1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/scatterlist.h>
9 #include <linux/swap.h>
10 #include <linux/radix-tree.h>
11 #include <linux/writeback.h>
12 #include <linux/buffer_head.h>
13 #include <linux/workqueue.h>
14 #include <linux/kthread.h>
15 #include <linux/slab.h>
16 #include <linux/migrate.h>
17 #include <linux/ratelimit.h>
18 #include <linux/uuid.h>
19 #include <linux/semaphore.h>
20 #include <linux/error-injection.h>
21 #include <linux/crc32c.h>
22 #include <asm/unaligned.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
45 #include <asm/cpufeature.h>
48 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
49 BTRFS_HEADER_FLAG_RELOC |\
50 BTRFS_SUPER_FLAG_ERROR |\
51 BTRFS_SUPER_FLAG_SEEDING |\
52 BTRFS_SUPER_FLAG_METADUMP |\
53 BTRFS_SUPER_FLAG_METADUMP_V2)
55 static const struct extent_io_ops btree_extent_io_ops;
56 static void end_workqueue_fn(struct btrfs_work *work);
57 static void free_fs_root(struct btrfs_root *root);
58 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
59 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
60 struct btrfs_fs_info *fs_info);
61 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
62 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *dirty_pages,
65 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
66 struct extent_io_tree *pinned_extents);
67 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
68 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
71 * btrfs_end_io_wq structs are used to do processing in task context when an IO
72 * is complete. This is used during reads to verify checksums, and it is used
73 * by writes to insert metadata for new file extents after IO is complete.
75 struct btrfs_end_io_wq {
79 struct btrfs_fs_info *info;
81 enum btrfs_wq_endio_type metadata;
82 struct btrfs_work work;
85 static struct kmem_cache *btrfs_end_io_wq_cache;
87 int __init btrfs_end_io_wq_init(void)
89 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
90 sizeof(struct btrfs_end_io_wq),
94 if (!btrfs_end_io_wq_cache)
99 void __cold btrfs_end_io_wq_exit(void)
101 kmem_cache_destroy(btrfs_end_io_wq_cache);
105 * async submit bios are used to offload expensive checksumming
106 * onto the worker threads. They checksum file and metadata bios
107 * just before they are sent down the IO stack.
109 struct async_submit_bio {
111 struct btrfs_fs_info *fs_info;
113 extent_submit_bio_start_t *submit_bio_start;
114 extent_submit_bio_done_t *submit_bio_done;
116 unsigned long bio_flags;
118 * bio_offset is optional, can be used if the pages in the bio
119 * can't tell us where in the file the bio should go
122 struct btrfs_work work;
127 * Lockdep class keys for extent_buffer->lock's in this root. For a given
128 * eb, the lockdep key is determined by the btrfs_root it belongs to and
129 * the level the eb occupies in the tree.
131 * Different roots are used for different purposes and may nest inside each
132 * other and they require separate keysets. As lockdep keys should be
133 * static, assign keysets according to the purpose of the root as indicated
134 * by btrfs_root->objectid. This ensures that all special purpose roots
135 * have separate keysets.
137 * Lock-nesting across peer nodes is always done with the immediate parent
138 * node locked thus preventing deadlock. As lockdep doesn't know this, use
139 * subclass to avoid triggering lockdep warning in such cases.
141 * The key is set by the readpage_end_io_hook after the buffer has passed
142 * csum validation but before the pages are unlocked. It is also set by
143 * btrfs_init_new_buffer on freshly allocated blocks.
145 * We also add a check to make sure the highest level of the tree is the
146 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
147 * needs update as well.
149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
150 # if BTRFS_MAX_LEVEL != 8
154 static struct btrfs_lockdep_keyset {
155 u64 id; /* root objectid */
156 const char *name_stem; /* lock name stem */
157 char names[BTRFS_MAX_LEVEL + 1][20];
158 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
159 } btrfs_lockdep_keysets[] = {
160 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
161 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
162 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
163 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
164 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
165 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
166 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
167 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
168 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
169 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
170 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
171 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
172 { .id = 0, .name_stem = "tree" },
175 void __init btrfs_init_lockdep(void)
179 /* initialize lockdep class names */
180 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
181 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
183 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
184 snprintf(ks->names[j], sizeof(ks->names[j]),
185 "btrfs-%s-%02d", ks->name_stem, j);
189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
192 struct btrfs_lockdep_keyset *ks;
194 BUG_ON(level >= ARRAY_SIZE(ks->keys));
196 /* find the matching keyset, id 0 is the default entry */
197 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
198 if (ks->id == objectid)
201 lockdep_set_class_and_name(&eb->lock,
202 &ks->keys[level], ks->names[level]);
208 * extents on the btree inode are pretty simple, there's one extent
209 * that covers the entire device
211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
212 struct page *page, size_t pg_offset, u64 start, u64 len,
215 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
216 struct extent_map_tree *em_tree = &inode->extent_tree;
217 struct extent_map *em;
220 read_lock(&em_tree->lock);
221 em = lookup_extent_mapping(em_tree, start, len);
223 em->bdev = fs_info->fs_devices->latest_bdev;
224 read_unlock(&em_tree->lock);
227 read_unlock(&em_tree->lock);
229 em = alloc_extent_map();
231 em = ERR_PTR(-ENOMEM);
236 em->block_len = (u64)-1;
238 em->bdev = fs_info->fs_devices->latest_bdev;
240 write_lock(&em_tree->lock);
241 ret = add_extent_mapping(em_tree, em, 0);
242 if (ret == -EEXIST) {
244 em = lookup_extent_mapping(em_tree, start, len);
251 write_unlock(&em_tree->lock);
257 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
259 return crc32c(seed, data, len);
262 void btrfs_csum_final(u32 crc, u8 *result)
264 put_unaligned_le32(~crc, result);
268 * compute the csum for a btree block, and either verify it or write it
269 * into the csum field of the block.
271 static int csum_tree_block(struct btrfs_fs_info *fs_info,
272 struct extent_buffer *buf,
275 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
276 char result[BTRFS_CSUM_SIZE];
278 unsigned long cur_len;
279 unsigned long offset = BTRFS_CSUM_SIZE;
281 unsigned long map_start;
282 unsigned long map_len;
286 len = buf->len - offset;
288 err = map_private_extent_buffer(buf, offset, 32,
289 &kaddr, &map_start, &map_len);
292 cur_len = min(len, map_len - (offset - map_start));
293 crc = btrfs_csum_data(kaddr + offset - map_start,
298 memset(result, 0, BTRFS_CSUM_SIZE);
300 btrfs_csum_final(crc, result);
303 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
306 memcpy(&found, result, csum_size);
308 read_extent_buffer(buf, &val, 0, csum_size);
309 btrfs_warn_rl(fs_info,
310 "%s checksum verify failed on %llu wanted %X found %X level %d",
311 fs_info->sb->s_id, buf->start,
312 val, found, btrfs_header_level(buf));
316 write_extent_buffer(buf, result, 0, csum_size);
323 * we can't consider a given block up to date unless the transid of the
324 * block matches the transid in the parent node's pointer. This is how we
325 * detect blocks that either didn't get written at all or got written
326 * in the wrong place.
328 static int verify_parent_transid(struct extent_io_tree *io_tree,
329 struct extent_buffer *eb, u64 parent_transid,
332 struct extent_state *cached_state = NULL;
334 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
336 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
343 btrfs_tree_read_lock(eb);
344 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
347 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
349 if (extent_buffer_uptodate(eb) &&
350 btrfs_header_generation(eb) == parent_transid) {
354 btrfs_err_rl(eb->fs_info,
355 "parent transid verify failed on %llu wanted %llu found %llu",
357 parent_transid, btrfs_header_generation(eb));
361 * Things reading via commit roots that don't have normal protection,
362 * like send, can have a really old block in cache that may point at a
363 * block that has been freed and re-allocated. So don't clear uptodate
364 * if we find an eb that is under IO (dirty/writeback) because we could
365 * end up reading in the stale data and then writing it back out and
366 * making everybody very sad.
368 if (!extent_buffer_under_io(eb))
369 clear_extent_buffer_uptodate(eb);
371 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
374 btrfs_tree_read_unlock_blocking(eb);
379 * Return 0 if the superblock checksum type matches the checksum value of that
380 * algorithm. Pass the raw disk superblock data.
382 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
385 struct btrfs_super_block *disk_sb =
386 (struct btrfs_super_block *)raw_disk_sb;
387 u16 csum_type = btrfs_super_csum_type(disk_sb);
390 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
392 char result[sizeof(crc)];
395 * The super_block structure does not span the whole
396 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
397 * is filled with zeros and is included in the checksum.
399 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
400 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
401 btrfs_csum_final(crc, result);
403 if (memcmp(raw_disk_sb, result, sizeof(result)))
407 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
408 btrfs_err(fs_info, "unsupported checksum algorithm %u",
416 static int verify_level_key(struct btrfs_fs_info *fs_info,
417 struct extent_buffer *eb, int level,
418 struct btrfs_key *first_key)
421 struct btrfs_key found_key;
424 found_level = btrfs_header_level(eb);
425 if (found_level != level) {
426 #ifdef CONFIG_BTRFS_DEBUG
429 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
430 eb->start, level, found_level);
439 * For live tree block (new tree blocks in current transaction),
440 * we need proper lock context to avoid race, which is impossible here.
441 * So we only checks tree blocks which is read from disk, whose
442 * generation <= fs_info->last_trans_committed.
444 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
447 btrfs_node_key_to_cpu(eb, &found_key, 0);
449 btrfs_item_key_to_cpu(eb, &found_key, 0);
450 ret = btrfs_comp_cpu_keys(first_key, &found_key);
452 #ifdef CONFIG_BTRFS_DEBUG
456 "tree first key mismatch detected, bytenr=%llu key expected=(%llu, %u, %llu) has=(%llu, %u, %llu)",
457 eb->start, first_key->objectid, first_key->type,
458 first_key->offset, found_key.objectid,
459 found_key.type, found_key.offset);
466 * helper to read a given tree block, doing retries as required when
467 * the checksums don't match and we have alternate mirrors to try.
469 * @parent_transid: expected transid, skip check if 0
470 * @level: expected level, mandatory check
471 * @first_key: expected key of first slot, skip check if NULL
473 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
474 struct extent_buffer *eb,
475 u64 parent_transid, int level,
476 struct btrfs_key *first_key)
478 struct extent_io_tree *io_tree;
483 int failed_mirror = 0;
485 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
486 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
488 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
491 if (verify_parent_transid(io_tree, eb,
494 else if (verify_level_key(fs_info, eb, level,
502 * This buffer's crc is fine, but its contents are corrupted, so
503 * there is no reason to read the other copies, they won't be
506 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
510 num_copies = btrfs_num_copies(fs_info,
515 if (!failed_mirror) {
517 failed_mirror = eb->read_mirror;
521 if (mirror_num == failed_mirror)
524 if (mirror_num > num_copies)
528 if (failed && !ret && failed_mirror)
529 repair_eb_io_failure(fs_info, eb, failed_mirror);
535 * checksum a dirty tree block before IO. This has extra checks to make sure
536 * we only fill in the checksum field in the first page of a multi-page block
539 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
541 u64 start = page_offset(page);
543 struct extent_buffer *eb;
545 eb = (struct extent_buffer *)page->private;
546 if (page != eb->pages[0])
549 found_start = btrfs_header_bytenr(eb);
551 * Please do not consolidate these warnings into a single if.
552 * It is useful to know what went wrong.
554 if (WARN_ON(found_start != start))
556 if (WARN_ON(!PageUptodate(page)))
559 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
560 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
562 return csum_tree_block(fs_info, eb, 0);
565 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
566 struct extent_buffer *eb)
568 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
569 u8 fsid[BTRFS_FSID_SIZE];
572 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
574 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
578 fs_devices = fs_devices->seed;
583 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
584 u64 phy_offset, struct page *page,
585 u64 start, u64 end, int mirror)
589 struct extent_buffer *eb;
590 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
591 struct btrfs_fs_info *fs_info = root->fs_info;
598 eb = (struct extent_buffer *)page->private;
600 /* the pending IO might have been the only thing that kept this buffer
601 * in memory. Make sure we have a ref for all this other checks
603 extent_buffer_get(eb);
605 reads_done = atomic_dec_and_test(&eb->io_pages);
609 eb->read_mirror = mirror;
610 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
615 found_start = btrfs_header_bytenr(eb);
616 if (found_start != eb->start) {
617 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
618 found_start, eb->start);
622 if (check_tree_block_fsid(fs_info, eb)) {
623 btrfs_err_rl(fs_info, "bad fsid on block %llu",
628 found_level = btrfs_header_level(eb);
629 if (found_level >= BTRFS_MAX_LEVEL) {
630 btrfs_err(fs_info, "bad tree block level %d",
631 (int)btrfs_header_level(eb));
636 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
639 ret = csum_tree_block(fs_info, eb, 1);
644 * If this is a leaf block and it is corrupt, set the corrupt bit so
645 * that we don't try and read the other copies of this block, just
648 if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
649 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
653 if (found_level > 0 && btrfs_check_node(fs_info, eb))
657 set_extent_buffer_uptodate(eb);
660 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
661 btree_readahead_hook(eb, ret);
665 * our io error hook is going to dec the io pages
666 * again, we have to make sure it has something
669 atomic_inc(&eb->io_pages);
670 clear_extent_buffer_uptodate(eb);
672 free_extent_buffer(eb);
677 static int btree_io_failed_hook(struct page *page, int failed_mirror)
679 struct extent_buffer *eb;
681 eb = (struct extent_buffer *)page->private;
682 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
683 eb->read_mirror = failed_mirror;
684 atomic_dec(&eb->io_pages);
685 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
686 btree_readahead_hook(eb, -EIO);
687 return -EIO; /* we fixed nothing */
690 static void end_workqueue_bio(struct bio *bio)
692 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
693 struct btrfs_fs_info *fs_info;
694 struct btrfs_workqueue *wq;
695 btrfs_work_func_t func;
697 fs_info = end_io_wq->info;
698 end_io_wq->status = bio->bi_status;
700 if (bio_op(bio) == REQ_OP_WRITE) {
701 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
702 wq = fs_info->endio_meta_write_workers;
703 func = btrfs_endio_meta_write_helper;
704 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
705 wq = fs_info->endio_freespace_worker;
706 func = btrfs_freespace_write_helper;
707 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
708 wq = fs_info->endio_raid56_workers;
709 func = btrfs_endio_raid56_helper;
711 wq = fs_info->endio_write_workers;
712 func = btrfs_endio_write_helper;
715 if (unlikely(end_io_wq->metadata ==
716 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
717 wq = fs_info->endio_repair_workers;
718 func = btrfs_endio_repair_helper;
719 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
720 wq = fs_info->endio_raid56_workers;
721 func = btrfs_endio_raid56_helper;
722 } else if (end_io_wq->metadata) {
723 wq = fs_info->endio_meta_workers;
724 func = btrfs_endio_meta_helper;
726 wq = fs_info->endio_workers;
727 func = btrfs_endio_helper;
731 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
732 btrfs_queue_work(wq, &end_io_wq->work);
735 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
736 enum btrfs_wq_endio_type metadata)
738 struct btrfs_end_io_wq *end_io_wq;
740 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
742 return BLK_STS_RESOURCE;
744 end_io_wq->private = bio->bi_private;
745 end_io_wq->end_io = bio->bi_end_io;
746 end_io_wq->info = info;
747 end_io_wq->status = 0;
748 end_io_wq->bio = bio;
749 end_io_wq->metadata = metadata;
751 bio->bi_private = end_io_wq;
752 bio->bi_end_io = end_workqueue_bio;
756 static void run_one_async_start(struct btrfs_work *work)
758 struct async_submit_bio *async;
761 async = container_of(work, struct async_submit_bio, work);
762 ret = async->submit_bio_start(async->private_data, async->bio,
768 static void run_one_async_done(struct btrfs_work *work)
770 struct async_submit_bio *async;
772 async = container_of(work, struct async_submit_bio, work);
774 /* If an error occurred we just want to clean up the bio and move on */
776 async->bio->bi_status = async->status;
777 bio_endio(async->bio);
781 async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
784 static void run_one_async_free(struct btrfs_work *work)
786 struct async_submit_bio *async;
788 async = container_of(work, struct async_submit_bio, work);
792 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
793 int mirror_num, unsigned long bio_flags,
794 u64 bio_offset, void *private_data,
795 extent_submit_bio_start_t *submit_bio_start,
796 extent_submit_bio_done_t *submit_bio_done)
798 struct async_submit_bio *async;
800 async = kmalloc(sizeof(*async), GFP_NOFS);
802 return BLK_STS_RESOURCE;
804 async->private_data = private_data;
805 async->fs_info = fs_info;
807 async->mirror_num = mirror_num;
808 async->submit_bio_start = submit_bio_start;
809 async->submit_bio_done = submit_bio_done;
811 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
812 run_one_async_done, run_one_async_free);
814 async->bio_flags = bio_flags;
815 async->bio_offset = bio_offset;
819 if (op_is_sync(bio->bi_opf))
820 btrfs_set_work_high_priority(&async->work);
822 btrfs_queue_work(fs_info->workers, &async->work);
826 static blk_status_t btree_csum_one_bio(struct bio *bio)
828 struct bio_vec *bvec;
829 struct btrfs_root *root;
832 ASSERT(!bio_flagged(bio, BIO_CLONED));
833 bio_for_each_segment_all(bvec, bio, i) {
834 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
835 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
840 return errno_to_blk_status(ret);
843 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
847 * when we're called for a write, we're already in the async
848 * submission context. Just jump into btrfs_map_bio
850 return btree_csum_one_bio(bio);
853 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
856 struct inode *inode = private_data;
860 * when we're called for a write, we're already in the async
861 * submission context. Just jump into btrfs_map_bio
863 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
865 bio->bi_status = ret;
871 static int check_async_write(struct btrfs_inode *bi)
873 if (atomic_read(&bi->sync_writers))
876 if (static_cpu_has(X86_FEATURE_XMM4_2))
882 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
883 int mirror_num, unsigned long bio_flags,
886 struct inode *inode = private_data;
887 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
888 int async = check_async_write(BTRFS_I(inode));
891 if (bio_op(bio) != REQ_OP_WRITE) {
893 * called for a read, do the setup so that checksum validation
894 * can happen in the async kernel threads
896 ret = btrfs_bio_wq_end_io(fs_info, bio,
897 BTRFS_WQ_ENDIO_METADATA);
900 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
902 ret = btree_csum_one_bio(bio);
905 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
908 * kthread helpers are used to submit writes so that
909 * checksumming can happen in parallel across all CPUs
911 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
912 bio_offset, private_data,
913 btree_submit_bio_start,
914 btree_submit_bio_done);
922 bio->bi_status = ret;
927 #ifdef CONFIG_MIGRATION
928 static int btree_migratepage(struct address_space *mapping,
929 struct page *newpage, struct page *page,
930 enum migrate_mode mode)
933 * we can't safely write a btree page from here,
934 * we haven't done the locking hook
939 * Buffers may be managed in a filesystem specific way.
940 * We must have no buffers or drop them.
942 if (page_has_private(page) &&
943 !try_to_release_page(page, GFP_KERNEL))
945 return migrate_page(mapping, newpage, page, mode);
950 static int btree_writepages(struct address_space *mapping,
951 struct writeback_control *wbc)
953 struct btrfs_fs_info *fs_info;
956 if (wbc->sync_mode == WB_SYNC_NONE) {
958 if (wbc->for_kupdate)
961 fs_info = BTRFS_I(mapping->host)->root->fs_info;
962 /* this is a bit racy, but that's ok */
963 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
964 BTRFS_DIRTY_METADATA_THRESH);
968 return btree_write_cache_pages(mapping, wbc);
971 static int btree_readpage(struct file *file, struct page *page)
973 struct extent_io_tree *tree;
974 tree = &BTRFS_I(page->mapping->host)->io_tree;
975 return extent_read_full_page(tree, page, btree_get_extent, 0);
978 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
980 if (PageWriteback(page) || PageDirty(page))
983 return try_release_extent_buffer(page);
986 static void btree_invalidatepage(struct page *page, unsigned int offset,
989 struct extent_io_tree *tree;
990 tree = &BTRFS_I(page->mapping->host)->io_tree;
991 extent_invalidatepage(tree, page, offset);
992 btree_releasepage(page, GFP_NOFS);
993 if (PagePrivate(page)) {
994 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
995 "page private not zero on page %llu",
996 (unsigned long long)page_offset(page));
997 ClearPagePrivate(page);
998 set_page_private(page, 0);
1003 static int btree_set_page_dirty(struct page *page)
1006 struct extent_buffer *eb;
1008 BUG_ON(!PagePrivate(page));
1009 eb = (struct extent_buffer *)page->private;
1011 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1012 BUG_ON(!atomic_read(&eb->refs));
1013 btrfs_assert_tree_locked(eb);
1015 return __set_page_dirty_nobuffers(page);
1018 static const struct address_space_operations btree_aops = {
1019 .readpage = btree_readpage,
1020 .writepages = btree_writepages,
1021 .releasepage = btree_releasepage,
1022 .invalidatepage = btree_invalidatepage,
1023 #ifdef CONFIG_MIGRATION
1024 .migratepage = btree_migratepage,
1026 .set_page_dirty = btree_set_page_dirty,
1029 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1031 struct extent_buffer *buf = NULL;
1032 struct inode *btree_inode = fs_info->btree_inode;
1034 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1037 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1039 free_extent_buffer(buf);
1042 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1043 int mirror_num, struct extent_buffer **eb)
1045 struct extent_buffer *buf = NULL;
1046 struct inode *btree_inode = fs_info->btree_inode;
1047 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1050 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1054 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1056 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1059 free_extent_buffer(buf);
1063 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1064 free_extent_buffer(buf);
1066 } else if (extent_buffer_uptodate(buf)) {
1069 free_extent_buffer(buf);
1074 struct extent_buffer *btrfs_find_create_tree_block(
1075 struct btrfs_fs_info *fs_info,
1078 if (btrfs_is_testing(fs_info))
1079 return alloc_test_extent_buffer(fs_info, bytenr);
1080 return alloc_extent_buffer(fs_info, bytenr);
1084 int btrfs_write_tree_block(struct extent_buffer *buf)
1086 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1087 buf->start + buf->len - 1);
1090 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1092 filemap_fdatawait_range(buf->pages[0]->mapping,
1093 buf->start, buf->start + buf->len - 1);
1097 * Read tree block at logical address @bytenr and do variant basic but critical
1100 * @parent_transid: expected transid of this tree block, skip check if 0
1101 * @level: expected level, mandatory check
1102 * @first_key: expected key in slot 0, skip check if NULL
1104 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1105 u64 parent_transid, int level,
1106 struct btrfs_key *first_key)
1108 struct extent_buffer *buf = NULL;
1111 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1115 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1118 free_extent_buffer(buf);
1119 return ERR_PTR(ret);
1125 void clean_tree_block(struct btrfs_fs_info *fs_info,
1126 struct extent_buffer *buf)
1128 if (btrfs_header_generation(buf) ==
1129 fs_info->running_transaction->transid) {
1130 btrfs_assert_tree_locked(buf);
1132 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1133 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1135 fs_info->dirty_metadata_batch);
1136 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1137 btrfs_set_lock_blocking(buf);
1138 clear_extent_buffer_dirty(buf);
1143 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1145 struct btrfs_subvolume_writers *writers;
1148 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1150 return ERR_PTR(-ENOMEM);
1152 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1155 return ERR_PTR(ret);
1158 init_waitqueue_head(&writers->wait);
1163 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1165 percpu_counter_destroy(&writers->counter);
1169 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1172 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1174 root->commit_root = NULL;
1176 root->orphan_cleanup_state = 0;
1178 root->objectid = objectid;
1179 root->last_trans = 0;
1180 root->highest_objectid = 0;
1181 root->nr_delalloc_inodes = 0;
1182 root->nr_ordered_extents = 0;
1184 root->inode_tree = RB_ROOT;
1185 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1186 root->block_rsv = NULL;
1187 root->orphan_block_rsv = NULL;
1189 INIT_LIST_HEAD(&root->dirty_list);
1190 INIT_LIST_HEAD(&root->root_list);
1191 INIT_LIST_HEAD(&root->delalloc_inodes);
1192 INIT_LIST_HEAD(&root->delalloc_root);
1193 INIT_LIST_HEAD(&root->ordered_extents);
1194 INIT_LIST_HEAD(&root->ordered_root);
1195 INIT_LIST_HEAD(&root->logged_list[0]);
1196 INIT_LIST_HEAD(&root->logged_list[1]);
1197 spin_lock_init(&root->orphan_lock);
1198 spin_lock_init(&root->inode_lock);
1199 spin_lock_init(&root->delalloc_lock);
1200 spin_lock_init(&root->ordered_extent_lock);
1201 spin_lock_init(&root->accounting_lock);
1202 spin_lock_init(&root->log_extents_lock[0]);
1203 spin_lock_init(&root->log_extents_lock[1]);
1204 spin_lock_init(&root->qgroup_meta_rsv_lock);
1205 mutex_init(&root->objectid_mutex);
1206 mutex_init(&root->log_mutex);
1207 mutex_init(&root->ordered_extent_mutex);
1208 mutex_init(&root->delalloc_mutex);
1209 init_waitqueue_head(&root->log_writer_wait);
1210 init_waitqueue_head(&root->log_commit_wait[0]);
1211 init_waitqueue_head(&root->log_commit_wait[1]);
1212 INIT_LIST_HEAD(&root->log_ctxs[0]);
1213 INIT_LIST_HEAD(&root->log_ctxs[1]);
1214 atomic_set(&root->log_commit[0], 0);
1215 atomic_set(&root->log_commit[1], 0);
1216 atomic_set(&root->log_writers, 0);
1217 atomic_set(&root->log_batch, 0);
1218 atomic_set(&root->orphan_inodes, 0);
1219 refcount_set(&root->refs, 1);
1220 atomic_set(&root->will_be_snapshotted, 0);
1221 root->log_transid = 0;
1222 root->log_transid_committed = -1;
1223 root->last_log_commit = 0;
1225 extent_io_tree_init(&root->dirty_log_pages, NULL);
1227 memset(&root->root_key, 0, sizeof(root->root_key));
1228 memset(&root->root_item, 0, sizeof(root->root_item));
1229 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1231 root->defrag_trans_start = fs_info->generation;
1233 root->defrag_trans_start = 0;
1234 root->root_key.objectid = objectid;
1237 spin_lock_init(&root->root_item_lock);
1240 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1243 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1245 root->fs_info = fs_info;
1249 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1250 /* Should only be used by the testing infrastructure */
1251 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1253 struct btrfs_root *root;
1256 return ERR_PTR(-EINVAL);
1258 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1260 return ERR_PTR(-ENOMEM);
1262 /* We don't use the stripesize in selftest, set it as sectorsize */
1263 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1264 root->alloc_bytenr = 0;
1270 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1271 struct btrfs_fs_info *fs_info,
1274 struct extent_buffer *leaf;
1275 struct btrfs_root *tree_root = fs_info->tree_root;
1276 struct btrfs_root *root;
1277 struct btrfs_key key;
1279 uuid_le uuid = NULL_UUID_LE;
1281 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1283 return ERR_PTR(-ENOMEM);
1285 __setup_root(root, fs_info, objectid);
1286 root->root_key.objectid = objectid;
1287 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1288 root->root_key.offset = 0;
1290 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1292 ret = PTR_ERR(leaf);
1297 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1298 btrfs_set_header_bytenr(leaf, leaf->start);
1299 btrfs_set_header_generation(leaf, trans->transid);
1300 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1301 btrfs_set_header_owner(leaf, objectid);
1304 write_extent_buffer_fsid(leaf, fs_info->fsid);
1305 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1306 btrfs_mark_buffer_dirty(leaf);
1308 root->commit_root = btrfs_root_node(root);
1309 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1311 root->root_item.flags = 0;
1312 root->root_item.byte_limit = 0;
1313 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1314 btrfs_set_root_generation(&root->root_item, trans->transid);
1315 btrfs_set_root_level(&root->root_item, 0);
1316 btrfs_set_root_refs(&root->root_item, 1);
1317 btrfs_set_root_used(&root->root_item, leaf->len);
1318 btrfs_set_root_last_snapshot(&root->root_item, 0);
1319 btrfs_set_root_dirid(&root->root_item, 0);
1320 if (is_fstree(objectid))
1322 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1323 root->root_item.drop_level = 0;
1325 key.objectid = objectid;
1326 key.type = BTRFS_ROOT_ITEM_KEY;
1328 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1332 btrfs_tree_unlock(leaf);
1338 btrfs_tree_unlock(leaf);
1339 free_extent_buffer(root->commit_root);
1340 free_extent_buffer(leaf);
1344 return ERR_PTR(ret);
1347 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1348 struct btrfs_fs_info *fs_info)
1350 struct btrfs_root *root;
1351 struct extent_buffer *leaf;
1353 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1355 return ERR_PTR(-ENOMEM);
1357 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1359 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1360 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1361 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1364 * DON'T set REF_COWS for log trees
1366 * log trees do not get reference counted because they go away
1367 * before a real commit is actually done. They do store pointers
1368 * to file data extents, and those reference counts still get
1369 * updated (along with back refs to the log tree).
1372 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1376 return ERR_CAST(leaf);
1379 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1380 btrfs_set_header_bytenr(leaf, leaf->start);
1381 btrfs_set_header_generation(leaf, trans->transid);
1382 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1383 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1386 write_extent_buffer_fsid(root->node, fs_info->fsid);
1387 btrfs_mark_buffer_dirty(root->node);
1388 btrfs_tree_unlock(root->node);
1392 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1393 struct btrfs_fs_info *fs_info)
1395 struct btrfs_root *log_root;
1397 log_root = alloc_log_tree(trans, fs_info);
1398 if (IS_ERR(log_root))
1399 return PTR_ERR(log_root);
1400 WARN_ON(fs_info->log_root_tree);
1401 fs_info->log_root_tree = log_root;
1405 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1406 struct btrfs_root *root)
1408 struct btrfs_fs_info *fs_info = root->fs_info;
1409 struct btrfs_root *log_root;
1410 struct btrfs_inode_item *inode_item;
1412 log_root = alloc_log_tree(trans, fs_info);
1413 if (IS_ERR(log_root))
1414 return PTR_ERR(log_root);
1416 log_root->last_trans = trans->transid;
1417 log_root->root_key.offset = root->root_key.objectid;
1419 inode_item = &log_root->root_item.inode;
1420 btrfs_set_stack_inode_generation(inode_item, 1);
1421 btrfs_set_stack_inode_size(inode_item, 3);
1422 btrfs_set_stack_inode_nlink(inode_item, 1);
1423 btrfs_set_stack_inode_nbytes(inode_item,
1425 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1427 btrfs_set_root_node(&log_root->root_item, log_root->node);
1429 WARN_ON(root->log_root);
1430 root->log_root = log_root;
1431 root->log_transid = 0;
1432 root->log_transid_committed = -1;
1433 root->last_log_commit = 0;
1437 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1438 struct btrfs_key *key)
1440 struct btrfs_root *root;
1441 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1442 struct btrfs_path *path;
1447 path = btrfs_alloc_path();
1449 return ERR_PTR(-ENOMEM);
1451 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1457 __setup_root(root, fs_info, key->objectid);
1459 ret = btrfs_find_root(tree_root, key, path,
1460 &root->root_item, &root->root_key);
1467 generation = btrfs_root_generation(&root->root_item);
1468 level = btrfs_root_level(&root->root_item);
1469 root->node = read_tree_block(fs_info,
1470 btrfs_root_bytenr(&root->root_item),
1471 generation, level, NULL);
1472 if (IS_ERR(root->node)) {
1473 ret = PTR_ERR(root->node);
1475 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1477 free_extent_buffer(root->node);
1480 root->commit_root = btrfs_root_node(root);
1482 btrfs_free_path(path);
1488 root = ERR_PTR(ret);
1492 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1493 struct btrfs_key *location)
1495 struct btrfs_root *root;
1497 root = btrfs_read_tree_root(tree_root, location);
1501 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1502 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1503 btrfs_check_and_init_root_item(&root->root_item);
1509 int btrfs_init_fs_root(struct btrfs_root *root)
1512 struct btrfs_subvolume_writers *writers;
1514 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1515 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1517 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1522 writers = btrfs_alloc_subvolume_writers();
1523 if (IS_ERR(writers)) {
1524 ret = PTR_ERR(writers);
1527 root->subv_writers = writers;
1529 btrfs_init_free_ino_ctl(root);
1530 spin_lock_init(&root->ino_cache_lock);
1531 init_waitqueue_head(&root->ino_cache_wait);
1533 ret = get_anon_bdev(&root->anon_dev);
1537 mutex_lock(&root->objectid_mutex);
1538 ret = btrfs_find_highest_objectid(root,
1539 &root->highest_objectid);
1541 mutex_unlock(&root->objectid_mutex);
1545 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1547 mutex_unlock(&root->objectid_mutex);
1551 /* the caller is responsible to call free_fs_root */
1555 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1558 struct btrfs_root *root;
1560 spin_lock(&fs_info->fs_roots_radix_lock);
1561 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1562 (unsigned long)root_id);
1563 spin_unlock(&fs_info->fs_roots_radix_lock);
1567 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1568 struct btrfs_root *root)
1572 ret = radix_tree_preload(GFP_NOFS);
1576 spin_lock(&fs_info->fs_roots_radix_lock);
1577 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1578 (unsigned long)root->root_key.objectid,
1581 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1582 spin_unlock(&fs_info->fs_roots_radix_lock);
1583 radix_tree_preload_end();
1588 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1589 struct btrfs_key *location,
1592 struct btrfs_root *root;
1593 struct btrfs_path *path;
1594 struct btrfs_key key;
1597 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1598 return fs_info->tree_root;
1599 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1600 return fs_info->extent_root;
1601 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1602 return fs_info->chunk_root;
1603 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1604 return fs_info->dev_root;
1605 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1606 return fs_info->csum_root;
1607 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1608 return fs_info->quota_root ? fs_info->quota_root :
1610 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1611 return fs_info->uuid_root ? fs_info->uuid_root :
1613 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1614 return fs_info->free_space_root ? fs_info->free_space_root :
1617 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1619 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1620 return ERR_PTR(-ENOENT);
1624 root = btrfs_read_fs_root(fs_info->tree_root, location);
1628 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1633 ret = btrfs_init_fs_root(root);
1637 path = btrfs_alloc_path();
1642 key.objectid = BTRFS_ORPHAN_OBJECTID;
1643 key.type = BTRFS_ORPHAN_ITEM_KEY;
1644 key.offset = location->objectid;
1646 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1647 btrfs_free_path(path);
1651 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1653 ret = btrfs_insert_fs_root(fs_info, root);
1655 if (ret == -EEXIST) {
1664 return ERR_PTR(ret);
1667 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1669 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1671 struct btrfs_device *device;
1672 struct backing_dev_info *bdi;
1675 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1678 bdi = device->bdev->bd_bdi;
1679 if (bdi_congested(bdi, bdi_bits)) {
1689 * called by the kthread helper functions to finally call the bio end_io
1690 * functions. This is where read checksum verification actually happens
1692 static void end_workqueue_fn(struct btrfs_work *work)
1695 struct btrfs_end_io_wq *end_io_wq;
1697 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1698 bio = end_io_wq->bio;
1700 bio->bi_status = end_io_wq->status;
1701 bio->bi_private = end_io_wq->private;
1702 bio->bi_end_io = end_io_wq->end_io;
1703 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1707 static int cleaner_kthread(void *arg)
1709 struct btrfs_root *root = arg;
1710 struct btrfs_fs_info *fs_info = root->fs_info;
1712 struct btrfs_trans_handle *trans;
1717 /* Make the cleaner go to sleep early. */
1718 if (btrfs_need_cleaner_sleep(fs_info))
1722 * Do not do anything if we might cause open_ctree() to block
1723 * before we have finished mounting the filesystem.
1725 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1728 if (!mutex_trylock(&fs_info->cleaner_mutex))
1732 * Avoid the problem that we change the status of the fs
1733 * during the above check and trylock.
1735 if (btrfs_need_cleaner_sleep(fs_info)) {
1736 mutex_unlock(&fs_info->cleaner_mutex);
1740 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1741 btrfs_run_delayed_iputs(fs_info);
1742 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1744 again = btrfs_clean_one_deleted_snapshot(root);
1745 mutex_unlock(&fs_info->cleaner_mutex);
1748 * The defragger has dealt with the R/O remount and umount,
1749 * needn't do anything special here.
1751 btrfs_run_defrag_inodes(fs_info);
1754 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1755 * with relocation (btrfs_relocate_chunk) and relocation
1756 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1757 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1758 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1759 * unused block groups.
1761 btrfs_delete_unused_bgs(fs_info);
1764 set_current_state(TASK_INTERRUPTIBLE);
1765 if (!kthread_should_stop())
1767 __set_current_state(TASK_RUNNING);
1769 } while (!kthread_should_stop());
1772 * Transaction kthread is stopped before us and wakes us up.
1773 * However we might have started a new transaction and COWed some
1774 * tree blocks when deleting unused block groups for example. So
1775 * make sure we commit the transaction we started to have a clean
1776 * shutdown when evicting the btree inode - if it has dirty pages
1777 * when we do the final iput() on it, eviction will trigger a
1778 * writeback for it which will fail with null pointer dereferences
1779 * since work queues and other resources were already released and
1780 * destroyed by the time the iput/eviction/writeback is made.
1782 trans = btrfs_attach_transaction(root);
1783 if (IS_ERR(trans)) {
1784 if (PTR_ERR(trans) != -ENOENT)
1786 "cleaner transaction attach returned %ld",
1791 ret = btrfs_commit_transaction(trans);
1794 "cleaner open transaction commit returned %d",
1801 static int transaction_kthread(void *arg)
1803 struct btrfs_root *root = arg;
1804 struct btrfs_fs_info *fs_info = root->fs_info;
1805 struct btrfs_trans_handle *trans;
1806 struct btrfs_transaction *cur;
1809 unsigned long delay;
1813 cannot_commit = false;
1814 delay = HZ * fs_info->commit_interval;
1815 mutex_lock(&fs_info->transaction_kthread_mutex);
1817 spin_lock(&fs_info->trans_lock);
1818 cur = fs_info->running_transaction;
1820 spin_unlock(&fs_info->trans_lock);
1824 now = get_seconds();
1825 if (cur->state < TRANS_STATE_BLOCKED &&
1826 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1827 (now < cur->start_time ||
1828 now - cur->start_time < fs_info->commit_interval)) {
1829 spin_unlock(&fs_info->trans_lock);
1833 transid = cur->transid;
1834 spin_unlock(&fs_info->trans_lock);
1836 /* If the file system is aborted, this will always fail. */
1837 trans = btrfs_attach_transaction(root);
1838 if (IS_ERR(trans)) {
1839 if (PTR_ERR(trans) != -ENOENT)
1840 cannot_commit = true;
1843 if (transid == trans->transid) {
1844 btrfs_commit_transaction(trans);
1846 btrfs_end_transaction(trans);
1849 wake_up_process(fs_info->cleaner_kthread);
1850 mutex_unlock(&fs_info->transaction_kthread_mutex);
1852 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1853 &fs_info->fs_state)))
1854 btrfs_cleanup_transaction(fs_info);
1855 if (!kthread_should_stop() &&
1856 (!btrfs_transaction_blocked(fs_info) ||
1858 schedule_timeout_interruptible(delay);
1859 } while (!kthread_should_stop());
1864 * this will find the highest generation in the array of
1865 * root backups. The index of the highest array is returned,
1866 * or -1 if we can't find anything.
1868 * We check to make sure the array is valid by comparing the
1869 * generation of the latest root in the array with the generation
1870 * in the super block. If they don't match we pitch it.
1872 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1875 int newest_index = -1;
1876 struct btrfs_root_backup *root_backup;
1879 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1880 root_backup = info->super_copy->super_roots + i;
1881 cur = btrfs_backup_tree_root_gen(root_backup);
1882 if (cur == newest_gen)
1886 /* check to see if we actually wrapped around */
1887 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1888 root_backup = info->super_copy->super_roots;
1889 cur = btrfs_backup_tree_root_gen(root_backup);
1890 if (cur == newest_gen)
1893 return newest_index;
1898 * find the oldest backup so we know where to store new entries
1899 * in the backup array. This will set the backup_root_index
1900 * field in the fs_info struct
1902 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1905 int newest_index = -1;
1907 newest_index = find_newest_super_backup(info, newest_gen);
1908 /* if there was garbage in there, just move along */
1909 if (newest_index == -1) {
1910 info->backup_root_index = 0;
1912 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1917 * copy all the root pointers into the super backup array.
1918 * this will bump the backup pointer by one when it is
1921 static void backup_super_roots(struct btrfs_fs_info *info)
1924 struct btrfs_root_backup *root_backup;
1927 next_backup = info->backup_root_index;
1928 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1929 BTRFS_NUM_BACKUP_ROOTS;
1932 * just overwrite the last backup if we're at the same generation
1933 * this happens only at umount
1935 root_backup = info->super_for_commit->super_roots + last_backup;
1936 if (btrfs_backup_tree_root_gen(root_backup) ==
1937 btrfs_header_generation(info->tree_root->node))
1938 next_backup = last_backup;
1940 root_backup = info->super_for_commit->super_roots + next_backup;
1943 * make sure all of our padding and empty slots get zero filled
1944 * regardless of which ones we use today
1946 memset(root_backup, 0, sizeof(*root_backup));
1948 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1950 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1951 btrfs_set_backup_tree_root_gen(root_backup,
1952 btrfs_header_generation(info->tree_root->node));
1954 btrfs_set_backup_tree_root_level(root_backup,
1955 btrfs_header_level(info->tree_root->node));
1957 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1958 btrfs_set_backup_chunk_root_gen(root_backup,
1959 btrfs_header_generation(info->chunk_root->node));
1960 btrfs_set_backup_chunk_root_level(root_backup,
1961 btrfs_header_level(info->chunk_root->node));
1963 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1964 btrfs_set_backup_extent_root_gen(root_backup,
1965 btrfs_header_generation(info->extent_root->node));
1966 btrfs_set_backup_extent_root_level(root_backup,
1967 btrfs_header_level(info->extent_root->node));
1970 * we might commit during log recovery, which happens before we set
1971 * the fs_root. Make sure it is valid before we fill it in.
1973 if (info->fs_root && info->fs_root->node) {
1974 btrfs_set_backup_fs_root(root_backup,
1975 info->fs_root->node->start);
1976 btrfs_set_backup_fs_root_gen(root_backup,
1977 btrfs_header_generation(info->fs_root->node));
1978 btrfs_set_backup_fs_root_level(root_backup,
1979 btrfs_header_level(info->fs_root->node));
1982 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1983 btrfs_set_backup_dev_root_gen(root_backup,
1984 btrfs_header_generation(info->dev_root->node));
1985 btrfs_set_backup_dev_root_level(root_backup,
1986 btrfs_header_level(info->dev_root->node));
1988 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1989 btrfs_set_backup_csum_root_gen(root_backup,
1990 btrfs_header_generation(info->csum_root->node));
1991 btrfs_set_backup_csum_root_level(root_backup,
1992 btrfs_header_level(info->csum_root->node));
1994 btrfs_set_backup_total_bytes(root_backup,
1995 btrfs_super_total_bytes(info->super_copy));
1996 btrfs_set_backup_bytes_used(root_backup,
1997 btrfs_super_bytes_used(info->super_copy));
1998 btrfs_set_backup_num_devices(root_backup,
1999 btrfs_super_num_devices(info->super_copy));
2002 * if we don't copy this out to the super_copy, it won't get remembered
2003 * for the next commit
2005 memcpy(&info->super_copy->super_roots,
2006 &info->super_for_commit->super_roots,
2007 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2011 * this copies info out of the root backup array and back into
2012 * the in-memory super block. It is meant to help iterate through
2013 * the array, so you send it the number of backups you've already
2014 * tried and the last backup index you used.
2016 * this returns -1 when it has tried all the backups
2018 static noinline int next_root_backup(struct btrfs_fs_info *info,
2019 struct btrfs_super_block *super,
2020 int *num_backups_tried, int *backup_index)
2022 struct btrfs_root_backup *root_backup;
2023 int newest = *backup_index;
2025 if (*num_backups_tried == 0) {
2026 u64 gen = btrfs_super_generation(super);
2028 newest = find_newest_super_backup(info, gen);
2032 *backup_index = newest;
2033 *num_backups_tried = 1;
2034 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2035 /* we've tried all the backups, all done */
2038 /* jump to the next oldest backup */
2039 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2040 BTRFS_NUM_BACKUP_ROOTS;
2041 *backup_index = newest;
2042 *num_backups_tried += 1;
2044 root_backup = super->super_roots + newest;
2046 btrfs_set_super_generation(super,
2047 btrfs_backup_tree_root_gen(root_backup));
2048 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2049 btrfs_set_super_root_level(super,
2050 btrfs_backup_tree_root_level(root_backup));
2051 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2054 * fixme: the total bytes and num_devices need to match or we should
2057 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2058 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2062 /* helper to cleanup workers */
2063 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2065 btrfs_destroy_workqueue(fs_info->fixup_workers);
2066 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2067 btrfs_destroy_workqueue(fs_info->workers);
2068 btrfs_destroy_workqueue(fs_info->endio_workers);
2069 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2070 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2071 btrfs_destroy_workqueue(fs_info->rmw_workers);
2072 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2073 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2074 btrfs_destroy_workqueue(fs_info->submit_workers);
2075 btrfs_destroy_workqueue(fs_info->delayed_workers);
2076 btrfs_destroy_workqueue(fs_info->caching_workers);
2077 btrfs_destroy_workqueue(fs_info->readahead_workers);
2078 btrfs_destroy_workqueue(fs_info->flush_workers);
2079 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2080 btrfs_destroy_workqueue(fs_info->extent_workers);
2082 * Now that all other work queues are destroyed, we can safely destroy
2083 * the queues used for metadata I/O, since tasks from those other work
2084 * queues can do metadata I/O operations.
2086 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2087 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2090 static void free_root_extent_buffers(struct btrfs_root *root)
2093 free_extent_buffer(root->node);
2094 free_extent_buffer(root->commit_root);
2096 root->commit_root = NULL;
2100 /* helper to cleanup tree roots */
2101 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2103 free_root_extent_buffers(info->tree_root);
2105 free_root_extent_buffers(info->dev_root);
2106 free_root_extent_buffers(info->extent_root);
2107 free_root_extent_buffers(info->csum_root);
2108 free_root_extent_buffers(info->quota_root);
2109 free_root_extent_buffers(info->uuid_root);
2111 free_root_extent_buffers(info->chunk_root);
2112 free_root_extent_buffers(info->free_space_root);
2115 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2118 struct btrfs_root *gang[8];
2121 while (!list_empty(&fs_info->dead_roots)) {
2122 gang[0] = list_entry(fs_info->dead_roots.next,
2123 struct btrfs_root, root_list);
2124 list_del(&gang[0]->root_list);
2126 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2127 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2129 free_extent_buffer(gang[0]->node);
2130 free_extent_buffer(gang[0]->commit_root);
2131 btrfs_put_fs_root(gang[0]);
2136 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2141 for (i = 0; i < ret; i++)
2142 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2145 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2146 btrfs_free_log_root_tree(NULL, fs_info);
2147 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2151 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2153 mutex_init(&fs_info->scrub_lock);
2154 atomic_set(&fs_info->scrubs_running, 0);
2155 atomic_set(&fs_info->scrub_pause_req, 0);
2156 atomic_set(&fs_info->scrubs_paused, 0);
2157 atomic_set(&fs_info->scrub_cancel_req, 0);
2158 init_waitqueue_head(&fs_info->scrub_pause_wait);
2159 fs_info->scrub_workers_refcnt = 0;
2162 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2164 spin_lock_init(&fs_info->balance_lock);
2165 mutex_init(&fs_info->balance_mutex);
2166 atomic_set(&fs_info->balance_pause_req, 0);
2167 atomic_set(&fs_info->balance_cancel_req, 0);
2168 fs_info->balance_ctl = NULL;
2169 init_waitqueue_head(&fs_info->balance_wait_q);
2172 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2174 struct inode *inode = fs_info->btree_inode;
2176 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2177 set_nlink(inode, 1);
2179 * we set the i_size on the btree inode to the max possible int.
2180 * the real end of the address space is determined by all of
2181 * the devices in the system
2183 inode->i_size = OFFSET_MAX;
2184 inode->i_mapping->a_ops = &btree_aops;
2186 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2187 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2188 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2189 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2191 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2193 BTRFS_I(inode)->root = fs_info->tree_root;
2194 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2195 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2196 btrfs_insert_inode_hash(inode);
2199 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2201 fs_info->dev_replace.lock_owner = 0;
2202 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2203 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2204 rwlock_init(&fs_info->dev_replace.lock);
2205 atomic_set(&fs_info->dev_replace.read_locks, 0);
2206 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2207 init_waitqueue_head(&fs_info->replace_wait);
2208 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2211 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2213 spin_lock_init(&fs_info->qgroup_lock);
2214 mutex_init(&fs_info->qgroup_ioctl_lock);
2215 fs_info->qgroup_tree = RB_ROOT;
2216 fs_info->qgroup_op_tree = RB_ROOT;
2217 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2218 fs_info->qgroup_seq = 1;
2219 fs_info->qgroup_ulist = NULL;
2220 fs_info->qgroup_rescan_running = false;
2221 mutex_init(&fs_info->qgroup_rescan_lock);
2224 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2225 struct btrfs_fs_devices *fs_devices)
2227 u32 max_active = fs_info->thread_pool_size;
2228 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2231 btrfs_alloc_workqueue(fs_info, "worker",
2232 flags | WQ_HIGHPRI, max_active, 16);
2234 fs_info->delalloc_workers =
2235 btrfs_alloc_workqueue(fs_info, "delalloc",
2236 flags, max_active, 2);
2238 fs_info->flush_workers =
2239 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2240 flags, max_active, 0);
2242 fs_info->caching_workers =
2243 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2246 * a higher idle thresh on the submit workers makes it much more
2247 * likely that bios will be send down in a sane order to the
2250 fs_info->submit_workers =
2251 btrfs_alloc_workqueue(fs_info, "submit", flags,
2252 min_t(u64, fs_devices->num_devices,
2255 fs_info->fixup_workers =
2256 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2259 * endios are largely parallel and should have a very
2262 fs_info->endio_workers =
2263 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2264 fs_info->endio_meta_workers =
2265 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2267 fs_info->endio_meta_write_workers =
2268 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2270 fs_info->endio_raid56_workers =
2271 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2273 fs_info->endio_repair_workers =
2274 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2275 fs_info->rmw_workers =
2276 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2277 fs_info->endio_write_workers =
2278 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2280 fs_info->endio_freespace_worker =
2281 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2283 fs_info->delayed_workers =
2284 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2286 fs_info->readahead_workers =
2287 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2289 fs_info->qgroup_rescan_workers =
2290 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2291 fs_info->extent_workers =
2292 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2293 min_t(u64, fs_devices->num_devices,
2296 if (!(fs_info->workers && fs_info->delalloc_workers &&
2297 fs_info->submit_workers && fs_info->flush_workers &&
2298 fs_info->endio_workers && fs_info->endio_meta_workers &&
2299 fs_info->endio_meta_write_workers &&
2300 fs_info->endio_repair_workers &&
2301 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2302 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2303 fs_info->caching_workers && fs_info->readahead_workers &&
2304 fs_info->fixup_workers && fs_info->delayed_workers &&
2305 fs_info->extent_workers &&
2306 fs_info->qgroup_rescan_workers)) {
2313 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2314 struct btrfs_fs_devices *fs_devices)
2317 struct btrfs_root *log_tree_root;
2318 struct btrfs_super_block *disk_super = fs_info->super_copy;
2319 u64 bytenr = btrfs_super_log_root(disk_super);
2320 int level = btrfs_super_log_root_level(disk_super);
2322 if (fs_devices->rw_devices == 0) {
2323 btrfs_warn(fs_info, "log replay required on RO media");
2327 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2331 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2333 log_tree_root->node = read_tree_block(fs_info, bytenr,
2334 fs_info->generation + 1,
2336 if (IS_ERR(log_tree_root->node)) {
2337 btrfs_warn(fs_info, "failed to read log tree");
2338 ret = PTR_ERR(log_tree_root->node);
2339 kfree(log_tree_root);
2341 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2342 btrfs_err(fs_info, "failed to read log tree");
2343 free_extent_buffer(log_tree_root->node);
2344 kfree(log_tree_root);
2347 /* returns with log_tree_root freed on success */
2348 ret = btrfs_recover_log_trees(log_tree_root);
2350 btrfs_handle_fs_error(fs_info, ret,
2351 "Failed to recover log tree");
2352 free_extent_buffer(log_tree_root->node);
2353 kfree(log_tree_root);
2357 if (sb_rdonly(fs_info->sb)) {
2358 ret = btrfs_commit_super(fs_info);
2366 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2368 struct btrfs_root *tree_root = fs_info->tree_root;
2369 struct btrfs_root *root;
2370 struct btrfs_key location;
2373 BUG_ON(!fs_info->tree_root);
2375 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2376 location.type = BTRFS_ROOT_ITEM_KEY;
2377 location.offset = 0;
2379 root = btrfs_read_tree_root(tree_root, &location);
2381 ret = PTR_ERR(root);
2384 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2385 fs_info->extent_root = root;
2387 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2388 root = btrfs_read_tree_root(tree_root, &location);
2390 ret = PTR_ERR(root);
2393 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2394 fs_info->dev_root = root;
2395 btrfs_init_devices_late(fs_info);
2397 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2398 root = btrfs_read_tree_root(tree_root, &location);
2400 ret = PTR_ERR(root);
2403 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2404 fs_info->csum_root = root;
2406 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2407 root = btrfs_read_tree_root(tree_root, &location);
2408 if (!IS_ERR(root)) {
2409 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2410 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2411 fs_info->quota_root = root;
2414 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2415 root = btrfs_read_tree_root(tree_root, &location);
2417 ret = PTR_ERR(root);
2421 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2422 fs_info->uuid_root = root;
2425 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2426 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2427 root = btrfs_read_tree_root(tree_root, &location);
2429 ret = PTR_ERR(root);
2432 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2433 fs_info->free_space_root = root;
2438 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2439 location.objectid, ret);
2443 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
2445 struct btrfs_super_block *sb = fs_info->super_copy;
2446 u64 nodesize = btrfs_super_nodesize(sb);
2447 u64 sectorsize = btrfs_super_sectorsize(sb);
2450 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2451 btrfs_err(fs_info, "no valid FS found");
2454 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2455 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2456 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2459 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2460 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2461 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2464 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2465 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2466 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2469 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2470 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2471 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2476 * Check sectorsize and nodesize first, other check will need it.
2477 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2479 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2480 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2481 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2484 /* Only PAGE SIZE is supported yet */
2485 if (sectorsize != PAGE_SIZE) {
2487 "sectorsize %llu not supported yet, only support %lu",
2488 sectorsize, PAGE_SIZE);
2491 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2492 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2493 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2496 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2497 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2498 le32_to_cpu(sb->__unused_leafsize), nodesize);
2502 /* Root alignment check */
2503 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2504 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2505 btrfs_super_root(sb));
2508 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2509 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2510 btrfs_super_chunk_root(sb));
2513 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2514 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2515 btrfs_super_log_root(sb));
2519 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
2521 "dev_item UUID does not match fsid: %pU != %pU",
2522 fs_info->fsid, sb->dev_item.fsid);
2527 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2530 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2531 btrfs_err(fs_info, "bytes_used is too small %llu",
2532 btrfs_super_bytes_used(sb));
2535 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2536 btrfs_err(fs_info, "invalid stripesize %u",
2537 btrfs_super_stripesize(sb));
2540 if (btrfs_super_num_devices(sb) > (1UL << 31))
2541 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2542 btrfs_super_num_devices(sb));
2543 if (btrfs_super_num_devices(sb) == 0) {
2544 btrfs_err(fs_info, "number of devices is 0");
2548 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
2549 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2550 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2555 * Obvious sys_chunk_array corruptions, it must hold at least one key
2558 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2559 btrfs_err(fs_info, "system chunk array too big %u > %u",
2560 btrfs_super_sys_array_size(sb),
2561 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2564 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2565 + sizeof(struct btrfs_chunk)) {
2566 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2567 btrfs_super_sys_array_size(sb),
2568 sizeof(struct btrfs_disk_key)
2569 + sizeof(struct btrfs_chunk));
2574 * The generation is a global counter, we'll trust it more than the others
2575 * but it's still possible that it's the one that's wrong.
2577 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2579 "suspicious: generation < chunk_root_generation: %llu < %llu",
2580 btrfs_super_generation(sb),
2581 btrfs_super_chunk_root_generation(sb));
2582 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2583 && btrfs_super_cache_generation(sb) != (u64)-1)
2585 "suspicious: generation < cache_generation: %llu < %llu",
2586 btrfs_super_generation(sb),
2587 btrfs_super_cache_generation(sb));
2592 int open_ctree(struct super_block *sb,
2593 struct btrfs_fs_devices *fs_devices,
2601 struct btrfs_key location;
2602 struct buffer_head *bh;
2603 struct btrfs_super_block *disk_super;
2604 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2605 struct btrfs_root *tree_root;
2606 struct btrfs_root *chunk_root;
2609 int num_backups_tried = 0;
2610 int backup_index = 0;
2611 int clear_free_space_tree = 0;
2614 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2615 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2616 if (!tree_root || !chunk_root) {
2621 ret = init_srcu_struct(&fs_info->subvol_srcu);
2627 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2632 fs_info->dirty_metadata_batch = PAGE_SIZE *
2633 (1 + ilog2(nr_cpu_ids));
2635 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2638 goto fail_dirty_metadata_bytes;
2641 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2644 goto fail_delalloc_bytes;
2647 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2648 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2649 INIT_LIST_HEAD(&fs_info->trans_list);
2650 INIT_LIST_HEAD(&fs_info->dead_roots);
2651 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2652 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2653 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2654 INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2655 spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2656 spin_lock_init(&fs_info->delalloc_root_lock);
2657 spin_lock_init(&fs_info->trans_lock);
2658 spin_lock_init(&fs_info->fs_roots_radix_lock);
2659 spin_lock_init(&fs_info->delayed_iput_lock);
2660 spin_lock_init(&fs_info->defrag_inodes_lock);
2661 spin_lock_init(&fs_info->tree_mod_seq_lock);
2662 spin_lock_init(&fs_info->super_lock);
2663 spin_lock_init(&fs_info->qgroup_op_lock);
2664 spin_lock_init(&fs_info->buffer_lock);
2665 spin_lock_init(&fs_info->unused_bgs_lock);
2666 rwlock_init(&fs_info->tree_mod_log_lock);
2667 mutex_init(&fs_info->unused_bg_unpin_mutex);
2668 mutex_init(&fs_info->delete_unused_bgs_mutex);
2669 mutex_init(&fs_info->reloc_mutex);
2670 mutex_init(&fs_info->delalloc_root_mutex);
2671 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2672 seqlock_init(&fs_info->profiles_lock);
2674 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2675 INIT_LIST_HEAD(&fs_info->space_info);
2676 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2677 INIT_LIST_HEAD(&fs_info->unused_bgs);
2678 btrfs_mapping_init(&fs_info->mapping_tree);
2679 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2680 BTRFS_BLOCK_RSV_GLOBAL);
2681 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2682 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2683 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2684 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2685 BTRFS_BLOCK_RSV_DELOPS);
2686 atomic_set(&fs_info->async_delalloc_pages, 0);
2687 atomic_set(&fs_info->defrag_running, 0);
2688 atomic_set(&fs_info->qgroup_op_seq, 0);
2689 atomic_set(&fs_info->reada_works_cnt, 0);
2690 atomic64_set(&fs_info->tree_mod_seq, 0);
2692 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2693 fs_info->metadata_ratio = 0;
2694 fs_info->defrag_inodes = RB_ROOT;
2695 atomic64_set(&fs_info->free_chunk_space, 0);
2696 fs_info->tree_mod_log = RB_ROOT;
2697 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2698 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2699 /* readahead state */
2700 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2701 spin_lock_init(&fs_info->reada_lock);
2702 btrfs_init_ref_verify(fs_info);
2704 fs_info->thread_pool_size = min_t(unsigned long,
2705 num_online_cpus() + 2, 8);
2707 INIT_LIST_HEAD(&fs_info->ordered_roots);
2708 spin_lock_init(&fs_info->ordered_root_lock);
2710 fs_info->btree_inode = new_inode(sb);
2711 if (!fs_info->btree_inode) {
2713 goto fail_bio_counter;
2715 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2717 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2719 if (!fs_info->delayed_root) {
2723 btrfs_init_delayed_root(fs_info->delayed_root);
2725 btrfs_init_scrub(fs_info);
2726 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2727 fs_info->check_integrity_print_mask = 0;
2729 btrfs_init_balance(fs_info);
2730 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2732 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2733 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2735 btrfs_init_btree_inode(fs_info);
2737 spin_lock_init(&fs_info->block_group_cache_lock);
2738 fs_info->block_group_cache_tree = RB_ROOT;
2739 fs_info->first_logical_byte = (u64)-1;
2741 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2742 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2743 fs_info->pinned_extents = &fs_info->freed_extents[0];
2744 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2746 mutex_init(&fs_info->ordered_operations_mutex);
2747 mutex_init(&fs_info->tree_log_mutex);
2748 mutex_init(&fs_info->chunk_mutex);
2749 mutex_init(&fs_info->transaction_kthread_mutex);
2750 mutex_init(&fs_info->cleaner_mutex);
2751 mutex_init(&fs_info->ro_block_group_mutex);
2752 init_rwsem(&fs_info->commit_root_sem);
2753 init_rwsem(&fs_info->cleanup_work_sem);
2754 init_rwsem(&fs_info->subvol_sem);
2755 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2757 btrfs_init_dev_replace_locks(fs_info);
2758 btrfs_init_qgroup(fs_info);
2760 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2761 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2763 init_waitqueue_head(&fs_info->transaction_throttle);
2764 init_waitqueue_head(&fs_info->transaction_wait);
2765 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2766 init_waitqueue_head(&fs_info->async_submit_wait);
2768 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2770 /* Usable values until the real ones are cached from the superblock */
2771 fs_info->nodesize = 4096;
2772 fs_info->sectorsize = 4096;
2773 fs_info->stripesize = 4096;
2775 ret = btrfs_alloc_stripe_hash_table(fs_info);
2781 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2783 invalidate_bdev(fs_devices->latest_bdev);
2786 * Read super block and check the signature bytes only
2788 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2795 * We want to check superblock checksum, the type is stored inside.
2796 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2798 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2799 btrfs_err(fs_info, "superblock checksum mismatch");
2806 * super_copy is zeroed at allocation time and we never touch the
2807 * following bytes up to INFO_SIZE, the checksum is calculated from
2808 * the whole block of INFO_SIZE
2810 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2811 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2812 sizeof(*fs_info->super_for_commit));
2815 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2817 ret = btrfs_check_super_valid(fs_info);
2819 btrfs_err(fs_info, "superblock contains fatal errors");
2824 disk_super = fs_info->super_copy;
2825 if (!btrfs_super_root(disk_super))
2828 /* check FS state, whether FS is broken. */
2829 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2830 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2833 * run through our array of backup supers and setup
2834 * our ring pointer to the oldest one
2836 generation = btrfs_super_generation(disk_super);
2837 find_oldest_super_backup(fs_info, generation);
2840 * In the long term, we'll store the compression type in the super
2841 * block, and it'll be used for per file compression control.
2843 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2845 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2851 features = btrfs_super_incompat_flags(disk_super) &
2852 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2855 "cannot mount because of unsupported optional features (%llx)",
2861 features = btrfs_super_incompat_flags(disk_super);
2862 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2863 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2864 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2865 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2866 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2868 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2869 btrfs_info(fs_info, "has skinny extents");
2872 * flag our filesystem as having big metadata blocks if
2873 * they are bigger than the page size
2875 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2876 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2878 "flagging fs with big metadata feature");
2879 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2882 nodesize = btrfs_super_nodesize(disk_super);
2883 sectorsize = btrfs_super_sectorsize(disk_super);
2884 stripesize = sectorsize;
2885 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2886 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2888 /* Cache block sizes */
2889 fs_info->nodesize = nodesize;
2890 fs_info->sectorsize = sectorsize;
2891 fs_info->stripesize = stripesize;
2894 * mixed block groups end up with duplicate but slightly offset
2895 * extent buffers for the same range. It leads to corruptions
2897 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2898 (sectorsize != nodesize)) {
2900 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2901 nodesize, sectorsize);
2906 * Needn't use the lock because there is no other task which will
2909 btrfs_set_super_incompat_flags(disk_super, features);
2911 features = btrfs_super_compat_ro_flags(disk_super) &
2912 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2913 if (!sb_rdonly(sb) && features) {
2915 "cannot mount read-write because of unsupported optional features (%llx)",
2921 ret = btrfs_init_workqueues(fs_info, fs_devices);
2924 goto fail_sb_buffer;
2927 sb->s_bdi->congested_fn = btrfs_congested_fn;
2928 sb->s_bdi->congested_data = fs_info;
2929 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2930 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2931 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2932 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2934 sb->s_blocksize = sectorsize;
2935 sb->s_blocksize_bits = blksize_bits(sectorsize);
2936 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2938 mutex_lock(&fs_info->chunk_mutex);
2939 ret = btrfs_read_sys_array(fs_info);
2940 mutex_unlock(&fs_info->chunk_mutex);
2942 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2943 goto fail_sb_buffer;
2946 generation = btrfs_super_chunk_root_generation(disk_super);
2947 level = btrfs_super_chunk_root_level(disk_super);
2949 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2951 chunk_root->node = read_tree_block(fs_info,
2952 btrfs_super_chunk_root(disk_super),
2953 generation, level, NULL);
2954 if (IS_ERR(chunk_root->node) ||
2955 !extent_buffer_uptodate(chunk_root->node)) {
2956 btrfs_err(fs_info, "failed to read chunk root");
2957 if (!IS_ERR(chunk_root->node))
2958 free_extent_buffer(chunk_root->node);
2959 chunk_root->node = NULL;
2960 goto fail_tree_roots;
2962 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2963 chunk_root->commit_root = btrfs_root_node(chunk_root);
2965 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2966 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2968 ret = btrfs_read_chunk_tree(fs_info);
2970 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2971 goto fail_tree_roots;
2975 * Keep the devid that is marked to be the target device for the
2976 * device replace procedure
2978 btrfs_free_extra_devids(fs_devices, 0);
2980 if (!fs_devices->latest_bdev) {
2981 btrfs_err(fs_info, "failed to read devices");
2982 goto fail_tree_roots;
2986 generation = btrfs_super_generation(disk_super);
2987 level = btrfs_super_root_level(disk_super);
2989 tree_root->node = read_tree_block(fs_info,
2990 btrfs_super_root(disk_super),
2991 generation, level, NULL);
2992 if (IS_ERR(tree_root->node) ||
2993 !extent_buffer_uptodate(tree_root->node)) {
2994 btrfs_warn(fs_info, "failed to read tree root");
2995 if (!IS_ERR(tree_root->node))
2996 free_extent_buffer(tree_root->node);
2997 tree_root->node = NULL;
2998 goto recovery_tree_root;
3001 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3002 tree_root->commit_root = btrfs_root_node(tree_root);
3003 btrfs_set_root_refs(&tree_root->root_item, 1);
3005 mutex_lock(&tree_root->objectid_mutex);
3006 ret = btrfs_find_highest_objectid(tree_root,
3007 &tree_root->highest_objectid);
3009 mutex_unlock(&tree_root->objectid_mutex);
3010 goto recovery_tree_root;
3013 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3015 mutex_unlock(&tree_root->objectid_mutex);
3017 ret = btrfs_read_roots(fs_info);
3019 goto recovery_tree_root;
3021 fs_info->generation = generation;
3022 fs_info->last_trans_committed = generation;
3024 ret = btrfs_recover_balance(fs_info);
3026 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3027 goto fail_block_groups;
3030 ret = btrfs_init_dev_stats(fs_info);
3032 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3033 goto fail_block_groups;
3036 ret = btrfs_init_dev_replace(fs_info);
3038 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3039 goto fail_block_groups;
3042 btrfs_free_extra_devids(fs_devices, 1);
3044 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3046 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3048 goto fail_block_groups;
3051 ret = btrfs_sysfs_add_device(fs_devices);
3053 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3055 goto fail_fsdev_sysfs;
3058 ret = btrfs_sysfs_add_mounted(fs_info);
3060 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3061 goto fail_fsdev_sysfs;
3064 ret = btrfs_init_space_info(fs_info);
3066 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3070 ret = btrfs_read_block_groups(fs_info);
3072 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3076 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3078 "writeable mount is not allowed due to too many missing devices");
3082 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3084 if (IS_ERR(fs_info->cleaner_kthread))
3087 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3089 "btrfs-transaction");
3090 if (IS_ERR(fs_info->transaction_kthread))
3093 if (!btrfs_test_opt(fs_info, NOSSD) &&
3094 !fs_info->fs_devices->rotating) {
3095 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3099 * Mount does not set all options immediately, we can do it now and do
3100 * not have to wait for transaction commit
3102 btrfs_apply_pending_changes(fs_info);
3104 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3105 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3106 ret = btrfsic_mount(fs_info, fs_devices,
3107 btrfs_test_opt(fs_info,
3108 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3110 fs_info->check_integrity_print_mask);
3113 "failed to initialize integrity check module: %d",
3117 ret = btrfs_read_qgroup_config(fs_info);
3119 goto fail_trans_kthread;
3121 if (btrfs_build_ref_tree(fs_info))
3122 btrfs_err(fs_info, "couldn't build ref tree");
3124 /* do not make disk changes in broken FS or nologreplay is given */
3125 if (btrfs_super_log_root(disk_super) != 0 &&
3126 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3127 ret = btrfs_replay_log(fs_info, fs_devices);
3134 ret = btrfs_find_orphan_roots(fs_info);
3138 if (!sb_rdonly(sb)) {
3139 ret = btrfs_cleanup_fs_roots(fs_info);
3143 mutex_lock(&fs_info->cleaner_mutex);
3144 ret = btrfs_recover_relocation(tree_root);
3145 mutex_unlock(&fs_info->cleaner_mutex);
3147 btrfs_warn(fs_info, "failed to recover relocation: %d",
3154 location.objectid = BTRFS_FS_TREE_OBJECTID;
3155 location.type = BTRFS_ROOT_ITEM_KEY;
3156 location.offset = 0;
3158 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3159 if (IS_ERR(fs_info->fs_root)) {
3160 err = PTR_ERR(fs_info->fs_root);
3161 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3168 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3169 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3170 clear_free_space_tree = 1;
3171 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3172 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3173 btrfs_warn(fs_info, "free space tree is invalid");
3174 clear_free_space_tree = 1;
3177 if (clear_free_space_tree) {
3178 btrfs_info(fs_info, "clearing free space tree");
3179 ret = btrfs_clear_free_space_tree(fs_info);
3182 "failed to clear free space tree: %d", ret);
3183 close_ctree(fs_info);
3188 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3189 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3190 btrfs_info(fs_info, "creating free space tree");
3191 ret = btrfs_create_free_space_tree(fs_info);
3194 "failed to create free space tree: %d", ret);
3195 close_ctree(fs_info);
3200 down_read(&fs_info->cleanup_work_sem);
3201 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3202 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3203 up_read(&fs_info->cleanup_work_sem);
3204 close_ctree(fs_info);
3207 up_read(&fs_info->cleanup_work_sem);
3209 ret = btrfs_resume_balance_async(fs_info);
3211 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3212 close_ctree(fs_info);
3216 ret = btrfs_resume_dev_replace_async(fs_info);
3218 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3219 close_ctree(fs_info);
3223 btrfs_qgroup_rescan_resume(fs_info);
3225 if (!fs_info->uuid_root) {
3226 btrfs_info(fs_info, "creating UUID tree");
3227 ret = btrfs_create_uuid_tree(fs_info);
3230 "failed to create the UUID tree: %d", ret);
3231 close_ctree(fs_info);
3234 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3235 fs_info->generation !=
3236 btrfs_super_uuid_tree_generation(disk_super)) {
3237 btrfs_info(fs_info, "checking UUID tree");
3238 ret = btrfs_check_uuid_tree(fs_info);
3241 "failed to check the UUID tree: %d", ret);
3242 close_ctree(fs_info);
3246 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3248 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3251 * backuproot only affect mount behavior, and if open_ctree succeeded,
3252 * no need to keep the flag
3254 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3259 btrfs_free_qgroup_config(fs_info);
3261 kthread_stop(fs_info->transaction_kthread);
3262 btrfs_cleanup_transaction(fs_info);
3263 btrfs_free_fs_roots(fs_info);
3265 kthread_stop(fs_info->cleaner_kthread);
3268 * make sure we're done with the btree inode before we stop our
3271 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3274 btrfs_sysfs_remove_mounted(fs_info);
3277 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3280 btrfs_put_block_group_cache(fs_info);
3283 free_root_pointers(fs_info, 1);
3284 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3287 btrfs_stop_all_workers(fs_info);
3288 btrfs_free_block_groups(fs_info);
3291 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3293 iput(fs_info->btree_inode);
3295 percpu_counter_destroy(&fs_info->bio_counter);
3296 fail_delalloc_bytes:
3297 percpu_counter_destroy(&fs_info->delalloc_bytes);
3298 fail_dirty_metadata_bytes:
3299 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3301 cleanup_srcu_struct(&fs_info->subvol_srcu);
3303 btrfs_free_stripe_hash_table(fs_info);
3304 btrfs_close_devices(fs_info->fs_devices);
3308 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3309 goto fail_tree_roots;
3311 free_root_pointers(fs_info, 0);
3313 /* don't use the log in recovery mode, it won't be valid */
3314 btrfs_set_super_log_root(disk_super, 0);
3316 /* we can't trust the free space cache either */
3317 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3319 ret = next_root_backup(fs_info, fs_info->super_copy,
3320 &num_backups_tried, &backup_index);
3322 goto fail_block_groups;
3323 goto retry_root_backup;
3325 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3327 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3330 set_buffer_uptodate(bh);
3332 struct btrfs_device *device = (struct btrfs_device *)
3335 btrfs_warn_rl_in_rcu(device->fs_info,
3336 "lost page write due to IO error on %s",
3337 rcu_str_deref(device->name));
3338 /* note, we don't set_buffer_write_io_error because we have
3339 * our own ways of dealing with the IO errors
3341 clear_buffer_uptodate(bh);
3342 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3348 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3349 struct buffer_head **bh_ret)
3351 struct buffer_head *bh;
3352 struct btrfs_super_block *super;
3355 bytenr = btrfs_sb_offset(copy_num);
3356 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3359 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3361 * If we fail to read from the underlying devices, as of now
3362 * the best option we have is to mark it EIO.
3367 super = (struct btrfs_super_block *)bh->b_data;
3368 if (btrfs_super_bytenr(super) != bytenr ||
3369 btrfs_super_magic(super) != BTRFS_MAGIC) {
3379 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3381 struct buffer_head *bh;
3382 struct buffer_head *latest = NULL;
3383 struct btrfs_super_block *super;
3388 /* we would like to check all the supers, but that would make
3389 * a btrfs mount succeed after a mkfs from a different FS.
3390 * So, we need to add a special mount option to scan for
3391 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3393 for (i = 0; i < 1; i++) {
3394 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3398 super = (struct btrfs_super_block *)bh->b_data;
3400 if (!latest || btrfs_super_generation(super) > transid) {
3403 transid = btrfs_super_generation(super);
3410 return ERR_PTR(ret);
3416 * Write superblock @sb to the @device. Do not wait for completion, all the
3417 * buffer heads we write are pinned.
3419 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3420 * the expected device size at commit time. Note that max_mirrors must be
3421 * same for write and wait phases.
3423 * Return number of errors when buffer head is not found or submission fails.
3425 static int write_dev_supers(struct btrfs_device *device,
3426 struct btrfs_super_block *sb, int max_mirrors)
3428 struct buffer_head *bh;
3436 if (max_mirrors == 0)
3437 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3439 for (i = 0; i < max_mirrors; i++) {
3440 bytenr = btrfs_sb_offset(i);
3441 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3442 device->commit_total_bytes)
3445 btrfs_set_super_bytenr(sb, bytenr);
3448 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3449 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3450 btrfs_csum_final(crc, sb->csum);
3452 /* One reference for us, and we leave it for the caller */
3453 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3454 BTRFS_SUPER_INFO_SIZE);
3456 btrfs_err(device->fs_info,
3457 "couldn't get super buffer head for bytenr %llu",
3463 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3465 /* one reference for submit_bh */
3468 set_buffer_uptodate(bh);
3470 bh->b_end_io = btrfs_end_buffer_write_sync;
3471 bh->b_private = device;
3474 * we fua the first super. The others we allow
3477 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3478 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3479 op_flags |= REQ_FUA;
3480 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3484 return errors < i ? 0 : -1;
3488 * Wait for write completion of superblocks done by write_dev_supers,
3489 * @max_mirrors same for write and wait phases.
3491 * Return number of errors when buffer head is not found or not marked up to
3494 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3496 struct buffer_head *bh;
3499 bool primary_failed = false;
3502 if (max_mirrors == 0)
3503 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3505 for (i = 0; i < max_mirrors; i++) {
3506 bytenr = btrfs_sb_offset(i);
3507 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3508 device->commit_total_bytes)
3511 bh = __find_get_block(device->bdev,
3512 bytenr / BTRFS_BDEV_BLOCKSIZE,
3513 BTRFS_SUPER_INFO_SIZE);
3517 primary_failed = true;
3521 if (!buffer_uptodate(bh)) {
3524 primary_failed = true;
3527 /* drop our reference */
3530 /* drop the reference from the writing run */
3534 /* log error, force error return */
3535 if (primary_failed) {
3536 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3541 return errors < i ? 0 : -1;
3545 * endio for the write_dev_flush, this will wake anyone waiting
3546 * for the barrier when it is done
3548 static void btrfs_end_empty_barrier(struct bio *bio)
3550 complete(bio->bi_private);
3554 * Submit a flush request to the device if it supports it. Error handling is
3555 * done in the waiting counterpart.
3557 static void write_dev_flush(struct btrfs_device *device)
3559 struct request_queue *q = bdev_get_queue(device->bdev);
3560 struct bio *bio = device->flush_bio;
3562 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3566 bio->bi_end_io = btrfs_end_empty_barrier;
3567 bio_set_dev(bio, device->bdev);
3568 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3569 init_completion(&device->flush_wait);
3570 bio->bi_private = &device->flush_wait;
3572 btrfsic_submit_bio(bio);
3573 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3577 * If the flush bio has been submitted by write_dev_flush, wait for it.
3579 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3581 struct bio *bio = device->flush_bio;
3583 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3586 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3587 wait_for_completion_io(&device->flush_wait);
3589 return bio->bi_status;
3592 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3594 if (!btrfs_check_rw_degradable(fs_info, NULL))
3600 * send an empty flush down to each device in parallel,
3601 * then wait for them
3603 static int barrier_all_devices(struct btrfs_fs_info *info)
3605 struct list_head *head;
3606 struct btrfs_device *dev;
3607 int errors_wait = 0;
3610 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3611 /* send down all the barriers */
3612 head = &info->fs_devices->devices;
3613 list_for_each_entry(dev, head, dev_list) {
3614 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3618 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3619 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3622 write_dev_flush(dev);
3623 dev->last_flush_error = BLK_STS_OK;
3626 /* wait for all the barriers */
3627 list_for_each_entry(dev, head, dev_list) {
3628 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3634 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3635 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3638 ret = wait_dev_flush(dev);
3640 dev->last_flush_error = ret;
3641 btrfs_dev_stat_inc_and_print(dev,
3642 BTRFS_DEV_STAT_FLUSH_ERRS);
3649 * At some point we need the status of all disks
3650 * to arrive at the volume status. So error checking
3651 * is being pushed to a separate loop.
3653 return check_barrier_error(info);
3658 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3661 int min_tolerated = INT_MAX;
3663 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3664 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3665 min_tolerated = min(min_tolerated,
3666 btrfs_raid_array[BTRFS_RAID_SINGLE].
3667 tolerated_failures);
3669 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3670 if (raid_type == BTRFS_RAID_SINGLE)
3672 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3674 min_tolerated = min(min_tolerated,
3675 btrfs_raid_array[raid_type].
3676 tolerated_failures);
3679 if (min_tolerated == INT_MAX) {
3680 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3684 return min_tolerated;
3687 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3689 struct list_head *head;
3690 struct btrfs_device *dev;
3691 struct btrfs_super_block *sb;
3692 struct btrfs_dev_item *dev_item;
3696 int total_errors = 0;
3699 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3702 * max_mirrors == 0 indicates we're from commit_transaction,
3703 * not from fsync where the tree roots in fs_info have not
3704 * been consistent on disk.
3706 if (max_mirrors == 0)
3707 backup_super_roots(fs_info);
3709 sb = fs_info->super_for_commit;
3710 dev_item = &sb->dev_item;
3712 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3713 head = &fs_info->fs_devices->devices;
3714 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3717 ret = barrier_all_devices(fs_info);
3720 &fs_info->fs_devices->device_list_mutex);
3721 btrfs_handle_fs_error(fs_info, ret,
3722 "errors while submitting device barriers.");
3727 list_for_each_entry(dev, head, dev_list) {
3732 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3733 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3736 btrfs_set_stack_device_generation(dev_item, 0);
3737 btrfs_set_stack_device_type(dev_item, dev->type);
3738 btrfs_set_stack_device_id(dev_item, dev->devid);
3739 btrfs_set_stack_device_total_bytes(dev_item,
3740 dev->commit_total_bytes);
3741 btrfs_set_stack_device_bytes_used(dev_item,
3742 dev->commit_bytes_used);
3743 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3744 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3745 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3746 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3747 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3749 flags = btrfs_super_flags(sb);
3750 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3752 ret = write_dev_supers(dev, sb, max_mirrors);
3756 if (total_errors > max_errors) {
3757 btrfs_err(fs_info, "%d errors while writing supers",
3759 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3761 /* FUA is masked off if unsupported and can't be the reason */
3762 btrfs_handle_fs_error(fs_info, -EIO,
3763 "%d errors while writing supers",
3769 list_for_each_entry(dev, head, dev_list) {
3772 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3773 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3776 ret = wait_dev_supers(dev, max_mirrors);
3780 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3781 if (total_errors > max_errors) {
3782 btrfs_handle_fs_error(fs_info, -EIO,
3783 "%d errors while writing supers",
3790 /* Drop a fs root from the radix tree and free it. */
3791 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3792 struct btrfs_root *root)
3794 spin_lock(&fs_info->fs_roots_radix_lock);
3795 radix_tree_delete(&fs_info->fs_roots_radix,
3796 (unsigned long)root->root_key.objectid);
3797 spin_unlock(&fs_info->fs_roots_radix_lock);
3799 if (btrfs_root_refs(&root->root_item) == 0)
3800 synchronize_srcu(&fs_info->subvol_srcu);
3802 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3803 btrfs_free_log(NULL, root);
3804 if (root->reloc_root) {
3805 free_extent_buffer(root->reloc_root->node);
3806 free_extent_buffer(root->reloc_root->commit_root);
3807 btrfs_put_fs_root(root->reloc_root);
3808 root->reloc_root = NULL;
3812 if (root->free_ino_pinned)
3813 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3814 if (root->free_ino_ctl)
3815 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3819 static void free_fs_root(struct btrfs_root *root)
3821 iput(root->ino_cache_inode);
3822 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3823 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3824 root->orphan_block_rsv = NULL;
3826 free_anon_bdev(root->anon_dev);
3827 if (root->subv_writers)
3828 btrfs_free_subvolume_writers(root->subv_writers);
3829 free_extent_buffer(root->node);
3830 free_extent_buffer(root->commit_root);
3831 kfree(root->free_ino_ctl);
3832 kfree(root->free_ino_pinned);
3834 btrfs_put_fs_root(root);
3837 void btrfs_free_fs_root(struct btrfs_root *root)
3842 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3844 u64 root_objectid = 0;
3845 struct btrfs_root *gang[8];
3848 unsigned int ret = 0;
3852 index = srcu_read_lock(&fs_info->subvol_srcu);
3853 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3854 (void **)gang, root_objectid,
3857 srcu_read_unlock(&fs_info->subvol_srcu, index);
3860 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3862 for (i = 0; i < ret; i++) {
3863 /* Avoid to grab roots in dead_roots */
3864 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3868 /* grab all the search result for later use */
3869 gang[i] = btrfs_grab_fs_root(gang[i]);
3871 srcu_read_unlock(&fs_info->subvol_srcu, index);
3873 for (i = 0; i < ret; i++) {
3876 root_objectid = gang[i]->root_key.objectid;
3877 err = btrfs_orphan_cleanup(gang[i]);
3880 btrfs_put_fs_root(gang[i]);
3885 /* release the uncleaned roots due to error */
3886 for (; i < ret; i++) {
3888 btrfs_put_fs_root(gang[i]);
3893 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3895 struct btrfs_root *root = fs_info->tree_root;
3896 struct btrfs_trans_handle *trans;
3898 mutex_lock(&fs_info->cleaner_mutex);
3899 btrfs_run_delayed_iputs(fs_info);
3900 mutex_unlock(&fs_info->cleaner_mutex);
3901 wake_up_process(fs_info->cleaner_kthread);
3903 /* wait until ongoing cleanup work done */
3904 down_write(&fs_info->cleanup_work_sem);
3905 up_write(&fs_info->cleanup_work_sem);
3907 trans = btrfs_join_transaction(root);
3909 return PTR_ERR(trans);
3910 return btrfs_commit_transaction(trans);
3913 void close_ctree(struct btrfs_fs_info *fs_info)
3915 struct btrfs_root *root = fs_info->tree_root;
3918 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3920 /* wait for the qgroup rescan worker to stop */
3921 btrfs_qgroup_wait_for_completion(fs_info, false);
3923 /* wait for the uuid_scan task to finish */
3924 down(&fs_info->uuid_tree_rescan_sem);
3925 /* avoid complains from lockdep et al., set sem back to initial state */
3926 up(&fs_info->uuid_tree_rescan_sem);
3928 /* pause restriper - we want to resume on mount */
3929 btrfs_pause_balance(fs_info);
3931 btrfs_dev_replace_suspend_for_unmount(fs_info);
3933 btrfs_scrub_cancel(fs_info);
3935 /* wait for any defraggers to finish */
3936 wait_event(fs_info->transaction_wait,
3937 (atomic_read(&fs_info->defrag_running) == 0));
3939 /* clear out the rbtree of defraggable inodes */
3940 btrfs_cleanup_defrag_inodes(fs_info);
3942 cancel_work_sync(&fs_info->async_reclaim_work);
3944 if (!sb_rdonly(fs_info->sb)) {
3946 * If the cleaner thread is stopped and there are
3947 * block groups queued for removal, the deletion will be
3948 * skipped when we quit the cleaner thread.
3950 btrfs_delete_unused_bgs(fs_info);
3952 ret = btrfs_commit_super(fs_info);
3954 btrfs_err(fs_info, "commit super ret %d", ret);
3957 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3958 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3959 btrfs_error_commit_super(fs_info);
3961 kthread_stop(fs_info->transaction_kthread);
3962 kthread_stop(fs_info->cleaner_kthread);
3964 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3966 btrfs_free_qgroup_config(fs_info);
3967 ASSERT(list_empty(&fs_info->delalloc_roots));
3969 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3970 btrfs_info(fs_info, "at unmount delalloc count %lld",
3971 percpu_counter_sum(&fs_info->delalloc_bytes));
3974 btrfs_sysfs_remove_mounted(fs_info);
3975 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3977 btrfs_free_fs_roots(fs_info);
3979 btrfs_put_block_group_cache(fs_info);
3982 * we must make sure there is not any read request to
3983 * submit after we stopping all workers.
3985 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3986 btrfs_stop_all_workers(fs_info);
3988 btrfs_free_block_groups(fs_info);
3990 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3991 free_root_pointers(fs_info, 1);
3993 iput(fs_info->btree_inode);
3995 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3996 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3997 btrfsic_unmount(fs_info->fs_devices);
4000 btrfs_close_devices(fs_info->fs_devices);
4001 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4003 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4004 percpu_counter_destroy(&fs_info->delalloc_bytes);
4005 percpu_counter_destroy(&fs_info->bio_counter);
4006 cleanup_srcu_struct(&fs_info->subvol_srcu);
4008 btrfs_free_stripe_hash_table(fs_info);
4009 btrfs_free_ref_cache(fs_info);
4011 __btrfs_free_block_rsv(root->orphan_block_rsv);
4012 root->orphan_block_rsv = NULL;
4014 while (!list_empty(&fs_info->pinned_chunks)) {
4015 struct extent_map *em;
4017 em = list_first_entry(&fs_info->pinned_chunks,
4018 struct extent_map, list);
4019 list_del_init(&em->list);
4020 free_extent_map(em);
4024 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4028 struct inode *btree_inode = buf->pages[0]->mapping->host;
4030 ret = extent_buffer_uptodate(buf);
4034 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4035 parent_transid, atomic);
4041 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4043 struct btrfs_fs_info *fs_info;
4044 struct btrfs_root *root;
4045 u64 transid = btrfs_header_generation(buf);
4048 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4050 * This is a fast path so only do this check if we have sanity tests
4051 * enabled. Normal people shouldn't be marking dummy buffers as dirty
4052 * outside of the sanity tests.
4054 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4057 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4058 fs_info = root->fs_info;
4059 btrfs_assert_tree_locked(buf);
4060 if (transid != fs_info->generation)
4061 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4062 buf->start, transid, fs_info->generation);
4063 was_dirty = set_extent_buffer_dirty(buf);
4065 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4067 fs_info->dirty_metadata_batch);
4068 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4070 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4071 * but item data not updated.
4072 * So here we should only check item pointers, not item data.
4074 if (btrfs_header_level(buf) == 0 &&
4075 btrfs_check_leaf_relaxed(fs_info, buf)) {
4076 btrfs_print_leaf(buf);
4082 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4086 * looks as though older kernels can get into trouble with
4087 * this code, they end up stuck in balance_dirty_pages forever
4091 if (current->flags & PF_MEMALLOC)
4095 btrfs_balance_delayed_items(fs_info);
4097 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4098 BTRFS_DIRTY_METADATA_THRESH);
4100 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4104 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4106 __btrfs_btree_balance_dirty(fs_info, 1);
4109 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4111 __btrfs_btree_balance_dirty(fs_info, 0);
4114 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4115 struct btrfs_key *first_key)
4117 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4118 struct btrfs_fs_info *fs_info = root->fs_info;
4120 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4124 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4126 /* cleanup FS via transaction */
4127 btrfs_cleanup_transaction(fs_info);
4129 mutex_lock(&fs_info->cleaner_mutex);
4130 btrfs_run_delayed_iputs(fs_info);
4131 mutex_unlock(&fs_info->cleaner_mutex);
4133 down_write(&fs_info->cleanup_work_sem);
4134 up_write(&fs_info->cleanup_work_sem);
4137 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4139 struct btrfs_ordered_extent *ordered;
4141 spin_lock(&root->ordered_extent_lock);
4143 * This will just short circuit the ordered completion stuff which will
4144 * make sure the ordered extent gets properly cleaned up.
4146 list_for_each_entry(ordered, &root->ordered_extents,
4148 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4149 spin_unlock(&root->ordered_extent_lock);
4152 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4154 struct btrfs_root *root;
4155 struct list_head splice;
4157 INIT_LIST_HEAD(&splice);
4159 spin_lock(&fs_info->ordered_root_lock);
4160 list_splice_init(&fs_info->ordered_roots, &splice);
4161 while (!list_empty(&splice)) {
4162 root = list_first_entry(&splice, struct btrfs_root,
4164 list_move_tail(&root->ordered_root,
4165 &fs_info->ordered_roots);
4167 spin_unlock(&fs_info->ordered_root_lock);
4168 btrfs_destroy_ordered_extents(root);
4171 spin_lock(&fs_info->ordered_root_lock);
4173 spin_unlock(&fs_info->ordered_root_lock);
4176 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4177 struct btrfs_fs_info *fs_info)
4179 struct rb_node *node;
4180 struct btrfs_delayed_ref_root *delayed_refs;
4181 struct btrfs_delayed_ref_node *ref;
4184 delayed_refs = &trans->delayed_refs;
4186 spin_lock(&delayed_refs->lock);
4187 if (atomic_read(&delayed_refs->num_entries) == 0) {
4188 spin_unlock(&delayed_refs->lock);
4189 btrfs_info(fs_info, "delayed_refs has NO entry");
4193 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4194 struct btrfs_delayed_ref_head *head;
4196 bool pin_bytes = false;
4198 head = rb_entry(node, struct btrfs_delayed_ref_head,
4200 if (!mutex_trylock(&head->mutex)) {
4201 refcount_inc(&head->refs);
4202 spin_unlock(&delayed_refs->lock);
4204 mutex_lock(&head->mutex);
4205 mutex_unlock(&head->mutex);
4206 btrfs_put_delayed_ref_head(head);
4207 spin_lock(&delayed_refs->lock);
4210 spin_lock(&head->lock);
4211 while ((n = rb_first(&head->ref_tree)) != NULL) {
4212 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4215 rb_erase(&ref->ref_node, &head->ref_tree);
4216 RB_CLEAR_NODE(&ref->ref_node);
4217 if (!list_empty(&ref->add_list))
4218 list_del(&ref->add_list);
4219 atomic_dec(&delayed_refs->num_entries);
4220 btrfs_put_delayed_ref(ref);
4222 if (head->must_insert_reserved)
4224 btrfs_free_delayed_extent_op(head->extent_op);
4225 delayed_refs->num_heads--;
4226 if (head->processing == 0)
4227 delayed_refs->num_heads_ready--;
4228 atomic_dec(&delayed_refs->num_entries);
4229 rb_erase(&head->href_node, &delayed_refs->href_root);
4230 RB_CLEAR_NODE(&head->href_node);
4231 spin_unlock(&head->lock);
4232 spin_unlock(&delayed_refs->lock);
4233 mutex_unlock(&head->mutex);
4236 btrfs_pin_extent(fs_info, head->bytenr,
4237 head->num_bytes, 1);
4238 btrfs_put_delayed_ref_head(head);
4240 spin_lock(&delayed_refs->lock);
4243 spin_unlock(&delayed_refs->lock);
4248 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4250 struct btrfs_inode *btrfs_inode;
4251 struct list_head splice;
4253 INIT_LIST_HEAD(&splice);
4255 spin_lock(&root->delalloc_lock);
4256 list_splice_init(&root->delalloc_inodes, &splice);
4258 while (!list_empty(&splice)) {
4259 struct inode *inode = NULL;
4260 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4262 __btrfs_del_delalloc_inode(root, btrfs_inode);
4263 spin_unlock(&root->delalloc_lock);
4266 * Make sure we get a live inode and that it'll not disappear
4269 inode = igrab(&btrfs_inode->vfs_inode);
4271 invalidate_inode_pages2(inode->i_mapping);
4274 spin_lock(&root->delalloc_lock);
4276 spin_unlock(&root->delalloc_lock);
4279 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4281 struct btrfs_root *root;
4282 struct list_head splice;
4284 INIT_LIST_HEAD(&splice);
4286 spin_lock(&fs_info->delalloc_root_lock);
4287 list_splice_init(&fs_info->delalloc_roots, &splice);
4288 while (!list_empty(&splice)) {
4289 root = list_first_entry(&splice, struct btrfs_root,
4291 root = btrfs_grab_fs_root(root);
4293 spin_unlock(&fs_info->delalloc_root_lock);
4295 btrfs_destroy_delalloc_inodes(root);
4296 btrfs_put_fs_root(root);
4298 spin_lock(&fs_info->delalloc_root_lock);
4300 spin_unlock(&fs_info->delalloc_root_lock);
4303 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4304 struct extent_io_tree *dirty_pages,
4308 struct extent_buffer *eb;
4313 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4318 clear_extent_bits(dirty_pages, start, end, mark);
4319 while (start <= end) {
4320 eb = find_extent_buffer(fs_info, start);
4321 start += fs_info->nodesize;
4324 wait_on_extent_buffer_writeback(eb);
4326 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4328 clear_extent_buffer_dirty(eb);
4329 free_extent_buffer_stale(eb);
4336 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4337 struct extent_io_tree *pinned_extents)
4339 struct extent_io_tree *unpin;
4345 unpin = pinned_extents;
4348 ret = find_first_extent_bit(unpin, 0, &start, &end,
4349 EXTENT_DIRTY, NULL);
4353 clear_extent_dirty(unpin, start, end);
4354 btrfs_error_unpin_extent_range(fs_info, start, end);
4359 if (unpin == &fs_info->freed_extents[0])
4360 unpin = &fs_info->freed_extents[1];
4362 unpin = &fs_info->freed_extents[0];
4370 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4372 struct inode *inode;
4374 inode = cache->io_ctl.inode;
4376 invalidate_inode_pages2(inode->i_mapping);
4377 BTRFS_I(inode)->generation = 0;
4378 cache->io_ctl.inode = NULL;
4381 btrfs_put_block_group(cache);
4384 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4385 struct btrfs_fs_info *fs_info)
4387 struct btrfs_block_group_cache *cache;
4389 spin_lock(&cur_trans->dirty_bgs_lock);
4390 while (!list_empty(&cur_trans->dirty_bgs)) {
4391 cache = list_first_entry(&cur_trans->dirty_bgs,
4392 struct btrfs_block_group_cache,
4395 if (!list_empty(&cache->io_list)) {
4396 spin_unlock(&cur_trans->dirty_bgs_lock);
4397 list_del_init(&cache->io_list);
4398 btrfs_cleanup_bg_io(cache);
4399 spin_lock(&cur_trans->dirty_bgs_lock);
4402 list_del_init(&cache->dirty_list);
4403 spin_lock(&cache->lock);
4404 cache->disk_cache_state = BTRFS_DC_ERROR;
4405 spin_unlock(&cache->lock);
4407 spin_unlock(&cur_trans->dirty_bgs_lock);
4408 btrfs_put_block_group(cache);
4409 spin_lock(&cur_trans->dirty_bgs_lock);
4411 spin_unlock(&cur_trans->dirty_bgs_lock);
4414 * Refer to the definition of io_bgs member for details why it's safe
4415 * to use it without any locking
4417 while (!list_empty(&cur_trans->io_bgs)) {
4418 cache = list_first_entry(&cur_trans->io_bgs,
4419 struct btrfs_block_group_cache,
4422 list_del_init(&cache->io_list);
4423 spin_lock(&cache->lock);
4424 cache->disk_cache_state = BTRFS_DC_ERROR;
4425 spin_unlock(&cache->lock);
4426 btrfs_cleanup_bg_io(cache);
4430 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4431 struct btrfs_fs_info *fs_info)
4433 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4434 ASSERT(list_empty(&cur_trans->dirty_bgs));
4435 ASSERT(list_empty(&cur_trans->io_bgs));
4437 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4439 cur_trans->state = TRANS_STATE_COMMIT_START;
4440 wake_up(&fs_info->transaction_blocked_wait);
4442 cur_trans->state = TRANS_STATE_UNBLOCKED;
4443 wake_up(&fs_info->transaction_wait);
4445 btrfs_destroy_delayed_inodes(fs_info);
4446 btrfs_assert_delayed_root_empty(fs_info);
4448 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4450 btrfs_destroy_pinned_extent(fs_info,
4451 fs_info->pinned_extents);
4453 cur_trans->state =TRANS_STATE_COMPLETED;
4454 wake_up(&cur_trans->commit_wait);
4457 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4459 struct btrfs_transaction *t;
4461 mutex_lock(&fs_info->transaction_kthread_mutex);
4463 spin_lock(&fs_info->trans_lock);
4464 while (!list_empty(&fs_info->trans_list)) {
4465 t = list_first_entry(&fs_info->trans_list,
4466 struct btrfs_transaction, list);
4467 if (t->state >= TRANS_STATE_COMMIT_START) {
4468 refcount_inc(&t->use_count);
4469 spin_unlock(&fs_info->trans_lock);
4470 btrfs_wait_for_commit(fs_info, t->transid);
4471 btrfs_put_transaction(t);
4472 spin_lock(&fs_info->trans_lock);
4475 if (t == fs_info->running_transaction) {
4476 t->state = TRANS_STATE_COMMIT_DOING;
4477 spin_unlock(&fs_info->trans_lock);
4479 * We wait for 0 num_writers since we don't hold a trans
4480 * handle open currently for this transaction.
4482 wait_event(t->writer_wait,
4483 atomic_read(&t->num_writers) == 0);
4485 spin_unlock(&fs_info->trans_lock);
4487 btrfs_cleanup_one_transaction(t, fs_info);
4489 spin_lock(&fs_info->trans_lock);
4490 if (t == fs_info->running_transaction)
4491 fs_info->running_transaction = NULL;
4492 list_del_init(&t->list);
4493 spin_unlock(&fs_info->trans_lock);
4495 btrfs_put_transaction(t);
4496 trace_btrfs_transaction_commit(fs_info->tree_root);
4497 spin_lock(&fs_info->trans_lock);
4499 spin_unlock(&fs_info->trans_lock);
4500 btrfs_destroy_all_ordered_extents(fs_info);
4501 btrfs_destroy_delayed_inodes(fs_info);
4502 btrfs_assert_delayed_root_empty(fs_info);
4503 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4504 btrfs_destroy_all_delalloc_inodes(fs_info);
4505 mutex_unlock(&fs_info->transaction_kthread_mutex);
4510 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4512 struct inode *inode = private_data;
4513 return btrfs_sb(inode->i_sb);
4516 static const struct extent_io_ops btree_extent_io_ops = {
4517 /* mandatory callbacks */
4518 .submit_bio_hook = btree_submit_bio_hook,
4519 .readpage_end_io_hook = btree_readpage_end_io_hook,
4520 /* note we're sharing with inode.c for the merge bio hook */
4521 .merge_bio_hook = btrfs_merge_bio_hook,
4522 .readpage_io_failed_hook = btree_io_failed_hook,
4523 .set_range_writeback = btrfs_set_range_writeback,
4524 .tree_fs_info = btree_fs_info,
4526 /* optional callbacks */