2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include "compression.h"
53 #include "tree-checker.h"
54 #include "ref-verify.h"
57 #include <asm/cpufeature.h>
60 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
61 BTRFS_HEADER_FLAG_RELOC |\
62 BTRFS_SUPER_FLAG_ERROR |\
63 BTRFS_SUPER_FLAG_SEEDING |\
64 BTRFS_SUPER_FLAG_METADUMP |\
65 BTRFS_SUPER_FLAG_METADUMP_V2)
67 static const struct extent_io_ops btree_extent_io_ops;
68 static void end_workqueue_fn(struct btrfs_work *work);
69 static void free_fs_root(struct btrfs_root *root);
70 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
71 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
72 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
73 struct btrfs_fs_info *fs_info);
74 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
75 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
76 struct extent_io_tree *dirty_pages,
78 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
79 struct extent_io_tree *pinned_extents);
80 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
81 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
84 * btrfs_end_io_wq structs are used to do processing in task context when an IO
85 * is complete. This is used during reads to verify checksums, and it is used
86 * by writes to insert metadata for new file extents after IO is complete.
88 struct btrfs_end_io_wq {
92 struct btrfs_fs_info *info;
94 enum btrfs_wq_endio_type metadata;
95 struct btrfs_work work;
98 static struct kmem_cache *btrfs_end_io_wq_cache;
100 int __init btrfs_end_io_wq_init(void)
102 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
103 sizeof(struct btrfs_end_io_wq),
107 if (!btrfs_end_io_wq_cache)
112 void btrfs_end_io_wq_exit(void)
114 kmem_cache_destroy(btrfs_end_io_wq_cache);
118 * async submit bios are used to offload expensive checksumming
119 * onto the worker threads. They checksum file and metadata bios
120 * just before they are sent down the IO stack.
122 struct async_submit_bio {
124 struct btrfs_fs_info *fs_info;
126 extent_submit_bio_hook_t *submit_bio_start;
127 extent_submit_bio_hook_t *submit_bio_done;
129 unsigned long bio_flags;
131 * bio_offset is optional, can be used if the pages in the bio
132 * can't tell us where in the file the bio should go
135 struct btrfs_work work;
140 * Lockdep class keys for extent_buffer->lock's in this root. For a given
141 * eb, the lockdep key is determined by the btrfs_root it belongs to and
142 * the level the eb occupies in the tree.
144 * Different roots are used for different purposes and may nest inside each
145 * other and they require separate keysets. As lockdep keys should be
146 * static, assign keysets according to the purpose of the root as indicated
147 * by btrfs_root->objectid. This ensures that all special purpose roots
148 * have separate keysets.
150 * Lock-nesting across peer nodes is always done with the immediate parent
151 * node locked thus preventing deadlock. As lockdep doesn't know this, use
152 * subclass to avoid triggering lockdep warning in such cases.
154 * The key is set by the readpage_end_io_hook after the buffer has passed
155 * csum validation but before the pages are unlocked. It is also set by
156 * btrfs_init_new_buffer on freshly allocated blocks.
158 * We also add a check to make sure the highest level of the tree is the
159 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
160 * needs update as well.
162 #ifdef CONFIG_DEBUG_LOCK_ALLOC
163 # if BTRFS_MAX_LEVEL != 8
167 static struct btrfs_lockdep_keyset {
168 u64 id; /* root objectid */
169 const char *name_stem; /* lock name stem */
170 char names[BTRFS_MAX_LEVEL + 1][20];
171 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
172 } btrfs_lockdep_keysets[] = {
173 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
174 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
175 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
176 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
177 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
178 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
179 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
180 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
181 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
182 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
183 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
184 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
185 { .id = 0, .name_stem = "tree" },
188 void __init btrfs_init_lockdep(void)
192 /* initialize lockdep class names */
193 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
194 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
196 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
197 snprintf(ks->names[j], sizeof(ks->names[j]),
198 "btrfs-%s-%02d", ks->name_stem, j);
202 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
205 struct btrfs_lockdep_keyset *ks;
207 BUG_ON(level >= ARRAY_SIZE(ks->keys));
209 /* find the matching keyset, id 0 is the default entry */
210 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
211 if (ks->id == objectid)
214 lockdep_set_class_and_name(&eb->lock,
215 &ks->keys[level], ks->names[level]);
221 * extents on the btree inode are pretty simple, there's one extent
222 * that covers the entire device
224 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
225 struct page *page, size_t pg_offset, u64 start, u64 len,
228 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
229 struct extent_map_tree *em_tree = &inode->extent_tree;
230 struct extent_map *em;
233 read_lock(&em_tree->lock);
234 em = lookup_extent_mapping(em_tree, start, len);
236 em->bdev = fs_info->fs_devices->latest_bdev;
237 read_unlock(&em_tree->lock);
240 read_unlock(&em_tree->lock);
242 em = alloc_extent_map();
244 em = ERR_PTR(-ENOMEM);
249 em->block_len = (u64)-1;
251 em->bdev = fs_info->fs_devices->latest_bdev;
253 write_lock(&em_tree->lock);
254 ret = add_extent_mapping(em_tree, em, 0);
255 if (ret == -EEXIST) {
257 em = lookup_extent_mapping(em_tree, start, len);
264 write_unlock(&em_tree->lock);
270 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
272 return btrfs_crc32c(seed, data, len);
275 void btrfs_csum_final(u32 crc, u8 *result)
277 put_unaligned_le32(~crc, result);
281 * compute the csum for a btree block, and either verify it or write it
282 * into the csum field of the block.
284 static int csum_tree_block(struct btrfs_fs_info *fs_info,
285 struct extent_buffer *buf,
288 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
289 char result[BTRFS_CSUM_SIZE];
291 unsigned long cur_len;
292 unsigned long offset = BTRFS_CSUM_SIZE;
294 unsigned long map_start;
295 unsigned long map_len;
299 len = buf->len - offset;
301 err = map_private_extent_buffer(buf, offset, 32,
302 &kaddr, &map_start, &map_len);
305 cur_len = min(len, map_len - (offset - map_start));
306 crc = btrfs_csum_data(kaddr + offset - map_start,
311 memset(result, 0, BTRFS_CSUM_SIZE);
313 btrfs_csum_final(crc, result);
316 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
319 memcpy(&found, result, csum_size);
321 read_extent_buffer(buf, &val, 0, csum_size);
322 btrfs_warn_rl(fs_info,
323 "%s checksum verify failed on %llu wanted %X found %X level %d",
324 fs_info->sb->s_id, buf->start,
325 val, found, btrfs_header_level(buf));
329 write_extent_buffer(buf, result, 0, csum_size);
336 * we can't consider a given block up to date unless the transid of the
337 * block matches the transid in the parent node's pointer. This is how we
338 * detect blocks that either didn't get written at all or got written
339 * in the wrong place.
341 static int verify_parent_transid(struct extent_io_tree *io_tree,
342 struct extent_buffer *eb, u64 parent_transid,
345 struct extent_state *cached_state = NULL;
347 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
349 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
356 btrfs_tree_read_lock(eb);
357 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
360 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
362 if (extent_buffer_uptodate(eb) &&
363 btrfs_header_generation(eb) == parent_transid) {
367 btrfs_err_rl(eb->fs_info,
368 "parent transid verify failed on %llu wanted %llu found %llu",
370 parent_transid, btrfs_header_generation(eb));
374 * Things reading via commit roots that don't have normal protection,
375 * like send, can have a really old block in cache that may point at a
376 * block that has been freed and re-allocated. So don't clear uptodate
377 * if we find an eb that is under IO (dirty/writeback) because we could
378 * end up reading in the stale data and then writing it back out and
379 * making everybody very sad.
381 if (!extent_buffer_under_io(eb))
382 clear_extent_buffer_uptodate(eb);
384 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
387 btrfs_tree_read_unlock_blocking(eb);
392 * Return 0 if the superblock checksum type matches the checksum value of that
393 * algorithm. Pass the raw disk superblock data.
395 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
398 struct btrfs_super_block *disk_sb =
399 (struct btrfs_super_block *)raw_disk_sb;
400 u16 csum_type = btrfs_super_csum_type(disk_sb);
403 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
405 const int csum_size = sizeof(crc);
406 char result[csum_size];
409 * The super_block structure does not span the whole
410 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
411 * is filled with zeros and is included in the checksum.
413 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
414 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
415 btrfs_csum_final(crc, result);
417 if (memcmp(raw_disk_sb, result, csum_size))
421 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
422 btrfs_err(fs_info, "unsupported checksum algorithm %u",
431 * helper to read a given tree block, doing retries as required when
432 * the checksums don't match and we have alternate mirrors to try.
434 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
435 struct extent_buffer *eb,
438 struct extent_io_tree *io_tree;
443 int failed_mirror = 0;
445 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
446 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
448 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
451 if (!verify_parent_transid(io_tree, eb,
459 * This buffer's crc is fine, but its contents are corrupted, so
460 * there is no reason to read the other copies, they won't be
463 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
466 num_copies = btrfs_num_copies(fs_info,
471 if (!failed_mirror) {
473 failed_mirror = eb->read_mirror;
477 if (mirror_num == failed_mirror)
480 if (mirror_num > num_copies)
484 if (failed && !ret && failed_mirror)
485 repair_eb_io_failure(fs_info, eb, failed_mirror);
491 * checksum a dirty tree block before IO. This has extra checks to make sure
492 * we only fill in the checksum field in the first page of a multi-page block
495 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
497 u64 start = page_offset(page);
499 struct extent_buffer *eb;
501 eb = (struct extent_buffer *)page->private;
502 if (page != eb->pages[0])
505 found_start = btrfs_header_bytenr(eb);
507 * Please do not consolidate these warnings into a single if.
508 * It is useful to know what went wrong.
510 if (WARN_ON(found_start != start))
512 if (WARN_ON(!PageUptodate(page)))
515 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
516 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
518 return csum_tree_block(fs_info, eb, 0);
521 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
522 struct extent_buffer *eb)
524 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
525 u8 fsid[BTRFS_FSID_SIZE];
528 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
530 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
534 fs_devices = fs_devices->seed;
539 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
540 u64 phy_offset, struct page *page,
541 u64 start, u64 end, int mirror)
545 struct extent_buffer *eb;
546 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
547 struct btrfs_fs_info *fs_info = root->fs_info;
554 eb = (struct extent_buffer *)page->private;
556 /* the pending IO might have been the only thing that kept this buffer
557 * in memory. Make sure we have a ref for all this other checks
559 extent_buffer_get(eb);
561 reads_done = atomic_dec_and_test(&eb->io_pages);
565 eb->read_mirror = mirror;
566 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
571 found_start = btrfs_header_bytenr(eb);
572 if (found_start != eb->start) {
573 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
574 found_start, eb->start);
578 if (check_tree_block_fsid(fs_info, eb)) {
579 btrfs_err_rl(fs_info, "bad fsid on block %llu",
584 found_level = btrfs_header_level(eb);
585 if (found_level >= BTRFS_MAX_LEVEL) {
586 btrfs_err(fs_info, "bad tree block level %d",
587 (int)btrfs_header_level(eb));
592 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
595 ret = csum_tree_block(fs_info, eb, 1);
600 * If this is a leaf block and it is corrupt, set the corrupt bit so
601 * that we don't try and read the other copies of this block, just
604 if (found_level == 0 && btrfs_check_leaf_full(root, eb)) {
605 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
609 if (found_level > 0 && btrfs_check_node(root, eb))
613 set_extent_buffer_uptodate(eb);
616 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
617 btree_readahead_hook(eb, ret);
621 * our io error hook is going to dec the io pages
622 * again, we have to make sure it has something
625 atomic_inc(&eb->io_pages);
626 clear_extent_buffer_uptodate(eb);
628 free_extent_buffer(eb);
633 static int btree_io_failed_hook(struct page *page, int failed_mirror)
635 struct extent_buffer *eb;
637 eb = (struct extent_buffer *)page->private;
638 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
639 eb->read_mirror = failed_mirror;
640 atomic_dec(&eb->io_pages);
641 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
642 btree_readahead_hook(eb, -EIO);
643 return -EIO; /* we fixed nothing */
646 static void end_workqueue_bio(struct bio *bio)
648 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
649 struct btrfs_fs_info *fs_info;
650 struct btrfs_workqueue *wq;
651 btrfs_work_func_t func;
653 fs_info = end_io_wq->info;
654 end_io_wq->status = bio->bi_status;
656 if (bio_op(bio) == REQ_OP_WRITE) {
657 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
658 wq = fs_info->endio_meta_write_workers;
659 func = btrfs_endio_meta_write_helper;
660 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
661 wq = fs_info->endio_freespace_worker;
662 func = btrfs_freespace_write_helper;
663 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
664 wq = fs_info->endio_raid56_workers;
665 func = btrfs_endio_raid56_helper;
667 wq = fs_info->endio_write_workers;
668 func = btrfs_endio_write_helper;
671 if (unlikely(end_io_wq->metadata ==
672 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
673 wq = fs_info->endio_repair_workers;
674 func = btrfs_endio_repair_helper;
675 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
676 wq = fs_info->endio_raid56_workers;
677 func = btrfs_endio_raid56_helper;
678 } else if (end_io_wq->metadata) {
679 wq = fs_info->endio_meta_workers;
680 func = btrfs_endio_meta_helper;
682 wq = fs_info->endio_workers;
683 func = btrfs_endio_helper;
687 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
688 btrfs_queue_work(wq, &end_io_wq->work);
691 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
692 enum btrfs_wq_endio_type metadata)
694 struct btrfs_end_io_wq *end_io_wq;
696 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
698 return BLK_STS_RESOURCE;
700 end_io_wq->private = bio->bi_private;
701 end_io_wq->end_io = bio->bi_end_io;
702 end_io_wq->info = info;
703 end_io_wq->status = 0;
704 end_io_wq->bio = bio;
705 end_io_wq->metadata = metadata;
707 bio->bi_private = end_io_wq;
708 bio->bi_end_io = end_workqueue_bio;
712 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
714 unsigned long limit = min_t(unsigned long,
715 info->thread_pool_size,
716 info->fs_devices->open_devices);
720 static void run_one_async_start(struct btrfs_work *work)
722 struct async_submit_bio *async;
725 async = container_of(work, struct async_submit_bio, work);
726 ret = async->submit_bio_start(async->private_data, async->bio,
727 async->mirror_num, async->bio_flags,
733 static void run_one_async_done(struct btrfs_work *work)
735 struct async_submit_bio *async;
737 async = container_of(work, struct async_submit_bio, work);
739 /* If an error occurred we just want to clean up the bio and move on */
741 async->bio->bi_status = async->status;
742 bio_endio(async->bio);
746 async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
747 async->bio_flags, async->bio_offset);
750 static void run_one_async_free(struct btrfs_work *work)
752 struct async_submit_bio *async;
754 async = container_of(work, struct async_submit_bio, work);
758 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
759 int mirror_num, unsigned long bio_flags,
760 u64 bio_offset, void *private_data,
761 extent_submit_bio_hook_t *submit_bio_start,
762 extent_submit_bio_hook_t *submit_bio_done)
764 struct async_submit_bio *async;
766 async = kmalloc(sizeof(*async), GFP_NOFS);
768 return BLK_STS_RESOURCE;
770 async->private_data = private_data;
771 async->fs_info = fs_info;
773 async->mirror_num = mirror_num;
774 async->submit_bio_start = submit_bio_start;
775 async->submit_bio_done = submit_bio_done;
777 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
778 run_one_async_done, run_one_async_free);
780 async->bio_flags = bio_flags;
781 async->bio_offset = bio_offset;
785 if (op_is_sync(bio->bi_opf))
786 btrfs_set_work_high_priority(&async->work);
788 btrfs_queue_work(fs_info->workers, &async->work);
792 static blk_status_t btree_csum_one_bio(struct bio *bio)
794 struct bio_vec *bvec;
795 struct btrfs_root *root;
798 ASSERT(!bio_flagged(bio, BIO_CLONED));
799 bio_for_each_segment_all(bvec, bio, i) {
800 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
801 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
806 return errno_to_blk_status(ret);
809 static blk_status_t __btree_submit_bio_start(void *private_data, struct bio *bio,
810 int mirror_num, unsigned long bio_flags,
814 * when we're called for a write, we're already in the async
815 * submission context. Just jump into btrfs_map_bio
817 return btree_csum_one_bio(bio);
820 static blk_status_t __btree_submit_bio_done(void *private_data, struct bio *bio,
821 int mirror_num, unsigned long bio_flags,
824 struct inode *inode = private_data;
828 * when we're called for a write, we're already in the async
829 * submission context. Just jump into btrfs_map_bio
831 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
833 bio->bi_status = ret;
839 static int check_async_write(struct btrfs_inode *bi)
841 if (atomic_read(&bi->sync_writers))
844 if (static_cpu_has(X86_FEATURE_XMM4_2))
850 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
851 int mirror_num, unsigned long bio_flags,
854 struct inode *inode = private_data;
855 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
856 int async = check_async_write(BTRFS_I(inode));
859 if (bio_op(bio) != REQ_OP_WRITE) {
861 * called for a read, do the setup so that checksum validation
862 * can happen in the async kernel threads
864 ret = btrfs_bio_wq_end_io(fs_info, bio,
865 BTRFS_WQ_ENDIO_METADATA);
868 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
870 ret = btree_csum_one_bio(bio);
873 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
876 * kthread helpers are used to submit writes so that
877 * checksumming can happen in parallel across all CPUs
879 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
880 bio_offset, private_data,
881 __btree_submit_bio_start,
882 __btree_submit_bio_done);
890 bio->bi_status = ret;
895 #ifdef CONFIG_MIGRATION
896 static int btree_migratepage(struct address_space *mapping,
897 struct page *newpage, struct page *page,
898 enum migrate_mode mode)
901 * we can't safely write a btree page from here,
902 * we haven't done the locking hook
907 * Buffers may be managed in a filesystem specific way.
908 * We must have no buffers or drop them.
910 if (page_has_private(page) &&
911 !try_to_release_page(page, GFP_KERNEL))
913 return migrate_page(mapping, newpage, page, mode);
918 static int btree_writepages(struct address_space *mapping,
919 struct writeback_control *wbc)
921 struct btrfs_fs_info *fs_info;
924 if (wbc->sync_mode == WB_SYNC_NONE) {
926 if (wbc->for_kupdate)
929 fs_info = BTRFS_I(mapping->host)->root->fs_info;
930 /* this is a bit racy, but that's ok */
931 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
932 BTRFS_DIRTY_METADATA_THRESH);
936 return btree_write_cache_pages(mapping, wbc);
939 static int btree_readpage(struct file *file, struct page *page)
941 struct extent_io_tree *tree;
942 tree = &BTRFS_I(page->mapping->host)->io_tree;
943 return extent_read_full_page(tree, page, btree_get_extent, 0);
946 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
948 if (PageWriteback(page) || PageDirty(page))
951 return try_release_extent_buffer(page);
954 static void btree_invalidatepage(struct page *page, unsigned int offset,
957 struct extent_io_tree *tree;
958 tree = &BTRFS_I(page->mapping->host)->io_tree;
959 extent_invalidatepage(tree, page, offset);
960 btree_releasepage(page, GFP_NOFS);
961 if (PagePrivate(page)) {
962 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
963 "page private not zero on page %llu",
964 (unsigned long long)page_offset(page));
965 ClearPagePrivate(page);
966 set_page_private(page, 0);
971 static int btree_set_page_dirty(struct page *page)
974 struct extent_buffer *eb;
976 BUG_ON(!PagePrivate(page));
977 eb = (struct extent_buffer *)page->private;
979 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
980 BUG_ON(!atomic_read(&eb->refs));
981 btrfs_assert_tree_locked(eb);
983 return __set_page_dirty_nobuffers(page);
986 static const struct address_space_operations btree_aops = {
987 .readpage = btree_readpage,
988 .writepages = btree_writepages,
989 .releasepage = btree_releasepage,
990 .invalidatepage = btree_invalidatepage,
991 #ifdef CONFIG_MIGRATION
992 .migratepage = btree_migratepage,
994 .set_page_dirty = btree_set_page_dirty,
997 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
999 struct extent_buffer *buf = NULL;
1000 struct inode *btree_inode = fs_info->btree_inode;
1002 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1005 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1007 free_extent_buffer(buf);
1010 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1011 int mirror_num, struct extent_buffer **eb)
1013 struct extent_buffer *buf = NULL;
1014 struct inode *btree_inode = fs_info->btree_inode;
1015 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1018 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1022 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1024 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1027 free_extent_buffer(buf);
1031 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1032 free_extent_buffer(buf);
1034 } else if (extent_buffer_uptodate(buf)) {
1037 free_extent_buffer(buf);
1042 struct extent_buffer *btrfs_find_create_tree_block(
1043 struct btrfs_fs_info *fs_info,
1046 if (btrfs_is_testing(fs_info))
1047 return alloc_test_extent_buffer(fs_info, bytenr);
1048 return alloc_extent_buffer(fs_info, bytenr);
1052 int btrfs_write_tree_block(struct extent_buffer *buf)
1054 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1055 buf->start + buf->len - 1);
1058 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1060 filemap_fdatawait_range(buf->pages[0]->mapping,
1061 buf->start, buf->start + buf->len - 1);
1064 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1067 struct extent_buffer *buf = NULL;
1070 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1074 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1076 free_extent_buffer(buf);
1077 return ERR_PTR(ret);
1083 void clean_tree_block(struct btrfs_fs_info *fs_info,
1084 struct extent_buffer *buf)
1086 if (btrfs_header_generation(buf) ==
1087 fs_info->running_transaction->transid) {
1088 btrfs_assert_tree_locked(buf);
1090 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1091 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1093 fs_info->dirty_metadata_batch);
1094 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1095 btrfs_set_lock_blocking(buf);
1096 clear_extent_buffer_dirty(buf);
1101 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1103 struct btrfs_subvolume_writers *writers;
1106 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1108 return ERR_PTR(-ENOMEM);
1110 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1113 return ERR_PTR(ret);
1116 init_waitqueue_head(&writers->wait);
1121 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1123 percpu_counter_destroy(&writers->counter);
1127 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1130 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1132 root->commit_root = NULL;
1134 root->orphan_cleanup_state = 0;
1136 root->objectid = objectid;
1137 root->last_trans = 0;
1138 root->highest_objectid = 0;
1139 root->nr_delalloc_inodes = 0;
1140 root->nr_ordered_extents = 0;
1142 root->inode_tree = RB_ROOT;
1143 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1144 root->block_rsv = NULL;
1145 root->orphan_block_rsv = NULL;
1147 INIT_LIST_HEAD(&root->dirty_list);
1148 INIT_LIST_HEAD(&root->root_list);
1149 INIT_LIST_HEAD(&root->delalloc_inodes);
1150 INIT_LIST_HEAD(&root->delalloc_root);
1151 INIT_LIST_HEAD(&root->ordered_extents);
1152 INIT_LIST_HEAD(&root->ordered_root);
1153 INIT_LIST_HEAD(&root->logged_list[0]);
1154 INIT_LIST_HEAD(&root->logged_list[1]);
1155 spin_lock_init(&root->orphan_lock);
1156 spin_lock_init(&root->inode_lock);
1157 spin_lock_init(&root->delalloc_lock);
1158 spin_lock_init(&root->ordered_extent_lock);
1159 spin_lock_init(&root->accounting_lock);
1160 spin_lock_init(&root->log_extents_lock[0]);
1161 spin_lock_init(&root->log_extents_lock[1]);
1162 mutex_init(&root->objectid_mutex);
1163 mutex_init(&root->log_mutex);
1164 mutex_init(&root->ordered_extent_mutex);
1165 mutex_init(&root->delalloc_mutex);
1166 init_waitqueue_head(&root->log_writer_wait);
1167 init_waitqueue_head(&root->log_commit_wait[0]);
1168 init_waitqueue_head(&root->log_commit_wait[1]);
1169 INIT_LIST_HEAD(&root->log_ctxs[0]);
1170 INIT_LIST_HEAD(&root->log_ctxs[1]);
1171 atomic_set(&root->log_commit[0], 0);
1172 atomic_set(&root->log_commit[1], 0);
1173 atomic_set(&root->log_writers, 0);
1174 atomic_set(&root->log_batch, 0);
1175 atomic_set(&root->orphan_inodes, 0);
1176 refcount_set(&root->refs, 1);
1177 atomic_set(&root->will_be_snapshotted, 0);
1178 atomic64_set(&root->qgroup_meta_rsv, 0);
1179 root->log_transid = 0;
1180 root->log_transid_committed = -1;
1181 root->last_log_commit = 0;
1183 extent_io_tree_init(&root->dirty_log_pages, NULL);
1185 memset(&root->root_key, 0, sizeof(root->root_key));
1186 memset(&root->root_item, 0, sizeof(root->root_item));
1187 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1189 root->defrag_trans_start = fs_info->generation;
1191 root->defrag_trans_start = 0;
1192 root->root_key.objectid = objectid;
1195 spin_lock_init(&root->root_item_lock);
1198 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1201 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1203 root->fs_info = fs_info;
1207 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1208 /* Should only be used by the testing infrastructure */
1209 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1211 struct btrfs_root *root;
1214 return ERR_PTR(-EINVAL);
1216 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1218 return ERR_PTR(-ENOMEM);
1220 /* We don't use the stripesize in selftest, set it as sectorsize */
1221 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1222 root->alloc_bytenr = 0;
1228 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1229 struct btrfs_fs_info *fs_info,
1232 struct extent_buffer *leaf;
1233 struct btrfs_root *tree_root = fs_info->tree_root;
1234 struct btrfs_root *root;
1235 struct btrfs_key key;
1237 uuid_le uuid = NULL_UUID_LE;
1239 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1241 return ERR_PTR(-ENOMEM);
1243 __setup_root(root, fs_info, objectid);
1244 root->root_key.objectid = objectid;
1245 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1246 root->root_key.offset = 0;
1248 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1250 ret = PTR_ERR(leaf);
1255 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1256 btrfs_set_header_bytenr(leaf, leaf->start);
1257 btrfs_set_header_generation(leaf, trans->transid);
1258 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1259 btrfs_set_header_owner(leaf, objectid);
1262 write_extent_buffer_fsid(leaf, fs_info->fsid);
1263 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1264 btrfs_mark_buffer_dirty(leaf);
1266 root->commit_root = btrfs_root_node(root);
1267 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1269 root->root_item.flags = 0;
1270 root->root_item.byte_limit = 0;
1271 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1272 btrfs_set_root_generation(&root->root_item, trans->transid);
1273 btrfs_set_root_level(&root->root_item, 0);
1274 btrfs_set_root_refs(&root->root_item, 1);
1275 btrfs_set_root_used(&root->root_item, leaf->len);
1276 btrfs_set_root_last_snapshot(&root->root_item, 0);
1277 btrfs_set_root_dirid(&root->root_item, 0);
1278 if (is_fstree(objectid))
1280 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1281 root->root_item.drop_level = 0;
1283 key.objectid = objectid;
1284 key.type = BTRFS_ROOT_ITEM_KEY;
1286 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1290 btrfs_tree_unlock(leaf);
1296 btrfs_tree_unlock(leaf);
1297 free_extent_buffer(root->commit_root);
1298 free_extent_buffer(leaf);
1302 return ERR_PTR(ret);
1305 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1306 struct btrfs_fs_info *fs_info)
1308 struct btrfs_root *root;
1309 struct extent_buffer *leaf;
1311 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1313 return ERR_PTR(-ENOMEM);
1315 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1317 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1318 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1319 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1322 * DON'T set REF_COWS for log trees
1324 * log trees do not get reference counted because they go away
1325 * before a real commit is actually done. They do store pointers
1326 * to file data extents, and those reference counts still get
1327 * updated (along with back refs to the log tree).
1330 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1334 return ERR_CAST(leaf);
1337 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1338 btrfs_set_header_bytenr(leaf, leaf->start);
1339 btrfs_set_header_generation(leaf, trans->transid);
1340 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1341 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1344 write_extent_buffer_fsid(root->node, fs_info->fsid);
1345 btrfs_mark_buffer_dirty(root->node);
1346 btrfs_tree_unlock(root->node);
1350 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1351 struct btrfs_fs_info *fs_info)
1353 struct btrfs_root *log_root;
1355 log_root = alloc_log_tree(trans, fs_info);
1356 if (IS_ERR(log_root))
1357 return PTR_ERR(log_root);
1358 WARN_ON(fs_info->log_root_tree);
1359 fs_info->log_root_tree = log_root;
1363 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1364 struct btrfs_root *root)
1366 struct btrfs_fs_info *fs_info = root->fs_info;
1367 struct btrfs_root *log_root;
1368 struct btrfs_inode_item *inode_item;
1370 log_root = alloc_log_tree(trans, fs_info);
1371 if (IS_ERR(log_root))
1372 return PTR_ERR(log_root);
1374 log_root->last_trans = trans->transid;
1375 log_root->root_key.offset = root->root_key.objectid;
1377 inode_item = &log_root->root_item.inode;
1378 btrfs_set_stack_inode_generation(inode_item, 1);
1379 btrfs_set_stack_inode_size(inode_item, 3);
1380 btrfs_set_stack_inode_nlink(inode_item, 1);
1381 btrfs_set_stack_inode_nbytes(inode_item,
1383 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1385 btrfs_set_root_node(&log_root->root_item, log_root->node);
1387 WARN_ON(root->log_root);
1388 root->log_root = log_root;
1389 root->log_transid = 0;
1390 root->log_transid_committed = -1;
1391 root->last_log_commit = 0;
1395 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1396 struct btrfs_key *key)
1398 struct btrfs_root *root;
1399 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1400 struct btrfs_path *path;
1404 path = btrfs_alloc_path();
1406 return ERR_PTR(-ENOMEM);
1408 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1414 __setup_root(root, fs_info, key->objectid);
1416 ret = btrfs_find_root(tree_root, key, path,
1417 &root->root_item, &root->root_key);
1424 generation = btrfs_root_generation(&root->root_item);
1425 root->node = read_tree_block(fs_info,
1426 btrfs_root_bytenr(&root->root_item),
1428 if (IS_ERR(root->node)) {
1429 ret = PTR_ERR(root->node);
1431 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1433 free_extent_buffer(root->node);
1436 root->commit_root = btrfs_root_node(root);
1438 btrfs_free_path(path);
1444 root = ERR_PTR(ret);
1448 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1449 struct btrfs_key *location)
1451 struct btrfs_root *root;
1453 root = btrfs_read_tree_root(tree_root, location);
1457 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1458 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1459 btrfs_check_and_init_root_item(&root->root_item);
1465 int btrfs_init_fs_root(struct btrfs_root *root)
1468 struct btrfs_subvolume_writers *writers;
1470 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1471 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1473 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1478 writers = btrfs_alloc_subvolume_writers();
1479 if (IS_ERR(writers)) {
1480 ret = PTR_ERR(writers);
1483 root->subv_writers = writers;
1485 btrfs_init_free_ino_ctl(root);
1486 spin_lock_init(&root->ino_cache_lock);
1487 init_waitqueue_head(&root->ino_cache_wait);
1489 ret = get_anon_bdev(&root->anon_dev);
1493 mutex_lock(&root->objectid_mutex);
1494 ret = btrfs_find_highest_objectid(root,
1495 &root->highest_objectid);
1497 mutex_unlock(&root->objectid_mutex);
1501 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1503 mutex_unlock(&root->objectid_mutex);
1507 /* the caller is responsible to call free_fs_root */
1511 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1514 struct btrfs_root *root;
1516 spin_lock(&fs_info->fs_roots_radix_lock);
1517 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1518 (unsigned long)root_id);
1519 spin_unlock(&fs_info->fs_roots_radix_lock);
1523 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1524 struct btrfs_root *root)
1528 ret = radix_tree_preload(GFP_NOFS);
1532 spin_lock(&fs_info->fs_roots_radix_lock);
1533 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1534 (unsigned long)root->root_key.objectid,
1537 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1538 spin_unlock(&fs_info->fs_roots_radix_lock);
1539 radix_tree_preload_end();
1544 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1545 struct btrfs_key *location,
1548 struct btrfs_root *root;
1549 struct btrfs_path *path;
1550 struct btrfs_key key;
1553 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1554 return fs_info->tree_root;
1555 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1556 return fs_info->extent_root;
1557 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1558 return fs_info->chunk_root;
1559 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1560 return fs_info->dev_root;
1561 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1562 return fs_info->csum_root;
1563 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1564 return fs_info->quota_root ? fs_info->quota_root :
1566 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1567 return fs_info->uuid_root ? fs_info->uuid_root :
1569 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1570 return fs_info->free_space_root ? fs_info->free_space_root :
1573 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1575 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1576 return ERR_PTR(-ENOENT);
1580 root = btrfs_read_fs_root(fs_info->tree_root, location);
1584 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1589 ret = btrfs_init_fs_root(root);
1593 path = btrfs_alloc_path();
1598 key.objectid = BTRFS_ORPHAN_OBJECTID;
1599 key.type = BTRFS_ORPHAN_ITEM_KEY;
1600 key.offset = location->objectid;
1602 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1603 btrfs_free_path(path);
1607 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1609 ret = btrfs_insert_fs_root(fs_info, root);
1611 if (ret == -EEXIST) {
1620 return ERR_PTR(ret);
1623 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1625 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1627 struct btrfs_device *device;
1628 struct backing_dev_info *bdi;
1631 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1634 bdi = device->bdev->bd_bdi;
1635 if (bdi_congested(bdi, bdi_bits)) {
1645 * called by the kthread helper functions to finally call the bio end_io
1646 * functions. This is where read checksum verification actually happens
1648 static void end_workqueue_fn(struct btrfs_work *work)
1651 struct btrfs_end_io_wq *end_io_wq;
1653 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1654 bio = end_io_wq->bio;
1656 bio->bi_status = end_io_wq->status;
1657 bio->bi_private = end_io_wq->private;
1658 bio->bi_end_io = end_io_wq->end_io;
1659 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1663 static int cleaner_kthread(void *arg)
1665 struct btrfs_root *root = arg;
1666 struct btrfs_fs_info *fs_info = root->fs_info;
1668 struct btrfs_trans_handle *trans;
1673 /* Make the cleaner go to sleep early. */
1674 if (btrfs_need_cleaner_sleep(fs_info))
1678 * Do not do anything if we might cause open_ctree() to block
1679 * before we have finished mounting the filesystem.
1681 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1684 if (!mutex_trylock(&fs_info->cleaner_mutex))
1688 * Avoid the problem that we change the status of the fs
1689 * during the above check and trylock.
1691 if (btrfs_need_cleaner_sleep(fs_info)) {
1692 mutex_unlock(&fs_info->cleaner_mutex);
1696 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1697 btrfs_run_delayed_iputs(fs_info);
1698 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1700 again = btrfs_clean_one_deleted_snapshot(root);
1701 mutex_unlock(&fs_info->cleaner_mutex);
1704 * The defragger has dealt with the R/O remount and umount,
1705 * needn't do anything special here.
1707 btrfs_run_defrag_inodes(fs_info);
1710 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1711 * with relocation (btrfs_relocate_chunk) and relocation
1712 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1713 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1714 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1715 * unused block groups.
1717 btrfs_delete_unused_bgs(fs_info);
1720 set_current_state(TASK_INTERRUPTIBLE);
1721 if (!kthread_should_stop())
1723 __set_current_state(TASK_RUNNING);
1725 } while (!kthread_should_stop());
1728 * Transaction kthread is stopped before us and wakes us up.
1729 * However we might have started a new transaction and COWed some
1730 * tree blocks when deleting unused block groups for example. So
1731 * make sure we commit the transaction we started to have a clean
1732 * shutdown when evicting the btree inode - if it has dirty pages
1733 * when we do the final iput() on it, eviction will trigger a
1734 * writeback for it which will fail with null pointer dereferences
1735 * since work queues and other resources were already released and
1736 * destroyed by the time the iput/eviction/writeback is made.
1738 trans = btrfs_attach_transaction(root);
1739 if (IS_ERR(trans)) {
1740 if (PTR_ERR(trans) != -ENOENT)
1742 "cleaner transaction attach returned %ld",
1747 ret = btrfs_commit_transaction(trans);
1750 "cleaner open transaction commit returned %d",
1757 static int transaction_kthread(void *arg)
1759 struct btrfs_root *root = arg;
1760 struct btrfs_fs_info *fs_info = root->fs_info;
1761 struct btrfs_trans_handle *trans;
1762 struct btrfs_transaction *cur;
1765 unsigned long delay;
1769 cannot_commit = false;
1770 delay = HZ * fs_info->commit_interval;
1771 mutex_lock(&fs_info->transaction_kthread_mutex);
1773 spin_lock(&fs_info->trans_lock);
1774 cur = fs_info->running_transaction;
1776 spin_unlock(&fs_info->trans_lock);
1780 now = get_seconds();
1781 if (cur->state < TRANS_STATE_BLOCKED &&
1782 (now < cur->start_time ||
1783 now - cur->start_time < fs_info->commit_interval)) {
1784 spin_unlock(&fs_info->trans_lock);
1788 transid = cur->transid;
1789 spin_unlock(&fs_info->trans_lock);
1791 /* If the file system is aborted, this will always fail. */
1792 trans = btrfs_attach_transaction(root);
1793 if (IS_ERR(trans)) {
1794 if (PTR_ERR(trans) != -ENOENT)
1795 cannot_commit = true;
1798 if (transid == trans->transid) {
1799 btrfs_commit_transaction(trans);
1801 btrfs_end_transaction(trans);
1804 wake_up_process(fs_info->cleaner_kthread);
1805 mutex_unlock(&fs_info->transaction_kthread_mutex);
1807 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1808 &fs_info->fs_state)))
1809 btrfs_cleanup_transaction(fs_info);
1810 set_current_state(TASK_INTERRUPTIBLE);
1811 if (!kthread_should_stop() &&
1812 (!btrfs_transaction_blocked(fs_info) ||
1814 schedule_timeout(delay);
1815 __set_current_state(TASK_RUNNING);
1816 } while (!kthread_should_stop());
1821 * this will find the highest generation in the array of
1822 * root backups. The index of the highest array is returned,
1823 * or -1 if we can't find anything.
1825 * We check to make sure the array is valid by comparing the
1826 * generation of the latest root in the array with the generation
1827 * in the super block. If they don't match we pitch it.
1829 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1832 int newest_index = -1;
1833 struct btrfs_root_backup *root_backup;
1836 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1837 root_backup = info->super_copy->super_roots + i;
1838 cur = btrfs_backup_tree_root_gen(root_backup);
1839 if (cur == newest_gen)
1843 /* check to see if we actually wrapped around */
1844 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1845 root_backup = info->super_copy->super_roots;
1846 cur = btrfs_backup_tree_root_gen(root_backup);
1847 if (cur == newest_gen)
1850 return newest_index;
1855 * find the oldest backup so we know where to store new entries
1856 * in the backup array. This will set the backup_root_index
1857 * field in the fs_info struct
1859 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1862 int newest_index = -1;
1864 newest_index = find_newest_super_backup(info, newest_gen);
1865 /* if there was garbage in there, just move along */
1866 if (newest_index == -1) {
1867 info->backup_root_index = 0;
1869 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1874 * copy all the root pointers into the super backup array.
1875 * this will bump the backup pointer by one when it is
1878 static void backup_super_roots(struct btrfs_fs_info *info)
1881 struct btrfs_root_backup *root_backup;
1884 next_backup = info->backup_root_index;
1885 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1886 BTRFS_NUM_BACKUP_ROOTS;
1889 * just overwrite the last backup if we're at the same generation
1890 * this happens only at umount
1892 root_backup = info->super_for_commit->super_roots + last_backup;
1893 if (btrfs_backup_tree_root_gen(root_backup) ==
1894 btrfs_header_generation(info->tree_root->node))
1895 next_backup = last_backup;
1897 root_backup = info->super_for_commit->super_roots + next_backup;
1900 * make sure all of our padding and empty slots get zero filled
1901 * regardless of which ones we use today
1903 memset(root_backup, 0, sizeof(*root_backup));
1905 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1907 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1908 btrfs_set_backup_tree_root_gen(root_backup,
1909 btrfs_header_generation(info->tree_root->node));
1911 btrfs_set_backup_tree_root_level(root_backup,
1912 btrfs_header_level(info->tree_root->node));
1914 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1915 btrfs_set_backup_chunk_root_gen(root_backup,
1916 btrfs_header_generation(info->chunk_root->node));
1917 btrfs_set_backup_chunk_root_level(root_backup,
1918 btrfs_header_level(info->chunk_root->node));
1920 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1921 btrfs_set_backup_extent_root_gen(root_backup,
1922 btrfs_header_generation(info->extent_root->node));
1923 btrfs_set_backup_extent_root_level(root_backup,
1924 btrfs_header_level(info->extent_root->node));
1927 * we might commit during log recovery, which happens before we set
1928 * the fs_root. Make sure it is valid before we fill it in.
1930 if (info->fs_root && info->fs_root->node) {
1931 btrfs_set_backup_fs_root(root_backup,
1932 info->fs_root->node->start);
1933 btrfs_set_backup_fs_root_gen(root_backup,
1934 btrfs_header_generation(info->fs_root->node));
1935 btrfs_set_backup_fs_root_level(root_backup,
1936 btrfs_header_level(info->fs_root->node));
1939 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1940 btrfs_set_backup_dev_root_gen(root_backup,
1941 btrfs_header_generation(info->dev_root->node));
1942 btrfs_set_backup_dev_root_level(root_backup,
1943 btrfs_header_level(info->dev_root->node));
1945 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1946 btrfs_set_backup_csum_root_gen(root_backup,
1947 btrfs_header_generation(info->csum_root->node));
1948 btrfs_set_backup_csum_root_level(root_backup,
1949 btrfs_header_level(info->csum_root->node));
1951 btrfs_set_backup_total_bytes(root_backup,
1952 btrfs_super_total_bytes(info->super_copy));
1953 btrfs_set_backup_bytes_used(root_backup,
1954 btrfs_super_bytes_used(info->super_copy));
1955 btrfs_set_backup_num_devices(root_backup,
1956 btrfs_super_num_devices(info->super_copy));
1959 * if we don't copy this out to the super_copy, it won't get remembered
1960 * for the next commit
1962 memcpy(&info->super_copy->super_roots,
1963 &info->super_for_commit->super_roots,
1964 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1968 * this copies info out of the root backup array and back into
1969 * the in-memory super block. It is meant to help iterate through
1970 * the array, so you send it the number of backups you've already
1971 * tried and the last backup index you used.
1973 * this returns -1 when it has tried all the backups
1975 static noinline int next_root_backup(struct btrfs_fs_info *info,
1976 struct btrfs_super_block *super,
1977 int *num_backups_tried, int *backup_index)
1979 struct btrfs_root_backup *root_backup;
1980 int newest = *backup_index;
1982 if (*num_backups_tried == 0) {
1983 u64 gen = btrfs_super_generation(super);
1985 newest = find_newest_super_backup(info, gen);
1989 *backup_index = newest;
1990 *num_backups_tried = 1;
1991 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1992 /* we've tried all the backups, all done */
1995 /* jump to the next oldest backup */
1996 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1997 BTRFS_NUM_BACKUP_ROOTS;
1998 *backup_index = newest;
1999 *num_backups_tried += 1;
2001 root_backup = super->super_roots + newest;
2003 btrfs_set_super_generation(super,
2004 btrfs_backup_tree_root_gen(root_backup));
2005 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2006 btrfs_set_super_root_level(super,
2007 btrfs_backup_tree_root_level(root_backup));
2008 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2011 * fixme: the total bytes and num_devices need to match or we should
2014 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2015 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2019 /* helper to cleanup workers */
2020 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2022 btrfs_destroy_workqueue(fs_info->fixup_workers);
2023 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2024 btrfs_destroy_workqueue(fs_info->workers);
2025 btrfs_destroy_workqueue(fs_info->endio_workers);
2026 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2027 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2028 btrfs_destroy_workqueue(fs_info->rmw_workers);
2029 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2030 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2031 btrfs_destroy_workqueue(fs_info->submit_workers);
2032 btrfs_destroy_workqueue(fs_info->delayed_workers);
2033 btrfs_destroy_workqueue(fs_info->caching_workers);
2034 btrfs_destroy_workqueue(fs_info->readahead_workers);
2035 btrfs_destroy_workqueue(fs_info->flush_workers);
2036 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2037 btrfs_destroy_workqueue(fs_info->extent_workers);
2039 * Now that all other work queues are destroyed, we can safely destroy
2040 * the queues used for metadata I/O, since tasks from those other work
2041 * queues can do metadata I/O operations.
2043 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2044 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2047 static void free_root_extent_buffers(struct btrfs_root *root)
2050 free_extent_buffer(root->node);
2051 free_extent_buffer(root->commit_root);
2053 root->commit_root = NULL;
2057 /* helper to cleanup tree roots */
2058 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2060 free_root_extent_buffers(info->tree_root);
2062 free_root_extent_buffers(info->dev_root);
2063 free_root_extent_buffers(info->extent_root);
2064 free_root_extent_buffers(info->csum_root);
2065 free_root_extent_buffers(info->quota_root);
2066 free_root_extent_buffers(info->uuid_root);
2068 free_root_extent_buffers(info->chunk_root);
2069 free_root_extent_buffers(info->free_space_root);
2072 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2075 struct btrfs_root *gang[8];
2078 while (!list_empty(&fs_info->dead_roots)) {
2079 gang[0] = list_entry(fs_info->dead_roots.next,
2080 struct btrfs_root, root_list);
2081 list_del(&gang[0]->root_list);
2083 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2084 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2086 free_extent_buffer(gang[0]->node);
2087 free_extent_buffer(gang[0]->commit_root);
2088 btrfs_put_fs_root(gang[0]);
2093 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2098 for (i = 0; i < ret; i++)
2099 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2102 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2103 btrfs_free_log_root_tree(NULL, fs_info);
2104 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2108 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2110 mutex_init(&fs_info->scrub_lock);
2111 atomic_set(&fs_info->scrubs_running, 0);
2112 atomic_set(&fs_info->scrub_pause_req, 0);
2113 atomic_set(&fs_info->scrubs_paused, 0);
2114 atomic_set(&fs_info->scrub_cancel_req, 0);
2115 init_waitqueue_head(&fs_info->scrub_pause_wait);
2116 fs_info->scrub_workers_refcnt = 0;
2119 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2121 spin_lock_init(&fs_info->balance_lock);
2122 mutex_init(&fs_info->balance_mutex);
2123 atomic_set(&fs_info->balance_running, 0);
2124 atomic_set(&fs_info->balance_pause_req, 0);
2125 atomic_set(&fs_info->balance_cancel_req, 0);
2126 fs_info->balance_ctl = NULL;
2127 init_waitqueue_head(&fs_info->balance_wait_q);
2130 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2132 struct inode *inode = fs_info->btree_inode;
2134 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2135 set_nlink(inode, 1);
2137 * we set the i_size on the btree inode to the max possible int.
2138 * the real end of the address space is determined by all of
2139 * the devices in the system
2141 inode->i_size = OFFSET_MAX;
2142 inode->i_mapping->a_ops = &btree_aops;
2144 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2145 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2146 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2147 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2149 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2151 BTRFS_I(inode)->root = fs_info->tree_root;
2152 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2153 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2154 btrfs_insert_inode_hash(inode);
2157 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2159 fs_info->dev_replace.lock_owner = 0;
2160 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2161 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2162 rwlock_init(&fs_info->dev_replace.lock);
2163 atomic_set(&fs_info->dev_replace.read_locks, 0);
2164 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2165 init_waitqueue_head(&fs_info->replace_wait);
2166 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2169 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2171 spin_lock_init(&fs_info->qgroup_lock);
2172 mutex_init(&fs_info->qgroup_ioctl_lock);
2173 fs_info->qgroup_tree = RB_ROOT;
2174 fs_info->qgroup_op_tree = RB_ROOT;
2175 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2176 fs_info->qgroup_seq = 1;
2177 fs_info->qgroup_ulist = NULL;
2178 fs_info->qgroup_rescan_running = false;
2179 mutex_init(&fs_info->qgroup_rescan_lock);
2182 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2183 struct btrfs_fs_devices *fs_devices)
2185 int max_active = fs_info->thread_pool_size;
2186 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2189 btrfs_alloc_workqueue(fs_info, "worker",
2190 flags | WQ_HIGHPRI, max_active, 16);
2192 fs_info->delalloc_workers =
2193 btrfs_alloc_workqueue(fs_info, "delalloc",
2194 flags, max_active, 2);
2196 fs_info->flush_workers =
2197 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2198 flags, max_active, 0);
2200 fs_info->caching_workers =
2201 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2204 * a higher idle thresh on the submit workers makes it much more
2205 * likely that bios will be send down in a sane order to the
2208 fs_info->submit_workers =
2209 btrfs_alloc_workqueue(fs_info, "submit", flags,
2210 min_t(u64, fs_devices->num_devices,
2213 fs_info->fixup_workers =
2214 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2217 * endios are largely parallel and should have a very
2220 fs_info->endio_workers =
2221 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2222 fs_info->endio_meta_workers =
2223 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2225 fs_info->endio_meta_write_workers =
2226 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2228 fs_info->endio_raid56_workers =
2229 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2231 fs_info->endio_repair_workers =
2232 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2233 fs_info->rmw_workers =
2234 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2235 fs_info->endio_write_workers =
2236 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2238 fs_info->endio_freespace_worker =
2239 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2241 fs_info->delayed_workers =
2242 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2244 fs_info->readahead_workers =
2245 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2247 fs_info->qgroup_rescan_workers =
2248 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2249 fs_info->extent_workers =
2250 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2251 min_t(u64, fs_devices->num_devices,
2254 if (!(fs_info->workers && fs_info->delalloc_workers &&
2255 fs_info->submit_workers && fs_info->flush_workers &&
2256 fs_info->endio_workers && fs_info->endio_meta_workers &&
2257 fs_info->endio_meta_write_workers &&
2258 fs_info->endio_repair_workers &&
2259 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2260 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2261 fs_info->caching_workers && fs_info->readahead_workers &&
2262 fs_info->fixup_workers && fs_info->delayed_workers &&
2263 fs_info->extent_workers &&
2264 fs_info->qgroup_rescan_workers)) {
2271 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2272 struct btrfs_fs_devices *fs_devices)
2275 struct btrfs_root *log_tree_root;
2276 struct btrfs_super_block *disk_super = fs_info->super_copy;
2277 u64 bytenr = btrfs_super_log_root(disk_super);
2279 if (fs_devices->rw_devices == 0) {
2280 btrfs_warn(fs_info, "log replay required on RO media");
2284 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2288 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2290 log_tree_root->node = read_tree_block(fs_info, bytenr,
2291 fs_info->generation + 1);
2292 if (IS_ERR(log_tree_root->node)) {
2293 btrfs_warn(fs_info, "failed to read log tree");
2294 ret = PTR_ERR(log_tree_root->node);
2295 kfree(log_tree_root);
2297 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2298 btrfs_err(fs_info, "failed to read log tree");
2299 free_extent_buffer(log_tree_root->node);
2300 kfree(log_tree_root);
2303 /* returns with log_tree_root freed on success */
2304 ret = btrfs_recover_log_trees(log_tree_root);
2306 btrfs_handle_fs_error(fs_info, ret,
2307 "Failed to recover log tree");
2308 free_extent_buffer(log_tree_root->node);
2309 kfree(log_tree_root);
2313 if (sb_rdonly(fs_info->sb)) {
2314 ret = btrfs_commit_super(fs_info);
2322 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2324 struct btrfs_root *tree_root = fs_info->tree_root;
2325 struct btrfs_root *root;
2326 struct btrfs_key location;
2329 BUG_ON(!fs_info->tree_root);
2331 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2332 location.type = BTRFS_ROOT_ITEM_KEY;
2333 location.offset = 0;
2335 root = btrfs_read_tree_root(tree_root, &location);
2337 return PTR_ERR(root);
2338 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2339 fs_info->extent_root = root;
2341 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2342 root = btrfs_read_tree_root(tree_root, &location);
2344 return PTR_ERR(root);
2345 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2346 fs_info->dev_root = root;
2347 btrfs_init_devices_late(fs_info);
2349 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2350 root = btrfs_read_tree_root(tree_root, &location);
2352 return PTR_ERR(root);
2353 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2354 fs_info->csum_root = root;
2356 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2357 root = btrfs_read_tree_root(tree_root, &location);
2358 if (!IS_ERR(root)) {
2359 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2360 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2361 fs_info->quota_root = root;
2364 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2365 root = btrfs_read_tree_root(tree_root, &location);
2367 ret = PTR_ERR(root);
2371 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2372 fs_info->uuid_root = root;
2375 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2376 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2377 root = btrfs_read_tree_root(tree_root, &location);
2379 return PTR_ERR(root);
2380 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2381 fs_info->free_space_root = root;
2387 int open_ctree(struct super_block *sb,
2388 struct btrfs_fs_devices *fs_devices,
2396 struct btrfs_key location;
2397 struct buffer_head *bh;
2398 struct btrfs_super_block *disk_super;
2399 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2400 struct btrfs_root *tree_root;
2401 struct btrfs_root *chunk_root;
2404 int num_backups_tried = 0;
2405 int backup_index = 0;
2407 int clear_free_space_tree = 0;
2409 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2410 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2411 if (!tree_root || !chunk_root) {
2416 ret = init_srcu_struct(&fs_info->subvol_srcu);
2422 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2427 fs_info->dirty_metadata_batch = PAGE_SIZE *
2428 (1 + ilog2(nr_cpu_ids));
2430 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2433 goto fail_dirty_metadata_bytes;
2436 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2439 goto fail_delalloc_bytes;
2442 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2443 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2444 INIT_LIST_HEAD(&fs_info->trans_list);
2445 INIT_LIST_HEAD(&fs_info->dead_roots);
2446 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2447 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2448 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2449 spin_lock_init(&fs_info->delalloc_root_lock);
2450 spin_lock_init(&fs_info->trans_lock);
2451 spin_lock_init(&fs_info->fs_roots_radix_lock);
2452 spin_lock_init(&fs_info->delayed_iput_lock);
2453 spin_lock_init(&fs_info->defrag_inodes_lock);
2454 spin_lock_init(&fs_info->tree_mod_seq_lock);
2455 spin_lock_init(&fs_info->super_lock);
2456 spin_lock_init(&fs_info->qgroup_op_lock);
2457 spin_lock_init(&fs_info->buffer_lock);
2458 spin_lock_init(&fs_info->unused_bgs_lock);
2459 rwlock_init(&fs_info->tree_mod_log_lock);
2460 mutex_init(&fs_info->unused_bg_unpin_mutex);
2461 mutex_init(&fs_info->delete_unused_bgs_mutex);
2462 mutex_init(&fs_info->reloc_mutex);
2463 mutex_init(&fs_info->delalloc_root_mutex);
2464 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2465 seqlock_init(&fs_info->profiles_lock);
2467 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2468 INIT_LIST_HEAD(&fs_info->space_info);
2469 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2470 INIT_LIST_HEAD(&fs_info->unused_bgs);
2471 btrfs_mapping_init(&fs_info->mapping_tree);
2472 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2473 BTRFS_BLOCK_RSV_GLOBAL);
2474 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2475 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2476 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2477 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2478 BTRFS_BLOCK_RSV_DELOPS);
2479 atomic_set(&fs_info->async_delalloc_pages, 0);
2480 atomic_set(&fs_info->defrag_running, 0);
2481 atomic_set(&fs_info->qgroup_op_seq, 0);
2482 atomic_set(&fs_info->reada_works_cnt, 0);
2483 atomic64_set(&fs_info->tree_mod_seq, 0);
2485 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2486 fs_info->metadata_ratio = 0;
2487 fs_info->defrag_inodes = RB_ROOT;
2488 atomic64_set(&fs_info->free_chunk_space, 0);
2489 fs_info->tree_mod_log = RB_ROOT;
2490 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2491 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2492 /* readahead state */
2493 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2494 spin_lock_init(&fs_info->reada_lock);
2495 btrfs_init_ref_verify(fs_info);
2497 fs_info->thread_pool_size = min_t(unsigned long,
2498 num_online_cpus() + 2, 8);
2500 INIT_LIST_HEAD(&fs_info->ordered_roots);
2501 spin_lock_init(&fs_info->ordered_root_lock);
2503 fs_info->btree_inode = new_inode(sb);
2504 if (!fs_info->btree_inode) {
2506 goto fail_bio_counter;
2508 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2510 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2512 if (!fs_info->delayed_root) {
2516 btrfs_init_delayed_root(fs_info->delayed_root);
2518 btrfs_init_scrub(fs_info);
2519 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2520 fs_info->check_integrity_print_mask = 0;
2522 btrfs_init_balance(fs_info);
2523 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2525 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2526 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2528 btrfs_init_btree_inode(fs_info);
2530 spin_lock_init(&fs_info->block_group_cache_lock);
2531 fs_info->block_group_cache_tree = RB_ROOT;
2532 fs_info->first_logical_byte = (u64)-1;
2534 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2535 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2536 fs_info->pinned_extents = &fs_info->freed_extents[0];
2537 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2539 mutex_init(&fs_info->ordered_operations_mutex);
2540 mutex_init(&fs_info->tree_log_mutex);
2541 mutex_init(&fs_info->chunk_mutex);
2542 mutex_init(&fs_info->transaction_kthread_mutex);
2543 mutex_init(&fs_info->cleaner_mutex);
2544 mutex_init(&fs_info->volume_mutex);
2545 mutex_init(&fs_info->ro_block_group_mutex);
2546 init_rwsem(&fs_info->commit_root_sem);
2547 init_rwsem(&fs_info->cleanup_work_sem);
2548 init_rwsem(&fs_info->subvol_sem);
2549 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2551 btrfs_init_dev_replace_locks(fs_info);
2552 btrfs_init_qgroup(fs_info);
2554 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2555 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2557 init_waitqueue_head(&fs_info->transaction_throttle);
2558 init_waitqueue_head(&fs_info->transaction_wait);
2559 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2560 init_waitqueue_head(&fs_info->async_submit_wait);
2562 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2564 /* Usable values until the real ones are cached from the superblock */
2565 fs_info->nodesize = 4096;
2566 fs_info->sectorsize = 4096;
2567 fs_info->stripesize = 4096;
2569 ret = btrfs_alloc_stripe_hash_table(fs_info);
2575 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2577 invalidate_bdev(fs_devices->latest_bdev);
2580 * Read super block and check the signature bytes only
2582 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2589 * We want to check superblock checksum, the type is stored inside.
2590 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2592 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2593 btrfs_err(fs_info, "superblock checksum mismatch");
2600 * super_copy is zeroed at allocation time and we never touch the
2601 * following bytes up to INFO_SIZE, the checksum is calculated from
2602 * the whole block of INFO_SIZE
2604 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2605 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2606 sizeof(*fs_info->super_for_commit));
2609 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2611 ret = btrfs_check_super_valid(fs_info);
2613 btrfs_err(fs_info, "superblock contains fatal errors");
2618 disk_super = fs_info->super_copy;
2619 if (!btrfs_super_root(disk_super))
2622 /* check FS state, whether FS is broken. */
2623 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2624 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2627 * run through our array of backup supers and setup
2628 * our ring pointer to the oldest one
2630 generation = btrfs_super_generation(disk_super);
2631 find_oldest_super_backup(fs_info, generation);
2634 * In the long term, we'll store the compression type in the super
2635 * block, and it'll be used for per file compression control.
2637 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2639 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2645 features = btrfs_super_incompat_flags(disk_super) &
2646 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2649 "cannot mount because of unsupported optional features (%llx)",
2655 features = btrfs_super_incompat_flags(disk_super);
2656 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2657 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2658 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2659 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2660 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2662 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2663 btrfs_info(fs_info, "has skinny extents");
2666 * flag our filesystem as having big metadata blocks if
2667 * they are bigger than the page size
2669 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2670 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2672 "flagging fs with big metadata feature");
2673 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2676 nodesize = btrfs_super_nodesize(disk_super);
2677 sectorsize = btrfs_super_sectorsize(disk_super);
2678 stripesize = sectorsize;
2679 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2680 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2682 /* Cache block sizes */
2683 fs_info->nodesize = nodesize;
2684 fs_info->sectorsize = sectorsize;
2685 fs_info->stripesize = stripesize;
2688 * mixed block groups end up with duplicate but slightly offset
2689 * extent buffers for the same range. It leads to corruptions
2691 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2692 (sectorsize != nodesize)) {
2694 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2695 nodesize, sectorsize);
2700 * Needn't use the lock because there is no other task which will
2703 btrfs_set_super_incompat_flags(disk_super, features);
2705 features = btrfs_super_compat_ro_flags(disk_super) &
2706 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2707 if (!sb_rdonly(sb) && features) {
2709 "cannot mount read-write because of unsupported optional features (%llx)",
2715 max_active = fs_info->thread_pool_size;
2717 ret = btrfs_init_workqueues(fs_info, fs_devices);
2720 goto fail_sb_buffer;
2723 sb->s_bdi->congested_fn = btrfs_congested_fn;
2724 sb->s_bdi->congested_data = fs_info;
2725 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2726 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2727 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2728 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2730 sb->s_blocksize = sectorsize;
2731 sb->s_blocksize_bits = blksize_bits(sectorsize);
2732 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2734 mutex_lock(&fs_info->chunk_mutex);
2735 ret = btrfs_read_sys_array(fs_info);
2736 mutex_unlock(&fs_info->chunk_mutex);
2738 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2739 goto fail_sb_buffer;
2742 generation = btrfs_super_chunk_root_generation(disk_super);
2744 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2746 chunk_root->node = read_tree_block(fs_info,
2747 btrfs_super_chunk_root(disk_super),
2749 if (IS_ERR(chunk_root->node) ||
2750 !extent_buffer_uptodate(chunk_root->node)) {
2751 btrfs_err(fs_info, "failed to read chunk root");
2752 if (!IS_ERR(chunk_root->node))
2753 free_extent_buffer(chunk_root->node);
2754 chunk_root->node = NULL;
2755 goto fail_tree_roots;
2757 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2758 chunk_root->commit_root = btrfs_root_node(chunk_root);
2760 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2761 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2763 ret = btrfs_read_chunk_tree(fs_info);
2765 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2766 goto fail_tree_roots;
2770 * keep the device that is marked to be the target device for the
2771 * dev_replace procedure
2773 btrfs_close_extra_devices(fs_devices, 0);
2775 if (!fs_devices->latest_bdev) {
2776 btrfs_err(fs_info, "failed to read devices");
2777 goto fail_tree_roots;
2781 generation = btrfs_super_generation(disk_super);
2783 tree_root->node = read_tree_block(fs_info,
2784 btrfs_super_root(disk_super),
2786 if (IS_ERR(tree_root->node) ||
2787 !extent_buffer_uptodate(tree_root->node)) {
2788 btrfs_warn(fs_info, "failed to read tree root");
2789 if (!IS_ERR(tree_root->node))
2790 free_extent_buffer(tree_root->node);
2791 tree_root->node = NULL;
2792 goto recovery_tree_root;
2795 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2796 tree_root->commit_root = btrfs_root_node(tree_root);
2797 btrfs_set_root_refs(&tree_root->root_item, 1);
2799 mutex_lock(&tree_root->objectid_mutex);
2800 ret = btrfs_find_highest_objectid(tree_root,
2801 &tree_root->highest_objectid);
2803 mutex_unlock(&tree_root->objectid_mutex);
2804 goto recovery_tree_root;
2807 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2809 mutex_unlock(&tree_root->objectid_mutex);
2811 ret = btrfs_read_roots(fs_info);
2813 goto recovery_tree_root;
2815 fs_info->generation = generation;
2816 fs_info->last_trans_committed = generation;
2818 ret = btrfs_recover_balance(fs_info);
2820 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2821 goto fail_block_groups;
2824 ret = btrfs_init_dev_stats(fs_info);
2826 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2827 goto fail_block_groups;
2830 ret = btrfs_init_dev_replace(fs_info);
2832 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2833 goto fail_block_groups;
2836 btrfs_close_extra_devices(fs_devices, 1);
2838 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2840 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2842 goto fail_block_groups;
2845 ret = btrfs_sysfs_add_device(fs_devices);
2847 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2849 goto fail_fsdev_sysfs;
2852 ret = btrfs_sysfs_add_mounted(fs_info);
2854 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2855 goto fail_fsdev_sysfs;
2858 ret = btrfs_init_space_info(fs_info);
2860 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2864 ret = btrfs_read_block_groups(fs_info);
2866 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2870 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
2872 "writeable mount is not allowed due to too many missing devices");
2876 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2878 if (IS_ERR(fs_info->cleaner_kthread))
2881 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2883 "btrfs-transaction");
2884 if (IS_ERR(fs_info->transaction_kthread))
2887 if (!btrfs_test_opt(fs_info, NOSSD) &&
2888 !fs_info->fs_devices->rotating) {
2889 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2893 * Mount does not set all options immediately, we can do it now and do
2894 * not have to wait for transaction commit
2896 btrfs_apply_pending_changes(fs_info);
2898 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2899 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2900 ret = btrfsic_mount(fs_info, fs_devices,
2901 btrfs_test_opt(fs_info,
2902 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2904 fs_info->check_integrity_print_mask);
2907 "failed to initialize integrity check module: %d",
2911 ret = btrfs_read_qgroup_config(fs_info);
2913 goto fail_trans_kthread;
2915 if (btrfs_build_ref_tree(fs_info))
2916 btrfs_err(fs_info, "couldn't build ref tree");
2918 /* do not make disk changes in broken FS or nologreplay is given */
2919 if (btrfs_super_log_root(disk_super) != 0 &&
2920 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2921 ret = btrfs_replay_log(fs_info, fs_devices);
2928 ret = btrfs_find_orphan_roots(fs_info);
2932 if (!sb_rdonly(sb)) {
2933 ret = btrfs_cleanup_fs_roots(fs_info);
2937 mutex_lock(&fs_info->cleaner_mutex);
2938 ret = btrfs_recover_relocation(tree_root);
2939 mutex_unlock(&fs_info->cleaner_mutex);
2941 btrfs_warn(fs_info, "failed to recover relocation: %d",
2948 location.objectid = BTRFS_FS_TREE_OBJECTID;
2949 location.type = BTRFS_ROOT_ITEM_KEY;
2950 location.offset = 0;
2952 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2953 if (IS_ERR(fs_info->fs_root)) {
2954 err = PTR_ERR(fs_info->fs_root);
2961 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2962 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2963 clear_free_space_tree = 1;
2964 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2965 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2966 btrfs_warn(fs_info, "free space tree is invalid");
2967 clear_free_space_tree = 1;
2970 if (clear_free_space_tree) {
2971 btrfs_info(fs_info, "clearing free space tree");
2972 ret = btrfs_clear_free_space_tree(fs_info);
2975 "failed to clear free space tree: %d", ret);
2976 close_ctree(fs_info);
2981 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
2982 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2983 btrfs_info(fs_info, "creating free space tree");
2984 ret = btrfs_create_free_space_tree(fs_info);
2987 "failed to create free space tree: %d", ret);
2988 close_ctree(fs_info);
2993 down_read(&fs_info->cleanup_work_sem);
2994 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2995 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2996 up_read(&fs_info->cleanup_work_sem);
2997 close_ctree(fs_info);
3000 up_read(&fs_info->cleanup_work_sem);
3002 ret = btrfs_resume_balance_async(fs_info);
3004 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3005 close_ctree(fs_info);
3009 ret = btrfs_resume_dev_replace_async(fs_info);
3011 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3012 close_ctree(fs_info);
3016 btrfs_qgroup_rescan_resume(fs_info);
3018 if (!fs_info->uuid_root) {
3019 btrfs_info(fs_info, "creating UUID tree");
3020 ret = btrfs_create_uuid_tree(fs_info);
3023 "failed to create the UUID tree: %d", ret);
3024 close_ctree(fs_info);
3027 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3028 fs_info->generation !=
3029 btrfs_super_uuid_tree_generation(disk_super)) {
3030 btrfs_info(fs_info, "checking UUID tree");
3031 ret = btrfs_check_uuid_tree(fs_info);
3034 "failed to check the UUID tree: %d", ret);
3035 close_ctree(fs_info);
3039 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3041 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3044 * backuproot only affect mount behavior, and if open_ctree succeeded,
3045 * no need to keep the flag
3047 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3052 btrfs_free_qgroup_config(fs_info);
3054 kthread_stop(fs_info->transaction_kthread);
3055 btrfs_cleanup_transaction(fs_info);
3056 btrfs_free_fs_roots(fs_info);
3058 kthread_stop(fs_info->cleaner_kthread);
3061 * make sure we're done with the btree inode before we stop our
3064 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3067 btrfs_sysfs_remove_mounted(fs_info);
3070 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3073 btrfs_put_block_group_cache(fs_info);
3076 free_root_pointers(fs_info, 1);
3077 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3080 btrfs_stop_all_workers(fs_info);
3081 btrfs_free_block_groups(fs_info);
3084 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3086 iput(fs_info->btree_inode);
3088 percpu_counter_destroy(&fs_info->bio_counter);
3089 fail_delalloc_bytes:
3090 percpu_counter_destroy(&fs_info->delalloc_bytes);
3091 fail_dirty_metadata_bytes:
3092 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3094 cleanup_srcu_struct(&fs_info->subvol_srcu);
3096 btrfs_free_stripe_hash_table(fs_info);
3097 btrfs_close_devices(fs_info->fs_devices);
3101 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3102 goto fail_tree_roots;
3104 free_root_pointers(fs_info, 0);
3106 /* don't use the log in recovery mode, it won't be valid */
3107 btrfs_set_super_log_root(disk_super, 0);
3109 /* we can't trust the free space cache either */
3110 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3112 ret = next_root_backup(fs_info, fs_info->super_copy,
3113 &num_backups_tried, &backup_index);
3115 goto fail_block_groups;
3116 goto retry_root_backup;
3119 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3122 set_buffer_uptodate(bh);
3124 struct btrfs_device *device = (struct btrfs_device *)
3127 btrfs_warn_rl_in_rcu(device->fs_info,
3128 "lost page write due to IO error on %s",
3129 rcu_str_deref(device->name));
3130 /* note, we don't set_buffer_write_io_error because we have
3131 * our own ways of dealing with the IO errors
3133 clear_buffer_uptodate(bh);
3134 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3140 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3141 struct buffer_head **bh_ret)
3143 struct buffer_head *bh;
3144 struct btrfs_super_block *super;
3147 bytenr = btrfs_sb_offset(copy_num);
3148 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3151 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3153 * If we fail to read from the underlying devices, as of now
3154 * the best option we have is to mark it EIO.
3159 super = (struct btrfs_super_block *)bh->b_data;
3160 if (btrfs_super_bytenr(super) != bytenr ||
3161 btrfs_super_magic(super) != BTRFS_MAGIC) {
3171 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3173 struct buffer_head *bh;
3174 struct buffer_head *latest = NULL;
3175 struct btrfs_super_block *super;
3180 /* we would like to check all the supers, but that would make
3181 * a btrfs mount succeed after a mkfs from a different FS.
3182 * So, we need to add a special mount option to scan for
3183 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3185 for (i = 0; i < 1; i++) {
3186 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3190 super = (struct btrfs_super_block *)bh->b_data;
3192 if (!latest || btrfs_super_generation(super) > transid) {
3195 transid = btrfs_super_generation(super);
3202 return ERR_PTR(ret);
3208 * Write superblock @sb to the @device. Do not wait for completion, all the
3209 * buffer heads we write are pinned.
3211 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3212 * the expected device size at commit time. Note that max_mirrors must be
3213 * same for write and wait phases.
3215 * Return number of errors when buffer head is not found or submission fails.
3217 static int write_dev_supers(struct btrfs_device *device,
3218 struct btrfs_super_block *sb, int max_mirrors)
3220 struct buffer_head *bh;
3228 if (max_mirrors == 0)
3229 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3231 for (i = 0; i < max_mirrors; i++) {
3232 bytenr = btrfs_sb_offset(i);
3233 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3234 device->commit_total_bytes)
3237 btrfs_set_super_bytenr(sb, bytenr);
3240 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3241 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3242 btrfs_csum_final(crc, sb->csum);
3244 /* One reference for us, and we leave it for the caller */
3245 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3246 BTRFS_SUPER_INFO_SIZE);
3248 btrfs_err(device->fs_info,
3249 "couldn't get super buffer head for bytenr %llu",
3255 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3257 /* one reference for submit_bh */
3260 set_buffer_uptodate(bh);
3262 bh->b_end_io = btrfs_end_buffer_write_sync;
3263 bh->b_private = device;
3266 * we fua the first super. The others we allow
3269 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3270 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3271 op_flags |= REQ_FUA;
3272 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3276 return errors < i ? 0 : -1;
3280 * Wait for write completion of superblocks done by write_dev_supers,
3281 * @max_mirrors same for write and wait phases.
3283 * Return number of errors when buffer head is not found or not marked up to
3286 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3288 struct buffer_head *bh;
3293 if (max_mirrors == 0)
3294 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3296 for (i = 0; i < max_mirrors; i++) {
3297 bytenr = btrfs_sb_offset(i);
3298 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3299 device->commit_total_bytes)
3302 bh = __find_get_block(device->bdev,
3303 bytenr / BTRFS_BDEV_BLOCKSIZE,
3304 BTRFS_SUPER_INFO_SIZE);
3310 if (!buffer_uptodate(bh))
3313 /* drop our reference */
3316 /* drop the reference from the writing run */
3320 return errors < i ? 0 : -1;
3324 * endio for the write_dev_flush, this will wake anyone waiting
3325 * for the barrier when it is done
3327 static void btrfs_end_empty_barrier(struct bio *bio)
3329 complete(bio->bi_private);
3333 * Submit a flush request to the device if it supports it. Error handling is
3334 * done in the waiting counterpart.
3336 static void write_dev_flush(struct btrfs_device *device)
3338 struct request_queue *q = bdev_get_queue(device->bdev);
3339 struct bio *bio = device->flush_bio;
3341 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3345 bio->bi_end_io = btrfs_end_empty_barrier;
3346 bio_set_dev(bio, device->bdev);
3347 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3348 init_completion(&device->flush_wait);
3349 bio->bi_private = &device->flush_wait;
3351 btrfsic_submit_bio(bio);
3352 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3356 * If the flush bio has been submitted by write_dev_flush, wait for it.
3358 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3360 struct bio *bio = device->flush_bio;
3362 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3365 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3366 wait_for_completion_io(&device->flush_wait);
3368 return bio->bi_status;
3371 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3373 if (!btrfs_check_rw_degradable(fs_info, NULL))
3379 * send an empty flush down to each device in parallel,
3380 * then wait for them
3382 static int barrier_all_devices(struct btrfs_fs_info *info)
3384 struct list_head *head;
3385 struct btrfs_device *dev;
3386 int errors_wait = 0;
3389 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3390 /* send down all the barriers */
3391 head = &info->fs_devices->devices;
3392 list_for_each_entry(dev, head, dev_list) {
3393 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3397 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3398 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3401 write_dev_flush(dev);
3402 dev->last_flush_error = BLK_STS_OK;
3405 /* wait for all the barriers */
3406 list_for_each_entry(dev, head, dev_list) {
3407 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3413 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3414 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3417 ret = wait_dev_flush(dev);
3419 dev->last_flush_error = ret;
3420 btrfs_dev_stat_inc_and_print(dev,
3421 BTRFS_DEV_STAT_FLUSH_ERRS);
3428 * At some point we need the status of all disks
3429 * to arrive at the volume status. So error checking
3430 * is being pushed to a separate loop.
3432 return check_barrier_error(info);
3437 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3440 int min_tolerated = INT_MAX;
3442 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3443 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3444 min_tolerated = min(min_tolerated,
3445 btrfs_raid_array[BTRFS_RAID_SINGLE].
3446 tolerated_failures);
3448 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3449 if (raid_type == BTRFS_RAID_SINGLE)
3451 if (!(flags & btrfs_raid_group[raid_type]))
3453 min_tolerated = min(min_tolerated,
3454 btrfs_raid_array[raid_type].
3455 tolerated_failures);
3458 if (min_tolerated == INT_MAX) {
3459 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3463 return min_tolerated;
3466 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3468 struct list_head *head;
3469 struct btrfs_device *dev;
3470 struct btrfs_super_block *sb;
3471 struct btrfs_dev_item *dev_item;
3475 int total_errors = 0;
3478 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3481 * max_mirrors == 0 indicates we're from commit_transaction,
3482 * not from fsync where the tree roots in fs_info have not
3483 * been consistent on disk.
3485 if (max_mirrors == 0)
3486 backup_super_roots(fs_info);
3488 sb = fs_info->super_for_commit;
3489 dev_item = &sb->dev_item;
3491 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3492 head = &fs_info->fs_devices->devices;
3493 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3496 ret = barrier_all_devices(fs_info);
3499 &fs_info->fs_devices->device_list_mutex);
3500 btrfs_handle_fs_error(fs_info, ret,
3501 "errors while submitting device barriers.");
3506 list_for_each_entry(dev, head, dev_list) {
3511 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3512 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3515 btrfs_set_stack_device_generation(dev_item, 0);
3516 btrfs_set_stack_device_type(dev_item, dev->type);
3517 btrfs_set_stack_device_id(dev_item, dev->devid);
3518 btrfs_set_stack_device_total_bytes(dev_item,
3519 dev->commit_total_bytes);
3520 btrfs_set_stack_device_bytes_used(dev_item,
3521 dev->commit_bytes_used);
3522 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3523 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3524 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3525 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3526 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3528 flags = btrfs_super_flags(sb);
3529 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3531 ret = write_dev_supers(dev, sb, max_mirrors);
3535 if (total_errors > max_errors) {
3536 btrfs_err(fs_info, "%d errors while writing supers",
3538 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3540 /* FUA is masked off if unsupported and can't be the reason */
3541 btrfs_handle_fs_error(fs_info, -EIO,
3542 "%d errors while writing supers",
3548 list_for_each_entry(dev, head, dev_list) {
3551 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3552 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3555 ret = wait_dev_supers(dev, max_mirrors);
3559 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3560 if (total_errors > max_errors) {
3561 btrfs_handle_fs_error(fs_info, -EIO,
3562 "%d errors while writing supers",
3569 /* Drop a fs root from the radix tree and free it. */
3570 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3571 struct btrfs_root *root)
3573 spin_lock(&fs_info->fs_roots_radix_lock);
3574 radix_tree_delete(&fs_info->fs_roots_radix,
3575 (unsigned long)root->root_key.objectid);
3576 spin_unlock(&fs_info->fs_roots_radix_lock);
3578 if (btrfs_root_refs(&root->root_item) == 0)
3579 synchronize_srcu(&fs_info->subvol_srcu);
3581 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3582 btrfs_free_log(NULL, root);
3583 if (root->reloc_root) {
3584 free_extent_buffer(root->reloc_root->node);
3585 free_extent_buffer(root->reloc_root->commit_root);
3586 btrfs_put_fs_root(root->reloc_root);
3587 root->reloc_root = NULL;
3591 if (root->free_ino_pinned)
3592 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3593 if (root->free_ino_ctl)
3594 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3598 static void free_fs_root(struct btrfs_root *root)
3600 iput(root->ino_cache_inode);
3601 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3602 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3603 root->orphan_block_rsv = NULL;
3605 free_anon_bdev(root->anon_dev);
3606 if (root->subv_writers)
3607 btrfs_free_subvolume_writers(root->subv_writers);
3608 free_extent_buffer(root->node);
3609 free_extent_buffer(root->commit_root);
3610 kfree(root->free_ino_ctl);
3611 kfree(root->free_ino_pinned);
3613 btrfs_put_fs_root(root);
3616 void btrfs_free_fs_root(struct btrfs_root *root)
3621 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3623 u64 root_objectid = 0;
3624 struct btrfs_root *gang[8];
3627 unsigned int ret = 0;
3631 index = srcu_read_lock(&fs_info->subvol_srcu);
3632 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3633 (void **)gang, root_objectid,
3636 srcu_read_unlock(&fs_info->subvol_srcu, index);
3639 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3641 for (i = 0; i < ret; i++) {
3642 /* Avoid to grab roots in dead_roots */
3643 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3647 /* grab all the search result for later use */
3648 gang[i] = btrfs_grab_fs_root(gang[i]);
3650 srcu_read_unlock(&fs_info->subvol_srcu, index);
3652 for (i = 0; i < ret; i++) {
3655 root_objectid = gang[i]->root_key.objectid;
3656 err = btrfs_orphan_cleanup(gang[i]);
3659 btrfs_put_fs_root(gang[i]);
3664 /* release the uncleaned roots due to error */
3665 for (; i < ret; i++) {
3667 btrfs_put_fs_root(gang[i]);
3672 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3674 struct btrfs_root *root = fs_info->tree_root;
3675 struct btrfs_trans_handle *trans;
3677 mutex_lock(&fs_info->cleaner_mutex);
3678 btrfs_run_delayed_iputs(fs_info);
3679 mutex_unlock(&fs_info->cleaner_mutex);
3680 wake_up_process(fs_info->cleaner_kthread);
3682 /* wait until ongoing cleanup work done */
3683 down_write(&fs_info->cleanup_work_sem);
3684 up_write(&fs_info->cleanup_work_sem);
3686 trans = btrfs_join_transaction(root);
3688 return PTR_ERR(trans);
3689 return btrfs_commit_transaction(trans);
3692 void close_ctree(struct btrfs_fs_info *fs_info)
3694 struct btrfs_root *root = fs_info->tree_root;
3697 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3699 /* wait for the qgroup rescan worker to stop */
3700 btrfs_qgroup_wait_for_completion(fs_info, false);
3702 /* wait for the uuid_scan task to finish */
3703 down(&fs_info->uuid_tree_rescan_sem);
3704 /* avoid complains from lockdep et al., set sem back to initial state */
3705 up(&fs_info->uuid_tree_rescan_sem);
3707 /* pause restriper - we want to resume on mount */
3708 btrfs_pause_balance(fs_info);
3710 btrfs_dev_replace_suspend_for_unmount(fs_info);
3712 btrfs_scrub_cancel(fs_info);
3714 /* wait for any defraggers to finish */
3715 wait_event(fs_info->transaction_wait,
3716 (atomic_read(&fs_info->defrag_running) == 0));
3718 /* clear out the rbtree of defraggable inodes */
3719 btrfs_cleanup_defrag_inodes(fs_info);
3721 cancel_work_sync(&fs_info->async_reclaim_work);
3723 if (!sb_rdonly(fs_info->sb)) {
3725 * If the cleaner thread is stopped and there are
3726 * block groups queued for removal, the deletion will be
3727 * skipped when we quit the cleaner thread.
3729 btrfs_delete_unused_bgs(fs_info);
3731 ret = btrfs_commit_super(fs_info);
3733 btrfs_err(fs_info, "commit super ret %d", ret);
3736 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3737 btrfs_error_commit_super(fs_info);
3739 kthread_stop(fs_info->transaction_kthread);
3740 kthread_stop(fs_info->cleaner_kthread);
3742 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3744 btrfs_free_qgroup_config(fs_info);
3746 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3747 btrfs_info(fs_info, "at unmount delalloc count %lld",
3748 percpu_counter_sum(&fs_info->delalloc_bytes));
3751 btrfs_sysfs_remove_mounted(fs_info);
3752 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3754 btrfs_free_fs_roots(fs_info);
3756 btrfs_put_block_group_cache(fs_info);
3759 * we must make sure there is not any read request to
3760 * submit after we stopping all workers.
3762 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3763 btrfs_stop_all_workers(fs_info);
3765 btrfs_free_block_groups(fs_info);
3767 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3768 free_root_pointers(fs_info, 1);
3770 iput(fs_info->btree_inode);
3772 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3773 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3774 btrfsic_unmount(fs_info->fs_devices);
3777 btrfs_close_devices(fs_info->fs_devices);
3778 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3780 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3781 percpu_counter_destroy(&fs_info->delalloc_bytes);
3782 percpu_counter_destroy(&fs_info->bio_counter);
3783 cleanup_srcu_struct(&fs_info->subvol_srcu);
3785 btrfs_free_stripe_hash_table(fs_info);
3786 btrfs_free_ref_cache(fs_info);
3788 __btrfs_free_block_rsv(root->orphan_block_rsv);
3789 root->orphan_block_rsv = NULL;
3791 while (!list_empty(&fs_info->pinned_chunks)) {
3792 struct extent_map *em;
3794 em = list_first_entry(&fs_info->pinned_chunks,
3795 struct extent_map, list);
3796 list_del_init(&em->list);
3797 free_extent_map(em);
3801 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3805 struct inode *btree_inode = buf->pages[0]->mapping->host;
3807 ret = extent_buffer_uptodate(buf);
3811 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3812 parent_transid, atomic);
3818 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3820 struct btrfs_fs_info *fs_info;
3821 struct btrfs_root *root;
3822 u64 transid = btrfs_header_generation(buf);
3825 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3827 * This is a fast path so only do this check if we have sanity tests
3828 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3829 * outside of the sanity tests.
3831 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3834 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3835 fs_info = root->fs_info;
3836 btrfs_assert_tree_locked(buf);
3837 if (transid != fs_info->generation)
3838 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3839 buf->start, transid, fs_info->generation);
3840 was_dirty = set_extent_buffer_dirty(buf);
3842 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3844 fs_info->dirty_metadata_batch);
3845 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3847 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3848 * but item data not updated.
3849 * So here we should only check item pointers, not item data.
3851 if (btrfs_header_level(buf) == 0 &&
3852 btrfs_check_leaf_relaxed(root, buf)) {
3853 btrfs_print_leaf(buf);
3859 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
3863 * looks as though older kernels can get into trouble with
3864 * this code, they end up stuck in balance_dirty_pages forever
3868 if (current->flags & PF_MEMALLOC)
3872 btrfs_balance_delayed_items(fs_info);
3874 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
3875 BTRFS_DIRTY_METADATA_THRESH);
3877 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
3881 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
3883 __btrfs_btree_balance_dirty(fs_info, 1);
3886 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
3888 __btrfs_btree_balance_dirty(fs_info, 0);
3891 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3893 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3894 struct btrfs_fs_info *fs_info = root->fs_info;
3896 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
3899 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
3901 struct btrfs_super_block *sb = fs_info->super_copy;
3902 u64 nodesize = btrfs_super_nodesize(sb);
3903 u64 sectorsize = btrfs_super_sectorsize(sb);
3906 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
3907 btrfs_err(fs_info, "no valid FS found");
3910 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
3911 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
3912 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
3915 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3916 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
3917 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3920 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3921 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
3922 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3925 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3926 btrfs_err(fs_info, "log_root level too big: %d >= %d",
3927 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3932 * Check sectorsize and nodesize first, other check will need it.
3933 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
3935 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
3936 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3937 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
3940 /* Only PAGE SIZE is supported yet */
3941 if (sectorsize != PAGE_SIZE) {
3943 "sectorsize %llu not supported yet, only support %lu",
3944 sectorsize, PAGE_SIZE);
3947 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
3948 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3949 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
3952 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
3953 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
3954 le32_to_cpu(sb->__unused_leafsize), nodesize);
3958 /* Root alignment check */
3959 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
3960 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
3961 btrfs_super_root(sb));
3964 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
3965 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
3966 btrfs_super_chunk_root(sb));
3969 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
3970 btrfs_warn(fs_info, "log_root block unaligned: %llu",
3971 btrfs_super_log_root(sb));
3975 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
3977 "dev_item UUID does not match fsid: %pU != %pU",
3978 fs_info->fsid, sb->dev_item.fsid);
3983 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3986 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
3987 btrfs_err(fs_info, "bytes_used is too small %llu",
3988 btrfs_super_bytes_used(sb));
3991 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
3992 btrfs_err(fs_info, "invalid stripesize %u",
3993 btrfs_super_stripesize(sb));
3996 if (btrfs_super_num_devices(sb) > (1UL << 31))
3997 btrfs_warn(fs_info, "suspicious number of devices: %llu",
3998 btrfs_super_num_devices(sb));
3999 if (btrfs_super_num_devices(sb) == 0) {
4000 btrfs_err(fs_info, "number of devices is 0");
4004 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4005 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4006 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4011 * Obvious sys_chunk_array corruptions, it must hold at least one key
4014 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4015 btrfs_err(fs_info, "system chunk array too big %u > %u",
4016 btrfs_super_sys_array_size(sb),
4017 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4020 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4021 + sizeof(struct btrfs_chunk)) {
4022 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4023 btrfs_super_sys_array_size(sb),
4024 sizeof(struct btrfs_disk_key)
4025 + sizeof(struct btrfs_chunk));
4030 * The generation is a global counter, we'll trust it more than the others
4031 * but it's still possible that it's the one that's wrong.
4033 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4035 "suspicious: generation < chunk_root_generation: %llu < %llu",
4036 btrfs_super_generation(sb),
4037 btrfs_super_chunk_root_generation(sb));
4038 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4039 && btrfs_super_cache_generation(sb) != (u64)-1)
4041 "suspicious: generation < cache_generation: %llu < %llu",
4042 btrfs_super_generation(sb),
4043 btrfs_super_cache_generation(sb));
4048 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4050 mutex_lock(&fs_info->cleaner_mutex);
4051 btrfs_run_delayed_iputs(fs_info);
4052 mutex_unlock(&fs_info->cleaner_mutex);
4054 down_write(&fs_info->cleanup_work_sem);
4055 up_write(&fs_info->cleanup_work_sem);
4057 /* cleanup FS via transaction */
4058 btrfs_cleanup_transaction(fs_info);
4061 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4063 struct btrfs_ordered_extent *ordered;
4065 spin_lock(&root->ordered_extent_lock);
4067 * This will just short circuit the ordered completion stuff which will
4068 * make sure the ordered extent gets properly cleaned up.
4070 list_for_each_entry(ordered, &root->ordered_extents,
4072 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4073 spin_unlock(&root->ordered_extent_lock);
4076 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4078 struct btrfs_root *root;
4079 struct list_head splice;
4081 INIT_LIST_HEAD(&splice);
4083 spin_lock(&fs_info->ordered_root_lock);
4084 list_splice_init(&fs_info->ordered_roots, &splice);
4085 while (!list_empty(&splice)) {
4086 root = list_first_entry(&splice, struct btrfs_root,
4088 list_move_tail(&root->ordered_root,
4089 &fs_info->ordered_roots);
4091 spin_unlock(&fs_info->ordered_root_lock);
4092 btrfs_destroy_ordered_extents(root);
4095 spin_lock(&fs_info->ordered_root_lock);
4097 spin_unlock(&fs_info->ordered_root_lock);
4100 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4101 struct btrfs_fs_info *fs_info)
4103 struct rb_node *node;
4104 struct btrfs_delayed_ref_root *delayed_refs;
4105 struct btrfs_delayed_ref_node *ref;
4108 delayed_refs = &trans->delayed_refs;
4110 spin_lock(&delayed_refs->lock);
4111 if (atomic_read(&delayed_refs->num_entries) == 0) {
4112 spin_unlock(&delayed_refs->lock);
4113 btrfs_info(fs_info, "delayed_refs has NO entry");
4117 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4118 struct btrfs_delayed_ref_head *head;
4120 bool pin_bytes = false;
4122 head = rb_entry(node, struct btrfs_delayed_ref_head,
4124 if (!mutex_trylock(&head->mutex)) {
4125 refcount_inc(&head->refs);
4126 spin_unlock(&delayed_refs->lock);
4128 mutex_lock(&head->mutex);
4129 mutex_unlock(&head->mutex);
4130 btrfs_put_delayed_ref_head(head);
4131 spin_lock(&delayed_refs->lock);
4134 spin_lock(&head->lock);
4135 while ((n = rb_first(&head->ref_tree)) != NULL) {
4136 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4139 rb_erase(&ref->ref_node, &head->ref_tree);
4140 RB_CLEAR_NODE(&ref->ref_node);
4141 if (!list_empty(&ref->add_list))
4142 list_del(&ref->add_list);
4143 atomic_dec(&delayed_refs->num_entries);
4144 btrfs_put_delayed_ref(ref);
4146 if (head->must_insert_reserved)
4148 btrfs_free_delayed_extent_op(head->extent_op);
4149 delayed_refs->num_heads--;
4150 if (head->processing == 0)
4151 delayed_refs->num_heads_ready--;
4152 atomic_dec(&delayed_refs->num_entries);
4153 rb_erase(&head->href_node, &delayed_refs->href_root);
4154 RB_CLEAR_NODE(&head->href_node);
4155 spin_unlock(&head->lock);
4156 spin_unlock(&delayed_refs->lock);
4157 mutex_unlock(&head->mutex);
4160 btrfs_pin_extent(fs_info, head->bytenr,
4161 head->num_bytes, 1);
4162 btrfs_put_delayed_ref_head(head);
4164 spin_lock(&delayed_refs->lock);
4167 spin_unlock(&delayed_refs->lock);
4172 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4174 struct btrfs_inode *btrfs_inode;
4175 struct list_head splice;
4177 INIT_LIST_HEAD(&splice);
4179 spin_lock(&root->delalloc_lock);
4180 list_splice_init(&root->delalloc_inodes, &splice);
4182 while (!list_empty(&splice)) {
4183 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4186 list_del_init(&btrfs_inode->delalloc_inodes);
4187 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4188 &btrfs_inode->runtime_flags);
4189 spin_unlock(&root->delalloc_lock);
4191 btrfs_invalidate_inodes(btrfs_inode->root);
4193 spin_lock(&root->delalloc_lock);
4196 spin_unlock(&root->delalloc_lock);
4199 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4201 struct btrfs_root *root;
4202 struct list_head splice;
4204 INIT_LIST_HEAD(&splice);
4206 spin_lock(&fs_info->delalloc_root_lock);
4207 list_splice_init(&fs_info->delalloc_roots, &splice);
4208 while (!list_empty(&splice)) {
4209 root = list_first_entry(&splice, struct btrfs_root,
4211 list_del_init(&root->delalloc_root);
4212 root = btrfs_grab_fs_root(root);
4214 spin_unlock(&fs_info->delalloc_root_lock);
4216 btrfs_destroy_delalloc_inodes(root);
4217 btrfs_put_fs_root(root);
4219 spin_lock(&fs_info->delalloc_root_lock);
4221 spin_unlock(&fs_info->delalloc_root_lock);
4224 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4225 struct extent_io_tree *dirty_pages,
4229 struct extent_buffer *eb;
4234 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4239 clear_extent_bits(dirty_pages, start, end, mark);
4240 while (start <= end) {
4241 eb = find_extent_buffer(fs_info, start);
4242 start += fs_info->nodesize;
4245 wait_on_extent_buffer_writeback(eb);
4247 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4249 clear_extent_buffer_dirty(eb);
4250 free_extent_buffer_stale(eb);
4257 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4258 struct extent_io_tree *pinned_extents)
4260 struct extent_io_tree *unpin;
4266 unpin = pinned_extents;
4269 ret = find_first_extent_bit(unpin, 0, &start, &end,
4270 EXTENT_DIRTY, NULL);
4274 clear_extent_dirty(unpin, start, end);
4275 btrfs_error_unpin_extent_range(fs_info, start, end);
4280 if (unpin == &fs_info->freed_extents[0])
4281 unpin = &fs_info->freed_extents[1];
4283 unpin = &fs_info->freed_extents[0];
4291 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4293 struct inode *inode;
4295 inode = cache->io_ctl.inode;
4297 invalidate_inode_pages2(inode->i_mapping);
4298 BTRFS_I(inode)->generation = 0;
4299 cache->io_ctl.inode = NULL;
4302 btrfs_put_block_group(cache);
4305 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4306 struct btrfs_fs_info *fs_info)
4308 struct btrfs_block_group_cache *cache;
4310 spin_lock(&cur_trans->dirty_bgs_lock);
4311 while (!list_empty(&cur_trans->dirty_bgs)) {
4312 cache = list_first_entry(&cur_trans->dirty_bgs,
4313 struct btrfs_block_group_cache,
4316 btrfs_err(fs_info, "orphan block group dirty_bgs list");
4317 spin_unlock(&cur_trans->dirty_bgs_lock);
4321 if (!list_empty(&cache->io_list)) {
4322 spin_unlock(&cur_trans->dirty_bgs_lock);
4323 list_del_init(&cache->io_list);
4324 btrfs_cleanup_bg_io(cache);
4325 spin_lock(&cur_trans->dirty_bgs_lock);
4328 list_del_init(&cache->dirty_list);
4329 spin_lock(&cache->lock);
4330 cache->disk_cache_state = BTRFS_DC_ERROR;
4331 spin_unlock(&cache->lock);
4333 spin_unlock(&cur_trans->dirty_bgs_lock);
4334 btrfs_put_block_group(cache);
4335 spin_lock(&cur_trans->dirty_bgs_lock);
4337 spin_unlock(&cur_trans->dirty_bgs_lock);
4339 while (!list_empty(&cur_trans->io_bgs)) {
4340 cache = list_first_entry(&cur_trans->io_bgs,
4341 struct btrfs_block_group_cache,
4344 btrfs_err(fs_info, "orphan block group on io_bgs list");
4348 list_del_init(&cache->io_list);
4349 spin_lock(&cache->lock);
4350 cache->disk_cache_state = BTRFS_DC_ERROR;
4351 spin_unlock(&cache->lock);
4352 btrfs_cleanup_bg_io(cache);
4356 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4357 struct btrfs_fs_info *fs_info)
4359 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4360 ASSERT(list_empty(&cur_trans->dirty_bgs));
4361 ASSERT(list_empty(&cur_trans->io_bgs));
4363 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4365 cur_trans->state = TRANS_STATE_COMMIT_START;
4366 wake_up(&fs_info->transaction_blocked_wait);
4368 cur_trans->state = TRANS_STATE_UNBLOCKED;
4369 wake_up(&fs_info->transaction_wait);
4371 btrfs_destroy_delayed_inodes(fs_info);
4372 btrfs_assert_delayed_root_empty(fs_info);
4374 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4376 btrfs_destroy_pinned_extent(fs_info,
4377 fs_info->pinned_extents);
4379 cur_trans->state =TRANS_STATE_COMPLETED;
4380 wake_up(&cur_trans->commit_wait);
4383 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4385 struct btrfs_transaction *t;
4387 mutex_lock(&fs_info->transaction_kthread_mutex);
4389 spin_lock(&fs_info->trans_lock);
4390 while (!list_empty(&fs_info->trans_list)) {
4391 t = list_first_entry(&fs_info->trans_list,
4392 struct btrfs_transaction, list);
4393 if (t->state >= TRANS_STATE_COMMIT_START) {
4394 refcount_inc(&t->use_count);
4395 spin_unlock(&fs_info->trans_lock);
4396 btrfs_wait_for_commit(fs_info, t->transid);
4397 btrfs_put_transaction(t);
4398 spin_lock(&fs_info->trans_lock);
4401 if (t == fs_info->running_transaction) {
4402 t->state = TRANS_STATE_COMMIT_DOING;
4403 spin_unlock(&fs_info->trans_lock);
4405 * We wait for 0 num_writers since we don't hold a trans
4406 * handle open currently for this transaction.
4408 wait_event(t->writer_wait,
4409 atomic_read(&t->num_writers) == 0);
4411 spin_unlock(&fs_info->trans_lock);
4413 btrfs_cleanup_one_transaction(t, fs_info);
4415 spin_lock(&fs_info->trans_lock);
4416 if (t == fs_info->running_transaction)
4417 fs_info->running_transaction = NULL;
4418 list_del_init(&t->list);
4419 spin_unlock(&fs_info->trans_lock);
4421 btrfs_put_transaction(t);
4422 trace_btrfs_transaction_commit(fs_info->tree_root);
4423 spin_lock(&fs_info->trans_lock);
4425 spin_unlock(&fs_info->trans_lock);
4426 btrfs_destroy_all_ordered_extents(fs_info);
4427 btrfs_destroy_delayed_inodes(fs_info);
4428 btrfs_assert_delayed_root_empty(fs_info);
4429 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4430 btrfs_destroy_all_delalloc_inodes(fs_info);
4431 mutex_unlock(&fs_info->transaction_kthread_mutex);
4436 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4438 struct inode *inode = private_data;
4439 return btrfs_sb(inode->i_sb);
4442 static const struct extent_io_ops btree_extent_io_ops = {
4443 /* mandatory callbacks */
4444 .submit_bio_hook = btree_submit_bio_hook,
4445 .readpage_end_io_hook = btree_readpage_end_io_hook,
4446 /* note we're sharing with inode.c for the merge bio hook */
4447 .merge_bio_hook = btrfs_merge_bio_hook,
4448 .readpage_io_failed_hook = btree_io_failed_hook,
4449 .set_range_writeback = btrfs_set_range_writeback,
4450 .tree_fs_info = btree_fs_info,
4452 /* optional callbacks */