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 = num_extent_pages(buf);
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;
245 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
251 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
253 if (extent_buffer_uptodate(eb) &&
254 btrfs_header_generation(eb) == parent_transid) {
258 btrfs_err_rl(eb->fs_info,
259 "parent transid verify failed on %llu wanted %llu found %llu",
261 parent_transid, btrfs_header_generation(eb));
263 clear_extent_buffer_uptodate(eb);
265 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
270 static bool btrfs_supported_super_csum(u16 csum_type)
273 case BTRFS_CSUM_TYPE_CRC32:
274 case BTRFS_CSUM_TYPE_XXHASH:
275 case BTRFS_CSUM_TYPE_SHA256:
276 case BTRFS_CSUM_TYPE_BLAKE2:
284 * Return 0 if the superblock checksum type matches the checksum value of that
285 * algorithm. Pass the raw disk superblock data.
287 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
290 struct btrfs_super_block *disk_sb =
291 (struct btrfs_super_block *)raw_disk_sb;
292 char result[BTRFS_CSUM_SIZE];
293 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
295 shash->tfm = fs_info->csum_shash;
298 * The super_block structure does not span the whole
299 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
300 * filled with zeros and is included in the checksum.
302 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
303 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
305 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
311 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
312 struct btrfs_key *first_key, u64 parent_transid)
314 struct btrfs_fs_info *fs_info = eb->fs_info;
316 struct btrfs_key found_key;
319 found_level = btrfs_header_level(eb);
320 if (found_level != level) {
321 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
322 KERN_ERR "BTRFS: tree level check failed\n");
324 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
325 eb->start, level, found_level);
333 * For live tree block (new tree blocks in current transaction),
334 * we need proper lock context to avoid race, which is impossible here.
335 * So we only checks tree blocks which is read from disk, whose
336 * generation <= fs_info->last_trans_committed.
338 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
341 /* We have @first_key, so this @eb must have at least one item */
342 if (btrfs_header_nritems(eb) == 0) {
344 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
346 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
351 btrfs_node_key_to_cpu(eb, &found_key, 0);
353 btrfs_item_key_to_cpu(eb, &found_key, 0);
354 ret = btrfs_comp_cpu_keys(first_key, &found_key);
357 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
358 KERN_ERR "BTRFS: tree first key check failed\n");
360 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
361 eb->start, parent_transid, first_key->objectid,
362 first_key->type, first_key->offset,
363 found_key.objectid, found_key.type,
370 * helper to read a given tree block, doing retries as required when
371 * the checksums don't match and we have alternate mirrors to try.
373 * @parent_transid: expected transid, skip check if 0
374 * @level: expected level, mandatory check
375 * @first_key: expected key of first slot, skip check if NULL
377 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
378 u64 parent_transid, int level,
379 struct btrfs_key *first_key)
381 struct btrfs_fs_info *fs_info = eb->fs_info;
382 struct extent_io_tree *io_tree;
387 int failed_mirror = 0;
389 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
391 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
392 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
394 if (verify_parent_transid(io_tree, eb,
397 else if (btrfs_verify_level_key(eb, level,
398 first_key, parent_transid))
404 num_copies = btrfs_num_copies(fs_info,
409 if (!failed_mirror) {
411 failed_mirror = eb->read_mirror;
415 if (mirror_num == failed_mirror)
418 if (mirror_num > num_copies)
422 if (failed && !ret && failed_mirror)
423 btrfs_repair_eb_io_failure(eb, failed_mirror);
428 static int csum_one_extent_buffer(struct extent_buffer *eb)
430 struct btrfs_fs_info *fs_info = eb->fs_info;
431 u8 result[BTRFS_CSUM_SIZE];
434 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
435 offsetof(struct btrfs_header, fsid),
436 BTRFS_FSID_SIZE) == 0);
437 csum_tree_block(eb, result);
439 if (btrfs_header_level(eb))
440 ret = btrfs_check_node(eb);
442 ret = btrfs_check_leaf_full(eb);
448 * Also check the generation, the eb reached here must be newer than
449 * last committed. Or something seriously wrong happened.
451 if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
454 "block=%llu bad generation, have %llu expect > %llu",
455 eb->start, btrfs_header_generation(eb),
456 fs_info->last_trans_committed);
459 write_extent_buffer(eb, result, 0, fs_info->csum_size);
464 btrfs_print_tree(eb, 0);
465 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
467 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
471 /* Checksum all dirty extent buffers in one bio_vec */
472 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
473 struct bio_vec *bvec)
475 struct page *page = bvec->bv_page;
476 u64 bvec_start = page_offset(page) + bvec->bv_offset;
480 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
481 cur += fs_info->nodesize) {
482 struct extent_buffer *eb;
485 eb = find_extent_buffer(fs_info, cur);
486 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
489 /* A dirty eb shouldn't disappear from buffer_radix */
493 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
494 free_extent_buffer(eb);
497 if (WARN_ON(!uptodate)) {
498 free_extent_buffer(eb);
502 ret = csum_one_extent_buffer(eb);
503 free_extent_buffer(eb);
511 * Checksum a dirty tree block before IO. This has extra checks to make sure
512 * we only fill in the checksum field in the first page of a multi-page block.
513 * For subpage extent buffers we need bvec to also read the offset in the page.
515 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
517 struct page *page = bvec->bv_page;
518 u64 start = page_offset(page);
520 struct extent_buffer *eb;
522 if (fs_info->sectorsize < PAGE_SIZE)
523 return csum_dirty_subpage_buffers(fs_info, bvec);
525 eb = (struct extent_buffer *)page->private;
526 if (page != eb->pages[0])
529 found_start = btrfs_header_bytenr(eb);
531 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
532 WARN_ON(found_start != 0);
537 * Please do not consolidate these warnings into a single if.
538 * It is useful to know what went wrong.
540 if (WARN_ON(found_start != start))
542 if (WARN_ON(!PageUptodate(page)))
545 return csum_one_extent_buffer(eb);
548 static int check_tree_block_fsid(struct extent_buffer *eb)
550 struct btrfs_fs_info *fs_info = eb->fs_info;
551 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
552 u8 fsid[BTRFS_FSID_SIZE];
555 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
558 * Checking the incompat flag is only valid for the current fs. For
559 * seed devices it's forbidden to have their uuid changed so reading
560 * ->fsid in this case is fine
562 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
563 metadata_uuid = fs_devices->metadata_uuid;
565 metadata_uuid = fs_devices->fsid;
567 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
570 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
571 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
577 /* Do basic extent buffer checks at read time */
578 static int validate_extent_buffer(struct extent_buffer *eb)
580 struct btrfs_fs_info *fs_info = eb->fs_info;
582 const u32 csum_size = fs_info->csum_size;
584 u8 result[BTRFS_CSUM_SIZE];
585 const u8 *header_csum;
588 found_start = btrfs_header_bytenr(eb);
589 if (found_start != eb->start) {
590 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
591 eb->start, found_start);
595 if (check_tree_block_fsid(eb)) {
596 btrfs_err_rl(fs_info, "bad fsid on block %llu",
601 found_level = btrfs_header_level(eb);
602 if (found_level >= BTRFS_MAX_LEVEL) {
603 btrfs_err(fs_info, "bad tree block level %d on %llu",
604 (int)btrfs_header_level(eb), eb->start);
609 csum_tree_block(eb, result);
610 header_csum = page_address(eb->pages[0]) +
611 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
613 if (memcmp(result, header_csum, csum_size) != 0) {
614 btrfs_warn_rl(fs_info,
615 "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
617 CSUM_FMT_VALUE(csum_size, header_csum),
618 CSUM_FMT_VALUE(csum_size, result),
619 btrfs_header_level(eb));
625 * If this is a leaf block and it is corrupt, set the corrupt bit so
626 * that we don't try and read the other copies of this block, just
629 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
630 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
634 if (found_level > 0 && btrfs_check_node(eb))
638 set_extent_buffer_uptodate(eb);
641 "block=%llu read time tree block corruption detected",
647 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
650 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
651 struct extent_buffer *eb;
656 * We don't allow bio merge for subpage metadata read, so we should
657 * only get one eb for each endio hook.
659 ASSERT(end == start + fs_info->nodesize - 1);
660 ASSERT(PagePrivate(page));
662 eb = find_extent_buffer(fs_info, start);
664 * When we are reading one tree block, eb must have been inserted into
665 * the radix tree. If not, something is wrong.
669 reads_done = atomic_dec_and_test(&eb->io_pages);
670 /* Subpage read must finish in page read */
673 eb->read_mirror = mirror;
674 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
678 ret = validate_extent_buffer(eb);
682 set_extent_buffer_uptodate(eb);
684 free_extent_buffer(eb);
688 * end_bio_extent_readpage decrements io_pages in case of error,
689 * make sure it has something to decrement.
691 atomic_inc(&eb->io_pages);
692 clear_extent_buffer_uptodate(eb);
693 free_extent_buffer(eb);
697 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
698 struct page *page, u64 start, u64 end,
701 struct extent_buffer *eb;
705 ASSERT(page->private);
707 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
708 return validate_subpage_buffer(page, start, end, mirror);
710 eb = (struct extent_buffer *)page->private;
713 * The pending IO might have been the only thing that kept this buffer
714 * in memory. Make sure we have a ref for all this other checks
716 atomic_inc(&eb->refs);
718 reads_done = atomic_dec_and_test(&eb->io_pages);
722 eb->read_mirror = mirror;
723 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
727 ret = validate_extent_buffer(eb);
731 * our io error hook is going to dec the io pages
732 * again, we have to make sure it has something
735 atomic_inc(&eb->io_pages);
736 clear_extent_buffer_uptodate(eb);
738 free_extent_buffer(eb);
743 static void end_workqueue_bio(struct bio *bio)
745 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
746 struct btrfs_fs_info *fs_info;
747 struct btrfs_workqueue *wq;
749 fs_info = end_io_wq->info;
750 end_io_wq->status = bio->bi_status;
752 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
753 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
754 wq = fs_info->endio_meta_write_workers;
755 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
756 wq = fs_info->endio_freespace_worker;
757 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
758 wq = fs_info->endio_raid56_workers;
760 wq = fs_info->endio_write_workers;
762 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
763 wq = fs_info->endio_raid56_workers;
764 else if (end_io_wq->metadata)
765 wq = fs_info->endio_meta_workers;
767 wq = fs_info->endio_workers;
770 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
771 btrfs_queue_work(wq, &end_io_wq->work);
774 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
775 enum btrfs_wq_endio_type metadata)
777 struct btrfs_end_io_wq *end_io_wq;
779 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
781 return BLK_STS_RESOURCE;
783 end_io_wq->private = bio->bi_private;
784 end_io_wq->end_io = bio->bi_end_io;
785 end_io_wq->info = info;
786 end_io_wq->status = 0;
787 end_io_wq->bio = bio;
788 end_io_wq->metadata = metadata;
790 bio->bi_private = end_io_wq;
791 bio->bi_end_io = end_workqueue_bio;
795 static void run_one_async_start(struct btrfs_work *work)
797 struct async_submit_bio *async;
800 async = container_of(work, struct async_submit_bio, work);
801 ret = async->submit_bio_start(async->inode, async->bio,
802 async->dio_file_offset);
808 * In order to insert checksums into the metadata in large chunks, we wait
809 * until bio submission time. All the pages in the bio are checksummed and
810 * sums are attached onto the ordered extent record.
812 * At IO completion time the csums attached on the ordered extent record are
813 * inserted into the tree.
815 static void run_one_async_done(struct btrfs_work *work)
817 struct async_submit_bio *async;
821 async = container_of(work, struct async_submit_bio, work);
822 inode = async->inode;
824 /* If an error occurred we just want to clean up the bio and move on */
826 async->bio->bi_status = async->status;
827 bio_endio(async->bio);
832 * All of the bios that pass through here are from async helpers.
833 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
834 * This changes nothing when cgroups aren't in use.
836 async->bio->bi_opf |= REQ_CGROUP_PUNT;
837 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
839 async->bio->bi_status = ret;
840 bio_endio(async->bio);
844 static void run_one_async_free(struct btrfs_work *work)
846 struct async_submit_bio *async;
848 async = container_of(work, struct async_submit_bio, work);
852 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
853 int mirror_num, unsigned long bio_flags,
855 extent_submit_bio_start_t *submit_bio_start)
857 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
858 struct async_submit_bio *async;
860 async = kmalloc(sizeof(*async), GFP_NOFS);
862 return BLK_STS_RESOURCE;
864 async->inode = inode;
866 async->mirror_num = mirror_num;
867 async->submit_bio_start = submit_bio_start;
869 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
872 async->dio_file_offset = dio_file_offset;
876 if (op_is_sync(bio->bi_opf))
877 btrfs_set_work_high_priority(&async->work);
879 btrfs_queue_work(fs_info->workers, &async->work);
883 static blk_status_t btree_csum_one_bio(struct bio *bio)
885 struct bio_vec *bvec;
886 struct btrfs_root *root;
888 struct bvec_iter_all iter_all;
890 ASSERT(!bio_flagged(bio, BIO_CLONED));
891 bio_for_each_segment_all(bvec, bio, iter_all) {
892 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
893 ret = csum_dirty_buffer(root->fs_info, bvec);
898 return errno_to_blk_status(ret);
901 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
905 * when we're called for a write, we're already in the async
906 * submission context. Just jump into btrfs_map_bio
908 return btree_csum_one_bio(bio);
911 static bool should_async_write(struct btrfs_fs_info *fs_info,
912 struct btrfs_inode *bi)
914 if (btrfs_is_zoned(fs_info))
916 if (atomic_read(&bi->sync_writers))
918 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
923 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
924 int mirror_num, unsigned long bio_flags)
926 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
929 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
931 * called for a read, do the setup so that checksum validation
932 * can happen in the async kernel threads
934 ret = btrfs_bio_wq_end_io(fs_info, bio,
935 BTRFS_WQ_ENDIO_METADATA);
938 ret = btrfs_map_bio(fs_info, bio, mirror_num);
939 } else if (!should_async_write(fs_info, BTRFS_I(inode))) {
940 ret = btree_csum_one_bio(bio);
943 ret = btrfs_map_bio(fs_info, bio, mirror_num);
946 * kthread helpers are used to submit writes so that
947 * checksumming can happen in parallel across all CPUs
949 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
950 0, btree_submit_bio_start);
958 bio->bi_status = ret;
963 #ifdef CONFIG_MIGRATION
964 static int btree_migratepage(struct address_space *mapping,
965 struct page *newpage, struct page *page,
966 enum migrate_mode mode)
969 * we can't safely write a btree page from here,
970 * we haven't done the locking hook
975 * Buffers may be managed in a filesystem specific way.
976 * We must have no buffers or drop them.
978 if (page_has_private(page) &&
979 !try_to_release_page(page, GFP_KERNEL))
981 return migrate_page(mapping, newpage, page, mode);
986 static int btree_writepages(struct address_space *mapping,
987 struct writeback_control *wbc)
989 struct btrfs_fs_info *fs_info;
992 if (wbc->sync_mode == WB_SYNC_NONE) {
994 if (wbc->for_kupdate)
997 fs_info = BTRFS_I(mapping->host)->root->fs_info;
998 /* this is a bit racy, but that's ok */
999 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1000 BTRFS_DIRTY_METADATA_THRESH,
1001 fs_info->dirty_metadata_batch);
1005 return btree_write_cache_pages(mapping, wbc);
1008 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1010 if (PageWriteback(page) || PageDirty(page))
1013 return try_release_extent_buffer(page);
1016 static void btree_invalidate_folio(struct folio *folio, size_t offset,
1019 struct extent_io_tree *tree;
1020 tree = &BTRFS_I(folio->mapping->host)->io_tree;
1021 extent_invalidate_folio(tree, folio, offset);
1022 btree_releasepage(&folio->page, GFP_NOFS);
1023 if (folio_get_private(folio)) {
1024 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
1025 "folio private not zero on folio %llu",
1026 (unsigned long long)folio_pos(folio));
1027 folio_detach_private(folio);
1032 static bool btree_dirty_folio(struct address_space *mapping,
1033 struct folio *folio)
1035 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
1036 struct btrfs_subpage *subpage;
1037 struct extent_buffer *eb;
1039 u64 page_start = folio_pos(folio);
1041 if (fs_info->sectorsize == PAGE_SIZE) {
1042 eb = folio_get_private(folio);
1044 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1045 BUG_ON(!atomic_read(&eb->refs));
1046 btrfs_assert_tree_write_locked(eb);
1047 return filemap_dirty_folio(mapping, folio);
1049 subpage = folio_get_private(folio);
1051 ASSERT(subpage->dirty_bitmap);
1052 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1053 unsigned long flags;
1055 u16 tmp = (1 << cur_bit);
1057 spin_lock_irqsave(&subpage->lock, flags);
1058 if (!(tmp & subpage->dirty_bitmap)) {
1059 spin_unlock_irqrestore(&subpage->lock, flags);
1063 spin_unlock_irqrestore(&subpage->lock, flags);
1064 cur = page_start + cur_bit * fs_info->sectorsize;
1066 eb = find_extent_buffer(fs_info, cur);
1068 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1069 ASSERT(atomic_read(&eb->refs));
1070 btrfs_assert_tree_write_locked(eb);
1071 free_extent_buffer(eb);
1073 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1075 return filemap_dirty_folio(mapping, folio);
1078 #define btree_dirty_folio filemap_dirty_folio
1081 static const struct address_space_operations btree_aops = {
1082 .writepages = btree_writepages,
1083 .releasepage = btree_releasepage,
1084 .invalidate_folio = btree_invalidate_folio,
1085 #ifdef CONFIG_MIGRATION
1086 .migratepage = btree_migratepage,
1088 .dirty_folio = btree_dirty_folio,
1091 struct extent_buffer *btrfs_find_create_tree_block(
1092 struct btrfs_fs_info *fs_info,
1093 u64 bytenr, u64 owner_root,
1096 if (btrfs_is_testing(fs_info))
1097 return alloc_test_extent_buffer(fs_info, bytenr);
1098 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1102 * Read tree block at logical address @bytenr and do variant basic but critical
1105 * @owner_root: the objectid of the root owner for this block.
1106 * @parent_transid: expected transid of this tree block, skip check if 0
1107 * @level: expected level, mandatory check
1108 * @first_key: expected key in slot 0, skip check if NULL
1110 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1111 u64 owner_root, u64 parent_transid,
1112 int level, struct btrfs_key *first_key)
1114 struct extent_buffer *buf = NULL;
1117 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1121 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1124 free_extent_buffer_stale(buf);
1125 return ERR_PTR(ret);
1131 void btrfs_clean_tree_block(struct extent_buffer *buf)
1133 struct btrfs_fs_info *fs_info = buf->fs_info;
1134 if (btrfs_header_generation(buf) ==
1135 fs_info->running_transaction->transid) {
1136 btrfs_assert_tree_write_locked(buf);
1138 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1139 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1141 fs_info->dirty_metadata_batch);
1142 clear_extent_buffer_dirty(buf);
1147 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1150 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1152 memset(&root->root_key, 0, sizeof(root->root_key));
1153 memset(&root->root_item, 0, sizeof(root->root_item));
1154 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1155 root->fs_info = fs_info;
1156 root->root_key.objectid = objectid;
1158 root->commit_root = NULL;
1160 RB_CLEAR_NODE(&root->rb_node);
1162 root->last_trans = 0;
1163 root->free_objectid = 0;
1164 root->nr_delalloc_inodes = 0;
1165 root->nr_ordered_extents = 0;
1166 root->inode_tree = RB_ROOT;
1167 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1169 btrfs_init_root_block_rsv(root);
1171 INIT_LIST_HEAD(&root->dirty_list);
1172 INIT_LIST_HEAD(&root->root_list);
1173 INIT_LIST_HEAD(&root->delalloc_inodes);
1174 INIT_LIST_HEAD(&root->delalloc_root);
1175 INIT_LIST_HEAD(&root->ordered_extents);
1176 INIT_LIST_HEAD(&root->ordered_root);
1177 INIT_LIST_HEAD(&root->reloc_dirty_list);
1178 INIT_LIST_HEAD(&root->logged_list[0]);
1179 INIT_LIST_HEAD(&root->logged_list[1]);
1180 spin_lock_init(&root->inode_lock);
1181 spin_lock_init(&root->delalloc_lock);
1182 spin_lock_init(&root->ordered_extent_lock);
1183 spin_lock_init(&root->accounting_lock);
1184 spin_lock_init(&root->log_extents_lock[0]);
1185 spin_lock_init(&root->log_extents_lock[1]);
1186 spin_lock_init(&root->qgroup_meta_rsv_lock);
1187 mutex_init(&root->objectid_mutex);
1188 mutex_init(&root->log_mutex);
1189 mutex_init(&root->ordered_extent_mutex);
1190 mutex_init(&root->delalloc_mutex);
1191 init_waitqueue_head(&root->qgroup_flush_wait);
1192 init_waitqueue_head(&root->log_writer_wait);
1193 init_waitqueue_head(&root->log_commit_wait[0]);
1194 init_waitqueue_head(&root->log_commit_wait[1]);
1195 INIT_LIST_HEAD(&root->log_ctxs[0]);
1196 INIT_LIST_HEAD(&root->log_ctxs[1]);
1197 atomic_set(&root->log_commit[0], 0);
1198 atomic_set(&root->log_commit[1], 0);
1199 atomic_set(&root->log_writers, 0);
1200 atomic_set(&root->log_batch, 0);
1201 refcount_set(&root->refs, 1);
1202 atomic_set(&root->snapshot_force_cow, 0);
1203 atomic_set(&root->nr_swapfiles, 0);
1204 root->log_transid = 0;
1205 root->log_transid_committed = -1;
1206 root->last_log_commit = 0;
1209 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1210 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1211 extent_io_tree_init(fs_info, &root->log_csum_range,
1212 IO_TREE_LOG_CSUM_RANGE, NULL);
1215 spin_lock_init(&root->root_item_lock);
1216 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1217 #ifdef CONFIG_BTRFS_DEBUG
1218 INIT_LIST_HEAD(&root->leak_list);
1219 spin_lock(&fs_info->fs_roots_radix_lock);
1220 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1221 spin_unlock(&fs_info->fs_roots_radix_lock);
1225 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1226 u64 objectid, gfp_t flags)
1228 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1230 __setup_root(root, fs_info, objectid);
1234 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1235 /* Should only be used by the testing infrastructure */
1236 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1238 struct btrfs_root *root;
1241 return ERR_PTR(-EINVAL);
1243 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1245 return ERR_PTR(-ENOMEM);
1247 /* We don't use the stripesize in selftest, set it as sectorsize */
1248 root->alloc_bytenr = 0;
1254 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1256 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1257 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1259 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1262 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1264 const struct btrfs_key *key = k;
1265 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1267 return btrfs_comp_cpu_keys(key, &root->root_key);
1270 int btrfs_global_root_insert(struct btrfs_root *root)
1272 struct btrfs_fs_info *fs_info = root->fs_info;
1273 struct rb_node *tmp;
1275 write_lock(&fs_info->global_root_lock);
1276 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1277 write_unlock(&fs_info->global_root_lock);
1280 return tmp ? -EEXIST : 0;
1283 void btrfs_global_root_delete(struct btrfs_root *root)
1285 struct btrfs_fs_info *fs_info = root->fs_info;
1287 write_lock(&fs_info->global_root_lock);
1288 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1289 write_unlock(&fs_info->global_root_lock);
1292 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1293 struct btrfs_key *key)
1295 struct rb_node *node;
1296 struct btrfs_root *root = NULL;
1298 read_lock(&fs_info->global_root_lock);
1299 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1301 root = container_of(node, struct btrfs_root, rb_node);
1302 read_unlock(&fs_info->global_root_lock);
1307 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1309 struct btrfs_block_group *block_group;
1312 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1316 block_group = btrfs_lookup_block_group(fs_info, bytenr);
1318 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1319 ASSERT(block_group);
1322 ret = block_group->global_root_id;
1323 btrfs_put_block_group(block_group);
1328 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1330 struct btrfs_key key = {
1331 .objectid = BTRFS_CSUM_TREE_OBJECTID,
1332 .type = BTRFS_ROOT_ITEM_KEY,
1333 .offset = btrfs_global_root_id(fs_info, bytenr),
1336 return btrfs_global_root(fs_info, &key);
1339 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1341 struct btrfs_key key = {
1342 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
1343 .type = BTRFS_ROOT_ITEM_KEY,
1344 .offset = btrfs_global_root_id(fs_info, bytenr),
1347 return btrfs_global_root(fs_info, &key);
1350 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1353 struct btrfs_fs_info *fs_info = trans->fs_info;
1354 struct extent_buffer *leaf;
1355 struct btrfs_root *tree_root = fs_info->tree_root;
1356 struct btrfs_root *root;
1357 struct btrfs_key key;
1358 unsigned int nofs_flag;
1362 * We're holding a transaction handle, so use a NOFS memory allocation
1363 * context to avoid deadlock if reclaim happens.
1365 nofs_flag = memalloc_nofs_save();
1366 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1367 memalloc_nofs_restore(nofs_flag);
1369 return ERR_PTR(-ENOMEM);
1371 root->root_key.objectid = objectid;
1372 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1373 root->root_key.offset = 0;
1375 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1376 BTRFS_NESTING_NORMAL);
1378 ret = PTR_ERR(leaf);
1384 btrfs_mark_buffer_dirty(leaf);
1386 root->commit_root = btrfs_root_node(root);
1387 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1389 btrfs_set_root_flags(&root->root_item, 0);
1390 btrfs_set_root_limit(&root->root_item, 0);
1391 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1392 btrfs_set_root_generation(&root->root_item, trans->transid);
1393 btrfs_set_root_level(&root->root_item, 0);
1394 btrfs_set_root_refs(&root->root_item, 1);
1395 btrfs_set_root_used(&root->root_item, leaf->len);
1396 btrfs_set_root_last_snapshot(&root->root_item, 0);
1397 btrfs_set_root_dirid(&root->root_item, 0);
1398 if (is_fstree(objectid))
1399 generate_random_guid(root->root_item.uuid);
1401 export_guid(root->root_item.uuid, &guid_null);
1402 btrfs_set_root_drop_level(&root->root_item, 0);
1404 btrfs_tree_unlock(leaf);
1406 key.objectid = objectid;
1407 key.type = BTRFS_ROOT_ITEM_KEY;
1409 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1417 btrfs_tree_unlock(leaf);
1419 btrfs_put_root(root);
1421 return ERR_PTR(ret);
1424 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1425 struct btrfs_fs_info *fs_info)
1427 struct btrfs_root *root;
1429 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1431 return ERR_PTR(-ENOMEM);
1433 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1434 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1435 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1440 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1441 struct btrfs_root *root)
1443 struct extent_buffer *leaf;
1446 * DON'T set SHAREABLE bit for log trees.
1448 * Log trees are not exposed to user space thus can't be snapshotted,
1449 * and they go away before a real commit is actually done.
1451 * They do store pointers to file data extents, and those reference
1452 * counts still get updated (along with back refs to the log tree).
1455 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1456 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1458 return PTR_ERR(leaf);
1462 btrfs_mark_buffer_dirty(root->node);
1463 btrfs_tree_unlock(root->node);
1468 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1469 struct btrfs_fs_info *fs_info)
1471 struct btrfs_root *log_root;
1473 log_root = alloc_log_tree(trans, fs_info);
1474 if (IS_ERR(log_root))
1475 return PTR_ERR(log_root);
1477 if (!btrfs_is_zoned(fs_info)) {
1478 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1481 btrfs_put_root(log_root);
1486 WARN_ON(fs_info->log_root_tree);
1487 fs_info->log_root_tree = log_root;
1491 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1492 struct btrfs_root *root)
1494 struct btrfs_fs_info *fs_info = root->fs_info;
1495 struct btrfs_root *log_root;
1496 struct btrfs_inode_item *inode_item;
1499 log_root = alloc_log_tree(trans, fs_info);
1500 if (IS_ERR(log_root))
1501 return PTR_ERR(log_root);
1503 ret = btrfs_alloc_log_tree_node(trans, log_root);
1505 btrfs_put_root(log_root);
1509 log_root->last_trans = trans->transid;
1510 log_root->root_key.offset = root->root_key.objectid;
1512 inode_item = &log_root->root_item.inode;
1513 btrfs_set_stack_inode_generation(inode_item, 1);
1514 btrfs_set_stack_inode_size(inode_item, 3);
1515 btrfs_set_stack_inode_nlink(inode_item, 1);
1516 btrfs_set_stack_inode_nbytes(inode_item,
1518 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1520 btrfs_set_root_node(&log_root->root_item, log_root->node);
1522 WARN_ON(root->log_root);
1523 root->log_root = log_root;
1524 root->log_transid = 0;
1525 root->log_transid_committed = -1;
1526 root->last_log_commit = 0;
1530 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1531 struct btrfs_path *path,
1532 struct btrfs_key *key)
1534 struct btrfs_root *root;
1535 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1540 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1542 return ERR_PTR(-ENOMEM);
1544 ret = btrfs_find_root(tree_root, key, path,
1545 &root->root_item, &root->root_key);
1552 generation = btrfs_root_generation(&root->root_item);
1553 level = btrfs_root_level(&root->root_item);
1554 root->node = read_tree_block(fs_info,
1555 btrfs_root_bytenr(&root->root_item),
1556 key->objectid, generation, level, NULL);
1557 if (IS_ERR(root->node)) {
1558 ret = PTR_ERR(root->node);
1562 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1566 root->commit_root = btrfs_root_node(root);
1569 btrfs_put_root(root);
1570 return ERR_PTR(ret);
1573 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1574 struct btrfs_key *key)
1576 struct btrfs_root *root;
1577 struct btrfs_path *path;
1579 path = btrfs_alloc_path();
1581 return ERR_PTR(-ENOMEM);
1582 root = read_tree_root_path(tree_root, path, key);
1583 btrfs_free_path(path);
1589 * Initialize subvolume root in-memory structure
1591 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1593 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1596 unsigned int nofs_flag;
1599 * We might be called under a transaction (e.g. indirect backref
1600 * resolution) which could deadlock if it triggers memory reclaim
1602 nofs_flag = memalloc_nofs_save();
1603 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1604 memalloc_nofs_restore(nofs_flag);
1608 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1609 !btrfs_is_data_reloc_root(root)) {
1610 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1611 btrfs_check_and_init_root_item(&root->root_item);
1615 * Don't assign anonymous block device to roots that are not exposed to
1616 * userspace, the id pool is limited to 1M
1618 if (is_fstree(root->root_key.objectid) &&
1619 btrfs_root_refs(&root->root_item) > 0) {
1621 ret = get_anon_bdev(&root->anon_dev);
1625 root->anon_dev = anon_dev;
1629 mutex_lock(&root->objectid_mutex);
1630 ret = btrfs_init_root_free_objectid(root);
1632 mutex_unlock(&root->objectid_mutex);
1636 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1638 mutex_unlock(&root->objectid_mutex);
1642 /* The caller is responsible to call btrfs_free_fs_root */
1646 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1649 struct btrfs_root *root;
1651 spin_lock(&fs_info->fs_roots_radix_lock);
1652 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1653 (unsigned long)root_id);
1655 root = btrfs_grab_root(root);
1656 spin_unlock(&fs_info->fs_roots_radix_lock);
1660 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1663 struct btrfs_key key = {
1664 .objectid = objectid,
1665 .type = BTRFS_ROOT_ITEM_KEY,
1669 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1670 return btrfs_grab_root(fs_info->tree_root);
1671 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1672 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1673 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1674 return btrfs_grab_root(fs_info->chunk_root);
1675 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1676 return btrfs_grab_root(fs_info->dev_root);
1677 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1678 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1679 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1680 return btrfs_grab_root(fs_info->quota_root) ?
1681 fs_info->quota_root : ERR_PTR(-ENOENT);
1682 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1683 return btrfs_grab_root(fs_info->uuid_root) ?
1684 fs_info->uuid_root : ERR_PTR(-ENOENT);
1685 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1686 struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1688 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1693 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1694 struct btrfs_root *root)
1698 ret = radix_tree_preload(GFP_NOFS);
1702 spin_lock(&fs_info->fs_roots_radix_lock);
1703 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1704 (unsigned long)root->root_key.objectid,
1707 btrfs_grab_root(root);
1708 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1710 spin_unlock(&fs_info->fs_roots_radix_lock);
1711 radix_tree_preload_end();
1716 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1718 #ifdef CONFIG_BTRFS_DEBUG
1719 struct btrfs_root *root;
1721 while (!list_empty(&fs_info->allocated_roots)) {
1722 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1724 root = list_first_entry(&fs_info->allocated_roots,
1725 struct btrfs_root, leak_list);
1726 btrfs_err(fs_info, "leaked root %s refcount %d",
1727 btrfs_root_name(&root->root_key, buf),
1728 refcount_read(&root->refs));
1729 while (refcount_read(&root->refs) > 1)
1730 btrfs_put_root(root);
1731 btrfs_put_root(root);
1736 static void free_global_roots(struct btrfs_fs_info *fs_info)
1738 struct btrfs_root *root;
1739 struct rb_node *node;
1741 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1742 root = rb_entry(node, struct btrfs_root, rb_node);
1743 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1744 btrfs_put_root(root);
1748 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1750 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1751 percpu_counter_destroy(&fs_info->delalloc_bytes);
1752 percpu_counter_destroy(&fs_info->ordered_bytes);
1753 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1754 btrfs_free_csum_hash(fs_info);
1755 btrfs_free_stripe_hash_table(fs_info);
1756 btrfs_free_ref_cache(fs_info);
1757 kfree(fs_info->balance_ctl);
1758 kfree(fs_info->delayed_root);
1759 free_global_roots(fs_info);
1760 btrfs_put_root(fs_info->tree_root);
1761 btrfs_put_root(fs_info->chunk_root);
1762 btrfs_put_root(fs_info->dev_root);
1763 btrfs_put_root(fs_info->quota_root);
1764 btrfs_put_root(fs_info->uuid_root);
1765 btrfs_put_root(fs_info->fs_root);
1766 btrfs_put_root(fs_info->data_reloc_root);
1767 btrfs_put_root(fs_info->block_group_root);
1768 btrfs_check_leaked_roots(fs_info);
1769 btrfs_extent_buffer_leak_debug_check(fs_info);
1770 kfree(fs_info->super_copy);
1771 kfree(fs_info->super_for_commit);
1772 kfree(fs_info->subpage_info);
1778 * Get an in-memory reference of a root structure.
1780 * For essential trees like root/extent tree, we grab it from fs_info directly.
1781 * For subvolume trees, we check the cached filesystem roots first. If not
1782 * found, then read it from disk and add it to cached fs roots.
1784 * Caller should release the root by calling btrfs_put_root() after the usage.
1786 * NOTE: Reloc and log trees can't be read by this function as they share the
1787 * same root objectid.
1789 * @objectid: root id
1790 * @anon_dev: preallocated anonymous block device number for new roots,
1791 * pass 0 for new allocation.
1792 * @check_ref: whether to check root item references, If true, return -ENOENT
1795 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1796 u64 objectid, dev_t anon_dev,
1799 struct btrfs_root *root;
1800 struct btrfs_path *path;
1801 struct btrfs_key key;
1804 root = btrfs_get_global_root(fs_info, objectid);
1808 root = btrfs_lookup_fs_root(fs_info, objectid);
1810 /* Shouldn't get preallocated anon_dev for cached roots */
1812 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1813 btrfs_put_root(root);
1814 return ERR_PTR(-ENOENT);
1819 key.objectid = objectid;
1820 key.type = BTRFS_ROOT_ITEM_KEY;
1821 key.offset = (u64)-1;
1822 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1826 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1831 ret = btrfs_init_fs_root(root, anon_dev);
1835 path = btrfs_alloc_path();
1840 key.objectid = BTRFS_ORPHAN_OBJECTID;
1841 key.type = BTRFS_ORPHAN_ITEM_KEY;
1842 key.offset = objectid;
1844 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1845 btrfs_free_path(path);
1849 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1851 ret = btrfs_insert_fs_root(fs_info, root);
1853 if (ret == -EEXIST) {
1854 btrfs_put_root(root);
1862 * If our caller provided us an anonymous device, then it's his
1863 * responsability to free it in case we fail. So we have to set our
1864 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1865 * and once again by our caller.
1869 btrfs_put_root(root);
1870 return ERR_PTR(ret);
1874 * Get in-memory reference of a root structure
1876 * @objectid: tree objectid
1877 * @check_ref: if set, verify that the tree exists and the item has at least
1880 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1881 u64 objectid, bool check_ref)
1883 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1887 * Get in-memory reference of a root structure, created as new, optionally pass
1888 * the anonymous block device id
1890 * @objectid: tree objectid
1891 * @anon_dev: if zero, allocate a new anonymous block device or use the
1894 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1895 u64 objectid, dev_t anon_dev)
1897 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1901 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1902 * @fs_info: the fs_info
1903 * @objectid: the objectid we need to lookup
1905 * This is exclusively used for backref walking, and exists specifically because
1906 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1907 * creation time, which means we may have to read the tree_root in order to look
1908 * up a fs root that is not in memory. If the root is not in memory we will
1909 * read the tree root commit root and look up the fs root from there. This is a
1910 * temporary root, it will not be inserted into the radix tree as it doesn't
1911 * have the most uptodate information, it'll simply be discarded once the
1912 * backref code is finished using the root.
1914 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1915 struct btrfs_path *path,
1918 struct btrfs_root *root;
1919 struct btrfs_key key;
1921 ASSERT(path->search_commit_root && path->skip_locking);
1924 * This can return -ENOENT if we ask for a root that doesn't exist, but
1925 * since this is called via the backref walking code we won't be looking
1926 * up a root that doesn't exist, unless there's corruption. So if root
1927 * != NULL just return it.
1929 root = btrfs_get_global_root(fs_info, objectid);
1933 root = btrfs_lookup_fs_root(fs_info, objectid);
1937 key.objectid = objectid;
1938 key.type = BTRFS_ROOT_ITEM_KEY;
1939 key.offset = (u64)-1;
1940 root = read_tree_root_path(fs_info->tree_root, path, &key);
1941 btrfs_release_path(path);
1947 * called by the kthread helper functions to finally call the bio end_io
1948 * functions. This is where read checksum verification actually happens
1950 static void end_workqueue_fn(struct btrfs_work *work)
1953 struct btrfs_end_io_wq *end_io_wq;
1955 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1956 bio = end_io_wq->bio;
1958 bio->bi_status = end_io_wq->status;
1959 bio->bi_private = end_io_wq->private;
1960 bio->bi_end_io = end_io_wq->end_io;
1962 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1965 static int cleaner_kthread(void *arg)
1967 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)arg;
1973 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1975 /* Make the cleaner go to sleep early. */
1976 if (btrfs_need_cleaner_sleep(fs_info))
1980 * Do not do anything if we might cause open_ctree() to block
1981 * before we have finished mounting the filesystem.
1983 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1986 if (!mutex_trylock(&fs_info->cleaner_mutex))
1990 * Avoid the problem that we change the status of the fs
1991 * during the above check and trylock.
1993 if (btrfs_need_cleaner_sleep(fs_info)) {
1994 mutex_unlock(&fs_info->cleaner_mutex);
1998 btrfs_run_delayed_iputs(fs_info);
2000 again = btrfs_clean_one_deleted_snapshot(fs_info);
2001 mutex_unlock(&fs_info->cleaner_mutex);
2004 * The defragger has dealt with the R/O remount and umount,
2005 * needn't do anything special here.
2007 btrfs_run_defrag_inodes(fs_info);
2010 * Acquires fs_info->reclaim_bgs_lock to avoid racing
2011 * with relocation (btrfs_relocate_chunk) and relocation
2012 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
2013 * after acquiring fs_info->reclaim_bgs_lock. So we
2014 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
2015 * unused block groups.
2017 btrfs_delete_unused_bgs(fs_info);
2020 * Reclaim block groups in the reclaim_bgs list after we deleted
2021 * all unused block_groups. This possibly gives us some more free
2024 btrfs_reclaim_bgs(fs_info);
2026 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
2027 if (kthread_should_park())
2029 if (kthread_should_stop())
2032 set_current_state(TASK_INTERRUPTIBLE);
2034 __set_current_state(TASK_RUNNING);
2039 static int transaction_kthread(void *arg)
2041 struct btrfs_root *root = arg;
2042 struct btrfs_fs_info *fs_info = root->fs_info;
2043 struct btrfs_trans_handle *trans;
2044 struct btrfs_transaction *cur;
2047 unsigned long delay;
2051 cannot_commit = false;
2052 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
2053 mutex_lock(&fs_info->transaction_kthread_mutex);
2055 spin_lock(&fs_info->trans_lock);
2056 cur = fs_info->running_transaction;
2058 spin_unlock(&fs_info->trans_lock);
2062 delta = ktime_get_seconds() - cur->start_time;
2063 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
2064 cur->state < TRANS_STATE_COMMIT_START &&
2065 delta < fs_info->commit_interval) {
2066 spin_unlock(&fs_info->trans_lock);
2067 delay -= msecs_to_jiffies((delta - 1) * 1000);
2069 msecs_to_jiffies(fs_info->commit_interval * 1000));
2072 transid = cur->transid;
2073 spin_unlock(&fs_info->trans_lock);
2075 /* If the file system is aborted, this will always fail. */
2076 trans = btrfs_attach_transaction(root);
2077 if (IS_ERR(trans)) {
2078 if (PTR_ERR(trans) != -ENOENT)
2079 cannot_commit = true;
2082 if (transid == trans->transid) {
2083 btrfs_commit_transaction(trans);
2085 btrfs_end_transaction(trans);
2088 wake_up_process(fs_info->cleaner_kthread);
2089 mutex_unlock(&fs_info->transaction_kthread_mutex);
2091 if (BTRFS_FS_ERROR(fs_info))
2092 btrfs_cleanup_transaction(fs_info);
2093 if (!kthread_should_stop() &&
2094 (!btrfs_transaction_blocked(fs_info) ||
2096 schedule_timeout_interruptible(delay);
2097 } while (!kthread_should_stop());
2102 * This will find the highest generation in the array of root backups. The
2103 * index of the highest array is returned, or -EINVAL if we can't find
2106 * We check to make sure the array is valid by comparing the
2107 * generation of the latest root in the array with the generation
2108 * in the super block. If they don't match we pitch it.
2110 static int find_newest_super_backup(struct btrfs_fs_info *info)
2112 const u64 newest_gen = btrfs_super_generation(info->super_copy);
2114 struct btrfs_root_backup *root_backup;
2117 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2118 root_backup = info->super_copy->super_roots + i;
2119 cur = btrfs_backup_tree_root_gen(root_backup);
2120 if (cur == newest_gen)
2128 * copy all the root pointers into the super backup array.
2129 * this will bump the backup pointer by one when it is
2132 static void backup_super_roots(struct btrfs_fs_info *info)
2134 const int next_backup = info->backup_root_index;
2135 struct btrfs_root_backup *root_backup;
2137 root_backup = info->super_for_commit->super_roots + next_backup;
2140 * make sure all of our padding and empty slots get zero filled
2141 * regardless of which ones we use today
2143 memset(root_backup, 0, sizeof(*root_backup));
2145 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2147 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2148 btrfs_set_backup_tree_root_gen(root_backup,
2149 btrfs_header_generation(info->tree_root->node));
2151 btrfs_set_backup_tree_root_level(root_backup,
2152 btrfs_header_level(info->tree_root->node));
2154 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2155 btrfs_set_backup_chunk_root_gen(root_backup,
2156 btrfs_header_generation(info->chunk_root->node));
2157 btrfs_set_backup_chunk_root_level(root_backup,
2158 btrfs_header_level(info->chunk_root->node));
2160 if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) {
2161 btrfs_set_backup_block_group_root(root_backup,
2162 info->block_group_root->node->start);
2163 btrfs_set_backup_block_group_root_gen(root_backup,
2164 btrfs_header_generation(info->block_group_root->node));
2165 btrfs_set_backup_block_group_root_level(root_backup,
2166 btrfs_header_level(info->block_group_root->node));
2168 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2169 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2171 btrfs_set_backup_extent_root(root_backup,
2172 extent_root->node->start);
2173 btrfs_set_backup_extent_root_gen(root_backup,
2174 btrfs_header_generation(extent_root->node));
2175 btrfs_set_backup_extent_root_level(root_backup,
2176 btrfs_header_level(extent_root->node));
2178 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2179 btrfs_set_backup_csum_root_gen(root_backup,
2180 btrfs_header_generation(csum_root->node));
2181 btrfs_set_backup_csum_root_level(root_backup,
2182 btrfs_header_level(csum_root->node));
2186 * we might commit during log recovery, which happens before we set
2187 * the fs_root. Make sure it is valid before we fill it in.
2189 if (info->fs_root && info->fs_root->node) {
2190 btrfs_set_backup_fs_root(root_backup,
2191 info->fs_root->node->start);
2192 btrfs_set_backup_fs_root_gen(root_backup,
2193 btrfs_header_generation(info->fs_root->node));
2194 btrfs_set_backup_fs_root_level(root_backup,
2195 btrfs_header_level(info->fs_root->node));
2198 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2199 btrfs_set_backup_dev_root_gen(root_backup,
2200 btrfs_header_generation(info->dev_root->node));
2201 btrfs_set_backup_dev_root_level(root_backup,
2202 btrfs_header_level(info->dev_root->node));
2204 btrfs_set_backup_total_bytes(root_backup,
2205 btrfs_super_total_bytes(info->super_copy));
2206 btrfs_set_backup_bytes_used(root_backup,
2207 btrfs_super_bytes_used(info->super_copy));
2208 btrfs_set_backup_num_devices(root_backup,
2209 btrfs_super_num_devices(info->super_copy));
2212 * if we don't copy this out to the super_copy, it won't get remembered
2213 * for the next commit
2215 memcpy(&info->super_copy->super_roots,
2216 &info->super_for_commit->super_roots,
2217 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2221 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2222 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2224 * fs_info - filesystem whose backup roots need to be read
2225 * priority - priority of backup root required
2227 * Returns backup root index on success and -EINVAL otherwise.
2229 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2231 int backup_index = find_newest_super_backup(fs_info);
2232 struct btrfs_super_block *super = fs_info->super_copy;
2233 struct btrfs_root_backup *root_backup;
2235 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2237 return backup_index;
2239 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2240 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2245 root_backup = super->super_roots + backup_index;
2247 btrfs_set_super_generation(super,
2248 btrfs_backup_tree_root_gen(root_backup));
2249 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2250 btrfs_set_super_root_level(super,
2251 btrfs_backup_tree_root_level(root_backup));
2252 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2255 * Fixme: the total bytes and num_devices need to match or we should
2258 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2259 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2261 return backup_index;
2264 /* helper to cleanup workers */
2265 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2267 btrfs_destroy_workqueue(fs_info->fixup_workers);
2268 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2269 btrfs_destroy_workqueue(fs_info->workers);
2270 btrfs_destroy_workqueue(fs_info->endio_workers);
2271 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2272 btrfs_destroy_workqueue(fs_info->rmw_workers);
2273 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2274 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2275 btrfs_destroy_workqueue(fs_info->delayed_workers);
2276 btrfs_destroy_workqueue(fs_info->caching_workers);
2277 btrfs_destroy_workqueue(fs_info->flush_workers);
2278 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2279 if (fs_info->discard_ctl.discard_workers)
2280 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2282 * Now that all other work queues are destroyed, we can safely destroy
2283 * the queues used for metadata I/O, since tasks from those other work
2284 * queues can do metadata I/O operations.
2286 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2287 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2290 static void free_root_extent_buffers(struct btrfs_root *root)
2293 free_extent_buffer(root->node);
2294 free_extent_buffer(root->commit_root);
2296 root->commit_root = NULL;
2300 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2302 struct btrfs_root *root, *tmp;
2304 rbtree_postorder_for_each_entry_safe(root, tmp,
2305 &fs_info->global_root_tree,
2307 free_root_extent_buffers(root);
2310 /* helper to cleanup tree roots */
2311 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2313 free_root_extent_buffers(info->tree_root);
2315 free_global_root_pointers(info);
2316 free_root_extent_buffers(info->dev_root);
2317 free_root_extent_buffers(info->quota_root);
2318 free_root_extent_buffers(info->uuid_root);
2319 free_root_extent_buffers(info->fs_root);
2320 free_root_extent_buffers(info->data_reloc_root);
2321 free_root_extent_buffers(info->block_group_root);
2322 if (free_chunk_root)
2323 free_root_extent_buffers(info->chunk_root);
2326 void btrfs_put_root(struct btrfs_root *root)
2331 if (refcount_dec_and_test(&root->refs)) {
2332 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2333 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2335 free_anon_bdev(root->anon_dev);
2336 btrfs_drew_lock_destroy(&root->snapshot_lock);
2337 free_root_extent_buffers(root);
2338 #ifdef CONFIG_BTRFS_DEBUG
2339 spin_lock(&root->fs_info->fs_roots_radix_lock);
2340 list_del_init(&root->leak_list);
2341 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2347 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2350 struct btrfs_root *gang[8];
2353 while (!list_empty(&fs_info->dead_roots)) {
2354 gang[0] = list_entry(fs_info->dead_roots.next,
2355 struct btrfs_root, root_list);
2356 list_del(&gang[0]->root_list);
2358 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2359 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2360 btrfs_put_root(gang[0]);
2364 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2369 for (i = 0; i < ret; i++)
2370 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2374 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2376 mutex_init(&fs_info->scrub_lock);
2377 atomic_set(&fs_info->scrubs_running, 0);
2378 atomic_set(&fs_info->scrub_pause_req, 0);
2379 atomic_set(&fs_info->scrubs_paused, 0);
2380 atomic_set(&fs_info->scrub_cancel_req, 0);
2381 init_waitqueue_head(&fs_info->scrub_pause_wait);
2382 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2385 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2387 spin_lock_init(&fs_info->balance_lock);
2388 mutex_init(&fs_info->balance_mutex);
2389 atomic_set(&fs_info->balance_pause_req, 0);
2390 atomic_set(&fs_info->balance_cancel_req, 0);
2391 fs_info->balance_ctl = NULL;
2392 init_waitqueue_head(&fs_info->balance_wait_q);
2393 atomic_set(&fs_info->reloc_cancel_req, 0);
2396 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2398 struct inode *inode = fs_info->btree_inode;
2400 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2401 set_nlink(inode, 1);
2403 * we set the i_size on the btree inode to the max possible int.
2404 * the real end of the address space is determined by all of
2405 * the devices in the system
2407 inode->i_size = OFFSET_MAX;
2408 inode->i_mapping->a_ops = &btree_aops;
2410 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2411 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2412 IO_TREE_BTREE_INODE_IO, inode);
2413 BTRFS_I(inode)->io_tree.track_uptodate = false;
2414 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2416 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2417 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2418 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2419 btrfs_insert_inode_hash(inode);
2422 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2424 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2425 init_rwsem(&fs_info->dev_replace.rwsem);
2426 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2429 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2431 spin_lock_init(&fs_info->qgroup_lock);
2432 mutex_init(&fs_info->qgroup_ioctl_lock);
2433 fs_info->qgroup_tree = RB_ROOT;
2434 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2435 fs_info->qgroup_seq = 1;
2436 fs_info->qgroup_ulist = NULL;
2437 fs_info->qgroup_rescan_running = false;
2438 mutex_init(&fs_info->qgroup_rescan_lock);
2441 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2443 u32 max_active = fs_info->thread_pool_size;
2444 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2447 btrfs_alloc_workqueue(fs_info, "worker",
2448 flags | WQ_HIGHPRI, max_active, 16);
2450 fs_info->delalloc_workers =
2451 btrfs_alloc_workqueue(fs_info, "delalloc",
2452 flags, max_active, 2);
2454 fs_info->flush_workers =
2455 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2456 flags, max_active, 0);
2458 fs_info->caching_workers =
2459 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2461 fs_info->fixup_workers =
2462 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2465 * endios are largely parallel and should have a very
2468 fs_info->endio_workers =
2469 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2470 fs_info->endio_meta_workers =
2471 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2473 fs_info->endio_meta_write_workers =
2474 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2476 fs_info->endio_raid56_workers =
2477 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2479 fs_info->rmw_workers =
2480 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2481 fs_info->endio_write_workers =
2482 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2484 fs_info->endio_freespace_worker =
2485 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2487 fs_info->delayed_workers =
2488 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2490 fs_info->qgroup_rescan_workers =
2491 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2492 fs_info->discard_ctl.discard_workers =
2493 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2495 if (!(fs_info->workers && fs_info->delalloc_workers &&
2496 fs_info->flush_workers &&
2497 fs_info->endio_workers && fs_info->endio_meta_workers &&
2498 fs_info->endio_meta_write_workers &&
2499 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2500 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2501 fs_info->caching_workers && fs_info->fixup_workers &&
2502 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2503 fs_info->discard_ctl.discard_workers)) {
2510 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2512 struct crypto_shash *csum_shash;
2513 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2515 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2517 if (IS_ERR(csum_shash)) {
2518 btrfs_err(fs_info, "error allocating %s hash for checksum",
2520 return PTR_ERR(csum_shash);
2523 fs_info->csum_shash = csum_shash;
2528 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2529 struct btrfs_fs_devices *fs_devices)
2532 struct btrfs_root *log_tree_root;
2533 struct btrfs_super_block *disk_super = fs_info->super_copy;
2534 u64 bytenr = btrfs_super_log_root(disk_super);
2535 int level = btrfs_super_log_root_level(disk_super);
2537 if (fs_devices->rw_devices == 0) {
2538 btrfs_warn(fs_info, "log replay required on RO media");
2542 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2547 log_tree_root->node = read_tree_block(fs_info, bytenr,
2548 BTRFS_TREE_LOG_OBJECTID,
2549 fs_info->generation + 1, level,
2551 if (IS_ERR(log_tree_root->node)) {
2552 btrfs_warn(fs_info, "failed to read log tree");
2553 ret = PTR_ERR(log_tree_root->node);
2554 log_tree_root->node = NULL;
2555 btrfs_put_root(log_tree_root);
2558 if (!extent_buffer_uptodate(log_tree_root->node)) {
2559 btrfs_err(fs_info, "failed to read log tree");
2560 btrfs_put_root(log_tree_root);
2564 /* returns with log_tree_root freed on success */
2565 ret = btrfs_recover_log_trees(log_tree_root);
2567 btrfs_handle_fs_error(fs_info, ret,
2568 "Failed to recover log tree");
2569 btrfs_put_root(log_tree_root);
2573 if (sb_rdonly(fs_info->sb)) {
2574 ret = btrfs_commit_super(fs_info);
2582 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2583 struct btrfs_path *path, u64 objectid,
2586 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2587 struct btrfs_root *root;
2588 u64 max_global_id = 0;
2590 struct btrfs_key key = {
2591 .objectid = objectid,
2592 .type = BTRFS_ROOT_ITEM_KEY,
2597 /* If we have IGNOREDATACSUMS skip loading these roots. */
2598 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2599 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2600 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2605 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2609 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2610 ret = btrfs_next_leaf(tree_root, path);
2619 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2620 if (key.objectid != objectid)
2622 btrfs_release_path(path);
2625 * Just worry about this for extent tree, it'll be the same for
2628 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2629 max_global_id = max(max_global_id, key.offset);
2632 root = read_tree_root_path(tree_root, path, &key);
2634 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2635 ret = PTR_ERR(root);
2638 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2639 ret = btrfs_global_root_insert(root);
2641 btrfs_put_root(root);
2646 btrfs_release_path(path);
2648 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2649 fs_info->nr_global_roots = max_global_id + 1;
2651 if (!found || ret) {
2652 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2653 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2655 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2656 ret = ret ? ret : -ENOENT;
2659 btrfs_err(fs_info, "failed to load root %s", name);
2664 static int load_global_roots(struct btrfs_root *tree_root)
2666 struct btrfs_path *path;
2669 path = btrfs_alloc_path();
2673 ret = load_global_roots_objectid(tree_root, path,
2674 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2677 ret = load_global_roots_objectid(tree_root, path,
2678 BTRFS_CSUM_TREE_OBJECTID, "csum");
2681 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2683 ret = load_global_roots_objectid(tree_root, path,
2684 BTRFS_FREE_SPACE_TREE_OBJECTID,
2687 btrfs_free_path(path);
2691 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2693 struct btrfs_root *tree_root = fs_info->tree_root;
2694 struct btrfs_root *root;
2695 struct btrfs_key location;
2698 BUG_ON(!fs_info->tree_root);
2700 ret = load_global_roots(tree_root);
2704 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2705 location.type = BTRFS_ROOT_ITEM_KEY;
2706 location.offset = 0;
2708 root = btrfs_read_tree_root(tree_root, &location);
2710 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2711 ret = PTR_ERR(root);
2715 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2716 fs_info->dev_root = root;
2718 /* Initialize fs_info for all devices in any case */
2719 btrfs_init_devices_late(fs_info);
2722 * This tree can share blocks with some other fs tree during relocation
2723 * and we need a proper setup by btrfs_get_fs_root
2725 root = btrfs_get_fs_root(tree_root->fs_info,
2726 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2728 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2729 ret = PTR_ERR(root);
2733 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2734 fs_info->data_reloc_root = root;
2737 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2738 root = btrfs_read_tree_root(tree_root, &location);
2739 if (!IS_ERR(root)) {
2740 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2741 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2742 fs_info->quota_root = root;
2745 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2746 root = btrfs_read_tree_root(tree_root, &location);
2748 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2749 ret = PTR_ERR(root);
2754 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2755 fs_info->uuid_root = root;
2760 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2761 location.objectid, ret);
2766 * Real super block validation
2767 * NOTE: super csum type and incompat features will not be checked here.
2769 * @sb: super block to check
2770 * @mirror_num: the super block number to check its bytenr:
2771 * 0 the primary (1st) sb
2772 * 1, 2 2nd and 3rd backup copy
2773 * -1 skip bytenr check
2775 static int validate_super(struct btrfs_fs_info *fs_info,
2776 struct btrfs_super_block *sb, int mirror_num)
2778 u64 nodesize = btrfs_super_nodesize(sb);
2779 u64 sectorsize = btrfs_super_sectorsize(sb);
2782 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2783 btrfs_err(fs_info, "no valid FS found");
2786 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2787 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2788 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2791 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2792 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2793 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2796 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2797 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2798 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2801 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2802 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2803 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2808 * Check sectorsize and nodesize first, other check will need it.
2809 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2811 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2812 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2813 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2818 * For 4K page size, we only support 4K sector size.
2819 * For 64K page size, we support 64K and 4K sector sizes.
2821 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2822 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2823 sectorsize != SZ_64K))) {
2825 "sectorsize %llu not yet supported for page size %lu",
2826 sectorsize, PAGE_SIZE);
2830 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2831 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2832 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2835 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2836 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2837 le32_to_cpu(sb->__unused_leafsize), nodesize);
2841 /* Root alignment check */
2842 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2843 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2844 btrfs_super_root(sb));
2847 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2848 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2849 btrfs_super_chunk_root(sb));
2852 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2853 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2854 btrfs_super_log_root(sb));
2858 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2861 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2862 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2866 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2867 memcmp(fs_info->fs_devices->metadata_uuid,
2868 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2870 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2871 fs_info->super_copy->metadata_uuid,
2872 fs_info->fs_devices->metadata_uuid);
2876 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2877 BTRFS_FSID_SIZE) != 0) {
2879 "dev_item UUID does not match metadata fsid: %pU != %pU",
2880 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2885 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2888 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2889 btrfs_err(fs_info, "bytes_used is too small %llu",
2890 btrfs_super_bytes_used(sb));
2893 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2894 btrfs_err(fs_info, "invalid stripesize %u",
2895 btrfs_super_stripesize(sb));
2898 if (btrfs_super_num_devices(sb) > (1UL << 31))
2899 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2900 btrfs_super_num_devices(sb));
2901 if (btrfs_super_num_devices(sb) == 0) {
2902 btrfs_err(fs_info, "number of devices is 0");
2906 if (mirror_num >= 0 &&
2907 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2908 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2909 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2914 * Obvious sys_chunk_array corruptions, it must hold at least one key
2917 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2918 btrfs_err(fs_info, "system chunk array too big %u > %u",
2919 btrfs_super_sys_array_size(sb),
2920 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2923 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2924 + sizeof(struct btrfs_chunk)) {
2925 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2926 btrfs_super_sys_array_size(sb),
2927 sizeof(struct btrfs_disk_key)
2928 + sizeof(struct btrfs_chunk));
2933 * The generation is a global counter, we'll trust it more than the others
2934 * but it's still possible that it's the one that's wrong.
2936 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2938 "suspicious: generation < chunk_root_generation: %llu < %llu",
2939 btrfs_super_generation(sb),
2940 btrfs_super_chunk_root_generation(sb));
2941 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2942 && btrfs_super_cache_generation(sb) != (u64)-1)
2944 "suspicious: generation < cache_generation: %llu < %llu",
2945 btrfs_super_generation(sb),
2946 btrfs_super_cache_generation(sb));
2952 * Validation of super block at mount time.
2953 * Some checks already done early at mount time, like csum type and incompat
2954 * flags will be skipped.
2956 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2958 return validate_super(fs_info, fs_info->super_copy, 0);
2962 * Validation of super block at write time.
2963 * Some checks like bytenr check will be skipped as their values will be
2965 * Extra checks like csum type and incompat flags will be done here.
2967 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2968 struct btrfs_super_block *sb)
2972 ret = validate_super(fs_info, sb, -1);
2975 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2977 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2978 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2981 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2984 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2985 btrfs_super_incompat_flags(sb),
2986 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2992 "super block corruption detected before writing it to disk");
2996 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
3000 root->node = read_tree_block(root->fs_info, bytenr,
3001 root->root_key.objectid, gen, level, NULL);
3002 if (IS_ERR(root->node)) {
3003 ret = PTR_ERR(root->node);
3007 if (!extent_buffer_uptodate(root->node)) {
3008 free_extent_buffer(root->node);
3013 btrfs_set_root_node(&root->root_item, root->node);
3014 root->commit_root = btrfs_root_node(root);
3015 btrfs_set_root_refs(&root->root_item, 1);
3019 static int load_important_roots(struct btrfs_fs_info *fs_info)
3021 struct btrfs_super_block *sb = fs_info->super_copy;
3025 bytenr = btrfs_super_root(sb);
3026 gen = btrfs_super_generation(sb);
3027 level = btrfs_super_root_level(sb);
3028 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
3030 btrfs_warn(fs_info, "couldn't read tree root");
3034 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
3037 bytenr = btrfs_super_block_group_root(sb);
3038 gen = btrfs_super_block_group_root_generation(sb);
3039 level = btrfs_super_block_group_root_level(sb);
3040 ret = load_super_root(fs_info->block_group_root, bytenr, gen, level);
3042 btrfs_warn(fs_info, "couldn't read block group root");
3046 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
3048 int backup_index = find_newest_super_backup(fs_info);
3049 struct btrfs_super_block *sb = fs_info->super_copy;
3050 struct btrfs_root *tree_root = fs_info->tree_root;
3051 bool handle_error = false;
3055 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3056 struct btrfs_root *root;
3058 root = btrfs_alloc_root(fs_info, BTRFS_BLOCK_GROUP_TREE_OBJECTID,
3062 fs_info->block_group_root = root;
3065 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
3067 if (!IS_ERR(tree_root->node))
3068 free_extent_buffer(tree_root->node);
3069 tree_root->node = NULL;
3071 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3074 free_root_pointers(fs_info, 0);
3077 * Don't use the log in recovery mode, it won't be
3080 btrfs_set_super_log_root(sb, 0);
3082 /* We can't trust the free space cache either */
3083 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3085 ret = read_backup_root(fs_info, i);
3091 ret = load_important_roots(fs_info);
3093 handle_error = true;
3098 * No need to hold btrfs_root::objectid_mutex since the fs
3099 * hasn't been fully initialised and we are the only user
3101 ret = btrfs_init_root_free_objectid(tree_root);
3103 handle_error = true;
3107 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
3109 ret = btrfs_read_roots(fs_info);
3111 handle_error = true;
3115 /* All successful */
3116 fs_info->generation = btrfs_header_generation(tree_root->node);
3117 fs_info->last_trans_committed = fs_info->generation;
3118 fs_info->last_reloc_trans = 0;
3120 /* Always begin writing backup roots after the one being used */
3121 if (backup_index < 0) {
3122 fs_info->backup_root_index = 0;
3124 fs_info->backup_root_index = backup_index + 1;
3125 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3133 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3135 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
3136 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
3137 INIT_LIST_HEAD(&fs_info->trans_list);
3138 INIT_LIST_HEAD(&fs_info->dead_roots);
3139 INIT_LIST_HEAD(&fs_info->delayed_iputs);
3140 INIT_LIST_HEAD(&fs_info->delalloc_roots);
3141 INIT_LIST_HEAD(&fs_info->caching_block_groups);
3142 spin_lock_init(&fs_info->delalloc_root_lock);
3143 spin_lock_init(&fs_info->trans_lock);
3144 spin_lock_init(&fs_info->fs_roots_radix_lock);
3145 spin_lock_init(&fs_info->delayed_iput_lock);
3146 spin_lock_init(&fs_info->defrag_inodes_lock);
3147 spin_lock_init(&fs_info->super_lock);
3148 spin_lock_init(&fs_info->buffer_lock);
3149 spin_lock_init(&fs_info->unused_bgs_lock);
3150 spin_lock_init(&fs_info->treelog_bg_lock);
3151 spin_lock_init(&fs_info->zone_active_bgs_lock);
3152 spin_lock_init(&fs_info->relocation_bg_lock);
3153 rwlock_init(&fs_info->tree_mod_log_lock);
3154 rwlock_init(&fs_info->global_root_lock);
3155 mutex_init(&fs_info->unused_bg_unpin_mutex);
3156 mutex_init(&fs_info->reclaim_bgs_lock);
3157 mutex_init(&fs_info->reloc_mutex);
3158 mutex_init(&fs_info->delalloc_root_mutex);
3159 mutex_init(&fs_info->zoned_meta_io_lock);
3160 mutex_init(&fs_info->zoned_data_reloc_io_lock);
3161 seqlock_init(&fs_info->profiles_lock);
3163 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3164 INIT_LIST_HEAD(&fs_info->space_info);
3165 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3166 INIT_LIST_HEAD(&fs_info->unused_bgs);
3167 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3168 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3169 #ifdef CONFIG_BTRFS_DEBUG
3170 INIT_LIST_HEAD(&fs_info->allocated_roots);
3171 INIT_LIST_HEAD(&fs_info->allocated_ebs);
3172 spin_lock_init(&fs_info->eb_leak_lock);
3174 extent_map_tree_init(&fs_info->mapping_tree);
3175 btrfs_init_block_rsv(&fs_info->global_block_rsv,
3176 BTRFS_BLOCK_RSV_GLOBAL);
3177 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3178 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3179 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3180 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3181 BTRFS_BLOCK_RSV_DELOPS);
3182 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3183 BTRFS_BLOCK_RSV_DELREFS);
3185 atomic_set(&fs_info->async_delalloc_pages, 0);
3186 atomic_set(&fs_info->defrag_running, 0);
3187 atomic_set(&fs_info->nr_delayed_iputs, 0);
3188 atomic64_set(&fs_info->tree_mod_seq, 0);
3189 fs_info->global_root_tree = RB_ROOT;
3190 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3191 fs_info->metadata_ratio = 0;
3192 fs_info->defrag_inodes = RB_ROOT;
3193 atomic64_set(&fs_info->free_chunk_space, 0);
3194 fs_info->tree_mod_log = RB_ROOT;
3195 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3196 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3197 btrfs_init_ref_verify(fs_info);
3199 fs_info->thread_pool_size = min_t(unsigned long,
3200 num_online_cpus() + 2, 8);
3202 INIT_LIST_HEAD(&fs_info->ordered_roots);
3203 spin_lock_init(&fs_info->ordered_root_lock);
3205 btrfs_init_scrub(fs_info);
3206 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3207 fs_info->check_integrity_print_mask = 0;
3209 btrfs_init_balance(fs_info);
3210 btrfs_init_async_reclaim_work(fs_info);
3212 spin_lock_init(&fs_info->block_group_cache_lock);
3213 fs_info->block_group_cache_tree = RB_ROOT;
3214 fs_info->first_logical_byte = (u64)-1;
3216 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3217 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
3219 mutex_init(&fs_info->ordered_operations_mutex);
3220 mutex_init(&fs_info->tree_log_mutex);
3221 mutex_init(&fs_info->chunk_mutex);
3222 mutex_init(&fs_info->transaction_kthread_mutex);
3223 mutex_init(&fs_info->cleaner_mutex);
3224 mutex_init(&fs_info->ro_block_group_mutex);
3225 init_rwsem(&fs_info->commit_root_sem);
3226 init_rwsem(&fs_info->cleanup_work_sem);
3227 init_rwsem(&fs_info->subvol_sem);
3228 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3230 btrfs_init_dev_replace_locks(fs_info);
3231 btrfs_init_qgroup(fs_info);
3232 btrfs_discard_init(fs_info);
3234 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3235 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3237 init_waitqueue_head(&fs_info->transaction_throttle);
3238 init_waitqueue_head(&fs_info->transaction_wait);
3239 init_waitqueue_head(&fs_info->transaction_blocked_wait);
3240 init_waitqueue_head(&fs_info->async_submit_wait);
3241 init_waitqueue_head(&fs_info->delayed_iputs_wait);
3243 /* Usable values until the real ones are cached from the superblock */
3244 fs_info->nodesize = 4096;
3245 fs_info->sectorsize = 4096;
3246 fs_info->sectorsize_bits = ilog2(4096);
3247 fs_info->stripesize = 4096;
3249 spin_lock_init(&fs_info->swapfile_pins_lock);
3250 fs_info->swapfile_pins = RB_ROOT;
3252 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3253 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3256 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3261 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3262 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3264 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3268 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3272 fs_info->dirty_metadata_batch = PAGE_SIZE *
3273 (1 + ilog2(nr_cpu_ids));
3275 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3279 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3284 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3286 if (!fs_info->delayed_root)
3288 btrfs_init_delayed_root(fs_info->delayed_root);
3291 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3293 return btrfs_alloc_stripe_hash_table(fs_info);
3296 static int btrfs_uuid_rescan_kthread(void *data)
3298 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3302 * 1st step is to iterate through the existing UUID tree and
3303 * to delete all entries that contain outdated data.
3304 * 2nd step is to add all missing entries to the UUID tree.
3306 ret = btrfs_uuid_tree_iterate(fs_info);
3309 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3311 up(&fs_info->uuid_tree_rescan_sem);
3314 return btrfs_uuid_scan_kthread(data);
3317 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3319 struct task_struct *task;
3321 down(&fs_info->uuid_tree_rescan_sem);
3322 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3324 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3325 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3326 up(&fs_info->uuid_tree_rescan_sem);
3327 return PTR_ERR(task);
3334 * Some options only have meaning at mount time and shouldn't persist across
3335 * remounts, or be displayed. Clear these at the end of mount and remount
3338 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3340 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3341 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3345 * Mounting logic specific to read-write file systems. Shared by open_ctree
3346 * and btrfs_remount when remounting from read-only to read-write.
3348 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3351 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3352 bool clear_free_space_tree = false;
3354 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3355 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3356 clear_free_space_tree = true;
3357 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3358 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3359 btrfs_warn(fs_info, "free space tree is invalid");
3360 clear_free_space_tree = true;
3363 if (clear_free_space_tree) {
3364 btrfs_info(fs_info, "clearing free space tree");
3365 ret = btrfs_clear_free_space_tree(fs_info);
3368 "failed to clear free space tree: %d", ret);
3374 * btrfs_find_orphan_roots() is responsible for finding all the dead
3375 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3376 * them into the fs_info->fs_roots_radix tree. This must be done before
3377 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3378 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3379 * item before the root's tree is deleted - this means that if we unmount
3380 * or crash before the deletion completes, on the next mount we will not
3381 * delete what remains of the tree because the orphan item does not
3382 * exists anymore, which is what tells us we have a pending deletion.
3384 ret = btrfs_find_orphan_roots(fs_info);
3388 ret = btrfs_cleanup_fs_roots(fs_info);
3392 down_read(&fs_info->cleanup_work_sem);
3393 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3394 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3395 up_read(&fs_info->cleanup_work_sem);
3398 up_read(&fs_info->cleanup_work_sem);
3400 mutex_lock(&fs_info->cleaner_mutex);
3401 ret = btrfs_recover_relocation(fs_info);
3402 mutex_unlock(&fs_info->cleaner_mutex);
3404 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3408 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3409 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3410 btrfs_info(fs_info, "creating free space tree");
3411 ret = btrfs_create_free_space_tree(fs_info);
3414 "failed to create free space tree: %d", ret);
3419 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3420 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3425 ret = btrfs_resume_balance_async(fs_info);
3429 ret = btrfs_resume_dev_replace_async(fs_info);
3431 btrfs_warn(fs_info, "failed to resume dev_replace");
3435 btrfs_qgroup_rescan_resume(fs_info);
3437 if (!fs_info->uuid_root) {
3438 btrfs_info(fs_info, "creating UUID tree");
3439 ret = btrfs_create_uuid_tree(fs_info);
3442 "failed to create the UUID tree %d", ret);
3451 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3460 struct btrfs_super_block *disk_super;
3461 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3462 struct btrfs_root *tree_root;
3463 struct btrfs_root *chunk_root;
3468 ret = init_mount_fs_info(fs_info, sb);
3474 /* These need to be init'ed before we start creating inodes and such. */
3475 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3477 fs_info->tree_root = tree_root;
3478 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3480 fs_info->chunk_root = chunk_root;
3481 if (!tree_root || !chunk_root) {
3486 fs_info->btree_inode = new_inode(sb);
3487 if (!fs_info->btree_inode) {
3491 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3492 btrfs_init_btree_inode(fs_info);
3494 invalidate_bdev(fs_devices->latest_dev->bdev);
3497 * Read super block and check the signature bytes only
3499 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3500 if (IS_ERR(disk_super)) {
3501 err = PTR_ERR(disk_super);
3506 * Verify the type first, if that or the checksum value are
3507 * corrupted, we'll find out
3509 csum_type = btrfs_super_csum_type(disk_super);
3510 if (!btrfs_supported_super_csum(csum_type)) {
3511 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3514 btrfs_release_disk_super(disk_super);
3518 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3520 ret = btrfs_init_csum_hash(fs_info, csum_type);
3523 btrfs_release_disk_super(disk_super);
3528 * We want to check superblock checksum, the type is stored inside.
3529 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3531 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3532 btrfs_err(fs_info, "superblock checksum mismatch");
3534 btrfs_release_disk_super(disk_super);
3539 * super_copy is zeroed at allocation time and we never touch the
3540 * following bytes up to INFO_SIZE, the checksum is calculated from
3541 * the whole block of INFO_SIZE
3543 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3544 btrfs_release_disk_super(disk_super);
3546 disk_super = fs_info->super_copy;
3549 features = btrfs_super_flags(disk_super);
3550 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3551 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3552 btrfs_set_super_flags(disk_super, features);
3554 "found metadata UUID change in progress flag, clearing");
3557 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3558 sizeof(*fs_info->super_for_commit));
3560 ret = btrfs_validate_mount_super(fs_info);
3562 btrfs_err(fs_info, "superblock contains fatal errors");
3567 if (!btrfs_super_root(disk_super))
3570 /* check FS state, whether FS is broken. */
3571 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3572 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3575 * In the long term, we'll store the compression type in the super
3576 * block, and it'll be used for per file compression control.
3578 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3581 * Flag our filesystem as having big metadata blocks if they are bigger
3582 * than the page size.
3584 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3585 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3587 "flagging fs with big metadata feature");
3588 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3591 /* Set up fs_info before parsing mount options */
3592 nodesize = btrfs_super_nodesize(disk_super);
3593 sectorsize = btrfs_super_sectorsize(disk_super);
3594 stripesize = sectorsize;
3595 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3596 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3598 fs_info->nodesize = nodesize;
3599 fs_info->sectorsize = sectorsize;
3600 fs_info->sectorsize_bits = ilog2(sectorsize);
3601 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3602 fs_info->stripesize = stripesize;
3604 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3610 features = btrfs_super_incompat_flags(disk_super) &
3611 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3614 "cannot mount because of unsupported optional features (%llx)",
3620 features = btrfs_super_incompat_flags(disk_super);
3621 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3622 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3623 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3624 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3625 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3627 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3628 btrfs_info(fs_info, "has skinny extents");
3631 * mixed block groups end up with duplicate but slightly offset
3632 * extent buffers for the same range. It leads to corruptions
3634 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3635 (sectorsize != nodesize)) {
3637 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3638 nodesize, sectorsize);
3643 * Needn't use the lock because there is no other task which will
3646 btrfs_set_super_incompat_flags(disk_super, features);
3648 features = btrfs_super_compat_ro_flags(disk_super) &
3649 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3650 if (!sb_rdonly(sb) && features) {
3652 "cannot mount read-write because of unsupported optional features (%llx)",
3658 if (sectorsize < PAGE_SIZE) {
3659 struct btrfs_subpage_info *subpage_info;
3662 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3663 * going to be deprecated.
3665 * Force to use v2 cache for subpage case.
3667 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3668 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3669 "forcing free space tree for sector size %u with page size %lu",
3670 sectorsize, PAGE_SIZE);
3673 "read-write for sector size %u with page size %lu is experimental",
3674 sectorsize, PAGE_SIZE);
3675 if (btrfs_super_incompat_flags(fs_info->super_copy) &
3676 BTRFS_FEATURE_INCOMPAT_RAID56) {
3678 "RAID56 is not yet supported for sector size %u with page size %lu",
3679 sectorsize, PAGE_SIZE);
3683 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3686 btrfs_init_subpage_info(subpage_info, sectorsize);
3687 fs_info->subpage_info = subpage_info;
3690 ret = btrfs_init_workqueues(fs_info);
3693 goto fail_sb_buffer;
3696 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3697 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3699 sb->s_blocksize = sectorsize;
3700 sb->s_blocksize_bits = blksize_bits(sectorsize);
3701 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3703 mutex_lock(&fs_info->chunk_mutex);
3704 ret = btrfs_read_sys_array(fs_info);
3705 mutex_unlock(&fs_info->chunk_mutex);
3707 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3708 goto fail_sb_buffer;
3711 generation = btrfs_super_chunk_root_generation(disk_super);
3712 level = btrfs_super_chunk_root_level(disk_super);
3713 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3716 btrfs_err(fs_info, "failed to read chunk root");
3717 goto fail_tree_roots;
3720 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3721 offsetof(struct btrfs_header, chunk_tree_uuid),
3724 ret = btrfs_read_chunk_tree(fs_info);
3726 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3727 goto fail_tree_roots;
3731 * At this point we know all the devices that make this filesystem,
3732 * including the seed devices but we don't know yet if the replace
3733 * target is required. So free devices that are not part of this
3734 * filesystem but skip the replace target device which is checked
3735 * below in btrfs_init_dev_replace().
3737 btrfs_free_extra_devids(fs_devices);
3738 if (!fs_devices->latest_dev->bdev) {
3739 btrfs_err(fs_info, "failed to read devices");
3740 goto fail_tree_roots;
3743 ret = init_tree_roots(fs_info);
3745 goto fail_tree_roots;
3748 * Get zone type information of zoned block devices. This will also
3749 * handle emulation of a zoned filesystem if a regular device has the
3750 * zoned incompat feature flag set.
3752 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3755 "zoned: failed to read device zone info: %d",
3757 goto fail_block_groups;
3761 * If we have a uuid root and we're not being told to rescan we need to
3762 * check the generation here so we can set the
3763 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3764 * transaction during a balance or the log replay without updating the
3765 * uuid generation, and then if we crash we would rescan the uuid tree,
3766 * even though it was perfectly fine.
3768 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3769 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3770 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3772 ret = btrfs_verify_dev_extents(fs_info);
3775 "failed to verify dev extents against chunks: %d",
3777 goto fail_block_groups;
3779 ret = btrfs_recover_balance(fs_info);
3781 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3782 goto fail_block_groups;
3785 ret = btrfs_init_dev_stats(fs_info);
3787 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3788 goto fail_block_groups;
3791 ret = btrfs_init_dev_replace(fs_info);
3793 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3794 goto fail_block_groups;
3797 ret = btrfs_check_zoned_mode(fs_info);
3799 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3801 goto fail_block_groups;
3804 ret = btrfs_sysfs_add_fsid(fs_devices);
3806 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3808 goto fail_block_groups;
3811 ret = btrfs_sysfs_add_mounted(fs_info);
3813 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3814 goto fail_fsdev_sysfs;
3817 ret = btrfs_init_space_info(fs_info);
3819 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3823 ret = btrfs_read_block_groups(fs_info);
3825 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3829 btrfs_free_zone_cache(fs_info);
3831 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3832 !btrfs_check_rw_degradable(fs_info, NULL)) {
3834 "writable mount is not allowed due to too many missing devices");
3838 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3840 if (IS_ERR(fs_info->cleaner_kthread))
3843 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3845 "btrfs-transaction");
3846 if (IS_ERR(fs_info->transaction_kthread))
3849 if (!btrfs_test_opt(fs_info, NOSSD) &&
3850 !fs_info->fs_devices->rotating) {
3851 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3855 * Mount does not set all options immediately, we can do it now and do
3856 * not have to wait for transaction commit
3858 btrfs_apply_pending_changes(fs_info);
3860 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3861 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3862 ret = btrfsic_mount(fs_info, fs_devices,
3863 btrfs_test_opt(fs_info,
3864 CHECK_INTEGRITY_DATA) ? 1 : 0,
3865 fs_info->check_integrity_print_mask);
3868 "failed to initialize integrity check module: %d",
3872 ret = btrfs_read_qgroup_config(fs_info);
3874 goto fail_trans_kthread;
3876 if (btrfs_build_ref_tree(fs_info))
3877 btrfs_err(fs_info, "couldn't build ref tree");
3879 /* do not make disk changes in broken FS or nologreplay is given */
3880 if (btrfs_super_log_root(disk_super) != 0 &&
3881 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3882 btrfs_info(fs_info, "start tree-log replay");
3883 ret = btrfs_replay_log(fs_info, fs_devices);
3890 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3891 if (IS_ERR(fs_info->fs_root)) {
3892 err = PTR_ERR(fs_info->fs_root);
3893 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3894 fs_info->fs_root = NULL;
3901 ret = btrfs_start_pre_rw_mount(fs_info);
3903 close_ctree(fs_info);
3906 btrfs_discard_resume(fs_info);
3908 if (fs_info->uuid_root &&
3909 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3910 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3911 btrfs_info(fs_info, "checking UUID tree");
3912 ret = btrfs_check_uuid_tree(fs_info);
3915 "failed to check the UUID tree: %d", ret);
3916 close_ctree(fs_info);
3921 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3923 /* Kick the cleaner thread so it'll start deleting snapshots. */
3924 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3925 wake_up_process(fs_info->cleaner_kthread);
3928 btrfs_clear_oneshot_options(fs_info);
3932 btrfs_free_qgroup_config(fs_info);
3934 kthread_stop(fs_info->transaction_kthread);
3935 btrfs_cleanup_transaction(fs_info);
3936 btrfs_free_fs_roots(fs_info);
3938 kthread_stop(fs_info->cleaner_kthread);
3941 * make sure we're done with the btree inode before we stop our
3944 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3947 btrfs_sysfs_remove_mounted(fs_info);
3950 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3953 btrfs_put_block_group_cache(fs_info);
3956 if (fs_info->data_reloc_root)
3957 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3958 free_root_pointers(fs_info, true);
3959 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3962 btrfs_stop_all_workers(fs_info);
3963 btrfs_free_block_groups(fs_info);
3965 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3967 iput(fs_info->btree_inode);
3969 btrfs_close_devices(fs_info->fs_devices);
3972 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3974 static void btrfs_end_super_write(struct bio *bio)
3976 struct btrfs_device *device = bio->bi_private;
3977 struct bio_vec *bvec;
3978 struct bvec_iter_all iter_all;
3981 bio_for_each_segment_all(bvec, bio, iter_all) {
3982 page = bvec->bv_page;
3984 if (bio->bi_status) {
3985 btrfs_warn_rl_in_rcu(device->fs_info,
3986 "lost page write due to IO error on %s (%d)",
3987 rcu_str_deref(device->name),
3988 blk_status_to_errno(bio->bi_status));
3989 ClearPageUptodate(page);
3991 btrfs_dev_stat_inc_and_print(device,
3992 BTRFS_DEV_STAT_WRITE_ERRS);
3994 SetPageUptodate(page);
4004 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
4007 struct btrfs_super_block *super;
4009 u64 bytenr, bytenr_orig;
4010 struct address_space *mapping = bdev->bd_inode->i_mapping;
4013 bytenr_orig = btrfs_sb_offset(copy_num);
4014 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
4016 return ERR_PTR(-EINVAL);
4018 return ERR_PTR(ret);
4020 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
4021 return ERR_PTR(-EINVAL);
4023 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
4025 return ERR_CAST(page);
4027 super = page_address(page);
4028 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
4029 btrfs_release_disk_super(super);
4030 return ERR_PTR(-ENODATA);
4033 if (btrfs_super_bytenr(super) != bytenr_orig) {
4034 btrfs_release_disk_super(super);
4035 return ERR_PTR(-EINVAL);
4042 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
4044 struct btrfs_super_block *super, *latest = NULL;
4048 /* we would like to check all the supers, but that would make
4049 * a btrfs mount succeed after a mkfs from a different FS.
4050 * So, we need to add a special mount option to scan for
4051 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
4053 for (i = 0; i < 1; i++) {
4054 super = btrfs_read_dev_one_super(bdev, i);
4058 if (!latest || btrfs_super_generation(super) > transid) {
4060 btrfs_release_disk_super(super);
4063 transid = btrfs_super_generation(super);
4071 * Write superblock @sb to the @device. Do not wait for completion, all the
4072 * pages we use for writing are locked.
4074 * Write @max_mirrors copies of the superblock, where 0 means default that fit
4075 * the expected device size at commit time. Note that max_mirrors must be
4076 * same for write and wait phases.
4078 * Return number of errors when page is not found or submission fails.
4080 static int write_dev_supers(struct btrfs_device *device,
4081 struct btrfs_super_block *sb, int max_mirrors)
4083 struct btrfs_fs_info *fs_info = device->fs_info;
4084 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
4085 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4089 u64 bytenr, bytenr_orig;
4091 if (max_mirrors == 0)
4092 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4094 shash->tfm = fs_info->csum_shash;
4096 for (i = 0; i < max_mirrors; i++) {
4099 struct btrfs_super_block *disk_super;
4101 bytenr_orig = btrfs_sb_offset(i);
4102 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4103 if (ret == -ENOENT) {
4105 } else if (ret < 0) {
4106 btrfs_err(device->fs_info,
4107 "couldn't get super block location for mirror %d",
4112 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4113 device->commit_total_bytes)
4116 btrfs_set_super_bytenr(sb, bytenr_orig);
4118 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4119 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4122 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4125 btrfs_err(device->fs_info,
4126 "couldn't get super block page for bytenr %llu",
4132 /* Bump the refcount for wait_dev_supers() */
4135 disk_super = page_address(page);
4136 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4139 * Directly use bios here instead of relying on the page cache
4140 * to do I/O, so we don't lose the ability to do integrity
4143 bio = bio_alloc(device->bdev, 1,
4144 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4146 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4147 bio->bi_private = device;
4148 bio->bi_end_io = btrfs_end_super_write;
4149 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4150 offset_in_page(bytenr));
4153 * We FUA only the first super block. The others we allow to
4154 * go down lazy and there's a short window where the on-disk
4155 * copies might still contain the older version.
4157 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4158 bio->bi_opf |= REQ_FUA;
4160 btrfsic_submit_bio(bio);
4162 if (btrfs_advance_sb_log(device, i))
4165 return errors < i ? 0 : -1;
4169 * Wait for write completion of superblocks done by write_dev_supers,
4170 * @max_mirrors same for write and wait phases.
4172 * Return number of errors when page is not found or not marked up to
4175 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4179 bool primary_failed = false;
4183 if (max_mirrors == 0)
4184 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4186 for (i = 0; i < max_mirrors; i++) {
4189 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4190 if (ret == -ENOENT) {
4192 } else if (ret < 0) {
4195 primary_failed = true;
4198 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4199 device->commit_total_bytes)
4202 page = find_get_page(device->bdev->bd_inode->i_mapping,
4203 bytenr >> PAGE_SHIFT);
4207 primary_failed = true;
4210 /* Page is submitted locked and unlocked once the IO completes */
4211 wait_on_page_locked(page);
4212 if (PageError(page)) {
4215 primary_failed = true;
4218 /* Drop our reference */
4221 /* Drop the reference from the writing run */
4225 /* log error, force error return */
4226 if (primary_failed) {
4227 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4232 return errors < i ? 0 : -1;
4236 * endio for the write_dev_flush, this will wake anyone waiting
4237 * for the barrier when it is done
4239 static void btrfs_end_empty_barrier(struct bio *bio)
4242 complete(bio->bi_private);
4246 * Submit a flush request to the device if it supports it. Error handling is
4247 * done in the waiting counterpart.
4249 static void write_dev_flush(struct btrfs_device *device)
4251 struct bio *bio = &device->flush_bio;
4253 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4255 * When a disk has write caching disabled, we skip submission of a bio
4256 * with flush and sync requests before writing the superblock, since
4257 * it's not needed. However when the integrity checker is enabled, this
4258 * results in reports that there are metadata blocks referred by a
4259 * superblock that were not properly flushed. So don't skip the bio
4260 * submission only when the integrity checker is enabled for the sake
4261 * of simplicity, since this is a debug tool and not meant for use in
4264 if (!bdev_write_cache(device->bdev))
4268 bio_init(bio, device->bdev, NULL, 0,
4269 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4270 bio->bi_end_io = btrfs_end_empty_barrier;
4271 init_completion(&device->flush_wait);
4272 bio->bi_private = &device->flush_wait;
4274 btrfsic_submit_bio(bio);
4275 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4279 * If the flush bio has been submitted by write_dev_flush, wait for it.
4281 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4283 struct bio *bio = &device->flush_bio;
4285 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4288 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4289 wait_for_completion_io(&device->flush_wait);
4291 return bio->bi_status;
4294 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4296 if (!btrfs_check_rw_degradable(fs_info, NULL))
4302 * send an empty flush down to each device in parallel,
4303 * then wait for them
4305 static int barrier_all_devices(struct btrfs_fs_info *info)
4307 struct list_head *head;
4308 struct btrfs_device *dev;
4309 int errors_wait = 0;
4312 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4313 /* send down all the barriers */
4314 head = &info->fs_devices->devices;
4315 list_for_each_entry(dev, head, dev_list) {
4316 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4320 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4321 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4324 write_dev_flush(dev);
4325 dev->last_flush_error = BLK_STS_OK;
4328 /* wait for all the barriers */
4329 list_for_each_entry(dev, head, dev_list) {
4330 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4336 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4337 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4340 ret = wait_dev_flush(dev);
4342 dev->last_flush_error = ret;
4343 btrfs_dev_stat_inc_and_print(dev,
4344 BTRFS_DEV_STAT_FLUSH_ERRS);
4351 * At some point we need the status of all disks
4352 * to arrive at the volume status. So error checking
4353 * is being pushed to a separate loop.
4355 return check_barrier_error(info);
4360 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4363 int min_tolerated = INT_MAX;
4365 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4366 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4367 min_tolerated = min_t(int, min_tolerated,
4368 btrfs_raid_array[BTRFS_RAID_SINGLE].
4369 tolerated_failures);
4371 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4372 if (raid_type == BTRFS_RAID_SINGLE)
4374 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4376 min_tolerated = min_t(int, min_tolerated,
4377 btrfs_raid_array[raid_type].
4378 tolerated_failures);
4381 if (min_tolerated == INT_MAX) {
4382 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4386 return min_tolerated;
4389 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4391 struct list_head *head;
4392 struct btrfs_device *dev;
4393 struct btrfs_super_block *sb;
4394 struct btrfs_dev_item *dev_item;
4398 int total_errors = 0;
4401 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4404 * max_mirrors == 0 indicates we're from commit_transaction,
4405 * not from fsync where the tree roots in fs_info have not
4406 * been consistent on disk.
4408 if (max_mirrors == 0)
4409 backup_super_roots(fs_info);
4411 sb = fs_info->super_for_commit;
4412 dev_item = &sb->dev_item;
4414 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4415 head = &fs_info->fs_devices->devices;
4416 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4419 ret = barrier_all_devices(fs_info);
4422 &fs_info->fs_devices->device_list_mutex);
4423 btrfs_handle_fs_error(fs_info, ret,
4424 "errors while submitting device barriers.");
4429 list_for_each_entry(dev, head, dev_list) {
4434 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4435 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4438 btrfs_set_stack_device_generation(dev_item, 0);
4439 btrfs_set_stack_device_type(dev_item, dev->type);
4440 btrfs_set_stack_device_id(dev_item, dev->devid);
4441 btrfs_set_stack_device_total_bytes(dev_item,
4442 dev->commit_total_bytes);
4443 btrfs_set_stack_device_bytes_used(dev_item,
4444 dev->commit_bytes_used);
4445 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4446 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4447 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4448 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4449 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4452 flags = btrfs_super_flags(sb);
4453 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4455 ret = btrfs_validate_write_super(fs_info, sb);
4457 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4458 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4459 "unexpected superblock corruption detected");
4463 ret = write_dev_supers(dev, sb, max_mirrors);
4467 if (total_errors > max_errors) {
4468 btrfs_err(fs_info, "%d errors while writing supers",
4470 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4472 /* FUA is masked off if unsupported and can't be the reason */
4473 btrfs_handle_fs_error(fs_info, -EIO,
4474 "%d errors while writing supers",
4480 list_for_each_entry(dev, head, dev_list) {
4483 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4484 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4487 ret = wait_dev_supers(dev, max_mirrors);
4491 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4492 if (total_errors > max_errors) {
4493 btrfs_handle_fs_error(fs_info, -EIO,
4494 "%d errors while writing supers",
4501 /* Drop a fs root from the radix tree and free it. */
4502 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4503 struct btrfs_root *root)
4505 bool drop_ref = false;
4507 spin_lock(&fs_info->fs_roots_radix_lock);
4508 radix_tree_delete(&fs_info->fs_roots_radix,
4509 (unsigned long)root->root_key.objectid);
4510 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4512 spin_unlock(&fs_info->fs_roots_radix_lock);
4514 if (BTRFS_FS_ERROR(fs_info)) {
4515 ASSERT(root->log_root == NULL);
4516 if (root->reloc_root) {
4517 btrfs_put_root(root->reloc_root);
4518 root->reloc_root = NULL;
4523 btrfs_put_root(root);
4526 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4528 u64 root_objectid = 0;
4529 struct btrfs_root *gang[8];
4532 unsigned int ret = 0;
4535 spin_lock(&fs_info->fs_roots_radix_lock);
4536 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4537 (void **)gang, root_objectid,
4540 spin_unlock(&fs_info->fs_roots_radix_lock);
4543 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4545 for (i = 0; i < ret; i++) {
4546 /* Avoid to grab roots in dead_roots */
4547 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4551 /* grab all the search result for later use */
4552 gang[i] = btrfs_grab_root(gang[i]);
4554 spin_unlock(&fs_info->fs_roots_radix_lock);
4556 for (i = 0; i < ret; i++) {
4559 root_objectid = gang[i]->root_key.objectid;
4560 err = btrfs_orphan_cleanup(gang[i]);
4563 btrfs_put_root(gang[i]);
4568 /* release the uncleaned roots due to error */
4569 for (; i < ret; i++) {
4571 btrfs_put_root(gang[i]);
4576 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4578 struct btrfs_root *root = fs_info->tree_root;
4579 struct btrfs_trans_handle *trans;
4581 mutex_lock(&fs_info->cleaner_mutex);
4582 btrfs_run_delayed_iputs(fs_info);
4583 mutex_unlock(&fs_info->cleaner_mutex);
4584 wake_up_process(fs_info->cleaner_kthread);
4586 /* wait until ongoing cleanup work done */
4587 down_write(&fs_info->cleanup_work_sem);
4588 up_write(&fs_info->cleanup_work_sem);
4590 trans = btrfs_join_transaction(root);
4592 return PTR_ERR(trans);
4593 return btrfs_commit_transaction(trans);
4596 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4598 struct btrfs_transaction *trans;
4599 struct btrfs_transaction *tmp;
4602 if (list_empty(&fs_info->trans_list))
4606 * This function is only called at the very end of close_ctree(),
4607 * thus no other running transaction, no need to take trans_lock.
4609 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4610 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4611 struct extent_state *cached = NULL;
4612 u64 dirty_bytes = 0;
4618 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4619 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4620 dirty_bytes += found_end + 1 - found_start;
4621 cur = found_end + 1;
4624 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4625 trans->transid, dirty_bytes);
4626 btrfs_cleanup_one_transaction(trans, fs_info);
4628 if (trans == fs_info->running_transaction)
4629 fs_info->running_transaction = NULL;
4630 list_del_init(&trans->list);
4632 btrfs_put_transaction(trans);
4633 trace_btrfs_transaction_commit(fs_info);
4638 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4642 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4644 * We don't want the cleaner to start new transactions, add more delayed
4645 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4646 * because that frees the task_struct, and the transaction kthread might
4647 * still try to wake up the cleaner.
4649 kthread_park(fs_info->cleaner_kthread);
4652 * If we had UNFINISHED_DROPS we could still be processing them, so
4653 * clear that bit and wake up relocation so it can stop.
4655 btrfs_wake_unfinished_drop(fs_info);
4657 /* wait for the qgroup rescan worker to stop */
4658 btrfs_qgroup_wait_for_completion(fs_info, false);
4660 /* wait for the uuid_scan task to finish */
4661 down(&fs_info->uuid_tree_rescan_sem);
4662 /* avoid complains from lockdep et al., set sem back to initial state */
4663 up(&fs_info->uuid_tree_rescan_sem);
4665 /* pause restriper - we want to resume on mount */
4666 btrfs_pause_balance(fs_info);
4668 btrfs_dev_replace_suspend_for_unmount(fs_info);
4670 btrfs_scrub_cancel(fs_info);
4672 /* wait for any defraggers to finish */
4673 wait_event(fs_info->transaction_wait,
4674 (atomic_read(&fs_info->defrag_running) == 0));
4676 /* clear out the rbtree of defraggable inodes */
4677 btrfs_cleanup_defrag_inodes(fs_info);
4679 cancel_work_sync(&fs_info->async_reclaim_work);
4680 cancel_work_sync(&fs_info->async_data_reclaim_work);
4681 cancel_work_sync(&fs_info->preempt_reclaim_work);
4683 cancel_work_sync(&fs_info->reclaim_bgs_work);
4685 /* Cancel or finish ongoing discard work */
4686 btrfs_discard_cleanup(fs_info);
4688 if (!sb_rdonly(fs_info->sb)) {
4690 * The cleaner kthread is stopped, so do one final pass over
4691 * unused block groups.
4693 btrfs_delete_unused_bgs(fs_info);
4696 * There might be existing delayed inode workers still running
4697 * and holding an empty delayed inode item. We must wait for
4698 * them to complete first because they can create a transaction.
4699 * This happens when someone calls btrfs_balance_delayed_items()
4700 * and then a transaction commit runs the same delayed nodes
4701 * before any delayed worker has done something with the nodes.
4702 * We must wait for any worker here and not at transaction
4703 * commit time since that could cause a deadlock.
4704 * This is a very rare case.
4706 btrfs_flush_workqueue(fs_info->delayed_workers);
4708 ret = btrfs_commit_super(fs_info);
4710 btrfs_err(fs_info, "commit super ret %d", ret);
4713 if (BTRFS_FS_ERROR(fs_info))
4714 btrfs_error_commit_super(fs_info);
4716 kthread_stop(fs_info->transaction_kthread);
4717 kthread_stop(fs_info->cleaner_kthread);
4719 ASSERT(list_empty(&fs_info->delayed_iputs));
4720 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4722 if (btrfs_check_quota_leak(fs_info)) {
4723 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4724 btrfs_err(fs_info, "qgroup reserved space leaked");
4727 btrfs_free_qgroup_config(fs_info);
4728 ASSERT(list_empty(&fs_info->delalloc_roots));
4730 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4731 btrfs_info(fs_info, "at unmount delalloc count %lld",
4732 percpu_counter_sum(&fs_info->delalloc_bytes));
4735 if (percpu_counter_sum(&fs_info->ordered_bytes))
4736 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4737 percpu_counter_sum(&fs_info->ordered_bytes));
4739 btrfs_sysfs_remove_mounted(fs_info);
4740 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4742 btrfs_put_block_group_cache(fs_info);
4745 * we must make sure there is not any read request to
4746 * submit after we stopping all workers.
4748 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4749 btrfs_stop_all_workers(fs_info);
4751 /* We shouldn't have any transaction open at this point */
4752 warn_about_uncommitted_trans(fs_info);
4754 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4755 free_root_pointers(fs_info, true);
4756 btrfs_free_fs_roots(fs_info);
4759 * We must free the block groups after dropping the fs_roots as we could
4760 * have had an IO error and have left over tree log blocks that aren't
4761 * cleaned up until the fs roots are freed. This makes the block group
4762 * accounting appear to be wrong because there's pending reserved bytes,
4763 * so make sure we do the block group cleanup afterwards.
4765 btrfs_free_block_groups(fs_info);
4767 iput(fs_info->btree_inode);
4769 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4770 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4771 btrfsic_unmount(fs_info->fs_devices);
4774 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4775 btrfs_close_devices(fs_info->fs_devices);
4778 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4782 struct inode *btree_inode = buf->pages[0]->mapping->host;
4784 ret = extent_buffer_uptodate(buf);
4788 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4789 parent_transid, atomic);
4795 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4797 struct btrfs_fs_info *fs_info = buf->fs_info;
4798 u64 transid = btrfs_header_generation(buf);
4801 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4803 * This is a fast path so only do this check if we have sanity tests
4804 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4805 * outside of the sanity tests.
4807 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4810 btrfs_assert_tree_write_locked(buf);
4811 if (transid != fs_info->generation)
4812 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4813 buf->start, transid, fs_info->generation);
4814 was_dirty = set_extent_buffer_dirty(buf);
4816 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4818 fs_info->dirty_metadata_batch);
4819 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4821 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4822 * but item data not updated.
4823 * So here we should only check item pointers, not item data.
4825 if (btrfs_header_level(buf) == 0 &&
4826 btrfs_check_leaf_relaxed(buf)) {
4827 btrfs_print_leaf(buf);
4833 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4837 * looks as though older kernels can get into trouble with
4838 * this code, they end up stuck in balance_dirty_pages forever
4842 if (current->flags & PF_MEMALLOC)
4846 btrfs_balance_delayed_items(fs_info);
4848 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4849 BTRFS_DIRTY_METADATA_THRESH,
4850 fs_info->dirty_metadata_batch);
4852 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4856 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4858 __btrfs_btree_balance_dirty(fs_info, 1);
4861 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4863 __btrfs_btree_balance_dirty(fs_info, 0);
4866 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4867 struct btrfs_key *first_key)
4869 return btree_read_extent_buffer_pages(buf, parent_transid,
4873 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4875 /* cleanup FS via transaction */
4876 btrfs_cleanup_transaction(fs_info);
4878 mutex_lock(&fs_info->cleaner_mutex);
4879 btrfs_run_delayed_iputs(fs_info);
4880 mutex_unlock(&fs_info->cleaner_mutex);
4882 down_write(&fs_info->cleanup_work_sem);
4883 up_write(&fs_info->cleanup_work_sem);
4886 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4888 struct btrfs_root *gang[8];
4889 u64 root_objectid = 0;
4892 spin_lock(&fs_info->fs_roots_radix_lock);
4893 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4894 (void **)gang, root_objectid,
4895 ARRAY_SIZE(gang))) != 0) {
4898 for (i = 0; i < ret; i++)
4899 gang[i] = btrfs_grab_root(gang[i]);
4900 spin_unlock(&fs_info->fs_roots_radix_lock);
4902 for (i = 0; i < ret; i++) {
4905 root_objectid = gang[i]->root_key.objectid;
4906 btrfs_free_log(NULL, gang[i]);
4907 btrfs_put_root(gang[i]);
4910 spin_lock(&fs_info->fs_roots_radix_lock);
4912 spin_unlock(&fs_info->fs_roots_radix_lock);
4913 btrfs_free_log_root_tree(NULL, fs_info);
4916 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4918 struct btrfs_ordered_extent *ordered;
4920 spin_lock(&root->ordered_extent_lock);
4922 * This will just short circuit the ordered completion stuff which will
4923 * make sure the ordered extent gets properly cleaned up.
4925 list_for_each_entry(ordered, &root->ordered_extents,
4927 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4928 spin_unlock(&root->ordered_extent_lock);
4931 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4933 struct btrfs_root *root;
4934 struct list_head splice;
4936 INIT_LIST_HEAD(&splice);
4938 spin_lock(&fs_info->ordered_root_lock);
4939 list_splice_init(&fs_info->ordered_roots, &splice);
4940 while (!list_empty(&splice)) {
4941 root = list_first_entry(&splice, struct btrfs_root,
4943 list_move_tail(&root->ordered_root,
4944 &fs_info->ordered_roots);
4946 spin_unlock(&fs_info->ordered_root_lock);
4947 btrfs_destroy_ordered_extents(root);
4950 spin_lock(&fs_info->ordered_root_lock);
4952 spin_unlock(&fs_info->ordered_root_lock);
4955 * We need this here because if we've been flipped read-only we won't
4956 * get sync() from the umount, so we need to make sure any ordered
4957 * extents that haven't had their dirty pages IO start writeout yet
4958 * actually get run and error out properly.
4960 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4963 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4964 struct btrfs_fs_info *fs_info)
4966 struct rb_node *node;
4967 struct btrfs_delayed_ref_root *delayed_refs;
4968 struct btrfs_delayed_ref_node *ref;
4971 delayed_refs = &trans->delayed_refs;
4973 spin_lock(&delayed_refs->lock);
4974 if (atomic_read(&delayed_refs->num_entries) == 0) {
4975 spin_unlock(&delayed_refs->lock);
4976 btrfs_debug(fs_info, "delayed_refs has NO entry");
4980 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4981 struct btrfs_delayed_ref_head *head;
4983 bool pin_bytes = false;
4985 head = rb_entry(node, struct btrfs_delayed_ref_head,
4987 if (btrfs_delayed_ref_lock(delayed_refs, head))
4990 spin_lock(&head->lock);
4991 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4992 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4995 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4996 RB_CLEAR_NODE(&ref->ref_node);
4997 if (!list_empty(&ref->add_list))
4998 list_del(&ref->add_list);
4999 atomic_dec(&delayed_refs->num_entries);
5000 btrfs_put_delayed_ref(ref);
5002 if (head->must_insert_reserved)
5004 btrfs_free_delayed_extent_op(head->extent_op);
5005 btrfs_delete_ref_head(delayed_refs, head);
5006 spin_unlock(&head->lock);
5007 spin_unlock(&delayed_refs->lock);
5008 mutex_unlock(&head->mutex);
5011 struct btrfs_block_group *cache;
5013 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
5016 spin_lock(&cache->space_info->lock);
5017 spin_lock(&cache->lock);
5018 cache->pinned += head->num_bytes;
5019 btrfs_space_info_update_bytes_pinned(fs_info,
5020 cache->space_info, head->num_bytes);
5021 cache->reserved -= head->num_bytes;
5022 cache->space_info->bytes_reserved -= head->num_bytes;
5023 spin_unlock(&cache->lock);
5024 spin_unlock(&cache->space_info->lock);
5026 btrfs_put_block_group(cache);
5028 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
5029 head->bytenr + head->num_bytes - 1);
5031 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
5032 btrfs_put_delayed_ref_head(head);
5034 spin_lock(&delayed_refs->lock);
5036 btrfs_qgroup_destroy_extent_records(trans);
5038 spin_unlock(&delayed_refs->lock);
5043 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
5045 struct btrfs_inode *btrfs_inode;
5046 struct list_head splice;
5048 INIT_LIST_HEAD(&splice);
5050 spin_lock(&root->delalloc_lock);
5051 list_splice_init(&root->delalloc_inodes, &splice);
5053 while (!list_empty(&splice)) {
5054 struct inode *inode = NULL;
5055 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5057 __btrfs_del_delalloc_inode(root, btrfs_inode);
5058 spin_unlock(&root->delalloc_lock);
5061 * Make sure we get a live inode and that it'll not disappear
5064 inode = igrab(&btrfs_inode->vfs_inode);
5066 invalidate_inode_pages2(inode->i_mapping);
5069 spin_lock(&root->delalloc_lock);
5071 spin_unlock(&root->delalloc_lock);
5074 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5076 struct btrfs_root *root;
5077 struct list_head splice;
5079 INIT_LIST_HEAD(&splice);
5081 spin_lock(&fs_info->delalloc_root_lock);
5082 list_splice_init(&fs_info->delalloc_roots, &splice);
5083 while (!list_empty(&splice)) {
5084 root = list_first_entry(&splice, struct btrfs_root,
5086 root = btrfs_grab_root(root);
5088 spin_unlock(&fs_info->delalloc_root_lock);
5090 btrfs_destroy_delalloc_inodes(root);
5091 btrfs_put_root(root);
5093 spin_lock(&fs_info->delalloc_root_lock);
5095 spin_unlock(&fs_info->delalloc_root_lock);
5098 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5099 struct extent_io_tree *dirty_pages,
5103 struct extent_buffer *eb;
5108 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5113 clear_extent_bits(dirty_pages, start, end, mark);
5114 while (start <= end) {
5115 eb = find_extent_buffer(fs_info, start);
5116 start += fs_info->nodesize;
5119 wait_on_extent_buffer_writeback(eb);
5121 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5123 clear_extent_buffer_dirty(eb);
5124 free_extent_buffer_stale(eb);
5131 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5132 struct extent_io_tree *unpin)
5139 struct extent_state *cached_state = NULL;
5142 * The btrfs_finish_extent_commit() may get the same range as
5143 * ours between find_first_extent_bit and clear_extent_dirty.
5144 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5145 * the same extent range.
5147 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5148 ret = find_first_extent_bit(unpin, 0, &start, &end,
5149 EXTENT_DIRTY, &cached_state);
5151 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5155 clear_extent_dirty(unpin, start, end, &cached_state);
5156 free_extent_state(cached_state);
5157 btrfs_error_unpin_extent_range(fs_info, start, end);
5158 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5165 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5167 struct inode *inode;
5169 inode = cache->io_ctl.inode;
5171 invalidate_inode_pages2(inode->i_mapping);
5172 BTRFS_I(inode)->generation = 0;
5173 cache->io_ctl.inode = NULL;
5176 ASSERT(cache->io_ctl.pages == NULL);
5177 btrfs_put_block_group(cache);
5180 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5181 struct btrfs_fs_info *fs_info)
5183 struct btrfs_block_group *cache;
5185 spin_lock(&cur_trans->dirty_bgs_lock);
5186 while (!list_empty(&cur_trans->dirty_bgs)) {
5187 cache = list_first_entry(&cur_trans->dirty_bgs,
5188 struct btrfs_block_group,
5191 if (!list_empty(&cache->io_list)) {
5192 spin_unlock(&cur_trans->dirty_bgs_lock);
5193 list_del_init(&cache->io_list);
5194 btrfs_cleanup_bg_io(cache);
5195 spin_lock(&cur_trans->dirty_bgs_lock);
5198 list_del_init(&cache->dirty_list);
5199 spin_lock(&cache->lock);
5200 cache->disk_cache_state = BTRFS_DC_ERROR;
5201 spin_unlock(&cache->lock);
5203 spin_unlock(&cur_trans->dirty_bgs_lock);
5204 btrfs_put_block_group(cache);
5205 btrfs_delayed_refs_rsv_release(fs_info, 1);
5206 spin_lock(&cur_trans->dirty_bgs_lock);
5208 spin_unlock(&cur_trans->dirty_bgs_lock);
5211 * Refer to the definition of io_bgs member for details why it's safe
5212 * to use it without any locking
5214 while (!list_empty(&cur_trans->io_bgs)) {
5215 cache = list_first_entry(&cur_trans->io_bgs,
5216 struct btrfs_block_group,
5219 list_del_init(&cache->io_list);
5220 spin_lock(&cache->lock);
5221 cache->disk_cache_state = BTRFS_DC_ERROR;
5222 spin_unlock(&cache->lock);
5223 btrfs_cleanup_bg_io(cache);
5227 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5228 struct btrfs_fs_info *fs_info)
5230 struct btrfs_device *dev, *tmp;
5232 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5233 ASSERT(list_empty(&cur_trans->dirty_bgs));
5234 ASSERT(list_empty(&cur_trans->io_bgs));
5236 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5238 list_del_init(&dev->post_commit_list);
5241 btrfs_destroy_delayed_refs(cur_trans, fs_info);
5243 cur_trans->state = TRANS_STATE_COMMIT_START;
5244 wake_up(&fs_info->transaction_blocked_wait);
5246 cur_trans->state = TRANS_STATE_UNBLOCKED;
5247 wake_up(&fs_info->transaction_wait);
5249 btrfs_destroy_delayed_inodes(fs_info);
5251 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5253 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5255 btrfs_free_redirty_list(cur_trans);
5257 cur_trans->state =TRANS_STATE_COMPLETED;
5258 wake_up(&cur_trans->commit_wait);
5261 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5263 struct btrfs_transaction *t;
5265 mutex_lock(&fs_info->transaction_kthread_mutex);
5267 spin_lock(&fs_info->trans_lock);
5268 while (!list_empty(&fs_info->trans_list)) {
5269 t = list_first_entry(&fs_info->trans_list,
5270 struct btrfs_transaction, list);
5271 if (t->state >= TRANS_STATE_COMMIT_START) {
5272 refcount_inc(&t->use_count);
5273 spin_unlock(&fs_info->trans_lock);
5274 btrfs_wait_for_commit(fs_info, t->transid);
5275 btrfs_put_transaction(t);
5276 spin_lock(&fs_info->trans_lock);
5279 if (t == fs_info->running_transaction) {
5280 t->state = TRANS_STATE_COMMIT_DOING;
5281 spin_unlock(&fs_info->trans_lock);
5283 * We wait for 0 num_writers since we don't hold a trans
5284 * handle open currently for this transaction.
5286 wait_event(t->writer_wait,
5287 atomic_read(&t->num_writers) == 0);
5289 spin_unlock(&fs_info->trans_lock);
5291 btrfs_cleanup_one_transaction(t, fs_info);
5293 spin_lock(&fs_info->trans_lock);
5294 if (t == fs_info->running_transaction)
5295 fs_info->running_transaction = NULL;
5296 list_del_init(&t->list);
5297 spin_unlock(&fs_info->trans_lock);
5299 btrfs_put_transaction(t);
5300 trace_btrfs_transaction_commit(fs_info);
5301 spin_lock(&fs_info->trans_lock);
5303 spin_unlock(&fs_info->trans_lock);
5304 btrfs_destroy_all_ordered_extents(fs_info);
5305 btrfs_destroy_delayed_inodes(fs_info);
5306 btrfs_assert_delayed_root_empty(fs_info);
5307 btrfs_destroy_all_delalloc_inodes(fs_info);
5308 btrfs_drop_all_logs(fs_info);
5309 mutex_unlock(&fs_info->transaction_kthread_mutex);
5314 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5316 struct btrfs_path *path;
5318 struct extent_buffer *l;
5319 struct btrfs_key search_key;
5320 struct btrfs_key found_key;
5323 path = btrfs_alloc_path();
5327 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5328 search_key.type = -1;
5329 search_key.offset = (u64)-1;
5330 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5333 BUG_ON(ret == 0); /* Corruption */
5334 if (path->slots[0] > 0) {
5335 slot = path->slots[0] - 1;
5337 btrfs_item_key_to_cpu(l, &found_key, slot);
5338 root->free_objectid = max_t(u64, found_key.objectid + 1,
5339 BTRFS_FIRST_FREE_OBJECTID);
5341 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5345 btrfs_free_path(path);
5349 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5352 mutex_lock(&root->objectid_mutex);
5354 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5355 btrfs_warn(root->fs_info,
5356 "the objectid of root %llu reaches its highest value",
5357 root->root_key.objectid);
5362 *objectid = root->free_objectid++;
5365 mutex_unlock(&root->objectid_mutex);