objectid;
4853 mutex_unlock(&root->objectid_mutex);