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/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include <crypto/hash.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
43 #include "block-group.h"
46 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
53 static const struct extent_io_ops btree_extent_io_ops;
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 * bio_offset is optional, can be used if the pages in the bio
119 * can't tell us where in the file the bio should go
122 struct btrfs_work work;
127 * Lockdep class keys for extent_buffer->lock's in this root. For a given
128 * eb, the lockdep key is determined by the btrfs_root it belongs to and
129 * the level the eb occupies in the tree.
131 * Different roots are used for different purposes and may nest inside each
132 * other and they require separate keysets. As lockdep keys should be
133 * static, assign keysets according to the purpose of the root as indicated
134 * by btrfs_root->root_key.objectid. This ensures that all special purpose
135 * roots have separate keysets.
137 * Lock-nesting across peer nodes is always done with the immediate parent
138 * node locked thus preventing deadlock. As lockdep doesn't know this, use
139 * subclass to avoid triggering lockdep warning in such cases.
141 * The key is set by the readpage_end_io_hook after the buffer has passed
142 * csum validation but before the pages are unlocked. It is also set by
143 * btrfs_init_new_buffer on freshly allocated blocks.
145 * We also add a check to make sure the highest level of the tree is the
146 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
147 * needs update as well.
149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
150 # if BTRFS_MAX_LEVEL != 8
154 static struct btrfs_lockdep_keyset {
155 u64 id; /* root objectid */
156 const char *name_stem; /* lock name stem */
157 char names[BTRFS_MAX_LEVEL + 1][20];
158 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
159 } btrfs_lockdep_keysets[] = {
160 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
161 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
162 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
163 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
164 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
165 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
166 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
167 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
168 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
169 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
170 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
171 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
172 { .id = 0, .name_stem = "tree" },
175 void __init btrfs_init_lockdep(void)
179 /* initialize lockdep class names */
180 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
181 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
183 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
184 snprintf(ks->names[j], sizeof(ks->names[j]),
185 "btrfs-%s-%02d", ks->name_stem, j);
189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
192 struct btrfs_lockdep_keyset *ks;
194 BUG_ON(level >= ARRAY_SIZE(ks->keys));
196 /* find the matching keyset, id 0 is the default entry */
197 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
198 if (ks->id == objectid)
201 lockdep_set_class_and_name(&eb->lock,
202 &ks->keys[level], ks->names[level]);
208 * extents on the btree inode are pretty simple, there's one extent
209 * that covers the entire device
211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
212 struct page *page, size_t pg_offset,
215 struct extent_map_tree *em_tree = &inode->extent_tree;
216 struct extent_map *em;
219 read_lock(&em_tree->lock);
220 em = lookup_extent_mapping(em_tree, start, len);
222 read_unlock(&em_tree->lock);
225 read_unlock(&em_tree->lock);
227 em = alloc_extent_map();
229 em = ERR_PTR(-ENOMEM);
234 em->block_len = (u64)-1;
237 write_lock(&em_tree->lock);
238 ret = add_extent_mapping(em_tree, em, 0);
239 if (ret == -EEXIST) {
241 em = lookup_extent_mapping(em_tree, start, len);
248 write_unlock(&em_tree->lock);
255 * Compute the csum of a btree block and store the result to provided buffer.
257 * Returns error if the extent buffer cannot be mapped.
259 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
261 struct btrfs_fs_info *fs_info = buf->fs_info;
262 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
267 unsigned long map_start;
268 unsigned long map_len;
271 shash->tfm = fs_info->csum_shash;
272 crypto_shash_init(shash);
274 len = buf->len - offset;
278 * Note: we don't need to check for the err == 1 case here, as
279 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
280 * and 'min_len = 32' and the currently implemented mapping
281 * algorithm we cannot cross a page boundary.
283 err = map_private_extent_buffer(buf, offset, 32,
284 &kaddr, &map_start, &map_len);
287 cur_len = min(len, map_len - (offset - map_start));
288 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
292 memset(result, 0, BTRFS_CSUM_SIZE);
294 crypto_shash_final(shash, result);
300 * we can't consider a given block up to date unless the transid of the
301 * block matches the transid in the parent node's pointer. This is how we
302 * detect blocks that either didn't get written at all or got written
303 * in the wrong place.
305 static int verify_parent_transid(struct extent_io_tree *io_tree,
306 struct extent_buffer *eb, u64 parent_transid,
309 struct extent_state *cached_state = NULL;
311 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
313 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
320 btrfs_tree_read_lock(eb);
321 btrfs_set_lock_blocking_read(eb);
324 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
326 if (extent_buffer_uptodate(eb) &&
327 btrfs_header_generation(eb) == parent_transid) {
331 btrfs_err_rl(eb->fs_info,
332 "parent transid verify failed on %llu wanted %llu found %llu",
334 parent_transid, btrfs_header_generation(eb));
338 * Things reading via commit roots that don't have normal protection,
339 * like send, can have a really old block in cache that may point at a
340 * block that has been freed and re-allocated. So don't clear uptodate
341 * if we find an eb that is under IO (dirty/writeback) because we could
342 * end up reading in the stale data and then writing it back out and
343 * making everybody very sad.
345 if (!extent_buffer_under_io(eb))
346 clear_extent_buffer_uptodate(eb);
348 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
351 btrfs_tree_read_unlock_blocking(eb);
355 static bool btrfs_supported_super_csum(u16 csum_type)
358 case BTRFS_CSUM_TYPE_CRC32:
359 case BTRFS_CSUM_TYPE_XXHASH:
360 case BTRFS_CSUM_TYPE_SHA256:
361 case BTRFS_CSUM_TYPE_BLAKE2:
369 * Return 0 if the superblock checksum type matches the checksum value of that
370 * algorithm. Pass the raw disk superblock data.
372 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
375 struct btrfs_super_block *disk_sb =
376 (struct btrfs_super_block *)raw_disk_sb;
377 char result[BTRFS_CSUM_SIZE];
378 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
380 shash->tfm = fs_info->csum_shash;
381 crypto_shash_init(shash);
384 * The super_block structure does not span the whole
385 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
386 * filled with zeros and is included in the checksum.
388 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
389 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
390 crypto_shash_final(shash, result);
392 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
398 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
399 struct btrfs_key *first_key, u64 parent_transid)
401 struct btrfs_fs_info *fs_info = eb->fs_info;
403 struct btrfs_key found_key;
406 found_level = btrfs_header_level(eb);
407 if (found_level != level) {
408 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
409 KERN_ERR "BTRFS: tree level check failed\n");
411 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
412 eb->start, level, found_level);
420 * For live tree block (new tree blocks in current transaction),
421 * we need proper lock context to avoid race, which is impossible here.
422 * So we only checks tree blocks which is read from disk, whose
423 * generation <= fs_info->last_trans_committed.
425 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
428 /* We have @first_key, so this @eb must have at least one item */
429 if (btrfs_header_nritems(eb) == 0) {
431 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
433 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
438 btrfs_node_key_to_cpu(eb, &found_key, 0);
440 btrfs_item_key_to_cpu(eb, &found_key, 0);
441 ret = btrfs_comp_cpu_keys(first_key, &found_key);
444 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
445 KERN_ERR "BTRFS: tree first key check failed\n");
447 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
448 eb->start, parent_transid, first_key->objectid,
449 first_key->type, first_key->offset,
450 found_key.objectid, found_key.type,
457 * helper to read a given tree block, doing retries as required when
458 * the checksums don't match and we have alternate mirrors to try.
460 * @parent_transid: expected transid, skip check if 0
461 * @level: expected level, mandatory check
462 * @first_key: expected key of first slot, skip check if NULL
464 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
465 u64 parent_transid, int level,
466 struct btrfs_key *first_key)
468 struct btrfs_fs_info *fs_info = eb->fs_info;
469 struct extent_io_tree *io_tree;
474 int failed_mirror = 0;
476 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
478 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
479 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
481 if (verify_parent_transid(io_tree, eb,
484 else if (btrfs_verify_level_key(eb, level,
485 first_key, parent_transid))
491 num_copies = btrfs_num_copies(fs_info,
496 if (!failed_mirror) {
498 failed_mirror = eb->read_mirror;
502 if (mirror_num == failed_mirror)
505 if (mirror_num > num_copies)
509 if (failed && !ret && failed_mirror)
510 btrfs_repair_eb_io_failure(eb, failed_mirror);
516 * checksum a dirty tree block before IO. This has extra checks to make sure
517 * we only fill in the checksum field in the first page of a multi-page block
520 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
522 u64 start = page_offset(page);
524 u8 result[BTRFS_CSUM_SIZE];
525 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
526 struct extent_buffer *eb;
529 eb = (struct extent_buffer *)page->private;
530 if (page != eb->pages[0])
533 found_start = btrfs_header_bytenr(eb);
535 * Please do not consolidate these warnings into a single if.
536 * It is useful to know what went wrong.
538 if (WARN_ON(found_start != start))
540 if (WARN_ON(!PageUptodate(page)))
543 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
544 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
546 if (csum_tree_block(eb, result))
549 if (btrfs_header_level(eb))
550 ret = btrfs_check_node(eb);
552 ret = btrfs_check_leaf_full(eb);
555 btrfs_print_tree(eb, 0);
557 "block=%llu write time tree block corruption detected",
559 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
562 write_extent_buffer(eb, result, 0, csum_size);
567 static int check_tree_block_fsid(struct extent_buffer *eb)
569 struct btrfs_fs_info *fs_info = eb->fs_info;
570 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
571 u8 fsid[BTRFS_FSID_SIZE];
574 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
579 * Checking the incompat flag is only valid for the current
580 * fs. For seed devices it's forbidden to have their uuid
581 * changed so reading ->fsid in this case is fine
583 if (fs_devices == fs_info->fs_devices &&
584 btrfs_fs_incompat(fs_info, METADATA_UUID))
585 metadata_uuid = fs_devices->metadata_uuid;
587 metadata_uuid = fs_devices->fsid;
589 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
593 fs_devices = fs_devices->seed;
598 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
599 u64 phy_offset, struct page *page,
600 u64 start, u64 end, int mirror)
604 struct extent_buffer *eb;
605 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
606 struct btrfs_fs_info *fs_info = root->fs_info;
607 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
609 u8 result[BTRFS_CSUM_SIZE];
615 eb = (struct extent_buffer *)page->private;
617 /* the pending IO might have been the only thing that kept this buffer
618 * in memory. Make sure we have a ref for all this other checks
620 atomic_inc(&eb->refs);
622 reads_done = atomic_dec_and_test(&eb->io_pages);
626 eb->read_mirror = mirror;
627 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
632 found_start = btrfs_header_bytenr(eb);
633 if (found_start != eb->start) {
634 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
635 eb->start, found_start);
639 if (check_tree_block_fsid(eb)) {
640 btrfs_err_rl(fs_info, "bad fsid on block %llu",
645 found_level = btrfs_header_level(eb);
646 if (found_level >= BTRFS_MAX_LEVEL) {
647 btrfs_err(fs_info, "bad tree block level %d on %llu",
648 (int)btrfs_header_level(eb), eb->start);
653 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
656 ret = csum_tree_block(eb, result);
660 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
664 memcpy(&found, result, csum_size);
666 read_extent_buffer(eb, &val, 0, csum_size);
667 btrfs_warn_rl(fs_info,
668 "%s checksum verify failed on %llu wanted %x found %x level %d",
669 fs_info->sb->s_id, eb->start,
670 val, found, btrfs_header_level(eb));
676 * If this is a leaf block and it is corrupt, set the corrupt bit so
677 * that we don't try and read the other copies of this block, just
680 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
681 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
685 if (found_level > 0 && btrfs_check_node(eb))
689 set_extent_buffer_uptodate(eb);
692 "block=%llu read time tree block corruption detected",
696 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
697 btree_readahead_hook(eb, ret);
701 * our io error hook is going to dec the io pages
702 * again, we have to make sure it has something
705 atomic_inc(&eb->io_pages);
706 clear_extent_buffer_uptodate(eb);
708 free_extent_buffer(eb);
713 static void end_workqueue_bio(struct bio *bio)
715 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
716 struct btrfs_fs_info *fs_info;
717 struct btrfs_workqueue *wq;
719 fs_info = end_io_wq->info;
720 end_io_wq->status = bio->bi_status;
722 if (bio_op(bio) == REQ_OP_WRITE) {
723 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
724 wq = fs_info->endio_meta_write_workers;
725 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
726 wq = fs_info->endio_freespace_worker;
727 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
728 wq = fs_info->endio_raid56_workers;
730 wq = fs_info->endio_write_workers;
732 if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
733 wq = fs_info->endio_repair_workers;
734 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
735 wq = fs_info->endio_raid56_workers;
736 else if (end_io_wq->metadata)
737 wq = fs_info->endio_meta_workers;
739 wq = fs_info->endio_workers;
742 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
743 btrfs_queue_work(wq, &end_io_wq->work);
746 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
747 enum btrfs_wq_endio_type metadata)
749 struct btrfs_end_io_wq *end_io_wq;
751 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
753 return BLK_STS_RESOURCE;
755 end_io_wq->private = bio->bi_private;
756 end_io_wq->end_io = bio->bi_end_io;
757 end_io_wq->info = info;
758 end_io_wq->status = 0;
759 end_io_wq->bio = bio;
760 end_io_wq->metadata = metadata;
762 bio->bi_private = end_io_wq;
763 bio->bi_end_io = end_workqueue_bio;
767 static void run_one_async_start(struct btrfs_work *work)
769 struct async_submit_bio *async;
772 async = container_of(work, struct async_submit_bio, work);
773 ret = async->submit_bio_start(async->private_data, async->bio,
780 * In order to insert checksums into the metadata in large chunks, we wait
781 * until bio submission time. All the pages in the bio are checksummed and
782 * sums are attached onto the ordered extent record.
784 * At IO completion time the csums attached on the ordered extent record are
785 * inserted into the tree.
787 static void run_one_async_done(struct btrfs_work *work)
789 struct async_submit_bio *async;
793 async = container_of(work, struct async_submit_bio, work);
794 inode = async->private_data;
796 /* If an error occurred we just want to clean up the bio and move on */
798 async->bio->bi_status = async->status;
799 bio_endio(async->bio);
804 * All of the bios that pass through here are from async helpers.
805 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
806 * This changes nothing when cgroups aren't in use.
808 async->bio->bi_opf |= REQ_CGROUP_PUNT;
809 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
811 async->bio->bi_status = ret;
812 bio_endio(async->bio);
816 static void run_one_async_free(struct btrfs_work *work)
818 struct async_submit_bio *async;
820 async = container_of(work, struct async_submit_bio, work);
824 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
825 int mirror_num, unsigned long bio_flags,
826 u64 bio_offset, void *private_data,
827 extent_submit_bio_start_t *submit_bio_start)
829 struct async_submit_bio *async;
831 async = kmalloc(sizeof(*async), GFP_NOFS);
833 return BLK_STS_RESOURCE;
835 async->private_data = private_data;
837 async->mirror_num = mirror_num;
838 async->submit_bio_start = submit_bio_start;
840 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
843 async->bio_offset = bio_offset;
847 if (op_is_sync(bio->bi_opf))
848 btrfs_set_work_high_priority(&async->work);
850 btrfs_queue_work(fs_info->workers, &async->work);
854 static blk_status_t btree_csum_one_bio(struct bio *bio)
856 struct bio_vec *bvec;
857 struct btrfs_root *root;
859 struct bvec_iter_all iter_all;
861 ASSERT(!bio_flagged(bio, BIO_CLONED));
862 bio_for_each_segment_all(bvec, bio, iter_all) {
863 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
864 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
869 return errno_to_blk_status(ret);
872 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
876 * when we're called for a write, we're already in the async
877 * submission context. Just jump into btrfs_map_bio
879 return btree_csum_one_bio(bio);
882 static int check_async_write(struct btrfs_fs_info *fs_info,
883 struct btrfs_inode *bi)
885 if (atomic_read(&bi->sync_writers))
887 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
892 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
894 unsigned long bio_flags)
896 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
897 int async = check_async_write(fs_info, BTRFS_I(inode));
900 if (bio_op(bio) != REQ_OP_WRITE) {
902 * called for a read, do the setup so that checksum validation
903 * can happen in the async kernel threads
905 ret = btrfs_bio_wq_end_io(fs_info, bio,
906 BTRFS_WQ_ENDIO_METADATA);
909 ret = btrfs_map_bio(fs_info, bio, mirror_num);
911 ret = btree_csum_one_bio(bio);
914 ret = btrfs_map_bio(fs_info, bio, mirror_num);
917 * kthread helpers are used to submit writes so that
918 * checksumming can happen in parallel across all CPUs
920 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
921 0, inode, btree_submit_bio_start);
929 bio->bi_status = ret;
934 #ifdef CONFIG_MIGRATION
935 static int btree_migratepage(struct address_space *mapping,
936 struct page *newpage, struct page *page,
937 enum migrate_mode mode)
940 * we can't safely write a btree page from here,
941 * we haven't done the locking hook
946 * Buffers may be managed in a filesystem specific way.
947 * We must have no buffers or drop them.
949 if (page_has_private(page) &&
950 !try_to_release_page(page, GFP_KERNEL))
952 return migrate_page(mapping, newpage, page, mode);
957 static int btree_writepages(struct address_space *mapping,
958 struct writeback_control *wbc)
960 struct btrfs_fs_info *fs_info;
963 if (wbc->sync_mode == WB_SYNC_NONE) {
965 if (wbc->for_kupdate)
968 fs_info = BTRFS_I(mapping->host)->root->fs_info;
969 /* this is a bit racy, but that's ok */
970 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
971 BTRFS_DIRTY_METADATA_THRESH,
972 fs_info->dirty_metadata_batch);
976 return btree_write_cache_pages(mapping, wbc);
979 static int btree_readpage(struct file *file, struct page *page)
981 struct extent_io_tree *tree;
982 tree = &BTRFS_I(page->mapping->host)->io_tree;
983 return extent_read_full_page(tree, page, btree_get_extent, 0);
986 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
988 if (PageWriteback(page) || PageDirty(page))
991 return try_release_extent_buffer(page);
994 static void btree_invalidatepage(struct page *page, unsigned int offset,
997 struct extent_io_tree *tree;
998 tree = &BTRFS_I(page->mapping->host)->io_tree;
999 extent_invalidatepage(tree, page, offset);
1000 btree_releasepage(page, GFP_NOFS);
1001 if (PagePrivate(page)) {
1002 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1003 "page private not zero on page %llu",
1004 (unsigned long long)page_offset(page));
1005 ClearPagePrivate(page);
1006 set_page_private(page, 0);
1011 static int btree_set_page_dirty(struct page *page)
1014 struct extent_buffer *eb;
1016 BUG_ON(!PagePrivate(page));
1017 eb = (struct extent_buffer *)page->private;
1019 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1020 BUG_ON(!atomic_read(&eb->refs));
1021 btrfs_assert_tree_locked(eb);
1023 return __set_page_dirty_nobuffers(page);
1026 static const struct address_space_operations btree_aops = {
1027 .readpage = btree_readpage,
1028 .writepages = btree_writepages,
1029 .releasepage = btree_releasepage,
1030 .invalidatepage = btree_invalidatepage,
1031 #ifdef CONFIG_MIGRATION
1032 .migratepage = btree_migratepage,
1034 .set_page_dirty = btree_set_page_dirty,
1037 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1039 struct extent_buffer *buf = NULL;
1042 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1046 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1048 free_extent_buffer_stale(buf);
1050 free_extent_buffer(buf);
1053 struct extent_buffer *btrfs_find_create_tree_block(
1054 struct btrfs_fs_info *fs_info,
1057 if (btrfs_is_testing(fs_info))
1058 return alloc_test_extent_buffer(fs_info, bytenr);
1059 return alloc_extent_buffer(fs_info, bytenr);
1063 * Read tree block at logical address @bytenr and do variant basic but critical
1066 * @parent_transid: expected transid of this tree block, skip check if 0
1067 * @level: expected level, mandatory check
1068 * @first_key: expected key in slot 0, skip check if NULL
1070 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1071 u64 parent_transid, int level,
1072 struct btrfs_key *first_key)
1074 struct extent_buffer *buf = NULL;
1077 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1081 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1084 free_extent_buffer_stale(buf);
1085 return ERR_PTR(ret);
1091 void btrfs_clean_tree_block(struct extent_buffer *buf)
1093 struct btrfs_fs_info *fs_info = buf->fs_info;
1094 if (btrfs_header_generation(buf) ==
1095 fs_info->running_transaction->transid) {
1096 btrfs_assert_tree_locked(buf);
1098 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1099 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1101 fs_info->dirty_metadata_batch);
1102 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1103 btrfs_set_lock_blocking_write(buf);
1104 clear_extent_buffer_dirty(buf);
1109 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1111 struct btrfs_subvolume_writers *writers;
1114 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1116 return ERR_PTR(-ENOMEM);
1118 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1121 return ERR_PTR(ret);
1124 init_waitqueue_head(&writers->wait);
1129 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1131 percpu_counter_destroy(&writers->counter);
1135 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1138 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1139 root->fs_info = fs_info;
1141 root->commit_root = NULL;
1143 root->orphan_cleanup_state = 0;
1145 root->last_trans = 0;
1146 root->highest_objectid = 0;
1147 root->nr_delalloc_inodes = 0;
1148 root->nr_ordered_extents = 0;
1149 root->inode_tree = RB_ROOT;
1150 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1151 root->block_rsv = NULL;
1153 INIT_LIST_HEAD(&root->dirty_list);
1154 INIT_LIST_HEAD(&root->root_list);
1155 INIT_LIST_HEAD(&root->delalloc_inodes);
1156 INIT_LIST_HEAD(&root->delalloc_root);
1157 INIT_LIST_HEAD(&root->ordered_extents);
1158 INIT_LIST_HEAD(&root->ordered_root);
1159 INIT_LIST_HEAD(&root->reloc_dirty_list);
1160 INIT_LIST_HEAD(&root->logged_list[0]);
1161 INIT_LIST_HEAD(&root->logged_list[1]);
1162 spin_lock_init(&root->inode_lock);
1163 spin_lock_init(&root->delalloc_lock);
1164 spin_lock_init(&root->ordered_extent_lock);
1165 spin_lock_init(&root->accounting_lock);
1166 spin_lock_init(&root->log_extents_lock[0]);
1167 spin_lock_init(&root->log_extents_lock[1]);
1168 spin_lock_init(&root->qgroup_meta_rsv_lock);
1169 mutex_init(&root->objectid_mutex);
1170 mutex_init(&root->log_mutex);
1171 mutex_init(&root->ordered_extent_mutex);
1172 mutex_init(&root->delalloc_mutex);
1173 init_waitqueue_head(&root->log_writer_wait);
1174 init_waitqueue_head(&root->log_commit_wait[0]);
1175 init_waitqueue_head(&root->log_commit_wait[1]);
1176 INIT_LIST_HEAD(&root->log_ctxs[0]);
1177 INIT_LIST_HEAD(&root->log_ctxs[1]);
1178 atomic_set(&root->log_commit[0], 0);
1179 atomic_set(&root->log_commit[1], 0);
1180 atomic_set(&root->log_writers, 0);
1181 atomic_set(&root->log_batch, 0);
1182 refcount_set(&root->refs, 1);
1183 atomic_set(&root->will_be_snapshotted, 0);
1184 atomic_set(&root->snapshot_force_cow, 0);
1185 atomic_set(&root->nr_swapfiles, 0);
1186 root->log_transid = 0;
1187 root->log_transid_committed = -1;
1188 root->last_log_commit = 0;
1190 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1191 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1193 memset(&root->root_key, 0, sizeof(root->root_key));
1194 memset(&root->root_item, 0, sizeof(root->root_item));
1195 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1197 root->defrag_trans_start = fs_info->generation;
1199 root->defrag_trans_start = 0;
1200 root->root_key.objectid = objectid;
1203 spin_lock_init(&root->root_item_lock);
1204 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1205 #ifdef CONFIG_BTRFS_DEBUG
1206 INIT_LIST_HEAD(&root->leak_list);
1207 spin_lock(&fs_info->fs_roots_radix_lock);
1208 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1209 spin_unlock(&fs_info->fs_roots_radix_lock);
1213 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1214 u64 objectid, gfp_t flags)
1216 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1218 __setup_root(root, fs_info, objectid);
1222 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1223 /* Should only be used by the testing infrastructure */
1224 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1226 struct btrfs_root *root;
1229 return ERR_PTR(-EINVAL);
1231 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1233 return ERR_PTR(-ENOMEM);
1235 /* We don't use the stripesize in selftest, set it as sectorsize */
1236 root->alloc_bytenr = 0;
1242 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1245 struct btrfs_fs_info *fs_info = trans->fs_info;
1246 struct extent_buffer *leaf;
1247 struct btrfs_root *tree_root = fs_info->tree_root;
1248 struct btrfs_root *root;
1249 struct btrfs_key key;
1250 unsigned int nofs_flag;
1252 uuid_le uuid = NULL_UUID_LE;
1255 * We're holding a transaction handle, so use a NOFS memory allocation
1256 * context to avoid deadlock if reclaim happens.
1258 nofs_flag = memalloc_nofs_save();
1259 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1260 memalloc_nofs_restore(nofs_flag);
1262 return ERR_PTR(-ENOMEM);
1264 root->root_key.objectid = objectid;
1265 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1266 root->root_key.offset = 0;
1268 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1270 ret = PTR_ERR(leaf);
1276 btrfs_mark_buffer_dirty(leaf);
1278 root->commit_root = btrfs_root_node(root);
1279 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1281 root->root_item.flags = 0;
1282 root->root_item.byte_limit = 0;
1283 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1284 btrfs_set_root_generation(&root->root_item, trans->transid);
1285 btrfs_set_root_level(&root->root_item, 0);
1286 btrfs_set_root_refs(&root->root_item, 1);
1287 btrfs_set_root_used(&root->root_item, leaf->len);
1288 btrfs_set_root_last_snapshot(&root->root_item, 0);
1289 btrfs_set_root_dirid(&root->root_item, 0);
1290 if (is_fstree(objectid))
1292 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1293 root->root_item.drop_level = 0;
1295 key.objectid = objectid;
1296 key.type = BTRFS_ROOT_ITEM_KEY;
1298 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1302 btrfs_tree_unlock(leaf);
1308 btrfs_tree_unlock(leaf);
1309 free_extent_buffer(root->commit_root);
1310 free_extent_buffer(leaf);
1312 btrfs_put_fs_root(root);
1314 return ERR_PTR(ret);
1317 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1318 struct btrfs_fs_info *fs_info)
1320 struct btrfs_root *root;
1321 struct extent_buffer *leaf;
1323 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1325 return ERR_PTR(-ENOMEM);
1327 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1328 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1329 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1332 * DON'T set REF_COWS for log trees
1334 * log trees do not get reference counted because they go away
1335 * before a real commit is actually done. They do store pointers
1336 * to file data extents, and those reference counts still get
1337 * updated (along with back refs to the log tree).
1340 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1343 btrfs_put_fs_root(root);
1344 return ERR_CAST(leaf);
1349 btrfs_mark_buffer_dirty(root->node);
1350 btrfs_tree_unlock(root->node);
1354 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1355 struct btrfs_fs_info *fs_info)
1357 struct btrfs_root *log_root;
1359 log_root = alloc_log_tree(trans, fs_info);
1360 if (IS_ERR(log_root))
1361 return PTR_ERR(log_root);
1362 WARN_ON(fs_info->log_root_tree);
1363 fs_info->log_root_tree = log_root;
1367 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1368 struct btrfs_root *root)
1370 struct btrfs_fs_info *fs_info = root->fs_info;
1371 struct btrfs_root *log_root;
1372 struct btrfs_inode_item *inode_item;
1374 log_root = alloc_log_tree(trans, fs_info);
1375 if (IS_ERR(log_root))
1376 return PTR_ERR(log_root);
1378 log_root->last_trans = trans->transid;
1379 log_root->root_key.offset = root->root_key.objectid;
1381 inode_item = &log_root->root_item.inode;
1382 btrfs_set_stack_inode_generation(inode_item, 1);
1383 btrfs_set_stack_inode_size(inode_item, 3);
1384 btrfs_set_stack_inode_nlink(inode_item, 1);
1385 btrfs_set_stack_inode_nbytes(inode_item,
1387 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1389 btrfs_set_root_node(&log_root->root_item, log_root->node);
1391 WARN_ON(root->log_root);
1392 root->log_root = log_root;
1393 root->log_transid = 0;
1394 root->log_transid_committed = -1;
1395 root->last_log_commit = 0;
1399 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1400 struct btrfs_key *key)
1402 struct btrfs_root *root;
1403 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1404 struct btrfs_path *path;
1409 path = btrfs_alloc_path();
1411 return ERR_PTR(-ENOMEM);
1413 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1419 ret = btrfs_find_root(tree_root, key, path,
1420 &root->root_item, &root->root_key);
1427 generation = btrfs_root_generation(&root->root_item);
1428 level = btrfs_root_level(&root->root_item);
1429 root->node = read_tree_block(fs_info,
1430 btrfs_root_bytenr(&root->root_item),
1431 generation, level, NULL);
1432 if (IS_ERR(root->node)) {
1433 ret = PTR_ERR(root->node);
1435 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1437 free_extent_buffer(root->node);
1440 root->commit_root = btrfs_root_node(root);
1442 btrfs_free_path(path);
1446 btrfs_put_fs_root(root);
1448 root = ERR_PTR(ret);
1452 static int btrfs_init_fs_root(struct btrfs_root *root)
1455 struct btrfs_subvolume_writers *writers;
1457 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1458 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1460 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1465 writers = btrfs_alloc_subvolume_writers();
1466 if (IS_ERR(writers)) {
1467 ret = PTR_ERR(writers);
1470 root->subv_writers = writers;
1472 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1473 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1474 btrfs_check_and_init_root_item(&root->root_item);
1477 btrfs_init_free_ino_ctl(root);
1478 spin_lock_init(&root->ino_cache_lock);
1479 init_waitqueue_head(&root->ino_cache_wait);
1481 ret = get_anon_bdev(&root->anon_dev);
1485 mutex_lock(&root->objectid_mutex);
1486 ret = btrfs_find_highest_objectid(root,
1487 &root->highest_objectid);
1489 mutex_unlock(&root->objectid_mutex);
1493 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1495 mutex_unlock(&root->objectid_mutex);
1499 /* The caller is responsible to call btrfs_free_fs_root */
1503 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1506 struct btrfs_root *root;
1508 spin_lock(&fs_info->fs_roots_radix_lock);
1509 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1510 (unsigned long)root_id);
1512 root = btrfs_grab_fs_root(root);
1513 spin_unlock(&fs_info->fs_roots_radix_lock);
1517 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1518 struct btrfs_root *root)
1522 ret = radix_tree_preload(GFP_NOFS);
1526 spin_lock(&fs_info->fs_roots_radix_lock);
1527 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1528 (unsigned long)root->root_key.objectid,
1531 btrfs_grab_fs_root(root);
1532 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1534 spin_unlock(&fs_info->fs_roots_radix_lock);
1535 radix_tree_preload_end();
1540 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1542 #ifdef CONFIG_BTRFS_DEBUG
1543 struct btrfs_root *root;
1545 while (!list_empty(&fs_info->allocated_roots)) {
1546 root = list_first_entry(&fs_info->allocated_roots,
1547 struct btrfs_root, leak_list);
1548 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1549 root->root_key.objectid, root->root_key.offset,
1550 refcount_read(&root->refs));
1551 while (refcount_read(&root->refs) > 1)
1552 btrfs_put_fs_root(root);
1553 btrfs_put_fs_root(root);
1558 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1560 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1561 percpu_counter_destroy(&fs_info->delalloc_bytes);
1562 percpu_counter_destroy(&fs_info->dio_bytes);
1563 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1564 btrfs_free_csum_hash(fs_info);
1565 btrfs_free_stripe_hash_table(fs_info);
1566 btrfs_free_ref_cache(fs_info);
1567 kfree(fs_info->balance_ctl);
1568 kfree(fs_info->delayed_root);
1569 btrfs_put_fs_root(fs_info->extent_root);
1570 btrfs_put_fs_root(fs_info->tree_root);
1571 btrfs_put_fs_root(fs_info->chunk_root);
1572 btrfs_put_fs_root(fs_info->dev_root);
1573 btrfs_put_fs_root(fs_info->csum_root);
1574 btrfs_put_fs_root(fs_info->quota_root);
1575 btrfs_put_fs_root(fs_info->uuid_root);
1576 btrfs_put_fs_root(fs_info->free_space_root);
1577 btrfs_put_fs_root(fs_info->fs_root);
1578 btrfs_check_leaked_roots(fs_info);
1579 kfree(fs_info->super_copy);
1580 kfree(fs_info->super_for_commit);
1585 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1586 struct btrfs_key *location,
1589 struct btrfs_root *root;
1590 struct btrfs_path *path;
1591 struct btrfs_key key;
1594 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1595 return btrfs_grab_fs_root(fs_info->tree_root);
1596 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1597 return btrfs_grab_fs_root(fs_info->extent_root);
1598 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1599 return btrfs_grab_fs_root(fs_info->chunk_root);
1600 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1601 return btrfs_grab_fs_root(fs_info->dev_root);
1602 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1603 return btrfs_grab_fs_root(fs_info->csum_root);
1604 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1605 return btrfs_grab_fs_root(fs_info->quota_root) ?
1606 fs_info->quota_root : ERR_PTR(-ENOENT);
1607 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1608 return btrfs_grab_fs_root(fs_info->uuid_root) ?
1609 fs_info->uuid_root : ERR_PTR(-ENOENT);
1610 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1611 return btrfs_grab_fs_root(fs_info->free_space_root) ?
1612 fs_info->free_space_root : ERR_PTR(-ENOENT);
1614 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1616 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1617 btrfs_put_fs_root(root);
1618 return ERR_PTR(-ENOENT);
1623 root = btrfs_read_tree_root(fs_info->tree_root, location);
1627 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1632 ret = btrfs_init_fs_root(root);
1636 path = btrfs_alloc_path();
1641 key.objectid = BTRFS_ORPHAN_OBJECTID;
1642 key.type = BTRFS_ORPHAN_ITEM_KEY;
1643 key.offset = location->objectid;
1645 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1646 btrfs_free_path(path);
1650 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1653 * All roots have two refs on them at all times, one for the mounted fs,
1654 * and one for being in the radix tree. This way we only free the root
1655 * when we are unmounting or deleting the subvolume. We get one ref
1656 * from __setup_root, one for inserting it into the radix tree, and then
1657 * we have the third for returning it, and the caller will put it when
1658 * it's done with the root.
1660 btrfs_grab_fs_root(root);
1661 ret = btrfs_insert_fs_root(fs_info, root);
1663 btrfs_put_fs_root(root);
1664 if (ret == -EEXIST) {
1665 btrfs_free_fs_root(root);
1672 btrfs_free_fs_root(root);
1673 return ERR_PTR(ret);
1676 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1678 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1680 struct btrfs_device *device;
1681 struct backing_dev_info *bdi;
1684 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1687 bdi = device->bdev->bd_bdi;
1688 if (bdi_congested(bdi, bdi_bits)) {
1698 * called by the kthread helper functions to finally call the bio end_io
1699 * functions. This is where read checksum verification actually happens
1701 static void end_workqueue_fn(struct btrfs_work *work)
1704 struct btrfs_end_io_wq *end_io_wq;
1706 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1707 bio = end_io_wq->bio;
1709 bio->bi_status = end_io_wq->status;
1710 bio->bi_private = end_io_wq->private;
1711 bio->bi_end_io = end_io_wq->end_io;
1713 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1716 static int cleaner_kthread(void *arg)
1718 struct btrfs_root *root = arg;
1719 struct btrfs_fs_info *fs_info = root->fs_info;
1725 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1727 /* Make the cleaner go to sleep early. */
1728 if (btrfs_need_cleaner_sleep(fs_info))
1732 * Do not do anything if we might cause open_ctree() to block
1733 * before we have finished mounting the filesystem.
1735 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1738 if (!mutex_trylock(&fs_info->cleaner_mutex))
1742 * Avoid the problem that we change the status of the fs
1743 * during the above check and trylock.
1745 if (btrfs_need_cleaner_sleep(fs_info)) {
1746 mutex_unlock(&fs_info->cleaner_mutex);
1750 btrfs_run_delayed_iputs(fs_info);
1752 again = btrfs_clean_one_deleted_snapshot(root);
1753 mutex_unlock(&fs_info->cleaner_mutex);
1756 * The defragger has dealt with the R/O remount and umount,
1757 * needn't do anything special here.
1759 btrfs_run_defrag_inodes(fs_info);
1762 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1763 * with relocation (btrfs_relocate_chunk) and relocation
1764 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1765 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1766 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1767 * unused block groups.
1769 btrfs_delete_unused_bgs(fs_info);
1771 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1772 if (kthread_should_park())
1774 if (kthread_should_stop())
1777 set_current_state(TASK_INTERRUPTIBLE);
1779 __set_current_state(TASK_RUNNING);
1784 static int transaction_kthread(void *arg)
1786 struct btrfs_root *root = arg;
1787 struct btrfs_fs_info *fs_info = root->fs_info;
1788 struct btrfs_trans_handle *trans;
1789 struct btrfs_transaction *cur;
1792 unsigned long delay;
1796 cannot_commit = false;
1797 delay = HZ * fs_info->commit_interval;
1798 mutex_lock(&fs_info->transaction_kthread_mutex);
1800 spin_lock(&fs_info->trans_lock);
1801 cur = fs_info->running_transaction;
1803 spin_unlock(&fs_info->trans_lock);
1807 now = ktime_get_seconds();
1808 if (cur->state < TRANS_STATE_COMMIT_START &&
1809 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1810 (now < cur->start_time ||
1811 now - cur->start_time < fs_info->commit_interval)) {
1812 spin_unlock(&fs_info->trans_lock);
1816 transid = cur->transid;
1817 spin_unlock(&fs_info->trans_lock);
1819 /* If the file system is aborted, this will always fail. */
1820 trans = btrfs_attach_transaction(root);
1821 if (IS_ERR(trans)) {
1822 if (PTR_ERR(trans) != -ENOENT)
1823 cannot_commit = true;
1826 if (transid == trans->transid) {
1827 btrfs_commit_transaction(trans);
1829 btrfs_end_transaction(trans);
1832 wake_up_process(fs_info->cleaner_kthread);
1833 mutex_unlock(&fs_info->transaction_kthread_mutex);
1835 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1836 &fs_info->fs_state)))
1837 btrfs_cleanup_transaction(fs_info);
1838 if (!kthread_should_stop() &&
1839 (!btrfs_transaction_blocked(fs_info) ||
1841 schedule_timeout_interruptible(delay);
1842 } while (!kthread_should_stop());
1847 * This will find the highest generation in the array of root backups. The
1848 * index of the highest array is returned, or -EINVAL if we can't find
1851 * We check to make sure the array is valid by comparing the
1852 * generation of the latest root in the array with the generation
1853 * in the super block. If they don't match we pitch it.
1855 static int find_newest_super_backup(struct btrfs_fs_info *info)
1857 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1859 struct btrfs_root_backup *root_backup;
1862 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1863 root_backup = info->super_copy->super_roots + i;
1864 cur = btrfs_backup_tree_root_gen(root_backup);
1865 if (cur == newest_gen)
1873 * copy all the root pointers into the super backup array.
1874 * this will bump the backup pointer by one when it is
1877 static void backup_super_roots(struct btrfs_fs_info *info)
1879 const int next_backup = info->backup_root_index;
1880 struct btrfs_root_backup *root_backup;
1882 root_backup = info->super_for_commit->super_roots + next_backup;
1885 * make sure all of our padding and empty slots get zero filled
1886 * regardless of which ones we use today
1888 memset(root_backup, 0, sizeof(*root_backup));
1890 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1892 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1893 btrfs_set_backup_tree_root_gen(root_backup,
1894 btrfs_header_generation(info->tree_root->node));
1896 btrfs_set_backup_tree_root_level(root_backup,
1897 btrfs_header_level(info->tree_root->node));
1899 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1900 btrfs_set_backup_chunk_root_gen(root_backup,
1901 btrfs_header_generation(info->chunk_root->node));
1902 btrfs_set_backup_chunk_root_level(root_backup,
1903 btrfs_header_level(info->chunk_root->node));
1905 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1906 btrfs_set_backup_extent_root_gen(root_backup,
1907 btrfs_header_generation(info->extent_root->node));
1908 btrfs_set_backup_extent_root_level(root_backup,
1909 btrfs_header_level(info->extent_root->node));
1912 * we might commit during log recovery, which happens before we set
1913 * the fs_root. Make sure it is valid before we fill it in.
1915 if (info->fs_root && info->fs_root->node) {
1916 btrfs_set_backup_fs_root(root_backup,
1917 info->fs_root->node->start);
1918 btrfs_set_backup_fs_root_gen(root_backup,
1919 btrfs_header_generation(info->fs_root->node));
1920 btrfs_set_backup_fs_root_level(root_backup,
1921 btrfs_header_level(info->fs_root->node));
1924 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1925 btrfs_set_backup_dev_root_gen(root_backup,
1926 btrfs_header_generation(info->dev_root->node));
1927 btrfs_set_backup_dev_root_level(root_backup,
1928 btrfs_header_level(info->dev_root->node));
1930 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1931 btrfs_set_backup_csum_root_gen(root_backup,
1932 btrfs_header_generation(info->csum_root->node));
1933 btrfs_set_backup_csum_root_level(root_backup,
1934 btrfs_header_level(info->csum_root->node));
1936 btrfs_set_backup_total_bytes(root_backup,
1937 btrfs_super_total_bytes(info->super_copy));
1938 btrfs_set_backup_bytes_used(root_backup,
1939 btrfs_super_bytes_used(info->super_copy));
1940 btrfs_set_backup_num_devices(root_backup,
1941 btrfs_super_num_devices(info->super_copy));
1944 * if we don't copy this out to the super_copy, it won't get remembered
1945 * for the next commit
1947 memcpy(&info->super_copy->super_roots,
1948 &info->super_for_commit->super_roots,
1949 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1953 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1954 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1956 * fs_info - filesystem whose backup roots need to be read
1957 * priority - priority of backup root required
1959 * Returns backup root index on success and -EINVAL otherwise.
1961 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1963 int backup_index = find_newest_super_backup(fs_info);
1964 struct btrfs_super_block *super = fs_info->super_copy;
1965 struct btrfs_root_backup *root_backup;
1967 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1969 return backup_index;
1971 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1972 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1977 root_backup = super->super_roots + backup_index;
1979 btrfs_set_super_generation(super,
1980 btrfs_backup_tree_root_gen(root_backup));
1981 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1982 btrfs_set_super_root_level(super,
1983 btrfs_backup_tree_root_level(root_backup));
1984 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1987 * Fixme: the total bytes and num_devices need to match or we should
1990 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1991 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1993 return backup_index;
1996 /* helper to cleanup workers */
1997 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1999 btrfs_destroy_workqueue(fs_info->fixup_workers);
2000 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2001 btrfs_destroy_workqueue(fs_info->workers);
2002 btrfs_destroy_workqueue(fs_info->endio_workers);
2003 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2004 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2005 btrfs_destroy_workqueue(fs_info->rmw_workers);
2006 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2007 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2008 btrfs_destroy_workqueue(fs_info->delayed_workers);
2009 btrfs_destroy_workqueue(fs_info->caching_workers);
2010 btrfs_destroy_workqueue(fs_info->readahead_workers);
2011 btrfs_destroy_workqueue(fs_info->flush_workers);
2012 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2013 if (fs_info->discard_ctl.discard_workers)
2014 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2016 * Now that all other work queues are destroyed, we can safely destroy
2017 * the queues used for metadata I/O, since tasks from those other work
2018 * queues can do metadata I/O operations.
2020 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2021 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2024 static void free_root_extent_buffers(struct btrfs_root *root)
2027 free_extent_buffer(root->node);
2028 free_extent_buffer(root->commit_root);
2030 root->commit_root = NULL;
2034 /* helper to cleanup tree roots */
2035 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2037 free_root_extent_buffers(info->tree_root);
2039 free_root_extent_buffers(info->dev_root);
2040 free_root_extent_buffers(info->extent_root);
2041 free_root_extent_buffers(info->csum_root);
2042 free_root_extent_buffers(info->quota_root);
2043 free_root_extent_buffers(info->uuid_root);
2044 if (free_chunk_root)
2045 free_root_extent_buffers(info->chunk_root);
2046 free_root_extent_buffers(info->free_space_root);
2049 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2052 struct btrfs_root *gang[8];
2055 while (!list_empty(&fs_info->dead_roots)) {
2056 gang[0] = list_entry(fs_info->dead_roots.next,
2057 struct btrfs_root, root_list);
2058 list_del(&gang[0]->root_list);
2060 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2061 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2063 free_extent_buffer(gang[0]->node);
2064 free_extent_buffer(gang[0]->commit_root);
2065 btrfs_put_fs_root(gang[0]);
2070 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2075 for (i = 0; i < ret; i++)
2076 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2079 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2080 btrfs_free_log_root_tree(NULL, fs_info);
2081 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2085 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2087 mutex_init(&fs_info->scrub_lock);
2088 atomic_set(&fs_info->scrubs_running, 0);
2089 atomic_set(&fs_info->scrub_pause_req, 0);
2090 atomic_set(&fs_info->scrubs_paused, 0);
2091 atomic_set(&fs_info->scrub_cancel_req, 0);
2092 init_waitqueue_head(&fs_info->scrub_pause_wait);
2093 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2096 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2098 spin_lock_init(&fs_info->balance_lock);
2099 mutex_init(&fs_info->balance_mutex);
2100 atomic_set(&fs_info->balance_pause_req, 0);
2101 atomic_set(&fs_info->balance_cancel_req, 0);
2102 fs_info->balance_ctl = NULL;
2103 init_waitqueue_head(&fs_info->balance_wait_q);
2106 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2108 struct inode *inode = fs_info->btree_inode;
2110 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2111 set_nlink(inode, 1);
2113 * we set the i_size on the btree inode to the max possible int.
2114 * the real end of the address space is determined by all of
2115 * the devices in the system
2117 inode->i_size = OFFSET_MAX;
2118 inode->i_mapping->a_ops = &btree_aops;
2120 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2121 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2122 IO_TREE_INODE_IO, inode);
2123 BTRFS_I(inode)->io_tree.track_uptodate = false;
2124 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2126 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2128 BTRFS_I(inode)->root = fs_info->tree_root;
2129 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2130 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2131 btrfs_insert_inode_hash(inode);
2134 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2136 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2137 init_rwsem(&fs_info->dev_replace.rwsem);
2138 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2141 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2143 spin_lock_init(&fs_info->qgroup_lock);
2144 mutex_init(&fs_info->qgroup_ioctl_lock);
2145 fs_info->qgroup_tree = RB_ROOT;
2146 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2147 fs_info->qgroup_seq = 1;
2148 fs_info->qgroup_ulist = NULL;
2149 fs_info->qgroup_rescan_running = false;
2150 mutex_init(&fs_info->qgroup_rescan_lock);
2153 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2154 struct btrfs_fs_devices *fs_devices)
2156 u32 max_active = fs_info->thread_pool_size;
2157 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2160 btrfs_alloc_workqueue(fs_info, "worker",
2161 flags | WQ_HIGHPRI, max_active, 16);
2163 fs_info->delalloc_workers =
2164 btrfs_alloc_workqueue(fs_info, "delalloc",
2165 flags, max_active, 2);
2167 fs_info->flush_workers =
2168 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2169 flags, max_active, 0);
2171 fs_info->caching_workers =
2172 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2174 fs_info->fixup_workers =
2175 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2178 * endios are largely parallel and should have a very
2181 fs_info->endio_workers =
2182 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2183 fs_info->endio_meta_workers =
2184 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2186 fs_info->endio_meta_write_workers =
2187 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2189 fs_info->endio_raid56_workers =
2190 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2192 fs_info->endio_repair_workers =
2193 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2194 fs_info->rmw_workers =
2195 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2196 fs_info->endio_write_workers =
2197 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2199 fs_info->endio_freespace_worker =
2200 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2202 fs_info->delayed_workers =
2203 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2205 fs_info->readahead_workers =
2206 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2208 fs_info->qgroup_rescan_workers =
2209 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2210 fs_info->discard_ctl.discard_workers =
2211 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2213 if (!(fs_info->workers && fs_info->delalloc_workers &&
2214 fs_info->flush_workers &&
2215 fs_info->endio_workers && fs_info->endio_meta_workers &&
2216 fs_info->endio_meta_write_workers &&
2217 fs_info->endio_repair_workers &&
2218 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2219 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2220 fs_info->caching_workers && fs_info->readahead_workers &&
2221 fs_info->fixup_workers && fs_info->delayed_workers &&
2222 fs_info->qgroup_rescan_workers &&
2223 fs_info->discard_ctl.discard_workers)) {
2230 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2232 struct crypto_shash *csum_shash;
2233 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2235 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2237 if (IS_ERR(csum_shash)) {
2238 btrfs_err(fs_info, "error allocating %s hash for checksum",
2240 return PTR_ERR(csum_shash);
2243 fs_info->csum_shash = csum_shash;
2248 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2249 struct btrfs_fs_devices *fs_devices)
2252 struct btrfs_root *log_tree_root;
2253 struct btrfs_super_block *disk_super = fs_info->super_copy;
2254 u64 bytenr = btrfs_super_log_root(disk_super);
2255 int level = btrfs_super_log_root_level(disk_super);
2257 if (fs_devices->rw_devices == 0) {
2258 btrfs_warn(fs_info, "log replay required on RO media");
2262 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2267 log_tree_root->node = read_tree_block(fs_info, bytenr,
2268 fs_info->generation + 1,
2270 if (IS_ERR(log_tree_root->node)) {
2271 btrfs_warn(fs_info, "failed to read log tree");
2272 ret = PTR_ERR(log_tree_root->node);
2273 btrfs_put_fs_root(log_tree_root);
2275 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2276 btrfs_err(fs_info, "failed to read log tree");
2277 free_extent_buffer(log_tree_root->node);
2278 btrfs_put_fs_root(log_tree_root);
2281 /* returns with log_tree_root freed on success */
2282 ret = btrfs_recover_log_trees(log_tree_root);
2284 btrfs_handle_fs_error(fs_info, ret,
2285 "Failed to recover log tree");
2286 free_extent_buffer(log_tree_root->node);
2287 btrfs_put_fs_root(log_tree_root);
2291 if (sb_rdonly(fs_info->sb)) {
2292 ret = btrfs_commit_super(fs_info);
2300 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2302 struct btrfs_root *tree_root = fs_info->tree_root;
2303 struct btrfs_root *root;
2304 struct btrfs_key location;
2307 BUG_ON(!fs_info->tree_root);
2309 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2310 location.type = BTRFS_ROOT_ITEM_KEY;
2311 location.offset = 0;
2313 root = btrfs_read_tree_root(tree_root, &location);
2315 ret = PTR_ERR(root);
2318 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2319 fs_info->extent_root = root;
2321 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2322 root = btrfs_read_tree_root(tree_root, &location);
2324 ret = PTR_ERR(root);
2327 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2328 fs_info->dev_root = root;
2329 btrfs_init_devices_late(fs_info);
2331 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2332 root = btrfs_read_tree_root(tree_root, &location);
2334 ret = PTR_ERR(root);
2337 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2338 fs_info->csum_root = root;
2340 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2341 root = btrfs_read_tree_root(tree_root, &location);
2342 if (!IS_ERR(root)) {
2343 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2344 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2345 fs_info->quota_root = root;
2348 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2349 root = btrfs_read_tree_root(tree_root, &location);
2351 ret = PTR_ERR(root);
2355 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2356 fs_info->uuid_root = root;
2359 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2360 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2361 root = btrfs_read_tree_root(tree_root, &location);
2363 ret = PTR_ERR(root);
2366 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2367 fs_info->free_space_root = root;
2372 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2373 location.objectid, ret);
2378 * Real super block validation
2379 * NOTE: super csum type and incompat features will not be checked here.
2381 * @sb: super block to check
2382 * @mirror_num: the super block number to check its bytenr:
2383 * 0 the primary (1st) sb
2384 * 1, 2 2nd and 3rd backup copy
2385 * -1 skip bytenr check
2387 static int validate_super(struct btrfs_fs_info *fs_info,
2388 struct btrfs_super_block *sb, int mirror_num)
2390 u64 nodesize = btrfs_super_nodesize(sb);
2391 u64 sectorsize = btrfs_super_sectorsize(sb);
2394 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2395 btrfs_err(fs_info, "no valid FS found");
2398 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2399 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2400 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2403 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2404 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2405 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2408 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2409 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2410 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2413 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2414 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2415 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2420 * Check sectorsize and nodesize first, other check will need it.
2421 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2423 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2424 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2425 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2428 /* Only PAGE SIZE is supported yet */
2429 if (sectorsize != PAGE_SIZE) {
2431 "sectorsize %llu not supported yet, only support %lu",
2432 sectorsize, PAGE_SIZE);
2435 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2436 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2437 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2440 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2441 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2442 le32_to_cpu(sb->__unused_leafsize), nodesize);
2446 /* Root alignment check */
2447 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2448 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2449 btrfs_super_root(sb));
2452 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2453 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2454 btrfs_super_chunk_root(sb));
2457 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2458 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2459 btrfs_super_log_root(sb));
2463 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2464 BTRFS_FSID_SIZE) != 0) {
2466 "dev_item UUID does not match metadata fsid: %pU != %pU",
2467 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2472 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2475 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2476 btrfs_err(fs_info, "bytes_used is too small %llu",
2477 btrfs_super_bytes_used(sb));
2480 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2481 btrfs_err(fs_info, "invalid stripesize %u",
2482 btrfs_super_stripesize(sb));
2485 if (btrfs_super_num_devices(sb) > (1UL << 31))
2486 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2487 btrfs_super_num_devices(sb));
2488 if (btrfs_super_num_devices(sb) == 0) {
2489 btrfs_err(fs_info, "number of devices is 0");
2493 if (mirror_num >= 0 &&
2494 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2495 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2496 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2501 * Obvious sys_chunk_array corruptions, it must hold at least one key
2504 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2505 btrfs_err(fs_info, "system chunk array too big %u > %u",
2506 btrfs_super_sys_array_size(sb),
2507 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2510 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2511 + sizeof(struct btrfs_chunk)) {
2512 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2513 btrfs_super_sys_array_size(sb),
2514 sizeof(struct btrfs_disk_key)
2515 + sizeof(struct btrfs_chunk));
2520 * The generation is a global counter, we'll trust it more than the others
2521 * but it's still possible that it's the one that's wrong.
2523 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2525 "suspicious: generation < chunk_root_generation: %llu < %llu",
2526 btrfs_super_generation(sb),
2527 btrfs_super_chunk_root_generation(sb));
2528 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2529 && btrfs_super_cache_generation(sb) != (u64)-1)
2531 "suspicious: generation < cache_generation: %llu < %llu",
2532 btrfs_super_generation(sb),
2533 btrfs_super_cache_generation(sb));
2539 * Validation of super block at mount time.
2540 * Some checks already done early at mount time, like csum type and incompat
2541 * flags will be skipped.
2543 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2545 return validate_super(fs_info, fs_info->super_copy, 0);
2549 * Validation of super block at write time.
2550 * Some checks like bytenr check will be skipped as their values will be
2552 * Extra checks like csum type and incompat flags will be done here.
2554 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2555 struct btrfs_super_block *sb)
2559 ret = validate_super(fs_info, sb, -1);
2562 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2564 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2565 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2568 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2571 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2572 btrfs_super_incompat_flags(sb),
2573 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2579 "super block corruption detected before writing it to disk");
2583 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2585 int backup_index = find_newest_super_backup(fs_info);
2586 struct btrfs_super_block *sb = fs_info->super_copy;
2587 struct btrfs_root *tree_root = fs_info->tree_root;
2588 bool handle_error = false;
2592 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2597 if (!IS_ERR(tree_root->node))
2598 free_extent_buffer(tree_root->node);
2599 tree_root->node = NULL;
2601 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2604 free_root_pointers(fs_info, 0);
2607 * Don't use the log in recovery mode, it won't be
2610 btrfs_set_super_log_root(sb, 0);
2612 /* We can't trust the free space cache either */
2613 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2615 ret = read_backup_root(fs_info, i);
2620 generation = btrfs_super_generation(sb);
2621 level = btrfs_super_root_level(sb);
2622 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2623 generation, level, NULL);
2624 if (IS_ERR(tree_root->node) ||
2625 !extent_buffer_uptodate(tree_root->node)) {
2626 handle_error = true;
2628 if (IS_ERR(tree_root->node))
2629 ret = PTR_ERR(tree_root->node);
2630 else if (!extent_buffer_uptodate(tree_root->node))
2633 btrfs_warn(fs_info, "failed to read tree root");
2637 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2638 tree_root->commit_root = btrfs_root_node(tree_root);
2639 btrfs_set_root_refs(&tree_root->root_item, 1);
2642 * No need to hold btrfs_root::objectid_mutex since the fs
2643 * hasn't been fully initialised and we are the only user
2645 ret = btrfs_find_highest_objectid(tree_root,
2646 &tree_root->highest_objectid);
2648 handle_error = true;
2652 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2654 ret = btrfs_read_roots(fs_info);
2656 handle_error = true;
2660 /* All successful */
2661 fs_info->generation = generation;
2662 fs_info->last_trans_committed = generation;
2664 /* Always begin writing backup roots after the one being used */
2665 if (backup_index < 0) {
2666 fs_info->backup_root_index = 0;
2668 fs_info->backup_root_index = backup_index + 1;
2669 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2677 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2679 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2680 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2681 INIT_LIST_HEAD(&fs_info->trans_list);
2682 INIT_LIST_HEAD(&fs_info->dead_roots);
2683 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2684 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2685 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2686 spin_lock_init(&fs_info->delalloc_root_lock);
2687 spin_lock_init(&fs_info->trans_lock);
2688 spin_lock_init(&fs_info->fs_roots_radix_lock);
2689 spin_lock_init(&fs_info->delayed_iput_lock);
2690 spin_lock_init(&fs_info->defrag_inodes_lock);
2691 spin_lock_init(&fs_info->super_lock);
2692 spin_lock_init(&fs_info->buffer_lock);
2693 spin_lock_init(&fs_info->unused_bgs_lock);
2694 rwlock_init(&fs_info->tree_mod_log_lock);
2695 mutex_init(&fs_info->unused_bg_unpin_mutex);
2696 mutex_init(&fs_info->delete_unused_bgs_mutex);
2697 mutex_init(&fs_info->reloc_mutex);
2698 mutex_init(&fs_info->delalloc_root_mutex);
2699 seqlock_init(&fs_info->profiles_lock);
2701 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2702 INIT_LIST_HEAD(&fs_info->space_info);
2703 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2704 INIT_LIST_HEAD(&fs_info->unused_bgs);
2705 #ifdef CONFIG_BTRFS_DEBUG
2706 INIT_LIST_HEAD(&fs_info->allocated_roots);
2708 extent_map_tree_init(&fs_info->mapping_tree);
2709 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2710 BTRFS_BLOCK_RSV_GLOBAL);
2711 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2712 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2713 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2714 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2715 BTRFS_BLOCK_RSV_DELOPS);
2716 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2717 BTRFS_BLOCK_RSV_DELREFS);
2719 atomic_set(&fs_info->async_delalloc_pages, 0);
2720 atomic_set(&fs_info->defrag_running, 0);
2721 atomic_set(&fs_info->reada_works_cnt, 0);
2722 atomic_set(&fs_info->nr_delayed_iputs, 0);
2723 atomic64_set(&fs_info->tree_mod_seq, 0);
2724 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2725 fs_info->metadata_ratio = 0;
2726 fs_info->defrag_inodes = RB_ROOT;
2727 atomic64_set(&fs_info->free_chunk_space, 0);
2728 fs_info->tree_mod_log = RB_ROOT;
2729 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2730 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2731 /* readahead state */
2732 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2733 spin_lock_init(&fs_info->reada_lock);
2734 btrfs_init_ref_verify(fs_info);
2736 fs_info->thread_pool_size = min_t(unsigned long,
2737 num_online_cpus() + 2, 8);
2739 INIT_LIST_HEAD(&fs_info->ordered_roots);
2740 spin_lock_init(&fs_info->ordered_root_lock);
2742 btrfs_init_scrub(fs_info);
2743 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2744 fs_info->check_integrity_print_mask = 0;
2746 btrfs_init_balance(fs_info);
2747 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2749 spin_lock_init(&fs_info->block_group_cache_lock);
2750 fs_info->block_group_cache_tree = RB_ROOT;
2751 fs_info->first_logical_byte = (u64)-1;
2753 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2754 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2755 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2756 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2757 fs_info->pinned_extents = &fs_info->freed_extents[0];
2758 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2760 mutex_init(&fs_info->ordered_operations_mutex);
2761 mutex_init(&fs_info->tree_log_mutex);
2762 mutex_init(&fs_info->chunk_mutex);
2763 mutex_init(&fs_info->transaction_kthread_mutex);
2764 mutex_init(&fs_info->cleaner_mutex);
2765 mutex_init(&fs_info->ro_block_group_mutex);
2766 init_rwsem(&fs_info->commit_root_sem);
2767 init_rwsem(&fs_info->cleanup_work_sem);
2768 init_rwsem(&fs_info->subvol_sem);
2769 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2771 btrfs_init_dev_replace_locks(fs_info);
2772 btrfs_init_qgroup(fs_info);
2773 btrfs_discard_init(fs_info);
2775 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2776 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2778 init_waitqueue_head(&fs_info->transaction_throttle);
2779 init_waitqueue_head(&fs_info->transaction_wait);
2780 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2781 init_waitqueue_head(&fs_info->async_submit_wait);
2782 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2784 /* Usable values until the real ones are cached from the superblock */
2785 fs_info->nodesize = 4096;
2786 fs_info->sectorsize = 4096;
2787 fs_info->stripesize = 4096;
2789 spin_lock_init(&fs_info->swapfile_pins_lock);
2790 fs_info->swapfile_pins = RB_ROOT;
2792 fs_info->send_in_progress = 0;
2795 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2800 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2801 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2803 ret = init_srcu_struct(&fs_info->subvol_srcu);
2807 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2811 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2815 fs_info->dirty_metadata_batch = PAGE_SIZE *
2816 (1 + ilog2(nr_cpu_ids));
2818 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2822 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2827 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2829 if (!fs_info->delayed_root) {
2833 btrfs_init_delayed_root(fs_info->delayed_root);
2835 ret = btrfs_alloc_stripe_hash_table(fs_info);
2841 cleanup_srcu_struct(&fs_info->subvol_srcu);
2845 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2854 struct btrfs_key location;
2855 struct buffer_head *bh;
2856 struct btrfs_super_block *disk_super;
2857 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2858 struct btrfs_root *tree_root;
2859 struct btrfs_root *chunk_root;
2862 int clear_free_space_tree = 0;
2865 ret = init_mount_fs_info(fs_info, sb);
2871 /* These need to be init'ed before we start creating inodes and such. */
2872 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2874 fs_info->tree_root = tree_root;
2875 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2877 fs_info->chunk_root = chunk_root;
2878 if (!tree_root || !chunk_root) {
2883 fs_info->btree_inode = new_inode(sb);
2884 if (!fs_info->btree_inode) {
2888 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2889 btrfs_init_btree_inode(fs_info);
2891 invalidate_bdev(fs_devices->latest_bdev);
2894 * Read super block and check the signature bytes only
2896 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2903 * Verify the type first, if that or the the checksum value are
2904 * corrupted, we'll find out
2906 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2907 if (!btrfs_supported_super_csum(csum_type)) {
2908 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2915 ret = btrfs_init_csum_hash(fs_info, csum_type);
2922 * We want to check superblock checksum, the type is stored inside.
2923 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2925 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2926 btrfs_err(fs_info, "superblock checksum mismatch");
2933 * super_copy is zeroed at allocation time and we never touch the
2934 * following bytes up to INFO_SIZE, the checksum is calculated from
2935 * the whole block of INFO_SIZE
2937 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2940 disk_super = fs_info->super_copy;
2942 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2945 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2946 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2947 fs_info->super_copy->metadata_uuid,
2951 features = btrfs_super_flags(disk_super);
2952 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2953 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2954 btrfs_set_super_flags(disk_super, features);
2956 "found metadata UUID change in progress flag, clearing");
2959 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2960 sizeof(*fs_info->super_for_commit));
2962 ret = btrfs_validate_mount_super(fs_info);
2964 btrfs_err(fs_info, "superblock contains fatal errors");
2969 if (!btrfs_super_root(disk_super))
2972 /* check FS state, whether FS is broken. */
2973 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2974 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2977 * In the long term, we'll store the compression type in the super
2978 * block, and it'll be used for per file compression control.
2980 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2982 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2988 features = btrfs_super_incompat_flags(disk_super) &
2989 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2992 "cannot mount because of unsupported optional features (%llx)",
2998 features = btrfs_super_incompat_flags(disk_super);
2999 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3000 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3001 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3002 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3003 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3005 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3006 btrfs_info(fs_info, "has skinny extents");
3009 * flag our filesystem as having big metadata blocks if
3010 * they are bigger than the page size
3012 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3013 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3015 "flagging fs with big metadata feature");
3016 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3019 nodesize = btrfs_super_nodesize(disk_super);
3020 sectorsize = btrfs_super_sectorsize(disk_super);
3021 stripesize = sectorsize;
3022 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3023 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3025 /* Cache block sizes */
3026 fs_info->nodesize = nodesize;
3027 fs_info->sectorsize = sectorsize;
3028 fs_info->stripesize = stripesize;
3031 * mixed block groups end up with duplicate but slightly offset
3032 * extent buffers for the same range. It leads to corruptions
3034 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3035 (sectorsize != nodesize)) {
3037 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3038 nodesize, sectorsize);
3043 * Needn't use the lock because there is no other task which will
3046 btrfs_set_super_incompat_flags(disk_super, features);
3048 features = btrfs_super_compat_ro_flags(disk_super) &
3049 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3050 if (!sb_rdonly(sb) && features) {
3052 "cannot mount read-write because of unsupported optional features (%llx)",
3058 ret = btrfs_init_workqueues(fs_info, fs_devices);
3061 goto fail_sb_buffer;
3064 sb->s_bdi->congested_fn = btrfs_congested_fn;
3065 sb->s_bdi->congested_data = fs_info;
3066 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3067 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3068 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3069 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3071 sb->s_blocksize = sectorsize;
3072 sb->s_blocksize_bits = blksize_bits(sectorsize);
3073 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3075 mutex_lock(&fs_info->chunk_mutex);
3076 ret = btrfs_read_sys_array(fs_info);
3077 mutex_unlock(&fs_info->chunk_mutex);
3079 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3080 goto fail_sb_buffer;
3083 generation = btrfs_super_chunk_root_generation(disk_super);
3084 level = btrfs_super_chunk_root_level(disk_super);
3086 chunk_root->node = read_tree_block(fs_info,
3087 btrfs_super_chunk_root(disk_super),
3088 generation, level, NULL);
3089 if (IS_ERR(chunk_root->node) ||
3090 !extent_buffer_uptodate(chunk_root->node)) {
3091 btrfs_err(fs_info, "failed to read chunk root");
3092 if (!IS_ERR(chunk_root->node))
3093 free_extent_buffer(chunk_root->node);
3094 chunk_root->node = NULL;
3095 goto fail_tree_roots;
3097 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3098 chunk_root->commit_root = btrfs_root_node(chunk_root);
3100 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3101 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3103 ret = btrfs_read_chunk_tree(fs_info);
3105 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3106 goto fail_tree_roots;
3110 * Keep the devid that is marked to be the target device for the
3111 * device replace procedure
3113 btrfs_free_extra_devids(fs_devices, 0);
3115 if (!fs_devices->latest_bdev) {
3116 btrfs_err(fs_info, "failed to read devices");
3117 goto fail_tree_roots;
3120 ret = init_tree_roots(fs_info);
3122 goto fail_tree_roots;
3124 ret = btrfs_verify_dev_extents(fs_info);
3127 "failed to verify dev extents against chunks: %d",
3129 goto fail_block_groups;
3131 ret = btrfs_recover_balance(fs_info);
3133 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3134 goto fail_block_groups;
3137 ret = btrfs_init_dev_stats(fs_info);
3139 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3140 goto fail_block_groups;
3143 ret = btrfs_init_dev_replace(fs_info);
3145 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3146 goto fail_block_groups;
3149 btrfs_free_extra_devids(fs_devices, 1);
3151 ret = btrfs_sysfs_add_fsid(fs_devices);
3153 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3155 goto fail_block_groups;
3158 ret = btrfs_sysfs_add_mounted(fs_info);
3160 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3161 goto fail_fsdev_sysfs;
3164 ret = btrfs_init_space_info(fs_info);
3166 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3170 ret = btrfs_read_block_groups(fs_info);
3172 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3176 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3178 "writable mount is not allowed due to too many missing devices");
3182 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3184 if (IS_ERR(fs_info->cleaner_kthread))
3187 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3189 "btrfs-transaction");
3190 if (IS_ERR(fs_info->transaction_kthread))
3193 if (!btrfs_test_opt(fs_info, NOSSD) &&
3194 !fs_info->fs_devices->rotating) {
3195 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3199 * Mount does not set all options immediately, we can do it now and do
3200 * not have to wait for transaction commit
3202 btrfs_apply_pending_changes(fs_info);
3204 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3205 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3206 ret = btrfsic_mount(fs_info, fs_devices,
3207 btrfs_test_opt(fs_info,
3208 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3210 fs_info->check_integrity_print_mask);
3213 "failed to initialize integrity check module: %d",
3217 ret = btrfs_read_qgroup_config(fs_info);
3219 goto fail_trans_kthread;
3221 if (btrfs_build_ref_tree(fs_info))
3222 btrfs_err(fs_info, "couldn't build ref tree");
3224 /* do not make disk changes in broken FS or nologreplay is given */
3225 if (btrfs_super_log_root(disk_super) != 0 &&
3226 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3227 btrfs_info(fs_info, "start tree-log replay");
3228 ret = btrfs_replay_log(fs_info, fs_devices);
3235 ret = btrfs_find_orphan_roots(fs_info);
3239 if (!sb_rdonly(sb)) {
3240 ret = btrfs_cleanup_fs_roots(fs_info);
3244 mutex_lock(&fs_info->cleaner_mutex);
3245 ret = btrfs_recover_relocation(tree_root);
3246 mutex_unlock(&fs_info->cleaner_mutex);
3248 btrfs_warn(fs_info, "failed to recover relocation: %d",
3255 location.objectid = BTRFS_FS_TREE_OBJECTID;
3256 location.type = BTRFS_ROOT_ITEM_KEY;
3257 location.offset = 0;
3259 fs_info->fs_root = btrfs_get_fs_root(fs_info, &location, true);
3260 if (IS_ERR(fs_info->fs_root)) {
3261 err = PTR_ERR(fs_info->fs_root);
3262 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3263 fs_info->fs_root = NULL;
3270 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3271 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3272 clear_free_space_tree = 1;
3273 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3274 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3275 btrfs_warn(fs_info, "free space tree is invalid");
3276 clear_free_space_tree = 1;
3279 if (clear_free_space_tree) {
3280 btrfs_info(fs_info, "clearing free space tree");
3281 ret = btrfs_clear_free_space_tree(fs_info);
3284 "failed to clear free space tree: %d", ret);
3285 close_ctree(fs_info);
3290 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3291 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3292 btrfs_info(fs_info, "creating free space tree");
3293 ret = btrfs_create_free_space_tree(fs_info);
3296 "failed to create free space tree: %d", ret);
3297 close_ctree(fs_info);
3302 down_read(&fs_info->cleanup_work_sem);
3303 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3304 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3305 up_read(&fs_info->cleanup_work_sem);
3306 close_ctree(fs_info);
3309 up_read(&fs_info->cleanup_work_sem);
3311 ret = btrfs_resume_balance_async(fs_info);
3313 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3314 close_ctree(fs_info);
3318 ret = btrfs_resume_dev_replace_async(fs_info);
3320 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3321 close_ctree(fs_info);
3325 btrfs_qgroup_rescan_resume(fs_info);
3326 btrfs_discard_resume(fs_info);
3328 if (!fs_info->uuid_root) {
3329 btrfs_info(fs_info, "creating UUID tree");
3330 ret = btrfs_create_uuid_tree(fs_info);
3333 "failed to create the UUID tree: %d", ret);
3334 close_ctree(fs_info);
3337 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3338 fs_info->generation !=
3339 btrfs_super_uuid_tree_generation(disk_super)) {
3340 btrfs_info(fs_info, "checking UUID tree");
3341 ret = btrfs_check_uuid_tree(fs_info);
3344 "failed to check the UUID tree: %d", ret);
3345 close_ctree(fs_info);
3349 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3351 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3354 * backuproot only affect mount behavior, and if open_ctree succeeded,
3355 * no need to keep the flag
3357 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3362 btrfs_free_qgroup_config(fs_info);
3364 kthread_stop(fs_info->transaction_kthread);
3365 btrfs_cleanup_transaction(fs_info);
3366 btrfs_free_fs_roots(fs_info);
3368 kthread_stop(fs_info->cleaner_kthread);
3371 * make sure we're done with the btree inode before we stop our
3374 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3377 btrfs_sysfs_remove_mounted(fs_info);
3380 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3383 btrfs_put_block_group_cache(fs_info);
3386 free_root_pointers(fs_info, true);
3387 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3390 btrfs_stop_all_workers(fs_info);
3391 btrfs_free_block_groups(fs_info);
3393 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3395 iput(fs_info->btree_inode);
3397 cleanup_srcu_struct(&fs_info->subvol_srcu);
3399 btrfs_close_devices(fs_info->fs_devices);
3402 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3404 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3407 set_buffer_uptodate(bh);
3409 struct btrfs_device *device = (struct btrfs_device *)
3412 btrfs_warn_rl_in_rcu(device->fs_info,
3413 "lost page write due to IO error on %s",
3414 rcu_str_deref(device->name));
3415 /* note, we don't set_buffer_write_io_error because we have
3416 * our own ways of dealing with the IO errors
3418 clear_buffer_uptodate(bh);
3419 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3425 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3426 struct buffer_head **bh_ret)
3428 struct buffer_head *bh;
3429 struct btrfs_super_block *super;
3432 bytenr = btrfs_sb_offset(copy_num);
3433 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3436 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3438 * If we fail to read from the underlying devices, as of now
3439 * the best option we have is to mark it EIO.
3444 super = (struct btrfs_super_block *)bh->b_data;
3445 if (btrfs_super_bytenr(super) != bytenr ||
3446 btrfs_super_magic(super) != BTRFS_MAGIC) {
3456 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3458 struct buffer_head *bh;
3459 struct buffer_head *latest = NULL;
3460 struct btrfs_super_block *super;
3465 /* we would like to check all the supers, but that would make
3466 * a btrfs mount succeed after a mkfs from a different FS.
3467 * So, we need to add a special mount option to scan for
3468 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3470 for (i = 0; i < 1; i++) {
3471 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3475 super = (struct btrfs_super_block *)bh->b_data;
3477 if (!latest || btrfs_super_generation(super) > transid) {
3480 transid = btrfs_super_generation(super);
3487 return ERR_PTR(ret);
3493 * Write superblock @sb to the @device. Do not wait for completion, all the
3494 * buffer heads we write are pinned.
3496 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3497 * the expected device size at commit time. Note that max_mirrors must be
3498 * same for write and wait phases.
3500 * Return number of errors when buffer head is not found or submission fails.
3502 static int write_dev_supers(struct btrfs_device *device,
3503 struct btrfs_super_block *sb, int max_mirrors)
3505 struct btrfs_fs_info *fs_info = device->fs_info;
3506 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3507 struct buffer_head *bh;
3514 if (max_mirrors == 0)
3515 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3517 shash->tfm = fs_info->csum_shash;
3519 for (i = 0; i < max_mirrors; i++) {
3520 bytenr = btrfs_sb_offset(i);
3521 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3522 device->commit_total_bytes)
3525 btrfs_set_super_bytenr(sb, bytenr);
3527 crypto_shash_init(shash);
3528 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3529 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3530 crypto_shash_final(shash, sb->csum);
3532 /* One reference for us, and we leave it for the caller */
3533 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3534 BTRFS_SUPER_INFO_SIZE);
3536 btrfs_err(device->fs_info,
3537 "couldn't get super buffer head for bytenr %llu",
3543 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3545 /* one reference for submit_bh */
3548 set_buffer_uptodate(bh);
3550 bh->b_end_io = btrfs_end_buffer_write_sync;
3551 bh->b_private = device;
3554 * we fua the first super. The others we allow
3557 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3558 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3559 op_flags |= REQ_FUA;
3560 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3564 return errors < i ? 0 : -1;
3568 * Wait for write completion of superblocks done by write_dev_supers,
3569 * @max_mirrors same for write and wait phases.
3571 * Return number of errors when buffer head is not found or not marked up to
3574 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3576 struct buffer_head *bh;
3579 bool primary_failed = false;
3582 if (max_mirrors == 0)
3583 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3585 for (i = 0; i < max_mirrors; i++) {
3586 bytenr = btrfs_sb_offset(i);
3587 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3588 device->commit_total_bytes)
3591 bh = __find_get_block(device->bdev,
3592 bytenr / BTRFS_BDEV_BLOCKSIZE,
3593 BTRFS_SUPER_INFO_SIZE);
3597 primary_failed = true;
3601 if (!buffer_uptodate(bh)) {
3604 primary_failed = true;
3607 /* drop our reference */
3610 /* drop the reference from the writing run */
3614 /* log error, force error return */
3615 if (primary_failed) {
3616 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3621 return errors < i ? 0 : -1;
3625 * endio for the write_dev_flush, this will wake anyone waiting
3626 * for the barrier when it is done
3628 static void btrfs_end_empty_barrier(struct bio *bio)
3630 complete(bio->bi_private);
3634 * Submit a flush request to the device if it supports it. Error handling is
3635 * done in the waiting counterpart.
3637 static void write_dev_flush(struct btrfs_device *device)
3639 struct request_queue *q = bdev_get_queue(device->bdev);
3640 struct bio *bio = device->flush_bio;
3642 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3646 bio->bi_end_io = btrfs_end_empty_barrier;
3647 bio_set_dev(bio, device->bdev);
3648 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3649 init_completion(&device->flush_wait);
3650 bio->bi_private = &device->flush_wait;
3652 btrfsic_submit_bio(bio);
3653 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3657 * If the flush bio has been submitted by write_dev_flush, wait for it.
3659 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3661 struct bio *bio = device->flush_bio;
3663 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3666 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3667 wait_for_completion_io(&device->flush_wait);
3669 return bio->bi_status;
3672 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3674 if (!btrfs_check_rw_degradable(fs_info, NULL))
3680 * send an empty flush down to each device in parallel,
3681 * then wait for them
3683 static int barrier_all_devices(struct btrfs_fs_info *info)
3685 struct list_head *head;
3686 struct btrfs_device *dev;
3687 int errors_wait = 0;
3690 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3691 /* send down all the barriers */
3692 head = &info->fs_devices->devices;
3693 list_for_each_entry(dev, head, dev_list) {
3694 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3698 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3699 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3702 write_dev_flush(dev);
3703 dev->last_flush_error = BLK_STS_OK;
3706 /* wait for all the barriers */
3707 list_for_each_entry(dev, head, dev_list) {
3708 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3714 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3715 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3718 ret = wait_dev_flush(dev);
3720 dev->last_flush_error = ret;
3721 btrfs_dev_stat_inc_and_print(dev,
3722 BTRFS_DEV_STAT_FLUSH_ERRS);
3729 * At some point we need the status of all disks
3730 * to arrive at the volume status. So error checking
3731 * is being pushed to a separate loop.
3733 return check_barrier_error(info);
3738 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3741 int min_tolerated = INT_MAX;
3743 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3744 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3745 min_tolerated = min_t(int, min_tolerated,
3746 btrfs_raid_array[BTRFS_RAID_SINGLE].
3747 tolerated_failures);
3749 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3750 if (raid_type == BTRFS_RAID_SINGLE)
3752 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3754 min_tolerated = min_t(int, min_tolerated,
3755 btrfs_raid_array[raid_type].
3756 tolerated_failures);
3759 if (min_tolerated == INT_MAX) {
3760 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3764 return min_tolerated;
3767 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3769 struct list_head *head;
3770 struct btrfs_device *dev;
3771 struct btrfs_super_block *sb;
3772 struct btrfs_dev_item *dev_item;
3776 int total_errors = 0;
3779 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3782 * max_mirrors == 0 indicates we're from commit_transaction,
3783 * not from fsync where the tree roots in fs_info have not
3784 * been consistent on disk.
3786 if (max_mirrors == 0)
3787 backup_super_roots(fs_info);
3789 sb = fs_info->super_for_commit;
3790 dev_item = &sb->dev_item;
3792 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3793 head = &fs_info->fs_devices->devices;
3794 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3797 ret = barrier_all_devices(fs_info);
3800 &fs_info->fs_devices->device_list_mutex);
3801 btrfs_handle_fs_error(fs_info, ret,
3802 "errors while submitting device barriers.");
3807 list_for_each_entry(dev, head, dev_list) {
3812 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3813 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3816 btrfs_set_stack_device_generation(dev_item, 0);
3817 btrfs_set_stack_device_type(dev_item, dev->type);
3818 btrfs_set_stack_device_id(dev_item, dev->devid);
3819 btrfs_set_stack_device_total_bytes(dev_item,
3820 dev->commit_total_bytes);
3821 btrfs_set_stack_device_bytes_used(dev_item,
3822 dev->commit_bytes_used);
3823 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3824 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3825 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3826 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3827 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3830 flags = btrfs_super_flags(sb);
3831 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3833 ret = btrfs_validate_write_super(fs_info, sb);
3835 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3836 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3837 "unexpected superblock corruption detected");
3841 ret = write_dev_supers(dev, sb, max_mirrors);
3845 if (total_errors > max_errors) {
3846 btrfs_err(fs_info, "%d errors while writing supers",
3848 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3850 /* FUA is masked off if unsupported and can't be the reason */
3851 btrfs_handle_fs_error(fs_info, -EIO,
3852 "%d errors while writing supers",
3858 list_for_each_entry(dev, head, dev_list) {
3861 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3862 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3865 ret = wait_dev_supers(dev, max_mirrors);
3869 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3870 if (total_errors > max_errors) {
3871 btrfs_handle_fs_error(fs_info, -EIO,
3872 "%d errors while writing supers",
3879 /* Drop a fs root from the radix tree and free it. */
3880 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3881 struct btrfs_root *root)
3883 spin_lock(&fs_info->fs_roots_radix_lock);
3884 radix_tree_delete(&fs_info->fs_roots_radix,
3885 (unsigned long)root->root_key.objectid);
3886 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3887 btrfs_put_fs_root(root);
3888 spin_unlock(&fs_info->fs_roots_radix_lock);
3890 if (btrfs_root_refs(&root->root_item) == 0)
3891 synchronize_srcu(&fs_info->subvol_srcu);
3893 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3894 btrfs_free_log(NULL, root);
3895 if (root->reloc_root) {
3896 free_extent_buffer(root->reloc_root->node);
3897 free_extent_buffer(root->reloc_root->commit_root);
3898 btrfs_put_fs_root(root->reloc_root);
3899 root->reloc_root = NULL;
3903 if (root->free_ino_pinned)
3904 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3905 if (root->free_ino_ctl)
3906 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3907 btrfs_free_fs_root(root);
3910 void btrfs_free_fs_root(struct btrfs_root *root)
3912 iput(root->ino_cache_inode);
3913 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3915 free_anon_bdev(root->anon_dev);
3916 if (root->subv_writers)
3917 btrfs_free_subvolume_writers(root->subv_writers);
3918 free_extent_buffer(root->node);
3919 free_extent_buffer(root->commit_root);
3920 kfree(root->free_ino_ctl);
3921 kfree(root->free_ino_pinned);
3922 btrfs_put_fs_root(root);
3925 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3927 u64 root_objectid = 0;
3928 struct btrfs_root *gang[8];
3931 unsigned int ret = 0;
3935 index = srcu_read_lock(&fs_info->subvol_srcu);
3936 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3937 (void **)gang, root_objectid,
3940 srcu_read_unlock(&fs_info->subvol_srcu, index);
3943 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3945 for (i = 0; i < ret; i++) {
3946 /* Avoid to grab roots in dead_roots */
3947 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3951 /* grab all the search result for later use */
3952 gang[i] = btrfs_grab_fs_root(gang[i]);
3954 srcu_read_unlock(&fs_info->subvol_srcu, index);
3956 for (i = 0; i < ret; i++) {
3959 root_objectid = gang[i]->root_key.objectid;
3960 err = btrfs_orphan_cleanup(gang[i]);
3963 btrfs_put_fs_root(gang[i]);
3968 /* release the uncleaned roots due to error */
3969 for (; i < ret; i++) {
3971 btrfs_put_fs_root(gang[i]);
3976 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3978 struct btrfs_root *root = fs_info->tree_root;
3979 struct btrfs_trans_handle *trans;
3981 mutex_lock(&fs_info->cleaner_mutex);
3982 btrfs_run_delayed_iputs(fs_info);
3983 mutex_unlock(&fs_info->cleaner_mutex);
3984 wake_up_process(fs_info->cleaner_kthread);
3986 /* wait until ongoing cleanup work done */
3987 down_write(&fs_info->cleanup_work_sem);
3988 up_write(&fs_info->cleanup_work_sem);
3990 trans = btrfs_join_transaction(root);
3992 return PTR_ERR(trans);
3993 return btrfs_commit_transaction(trans);
3996 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4000 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4002 * We don't want the cleaner to start new transactions, add more delayed
4003 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4004 * because that frees the task_struct, and the transaction kthread might
4005 * still try to wake up the cleaner.
4007 kthread_park(fs_info->cleaner_kthread);
4009 /* wait for the qgroup rescan worker to stop */
4010 btrfs_qgroup_wait_for_completion(fs_info, false);
4012 /* wait for the uuid_scan task to finish */
4013 down(&fs_info->uuid_tree_rescan_sem);
4014 /* avoid complains from lockdep et al., set sem back to initial state */
4015 up(&fs_info->uuid_tree_rescan_sem);
4017 /* pause restriper - we want to resume on mount */
4018 btrfs_pause_balance(fs_info);
4020 btrfs_dev_replace_suspend_for_unmount(fs_info);
4022 btrfs_scrub_cancel(fs_info);
4024 /* wait for any defraggers to finish */
4025 wait_event(fs_info->transaction_wait,
4026 (atomic_read(&fs_info->defrag_running) == 0));
4028 /* clear out the rbtree of defraggable inodes */
4029 btrfs_cleanup_defrag_inodes(fs_info);
4031 cancel_work_sync(&fs_info->async_reclaim_work);
4033 /* Cancel or finish ongoing discard work */
4034 btrfs_discard_cleanup(fs_info);
4036 if (!sb_rdonly(fs_info->sb)) {
4038 * The cleaner kthread is stopped, so do one final pass over
4039 * unused block groups.
4041 btrfs_delete_unused_bgs(fs_info);
4043 ret = btrfs_commit_super(fs_info);
4045 btrfs_err(fs_info, "commit super ret %d", ret);
4048 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4049 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4050 btrfs_error_commit_super(fs_info);
4052 kthread_stop(fs_info->transaction_kthread);
4053 kthread_stop(fs_info->cleaner_kthread);
4055 ASSERT(list_empty(&fs_info->delayed_iputs));
4056 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4058 btrfs_free_qgroup_config(fs_info);
4059 ASSERT(list_empty(&fs_info->delalloc_roots));
4061 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4062 btrfs_info(fs_info, "at unmount delalloc count %lld",
4063 percpu_counter_sum(&fs_info->delalloc_bytes));
4066 if (percpu_counter_sum(&fs_info->dio_bytes))
4067 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4068 percpu_counter_sum(&fs_info->dio_bytes));
4070 btrfs_sysfs_remove_mounted(fs_info);
4071 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4073 btrfs_free_fs_roots(fs_info);
4075 btrfs_put_block_group_cache(fs_info);
4078 * we must make sure there is not any read request to
4079 * submit after we stopping all workers.
4081 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4082 btrfs_stop_all_workers(fs_info);
4084 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4085 free_root_pointers(fs_info, true);
4088 * We must free the block groups after dropping the fs_roots as we could
4089 * have had an IO error and have left over tree log blocks that aren't
4090 * cleaned up until the fs roots are freed. This makes the block group
4091 * accounting appear to be wrong because there's pending reserved bytes,
4092 * so make sure we do the block group cleanup afterwards.
4094 btrfs_free_block_groups(fs_info);
4096 iput(fs_info->btree_inode);
4098 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4099 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4100 btrfsic_unmount(fs_info->fs_devices);
4103 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4104 btrfs_close_devices(fs_info->fs_devices);
4105 cleanup_srcu_struct(&fs_info->subvol_srcu);
4108 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4112 struct inode *btree_inode = buf->pages[0]->mapping->host;
4114 ret = extent_buffer_uptodate(buf);
4118 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4119 parent_transid, atomic);
4125 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4127 struct btrfs_fs_info *fs_info;
4128 struct btrfs_root *root;
4129 u64 transid = btrfs_header_generation(buf);
4132 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4134 * This is a fast path so only do this check if we have sanity tests
4135 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4136 * outside of the sanity tests.
4138 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4141 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4142 fs_info = root->fs_info;
4143 btrfs_assert_tree_locked(buf);
4144 if (transid != fs_info->generation)
4145 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4146 buf->start, transid, fs_info->generation);
4147 was_dirty = set_extent_buffer_dirty(buf);
4149 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4151 fs_info->dirty_metadata_batch);
4152 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4154 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4155 * but item data not updated.
4156 * So here we should only check item pointers, not item data.
4158 if (btrfs_header_level(buf) == 0 &&
4159 btrfs_check_leaf_relaxed(buf)) {
4160 btrfs_print_leaf(buf);
4166 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4170 * looks as though older kernels can get into trouble with
4171 * this code, they end up stuck in balance_dirty_pages forever
4175 if (current->flags & PF_MEMALLOC)
4179 btrfs_balance_delayed_items(fs_info);
4181 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4182 BTRFS_DIRTY_METADATA_THRESH,
4183 fs_info->dirty_metadata_batch);
4185 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4189 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4191 __btrfs_btree_balance_dirty(fs_info, 1);
4194 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4196 __btrfs_btree_balance_dirty(fs_info, 0);
4199 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4200 struct btrfs_key *first_key)
4202 return btree_read_extent_buffer_pages(buf, parent_transid,
4206 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4208 /* cleanup FS via transaction */
4209 btrfs_cleanup_transaction(fs_info);
4211 mutex_lock(&fs_info->cleaner_mutex);
4212 btrfs_run_delayed_iputs(fs_info);
4213 mutex_unlock(&fs_info->cleaner_mutex);
4215 down_write(&fs_info->cleanup_work_sem);
4216 up_write(&fs_info->cleanup_work_sem);
4219 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4221 struct btrfs_ordered_extent *ordered;
4223 spin_lock(&root->ordered_extent_lock);
4225 * This will just short circuit the ordered completion stuff which will
4226 * make sure the ordered extent gets properly cleaned up.
4228 list_for_each_entry(ordered, &root->ordered_extents,
4230 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4231 spin_unlock(&root->ordered_extent_lock);
4234 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4236 struct btrfs_root *root;
4237 struct list_head splice;
4239 INIT_LIST_HEAD(&splice);
4241 spin_lock(&fs_info->ordered_root_lock);
4242 list_splice_init(&fs_info->ordered_roots, &splice);
4243 while (!list_empty(&splice)) {
4244 root = list_first_entry(&splice, struct btrfs_root,
4246 list_move_tail(&root->ordered_root,
4247 &fs_info->ordered_roots);
4249 spin_unlock(&fs_info->ordered_root_lock);
4250 btrfs_destroy_ordered_extents(root);
4253 spin_lock(&fs_info->ordered_root_lock);
4255 spin_unlock(&fs_info->ordered_root_lock);
4258 * We need this here because if we've been flipped read-only we won't
4259 * get sync() from the umount, so we need to make sure any ordered
4260 * extents that haven't had their dirty pages IO start writeout yet
4261 * actually get run and error out properly.
4263 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4266 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4267 struct btrfs_fs_info *fs_info)
4269 struct rb_node *node;
4270 struct btrfs_delayed_ref_root *delayed_refs;
4271 struct btrfs_delayed_ref_node *ref;
4274 delayed_refs = &trans->delayed_refs;
4276 spin_lock(&delayed_refs->lock);
4277 if (atomic_read(&delayed_refs->num_entries) == 0) {
4278 spin_unlock(&delayed_refs->lock);
4279 btrfs_info(fs_info, "delayed_refs has NO entry");
4283 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4284 struct btrfs_delayed_ref_head *head;
4286 bool pin_bytes = false;
4288 head = rb_entry(node, struct btrfs_delayed_ref_head,
4290 if (btrfs_delayed_ref_lock(delayed_refs, head))
4293 spin_lock(&head->lock);
4294 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4295 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4298 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4299 RB_CLEAR_NODE(&ref->ref_node);
4300 if (!list_empty(&ref->add_list))
4301 list_del(&ref->add_list);
4302 atomic_dec(&delayed_refs->num_entries);
4303 btrfs_put_delayed_ref(ref);
4305 if (head->must_insert_reserved)
4307 btrfs_free_delayed_extent_op(head->extent_op);
4308 btrfs_delete_ref_head(delayed_refs, head);
4309 spin_unlock(&head->lock);
4310 spin_unlock(&delayed_refs->lock);
4311 mutex_unlock(&head->mutex);
4314 btrfs_pin_extent(fs_info, head->bytenr,
4315 head->num_bytes, 1);
4316 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4317 btrfs_put_delayed_ref_head(head);
4319 spin_lock(&delayed_refs->lock);
4321 btrfs_qgroup_destroy_extent_records(trans);
4323 spin_unlock(&delayed_refs->lock);
4328 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4330 struct btrfs_inode *btrfs_inode;
4331 struct list_head splice;
4333 INIT_LIST_HEAD(&splice);
4335 spin_lock(&root->delalloc_lock);
4336 list_splice_init(&root->delalloc_inodes, &splice);
4338 while (!list_empty(&splice)) {
4339 struct inode *inode = NULL;
4340 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4342 __btrfs_del_delalloc_inode(root, btrfs_inode);
4343 spin_unlock(&root->delalloc_lock);
4346 * Make sure we get a live inode and that it'll not disappear
4349 inode = igrab(&btrfs_inode->vfs_inode);
4351 invalidate_inode_pages2(inode->i_mapping);
4354 spin_lock(&root->delalloc_lock);
4356 spin_unlock(&root->delalloc_lock);
4359 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4361 struct btrfs_root *root;
4362 struct list_head splice;
4364 INIT_LIST_HEAD(&splice);
4366 spin_lock(&fs_info->delalloc_root_lock);
4367 list_splice_init(&fs_info->delalloc_roots, &splice);
4368 while (!list_empty(&splice)) {
4369 root = list_first_entry(&splice, struct btrfs_root,
4371 root = btrfs_grab_fs_root(root);
4373 spin_unlock(&fs_info->delalloc_root_lock);
4375 btrfs_destroy_delalloc_inodes(root);
4376 btrfs_put_fs_root(root);
4378 spin_lock(&fs_info->delalloc_root_lock);
4380 spin_unlock(&fs_info->delalloc_root_lock);
4383 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4384 struct extent_io_tree *dirty_pages,
4388 struct extent_buffer *eb;
4393 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4398 clear_extent_bits(dirty_pages, start, end, mark);
4399 while (start <= end) {
4400 eb = find_extent_buffer(fs_info, start);
4401 start += fs_info->nodesize;
4404 wait_on_extent_buffer_writeback(eb);
4406 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4408 clear_extent_buffer_dirty(eb);
4409 free_extent_buffer_stale(eb);
4416 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4417 struct extent_io_tree *pinned_extents)
4419 struct extent_io_tree *unpin;
4425 unpin = pinned_extents;
4428 struct extent_state *cached_state = NULL;
4431 * The btrfs_finish_extent_commit() may get the same range as
4432 * ours between find_first_extent_bit and clear_extent_dirty.
4433 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4434 * the same extent range.
4436 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4437 ret = find_first_extent_bit(unpin, 0, &start, &end,
4438 EXTENT_DIRTY, &cached_state);
4440 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4444 clear_extent_dirty(unpin, start, end, &cached_state);
4445 free_extent_state(cached_state);
4446 btrfs_error_unpin_extent_range(fs_info, start, end);
4447 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4452 if (unpin == &fs_info->freed_extents[0])
4453 unpin = &fs_info->freed_extents[1];
4455 unpin = &fs_info->freed_extents[0];
4463 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4465 struct inode *inode;
4467 inode = cache->io_ctl.inode;
4469 invalidate_inode_pages2(inode->i_mapping);
4470 BTRFS_I(inode)->generation = 0;
4471 cache->io_ctl.inode = NULL;
4474 btrfs_put_block_group(cache);
4477 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4478 struct btrfs_fs_info *fs_info)
4480 struct btrfs_block_group *cache;
4482 spin_lock(&cur_trans->dirty_bgs_lock);
4483 while (!list_empty(&cur_trans->dirty_bgs)) {
4484 cache = list_first_entry(&cur_trans->dirty_bgs,
4485 struct btrfs_block_group,
4488 if (!list_empty(&cache->io_list)) {
4489 spin_unlock(&cur_trans->dirty_bgs_lock);
4490 list_del_init(&cache->io_list);
4491 btrfs_cleanup_bg_io(cache);
4492 spin_lock(&cur_trans->dirty_bgs_lock);
4495 list_del_init(&cache->dirty_list);
4496 spin_lock(&cache->lock);
4497 cache->disk_cache_state = BTRFS_DC_ERROR;
4498 spin_unlock(&cache->lock);
4500 spin_unlock(&cur_trans->dirty_bgs_lock);
4501 btrfs_put_block_group(cache);
4502 btrfs_delayed_refs_rsv_release(fs_info, 1);
4503 spin_lock(&cur_trans->dirty_bgs_lock);
4505 spin_unlock(&cur_trans->dirty_bgs_lock);
4508 * Refer to the definition of io_bgs member for details why it's safe
4509 * to use it without any locking
4511 while (!list_empty(&cur_trans->io_bgs)) {
4512 cache = list_first_entry(&cur_trans->io_bgs,
4513 struct btrfs_block_group,
4516 list_del_init(&cache->io_list);
4517 spin_lock(&cache->lock);
4518 cache->disk_cache_state = BTRFS_DC_ERROR;
4519 spin_unlock(&cache->lock);
4520 btrfs_cleanup_bg_io(cache);
4524 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4525 struct btrfs_fs_info *fs_info)
4527 struct btrfs_device *dev, *tmp;
4529 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4530 ASSERT(list_empty(&cur_trans->dirty_bgs));
4531 ASSERT(list_empty(&cur_trans->io_bgs));
4533 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4535 list_del_init(&dev->post_commit_list);
4538 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4540 cur_trans->state = TRANS_STATE_COMMIT_START;
4541 wake_up(&fs_info->transaction_blocked_wait);
4543 cur_trans->state = TRANS_STATE_UNBLOCKED;
4544 wake_up(&fs_info->transaction_wait);
4546 btrfs_destroy_delayed_inodes(fs_info);
4548 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4550 btrfs_destroy_pinned_extent(fs_info,
4551 fs_info->pinned_extents);
4553 cur_trans->state =TRANS_STATE_COMPLETED;
4554 wake_up(&cur_trans->commit_wait);
4557 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4559 struct btrfs_transaction *t;
4561 mutex_lock(&fs_info->transaction_kthread_mutex);
4563 spin_lock(&fs_info->trans_lock);
4564 while (!list_empty(&fs_info->trans_list)) {
4565 t = list_first_entry(&fs_info->trans_list,
4566 struct btrfs_transaction, list);
4567 if (t->state >= TRANS_STATE_COMMIT_START) {
4568 refcount_inc(&t->use_count);
4569 spin_unlock(&fs_info->trans_lock);
4570 btrfs_wait_for_commit(fs_info, t->transid);
4571 btrfs_put_transaction(t);
4572 spin_lock(&fs_info->trans_lock);
4575 if (t == fs_info->running_transaction) {
4576 t->state = TRANS_STATE_COMMIT_DOING;
4577 spin_unlock(&fs_info->trans_lock);
4579 * We wait for 0 num_writers since we don't hold a trans
4580 * handle open currently for this transaction.
4582 wait_event(t->writer_wait,
4583 atomic_read(&t->num_writers) == 0);
4585 spin_unlock(&fs_info->trans_lock);
4587 btrfs_cleanup_one_transaction(t, fs_info);
4589 spin_lock(&fs_info->trans_lock);
4590 if (t == fs_info->running_transaction)
4591 fs_info->running_transaction = NULL;
4592 list_del_init(&t->list);
4593 spin_unlock(&fs_info->trans_lock);
4595 btrfs_put_transaction(t);
4596 trace_btrfs_transaction_commit(fs_info->tree_root);
4597 spin_lock(&fs_info->trans_lock);
4599 spin_unlock(&fs_info->trans_lock);
4600 btrfs_destroy_all_ordered_extents(fs_info);
4601 btrfs_destroy_delayed_inodes(fs_info);
4602 btrfs_assert_delayed_root_empty(fs_info);
4603 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4604 btrfs_destroy_all_delalloc_inodes(fs_info);
4605 mutex_unlock(&fs_info->transaction_kthread_mutex);
4610 static const struct extent_io_ops btree_extent_io_ops = {
4611 /* mandatory callbacks */
4612 .submit_bio_hook = btree_submit_bio_hook,
4613 .readpage_end_io_hook = btree_readpage_end_io_hook,