1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <asm/unaligned.h>
23 #include "transaction.h"
24 #include "btrfs_inode.h"
26 #include "print-tree.h"
29 #include "free-space-cache.h"
30 #include "free-space-tree.h"
31 #include "inode-map.h"
32 #include "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
43 #include <asm/cpufeature.h>
46 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
53 static const struct extent_io_ops btree_extent_io_ops;
54 static void end_workqueue_fn(struct btrfs_work *work);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_fs_info *fs_info);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 struct extent_io_tree *dirty_pages,
62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
72 struct btrfs_end_io_wq {
76 struct btrfs_fs_info *info;
78 enum btrfs_wq_endio_type metadata;
79 struct btrfs_work work;
82 static struct kmem_cache *btrfs_end_io_wq_cache;
84 int __init btrfs_end_io_wq_init(void)
86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq),
91 if (!btrfs_end_io_wq_cache)
96 void __cold btrfs_end_io_wq_exit(void)
98 kmem_cache_destroy(btrfs_end_io_wq_cache);
102 * async submit bios are used to offload expensive checksumming
103 * onto the worker threads. They checksum file and metadata bios
104 * just before they are sent down the IO stack.
106 struct async_submit_bio {
109 extent_submit_bio_start_t *submit_bio_start;
112 * bio_offset is optional, can be used if the pages in the bio
113 * can't tell us where in the file the bio should go
116 struct btrfs_work work;
121 * Lockdep class keys for extent_buffer->lock's in this root. For a given
122 * eb, the lockdep key is determined by the btrfs_root it belongs to and
123 * the level the eb occupies in the tree.
125 * Different roots are used for different purposes and may nest inside each
126 * other and they require separate keysets. As lockdep keys should be
127 * static, assign keysets according to the purpose of the root as indicated
128 * by btrfs_root->root_key.objectid. This ensures that all special purpose
129 * roots have separate keysets.
131 * Lock-nesting across peer nodes is always done with the immediate parent
132 * node locked thus preventing deadlock. As lockdep doesn't know this, use
133 * subclass to avoid triggering lockdep warning in such cases.
135 * The key is set by the readpage_end_io_hook after the buffer has passed
136 * csum validation but before the pages are unlocked. It is also set by
137 * btrfs_init_new_buffer on freshly allocated blocks.
139 * We also add a check to make sure the highest level of the tree is the
140 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
141 * needs update as well.
143 #ifdef CONFIG_DEBUG_LOCK_ALLOC
144 # if BTRFS_MAX_LEVEL != 8
148 static struct btrfs_lockdep_keyset {
149 u64 id; /* root objectid */
150 const char *name_stem; /* lock name stem */
151 char names[BTRFS_MAX_LEVEL + 1][20];
152 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
153 } btrfs_lockdep_keysets[] = {
154 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
155 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
156 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
157 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
158 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
159 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
160 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
161 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
162 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
163 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
164 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
165 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
166 { .id = 0, .name_stem = "tree" },
169 void __init btrfs_init_lockdep(void)
173 /* initialize lockdep class names */
174 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
175 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
177 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
178 snprintf(ks->names[j], sizeof(ks->names[j]),
179 "btrfs-%s-%02d", ks->name_stem, j);
183 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
186 struct btrfs_lockdep_keyset *ks;
188 BUG_ON(level >= ARRAY_SIZE(ks->keys));
190 /* find the matching keyset, id 0 is the default entry */
191 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
192 if (ks->id == objectid)
195 lockdep_set_class_and_name(&eb->lock,
196 &ks->keys[level], ks->names[level]);
202 * extents on the btree inode are pretty simple, there's one extent
203 * that covers the entire device
205 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
206 struct page *page, size_t pg_offset, u64 start, u64 len,
209 struct btrfs_fs_info *fs_info = inode->root->fs_info;
210 struct extent_map_tree *em_tree = &inode->extent_tree;
211 struct extent_map *em;
214 read_lock(&em_tree->lock);
215 em = lookup_extent_mapping(em_tree, start, len);
217 em->bdev = fs_info->fs_devices->latest_bdev;
218 read_unlock(&em_tree->lock);
221 read_unlock(&em_tree->lock);
223 em = alloc_extent_map();
225 em = ERR_PTR(-ENOMEM);
230 em->block_len = (u64)-1;
232 em->bdev = fs_info->fs_devices->latest_bdev;
234 write_lock(&em_tree->lock);
235 ret = add_extent_mapping(em_tree, em, 0);
236 if (ret == -EEXIST) {
238 em = lookup_extent_mapping(em_tree, start, len);
245 write_unlock(&em_tree->lock);
251 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
253 return crc32c(seed, data, len);
256 void btrfs_csum_final(u32 crc, u8 *result)
258 put_unaligned_le32(~crc, result);
262 * compute the csum for a btree block, and either verify it or write it
263 * into the csum field of the block.
265 static int csum_tree_block(struct btrfs_fs_info *fs_info,
266 struct extent_buffer *buf,
269 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
270 char result[BTRFS_CSUM_SIZE];
272 unsigned long cur_len;
273 unsigned long offset = BTRFS_CSUM_SIZE;
275 unsigned long map_start;
276 unsigned long map_len;
280 len = buf->len - offset;
282 err = map_private_extent_buffer(buf, offset, 32,
283 &kaddr, &map_start, &map_len);
286 cur_len = min(len, map_len - (offset - map_start));
287 crc = btrfs_csum_data(kaddr + offset - map_start,
292 memset(result, 0, BTRFS_CSUM_SIZE);
294 btrfs_csum_final(crc, result);
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
300 memcpy(&found, result, csum_size);
302 read_extent_buffer(buf, &val, 0, csum_size);
303 btrfs_warn_rl(fs_info,
304 "%s checksum verify failed on %llu wanted %X found %X level %d",
305 fs_info->sb->s_id, buf->start,
306 val, found, btrfs_header_level(buf));
310 write_extent_buffer(buf, result, 0, csum_size);
317 * we can't consider a given block up to date unless the transid of the
318 * block matches the transid in the parent node's pointer. This is how we
319 * detect blocks that either didn't get written at all or got written
320 * in the wrong place.
322 static int verify_parent_transid(struct extent_io_tree *io_tree,
323 struct extent_buffer *eb, u64 parent_transid,
326 struct extent_state *cached_state = NULL;
328 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
330 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
337 btrfs_tree_read_lock(eb);
338 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
341 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
343 if (extent_buffer_uptodate(eb) &&
344 btrfs_header_generation(eb) == parent_transid) {
348 btrfs_err_rl(eb->fs_info,
349 "parent transid verify failed on %llu wanted %llu found %llu",
351 parent_transid, btrfs_header_generation(eb));
355 * Things reading via commit roots that don't have normal protection,
356 * like send, can have a really old block in cache that may point at a
357 * block that has been freed and re-allocated. So don't clear uptodate
358 * if we find an eb that is under IO (dirty/writeback) because we could
359 * end up reading in the stale data and then writing it back out and
360 * making everybody very sad.
362 if (!extent_buffer_under_io(eb))
363 clear_extent_buffer_uptodate(eb);
365 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
368 btrfs_tree_read_unlock_blocking(eb);
373 * Return 0 if the superblock checksum type matches the checksum value of that
374 * algorithm. Pass the raw disk superblock data.
376 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
379 struct btrfs_super_block *disk_sb =
380 (struct btrfs_super_block *)raw_disk_sb;
381 u16 csum_type = btrfs_super_csum_type(disk_sb);
384 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
386 char result[sizeof(crc)];
389 * The super_block structure does not span the whole
390 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
391 * is filled with zeros and is included in the checksum.
393 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
394 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
395 btrfs_csum_final(crc, result);
397 if (memcmp(raw_disk_sb, result, sizeof(result)))
401 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
402 btrfs_err(fs_info, "unsupported checksum algorithm %u",
410 static int verify_level_key(struct btrfs_fs_info *fs_info,
411 struct extent_buffer *eb, int level,
412 struct btrfs_key *first_key, u64 parent_transid)
415 struct btrfs_key found_key;
418 found_level = btrfs_header_level(eb);
419 if (found_level != level) {
420 #ifdef CONFIG_BTRFS_DEBUG
423 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
424 eb->start, level, found_level);
433 * For live tree block (new tree blocks in current transaction),
434 * we need proper lock context to avoid race, which is impossible here.
435 * So we only checks tree blocks which is read from disk, whose
436 * generation <= fs_info->last_trans_committed.
438 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
441 btrfs_node_key_to_cpu(eb, &found_key, 0);
443 btrfs_item_key_to_cpu(eb, &found_key, 0);
444 ret = btrfs_comp_cpu_keys(first_key, &found_key);
446 #ifdef CONFIG_BTRFS_DEBUG
450 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
451 eb->start, parent_transid, first_key->objectid,
452 first_key->type, first_key->offset,
453 found_key.objectid, found_key.type,
461 * helper to read a given tree block, doing retries as required when
462 * the checksums don't match and we have alternate mirrors to try.
464 * @parent_transid: expected transid, skip check if 0
465 * @level: expected level, mandatory check
466 * @first_key: expected key of first slot, skip check if NULL
468 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
469 struct extent_buffer *eb,
470 u64 parent_transid, int level,
471 struct btrfs_key *first_key)
473 struct extent_io_tree *io_tree;
478 int failed_mirror = 0;
480 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
482 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
483 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
486 if (verify_parent_transid(io_tree, eb,
489 else if (verify_level_key(fs_info, eb, level,
490 first_key, parent_transid))
496 num_copies = btrfs_num_copies(fs_info,
501 if (!failed_mirror) {
503 failed_mirror = eb->read_mirror;
507 if (mirror_num == failed_mirror)
510 if (mirror_num > num_copies)
514 if (failed && !ret && failed_mirror)
515 repair_eb_io_failure(fs_info, eb, failed_mirror);
521 * checksum a dirty tree block before IO. This has extra checks to make sure
522 * we only fill in the checksum field in the first page of a multi-page block
525 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
527 u64 start = page_offset(page);
529 struct extent_buffer *eb;
531 eb = (struct extent_buffer *)page->private;
532 if (page != eb->pages[0])
535 found_start = btrfs_header_bytenr(eb);
537 * Please do not consolidate these warnings into a single if.
538 * It is useful to know what went wrong.
540 if (WARN_ON(found_start != start))
542 if (WARN_ON(!PageUptodate(page)))
545 ASSERT(memcmp_extent_buffer(eb, fs_info->metadata_fsid,
546 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
548 return csum_tree_block(fs_info, eb, 0);
551 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
552 struct extent_buffer *eb)
554 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
555 u8 fsid[BTRFS_FSID_SIZE];
558 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
563 * Checking the incompat flag is only valid for the current
564 * fs. For seed devices it's forbidden to have their uuid
565 * changed so reading ->fsid in this case is fine
567 if (fs_devices == fs_info->fs_devices &&
568 btrfs_fs_incompat(fs_info, METADATA_UUID))
569 metadata_uuid = fs_devices->metadata_uuid;
571 metadata_uuid = fs_devices->fsid;
573 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
577 fs_devices = fs_devices->seed;
582 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
583 u64 phy_offset, struct page *page,
584 u64 start, u64 end, int mirror)
588 struct extent_buffer *eb;
589 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
590 struct btrfs_fs_info *fs_info = root->fs_info;
597 eb = (struct extent_buffer *)page->private;
599 /* the pending IO might have been the only thing that kept this buffer
600 * in memory. Make sure we have a ref for all this other checks
602 extent_buffer_get(eb);
604 reads_done = atomic_dec_and_test(&eb->io_pages);
608 eb->read_mirror = mirror;
609 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
614 found_start = btrfs_header_bytenr(eb);
615 if (found_start != eb->start) {
616 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
617 eb->start, found_start);
621 if (check_tree_block_fsid(fs_info, eb)) {
622 btrfs_err_rl(fs_info, "bad fsid on block %llu",
627 found_level = btrfs_header_level(eb);
628 if (found_level >= BTRFS_MAX_LEVEL) {
629 btrfs_err(fs_info, "bad tree block level %d on %llu",
630 (int)btrfs_header_level(eb), eb->start);
635 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
638 ret = csum_tree_block(fs_info, eb, 1);
643 * If this is a leaf block and it is corrupt, set the corrupt bit so
644 * that we don't try and read the other copies of this block, just
647 if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
648 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
652 if (found_level > 0 && btrfs_check_node(fs_info, eb))
656 set_extent_buffer_uptodate(eb);
659 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
660 btree_readahead_hook(eb, ret);
664 * our io error hook is going to dec the io pages
665 * again, we have to make sure it has something
668 atomic_inc(&eb->io_pages);
669 clear_extent_buffer_uptodate(eb);
671 free_extent_buffer(eb);
676 static int btree_io_failed_hook(struct page *page, int failed_mirror)
678 struct extent_buffer *eb;
680 eb = (struct extent_buffer *)page->private;
681 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
682 eb->read_mirror = failed_mirror;
683 atomic_dec(&eb->io_pages);
684 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
685 btree_readahead_hook(eb, -EIO);
686 return -EIO; /* we fixed nothing */
689 static void end_workqueue_bio(struct bio *bio)
691 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
692 struct btrfs_fs_info *fs_info;
693 struct btrfs_workqueue *wq;
694 btrfs_work_func_t func;
696 fs_info = end_io_wq->info;
697 end_io_wq->status = bio->bi_status;
699 if (bio_op(bio) == REQ_OP_WRITE) {
700 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
701 wq = fs_info->endio_meta_write_workers;
702 func = btrfs_endio_meta_write_helper;
703 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
704 wq = fs_info->endio_freespace_worker;
705 func = btrfs_freespace_write_helper;
706 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
707 wq = fs_info->endio_raid56_workers;
708 func = btrfs_endio_raid56_helper;
710 wq = fs_info->endio_write_workers;
711 func = btrfs_endio_write_helper;
714 if (unlikely(end_io_wq->metadata ==
715 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
716 wq = fs_info->endio_repair_workers;
717 func = btrfs_endio_repair_helper;
718 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
719 wq = fs_info->endio_raid56_workers;
720 func = btrfs_endio_raid56_helper;
721 } else if (end_io_wq->metadata) {
722 wq = fs_info->endio_meta_workers;
723 func = btrfs_endio_meta_helper;
725 wq = fs_info->endio_workers;
726 func = btrfs_endio_helper;
730 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
731 btrfs_queue_work(wq, &end_io_wq->work);
734 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
735 enum btrfs_wq_endio_type metadata)
737 struct btrfs_end_io_wq *end_io_wq;
739 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
741 return BLK_STS_RESOURCE;
743 end_io_wq->private = bio->bi_private;
744 end_io_wq->end_io = bio->bi_end_io;
745 end_io_wq->info = info;
746 end_io_wq->status = 0;
747 end_io_wq->bio = bio;
748 end_io_wq->metadata = metadata;
750 bio->bi_private = end_io_wq;
751 bio->bi_end_io = end_workqueue_bio;
755 static void run_one_async_start(struct btrfs_work *work)
757 struct async_submit_bio *async;
760 async = container_of(work, struct async_submit_bio, work);
761 ret = async->submit_bio_start(async->private_data, async->bio,
767 static void run_one_async_done(struct btrfs_work *work)
769 struct async_submit_bio *async;
771 async = container_of(work, struct async_submit_bio, work);
773 /* If an error occurred we just want to clean up the bio and move on */
775 async->bio->bi_status = async->status;
776 bio_endio(async->bio);
780 btrfs_submit_bio_done(async->private_data, async->bio, async->mirror_num);
783 static void run_one_async_free(struct btrfs_work *work)
785 struct async_submit_bio *async;
787 async = container_of(work, struct async_submit_bio, work);
791 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
792 int mirror_num, unsigned long bio_flags,
793 u64 bio_offset, void *private_data,
794 extent_submit_bio_start_t *submit_bio_start)
796 struct async_submit_bio *async;
798 async = kmalloc(sizeof(*async), GFP_NOFS);
800 return BLK_STS_RESOURCE;
802 async->private_data = private_data;
804 async->mirror_num = mirror_num;
805 async->submit_bio_start = submit_bio_start;
807 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
808 run_one_async_done, run_one_async_free);
810 async->bio_offset = bio_offset;
814 if (op_is_sync(bio->bi_opf))
815 btrfs_set_work_high_priority(&async->work);
817 btrfs_queue_work(fs_info->workers, &async->work);
821 static blk_status_t btree_csum_one_bio(struct bio *bio)
823 struct bio_vec *bvec;
824 struct btrfs_root *root;
827 ASSERT(!bio_flagged(bio, BIO_CLONED));
828 bio_for_each_segment_all(bvec, bio, i) {
829 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
830 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
835 return errno_to_blk_status(ret);
838 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
842 * when we're called for a write, we're already in the async
843 * submission context. Just jump into btrfs_map_bio
845 return btree_csum_one_bio(bio);
848 static int check_async_write(struct btrfs_inode *bi)
850 if (atomic_read(&bi->sync_writers))
853 if (static_cpu_has(X86_FEATURE_XMM4_2))
859 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
860 int mirror_num, unsigned long bio_flags,
863 struct inode *inode = private_data;
864 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
865 int async = check_async_write(BTRFS_I(inode));
868 if (bio_op(bio) != REQ_OP_WRITE) {
870 * called for a read, do the setup so that checksum validation
871 * can happen in the async kernel threads
873 ret = btrfs_bio_wq_end_io(fs_info, bio,
874 BTRFS_WQ_ENDIO_METADATA);
877 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
879 ret = btree_csum_one_bio(bio);
882 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
885 * kthread helpers are used to submit writes so that
886 * checksumming can happen in parallel across all CPUs
888 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
889 bio_offset, private_data,
890 btree_submit_bio_start);
898 bio->bi_status = ret;
903 #ifdef CONFIG_MIGRATION
904 static int btree_migratepage(struct address_space *mapping,
905 struct page *newpage, struct page *page,
906 enum migrate_mode mode)
909 * we can't safely write a btree page from here,
910 * we haven't done the locking hook
915 * Buffers may be managed in a filesystem specific way.
916 * We must have no buffers or drop them.
918 if (page_has_private(page) &&
919 !try_to_release_page(page, GFP_KERNEL))
921 return migrate_page(mapping, newpage, page, mode);
926 static int btree_writepages(struct address_space *mapping,
927 struct writeback_control *wbc)
929 struct btrfs_fs_info *fs_info;
932 if (wbc->sync_mode == WB_SYNC_NONE) {
934 if (wbc->for_kupdate)
937 fs_info = BTRFS_I(mapping->host)->root->fs_info;
938 /* this is a bit racy, but that's ok */
939 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
940 BTRFS_DIRTY_METADATA_THRESH,
941 fs_info->dirty_metadata_batch);
945 return btree_write_cache_pages(mapping, wbc);
948 static int btree_readpage(struct file *file, struct page *page)
950 struct extent_io_tree *tree;
951 tree = &BTRFS_I(page->mapping->host)->io_tree;
952 return extent_read_full_page(tree, page, btree_get_extent, 0);
955 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
957 if (PageWriteback(page) || PageDirty(page))
960 return try_release_extent_buffer(page);
963 static void btree_invalidatepage(struct page *page, unsigned int offset,
966 struct extent_io_tree *tree;
967 tree = &BTRFS_I(page->mapping->host)->io_tree;
968 extent_invalidatepage(tree, page, offset);
969 btree_releasepage(page, GFP_NOFS);
970 if (PagePrivate(page)) {
971 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
972 "page private not zero on page %llu",
973 (unsigned long long)page_offset(page));
974 ClearPagePrivate(page);
975 set_page_private(page, 0);
980 static int btree_set_page_dirty(struct page *page)
983 struct extent_buffer *eb;
985 BUG_ON(!PagePrivate(page));
986 eb = (struct extent_buffer *)page->private;
988 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
989 BUG_ON(!atomic_read(&eb->refs));
990 btrfs_assert_tree_locked(eb);
992 return __set_page_dirty_nobuffers(page);
995 static const struct address_space_operations btree_aops = {
996 .readpage = btree_readpage,
997 .writepages = btree_writepages,
998 .releasepage = btree_releasepage,
999 .invalidatepage = btree_invalidatepage,
1000 #ifdef CONFIG_MIGRATION
1001 .migratepage = btree_migratepage,
1003 .set_page_dirty = btree_set_page_dirty,
1006 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1008 struct extent_buffer *buf = NULL;
1009 struct inode *btree_inode = fs_info->btree_inode;
1011 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1014 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016 free_extent_buffer(buf);
1019 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1020 int mirror_num, struct extent_buffer **eb)
1022 struct extent_buffer *buf = NULL;
1023 struct inode *btree_inode = fs_info->btree_inode;
1024 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1027 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1031 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1033 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1036 free_extent_buffer(buf);
1040 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1041 free_extent_buffer(buf);
1043 } else if (extent_buffer_uptodate(buf)) {
1046 free_extent_buffer(buf);
1051 struct extent_buffer *btrfs_find_create_tree_block(
1052 struct btrfs_fs_info *fs_info,
1055 if (btrfs_is_testing(fs_info))
1056 return alloc_test_extent_buffer(fs_info, bytenr);
1057 return alloc_extent_buffer(fs_info, bytenr);
1061 int btrfs_write_tree_block(struct extent_buffer *buf)
1063 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1064 buf->start + buf->len - 1);
1067 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1069 filemap_fdatawait_range(buf->pages[0]->mapping,
1070 buf->start, buf->start + buf->len - 1);
1074 * Read tree block at logical address @bytenr and do variant basic but critical
1077 * @parent_transid: expected transid of this tree block, skip check if 0
1078 * @level: expected level, mandatory check
1079 * @first_key: expected key in slot 0, skip check if NULL
1081 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1082 u64 parent_transid, int level,
1083 struct btrfs_key *first_key)
1085 struct extent_buffer *buf = NULL;
1088 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1092 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1095 free_extent_buffer(buf);
1096 return ERR_PTR(ret);
1102 void clean_tree_block(struct btrfs_fs_info *fs_info,
1103 struct extent_buffer *buf)
1105 if (btrfs_header_generation(buf) ==
1106 fs_info->running_transaction->transid) {
1107 btrfs_assert_tree_locked(buf);
1109 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1110 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1112 fs_info->dirty_metadata_batch);
1113 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1114 btrfs_set_lock_blocking(buf);
1115 clear_extent_buffer_dirty(buf);
1120 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1122 struct btrfs_subvolume_writers *writers;
1125 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1127 return ERR_PTR(-ENOMEM);
1129 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1132 return ERR_PTR(ret);
1135 init_waitqueue_head(&writers->wait);
1140 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1142 percpu_counter_destroy(&writers->counter);
1146 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1149 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1151 root->commit_root = NULL;
1153 root->orphan_cleanup_state = 0;
1155 root->last_trans = 0;
1156 root->highest_objectid = 0;
1157 root->nr_delalloc_inodes = 0;
1158 root->nr_ordered_extents = 0;
1159 root->inode_tree = RB_ROOT;
1160 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1161 root->block_rsv = NULL;
1163 INIT_LIST_HEAD(&root->dirty_list);
1164 INIT_LIST_HEAD(&root->root_list);
1165 INIT_LIST_HEAD(&root->delalloc_inodes);
1166 INIT_LIST_HEAD(&root->delalloc_root);
1167 INIT_LIST_HEAD(&root->ordered_extents);
1168 INIT_LIST_HEAD(&root->ordered_root);
1169 INIT_LIST_HEAD(&root->logged_list[0]);
1170 INIT_LIST_HEAD(&root->logged_list[1]);
1171 spin_lock_init(&root->inode_lock);
1172 spin_lock_init(&root->delalloc_lock);
1173 spin_lock_init(&root->ordered_extent_lock);
1174 spin_lock_init(&root->accounting_lock);
1175 spin_lock_init(&root->log_extents_lock[0]);
1176 spin_lock_init(&root->log_extents_lock[1]);
1177 spin_lock_init(&root->qgroup_meta_rsv_lock);
1178 mutex_init(&root->objectid_mutex);
1179 mutex_init(&root->log_mutex);
1180 mutex_init(&root->ordered_extent_mutex);
1181 mutex_init(&root->delalloc_mutex);
1182 init_waitqueue_head(&root->log_writer_wait);
1183 init_waitqueue_head(&root->log_commit_wait[0]);
1184 init_waitqueue_head(&root->log_commit_wait[1]);
1185 INIT_LIST_HEAD(&root->log_ctxs[0]);
1186 INIT_LIST_HEAD(&root->log_ctxs[1]);
1187 atomic_set(&root->log_commit[0], 0);
1188 atomic_set(&root->log_commit[1], 0);
1189 atomic_set(&root->log_writers, 0);
1190 atomic_set(&root->log_batch, 0);
1191 refcount_set(&root->refs, 1);
1192 atomic_set(&root->will_be_snapshotted, 0);
1193 atomic_set(&root->snapshot_force_cow, 0);
1194 atomic_set(&root->nr_swapfiles, 0);
1195 root->log_transid = 0;
1196 root->log_transid_committed = -1;
1197 root->last_log_commit = 0;
1199 extent_io_tree_init(&root->dirty_log_pages, NULL);
1201 memset(&root->root_key, 0, sizeof(root->root_key));
1202 memset(&root->root_item, 0, sizeof(root->root_item));
1203 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1205 root->defrag_trans_start = fs_info->generation;
1207 root->defrag_trans_start = 0;
1208 root->root_key.objectid = objectid;
1211 spin_lock_init(&root->root_item_lock);
1214 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1217 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1219 root->fs_info = fs_info;
1223 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1224 /* Should only be used by the testing infrastructure */
1225 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1227 struct btrfs_root *root;
1230 return ERR_PTR(-EINVAL);
1232 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1234 return ERR_PTR(-ENOMEM);
1236 /* We don't use the stripesize in selftest, set it as sectorsize */
1237 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1238 root->alloc_bytenr = 0;
1244 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1245 struct btrfs_fs_info *fs_info,
1248 struct extent_buffer *leaf;
1249 struct btrfs_root *tree_root = fs_info->tree_root;
1250 struct btrfs_root *root;
1251 struct btrfs_key key;
1253 uuid_le uuid = NULL_UUID_LE;
1255 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1257 return ERR_PTR(-ENOMEM);
1259 __setup_root(root, fs_info, objectid);
1260 root->root_key.objectid = objectid;
1261 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1262 root->root_key.offset = 0;
1264 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1266 ret = PTR_ERR(leaf);
1272 btrfs_mark_buffer_dirty(leaf);
1274 root->commit_root = btrfs_root_node(root);
1275 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1277 root->root_item.flags = 0;
1278 root->root_item.byte_limit = 0;
1279 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1280 btrfs_set_root_generation(&root->root_item, trans->transid);
1281 btrfs_set_root_level(&root->root_item, 0);
1282 btrfs_set_root_refs(&root->root_item, 1);
1283 btrfs_set_root_used(&root->root_item, leaf->len);
1284 btrfs_set_root_last_snapshot(&root->root_item, 0);
1285 btrfs_set_root_dirid(&root->root_item, 0);
1286 if (is_fstree(objectid))
1288 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1289 root->root_item.drop_level = 0;
1291 key.objectid = objectid;
1292 key.type = BTRFS_ROOT_ITEM_KEY;
1294 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1298 btrfs_tree_unlock(leaf);
1304 btrfs_tree_unlock(leaf);
1305 free_extent_buffer(root->commit_root);
1306 free_extent_buffer(leaf);
1310 return ERR_PTR(ret);
1313 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1314 struct btrfs_fs_info *fs_info)
1316 struct btrfs_root *root;
1317 struct extent_buffer *leaf;
1319 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1321 return ERR_PTR(-ENOMEM);
1323 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1325 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1326 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1327 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1330 * DON'T set REF_COWS for log trees
1332 * log trees do not get reference counted because they go away
1333 * before a real commit is actually done. They do store pointers
1334 * to file data extents, and those reference counts still get
1335 * updated (along with back refs to the log tree).
1338 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1342 return ERR_CAST(leaf);
1347 btrfs_mark_buffer_dirty(root->node);
1348 btrfs_tree_unlock(root->node);
1352 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1353 struct btrfs_fs_info *fs_info)
1355 struct btrfs_root *log_root;
1357 log_root = alloc_log_tree(trans, fs_info);
1358 if (IS_ERR(log_root))
1359 return PTR_ERR(log_root);
1360 WARN_ON(fs_info->log_root_tree);
1361 fs_info->log_root_tree = log_root;
1365 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1366 struct btrfs_root *root)
1368 struct btrfs_fs_info *fs_info = root->fs_info;
1369 struct btrfs_root *log_root;
1370 struct btrfs_inode_item *inode_item;
1372 log_root = alloc_log_tree(trans, fs_info);
1373 if (IS_ERR(log_root))
1374 return PTR_ERR(log_root);
1376 log_root->last_trans = trans->transid;
1377 log_root->root_key.offset = root->root_key.objectid;
1379 inode_item = &log_root->root_item.inode;
1380 btrfs_set_stack_inode_generation(inode_item, 1);
1381 btrfs_set_stack_inode_size(inode_item, 3);
1382 btrfs_set_stack_inode_nlink(inode_item, 1);
1383 btrfs_set_stack_inode_nbytes(inode_item,
1385 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1387 btrfs_set_root_node(&log_root->root_item, log_root->node);
1389 WARN_ON(root->log_root);
1390 root->log_root = log_root;
1391 root->log_transid = 0;
1392 root->log_transid_committed = -1;
1393 root->last_log_commit = 0;
1397 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1398 struct btrfs_key *key)
1400 struct btrfs_root *root;
1401 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1402 struct btrfs_path *path;
1407 path = btrfs_alloc_path();
1409 return ERR_PTR(-ENOMEM);
1411 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1417 __setup_root(root, fs_info, key->objectid);
1419 ret = btrfs_find_root(tree_root, key, path,
1420 &root->root_item, &root->root_key);
1427 generation = btrfs_root_generation(&root->root_item);
1428 level = btrfs_root_level(&root->root_item);
1429 root->node = read_tree_block(fs_info,
1430 btrfs_root_bytenr(&root->root_item),
1431 generation, level, NULL);
1432 if (IS_ERR(root->node)) {
1433 ret = PTR_ERR(root->node);
1435 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1437 free_extent_buffer(root->node);
1440 root->commit_root = btrfs_root_node(root);
1442 btrfs_free_path(path);
1448 root = ERR_PTR(ret);
1452 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1453 struct btrfs_key *location)
1455 struct btrfs_root *root;
1457 root = btrfs_read_tree_root(tree_root, location);
1461 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1462 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1463 btrfs_check_and_init_root_item(&root->root_item);
1469 int btrfs_init_fs_root(struct btrfs_root *root)
1472 struct btrfs_subvolume_writers *writers;
1474 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1475 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1477 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1482 writers = btrfs_alloc_subvolume_writers();
1483 if (IS_ERR(writers)) {
1484 ret = PTR_ERR(writers);
1487 root->subv_writers = writers;
1489 btrfs_init_free_ino_ctl(root);
1490 spin_lock_init(&root->ino_cache_lock);
1491 init_waitqueue_head(&root->ino_cache_wait);
1493 ret = get_anon_bdev(&root->anon_dev);
1497 mutex_lock(&root->objectid_mutex);
1498 ret = btrfs_find_highest_objectid(root,
1499 &root->highest_objectid);
1501 mutex_unlock(&root->objectid_mutex);
1505 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1507 mutex_unlock(&root->objectid_mutex);
1511 /* The caller is responsible to call btrfs_free_fs_root */
1515 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1518 struct btrfs_root *root;
1520 spin_lock(&fs_info->fs_roots_radix_lock);
1521 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1522 (unsigned long)root_id);
1523 spin_unlock(&fs_info->fs_roots_radix_lock);
1527 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1528 struct btrfs_root *root)
1532 ret = radix_tree_preload(GFP_NOFS);
1536 spin_lock(&fs_info->fs_roots_radix_lock);
1537 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1538 (unsigned long)root->root_key.objectid,
1541 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1542 spin_unlock(&fs_info->fs_roots_radix_lock);
1543 radix_tree_preload_end();
1548 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1549 struct btrfs_key *location,
1552 struct btrfs_root *root;
1553 struct btrfs_path *path;
1554 struct btrfs_key key;
1557 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1558 return fs_info->tree_root;
1559 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1560 return fs_info->extent_root;
1561 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1562 return fs_info->chunk_root;
1563 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1564 return fs_info->dev_root;
1565 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1566 return fs_info->csum_root;
1567 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1568 return fs_info->quota_root ? fs_info->quota_root :
1570 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1571 return fs_info->uuid_root ? fs_info->uuid_root :
1573 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1574 return fs_info->free_space_root ? fs_info->free_space_root :
1577 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1579 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1580 return ERR_PTR(-ENOENT);
1584 root = btrfs_read_fs_root(fs_info->tree_root, location);
1588 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1593 ret = btrfs_init_fs_root(root);
1597 path = btrfs_alloc_path();
1602 key.objectid = BTRFS_ORPHAN_OBJECTID;
1603 key.type = BTRFS_ORPHAN_ITEM_KEY;
1604 key.offset = location->objectid;
1606 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1607 btrfs_free_path(path);
1611 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1613 ret = btrfs_insert_fs_root(fs_info, root);
1615 if (ret == -EEXIST) {
1616 btrfs_free_fs_root(root);
1623 btrfs_free_fs_root(root);
1624 return ERR_PTR(ret);
1627 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1629 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1631 struct btrfs_device *device;
1632 struct backing_dev_info *bdi;
1635 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1638 bdi = device->bdev->bd_bdi;
1639 if (bdi_congested(bdi, bdi_bits)) {
1649 * called by the kthread helper functions to finally call the bio end_io
1650 * functions. This is where read checksum verification actually happens
1652 static void end_workqueue_fn(struct btrfs_work *work)
1655 struct btrfs_end_io_wq *end_io_wq;
1657 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1658 bio = end_io_wq->bio;
1660 bio->bi_status = end_io_wq->status;
1661 bio->bi_private = end_io_wq->private;
1662 bio->bi_end_io = end_io_wq->end_io;
1663 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1667 static int cleaner_kthread(void *arg)
1669 struct btrfs_root *root = arg;
1670 struct btrfs_fs_info *fs_info = root->fs_info;
1676 /* Make the cleaner go to sleep early. */
1677 if (btrfs_need_cleaner_sleep(fs_info))
1681 * Do not do anything if we might cause open_ctree() to block
1682 * before we have finished mounting the filesystem.
1684 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1687 if (!mutex_trylock(&fs_info->cleaner_mutex))
1691 * Avoid the problem that we change the status of the fs
1692 * during the above check and trylock.
1694 if (btrfs_need_cleaner_sleep(fs_info)) {
1695 mutex_unlock(&fs_info->cleaner_mutex);
1699 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1700 btrfs_run_delayed_iputs(fs_info);
1701 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1703 again = btrfs_clean_one_deleted_snapshot(root);
1704 mutex_unlock(&fs_info->cleaner_mutex);
1707 * The defragger has dealt with the R/O remount and umount,
1708 * needn't do anything special here.
1710 btrfs_run_defrag_inodes(fs_info);
1713 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1714 * with relocation (btrfs_relocate_chunk) and relocation
1715 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1716 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1717 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1718 * unused block groups.
1720 btrfs_delete_unused_bgs(fs_info);
1722 if (kthread_should_park())
1724 if (kthread_should_stop())
1727 set_current_state(TASK_INTERRUPTIBLE);
1729 __set_current_state(TASK_RUNNING);
1734 static int transaction_kthread(void *arg)
1736 struct btrfs_root *root = arg;
1737 struct btrfs_fs_info *fs_info = root->fs_info;
1738 struct btrfs_trans_handle *trans;
1739 struct btrfs_transaction *cur;
1742 unsigned long delay;
1746 cannot_commit = false;
1747 delay = HZ * fs_info->commit_interval;
1748 mutex_lock(&fs_info->transaction_kthread_mutex);
1750 spin_lock(&fs_info->trans_lock);
1751 cur = fs_info->running_transaction;
1753 spin_unlock(&fs_info->trans_lock);
1757 now = ktime_get_seconds();
1758 if (cur->state < TRANS_STATE_BLOCKED &&
1759 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1760 (now < cur->start_time ||
1761 now - cur->start_time < fs_info->commit_interval)) {
1762 spin_unlock(&fs_info->trans_lock);
1766 transid = cur->transid;
1767 spin_unlock(&fs_info->trans_lock);
1769 /* If the file system is aborted, this will always fail. */
1770 trans = btrfs_attach_transaction(root);
1771 if (IS_ERR(trans)) {
1772 if (PTR_ERR(trans) != -ENOENT)
1773 cannot_commit = true;
1776 if (transid == trans->transid) {
1777 btrfs_commit_transaction(trans);
1779 btrfs_end_transaction(trans);
1782 wake_up_process(fs_info->cleaner_kthread);
1783 mutex_unlock(&fs_info->transaction_kthread_mutex);
1785 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1786 &fs_info->fs_state)))
1787 btrfs_cleanup_transaction(fs_info);
1788 if (!kthread_should_stop() &&
1789 (!btrfs_transaction_blocked(fs_info) ||
1791 schedule_timeout_interruptible(delay);
1792 } while (!kthread_should_stop());
1797 * this will find the highest generation in the array of
1798 * root backups. The index of the highest array is returned,
1799 * or -1 if we can't find anything.
1801 * We check to make sure the array is valid by comparing the
1802 * generation of the latest root in the array with the generation
1803 * in the super block. If they don't match we pitch it.
1805 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1808 int newest_index = -1;
1809 struct btrfs_root_backup *root_backup;
1812 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1813 root_backup = info->super_copy->super_roots + i;
1814 cur = btrfs_backup_tree_root_gen(root_backup);
1815 if (cur == newest_gen)
1819 /* check to see if we actually wrapped around */
1820 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1821 root_backup = info->super_copy->super_roots;
1822 cur = btrfs_backup_tree_root_gen(root_backup);
1823 if (cur == newest_gen)
1826 return newest_index;
1831 * find the oldest backup so we know where to store new entries
1832 * in the backup array. This will set the backup_root_index
1833 * field in the fs_info struct
1835 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1838 int newest_index = -1;
1840 newest_index = find_newest_super_backup(info, newest_gen);
1841 /* if there was garbage in there, just move along */
1842 if (newest_index == -1) {
1843 info->backup_root_index = 0;
1845 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1850 * copy all the root pointers into the super backup array.
1851 * this will bump the backup pointer by one when it is
1854 static void backup_super_roots(struct btrfs_fs_info *info)
1857 struct btrfs_root_backup *root_backup;
1860 next_backup = info->backup_root_index;
1861 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1862 BTRFS_NUM_BACKUP_ROOTS;
1865 * just overwrite the last backup if we're at the same generation
1866 * this happens only at umount
1868 root_backup = info->super_for_commit->super_roots + last_backup;
1869 if (btrfs_backup_tree_root_gen(root_backup) ==
1870 btrfs_header_generation(info->tree_root->node))
1871 next_backup = last_backup;
1873 root_backup = info->super_for_commit->super_roots + next_backup;
1876 * make sure all of our padding and empty slots get zero filled
1877 * regardless of which ones we use today
1879 memset(root_backup, 0, sizeof(*root_backup));
1881 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1883 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1884 btrfs_set_backup_tree_root_gen(root_backup,
1885 btrfs_header_generation(info->tree_root->node));
1887 btrfs_set_backup_tree_root_level(root_backup,
1888 btrfs_header_level(info->tree_root->node));
1890 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1891 btrfs_set_backup_chunk_root_gen(root_backup,
1892 btrfs_header_generation(info->chunk_root->node));
1893 btrfs_set_backup_chunk_root_level(root_backup,
1894 btrfs_header_level(info->chunk_root->node));
1896 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1897 btrfs_set_backup_extent_root_gen(root_backup,
1898 btrfs_header_generation(info->extent_root->node));
1899 btrfs_set_backup_extent_root_level(root_backup,
1900 btrfs_header_level(info->extent_root->node));
1903 * we might commit during log recovery, which happens before we set
1904 * the fs_root. Make sure it is valid before we fill it in.
1906 if (info->fs_root && info->fs_root->node) {
1907 btrfs_set_backup_fs_root(root_backup,
1908 info->fs_root->node->start);
1909 btrfs_set_backup_fs_root_gen(root_backup,
1910 btrfs_header_generation(info->fs_root->node));
1911 btrfs_set_backup_fs_root_level(root_backup,
1912 btrfs_header_level(info->fs_root->node));
1915 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1916 btrfs_set_backup_dev_root_gen(root_backup,
1917 btrfs_header_generation(info->dev_root->node));
1918 btrfs_set_backup_dev_root_level(root_backup,
1919 btrfs_header_level(info->dev_root->node));
1921 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1922 btrfs_set_backup_csum_root_gen(root_backup,
1923 btrfs_header_generation(info->csum_root->node));
1924 btrfs_set_backup_csum_root_level(root_backup,
1925 btrfs_header_level(info->csum_root->node));
1927 btrfs_set_backup_total_bytes(root_backup,
1928 btrfs_super_total_bytes(info->super_copy));
1929 btrfs_set_backup_bytes_used(root_backup,
1930 btrfs_super_bytes_used(info->super_copy));
1931 btrfs_set_backup_num_devices(root_backup,
1932 btrfs_super_num_devices(info->super_copy));
1935 * if we don't copy this out to the super_copy, it won't get remembered
1936 * for the next commit
1938 memcpy(&info->super_copy->super_roots,
1939 &info->super_for_commit->super_roots,
1940 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1944 * this copies info out of the root backup array and back into
1945 * the in-memory super block. It is meant to help iterate through
1946 * the array, so you send it the number of backups you've already
1947 * tried and the last backup index you used.
1949 * this returns -1 when it has tried all the backups
1951 static noinline int next_root_backup(struct btrfs_fs_info *info,
1952 struct btrfs_super_block *super,
1953 int *num_backups_tried, int *backup_index)
1955 struct btrfs_root_backup *root_backup;
1956 int newest = *backup_index;
1958 if (*num_backups_tried == 0) {
1959 u64 gen = btrfs_super_generation(super);
1961 newest = find_newest_super_backup(info, gen);
1965 *backup_index = newest;
1966 *num_backups_tried = 1;
1967 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1968 /* we've tried all the backups, all done */
1971 /* jump to the next oldest backup */
1972 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1973 BTRFS_NUM_BACKUP_ROOTS;
1974 *backup_index = newest;
1975 *num_backups_tried += 1;
1977 root_backup = super->super_roots + newest;
1979 btrfs_set_super_generation(super,
1980 btrfs_backup_tree_root_gen(root_backup));
1981 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1982 btrfs_set_super_root_level(super,
1983 btrfs_backup_tree_root_level(root_backup));
1984 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1987 * fixme: the total bytes and num_devices need to match or we should
1990 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1991 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1995 /* helper to cleanup workers */
1996 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1998 btrfs_destroy_workqueue(fs_info->fixup_workers);
1999 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2000 btrfs_destroy_workqueue(fs_info->workers);
2001 btrfs_destroy_workqueue(fs_info->endio_workers);
2002 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2003 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2004 btrfs_destroy_workqueue(fs_info->rmw_workers);
2005 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2006 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2007 btrfs_destroy_workqueue(fs_info->submit_workers);
2008 btrfs_destroy_workqueue(fs_info->delayed_workers);
2009 btrfs_destroy_workqueue(fs_info->caching_workers);
2010 btrfs_destroy_workqueue(fs_info->readahead_workers);
2011 btrfs_destroy_workqueue(fs_info->flush_workers);
2012 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2013 btrfs_destroy_workqueue(fs_info->extent_workers);
2015 * Now that all other work queues are destroyed, we can safely destroy
2016 * the queues used for metadata I/O, since tasks from those other work
2017 * queues can do metadata I/O operations.
2019 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2020 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2023 static void free_root_extent_buffers(struct btrfs_root *root)
2026 free_extent_buffer(root->node);
2027 free_extent_buffer(root->commit_root);
2029 root->commit_root = NULL;
2033 /* helper to cleanup tree roots */
2034 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2036 free_root_extent_buffers(info->tree_root);
2038 free_root_extent_buffers(info->dev_root);
2039 free_root_extent_buffers(info->extent_root);
2040 free_root_extent_buffers(info->csum_root);
2041 free_root_extent_buffers(info->quota_root);
2042 free_root_extent_buffers(info->uuid_root);
2044 free_root_extent_buffers(info->chunk_root);
2045 free_root_extent_buffers(info->free_space_root);
2048 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2051 struct btrfs_root *gang[8];
2054 while (!list_empty(&fs_info->dead_roots)) {
2055 gang[0] = list_entry(fs_info->dead_roots.next,
2056 struct btrfs_root, root_list);
2057 list_del(&gang[0]->root_list);
2059 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2060 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2062 free_extent_buffer(gang[0]->node);
2063 free_extent_buffer(gang[0]->commit_root);
2064 btrfs_put_fs_root(gang[0]);
2069 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2074 for (i = 0; i < ret; i++)
2075 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2078 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2079 btrfs_free_log_root_tree(NULL, fs_info);
2080 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2084 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2086 mutex_init(&fs_info->scrub_lock);
2087 atomic_set(&fs_info->scrubs_running, 0);
2088 atomic_set(&fs_info->scrub_pause_req, 0);
2089 atomic_set(&fs_info->scrubs_paused, 0);
2090 atomic_set(&fs_info->scrub_cancel_req, 0);
2091 init_waitqueue_head(&fs_info->scrub_pause_wait);
2092 fs_info->scrub_workers_refcnt = 0;
2095 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2097 spin_lock_init(&fs_info->balance_lock);
2098 mutex_init(&fs_info->balance_mutex);
2099 atomic_set(&fs_info->balance_pause_req, 0);
2100 atomic_set(&fs_info->balance_cancel_req, 0);
2101 fs_info->balance_ctl = NULL;
2102 init_waitqueue_head(&fs_info->balance_wait_q);
2105 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2107 struct inode *inode = fs_info->btree_inode;
2109 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2110 set_nlink(inode, 1);
2112 * we set the i_size on the btree inode to the max possible int.
2113 * the real end of the address space is determined by all of
2114 * the devices in the system
2116 inode->i_size = OFFSET_MAX;
2117 inode->i_mapping->a_ops = &btree_aops;
2119 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2120 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2121 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2122 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2124 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2126 BTRFS_I(inode)->root = fs_info->tree_root;
2127 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2128 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2129 btrfs_insert_inode_hash(inode);
2132 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2134 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2135 rwlock_init(&fs_info->dev_replace.lock);
2136 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2137 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2138 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2141 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2143 spin_lock_init(&fs_info->qgroup_lock);
2144 mutex_init(&fs_info->qgroup_ioctl_lock);
2145 fs_info->qgroup_tree = RB_ROOT;
2146 fs_info->qgroup_op_tree = RB_ROOT;
2147 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2148 fs_info->qgroup_seq = 1;
2149 fs_info->qgroup_ulist = NULL;
2150 fs_info->qgroup_rescan_running = false;
2151 mutex_init(&fs_info->qgroup_rescan_lock);
2154 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2155 struct btrfs_fs_devices *fs_devices)
2157 u32 max_active = fs_info->thread_pool_size;
2158 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2161 btrfs_alloc_workqueue(fs_info, "worker",
2162 flags | WQ_HIGHPRI, max_active, 16);
2164 fs_info->delalloc_workers =
2165 btrfs_alloc_workqueue(fs_info, "delalloc",
2166 flags, max_active, 2);
2168 fs_info->flush_workers =
2169 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2170 flags, max_active, 0);
2172 fs_info->caching_workers =
2173 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2176 * a higher idle thresh on the submit workers makes it much more
2177 * likely that bios will be send down in a sane order to the
2180 fs_info->submit_workers =
2181 btrfs_alloc_workqueue(fs_info, "submit", flags,
2182 min_t(u64, fs_devices->num_devices,
2185 fs_info->fixup_workers =
2186 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2189 * endios are largely parallel and should have a very
2192 fs_info->endio_workers =
2193 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2194 fs_info->endio_meta_workers =
2195 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2197 fs_info->endio_meta_write_workers =
2198 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2200 fs_info->endio_raid56_workers =
2201 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2203 fs_info->endio_repair_workers =
2204 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2205 fs_info->rmw_workers =
2206 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2207 fs_info->endio_write_workers =
2208 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2210 fs_info->endio_freespace_worker =
2211 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2213 fs_info->delayed_workers =
2214 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2216 fs_info->readahead_workers =
2217 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2219 fs_info->qgroup_rescan_workers =
2220 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2221 fs_info->extent_workers =
2222 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2223 min_t(u64, fs_devices->num_devices,
2226 if (!(fs_info->workers && fs_info->delalloc_workers &&
2227 fs_info->submit_workers && fs_info->flush_workers &&
2228 fs_info->endio_workers && fs_info->endio_meta_workers &&
2229 fs_info->endio_meta_write_workers &&
2230 fs_info->endio_repair_workers &&
2231 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2232 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2233 fs_info->caching_workers && fs_info->readahead_workers &&
2234 fs_info->fixup_workers && fs_info->delayed_workers &&
2235 fs_info->extent_workers &&
2236 fs_info->qgroup_rescan_workers)) {
2243 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2244 struct btrfs_fs_devices *fs_devices)
2247 struct btrfs_root *log_tree_root;
2248 struct btrfs_super_block *disk_super = fs_info->super_copy;
2249 u64 bytenr = btrfs_super_log_root(disk_super);
2250 int level = btrfs_super_log_root_level(disk_super);
2252 if (fs_devices->rw_devices == 0) {
2253 btrfs_warn(fs_info, "log replay required on RO media");
2257 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2261 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2263 log_tree_root->node = read_tree_block(fs_info, bytenr,
2264 fs_info->generation + 1,
2266 if (IS_ERR(log_tree_root->node)) {
2267 btrfs_warn(fs_info, "failed to read log tree");
2268 ret = PTR_ERR(log_tree_root->node);
2269 kfree(log_tree_root);
2271 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2272 btrfs_err(fs_info, "failed to read log tree");
2273 free_extent_buffer(log_tree_root->node);
2274 kfree(log_tree_root);
2277 /* returns with log_tree_root freed on success */
2278 ret = btrfs_recover_log_trees(log_tree_root);
2280 btrfs_handle_fs_error(fs_info, ret,
2281 "Failed to recover log tree");
2282 free_extent_buffer(log_tree_root->node);
2283 kfree(log_tree_root);
2287 if (sb_rdonly(fs_info->sb)) {
2288 ret = btrfs_commit_super(fs_info);
2296 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2298 struct btrfs_root *tree_root = fs_info->tree_root;
2299 struct btrfs_root *root;
2300 struct btrfs_key location;
2303 BUG_ON(!fs_info->tree_root);
2305 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2306 location.type = BTRFS_ROOT_ITEM_KEY;
2307 location.offset = 0;
2309 root = btrfs_read_tree_root(tree_root, &location);
2311 ret = PTR_ERR(root);
2314 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2315 fs_info->extent_root = root;
2317 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2318 root = btrfs_read_tree_root(tree_root, &location);
2320 ret = PTR_ERR(root);
2323 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2324 fs_info->dev_root = root;
2325 btrfs_init_devices_late(fs_info);
2327 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2328 root = btrfs_read_tree_root(tree_root, &location);
2330 ret = PTR_ERR(root);
2333 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2334 fs_info->csum_root = root;
2336 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2337 root = btrfs_read_tree_root(tree_root, &location);
2338 if (!IS_ERR(root)) {
2339 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2340 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2341 fs_info->quota_root = root;
2344 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2345 root = btrfs_read_tree_root(tree_root, &location);
2347 ret = PTR_ERR(root);
2351 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2352 fs_info->uuid_root = root;
2355 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2356 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2357 root = btrfs_read_tree_root(tree_root, &location);
2359 ret = PTR_ERR(root);
2362 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2363 fs_info->free_space_root = root;
2368 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2369 location.objectid, ret);
2374 * Real super block validation
2375 * NOTE: super csum type and incompat features will not be checked here.
2377 * @sb: super block to check
2378 * @mirror_num: the super block number to check its bytenr:
2379 * 0 the primary (1st) sb
2380 * 1, 2 2nd and 3rd backup copy
2381 * -1 skip bytenr check
2383 static int validate_super(struct btrfs_fs_info *fs_info,
2384 struct btrfs_super_block *sb, int mirror_num)
2386 u64 nodesize = btrfs_super_nodesize(sb);
2387 u64 sectorsize = btrfs_super_sectorsize(sb);
2390 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2391 btrfs_err(fs_info, "no valid FS found");
2394 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2395 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2396 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2399 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2400 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2401 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2404 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2405 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2406 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2409 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2410 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2411 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2416 * Check sectorsize and nodesize first, other check will need it.
2417 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2419 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2420 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2421 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2424 /* Only PAGE SIZE is supported yet */
2425 if (sectorsize != PAGE_SIZE) {
2427 "sectorsize %llu not supported yet, only support %lu",
2428 sectorsize, PAGE_SIZE);
2431 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2432 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2433 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2436 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2437 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2438 le32_to_cpu(sb->__unused_leafsize), nodesize);
2442 /* Root alignment check */
2443 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2444 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2445 btrfs_super_root(sb));
2448 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2449 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2450 btrfs_super_chunk_root(sb));
2453 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2454 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2455 btrfs_super_log_root(sb));
2459 if (memcmp(fs_info->metadata_fsid, sb->dev_item.fsid,
2460 BTRFS_FSID_SIZE) != 0) {
2462 "dev_item UUID does not match metadata fsid: %pU != %pU",
2463 fs_info->metadata_fsid, sb->dev_item.fsid);
2468 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2471 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2472 btrfs_err(fs_info, "bytes_used is too small %llu",
2473 btrfs_super_bytes_used(sb));
2476 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2477 btrfs_err(fs_info, "invalid stripesize %u",
2478 btrfs_super_stripesize(sb));
2481 if (btrfs_super_num_devices(sb) > (1UL << 31))
2482 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2483 btrfs_super_num_devices(sb));
2484 if (btrfs_super_num_devices(sb) == 0) {
2485 btrfs_err(fs_info, "number of devices is 0");
2489 if (mirror_num >= 0 &&
2490 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2491 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2492 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2497 * Obvious sys_chunk_array corruptions, it must hold at least one key
2500 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2501 btrfs_err(fs_info, "system chunk array too big %u > %u",
2502 btrfs_super_sys_array_size(sb),
2503 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2506 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2507 + sizeof(struct btrfs_chunk)) {
2508 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2509 btrfs_super_sys_array_size(sb),
2510 sizeof(struct btrfs_disk_key)
2511 + sizeof(struct btrfs_chunk));
2516 * The generation is a global counter, we'll trust it more than the others
2517 * but it's still possible that it's the one that's wrong.
2519 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2521 "suspicious: generation < chunk_root_generation: %llu < %llu",
2522 btrfs_super_generation(sb),
2523 btrfs_super_chunk_root_generation(sb));
2524 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2525 && btrfs_super_cache_generation(sb) != (u64)-1)
2527 "suspicious: generation < cache_generation: %llu < %llu",
2528 btrfs_super_generation(sb),
2529 btrfs_super_cache_generation(sb));
2535 * Validation of super block at mount time.
2536 * Some checks already done early at mount time, like csum type and incompat
2537 * flags will be skipped.
2539 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2541 return validate_super(fs_info, fs_info->super_copy, 0);
2545 * Validation of super block at write time.
2546 * Some checks like bytenr check will be skipped as their values will be
2548 * Extra checks like csum type and incompat flags will be done here.
2550 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2551 struct btrfs_super_block *sb)
2555 ret = validate_super(fs_info, sb, -1);
2558 if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2560 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2561 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2564 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2567 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2568 btrfs_super_incompat_flags(sb),
2569 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2575 "super block corruption detected before writing it to disk");
2579 int open_ctree(struct super_block *sb,
2580 struct btrfs_fs_devices *fs_devices,
2588 struct btrfs_key location;
2589 struct buffer_head *bh;
2590 struct btrfs_super_block *disk_super;
2591 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2592 struct btrfs_root *tree_root;
2593 struct btrfs_root *chunk_root;
2596 int num_backups_tried = 0;
2597 int backup_index = 0;
2598 int clear_free_space_tree = 0;
2601 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2602 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2603 if (!tree_root || !chunk_root) {
2608 ret = init_srcu_struct(&fs_info->subvol_srcu);
2614 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2619 fs_info->dirty_metadata_batch = PAGE_SIZE *
2620 (1 + ilog2(nr_cpu_ids));
2622 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2625 goto fail_dirty_metadata_bytes;
2628 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2632 goto fail_delalloc_bytes;
2635 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2636 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2637 INIT_LIST_HEAD(&fs_info->trans_list);
2638 INIT_LIST_HEAD(&fs_info->dead_roots);
2639 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2640 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2641 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2642 INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2643 spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2644 spin_lock_init(&fs_info->delalloc_root_lock);
2645 spin_lock_init(&fs_info->trans_lock);
2646 spin_lock_init(&fs_info->fs_roots_radix_lock);
2647 spin_lock_init(&fs_info->delayed_iput_lock);
2648 spin_lock_init(&fs_info->defrag_inodes_lock);
2649 spin_lock_init(&fs_info->tree_mod_seq_lock);
2650 spin_lock_init(&fs_info->super_lock);
2651 spin_lock_init(&fs_info->qgroup_op_lock);
2652 spin_lock_init(&fs_info->buffer_lock);
2653 spin_lock_init(&fs_info->unused_bgs_lock);
2654 rwlock_init(&fs_info->tree_mod_log_lock);
2655 mutex_init(&fs_info->unused_bg_unpin_mutex);
2656 mutex_init(&fs_info->delete_unused_bgs_mutex);
2657 mutex_init(&fs_info->reloc_mutex);
2658 mutex_init(&fs_info->delalloc_root_mutex);
2659 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2660 seqlock_init(&fs_info->profiles_lock);
2662 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2663 INIT_LIST_HEAD(&fs_info->space_info);
2664 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2665 INIT_LIST_HEAD(&fs_info->unused_bgs);
2666 btrfs_mapping_init(&fs_info->mapping_tree);
2667 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2668 BTRFS_BLOCK_RSV_GLOBAL);
2669 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2670 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2671 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2672 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2673 BTRFS_BLOCK_RSV_DELOPS);
2674 atomic_set(&fs_info->async_delalloc_pages, 0);
2675 atomic_set(&fs_info->defrag_running, 0);
2676 atomic_set(&fs_info->qgroup_op_seq, 0);
2677 atomic_set(&fs_info->reada_works_cnt, 0);
2678 atomic64_set(&fs_info->tree_mod_seq, 0);
2680 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2681 fs_info->metadata_ratio = 0;
2682 fs_info->defrag_inodes = RB_ROOT;
2683 atomic64_set(&fs_info->free_chunk_space, 0);
2684 fs_info->tree_mod_log = RB_ROOT;
2685 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2686 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2687 /* readahead state */
2688 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2689 spin_lock_init(&fs_info->reada_lock);
2690 btrfs_init_ref_verify(fs_info);
2692 fs_info->thread_pool_size = min_t(unsigned long,
2693 num_online_cpus() + 2, 8);
2695 INIT_LIST_HEAD(&fs_info->ordered_roots);
2696 spin_lock_init(&fs_info->ordered_root_lock);
2698 fs_info->btree_inode = new_inode(sb);
2699 if (!fs_info->btree_inode) {
2701 goto fail_bio_counter;
2703 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2705 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2707 if (!fs_info->delayed_root) {
2711 btrfs_init_delayed_root(fs_info->delayed_root);
2713 btrfs_init_scrub(fs_info);
2714 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2715 fs_info->check_integrity_print_mask = 0;
2717 btrfs_init_balance(fs_info);
2718 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2720 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2721 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2723 btrfs_init_btree_inode(fs_info);
2725 spin_lock_init(&fs_info->block_group_cache_lock);
2726 fs_info->block_group_cache_tree = RB_ROOT;
2727 fs_info->first_logical_byte = (u64)-1;
2729 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2730 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2731 fs_info->pinned_extents = &fs_info->freed_extents[0];
2732 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2734 mutex_init(&fs_info->ordered_operations_mutex);
2735 mutex_init(&fs_info->tree_log_mutex);
2736 mutex_init(&fs_info->chunk_mutex);
2737 mutex_init(&fs_info->transaction_kthread_mutex);
2738 mutex_init(&fs_info->cleaner_mutex);
2739 mutex_init(&fs_info->ro_block_group_mutex);
2740 init_rwsem(&fs_info->commit_root_sem);
2741 init_rwsem(&fs_info->cleanup_work_sem);
2742 init_rwsem(&fs_info->subvol_sem);
2743 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2745 btrfs_init_dev_replace_locks(fs_info);
2746 btrfs_init_qgroup(fs_info);
2748 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2749 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2751 init_waitqueue_head(&fs_info->transaction_throttle);
2752 init_waitqueue_head(&fs_info->transaction_wait);
2753 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2754 init_waitqueue_head(&fs_info->async_submit_wait);
2756 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2758 /* Usable values until the real ones are cached from the superblock */
2759 fs_info->nodesize = 4096;
2760 fs_info->sectorsize = 4096;
2761 fs_info->stripesize = 4096;
2763 spin_lock_init(&fs_info->swapfile_pins_lock);
2764 fs_info->swapfile_pins = RB_ROOT;
2766 ret = btrfs_alloc_stripe_hash_table(fs_info);
2772 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2774 invalidate_bdev(fs_devices->latest_bdev);
2777 * Read super block and check the signature bytes only
2779 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2786 * We want to check superblock checksum, the type is stored inside.
2787 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2789 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2790 btrfs_err(fs_info, "superblock checksum mismatch");
2797 * super_copy is zeroed at allocation time and we never touch the
2798 * following bytes up to INFO_SIZE, the checksum is calculated from
2799 * the whole block of INFO_SIZE
2801 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2802 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2803 sizeof(*fs_info->super_for_commit));
2806 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2807 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2808 memcpy(fs_info->metadata_fsid,
2809 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE);
2811 memcpy(fs_info->metadata_fsid, fs_info->fsid, BTRFS_FSID_SIZE);
2814 ret = btrfs_validate_mount_super(fs_info);
2816 btrfs_err(fs_info, "superblock contains fatal errors");
2821 disk_super = fs_info->super_copy;
2822 if (!btrfs_super_root(disk_super))
2825 /* check FS state, whether FS is broken. */
2826 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2827 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2830 * run through our array of backup supers and setup
2831 * our ring pointer to the oldest one
2833 generation = btrfs_super_generation(disk_super);
2834 find_oldest_super_backup(fs_info, generation);
2837 * In the long term, we'll store the compression type in the super
2838 * block, and it'll be used for per file compression control.
2840 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2842 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2848 features = btrfs_super_incompat_flags(disk_super) &
2849 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2852 "cannot mount because of unsupported optional features (%llx)",
2858 features = btrfs_super_incompat_flags(disk_super);
2859 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2860 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2861 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2862 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2863 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2865 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2866 btrfs_info(fs_info, "has skinny extents");
2869 * flag our filesystem as having big metadata blocks if
2870 * they are bigger than the page size
2872 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2873 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2875 "flagging fs with big metadata feature");
2876 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2879 nodesize = btrfs_super_nodesize(disk_super);
2880 sectorsize = btrfs_super_sectorsize(disk_super);
2881 stripesize = sectorsize;
2882 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2883 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2885 /* Cache block sizes */
2886 fs_info->nodesize = nodesize;
2887 fs_info->sectorsize = sectorsize;
2888 fs_info->stripesize = stripesize;
2891 * mixed block groups end up with duplicate but slightly offset
2892 * extent buffers for the same range. It leads to corruptions
2894 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2895 (sectorsize != nodesize)) {
2897 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2898 nodesize, sectorsize);
2903 * Needn't use the lock because there is no other task which will
2906 btrfs_set_super_incompat_flags(disk_super, features);
2908 features = btrfs_super_compat_ro_flags(disk_super) &
2909 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2910 if (!sb_rdonly(sb) && features) {
2912 "cannot mount read-write because of unsupported optional features (%llx)",
2918 ret = btrfs_init_workqueues(fs_info, fs_devices);
2921 goto fail_sb_buffer;
2924 sb->s_bdi->congested_fn = btrfs_congested_fn;
2925 sb->s_bdi->congested_data = fs_info;
2926 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2927 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2928 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2929 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2931 sb->s_blocksize = sectorsize;
2932 sb->s_blocksize_bits = blksize_bits(sectorsize);
2933 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2935 mutex_lock(&fs_info->chunk_mutex);
2936 ret = btrfs_read_sys_array(fs_info);
2937 mutex_unlock(&fs_info->chunk_mutex);
2939 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2940 goto fail_sb_buffer;
2943 generation = btrfs_super_chunk_root_generation(disk_super);
2944 level = btrfs_super_chunk_root_level(disk_super);
2946 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2948 chunk_root->node = read_tree_block(fs_info,
2949 btrfs_super_chunk_root(disk_super),
2950 generation, level, NULL);
2951 if (IS_ERR(chunk_root->node) ||
2952 !extent_buffer_uptodate(chunk_root->node)) {
2953 btrfs_err(fs_info, "failed to read chunk root");
2954 if (!IS_ERR(chunk_root->node))
2955 free_extent_buffer(chunk_root->node);
2956 chunk_root->node = NULL;
2957 goto fail_tree_roots;
2959 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2960 chunk_root->commit_root = btrfs_root_node(chunk_root);
2962 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2963 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2965 ret = btrfs_read_chunk_tree(fs_info);
2967 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2968 goto fail_tree_roots;
2972 * Keep the devid that is marked to be the target device for the
2973 * device replace procedure
2975 btrfs_free_extra_devids(fs_devices, 0);
2977 if (!fs_devices->latest_bdev) {
2978 btrfs_err(fs_info, "failed to read devices");
2979 goto fail_tree_roots;
2983 generation = btrfs_super_generation(disk_super);
2984 level = btrfs_super_root_level(disk_super);
2986 tree_root->node = read_tree_block(fs_info,
2987 btrfs_super_root(disk_super),
2988 generation, level, NULL);
2989 if (IS_ERR(tree_root->node) ||
2990 !extent_buffer_uptodate(tree_root->node)) {
2991 btrfs_warn(fs_info, "failed to read tree root");
2992 if (!IS_ERR(tree_root->node))
2993 free_extent_buffer(tree_root->node);
2994 tree_root->node = NULL;
2995 goto recovery_tree_root;
2998 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2999 tree_root->commit_root = btrfs_root_node(tree_root);
3000 btrfs_set_root_refs(&tree_root->root_item, 1);
3002 mutex_lock(&tree_root->objectid_mutex);
3003 ret = btrfs_find_highest_objectid(tree_root,
3004 &tree_root->highest_objectid);
3006 mutex_unlock(&tree_root->objectid_mutex);
3007 goto recovery_tree_root;
3010 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3012 mutex_unlock(&tree_root->objectid_mutex);
3014 ret = btrfs_read_roots(fs_info);
3016 goto recovery_tree_root;
3018 fs_info->generation = generation;
3019 fs_info->last_trans_committed = generation;
3021 ret = btrfs_verify_dev_extents(fs_info);
3024 "failed to verify dev extents against chunks: %d",
3026 goto fail_block_groups;
3028 ret = btrfs_recover_balance(fs_info);
3030 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3031 goto fail_block_groups;
3034 ret = btrfs_init_dev_stats(fs_info);
3036 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3037 goto fail_block_groups;
3040 ret = btrfs_init_dev_replace(fs_info);
3042 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3043 goto fail_block_groups;
3046 btrfs_free_extra_devids(fs_devices, 1);
3048 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3050 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3052 goto fail_block_groups;
3055 ret = btrfs_sysfs_add_device(fs_devices);
3057 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3059 goto fail_fsdev_sysfs;
3062 ret = btrfs_sysfs_add_mounted(fs_info);
3064 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3065 goto fail_fsdev_sysfs;
3068 ret = btrfs_init_space_info(fs_info);
3070 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3074 ret = btrfs_read_block_groups(fs_info);
3076 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3080 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3082 "writeable mount is not allowed due to too many missing devices");
3086 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3088 if (IS_ERR(fs_info->cleaner_kthread))
3091 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3093 "btrfs-transaction");
3094 if (IS_ERR(fs_info->transaction_kthread))
3097 if (!btrfs_test_opt(fs_info, NOSSD) &&
3098 !fs_info->fs_devices->rotating) {
3099 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3103 * Mount does not set all options immediately, we can do it now and do
3104 * not have to wait for transaction commit
3106 btrfs_apply_pending_changes(fs_info);
3108 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3109 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3110 ret = btrfsic_mount(fs_info, fs_devices,
3111 btrfs_test_opt(fs_info,
3112 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3114 fs_info->check_integrity_print_mask);
3117 "failed to initialize integrity check module: %d",
3121 ret = btrfs_read_qgroup_config(fs_info);
3123 goto fail_trans_kthread;
3125 if (btrfs_build_ref_tree(fs_info))
3126 btrfs_err(fs_info, "couldn't build ref tree");
3128 /* do not make disk changes in broken FS or nologreplay is given */
3129 if (btrfs_super_log_root(disk_super) != 0 &&
3130 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3131 ret = btrfs_replay_log(fs_info, fs_devices);
3138 ret = btrfs_find_orphan_roots(fs_info);
3142 if (!sb_rdonly(sb)) {
3143 ret = btrfs_cleanup_fs_roots(fs_info);
3147 mutex_lock(&fs_info->cleaner_mutex);
3148 ret = btrfs_recover_relocation(tree_root);
3149 mutex_unlock(&fs_info->cleaner_mutex);
3151 btrfs_warn(fs_info, "failed to recover relocation: %d",
3158 location.objectid = BTRFS_FS_TREE_OBJECTID;
3159 location.type = BTRFS_ROOT_ITEM_KEY;
3160 location.offset = 0;
3162 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3163 if (IS_ERR(fs_info->fs_root)) {
3164 err = PTR_ERR(fs_info->fs_root);
3165 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3172 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3173 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3174 clear_free_space_tree = 1;
3175 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3176 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3177 btrfs_warn(fs_info, "free space tree is invalid");
3178 clear_free_space_tree = 1;
3181 if (clear_free_space_tree) {
3182 btrfs_info(fs_info, "clearing free space tree");
3183 ret = btrfs_clear_free_space_tree(fs_info);
3186 "failed to clear free space tree: %d", ret);
3187 close_ctree(fs_info);
3192 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3193 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3194 btrfs_info(fs_info, "creating free space tree");
3195 ret = btrfs_create_free_space_tree(fs_info);
3198 "failed to create free space tree: %d", ret);
3199 close_ctree(fs_info);
3204 down_read(&fs_info->cleanup_work_sem);
3205 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3206 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3207 up_read(&fs_info->cleanup_work_sem);
3208 close_ctree(fs_info);
3211 up_read(&fs_info->cleanup_work_sem);
3213 ret = btrfs_resume_balance_async(fs_info);
3215 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3216 close_ctree(fs_info);
3220 ret = btrfs_resume_dev_replace_async(fs_info);
3222 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3223 close_ctree(fs_info);
3227 btrfs_qgroup_rescan_resume(fs_info);
3229 if (!fs_info->uuid_root) {
3230 btrfs_info(fs_info, "creating UUID tree");
3231 ret = btrfs_create_uuid_tree(fs_info);
3234 "failed to create the UUID tree: %d", ret);
3235 close_ctree(fs_info);
3238 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3239 fs_info->generation !=
3240 btrfs_super_uuid_tree_generation(disk_super)) {
3241 btrfs_info(fs_info, "checking UUID tree");
3242 ret = btrfs_check_uuid_tree(fs_info);
3245 "failed to check the UUID tree: %d", ret);
3246 close_ctree(fs_info);
3250 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3252 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3255 * backuproot only affect mount behavior, and if open_ctree succeeded,
3256 * no need to keep the flag
3258 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3263 btrfs_free_qgroup_config(fs_info);
3265 kthread_stop(fs_info->transaction_kthread);
3266 btrfs_cleanup_transaction(fs_info);
3267 btrfs_free_fs_roots(fs_info);
3269 kthread_stop(fs_info->cleaner_kthread);
3272 * make sure we're done with the btree inode before we stop our
3275 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3278 btrfs_sysfs_remove_mounted(fs_info);
3281 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3284 btrfs_put_block_group_cache(fs_info);
3287 free_root_pointers(fs_info, 1);
3288 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3291 btrfs_stop_all_workers(fs_info);
3292 btrfs_free_block_groups(fs_info);
3295 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3297 iput(fs_info->btree_inode);
3299 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3300 fail_delalloc_bytes:
3301 percpu_counter_destroy(&fs_info->delalloc_bytes);
3302 fail_dirty_metadata_bytes:
3303 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3305 cleanup_srcu_struct(&fs_info->subvol_srcu);
3307 btrfs_free_stripe_hash_table(fs_info);
3308 btrfs_close_devices(fs_info->fs_devices);
3312 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3313 goto fail_tree_roots;
3315 free_root_pointers(fs_info, 0);
3317 /* don't use the log in recovery mode, it won't be valid */
3318 btrfs_set_super_log_root(disk_super, 0);
3320 /* we can't trust the free space cache either */
3321 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3323 ret = next_root_backup(fs_info, fs_info->super_copy,
3324 &num_backups_tried, &backup_index);
3326 goto fail_block_groups;
3327 goto retry_root_backup;
3329 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3331 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3334 set_buffer_uptodate(bh);
3336 struct btrfs_device *device = (struct btrfs_device *)
3339 btrfs_warn_rl_in_rcu(device->fs_info,
3340 "lost page write due to IO error on %s",
3341 rcu_str_deref(device->name));
3342 /* note, we don't set_buffer_write_io_error because we have
3343 * our own ways of dealing with the IO errors
3345 clear_buffer_uptodate(bh);
3346 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3352 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3353 struct buffer_head **bh_ret)
3355 struct buffer_head *bh;
3356 struct btrfs_super_block *super;
3359 bytenr = btrfs_sb_offset(copy_num);
3360 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3363 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3365 * If we fail to read from the underlying devices, as of now
3366 * the best option we have is to mark it EIO.
3371 super = (struct btrfs_super_block *)bh->b_data;
3372 if (btrfs_super_bytenr(super) != bytenr ||
3373 btrfs_super_magic(super) != BTRFS_MAGIC) {
3383 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3385 struct buffer_head *bh;
3386 struct buffer_head *latest = NULL;
3387 struct btrfs_super_block *super;
3392 /* we would like to check all the supers, but that would make
3393 * a btrfs mount succeed after a mkfs from a different FS.
3394 * So, we need to add a special mount option to scan for
3395 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3397 for (i = 0; i < 1; i++) {
3398 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3402 super = (struct btrfs_super_block *)bh->b_data;
3404 if (!latest || btrfs_super_generation(super) > transid) {
3407 transid = btrfs_super_generation(super);
3414 return ERR_PTR(ret);
3420 * Write superblock @sb to the @device. Do not wait for completion, all the
3421 * buffer heads we write are pinned.
3423 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3424 * the expected device size at commit time. Note that max_mirrors must be
3425 * same for write and wait phases.
3427 * Return number of errors when buffer head is not found or submission fails.
3429 static int write_dev_supers(struct btrfs_device *device,
3430 struct btrfs_super_block *sb, int max_mirrors)
3432 struct buffer_head *bh;
3440 if (max_mirrors == 0)
3441 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3443 for (i = 0; i < max_mirrors; i++) {
3444 bytenr = btrfs_sb_offset(i);
3445 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3446 device->commit_total_bytes)
3449 btrfs_set_super_bytenr(sb, bytenr);
3452 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3453 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3454 btrfs_csum_final(crc, sb->csum);
3456 /* One reference for us, and we leave it for the caller */
3457 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3458 BTRFS_SUPER_INFO_SIZE);
3460 btrfs_err(device->fs_info,
3461 "couldn't get super buffer head for bytenr %llu",
3467 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3469 /* one reference for submit_bh */
3472 set_buffer_uptodate(bh);
3474 bh->b_end_io = btrfs_end_buffer_write_sync;
3475 bh->b_private = device;
3478 * we fua the first super. The others we allow
3481 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3482 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3483 op_flags |= REQ_FUA;
3484 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3488 return errors < i ? 0 : -1;
3492 * Wait for write completion of superblocks done by write_dev_supers,
3493 * @max_mirrors same for write and wait phases.
3495 * Return number of errors when buffer head is not found or not marked up to
3498 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3500 struct buffer_head *bh;
3503 bool primary_failed = false;
3506 if (max_mirrors == 0)
3507 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3509 for (i = 0; i < max_mirrors; i++) {
3510 bytenr = btrfs_sb_offset(i);
3511 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3512 device->commit_total_bytes)
3515 bh = __find_get_block(device->bdev,
3516 bytenr / BTRFS_BDEV_BLOCKSIZE,
3517 BTRFS_SUPER_INFO_SIZE);
3521 primary_failed = true;
3525 if (!buffer_uptodate(bh)) {
3528 primary_failed = true;
3531 /* drop our reference */
3534 /* drop the reference from the writing run */
3538 /* log error, force error return */
3539 if (primary_failed) {
3540 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3545 return errors < i ? 0 : -1;
3549 * endio for the write_dev_flush, this will wake anyone waiting
3550 * for the barrier when it is done
3552 static void btrfs_end_empty_barrier(struct bio *bio)
3554 complete(bio->bi_private);
3558 * Submit a flush request to the device if it supports it. Error handling is
3559 * done in the waiting counterpart.
3561 static void write_dev_flush(struct btrfs_device *device)
3563 struct request_queue *q = bdev_get_queue(device->bdev);
3564 struct bio *bio = device->flush_bio;
3566 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3570 bio->bi_end_io = btrfs_end_empty_barrier;
3571 bio_set_dev(bio, device->bdev);
3572 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3573 init_completion(&device->flush_wait);
3574 bio->bi_private = &device->flush_wait;
3576 btrfsic_submit_bio(bio);
3577 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3581 * If the flush bio has been submitted by write_dev_flush, wait for it.
3583 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3585 struct bio *bio = device->flush_bio;
3587 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3590 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3591 wait_for_completion_io(&device->flush_wait);
3593 return bio->bi_status;
3596 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3598 if (!btrfs_check_rw_degradable(fs_info, NULL))
3604 * send an empty flush down to each device in parallel,
3605 * then wait for them
3607 static int barrier_all_devices(struct btrfs_fs_info *info)
3609 struct list_head *head;
3610 struct btrfs_device *dev;
3611 int errors_wait = 0;
3614 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3615 /* send down all the barriers */
3616 head = &info->fs_devices->devices;
3617 list_for_each_entry(dev, head, dev_list) {
3618 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3622 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3623 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3626 write_dev_flush(dev);
3627 dev->last_flush_error = BLK_STS_OK;
3630 /* wait for all the barriers */
3631 list_for_each_entry(dev, head, dev_list) {
3632 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3638 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3639 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3642 ret = wait_dev_flush(dev);
3644 dev->last_flush_error = ret;
3645 btrfs_dev_stat_inc_and_print(dev,
3646 BTRFS_DEV_STAT_FLUSH_ERRS);
3653 * At some point we need the status of all disks
3654 * to arrive at the volume status. So error checking
3655 * is being pushed to a separate loop.
3657 return check_barrier_error(info);
3662 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3665 int min_tolerated = INT_MAX;
3667 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3668 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3669 min_tolerated = min(min_tolerated,
3670 btrfs_raid_array[BTRFS_RAID_SINGLE].
3671 tolerated_failures);
3673 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3674 if (raid_type == BTRFS_RAID_SINGLE)
3676 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3678 min_tolerated = min(min_tolerated,
3679 btrfs_raid_array[raid_type].
3680 tolerated_failures);
3683 if (min_tolerated == INT_MAX) {
3684 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3688 return min_tolerated;
3691 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3693 struct list_head *head;
3694 struct btrfs_device *dev;
3695 struct btrfs_super_block *sb;
3696 struct btrfs_dev_item *dev_item;
3700 int total_errors = 0;
3703 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3706 * max_mirrors == 0 indicates we're from commit_transaction,
3707 * not from fsync where the tree roots in fs_info have not
3708 * been consistent on disk.
3710 if (max_mirrors == 0)
3711 backup_super_roots(fs_info);
3713 sb = fs_info->super_for_commit;
3714 dev_item = &sb->dev_item;
3716 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3717 head = &fs_info->fs_devices->devices;
3718 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3721 ret = barrier_all_devices(fs_info);
3724 &fs_info->fs_devices->device_list_mutex);
3725 btrfs_handle_fs_error(fs_info, ret,
3726 "errors while submitting device barriers.");
3731 list_for_each_entry(dev, head, dev_list) {
3736 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3737 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3740 btrfs_set_stack_device_generation(dev_item, 0);
3741 btrfs_set_stack_device_type(dev_item, dev->type);
3742 btrfs_set_stack_device_id(dev_item, dev->devid);
3743 btrfs_set_stack_device_total_bytes(dev_item,
3744 dev->commit_total_bytes);
3745 btrfs_set_stack_device_bytes_used(dev_item,
3746 dev->commit_bytes_used);
3747 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3748 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3749 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3750 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3751 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3754 flags = btrfs_super_flags(sb);
3755 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3757 ret = btrfs_validate_write_super(fs_info, sb);
3759 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3760 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3761 "unexpected superblock corruption detected");
3765 ret = write_dev_supers(dev, sb, max_mirrors);
3769 if (total_errors > max_errors) {
3770 btrfs_err(fs_info, "%d errors while writing supers",
3772 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3774 /* FUA is masked off if unsupported and can't be the reason */
3775 btrfs_handle_fs_error(fs_info, -EIO,
3776 "%d errors while writing supers",
3782 list_for_each_entry(dev, head, dev_list) {
3785 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3786 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3789 ret = wait_dev_supers(dev, max_mirrors);
3793 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3794 if (total_errors > max_errors) {
3795 btrfs_handle_fs_error(fs_info, -EIO,
3796 "%d errors while writing supers",
3803 /* Drop a fs root from the radix tree and free it. */
3804 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3805 struct btrfs_root *root)
3807 spin_lock(&fs_info->fs_roots_radix_lock);
3808 radix_tree_delete(&fs_info->fs_roots_radix,
3809 (unsigned long)root->root_key.objectid);
3810 spin_unlock(&fs_info->fs_roots_radix_lock);
3812 if (btrfs_root_refs(&root->root_item) == 0)
3813 synchronize_srcu(&fs_info->subvol_srcu);
3815 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3816 btrfs_free_log(NULL, root);
3817 if (root->reloc_root) {
3818 free_extent_buffer(root->reloc_root->node);
3819 free_extent_buffer(root->reloc_root->commit_root);
3820 btrfs_put_fs_root(root->reloc_root);
3821 root->reloc_root = NULL;
3825 if (root->free_ino_pinned)
3826 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3827 if (root->free_ino_ctl)
3828 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3829 btrfs_free_fs_root(root);
3832 void btrfs_free_fs_root(struct btrfs_root *root)
3834 iput(root->ino_cache_inode);
3835 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3837 free_anon_bdev(root->anon_dev);
3838 if (root->subv_writers)
3839 btrfs_free_subvolume_writers(root->subv_writers);
3840 free_extent_buffer(root->node);
3841 free_extent_buffer(root->commit_root);
3842 kfree(root->free_ino_ctl);
3843 kfree(root->free_ino_pinned);
3844 btrfs_put_fs_root(root);
3847 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3849 u64 root_objectid = 0;
3850 struct btrfs_root *gang[8];
3853 unsigned int ret = 0;
3857 index = srcu_read_lock(&fs_info->subvol_srcu);
3858 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3859 (void **)gang, root_objectid,
3862 srcu_read_unlock(&fs_info->subvol_srcu, index);
3865 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3867 for (i = 0; i < ret; i++) {
3868 /* Avoid to grab roots in dead_roots */
3869 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3873 /* grab all the search result for later use */
3874 gang[i] = btrfs_grab_fs_root(gang[i]);
3876 srcu_read_unlock(&fs_info->subvol_srcu, index);
3878 for (i = 0; i < ret; i++) {
3881 root_objectid = gang[i]->root_key.objectid;
3882 err = btrfs_orphan_cleanup(gang[i]);
3885 btrfs_put_fs_root(gang[i]);
3890 /* release the uncleaned roots due to error */
3891 for (; i < ret; i++) {
3893 btrfs_put_fs_root(gang[i]);
3898 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3900 struct btrfs_root *root = fs_info->tree_root;
3901 struct btrfs_trans_handle *trans;
3903 mutex_lock(&fs_info->cleaner_mutex);
3904 btrfs_run_delayed_iputs(fs_info);
3905 mutex_unlock(&fs_info->cleaner_mutex);
3906 wake_up_process(fs_info->cleaner_kthread);
3908 /* wait until ongoing cleanup work done */
3909 down_write(&fs_info->cleanup_work_sem);
3910 up_write(&fs_info->cleanup_work_sem);
3912 trans = btrfs_join_transaction(root);
3914 return PTR_ERR(trans);
3915 return btrfs_commit_transaction(trans);
3918 void close_ctree(struct btrfs_fs_info *fs_info)
3922 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3924 * We don't want the cleaner to start new transactions, add more delayed
3925 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3926 * because that frees the task_struct, and the transaction kthread might
3927 * still try to wake up the cleaner.
3929 kthread_park(fs_info->cleaner_kthread);
3931 /* wait for the qgroup rescan worker to stop */
3932 btrfs_qgroup_wait_for_completion(fs_info, false);
3934 /* wait for the uuid_scan task to finish */
3935 down(&fs_info->uuid_tree_rescan_sem);
3936 /* avoid complains from lockdep et al., set sem back to initial state */
3937 up(&fs_info->uuid_tree_rescan_sem);
3939 /* pause restriper - we want to resume on mount */
3940 btrfs_pause_balance(fs_info);
3942 btrfs_dev_replace_suspend_for_unmount(fs_info);
3944 btrfs_scrub_cancel(fs_info);
3946 /* wait for any defraggers to finish */
3947 wait_event(fs_info->transaction_wait,
3948 (atomic_read(&fs_info->defrag_running) == 0));
3950 /* clear out the rbtree of defraggable inodes */
3951 btrfs_cleanup_defrag_inodes(fs_info);
3953 cancel_work_sync(&fs_info->async_reclaim_work);
3955 if (!sb_rdonly(fs_info->sb)) {
3957 * The cleaner kthread is stopped, so do one final pass over
3958 * unused block groups.
3960 btrfs_delete_unused_bgs(fs_info);
3962 ret = btrfs_commit_super(fs_info);
3964 btrfs_err(fs_info, "commit super ret %d", ret);
3967 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3968 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3969 btrfs_error_commit_super(fs_info);
3971 kthread_stop(fs_info->transaction_kthread);
3972 kthread_stop(fs_info->cleaner_kthread);
3974 ASSERT(list_empty(&fs_info->delayed_iputs));
3975 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3977 btrfs_free_qgroup_config(fs_info);
3978 ASSERT(list_empty(&fs_info->delalloc_roots));
3980 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3981 btrfs_info(fs_info, "at unmount delalloc count %lld",
3982 percpu_counter_sum(&fs_info->delalloc_bytes));
3985 btrfs_sysfs_remove_mounted(fs_info);
3986 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3988 btrfs_free_fs_roots(fs_info);
3990 btrfs_put_block_group_cache(fs_info);
3993 * we must make sure there is not any read request to
3994 * submit after we stopping all workers.
3996 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3997 btrfs_stop_all_workers(fs_info);
3999 btrfs_free_block_groups(fs_info);
4001 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4002 free_root_pointers(fs_info, 1);
4004 iput(fs_info->btree_inode);
4006 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4007 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4008 btrfsic_unmount(fs_info->fs_devices);
4011 btrfs_close_devices(fs_info->fs_devices);
4012 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4014 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4015 percpu_counter_destroy(&fs_info->delalloc_bytes);
4016 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4017 cleanup_srcu_struct(&fs_info->subvol_srcu);
4019 btrfs_free_stripe_hash_table(fs_info);
4020 btrfs_free_ref_cache(fs_info);
4022 while (!list_empty(&fs_info->pinned_chunks)) {
4023 struct extent_map *em;
4025 em = list_first_entry(&fs_info->pinned_chunks,
4026 struct extent_map, list);
4027 list_del_init(&em->list);
4028 free_extent_map(em);
4032 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4036 struct inode *btree_inode = buf->pages[0]->mapping->host;
4038 ret = extent_buffer_uptodate(buf);
4042 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4043 parent_transid, atomic);
4049 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4051 struct btrfs_fs_info *fs_info;
4052 struct btrfs_root *root;
4053 u64 transid = btrfs_header_generation(buf);
4056 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4058 * This is a fast path so only do this check if we have sanity tests
4059 * enabled. Normal people shouldn't be using umapped buffers as dirty
4060 * outside of the sanity tests.
4062 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4065 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4066 fs_info = root->fs_info;
4067 btrfs_assert_tree_locked(buf);
4068 if (transid != fs_info->generation)
4069 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4070 buf->start, transid, fs_info->generation);
4071 was_dirty = set_extent_buffer_dirty(buf);
4073 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4075 fs_info->dirty_metadata_batch);
4076 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4078 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4079 * but item data not updated.
4080 * So here we should only check item pointers, not item data.
4082 if (btrfs_header_level(buf) == 0 &&
4083 btrfs_check_leaf_relaxed(fs_info, buf)) {
4084 btrfs_print_leaf(buf);
4090 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4094 * looks as though older kernels can get into trouble with
4095 * this code, they end up stuck in balance_dirty_pages forever
4099 if (current->flags & PF_MEMALLOC)
4103 btrfs_balance_delayed_items(fs_info);
4105 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4106 BTRFS_DIRTY_METADATA_THRESH,
4107 fs_info->dirty_metadata_batch);
4109 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4113 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4115 __btrfs_btree_balance_dirty(fs_info, 1);
4118 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4120 __btrfs_btree_balance_dirty(fs_info, 0);
4123 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4124 struct btrfs_key *first_key)
4126 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4127 struct btrfs_fs_info *fs_info = root->fs_info;
4129 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4133 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4135 /* cleanup FS via transaction */
4136 btrfs_cleanup_transaction(fs_info);
4138 mutex_lock(&fs_info->cleaner_mutex);
4139 btrfs_run_delayed_iputs(fs_info);
4140 mutex_unlock(&fs_info->cleaner_mutex);
4142 down_write(&fs_info->cleanup_work_sem);
4143 up_write(&fs_info->cleanup_work_sem);
4146 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4148 struct btrfs_ordered_extent *ordered;
4150 spin_lock(&root->ordered_extent_lock);
4152 * This will just short circuit the ordered completion stuff which will
4153 * make sure the ordered extent gets properly cleaned up.
4155 list_for_each_entry(ordered, &root->ordered_extents,
4157 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4158 spin_unlock(&root->ordered_extent_lock);
4161 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4163 struct btrfs_root *root;
4164 struct list_head splice;
4166 INIT_LIST_HEAD(&splice);
4168 spin_lock(&fs_info->ordered_root_lock);
4169 list_splice_init(&fs_info->ordered_roots, &splice);
4170 while (!list_empty(&splice)) {
4171 root = list_first_entry(&splice, struct btrfs_root,
4173 list_move_tail(&root->ordered_root,
4174 &fs_info->ordered_roots);
4176 spin_unlock(&fs_info->ordered_root_lock);
4177 btrfs_destroy_ordered_extents(root);
4180 spin_lock(&fs_info->ordered_root_lock);
4182 spin_unlock(&fs_info->ordered_root_lock);
4185 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4186 struct btrfs_fs_info *fs_info)
4188 struct rb_node *node;
4189 struct btrfs_delayed_ref_root *delayed_refs;
4190 struct btrfs_delayed_ref_node *ref;
4193 delayed_refs = &trans->delayed_refs;
4195 spin_lock(&delayed_refs->lock);
4196 if (atomic_read(&delayed_refs->num_entries) == 0) {
4197 spin_unlock(&delayed_refs->lock);
4198 btrfs_info(fs_info, "delayed_refs has NO entry");
4202 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4203 struct btrfs_delayed_ref_head *head;
4205 bool pin_bytes = false;
4207 head = rb_entry(node, struct btrfs_delayed_ref_head,
4209 if (!mutex_trylock(&head->mutex)) {
4210 refcount_inc(&head->refs);
4211 spin_unlock(&delayed_refs->lock);
4213 mutex_lock(&head->mutex);
4214 mutex_unlock(&head->mutex);
4215 btrfs_put_delayed_ref_head(head);
4216 spin_lock(&delayed_refs->lock);
4219 spin_lock(&head->lock);
4220 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4221 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4224 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4225 RB_CLEAR_NODE(&ref->ref_node);
4226 if (!list_empty(&ref->add_list))
4227 list_del(&ref->add_list);
4228 atomic_dec(&delayed_refs->num_entries);
4229 btrfs_put_delayed_ref(ref);
4231 if (head->must_insert_reserved)
4233 btrfs_free_delayed_extent_op(head->extent_op);
4234 delayed_refs->num_heads--;
4235 if (head->processing == 0)
4236 delayed_refs->num_heads_ready--;
4237 atomic_dec(&delayed_refs->num_entries);
4238 rb_erase_cached(&head->href_node, &delayed_refs->href_root);
4239 RB_CLEAR_NODE(&head->href_node);
4240 spin_unlock(&head->lock);
4241 spin_unlock(&delayed_refs->lock);
4242 mutex_unlock(&head->mutex);
4245 btrfs_pin_extent(fs_info, head->bytenr,
4246 head->num_bytes, 1);
4247 btrfs_put_delayed_ref_head(head);
4249 spin_lock(&delayed_refs->lock);
4252 spin_unlock(&delayed_refs->lock);
4257 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4259 struct btrfs_inode *btrfs_inode;
4260 struct list_head splice;
4262 INIT_LIST_HEAD(&splice);
4264 spin_lock(&root->delalloc_lock);
4265 list_splice_init(&root->delalloc_inodes, &splice);
4267 while (!list_empty(&splice)) {
4268 struct inode *inode = NULL;
4269 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4271 __btrfs_del_delalloc_inode(root, btrfs_inode);
4272 spin_unlock(&root->delalloc_lock);
4275 * Make sure we get a live inode and that it'll not disappear
4278 inode = igrab(&btrfs_inode->vfs_inode);
4280 invalidate_inode_pages2(inode->i_mapping);
4283 spin_lock(&root->delalloc_lock);
4285 spin_unlock(&root->delalloc_lock);
4288 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4290 struct btrfs_root *root;
4291 struct list_head splice;
4293 INIT_LIST_HEAD(&splice);
4295 spin_lock(&fs_info->delalloc_root_lock);
4296 list_splice_init(&fs_info->delalloc_roots, &splice);
4297 while (!list_empty(&splice)) {
4298 root = list_first_entry(&splice, struct btrfs_root,
4300 root = btrfs_grab_fs_root(root);
4302 spin_unlock(&fs_info->delalloc_root_lock);
4304 btrfs_destroy_delalloc_inodes(root);
4305 btrfs_put_fs_root(root);
4307 spin_lock(&fs_info->delalloc_root_lock);
4309 spin_unlock(&fs_info->delalloc_root_lock);
4312 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4313 struct extent_io_tree *dirty_pages,
4317 struct extent_buffer *eb;
4322 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4327 clear_extent_bits(dirty_pages, start, end, mark);
4328 while (start <= end) {
4329 eb = find_extent_buffer(fs_info, start);
4330 start += fs_info->nodesize;
4333 wait_on_extent_buffer_writeback(eb);
4335 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4337 clear_extent_buffer_dirty(eb);
4338 free_extent_buffer_stale(eb);
4345 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4346 struct extent_io_tree *pinned_extents)
4348 struct extent_io_tree *unpin;
4354 unpin = pinned_extents;
4358 * The btrfs_finish_extent_commit() may get the same range as
4359 * ours between find_first_extent_bit and clear_extent_dirty.
4360 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4361 * the same extent range.
4363 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4364 ret = find_first_extent_bit(unpin, 0, &start, &end,
4365 EXTENT_DIRTY, NULL);
4367 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4371 clear_extent_dirty(unpin, start, end);
4372 btrfs_error_unpin_extent_range(fs_info, start, end);
4373 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4378 if (unpin == &fs_info->freed_extents[0])
4379 unpin = &fs_info->freed_extents[1];
4381 unpin = &fs_info->freed_extents[0];
4389 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4391 struct inode *inode;
4393 inode = cache->io_ctl.inode;
4395 invalidate_inode_pages2(inode->i_mapping);
4396 BTRFS_I(inode)->generation = 0;
4397 cache->io_ctl.inode = NULL;
4400 btrfs_put_block_group(cache);
4403 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4404 struct btrfs_fs_info *fs_info)
4406 struct btrfs_block_group_cache *cache;
4408 spin_lock(&cur_trans->dirty_bgs_lock);
4409 while (!list_empty(&cur_trans->dirty_bgs)) {
4410 cache = list_first_entry(&cur_trans->dirty_bgs,
4411 struct btrfs_block_group_cache,
4414 if (!list_empty(&cache->io_list)) {
4415 spin_unlock(&cur_trans->dirty_bgs_lock);
4416 list_del_init(&cache->io_list);
4417 btrfs_cleanup_bg_io(cache);
4418 spin_lock(&cur_trans->dirty_bgs_lock);
4421 list_del_init(&cache->dirty_list);
4422 spin_lock(&cache->lock);
4423 cache->disk_cache_state = BTRFS_DC_ERROR;
4424 spin_unlock(&cache->lock);
4426 spin_unlock(&cur_trans->dirty_bgs_lock);
4427 btrfs_put_block_group(cache);
4428 spin_lock(&cur_trans->dirty_bgs_lock);
4430 spin_unlock(&cur_trans->dirty_bgs_lock);
4433 * Refer to the definition of io_bgs member for details why it's safe
4434 * to use it without any locking
4436 while (!list_empty(&cur_trans->io_bgs)) {
4437 cache = list_first_entry(&cur_trans->io_bgs,
4438 struct btrfs_block_group_cache,
4441 list_del_init(&cache->io_list);
4442 spin_lock(&cache->lock);
4443 cache->disk_cache_state = BTRFS_DC_ERROR;
4444 spin_unlock(&cache->lock);
4445 btrfs_cleanup_bg_io(cache);
4449 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4450 struct btrfs_fs_info *fs_info)
4452 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4453 ASSERT(list_empty(&cur_trans->dirty_bgs));
4454 ASSERT(list_empty(&cur_trans->io_bgs));
4456 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4458 cur_trans->state = TRANS_STATE_COMMIT_START;
4459 wake_up(&fs_info->transaction_blocked_wait);
4461 cur_trans->state = TRANS_STATE_UNBLOCKED;
4462 wake_up(&fs_info->transaction_wait);
4464 btrfs_destroy_delayed_inodes(fs_info);
4465 btrfs_assert_delayed_root_empty(fs_info);
4467 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4469 btrfs_destroy_pinned_extent(fs_info,
4470 fs_info->pinned_extents);
4472 cur_trans->state =TRANS_STATE_COMPLETED;
4473 wake_up(&cur_trans->commit_wait);
4476 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4478 struct btrfs_transaction *t;
4480 mutex_lock(&fs_info->transaction_kthread_mutex);
4482 spin_lock(&fs_info->trans_lock);
4483 while (!list_empty(&fs_info->trans_list)) {
4484 t = list_first_entry(&fs_info->trans_list,
4485 struct btrfs_transaction, list);
4486 if (t->state >= TRANS_STATE_COMMIT_START) {
4487 refcount_inc(&t->use_count);
4488 spin_unlock(&fs_info->trans_lock);
4489 btrfs_wait_for_commit(fs_info, t->transid);
4490 btrfs_put_transaction(t);
4491 spin_lock(&fs_info->trans_lock);
4494 if (t == fs_info->running_transaction) {
4495 t->state = TRANS_STATE_COMMIT_DOING;
4496 spin_unlock(&fs_info->trans_lock);
4498 * We wait for 0 num_writers since we don't hold a trans
4499 * handle open currently for this transaction.
4501 wait_event(t->writer_wait,
4502 atomic_read(&t->num_writers) == 0);
4504 spin_unlock(&fs_info->trans_lock);
4506 btrfs_cleanup_one_transaction(t, fs_info);
4508 spin_lock(&fs_info->trans_lock);
4509 if (t == fs_info->running_transaction)
4510 fs_info->running_transaction = NULL;
4511 list_del_init(&t->list);
4512 spin_unlock(&fs_info->trans_lock);
4514 btrfs_put_transaction(t);
4515 trace_btrfs_transaction_commit(fs_info->tree_root);
4516 spin_lock(&fs_info->trans_lock);
4518 spin_unlock(&fs_info->trans_lock);
4519 btrfs_destroy_all_ordered_extents(fs_info);
4520 btrfs_destroy_delayed_inodes(fs_info);
4521 btrfs_assert_delayed_root_empty(fs_info);
4522 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4523 btrfs_destroy_all_delalloc_inodes(fs_info);
4524 mutex_unlock(&fs_info->transaction_kthread_mutex);
4529 static const struct extent_io_ops btree_extent_io_ops = {
4530 /* mandatory callbacks */
4531 .submit_bio_hook = btree_submit_bio_hook,
4532 .readpage_end_io_hook = btree_readpage_end_io_hook,
4533 .readpage_io_failed_hook = btree_io_failed_hook,
4535 /* optional callbacks */