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);
445 btrfs_print_tree(eb, 0);
447 "block=%llu write time tree block corruption detected",
449 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
452 write_extent_buffer(eb, result, 0, fs_info->csum_size);
457 /* Checksum all dirty extent buffers in one bio_vec */
458 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
459 struct bio_vec *bvec)
461 struct page *page = bvec->bv_page;
462 u64 bvec_start = page_offset(page) + bvec->bv_offset;
466 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
467 cur += fs_info->nodesize) {
468 struct extent_buffer *eb;
471 eb = find_extent_buffer(fs_info, cur);
472 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
475 /* A dirty eb shouldn't disappear from buffer_radix */
479 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
480 free_extent_buffer(eb);
483 if (WARN_ON(!uptodate)) {
484 free_extent_buffer(eb);
488 ret = csum_one_extent_buffer(eb);
489 free_extent_buffer(eb);
497 * Checksum a dirty tree block before IO. This has extra checks to make sure
498 * we only fill in the checksum field in the first page of a multi-page block.
499 * For subpage extent buffers we need bvec to also read the offset in the page.
501 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
503 struct page *page = bvec->bv_page;
504 u64 start = page_offset(page);
506 struct extent_buffer *eb;
508 if (fs_info->sectorsize < PAGE_SIZE)
509 return csum_dirty_subpage_buffers(fs_info, bvec);
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
515 found_start = btrfs_header_bytenr(eb);
517 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
518 WARN_ON(found_start != 0);
523 * Please do not consolidate these warnings into a single if.
524 * It is useful to know what went wrong.
526 if (WARN_ON(found_start != start))
528 if (WARN_ON(!PageUptodate(page)))
531 return csum_one_extent_buffer(eb);
534 static int check_tree_block_fsid(struct extent_buffer *eb)
536 struct btrfs_fs_info *fs_info = eb->fs_info;
537 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
538 u8 fsid[BTRFS_FSID_SIZE];
541 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
544 * Checking the incompat flag is only valid for the current fs. For
545 * seed devices it's forbidden to have their uuid changed so reading
546 * ->fsid in this case is fine
548 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
549 metadata_uuid = fs_devices->metadata_uuid;
551 metadata_uuid = fs_devices->fsid;
553 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
556 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
557 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
563 /* Do basic extent buffer checks at read time */
564 static int validate_extent_buffer(struct extent_buffer *eb)
566 struct btrfs_fs_info *fs_info = eb->fs_info;
568 const u32 csum_size = fs_info->csum_size;
570 u8 result[BTRFS_CSUM_SIZE];
571 const u8 *header_csum;
574 found_start = btrfs_header_bytenr(eb);
575 if (found_start != eb->start) {
576 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
577 eb->start, found_start);
581 if (check_tree_block_fsid(eb)) {
582 btrfs_err_rl(fs_info, "bad fsid on block %llu",
587 found_level = btrfs_header_level(eb);
588 if (found_level >= BTRFS_MAX_LEVEL) {
589 btrfs_err(fs_info, "bad tree block level %d on %llu",
590 (int)btrfs_header_level(eb), eb->start);
595 csum_tree_block(eb, result);
596 header_csum = page_address(eb->pages[0]) +
597 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
599 if (memcmp(result, header_csum, csum_size) != 0) {
600 btrfs_warn_rl(fs_info,
601 "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
603 CSUM_FMT_VALUE(csum_size, header_csum),
604 CSUM_FMT_VALUE(csum_size, result),
605 btrfs_header_level(eb));
611 * If this is a leaf block and it is corrupt, set the corrupt bit so
612 * that we don't try and read the other copies of this block, just
615 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
616 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
620 if (found_level > 0 && btrfs_check_node(eb))
624 set_extent_buffer_uptodate(eb);
627 "block=%llu read time tree block corruption detected",
633 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
636 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
637 struct extent_buffer *eb;
642 * We don't allow bio merge for subpage metadata read, so we should
643 * only get one eb for each endio hook.
645 ASSERT(end == start + fs_info->nodesize - 1);
646 ASSERT(PagePrivate(page));
648 eb = find_extent_buffer(fs_info, start);
650 * When we are reading one tree block, eb must have been inserted into
651 * the radix tree. If not, something is wrong.
655 reads_done = atomic_dec_and_test(&eb->io_pages);
656 /* Subpage read must finish in page read */
659 eb->read_mirror = mirror;
660 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
664 ret = validate_extent_buffer(eb);
668 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
669 btree_readahead_hook(eb, ret);
671 set_extent_buffer_uptodate(eb);
673 free_extent_buffer(eb);
677 * end_bio_extent_readpage decrements io_pages in case of error,
678 * make sure it has something to decrement.
680 atomic_inc(&eb->io_pages);
681 clear_extent_buffer_uptodate(eb);
682 free_extent_buffer(eb);
686 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
687 struct page *page, u64 start, u64 end,
690 struct extent_buffer *eb;
694 ASSERT(page->private);
696 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
697 return validate_subpage_buffer(page, start, end, mirror);
699 eb = (struct extent_buffer *)page->private;
702 * The pending IO might have been the only thing that kept this buffer
703 * in memory. Make sure we have a ref for all this other checks
705 atomic_inc(&eb->refs);
707 reads_done = atomic_dec_and_test(&eb->io_pages);
711 eb->read_mirror = mirror;
712 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
716 ret = validate_extent_buffer(eb);
719 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
720 btree_readahead_hook(eb, ret);
724 * our io error hook is going to dec the io pages
725 * again, we have to make sure it has something
728 atomic_inc(&eb->io_pages);
729 clear_extent_buffer_uptodate(eb);
731 free_extent_buffer(eb);
736 static void end_workqueue_bio(struct bio *bio)
738 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
739 struct btrfs_fs_info *fs_info;
740 struct btrfs_workqueue *wq;
742 fs_info = end_io_wq->info;
743 end_io_wq->status = bio->bi_status;
745 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
746 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
747 wq = fs_info->endio_meta_write_workers;
748 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
749 wq = fs_info->endio_freespace_worker;
750 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
751 wq = fs_info->endio_raid56_workers;
753 wq = fs_info->endio_write_workers;
755 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
756 wq = fs_info->endio_raid56_workers;
757 else if (end_io_wq->metadata)
758 wq = fs_info->endio_meta_workers;
760 wq = fs_info->endio_workers;
763 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
764 btrfs_queue_work(wq, &end_io_wq->work);
767 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
768 enum btrfs_wq_endio_type metadata)
770 struct btrfs_end_io_wq *end_io_wq;
772 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
774 return BLK_STS_RESOURCE;
776 end_io_wq->private = bio->bi_private;
777 end_io_wq->end_io = bio->bi_end_io;
778 end_io_wq->info = info;
779 end_io_wq->status = 0;
780 end_io_wq->bio = bio;
781 end_io_wq->metadata = metadata;
783 bio->bi_private = end_io_wq;
784 bio->bi_end_io = end_workqueue_bio;
788 static void run_one_async_start(struct btrfs_work *work)
790 struct async_submit_bio *async;
793 async = container_of(work, struct async_submit_bio, work);
794 ret = async->submit_bio_start(async->inode, async->bio,
795 async->dio_file_offset);
801 * In order to insert checksums into the metadata in large chunks, we wait
802 * until bio submission time. All the pages in the bio are checksummed and
803 * sums are attached onto the ordered extent record.
805 * At IO completion time the csums attached on the ordered extent record are
806 * inserted into the tree.
808 static void run_one_async_done(struct btrfs_work *work)
810 struct async_submit_bio *async;
814 async = container_of(work, struct async_submit_bio, work);
815 inode = async->inode;
817 /* If an error occurred we just want to clean up the bio and move on */
819 async->bio->bi_status = async->status;
820 bio_endio(async->bio);
825 * All of the bios that pass through here are from async helpers.
826 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
827 * This changes nothing when cgroups aren't in use.
829 async->bio->bi_opf |= REQ_CGROUP_PUNT;
830 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
832 async->bio->bi_status = ret;
833 bio_endio(async->bio);
837 static void run_one_async_free(struct btrfs_work *work)
839 struct async_submit_bio *async;
841 async = container_of(work, struct async_submit_bio, work);
845 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
846 int mirror_num, unsigned long bio_flags,
848 extent_submit_bio_start_t *submit_bio_start)
850 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
851 struct async_submit_bio *async;
853 async = kmalloc(sizeof(*async), GFP_NOFS);
855 return BLK_STS_RESOURCE;
857 async->inode = inode;
859 async->mirror_num = mirror_num;
860 async->submit_bio_start = submit_bio_start;
862 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
865 async->dio_file_offset = dio_file_offset;
869 if (op_is_sync(bio->bi_opf))
870 btrfs_set_work_high_priority(&async->work);
872 btrfs_queue_work(fs_info->workers, &async->work);
876 static blk_status_t btree_csum_one_bio(struct bio *bio)
878 struct bio_vec *bvec;
879 struct btrfs_root *root;
881 struct bvec_iter_all iter_all;
883 ASSERT(!bio_flagged(bio, BIO_CLONED));
884 bio_for_each_segment_all(bvec, bio, iter_all) {
885 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
886 ret = csum_dirty_buffer(root->fs_info, bvec);
891 return errno_to_blk_status(ret);
894 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
898 * when we're called for a write, we're already in the async
899 * submission context. Just jump into btrfs_map_bio
901 return btree_csum_one_bio(bio);
904 static bool should_async_write(struct btrfs_fs_info *fs_info,
905 struct btrfs_inode *bi)
907 if (btrfs_is_zoned(fs_info))
909 if (atomic_read(&bi->sync_writers))
911 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
916 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
917 int mirror_num, unsigned long bio_flags)
919 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
922 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
924 * called for a read, do the setup so that checksum validation
925 * can happen in the async kernel threads
927 ret = btrfs_bio_wq_end_io(fs_info, bio,
928 BTRFS_WQ_ENDIO_METADATA);
931 ret = btrfs_map_bio(fs_info, bio, mirror_num);
932 } else if (!should_async_write(fs_info, BTRFS_I(inode))) {
933 ret = btree_csum_one_bio(bio);
936 ret = btrfs_map_bio(fs_info, bio, mirror_num);
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
942 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
943 0, btree_submit_bio_start);
951 bio->bi_status = ret;
956 #ifdef CONFIG_MIGRATION
957 static int btree_migratepage(struct address_space *mapping,
958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
974 return migrate_page(mapping, newpage, page, mode);
979 static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
982 struct btrfs_fs_info *fs_info;
985 if (wbc->sync_mode == WB_SYNC_NONE) {
987 if (wbc->for_kupdate)
990 fs_info = BTRFS_I(mapping->host)->root->fs_info;
991 /* this is a bit racy, but that's ok */
992 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
993 BTRFS_DIRTY_METADATA_THRESH,
994 fs_info->dirty_metadata_batch);
998 return btree_write_cache_pages(mapping, wbc);
1001 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1003 if (PageWriteback(page) || PageDirty(page))
1006 return try_release_extent_buffer(page);
1009 static void btree_invalidatepage(struct page *page, unsigned int offset,
1010 unsigned int length)
1012 struct extent_io_tree *tree;
1013 tree = &BTRFS_I(page->mapping->host)->io_tree;
1014 extent_invalidatepage(tree, page, offset);
1015 btree_releasepage(page, GFP_NOFS);
1016 if (PagePrivate(page)) {
1017 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1018 "page private not zero on page %llu",
1019 (unsigned long long)page_offset(page));
1020 detach_page_private(page);
1024 static int btree_set_page_dirty(struct page *page)
1027 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1028 struct btrfs_subpage *subpage;
1029 struct extent_buffer *eb;
1031 u64 page_start = page_offset(page);
1033 if (fs_info->sectorsize == PAGE_SIZE) {
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_write_locked(eb);
1040 return __set_page_dirty_nobuffers(page);
1042 ASSERT(PagePrivate(page) && page->private);
1043 subpage = (struct btrfs_subpage *)page->private;
1045 ASSERT(subpage->dirty_bitmap);
1046 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1047 unsigned long flags;
1049 u16 tmp = (1 << cur_bit);
1051 spin_lock_irqsave(&subpage->lock, flags);
1052 if (!(tmp & subpage->dirty_bitmap)) {
1053 spin_unlock_irqrestore(&subpage->lock, flags);
1057 spin_unlock_irqrestore(&subpage->lock, flags);
1058 cur = page_start + cur_bit * fs_info->sectorsize;
1060 eb = find_extent_buffer(fs_info, cur);
1062 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063 ASSERT(atomic_read(&eb->refs));
1064 btrfs_assert_tree_write_locked(eb);
1065 free_extent_buffer(eb);
1067 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1070 return __set_page_dirty_nobuffers(page);
1073 static const struct address_space_operations btree_aops = {
1074 .writepages = btree_writepages,
1075 .releasepage = btree_releasepage,
1076 .invalidatepage = btree_invalidatepage,
1077 #ifdef CONFIG_MIGRATION
1078 .migratepage = btree_migratepage,
1080 .set_page_dirty = btree_set_page_dirty,
1083 struct extent_buffer *btrfs_find_create_tree_block(
1084 struct btrfs_fs_info *fs_info,
1085 u64 bytenr, u64 owner_root,
1088 if (btrfs_is_testing(fs_info))
1089 return alloc_test_extent_buffer(fs_info, bytenr);
1090 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1094 * Read tree block at logical address @bytenr and do variant basic but critical
1097 * @owner_root: the objectid of the root owner for this block.
1098 * @parent_transid: expected transid of this tree block, skip check if 0
1099 * @level: expected level, mandatory check
1100 * @first_key: expected key in slot 0, skip check if NULL
1102 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1103 u64 owner_root, u64 parent_transid,
1104 int level, struct btrfs_key *first_key)
1106 struct extent_buffer *buf = NULL;
1109 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1113 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1116 free_extent_buffer_stale(buf);
1117 return ERR_PTR(ret);
1123 void btrfs_clean_tree_block(struct extent_buffer *buf)
1125 struct btrfs_fs_info *fs_info = buf->fs_info;
1126 if (btrfs_header_generation(buf) ==
1127 fs_info->running_transaction->transid) {
1128 btrfs_assert_tree_write_locked(buf);
1130 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1131 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1133 fs_info->dirty_metadata_batch);
1134 clear_extent_buffer_dirty(buf);
1139 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1142 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1143 root->fs_info = fs_info;
1145 root->commit_root = NULL;
1147 root->orphan_cleanup_state = 0;
1149 root->last_trans = 0;
1150 root->free_objectid = 0;
1151 root->nr_delalloc_inodes = 0;
1152 root->nr_ordered_extents = 0;
1153 root->inode_tree = RB_ROOT;
1154 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1155 root->block_rsv = NULL;
1157 INIT_LIST_HEAD(&root->dirty_list);
1158 INIT_LIST_HEAD(&root->root_list);
1159 INIT_LIST_HEAD(&root->delalloc_inodes);
1160 INIT_LIST_HEAD(&root->delalloc_root);
1161 INIT_LIST_HEAD(&root->ordered_extents);
1162 INIT_LIST_HEAD(&root->ordered_root);
1163 INIT_LIST_HEAD(&root->reloc_dirty_list);
1164 INIT_LIST_HEAD(&root->logged_list[0]);
1165 INIT_LIST_HEAD(&root->logged_list[1]);
1166 spin_lock_init(&root->inode_lock);
1167 spin_lock_init(&root->delalloc_lock);
1168 spin_lock_init(&root->ordered_extent_lock);
1169 spin_lock_init(&root->accounting_lock);
1170 spin_lock_init(&root->log_extents_lock[0]);
1171 spin_lock_init(&root->log_extents_lock[1]);
1172 spin_lock_init(&root->qgroup_meta_rsv_lock);
1173 mutex_init(&root->objectid_mutex);
1174 mutex_init(&root->log_mutex);
1175 mutex_init(&root->ordered_extent_mutex);
1176 mutex_init(&root->delalloc_mutex);
1177 init_waitqueue_head(&root->qgroup_flush_wait);
1178 init_waitqueue_head(&root->log_writer_wait);
1179 init_waitqueue_head(&root->log_commit_wait[0]);
1180 init_waitqueue_head(&root->log_commit_wait[1]);
1181 INIT_LIST_HEAD(&root->log_ctxs[0]);
1182 INIT_LIST_HEAD(&root->log_ctxs[1]);
1183 atomic_set(&root->log_commit[0], 0);
1184 atomic_set(&root->log_commit[1], 0);
1185 atomic_set(&root->log_writers, 0);
1186 atomic_set(&root->log_batch, 0);
1187 refcount_set(&root->refs, 1);
1188 atomic_set(&root->snapshot_force_cow, 0);
1189 atomic_set(&root->nr_swapfiles, 0);
1190 root->log_transid = 0;
1191 root->log_transid_committed = -1;
1192 root->last_log_commit = 0;
1194 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1195 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1196 extent_io_tree_init(fs_info, &root->log_csum_range,
1197 IO_TREE_LOG_CSUM_RANGE, NULL);
1200 memset(&root->root_key, 0, sizeof(root->root_key));
1201 memset(&root->root_item, 0, sizeof(root->root_item));
1202 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1203 root->root_key.objectid = objectid;
1206 spin_lock_init(&root->root_item_lock);
1207 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1208 #ifdef CONFIG_BTRFS_DEBUG
1209 INIT_LIST_HEAD(&root->leak_list);
1210 spin_lock(&fs_info->fs_roots_radix_lock);
1211 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1212 spin_unlock(&fs_info->fs_roots_radix_lock);
1216 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1217 u64 objectid, gfp_t flags)
1219 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1221 __setup_root(root, fs_info, objectid);
1225 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1226 /* Should only be used by the testing infrastructure */
1227 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1229 struct btrfs_root *root;
1232 return ERR_PTR(-EINVAL);
1234 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1236 return ERR_PTR(-ENOMEM);
1238 /* We don't use the stripesize in selftest, set it as sectorsize */
1239 root->alloc_bytenr = 0;
1245 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1248 struct btrfs_fs_info *fs_info = trans->fs_info;
1249 struct extent_buffer *leaf;
1250 struct btrfs_root *tree_root = fs_info->tree_root;
1251 struct btrfs_root *root;
1252 struct btrfs_key key;
1253 unsigned int nofs_flag;
1257 * We're holding a transaction handle, so use a NOFS memory allocation
1258 * context to avoid deadlock if reclaim happens.
1260 nofs_flag = memalloc_nofs_save();
1261 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1262 memalloc_nofs_restore(nofs_flag);
1264 return ERR_PTR(-ENOMEM);
1266 root->root_key.objectid = objectid;
1267 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1268 root->root_key.offset = 0;
1270 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1271 BTRFS_NESTING_NORMAL);
1273 ret = PTR_ERR(leaf);
1279 btrfs_mark_buffer_dirty(leaf);
1281 root->commit_root = btrfs_root_node(root);
1282 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1284 btrfs_set_root_flags(&root->root_item, 0);
1285 btrfs_set_root_limit(&root->root_item, 0);
1286 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1287 btrfs_set_root_generation(&root->root_item, trans->transid);
1288 btrfs_set_root_level(&root->root_item, 0);
1289 btrfs_set_root_refs(&root->root_item, 1);
1290 btrfs_set_root_used(&root->root_item, leaf->len);
1291 btrfs_set_root_last_snapshot(&root->root_item, 0);
1292 btrfs_set_root_dirid(&root->root_item, 0);
1293 if (is_fstree(objectid))
1294 generate_random_guid(root->root_item.uuid);
1296 export_guid(root->root_item.uuid, &guid_null);
1297 btrfs_set_root_drop_level(&root->root_item, 0);
1299 btrfs_tree_unlock(leaf);
1301 key.objectid = objectid;
1302 key.type = BTRFS_ROOT_ITEM_KEY;
1304 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1312 btrfs_tree_unlock(leaf);
1314 btrfs_put_root(root);
1316 return ERR_PTR(ret);
1319 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1320 struct btrfs_fs_info *fs_info)
1322 struct btrfs_root *root;
1324 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1326 return ERR_PTR(-ENOMEM);
1328 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1329 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1330 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1335 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1336 struct btrfs_root *root)
1338 struct extent_buffer *leaf;
1341 * DON'T set SHAREABLE bit for log trees.
1343 * Log trees are not exposed to user space thus can't be snapshotted,
1344 * and they go away before a real commit is actually done.
1346 * They do store pointers to file data extents, and those reference
1347 * counts still get updated (along with back refs to the log tree).
1350 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1351 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1353 return PTR_ERR(leaf);
1357 btrfs_mark_buffer_dirty(root->node);
1358 btrfs_tree_unlock(root->node);
1363 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1364 struct btrfs_fs_info *fs_info)
1366 struct btrfs_root *log_root;
1368 log_root = alloc_log_tree(trans, fs_info);
1369 if (IS_ERR(log_root))
1370 return PTR_ERR(log_root);
1372 if (!btrfs_is_zoned(fs_info)) {
1373 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1376 btrfs_put_root(log_root);
1381 WARN_ON(fs_info->log_root_tree);
1382 fs_info->log_root_tree = log_root;
1386 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root)
1389 struct btrfs_fs_info *fs_info = root->fs_info;
1390 struct btrfs_root *log_root;
1391 struct btrfs_inode_item *inode_item;
1394 log_root = alloc_log_tree(trans, fs_info);
1395 if (IS_ERR(log_root))
1396 return PTR_ERR(log_root);
1398 ret = btrfs_alloc_log_tree_node(trans, log_root);
1400 btrfs_put_root(log_root);
1404 log_root->last_trans = trans->transid;
1405 log_root->root_key.offset = root->root_key.objectid;
1407 inode_item = &log_root->root_item.inode;
1408 btrfs_set_stack_inode_generation(inode_item, 1);
1409 btrfs_set_stack_inode_size(inode_item, 3);
1410 btrfs_set_stack_inode_nlink(inode_item, 1);
1411 btrfs_set_stack_inode_nbytes(inode_item,
1413 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1415 btrfs_set_root_node(&log_root->root_item, log_root->node);
1417 WARN_ON(root->log_root);
1418 root->log_root = log_root;
1419 root->log_transid = 0;
1420 root->log_transid_committed = -1;
1421 root->last_log_commit = 0;
1425 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1426 struct btrfs_path *path,
1427 struct btrfs_key *key)
1429 struct btrfs_root *root;
1430 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1435 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1437 return ERR_PTR(-ENOMEM);
1439 ret = btrfs_find_root(tree_root, key, path,
1440 &root->root_item, &root->root_key);
1447 generation = btrfs_root_generation(&root->root_item);
1448 level = btrfs_root_level(&root->root_item);
1449 root->node = read_tree_block(fs_info,
1450 btrfs_root_bytenr(&root->root_item),
1451 key->objectid, generation, level, NULL);
1452 if (IS_ERR(root->node)) {
1453 ret = PTR_ERR(root->node);
1456 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1460 root->commit_root = btrfs_root_node(root);
1463 btrfs_put_root(root);
1464 return ERR_PTR(ret);
1467 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1468 struct btrfs_key *key)
1470 struct btrfs_root *root;
1471 struct btrfs_path *path;
1473 path = btrfs_alloc_path();
1475 return ERR_PTR(-ENOMEM);
1476 root = read_tree_root_path(tree_root, path, key);
1477 btrfs_free_path(path);
1483 * Initialize subvolume root in-memory structure
1485 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1487 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1490 unsigned int nofs_flag;
1493 * We might be called under a transaction (e.g. indirect backref
1494 * resolution) which could deadlock if it triggers memory reclaim
1496 nofs_flag = memalloc_nofs_save();
1497 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1498 memalloc_nofs_restore(nofs_flag);
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1503 !btrfs_is_data_reloc_root(root)) {
1504 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1505 btrfs_check_and_init_root_item(&root->root_item);
1509 * Don't assign anonymous block device to roots that are not exposed to
1510 * userspace, the id pool is limited to 1M
1512 if (is_fstree(root->root_key.objectid) &&
1513 btrfs_root_refs(&root->root_item) > 0) {
1515 ret = get_anon_bdev(&root->anon_dev);
1519 root->anon_dev = anon_dev;
1523 mutex_lock(&root->objectid_mutex);
1524 ret = btrfs_init_root_free_objectid(root);
1526 mutex_unlock(&root->objectid_mutex);
1530 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1532 mutex_unlock(&root->objectid_mutex);
1536 /* The caller is responsible to call btrfs_free_fs_root */
1540 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1543 struct btrfs_root *root;
1545 spin_lock(&fs_info->fs_roots_radix_lock);
1546 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1547 (unsigned long)root_id);
1549 root = btrfs_grab_root(root);
1550 spin_unlock(&fs_info->fs_roots_radix_lock);
1554 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1557 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1558 return btrfs_grab_root(fs_info->tree_root);
1559 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1560 return btrfs_grab_root(fs_info->extent_root);
1561 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1562 return btrfs_grab_root(fs_info->chunk_root);
1563 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1564 return btrfs_grab_root(fs_info->dev_root);
1565 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1566 return btrfs_grab_root(fs_info->csum_root);
1567 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1568 return btrfs_grab_root(fs_info->quota_root) ?
1569 fs_info->quota_root : ERR_PTR(-ENOENT);
1570 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1571 return btrfs_grab_root(fs_info->uuid_root) ?
1572 fs_info->uuid_root : ERR_PTR(-ENOENT);
1573 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1574 return btrfs_grab_root(fs_info->free_space_root) ?
1575 fs_info->free_space_root : ERR_PTR(-ENOENT);
1579 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1580 struct btrfs_root *root)
1584 ret = radix_tree_preload(GFP_NOFS);
1588 spin_lock(&fs_info->fs_roots_radix_lock);
1589 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1590 (unsigned long)root->root_key.objectid,
1593 btrfs_grab_root(root);
1594 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1596 spin_unlock(&fs_info->fs_roots_radix_lock);
1597 radix_tree_preload_end();
1602 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1604 #ifdef CONFIG_BTRFS_DEBUG
1605 struct btrfs_root *root;
1607 while (!list_empty(&fs_info->allocated_roots)) {
1608 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1610 root = list_first_entry(&fs_info->allocated_roots,
1611 struct btrfs_root, leak_list);
1612 btrfs_err(fs_info, "leaked root %s refcount %d",
1613 btrfs_root_name(&root->root_key, buf),
1614 refcount_read(&root->refs));
1615 while (refcount_read(&root->refs) > 1)
1616 btrfs_put_root(root);
1617 btrfs_put_root(root);
1622 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1624 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1625 percpu_counter_destroy(&fs_info->delalloc_bytes);
1626 percpu_counter_destroy(&fs_info->ordered_bytes);
1627 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1628 btrfs_free_csum_hash(fs_info);
1629 btrfs_free_stripe_hash_table(fs_info);
1630 btrfs_free_ref_cache(fs_info);
1631 kfree(fs_info->balance_ctl);
1632 kfree(fs_info->delayed_root);
1633 btrfs_put_root(fs_info->extent_root);
1634 btrfs_put_root(fs_info->tree_root);
1635 btrfs_put_root(fs_info->chunk_root);
1636 btrfs_put_root(fs_info->dev_root);
1637 btrfs_put_root(fs_info->csum_root);
1638 btrfs_put_root(fs_info->quota_root);
1639 btrfs_put_root(fs_info->uuid_root);
1640 btrfs_put_root(fs_info->free_space_root);
1641 btrfs_put_root(fs_info->fs_root);
1642 btrfs_put_root(fs_info->data_reloc_root);
1643 btrfs_check_leaked_roots(fs_info);
1644 btrfs_extent_buffer_leak_debug_check(fs_info);
1645 kfree(fs_info->super_copy);
1646 kfree(fs_info->super_for_commit);
1647 kfree(fs_info->subpage_info);
1653 * Get an in-memory reference of a root structure.
1655 * For essential trees like root/extent tree, we grab it from fs_info directly.
1656 * For subvolume trees, we check the cached filesystem roots first. If not
1657 * found, then read it from disk and add it to cached fs roots.
1659 * Caller should release the root by calling btrfs_put_root() after the usage.
1661 * NOTE: Reloc and log trees can't be read by this function as they share the
1662 * same root objectid.
1664 * @objectid: root id
1665 * @anon_dev: preallocated anonymous block device number for new roots,
1666 * pass 0 for new allocation.
1667 * @check_ref: whether to check root item references, If true, return -ENOENT
1670 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1671 u64 objectid, dev_t anon_dev,
1674 struct btrfs_root *root;
1675 struct btrfs_path *path;
1676 struct btrfs_key key;
1679 root = btrfs_get_global_root(fs_info, objectid);
1683 root = btrfs_lookup_fs_root(fs_info, objectid);
1685 /* Shouldn't get preallocated anon_dev for cached roots */
1687 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1688 btrfs_put_root(root);
1689 return ERR_PTR(-ENOENT);
1694 key.objectid = objectid;
1695 key.type = BTRFS_ROOT_ITEM_KEY;
1696 key.offset = (u64)-1;
1697 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1701 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1706 ret = btrfs_init_fs_root(root, anon_dev);
1710 path = btrfs_alloc_path();
1715 key.objectid = BTRFS_ORPHAN_OBJECTID;
1716 key.type = BTRFS_ORPHAN_ITEM_KEY;
1717 key.offset = objectid;
1719 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1720 btrfs_free_path(path);
1724 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1726 ret = btrfs_insert_fs_root(fs_info, root);
1728 btrfs_put_root(root);
1736 * If our caller provided us an anonymous device, then it's his
1737 * responsability to free it in case we fail. So we have to set our
1738 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1739 * and once again by our caller.
1743 btrfs_put_root(root);
1744 return ERR_PTR(ret);
1748 * Get in-memory reference of a root structure
1750 * @objectid: tree objectid
1751 * @check_ref: if set, verify that the tree exists and the item has at least
1754 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1755 u64 objectid, bool check_ref)
1757 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1761 * Get in-memory reference of a root structure, created as new, optionally pass
1762 * the anonymous block device id
1764 * @objectid: tree objectid
1765 * @anon_dev: if zero, allocate a new anonymous block device or use the
1768 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1769 u64 objectid, dev_t anon_dev)
1771 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1775 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1776 * @fs_info: the fs_info
1777 * @objectid: the objectid we need to lookup
1779 * This is exclusively used for backref walking, and exists specifically because
1780 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1781 * creation time, which means we may have to read the tree_root in order to look
1782 * up a fs root that is not in memory. If the root is not in memory we will
1783 * read the tree root commit root and look up the fs root from there. This is a
1784 * temporary root, it will not be inserted into the radix tree as it doesn't
1785 * have the most uptodate information, it'll simply be discarded once the
1786 * backref code is finished using the root.
1788 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1789 struct btrfs_path *path,
1792 struct btrfs_root *root;
1793 struct btrfs_key key;
1795 ASSERT(path->search_commit_root && path->skip_locking);
1798 * This can return -ENOENT if we ask for a root that doesn't exist, but
1799 * since this is called via the backref walking code we won't be looking
1800 * up a root that doesn't exist, unless there's corruption. So if root
1801 * != NULL just return it.
1803 root = btrfs_get_global_root(fs_info, objectid);
1807 root = btrfs_lookup_fs_root(fs_info, objectid);
1811 key.objectid = objectid;
1812 key.type = BTRFS_ROOT_ITEM_KEY;
1813 key.offset = (u64)-1;
1814 root = read_tree_root_path(fs_info->tree_root, path, &key);
1815 btrfs_release_path(path);
1821 * called by the kthread helper functions to finally call the bio end_io
1822 * functions. This is where read checksum verification actually happens
1824 static void end_workqueue_fn(struct btrfs_work *work)
1827 struct btrfs_end_io_wq *end_io_wq;
1829 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1830 bio = end_io_wq->bio;
1832 bio->bi_status = end_io_wq->status;
1833 bio->bi_private = end_io_wq->private;
1834 bio->bi_end_io = end_io_wq->end_io;
1836 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1839 static int cleaner_kthread(void *arg)
1841 struct btrfs_root *root = arg;
1842 struct btrfs_fs_info *fs_info = root->fs_info;
1848 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1850 /* Make the cleaner go to sleep early. */
1851 if (btrfs_need_cleaner_sleep(fs_info))
1855 * Do not do anything if we might cause open_ctree() to block
1856 * before we have finished mounting the filesystem.
1858 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1861 if (!mutex_trylock(&fs_info->cleaner_mutex))
1865 * Avoid the problem that we change the status of the fs
1866 * during the above check and trylock.
1868 if (btrfs_need_cleaner_sleep(fs_info)) {
1869 mutex_unlock(&fs_info->cleaner_mutex);
1873 btrfs_run_delayed_iputs(fs_info);
1875 again = btrfs_clean_one_deleted_snapshot(root);
1876 mutex_unlock(&fs_info->cleaner_mutex);
1879 * The defragger has dealt with the R/O remount and umount,
1880 * needn't do anything special here.
1882 btrfs_run_defrag_inodes(fs_info);
1885 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1886 * with relocation (btrfs_relocate_chunk) and relocation
1887 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1888 * after acquiring fs_info->reclaim_bgs_lock. So we
1889 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1890 * unused block groups.
1892 btrfs_delete_unused_bgs(fs_info);
1895 * Reclaim block groups in the reclaim_bgs list after we deleted
1896 * all unused block_groups. This possibly gives us some more free
1899 btrfs_reclaim_bgs(fs_info);
1901 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1902 if (kthread_should_park())
1904 if (kthread_should_stop())
1907 set_current_state(TASK_INTERRUPTIBLE);
1909 __set_current_state(TASK_RUNNING);
1914 static int transaction_kthread(void *arg)
1916 struct btrfs_root *root = arg;
1917 struct btrfs_fs_info *fs_info = root->fs_info;
1918 struct btrfs_trans_handle *trans;
1919 struct btrfs_transaction *cur;
1922 unsigned long delay;
1926 cannot_commit = false;
1927 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1928 mutex_lock(&fs_info->transaction_kthread_mutex);
1930 spin_lock(&fs_info->trans_lock);
1931 cur = fs_info->running_transaction;
1933 spin_unlock(&fs_info->trans_lock);
1937 delta = ktime_get_seconds() - cur->start_time;
1938 if (cur->state < TRANS_STATE_COMMIT_START &&
1939 delta < fs_info->commit_interval) {
1940 spin_unlock(&fs_info->trans_lock);
1941 delay -= msecs_to_jiffies((delta - 1) * 1000);
1943 msecs_to_jiffies(fs_info->commit_interval * 1000));
1946 transid = cur->transid;
1947 spin_unlock(&fs_info->trans_lock);
1949 /* If the file system is aborted, this will always fail. */
1950 trans = btrfs_attach_transaction(root);
1951 if (IS_ERR(trans)) {
1952 if (PTR_ERR(trans) != -ENOENT)
1953 cannot_commit = true;
1956 if (transid == trans->transid) {
1957 btrfs_commit_transaction(trans);
1959 btrfs_end_transaction(trans);
1962 wake_up_process(fs_info->cleaner_kthread);
1963 mutex_unlock(&fs_info->transaction_kthread_mutex);
1965 if (BTRFS_FS_ERROR(fs_info))
1966 btrfs_cleanup_transaction(fs_info);
1967 if (!kthread_should_stop() &&
1968 (!btrfs_transaction_blocked(fs_info) ||
1970 schedule_timeout_interruptible(delay);
1971 } while (!kthread_should_stop());
1976 * This will find the highest generation in the array of root backups. The
1977 * index of the highest array is returned, or -EINVAL if we can't find
1980 * We check to make sure the array is valid by comparing the
1981 * generation of the latest root in the array with the generation
1982 * in the super block. If they don't match we pitch it.
1984 static int find_newest_super_backup(struct btrfs_fs_info *info)
1986 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1988 struct btrfs_root_backup *root_backup;
1991 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1992 root_backup = info->super_copy->super_roots + i;
1993 cur = btrfs_backup_tree_root_gen(root_backup);
1994 if (cur == newest_gen)
2002 * copy all the root pointers into the super backup array.
2003 * this will bump the backup pointer by one when it is
2006 static void backup_super_roots(struct btrfs_fs_info *info)
2008 const int next_backup = info->backup_root_index;
2009 struct btrfs_root_backup *root_backup;
2011 root_backup = info->super_for_commit->super_roots + next_backup;
2014 * make sure all of our padding and empty slots get zero filled
2015 * regardless of which ones we use today
2017 memset(root_backup, 0, sizeof(*root_backup));
2019 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2021 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2022 btrfs_set_backup_tree_root_gen(root_backup,
2023 btrfs_header_generation(info->tree_root->node));
2025 btrfs_set_backup_tree_root_level(root_backup,
2026 btrfs_header_level(info->tree_root->node));
2028 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2029 btrfs_set_backup_chunk_root_gen(root_backup,
2030 btrfs_header_generation(info->chunk_root->node));
2031 btrfs_set_backup_chunk_root_level(root_backup,
2032 btrfs_header_level(info->chunk_root->node));
2034 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2035 btrfs_set_backup_extent_root_gen(root_backup,
2036 btrfs_header_generation(info->extent_root->node));
2037 btrfs_set_backup_extent_root_level(root_backup,
2038 btrfs_header_level(info->extent_root->node));
2041 * we might commit during log recovery, which happens before we set
2042 * the fs_root. Make sure it is valid before we fill it in.
2044 if (info->fs_root && info->fs_root->node) {
2045 btrfs_set_backup_fs_root(root_backup,
2046 info->fs_root->node->start);
2047 btrfs_set_backup_fs_root_gen(root_backup,
2048 btrfs_header_generation(info->fs_root->node));
2049 btrfs_set_backup_fs_root_level(root_backup,
2050 btrfs_header_level(info->fs_root->node));
2053 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2054 btrfs_set_backup_dev_root_gen(root_backup,
2055 btrfs_header_generation(info->dev_root->node));
2056 btrfs_set_backup_dev_root_level(root_backup,
2057 btrfs_header_level(info->dev_root->node));
2059 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2060 btrfs_set_backup_csum_root_gen(root_backup,
2061 btrfs_header_generation(info->csum_root->node));
2062 btrfs_set_backup_csum_root_level(root_backup,
2063 btrfs_header_level(info->csum_root->node));
2065 btrfs_set_backup_total_bytes(root_backup,
2066 btrfs_super_total_bytes(info->super_copy));
2067 btrfs_set_backup_bytes_used(root_backup,
2068 btrfs_super_bytes_used(info->super_copy));
2069 btrfs_set_backup_num_devices(root_backup,
2070 btrfs_super_num_devices(info->super_copy));
2073 * if we don't copy this out to the super_copy, it won't get remembered
2074 * for the next commit
2076 memcpy(&info->super_copy->super_roots,
2077 &info->super_for_commit->super_roots,
2078 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2082 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2083 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2085 * fs_info - filesystem whose backup roots need to be read
2086 * priority - priority of backup root required
2088 * Returns backup root index on success and -EINVAL otherwise.
2090 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2092 int backup_index = find_newest_super_backup(fs_info);
2093 struct btrfs_super_block *super = fs_info->super_copy;
2094 struct btrfs_root_backup *root_backup;
2096 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2098 return backup_index;
2100 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2101 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2106 root_backup = super->super_roots + backup_index;
2108 btrfs_set_super_generation(super,
2109 btrfs_backup_tree_root_gen(root_backup));
2110 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2111 btrfs_set_super_root_level(super,
2112 btrfs_backup_tree_root_level(root_backup));
2113 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2116 * Fixme: the total bytes and num_devices need to match or we should
2119 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2120 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2122 return backup_index;
2125 /* helper to cleanup workers */
2126 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2128 btrfs_destroy_workqueue(fs_info->fixup_workers);
2129 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2130 btrfs_destroy_workqueue(fs_info->workers);
2131 btrfs_destroy_workqueue(fs_info->endio_workers);
2132 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2133 btrfs_destroy_workqueue(fs_info->rmw_workers);
2134 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2135 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2136 btrfs_destroy_workqueue(fs_info->delayed_workers);
2137 btrfs_destroy_workqueue(fs_info->caching_workers);
2138 btrfs_destroy_workqueue(fs_info->readahead_workers);
2139 btrfs_destroy_workqueue(fs_info->flush_workers);
2140 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2141 if (fs_info->discard_ctl.discard_workers)
2142 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2144 * Now that all other work queues are destroyed, we can safely destroy
2145 * the queues used for metadata I/O, since tasks from those other work
2146 * queues can do metadata I/O operations.
2148 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2149 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2152 static void free_root_extent_buffers(struct btrfs_root *root)
2155 free_extent_buffer(root->node);
2156 free_extent_buffer(root->commit_root);
2158 root->commit_root = NULL;
2162 /* helper to cleanup tree roots */
2163 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2165 free_root_extent_buffers(info->tree_root);
2167 free_root_extent_buffers(info->dev_root);
2168 free_root_extent_buffers(info->extent_root);
2169 free_root_extent_buffers(info->csum_root);
2170 free_root_extent_buffers(info->quota_root);
2171 free_root_extent_buffers(info->uuid_root);
2172 free_root_extent_buffers(info->fs_root);
2173 free_root_extent_buffers(info->data_reloc_root);
2174 if (free_chunk_root)
2175 free_root_extent_buffers(info->chunk_root);
2176 free_root_extent_buffers(info->free_space_root);
2179 void btrfs_put_root(struct btrfs_root *root)
2184 if (refcount_dec_and_test(&root->refs)) {
2185 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2186 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2188 free_anon_bdev(root->anon_dev);
2189 btrfs_drew_lock_destroy(&root->snapshot_lock);
2190 free_root_extent_buffers(root);
2191 #ifdef CONFIG_BTRFS_DEBUG
2192 spin_lock(&root->fs_info->fs_roots_radix_lock);
2193 list_del_init(&root->leak_list);
2194 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2200 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2203 struct btrfs_root *gang[8];
2206 while (!list_empty(&fs_info->dead_roots)) {
2207 gang[0] = list_entry(fs_info->dead_roots.next,
2208 struct btrfs_root, root_list);
2209 list_del(&gang[0]->root_list);
2211 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2212 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2213 btrfs_put_root(gang[0]);
2217 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2222 for (i = 0; i < ret; i++)
2223 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2227 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2229 mutex_init(&fs_info->scrub_lock);
2230 atomic_set(&fs_info->scrubs_running, 0);
2231 atomic_set(&fs_info->scrub_pause_req, 0);
2232 atomic_set(&fs_info->scrubs_paused, 0);
2233 atomic_set(&fs_info->scrub_cancel_req, 0);
2234 init_waitqueue_head(&fs_info->scrub_pause_wait);
2235 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2238 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2240 spin_lock_init(&fs_info->balance_lock);
2241 mutex_init(&fs_info->balance_mutex);
2242 atomic_set(&fs_info->balance_pause_req, 0);
2243 atomic_set(&fs_info->balance_cancel_req, 0);
2244 fs_info->balance_ctl = NULL;
2245 init_waitqueue_head(&fs_info->balance_wait_q);
2246 atomic_set(&fs_info->reloc_cancel_req, 0);
2249 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2251 struct inode *inode = fs_info->btree_inode;
2253 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2254 set_nlink(inode, 1);
2256 * we set the i_size on the btree inode to the max possible int.
2257 * the real end of the address space is determined by all of
2258 * the devices in the system
2260 inode->i_size = OFFSET_MAX;
2261 inode->i_mapping->a_ops = &btree_aops;
2263 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2264 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2265 IO_TREE_BTREE_INODE_IO, inode);
2266 BTRFS_I(inode)->io_tree.track_uptodate = false;
2267 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2269 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2270 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2271 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2272 btrfs_insert_inode_hash(inode);
2275 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2277 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2278 init_rwsem(&fs_info->dev_replace.rwsem);
2279 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2282 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2284 spin_lock_init(&fs_info->qgroup_lock);
2285 mutex_init(&fs_info->qgroup_ioctl_lock);
2286 fs_info->qgroup_tree = RB_ROOT;
2287 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2288 fs_info->qgroup_seq = 1;
2289 fs_info->qgroup_ulist = NULL;
2290 fs_info->qgroup_rescan_running = false;
2291 mutex_init(&fs_info->qgroup_rescan_lock);
2294 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2295 struct btrfs_fs_devices *fs_devices)
2297 u32 max_active = fs_info->thread_pool_size;
2298 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2301 btrfs_alloc_workqueue(fs_info, "worker",
2302 flags | WQ_HIGHPRI, max_active, 16);
2304 fs_info->delalloc_workers =
2305 btrfs_alloc_workqueue(fs_info, "delalloc",
2306 flags, max_active, 2);
2308 fs_info->flush_workers =
2309 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2310 flags, max_active, 0);
2312 fs_info->caching_workers =
2313 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2315 fs_info->fixup_workers =
2316 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2319 * endios are largely parallel and should have a very
2322 fs_info->endio_workers =
2323 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2324 fs_info->endio_meta_workers =
2325 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2327 fs_info->endio_meta_write_workers =
2328 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2330 fs_info->endio_raid56_workers =
2331 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2333 fs_info->rmw_workers =
2334 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2335 fs_info->endio_write_workers =
2336 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2338 fs_info->endio_freespace_worker =
2339 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2341 fs_info->delayed_workers =
2342 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2344 fs_info->readahead_workers =
2345 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2347 fs_info->qgroup_rescan_workers =
2348 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2349 fs_info->discard_ctl.discard_workers =
2350 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2352 if (!(fs_info->workers && fs_info->delalloc_workers &&
2353 fs_info->flush_workers &&
2354 fs_info->endio_workers && fs_info->endio_meta_workers &&
2355 fs_info->endio_meta_write_workers &&
2356 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2357 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2358 fs_info->caching_workers && fs_info->readahead_workers &&
2359 fs_info->fixup_workers && fs_info->delayed_workers &&
2360 fs_info->qgroup_rescan_workers &&
2361 fs_info->discard_ctl.discard_workers)) {
2368 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2370 struct crypto_shash *csum_shash;
2371 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2373 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2375 if (IS_ERR(csum_shash)) {
2376 btrfs_err(fs_info, "error allocating %s hash for checksum",
2378 return PTR_ERR(csum_shash);
2381 fs_info->csum_shash = csum_shash;
2386 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2387 struct btrfs_fs_devices *fs_devices)
2390 struct btrfs_root *log_tree_root;
2391 struct btrfs_super_block *disk_super = fs_info->super_copy;
2392 u64 bytenr = btrfs_super_log_root(disk_super);
2393 int level = btrfs_super_log_root_level(disk_super);
2395 if (fs_devices->rw_devices == 0) {
2396 btrfs_warn(fs_info, "log replay required on RO media");
2400 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2405 log_tree_root->node = read_tree_block(fs_info, bytenr,
2406 BTRFS_TREE_LOG_OBJECTID,
2407 fs_info->generation + 1, level,
2409 if (IS_ERR(log_tree_root->node)) {
2410 btrfs_warn(fs_info, "failed to read log tree");
2411 ret = PTR_ERR(log_tree_root->node);
2412 log_tree_root->node = NULL;
2413 btrfs_put_root(log_tree_root);
2415 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2416 btrfs_err(fs_info, "failed to read log tree");
2417 btrfs_put_root(log_tree_root);
2420 /* returns with log_tree_root freed on success */
2421 ret = btrfs_recover_log_trees(log_tree_root);
2423 btrfs_handle_fs_error(fs_info, ret,
2424 "Failed to recover log tree");
2425 btrfs_put_root(log_tree_root);
2429 if (sb_rdonly(fs_info->sb)) {
2430 ret = btrfs_commit_super(fs_info);
2438 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2440 struct btrfs_root *tree_root = fs_info->tree_root;
2441 struct btrfs_root *root;
2442 struct btrfs_key location;
2445 BUG_ON(!fs_info->tree_root);
2447 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2448 location.type = BTRFS_ROOT_ITEM_KEY;
2449 location.offset = 0;
2451 root = btrfs_read_tree_root(tree_root, &location);
2453 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2454 ret = PTR_ERR(root);
2458 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2459 fs_info->extent_root = root;
2462 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2463 root = btrfs_read_tree_root(tree_root, &location);
2465 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2466 ret = PTR_ERR(root);
2470 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2471 fs_info->dev_root = root;
2473 /* Initialize fs_info for all devices in any case */
2474 btrfs_init_devices_late(fs_info);
2476 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2477 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2478 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2479 root = btrfs_read_tree_root(tree_root, &location);
2481 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2482 ret = PTR_ERR(root);
2486 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2487 fs_info->csum_root = root;
2492 * This tree can share blocks with some other fs tree during relocation
2493 * and we need a proper setup by btrfs_get_fs_root
2495 root = btrfs_get_fs_root(tree_root->fs_info,
2496 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2498 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2499 ret = PTR_ERR(root);
2503 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2504 fs_info->data_reloc_root = root;
2507 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2508 root = btrfs_read_tree_root(tree_root, &location);
2509 if (!IS_ERR(root)) {
2510 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2511 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2512 fs_info->quota_root = root;
2515 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2516 root = btrfs_read_tree_root(tree_root, &location);
2518 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2519 ret = PTR_ERR(root);
2524 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2525 fs_info->uuid_root = root;
2528 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2529 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2530 root = btrfs_read_tree_root(tree_root, &location);
2532 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2533 ret = PTR_ERR(root);
2537 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2538 fs_info->free_space_root = root;
2544 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2545 location.objectid, ret);
2550 * Real super block validation
2551 * NOTE: super csum type and incompat features will not be checked here.
2553 * @sb: super block to check
2554 * @mirror_num: the super block number to check its bytenr:
2555 * 0 the primary (1st) sb
2556 * 1, 2 2nd and 3rd backup copy
2557 * -1 skip bytenr check
2559 static int validate_super(struct btrfs_fs_info *fs_info,
2560 struct btrfs_super_block *sb, int mirror_num)
2562 u64 nodesize = btrfs_super_nodesize(sb);
2563 u64 sectorsize = btrfs_super_sectorsize(sb);
2566 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2567 btrfs_err(fs_info, "no valid FS found");
2570 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2571 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2572 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2575 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2576 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2577 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2580 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2581 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2582 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2585 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2586 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2587 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2592 * Check sectorsize and nodesize first, other check will need it.
2593 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2595 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2596 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2597 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2602 * For 4K page size, we only support 4K sector size.
2603 * For 64K page size, we support 64K and 4K sector sizes.
2605 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2606 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2607 sectorsize != SZ_64K))) {
2609 "sectorsize %llu not yet supported for page size %lu",
2610 sectorsize, PAGE_SIZE);
2614 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2615 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2616 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2619 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2620 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2621 le32_to_cpu(sb->__unused_leafsize), nodesize);
2625 /* Root alignment check */
2626 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2627 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2628 btrfs_super_root(sb));
2631 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2632 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2633 btrfs_super_chunk_root(sb));
2636 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2637 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2638 btrfs_super_log_root(sb));
2642 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2645 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2646 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2650 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2651 memcmp(fs_info->fs_devices->metadata_uuid,
2652 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2654 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2655 fs_info->super_copy->metadata_uuid,
2656 fs_info->fs_devices->metadata_uuid);
2660 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2661 BTRFS_FSID_SIZE) != 0) {
2663 "dev_item UUID does not match metadata fsid: %pU != %pU",
2664 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2669 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2672 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2673 btrfs_err(fs_info, "bytes_used is too small %llu",
2674 btrfs_super_bytes_used(sb));
2677 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2678 btrfs_err(fs_info, "invalid stripesize %u",
2679 btrfs_super_stripesize(sb));
2682 if (btrfs_super_num_devices(sb) > (1UL << 31))
2683 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2684 btrfs_super_num_devices(sb));
2685 if (btrfs_super_num_devices(sb) == 0) {
2686 btrfs_err(fs_info, "number of devices is 0");
2690 if (mirror_num >= 0 &&
2691 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2692 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2693 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2698 * Obvious sys_chunk_array corruptions, it must hold at least one key
2701 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2702 btrfs_err(fs_info, "system chunk array too big %u > %u",
2703 btrfs_super_sys_array_size(sb),
2704 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2707 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2708 + sizeof(struct btrfs_chunk)) {
2709 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2710 btrfs_super_sys_array_size(sb),
2711 sizeof(struct btrfs_disk_key)
2712 + sizeof(struct btrfs_chunk));
2717 * The generation is a global counter, we'll trust it more than the others
2718 * but it's still possible that it's the one that's wrong.
2720 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2722 "suspicious: generation < chunk_root_generation: %llu < %llu",
2723 btrfs_super_generation(sb),
2724 btrfs_super_chunk_root_generation(sb));
2725 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2726 && btrfs_super_cache_generation(sb) != (u64)-1)
2728 "suspicious: generation < cache_generation: %llu < %llu",
2729 btrfs_super_generation(sb),
2730 btrfs_super_cache_generation(sb));
2736 * Validation of super block at mount time.
2737 * Some checks already done early at mount time, like csum type and incompat
2738 * flags will be skipped.
2740 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2742 return validate_super(fs_info, fs_info->super_copy, 0);
2746 * Validation of super block at write time.
2747 * Some checks like bytenr check will be skipped as their values will be
2749 * Extra checks like csum type and incompat flags will be done here.
2751 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2752 struct btrfs_super_block *sb)
2756 ret = validate_super(fs_info, sb, -1);
2759 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2761 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2762 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2765 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2768 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2769 btrfs_super_incompat_flags(sb),
2770 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2776 "super block corruption detected before writing it to disk");
2780 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2782 int backup_index = find_newest_super_backup(fs_info);
2783 struct btrfs_super_block *sb = fs_info->super_copy;
2784 struct btrfs_root *tree_root = fs_info->tree_root;
2785 bool handle_error = false;
2789 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2794 if (!IS_ERR(tree_root->node))
2795 free_extent_buffer(tree_root->node);
2796 tree_root->node = NULL;
2798 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2801 free_root_pointers(fs_info, 0);
2804 * Don't use the log in recovery mode, it won't be
2807 btrfs_set_super_log_root(sb, 0);
2809 /* We can't trust the free space cache either */
2810 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2812 ret = read_backup_root(fs_info, i);
2817 generation = btrfs_super_generation(sb);
2818 level = btrfs_super_root_level(sb);
2819 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2820 BTRFS_ROOT_TREE_OBJECTID,
2821 generation, level, NULL);
2822 if (IS_ERR(tree_root->node)) {
2823 handle_error = true;
2824 ret = PTR_ERR(tree_root->node);
2825 tree_root->node = NULL;
2826 btrfs_warn(fs_info, "couldn't read tree root");
2829 } else if (!extent_buffer_uptodate(tree_root->node)) {
2830 handle_error = true;
2832 btrfs_warn(fs_info, "error while reading tree root");
2836 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2837 tree_root->commit_root = btrfs_root_node(tree_root);
2838 btrfs_set_root_refs(&tree_root->root_item, 1);
2841 * No need to hold btrfs_root::objectid_mutex since the fs
2842 * hasn't been fully initialised and we are the only user
2844 ret = btrfs_init_root_free_objectid(tree_root);
2846 handle_error = true;
2850 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2852 ret = btrfs_read_roots(fs_info);
2854 handle_error = true;
2858 /* All successful */
2859 fs_info->generation = generation;
2860 fs_info->last_trans_committed = generation;
2862 /* Always begin writing backup roots after the one being used */
2863 if (backup_index < 0) {
2864 fs_info->backup_root_index = 0;
2866 fs_info->backup_root_index = backup_index + 1;
2867 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2875 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2877 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2878 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2879 INIT_LIST_HEAD(&fs_info->trans_list);
2880 INIT_LIST_HEAD(&fs_info->dead_roots);
2881 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2882 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2883 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2884 spin_lock_init(&fs_info->delalloc_root_lock);
2885 spin_lock_init(&fs_info->trans_lock);
2886 spin_lock_init(&fs_info->fs_roots_radix_lock);
2887 spin_lock_init(&fs_info->delayed_iput_lock);
2888 spin_lock_init(&fs_info->defrag_inodes_lock);
2889 spin_lock_init(&fs_info->super_lock);
2890 spin_lock_init(&fs_info->buffer_lock);
2891 spin_lock_init(&fs_info->unused_bgs_lock);
2892 spin_lock_init(&fs_info->treelog_bg_lock);
2893 spin_lock_init(&fs_info->zone_active_bgs_lock);
2894 spin_lock_init(&fs_info->relocation_bg_lock);
2895 rwlock_init(&fs_info->tree_mod_log_lock);
2896 mutex_init(&fs_info->unused_bg_unpin_mutex);
2897 mutex_init(&fs_info->reclaim_bgs_lock);
2898 mutex_init(&fs_info->reloc_mutex);
2899 mutex_init(&fs_info->delalloc_root_mutex);
2900 mutex_init(&fs_info->zoned_meta_io_lock);
2901 seqlock_init(&fs_info->profiles_lock);
2903 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2904 INIT_LIST_HEAD(&fs_info->space_info);
2905 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2906 INIT_LIST_HEAD(&fs_info->unused_bgs);
2907 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2908 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2909 #ifdef CONFIG_BTRFS_DEBUG
2910 INIT_LIST_HEAD(&fs_info->allocated_roots);
2911 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2912 spin_lock_init(&fs_info->eb_leak_lock);
2914 extent_map_tree_init(&fs_info->mapping_tree);
2915 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2916 BTRFS_BLOCK_RSV_GLOBAL);
2917 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2918 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2919 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2920 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2921 BTRFS_BLOCK_RSV_DELOPS);
2922 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2923 BTRFS_BLOCK_RSV_DELREFS);
2925 atomic_set(&fs_info->async_delalloc_pages, 0);
2926 atomic_set(&fs_info->defrag_running, 0);
2927 atomic_set(&fs_info->reada_works_cnt, 0);
2928 atomic_set(&fs_info->nr_delayed_iputs, 0);
2929 atomic64_set(&fs_info->tree_mod_seq, 0);
2930 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2931 fs_info->metadata_ratio = 0;
2932 fs_info->defrag_inodes = RB_ROOT;
2933 atomic64_set(&fs_info->free_chunk_space, 0);
2934 fs_info->tree_mod_log = RB_ROOT;
2935 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2936 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2937 /* readahead state */
2938 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2939 spin_lock_init(&fs_info->reada_lock);
2940 btrfs_init_ref_verify(fs_info);
2942 fs_info->thread_pool_size = min_t(unsigned long,
2943 num_online_cpus() + 2, 8);
2945 INIT_LIST_HEAD(&fs_info->ordered_roots);
2946 spin_lock_init(&fs_info->ordered_root_lock);
2948 btrfs_init_scrub(fs_info);
2949 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2950 fs_info->check_integrity_print_mask = 0;
2952 btrfs_init_balance(fs_info);
2953 btrfs_init_async_reclaim_work(fs_info);
2955 spin_lock_init(&fs_info->block_group_cache_lock);
2956 fs_info->block_group_cache_tree = RB_ROOT;
2957 fs_info->first_logical_byte = (u64)-1;
2959 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2960 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2961 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2963 mutex_init(&fs_info->ordered_operations_mutex);
2964 mutex_init(&fs_info->tree_log_mutex);
2965 mutex_init(&fs_info->chunk_mutex);
2966 mutex_init(&fs_info->transaction_kthread_mutex);
2967 mutex_init(&fs_info->cleaner_mutex);
2968 mutex_init(&fs_info->ro_block_group_mutex);
2969 init_rwsem(&fs_info->commit_root_sem);
2970 init_rwsem(&fs_info->cleanup_work_sem);
2971 init_rwsem(&fs_info->subvol_sem);
2972 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2974 btrfs_init_dev_replace_locks(fs_info);
2975 btrfs_init_qgroup(fs_info);
2976 btrfs_discard_init(fs_info);
2978 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2979 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2981 init_waitqueue_head(&fs_info->transaction_throttle);
2982 init_waitqueue_head(&fs_info->transaction_wait);
2983 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2984 init_waitqueue_head(&fs_info->async_submit_wait);
2985 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2987 /* Usable values until the real ones are cached from the superblock */
2988 fs_info->nodesize = 4096;
2989 fs_info->sectorsize = 4096;
2990 fs_info->sectorsize_bits = ilog2(4096);
2991 fs_info->stripesize = 4096;
2993 spin_lock_init(&fs_info->swapfile_pins_lock);
2994 fs_info->swapfile_pins = RB_ROOT;
2996 spin_lock_init(&fs_info->send_reloc_lock);
2997 fs_info->send_in_progress = 0;
2999 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3000 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3003 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3008 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3009 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3011 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3015 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3019 fs_info->dirty_metadata_batch = PAGE_SIZE *
3020 (1 + ilog2(nr_cpu_ids));
3022 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3026 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3031 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3033 if (!fs_info->delayed_root)
3035 btrfs_init_delayed_root(fs_info->delayed_root);
3038 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3040 return btrfs_alloc_stripe_hash_table(fs_info);
3043 static int btrfs_uuid_rescan_kthread(void *data)
3045 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3049 * 1st step is to iterate through the existing UUID tree and
3050 * to delete all entries that contain outdated data.
3051 * 2nd step is to add all missing entries to the UUID tree.
3053 ret = btrfs_uuid_tree_iterate(fs_info);
3056 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3058 up(&fs_info->uuid_tree_rescan_sem);
3061 return btrfs_uuid_scan_kthread(data);
3064 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3066 struct task_struct *task;
3068 down(&fs_info->uuid_tree_rescan_sem);
3069 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3071 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3072 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3073 up(&fs_info->uuid_tree_rescan_sem);
3074 return PTR_ERR(task);
3081 * Some options only have meaning at mount time and shouldn't persist across
3082 * remounts, or be displayed. Clear these at the end of mount and remount
3085 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3087 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3088 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3092 * Mounting logic specific to read-write file systems. Shared by open_ctree
3093 * and btrfs_remount when remounting from read-only to read-write.
3095 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3098 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3099 bool clear_free_space_tree = false;
3101 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3102 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3103 clear_free_space_tree = true;
3104 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3105 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3106 btrfs_warn(fs_info, "free space tree is invalid");
3107 clear_free_space_tree = true;
3110 if (clear_free_space_tree) {
3111 btrfs_info(fs_info, "clearing free space tree");
3112 ret = btrfs_clear_free_space_tree(fs_info);
3115 "failed to clear free space tree: %d", ret);
3121 * btrfs_find_orphan_roots() is responsible for finding all the dead
3122 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3123 * them into the fs_info->fs_roots_radix tree. This must be done before
3124 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3125 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3126 * item before the root's tree is deleted - this means that if we unmount
3127 * or crash before the deletion completes, on the next mount we will not
3128 * delete what remains of the tree because the orphan item does not
3129 * exists anymore, which is what tells us we have a pending deletion.
3131 ret = btrfs_find_orphan_roots(fs_info);
3135 ret = btrfs_cleanup_fs_roots(fs_info);
3139 down_read(&fs_info->cleanup_work_sem);
3140 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3141 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3142 up_read(&fs_info->cleanup_work_sem);
3145 up_read(&fs_info->cleanup_work_sem);
3147 mutex_lock(&fs_info->cleaner_mutex);
3148 ret = btrfs_recover_relocation(fs_info->tree_root);
3149 mutex_unlock(&fs_info->cleaner_mutex);
3151 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3155 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3156 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3157 btrfs_info(fs_info, "creating free space tree");
3158 ret = btrfs_create_free_space_tree(fs_info);
3161 "failed to create free space tree: %d", ret);
3166 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3167 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3172 ret = btrfs_resume_balance_async(fs_info);
3176 ret = btrfs_resume_dev_replace_async(fs_info);
3178 btrfs_warn(fs_info, "failed to resume dev_replace");
3182 btrfs_qgroup_rescan_resume(fs_info);
3184 if (!fs_info->uuid_root) {
3185 btrfs_info(fs_info, "creating UUID tree");
3186 ret = btrfs_create_uuid_tree(fs_info);
3189 "failed to create the UUID tree %d", ret);
3198 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3207 struct btrfs_super_block *disk_super;
3208 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3209 struct btrfs_root *tree_root;
3210 struct btrfs_root *chunk_root;
3215 ret = init_mount_fs_info(fs_info, sb);
3221 /* These need to be init'ed before we start creating inodes and such. */
3222 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3224 fs_info->tree_root = tree_root;
3225 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3227 fs_info->chunk_root = chunk_root;
3228 if (!tree_root || !chunk_root) {
3233 fs_info->btree_inode = new_inode(sb);
3234 if (!fs_info->btree_inode) {
3238 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3239 btrfs_init_btree_inode(fs_info);
3241 invalidate_bdev(fs_devices->latest_dev->bdev);
3244 * Read super block and check the signature bytes only
3246 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3247 if (IS_ERR(disk_super)) {
3248 err = PTR_ERR(disk_super);
3253 * Verify the type first, if that or the checksum value are
3254 * corrupted, we'll find out
3256 csum_type = btrfs_super_csum_type(disk_super);
3257 if (!btrfs_supported_super_csum(csum_type)) {
3258 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3261 btrfs_release_disk_super(disk_super);
3265 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3267 ret = btrfs_init_csum_hash(fs_info, csum_type);
3270 btrfs_release_disk_super(disk_super);
3275 * We want to check superblock checksum, the type is stored inside.
3276 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3278 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3279 btrfs_err(fs_info, "superblock checksum mismatch");
3281 btrfs_release_disk_super(disk_super);
3286 * super_copy is zeroed at allocation time and we never touch the
3287 * following bytes up to INFO_SIZE, the checksum is calculated from
3288 * the whole block of INFO_SIZE
3290 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3291 btrfs_release_disk_super(disk_super);
3293 disk_super = fs_info->super_copy;
3296 features = btrfs_super_flags(disk_super);
3297 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3298 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3299 btrfs_set_super_flags(disk_super, features);
3301 "found metadata UUID change in progress flag, clearing");
3304 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3305 sizeof(*fs_info->super_for_commit));
3307 ret = btrfs_validate_mount_super(fs_info);
3309 btrfs_err(fs_info, "superblock contains fatal errors");
3314 if (!btrfs_super_root(disk_super))
3317 /* check FS state, whether FS is broken. */
3318 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3319 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3322 * In the long term, we'll store the compression type in the super
3323 * block, and it'll be used for per file compression control.
3325 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3328 * Flag our filesystem as having big metadata blocks if they are bigger
3329 * than the page size.
3331 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3332 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3334 "flagging fs with big metadata feature");
3335 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3338 /* Set up fs_info before parsing mount options */
3339 nodesize = btrfs_super_nodesize(disk_super);
3340 sectorsize = btrfs_super_sectorsize(disk_super);
3341 stripesize = sectorsize;
3342 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3343 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3345 fs_info->nodesize = nodesize;
3346 fs_info->sectorsize = sectorsize;
3347 fs_info->sectorsize_bits = ilog2(sectorsize);
3348 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3349 fs_info->stripesize = stripesize;
3351 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3357 features = btrfs_super_incompat_flags(disk_super) &
3358 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3361 "cannot mount because of unsupported optional features (%llx)",
3367 features = btrfs_super_incompat_flags(disk_super);
3368 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3369 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3370 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3371 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3372 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3374 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3375 btrfs_info(fs_info, "has skinny extents");
3378 * mixed block groups end up with duplicate but slightly offset
3379 * extent buffers for the same range. It leads to corruptions
3381 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3382 (sectorsize != nodesize)) {
3384 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3385 nodesize, sectorsize);
3390 * Needn't use the lock because there is no other task which will
3393 btrfs_set_super_incompat_flags(disk_super, features);
3395 features = btrfs_super_compat_ro_flags(disk_super) &
3396 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3397 if (!sb_rdonly(sb) && features) {
3399 "cannot mount read-write because of unsupported optional features (%llx)",
3405 if (sectorsize < PAGE_SIZE) {
3406 struct btrfs_subpage_info *subpage_info;
3409 "read-write for sector size %u with page size %lu is experimental",
3410 sectorsize, PAGE_SIZE);
3411 if (btrfs_super_incompat_flags(fs_info->super_copy) &
3412 BTRFS_FEATURE_INCOMPAT_RAID56) {
3414 "RAID56 is not yet supported for sector size %u with page size %lu",
3415 sectorsize, PAGE_SIZE);
3419 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3422 btrfs_init_subpage_info(subpage_info, sectorsize);
3423 fs_info->subpage_info = subpage_info;
3426 ret = btrfs_init_workqueues(fs_info, fs_devices);
3429 goto fail_sb_buffer;
3432 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3433 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3435 sb->s_blocksize = sectorsize;
3436 sb->s_blocksize_bits = blksize_bits(sectorsize);
3437 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3439 mutex_lock(&fs_info->chunk_mutex);
3440 ret = btrfs_read_sys_array(fs_info);
3441 mutex_unlock(&fs_info->chunk_mutex);
3443 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3444 goto fail_sb_buffer;
3447 generation = btrfs_super_chunk_root_generation(disk_super);
3448 level = btrfs_super_chunk_root_level(disk_super);
3450 chunk_root->node = read_tree_block(fs_info,
3451 btrfs_super_chunk_root(disk_super),
3452 BTRFS_CHUNK_TREE_OBJECTID,
3453 generation, level, NULL);
3454 if (IS_ERR(chunk_root->node) ||
3455 !extent_buffer_uptodate(chunk_root->node)) {
3456 btrfs_err(fs_info, "failed to read chunk root");
3457 if (!IS_ERR(chunk_root->node))
3458 free_extent_buffer(chunk_root->node);
3459 chunk_root->node = NULL;
3460 goto fail_tree_roots;
3462 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3463 chunk_root->commit_root = btrfs_root_node(chunk_root);
3465 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3466 offsetof(struct btrfs_header, chunk_tree_uuid),
3469 ret = btrfs_read_chunk_tree(fs_info);
3471 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3472 goto fail_tree_roots;
3476 * At this point we know all the devices that make this filesystem,
3477 * including the seed devices but we don't know yet if the replace
3478 * target is required. So free devices that are not part of this
3479 * filesystem but skip the replace target device which is checked
3480 * below in btrfs_init_dev_replace().
3482 btrfs_free_extra_devids(fs_devices);
3483 if (!fs_devices->latest_dev->bdev) {
3484 btrfs_err(fs_info, "failed to read devices");
3485 goto fail_tree_roots;
3488 ret = init_tree_roots(fs_info);
3490 goto fail_tree_roots;
3493 * Get zone type information of zoned block devices. This will also
3494 * handle emulation of a zoned filesystem if a regular device has the
3495 * zoned incompat feature flag set.
3497 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3500 "zoned: failed to read device zone info: %d",
3502 goto fail_block_groups;
3506 * If we have a uuid root and we're not being told to rescan we need to
3507 * check the generation here so we can set the
3508 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3509 * transaction during a balance or the log replay without updating the
3510 * uuid generation, and then if we crash we would rescan the uuid tree,
3511 * even though it was perfectly fine.
3513 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3514 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3515 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3517 ret = btrfs_verify_dev_extents(fs_info);
3520 "failed to verify dev extents against chunks: %d",
3522 goto fail_block_groups;
3524 ret = btrfs_recover_balance(fs_info);
3526 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3527 goto fail_block_groups;
3530 ret = btrfs_init_dev_stats(fs_info);
3532 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3533 goto fail_block_groups;
3536 ret = btrfs_init_dev_replace(fs_info);
3538 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3539 goto fail_block_groups;
3542 ret = btrfs_check_zoned_mode(fs_info);
3544 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3546 goto fail_block_groups;
3549 ret = btrfs_sysfs_add_fsid(fs_devices);
3551 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3553 goto fail_block_groups;
3556 ret = btrfs_sysfs_add_mounted(fs_info);
3558 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3559 goto fail_fsdev_sysfs;
3562 ret = btrfs_init_space_info(fs_info);
3564 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3568 ret = btrfs_read_block_groups(fs_info);
3570 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3574 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3575 !btrfs_check_rw_degradable(fs_info, NULL)) {
3577 "writable mount is not allowed due to too many missing devices");
3581 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3583 if (IS_ERR(fs_info->cleaner_kthread))
3586 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3588 "btrfs-transaction");
3589 if (IS_ERR(fs_info->transaction_kthread))
3592 if (!btrfs_test_opt(fs_info, NOSSD) &&
3593 !fs_info->fs_devices->rotating) {
3594 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3598 * Mount does not set all options immediately, we can do it now and do
3599 * not have to wait for transaction commit
3601 btrfs_apply_pending_changes(fs_info);
3603 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3604 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3605 ret = btrfsic_mount(fs_info, fs_devices,
3606 btrfs_test_opt(fs_info,
3607 CHECK_INTEGRITY_DATA) ? 1 : 0,
3608 fs_info->check_integrity_print_mask);
3611 "failed to initialize integrity check module: %d",
3615 ret = btrfs_read_qgroup_config(fs_info);
3617 goto fail_trans_kthread;
3619 if (btrfs_build_ref_tree(fs_info))
3620 btrfs_err(fs_info, "couldn't build ref tree");
3622 /* do not make disk changes in broken FS or nologreplay is given */
3623 if (btrfs_super_log_root(disk_super) != 0 &&
3624 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3625 btrfs_info(fs_info, "start tree-log replay");
3626 ret = btrfs_replay_log(fs_info, fs_devices);
3633 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3634 if (IS_ERR(fs_info->fs_root)) {
3635 err = PTR_ERR(fs_info->fs_root);
3636 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3637 fs_info->fs_root = NULL;
3644 ret = btrfs_start_pre_rw_mount(fs_info);
3646 close_ctree(fs_info);
3649 btrfs_discard_resume(fs_info);
3651 if (fs_info->uuid_root &&
3652 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3653 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3654 btrfs_info(fs_info, "checking UUID tree");
3655 ret = btrfs_check_uuid_tree(fs_info);
3658 "failed to check the UUID tree: %d", ret);
3659 close_ctree(fs_info);
3664 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3667 btrfs_clear_oneshot_options(fs_info);
3671 btrfs_free_qgroup_config(fs_info);
3673 kthread_stop(fs_info->transaction_kthread);
3674 btrfs_cleanup_transaction(fs_info);
3675 btrfs_free_fs_roots(fs_info);
3677 kthread_stop(fs_info->cleaner_kthread);
3680 * make sure we're done with the btree inode before we stop our
3683 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3686 btrfs_sysfs_remove_mounted(fs_info);
3689 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3692 btrfs_put_block_group_cache(fs_info);
3695 if (fs_info->data_reloc_root)
3696 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3697 free_root_pointers(fs_info, true);
3698 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3701 btrfs_stop_all_workers(fs_info);
3702 btrfs_free_block_groups(fs_info);
3704 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3706 iput(fs_info->btree_inode);
3708 btrfs_close_devices(fs_info->fs_devices);
3711 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3713 static void btrfs_end_super_write(struct bio *bio)
3715 struct btrfs_device *device = bio->bi_private;
3716 struct bio_vec *bvec;
3717 struct bvec_iter_all iter_all;
3720 bio_for_each_segment_all(bvec, bio, iter_all) {
3721 page = bvec->bv_page;
3723 if (bio->bi_status) {
3724 btrfs_warn_rl_in_rcu(device->fs_info,
3725 "lost page write due to IO error on %s (%d)",
3726 rcu_str_deref(device->name),
3727 blk_status_to_errno(bio->bi_status));
3728 ClearPageUptodate(page);
3730 btrfs_dev_stat_inc_and_print(device,
3731 BTRFS_DEV_STAT_WRITE_ERRS);
3733 SetPageUptodate(page);
3743 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3746 struct btrfs_super_block *super;
3748 u64 bytenr, bytenr_orig;
3749 struct address_space *mapping = bdev->bd_inode->i_mapping;
3752 bytenr_orig = btrfs_sb_offset(copy_num);
3753 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3755 return ERR_PTR(-EINVAL);
3757 return ERR_PTR(ret);
3759 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3760 return ERR_PTR(-EINVAL);
3762 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3764 return ERR_CAST(page);
3766 super = page_address(page);
3767 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3768 btrfs_release_disk_super(super);
3769 return ERR_PTR(-ENODATA);
3772 if (btrfs_super_bytenr(super) != bytenr_orig) {
3773 btrfs_release_disk_super(super);
3774 return ERR_PTR(-EINVAL);
3781 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3783 struct btrfs_super_block *super, *latest = NULL;
3787 /* we would like to check all the supers, but that would make
3788 * a btrfs mount succeed after a mkfs from a different FS.
3789 * So, we need to add a special mount option to scan for
3790 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3792 for (i = 0; i < 1; i++) {
3793 super = btrfs_read_dev_one_super(bdev, i);
3797 if (!latest || btrfs_super_generation(super) > transid) {
3799 btrfs_release_disk_super(super);
3802 transid = btrfs_super_generation(super);
3810 * Write superblock @sb to the @device. Do not wait for completion, all the
3811 * pages we use for writing are locked.
3813 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3814 * the expected device size at commit time. Note that max_mirrors must be
3815 * same for write and wait phases.
3817 * Return number of errors when page is not found or submission fails.
3819 static int write_dev_supers(struct btrfs_device *device,
3820 struct btrfs_super_block *sb, int max_mirrors)
3822 struct btrfs_fs_info *fs_info = device->fs_info;
3823 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3824 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3828 u64 bytenr, bytenr_orig;
3830 if (max_mirrors == 0)
3831 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3833 shash->tfm = fs_info->csum_shash;
3835 for (i = 0; i < max_mirrors; i++) {
3838 struct btrfs_super_block *disk_super;
3840 bytenr_orig = btrfs_sb_offset(i);
3841 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3842 if (ret == -ENOENT) {
3844 } else if (ret < 0) {
3845 btrfs_err(device->fs_info,
3846 "couldn't get super block location for mirror %d",
3851 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3852 device->commit_total_bytes)
3855 btrfs_set_super_bytenr(sb, bytenr_orig);
3857 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3858 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3861 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3864 btrfs_err(device->fs_info,
3865 "couldn't get super block page for bytenr %llu",
3871 /* Bump the refcount for wait_dev_supers() */
3874 disk_super = page_address(page);
3875 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3878 * Directly use bios here instead of relying on the page cache
3879 * to do I/O, so we don't lose the ability to do integrity
3882 bio = bio_alloc(GFP_NOFS, 1);
3883 bio_set_dev(bio, device->bdev);
3884 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3885 bio->bi_private = device;
3886 bio->bi_end_io = btrfs_end_super_write;
3887 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3888 offset_in_page(bytenr));
3891 * We FUA only the first super block. The others we allow to
3892 * go down lazy and there's a short window where the on-disk
3893 * copies might still contain the older version.
3895 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3896 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3897 bio->bi_opf |= REQ_FUA;
3899 btrfsic_submit_bio(bio);
3901 if (btrfs_advance_sb_log(device, i))
3904 return errors < i ? 0 : -1;
3908 * Wait for write completion of superblocks done by write_dev_supers,
3909 * @max_mirrors same for write and wait phases.
3911 * Return number of errors when page is not found or not marked up to
3914 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3918 bool primary_failed = false;
3922 if (max_mirrors == 0)
3923 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3925 for (i = 0; i < max_mirrors; i++) {
3928 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3929 if (ret == -ENOENT) {
3931 } else if (ret < 0) {
3934 primary_failed = true;
3937 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3938 device->commit_total_bytes)
3941 page = find_get_page(device->bdev->bd_inode->i_mapping,
3942 bytenr >> PAGE_SHIFT);
3946 primary_failed = true;
3949 /* Page is submitted locked and unlocked once the IO completes */
3950 wait_on_page_locked(page);
3951 if (PageError(page)) {
3954 primary_failed = true;
3957 /* Drop our reference */
3960 /* Drop the reference from the writing run */
3964 /* log error, force error return */
3965 if (primary_failed) {
3966 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3971 return errors < i ? 0 : -1;
3975 * endio for the write_dev_flush, this will wake anyone waiting
3976 * for the barrier when it is done
3978 static void btrfs_end_empty_barrier(struct bio *bio)
3980 complete(bio->bi_private);
3984 * Submit a flush request to the device if it supports it. Error handling is
3985 * done in the waiting counterpart.
3987 static void write_dev_flush(struct btrfs_device *device)
3989 struct bio *bio = device->flush_bio;
3991 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3993 * When a disk has write caching disabled, we skip submission of a bio
3994 * with flush and sync requests before writing the superblock, since
3995 * it's not needed. However when the integrity checker is enabled, this
3996 * results in reports that there are metadata blocks referred by a
3997 * superblock that were not properly flushed. So don't skip the bio
3998 * submission only when the integrity checker is enabled for the sake
3999 * of simplicity, since this is a debug tool and not meant for use in
4002 struct request_queue *q = bdev_get_queue(device->bdev);
4003 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
4008 bio->bi_end_io = btrfs_end_empty_barrier;
4009 bio_set_dev(bio, device->bdev);
4010 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
4011 init_completion(&device->flush_wait);
4012 bio->bi_private = &device->flush_wait;
4014 btrfsic_submit_bio(bio);
4015 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4019 * If the flush bio has been submitted by write_dev_flush, wait for it.
4021 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4023 struct bio *bio = device->flush_bio;
4025 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4028 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4029 wait_for_completion_io(&device->flush_wait);
4031 return bio->bi_status;
4034 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4036 if (!btrfs_check_rw_degradable(fs_info, NULL))
4042 * send an empty flush down to each device in parallel,
4043 * then wait for them
4045 static int barrier_all_devices(struct btrfs_fs_info *info)
4047 struct list_head *head;
4048 struct btrfs_device *dev;
4049 int errors_wait = 0;
4052 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4053 /* send down all the barriers */
4054 head = &info->fs_devices->devices;
4055 list_for_each_entry(dev, head, dev_list) {
4056 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4060 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4061 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4064 write_dev_flush(dev);
4065 dev->last_flush_error = BLK_STS_OK;
4068 /* wait for all the barriers */
4069 list_for_each_entry(dev, head, dev_list) {
4070 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4076 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4077 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4080 ret = wait_dev_flush(dev);
4082 dev->last_flush_error = ret;
4083 btrfs_dev_stat_inc_and_print(dev,
4084 BTRFS_DEV_STAT_FLUSH_ERRS);
4091 * At some point we need the status of all disks
4092 * to arrive at the volume status. So error checking
4093 * is being pushed to a separate loop.
4095 return check_barrier_error(info);
4100 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4103 int min_tolerated = INT_MAX;
4105 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4106 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4107 min_tolerated = min_t(int, min_tolerated,
4108 btrfs_raid_array[BTRFS_RAID_SINGLE].
4109 tolerated_failures);
4111 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4112 if (raid_type == BTRFS_RAID_SINGLE)
4114 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4116 min_tolerated = min_t(int, min_tolerated,
4117 btrfs_raid_array[raid_type].
4118 tolerated_failures);
4121 if (min_tolerated == INT_MAX) {
4122 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4126 return min_tolerated;
4129 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4131 struct list_head *head;
4132 struct btrfs_device *dev;
4133 struct btrfs_super_block *sb;
4134 struct btrfs_dev_item *dev_item;
4138 int total_errors = 0;
4141 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4144 * max_mirrors == 0 indicates we're from commit_transaction,
4145 * not from fsync where the tree roots in fs_info have not
4146 * been consistent on disk.
4148 if (max_mirrors == 0)
4149 backup_super_roots(fs_info);
4151 sb = fs_info->super_for_commit;
4152 dev_item = &sb->dev_item;
4154 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4155 head = &fs_info->fs_devices->devices;
4156 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4159 ret = barrier_all_devices(fs_info);
4162 &fs_info->fs_devices->device_list_mutex);
4163 btrfs_handle_fs_error(fs_info, ret,
4164 "errors while submitting device barriers.");
4169 list_for_each_entry(dev, head, dev_list) {
4174 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4175 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4178 btrfs_set_stack_device_generation(dev_item, 0);
4179 btrfs_set_stack_device_type(dev_item, dev->type);
4180 btrfs_set_stack_device_id(dev_item, dev->devid);
4181 btrfs_set_stack_device_total_bytes(dev_item,
4182 dev->commit_total_bytes);
4183 btrfs_set_stack_device_bytes_used(dev_item,
4184 dev->commit_bytes_used);
4185 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4186 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4187 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4188 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4189 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4192 flags = btrfs_super_flags(sb);
4193 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4195 ret = btrfs_validate_write_super(fs_info, sb);
4197 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4198 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4199 "unexpected superblock corruption detected");
4203 ret = write_dev_supers(dev, sb, max_mirrors);
4207 if (total_errors > max_errors) {
4208 btrfs_err(fs_info, "%d errors while writing supers",
4210 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4212 /* FUA is masked off if unsupported and can't be the reason */
4213 btrfs_handle_fs_error(fs_info, -EIO,
4214 "%d errors while writing supers",
4220 list_for_each_entry(dev, head, dev_list) {
4223 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4224 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4227 ret = wait_dev_supers(dev, max_mirrors);
4231 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4232 if (total_errors > max_errors) {
4233 btrfs_handle_fs_error(fs_info, -EIO,
4234 "%d errors while writing supers",
4241 /* Drop a fs root from the radix tree and free it. */
4242 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4243 struct btrfs_root *root)
4245 bool drop_ref = false;
4247 spin_lock(&fs_info->fs_roots_radix_lock);
4248 radix_tree_delete(&fs_info->fs_roots_radix,
4249 (unsigned long)root->root_key.objectid);
4250 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4252 spin_unlock(&fs_info->fs_roots_radix_lock);
4254 if (BTRFS_FS_ERROR(fs_info)) {
4255 ASSERT(root->log_root == NULL);
4256 if (root->reloc_root) {
4257 btrfs_put_root(root->reloc_root);
4258 root->reloc_root = NULL;
4263 btrfs_put_root(root);
4266 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4268 u64 root_objectid = 0;
4269 struct btrfs_root *gang[8];
4272 unsigned int ret = 0;
4275 spin_lock(&fs_info->fs_roots_radix_lock);
4276 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4277 (void **)gang, root_objectid,
4280 spin_unlock(&fs_info->fs_roots_radix_lock);
4283 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4285 for (i = 0; i < ret; i++) {
4286 /* Avoid to grab roots in dead_roots */
4287 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4291 /* grab all the search result for later use */
4292 gang[i] = btrfs_grab_root(gang[i]);
4294 spin_unlock(&fs_info->fs_roots_radix_lock);
4296 for (i = 0; i < ret; i++) {
4299 root_objectid = gang[i]->root_key.objectid;
4300 err = btrfs_orphan_cleanup(gang[i]);
4303 btrfs_put_root(gang[i]);
4308 /* release the uncleaned roots due to error */
4309 for (; i < ret; i++) {
4311 btrfs_put_root(gang[i]);
4316 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4318 struct btrfs_root *root = fs_info->tree_root;
4319 struct btrfs_trans_handle *trans;
4321 mutex_lock(&fs_info->cleaner_mutex);
4322 btrfs_run_delayed_iputs(fs_info);
4323 mutex_unlock(&fs_info->cleaner_mutex);
4324 wake_up_process(fs_info->cleaner_kthread);
4326 /* wait until ongoing cleanup work done */
4327 down_write(&fs_info->cleanup_work_sem);
4328 up_write(&fs_info->cleanup_work_sem);
4330 trans = btrfs_join_transaction(root);
4332 return PTR_ERR(trans);
4333 return btrfs_commit_transaction(trans);
4336 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4340 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4342 * We don't want the cleaner to start new transactions, add more delayed
4343 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4344 * because that frees the task_struct, and the transaction kthread might
4345 * still try to wake up the cleaner.
4347 kthread_park(fs_info->cleaner_kthread);
4349 /* wait for the qgroup rescan worker to stop */
4350 btrfs_qgroup_wait_for_completion(fs_info, false);
4352 /* wait for the uuid_scan task to finish */
4353 down(&fs_info->uuid_tree_rescan_sem);
4354 /* avoid complains from lockdep et al., set sem back to initial state */
4355 up(&fs_info->uuid_tree_rescan_sem);
4357 /* pause restriper - we want to resume on mount */
4358 btrfs_pause_balance(fs_info);
4360 btrfs_dev_replace_suspend_for_unmount(fs_info);
4362 btrfs_scrub_cancel(fs_info);
4364 /* wait for any defraggers to finish */
4365 wait_event(fs_info->transaction_wait,
4366 (atomic_read(&fs_info->defrag_running) == 0));
4368 /* clear out the rbtree of defraggable inodes */
4369 btrfs_cleanup_defrag_inodes(fs_info);
4371 cancel_work_sync(&fs_info->async_reclaim_work);
4372 cancel_work_sync(&fs_info->async_data_reclaim_work);
4373 cancel_work_sync(&fs_info->preempt_reclaim_work);
4375 cancel_work_sync(&fs_info->reclaim_bgs_work);
4377 /* Cancel or finish ongoing discard work */
4378 btrfs_discard_cleanup(fs_info);
4380 if (!sb_rdonly(fs_info->sb)) {
4382 * The cleaner kthread is stopped, so do one final pass over
4383 * unused block groups.
4385 btrfs_delete_unused_bgs(fs_info);
4388 * There might be existing delayed inode workers still running
4389 * and holding an empty delayed inode item. We must wait for
4390 * them to complete first because they can create a transaction.
4391 * This happens when someone calls btrfs_balance_delayed_items()
4392 * and then a transaction commit runs the same delayed nodes
4393 * before any delayed worker has done something with the nodes.
4394 * We must wait for any worker here and not at transaction
4395 * commit time since that could cause a deadlock.
4396 * This is a very rare case.
4398 btrfs_flush_workqueue(fs_info->delayed_workers);
4400 ret = btrfs_commit_super(fs_info);
4402 btrfs_err(fs_info, "commit super ret %d", ret);
4405 if (BTRFS_FS_ERROR(fs_info))
4406 btrfs_error_commit_super(fs_info);
4408 kthread_stop(fs_info->transaction_kthread);
4409 kthread_stop(fs_info->cleaner_kthread);
4411 ASSERT(list_empty(&fs_info->delayed_iputs));
4412 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4414 if (btrfs_check_quota_leak(fs_info)) {
4415 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4416 btrfs_err(fs_info, "qgroup reserved space leaked");
4419 btrfs_free_qgroup_config(fs_info);
4420 ASSERT(list_empty(&fs_info->delalloc_roots));
4422 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4423 btrfs_info(fs_info, "at unmount delalloc count %lld",
4424 percpu_counter_sum(&fs_info->delalloc_bytes));
4427 if (percpu_counter_sum(&fs_info->ordered_bytes))
4428 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4429 percpu_counter_sum(&fs_info->ordered_bytes));
4431 btrfs_sysfs_remove_mounted(fs_info);
4432 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4434 btrfs_put_block_group_cache(fs_info);
4437 * we must make sure there is not any read request to
4438 * submit after we stopping all workers.
4440 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4441 btrfs_stop_all_workers(fs_info);
4443 /* We shouldn't have any transaction open at this point */
4444 ASSERT(list_empty(&fs_info->trans_list));
4446 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4447 free_root_pointers(fs_info, true);
4448 btrfs_free_fs_roots(fs_info);
4451 * We must free the block groups after dropping the fs_roots as we could
4452 * have had an IO error and have left over tree log blocks that aren't
4453 * cleaned up until the fs roots are freed. This makes the block group
4454 * accounting appear to be wrong because there's pending reserved bytes,
4455 * so make sure we do the block group cleanup afterwards.
4457 btrfs_free_block_groups(fs_info);
4459 iput(fs_info->btree_inode);
4461 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4462 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4463 btrfsic_unmount(fs_info->fs_devices);
4466 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4467 btrfs_close_devices(fs_info->fs_devices);
4470 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4474 struct inode *btree_inode = buf->pages[0]->mapping->host;
4476 ret = extent_buffer_uptodate(buf);
4480 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4481 parent_transid, atomic);
4487 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4489 struct btrfs_fs_info *fs_info = buf->fs_info;
4490 u64 transid = btrfs_header_generation(buf);
4493 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4495 * This is a fast path so only do this check if we have sanity tests
4496 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4497 * outside of the sanity tests.
4499 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4502 btrfs_assert_tree_write_locked(buf);
4503 if (transid != fs_info->generation)
4504 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4505 buf->start, transid, fs_info->generation);
4506 was_dirty = set_extent_buffer_dirty(buf);
4508 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4510 fs_info->dirty_metadata_batch);
4511 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4513 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4514 * but item data not updated.
4515 * So here we should only check item pointers, not item data.
4517 if (btrfs_header_level(buf) == 0 &&
4518 btrfs_check_leaf_relaxed(buf)) {
4519 btrfs_print_leaf(buf);
4525 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4529 * looks as though older kernels can get into trouble with
4530 * this code, they end up stuck in balance_dirty_pages forever
4534 if (current->flags & PF_MEMALLOC)
4538 btrfs_balance_delayed_items(fs_info);
4540 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4541 BTRFS_DIRTY_METADATA_THRESH,
4542 fs_info->dirty_metadata_batch);
4544 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4548 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4550 __btrfs_btree_balance_dirty(fs_info, 1);
4553 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4555 __btrfs_btree_balance_dirty(fs_info, 0);
4558 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4559 struct btrfs_key *first_key)
4561 return btree_read_extent_buffer_pages(buf, parent_transid,
4565 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4567 /* cleanup FS via transaction */
4568 btrfs_cleanup_transaction(fs_info);
4570 mutex_lock(&fs_info->cleaner_mutex);
4571 btrfs_run_delayed_iputs(fs_info);
4572 mutex_unlock(&fs_info->cleaner_mutex);
4574 down_write(&fs_info->cleanup_work_sem);
4575 up_write(&fs_info->cleanup_work_sem);
4578 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4580 struct btrfs_root *gang[8];
4581 u64 root_objectid = 0;
4584 spin_lock(&fs_info->fs_roots_radix_lock);
4585 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4586 (void **)gang, root_objectid,
4587 ARRAY_SIZE(gang))) != 0) {
4590 for (i = 0; i < ret; i++)
4591 gang[i] = btrfs_grab_root(gang[i]);
4592 spin_unlock(&fs_info->fs_roots_radix_lock);
4594 for (i = 0; i < ret; i++) {
4597 root_objectid = gang[i]->root_key.objectid;
4598 btrfs_free_log(NULL, gang[i]);
4599 btrfs_put_root(gang[i]);
4602 spin_lock(&fs_info->fs_roots_radix_lock);
4604 spin_unlock(&fs_info->fs_roots_radix_lock);
4605 btrfs_free_log_root_tree(NULL, fs_info);
4608 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4610 struct btrfs_ordered_extent *ordered;
4612 spin_lock(&root->ordered_extent_lock);
4614 * This will just short circuit the ordered completion stuff which will
4615 * make sure the ordered extent gets properly cleaned up.
4617 list_for_each_entry(ordered, &root->ordered_extents,
4619 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4620 spin_unlock(&root->ordered_extent_lock);
4623 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4625 struct btrfs_root *root;
4626 struct list_head splice;
4628 INIT_LIST_HEAD(&splice);
4630 spin_lock(&fs_info->ordered_root_lock);
4631 list_splice_init(&fs_info->ordered_roots, &splice);
4632 while (!list_empty(&splice)) {
4633 root = list_first_entry(&splice, struct btrfs_root,
4635 list_move_tail(&root->ordered_root,
4636 &fs_info->ordered_roots);
4638 spin_unlock(&fs_info->ordered_root_lock);
4639 btrfs_destroy_ordered_extents(root);
4642 spin_lock(&fs_info->ordered_root_lock);
4644 spin_unlock(&fs_info->ordered_root_lock);
4647 * We need this here because if we've been flipped read-only we won't
4648 * get sync() from the umount, so we need to make sure any ordered
4649 * extents that haven't had their dirty pages IO start writeout yet
4650 * actually get run and error out properly.
4652 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4655 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4656 struct btrfs_fs_info *fs_info)
4658 struct rb_node *node;
4659 struct btrfs_delayed_ref_root *delayed_refs;
4660 struct btrfs_delayed_ref_node *ref;
4663 delayed_refs = &trans->delayed_refs;
4665 spin_lock(&delayed_refs->lock);
4666 if (atomic_read(&delayed_refs->num_entries) == 0) {
4667 spin_unlock(&delayed_refs->lock);
4668 btrfs_debug(fs_info, "delayed_refs has NO entry");
4672 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4673 struct btrfs_delayed_ref_head *head;
4675 bool pin_bytes = false;
4677 head = rb_entry(node, struct btrfs_delayed_ref_head,
4679 if (btrfs_delayed_ref_lock(delayed_refs, head))
4682 spin_lock(&head->lock);
4683 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4684 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4687 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4688 RB_CLEAR_NODE(&ref->ref_node);
4689 if (!list_empty(&ref->add_list))
4690 list_del(&ref->add_list);
4691 atomic_dec(&delayed_refs->num_entries);
4692 btrfs_put_delayed_ref(ref);
4694 if (head->must_insert_reserved)
4696 btrfs_free_delayed_extent_op(head->extent_op);
4697 btrfs_delete_ref_head(delayed_refs, head);
4698 spin_unlock(&head->lock);
4699 spin_unlock(&delayed_refs->lock);
4700 mutex_unlock(&head->mutex);
4703 struct btrfs_block_group *cache;
4705 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4708 spin_lock(&cache->space_info->lock);
4709 spin_lock(&cache->lock);
4710 cache->pinned += head->num_bytes;
4711 btrfs_space_info_update_bytes_pinned(fs_info,
4712 cache->space_info, head->num_bytes);
4713 cache->reserved -= head->num_bytes;
4714 cache->space_info->bytes_reserved -= head->num_bytes;
4715 spin_unlock(&cache->lock);
4716 spin_unlock(&cache->space_info->lock);
4718 btrfs_put_block_group(cache);
4720 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4721 head->bytenr + head->num_bytes - 1);
4723 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4724 btrfs_put_delayed_ref_head(head);
4726 spin_lock(&delayed_refs->lock);
4728 btrfs_qgroup_destroy_extent_records(trans);
4730 spin_unlock(&delayed_refs->lock);
4735 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4737 struct btrfs_inode *btrfs_inode;
4738 struct list_head splice;
4740 INIT_LIST_HEAD(&splice);
4742 spin_lock(&root->delalloc_lock);
4743 list_splice_init(&root->delalloc_inodes, &splice);
4745 while (!list_empty(&splice)) {
4746 struct inode *inode = NULL;
4747 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4749 __btrfs_del_delalloc_inode(root, btrfs_inode);
4750 spin_unlock(&root->delalloc_lock);
4753 * Make sure we get a live inode and that it'll not disappear
4756 inode = igrab(&btrfs_inode->vfs_inode);
4758 invalidate_inode_pages2(inode->i_mapping);
4761 spin_lock(&root->delalloc_lock);
4763 spin_unlock(&root->delalloc_lock);
4766 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4768 struct btrfs_root *root;
4769 struct list_head splice;
4771 INIT_LIST_HEAD(&splice);
4773 spin_lock(&fs_info->delalloc_root_lock);
4774 list_splice_init(&fs_info->delalloc_roots, &splice);
4775 while (!list_empty(&splice)) {
4776 root = list_first_entry(&splice, struct btrfs_root,
4778 root = btrfs_grab_root(root);
4780 spin_unlock(&fs_info->delalloc_root_lock);
4782 btrfs_destroy_delalloc_inodes(root);
4783 btrfs_put_root(root);
4785 spin_lock(&fs_info->delalloc_root_lock);
4787 spin_unlock(&fs_info->delalloc_root_lock);
4790 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4791 struct extent_io_tree *dirty_pages,
4795 struct extent_buffer *eb;
4800 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4805 clear_extent_bits(dirty_pages, start, end, mark);
4806 while (start <= end) {
4807 eb = find_extent_buffer(fs_info, start);
4808 start += fs_info->nodesize;
4811 wait_on_extent_buffer_writeback(eb);
4813 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4815 clear_extent_buffer_dirty(eb);
4816 free_extent_buffer_stale(eb);
4823 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4824 struct extent_io_tree *unpin)
4831 struct extent_state *cached_state = NULL;
4834 * The btrfs_finish_extent_commit() may get the same range as
4835 * ours between find_first_extent_bit and clear_extent_dirty.
4836 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4837 * the same extent range.
4839 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4840 ret = find_first_extent_bit(unpin, 0, &start, &end,
4841 EXTENT_DIRTY, &cached_state);
4843 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4847 clear_extent_dirty(unpin, start, end, &cached_state);
4848 free_extent_state(cached_state);
4849 btrfs_error_unpin_extent_range(fs_info, start, end);
4850 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4857 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4859 struct inode *inode;
4861 inode = cache->io_ctl.inode;
4863 invalidate_inode_pages2(inode->i_mapping);
4864 BTRFS_I(inode)->generation = 0;
4865 cache->io_ctl.inode = NULL;
4868 ASSERT(cache->io_ctl.pages == NULL);
4869 btrfs_put_block_group(cache);
4872 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4873 struct btrfs_fs_info *fs_info)
4875 struct btrfs_block_group *cache;
4877 spin_lock(&cur_trans->dirty_bgs_lock);
4878 while (!list_empty(&cur_trans->dirty_bgs)) {
4879 cache = list_first_entry(&cur_trans->dirty_bgs,
4880 struct btrfs_block_group,
4883 if (!list_empty(&cache->io_list)) {
4884 spin_unlock(&cur_trans->dirty_bgs_lock);
4885 list_del_init(&cache->io_list);
4886 btrfs_cleanup_bg_io(cache);
4887 spin_lock(&cur_trans->dirty_bgs_lock);
4890 list_del_init(&cache->dirty_list);
4891 spin_lock(&cache->lock);
4892 cache->disk_cache_state = BTRFS_DC_ERROR;
4893 spin_unlock(&cache->lock);
4895 spin_unlock(&cur_trans->dirty_bgs_lock);
4896 btrfs_put_block_group(cache);
4897 btrfs_delayed_refs_rsv_release(fs_info, 1);
4898 spin_lock(&cur_trans->dirty_bgs_lock);
4900 spin_unlock(&cur_trans->dirty_bgs_lock);
4903 * Refer to the definition of io_bgs member for details why it's safe
4904 * to use it without any locking
4906 while (!list_empty(&cur_trans->io_bgs)) {
4907 cache = list_first_entry(&cur_trans->io_bgs,
4908 struct btrfs_block_group,
4911 list_del_init(&cache->io_list);
4912 spin_lock(&cache->lock);
4913 cache->disk_cache_state = BTRFS_DC_ERROR;
4914 spin_unlock(&cache->lock);
4915 btrfs_cleanup_bg_io(cache);
4919 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4920 struct btrfs_fs_info *fs_info)
4922 struct btrfs_device *dev, *tmp;
4924 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4925 ASSERT(list_empty(&cur_trans->dirty_bgs));
4926 ASSERT(list_empty(&cur_trans->io_bgs));
4928 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4930 list_del_init(&dev->post_commit_list);
4933 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4935 cur_trans->state = TRANS_STATE_COMMIT_START;
4936 wake_up(&fs_info->transaction_blocked_wait);
4938 cur_trans->state = TRANS_STATE_UNBLOCKED;
4939 wake_up(&fs_info->transaction_wait);
4941 btrfs_destroy_delayed_inodes(fs_info);
4943 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4945 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4947 btrfs_free_redirty_list(cur_trans);
4949 cur_trans->state =TRANS_STATE_COMPLETED;
4950 wake_up(&cur_trans->commit_wait);
4953 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4955 struct btrfs_transaction *t;
4957 mutex_lock(&fs_info->transaction_kthread_mutex);
4959 spin_lock(&fs_info->trans_lock);
4960 while (!list_empty(&fs_info->trans_list)) {
4961 t = list_first_entry(&fs_info->trans_list,
4962 struct btrfs_transaction, list);
4963 if (t->state >= TRANS_STATE_COMMIT_START) {
4964 refcount_inc(&t->use_count);
4965 spin_unlock(&fs_info->trans_lock);
4966 btrfs_wait_for_commit(fs_info, t->transid);
4967 btrfs_put_transaction(t);
4968 spin_lock(&fs_info->trans_lock);
4971 if (t == fs_info->running_transaction) {
4972 t->state = TRANS_STATE_COMMIT_DOING;
4973 spin_unlock(&fs_info->trans_lock);
4975 * We wait for 0 num_writers since we don't hold a trans
4976 * handle open currently for this transaction.
4978 wait_event(t->writer_wait,
4979 atomic_read(&t->num_writers) == 0);
4981 spin_unlock(&fs_info->trans_lock);
4983 btrfs_cleanup_one_transaction(t, fs_info);
4985 spin_lock(&fs_info->trans_lock);
4986 if (t == fs_info->running_transaction)
4987 fs_info->running_transaction = NULL;
4988 list_del_init(&t->list);
4989 spin_unlock(&fs_info->trans_lock);
4991 btrfs_put_transaction(t);
4992 trace_btrfs_transaction_commit(fs_info->tree_root);
4993 spin_lock(&fs_info->trans_lock);
4995 spin_unlock(&fs_info->trans_lock);
4996 btrfs_destroy_all_ordered_extents(fs_info);
4997 btrfs_destroy_delayed_inodes(fs_info);
4998 btrfs_assert_delayed_root_empty(fs_info);
4999 btrfs_destroy_all_delalloc_inodes(fs_info);
5000 btrfs_drop_all_logs(fs_info);
5001 mutex_unlock(&fs_info->transaction_kthread_mutex);
5006 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5008 struct btrfs_path *path;
5010 struct extent_buffer *l;
5011 struct btrfs_key search_key;
5012 struct btrfs_key found_key;
5015 path = btrfs_alloc_path();
5019 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5020 search_key.type = -1;
5021 search_key.offset = (u64)-1;
5022 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5025 BUG_ON(ret == 0); /* Corruption */
5026 if (path->slots[0] > 0) {
5027 slot = path->slots[0] - 1;
5029 btrfs_item_key_to_cpu(l, &found_key, slot);
5030 root->free_objectid = max_t(u64, found_key.objectid + 1,
5031 BTRFS_FIRST_FREE_OBJECTID);
5033 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5037 btrfs_free_path(path);
5041 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5044 mutex_lock(&root->objectid_mutex);
5046 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5047 btrfs_warn(root->fs_info,
5048 "the objectid of root %llu reaches its highest value",
5049 root->root_key.objectid);
5054 *objectid = root->free_objectid++;
5057 mutex_unlock(&root->objectid_mutex);