1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/scatterlist.h>
9 #include <linux/swap.h>
10 #include <linux/radix-tree.h>
11 #include <linux/writeback.h>
12 #include <linux/buffer_head.h>
13 #include <linux/workqueue.h>
14 #include <linux/kthread.h>
15 #include <linux/slab.h>
16 #include <linux/migrate.h>
17 #include <linux/ratelimit.h>
18 #include <linux/uuid.h>
19 #include <linux/semaphore.h>
20 #include <linux/error-injection.h>
21 #include <linux/crc32c.h>
22 #include <asm/unaligned.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"
45 #include <asm/cpufeature.h>
48 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
49 BTRFS_HEADER_FLAG_RELOC |\
50 BTRFS_SUPER_FLAG_ERROR |\
51 BTRFS_SUPER_FLAG_SEEDING |\
52 BTRFS_SUPER_FLAG_METADUMP |\
53 BTRFS_SUPER_FLAG_METADUMP_V2)
55 static const struct extent_io_ops btree_extent_io_ops;
56 static void end_workqueue_fn(struct btrfs_work *work);
57 static void free_fs_root(struct btrfs_root *root);
58 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
59 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
60 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
61 struct btrfs_fs_info *fs_info);
62 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
63 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
64 struct extent_io_tree *dirty_pages,
66 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
67 struct extent_io_tree *pinned_extents);
68 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
69 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
72 * btrfs_end_io_wq structs are used to do processing in task context when an IO
73 * is complete. This is used during reads to verify checksums, and it is used
74 * by writes to insert metadata for new file extents after IO is complete.
76 struct btrfs_end_io_wq {
80 struct btrfs_fs_info *info;
82 enum btrfs_wq_endio_type metadata;
83 struct btrfs_work work;
86 static struct kmem_cache *btrfs_end_io_wq_cache;
88 int __init btrfs_end_io_wq_init(void)
90 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
91 sizeof(struct btrfs_end_io_wq),
95 if (!btrfs_end_io_wq_cache)
100 void __cold btrfs_end_io_wq_exit(void)
102 kmem_cache_destroy(btrfs_end_io_wq_cache);
106 * async submit bios are used to offload expensive checksumming
107 * onto the worker threads. They checksum file and metadata bios
108 * just before they are sent down the IO stack.
110 struct async_submit_bio {
112 struct btrfs_fs_info *fs_info;
114 extent_submit_bio_start_t *submit_bio_start;
115 extent_submit_bio_done_t *submit_bio_done;
117 unsigned long bio_flags;
119 * bio_offset is optional, can be used if the pages in the bio
120 * can't tell us where in the file the bio should go
123 struct btrfs_work work;
128 * Lockdep class keys for extent_buffer->lock's in this root. For a given
129 * eb, the lockdep key is determined by the btrfs_root it belongs to and
130 * the level the eb occupies in the tree.
132 * Different roots are used for different purposes and may nest inside each
133 * other and they require separate keysets. As lockdep keys should be
134 * static, assign keysets according to the purpose of the root as indicated
135 * by btrfs_root->objectid. This ensures that all special purpose roots
136 * have separate keysets.
138 * Lock-nesting across peer nodes is always done with the immediate parent
139 * node locked thus preventing deadlock. As lockdep doesn't know this, use
140 * subclass to avoid triggering lockdep warning in such cases.
142 * The key is set by the readpage_end_io_hook after the buffer has passed
143 * csum validation but before the pages are unlocked. It is also set by
144 * btrfs_init_new_buffer on freshly allocated blocks.
146 * We also add a check to make sure the highest level of the tree is the
147 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
148 * needs update as well.
150 #ifdef CONFIG_DEBUG_LOCK_ALLOC
151 # if BTRFS_MAX_LEVEL != 8
155 static struct btrfs_lockdep_keyset {
156 u64 id; /* root objectid */
157 const char *name_stem; /* lock name stem */
158 char names[BTRFS_MAX_LEVEL + 1][20];
159 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
160 } btrfs_lockdep_keysets[] = {
161 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
162 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
163 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
164 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
165 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
166 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
167 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
168 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
169 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
170 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
171 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
172 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
173 { .id = 0, .name_stem = "tree" },
176 void __init btrfs_init_lockdep(void)
180 /* initialize lockdep class names */
181 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
182 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
184 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
185 snprintf(ks->names[j], sizeof(ks->names[j]),
186 "btrfs-%s-%02d", ks->name_stem, j);
190 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
193 struct btrfs_lockdep_keyset *ks;
195 BUG_ON(level >= ARRAY_SIZE(ks->keys));
197 /* find the matching keyset, id 0 is the default entry */
198 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
199 if (ks->id == objectid)
202 lockdep_set_class_and_name(&eb->lock,
203 &ks->keys[level], ks->names[level]);
209 * extents on the btree inode are pretty simple, there's one extent
210 * that covers the entire device
212 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
213 struct page *page, size_t pg_offset, u64 start, u64 len,
216 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
217 struct extent_map_tree *em_tree = &inode->extent_tree;
218 struct extent_map *em;
221 read_lock(&em_tree->lock);
222 em = lookup_extent_mapping(em_tree, start, len);
224 em->bdev = fs_info->fs_devices->latest_bdev;
225 read_unlock(&em_tree->lock);
228 read_unlock(&em_tree->lock);
230 em = alloc_extent_map();
232 em = ERR_PTR(-ENOMEM);
237 em->block_len = (u64)-1;
239 em->bdev = fs_info->fs_devices->latest_bdev;
241 write_lock(&em_tree->lock);
242 ret = add_extent_mapping(em_tree, em, 0);
243 if (ret == -EEXIST) {
245 em = lookup_extent_mapping(em_tree, start, len);
252 write_unlock(&em_tree->lock);
258 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
260 return crc32c(seed, data, len);
263 void btrfs_csum_final(u32 crc, u8 *result)
265 put_unaligned_le32(~crc, result);
269 * compute the csum for a btree block, and either verify it or write it
270 * into the csum field of the block.
272 static int csum_tree_block(struct btrfs_fs_info *fs_info,
273 struct extent_buffer *buf,
276 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
277 char result[BTRFS_CSUM_SIZE];
279 unsigned long cur_len;
280 unsigned long offset = BTRFS_CSUM_SIZE;
282 unsigned long map_start;
283 unsigned long map_len;
287 len = buf->len - offset;
289 err = map_private_extent_buffer(buf, offset, 32,
290 &kaddr, &map_start, &map_len);
293 cur_len = min(len, map_len - (offset - map_start));
294 crc = btrfs_csum_data(kaddr + offset - map_start,
299 memset(result, 0, BTRFS_CSUM_SIZE);
301 btrfs_csum_final(crc, result);
304 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
307 memcpy(&found, result, csum_size);
309 read_extent_buffer(buf, &val, 0, csum_size);
310 btrfs_warn_rl(fs_info,
311 "%s checksum verify failed on %llu wanted %X found %X level %d",
312 fs_info->sb->s_id, buf->start,
313 val, found, btrfs_header_level(buf));
317 write_extent_buffer(buf, result, 0, csum_size);
324 * we can't consider a given block up to date unless the transid of the
325 * block matches the transid in the parent node's pointer. This is how we
326 * detect blocks that either didn't get written at all or got written
327 * in the wrong place.
329 static int verify_parent_transid(struct extent_io_tree *io_tree,
330 struct extent_buffer *eb, u64 parent_transid,
333 struct extent_state *cached_state = NULL;
335 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
337 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
344 btrfs_tree_read_lock(eb);
345 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
348 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
350 if (extent_buffer_uptodate(eb) &&
351 btrfs_header_generation(eb) == parent_transid) {
355 btrfs_err_rl(eb->fs_info,
356 "parent transid verify failed on %llu wanted %llu found %llu",
358 parent_transid, btrfs_header_generation(eb));
362 * Things reading via commit roots that don't have normal protection,
363 * like send, can have a really old block in cache that may point at a
364 * block that has been freed and re-allocated. So don't clear uptodate
365 * if we find an eb that is under IO (dirty/writeback) because we could
366 * end up reading in the stale data and then writing it back out and
367 * making everybody very sad.
369 if (!extent_buffer_under_io(eb))
370 clear_extent_buffer_uptodate(eb);
372 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
375 btrfs_tree_read_unlock_blocking(eb);
380 * Return 0 if the superblock checksum type matches the checksum value of that
381 * algorithm. Pass the raw disk superblock data.
383 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
386 struct btrfs_super_block *disk_sb =
387 (struct btrfs_super_block *)raw_disk_sb;
388 u16 csum_type = btrfs_super_csum_type(disk_sb);
391 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
393 char result[sizeof(crc)];
396 * The super_block structure does not span the whole
397 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
398 * is filled with zeros and is included in the checksum.
400 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
401 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
402 btrfs_csum_final(crc, result);
404 if (memcmp(raw_disk_sb, result, sizeof(result)))
408 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
409 btrfs_err(fs_info, "unsupported checksum algorithm %u",
417 static int verify_level_key(struct btrfs_fs_info *fs_info,
418 struct extent_buffer *eb, int level,
419 struct btrfs_key *first_key)
422 struct btrfs_key found_key;
425 found_level = btrfs_header_level(eb);
426 if (found_level != level) {
427 #ifdef CONFIG_BTRFS_DEBUG
430 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
431 eb->start, level, found_level);
440 * For live tree block (new tree blocks in current transaction),
441 * we need proper lock context to avoid race, which is impossible here.
442 * So we only checks tree blocks which is read from disk, whose
443 * generation <= fs_info->last_trans_committed.
445 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
448 btrfs_node_key_to_cpu(eb, &found_key, 0);
450 btrfs_item_key_to_cpu(eb, &found_key, 0);
451 ret = btrfs_comp_cpu_keys(first_key, &found_key);
453 #ifdef CONFIG_BTRFS_DEBUG
457 "tree first key mismatch detected, bytenr=%llu key expected=(%llu, %u, %llu) has=(%llu, %u, %llu)",
458 eb->start, first_key->objectid, first_key->type,
459 first_key->offset, found_key.objectid,
460 found_key.type, found_key.offset);
467 * helper to read a given tree block, doing retries as required when
468 * the checksums don't match and we have alternate mirrors to try.
470 * @parent_transid: expected transid, skip check if 0
471 * @level: expected level, mandatory check
472 * @first_key: expected key of first slot, skip check if NULL
474 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
475 struct extent_buffer *eb,
476 u64 parent_transid, int level,
477 struct btrfs_key *first_key)
479 struct extent_io_tree *io_tree;
484 int failed_mirror = 0;
486 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
487 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
489 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
492 if (verify_parent_transid(io_tree, eb,
495 else if (verify_level_key(fs_info, eb, level,
503 * This buffer's crc is fine, but its contents are corrupted, so
504 * there is no reason to read the other copies, they won't be
507 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
511 num_copies = btrfs_num_copies(fs_info,
516 if (!failed_mirror) {
518 failed_mirror = eb->read_mirror;
522 if (mirror_num == failed_mirror)
525 if (mirror_num > num_copies)
529 if (failed && !ret && failed_mirror)
530 repair_eb_io_failure(fs_info, eb, failed_mirror);
536 * checksum a dirty tree block before IO. This has extra checks to make sure
537 * we only fill in the checksum field in the first page of a multi-page block
540 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
542 u64 start = page_offset(page);
544 struct extent_buffer *eb;
546 eb = (struct extent_buffer *)page->private;
547 if (page != eb->pages[0])
550 found_start = btrfs_header_bytenr(eb);
552 * Please do not consolidate these warnings into a single if.
553 * It is useful to know what went wrong.
555 if (WARN_ON(found_start != start))
557 if (WARN_ON(!PageUptodate(page)))
560 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
561 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
563 return csum_tree_block(fs_info, eb, 0);
566 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
567 struct extent_buffer *eb)
569 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
570 u8 fsid[BTRFS_FSID_SIZE];
573 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
575 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
579 fs_devices = fs_devices->seed;
584 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
585 u64 phy_offset, struct page *page,
586 u64 start, u64 end, int mirror)
590 struct extent_buffer *eb;
591 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
592 struct btrfs_fs_info *fs_info = root->fs_info;
599 eb = (struct extent_buffer *)page->private;
601 /* the pending IO might have been the only thing that kept this buffer
602 * in memory. Make sure we have a ref for all this other checks
604 extent_buffer_get(eb);
606 reads_done = atomic_dec_and_test(&eb->io_pages);
610 eb->read_mirror = mirror;
611 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
616 found_start = btrfs_header_bytenr(eb);
617 if (found_start != eb->start) {
618 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
619 found_start, eb->start);
623 if (check_tree_block_fsid(fs_info, eb)) {
624 btrfs_err_rl(fs_info, "bad fsid on block %llu",
629 found_level = btrfs_header_level(eb);
630 if (found_level >= BTRFS_MAX_LEVEL) {
631 btrfs_err(fs_info, "bad tree block level %d",
632 (int)btrfs_header_level(eb));
637 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
640 ret = csum_tree_block(fs_info, eb, 1);
645 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 * that we don't try and read the other copies of this block, just
649 if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
650 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
654 if (found_level > 0 && btrfs_check_node(fs_info, eb))
658 set_extent_buffer_uptodate(eb);
661 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 btree_readahead_hook(eb, ret);
666 * our io error hook is going to dec the io pages
667 * again, we have to make sure it has something
670 atomic_inc(&eb->io_pages);
671 clear_extent_buffer_uptodate(eb);
673 free_extent_buffer(eb);
678 static int btree_io_failed_hook(struct page *page, int failed_mirror)
680 struct extent_buffer *eb;
682 eb = (struct extent_buffer *)page->private;
683 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
684 eb->read_mirror = failed_mirror;
685 atomic_dec(&eb->io_pages);
686 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
687 btree_readahead_hook(eb, -EIO);
688 return -EIO; /* we fixed nothing */
691 static void end_workqueue_bio(struct bio *bio)
693 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
694 struct btrfs_fs_info *fs_info;
695 struct btrfs_workqueue *wq;
696 btrfs_work_func_t func;
698 fs_info = end_io_wq->info;
699 end_io_wq->status = bio->bi_status;
701 if (bio_op(bio) == REQ_OP_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
703 wq = fs_info->endio_meta_write_workers;
704 func = btrfs_endio_meta_write_helper;
705 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
706 wq = fs_info->endio_freespace_worker;
707 func = btrfs_freespace_write_helper;
708 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
709 wq = fs_info->endio_raid56_workers;
710 func = btrfs_endio_raid56_helper;
712 wq = fs_info->endio_write_workers;
713 func = btrfs_endio_write_helper;
716 if (unlikely(end_io_wq->metadata ==
717 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
718 wq = fs_info->endio_repair_workers;
719 func = btrfs_endio_repair_helper;
720 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
721 wq = fs_info->endio_raid56_workers;
722 func = btrfs_endio_raid56_helper;
723 } else if (end_io_wq->metadata) {
724 wq = fs_info->endio_meta_workers;
725 func = btrfs_endio_meta_helper;
727 wq = fs_info->endio_workers;
728 func = btrfs_endio_helper;
732 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
733 btrfs_queue_work(wq, &end_io_wq->work);
736 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
737 enum btrfs_wq_endio_type metadata)
739 struct btrfs_end_io_wq *end_io_wq;
741 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
743 return BLK_STS_RESOURCE;
745 end_io_wq->private = bio->bi_private;
746 end_io_wq->end_io = bio->bi_end_io;
747 end_io_wq->info = info;
748 end_io_wq->status = 0;
749 end_io_wq->bio = bio;
750 end_io_wq->metadata = metadata;
752 bio->bi_private = end_io_wq;
753 bio->bi_end_io = end_workqueue_bio;
757 static void run_one_async_start(struct btrfs_work *work)
759 struct async_submit_bio *async;
762 async = container_of(work, struct async_submit_bio, work);
763 ret = async->submit_bio_start(async->private_data, async->bio,
769 static void run_one_async_done(struct btrfs_work *work)
771 struct async_submit_bio *async;
773 async = container_of(work, struct async_submit_bio, work);
775 /* If an error occurred we just want to clean up the bio and move on */
777 async->bio->bi_status = async->status;
778 bio_endio(async->bio);
782 async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
785 static void run_one_async_free(struct btrfs_work *work)
787 struct async_submit_bio *async;
789 async = container_of(work, struct async_submit_bio, work);
793 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
794 int mirror_num, unsigned long bio_flags,
795 u64 bio_offset, void *private_data,
796 extent_submit_bio_start_t *submit_bio_start,
797 extent_submit_bio_done_t *submit_bio_done)
799 struct async_submit_bio *async;
801 async = kmalloc(sizeof(*async), GFP_NOFS);
803 return BLK_STS_RESOURCE;
805 async->private_data = private_data;
806 async->fs_info = fs_info;
808 async->mirror_num = mirror_num;
809 async->submit_bio_start = submit_bio_start;
810 async->submit_bio_done = submit_bio_done;
812 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
813 run_one_async_done, run_one_async_free);
815 async->bio_flags = bio_flags;
816 async->bio_offset = bio_offset;
820 if (op_is_sync(bio->bi_opf))
821 btrfs_set_work_high_priority(&async->work);
823 btrfs_queue_work(fs_info->workers, &async->work);
827 static blk_status_t btree_csum_one_bio(struct bio *bio)
829 struct bio_vec *bvec;
830 struct btrfs_root *root;
833 ASSERT(!bio_flagged(bio, BIO_CLONED));
834 bio_for_each_segment_all(bvec, bio, i) {
835 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
836 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
841 return errno_to_blk_status(ret);
844 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
848 * when we're called for a write, we're already in the async
849 * submission context. Just jump into btrfs_map_bio
851 return btree_csum_one_bio(bio);
854 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
857 struct inode *inode = private_data;
861 * when we're called for a write, we're already in the async
862 * submission context. Just jump into btrfs_map_bio
864 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
866 bio->bi_status = ret;
872 static int check_async_write(struct btrfs_inode *bi)
874 if (atomic_read(&bi->sync_writers))
877 if (static_cpu_has(X86_FEATURE_XMM4_2))
883 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
884 int mirror_num, unsigned long bio_flags,
887 struct inode *inode = private_data;
888 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
889 int async = check_async_write(BTRFS_I(inode));
892 if (bio_op(bio) != REQ_OP_WRITE) {
894 * called for a read, do the setup so that checksum validation
895 * can happen in the async kernel threads
897 ret = btrfs_bio_wq_end_io(fs_info, bio,
898 BTRFS_WQ_ENDIO_METADATA);
901 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
903 ret = btree_csum_one_bio(bio);
906 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
909 * kthread helpers are used to submit writes so that
910 * checksumming can happen in parallel across all CPUs
912 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
913 bio_offset, private_data,
914 btree_submit_bio_start,
915 btree_submit_bio_done);
923 bio->bi_status = ret;
928 #ifdef CONFIG_MIGRATION
929 static int btree_migratepage(struct address_space *mapping,
930 struct page *newpage, struct page *page,
931 enum migrate_mode mode)
934 * we can't safely write a btree page from here,
935 * we haven't done the locking hook
940 * Buffers may be managed in a filesystem specific way.
941 * We must have no buffers or drop them.
943 if (page_has_private(page) &&
944 !try_to_release_page(page, GFP_KERNEL))
946 return migrate_page(mapping, newpage, page, mode);
951 static int btree_writepages(struct address_space *mapping,
952 struct writeback_control *wbc)
954 struct btrfs_fs_info *fs_info;
957 if (wbc->sync_mode == WB_SYNC_NONE) {
959 if (wbc->for_kupdate)
962 fs_info = BTRFS_I(mapping->host)->root->fs_info;
963 /* this is a bit racy, but that's ok */
964 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
965 BTRFS_DIRTY_METADATA_THRESH);
969 return btree_write_cache_pages(mapping, wbc);
972 static int btree_readpage(struct file *file, struct page *page)
974 struct extent_io_tree *tree;
975 tree = &BTRFS_I(page->mapping->host)->io_tree;
976 return extent_read_full_page(tree, page, btree_get_extent, 0);
979 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
981 if (PageWriteback(page) || PageDirty(page))
984 return try_release_extent_buffer(page);
987 static void btree_invalidatepage(struct page *page, unsigned int offset,
990 struct extent_io_tree *tree;
991 tree = &BTRFS_I(page->mapping->host)->io_tree;
992 extent_invalidatepage(tree, page, offset);
993 btree_releasepage(page, GFP_NOFS);
994 if (PagePrivate(page)) {
995 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
996 "page private not zero on page %llu",
997 (unsigned long long)page_offset(page));
998 ClearPagePrivate(page);
999 set_page_private(page, 0);
1004 static int btree_set_page_dirty(struct page *page)
1007 struct extent_buffer *eb;
1009 BUG_ON(!PagePrivate(page));
1010 eb = (struct extent_buffer *)page->private;
1012 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1013 BUG_ON(!atomic_read(&eb->refs));
1014 btrfs_assert_tree_locked(eb);
1016 return __set_page_dirty_nobuffers(page);
1019 static const struct address_space_operations btree_aops = {
1020 .readpage = btree_readpage,
1021 .writepages = btree_writepages,
1022 .releasepage = btree_releasepage,
1023 .invalidatepage = btree_invalidatepage,
1024 #ifdef CONFIG_MIGRATION
1025 .migratepage = btree_migratepage,
1027 .set_page_dirty = btree_set_page_dirty,
1030 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1032 struct extent_buffer *buf = NULL;
1033 struct inode *btree_inode = fs_info->btree_inode;
1035 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1038 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1040 free_extent_buffer(buf);
1043 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1044 int mirror_num, struct extent_buffer **eb)
1046 struct extent_buffer *buf = NULL;
1047 struct inode *btree_inode = fs_info->btree_inode;
1048 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1051 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1055 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1057 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1060 free_extent_buffer(buf);
1064 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1065 free_extent_buffer(buf);
1067 } else if (extent_buffer_uptodate(buf)) {
1070 free_extent_buffer(buf);
1075 struct extent_buffer *btrfs_find_create_tree_block(
1076 struct btrfs_fs_info *fs_info,
1079 if (btrfs_is_testing(fs_info))
1080 return alloc_test_extent_buffer(fs_info, bytenr);
1081 return alloc_extent_buffer(fs_info, bytenr);
1085 int btrfs_write_tree_block(struct extent_buffer *buf)
1087 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1088 buf->start + buf->len - 1);
1091 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1093 filemap_fdatawait_range(buf->pages[0]->mapping,
1094 buf->start, buf->start + buf->len - 1);
1098 * Read tree block at logical address @bytenr and do variant basic but critical
1101 * @parent_transid: expected transid of this tree block, skip check if 0
1102 * @level: expected level, mandatory check
1103 * @first_key: expected key in slot 0, skip check if NULL
1105 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1106 u64 parent_transid, int level,
1107 struct btrfs_key *first_key)
1109 struct extent_buffer *buf = NULL;
1112 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1116 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1119 free_extent_buffer(buf);
1120 return ERR_PTR(ret);
1126 void clean_tree_block(struct btrfs_fs_info *fs_info,
1127 struct extent_buffer *buf)
1129 if (btrfs_header_generation(buf) ==
1130 fs_info->running_transaction->transid) {
1131 btrfs_assert_tree_locked(buf);
1133 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1134 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1136 fs_info->dirty_metadata_batch);
1137 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1138 btrfs_set_lock_blocking(buf);
1139 clear_extent_buffer_dirty(buf);
1144 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1146 struct btrfs_subvolume_writers *writers;
1149 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1151 return ERR_PTR(-ENOMEM);
1153 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1156 return ERR_PTR(ret);
1159 init_waitqueue_head(&writers->wait);
1164 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1166 percpu_counter_destroy(&writers->counter);
1170 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1173 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1175 root->commit_root = NULL;
1177 root->orphan_cleanup_state = 0;
1179 root->objectid = objectid;
1180 root->last_trans = 0;
1181 root->highest_objectid = 0;
1182 root->nr_delalloc_inodes = 0;
1183 root->nr_ordered_extents = 0;
1185 root->inode_tree = RB_ROOT;
1186 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1187 root->block_rsv = NULL;
1188 root->orphan_block_rsv = NULL;
1190 INIT_LIST_HEAD(&root->dirty_list);
1191 INIT_LIST_HEAD(&root->root_list);
1192 INIT_LIST_HEAD(&root->delalloc_inodes);
1193 INIT_LIST_HEAD(&root->delalloc_root);
1194 INIT_LIST_HEAD(&root->ordered_extents);
1195 INIT_LIST_HEAD(&root->ordered_root);
1196 INIT_LIST_HEAD(&root->logged_list[0]);
1197 INIT_LIST_HEAD(&root->logged_list[1]);
1198 spin_lock_init(&root->orphan_lock);
1199 spin_lock_init(&root->inode_lock);
1200 spin_lock_init(&root->delalloc_lock);
1201 spin_lock_init(&root->ordered_extent_lock);
1202 spin_lock_init(&root->accounting_lock);
1203 spin_lock_init(&root->log_extents_lock[0]);
1204 spin_lock_init(&root->log_extents_lock[1]);
1205 spin_lock_init(&root->qgroup_meta_rsv_lock);
1206 mutex_init(&root->objectid_mutex);
1207 mutex_init(&root->log_mutex);
1208 mutex_init(&root->ordered_extent_mutex);
1209 mutex_init(&root->delalloc_mutex);
1210 init_waitqueue_head(&root->log_writer_wait);
1211 init_waitqueue_head(&root->log_commit_wait[0]);
1212 init_waitqueue_head(&root->log_commit_wait[1]);
1213 INIT_LIST_HEAD(&root->log_ctxs[0]);
1214 INIT_LIST_HEAD(&root->log_ctxs[1]);
1215 atomic_set(&root->log_commit[0], 0);
1216 atomic_set(&root->log_commit[1], 0);
1217 atomic_set(&root->log_writers, 0);
1218 atomic_set(&root->log_batch, 0);
1219 atomic_set(&root->orphan_inodes, 0);
1220 refcount_set(&root->refs, 1);
1221 atomic_set(&root->will_be_snapshotted, 0);
1222 root->log_transid = 0;
1223 root->log_transid_committed = -1;
1224 root->last_log_commit = 0;
1226 extent_io_tree_init(&root->dirty_log_pages, NULL);
1228 memset(&root->root_key, 0, sizeof(root->root_key));
1229 memset(&root->root_item, 0, sizeof(root->root_item));
1230 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1232 root->defrag_trans_start = fs_info->generation;
1234 root->defrag_trans_start = 0;
1235 root->root_key.objectid = objectid;
1238 spin_lock_init(&root->root_item_lock);
1241 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1244 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1246 root->fs_info = fs_info;
1250 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1251 /* Should only be used by the testing infrastructure */
1252 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1254 struct btrfs_root *root;
1257 return ERR_PTR(-EINVAL);
1259 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1261 return ERR_PTR(-ENOMEM);
1263 /* We don't use the stripesize in selftest, set it as sectorsize */
1264 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1265 root->alloc_bytenr = 0;
1271 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1272 struct btrfs_fs_info *fs_info,
1275 struct extent_buffer *leaf;
1276 struct btrfs_root *tree_root = fs_info->tree_root;
1277 struct btrfs_root *root;
1278 struct btrfs_key key;
1280 uuid_le uuid = NULL_UUID_LE;
1282 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1284 return ERR_PTR(-ENOMEM);
1286 __setup_root(root, fs_info, objectid);
1287 root->root_key.objectid = objectid;
1288 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1289 root->root_key.offset = 0;
1291 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1293 ret = PTR_ERR(leaf);
1298 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1299 btrfs_set_header_bytenr(leaf, leaf->start);
1300 btrfs_set_header_generation(leaf, trans->transid);
1301 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1302 btrfs_set_header_owner(leaf, objectid);
1305 write_extent_buffer_fsid(leaf, fs_info->fsid);
1306 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1307 btrfs_mark_buffer_dirty(leaf);
1309 root->commit_root = btrfs_root_node(root);
1310 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1312 root->root_item.flags = 0;
1313 root->root_item.byte_limit = 0;
1314 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1315 btrfs_set_root_generation(&root->root_item, trans->transid);
1316 btrfs_set_root_level(&root->root_item, 0);
1317 btrfs_set_root_refs(&root->root_item, 1);
1318 btrfs_set_root_used(&root->root_item, leaf->len);
1319 btrfs_set_root_last_snapshot(&root->root_item, 0);
1320 btrfs_set_root_dirid(&root->root_item, 0);
1321 if (is_fstree(objectid))
1323 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1324 root->root_item.drop_level = 0;
1326 key.objectid = objectid;
1327 key.type = BTRFS_ROOT_ITEM_KEY;
1329 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1333 btrfs_tree_unlock(leaf);
1339 btrfs_tree_unlock(leaf);
1340 free_extent_buffer(root->commit_root);
1341 free_extent_buffer(leaf);
1345 return ERR_PTR(ret);
1348 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1349 struct btrfs_fs_info *fs_info)
1351 struct btrfs_root *root;
1352 struct extent_buffer *leaf;
1354 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1356 return ERR_PTR(-ENOMEM);
1358 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1360 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1361 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1362 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1365 * DON'T set REF_COWS for log trees
1367 * log trees do not get reference counted because they go away
1368 * before a real commit is actually done. They do store pointers
1369 * to file data extents, and those reference counts still get
1370 * updated (along with back refs to the log tree).
1373 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1377 return ERR_CAST(leaf);
1380 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1381 btrfs_set_header_bytenr(leaf, leaf->start);
1382 btrfs_set_header_generation(leaf, trans->transid);
1383 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1384 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1387 write_extent_buffer_fsid(root->node, fs_info->fsid);
1388 btrfs_mark_buffer_dirty(root->node);
1389 btrfs_tree_unlock(root->node);
1393 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1394 struct btrfs_fs_info *fs_info)
1396 struct btrfs_root *log_root;
1398 log_root = alloc_log_tree(trans, fs_info);
1399 if (IS_ERR(log_root))
1400 return PTR_ERR(log_root);
1401 WARN_ON(fs_info->log_root_tree);
1402 fs_info->log_root_tree = log_root;
1406 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1407 struct btrfs_root *root)
1409 struct btrfs_fs_info *fs_info = root->fs_info;
1410 struct btrfs_root *log_root;
1411 struct btrfs_inode_item *inode_item;
1413 log_root = alloc_log_tree(trans, fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1417 log_root->last_trans = trans->transid;
1418 log_root->root_key.offset = root->root_key.objectid;
1420 inode_item = &log_root->root_item.inode;
1421 btrfs_set_stack_inode_generation(inode_item, 1);
1422 btrfs_set_stack_inode_size(inode_item, 3);
1423 btrfs_set_stack_inode_nlink(inode_item, 1);
1424 btrfs_set_stack_inode_nbytes(inode_item,
1426 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1428 btrfs_set_root_node(&log_root->root_item, log_root->node);
1430 WARN_ON(root->log_root);
1431 root->log_root = log_root;
1432 root->log_transid = 0;
1433 root->log_transid_committed = -1;
1434 root->last_log_commit = 0;
1438 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1439 struct btrfs_key *key)
1441 struct btrfs_root *root;
1442 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1443 struct btrfs_path *path;
1448 path = btrfs_alloc_path();
1450 return ERR_PTR(-ENOMEM);
1452 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1458 __setup_root(root, fs_info, key->objectid);
1460 ret = btrfs_find_root(tree_root, key, path,
1461 &root->root_item, &root->root_key);
1468 generation = btrfs_root_generation(&root->root_item);
1469 level = btrfs_root_level(&root->root_item);
1470 root->node = read_tree_block(fs_info,
1471 btrfs_root_bytenr(&root->root_item),
1472 generation, level, NULL);
1473 if (IS_ERR(root->node)) {
1474 ret = PTR_ERR(root->node);
1476 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1478 free_extent_buffer(root->node);
1481 root->commit_root = btrfs_root_node(root);
1483 btrfs_free_path(path);
1489 root = ERR_PTR(ret);
1493 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1494 struct btrfs_key *location)
1496 struct btrfs_root *root;
1498 root = btrfs_read_tree_root(tree_root, location);
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1503 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1504 btrfs_check_and_init_root_item(&root->root_item);
1510 int btrfs_init_fs_root(struct btrfs_root *root)
1513 struct btrfs_subvolume_writers *writers;
1515 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1516 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1518 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1523 writers = btrfs_alloc_subvolume_writers();
1524 if (IS_ERR(writers)) {
1525 ret = PTR_ERR(writers);
1528 root->subv_writers = writers;
1530 btrfs_init_free_ino_ctl(root);
1531 spin_lock_init(&root->ino_cache_lock);
1532 init_waitqueue_head(&root->ino_cache_wait);
1534 ret = get_anon_bdev(&root->anon_dev);
1538 mutex_lock(&root->objectid_mutex);
1539 ret = btrfs_find_highest_objectid(root,
1540 &root->highest_objectid);
1542 mutex_unlock(&root->objectid_mutex);
1546 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1548 mutex_unlock(&root->objectid_mutex);
1552 /* the caller is responsible to call free_fs_root */
1556 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1559 struct btrfs_root *root;
1561 spin_lock(&fs_info->fs_roots_radix_lock);
1562 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1563 (unsigned long)root_id);
1564 spin_unlock(&fs_info->fs_roots_radix_lock);
1568 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1569 struct btrfs_root *root)
1573 ret = radix_tree_preload(GFP_NOFS);
1577 spin_lock(&fs_info->fs_roots_radix_lock);
1578 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1579 (unsigned long)root->root_key.objectid,
1582 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1583 spin_unlock(&fs_info->fs_roots_radix_lock);
1584 radix_tree_preload_end();
1589 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1590 struct btrfs_key *location,
1593 struct btrfs_root *root;
1594 struct btrfs_path *path;
1595 struct btrfs_key key;
1598 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1599 return fs_info->tree_root;
1600 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1601 return fs_info->extent_root;
1602 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1603 return fs_info->chunk_root;
1604 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1605 return fs_info->dev_root;
1606 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1607 return fs_info->csum_root;
1608 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1609 return fs_info->quota_root ? fs_info->quota_root :
1611 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1612 return fs_info->uuid_root ? fs_info->uuid_root :
1614 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1615 return fs_info->free_space_root ? fs_info->free_space_root :
1618 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1620 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1621 return ERR_PTR(-ENOENT);
1625 root = btrfs_read_fs_root(fs_info->tree_root, location);
1629 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1634 ret = btrfs_init_fs_root(root);
1638 path = btrfs_alloc_path();
1643 key.objectid = BTRFS_ORPHAN_OBJECTID;
1644 key.type = BTRFS_ORPHAN_ITEM_KEY;
1645 key.offset = location->objectid;
1647 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1648 btrfs_free_path(path);
1652 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1654 ret = btrfs_insert_fs_root(fs_info, root);
1656 if (ret == -EEXIST) {
1665 return ERR_PTR(ret);
1668 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1670 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1672 struct btrfs_device *device;
1673 struct backing_dev_info *bdi;
1676 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1679 bdi = device->bdev->bd_bdi;
1680 if (bdi_congested(bdi, bdi_bits)) {
1690 * called by the kthread helper functions to finally call the bio end_io
1691 * functions. This is where read checksum verification actually happens
1693 static void end_workqueue_fn(struct btrfs_work *work)
1696 struct btrfs_end_io_wq *end_io_wq;
1698 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1699 bio = end_io_wq->bio;
1701 bio->bi_status = end_io_wq->status;
1702 bio->bi_private = end_io_wq->private;
1703 bio->bi_end_io = end_io_wq->end_io;
1704 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1708 static int cleaner_kthread(void *arg)
1710 struct btrfs_root *root = arg;
1711 struct btrfs_fs_info *fs_info = root->fs_info;
1713 struct btrfs_trans_handle *trans;
1718 /* Make the cleaner go to sleep early. */
1719 if (btrfs_need_cleaner_sleep(fs_info))
1723 * Do not do anything if we might cause open_ctree() to block
1724 * before we have finished mounting the filesystem.
1726 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1729 if (!mutex_trylock(&fs_info->cleaner_mutex))
1733 * Avoid the problem that we change the status of the fs
1734 * during the above check and trylock.
1736 if (btrfs_need_cleaner_sleep(fs_info)) {
1737 mutex_unlock(&fs_info->cleaner_mutex);
1741 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1742 btrfs_run_delayed_iputs(fs_info);
1743 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1745 again = btrfs_clean_one_deleted_snapshot(root);
1746 mutex_unlock(&fs_info->cleaner_mutex);
1749 * The defragger has dealt with the R/O remount and umount,
1750 * needn't do anything special here.
1752 btrfs_run_defrag_inodes(fs_info);
1755 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1756 * with relocation (btrfs_relocate_chunk) and relocation
1757 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1758 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1759 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1760 * unused block groups.
1762 btrfs_delete_unused_bgs(fs_info);
1765 set_current_state(TASK_INTERRUPTIBLE);
1766 if (!kthread_should_stop())
1768 __set_current_state(TASK_RUNNING);
1770 } while (!kthread_should_stop());
1773 * Transaction kthread is stopped before us and wakes us up.
1774 * However we might have started a new transaction and COWed some
1775 * tree blocks when deleting unused block groups for example. So
1776 * make sure we commit the transaction we started to have a clean
1777 * shutdown when evicting the btree inode - if it has dirty pages
1778 * when we do the final iput() on it, eviction will trigger a
1779 * writeback for it which will fail with null pointer dereferences
1780 * since work queues and other resources were already released and
1781 * destroyed by the time the iput/eviction/writeback is made.
1783 trans = btrfs_attach_transaction(root);
1784 if (IS_ERR(trans)) {
1785 if (PTR_ERR(trans) != -ENOENT)
1787 "cleaner transaction attach returned %ld",
1792 ret = btrfs_commit_transaction(trans);
1795 "cleaner open transaction commit returned %d",
1802 static int transaction_kthread(void *arg)
1804 struct btrfs_root *root = arg;
1805 struct btrfs_fs_info *fs_info = root->fs_info;
1806 struct btrfs_trans_handle *trans;
1807 struct btrfs_transaction *cur;
1810 unsigned long delay;
1814 cannot_commit = false;
1815 delay = HZ * fs_info->commit_interval;
1816 mutex_lock(&fs_info->transaction_kthread_mutex);
1818 spin_lock(&fs_info->trans_lock);
1819 cur = fs_info->running_transaction;
1821 spin_unlock(&fs_info->trans_lock);
1825 now = get_seconds();
1826 if (cur->state < TRANS_STATE_BLOCKED &&
1827 (now < cur->start_time ||
1828 now - cur->start_time < fs_info->commit_interval)) {
1829 spin_unlock(&fs_info->trans_lock);
1833 transid = cur->transid;
1834 spin_unlock(&fs_info->trans_lock);
1836 /* If the file system is aborted, this will always fail. */
1837 trans = btrfs_attach_transaction(root);
1838 if (IS_ERR(trans)) {
1839 if (PTR_ERR(trans) != -ENOENT)
1840 cannot_commit = true;
1843 if (transid == trans->transid) {
1844 btrfs_commit_transaction(trans);
1846 btrfs_end_transaction(trans);
1849 wake_up_process(fs_info->cleaner_kthread);
1850 mutex_unlock(&fs_info->transaction_kthread_mutex);
1852 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1853 &fs_info->fs_state)))
1854 btrfs_cleanup_transaction(fs_info);
1855 if (!kthread_should_stop() &&
1856 (!btrfs_transaction_blocked(fs_info) ||
1858 schedule_timeout_interruptible(delay);
1859 } while (!kthread_should_stop());
1864 * this will find the highest generation in the array of
1865 * root backups. The index of the highest array is returned,
1866 * or -1 if we can't find anything.
1868 * We check to make sure the array is valid by comparing the
1869 * generation of the latest root in the array with the generation
1870 * in the super block. If they don't match we pitch it.
1872 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1875 int newest_index = -1;
1876 struct btrfs_root_backup *root_backup;
1879 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1880 root_backup = info->super_copy->super_roots + i;
1881 cur = btrfs_backup_tree_root_gen(root_backup);
1882 if (cur == newest_gen)
1886 /* check to see if we actually wrapped around */
1887 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1888 root_backup = info->super_copy->super_roots;
1889 cur = btrfs_backup_tree_root_gen(root_backup);
1890 if (cur == newest_gen)
1893 return newest_index;
1898 * find the oldest backup so we know where to store new entries
1899 * in the backup array. This will set the backup_root_index
1900 * field in the fs_info struct
1902 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1905 int newest_index = -1;
1907 newest_index = find_newest_super_backup(info, newest_gen);
1908 /* if there was garbage in there, just move along */
1909 if (newest_index == -1) {
1910 info->backup_root_index = 0;
1912 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1917 * copy all the root pointers into the super backup array.
1918 * this will bump the backup pointer by one when it is
1921 static void backup_super_roots(struct btrfs_fs_info *info)
1924 struct btrfs_root_backup *root_backup;
1927 next_backup = info->backup_root_index;
1928 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1929 BTRFS_NUM_BACKUP_ROOTS;
1932 * just overwrite the last backup if we're at the same generation
1933 * this happens only at umount
1935 root_backup = info->super_for_commit->super_roots + last_backup;
1936 if (btrfs_backup_tree_root_gen(root_backup) ==
1937 btrfs_header_generation(info->tree_root->node))
1938 next_backup = last_backup;
1940 root_backup = info->super_for_commit->super_roots + next_backup;
1943 * make sure all of our padding and empty slots get zero filled
1944 * regardless of which ones we use today
1946 memset(root_backup, 0, sizeof(*root_backup));
1948 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1950 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1951 btrfs_set_backup_tree_root_gen(root_backup,
1952 btrfs_header_generation(info->tree_root->node));
1954 btrfs_set_backup_tree_root_level(root_backup,
1955 btrfs_header_level(info->tree_root->node));
1957 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1958 btrfs_set_backup_chunk_root_gen(root_backup,
1959 btrfs_header_generation(info->chunk_root->node));
1960 btrfs_set_backup_chunk_root_level(root_backup,
1961 btrfs_header_level(info->chunk_root->node));
1963 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1964 btrfs_set_backup_extent_root_gen(root_backup,
1965 btrfs_header_generation(info->extent_root->node));
1966 btrfs_set_backup_extent_root_level(root_backup,
1967 btrfs_header_level(info->extent_root->node));
1970 * we might commit during log recovery, which happens before we set
1971 * the fs_root. Make sure it is valid before we fill it in.
1973 if (info->fs_root && info->fs_root->node) {
1974 btrfs_set_backup_fs_root(root_backup,
1975 info->fs_root->node->start);
1976 btrfs_set_backup_fs_root_gen(root_backup,
1977 btrfs_header_generation(info->fs_root->node));
1978 btrfs_set_backup_fs_root_level(root_backup,
1979 btrfs_header_level(info->fs_root->node));
1982 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1983 btrfs_set_backup_dev_root_gen(root_backup,
1984 btrfs_header_generation(info->dev_root->node));
1985 btrfs_set_backup_dev_root_level(root_backup,
1986 btrfs_header_level(info->dev_root->node));
1988 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1989 btrfs_set_backup_csum_root_gen(root_backup,
1990 btrfs_header_generation(info->csum_root->node));
1991 btrfs_set_backup_csum_root_level(root_backup,
1992 btrfs_header_level(info->csum_root->node));
1994 btrfs_set_backup_total_bytes(root_backup,
1995 btrfs_super_total_bytes(info->super_copy));
1996 btrfs_set_backup_bytes_used(root_backup,
1997 btrfs_super_bytes_used(info->super_copy));
1998 btrfs_set_backup_num_devices(root_backup,
1999 btrfs_super_num_devices(info->super_copy));
2002 * if we don't copy this out to the super_copy, it won't get remembered
2003 * for the next commit
2005 memcpy(&info->super_copy->super_roots,
2006 &info->super_for_commit->super_roots,
2007 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2011 * this copies info out of the root backup array and back into
2012 * the in-memory super block. It is meant to help iterate through
2013 * the array, so you send it the number of backups you've already
2014 * tried and the last backup index you used.
2016 * this returns -1 when it has tried all the backups
2018 static noinline int next_root_backup(struct btrfs_fs_info *info,
2019 struct btrfs_super_block *super,
2020 int *num_backups_tried, int *backup_index)
2022 struct btrfs_root_backup *root_backup;
2023 int newest = *backup_index;
2025 if (*num_backups_tried == 0) {
2026 u64 gen = btrfs_super_generation(super);
2028 newest = find_newest_super_backup(info, gen);
2032 *backup_index = newest;
2033 *num_backups_tried = 1;
2034 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2035 /* we've tried all the backups, all done */
2038 /* jump to the next oldest backup */
2039 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2040 BTRFS_NUM_BACKUP_ROOTS;
2041 *backup_index = newest;
2042 *num_backups_tried += 1;
2044 root_backup = super->super_roots + newest;
2046 btrfs_set_super_generation(super,
2047 btrfs_backup_tree_root_gen(root_backup));
2048 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2049 btrfs_set_super_root_level(super,
2050 btrfs_backup_tree_root_level(root_backup));
2051 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2054 * fixme: the total bytes and num_devices need to match or we should
2057 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2058 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2062 /* helper to cleanup workers */
2063 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2065 btrfs_destroy_workqueue(fs_info->fixup_workers);
2066 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2067 btrfs_destroy_workqueue(fs_info->workers);
2068 btrfs_destroy_workqueue(fs_info->endio_workers);
2069 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2070 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2071 btrfs_destroy_workqueue(fs_info->rmw_workers);
2072 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2073 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2074 btrfs_destroy_workqueue(fs_info->submit_workers);
2075 btrfs_destroy_workqueue(fs_info->delayed_workers);
2076 btrfs_destroy_workqueue(fs_info->caching_workers);
2077 btrfs_destroy_workqueue(fs_info->readahead_workers);
2078 btrfs_destroy_workqueue(fs_info->flush_workers);
2079 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2080 btrfs_destroy_workqueue(fs_info->extent_workers);
2082 * Now that all other work queues are destroyed, we can safely destroy
2083 * the queues used for metadata I/O, since tasks from those other work
2084 * queues can do metadata I/O operations.
2086 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2087 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2090 static void free_root_extent_buffers(struct btrfs_root *root)
2093 free_extent_buffer(root->node);
2094 free_extent_buffer(root->commit_root);
2096 root->commit_root = NULL;
2100 /* helper to cleanup tree roots */
2101 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2103 free_root_extent_buffers(info->tree_root);
2105 free_root_extent_buffers(info->dev_root);
2106 free_root_extent_buffers(info->extent_root);
2107 free_root_extent_buffers(info->csum_root);
2108 free_root_extent_buffers(info->quota_root);
2109 free_root_extent_buffers(info->uuid_root);
2111 free_root_extent_buffers(info->chunk_root);
2112 free_root_extent_buffers(info->free_space_root);
2115 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2118 struct btrfs_root *gang[8];
2121 while (!list_empty(&fs_info->dead_roots)) {
2122 gang[0] = list_entry(fs_info->dead_roots.next,
2123 struct btrfs_root, root_list);
2124 list_del(&gang[0]->root_list);
2126 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2127 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2129 free_extent_buffer(gang[0]->node);
2130 free_extent_buffer(gang[0]->commit_root);
2131 btrfs_put_fs_root(gang[0]);
2136 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2141 for (i = 0; i < ret; i++)
2142 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2145 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2146 btrfs_free_log_root_tree(NULL, fs_info);
2147 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2151 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2153 mutex_init(&fs_info->scrub_lock);
2154 atomic_set(&fs_info->scrubs_running, 0);
2155 atomic_set(&fs_info->scrub_pause_req, 0);
2156 atomic_set(&fs_info->scrubs_paused, 0);
2157 atomic_set(&fs_info->scrub_cancel_req, 0);
2158 init_waitqueue_head(&fs_info->scrub_pause_wait);
2159 fs_info->scrub_workers_refcnt = 0;
2162 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2164 spin_lock_init(&fs_info->balance_lock);
2165 mutex_init(&fs_info->balance_mutex);
2166 atomic_set(&fs_info->balance_running, 0);
2167 atomic_set(&fs_info->balance_pause_req, 0);
2168 atomic_set(&fs_info->balance_cancel_req, 0);
2169 fs_info->balance_ctl = NULL;
2170 init_waitqueue_head(&fs_info->balance_wait_q);
2173 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2175 struct inode *inode = fs_info->btree_inode;
2177 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2178 set_nlink(inode, 1);
2180 * we set the i_size on the btree inode to the max possible int.
2181 * the real end of the address space is determined by all of
2182 * the devices in the system
2184 inode->i_size = OFFSET_MAX;
2185 inode->i_mapping->a_ops = &btree_aops;
2187 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2188 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2189 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2190 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2192 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2194 BTRFS_I(inode)->root = fs_info->tree_root;
2195 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2196 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2197 btrfs_insert_inode_hash(inode);
2200 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2202 fs_info->dev_replace.lock_owner = 0;
2203 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2204 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2205 rwlock_init(&fs_info->dev_replace.lock);
2206 atomic_set(&fs_info->dev_replace.read_locks, 0);
2207 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2208 init_waitqueue_head(&fs_info->replace_wait);
2209 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2212 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2214 spin_lock_init(&fs_info->qgroup_lock);
2215 mutex_init(&fs_info->qgroup_ioctl_lock);
2216 fs_info->qgroup_tree = RB_ROOT;
2217 fs_info->qgroup_op_tree = RB_ROOT;
2218 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2219 fs_info->qgroup_seq = 1;
2220 fs_info->qgroup_ulist = NULL;
2221 fs_info->qgroup_rescan_running = false;
2222 mutex_init(&fs_info->qgroup_rescan_lock);
2225 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2226 struct btrfs_fs_devices *fs_devices)
2228 u32 max_active = fs_info->thread_pool_size;
2229 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2232 btrfs_alloc_workqueue(fs_info, "worker",
2233 flags | WQ_HIGHPRI, max_active, 16);
2235 fs_info->delalloc_workers =
2236 btrfs_alloc_workqueue(fs_info, "delalloc",
2237 flags, max_active, 2);
2239 fs_info->flush_workers =
2240 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2241 flags, max_active, 0);
2243 fs_info->caching_workers =
2244 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2247 * a higher idle thresh on the submit workers makes it much more
2248 * likely that bios will be send down in a sane order to the
2251 fs_info->submit_workers =
2252 btrfs_alloc_workqueue(fs_info, "submit", flags,
2253 min_t(u64, fs_devices->num_devices,
2256 fs_info->fixup_workers =
2257 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2260 * endios are largely parallel and should have a very
2263 fs_info->endio_workers =
2264 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2265 fs_info->endio_meta_workers =
2266 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2268 fs_info->endio_meta_write_workers =
2269 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2271 fs_info->endio_raid56_workers =
2272 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2274 fs_info->endio_repair_workers =
2275 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2276 fs_info->rmw_workers =
2277 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2278 fs_info->endio_write_workers =
2279 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2281 fs_info->endio_freespace_worker =
2282 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2284 fs_info->delayed_workers =
2285 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2287 fs_info->readahead_workers =
2288 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2290 fs_info->qgroup_rescan_workers =
2291 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2292 fs_info->extent_workers =
2293 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2294 min_t(u64, fs_devices->num_devices,
2297 if (!(fs_info->workers && fs_info->delalloc_workers &&
2298 fs_info->submit_workers && fs_info->flush_workers &&
2299 fs_info->endio_workers && fs_info->endio_meta_workers &&
2300 fs_info->endio_meta_write_workers &&
2301 fs_info->endio_repair_workers &&
2302 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2303 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2304 fs_info->caching_workers && fs_info->readahead_workers &&
2305 fs_info->fixup_workers && fs_info->delayed_workers &&
2306 fs_info->extent_workers &&
2307 fs_info->qgroup_rescan_workers)) {
2314 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2315 struct btrfs_fs_devices *fs_devices)
2318 struct btrfs_root *log_tree_root;
2319 struct btrfs_super_block *disk_super = fs_info->super_copy;
2320 u64 bytenr = btrfs_super_log_root(disk_super);
2321 int level = btrfs_super_log_root_level(disk_super);
2323 if (fs_devices->rw_devices == 0) {
2324 btrfs_warn(fs_info, "log replay required on RO media");
2328 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2332 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2334 log_tree_root->node = read_tree_block(fs_info, bytenr,
2335 fs_info->generation + 1,
2337 if (IS_ERR(log_tree_root->node)) {
2338 btrfs_warn(fs_info, "failed to read log tree");
2339 ret = PTR_ERR(log_tree_root->node);
2340 kfree(log_tree_root);
2342 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2343 btrfs_err(fs_info, "failed to read log tree");
2344 free_extent_buffer(log_tree_root->node);
2345 kfree(log_tree_root);
2348 /* returns with log_tree_root freed on success */
2349 ret = btrfs_recover_log_trees(log_tree_root);
2351 btrfs_handle_fs_error(fs_info, ret,
2352 "Failed to recover log tree");
2353 free_extent_buffer(log_tree_root->node);
2354 kfree(log_tree_root);
2358 if (sb_rdonly(fs_info->sb)) {
2359 ret = btrfs_commit_super(fs_info);
2367 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2369 struct btrfs_root *tree_root = fs_info->tree_root;
2370 struct btrfs_root *root;
2371 struct btrfs_key location;
2374 BUG_ON(!fs_info->tree_root);
2376 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2377 location.type = BTRFS_ROOT_ITEM_KEY;
2378 location.offset = 0;
2380 root = btrfs_read_tree_root(tree_root, &location);
2382 ret = PTR_ERR(root);
2385 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2386 fs_info->extent_root = root;
2388 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2389 root = btrfs_read_tree_root(tree_root, &location);
2391 ret = PTR_ERR(root);
2394 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2395 fs_info->dev_root = root;
2396 btrfs_init_devices_late(fs_info);
2398 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2399 root = btrfs_read_tree_root(tree_root, &location);
2401 ret = PTR_ERR(root);
2404 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2405 fs_info->csum_root = root;
2407 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2408 root = btrfs_read_tree_root(tree_root, &location);
2409 if (!IS_ERR(root)) {
2410 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2411 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2412 fs_info->quota_root = root;
2415 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2416 root = btrfs_read_tree_root(tree_root, &location);
2418 ret = PTR_ERR(root);
2422 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2423 fs_info->uuid_root = root;
2426 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2427 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2428 root = btrfs_read_tree_root(tree_root, &location);
2430 ret = PTR_ERR(root);
2433 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2434 fs_info->free_space_root = root;
2439 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2440 location.objectid, ret);
2444 int open_ctree(struct super_block *sb,
2445 struct btrfs_fs_devices *fs_devices,
2453 struct btrfs_key location;
2454 struct buffer_head *bh;
2455 struct btrfs_super_block *disk_super;
2456 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2457 struct btrfs_root *tree_root;
2458 struct btrfs_root *chunk_root;
2461 int num_backups_tried = 0;
2462 int backup_index = 0;
2463 int clear_free_space_tree = 0;
2466 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2467 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2468 if (!tree_root || !chunk_root) {
2473 ret = init_srcu_struct(&fs_info->subvol_srcu);
2479 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2484 fs_info->dirty_metadata_batch = PAGE_SIZE *
2485 (1 + ilog2(nr_cpu_ids));
2487 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2490 goto fail_dirty_metadata_bytes;
2493 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2496 goto fail_delalloc_bytes;
2499 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2500 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2501 INIT_LIST_HEAD(&fs_info->trans_list);
2502 INIT_LIST_HEAD(&fs_info->dead_roots);
2503 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2504 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2505 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2506 INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2507 spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2508 spin_lock_init(&fs_info->delalloc_root_lock);
2509 spin_lock_init(&fs_info->trans_lock);
2510 spin_lock_init(&fs_info->fs_roots_radix_lock);
2511 spin_lock_init(&fs_info->delayed_iput_lock);
2512 spin_lock_init(&fs_info->defrag_inodes_lock);
2513 spin_lock_init(&fs_info->tree_mod_seq_lock);
2514 spin_lock_init(&fs_info->super_lock);
2515 spin_lock_init(&fs_info->qgroup_op_lock);
2516 spin_lock_init(&fs_info->buffer_lock);
2517 spin_lock_init(&fs_info->unused_bgs_lock);
2518 rwlock_init(&fs_info->tree_mod_log_lock);
2519 mutex_init(&fs_info->unused_bg_unpin_mutex);
2520 mutex_init(&fs_info->delete_unused_bgs_mutex);
2521 mutex_init(&fs_info->reloc_mutex);
2522 mutex_init(&fs_info->delalloc_root_mutex);
2523 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2524 seqlock_init(&fs_info->profiles_lock);
2526 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2527 INIT_LIST_HEAD(&fs_info->space_info);
2528 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2529 INIT_LIST_HEAD(&fs_info->unused_bgs);
2530 btrfs_mapping_init(&fs_info->mapping_tree);
2531 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2532 BTRFS_BLOCK_RSV_GLOBAL);
2533 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2534 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2535 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2536 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2537 BTRFS_BLOCK_RSV_DELOPS);
2538 atomic_set(&fs_info->async_delalloc_pages, 0);
2539 atomic_set(&fs_info->defrag_running, 0);
2540 atomic_set(&fs_info->qgroup_op_seq, 0);
2541 atomic_set(&fs_info->reada_works_cnt, 0);
2542 atomic64_set(&fs_info->tree_mod_seq, 0);
2544 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2545 fs_info->metadata_ratio = 0;
2546 fs_info->defrag_inodes = RB_ROOT;
2547 atomic64_set(&fs_info->free_chunk_space, 0);
2548 fs_info->tree_mod_log = RB_ROOT;
2549 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2550 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2551 /* readahead state */
2552 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2553 spin_lock_init(&fs_info->reada_lock);
2554 btrfs_init_ref_verify(fs_info);
2556 fs_info->thread_pool_size = min_t(unsigned long,
2557 num_online_cpus() + 2, 8);
2559 INIT_LIST_HEAD(&fs_info->ordered_roots);
2560 spin_lock_init(&fs_info->ordered_root_lock);
2562 fs_info->btree_inode = new_inode(sb);
2563 if (!fs_info->btree_inode) {
2565 goto fail_bio_counter;
2567 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2569 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2571 if (!fs_info->delayed_root) {
2575 btrfs_init_delayed_root(fs_info->delayed_root);
2577 btrfs_init_scrub(fs_info);
2578 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2579 fs_info->check_integrity_print_mask = 0;
2581 btrfs_init_balance(fs_info);
2582 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2584 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2585 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2587 btrfs_init_btree_inode(fs_info);
2589 spin_lock_init(&fs_info->block_group_cache_lock);
2590 fs_info->block_group_cache_tree = RB_ROOT;
2591 fs_info->first_logical_byte = (u64)-1;
2593 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2594 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2595 fs_info->pinned_extents = &fs_info->freed_extents[0];
2596 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2598 mutex_init(&fs_info->ordered_operations_mutex);
2599 mutex_init(&fs_info->tree_log_mutex);
2600 mutex_init(&fs_info->chunk_mutex);
2601 mutex_init(&fs_info->transaction_kthread_mutex);
2602 mutex_init(&fs_info->cleaner_mutex);
2603 mutex_init(&fs_info->volume_mutex);
2604 mutex_init(&fs_info->ro_block_group_mutex);
2605 init_rwsem(&fs_info->commit_root_sem);
2606 init_rwsem(&fs_info->cleanup_work_sem);
2607 init_rwsem(&fs_info->subvol_sem);
2608 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2610 btrfs_init_dev_replace_locks(fs_info);
2611 btrfs_init_qgroup(fs_info);
2613 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2614 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2616 init_waitqueue_head(&fs_info->transaction_throttle);
2617 init_waitqueue_head(&fs_info->transaction_wait);
2618 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2619 init_waitqueue_head(&fs_info->async_submit_wait);
2621 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2623 /* Usable values until the real ones are cached from the superblock */
2624 fs_info->nodesize = 4096;
2625 fs_info->sectorsize = 4096;
2626 fs_info->stripesize = 4096;
2628 ret = btrfs_alloc_stripe_hash_table(fs_info);
2634 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2636 invalidate_bdev(fs_devices->latest_bdev);
2639 * Read super block and check the signature bytes only
2641 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2648 * We want to check superblock checksum, the type is stored inside.
2649 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2651 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2652 btrfs_err(fs_info, "superblock checksum mismatch");
2659 * super_copy is zeroed at allocation time and we never touch the
2660 * following bytes up to INFO_SIZE, the checksum is calculated from
2661 * the whole block of INFO_SIZE
2663 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2664 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2665 sizeof(*fs_info->super_for_commit));
2668 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2670 ret = btrfs_check_super_valid(fs_info);
2672 btrfs_err(fs_info, "superblock contains fatal errors");
2677 disk_super = fs_info->super_copy;
2678 if (!btrfs_super_root(disk_super))
2681 /* check FS state, whether FS is broken. */
2682 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2683 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2686 * run through our array of backup supers and setup
2687 * our ring pointer to the oldest one
2689 generation = btrfs_super_generation(disk_super);
2690 find_oldest_super_backup(fs_info, generation);
2693 * In the long term, we'll store the compression type in the super
2694 * block, and it'll be used for per file compression control.
2696 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2698 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2704 features = btrfs_super_incompat_flags(disk_super) &
2705 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2708 "cannot mount because of unsupported optional features (%llx)",
2714 features = btrfs_super_incompat_flags(disk_super);
2715 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2716 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2717 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2718 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2719 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2721 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2722 btrfs_info(fs_info, "has skinny extents");
2725 * flag our filesystem as having big metadata blocks if
2726 * they are bigger than the page size
2728 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2729 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2731 "flagging fs with big metadata feature");
2732 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2735 nodesize = btrfs_super_nodesize(disk_super);
2736 sectorsize = btrfs_super_sectorsize(disk_super);
2737 stripesize = sectorsize;
2738 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2739 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2741 /* Cache block sizes */
2742 fs_info->nodesize = nodesize;
2743 fs_info->sectorsize = sectorsize;
2744 fs_info->stripesize = stripesize;
2747 * mixed block groups end up with duplicate but slightly offset
2748 * extent buffers for the same range. It leads to corruptions
2750 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2751 (sectorsize != nodesize)) {
2753 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2754 nodesize, sectorsize);
2759 * Needn't use the lock because there is no other task which will
2762 btrfs_set_super_incompat_flags(disk_super, features);
2764 features = btrfs_super_compat_ro_flags(disk_super) &
2765 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2766 if (!sb_rdonly(sb) && features) {
2768 "cannot mount read-write because of unsupported optional features (%llx)",
2774 ret = btrfs_init_workqueues(fs_info, fs_devices);
2777 goto fail_sb_buffer;
2780 sb->s_bdi->congested_fn = btrfs_congested_fn;
2781 sb->s_bdi->congested_data = fs_info;
2782 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2783 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2784 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2785 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2787 sb->s_blocksize = sectorsize;
2788 sb->s_blocksize_bits = blksize_bits(sectorsize);
2789 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2791 mutex_lock(&fs_info->chunk_mutex);
2792 ret = btrfs_read_sys_array(fs_info);
2793 mutex_unlock(&fs_info->chunk_mutex);
2795 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2796 goto fail_sb_buffer;
2799 generation = btrfs_super_chunk_root_generation(disk_super);
2800 level = btrfs_super_chunk_root_level(disk_super);
2802 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2804 chunk_root->node = read_tree_block(fs_info,
2805 btrfs_super_chunk_root(disk_super),
2806 generation, level, NULL);
2807 if (IS_ERR(chunk_root->node) ||
2808 !extent_buffer_uptodate(chunk_root->node)) {
2809 btrfs_err(fs_info, "failed to read chunk root");
2810 if (!IS_ERR(chunk_root->node))
2811 free_extent_buffer(chunk_root->node);
2812 chunk_root->node = NULL;
2813 goto fail_tree_roots;
2815 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2816 chunk_root->commit_root = btrfs_root_node(chunk_root);
2818 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2819 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2821 ret = btrfs_read_chunk_tree(fs_info);
2823 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2824 goto fail_tree_roots;
2828 * Keep the devid that is marked to be the target device for the
2829 * device replace procedure
2831 btrfs_free_extra_devids(fs_devices, 0);
2833 if (!fs_devices->latest_bdev) {
2834 btrfs_err(fs_info, "failed to read devices");
2835 goto fail_tree_roots;
2839 generation = btrfs_super_generation(disk_super);
2840 level = btrfs_super_root_level(disk_super);
2842 tree_root->node = read_tree_block(fs_info,
2843 btrfs_super_root(disk_super),
2844 generation, level, NULL);
2845 if (IS_ERR(tree_root->node) ||
2846 !extent_buffer_uptodate(tree_root->node)) {
2847 btrfs_warn(fs_info, "failed to read tree root");
2848 if (!IS_ERR(tree_root->node))
2849 free_extent_buffer(tree_root->node);
2850 tree_root->node = NULL;
2851 goto recovery_tree_root;
2854 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2855 tree_root->commit_root = btrfs_root_node(tree_root);
2856 btrfs_set_root_refs(&tree_root->root_item, 1);
2858 mutex_lock(&tree_root->objectid_mutex);
2859 ret = btrfs_find_highest_objectid(tree_root,
2860 &tree_root->highest_objectid);
2862 mutex_unlock(&tree_root->objectid_mutex);
2863 goto recovery_tree_root;
2866 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2868 mutex_unlock(&tree_root->objectid_mutex);
2870 ret = btrfs_read_roots(fs_info);
2872 goto recovery_tree_root;
2874 fs_info->generation = generation;
2875 fs_info->last_trans_committed = generation;
2877 ret = btrfs_recover_balance(fs_info);
2879 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2880 goto fail_block_groups;
2883 ret = btrfs_init_dev_stats(fs_info);
2885 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2886 goto fail_block_groups;
2889 ret = btrfs_init_dev_replace(fs_info);
2891 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2892 goto fail_block_groups;
2895 btrfs_free_extra_devids(fs_devices, 1);
2897 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2899 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2901 goto fail_block_groups;
2904 ret = btrfs_sysfs_add_device(fs_devices);
2906 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2908 goto fail_fsdev_sysfs;
2911 ret = btrfs_sysfs_add_mounted(fs_info);
2913 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2914 goto fail_fsdev_sysfs;
2917 ret = btrfs_init_space_info(fs_info);
2919 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2923 ret = btrfs_read_block_groups(fs_info);
2925 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2929 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
2931 "writeable mount is not allowed due to too many missing devices");
2935 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2937 if (IS_ERR(fs_info->cleaner_kthread))
2940 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2942 "btrfs-transaction");
2943 if (IS_ERR(fs_info->transaction_kthread))
2946 if (!btrfs_test_opt(fs_info, NOSSD) &&
2947 !fs_info->fs_devices->rotating) {
2948 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2952 * Mount does not set all options immediately, we can do it now and do
2953 * not have to wait for transaction commit
2955 btrfs_apply_pending_changes(fs_info);
2957 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2958 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2959 ret = btrfsic_mount(fs_info, fs_devices,
2960 btrfs_test_opt(fs_info,
2961 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2963 fs_info->check_integrity_print_mask);
2966 "failed to initialize integrity check module: %d",
2970 ret = btrfs_read_qgroup_config(fs_info);
2972 goto fail_trans_kthread;
2974 if (btrfs_build_ref_tree(fs_info))
2975 btrfs_err(fs_info, "couldn't build ref tree");
2977 /* do not make disk changes in broken FS or nologreplay is given */
2978 if (btrfs_super_log_root(disk_super) != 0 &&
2979 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2980 ret = btrfs_replay_log(fs_info, fs_devices);
2987 ret = btrfs_find_orphan_roots(fs_info);
2991 if (!sb_rdonly(sb)) {
2992 ret = btrfs_cleanup_fs_roots(fs_info);
2996 mutex_lock(&fs_info->cleaner_mutex);
2997 ret = btrfs_recover_relocation(tree_root);
2998 mutex_unlock(&fs_info->cleaner_mutex);
3000 btrfs_warn(fs_info, "failed to recover relocation: %d",
3007 location.objectid = BTRFS_FS_TREE_OBJECTID;
3008 location.type = BTRFS_ROOT_ITEM_KEY;
3009 location.offset = 0;
3011 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3012 if (IS_ERR(fs_info->fs_root)) {
3013 err = PTR_ERR(fs_info->fs_root);
3014 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3021 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3022 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3023 clear_free_space_tree = 1;
3024 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3025 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3026 btrfs_warn(fs_info, "free space tree is invalid");
3027 clear_free_space_tree = 1;
3030 if (clear_free_space_tree) {
3031 btrfs_info(fs_info, "clearing free space tree");
3032 ret = btrfs_clear_free_space_tree(fs_info);
3035 "failed to clear free space tree: %d", ret);
3036 close_ctree(fs_info);
3041 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3042 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3043 btrfs_info(fs_info, "creating free space tree");
3044 ret = btrfs_create_free_space_tree(fs_info);
3047 "failed to create free space tree: %d", ret);
3048 close_ctree(fs_info);
3053 down_read(&fs_info->cleanup_work_sem);
3054 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3055 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3056 up_read(&fs_info->cleanup_work_sem);
3057 close_ctree(fs_info);
3060 up_read(&fs_info->cleanup_work_sem);
3062 ret = btrfs_resume_balance_async(fs_info);
3064 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3065 close_ctree(fs_info);
3069 ret = btrfs_resume_dev_replace_async(fs_info);
3071 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3072 close_ctree(fs_info);
3076 btrfs_qgroup_rescan_resume(fs_info);
3078 if (!fs_info->uuid_root) {
3079 btrfs_info(fs_info, "creating UUID tree");
3080 ret = btrfs_create_uuid_tree(fs_info);
3083 "failed to create the UUID tree: %d", ret);
3084 close_ctree(fs_info);
3087 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3088 fs_info->generation !=
3089 btrfs_super_uuid_tree_generation(disk_super)) {
3090 btrfs_info(fs_info, "checking UUID tree");
3091 ret = btrfs_check_uuid_tree(fs_info);
3094 "failed to check the UUID tree: %d", ret);
3095 close_ctree(fs_info);
3099 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3101 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3104 * backuproot only affect mount behavior, and if open_ctree succeeded,
3105 * no need to keep the flag
3107 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3112 btrfs_free_qgroup_config(fs_info);
3114 kthread_stop(fs_info->transaction_kthread);
3115 btrfs_cleanup_transaction(fs_info);
3116 btrfs_free_fs_roots(fs_info);
3118 kthread_stop(fs_info->cleaner_kthread);
3121 * make sure we're done with the btree inode before we stop our
3124 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3127 btrfs_sysfs_remove_mounted(fs_info);
3130 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3133 btrfs_put_block_group_cache(fs_info);
3136 free_root_pointers(fs_info, 1);
3137 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3140 btrfs_stop_all_workers(fs_info);
3141 btrfs_free_block_groups(fs_info);
3144 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3146 iput(fs_info->btree_inode);
3148 percpu_counter_destroy(&fs_info->bio_counter);
3149 fail_delalloc_bytes:
3150 percpu_counter_destroy(&fs_info->delalloc_bytes);
3151 fail_dirty_metadata_bytes:
3152 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3154 cleanup_srcu_struct(&fs_info->subvol_srcu);
3156 btrfs_free_stripe_hash_table(fs_info);
3157 btrfs_close_devices(fs_info->fs_devices);
3161 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3162 goto fail_tree_roots;
3164 free_root_pointers(fs_info, 0);
3166 /* don't use the log in recovery mode, it won't be valid */
3167 btrfs_set_super_log_root(disk_super, 0);
3169 /* we can't trust the free space cache either */
3170 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3172 ret = next_root_backup(fs_info, fs_info->super_copy,
3173 &num_backups_tried, &backup_index);
3175 goto fail_block_groups;
3176 goto retry_root_backup;
3178 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3180 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3183 set_buffer_uptodate(bh);
3185 struct btrfs_device *device = (struct btrfs_device *)
3188 btrfs_warn_rl_in_rcu(device->fs_info,
3189 "lost page write due to IO error on %s",
3190 rcu_str_deref(device->name));
3191 /* note, we don't set_buffer_write_io_error because we have
3192 * our own ways of dealing with the IO errors
3194 clear_buffer_uptodate(bh);
3195 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3201 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3202 struct buffer_head **bh_ret)
3204 struct buffer_head *bh;
3205 struct btrfs_super_block *super;
3208 bytenr = btrfs_sb_offset(copy_num);
3209 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3212 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3214 * If we fail to read from the underlying devices, as of now
3215 * the best option we have is to mark it EIO.
3220 super = (struct btrfs_super_block *)bh->b_data;
3221 if (btrfs_super_bytenr(super) != bytenr ||
3222 btrfs_super_magic(super) != BTRFS_MAGIC) {
3232 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3234 struct buffer_head *bh;
3235 struct buffer_head *latest = NULL;
3236 struct btrfs_super_block *super;
3241 /* we would like to check all the supers, but that would make
3242 * a btrfs mount succeed after a mkfs from a different FS.
3243 * So, we need to add a special mount option to scan for
3244 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3246 for (i = 0; i < 1; i++) {
3247 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3251 super = (struct btrfs_super_block *)bh->b_data;
3253 if (!latest || btrfs_super_generation(super) > transid) {
3256 transid = btrfs_super_generation(super);
3263 return ERR_PTR(ret);
3269 * Write superblock @sb to the @device. Do not wait for completion, all the
3270 * buffer heads we write are pinned.
3272 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3273 * the expected device size at commit time. Note that max_mirrors must be
3274 * same for write and wait phases.
3276 * Return number of errors when buffer head is not found or submission fails.
3278 static int write_dev_supers(struct btrfs_device *device,
3279 struct btrfs_super_block *sb, int max_mirrors)
3281 struct buffer_head *bh;
3289 if (max_mirrors == 0)
3290 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3292 for (i = 0; i < max_mirrors; i++) {
3293 bytenr = btrfs_sb_offset(i);
3294 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3295 device->commit_total_bytes)
3298 btrfs_set_super_bytenr(sb, bytenr);
3301 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3302 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3303 btrfs_csum_final(crc, sb->csum);
3305 /* One reference for us, and we leave it for the caller */
3306 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3307 BTRFS_SUPER_INFO_SIZE);
3309 btrfs_err(device->fs_info,
3310 "couldn't get super buffer head for bytenr %llu",
3316 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3318 /* one reference for submit_bh */
3321 set_buffer_uptodate(bh);
3323 bh->b_end_io = btrfs_end_buffer_write_sync;
3324 bh->b_private = device;
3327 * we fua the first super. The others we allow
3330 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3331 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3332 op_flags |= REQ_FUA;
3333 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3337 return errors < i ? 0 : -1;
3341 * Wait for write completion of superblocks done by write_dev_supers,
3342 * @max_mirrors same for write and wait phases.
3344 * Return number of errors when buffer head is not found or not marked up to
3347 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3349 struct buffer_head *bh;
3352 bool primary_failed = false;
3355 if (max_mirrors == 0)
3356 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3358 for (i = 0; i < max_mirrors; i++) {
3359 bytenr = btrfs_sb_offset(i);
3360 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3361 device->commit_total_bytes)
3364 bh = __find_get_block(device->bdev,
3365 bytenr / BTRFS_BDEV_BLOCKSIZE,
3366 BTRFS_SUPER_INFO_SIZE);
3370 primary_failed = true;
3374 if (!buffer_uptodate(bh)) {
3377 primary_failed = true;
3380 /* drop our reference */
3383 /* drop the reference from the writing run */
3387 /* log error, force error return */
3388 if (primary_failed) {
3389 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3394 return errors < i ? 0 : -1;
3398 * endio for the write_dev_flush, this will wake anyone waiting
3399 * for the barrier when it is done
3401 static void btrfs_end_empty_barrier(struct bio *bio)
3403 complete(bio->bi_private);
3407 * Submit a flush request to the device if it supports it. Error handling is
3408 * done in the waiting counterpart.
3410 static void write_dev_flush(struct btrfs_device *device)
3412 struct request_queue *q = bdev_get_queue(device->bdev);
3413 struct bio *bio = device->flush_bio;
3415 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3419 bio->bi_end_io = btrfs_end_empty_barrier;
3420 bio_set_dev(bio, device->bdev);
3421 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3422 init_completion(&device->flush_wait);
3423 bio->bi_private = &device->flush_wait;
3425 btrfsic_submit_bio(bio);
3426 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3430 * If the flush bio has been submitted by write_dev_flush, wait for it.
3432 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3434 struct bio *bio = device->flush_bio;
3436 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3439 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3440 wait_for_completion_io(&device->flush_wait);
3442 return bio->bi_status;
3445 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3447 if (!btrfs_check_rw_degradable(fs_info, NULL))
3453 * send an empty flush down to each device in parallel,
3454 * then wait for them
3456 static int barrier_all_devices(struct btrfs_fs_info *info)
3458 struct list_head *head;
3459 struct btrfs_device *dev;
3460 int errors_wait = 0;
3463 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3464 /* send down all the barriers */
3465 head = &info->fs_devices->devices;
3466 list_for_each_entry(dev, head, dev_list) {
3467 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3471 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3472 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3475 write_dev_flush(dev);
3476 dev->last_flush_error = BLK_STS_OK;
3479 /* wait for all the barriers */
3480 list_for_each_entry(dev, head, dev_list) {
3481 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3487 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3488 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3491 ret = wait_dev_flush(dev);
3493 dev->last_flush_error = ret;
3494 btrfs_dev_stat_inc_and_print(dev,
3495 BTRFS_DEV_STAT_FLUSH_ERRS);
3502 * At some point we need the status of all disks
3503 * to arrive at the volume status. So error checking
3504 * is being pushed to a separate loop.
3506 return check_barrier_error(info);
3511 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3514 int min_tolerated = INT_MAX;
3516 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3517 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3518 min_tolerated = min(min_tolerated,
3519 btrfs_raid_array[BTRFS_RAID_SINGLE].
3520 tolerated_failures);
3522 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3523 if (raid_type == BTRFS_RAID_SINGLE)
3525 if (!(flags & btrfs_raid_group[raid_type]))
3527 min_tolerated = min(min_tolerated,
3528 btrfs_raid_array[raid_type].
3529 tolerated_failures);
3532 if (min_tolerated == INT_MAX) {
3533 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3537 return min_tolerated;
3540 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3542 struct list_head *head;
3543 struct btrfs_device *dev;
3544 struct btrfs_super_block *sb;
3545 struct btrfs_dev_item *dev_item;
3549 int total_errors = 0;
3552 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3555 * max_mirrors == 0 indicates we're from commit_transaction,
3556 * not from fsync where the tree roots in fs_info have not
3557 * been consistent on disk.
3559 if (max_mirrors == 0)
3560 backup_super_roots(fs_info);
3562 sb = fs_info->super_for_commit;
3563 dev_item = &sb->dev_item;
3565 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3566 head = &fs_info->fs_devices->devices;
3567 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3570 ret = barrier_all_devices(fs_info);
3573 &fs_info->fs_devices->device_list_mutex);
3574 btrfs_handle_fs_error(fs_info, ret,
3575 "errors while submitting device barriers.");
3580 list_for_each_entry(dev, head, dev_list) {
3585 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3586 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3589 btrfs_set_stack_device_generation(dev_item, 0);
3590 btrfs_set_stack_device_type(dev_item, dev->type);
3591 btrfs_set_stack_device_id(dev_item, dev->devid);
3592 btrfs_set_stack_device_total_bytes(dev_item,
3593 dev->commit_total_bytes);
3594 btrfs_set_stack_device_bytes_used(dev_item,
3595 dev->commit_bytes_used);
3596 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3597 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3598 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3599 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3600 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3602 flags = btrfs_super_flags(sb);
3603 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3605 ret = write_dev_supers(dev, sb, max_mirrors);
3609 if (total_errors > max_errors) {
3610 btrfs_err(fs_info, "%d errors while writing supers",
3612 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3614 /* FUA is masked off if unsupported and can't be the reason */
3615 btrfs_handle_fs_error(fs_info, -EIO,
3616 "%d errors while writing supers",
3622 list_for_each_entry(dev, head, dev_list) {
3625 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3626 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3629 ret = wait_dev_supers(dev, max_mirrors);
3633 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3634 if (total_errors > max_errors) {
3635 btrfs_handle_fs_error(fs_info, -EIO,
3636 "%d errors while writing supers",
3643 /* Drop a fs root from the radix tree and free it. */
3644 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3645 struct btrfs_root *root)
3647 spin_lock(&fs_info->fs_roots_radix_lock);
3648 radix_tree_delete(&fs_info->fs_roots_radix,
3649 (unsigned long)root->root_key.objectid);
3650 spin_unlock(&fs_info->fs_roots_radix_lock);
3652 if (btrfs_root_refs(&root->root_item) == 0)
3653 synchronize_srcu(&fs_info->subvol_srcu);
3655 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3656 btrfs_free_log(NULL, root);
3657 if (root->reloc_root) {
3658 free_extent_buffer(root->reloc_root->node);
3659 free_extent_buffer(root->reloc_root->commit_root);
3660 btrfs_put_fs_root(root->reloc_root);
3661 root->reloc_root = NULL;
3665 if (root->free_ino_pinned)
3666 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3667 if (root->free_ino_ctl)
3668 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3672 static void free_fs_root(struct btrfs_root *root)
3674 iput(root->ino_cache_inode);
3675 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3676 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3677 root->orphan_block_rsv = NULL;
3679 free_anon_bdev(root->anon_dev);
3680 if (root->subv_writers)
3681 btrfs_free_subvolume_writers(root->subv_writers);
3682 free_extent_buffer(root->node);
3683 free_extent_buffer(root->commit_root);
3684 kfree(root->free_ino_ctl);
3685 kfree(root->free_ino_pinned);
3687 btrfs_put_fs_root(root);
3690 void btrfs_free_fs_root(struct btrfs_root *root)
3695 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3697 u64 root_objectid = 0;
3698 struct btrfs_root *gang[8];
3701 unsigned int ret = 0;
3705 index = srcu_read_lock(&fs_info->subvol_srcu);
3706 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3707 (void **)gang, root_objectid,
3710 srcu_read_unlock(&fs_info->subvol_srcu, index);
3713 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3715 for (i = 0; i < ret; i++) {
3716 /* Avoid to grab roots in dead_roots */
3717 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3721 /* grab all the search result for later use */
3722 gang[i] = btrfs_grab_fs_root(gang[i]);
3724 srcu_read_unlock(&fs_info->subvol_srcu, index);
3726 for (i = 0; i < ret; i++) {
3729 root_objectid = gang[i]->root_key.objectid;
3730 err = btrfs_orphan_cleanup(gang[i]);
3733 btrfs_put_fs_root(gang[i]);
3738 /* release the uncleaned roots due to error */
3739 for (; i < ret; i++) {
3741 btrfs_put_fs_root(gang[i]);
3746 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3748 struct btrfs_root *root = fs_info->tree_root;
3749 struct btrfs_trans_handle *trans;
3751 mutex_lock(&fs_info->cleaner_mutex);
3752 btrfs_run_delayed_iputs(fs_info);
3753 mutex_unlock(&fs_info->cleaner_mutex);
3754 wake_up_process(fs_info->cleaner_kthread);
3756 /* wait until ongoing cleanup work done */
3757 down_write(&fs_info->cleanup_work_sem);
3758 up_write(&fs_info->cleanup_work_sem);
3760 trans = btrfs_join_transaction(root);
3762 return PTR_ERR(trans);
3763 return btrfs_commit_transaction(trans);
3766 void close_ctree(struct btrfs_fs_info *fs_info)
3768 struct btrfs_root *root = fs_info->tree_root;
3771 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3773 /* wait for the qgroup rescan worker to stop */
3774 btrfs_qgroup_wait_for_completion(fs_info, false);
3776 /* wait for the uuid_scan task to finish */
3777 down(&fs_info->uuid_tree_rescan_sem);
3778 /* avoid complains from lockdep et al., set sem back to initial state */
3779 up(&fs_info->uuid_tree_rescan_sem);
3781 /* pause restriper - we want to resume on mount */
3782 btrfs_pause_balance(fs_info);
3784 btrfs_dev_replace_suspend_for_unmount(fs_info);
3786 btrfs_scrub_cancel(fs_info);
3788 /* wait for any defraggers to finish */
3789 wait_event(fs_info->transaction_wait,
3790 (atomic_read(&fs_info->defrag_running) == 0));
3792 /* clear out the rbtree of defraggable inodes */
3793 btrfs_cleanup_defrag_inodes(fs_info);
3795 cancel_work_sync(&fs_info->async_reclaim_work);
3797 if (!sb_rdonly(fs_info->sb)) {
3799 * If the cleaner thread is stopped and there are
3800 * block groups queued for removal, the deletion will be
3801 * skipped when we quit the cleaner thread.
3803 btrfs_delete_unused_bgs(fs_info);
3805 ret = btrfs_commit_super(fs_info);
3807 btrfs_err(fs_info, "commit super ret %d", ret);
3810 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3811 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3812 btrfs_error_commit_super(fs_info);
3814 kthread_stop(fs_info->transaction_kthread);
3815 kthread_stop(fs_info->cleaner_kthread);
3817 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3819 btrfs_free_qgroup_config(fs_info);
3821 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3822 btrfs_info(fs_info, "at unmount delalloc count %lld",
3823 percpu_counter_sum(&fs_info->delalloc_bytes));
3826 btrfs_sysfs_remove_mounted(fs_info);
3827 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3829 btrfs_free_fs_roots(fs_info);
3831 btrfs_put_block_group_cache(fs_info);
3834 * we must make sure there is not any read request to
3835 * submit after we stopping all workers.
3837 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3838 btrfs_stop_all_workers(fs_info);
3840 btrfs_free_block_groups(fs_info);
3842 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3843 free_root_pointers(fs_info, 1);
3845 iput(fs_info->btree_inode);
3847 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3848 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3849 btrfsic_unmount(fs_info->fs_devices);
3852 btrfs_close_devices(fs_info->fs_devices);
3853 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3855 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3856 percpu_counter_destroy(&fs_info->delalloc_bytes);
3857 percpu_counter_destroy(&fs_info->bio_counter);
3858 cleanup_srcu_struct(&fs_info->subvol_srcu);
3860 btrfs_free_stripe_hash_table(fs_info);
3861 btrfs_free_ref_cache(fs_info);
3863 __btrfs_free_block_rsv(root->orphan_block_rsv);
3864 root->orphan_block_rsv = NULL;
3866 while (!list_empty(&fs_info->pinned_chunks)) {
3867 struct extent_map *em;
3869 em = list_first_entry(&fs_info->pinned_chunks,
3870 struct extent_map, list);
3871 list_del_init(&em->list);
3872 free_extent_map(em);
3876 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3880 struct inode *btree_inode = buf->pages[0]->mapping->host;
3882 ret = extent_buffer_uptodate(buf);
3886 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3887 parent_transid, atomic);
3893 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3895 struct btrfs_fs_info *fs_info;
3896 struct btrfs_root *root;
3897 u64 transid = btrfs_header_generation(buf);
3900 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3902 * This is a fast path so only do this check if we have sanity tests
3903 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3904 * outside of the sanity tests.
3906 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3909 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3910 fs_info = root->fs_info;
3911 btrfs_assert_tree_locked(buf);
3912 if (transid != fs_info->generation)
3913 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3914 buf->start, transid, fs_info->generation);
3915 was_dirty = set_extent_buffer_dirty(buf);
3917 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3919 fs_info->dirty_metadata_batch);
3920 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3922 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3923 * but item data not updated.
3924 * So here we should only check item pointers, not item data.
3926 if (btrfs_header_level(buf) == 0 &&
3927 btrfs_check_leaf_relaxed(fs_info, buf)) {
3928 btrfs_print_leaf(buf);
3934 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
3938 * looks as though older kernels can get into trouble with
3939 * this code, they end up stuck in balance_dirty_pages forever
3943 if (current->flags & PF_MEMALLOC)
3947 btrfs_balance_delayed_items(fs_info);
3949 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
3950 BTRFS_DIRTY_METADATA_THRESH);
3952 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
3956 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
3958 __btrfs_btree_balance_dirty(fs_info, 1);
3961 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
3963 __btrfs_btree_balance_dirty(fs_info, 0);
3966 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
3967 struct btrfs_key *first_key)
3969 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3970 struct btrfs_fs_info *fs_info = root->fs_info;
3972 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
3976 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
3978 struct btrfs_super_block *sb = fs_info->super_copy;
3979 u64 nodesize = btrfs_super_nodesize(sb);
3980 u64 sectorsize = btrfs_super_sectorsize(sb);
3983 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
3984 btrfs_err(fs_info, "no valid FS found");
3987 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
3988 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
3989 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
3992 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3993 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
3994 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3997 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3998 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
3999 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4002 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4003 btrfs_err(fs_info, "log_root level too big: %d >= %d",
4004 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4009 * Check sectorsize and nodesize first, other check will need it.
4010 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4012 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4013 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4014 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4017 /* Only PAGE SIZE is supported yet */
4018 if (sectorsize != PAGE_SIZE) {
4020 "sectorsize %llu not supported yet, only support %lu",
4021 sectorsize, PAGE_SIZE);
4024 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4025 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4026 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4029 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4030 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4031 le32_to_cpu(sb->__unused_leafsize), nodesize);
4035 /* Root alignment check */
4036 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4037 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4038 btrfs_super_root(sb));
4041 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4042 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4043 btrfs_super_chunk_root(sb));
4046 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4047 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4048 btrfs_super_log_root(sb));
4052 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
4054 "dev_item UUID does not match fsid: %pU != %pU",
4055 fs_info->fsid, sb->dev_item.fsid);
4060 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4063 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4064 btrfs_err(fs_info, "bytes_used is too small %llu",
4065 btrfs_super_bytes_used(sb));
4068 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4069 btrfs_err(fs_info, "invalid stripesize %u",
4070 btrfs_super_stripesize(sb));
4073 if (btrfs_super_num_devices(sb) > (1UL << 31))
4074 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4075 btrfs_super_num_devices(sb));
4076 if (btrfs_super_num_devices(sb) == 0) {
4077 btrfs_err(fs_info, "number of devices is 0");
4081 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4082 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4083 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4088 * Obvious sys_chunk_array corruptions, it must hold at least one key
4091 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4092 btrfs_err(fs_info, "system chunk array too big %u > %u",
4093 btrfs_super_sys_array_size(sb),
4094 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4097 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4098 + sizeof(struct btrfs_chunk)) {
4099 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4100 btrfs_super_sys_array_size(sb),
4101 sizeof(struct btrfs_disk_key)
4102 + sizeof(struct btrfs_chunk));
4107 * The generation is a global counter, we'll trust it more than the others
4108 * but it's still possible that it's the one that's wrong.
4110 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4112 "suspicious: generation < chunk_root_generation: %llu < %llu",
4113 btrfs_super_generation(sb),
4114 btrfs_super_chunk_root_generation(sb));
4115 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4116 && btrfs_super_cache_generation(sb) != (u64)-1)
4118 "suspicious: generation < cache_generation: %llu < %llu",
4119 btrfs_super_generation(sb),
4120 btrfs_super_cache_generation(sb));
4125 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4127 mutex_lock(&fs_info->cleaner_mutex);
4128 btrfs_run_delayed_iputs(fs_info);
4129 mutex_unlock(&fs_info->cleaner_mutex);
4131 down_write(&fs_info->cleanup_work_sem);
4132 up_write(&fs_info->cleanup_work_sem);
4134 /* cleanup FS via transaction */
4135 btrfs_cleanup_transaction(fs_info);
4138 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4140 struct btrfs_ordered_extent *ordered;
4142 spin_lock(&root->ordered_extent_lock);
4144 * This will just short circuit the ordered completion stuff which will
4145 * make sure the ordered extent gets properly cleaned up.
4147 list_for_each_entry(ordered, &root->ordered_extents,
4149 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4150 spin_unlock(&root->ordered_extent_lock);
4153 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4155 struct btrfs_root *root;
4156 struct list_head splice;
4158 INIT_LIST_HEAD(&splice);
4160 spin_lock(&fs_info->ordered_root_lock);
4161 list_splice_init(&fs_info->ordered_roots, &splice);
4162 while (!list_empty(&splice)) {
4163 root = list_first_entry(&splice, struct btrfs_root,
4165 list_move_tail(&root->ordered_root,
4166 &fs_info->ordered_roots);
4168 spin_unlock(&fs_info->ordered_root_lock);
4169 btrfs_destroy_ordered_extents(root);
4172 spin_lock(&fs_info->ordered_root_lock);
4174 spin_unlock(&fs_info->ordered_root_lock);
4177 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4178 struct btrfs_fs_info *fs_info)
4180 struct rb_node *node;
4181 struct btrfs_delayed_ref_root *delayed_refs;
4182 struct btrfs_delayed_ref_node *ref;
4185 delayed_refs = &trans->delayed_refs;
4187 spin_lock(&delayed_refs->lock);
4188 if (atomic_read(&delayed_refs->num_entries) == 0) {
4189 spin_unlock(&delayed_refs->lock);
4190 btrfs_info(fs_info, "delayed_refs has NO entry");
4194 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4195 struct btrfs_delayed_ref_head *head;
4197 bool pin_bytes = false;
4199 head = rb_entry(node, struct btrfs_delayed_ref_head,
4201 if (!mutex_trylock(&head->mutex)) {
4202 refcount_inc(&head->refs);
4203 spin_unlock(&delayed_refs->lock);
4205 mutex_lock(&head->mutex);
4206 mutex_unlock(&head->mutex);
4207 btrfs_put_delayed_ref_head(head);
4208 spin_lock(&delayed_refs->lock);
4211 spin_lock(&head->lock);
4212 while ((n = rb_first(&head->ref_tree)) != NULL) {
4213 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4216 rb_erase(&ref->ref_node, &head->ref_tree);
4217 RB_CLEAR_NODE(&ref->ref_node);
4218 if (!list_empty(&ref->add_list))
4219 list_del(&ref->add_list);
4220 atomic_dec(&delayed_refs->num_entries);
4221 btrfs_put_delayed_ref(ref);
4223 if (head->must_insert_reserved)
4225 btrfs_free_delayed_extent_op(head->extent_op);
4226 delayed_refs->num_heads--;
4227 if (head->processing == 0)
4228 delayed_refs->num_heads_ready--;
4229 atomic_dec(&delayed_refs->num_entries);
4230 rb_erase(&head->href_node, &delayed_refs->href_root);
4231 RB_CLEAR_NODE(&head->href_node);
4232 spin_unlock(&head->lock);
4233 spin_unlock(&delayed_refs->lock);
4234 mutex_unlock(&head->mutex);
4237 btrfs_pin_extent(fs_info, head->bytenr,
4238 head->num_bytes, 1);
4239 btrfs_put_delayed_ref_head(head);
4241 spin_lock(&delayed_refs->lock);
4244 spin_unlock(&delayed_refs->lock);
4249 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4251 struct btrfs_inode *btrfs_inode;
4252 struct list_head splice;
4254 INIT_LIST_HEAD(&splice);
4256 spin_lock(&root->delalloc_lock);
4257 list_splice_init(&root->delalloc_inodes, &splice);
4259 while (!list_empty(&splice)) {
4260 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4263 list_del_init(&btrfs_inode->delalloc_inodes);
4264 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4265 &btrfs_inode->runtime_flags);
4266 spin_unlock(&root->delalloc_lock);
4268 btrfs_invalidate_inodes(btrfs_inode->root);
4270 spin_lock(&root->delalloc_lock);
4273 spin_unlock(&root->delalloc_lock);
4276 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4278 struct btrfs_root *root;
4279 struct list_head splice;
4281 INIT_LIST_HEAD(&splice);
4283 spin_lock(&fs_info->delalloc_root_lock);
4284 list_splice_init(&fs_info->delalloc_roots, &splice);
4285 while (!list_empty(&splice)) {
4286 root = list_first_entry(&splice, struct btrfs_root,
4288 list_del_init(&root->delalloc_root);
4289 root = btrfs_grab_fs_root(root);
4291 spin_unlock(&fs_info->delalloc_root_lock);
4293 btrfs_destroy_delalloc_inodes(root);
4294 btrfs_put_fs_root(root);
4296 spin_lock(&fs_info->delalloc_root_lock);
4298 spin_unlock(&fs_info->delalloc_root_lock);
4301 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4302 struct extent_io_tree *dirty_pages,
4306 struct extent_buffer *eb;
4311 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4316 clear_extent_bits(dirty_pages, start, end, mark);
4317 while (start <= end) {
4318 eb = find_extent_buffer(fs_info, start);
4319 start += fs_info->nodesize;
4322 wait_on_extent_buffer_writeback(eb);
4324 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4326 clear_extent_buffer_dirty(eb);
4327 free_extent_buffer_stale(eb);
4334 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4335 struct extent_io_tree *pinned_extents)
4337 struct extent_io_tree *unpin;
4343 unpin = pinned_extents;
4346 ret = find_first_extent_bit(unpin, 0, &start, &end,
4347 EXTENT_DIRTY, NULL);
4351 clear_extent_dirty(unpin, start, end);
4352 btrfs_error_unpin_extent_range(fs_info, start, end);
4357 if (unpin == &fs_info->freed_extents[0])
4358 unpin = &fs_info->freed_extents[1];
4360 unpin = &fs_info->freed_extents[0];
4368 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4370 struct inode *inode;
4372 inode = cache->io_ctl.inode;
4374 invalidate_inode_pages2(inode->i_mapping);
4375 BTRFS_I(inode)->generation = 0;
4376 cache->io_ctl.inode = NULL;
4379 btrfs_put_block_group(cache);
4382 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4383 struct btrfs_fs_info *fs_info)
4385 struct btrfs_block_group_cache *cache;
4387 spin_lock(&cur_trans->dirty_bgs_lock);
4388 while (!list_empty(&cur_trans->dirty_bgs)) {
4389 cache = list_first_entry(&cur_trans->dirty_bgs,
4390 struct btrfs_block_group_cache,
4393 if (!list_empty(&cache->io_list)) {
4394 spin_unlock(&cur_trans->dirty_bgs_lock);
4395 list_del_init(&cache->io_list);
4396 btrfs_cleanup_bg_io(cache);
4397 spin_lock(&cur_trans->dirty_bgs_lock);
4400 list_del_init(&cache->dirty_list);
4401 spin_lock(&cache->lock);
4402 cache->disk_cache_state = BTRFS_DC_ERROR;
4403 spin_unlock(&cache->lock);
4405 spin_unlock(&cur_trans->dirty_bgs_lock);
4406 btrfs_put_block_group(cache);
4407 spin_lock(&cur_trans->dirty_bgs_lock);
4409 spin_unlock(&cur_trans->dirty_bgs_lock);
4412 * Refer to the definition of io_bgs member for details why it's safe
4413 * to use it without any locking
4415 while (!list_empty(&cur_trans->io_bgs)) {
4416 cache = list_first_entry(&cur_trans->io_bgs,
4417 struct btrfs_block_group_cache,
4420 list_del_init(&cache->io_list);
4421 spin_lock(&cache->lock);
4422 cache->disk_cache_state = BTRFS_DC_ERROR;
4423 spin_unlock(&cache->lock);
4424 btrfs_cleanup_bg_io(cache);
4428 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4429 struct btrfs_fs_info *fs_info)
4431 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4432 ASSERT(list_empty(&cur_trans->dirty_bgs));
4433 ASSERT(list_empty(&cur_trans->io_bgs));
4435 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4437 cur_trans->state = TRANS_STATE_COMMIT_START;
4438 wake_up(&fs_info->transaction_blocked_wait);
4440 cur_trans->state = TRANS_STATE_UNBLOCKED;
4441 wake_up(&fs_info->transaction_wait);
4443 btrfs_destroy_delayed_inodes(fs_info);
4444 btrfs_assert_delayed_root_empty(fs_info);
4446 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4448 btrfs_destroy_pinned_extent(fs_info,
4449 fs_info->pinned_extents);
4451 cur_trans->state =TRANS_STATE_COMPLETED;
4452 wake_up(&cur_trans->commit_wait);
4455 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4457 struct btrfs_transaction *t;
4459 mutex_lock(&fs_info->transaction_kthread_mutex);
4461 spin_lock(&fs_info->trans_lock);
4462 while (!list_empty(&fs_info->trans_list)) {
4463 t = list_first_entry(&fs_info->trans_list,
4464 struct btrfs_transaction, list);
4465 if (t->state >= TRANS_STATE_COMMIT_START) {
4466 refcount_inc(&t->use_count);
4467 spin_unlock(&fs_info->trans_lock);
4468 btrfs_wait_for_commit(fs_info, t->transid);
4469 btrfs_put_transaction(t);
4470 spin_lock(&fs_info->trans_lock);
4473 if (t == fs_info->running_transaction) {
4474 t->state = TRANS_STATE_COMMIT_DOING;
4475 spin_unlock(&fs_info->trans_lock);
4477 * We wait for 0 num_writers since we don't hold a trans
4478 * handle open currently for this transaction.
4480 wait_event(t->writer_wait,
4481 atomic_read(&t->num_writers) == 0);
4483 spin_unlock(&fs_info->trans_lock);
4485 btrfs_cleanup_one_transaction(t, fs_info);
4487 spin_lock(&fs_info->trans_lock);
4488 if (t == fs_info->running_transaction)
4489 fs_info->running_transaction = NULL;
4490 list_del_init(&t->list);
4491 spin_unlock(&fs_info->trans_lock);
4493 btrfs_put_transaction(t);
4494 trace_btrfs_transaction_commit(fs_info->tree_root);
4495 spin_lock(&fs_info->trans_lock);
4497 spin_unlock(&fs_info->trans_lock);
4498 btrfs_destroy_all_ordered_extents(fs_info);
4499 btrfs_destroy_delayed_inodes(fs_info);
4500 btrfs_assert_delayed_root_empty(fs_info);
4501 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4502 btrfs_destroy_all_delalloc_inodes(fs_info);
4503 mutex_unlock(&fs_info->transaction_kthread_mutex);
4508 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4510 struct inode *inode = private_data;
4511 return btrfs_sb(inode->i_sb);
4514 static const struct extent_io_ops btree_extent_io_ops = {
4515 /* mandatory callbacks */
4516 .submit_bio_hook = btree_submit_bio_hook,
4517 .readpage_end_io_hook = btree_readpage_end_io_hook,
4518 /* note we're sharing with inode.c for the merge bio hook */
4519 .merge_bio_hook = btrfs_merge_bio_hook,
4520 .readpage_io_failed_hook = btree_io_failed_hook,
4521 .set_range_writeback = btrfs_set_range_writeback,
4522 .tree_fs_info = btree_fs_info,
4524 /* optional callbacks */