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 "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41 #include "block-group.h"
43 #include "space-info.h"
47 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
48 BTRFS_HEADER_FLAG_RELOC |\
49 BTRFS_SUPER_FLAG_ERROR |\
50 BTRFS_SUPER_FLAG_SEEDING |\
51 BTRFS_SUPER_FLAG_METADUMP |\
52 BTRFS_SUPER_FLAG_METADUMP_V2)
54 static void end_workqueue_fn(struct btrfs_work *work);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_fs_info *fs_info);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 struct extent_io_tree *dirty_pages,
62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
72 struct btrfs_end_io_wq {
76 struct btrfs_fs_info *info;
78 enum btrfs_wq_endio_type metadata;
79 struct btrfs_work work;
82 static struct kmem_cache *btrfs_end_io_wq_cache;
84 int __init btrfs_end_io_wq_init(void)
86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq),
91 if (!btrfs_end_io_wq_cache)
96 void __cold btrfs_end_io_wq_exit(void)
98 kmem_cache_destroy(btrfs_end_io_wq_cache);
101 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
103 if (fs_info->csum_shash)
104 crypto_free_shash(fs_info->csum_shash);
108 * async submit bios are used to offload expensive checksumming
109 * onto the worker threads. They checksum file and metadata bios
110 * just before they are sent down the IO stack.
112 struct async_submit_bio {
115 extent_submit_bio_start_t *submit_bio_start;
118 /* Optional parameter for submit_bio_start used by direct io */
120 struct btrfs_work work;
125 * Lockdep class keys for extent_buffer->lock's in this root. For a given
126 * eb, the lockdep key is determined by the btrfs_root it belongs to and
127 * the level the eb occupies in the tree.
129 * Different roots are used for different purposes and may nest inside each
130 * other and they require separate keysets. As lockdep keys should be
131 * static, assign keysets according to the purpose of the root as indicated
132 * by btrfs_root->root_key.objectid. This ensures that all special purpose
133 * roots have separate keysets.
135 * Lock-nesting across peer nodes is always done with the immediate parent
136 * node locked thus preventing deadlock. As lockdep doesn't know this, use
137 * subclass to avoid triggering lockdep warning in such cases.
139 * The key is set by the readpage_end_io_hook after the buffer has passed
140 * csum validation but before the pages are unlocked. It is also set by
141 * btrfs_init_new_buffer on freshly allocated blocks.
143 * We also add a check to make sure the highest level of the tree is the
144 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
145 * needs update as well.
147 #ifdef CONFIG_DEBUG_LOCK_ALLOC
148 # if BTRFS_MAX_LEVEL != 8
152 #define DEFINE_LEVEL(stem, level) \
153 .names[level] = "btrfs-" stem "-0" #level,
155 #define DEFINE_NAME(stem) \
156 DEFINE_LEVEL(stem, 0) \
157 DEFINE_LEVEL(stem, 1) \
158 DEFINE_LEVEL(stem, 2) \
159 DEFINE_LEVEL(stem, 3) \
160 DEFINE_LEVEL(stem, 4) \
161 DEFINE_LEVEL(stem, 5) \
162 DEFINE_LEVEL(stem, 6) \
163 DEFINE_LEVEL(stem, 7)
165 static struct btrfs_lockdep_keyset {
166 u64 id; /* root objectid */
167 /* Longest entry: btrfs-free-space-00 */
168 char names[BTRFS_MAX_LEVEL][20];
169 struct lock_class_key keys[BTRFS_MAX_LEVEL];
170 } btrfs_lockdep_keysets[] = {
171 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") },
172 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") },
173 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") },
174 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") },
175 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
176 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
177 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
178 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
179 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
180 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
181 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
182 { .id = 0, DEFINE_NAME("tree") },
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 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
214 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
218 shash->tfm = fs_info->csum_shash;
219 crypto_shash_init(shash);
220 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
221 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
222 first_page_part - BTRFS_CSUM_SIZE);
224 for (i = 1; i < num_pages; i++) {
225 kaddr = page_address(buf->pages[i]);
226 crypto_shash_update(shash, kaddr, PAGE_SIZE);
228 memset(result, 0, BTRFS_CSUM_SIZE);
229 crypto_shash_final(shash, result);
233 * we can't consider a given block up to date unless the transid of the
234 * block matches the transid in the parent node's pointer. This is how we
235 * detect blocks that either didn't get written at all or got written
236 * in the wrong place.
238 static int verify_parent_transid(struct extent_io_tree *io_tree,
239 struct extent_buffer *eb, u64 parent_transid,
242 struct extent_state *cached_state = NULL;
244 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
246 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
253 btrfs_tree_read_lock(eb);
255 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
257 if (extent_buffer_uptodate(eb) &&
258 btrfs_header_generation(eb) == parent_transid) {
262 btrfs_err_rl(eb->fs_info,
263 "parent transid verify failed on %llu wanted %llu found %llu",
265 parent_transid, btrfs_header_generation(eb));
269 * Things reading via commit roots that don't have normal protection,
270 * like send, can have a really old block in cache that may point at a
271 * block that has been freed and re-allocated. So don't clear uptodate
272 * if we find an eb that is under IO (dirty/writeback) because we could
273 * end up reading in the stale data and then writing it back out and
274 * making everybody very sad.
276 if (!extent_buffer_under_io(eb))
277 clear_extent_buffer_uptodate(eb);
279 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
282 btrfs_tree_read_unlock(eb);
286 static bool btrfs_supported_super_csum(u16 csum_type)
289 case BTRFS_CSUM_TYPE_CRC32:
290 case BTRFS_CSUM_TYPE_XXHASH:
291 case BTRFS_CSUM_TYPE_SHA256:
292 case BTRFS_CSUM_TYPE_BLAKE2:
300 * Return 0 if the superblock checksum type matches the checksum value of that
301 * algorithm. Pass the raw disk superblock data.
303 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
306 struct btrfs_super_block *disk_sb =
307 (struct btrfs_super_block *)raw_disk_sb;
308 char result[BTRFS_CSUM_SIZE];
309 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
311 shash->tfm = fs_info->csum_shash;
314 * The super_block structure does not span the whole
315 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
316 * filled with zeros and is included in the checksum.
318 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
319 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
321 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
327 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
328 struct btrfs_key *first_key, u64 parent_transid)
330 struct btrfs_fs_info *fs_info = eb->fs_info;
332 struct btrfs_key found_key;
335 found_level = btrfs_header_level(eb);
336 if (found_level != level) {
337 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
338 KERN_ERR "BTRFS: tree level check failed\n");
340 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
341 eb->start, level, found_level);
349 * For live tree block (new tree blocks in current transaction),
350 * we need proper lock context to avoid race, which is impossible here.
351 * So we only checks tree blocks which is read from disk, whose
352 * generation <= fs_info->last_trans_committed.
354 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
357 /* We have @first_key, so this @eb must have at least one item */
358 if (btrfs_header_nritems(eb) == 0) {
360 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
362 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
367 btrfs_node_key_to_cpu(eb, &found_key, 0);
369 btrfs_item_key_to_cpu(eb, &found_key, 0);
370 ret = btrfs_comp_cpu_keys(first_key, &found_key);
373 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
374 KERN_ERR "BTRFS: tree first key check failed\n");
376 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
377 eb->start, parent_transid, first_key->objectid,
378 first_key->type, first_key->offset,
379 found_key.objectid, found_key.type,
386 * helper to read a given tree block, doing retries as required when
387 * the checksums don't match and we have alternate mirrors to try.
389 * @parent_transid: expected transid, skip check if 0
390 * @level: expected level, mandatory check
391 * @first_key: expected key of first slot, skip check if NULL
393 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
394 u64 parent_transid, int level,
395 struct btrfs_key *first_key)
397 struct btrfs_fs_info *fs_info = eb->fs_info;
398 struct extent_io_tree *io_tree;
403 int failed_mirror = 0;
405 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
407 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
408 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
410 if (verify_parent_transid(io_tree, eb,
413 else if (btrfs_verify_level_key(eb, level,
414 first_key, parent_transid))
420 num_copies = btrfs_num_copies(fs_info,
425 if (!failed_mirror) {
427 failed_mirror = eb->read_mirror;
431 if (mirror_num == failed_mirror)
434 if (mirror_num > num_copies)
438 if (failed && !ret && failed_mirror)
439 btrfs_repair_eb_io_failure(eb, failed_mirror);
444 static int csum_one_extent_buffer(struct extent_buffer *eb)
446 struct btrfs_fs_info *fs_info = eb->fs_info;
447 u8 result[BTRFS_CSUM_SIZE];
450 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
451 offsetof(struct btrfs_header, fsid),
452 BTRFS_FSID_SIZE) == 0);
453 csum_tree_block(eb, result);
455 if (btrfs_header_level(eb))
456 ret = btrfs_check_node(eb);
458 ret = btrfs_check_leaf_full(eb);
461 btrfs_print_tree(eb, 0);
463 "block=%llu write time tree block corruption detected",
465 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
468 write_extent_buffer(eb, result, 0, fs_info->csum_size);
473 /* Checksum all dirty extent buffers in one bio_vec */
474 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
475 struct bio_vec *bvec)
477 struct page *page = bvec->bv_page;
478 u64 bvec_start = page_offset(page) + bvec->bv_offset;
482 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
483 cur += fs_info->nodesize) {
484 struct extent_buffer *eb;
487 eb = find_extent_buffer(fs_info, cur);
488 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
491 /* A dirty eb shouldn't disappear from buffer_radix */
495 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
496 free_extent_buffer(eb);
499 if (WARN_ON(!uptodate)) {
500 free_extent_buffer(eb);
504 ret = csum_one_extent_buffer(eb);
505 free_extent_buffer(eb);
513 * Checksum a dirty tree block before IO. This has extra checks to make sure
514 * we only fill in the checksum field in the first page of a multi-page block.
515 * For subpage extent buffers we need bvec to also read the offset in the page.
517 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
519 struct page *page = bvec->bv_page;
520 u64 start = page_offset(page);
522 struct extent_buffer *eb;
524 if (fs_info->sectorsize < PAGE_SIZE)
525 return csum_dirty_subpage_buffers(fs_info, bvec);
527 eb = (struct extent_buffer *)page->private;
528 if (page != eb->pages[0])
531 found_start = btrfs_header_bytenr(eb);
533 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
534 WARN_ON(found_start != 0);
539 * Please do not consolidate these warnings into a single if.
540 * It is useful to know what went wrong.
542 if (WARN_ON(found_start != start))
544 if (WARN_ON(!PageUptodate(page)))
547 return csum_one_extent_buffer(eb);
550 static int check_tree_block_fsid(struct extent_buffer *eb)
552 struct btrfs_fs_info *fs_info = eb->fs_info;
553 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
554 u8 fsid[BTRFS_FSID_SIZE];
557 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
560 * Checking the incompat flag is only valid for the current fs. For
561 * seed devices it's forbidden to have their uuid changed so reading
562 * ->fsid in this case is fine
564 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
565 metadata_uuid = fs_devices->metadata_uuid;
567 metadata_uuid = fs_devices->fsid;
569 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
572 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
573 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
579 /* Do basic extent buffer checks at read time */
580 static int validate_extent_buffer(struct extent_buffer *eb)
582 struct btrfs_fs_info *fs_info = eb->fs_info;
584 const u32 csum_size = fs_info->csum_size;
586 u8 result[BTRFS_CSUM_SIZE];
589 found_start = btrfs_header_bytenr(eb);
590 if (found_start != eb->start) {
591 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
592 eb->start, found_start);
596 if (check_tree_block_fsid(eb)) {
597 btrfs_err_rl(fs_info, "bad fsid on block %llu",
602 found_level = btrfs_header_level(eb);
603 if (found_level >= BTRFS_MAX_LEVEL) {
604 btrfs_err(fs_info, "bad tree block level %d on %llu",
605 (int)btrfs_header_level(eb), eb->start);
610 csum_tree_block(eb, result);
612 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
613 u8 val[BTRFS_CSUM_SIZE] = { 0 };
615 read_extent_buffer(eb, &val, 0, csum_size);
616 btrfs_warn_rl(fs_info,
617 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
618 fs_info->sb->s_id, eb->start,
619 CSUM_FMT_VALUE(csum_size, val),
620 CSUM_FMT_VALUE(csum_size, result),
621 btrfs_header_level(eb));
627 * If this is a leaf block and it is corrupt, set the corrupt bit so
628 * that we don't try and read the other copies of this block, just
631 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
632 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636 if (found_level > 0 && btrfs_check_node(eb))
640 set_extent_buffer_uptodate(eb);
643 "block=%llu read time tree block corruption detected",
649 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
652 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
653 struct extent_buffer *eb;
658 * We don't allow bio merge for subpage metadata read, so we should
659 * only get one eb for each endio hook.
661 ASSERT(end == start + fs_info->nodesize - 1);
662 ASSERT(PagePrivate(page));
664 eb = find_extent_buffer(fs_info, start);
666 * When we are reading one tree block, eb must have been inserted into
667 * the radix tree. If not, something is wrong.
671 reads_done = atomic_dec_and_test(&eb->io_pages);
672 /* Subpage read must finish in page read */
675 eb->read_mirror = mirror;
676 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
680 ret = validate_extent_buffer(eb);
684 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
685 btree_readahead_hook(eb, ret);
687 set_extent_buffer_uptodate(eb);
689 free_extent_buffer(eb);
693 * end_bio_extent_readpage decrements io_pages in case of error,
694 * make sure it has something to decrement.
696 atomic_inc(&eb->io_pages);
697 clear_extent_buffer_uptodate(eb);
698 free_extent_buffer(eb);
702 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
703 struct page *page, u64 start, u64 end,
706 struct extent_buffer *eb;
710 ASSERT(page->private);
712 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
713 return validate_subpage_buffer(page, start, end, mirror);
715 eb = (struct extent_buffer *)page->private;
718 * The pending IO might have been the only thing that kept this buffer
719 * in memory. Make sure we have a ref for all this other checks
721 atomic_inc(&eb->refs);
723 reads_done = atomic_dec_and_test(&eb->io_pages);
727 eb->read_mirror = mirror;
728 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
732 ret = validate_extent_buffer(eb);
735 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
736 btree_readahead_hook(eb, ret);
740 * our io error hook is going to dec the io pages
741 * again, we have to make sure it has something
744 atomic_inc(&eb->io_pages);
745 clear_extent_buffer_uptodate(eb);
747 free_extent_buffer(eb);
752 static void end_workqueue_bio(struct bio *bio)
754 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
755 struct btrfs_fs_info *fs_info;
756 struct btrfs_workqueue *wq;
758 fs_info = end_io_wq->info;
759 end_io_wq->status = bio->bi_status;
761 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
762 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
763 wq = fs_info->endio_meta_write_workers;
764 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
765 wq = fs_info->endio_freespace_worker;
766 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
767 wq = fs_info->endio_raid56_workers;
769 wq = fs_info->endio_write_workers;
771 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
772 wq = fs_info->endio_raid56_workers;
773 else if (end_io_wq->metadata)
774 wq = fs_info->endio_meta_workers;
776 wq = fs_info->endio_workers;
779 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
780 btrfs_queue_work(wq, &end_io_wq->work);
783 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
784 enum btrfs_wq_endio_type metadata)
786 struct btrfs_end_io_wq *end_io_wq;
788 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
790 return BLK_STS_RESOURCE;
792 end_io_wq->private = bio->bi_private;
793 end_io_wq->end_io = bio->bi_end_io;
794 end_io_wq->info = info;
795 end_io_wq->status = 0;
796 end_io_wq->bio = bio;
797 end_io_wq->metadata = metadata;
799 bio->bi_private = end_io_wq;
800 bio->bi_end_io = end_workqueue_bio;
804 static void run_one_async_start(struct btrfs_work *work)
806 struct async_submit_bio *async;
809 async = container_of(work, struct async_submit_bio, work);
810 ret = async->submit_bio_start(async->inode, async->bio,
811 async->dio_file_offset);
817 * In order to insert checksums into the metadata in large chunks, we wait
818 * until bio submission time. All the pages in the bio are checksummed and
819 * sums are attached onto the ordered extent record.
821 * At IO completion time the csums attached on the ordered extent record are
822 * inserted into the tree.
824 static void run_one_async_done(struct btrfs_work *work)
826 struct async_submit_bio *async;
830 async = container_of(work, struct async_submit_bio, work);
831 inode = async->inode;
833 /* If an error occurred we just want to clean up the bio and move on */
835 async->bio->bi_status = async->status;
836 bio_endio(async->bio);
841 * All of the bios that pass through here are from async helpers.
842 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
843 * This changes nothing when cgroups aren't in use.
845 async->bio->bi_opf |= REQ_CGROUP_PUNT;
846 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
848 async->bio->bi_status = ret;
849 bio_endio(async->bio);
853 static void run_one_async_free(struct btrfs_work *work)
855 struct async_submit_bio *async;
857 async = container_of(work, struct async_submit_bio, work);
861 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
862 int mirror_num, unsigned long bio_flags,
864 extent_submit_bio_start_t *submit_bio_start)
866 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
867 struct async_submit_bio *async;
869 async = kmalloc(sizeof(*async), GFP_NOFS);
871 return BLK_STS_RESOURCE;
873 async->inode = inode;
875 async->mirror_num = mirror_num;
876 async->submit_bio_start = submit_bio_start;
878 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
881 async->dio_file_offset = dio_file_offset;
885 if (op_is_sync(bio->bi_opf))
886 btrfs_set_work_high_priority(&async->work);
888 btrfs_queue_work(fs_info->workers, &async->work);
892 static blk_status_t btree_csum_one_bio(struct bio *bio)
894 struct bio_vec *bvec;
895 struct btrfs_root *root;
897 struct bvec_iter_all iter_all;
899 ASSERT(!bio_flagged(bio, BIO_CLONED));
900 bio_for_each_segment_all(bvec, bio, iter_all) {
901 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
902 ret = csum_dirty_buffer(root->fs_info, bvec);
907 return errno_to_blk_status(ret);
910 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
914 * when we're called for a write, we're already in the async
915 * submission context. Just jump into btrfs_map_bio
917 return btree_csum_one_bio(bio);
920 static int check_async_write(struct btrfs_fs_info *fs_info,
921 struct btrfs_inode *bi)
923 if (btrfs_is_zoned(fs_info))
925 if (atomic_read(&bi->sync_writers))
927 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
932 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
933 int mirror_num, unsigned long bio_flags)
935 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
936 int async = check_async_write(fs_info, BTRFS_I(inode));
939 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
941 * called for a read, do the setup so that checksum validation
942 * can happen in the async kernel threads
944 ret = btrfs_bio_wq_end_io(fs_info, bio,
945 BTRFS_WQ_ENDIO_METADATA);
948 ret = btrfs_map_bio(fs_info, bio, mirror_num);
950 ret = btree_csum_one_bio(bio);
953 ret = btrfs_map_bio(fs_info, bio, mirror_num);
956 * kthread helpers are used to submit writes so that
957 * checksumming can happen in parallel across all CPUs
959 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
960 0, btree_submit_bio_start);
968 bio->bi_status = ret;
973 #ifdef CONFIG_MIGRATION
974 static int btree_migratepage(struct address_space *mapping,
975 struct page *newpage, struct page *page,
976 enum migrate_mode mode)
979 * we can't safely write a btree page from here,
980 * we haven't done the locking hook
985 * Buffers may be managed in a filesystem specific way.
986 * We must have no buffers or drop them.
988 if (page_has_private(page) &&
989 !try_to_release_page(page, GFP_KERNEL))
991 return migrate_page(mapping, newpage, page, mode);
996 static int btree_writepages(struct address_space *mapping,
997 struct writeback_control *wbc)
999 struct btrfs_fs_info *fs_info;
1002 if (wbc->sync_mode == WB_SYNC_NONE) {
1004 if (wbc->for_kupdate)
1007 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1008 /* this is a bit racy, but that's ok */
1009 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1010 BTRFS_DIRTY_METADATA_THRESH,
1011 fs_info->dirty_metadata_batch);
1015 return btree_write_cache_pages(mapping, wbc);
1018 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1020 if (PageWriteback(page) || PageDirty(page))
1023 return try_release_extent_buffer(page);
1026 static void btree_invalidatepage(struct page *page, unsigned int offset,
1027 unsigned int length)
1029 struct extent_io_tree *tree;
1030 tree = &BTRFS_I(page->mapping->host)->io_tree;
1031 extent_invalidatepage(tree, page, offset);
1032 btree_releasepage(page, GFP_NOFS);
1033 if (PagePrivate(page)) {
1034 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1035 "page private not zero on page %llu",
1036 (unsigned long long)page_offset(page));
1037 detach_page_private(page);
1041 static int btree_set_page_dirty(struct page *page)
1044 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1045 struct btrfs_subpage *subpage;
1046 struct extent_buffer *eb;
1048 u64 page_start = page_offset(page);
1050 if (fs_info->sectorsize == PAGE_SIZE) {
1051 BUG_ON(!PagePrivate(page));
1052 eb = (struct extent_buffer *)page->private;
1054 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1055 BUG_ON(!atomic_read(&eb->refs));
1056 btrfs_assert_tree_locked(eb);
1057 return __set_page_dirty_nobuffers(page);
1059 ASSERT(PagePrivate(page) && page->private);
1060 subpage = (struct btrfs_subpage *)page->private;
1062 ASSERT(subpage->dirty_bitmap);
1063 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1064 unsigned long flags;
1066 u16 tmp = (1 << cur_bit);
1068 spin_lock_irqsave(&subpage->lock, flags);
1069 if (!(tmp & subpage->dirty_bitmap)) {
1070 spin_unlock_irqrestore(&subpage->lock, flags);
1074 spin_unlock_irqrestore(&subpage->lock, flags);
1075 cur = page_start + cur_bit * fs_info->sectorsize;
1077 eb = find_extent_buffer(fs_info, cur);
1079 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1080 ASSERT(atomic_read(&eb->refs));
1081 btrfs_assert_tree_locked(eb);
1082 free_extent_buffer(eb);
1084 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1087 return __set_page_dirty_nobuffers(page);
1090 static const struct address_space_operations btree_aops = {
1091 .writepages = btree_writepages,
1092 .releasepage = btree_releasepage,
1093 .invalidatepage = btree_invalidatepage,
1094 #ifdef CONFIG_MIGRATION
1095 .migratepage = btree_migratepage,
1097 .set_page_dirty = btree_set_page_dirty,
1100 struct extent_buffer *btrfs_find_create_tree_block(
1101 struct btrfs_fs_info *fs_info,
1102 u64 bytenr, u64 owner_root,
1105 if (btrfs_is_testing(fs_info))
1106 return alloc_test_extent_buffer(fs_info, bytenr);
1107 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1111 * Read tree block at logical address @bytenr and do variant basic but critical
1114 * @owner_root: the objectid of the root owner for this block.
1115 * @parent_transid: expected transid of this tree block, skip check if 0
1116 * @level: expected level, mandatory check
1117 * @first_key: expected key in slot 0, skip check if NULL
1119 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1120 u64 owner_root, u64 parent_transid,
1121 int level, struct btrfs_key *first_key)
1123 struct extent_buffer *buf = NULL;
1126 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1130 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1133 free_extent_buffer_stale(buf);
1134 return ERR_PTR(ret);
1140 void btrfs_clean_tree_block(struct extent_buffer *buf)
1142 struct btrfs_fs_info *fs_info = buf->fs_info;
1143 if (btrfs_header_generation(buf) ==
1144 fs_info->running_transaction->transid) {
1145 btrfs_assert_tree_locked(buf);
1147 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1148 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1150 fs_info->dirty_metadata_batch);
1151 clear_extent_buffer_dirty(buf);
1156 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1159 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1160 root->fs_info = fs_info;
1162 root->commit_root = NULL;
1164 root->orphan_cleanup_state = 0;
1166 root->last_trans = 0;
1167 root->free_objectid = 0;
1168 root->nr_delalloc_inodes = 0;
1169 root->nr_ordered_extents = 0;
1170 root->inode_tree = RB_ROOT;
1171 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1172 root->block_rsv = NULL;
1174 INIT_LIST_HEAD(&root->dirty_list);
1175 INIT_LIST_HEAD(&root->root_list);
1176 INIT_LIST_HEAD(&root->delalloc_inodes);
1177 INIT_LIST_HEAD(&root->delalloc_root);
1178 INIT_LIST_HEAD(&root->ordered_extents);
1179 INIT_LIST_HEAD(&root->ordered_root);
1180 INIT_LIST_HEAD(&root->reloc_dirty_list);
1181 INIT_LIST_HEAD(&root->logged_list[0]);
1182 INIT_LIST_HEAD(&root->logged_list[1]);
1183 spin_lock_init(&root->inode_lock);
1184 spin_lock_init(&root->delalloc_lock);
1185 spin_lock_init(&root->ordered_extent_lock);
1186 spin_lock_init(&root->accounting_lock);
1187 spin_lock_init(&root->log_extents_lock[0]);
1188 spin_lock_init(&root->log_extents_lock[1]);
1189 spin_lock_init(&root->qgroup_meta_rsv_lock);
1190 mutex_init(&root->objectid_mutex);
1191 mutex_init(&root->log_mutex);
1192 mutex_init(&root->ordered_extent_mutex);
1193 mutex_init(&root->delalloc_mutex);
1194 init_waitqueue_head(&root->qgroup_flush_wait);
1195 init_waitqueue_head(&root->log_writer_wait);
1196 init_waitqueue_head(&root->log_commit_wait[0]);
1197 init_waitqueue_head(&root->log_commit_wait[1]);
1198 INIT_LIST_HEAD(&root->log_ctxs[0]);
1199 INIT_LIST_HEAD(&root->log_ctxs[1]);
1200 atomic_set(&root->log_commit[0], 0);
1201 atomic_set(&root->log_commit[1], 0);
1202 atomic_set(&root->log_writers, 0);
1203 atomic_set(&root->log_batch, 0);
1204 refcount_set(&root->refs, 1);
1205 atomic_set(&root->snapshot_force_cow, 0);
1206 atomic_set(&root->nr_swapfiles, 0);
1207 root->log_transid = 0;
1208 root->log_transid_committed = -1;
1209 root->last_log_commit = 0;
1211 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1212 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1213 extent_io_tree_init(fs_info, &root->log_csum_range,
1214 IO_TREE_LOG_CSUM_RANGE, NULL);
1217 memset(&root->root_key, 0, sizeof(root->root_key));
1218 memset(&root->root_item, 0, sizeof(root->root_item));
1219 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1220 root->root_key.objectid = objectid;
1223 spin_lock_init(&root->root_item_lock);
1224 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1225 #ifdef CONFIG_BTRFS_DEBUG
1226 INIT_LIST_HEAD(&root->leak_list);
1227 spin_lock(&fs_info->fs_roots_radix_lock);
1228 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1229 spin_unlock(&fs_info->fs_roots_radix_lock);
1233 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1234 u64 objectid, gfp_t flags)
1236 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1238 __setup_root(root, fs_info, objectid);
1242 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1243 /* Should only be used by the testing infrastructure */
1244 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1246 struct btrfs_root *root;
1249 return ERR_PTR(-EINVAL);
1251 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1253 return ERR_PTR(-ENOMEM);
1255 /* We don't use the stripesize in selftest, set it as sectorsize */
1256 root->alloc_bytenr = 0;
1262 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1265 struct btrfs_fs_info *fs_info = trans->fs_info;
1266 struct extent_buffer *leaf;
1267 struct btrfs_root *tree_root = fs_info->tree_root;
1268 struct btrfs_root *root;
1269 struct btrfs_key key;
1270 unsigned int nofs_flag;
1274 * We're holding a transaction handle, so use a NOFS memory allocation
1275 * context to avoid deadlock if reclaim happens.
1277 nofs_flag = memalloc_nofs_save();
1278 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1279 memalloc_nofs_restore(nofs_flag);
1281 return ERR_PTR(-ENOMEM);
1283 root->root_key.objectid = objectid;
1284 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1285 root->root_key.offset = 0;
1287 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1288 BTRFS_NESTING_NORMAL);
1290 ret = PTR_ERR(leaf);
1296 btrfs_mark_buffer_dirty(leaf);
1298 root->commit_root = btrfs_root_node(root);
1299 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1301 btrfs_set_root_flags(&root->root_item, 0);
1302 btrfs_set_root_limit(&root->root_item, 0);
1303 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1304 btrfs_set_root_generation(&root->root_item, trans->transid);
1305 btrfs_set_root_level(&root->root_item, 0);
1306 btrfs_set_root_refs(&root->root_item, 1);
1307 btrfs_set_root_used(&root->root_item, leaf->len);
1308 btrfs_set_root_last_snapshot(&root->root_item, 0);
1309 btrfs_set_root_dirid(&root->root_item, 0);
1310 if (is_fstree(objectid))
1311 generate_random_guid(root->root_item.uuid);
1313 export_guid(root->root_item.uuid, &guid_null);
1314 btrfs_set_root_drop_level(&root->root_item, 0);
1316 btrfs_tree_unlock(leaf);
1318 key.objectid = objectid;
1319 key.type = BTRFS_ROOT_ITEM_KEY;
1321 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1329 btrfs_tree_unlock(leaf);
1331 btrfs_put_root(root);
1333 return ERR_PTR(ret);
1336 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1337 struct btrfs_fs_info *fs_info)
1339 struct btrfs_root *root;
1341 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1343 return ERR_PTR(-ENOMEM);
1345 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1346 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1347 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1352 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1353 struct btrfs_root *root)
1355 struct extent_buffer *leaf;
1358 * DON'T set SHAREABLE bit for log trees.
1360 * Log trees are not exposed to user space thus can't be snapshotted,
1361 * and they go away before a real commit is actually done.
1363 * They do store pointers to file data extents, and those reference
1364 * counts still get updated (along with back refs to the log tree).
1367 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1368 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1370 return PTR_ERR(leaf);
1374 btrfs_mark_buffer_dirty(root->node);
1375 btrfs_tree_unlock(root->node);
1380 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1381 struct btrfs_fs_info *fs_info)
1383 struct btrfs_root *log_root;
1385 log_root = alloc_log_tree(trans, fs_info);
1386 if (IS_ERR(log_root))
1387 return PTR_ERR(log_root);
1389 if (!btrfs_is_zoned(fs_info)) {
1390 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1393 btrfs_put_root(log_root);
1398 WARN_ON(fs_info->log_root_tree);
1399 fs_info->log_root_tree = log_root;
1403 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root)
1406 struct btrfs_fs_info *fs_info = root->fs_info;
1407 struct btrfs_root *log_root;
1408 struct btrfs_inode_item *inode_item;
1411 log_root = alloc_log_tree(trans, fs_info);
1412 if (IS_ERR(log_root))
1413 return PTR_ERR(log_root);
1415 ret = btrfs_alloc_log_tree_node(trans, log_root);
1417 btrfs_put_root(log_root);
1421 log_root->last_trans = trans->transid;
1422 log_root->root_key.offset = root->root_key.objectid;
1424 inode_item = &log_root->root_item.inode;
1425 btrfs_set_stack_inode_generation(inode_item, 1);
1426 btrfs_set_stack_inode_size(inode_item, 3);
1427 btrfs_set_stack_inode_nlink(inode_item, 1);
1428 btrfs_set_stack_inode_nbytes(inode_item,
1430 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1432 btrfs_set_root_node(&log_root->root_item, log_root->node);
1434 WARN_ON(root->log_root);
1435 root->log_root = log_root;
1436 root->log_transid = 0;
1437 root->log_transid_committed = -1;
1438 root->last_log_commit = 0;
1442 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1443 struct btrfs_path *path,
1444 struct btrfs_key *key)
1446 struct btrfs_root *root;
1447 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1452 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1454 return ERR_PTR(-ENOMEM);
1456 ret = btrfs_find_root(tree_root, key, path,
1457 &root->root_item, &root->root_key);
1464 generation = btrfs_root_generation(&root->root_item);
1465 level = btrfs_root_level(&root->root_item);
1466 root->node = read_tree_block(fs_info,
1467 btrfs_root_bytenr(&root->root_item),
1468 key->objectid, generation, level, NULL);
1469 if (IS_ERR(root->node)) {
1470 ret = PTR_ERR(root->node);
1473 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1477 root->commit_root = btrfs_root_node(root);
1480 btrfs_put_root(root);
1481 return ERR_PTR(ret);
1484 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1485 struct btrfs_key *key)
1487 struct btrfs_root *root;
1488 struct btrfs_path *path;
1490 path = btrfs_alloc_path();
1492 return ERR_PTR(-ENOMEM);
1493 root = read_tree_root_path(tree_root, path, key);
1494 btrfs_free_path(path);
1500 * Initialize subvolume root in-memory structure
1502 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1504 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1507 unsigned int nofs_flag;
1510 * We might be called under a transaction (e.g. indirect backref
1511 * resolution) which could deadlock if it triggers memory reclaim
1513 nofs_flag = memalloc_nofs_save();
1514 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1515 memalloc_nofs_restore(nofs_flag);
1519 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1520 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1521 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1522 btrfs_check_and_init_root_item(&root->root_item);
1526 * Don't assign anonymous block device to roots that are not exposed to
1527 * userspace, the id pool is limited to 1M
1529 if (is_fstree(root->root_key.objectid) &&
1530 btrfs_root_refs(&root->root_item) > 0) {
1532 ret = get_anon_bdev(&root->anon_dev);
1536 root->anon_dev = anon_dev;
1540 mutex_lock(&root->objectid_mutex);
1541 ret = btrfs_init_root_free_objectid(root);
1543 mutex_unlock(&root->objectid_mutex);
1547 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1549 mutex_unlock(&root->objectid_mutex);
1553 /* The caller is responsible to call btrfs_free_fs_root */
1557 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1560 struct btrfs_root *root;
1562 spin_lock(&fs_info->fs_roots_radix_lock);
1563 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1564 (unsigned long)root_id);
1566 root = btrfs_grab_root(root);
1567 spin_unlock(&fs_info->fs_roots_radix_lock);
1571 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1574 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1575 return btrfs_grab_root(fs_info->tree_root);
1576 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1577 return btrfs_grab_root(fs_info->extent_root);
1578 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1579 return btrfs_grab_root(fs_info->chunk_root);
1580 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1581 return btrfs_grab_root(fs_info->dev_root);
1582 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1583 return btrfs_grab_root(fs_info->csum_root);
1584 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1585 return btrfs_grab_root(fs_info->quota_root) ?
1586 fs_info->quota_root : ERR_PTR(-ENOENT);
1587 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1588 return btrfs_grab_root(fs_info->uuid_root) ?
1589 fs_info->uuid_root : ERR_PTR(-ENOENT);
1590 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1591 return btrfs_grab_root(fs_info->free_space_root) ?
1592 fs_info->free_space_root : ERR_PTR(-ENOENT);
1596 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1597 struct btrfs_root *root)
1601 ret = radix_tree_preload(GFP_NOFS);
1605 spin_lock(&fs_info->fs_roots_radix_lock);
1606 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1607 (unsigned long)root->root_key.objectid,
1610 btrfs_grab_root(root);
1611 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1613 spin_unlock(&fs_info->fs_roots_radix_lock);
1614 radix_tree_preload_end();
1619 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1621 #ifdef CONFIG_BTRFS_DEBUG
1622 struct btrfs_root *root;
1624 while (!list_empty(&fs_info->allocated_roots)) {
1625 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1627 root = list_first_entry(&fs_info->allocated_roots,
1628 struct btrfs_root, leak_list);
1629 btrfs_err(fs_info, "leaked root %s refcount %d",
1630 btrfs_root_name(&root->root_key, buf),
1631 refcount_read(&root->refs));
1632 while (refcount_read(&root->refs) > 1)
1633 btrfs_put_root(root);
1634 btrfs_put_root(root);
1639 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1641 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1642 percpu_counter_destroy(&fs_info->delalloc_bytes);
1643 percpu_counter_destroy(&fs_info->ordered_bytes);
1644 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1645 btrfs_free_csum_hash(fs_info);
1646 btrfs_free_stripe_hash_table(fs_info);
1647 btrfs_free_ref_cache(fs_info);
1648 kfree(fs_info->balance_ctl);
1649 kfree(fs_info->delayed_root);
1650 btrfs_put_root(fs_info->extent_root);
1651 btrfs_put_root(fs_info->tree_root);
1652 btrfs_put_root(fs_info->chunk_root);
1653 btrfs_put_root(fs_info->dev_root);
1654 btrfs_put_root(fs_info->csum_root);
1655 btrfs_put_root(fs_info->quota_root);
1656 btrfs_put_root(fs_info->uuid_root);
1657 btrfs_put_root(fs_info->free_space_root);
1658 btrfs_put_root(fs_info->fs_root);
1659 btrfs_put_root(fs_info->data_reloc_root);
1660 btrfs_check_leaked_roots(fs_info);
1661 btrfs_extent_buffer_leak_debug_check(fs_info);
1662 kfree(fs_info->super_copy);
1663 kfree(fs_info->super_for_commit);
1669 * Get an in-memory reference of a root structure.
1671 * For essential trees like root/extent tree, we grab it from fs_info directly.
1672 * For subvolume trees, we check the cached filesystem roots first. If not
1673 * found, then read it from disk and add it to cached fs roots.
1675 * Caller should release the root by calling btrfs_put_root() after the usage.
1677 * NOTE: Reloc and log trees can't be read by this function as they share the
1678 * same root objectid.
1680 * @objectid: root id
1681 * @anon_dev: preallocated anonymous block device number for new roots,
1682 * pass 0 for new allocation.
1683 * @check_ref: whether to check root item references, If true, return -ENOENT
1686 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1687 u64 objectid, dev_t anon_dev,
1690 struct btrfs_root *root;
1691 struct btrfs_path *path;
1692 struct btrfs_key key;
1695 root = btrfs_get_global_root(fs_info, objectid);
1699 root = btrfs_lookup_fs_root(fs_info, objectid);
1701 /* Shouldn't get preallocated anon_dev for cached roots */
1703 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1704 btrfs_put_root(root);
1705 return ERR_PTR(-ENOENT);
1710 key.objectid = objectid;
1711 key.type = BTRFS_ROOT_ITEM_KEY;
1712 key.offset = (u64)-1;
1713 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1717 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1722 ret = btrfs_init_fs_root(root, anon_dev);
1726 path = btrfs_alloc_path();
1731 key.objectid = BTRFS_ORPHAN_OBJECTID;
1732 key.type = BTRFS_ORPHAN_ITEM_KEY;
1733 key.offset = objectid;
1735 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1736 btrfs_free_path(path);
1740 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1742 ret = btrfs_insert_fs_root(fs_info, root);
1744 btrfs_put_root(root);
1751 btrfs_put_root(root);
1752 return ERR_PTR(ret);
1756 * Get in-memory reference of a root structure
1758 * @objectid: tree objectid
1759 * @check_ref: if set, verify that the tree exists and the item has at least
1762 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1763 u64 objectid, bool check_ref)
1765 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1769 * Get in-memory reference of a root structure, created as new, optionally pass
1770 * the anonymous block device id
1772 * @objectid: tree objectid
1773 * @anon_dev: if zero, allocate a new anonymous block device or use the
1776 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1777 u64 objectid, dev_t anon_dev)
1779 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1783 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1784 * @fs_info: the fs_info
1785 * @objectid: the objectid we need to lookup
1787 * This is exclusively used for backref walking, and exists specifically because
1788 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1789 * creation time, which means we may have to read the tree_root in order to look
1790 * up a fs root that is not in memory. If the root is not in memory we will
1791 * read the tree root commit root and look up the fs root from there. This is a
1792 * temporary root, it will not be inserted into the radix tree as it doesn't
1793 * have the most uptodate information, it'll simply be discarded once the
1794 * backref code is finished using the root.
1796 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1797 struct btrfs_path *path,
1800 struct btrfs_root *root;
1801 struct btrfs_key key;
1803 ASSERT(path->search_commit_root && path->skip_locking);
1806 * This can return -ENOENT if we ask for a root that doesn't exist, but
1807 * since this is called via the backref walking code we won't be looking
1808 * up a root that doesn't exist, unless there's corruption. So if root
1809 * != NULL just return it.
1811 root = btrfs_get_global_root(fs_info, objectid);
1815 root = btrfs_lookup_fs_root(fs_info, objectid);
1819 key.objectid = objectid;
1820 key.type = BTRFS_ROOT_ITEM_KEY;
1821 key.offset = (u64)-1;
1822 root = read_tree_root_path(fs_info->tree_root, path, &key);
1823 btrfs_release_path(path);
1829 * called by the kthread helper functions to finally call the bio end_io
1830 * functions. This is where read checksum verification actually happens
1832 static void end_workqueue_fn(struct btrfs_work *work)
1835 struct btrfs_end_io_wq *end_io_wq;
1837 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1838 bio = end_io_wq->bio;
1840 bio->bi_status = end_io_wq->status;
1841 bio->bi_private = end_io_wq->private;
1842 bio->bi_end_io = end_io_wq->end_io;
1844 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1847 static int cleaner_kthread(void *arg)
1849 struct btrfs_root *root = arg;
1850 struct btrfs_fs_info *fs_info = root->fs_info;
1856 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1858 /* Make the cleaner go to sleep early. */
1859 if (btrfs_need_cleaner_sleep(fs_info))
1863 * Do not do anything if we might cause open_ctree() to block
1864 * before we have finished mounting the filesystem.
1866 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1869 if (!mutex_trylock(&fs_info->cleaner_mutex))
1873 * Avoid the problem that we change the status of the fs
1874 * during the above check and trylock.
1876 if (btrfs_need_cleaner_sleep(fs_info)) {
1877 mutex_unlock(&fs_info->cleaner_mutex);
1881 btrfs_run_delayed_iputs(fs_info);
1883 again = btrfs_clean_one_deleted_snapshot(root);
1884 mutex_unlock(&fs_info->cleaner_mutex);
1887 * The defragger has dealt with the R/O remount and umount,
1888 * needn't do anything special here.
1890 btrfs_run_defrag_inodes(fs_info);
1893 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1894 * with relocation (btrfs_relocate_chunk) and relocation
1895 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1896 * after acquiring fs_info->reclaim_bgs_lock. So we
1897 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1898 * unused block groups.
1900 btrfs_delete_unused_bgs(fs_info);
1903 * Reclaim block groups in the reclaim_bgs list after we deleted
1904 * all unused block_groups. This possibly gives us some more free
1907 btrfs_reclaim_bgs(fs_info);
1909 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1910 if (kthread_should_park())
1912 if (kthread_should_stop())
1915 set_current_state(TASK_INTERRUPTIBLE);
1917 __set_current_state(TASK_RUNNING);
1922 static int transaction_kthread(void *arg)
1924 struct btrfs_root *root = arg;
1925 struct btrfs_fs_info *fs_info = root->fs_info;
1926 struct btrfs_trans_handle *trans;
1927 struct btrfs_transaction *cur;
1930 unsigned long delay;
1934 cannot_commit = false;
1935 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1936 mutex_lock(&fs_info->transaction_kthread_mutex);
1938 spin_lock(&fs_info->trans_lock);
1939 cur = fs_info->running_transaction;
1941 spin_unlock(&fs_info->trans_lock);
1945 delta = ktime_get_seconds() - cur->start_time;
1946 if (cur->state < TRANS_STATE_COMMIT_START &&
1947 delta < fs_info->commit_interval) {
1948 spin_unlock(&fs_info->trans_lock);
1949 delay -= msecs_to_jiffies((delta - 1) * 1000);
1951 msecs_to_jiffies(fs_info->commit_interval * 1000));
1954 transid = cur->transid;
1955 spin_unlock(&fs_info->trans_lock);
1957 /* If the file system is aborted, this will always fail. */
1958 trans = btrfs_attach_transaction(root);
1959 if (IS_ERR(trans)) {
1960 if (PTR_ERR(trans) != -ENOENT)
1961 cannot_commit = true;
1964 if (transid == trans->transid) {
1965 btrfs_commit_transaction(trans);
1967 btrfs_end_transaction(trans);
1970 wake_up_process(fs_info->cleaner_kthread);
1971 mutex_unlock(&fs_info->transaction_kthread_mutex);
1973 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1974 &fs_info->fs_state)))
1975 btrfs_cleanup_transaction(fs_info);
1976 if (!kthread_should_stop() &&
1977 (!btrfs_transaction_blocked(fs_info) ||
1979 schedule_timeout_interruptible(delay);
1980 } while (!kthread_should_stop());
1985 * This will find the highest generation in the array of root backups. The
1986 * index of the highest array is returned, or -EINVAL if we can't find
1989 * We check to make sure the array is valid by comparing the
1990 * generation of the latest root in the array with the generation
1991 * in the super block. If they don't match we pitch it.
1993 static int find_newest_super_backup(struct btrfs_fs_info *info)
1995 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1997 struct btrfs_root_backup *root_backup;
2000 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2001 root_backup = info->super_copy->super_roots + i;
2002 cur = btrfs_backup_tree_root_gen(root_backup);
2003 if (cur == newest_gen)
2011 * copy all the root pointers into the super backup array.
2012 * this will bump the backup pointer by one when it is
2015 static void backup_super_roots(struct btrfs_fs_info *info)
2017 const int next_backup = info->backup_root_index;
2018 struct btrfs_root_backup *root_backup;
2020 root_backup = info->super_for_commit->super_roots + next_backup;
2023 * make sure all of our padding and empty slots get zero filled
2024 * regardless of which ones we use today
2026 memset(root_backup, 0, sizeof(*root_backup));
2028 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2030 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2031 btrfs_set_backup_tree_root_gen(root_backup,
2032 btrfs_header_generation(info->tree_root->node));
2034 btrfs_set_backup_tree_root_level(root_backup,
2035 btrfs_header_level(info->tree_root->node));
2037 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2038 btrfs_set_backup_chunk_root_gen(root_backup,
2039 btrfs_header_generation(info->chunk_root->node));
2040 btrfs_set_backup_chunk_root_level(root_backup,
2041 btrfs_header_level(info->chunk_root->node));
2043 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2044 btrfs_set_backup_extent_root_gen(root_backup,
2045 btrfs_header_generation(info->extent_root->node));
2046 btrfs_set_backup_extent_root_level(root_backup,
2047 btrfs_header_level(info->extent_root->node));
2050 * we might commit during log recovery, which happens before we set
2051 * the fs_root. Make sure it is valid before we fill it in.
2053 if (info->fs_root && info->fs_root->node) {
2054 btrfs_set_backup_fs_root(root_backup,
2055 info->fs_root->node->start);
2056 btrfs_set_backup_fs_root_gen(root_backup,
2057 btrfs_header_generation(info->fs_root->node));
2058 btrfs_set_backup_fs_root_level(root_backup,
2059 btrfs_header_level(info->fs_root->node));
2062 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2063 btrfs_set_backup_dev_root_gen(root_backup,
2064 btrfs_header_generation(info->dev_root->node));
2065 btrfs_set_backup_dev_root_level(root_backup,
2066 btrfs_header_level(info->dev_root->node));
2068 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2069 btrfs_set_backup_csum_root_gen(root_backup,
2070 btrfs_header_generation(info->csum_root->node));
2071 btrfs_set_backup_csum_root_level(root_backup,
2072 btrfs_header_level(info->csum_root->node));
2074 btrfs_set_backup_total_bytes(root_backup,
2075 btrfs_super_total_bytes(info->super_copy));
2076 btrfs_set_backup_bytes_used(root_backup,
2077 btrfs_super_bytes_used(info->super_copy));
2078 btrfs_set_backup_num_devices(root_backup,
2079 btrfs_super_num_devices(info->super_copy));
2082 * if we don't copy this out to the super_copy, it won't get remembered
2083 * for the next commit
2085 memcpy(&info->super_copy->super_roots,
2086 &info->super_for_commit->super_roots,
2087 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2091 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2092 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2094 * fs_info - filesystem whose backup roots need to be read
2095 * priority - priority of backup root required
2097 * Returns backup root index on success and -EINVAL otherwise.
2099 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2101 int backup_index = find_newest_super_backup(fs_info);
2102 struct btrfs_super_block *super = fs_info->super_copy;
2103 struct btrfs_root_backup *root_backup;
2105 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2107 return backup_index;
2109 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2110 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2115 root_backup = super->super_roots + backup_index;
2117 btrfs_set_super_generation(super,
2118 btrfs_backup_tree_root_gen(root_backup));
2119 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2120 btrfs_set_super_root_level(super,
2121 btrfs_backup_tree_root_level(root_backup));
2122 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2125 * Fixme: the total bytes and num_devices need to match or we should
2128 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2129 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2131 return backup_index;
2134 /* helper to cleanup workers */
2135 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2137 btrfs_destroy_workqueue(fs_info->fixup_workers);
2138 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2139 btrfs_destroy_workqueue(fs_info->workers);
2140 btrfs_destroy_workqueue(fs_info->endio_workers);
2141 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2142 btrfs_destroy_workqueue(fs_info->rmw_workers);
2143 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2144 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2145 btrfs_destroy_workqueue(fs_info->delayed_workers);
2146 btrfs_destroy_workqueue(fs_info->caching_workers);
2147 btrfs_destroy_workqueue(fs_info->readahead_workers);
2148 btrfs_destroy_workqueue(fs_info->flush_workers);
2149 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2150 if (fs_info->discard_ctl.discard_workers)
2151 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2153 * Now that all other work queues are destroyed, we can safely destroy
2154 * the queues used for metadata I/O, since tasks from those other work
2155 * queues can do metadata I/O operations.
2157 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2158 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2161 static void free_root_extent_buffers(struct btrfs_root *root)
2164 free_extent_buffer(root->node);
2165 free_extent_buffer(root->commit_root);
2167 root->commit_root = NULL;
2171 /* helper to cleanup tree roots */
2172 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2174 free_root_extent_buffers(info->tree_root);
2176 free_root_extent_buffers(info->dev_root);
2177 free_root_extent_buffers(info->extent_root);
2178 free_root_extent_buffers(info->csum_root);
2179 free_root_extent_buffers(info->quota_root);
2180 free_root_extent_buffers(info->uuid_root);
2181 free_root_extent_buffers(info->fs_root);
2182 free_root_extent_buffers(info->data_reloc_root);
2183 if (free_chunk_root)
2184 free_root_extent_buffers(info->chunk_root);
2185 free_root_extent_buffers(info->free_space_root);
2188 void btrfs_put_root(struct btrfs_root *root)
2193 if (refcount_dec_and_test(&root->refs)) {
2194 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2195 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2197 free_anon_bdev(root->anon_dev);
2198 btrfs_drew_lock_destroy(&root->snapshot_lock);
2199 free_root_extent_buffers(root);
2200 #ifdef CONFIG_BTRFS_DEBUG
2201 spin_lock(&root->fs_info->fs_roots_radix_lock);
2202 list_del_init(&root->leak_list);
2203 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2209 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2212 struct btrfs_root *gang[8];
2215 while (!list_empty(&fs_info->dead_roots)) {
2216 gang[0] = list_entry(fs_info->dead_roots.next,
2217 struct btrfs_root, root_list);
2218 list_del(&gang[0]->root_list);
2220 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2221 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2222 btrfs_put_root(gang[0]);
2226 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2231 for (i = 0; i < ret; i++)
2232 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2236 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2238 mutex_init(&fs_info->scrub_lock);
2239 atomic_set(&fs_info->scrubs_running, 0);
2240 atomic_set(&fs_info->scrub_pause_req, 0);
2241 atomic_set(&fs_info->scrubs_paused, 0);
2242 atomic_set(&fs_info->scrub_cancel_req, 0);
2243 init_waitqueue_head(&fs_info->scrub_pause_wait);
2244 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2247 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2249 spin_lock_init(&fs_info->balance_lock);
2250 mutex_init(&fs_info->balance_mutex);
2251 atomic_set(&fs_info->balance_pause_req, 0);
2252 atomic_set(&fs_info->balance_cancel_req, 0);
2253 fs_info->balance_ctl = NULL;
2254 init_waitqueue_head(&fs_info->balance_wait_q);
2257 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2259 struct inode *inode = fs_info->btree_inode;
2261 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2262 set_nlink(inode, 1);
2264 * we set the i_size on the btree inode to the max possible int.
2265 * the real end of the address space is determined by all of
2266 * the devices in the system
2268 inode->i_size = OFFSET_MAX;
2269 inode->i_mapping->a_ops = &btree_aops;
2271 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2272 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2273 IO_TREE_BTREE_INODE_IO, inode);
2274 BTRFS_I(inode)->io_tree.track_uptodate = false;
2275 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2277 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2278 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2279 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2280 btrfs_insert_inode_hash(inode);
2283 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2285 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2286 init_rwsem(&fs_info->dev_replace.rwsem);
2287 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2290 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2292 spin_lock_init(&fs_info->qgroup_lock);
2293 mutex_init(&fs_info->qgroup_ioctl_lock);
2294 fs_info->qgroup_tree = RB_ROOT;
2295 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2296 fs_info->qgroup_seq = 1;
2297 fs_info->qgroup_ulist = NULL;
2298 fs_info->qgroup_rescan_running = false;
2299 mutex_init(&fs_info->qgroup_rescan_lock);
2302 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2303 struct btrfs_fs_devices *fs_devices)
2305 u32 max_active = fs_info->thread_pool_size;
2306 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2309 btrfs_alloc_workqueue(fs_info, "worker",
2310 flags | WQ_HIGHPRI, max_active, 16);
2312 fs_info->delalloc_workers =
2313 btrfs_alloc_workqueue(fs_info, "delalloc",
2314 flags, max_active, 2);
2316 fs_info->flush_workers =
2317 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2318 flags, max_active, 0);
2320 fs_info->caching_workers =
2321 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2323 fs_info->fixup_workers =
2324 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2327 * endios are largely parallel and should have a very
2330 fs_info->endio_workers =
2331 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2332 fs_info->endio_meta_workers =
2333 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2335 fs_info->endio_meta_write_workers =
2336 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2338 fs_info->endio_raid56_workers =
2339 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2341 fs_info->rmw_workers =
2342 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2343 fs_info->endio_write_workers =
2344 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2346 fs_info->endio_freespace_worker =
2347 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2349 fs_info->delayed_workers =
2350 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2352 fs_info->readahead_workers =
2353 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2355 fs_info->qgroup_rescan_workers =
2356 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2357 fs_info->discard_ctl.discard_workers =
2358 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2360 if (!(fs_info->workers && fs_info->delalloc_workers &&
2361 fs_info->flush_workers &&
2362 fs_info->endio_workers && fs_info->endio_meta_workers &&
2363 fs_info->endio_meta_write_workers &&
2364 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2365 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2366 fs_info->caching_workers && fs_info->readahead_workers &&
2367 fs_info->fixup_workers && fs_info->delayed_workers &&
2368 fs_info->qgroup_rescan_workers &&
2369 fs_info->discard_ctl.discard_workers)) {
2376 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2378 struct crypto_shash *csum_shash;
2379 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2381 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2383 if (IS_ERR(csum_shash)) {
2384 btrfs_err(fs_info, "error allocating %s hash for checksum",
2386 return PTR_ERR(csum_shash);
2389 fs_info->csum_shash = csum_shash;
2394 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2395 struct btrfs_fs_devices *fs_devices)
2398 struct btrfs_root *log_tree_root;
2399 struct btrfs_super_block *disk_super = fs_info->super_copy;
2400 u64 bytenr = btrfs_super_log_root(disk_super);
2401 int level = btrfs_super_log_root_level(disk_super);
2403 if (fs_devices->rw_devices == 0) {
2404 btrfs_warn(fs_info, "log replay required on RO media");
2408 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2413 log_tree_root->node = read_tree_block(fs_info, bytenr,
2414 BTRFS_TREE_LOG_OBJECTID,
2415 fs_info->generation + 1, level,
2417 if (IS_ERR(log_tree_root->node)) {
2418 btrfs_warn(fs_info, "failed to read log tree");
2419 ret = PTR_ERR(log_tree_root->node);
2420 log_tree_root->node = NULL;
2421 btrfs_put_root(log_tree_root);
2423 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2424 btrfs_err(fs_info, "failed to read log tree");
2425 btrfs_put_root(log_tree_root);
2428 /* returns with log_tree_root freed on success */
2429 ret = btrfs_recover_log_trees(log_tree_root);
2431 btrfs_handle_fs_error(fs_info, ret,
2432 "Failed to recover log tree");
2433 btrfs_put_root(log_tree_root);
2437 if (sb_rdonly(fs_info->sb)) {
2438 ret = btrfs_commit_super(fs_info);
2446 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2448 struct btrfs_root *tree_root = fs_info->tree_root;
2449 struct btrfs_root *root;
2450 struct btrfs_key location;
2453 BUG_ON(!fs_info->tree_root);
2455 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2456 location.type = BTRFS_ROOT_ITEM_KEY;
2457 location.offset = 0;
2459 root = btrfs_read_tree_root(tree_root, &location);
2461 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2462 ret = PTR_ERR(root);
2466 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2467 fs_info->extent_root = root;
2470 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2471 root = btrfs_read_tree_root(tree_root, &location);
2473 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2474 ret = PTR_ERR(root);
2478 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2479 fs_info->dev_root = root;
2481 /* Initialize fs_info for all devices in any case */
2482 btrfs_init_devices_late(fs_info);
2484 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2485 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2486 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2487 root = btrfs_read_tree_root(tree_root, &location);
2489 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2490 ret = PTR_ERR(root);
2494 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2495 fs_info->csum_root = root;
2500 * This tree can share blocks with some other fs tree during relocation
2501 * and we need a proper setup by btrfs_get_fs_root
2503 root = btrfs_get_fs_root(tree_root->fs_info,
2504 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2506 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2507 ret = PTR_ERR(root);
2511 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2512 fs_info->data_reloc_root = root;
2515 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2516 root = btrfs_read_tree_root(tree_root, &location);
2517 if (!IS_ERR(root)) {
2518 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2519 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2520 fs_info->quota_root = root;
2523 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2524 root = btrfs_read_tree_root(tree_root, &location);
2526 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2527 ret = PTR_ERR(root);
2532 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2533 fs_info->uuid_root = root;
2536 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2537 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2538 root = btrfs_read_tree_root(tree_root, &location);
2540 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2541 ret = PTR_ERR(root);
2545 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2546 fs_info->free_space_root = root;
2552 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2553 location.objectid, ret);
2558 * Real super block validation
2559 * NOTE: super csum type and incompat features will not be checked here.
2561 * @sb: super block to check
2562 * @mirror_num: the super block number to check its bytenr:
2563 * 0 the primary (1st) sb
2564 * 1, 2 2nd and 3rd backup copy
2565 * -1 skip bytenr check
2567 static int validate_super(struct btrfs_fs_info *fs_info,
2568 struct btrfs_super_block *sb, int mirror_num)
2570 u64 nodesize = btrfs_super_nodesize(sb);
2571 u64 sectorsize = btrfs_super_sectorsize(sb);
2574 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2575 btrfs_err(fs_info, "no valid FS found");
2578 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2579 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2580 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2583 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2584 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2585 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2588 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2589 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2590 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2593 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2594 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2595 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2600 * Check sectorsize and nodesize first, other check will need it.
2601 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2603 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2604 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2605 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2610 * For 4K page size, we only support 4K sector size.
2611 * For 64K page size, we support read-write for 64K sector size, and
2612 * read-only for 4K sector size.
2614 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2615 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2616 sectorsize != SZ_64K))) {
2618 "sectorsize %llu not yet supported for page size %lu",
2619 sectorsize, PAGE_SIZE);
2623 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2624 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2625 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2628 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2629 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2630 le32_to_cpu(sb->__unused_leafsize), nodesize);
2634 /* Root alignment check */
2635 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2636 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2637 btrfs_super_root(sb));
2640 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2641 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2642 btrfs_super_chunk_root(sb));
2645 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2646 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2647 btrfs_super_log_root(sb));
2651 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2654 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2655 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2659 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2660 memcmp(fs_info->fs_devices->metadata_uuid,
2661 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2663 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2664 fs_info->super_copy->metadata_uuid,
2665 fs_info->fs_devices->metadata_uuid);
2669 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2670 BTRFS_FSID_SIZE) != 0) {
2672 "dev_item UUID does not match metadata fsid: %pU != %pU",
2673 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2678 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2681 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2682 btrfs_err(fs_info, "bytes_used is too small %llu",
2683 btrfs_super_bytes_used(sb));
2686 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2687 btrfs_err(fs_info, "invalid stripesize %u",
2688 btrfs_super_stripesize(sb));
2691 if (btrfs_super_num_devices(sb) > (1UL << 31))
2692 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2693 btrfs_super_num_devices(sb));
2694 if (btrfs_super_num_devices(sb) == 0) {
2695 btrfs_err(fs_info, "number of devices is 0");
2699 if (mirror_num >= 0 &&
2700 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2701 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2702 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2707 * Obvious sys_chunk_array corruptions, it must hold at least one key
2710 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2711 btrfs_err(fs_info, "system chunk array too big %u > %u",
2712 btrfs_super_sys_array_size(sb),
2713 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2716 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2717 + sizeof(struct btrfs_chunk)) {
2718 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2719 btrfs_super_sys_array_size(sb),
2720 sizeof(struct btrfs_disk_key)
2721 + sizeof(struct btrfs_chunk));
2726 * The generation is a global counter, we'll trust it more than the others
2727 * but it's still possible that it's the one that's wrong.
2729 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2731 "suspicious: generation < chunk_root_generation: %llu < %llu",
2732 btrfs_super_generation(sb),
2733 btrfs_super_chunk_root_generation(sb));
2734 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2735 && btrfs_super_cache_generation(sb) != (u64)-1)
2737 "suspicious: generation < cache_generation: %llu < %llu",
2738 btrfs_super_generation(sb),
2739 btrfs_super_cache_generation(sb));
2745 * Validation of super block at mount time.
2746 * Some checks already done early at mount time, like csum type and incompat
2747 * flags will be skipped.
2749 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2751 return validate_super(fs_info, fs_info->super_copy, 0);
2755 * Validation of super block at write time.
2756 * Some checks like bytenr check will be skipped as their values will be
2758 * Extra checks like csum type and incompat flags will be done here.
2760 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2761 struct btrfs_super_block *sb)
2765 ret = validate_super(fs_info, sb, -1);
2768 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2770 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2771 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2774 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2777 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2778 btrfs_super_incompat_flags(sb),
2779 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2785 "super block corruption detected before writing it to disk");
2789 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2791 int backup_index = find_newest_super_backup(fs_info);
2792 struct btrfs_super_block *sb = fs_info->super_copy;
2793 struct btrfs_root *tree_root = fs_info->tree_root;
2794 bool handle_error = false;
2798 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2803 if (!IS_ERR(tree_root->node))
2804 free_extent_buffer(tree_root->node);
2805 tree_root->node = NULL;
2807 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2810 free_root_pointers(fs_info, 0);
2813 * Don't use the log in recovery mode, it won't be
2816 btrfs_set_super_log_root(sb, 0);
2818 /* We can't trust the free space cache either */
2819 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2821 ret = read_backup_root(fs_info, i);
2826 generation = btrfs_super_generation(sb);
2827 level = btrfs_super_root_level(sb);
2828 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2829 BTRFS_ROOT_TREE_OBJECTID,
2830 generation, level, NULL);
2831 if (IS_ERR(tree_root->node)) {
2832 handle_error = true;
2833 ret = PTR_ERR(tree_root->node);
2834 tree_root->node = NULL;
2835 btrfs_warn(fs_info, "couldn't read tree root");
2838 } else if (!extent_buffer_uptodate(tree_root->node)) {
2839 handle_error = true;
2841 btrfs_warn(fs_info, "error while reading tree root");
2845 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2846 tree_root->commit_root = btrfs_root_node(tree_root);
2847 btrfs_set_root_refs(&tree_root->root_item, 1);
2850 * No need to hold btrfs_root::objectid_mutex since the fs
2851 * hasn't been fully initialised and we are the only user
2853 ret = btrfs_init_root_free_objectid(tree_root);
2855 handle_error = true;
2859 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2861 ret = btrfs_read_roots(fs_info);
2863 handle_error = true;
2867 /* All successful */
2868 fs_info->generation = generation;
2869 fs_info->last_trans_committed = generation;
2871 /* Always begin writing backup roots after the one being used */
2872 if (backup_index < 0) {
2873 fs_info->backup_root_index = 0;
2875 fs_info->backup_root_index = backup_index + 1;
2876 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2884 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2886 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2887 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2888 INIT_LIST_HEAD(&fs_info->trans_list);
2889 INIT_LIST_HEAD(&fs_info->dead_roots);
2890 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2891 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2892 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2893 spin_lock_init(&fs_info->delalloc_root_lock);
2894 spin_lock_init(&fs_info->trans_lock);
2895 spin_lock_init(&fs_info->fs_roots_radix_lock);
2896 spin_lock_init(&fs_info->delayed_iput_lock);
2897 spin_lock_init(&fs_info->defrag_inodes_lock);
2898 spin_lock_init(&fs_info->super_lock);
2899 spin_lock_init(&fs_info->buffer_lock);
2900 spin_lock_init(&fs_info->unused_bgs_lock);
2901 spin_lock_init(&fs_info->treelog_bg_lock);
2902 rwlock_init(&fs_info->tree_mod_log_lock);
2903 mutex_init(&fs_info->unused_bg_unpin_mutex);
2904 mutex_init(&fs_info->reclaim_bgs_lock);
2905 mutex_init(&fs_info->reloc_mutex);
2906 mutex_init(&fs_info->delalloc_root_mutex);
2907 mutex_init(&fs_info->zoned_meta_io_lock);
2908 seqlock_init(&fs_info->profiles_lock);
2910 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2911 INIT_LIST_HEAD(&fs_info->space_info);
2912 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2913 INIT_LIST_HEAD(&fs_info->unused_bgs);
2914 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2915 #ifdef CONFIG_BTRFS_DEBUG
2916 INIT_LIST_HEAD(&fs_info->allocated_roots);
2917 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2918 spin_lock_init(&fs_info->eb_leak_lock);
2920 extent_map_tree_init(&fs_info->mapping_tree);
2921 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2922 BTRFS_BLOCK_RSV_GLOBAL);
2923 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2924 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2925 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2926 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2927 BTRFS_BLOCK_RSV_DELOPS);
2928 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2929 BTRFS_BLOCK_RSV_DELREFS);
2931 atomic_set(&fs_info->async_delalloc_pages, 0);
2932 atomic_set(&fs_info->defrag_running, 0);
2933 atomic_set(&fs_info->reada_works_cnt, 0);
2934 atomic_set(&fs_info->nr_delayed_iputs, 0);
2935 atomic64_set(&fs_info->tree_mod_seq, 0);
2936 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2937 fs_info->metadata_ratio = 0;
2938 fs_info->defrag_inodes = RB_ROOT;
2939 atomic64_set(&fs_info->free_chunk_space, 0);
2940 fs_info->tree_mod_log = RB_ROOT;
2941 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2942 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2943 /* readahead state */
2944 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2945 spin_lock_init(&fs_info->reada_lock);
2946 btrfs_init_ref_verify(fs_info);
2948 fs_info->thread_pool_size = min_t(unsigned long,
2949 num_online_cpus() + 2, 8);
2951 INIT_LIST_HEAD(&fs_info->ordered_roots);
2952 spin_lock_init(&fs_info->ordered_root_lock);
2954 btrfs_init_scrub(fs_info);
2955 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2956 fs_info->check_integrity_print_mask = 0;
2958 btrfs_init_balance(fs_info);
2959 btrfs_init_async_reclaim_work(fs_info);
2961 spin_lock_init(&fs_info->block_group_cache_lock);
2962 fs_info->block_group_cache_tree = RB_ROOT;
2963 fs_info->first_logical_byte = (u64)-1;
2965 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2966 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2967 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2969 mutex_init(&fs_info->ordered_operations_mutex);
2970 mutex_init(&fs_info->tree_log_mutex);
2971 mutex_init(&fs_info->chunk_mutex);
2972 mutex_init(&fs_info->transaction_kthread_mutex);
2973 mutex_init(&fs_info->cleaner_mutex);
2974 mutex_init(&fs_info->ro_block_group_mutex);
2975 init_rwsem(&fs_info->commit_root_sem);
2976 init_rwsem(&fs_info->cleanup_work_sem);
2977 init_rwsem(&fs_info->subvol_sem);
2978 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2980 btrfs_init_dev_replace_locks(fs_info);
2981 btrfs_init_qgroup(fs_info);
2982 btrfs_discard_init(fs_info);
2984 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2985 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2987 init_waitqueue_head(&fs_info->transaction_throttle);
2988 init_waitqueue_head(&fs_info->transaction_wait);
2989 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2990 init_waitqueue_head(&fs_info->async_submit_wait);
2991 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2993 /* Usable values until the real ones are cached from the superblock */
2994 fs_info->nodesize = 4096;
2995 fs_info->sectorsize = 4096;
2996 fs_info->sectorsize_bits = ilog2(4096);
2997 fs_info->stripesize = 4096;
2999 spin_lock_init(&fs_info->swapfile_pins_lock);
3000 fs_info->swapfile_pins = RB_ROOT;
3002 fs_info->send_in_progress = 0;
3004 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3005 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3008 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3013 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3014 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3016 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3020 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3024 fs_info->dirty_metadata_batch = PAGE_SIZE *
3025 (1 + ilog2(nr_cpu_ids));
3027 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3031 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3036 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3038 if (!fs_info->delayed_root)
3040 btrfs_init_delayed_root(fs_info->delayed_root);
3043 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3045 return btrfs_alloc_stripe_hash_table(fs_info);
3048 static int btrfs_uuid_rescan_kthread(void *data)
3050 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3054 * 1st step is to iterate through the existing UUID tree and
3055 * to delete all entries that contain outdated data.
3056 * 2nd step is to add all missing entries to the UUID tree.
3058 ret = btrfs_uuid_tree_iterate(fs_info);
3061 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3063 up(&fs_info->uuid_tree_rescan_sem);
3066 return btrfs_uuid_scan_kthread(data);
3069 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3071 struct task_struct *task;
3073 down(&fs_info->uuid_tree_rescan_sem);
3074 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3076 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3077 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3078 up(&fs_info->uuid_tree_rescan_sem);
3079 return PTR_ERR(task);
3086 * Some options only have meaning at mount time and shouldn't persist across
3087 * remounts, or be displayed. Clear these at the end of mount and remount
3090 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3092 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3093 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3097 * Mounting logic specific to read-write file systems. Shared by open_ctree
3098 * and btrfs_remount when remounting from read-only to read-write.
3100 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3103 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3104 bool clear_free_space_tree = false;
3106 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3107 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3108 clear_free_space_tree = true;
3109 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3110 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3111 btrfs_warn(fs_info, "free space tree is invalid");
3112 clear_free_space_tree = true;
3115 if (clear_free_space_tree) {
3116 btrfs_info(fs_info, "clearing free space tree");
3117 ret = btrfs_clear_free_space_tree(fs_info);
3120 "failed to clear free space tree: %d", ret);
3126 * btrfs_find_orphan_roots() is responsible for finding all the dead
3127 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3128 * them into the fs_info->fs_roots_radix tree. This must be done before
3129 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3130 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3131 * item before the root's tree is deleted - this means that if we unmount
3132 * or crash before the deletion completes, on the next mount we will not
3133 * delete what remains of the tree because the orphan item does not
3134 * exists anymore, which is what tells us we have a pending deletion.
3136 ret = btrfs_find_orphan_roots(fs_info);
3140 ret = btrfs_cleanup_fs_roots(fs_info);
3144 down_read(&fs_info->cleanup_work_sem);
3145 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3146 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3147 up_read(&fs_info->cleanup_work_sem);
3150 up_read(&fs_info->cleanup_work_sem);
3152 mutex_lock(&fs_info->cleaner_mutex);
3153 ret = btrfs_recover_relocation(fs_info->tree_root);
3154 mutex_unlock(&fs_info->cleaner_mutex);
3156 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3160 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3161 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3162 btrfs_info(fs_info, "creating free space tree");
3163 ret = btrfs_create_free_space_tree(fs_info);
3166 "failed to create free space tree: %d", ret);
3171 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3172 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3177 ret = btrfs_resume_balance_async(fs_info);
3181 ret = btrfs_resume_dev_replace_async(fs_info);
3183 btrfs_warn(fs_info, "failed to resume dev_replace");
3187 btrfs_qgroup_rescan_resume(fs_info);
3189 if (!fs_info->uuid_root) {
3190 btrfs_info(fs_info, "creating UUID tree");
3191 ret = btrfs_create_uuid_tree(fs_info);
3194 "failed to create the UUID tree %d", ret);
3203 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3212 struct btrfs_super_block *disk_super;
3213 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3214 struct btrfs_root *tree_root;
3215 struct btrfs_root *chunk_root;
3220 ret = init_mount_fs_info(fs_info, sb);
3226 /* These need to be init'ed before we start creating inodes and such. */
3227 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3229 fs_info->tree_root = tree_root;
3230 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3232 fs_info->chunk_root = chunk_root;
3233 if (!tree_root || !chunk_root) {
3238 fs_info->btree_inode = new_inode(sb);
3239 if (!fs_info->btree_inode) {
3243 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3244 btrfs_init_btree_inode(fs_info);
3246 invalidate_bdev(fs_devices->latest_bdev);
3249 * Read super block and check the signature bytes only
3251 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
3252 if (IS_ERR(disk_super)) {
3253 err = PTR_ERR(disk_super);
3258 * Verify the type first, if that or the checksum value are
3259 * corrupted, we'll find out
3261 csum_type = btrfs_super_csum_type(disk_super);
3262 if (!btrfs_supported_super_csum(csum_type)) {
3263 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3266 btrfs_release_disk_super(disk_super);
3270 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3272 ret = btrfs_init_csum_hash(fs_info, csum_type);
3275 btrfs_release_disk_super(disk_super);
3280 * We want to check superblock checksum, the type is stored inside.
3281 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3283 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3284 btrfs_err(fs_info, "superblock checksum mismatch");
3286 btrfs_release_disk_super(disk_super);
3291 * super_copy is zeroed at allocation time and we never touch the
3292 * following bytes up to INFO_SIZE, the checksum is calculated from
3293 * the whole block of INFO_SIZE
3295 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3296 btrfs_release_disk_super(disk_super);
3298 disk_super = fs_info->super_copy;
3301 features = btrfs_super_flags(disk_super);
3302 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3303 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3304 btrfs_set_super_flags(disk_super, features);
3306 "found metadata UUID change in progress flag, clearing");
3309 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3310 sizeof(*fs_info->super_for_commit));
3312 ret = btrfs_validate_mount_super(fs_info);
3314 btrfs_err(fs_info, "superblock contains fatal errors");
3319 if (!btrfs_super_root(disk_super))
3322 /* check FS state, whether FS is broken. */
3323 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3324 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3327 * In the long term, we'll store the compression type in the super
3328 * block, and it'll be used for per file compression control.
3330 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3332 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3338 features = btrfs_super_incompat_flags(disk_super) &
3339 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3342 "cannot mount because of unsupported optional features (%llx)",
3348 features = btrfs_super_incompat_flags(disk_super);
3349 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3350 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3351 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3352 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3353 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3355 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3356 btrfs_info(fs_info, "has skinny extents");
3359 * flag our filesystem as having big metadata blocks if
3360 * they are bigger than the page size
3362 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3363 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3365 "flagging fs with big metadata feature");
3366 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3369 nodesize = btrfs_super_nodesize(disk_super);
3370 sectorsize = btrfs_super_sectorsize(disk_super);
3371 stripesize = sectorsize;
3372 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3373 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3375 /* Cache block sizes */
3376 fs_info->nodesize = nodesize;
3377 fs_info->sectorsize = sectorsize;
3378 fs_info->sectorsize_bits = ilog2(sectorsize);
3379 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3380 fs_info->stripesize = stripesize;
3383 * mixed block groups end up with duplicate but slightly offset
3384 * extent buffers for the same range. It leads to corruptions
3386 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3387 (sectorsize != nodesize)) {
3389 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3390 nodesize, sectorsize);
3395 * Needn't use the lock because there is no other task which will
3398 btrfs_set_super_incompat_flags(disk_super, features);
3400 features = btrfs_super_compat_ro_flags(disk_super) &
3401 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3402 if (!sb_rdonly(sb) && features) {
3404 "cannot mount read-write because of unsupported optional features (%llx)",
3410 /* For 4K sector size support, it's only read-only */
3411 if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) {
3412 if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) {
3414 "subpage sectorsize %u only supported read-only for page size %lu",
3415 sectorsize, PAGE_SIZE);
3421 ret = btrfs_init_workqueues(fs_info, fs_devices);
3424 goto fail_sb_buffer;
3427 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3428 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3430 sb->s_blocksize = sectorsize;
3431 sb->s_blocksize_bits = blksize_bits(sectorsize);
3432 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3434 mutex_lock(&fs_info->chunk_mutex);
3435 ret = btrfs_read_sys_array(fs_info);
3436 mutex_unlock(&fs_info->chunk_mutex);
3438 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3439 goto fail_sb_buffer;
3442 generation = btrfs_super_chunk_root_generation(disk_super);
3443 level = btrfs_super_chunk_root_level(disk_super);
3445 chunk_root->node = read_tree_block(fs_info,
3446 btrfs_super_chunk_root(disk_super),
3447 BTRFS_CHUNK_TREE_OBJECTID,
3448 generation, level, NULL);
3449 if (IS_ERR(chunk_root->node) ||
3450 !extent_buffer_uptodate(chunk_root->node)) {
3451 btrfs_err(fs_info, "failed to read chunk root");
3452 if (!IS_ERR(chunk_root->node))
3453 free_extent_buffer(chunk_root->node);
3454 chunk_root->node = NULL;
3455 goto fail_tree_roots;
3457 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3458 chunk_root->commit_root = btrfs_root_node(chunk_root);
3460 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3461 offsetof(struct btrfs_header, chunk_tree_uuid),
3464 ret = btrfs_read_chunk_tree(fs_info);
3466 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3467 goto fail_tree_roots;
3471 * At this point we know all the devices that make this filesystem,
3472 * including the seed devices but we don't know yet if the replace
3473 * target is required. So free devices that are not part of this
3474 * filesystem but skip the replace traget device which is checked
3475 * below in btrfs_init_dev_replace().
3477 btrfs_free_extra_devids(fs_devices);
3478 if (!fs_devices->latest_bdev) {
3479 btrfs_err(fs_info, "failed to read devices");
3480 goto fail_tree_roots;
3483 ret = init_tree_roots(fs_info);
3485 goto fail_tree_roots;
3488 * Get zone type information of zoned block devices. This will also
3489 * handle emulation of a zoned filesystem if a regular device has the
3490 * zoned incompat feature flag set.
3492 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3495 "zoned: failed to read device zone info: %d",
3497 goto fail_block_groups;
3501 * If we have a uuid root and we're not being told to rescan we need to
3502 * check the generation here so we can set the
3503 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3504 * transaction during a balance or the log replay without updating the
3505 * uuid generation, and then if we crash we would rescan the uuid tree,
3506 * even though it was perfectly fine.
3508 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3509 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3510 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3512 ret = btrfs_verify_dev_extents(fs_info);
3515 "failed to verify dev extents against chunks: %d",
3517 goto fail_block_groups;
3519 ret = btrfs_recover_balance(fs_info);
3521 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3522 goto fail_block_groups;
3525 ret = btrfs_init_dev_stats(fs_info);
3527 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3528 goto fail_block_groups;
3531 ret = btrfs_init_dev_replace(fs_info);
3533 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3534 goto fail_block_groups;
3537 ret = btrfs_check_zoned_mode(fs_info);
3539 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3541 goto fail_block_groups;
3544 ret = btrfs_sysfs_add_fsid(fs_devices);
3546 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3548 goto fail_block_groups;
3551 ret = btrfs_sysfs_add_mounted(fs_info);
3553 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3554 goto fail_fsdev_sysfs;
3557 ret = btrfs_init_space_info(fs_info);
3559 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3563 ret = btrfs_read_block_groups(fs_info);
3565 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3569 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3571 "writable mount is not allowed due to too many missing devices");
3575 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3577 if (IS_ERR(fs_info->cleaner_kthread))
3580 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3582 "btrfs-transaction");
3583 if (IS_ERR(fs_info->transaction_kthread))
3586 if (!btrfs_test_opt(fs_info, NOSSD) &&
3587 !fs_info->fs_devices->rotating) {
3588 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3592 * Mount does not set all options immediately, we can do it now and do
3593 * not have to wait for transaction commit
3595 btrfs_apply_pending_changes(fs_info);
3597 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3598 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3599 ret = btrfsic_mount(fs_info, fs_devices,
3600 btrfs_test_opt(fs_info,
3601 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3603 fs_info->check_integrity_print_mask);
3606 "failed to initialize integrity check module: %d",
3610 ret = btrfs_read_qgroup_config(fs_info);
3612 goto fail_trans_kthread;
3614 if (btrfs_build_ref_tree(fs_info))
3615 btrfs_err(fs_info, "couldn't build ref tree");
3617 /* do not make disk changes in broken FS or nologreplay is given */
3618 if (btrfs_super_log_root(disk_super) != 0 &&
3619 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3620 btrfs_info(fs_info, "start tree-log replay");
3621 ret = btrfs_replay_log(fs_info, fs_devices);
3628 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3629 if (IS_ERR(fs_info->fs_root)) {
3630 err = PTR_ERR(fs_info->fs_root);
3631 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3632 fs_info->fs_root = NULL;
3639 ret = btrfs_start_pre_rw_mount(fs_info);
3641 close_ctree(fs_info);
3644 btrfs_discard_resume(fs_info);
3646 if (fs_info->uuid_root &&
3647 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3648 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3649 btrfs_info(fs_info, "checking UUID tree");
3650 ret = btrfs_check_uuid_tree(fs_info);
3653 "failed to check the UUID tree: %d", ret);
3654 close_ctree(fs_info);
3659 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3662 btrfs_clear_oneshot_options(fs_info);
3666 btrfs_free_qgroup_config(fs_info);
3668 kthread_stop(fs_info->transaction_kthread);
3669 btrfs_cleanup_transaction(fs_info);
3670 btrfs_free_fs_roots(fs_info);
3672 kthread_stop(fs_info->cleaner_kthread);
3675 * make sure we're done with the btree inode before we stop our
3678 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3681 btrfs_sysfs_remove_mounted(fs_info);
3684 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3687 btrfs_put_block_group_cache(fs_info);
3690 if (fs_info->data_reloc_root)
3691 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3692 free_root_pointers(fs_info, true);
3693 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3696 btrfs_stop_all_workers(fs_info);
3697 btrfs_free_block_groups(fs_info);
3699 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3701 iput(fs_info->btree_inode);
3703 btrfs_close_devices(fs_info->fs_devices);
3706 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3708 static void btrfs_end_super_write(struct bio *bio)
3710 struct btrfs_device *device = bio->bi_private;
3711 struct bio_vec *bvec;
3712 struct bvec_iter_all iter_all;
3715 bio_for_each_segment_all(bvec, bio, iter_all) {
3716 page = bvec->bv_page;
3718 if (bio->bi_status) {
3719 btrfs_warn_rl_in_rcu(device->fs_info,
3720 "lost page write due to IO error on %s (%d)",
3721 rcu_str_deref(device->name),
3722 blk_status_to_errno(bio->bi_status));
3723 ClearPageUptodate(page);
3725 btrfs_dev_stat_inc_and_print(device,
3726 BTRFS_DEV_STAT_WRITE_ERRS);
3728 SetPageUptodate(page);
3738 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3741 struct btrfs_super_block *super;
3743 u64 bytenr, bytenr_orig;
3744 struct address_space *mapping = bdev->bd_inode->i_mapping;
3747 bytenr_orig = btrfs_sb_offset(copy_num);
3748 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3750 return ERR_PTR(-EINVAL);
3752 return ERR_PTR(ret);
3754 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3755 return ERR_PTR(-EINVAL);
3757 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3759 return ERR_CAST(page);
3761 super = page_address(page);
3762 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3763 btrfs_release_disk_super(super);
3764 return ERR_PTR(-ENODATA);
3767 if (btrfs_super_bytenr(super) != bytenr_orig) {
3768 btrfs_release_disk_super(super);
3769 return ERR_PTR(-EINVAL);
3776 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3778 struct btrfs_super_block *super, *latest = NULL;
3782 /* we would like to check all the supers, but that would make
3783 * a btrfs mount succeed after a mkfs from a different FS.
3784 * So, we need to add a special mount option to scan for
3785 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3787 for (i = 0; i < 1; i++) {
3788 super = btrfs_read_dev_one_super(bdev, i);
3792 if (!latest || btrfs_super_generation(super) > transid) {
3794 btrfs_release_disk_super(super);
3797 transid = btrfs_super_generation(super);
3805 * Write superblock @sb to the @device. Do not wait for completion, all the
3806 * pages we use for writing are locked.
3808 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3809 * the expected device size at commit time. Note that max_mirrors must be
3810 * same for write and wait phases.
3812 * Return number of errors when page is not found or submission fails.
3814 static int write_dev_supers(struct btrfs_device *device,
3815 struct btrfs_super_block *sb, int max_mirrors)
3817 struct btrfs_fs_info *fs_info = device->fs_info;
3818 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3819 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3823 u64 bytenr, bytenr_orig;
3825 if (max_mirrors == 0)
3826 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3828 shash->tfm = fs_info->csum_shash;
3830 for (i = 0; i < max_mirrors; i++) {
3833 struct btrfs_super_block *disk_super;
3835 bytenr_orig = btrfs_sb_offset(i);
3836 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3837 if (ret == -ENOENT) {
3839 } else if (ret < 0) {
3840 btrfs_err(device->fs_info,
3841 "couldn't get super block location for mirror %d",
3846 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3847 device->commit_total_bytes)
3850 btrfs_set_super_bytenr(sb, bytenr_orig);
3852 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3853 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3856 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3859 btrfs_err(device->fs_info,
3860 "couldn't get super block page for bytenr %llu",
3866 /* Bump the refcount for wait_dev_supers() */
3869 disk_super = page_address(page);
3870 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3873 * Directly use bios here instead of relying on the page cache
3874 * to do I/O, so we don't lose the ability to do integrity
3877 bio = bio_alloc(GFP_NOFS, 1);
3878 bio_set_dev(bio, device->bdev);
3879 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3880 bio->bi_private = device;
3881 bio->bi_end_io = btrfs_end_super_write;
3882 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3883 offset_in_page(bytenr));
3886 * We FUA only the first super block. The others we allow to
3887 * go down lazy and there's a short window where the on-disk
3888 * copies might still contain the older version.
3890 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3891 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3892 bio->bi_opf |= REQ_FUA;
3894 btrfsic_submit_bio(bio);
3895 btrfs_advance_sb_log(device, i);
3897 return errors < i ? 0 : -1;
3901 * Wait for write completion of superblocks done by write_dev_supers,
3902 * @max_mirrors same for write and wait phases.
3904 * Return number of errors when page is not found or not marked up to
3907 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3911 bool primary_failed = false;
3915 if (max_mirrors == 0)
3916 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3918 for (i = 0; i < max_mirrors; i++) {
3921 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3922 if (ret == -ENOENT) {
3924 } else if (ret < 0) {
3927 primary_failed = true;
3930 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3931 device->commit_total_bytes)
3934 page = find_get_page(device->bdev->bd_inode->i_mapping,
3935 bytenr >> PAGE_SHIFT);
3939 primary_failed = true;
3942 /* Page is submitted locked and unlocked once the IO completes */
3943 wait_on_page_locked(page);
3944 if (PageError(page)) {
3947 primary_failed = true;
3950 /* Drop our reference */
3953 /* Drop the reference from the writing run */
3957 /* log error, force error return */
3958 if (primary_failed) {
3959 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3964 return errors < i ? 0 : -1;
3968 * endio for the write_dev_flush, this will wake anyone waiting
3969 * for the barrier when it is done
3971 static void btrfs_end_empty_barrier(struct bio *bio)
3973 complete(bio->bi_private);
3977 * Submit a flush request to the device if it supports it. Error handling is
3978 * done in the waiting counterpart.
3980 static void write_dev_flush(struct btrfs_device *device)
3982 struct request_queue *q = bdev_get_queue(device->bdev);
3983 struct bio *bio = device->flush_bio;
3985 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3989 bio->bi_end_io = btrfs_end_empty_barrier;
3990 bio_set_dev(bio, device->bdev);
3991 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3992 init_completion(&device->flush_wait);
3993 bio->bi_private = &device->flush_wait;
3995 btrfsic_submit_bio(bio);
3996 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4000 * If the flush bio has been submitted by write_dev_flush, wait for it.
4002 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4004 struct bio *bio = device->flush_bio;
4006 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4009 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4010 wait_for_completion_io(&device->flush_wait);
4012 return bio->bi_status;
4015 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4017 if (!btrfs_check_rw_degradable(fs_info, NULL))
4023 * send an empty flush down to each device in parallel,
4024 * then wait for them
4026 static int barrier_all_devices(struct btrfs_fs_info *info)
4028 struct list_head *head;
4029 struct btrfs_device *dev;
4030 int errors_wait = 0;
4033 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4034 /* send down all the barriers */
4035 head = &info->fs_devices->devices;
4036 list_for_each_entry(dev, head, dev_list) {
4037 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4041 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4042 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4045 write_dev_flush(dev);
4046 dev->last_flush_error = BLK_STS_OK;
4049 /* wait for all the barriers */
4050 list_for_each_entry(dev, head, dev_list) {
4051 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4057 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4058 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4061 ret = wait_dev_flush(dev);
4063 dev->last_flush_error = ret;
4064 btrfs_dev_stat_inc_and_print(dev,
4065 BTRFS_DEV_STAT_FLUSH_ERRS);
4072 * At some point we need the status of all disks
4073 * to arrive at the volume status. So error checking
4074 * is being pushed to a separate loop.
4076 return check_barrier_error(info);
4081 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4084 int min_tolerated = INT_MAX;
4086 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4087 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4088 min_tolerated = min_t(int, min_tolerated,
4089 btrfs_raid_array[BTRFS_RAID_SINGLE].
4090 tolerated_failures);
4092 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4093 if (raid_type == BTRFS_RAID_SINGLE)
4095 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4097 min_tolerated = min_t(int, min_tolerated,
4098 btrfs_raid_array[raid_type].
4099 tolerated_failures);
4102 if (min_tolerated == INT_MAX) {
4103 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4107 return min_tolerated;
4110 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4112 struct list_head *head;
4113 struct btrfs_device *dev;
4114 struct btrfs_super_block *sb;
4115 struct btrfs_dev_item *dev_item;
4119 int total_errors = 0;
4122 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4125 * max_mirrors == 0 indicates we're from commit_transaction,
4126 * not from fsync where the tree roots in fs_info have not
4127 * been consistent on disk.
4129 if (max_mirrors == 0)
4130 backup_super_roots(fs_info);
4132 sb = fs_info->super_for_commit;
4133 dev_item = &sb->dev_item;
4135 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4136 head = &fs_info->fs_devices->devices;
4137 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4140 ret = barrier_all_devices(fs_info);
4143 &fs_info->fs_devices->device_list_mutex);
4144 btrfs_handle_fs_error(fs_info, ret,
4145 "errors while submitting device barriers.");
4150 list_for_each_entry(dev, head, dev_list) {
4155 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4156 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4159 btrfs_set_stack_device_generation(dev_item, 0);
4160 btrfs_set_stack_device_type(dev_item, dev->type);
4161 btrfs_set_stack_device_id(dev_item, dev->devid);
4162 btrfs_set_stack_device_total_bytes(dev_item,
4163 dev->commit_total_bytes);
4164 btrfs_set_stack_device_bytes_used(dev_item,
4165 dev->commit_bytes_used);
4166 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4167 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4168 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4169 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4170 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4173 flags = btrfs_super_flags(sb);
4174 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4176 ret = btrfs_validate_write_super(fs_info, sb);
4178 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4179 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4180 "unexpected superblock corruption detected");
4184 ret = write_dev_supers(dev, sb, max_mirrors);
4188 if (total_errors > max_errors) {
4189 btrfs_err(fs_info, "%d errors while writing supers",
4191 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4193 /* FUA is masked off if unsupported and can't be the reason */
4194 btrfs_handle_fs_error(fs_info, -EIO,
4195 "%d errors while writing supers",
4201 list_for_each_entry(dev, head, dev_list) {
4204 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4205 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4208 ret = wait_dev_supers(dev, max_mirrors);
4212 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4213 if (total_errors > max_errors) {
4214 btrfs_handle_fs_error(fs_info, -EIO,
4215 "%d errors while writing supers",
4222 /* Drop a fs root from the radix tree and free it. */
4223 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4224 struct btrfs_root *root)
4226 bool drop_ref = false;
4228 spin_lock(&fs_info->fs_roots_radix_lock);
4229 radix_tree_delete(&fs_info->fs_roots_radix,
4230 (unsigned long)root->root_key.objectid);
4231 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4233 spin_unlock(&fs_info->fs_roots_radix_lock);
4235 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4236 ASSERT(root->log_root == NULL);
4237 if (root->reloc_root) {
4238 btrfs_put_root(root->reloc_root);
4239 root->reloc_root = NULL;
4244 btrfs_put_root(root);
4247 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4249 u64 root_objectid = 0;
4250 struct btrfs_root *gang[8];
4253 unsigned int ret = 0;
4256 spin_lock(&fs_info->fs_roots_radix_lock);
4257 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4258 (void **)gang, root_objectid,
4261 spin_unlock(&fs_info->fs_roots_radix_lock);
4264 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4266 for (i = 0; i < ret; i++) {
4267 /* Avoid to grab roots in dead_roots */
4268 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4272 /* grab all the search result for later use */
4273 gang[i] = btrfs_grab_root(gang[i]);
4275 spin_unlock(&fs_info->fs_roots_radix_lock);
4277 for (i = 0; i < ret; i++) {
4280 root_objectid = gang[i]->root_key.objectid;
4281 err = btrfs_orphan_cleanup(gang[i]);
4284 btrfs_put_root(gang[i]);
4289 /* release the uncleaned roots due to error */
4290 for (; i < ret; i++) {
4292 btrfs_put_root(gang[i]);
4297 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4299 struct btrfs_root *root = fs_info->tree_root;
4300 struct btrfs_trans_handle *trans;
4302 mutex_lock(&fs_info->cleaner_mutex);
4303 btrfs_run_delayed_iputs(fs_info);
4304 mutex_unlock(&fs_info->cleaner_mutex);
4305 wake_up_process(fs_info->cleaner_kthread);
4307 /* wait until ongoing cleanup work done */
4308 down_write(&fs_info->cleanup_work_sem);
4309 up_write(&fs_info->cleanup_work_sem);
4311 trans = btrfs_join_transaction(root);
4313 return PTR_ERR(trans);
4314 return btrfs_commit_transaction(trans);
4317 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4321 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4323 * We don't want the cleaner to start new transactions, add more delayed
4324 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4325 * because that frees the task_struct, and the transaction kthread might
4326 * still try to wake up the cleaner.
4328 kthread_park(fs_info->cleaner_kthread);
4330 /* wait for the qgroup rescan worker to stop */
4331 btrfs_qgroup_wait_for_completion(fs_info, false);
4333 /* wait for the uuid_scan task to finish */
4334 down(&fs_info->uuid_tree_rescan_sem);
4335 /* avoid complains from lockdep et al., set sem back to initial state */
4336 up(&fs_info->uuid_tree_rescan_sem);
4338 /* pause restriper - we want to resume on mount */
4339 btrfs_pause_balance(fs_info);
4341 btrfs_dev_replace_suspend_for_unmount(fs_info);
4343 btrfs_scrub_cancel(fs_info);
4345 /* wait for any defraggers to finish */
4346 wait_event(fs_info->transaction_wait,
4347 (atomic_read(&fs_info->defrag_running) == 0));
4349 /* clear out the rbtree of defraggable inodes */
4350 btrfs_cleanup_defrag_inodes(fs_info);
4352 cancel_work_sync(&fs_info->async_reclaim_work);
4353 cancel_work_sync(&fs_info->async_data_reclaim_work);
4354 cancel_work_sync(&fs_info->preempt_reclaim_work);
4356 cancel_work_sync(&fs_info->reclaim_bgs_work);
4358 /* Cancel or finish ongoing discard work */
4359 btrfs_discard_cleanup(fs_info);
4361 if (!sb_rdonly(fs_info->sb)) {
4363 * The cleaner kthread is stopped, so do one final pass over
4364 * unused block groups.
4366 btrfs_delete_unused_bgs(fs_info);
4369 * There might be existing delayed inode workers still running
4370 * and holding an empty delayed inode item. We must wait for
4371 * them to complete first because they can create a transaction.
4372 * This happens when someone calls btrfs_balance_delayed_items()
4373 * and then a transaction commit runs the same delayed nodes
4374 * before any delayed worker has done something with the nodes.
4375 * We must wait for any worker here and not at transaction
4376 * commit time since that could cause a deadlock.
4377 * This is a very rare case.
4379 btrfs_flush_workqueue(fs_info->delayed_workers);
4381 ret = btrfs_commit_super(fs_info);
4383 btrfs_err(fs_info, "commit super ret %d", ret);
4386 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4387 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4388 btrfs_error_commit_super(fs_info);
4390 kthread_stop(fs_info->transaction_kthread);
4391 kthread_stop(fs_info->cleaner_kthread);
4393 ASSERT(list_empty(&fs_info->delayed_iputs));
4394 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4396 if (btrfs_check_quota_leak(fs_info)) {
4397 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4398 btrfs_err(fs_info, "qgroup reserved space leaked");
4401 btrfs_free_qgroup_config(fs_info);
4402 ASSERT(list_empty(&fs_info->delalloc_roots));
4404 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4405 btrfs_info(fs_info, "at unmount delalloc count %lld",
4406 percpu_counter_sum(&fs_info->delalloc_bytes));
4409 if (percpu_counter_sum(&fs_info->ordered_bytes))
4410 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4411 percpu_counter_sum(&fs_info->ordered_bytes));
4413 btrfs_sysfs_remove_mounted(fs_info);
4414 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4416 btrfs_put_block_group_cache(fs_info);
4419 * we must make sure there is not any read request to
4420 * submit after we stopping all workers.
4422 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4423 btrfs_stop_all_workers(fs_info);
4425 /* We shouldn't have any transaction open at this point */
4426 ASSERT(list_empty(&fs_info->trans_list));
4428 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4429 free_root_pointers(fs_info, true);
4430 btrfs_free_fs_roots(fs_info);
4433 * We must free the block groups after dropping the fs_roots as we could
4434 * have had an IO error and have left over tree log blocks that aren't
4435 * cleaned up until the fs roots are freed. This makes the block group
4436 * accounting appear to be wrong because there's pending reserved bytes,
4437 * so make sure we do the block group cleanup afterwards.
4439 btrfs_free_block_groups(fs_info);
4441 iput(fs_info->btree_inode);
4443 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4444 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4445 btrfsic_unmount(fs_info->fs_devices);
4448 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4449 btrfs_close_devices(fs_info->fs_devices);
4452 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4456 struct inode *btree_inode = buf->pages[0]->mapping->host;
4458 ret = extent_buffer_uptodate(buf);
4462 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4463 parent_transid, atomic);
4469 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4471 struct btrfs_fs_info *fs_info = buf->fs_info;
4472 u64 transid = btrfs_header_generation(buf);
4475 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4477 * This is a fast path so only do this check if we have sanity tests
4478 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4479 * outside of the sanity tests.
4481 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4484 btrfs_assert_tree_locked(buf);
4485 if (transid != fs_info->generation)
4486 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4487 buf->start, transid, fs_info->generation);
4488 was_dirty = set_extent_buffer_dirty(buf);
4490 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4492 fs_info->dirty_metadata_batch);
4493 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4495 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4496 * but item data not updated.
4497 * So here we should only check item pointers, not item data.
4499 if (btrfs_header_level(buf) == 0 &&
4500 btrfs_check_leaf_relaxed(buf)) {
4501 btrfs_print_leaf(buf);
4507 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4511 * looks as though older kernels can get into trouble with
4512 * this code, they end up stuck in balance_dirty_pages forever
4516 if (current->flags & PF_MEMALLOC)
4520 btrfs_balance_delayed_items(fs_info);
4522 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4523 BTRFS_DIRTY_METADATA_THRESH,
4524 fs_info->dirty_metadata_batch);
4526 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4530 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4532 __btrfs_btree_balance_dirty(fs_info, 1);
4535 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4537 __btrfs_btree_balance_dirty(fs_info, 0);
4540 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4541 struct btrfs_key *first_key)
4543 return btree_read_extent_buffer_pages(buf, parent_transid,
4547 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4549 /* cleanup FS via transaction */
4550 btrfs_cleanup_transaction(fs_info);
4552 mutex_lock(&fs_info->cleaner_mutex);
4553 btrfs_run_delayed_iputs(fs_info);
4554 mutex_unlock(&fs_info->cleaner_mutex);
4556 down_write(&fs_info->cleanup_work_sem);
4557 up_write(&fs_info->cleanup_work_sem);
4560 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4562 struct btrfs_root *gang[8];
4563 u64 root_objectid = 0;
4566 spin_lock(&fs_info->fs_roots_radix_lock);
4567 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4568 (void **)gang, root_objectid,
4569 ARRAY_SIZE(gang))) != 0) {
4572 for (i = 0; i < ret; i++)
4573 gang[i] = btrfs_grab_root(gang[i]);
4574 spin_unlock(&fs_info->fs_roots_radix_lock);
4576 for (i = 0; i < ret; i++) {
4579 root_objectid = gang[i]->root_key.objectid;
4580 btrfs_free_log(NULL, gang[i]);
4581 btrfs_put_root(gang[i]);
4584 spin_lock(&fs_info->fs_roots_radix_lock);
4586 spin_unlock(&fs_info->fs_roots_radix_lock);
4587 btrfs_free_log_root_tree(NULL, fs_info);
4590 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4592 struct btrfs_ordered_extent *ordered;
4594 spin_lock(&root->ordered_extent_lock);
4596 * This will just short circuit the ordered completion stuff which will
4597 * make sure the ordered extent gets properly cleaned up.
4599 list_for_each_entry(ordered, &root->ordered_extents,
4601 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4602 spin_unlock(&root->ordered_extent_lock);
4605 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4607 struct btrfs_root *root;
4608 struct list_head splice;
4610 INIT_LIST_HEAD(&splice);
4612 spin_lock(&fs_info->ordered_root_lock);
4613 list_splice_init(&fs_info->ordered_roots, &splice);
4614 while (!list_empty(&splice)) {
4615 root = list_first_entry(&splice, struct btrfs_root,
4617 list_move_tail(&root->ordered_root,
4618 &fs_info->ordered_roots);
4620 spin_unlock(&fs_info->ordered_root_lock);
4621 btrfs_destroy_ordered_extents(root);
4624 spin_lock(&fs_info->ordered_root_lock);
4626 spin_unlock(&fs_info->ordered_root_lock);
4629 * We need this here because if we've been flipped read-only we won't
4630 * get sync() from the umount, so we need to make sure any ordered
4631 * extents that haven't had their dirty pages IO start writeout yet
4632 * actually get run and error out properly.
4634 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4637 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4638 struct btrfs_fs_info *fs_info)
4640 struct rb_node *node;
4641 struct btrfs_delayed_ref_root *delayed_refs;
4642 struct btrfs_delayed_ref_node *ref;
4645 delayed_refs = &trans->delayed_refs;
4647 spin_lock(&delayed_refs->lock);
4648 if (atomic_read(&delayed_refs->num_entries) == 0) {
4649 spin_unlock(&delayed_refs->lock);
4650 btrfs_debug(fs_info, "delayed_refs has NO entry");
4654 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4655 struct btrfs_delayed_ref_head *head;
4657 bool pin_bytes = false;
4659 head = rb_entry(node, struct btrfs_delayed_ref_head,
4661 if (btrfs_delayed_ref_lock(delayed_refs, head))
4664 spin_lock(&head->lock);
4665 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4666 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4669 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4670 RB_CLEAR_NODE(&ref->ref_node);
4671 if (!list_empty(&ref->add_list))
4672 list_del(&ref->add_list);
4673 atomic_dec(&delayed_refs->num_entries);
4674 btrfs_put_delayed_ref(ref);
4676 if (head->must_insert_reserved)
4678 btrfs_free_delayed_extent_op(head->extent_op);
4679 btrfs_delete_ref_head(delayed_refs, head);
4680 spin_unlock(&head->lock);
4681 spin_unlock(&delayed_refs->lock);
4682 mutex_unlock(&head->mutex);
4685 struct btrfs_block_group *cache;
4687 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4690 spin_lock(&cache->space_info->lock);
4691 spin_lock(&cache->lock);
4692 cache->pinned += head->num_bytes;
4693 btrfs_space_info_update_bytes_pinned(fs_info,
4694 cache->space_info, head->num_bytes);
4695 cache->reserved -= head->num_bytes;
4696 cache->space_info->bytes_reserved -= head->num_bytes;
4697 spin_unlock(&cache->lock);
4698 spin_unlock(&cache->space_info->lock);
4699 percpu_counter_add_batch(
4700 &cache->space_info->total_bytes_pinned,
4701 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4703 btrfs_put_block_group(cache);
4705 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4706 head->bytenr + head->num_bytes - 1);
4708 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4709 btrfs_put_delayed_ref_head(head);
4711 spin_lock(&delayed_refs->lock);
4713 btrfs_qgroup_destroy_extent_records(trans);
4715 spin_unlock(&delayed_refs->lock);
4720 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4722 struct btrfs_inode *btrfs_inode;
4723 struct list_head splice;
4725 INIT_LIST_HEAD(&splice);
4727 spin_lock(&root->delalloc_lock);
4728 list_splice_init(&root->delalloc_inodes, &splice);
4730 while (!list_empty(&splice)) {
4731 struct inode *inode = NULL;
4732 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4734 __btrfs_del_delalloc_inode(root, btrfs_inode);
4735 spin_unlock(&root->delalloc_lock);
4738 * Make sure we get a live inode and that it'll not disappear
4741 inode = igrab(&btrfs_inode->vfs_inode);
4743 invalidate_inode_pages2(inode->i_mapping);
4746 spin_lock(&root->delalloc_lock);
4748 spin_unlock(&root->delalloc_lock);
4751 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4753 struct btrfs_root *root;
4754 struct list_head splice;
4756 INIT_LIST_HEAD(&splice);
4758 spin_lock(&fs_info->delalloc_root_lock);
4759 list_splice_init(&fs_info->delalloc_roots, &splice);
4760 while (!list_empty(&splice)) {
4761 root = list_first_entry(&splice, struct btrfs_root,
4763 root = btrfs_grab_root(root);
4765 spin_unlock(&fs_info->delalloc_root_lock);
4767 btrfs_destroy_delalloc_inodes(root);
4768 btrfs_put_root(root);
4770 spin_lock(&fs_info->delalloc_root_lock);
4772 spin_unlock(&fs_info->delalloc_root_lock);
4775 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4776 struct extent_io_tree *dirty_pages,
4780 struct extent_buffer *eb;
4785 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4790 clear_extent_bits(dirty_pages, start, end, mark);
4791 while (start <= end) {
4792 eb = find_extent_buffer(fs_info, start);
4793 start += fs_info->nodesize;
4796 wait_on_extent_buffer_writeback(eb);
4798 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4800 clear_extent_buffer_dirty(eb);
4801 free_extent_buffer_stale(eb);
4808 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4809 struct extent_io_tree *unpin)
4816 struct extent_state *cached_state = NULL;
4819 * The btrfs_finish_extent_commit() may get the same range as
4820 * ours between find_first_extent_bit and clear_extent_dirty.
4821 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4822 * the same extent range.
4824 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4825 ret = find_first_extent_bit(unpin, 0, &start, &end,
4826 EXTENT_DIRTY, &cached_state);
4828 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4832 clear_extent_dirty(unpin, start, end, &cached_state);
4833 free_extent_state(cached_state);
4834 btrfs_error_unpin_extent_range(fs_info, start, end);
4835 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4842 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4844 struct inode *inode;
4846 inode = cache->io_ctl.inode;
4848 invalidate_inode_pages2(inode->i_mapping);
4849 BTRFS_I(inode)->generation = 0;
4850 cache->io_ctl.inode = NULL;
4853 ASSERT(cache->io_ctl.pages == NULL);
4854 btrfs_put_block_group(cache);
4857 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4858 struct btrfs_fs_info *fs_info)
4860 struct btrfs_block_group *cache;
4862 spin_lock(&cur_trans->dirty_bgs_lock);
4863 while (!list_empty(&cur_trans->dirty_bgs)) {
4864 cache = list_first_entry(&cur_trans->dirty_bgs,
4865 struct btrfs_block_group,
4868 if (!list_empty(&cache->io_list)) {
4869 spin_unlock(&cur_trans->dirty_bgs_lock);
4870 list_del_init(&cache->io_list);
4871 btrfs_cleanup_bg_io(cache);
4872 spin_lock(&cur_trans->dirty_bgs_lock);
4875 list_del_init(&cache->dirty_list);
4876 spin_lock(&cache->lock);
4877 cache->disk_cache_state = BTRFS_DC_ERROR;
4878 spin_unlock(&cache->lock);
4880 spin_unlock(&cur_trans->dirty_bgs_lock);
4881 btrfs_put_block_group(cache);
4882 btrfs_delayed_refs_rsv_release(fs_info, 1);
4883 spin_lock(&cur_trans->dirty_bgs_lock);
4885 spin_unlock(&cur_trans->dirty_bgs_lock);
4888 * Refer to the definition of io_bgs member for details why it's safe
4889 * to use it without any locking
4891 while (!list_empty(&cur_trans->io_bgs)) {
4892 cache = list_first_entry(&cur_trans->io_bgs,
4893 struct btrfs_block_group,
4896 list_del_init(&cache->io_list);
4897 spin_lock(&cache->lock);
4898 cache->disk_cache_state = BTRFS_DC_ERROR;
4899 spin_unlock(&cache->lock);
4900 btrfs_cleanup_bg_io(cache);
4904 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4905 struct btrfs_fs_info *fs_info)
4907 struct btrfs_device *dev, *tmp;
4909 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4910 ASSERT(list_empty(&cur_trans->dirty_bgs));
4911 ASSERT(list_empty(&cur_trans->io_bgs));
4913 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4915 list_del_init(&dev->post_commit_list);
4918 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4920 cur_trans->state = TRANS_STATE_COMMIT_START;
4921 wake_up(&fs_info->transaction_blocked_wait);
4923 cur_trans->state = TRANS_STATE_UNBLOCKED;
4924 wake_up(&fs_info->transaction_wait);
4926 btrfs_destroy_delayed_inodes(fs_info);
4928 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4930 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4932 btrfs_free_redirty_list(cur_trans);
4934 cur_trans->state =TRANS_STATE_COMPLETED;
4935 wake_up(&cur_trans->commit_wait);
4938 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4940 struct btrfs_transaction *t;
4942 mutex_lock(&fs_info->transaction_kthread_mutex);
4944 spin_lock(&fs_info->trans_lock);
4945 while (!list_empty(&fs_info->trans_list)) {
4946 t = list_first_entry(&fs_info->trans_list,
4947 struct btrfs_transaction, list);
4948 if (t->state >= TRANS_STATE_COMMIT_START) {
4949 refcount_inc(&t->use_count);
4950 spin_unlock(&fs_info->trans_lock);
4951 btrfs_wait_for_commit(fs_info, t->transid);
4952 btrfs_put_transaction(t);
4953 spin_lock(&fs_info->trans_lock);
4956 if (t == fs_info->running_transaction) {
4957 t->state = TRANS_STATE_COMMIT_DOING;
4958 spin_unlock(&fs_info->trans_lock);
4960 * We wait for 0 num_writers since we don't hold a trans
4961 * handle open currently for this transaction.
4963 wait_event(t->writer_wait,
4964 atomic_read(&t->num_writers) == 0);
4966 spin_unlock(&fs_info->trans_lock);
4968 btrfs_cleanup_one_transaction(t, fs_info);
4970 spin_lock(&fs_info->trans_lock);
4971 if (t == fs_info->running_transaction)
4972 fs_info->running_transaction = NULL;
4973 list_del_init(&t->list);
4974 spin_unlock(&fs_info->trans_lock);
4976 btrfs_put_transaction(t);
4977 trace_btrfs_transaction_commit(fs_info->tree_root);
4978 spin_lock(&fs_info->trans_lock);
4980 spin_unlock(&fs_info->trans_lock);
4981 btrfs_destroy_all_ordered_extents(fs_info);
4982 btrfs_destroy_delayed_inodes(fs_info);
4983 btrfs_assert_delayed_root_empty(fs_info);
4984 btrfs_destroy_all_delalloc_inodes(fs_info);
4985 btrfs_drop_all_logs(fs_info);
4986 mutex_unlock(&fs_info->transaction_kthread_mutex);
4991 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4993 struct btrfs_path *path;
4995 struct extent_buffer *l;
4996 struct btrfs_key search_key;
4997 struct btrfs_key found_key;
5000 path = btrfs_alloc_path();
5004 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5005 search_key.type = -1;
5006 search_key.offset = (u64)-1;
5007 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5010 BUG_ON(ret == 0); /* Corruption */
5011 if (path->slots[0] > 0) {
5012 slot = path->slots[0] - 1;
5014 btrfs_item_key_to_cpu(l, &found_key, slot);
5015 root->free_objectid = max_t(u64, found_key.objectid + 1,
5016 BTRFS_FIRST_FREE_OBJECTID);
5018 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5022 btrfs_free_path(path);
5026 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5029 mutex_lock(&root->objectid_mutex);
5031 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5032 btrfs_warn(root->fs_info,
5033 "the objectid of root %llu reaches its highest value",
5034 root->root_key.objectid);
5039 *objectid = root->free_objectid++;
5042 mutex_unlock(&root->objectid_mutex);