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 <linux/bpf.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "free-space-tree.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
53 #include "compression.h"
54 #include "tree-checker.h"
55 #include "ref-verify.h"
58 #include <asm/cpufeature.h>
61 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
62 BTRFS_HEADER_FLAG_RELOC |\
63 BTRFS_SUPER_FLAG_ERROR |\
64 BTRFS_SUPER_FLAG_SEEDING |\
65 BTRFS_SUPER_FLAG_METADUMP)
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 static 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);
291 unsigned long cur_len;
292 unsigned long offset = BTRFS_CSUM_SIZE;
294 unsigned long map_start;
295 unsigned long map_len;
298 unsigned long inline_result;
300 len = buf->len - offset;
302 err = map_private_extent_buffer(buf, offset, 32,
303 &kaddr, &map_start, &map_len);
306 cur_len = min(len, map_len - (offset - map_start));
307 crc = btrfs_csum_data(kaddr + offset - map_start,
312 if (csum_size > sizeof(inline_result)) {
313 result = kzalloc(csum_size, GFP_NOFS);
317 result = (char *)&inline_result;
320 btrfs_csum_final(crc, result);
323 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
326 memcpy(&found, result, csum_size);
328 read_extent_buffer(buf, &val, 0, csum_size);
329 btrfs_warn_rl(fs_info,
330 "%s checksum verify failed on %llu wanted %X found %X level %d",
331 fs_info->sb->s_id, buf->start,
332 val, found, btrfs_header_level(buf));
333 if (result != (char *)&inline_result)
338 write_extent_buffer(buf, result, 0, csum_size);
340 if (result != (char *)&inline_result)
346 * we can't consider a given block up to date unless the transid of the
347 * block matches the transid in the parent node's pointer. This is how we
348 * detect blocks that either didn't get written at all or got written
349 * in the wrong place.
351 static int verify_parent_transid(struct extent_io_tree *io_tree,
352 struct extent_buffer *eb, u64 parent_transid,
355 struct extent_state *cached_state = NULL;
357 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
359 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
366 btrfs_tree_read_lock(eb);
367 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
370 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
372 if (extent_buffer_uptodate(eb) &&
373 btrfs_header_generation(eb) == parent_transid) {
377 btrfs_err_rl(eb->fs_info,
378 "parent transid verify failed on %llu wanted %llu found %llu",
380 parent_transid, btrfs_header_generation(eb));
384 * Things reading via commit roots that don't have normal protection,
385 * like send, can have a really old block in cache that may point at a
386 * block that has been freed and re-allocated. So don't clear uptodate
387 * if we find an eb that is under IO (dirty/writeback) because we could
388 * end up reading in the stale data and then writing it back out and
389 * making everybody very sad.
391 if (!extent_buffer_under_io(eb))
392 clear_extent_buffer_uptodate(eb);
394 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
395 &cached_state, GFP_NOFS);
397 btrfs_tree_read_unlock_blocking(eb);
402 * Return 0 if the superblock checksum type matches the checksum value of that
403 * algorithm. Pass the raw disk superblock data.
405 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
408 struct btrfs_super_block *disk_sb =
409 (struct btrfs_super_block *)raw_disk_sb;
410 u16 csum_type = btrfs_super_csum_type(disk_sb);
413 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
415 const int csum_size = sizeof(crc);
416 char result[csum_size];
419 * The super_block structure does not span the whole
420 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
421 * is filled with zeros and is included in the checksum.
423 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
424 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
425 btrfs_csum_final(crc, result);
427 if (memcmp(raw_disk_sb, result, csum_size))
431 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
432 btrfs_err(fs_info, "unsupported checksum algorithm %u",
441 * helper to read a given tree block, doing retries as required when
442 * the checksums don't match and we have alternate mirrors to try.
444 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
445 struct extent_buffer *eb,
448 struct extent_io_tree *io_tree;
453 int failed_mirror = 0;
455 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
456 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
458 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
459 btree_get_extent, mirror_num);
461 if (!verify_parent_transid(io_tree, eb,
469 * This buffer's crc is fine, but its contents are corrupted, so
470 * there is no reason to read the other copies, they won't be
473 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
476 num_copies = btrfs_num_copies(fs_info,
481 if (!failed_mirror) {
483 failed_mirror = eb->read_mirror;
487 if (mirror_num == failed_mirror)
490 if (mirror_num > num_copies)
494 if (failed && !ret && failed_mirror)
495 repair_eb_io_failure(fs_info, eb, failed_mirror);
501 * checksum a dirty tree block before IO. This has extra checks to make sure
502 * we only fill in the checksum field in the first page of a multi-page block
505 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
507 u64 start = page_offset(page);
509 struct extent_buffer *eb;
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
515 found_start = btrfs_header_bytenr(eb);
517 * Please do not consolidate these warnings into a single if.
518 * It is useful to know what went wrong.
520 if (WARN_ON(found_start != start))
522 if (WARN_ON(!PageUptodate(page)))
525 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
526 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
528 return csum_tree_block(fs_info, eb, 0);
531 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
532 struct extent_buffer *eb)
534 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
535 u8 fsid[BTRFS_FSID_SIZE];
538 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
540 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
544 fs_devices = fs_devices->seed;
549 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
550 u64 phy_offset, struct page *page,
551 u64 start, u64 end, int mirror)
555 struct extent_buffer *eb;
556 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
557 struct btrfs_fs_info *fs_info = root->fs_info;
564 eb = (struct extent_buffer *)page->private;
566 /* the pending IO might have been the only thing that kept this buffer
567 * in memory. Make sure we have a ref for all this other checks
569 extent_buffer_get(eb);
571 reads_done = atomic_dec_and_test(&eb->io_pages);
575 eb->read_mirror = mirror;
576 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
581 found_start = btrfs_header_bytenr(eb);
582 if (found_start != eb->start) {
583 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
584 found_start, eb->start);
588 if (check_tree_block_fsid(fs_info, eb)) {
589 btrfs_err_rl(fs_info, "bad fsid on block %llu",
594 found_level = btrfs_header_level(eb);
595 if (found_level >= BTRFS_MAX_LEVEL) {
596 btrfs_err(fs_info, "bad tree block level %d",
597 (int)btrfs_header_level(eb));
602 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
605 ret = csum_tree_block(fs_info, eb, 1);
610 * If this is a leaf block and it is corrupt, set the corrupt bit so
611 * that we don't try and read the other copies of this block, just
614 if (found_level == 0 && btrfs_check_leaf_full(root, eb)) {
615 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
619 if (found_level > 0 && btrfs_check_node(root, eb))
623 set_extent_buffer_uptodate(eb);
626 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
627 btree_readahead_hook(eb, ret);
631 * our io error hook is going to dec the io pages
632 * again, we have to make sure it has something
635 atomic_inc(&eb->io_pages);
636 clear_extent_buffer_uptodate(eb);
638 free_extent_buffer(eb);
643 static int btree_io_failed_hook(struct page *page, int failed_mirror)
645 struct extent_buffer *eb;
647 eb = (struct extent_buffer *)page->private;
648 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
649 eb->read_mirror = failed_mirror;
650 atomic_dec(&eb->io_pages);
651 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
652 btree_readahead_hook(eb, -EIO);
653 return -EIO; /* we fixed nothing */
656 static void end_workqueue_bio(struct bio *bio)
658 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
659 struct btrfs_fs_info *fs_info;
660 struct btrfs_workqueue *wq;
661 btrfs_work_func_t func;
663 fs_info = end_io_wq->info;
664 end_io_wq->status = bio->bi_status;
666 if (bio_op(bio) == REQ_OP_WRITE) {
667 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
668 wq = fs_info->endio_meta_write_workers;
669 func = btrfs_endio_meta_write_helper;
670 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
671 wq = fs_info->endio_freespace_worker;
672 func = btrfs_freespace_write_helper;
673 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
674 wq = fs_info->endio_raid56_workers;
675 func = btrfs_endio_raid56_helper;
677 wq = fs_info->endio_write_workers;
678 func = btrfs_endio_write_helper;
681 if (unlikely(end_io_wq->metadata ==
682 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
683 wq = fs_info->endio_repair_workers;
684 func = btrfs_endio_repair_helper;
685 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
686 wq = fs_info->endio_raid56_workers;
687 func = btrfs_endio_raid56_helper;
688 } else if (end_io_wq->metadata) {
689 wq = fs_info->endio_meta_workers;
690 func = btrfs_endio_meta_helper;
692 wq = fs_info->endio_workers;
693 func = btrfs_endio_helper;
697 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
698 btrfs_queue_work(wq, &end_io_wq->work);
701 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
702 enum btrfs_wq_endio_type metadata)
704 struct btrfs_end_io_wq *end_io_wq;
706 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
708 return BLK_STS_RESOURCE;
710 end_io_wq->private = bio->bi_private;
711 end_io_wq->end_io = bio->bi_end_io;
712 end_io_wq->info = info;
713 end_io_wq->status = 0;
714 end_io_wq->bio = bio;
715 end_io_wq->metadata = metadata;
717 bio->bi_private = end_io_wq;
718 bio->bi_end_io = end_workqueue_bio;
722 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
724 unsigned long limit = min_t(unsigned long,
725 info->thread_pool_size,
726 info->fs_devices->open_devices);
730 static void run_one_async_start(struct btrfs_work *work)
732 struct async_submit_bio *async;
735 async = container_of(work, struct async_submit_bio, work);
736 ret = async->submit_bio_start(async->private_data, async->bio,
737 async->mirror_num, async->bio_flags,
743 static void run_one_async_done(struct btrfs_work *work)
745 struct async_submit_bio *async;
747 async = container_of(work, struct async_submit_bio, work);
749 /* If an error occurred we just want to clean up the bio and move on */
751 async->bio->bi_status = async->status;
752 bio_endio(async->bio);
756 async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
757 async->bio_flags, async->bio_offset);
760 static void run_one_async_free(struct btrfs_work *work)
762 struct async_submit_bio *async;
764 async = container_of(work, struct async_submit_bio, work);
768 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
769 int mirror_num, unsigned long bio_flags,
770 u64 bio_offset, void *private_data,
771 extent_submit_bio_hook_t *submit_bio_start,
772 extent_submit_bio_hook_t *submit_bio_done)
774 struct async_submit_bio *async;
776 async = kmalloc(sizeof(*async), GFP_NOFS);
778 return BLK_STS_RESOURCE;
780 async->private_data = private_data;
781 async->fs_info = fs_info;
783 async->mirror_num = mirror_num;
784 async->submit_bio_start = submit_bio_start;
785 async->submit_bio_done = submit_bio_done;
787 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
788 run_one_async_done, run_one_async_free);
790 async->bio_flags = bio_flags;
791 async->bio_offset = bio_offset;
795 if (op_is_sync(bio->bi_opf))
796 btrfs_set_work_high_priority(&async->work);
798 btrfs_queue_work(fs_info->workers, &async->work);
802 static blk_status_t btree_csum_one_bio(struct bio *bio)
804 struct bio_vec *bvec;
805 struct btrfs_root *root;
808 ASSERT(!bio_flagged(bio, BIO_CLONED));
809 bio_for_each_segment_all(bvec, bio, i) {
810 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
811 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
816 return errno_to_blk_status(ret);
819 static blk_status_t __btree_submit_bio_start(void *private_data, struct bio *bio,
820 int mirror_num, unsigned long bio_flags,
824 * when we're called for a write, we're already in the async
825 * submission context. Just jump into btrfs_map_bio
827 return btree_csum_one_bio(bio);
830 static blk_status_t __btree_submit_bio_done(void *private_data, struct bio *bio,
831 int mirror_num, unsigned long bio_flags,
834 struct inode *inode = private_data;
838 * when we're called for a write, we're already in the async
839 * submission context. Just jump into btrfs_map_bio
841 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
843 bio->bi_status = ret;
849 static int check_async_write(struct btrfs_inode *bi)
851 if (atomic_read(&bi->sync_writers))
854 if (static_cpu_has(X86_FEATURE_XMM4_2))
860 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
861 int mirror_num, unsigned long bio_flags,
864 struct inode *inode = private_data;
865 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
866 int async = check_async_write(BTRFS_I(inode));
869 if (bio_op(bio) != REQ_OP_WRITE) {
871 * called for a read, do the setup so that checksum validation
872 * can happen in the async kernel threads
874 ret = btrfs_bio_wq_end_io(fs_info, bio,
875 BTRFS_WQ_ENDIO_METADATA);
878 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
880 ret = btree_csum_one_bio(bio);
883 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
886 * kthread helpers are used to submit writes so that
887 * checksumming can happen in parallel across all CPUs
889 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
890 bio_offset, private_data,
891 __btree_submit_bio_start,
892 __btree_submit_bio_done);
900 bio->bi_status = ret;
905 #ifdef CONFIG_MIGRATION
906 static int btree_migratepage(struct address_space *mapping,
907 struct page *newpage, struct page *page,
908 enum migrate_mode mode)
911 * we can't safely write a btree page from here,
912 * we haven't done the locking hook
917 * Buffers may be managed in a filesystem specific way.
918 * We must have no buffers or drop them.
920 if (page_has_private(page) &&
921 !try_to_release_page(page, GFP_KERNEL))
923 return migrate_page(mapping, newpage, page, mode);
928 static int btree_writepages(struct address_space *mapping,
929 struct writeback_control *wbc)
931 struct btrfs_fs_info *fs_info;
934 if (wbc->sync_mode == WB_SYNC_NONE) {
936 if (wbc->for_kupdate)
939 fs_info = BTRFS_I(mapping->host)->root->fs_info;
940 /* this is a bit racy, but that's ok */
941 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
942 BTRFS_DIRTY_METADATA_THRESH);
946 return btree_write_cache_pages(mapping, wbc);
949 static int btree_readpage(struct file *file, struct page *page)
951 struct extent_io_tree *tree;
952 tree = &BTRFS_I(page->mapping->host)->io_tree;
953 return extent_read_full_page(tree, page, btree_get_extent, 0);
956 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
958 if (PageWriteback(page) || PageDirty(page))
961 return try_release_extent_buffer(page);
964 static void btree_invalidatepage(struct page *page, unsigned int offset,
967 struct extent_io_tree *tree;
968 tree = &BTRFS_I(page->mapping->host)->io_tree;
969 extent_invalidatepage(tree, page, offset);
970 btree_releasepage(page, GFP_NOFS);
971 if (PagePrivate(page)) {
972 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
973 "page private not zero on page %llu",
974 (unsigned long long)page_offset(page));
975 ClearPagePrivate(page);
976 set_page_private(page, 0);
981 static int btree_set_page_dirty(struct page *page)
984 struct extent_buffer *eb;
986 BUG_ON(!PagePrivate(page));
987 eb = (struct extent_buffer *)page->private;
989 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
990 BUG_ON(!atomic_read(&eb->refs));
991 btrfs_assert_tree_locked(eb);
993 return __set_page_dirty_nobuffers(page);
996 static const struct address_space_operations btree_aops = {
997 .readpage = btree_readpage,
998 .writepages = btree_writepages,
999 .releasepage = btree_releasepage,
1000 .invalidatepage = btree_invalidatepage,
1001 #ifdef CONFIG_MIGRATION
1002 .migratepage = btree_migratepage,
1004 .set_page_dirty = btree_set_page_dirty,
1007 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1009 struct extent_buffer *buf = NULL;
1010 struct inode *btree_inode = fs_info->btree_inode;
1012 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1015 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016 buf, WAIT_NONE, btree_get_extent, 0);
1017 free_extent_buffer(buf);
1020 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1021 int mirror_num, struct extent_buffer **eb)
1023 struct extent_buffer *buf = NULL;
1024 struct inode *btree_inode = fs_info->btree_inode;
1025 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1028 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1032 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1034 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1035 btree_get_extent, mirror_num);
1037 free_extent_buffer(buf);
1041 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1042 free_extent_buffer(buf);
1044 } else if (extent_buffer_uptodate(buf)) {
1047 free_extent_buffer(buf);
1052 struct extent_buffer *btrfs_find_create_tree_block(
1053 struct btrfs_fs_info *fs_info,
1056 if (btrfs_is_testing(fs_info))
1057 return alloc_test_extent_buffer(fs_info, bytenr);
1058 return alloc_extent_buffer(fs_info, bytenr);
1062 int btrfs_write_tree_block(struct extent_buffer *buf)
1064 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1065 buf->start + buf->len - 1);
1068 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1070 filemap_fdatawait_range(buf->pages[0]->mapping,
1071 buf->start, buf->start + buf->len - 1);
1074 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1077 struct extent_buffer *buf = NULL;
1080 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1084 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1086 free_extent_buffer(buf);
1087 return ERR_PTR(ret);
1093 void clean_tree_block(struct btrfs_fs_info *fs_info,
1094 struct extent_buffer *buf)
1096 if (btrfs_header_generation(buf) ==
1097 fs_info->running_transaction->transid) {
1098 btrfs_assert_tree_locked(buf);
1100 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1101 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1103 fs_info->dirty_metadata_batch);
1104 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1105 btrfs_set_lock_blocking(buf);
1106 clear_extent_buffer_dirty(buf);
1111 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1113 struct btrfs_subvolume_writers *writers;
1116 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1118 return ERR_PTR(-ENOMEM);
1120 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1123 return ERR_PTR(ret);
1126 init_waitqueue_head(&writers->wait);
1131 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1133 percpu_counter_destroy(&writers->counter);
1137 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1140 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1142 root->commit_root = NULL;
1144 root->orphan_cleanup_state = 0;
1146 root->objectid = objectid;
1147 root->last_trans = 0;
1148 root->highest_objectid = 0;
1149 root->nr_delalloc_inodes = 0;
1150 root->nr_ordered_extents = 0;
1152 root->inode_tree = RB_ROOT;
1153 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1154 root->block_rsv = NULL;
1155 root->orphan_block_rsv = NULL;
1157 INIT_LIST_HEAD(&root->dirty_list);
1158 INIT_LIST_HEAD(&root->root_list);
1159 INIT_LIST_HEAD(&root->delalloc_inodes);
1160 INIT_LIST_HEAD(&root->delalloc_root);
1161 INIT_LIST_HEAD(&root->ordered_extents);
1162 INIT_LIST_HEAD(&root->ordered_root);
1163 INIT_LIST_HEAD(&root->logged_list[0]);
1164 INIT_LIST_HEAD(&root->logged_list[1]);
1165 spin_lock_init(&root->orphan_lock);
1166 spin_lock_init(&root->inode_lock);
1167 spin_lock_init(&root->delalloc_lock);
1168 spin_lock_init(&root->ordered_extent_lock);
1169 spin_lock_init(&root->accounting_lock);
1170 spin_lock_init(&root->log_extents_lock[0]);
1171 spin_lock_init(&root->log_extents_lock[1]);
1172 mutex_init(&root->objectid_mutex);
1173 mutex_init(&root->log_mutex);
1174 mutex_init(&root->ordered_extent_mutex);
1175 mutex_init(&root->delalloc_mutex);
1176 init_waitqueue_head(&root->log_writer_wait);
1177 init_waitqueue_head(&root->log_commit_wait[0]);
1178 init_waitqueue_head(&root->log_commit_wait[1]);
1179 INIT_LIST_HEAD(&root->log_ctxs[0]);
1180 INIT_LIST_HEAD(&root->log_ctxs[1]);
1181 atomic_set(&root->log_commit[0], 0);
1182 atomic_set(&root->log_commit[1], 0);
1183 atomic_set(&root->log_writers, 0);
1184 atomic_set(&root->log_batch, 0);
1185 atomic_set(&root->orphan_inodes, 0);
1186 refcount_set(&root->refs, 1);
1187 atomic_set(&root->will_be_snapshotted, 0);
1188 atomic64_set(&root->qgroup_meta_rsv, 0);
1189 root->log_transid = 0;
1190 root->log_transid_committed = -1;
1191 root->last_log_commit = 0;
1193 extent_io_tree_init(&root->dirty_log_pages, NULL);
1195 memset(&root->root_key, 0, sizeof(root->root_key));
1196 memset(&root->root_item, 0, sizeof(root->root_item));
1197 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1199 root->defrag_trans_start = fs_info->generation;
1201 root->defrag_trans_start = 0;
1202 root->root_key.objectid = objectid;
1205 spin_lock_init(&root->root_item_lock);
1208 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1211 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1213 root->fs_info = fs_info;
1217 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1218 /* Should only be used by the testing infrastructure */
1219 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1221 struct btrfs_root *root;
1224 return ERR_PTR(-EINVAL);
1226 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1228 return ERR_PTR(-ENOMEM);
1230 /* We don't use the stripesize in selftest, set it as sectorsize */
1231 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1232 root->alloc_bytenr = 0;
1238 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1239 struct btrfs_fs_info *fs_info,
1242 struct extent_buffer *leaf;
1243 struct btrfs_root *tree_root = fs_info->tree_root;
1244 struct btrfs_root *root;
1245 struct btrfs_key key;
1249 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1251 return ERR_PTR(-ENOMEM);
1253 __setup_root(root, fs_info, objectid);
1254 root->root_key.objectid = objectid;
1255 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1256 root->root_key.offset = 0;
1258 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1260 ret = PTR_ERR(leaf);
1265 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1266 btrfs_set_header_bytenr(leaf, leaf->start);
1267 btrfs_set_header_generation(leaf, trans->transid);
1268 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1269 btrfs_set_header_owner(leaf, objectid);
1272 write_extent_buffer_fsid(leaf, fs_info->fsid);
1273 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1274 btrfs_mark_buffer_dirty(leaf);
1276 root->commit_root = btrfs_root_node(root);
1277 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1279 root->root_item.flags = 0;
1280 root->root_item.byte_limit = 0;
1281 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1282 btrfs_set_root_generation(&root->root_item, trans->transid);
1283 btrfs_set_root_level(&root->root_item, 0);
1284 btrfs_set_root_refs(&root->root_item, 1);
1285 btrfs_set_root_used(&root->root_item, leaf->len);
1286 btrfs_set_root_last_snapshot(&root->root_item, 0);
1287 btrfs_set_root_dirid(&root->root_item, 0);
1289 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1290 root->root_item.drop_level = 0;
1292 key.objectid = objectid;
1293 key.type = BTRFS_ROOT_ITEM_KEY;
1295 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1299 btrfs_tree_unlock(leaf);
1305 btrfs_tree_unlock(leaf);
1306 free_extent_buffer(root->commit_root);
1307 free_extent_buffer(leaf);
1311 return ERR_PTR(ret);
1314 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1315 struct btrfs_fs_info *fs_info)
1317 struct btrfs_root *root;
1318 struct extent_buffer *leaf;
1320 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1322 return ERR_PTR(-ENOMEM);
1324 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1326 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1327 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1328 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1331 * DON'T set REF_COWS for log trees
1333 * log trees do not get reference counted because they go away
1334 * before a real commit is actually done. They do store pointers
1335 * to file data extents, and those reference counts still get
1336 * updated (along with back refs to the log tree).
1339 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1343 return ERR_CAST(leaf);
1346 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1347 btrfs_set_header_bytenr(leaf, leaf->start);
1348 btrfs_set_header_generation(leaf, trans->transid);
1349 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1350 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1353 write_extent_buffer_fsid(root->node, fs_info->fsid);
1354 btrfs_mark_buffer_dirty(root->node);
1355 btrfs_tree_unlock(root->node);
1359 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1360 struct btrfs_fs_info *fs_info)
1362 struct btrfs_root *log_root;
1364 log_root = alloc_log_tree(trans, fs_info);
1365 if (IS_ERR(log_root))
1366 return PTR_ERR(log_root);
1367 WARN_ON(fs_info->log_root_tree);
1368 fs_info->log_root_tree = log_root;
1372 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1373 struct btrfs_root *root)
1375 struct btrfs_fs_info *fs_info = root->fs_info;
1376 struct btrfs_root *log_root;
1377 struct btrfs_inode_item *inode_item;
1379 log_root = alloc_log_tree(trans, fs_info);
1380 if (IS_ERR(log_root))
1381 return PTR_ERR(log_root);
1383 log_root->last_trans = trans->transid;
1384 log_root->root_key.offset = root->root_key.objectid;
1386 inode_item = &log_root->root_item.inode;
1387 btrfs_set_stack_inode_generation(inode_item, 1);
1388 btrfs_set_stack_inode_size(inode_item, 3);
1389 btrfs_set_stack_inode_nlink(inode_item, 1);
1390 btrfs_set_stack_inode_nbytes(inode_item,
1392 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1394 btrfs_set_root_node(&log_root->root_item, log_root->node);
1396 WARN_ON(root->log_root);
1397 root->log_root = log_root;
1398 root->log_transid = 0;
1399 root->log_transid_committed = -1;
1400 root->last_log_commit = 0;
1404 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1405 struct btrfs_key *key)
1407 struct btrfs_root *root;
1408 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1409 struct btrfs_path *path;
1413 path = btrfs_alloc_path();
1415 return ERR_PTR(-ENOMEM);
1417 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1423 __setup_root(root, fs_info, key->objectid);
1425 ret = btrfs_find_root(tree_root, key, path,
1426 &root->root_item, &root->root_key);
1433 generation = btrfs_root_generation(&root->root_item);
1434 root->node = read_tree_block(fs_info,
1435 btrfs_root_bytenr(&root->root_item),
1437 if (IS_ERR(root->node)) {
1438 ret = PTR_ERR(root->node);
1440 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1442 free_extent_buffer(root->node);
1445 root->commit_root = btrfs_root_node(root);
1447 btrfs_free_path(path);
1453 root = ERR_PTR(ret);
1457 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1458 struct btrfs_key *location)
1460 struct btrfs_root *root;
1462 root = btrfs_read_tree_root(tree_root, location);
1466 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1467 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1468 btrfs_check_and_init_root_item(&root->root_item);
1474 int btrfs_init_fs_root(struct btrfs_root *root)
1477 struct btrfs_subvolume_writers *writers;
1479 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1480 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1482 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1487 writers = btrfs_alloc_subvolume_writers();
1488 if (IS_ERR(writers)) {
1489 ret = PTR_ERR(writers);
1492 root->subv_writers = writers;
1494 btrfs_init_free_ino_ctl(root);
1495 spin_lock_init(&root->ino_cache_lock);
1496 init_waitqueue_head(&root->ino_cache_wait);
1498 ret = get_anon_bdev(&root->anon_dev);
1502 mutex_lock(&root->objectid_mutex);
1503 ret = btrfs_find_highest_objectid(root,
1504 &root->highest_objectid);
1506 mutex_unlock(&root->objectid_mutex);
1510 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1512 mutex_unlock(&root->objectid_mutex);
1516 /* the caller is responsible to call free_fs_root */
1520 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1523 struct btrfs_root *root;
1525 spin_lock(&fs_info->fs_roots_radix_lock);
1526 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1527 (unsigned long)root_id);
1528 spin_unlock(&fs_info->fs_roots_radix_lock);
1532 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1533 struct btrfs_root *root)
1537 ret = radix_tree_preload(GFP_NOFS);
1541 spin_lock(&fs_info->fs_roots_radix_lock);
1542 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1543 (unsigned long)root->root_key.objectid,
1546 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1547 spin_unlock(&fs_info->fs_roots_radix_lock);
1548 radix_tree_preload_end();
1553 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1554 struct btrfs_key *location,
1557 struct btrfs_root *root;
1558 struct btrfs_path *path;
1559 struct btrfs_key key;
1562 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1563 return fs_info->tree_root;
1564 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1565 return fs_info->extent_root;
1566 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1567 return fs_info->chunk_root;
1568 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1569 return fs_info->dev_root;
1570 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1571 return fs_info->csum_root;
1572 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1573 return fs_info->quota_root ? fs_info->quota_root :
1575 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1576 return fs_info->uuid_root ? fs_info->uuid_root :
1578 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1579 return fs_info->free_space_root ? fs_info->free_space_root :
1582 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1584 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1585 return ERR_PTR(-ENOENT);
1589 root = btrfs_read_fs_root(fs_info->tree_root, location);
1593 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1598 ret = btrfs_init_fs_root(root);
1602 path = btrfs_alloc_path();
1607 key.objectid = BTRFS_ORPHAN_OBJECTID;
1608 key.type = BTRFS_ORPHAN_ITEM_KEY;
1609 key.offset = location->objectid;
1611 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1612 btrfs_free_path(path);
1616 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1618 ret = btrfs_insert_fs_root(fs_info, root);
1620 if (ret == -EEXIST) {
1629 return ERR_PTR(ret);
1632 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1634 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1636 struct btrfs_device *device;
1637 struct backing_dev_info *bdi;
1640 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1643 bdi = device->bdev->bd_bdi;
1644 if (bdi_congested(bdi, bdi_bits)) {
1654 * called by the kthread helper functions to finally call the bio end_io
1655 * functions. This is where read checksum verification actually happens
1657 static void end_workqueue_fn(struct btrfs_work *work)
1660 struct btrfs_end_io_wq *end_io_wq;
1662 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1663 bio = end_io_wq->bio;
1665 bio->bi_status = end_io_wq->status;
1666 bio->bi_private = end_io_wq->private;
1667 bio->bi_end_io = end_io_wq->end_io;
1668 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1672 static int cleaner_kthread(void *arg)
1674 struct btrfs_root *root = arg;
1675 struct btrfs_fs_info *fs_info = root->fs_info;
1677 struct btrfs_trans_handle *trans;
1682 /* Make the cleaner go to sleep early. */
1683 if (btrfs_need_cleaner_sleep(fs_info))
1687 * Do not do anything if we might cause open_ctree() to block
1688 * before we have finished mounting the filesystem.
1690 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1693 if (!mutex_trylock(&fs_info->cleaner_mutex))
1697 * Avoid the problem that we change the status of the fs
1698 * during the above check and trylock.
1700 if (btrfs_need_cleaner_sleep(fs_info)) {
1701 mutex_unlock(&fs_info->cleaner_mutex);
1705 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1706 btrfs_run_delayed_iputs(fs_info);
1707 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1709 again = btrfs_clean_one_deleted_snapshot(root);
1710 mutex_unlock(&fs_info->cleaner_mutex);
1713 * The defragger has dealt with the R/O remount and umount,
1714 * needn't do anything special here.
1716 btrfs_run_defrag_inodes(fs_info);
1719 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1720 * with relocation (btrfs_relocate_chunk) and relocation
1721 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1722 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1723 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1724 * unused block groups.
1726 btrfs_delete_unused_bgs(fs_info);
1729 set_current_state(TASK_INTERRUPTIBLE);
1730 if (!kthread_should_stop())
1732 __set_current_state(TASK_RUNNING);
1734 } while (!kthread_should_stop());
1737 * Transaction kthread is stopped before us and wakes us up.
1738 * However we might have started a new transaction and COWed some
1739 * tree blocks when deleting unused block groups for example. So
1740 * make sure we commit the transaction we started to have a clean
1741 * shutdown when evicting the btree inode - if it has dirty pages
1742 * when we do the final iput() on it, eviction will trigger a
1743 * writeback for it which will fail with null pointer dereferences
1744 * since work queues and other resources were already released and
1745 * destroyed by the time the iput/eviction/writeback is made.
1747 trans = btrfs_attach_transaction(root);
1748 if (IS_ERR(trans)) {
1749 if (PTR_ERR(trans) != -ENOENT)
1751 "cleaner transaction attach returned %ld",
1756 ret = btrfs_commit_transaction(trans);
1759 "cleaner open transaction commit returned %d",
1766 static int transaction_kthread(void *arg)
1768 struct btrfs_root *root = arg;
1769 struct btrfs_fs_info *fs_info = root->fs_info;
1770 struct btrfs_trans_handle *trans;
1771 struct btrfs_transaction *cur;
1774 unsigned long delay;
1778 cannot_commit = false;
1779 delay = HZ * fs_info->commit_interval;
1780 mutex_lock(&fs_info->transaction_kthread_mutex);
1782 spin_lock(&fs_info->trans_lock);
1783 cur = fs_info->running_transaction;
1785 spin_unlock(&fs_info->trans_lock);
1789 now = get_seconds();
1790 if (cur->state < TRANS_STATE_BLOCKED &&
1791 (now < cur->start_time ||
1792 now - cur->start_time < fs_info->commit_interval)) {
1793 spin_unlock(&fs_info->trans_lock);
1797 transid = cur->transid;
1798 spin_unlock(&fs_info->trans_lock);
1800 /* If the file system is aborted, this will always fail. */
1801 trans = btrfs_attach_transaction(root);
1802 if (IS_ERR(trans)) {
1803 if (PTR_ERR(trans) != -ENOENT)
1804 cannot_commit = true;
1807 if (transid == trans->transid) {
1808 btrfs_commit_transaction(trans);
1810 btrfs_end_transaction(trans);
1813 wake_up_process(fs_info->cleaner_kthread);
1814 mutex_unlock(&fs_info->transaction_kthread_mutex);
1816 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1817 &fs_info->fs_state)))
1818 btrfs_cleanup_transaction(fs_info);
1819 set_current_state(TASK_INTERRUPTIBLE);
1820 if (!kthread_should_stop() &&
1821 (!btrfs_transaction_blocked(fs_info) ||
1823 schedule_timeout(delay);
1824 __set_current_state(TASK_RUNNING);
1825 } while (!kthread_should_stop());
1830 * this will find the highest generation in the array of
1831 * root backups. The index of the highest array is returned,
1832 * or -1 if we can't find anything.
1834 * We check to make sure the array is valid by comparing the
1835 * generation of the latest root in the array with the generation
1836 * in the super block. If they don't match we pitch it.
1838 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1841 int newest_index = -1;
1842 struct btrfs_root_backup *root_backup;
1845 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1846 root_backup = info->super_copy->super_roots + i;
1847 cur = btrfs_backup_tree_root_gen(root_backup);
1848 if (cur == newest_gen)
1852 /* check to see if we actually wrapped around */
1853 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1854 root_backup = info->super_copy->super_roots;
1855 cur = btrfs_backup_tree_root_gen(root_backup);
1856 if (cur == newest_gen)
1859 return newest_index;
1864 * find the oldest backup so we know where to store new entries
1865 * in the backup array. This will set the backup_root_index
1866 * field in the fs_info struct
1868 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1871 int newest_index = -1;
1873 newest_index = find_newest_super_backup(info, newest_gen);
1874 /* if there was garbage in there, just move along */
1875 if (newest_index == -1) {
1876 info->backup_root_index = 0;
1878 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1883 * copy all the root pointers into the super backup array.
1884 * this will bump the backup pointer by one when it is
1887 static void backup_super_roots(struct btrfs_fs_info *info)
1890 struct btrfs_root_backup *root_backup;
1893 next_backup = info->backup_root_index;
1894 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1895 BTRFS_NUM_BACKUP_ROOTS;
1898 * just overwrite the last backup if we're at the same generation
1899 * this happens only at umount
1901 root_backup = info->super_for_commit->super_roots + last_backup;
1902 if (btrfs_backup_tree_root_gen(root_backup) ==
1903 btrfs_header_generation(info->tree_root->node))
1904 next_backup = last_backup;
1906 root_backup = info->super_for_commit->super_roots + next_backup;
1909 * make sure all of our padding and empty slots get zero filled
1910 * regardless of which ones we use today
1912 memset(root_backup, 0, sizeof(*root_backup));
1914 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1916 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1917 btrfs_set_backup_tree_root_gen(root_backup,
1918 btrfs_header_generation(info->tree_root->node));
1920 btrfs_set_backup_tree_root_level(root_backup,
1921 btrfs_header_level(info->tree_root->node));
1923 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1924 btrfs_set_backup_chunk_root_gen(root_backup,
1925 btrfs_header_generation(info->chunk_root->node));
1926 btrfs_set_backup_chunk_root_level(root_backup,
1927 btrfs_header_level(info->chunk_root->node));
1929 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1930 btrfs_set_backup_extent_root_gen(root_backup,
1931 btrfs_header_generation(info->extent_root->node));
1932 btrfs_set_backup_extent_root_level(root_backup,
1933 btrfs_header_level(info->extent_root->node));
1936 * we might commit during log recovery, which happens before we set
1937 * the fs_root. Make sure it is valid before we fill it in.
1939 if (info->fs_root && info->fs_root->node) {
1940 btrfs_set_backup_fs_root(root_backup,
1941 info->fs_root->node->start);
1942 btrfs_set_backup_fs_root_gen(root_backup,
1943 btrfs_header_generation(info->fs_root->node));
1944 btrfs_set_backup_fs_root_level(root_backup,
1945 btrfs_header_level(info->fs_root->node));
1948 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1949 btrfs_set_backup_dev_root_gen(root_backup,
1950 btrfs_header_generation(info->dev_root->node));
1951 btrfs_set_backup_dev_root_level(root_backup,
1952 btrfs_header_level(info->dev_root->node));
1954 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1955 btrfs_set_backup_csum_root_gen(root_backup,
1956 btrfs_header_generation(info->csum_root->node));
1957 btrfs_set_backup_csum_root_level(root_backup,
1958 btrfs_header_level(info->csum_root->node));
1960 btrfs_set_backup_total_bytes(root_backup,
1961 btrfs_super_total_bytes(info->super_copy));
1962 btrfs_set_backup_bytes_used(root_backup,
1963 btrfs_super_bytes_used(info->super_copy));
1964 btrfs_set_backup_num_devices(root_backup,
1965 btrfs_super_num_devices(info->super_copy));
1968 * if we don't copy this out to the super_copy, it won't get remembered
1969 * for the next commit
1971 memcpy(&info->super_copy->super_roots,
1972 &info->super_for_commit->super_roots,
1973 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1977 * this copies info out of the root backup array and back into
1978 * the in-memory super block. It is meant to help iterate through
1979 * the array, so you send it the number of backups you've already
1980 * tried and the last backup index you used.
1982 * this returns -1 when it has tried all the backups
1984 static noinline int next_root_backup(struct btrfs_fs_info *info,
1985 struct btrfs_super_block *super,
1986 int *num_backups_tried, int *backup_index)
1988 struct btrfs_root_backup *root_backup;
1989 int newest = *backup_index;
1991 if (*num_backups_tried == 0) {
1992 u64 gen = btrfs_super_generation(super);
1994 newest = find_newest_super_backup(info, gen);
1998 *backup_index = newest;
1999 *num_backups_tried = 1;
2000 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2001 /* we've tried all the backups, all done */
2004 /* jump to the next oldest backup */
2005 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2006 BTRFS_NUM_BACKUP_ROOTS;
2007 *backup_index = newest;
2008 *num_backups_tried += 1;
2010 root_backup = super->super_roots + newest;
2012 btrfs_set_super_generation(super,
2013 btrfs_backup_tree_root_gen(root_backup));
2014 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2015 btrfs_set_super_root_level(super,
2016 btrfs_backup_tree_root_level(root_backup));
2017 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2020 * fixme: the total bytes and num_devices need to match or we should
2023 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2024 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2028 /* helper to cleanup workers */
2029 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2031 btrfs_destroy_workqueue(fs_info->fixup_workers);
2032 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2033 btrfs_destroy_workqueue(fs_info->workers);
2034 btrfs_destroy_workqueue(fs_info->endio_workers);
2035 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2036 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2037 btrfs_destroy_workqueue(fs_info->rmw_workers);
2038 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2039 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2040 btrfs_destroy_workqueue(fs_info->submit_workers);
2041 btrfs_destroy_workqueue(fs_info->delayed_workers);
2042 btrfs_destroy_workqueue(fs_info->caching_workers);
2043 btrfs_destroy_workqueue(fs_info->readahead_workers);
2044 btrfs_destroy_workqueue(fs_info->flush_workers);
2045 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2046 btrfs_destroy_workqueue(fs_info->extent_workers);
2048 * Now that all other work queues are destroyed, we can safely destroy
2049 * the queues used for metadata I/O, since tasks from those other work
2050 * queues can do metadata I/O operations.
2052 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2053 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2056 static void free_root_extent_buffers(struct btrfs_root *root)
2059 free_extent_buffer(root->node);
2060 free_extent_buffer(root->commit_root);
2062 root->commit_root = NULL;
2066 /* helper to cleanup tree roots */
2067 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2069 free_root_extent_buffers(info->tree_root);
2071 free_root_extent_buffers(info->dev_root);
2072 free_root_extent_buffers(info->extent_root);
2073 free_root_extent_buffers(info->csum_root);
2074 free_root_extent_buffers(info->quota_root);
2075 free_root_extent_buffers(info->uuid_root);
2077 free_root_extent_buffers(info->chunk_root);
2078 free_root_extent_buffers(info->free_space_root);
2081 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2084 struct btrfs_root *gang[8];
2087 while (!list_empty(&fs_info->dead_roots)) {
2088 gang[0] = list_entry(fs_info->dead_roots.next,
2089 struct btrfs_root, root_list);
2090 list_del(&gang[0]->root_list);
2092 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2093 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2095 free_extent_buffer(gang[0]->node);
2096 free_extent_buffer(gang[0]->commit_root);
2097 btrfs_put_fs_root(gang[0]);
2102 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2107 for (i = 0; i < ret; i++)
2108 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2111 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2112 btrfs_free_log_root_tree(NULL, fs_info);
2113 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2117 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2119 mutex_init(&fs_info->scrub_lock);
2120 atomic_set(&fs_info->scrubs_running, 0);
2121 atomic_set(&fs_info->scrub_pause_req, 0);
2122 atomic_set(&fs_info->scrubs_paused, 0);
2123 atomic_set(&fs_info->scrub_cancel_req, 0);
2124 init_waitqueue_head(&fs_info->scrub_pause_wait);
2125 fs_info->scrub_workers_refcnt = 0;
2128 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2130 spin_lock_init(&fs_info->balance_lock);
2131 mutex_init(&fs_info->balance_mutex);
2132 atomic_set(&fs_info->balance_running, 0);
2133 atomic_set(&fs_info->balance_pause_req, 0);
2134 atomic_set(&fs_info->balance_cancel_req, 0);
2135 fs_info->balance_ctl = NULL;
2136 init_waitqueue_head(&fs_info->balance_wait_q);
2139 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2141 struct inode *inode = fs_info->btree_inode;
2143 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2144 set_nlink(inode, 1);
2146 * we set the i_size on the btree inode to the max possible int.
2147 * the real end of the address space is determined by all of
2148 * the devices in the system
2150 inode->i_size = OFFSET_MAX;
2151 inode->i_mapping->a_ops = &btree_aops;
2153 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2154 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2155 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2156 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2158 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2160 BTRFS_I(inode)->root = fs_info->tree_root;
2161 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2162 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2163 btrfs_insert_inode_hash(inode);
2166 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2168 fs_info->dev_replace.lock_owner = 0;
2169 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2170 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2171 rwlock_init(&fs_info->dev_replace.lock);
2172 atomic_set(&fs_info->dev_replace.read_locks, 0);
2173 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2174 init_waitqueue_head(&fs_info->replace_wait);
2175 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2178 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2180 spin_lock_init(&fs_info->qgroup_lock);
2181 mutex_init(&fs_info->qgroup_ioctl_lock);
2182 fs_info->qgroup_tree = RB_ROOT;
2183 fs_info->qgroup_op_tree = RB_ROOT;
2184 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2185 fs_info->qgroup_seq = 1;
2186 fs_info->qgroup_ulist = NULL;
2187 fs_info->qgroup_rescan_running = false;
2188 mutex_init(&fs_info->qgroup_rescan_lock);
2191 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2192 struct btrfs_fs_devices *fs_devices)
2194 int max_active = fs_info->thread_pool_size;
2195 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2198 btrfs_alloc_workqueue(fs_info, "worker",
2199 flags | WQ_HIGHPRI, max_active, 16);
2201 fs_info->delalloc_workers =
2202 btrfs_alloc_workqueue(fs_info, "delalloc",
2203 flags, max_active, 2);
2205 fs_info->flush_workers =
2206 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2207 flags, max_active, 0);
2209 fs_info->caching_workers =
2210 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2213 * a higher idle thresh on the submit workers makes it much more
2214 * likely that bios will be send down in a sane order to the
2217 fs_info->submit_workers =
2218 btrfs_alloc_workqueue(fs_info, "submit", flags,
2219 min_t(u64, fs_devices->num_devices,
2222 fs_info->fixup_workers =
2223 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2226 * endios are largely parallel and should have a very
2229 fs_info->endio_workers =
2230 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2231 fs_info->endio_meta_workers =
2232 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2234 fs_info->endio_meta_write_workers =
2235 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2237 fs_info->endio_raid56_workers =
2238 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2240 fs_info->endio_repair_workers =
2241 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2242 fs_info->rmw_workers =
2243 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2244 fs_info->endio_write_workers =
2245 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2247 fs_info->endio_freespace_worker =
2248 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2250 fs_info->delayed_workers =
2251 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2253 fs_info->readahead_workers =
2254 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2256 fs_info->qgroup_rescan_workers =
2257 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2258 fs_info->extent_workers =
2259 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2260 min_t(u64, fs_devices->num_devices,
2263 if (!(fs_info->workers && fs_info->delalloc_workers &&
2264 fs_info->submit_workers && fs_info->flush_workers &&
2265 fs_info->endio_workers && fs_info->endio_meta_workers &&
2266 fs_info->endio_meta_write_workers &&
2267 fs_info->endio_repair_workers &&
2268 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2269 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2270 fs_info->caching_workers && fs_info->readahead_workers &&
2271 fs_info->fixup_workers && fs_info->delayed_workers &&
2272 fs_info->extent_workers &&
2273 fs_info->qgroup_rescan_workers)) {
2280 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2281 struct btrfs_fs_devices *fs_devices)
2284 struct btrfs_root *log_tree_root;
2285 struct btrfs_super_block *disk_super = fs_info->super_copy;
2286 u64 bytenr = btrfs_super_log_root(disk_super);
2288 if (fs_devices->rw_devices == 0) {
2289 btrfs_warn(fs_info, "log replay required on RO media");
2293 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2297 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2299 log_tree_root->node = read_tree_block(fs_info, bytenr,
2300 fs_info->generation + 1);
2301 if (IS_ERR(log_tree_root->node)) {
2302 btrfs_warn(fs_info, "failed to read log tree");
2303 ret = PTR_ERR(log_tree_root->node);
2304 kfree(log_tree_root);
2306 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2307 btrfs_err(fs_info, "failed to read log tree");
2308 free_extent_buffer(log_tree_root->node);
2309 kfree(log_tree_root);
2312 /* returns with log_tree_root freed on success */
2313 ret = btrfs_recover_log_trees(log_tree_root);
2315 btrfs_handle_fs_error(fs_info, ret,
2316 "Failed to recover log tree");
2317 free_extent_buffer(log_tree_root->node);
2318 kfree(log_tree_root);
2322 if (sb_rdonly(fs_info->sb)) {
2323 ret = btrfs_commit_super(fs_info);
2331 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2333 struct btrfs_root *tree_root = fs_info->tree_root;
2334 struct btrfs_root *root;
2335 struct btrfs_key location;
2338 BUG_ON(!fs_info->tree_root);
2340 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2341 location.type = BTRFS_ROOT_ITEM_KEY;
2342 location.offset = 0;
2344 root = btrfs_read_tree_root(tree_root, &location);
2346 return PTR_ERR(root);
2347 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2348 fs_info->extent_root = root;
2350 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2351 root = btrfs_read_tree_root(tree_root, &location);
2353 return PTR_ERR(root);
2354 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2355 fs_info->dev_root = root;
2356 btrfs_init_devices_late(fs_info);
2358 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2359 root = btrfs_read_tree_root(tree_root, &location);
2361 return PTR_ERR(root);
2362 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2363 fs_info->csum_root = root;
2365 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2366 root = btrfs_read_tree_root(tree_root, &location);
2367 if (!IS_ERR(root)) {
2368 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2369 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2370 fs_info->quota_root = root;
2373 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2374 root = btrfs_read_tree_root(tree_root, &location);
2376 ret = PTR_ERR(root);
2380 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2381 fs_info->uuid_root = root;
2384 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2385 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2386 root = btrfs_read_tree_root(tree_root, &location);
2388 return PTR_ERR(root);
2389 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2390 fs_info->free_space_root = root;
2396 int open_ctree(struct super_block *sb,
2397 struct btrfs_fs_devices *fs_devices,
2405 struct btrfs_key location;
2406 struct buffer_head *bh;
2407 struct btrfs_super_block *disk_super;
2408 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2409 struct btrfs_root *tree_root;
2410 struct btrfs_root *chunk_root;
2413 int num_backups_tried = 0;
2414 int backup_index = 0;
2416 int clear_free_space_tree = 0;
2418 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2419 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2420 if (!tree_root || !chunk_root) {
2425 ret = init_srcu_struct(&fs_info->subvol_srcu);
2431 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2436 fs_info->dirty_metadata_batch = PAGE_SIZE *
2437 (1 + ilog2(nr_cpu_ids));
2439 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2442 goto fail_dirty_metadata_bytes;
2445 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2448 goto fail_delalloc_bytes;
2451 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2452 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2453 INIT_LIST_HEAD(&fs_info->trans_list);
2454 INIT_LIST_HEAD(&fs_info->dead_roots);
2455 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2456 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2457 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2458 spin_lock_init(&fs_info->delalloc_root_lock);
2459 spin_lock_init(&fs_info->trans_lock);
2460 spin_lock_init(&fs_info->fs_roots_radix_lock);
2461 spin_lock_init(&fs_info->delayed_iput_lock);
2462 spin_lock_init(&fs_info->defrag_inodes_lock);
2463 spin_lock_init(&fs_info->tree_mod_seq_lock);
2464 spin_lock_init(&fs_info->super_lock);
2465 spin_lock_init(&fs_info->qgroup_op_lock);
2466 spin_lock_init(&fs_info->buffer_lock);
2467 spin_lock_init(&fs_info->unused_bgs_lock);
2468 rwlock_init(&fs_info->tree_mod_log_lock);
2469 mutex_init(&fs_info->unused_bg_unpin_mutex);
2470 mutex_init(&fs_info->delete_unused_bgs_mutex);
2471 mutex_init(&fs_info->reloc_mutex);
2472 mutex_init(&fs_info->delalloc_root_mutex);
2473 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2474 seqlock_init(&fs_info->profiles_lock);
2476 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2477 INIT_LIST_HEAD(&fs_info->space_info);
2478 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2479 INIT_LIST_HEAD(&fs_info->unused_bgs);
2480 btrfs_mapping_init(&fs_info->mapping_tree);
2481 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2482 BTRFS_BLOCK_RSV_GLOBAL);
2483 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2484 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2485 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2486 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2487 BTRFS_BLOCK_RSV_DELOPS);
2488 atomic_set(&fs_info->async_delalloc_pages, 0);
2489 atomic_set(&fs_info->defrag_running, 0);
2490 atomic_set(&fs_info->qgroup_op_seq, 0);
2491 atomic_set(&fs_info->reada_works_cnt, 0);
2492 atomic64_set(&fs_info->tree_mod_seq, 0);
2494 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2495 fs_info->metadata_ratio = 0;
2496 fs_info->defrag_inodes = RB_ROOT;
2497 atomic64_set(&fs_info->free_chunk_space, 0);
2498 fs_info->tree_mod_log = RB_ROOT;
2499 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2500 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2501 /* readahead state */
2502 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2503 spin_lock_init(&fs_info->reada_lock);
2504 btrfs_init_ref_verify(fs_info);
2506 fs_info->thread_pool_size = min_t(unsigned long,
2507 num_online_cpus() + 2, 8);
2509 INIT_LIST_HEAD(&fs_info->ordered_roots);
2510 spin_lock_init(&fs_info->ordered_root_lock);
2512 fs_info->btree_inode = new_inode(sb);
2513 if (!fs_info->btree_inode) {
2515 goto fail_bio_counter;
2517 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2519 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2521 if (!fs_info->delayed_root) {
2525 btrfs_init_delayed_root(fs_info->delayed_root);
2527 btrfs_init_scrub(fs_info);
2528 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2529 fs_info->check_integrity_print_mask = 0;
2531 btrfs_init_balance(fs_info);
2532 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2534 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2535 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2537 btrfs_init_btree_inode(fs_info);
2539 spin_lock_init(&fs_info->block_group_cache_lock);
2540 fs_info->block_group_cache_tree = RB_ROOT;
2541 fs_info->first_logical_byte = (u64)-1;
2543 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2544 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2545 fs_info->pinned_extents = &fs_info->freed_extents[0];
2546 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2548 mutex_init(&fs_info->ordered_operations_mutex);
2549 mutex_init(&fs_info->tree_log_mutex);
2550 mutex_init(&fs_info->chunk_mutex);
2551 mutex_init(&fs_info->transaction_kthread_mutex);
2552 mutex_init(&fs_info->cleaner_mutex);
2553 mutex_init(&fs_info->volume_mutex);
2554 mutex_init(&fs_info->ro_block_group_mutex);
2555 init_rwsem(&fs_info->commit_root_sem);
2556 init_rwsem(&fs_info->cleanup_work_sem);
2557 init_rwsem(&fs_info->subvol_sem);
2558 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2560 btrfs_init_dev_replace_locks(fs_info);
2561 btrfs_init_qgroup(fs_info);
2563 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2564 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2566 init_waitqueue_head(&fs_info->transaction_throttle);
2567 init_waitqueue_head(&fs_info->transaction_wait);
2568 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2569 init_waitqueue_head(&fs_info->async_submit_wait);
2571 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2573 /* Usable values until the real ones are cached from the superblock */
2574 fs_info->nodesize = 4096;
2575 fs_info->sectorsize = 4096;
2576 fs_info->stripesize = 4096;
2578 ret = btrfs_alloc_stripe_hash_table(fs_info);
2584 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2586 invalidate_bdev(fs_devices->latest_bdev);
2589 * Read super block and check the signature bytes only
2591 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2598 * We want to check superblock checksum, the type is stored inside.
2599 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2601 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2602 btrfs_err(fs_info, "superblock checksum mismatch");
2609 * super_copy is zeroed at allocation time and we never touch the
2610 * following bytes up to INFO_SIZE, the checksum is calculated from
2611 * the whole block of INFO_SIZE
2613 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2614 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2615 sizeof(*fs_info->super_for_commit));
2618 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2620 ret = btrfs_check_super_valid(fs_info);
2622 btrfs_err(fs_info, "superblock contains fatal errors");
2627 disk_super = fs_info->super_copy;
2628 if (!btrfs_super_root(disk_super))
2631 /* check FS state, whether FS is broken. */
2632 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2633 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2636 * run through our array of backup supers and setup
2637 * our ring pointer to the oldest one
2639 generation = btrfs_super_generation(disk_super);
2640 find_oldest_super_backup(fs_info, generation);
2643 * In the long term, we'll store the compression type in the super
2644 * block, and it'll be used for per file compression control.
2646 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2648 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2654 features = btrfs_super_incompat_flags(disk_super) &
2655 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2658 "cannot mount because of unsupported optional features (%llx)",
2664 features = btrfs_super_incompat_flags(disk_super);
2665 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2666 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2667 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2668 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2669 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2671 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2672 btrfs_info(fs_info, "has skinny extents");
2675 * flag our filesystem as having big metadata blocks if
2676 * they are bigger than the page size
2678 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2679 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2681 "flagging fs with big metadata feature");
2682 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2685 nodesize = btrfs_super_nodesize(disk_super);
2686 sectorsize = btrfs_super_sectorsize(disk_super);
2687 stripesize = sectorsize;
2688 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2689 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2691 /* Cache block sizes */
2692 fs_info->nodesize = nodesize;
2693 fs_info->sectorsize = sectorsize;
2694 fs_info->stripesize = stripesize;
2697 * mixed block groups end up with duplicate but slightly offset
2698 * extent buffers for the same range. It leads to corruptions
2700 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2701 (sectorsize != nodesize)) {
2703 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2704 nodesize, sectorsize);
2709 * Needn't use the lock because there is no other task which will
2712 btrfs_set_super_incompat_flags(disk_super, features);
2714 features = btrfs_super_compat_ro_flags(disk_super) &
2715 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2716 if (!sb_rdonly(sb) && features) {
2718 "cannot mount read-write because of unsupported optional features (%llx)",
2724 max_active = fs_info->thread_pool_size;
2726 ret = btrfs_init_workqueues(fs_info, fs_devices);
2729 goto fail_sb_buffer;
2732 sb->s_bdi->congested_fn = btrfs_congested_fn;
2733 sb->s_bdi->congested_data = fs_info;
2734 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2735 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2736 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2737 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2739 sb->s_blocksize = sectorsize;
2740 sb->s_blocksize_bits = blksize_bits(sectorsize);
2741 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2743 mutex_lock(&fs_info->chunk_mutex);
2744 ret = btrfs_read_sys_array(fs_info);
2745 mutex_unlock(&fs_info->chunk_mutex);
2747 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2748 goto fail_sb_buffer;
2751 generation = btrfs_super_chunk_root_generation(disk_super);
2753 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2755 chunk_root->node = read_tree_block(fs_info,
2756 btrfs_super_chunk_root(disk_super),
2758 if (IS_ERR(chunk_root->node) ||
2759 !extent_buffer_uptodate(chunk_root->node)) {
2760 btrfs_err(fs_info, "failed to read chunk root");
2761 if (!IS_ERR(chunk_root->node))
2762 free_extent_buffer(chunk_root->node);
2763 chunk_root->node = NULL;
2764 goto fail_tree_roots;
2766 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2767 chunk_root->commit_root = btrfs_root_node(chunk_root);
2769 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2770 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2772 ret = btrfs_read_chunk_tree(fs_info);
2774 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2775 goto fail_tree_roots;
2779 * keep the device that is marked to be the target device for the
2780 * dev_replace procedure
2782 btrfs_close_extra_devices(fs_devices, 0);
2784 if (!fs_devices->latest_bdev) {
2785 btrfs_err(fs_info, "failed to read devices");
2786 goto fail_tree_roots;
2790 generation = btrfs_super_generation(disk_super);
2792 tree_root->node = read_tree_block(fs_info,
2793 btrfs_super_root(disk_super),
2795 if (IS_ERR(tree_root->node) ||
2796 !extent_buffer_uptodate(tree_root->node)) {
2797 btrfs_warn(fs_info, "failed to read tree root");
2798 if (!IS_ERR(tree_root->node))
2799 free_extent_buffer(tree_root->node);
2800 tree_root->node = NULL;
2801 goto recovery_tree_root;
2804 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2805 tree_root->commit_root = btrfs_root_node(tree_root);
2806 btrfs_set_root_refs(&tree_root->root_item, 1);
2808 mutex_lock(&tree_root->objectid_mutex);
2809 ret = btrfs_find_highest_objectid(tree_root,
2810 &tree_root->highest_objectid);
2812 mutex_unlock(&tree_root->objectid_mutex);
2813 goto recovery_tree_root;
2816 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2818 mutex_unlock(&tree_root->objectid_mutex);
2820 ret = btrfs_read_roots(fs_info);
2822 goto recovery_tree_root;
2824 fs_info->generation = generation;
2825 fs_info->last_trans_committed = generation;
2827 ret = btrfs_recover_balance(fs_info);
2829 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2830 goto fail_block_groups;
2833 ret = btrfs_init_dev_stats(fs_info);
2835 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2836 goto fail_block_groups;
2839 ret = btrfs_init_dev_replace(fs_info);
2841 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2842 goto fail_block_groups;
2845 btrfs_close_extra_devices(fs_devices, 1);
2847 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2849 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2851 goto fail_block_groups;
2854 ret = btrfs_sysfs_add_device(fs_devices);
2856 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2858 goto fail_fsdev_sysfs;
2861 ret = btrfs_sysfs_add_mounted(fs_info);
2863 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2864 goto fail_fsdev_sysfs;
2867 ret = btrfs_init_space_info(fs_info);
2869 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2873 ret = btrfs_read_block_groups(fs_info);
2875 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2879 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info)) {
2881 "writeable mount is not allowed due to too many missing devices");
2885 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2887 if (IS_ERR(fs_info->cleaner_kthread))
2890 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2892 "btrfs-transaction");
2893 if (IS_ERR(fs_info->transaction_kthread))
2896 if (!btrfs_test_opt(fs_info, NOSSD) &&
2897 !fs_info->fs_devices->rotating) {
2898 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2902 * Mount does not set all options immediately, we can do it now and do
2903 * not have to wait for transaction commit
2905 btrfs_apply_pending_changes(fs_info);
2907 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2908 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2909 ret = btrfsic_mount(fs_info, fs_devices,
2910 btrfs_test_opt(fs_info,
2911 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2913 fs_info->check_integrity_print_mask);
2916 "failed to initialize integrity check module: %d",
2920 ret = btrfs_read_qgroup_config(fs_info);
2922 goto fail_trans_kthread;
2924 if (btrfs_build_ref_tree(fs_info))
2925 btrfs_err(fs_info, "couldn't build ref tree");
2927 /* do not make disk changes in broken FS or nologreplay is given */
2928 if (btrfs_super_log_root(disk_super) != 0 &&
2929 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2930 ret = btrfs_replay_log(fs_info, fs_devices);
2937 ret = btrfs_find_orphan_roots(fs_info);
2941 if (!sb_rdonly(sb)) {
2942 ret = btrfs_cleanup_fs_roots(fs_info);
2946 mutex_lock(&fs_info->cleaner_mutex);
2947 ret = btrfs_recover_relocation(tree_root);
2948 mutex_unlock(&fs_info->cleaner_mutex);
2950 btrfs_warn(fs_info, "failed to recover relocation: %d",
2957 location.objectid = BTRFS_FS_TREE_OBJECTID;
2958 location.type = BTRFS_ROOT_ITEM_KEY;
2959 location.offset = 0;
2961 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2962 if (IS_ERR(fs_info->fs_root)) {
2963 err = PTR_ERR(fs_info->fs_root);
2970 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2971 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2972 clear_free_space_tree = 1;
2973 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2974 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2975 btrfs_warn(fs_info, "free space tree is invalid");
2976 clear_free_space_tree = 1;
2979 if (clear_free_space_tree) {
2980 btrfs_info(fs_info, "clearing free space tree");
2981 ret = btrfs_clear_free_space_tree(fs_info);
2984 "failed to clear free space tree: %d", ret);
2985 close_ctree(fs_info);
2990 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
2991 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2992 btrfs_info(fs_info, "creating free space tree");
2993 ret = btrfs_create_free_space_tree(fs_info);
2996 "failed to create free space tree: %d", ret);
2997 close_ctree(fs_info);
3002 down_read(&fs_info->cleanup_work_sem);
3003 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3004 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3005 up_read(&fs_info->cleanup_work_sem);
3006 close_ctree(fs_info);
3009 up_read(&fs_info->cleanup_work_sem);
3011 ret = btrfs_resume_balance_async(fs_info);
3013 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3014 close_ctree(fs_info);
3018 ret = btrfs_resume_dev_replace_async(fs_info);
3020 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3021 close_ctree(fs_info);
3025 btrfs_qgroup_rescan_resume(fs_info);
3027 if (!fs_info->uuid_root) {
3028 btrfs_info(fs_info, "creating UUID tree");
3029 ret = btrfs_create_uuid_tree(fs_info);
3032 "failed to create the UUID tree: %d", ret);
3033 close_ctree(fs_info);
3036 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3037 fs_info->generation !=
3038 btrfs_super_uuid_tree_generation(disk_super)) {
3039 btrfs_info(fs_info, "checking UUID tree");
3040 ret = btrfs_check_uuid_tree(fs_info);
3043 "failed to check the UUID tree: %d", ret);
3044 close_ctree(fs_info);
3048 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3050 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3053 * backuproot only affect mount behavior, and if open_ctree succeeded,
3054 * no need to keep the flag
3056 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3061 btrfs_free_qgroup_config(fs_info);
3063 kthread_stop(fs_info->transaction_kthread);
3064 btrfs_cleanup_transaction(fs_info);
3065 btrfs_free_fs_roots(fs_info);
3067 kthread_stop(fs_info->cleaner_kthread);
3070 * make sure we're done with the btree inode before we stop our
3073 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3076 btrfs_sysfs_remove_mounted(fs_info);
3079 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3082 btrfs_put_block_group_cache(fs_info);
3085 free_root_pointers(fs_info, 1);
3086 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3089 btrfs_stop_all_workers(fs_info);
3090 btrfs_free_block_groups(fs_info);
3093 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3095 iput(fs_info->btree_inode);
3097 percpu_counter_destroy(&fs_info->bio_counter);
3098 fail_delalloc_bytes:
3099 percpu_counter_destroy(&fs_info->delalloc_bytes);
3100 fail_dirty_metadata_bytes:
3101 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3103 cleanup_srcu_struct(&fs_info->subvol_srcu);
3105 btrfs_free_stripe_hash_table(fs_info);
3106 btrfs_close_devices(fs_info->fs_devices);
3110 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3111 goto fail_tree_roots;
3113 free_root_pointers(fs_info, 0);
3115 /* don't use the log in recovery mode, it won't be valid */
3116 btrfs_set_super_log_root(disk_super, 0);
3118 /* we can't trust the free space cache either */
3119 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3121 ret = next_root_backup(fs_info, fs_info->super_copy,
3122 &num_backups_tried, &backup_index);
3124 goto fail_block_groups;
3125 goto retry_root_backup;
3127 BPF_ALLOW_ERROR_INJECTION(open_ctree);
3129 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3132 set_buffer_uptodate(bh);
3134 struct btrfs_device *device = (struct btrfs_device *)
3137 btrfs_warn_rl_in_rcu(device->fs_info,
3138 "lost page write due to IO error on %s",
3139 rcu_str_deref(device->name));
3140 /* note, we don't set_buffer_write_io_error because we have
3141 * our own ways of dealing with the IO errors
3143 clear_buffer_uptodate(bh);
3144 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3150 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3151 struct buffer_head **bh_ret)
3153 struct buffer_head *bh;
3154 struct btrfs_super_block *super;
3157 bytenr = btrfs_sb_offset(copy_num);
3158 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3161 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3163 * If we fail to read from the underlying devices, as of now
3164 * the best option we have is to mark it EIO.
3169 super = (struct btrfs_super_block *)bh->b_data;
3170 if (btrfs_super_bytenr(super) != bytenr ||
3171 btrfs_super_magic(super) != BTRFS_MAGIC) {
3181 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3183 struct buffer_head *bh;
3184 struct buffer_head *latest = NULL;
3185 struct btrfs_super_block *super;
3190 /* we would like to check all the supers, but that would make
3191 * a btrfs mount succeed after a mkfs from a different FS.
3192 * So, we need to add a special mount option to scan for
3193 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3195 for (i = 0; i < 1; i++) {
3196 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3200 super = (struct btrfs_super_block *)bh->b_data;
3202 if (!latest || btrfs_super_generation(super) > transid) {
3205 transid = btrfs_super_generation(super);
3212 return ERR_PTR(ret);
3218 * Write superblock @sb to the @device. Do not wait for completion, all the
3219 * buffer heads we write are pinned.
3221 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3222 * the expected device size at commit time. Note that max_mirrors must be
3223 * same for write and wait phases.
3225 * Return number of errors when buffer head is not found or submission fails.
3227 static int write_dev_supers(struct btrfs_device *device,
3228 struct btrfs_super_block *sb, int max_mirrors)
3230 struct buffer_head *bh;
3238 if (max_mirrors == 0)
3239 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3241 for (i = 0; i < max_mirrors; i++) {
3242 bytenr = btrfs_sb_offset(i);
3243 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3244 device->commit_total_bytes)
3247 btrfs_set_super_bytenr(sb, bytenr);
3250 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3251 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3252 btrfs_csum_final(crc, sb->csum);
3254 /* One reference for us, and we leave it for the caller */
3255 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3256 BTRFS_SUPER_INFO_SIZE);
3258 btrfs_err(device->fs_info,
3259 "couldn't get super buffer head for bytenr %llu",
3265 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3267 /* one reference for submit_bh */
3270 set_buffer_uptodate(bh);
3272 bh->b_end_io = btrfs_end_buffer_write_sync;
3273 bh->b_private = device;
3276 * we fua the first super. The others we allow
3279 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3280 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3281 op_flags |= REQ_FUA;
3282 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3286 return errors < i ? 0 : -1;
3290 * Wait for write completion of superblocks done by write_dev_supers,
3291 * @max_mirrors same for write and wait phases.
3293 * Return number of errors when buffer head is not found or not marked up to
3296 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3298 struct buffer_head *bh;
3303 if (max_mirrors == 0)
3304 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3306 for (i = 0; i < max_mirrors; i++) {
3307 bytenr = btrfs_sb_offset(i);
3308 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3309 device->commit_total_bytes)
3312 bh = __find_get_block(device->bdev,
3313 bytenr / BTRFS_BDEV_BLOCKSIZE,
3314 BTRFS_SUPER_INFO_SIZE);
3320 if (!buffer_uptodate(bh))
3323 /* drop our reference */
3326 /* drop the reference from the writing run */
3330 return errors < i ? 0 : -1;
3334 * endio for the write_dev_flush, this will wake anyone waiting
3335 * for the barrier when it is done
3337 static void btrfs_end_empty_barrier(struct bio *bio)
3339 complete(bio->bi_private);
3343 * Submit a flush request to the device if it supports it. Error handling is
3344 * done in the waiting counterpart.
3346 static void write_dev_flush(struct btrfs_device *device)
3348 struct request_queue *q = bdev_get_queue(device->bdev);
3349 struct bio *bio = device->flush_bio;
3351 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3355 bio->bi_end_io = btrfs_end_empty_barrier;
3356 bio_set_dev(bio, device->bdev);
3357 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3358 init_completion(&device->flush_wait);
3359 bio->bi_private = &device->flush_wait;
3361 btrfsic_submit_bio(bio);
3362 device->flush_bio_sent = 1;
3366 * If the flush bio has been submitted by write_dev_flush, wait for it.
3368 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3370 struct bio *bio = device->flush_bio;
3372 if (!device->flush_bio_sent)
3375 device->flush_bio_sent = 0;
3376 wait_for_completion_io(&device->flush_wait);
3378 return bio->bi_status;
3381 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3383 if (!btrfs_check_rw_degradable(fs_info))
3389 * send an empty flush down to each device in parallel,
3390 * then wait for them
3392 static int barrier_all_devices(struct btrfs_fs_info *info)
3394 struct list_head *head;
3395 struct btrfs_device *dev;
3396 int errors_wait = 0;
3399 /* send down all the barriers */
3400 head = &info->fs_devices->devices;
3401 list_for_each_entry_rcu(dev, head, dev_list) {
3406 if (!dev->in_fs_metadata || !dev->writeable)
3409 write_dev_flush(dev);
3410 dev->last_flush_error = BLK_STS_OK;
3413 /* wait for all the barriers */
3414 list_for_each_entry_rcu(dev, head, dev_list) {
3421 if (!dev->in_fs_metadata || !dev->writeable)
3424 ret = wait_dev_flush(dev);
3426 dev->last_flush_error = ret;
3427 btrfs_dev_stat_inc_and_print(dev,
3428 BTRFS_DEV_STAT_FLUSH_ERRS);
3435 * At some point we need the status of all disks
3436 * to arrive at the volume status. So error checking
3437 * is being pushed to a separate loop.
3439 return check_barrier_error(info);
3444 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3447 int min_tolerated = INT_MAX;
3449 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3450 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3451 min_tolerated = min(min_tolerated,
3452 btrfs_raid_array[BTRFS_RAID_SINGLE].
3453 tolerated_failures);
3455 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3456 if (raid_type == BTRFS_RAID_SINGLE)
3458 if (!(flags & btrfs_raid_group[raid_type]))
3460 min_tolerated = min(min_tolerated,
3461 btrfs_raid_array[raid_type].
3462 tolerated_failures);
3465 if (min_tolerated == INT_MAX) {
3466 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3470 return min_tolerated;
3473 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3475 struct list_head *head;
3476 struct btrfs_device *dev;
3477 struct btrfs_super_block *sb;
3478 struct btrfs_dev_item *dev_item;
3482 int total_errors = 0;
3485 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3488 * max_mirrors == 0 indicates we're from commit_transaction,
3489 * not from fsync where the tree roots in fs_info have not
3490 * been consistent on disk.
3492 if (max_mirrors == 0)
3493 backup_super_roots(fs_info);
3495 sb = fs_info->super_for_commit;
3496 dev_item = &sb->dev_item;
3498 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3499 head = &fs_info->fs_devices->devices;
3500 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3503 ret = barrier_all_devices(fs_info);
3506 &fs_info->fs_devices->device_list_mutex);
3507 btrfs_handle_fs_error(fs_info, ret,
3508 "errors while submitting device barriers.");
3513 list_for_each_entry_rcu(dev, head, dev_list) {
3518 if (!dev->in_fs_metadata || !dev->writeable)
3521 btrfs_set_stack_device_generation(dev_item, 0);
3522 btrfs_set_stack_device_type(dev_item, dev->type);
3523 btrfs_set_stack_device_id(dev_item, dev->devid);
3524 btrfs_set_stack_device_total_bytes(dev_item,
3525 dev->commit_total_bytes);
3526 btrfs_set_stack_device_bytes_used(dev_item,
3527 dev->commit_bytes_used);
3528 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3529 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3530 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3531 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3532 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3534 flags = btrfs_super_flags(sb);
3535 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3537 ret = write_dev_supers(dev, sb, max_mirrors);
3541 if (total_errors > max_errors) {
3542 btrfs_err(fs_info, "%d errors while writing supers",
3544 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3546 /* FUA is masked off if unsupported and can't be the reason */
3547 btrfs_handle_fs_error(fs_info, -EIO,
3548 "%d errors while writing supers",
3554 list_for_each_entry_rcu(dev, head, dev_list) {
3557 if (!dev->in_fs_metadata || !dev->writeable)
3560 ret = wait_dev_supers(dev, max_mirrors);
3564 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3565 if (total_errors > max_errors) {
3566 btrfs_handle_fs_error(fs_info, -EIO,
3567 "%d errors while writing supers",
3574 /* Drop a fs root from the radix tree and free it. */
3575 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3576 struct btrfs_root *root)
3578 spin_lock(&fs_info->fs_roots_radix_lock);
3579 radix_tree_delete(&fs_info->fs_roots_radix,
3580 (unsigned long)root->root_key.objectid);
3581 spin_unlock(&fs_info->fs_roots_radix_lock);
3583 if (btrfs_root_refs(&root->root_item) == 0)
3584 synchronize_srcu(&fs_info->subvol_srcu);
3586 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3587 btrfs_free_log(NULL, root);
3588 if (root->reloc_root) {
3589 free_extent_buffer(root->reloc_root->node);
3590 free_extent_buffer(root->reloc_root->commit_root);
3591 btrfs_put_fs_root(root->reloc_root);
3592 root->reloc_root = NULL;
3596 if (root->free_ino_pinned)
3597 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3598 if (root->free_ino_ctl)
3599 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3603 static void free_fs_root(struct btrfs_root *root)
3605 iput(root->ino_cache_inode);
3606 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3607 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3608 root->orphan_block_rsv = NULL;
3610 free_anon_bdev(root->anon_dev);
3611 if (root->subv_writers)
3612 btrfs_free_subvolume_writers(root->subv_writers);
3613 free_extent_buffer(root->node);
3614 free_extent_buffer(root->commit_root);
3615 kfree(root->free_ino_ctl);
3616 kfree(root->free_ino_pinned);
3618 btrfs_put_fs_root(root);
3621 void btrfs_free_fs_root(struct btrfs_root *root)
3626 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3628 u64 root_objectid = 0;
3629 struct btrfs_root *gang[8];
3632 unsigned int ret = 0;
3636 index = srcu_read_lock(&fs_info->subvol_srcu);
3637 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3638 (void **)gang, root_objectid,
3641 srcu_read_unlock(&fs_info->subvol_srcu, index);
3644 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3646 for (i = 0; i < ret; i++) {
3647 /* Avoid to grab roots in dead_roots */
3648 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3652 /* grab all the search result for later use */
3653 gang[i] = btrfs_grab_fs_root(gang[i]);
3655 srcu_read_unlock(&fs_info->subvol_srcu, index);
3657 for (i = 0; i < ret; i++) {
3660 root_objectid = gang[i]->root_key.objectid;
3661 err = btrfs_orphan_cleanup(gang[i]);
3664 btrfs_put_fs_root(gang[i]);
3669 /* release the uncleaned roots due to error */
3670 for (; i < ret; i++) {
3672 btrfs_put_fs_root(gang[i]);
3677 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3679 struct btrfs_root *root = fs_info->tree_root;
3680 struct btrfs_trans_handle *trans;
3682 mutex_lock(&fs_info->cleaner_mutex);
3683 btrfs_run_delayed_iputs(fs_info);
3684 mutex_unlock(&fs_info->cleaner_mutex);
3685 wake_up_process(fs_info->cleaner_kthread);
3687 /* wait until ongoing cleanup work done */
3688 down_write(&fs_info->cleanup_work_sem);
3689 up_write(&fs_info->cleanup_work_sem);
3691 trans = btrfs_join_transaction(root);
3693 return PTR_ERR(trans);
3694 return btrfs_commit_transaction(trans);
3697 void close_ctree(struct btrfs_fs_info *fs_info)
3699 struct btrfs_root *root = fs_info->tree_root;
3702 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3704 /* wait for the qgroup rescan worker to stop */
3705 btrfs_qgroup_wait_for_completion(fs_info, false);
3707 /* wait for the uuid_scan task to finish */
3708 down(&fs_info->uuid_tree_rescan_sem);
3709 /* avoid complains from lockdep et al., set sem back to initial state */
3710 up(&fs_info->uuid_tree_rescan_sem);
3712 /* pause restriper - we want to resume on mount */
3713 btrfs_pause_balance(fs_info);
3715 btrfs_dev_replace_suspend_for_unmount(fs_info);
3717 btrfs_scrub_cancel(fs_info);
3719 /* wait for any defraggers to finish */
3720 wait_event(fs_info->transaction_wait,
3721 (atomic_read(&fs_info->defrag_running) == 0));
3723 /* clear out the rbtree of defraggable inodes */
3724 btrfs_cleanup_defrag_inodes(fs_info);
3726 cancel_work_sync(&fs_info->async_reclaim_work);
3728 if (!sb_rdonly(fs_info->sb)) {
3730 * If the cleaner thread is stopped and there are
3731 * block groups queued for removal, the deletion will be
3732 * skipped when we quit the cleaner thread.
3734 btrfs_delete_unused_bgs(fs_info);
3736 ret = btrfs_commit_super(fs_info);
3738 btrfs_err(fs_info, "commit super ret %d", ret);
3741 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3742 btrfs_error_commit_super(fs_info);
3744 kthread_stop(fs_info->transaction_kthread);
3745 kthread_stop(fs_info->cleaner_kthread);
3747 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3749 btrfs_free_qgroup_config(fs_info);
3751 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3752 btrfs_info(fs_info, "at unmount delalloc count %lld",
3753 percpu_counter_sum(&fs_info->delalloc_bytes));
3756 btrfs_sysfs_remove_mounted(fs_info);
3757 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3759 btrfs_free_fs_roots(fs_info);
3761 btrfs_put_block_group_cache(fs_info);
3764 * we must make sure there is not any read request to
3765 * submit after we stopping all workers.
3767 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3768 btrfs_stop_all_workers(fs_info);
3770 btrfs_free_block_groups(fs_info);
3772 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3773 free_root_pointers(fs_info, 1);
3775 iput(fs_info->btree_inode);
3777 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3778 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3779 btrfsic_unmount(fs_info->fs_devices);
3782 btrfs_close_devices(fs_info->fs_devices);
3783 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3785 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3786 percpu_counter_destroy(&fs_info->delalloc_bytes);
3787 percpu_counter_destroy(&fs_info->bio_counter);
3788 cleanup_srcu_struct(&fs_info->subvol_srcu);
3790 btrfs_free_stripe_hash_table(fs_info);
3791 btrfs_free_ref_cache(fs_info);
3793 __btrfs_free_block_rsv(root->orphan_block_rsv);
3794 root->orphan_block_rsv = NULL;
3796 while (!list_empty(&fs_info->pinned_chunks)) {
3797 struct extent_map *em;
3799 em = list_first_entry(&fs_info->pinned_chunks,
3800 struct extent_map, list);
3801 list_del_init(&em->list);
3802 free_extent_map(em);
3806 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3810 struct inode *btree_inode = buf->pages[0]->mapping->host;
3812 ret = extent_buffer_uptodate(buf);
3816 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3817 parent_transid, atomic);
3823 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3825 struct btrfs_fs_info *fs_info;
3826 struct btrfs_root *root;
3827 u64 transid = btrfs_header_generation(buf);
3830 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3832 * This is a fast path so only do this check if we have sanity tests
3833 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3834 * outside of the sanity tests.
3836 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3839 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3840 fs_info = root->fs_info;
3841 btrfs_assert_tree_locked(buf);
3842 if (transid != fs_info->generation)
3843 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3844 buf->start, transid, fs_info->generation);
3845 was_dirty = set_extent_buffer_dirty(buf);
3847 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3849 fs_info->dirty_metadata_batch);
3850 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3852 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3853 * but item data not updated.
3854 * So here we should only check item pointers, not item data.
3856 if (btrfs_header_level(buf) == 0 &&
3857 btrfs_check_leaf_relaxed(root, buf)) {
3858 btrfs_print_leaf(buf);
3864 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
3868 * looks as though older kernels can get into trouble with
3869 * this code, they end up stuck in balance_dirty_pages forever
3873 if (current->flags & PF_MEMALLOC)
3877 btrfs_balance_delayed_items(fs_info);
3879 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
3880 BTRFS_DIRTY_METADATA_THRESH);
3882 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
3886 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
3888 __btrfs_btree_balance_dirty(fs_info, 1);
3891 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
3893 __btrfs_btree_balance_dirty(fs_info, 0);
3896 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3898 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3899 struct btrfs_fs_info *fs_info = root->fs_info;
3901 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
3904 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
3906 struct btrfs_super_block *sb = fs_info->super_copy;
3907 u64 nodesize = btrfs_super_nodesize(sb);
3908 u64 sectorsize = btrfs_super_sectorsize(sb);
3911 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
3912 btrfs_err(fs_info, "no valid FS found");
3915 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
3916 btrfs_warn(fs_info, "unrecognized super flag: %llu",
3917 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
3918 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3919 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
3920 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3923 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3924 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
3925 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3928 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3929 btrfs_err(fs_info, "log_root level too big: %d >= %d",
3930 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3935 * Check sectorsize and nodesize first, other check will need it.
3936 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
3938 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
3939 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3940 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
3943 /* Only PAGE SIZE is supported yet */
3944 if (sectorsize != PAGE_SIZE) {
3946 "sectorsize %llu not supported yet, only support %lu",
3947 sectorsize, PAGE_SIZE);
3950 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
3951 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3952 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
3955 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
3956 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
3957 le32_to_cpu(sb->__unused_leafsize), nodesize);
3961 /* Root alignment check */
3962 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
3963 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
3964 btrfs_super_root(sb));
3967 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
3968 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
3969 btrfs_super_chunk_root(sb));
3972 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
3973 btrfs_warn(fs_info, "log_root block unaligned: %llu",
3974 btrfs_super_log_root(sb));
3978 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
3980 "dev_item UUID does not match fsid: %pU != %pU",
3981 fs_info->fsid, sb->dev_item.fsid);
3986 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3989 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
3990 btrfs_err(fs_info, "bytes_used is too small %llu",
3991 btrfs_super_bytes_used(sb));
3994 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
3995 btrfs_err(fs_info, "invalid stripesize %u",
3996 btrfs_super_stripesize(sb));
3999 if (btrfs_super_num_devices(sb) > (1UL << 31))
4000 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4001 btrfs_super_num_devices(sb));
4002 if (btrfs_super_num_devices(sb) == 0) {
4003 btrfs_err(fs_info, "number of devices is 0");
4007 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4008 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4009 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4014 * Obvious sys_chunk_array corruptions, it must hold at least one key
4017 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4018 btrfs_err(fs_info, "system chunk array too big %u > %u",
4019 btrfs_super_sys_array_size(sb),
4020 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4023 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4024 + sizeof(struct btrfs_chunk)) {
4025 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4026 btrfs_super_sys_array_size(sb),
4027 sizeof(struct btrfs_disk_key)
4028 + sizeof(struct btrfs_chunk));
4033 * The generation is a global counter, we'll trust it more than the others
4034 * but it's still possible that it's the one that's wrong.
4036 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4038 "suspicious: generation < chunk_root_generation: %llu < %llu",
4039 btrfs_super_generation(sb),
4040 btrfs_super_chunk_root_generation(sb));
4041 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4042 && btrfs_super_cache_generation(sb) != (u64)-1)
4044 "suspicious: generation < cache_generation: %llu < %llu",
4045 btrfs_super_generation(sb),
4046 btrfs_super_cache_generation(sb));
4051 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4053 mutex_lock(&fs_info->cleaner_mutex);
4054 btrfs_run_delayed_iputs(fs_info);
4055 mutex_unlock(&fs_info->cleaner_mutex);
4057 down_write(&fs_info->cleanup_work_sem);
4058 up_write(&fs_info->cleanup_work_sem);
4060 /* cleanup FS via transaction */
4061 btrfs_cleanup_transaction(fs_info);
4064 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4066 struct btrfs_ordered_extent *ordered;
4068 spin_lock(&root->ordered_extent_lock);
4070 * This will just short circuit the ordered completion stuff which will
4071 * make sure the ordered extent gets properly cleaned up.
4073 list_for_each_entry(ordered, &root->ordered_extents,
4075 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4076 spin_unlock(&root->ordered_extent_lock);
4079 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4081 struct btrfs_root *root;
4082 struct list_head splice;
4084 INIT_LIST_HEAD(&splice);
4086 spin_lock(&fs_info->ordered_root_lock);
4087 list_splice_init(&fs_info->ordered_roots, &splice);
4088 while (!list_empty(&splice)) {
4089 root = list_first_entry(&splice, struct btrfs_root,
4091 list_move_tail(&root->ordered_root,
4092 &fs_info->ordered_roots);
4094 spin_unlock(&fs_info->ordered_root_lock);
4095 btrfs_destroy_ordered_extents(root);
4098 spin_lock(&fs_info->ordered_root_lock);
4100 spin_unlock(&fs_info->ordered_root_lock);
4103 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4104 struct btrfs_fs_info *fs_info)
4106 struct rb_node *node;
4107 struct btrfs_delayed_ref_root *delayed_refs;
4108 struct btrfs_delayed_ref_node *ref;
4111 delayed_refs = &trans->delayed_refs;
4113 spin_lock(&delayed_refs->lock);
4114 if (atomic_read(&delayed_refs->num_entries) == 0) {
4115 spin_unlock(&delayed_refs->lock);
4116 btrfs_info(fs_info, "delayed_refs has NO entry");
4120 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4121 struct btrfs_delayed_ref_head *head;
4123 bool pin_bytes = false;
4125 head = rb_entry(node, struct btrfs_delayed_ref_head,
4127 if (!mutex_trylock(&head->mutex)) {
4128 refcount_inc(&head->refs);
4129 spin_unlock(&delayed_refs->lock);
4131 mutex_lock(&head->mutex);
4132 mutex_unlock(&head->mutex);
4133 btrfs_put_delayed_ref_head(head);
4134 spin_lock(&delayed_refs->lock);
4137 spin_lock(&head->lock);
4138 while ((n = rb_first(&head->ref_tree)) != NULL) {
4139 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4142 rb_erase(&ref->ref_node, &head->ref_tree);
4143 RB_CLEAR_NODE(&ref->ref_node);
4144 if (!list_empty(&ref->add_list))
4145 list_del(&ref->add_list);
4146 atomic_dec(&delayed_refs->num_entries);
4147 btrfs_put_delayed_ref(ref);
4149 if (head->must_insert_reserved)
4151 btrfs_free_delayed_extent_op(head->extent_op);
4152 delayed_refs->num_heads--;
4153 if (head->processing == 0)
4154 delayed_refs->num_heads_ready--;
4155 atomic_dec(&delayed_refs->num_entries);
4156 rb_erase(&head->href_node, &delayed_refs->href_root);
4157 RB_CLEAR_NODE(&head->href_node);
4158 spin_unlock(&head->lock);
4159 spin_unlock(&delayed_refs->lock);
4160 mutex_unlock(&head->mutex);
4163 btrfs_pin_extent(fs_info, head->bytenr,
4164 head->num_bytes, 1);
4165 btrfs_put_delayed_ref_head(head);
4167 spin_lock(&delayed_refs->lock);
4170 spin_unlock(&delayed_refs->lock);
4175 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4177 struct btrfs_inode *btrfs_inode;
4178 struct list_head splice;
4180 INIT_LIST_HEAD(&splice);
4182 spin_lock(&root->delalloc_lock);
4183 list_splice_init(&root->delalloc_inodes, &splice);
4185 while (!list_empty(&splice)) {
4186 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4189 list_del_init(&btrfs_inode->delalloc_inodes);
4190 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4191 &btrfs_inode->runtime_flags);
4192 spin_unlock(&root->delalloc_lock);
4194 btrfs_invalidate_inodes(btrfs_inode->root);
4196 spin_lock(&root->delalloc_lock);
4199 spin_unlock(&root->delalloc_lock);
4202 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4204 struct btrfs_root *root;
4205 struct list_head splice;
4207 INIT_LIST_HEAD(&splice);
4209 spin_lock(&fs_info->delalloc_root_lock);
4210 list_splice_init(&fs_info->delalloc_roots, &splice);
4211 while (!list_empty(&splice)) {
4212 root = list_first_entry(&splice, struct btrfs_root,
4214 list_del_init(&root->delalloc_root);
4215 root = btrfs_grab_fs_root(root);
4217 spin_unlock(&fs_info->delalloc_root_lock);
4219 btrfs_destroy_delalloc_inodes(root);
4220 btrfs_put_fs_root(root);
4222 spin_lock(&fs_info->delalloc_root_lock);
4224 spin_unlock(&fs_info->delalloc_root_lock);
4227 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4228 struct extent_io_tree *dirty_pages,
4232 struct extent_buffer *eb;
4237 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4242 clear_extent_bits(dirty_pages, start, end, mark);
4243 while (start <= end) {
4244 eb = find_extent_buffer(fs_info, start);
4245 start += fs_info->nodesize;
4248 wait_on_extent_buffer_writeback(eb);
4250 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4252 clear_extent_buffer_dirty(eb);
4253 free_extent_buffer_stale(eb);
4260 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4261 struct extent_io_tree *pinned_extents)
4263 struct extent_io_tree *unpin;
4269 unpin = pinned_extents;
4272 ret = find_first_extent_bit(unpin, 0, &start, &end,
4273 EXTENT_DIRTY, NULL);
4277 clear_extent_dirty(unpin, start, end);
4278 btrfs_error_unpin_extent_range(fs_info, start, end);
4283 if (unpin == &fs_info->freed_extents[0])
4284 unpin = &fs_info->freed_extents[1];
4286 unpin = &fs_info->freed_extents[0];
4294 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4296 struct inode *inode;
4298 inode = cache->io_ctl.inode;
4300 invalidate_inode_pages2(inode->i_mapping);
4301 BTRFS_I(inode)->generation = 0;
4302 cache->io_ctl.inode = NULL;
4305 btrfs_put_block_group(cache);
4308 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4309 struct btrfs_fs_info *fs_info)
4311 struct btrfs_block_group_cache *cache;
4313 spin_lock(&cur_trans->dirty_bgs_lock);
4314 while (!list_empty(&cur_trans->dirty_bgs)) {
4315 cache = list_first_entry(&cur_trans->dirty_bgs,
4316 struct btrfs_block_group_cache,
4319 btrfs_err(fs_info, "orphan block group dirty_bgs list");
4320 spin_unlock(&cur_trans->dirty_bgs_lock);
4324 if (!list_empty(&cache->io_list)) {
4325 spin_unlock(&cur_trans->dirty_bgs_lock);
4326 list_del_init(&cache->io_list);
4327 btrfs_cleanup_bg_io(cache);
4328 spin_lock(&cur_trans->dirty_bgs_lock);
4331 list_del_init(&cache->dirty_list);
4332 spin_lock(&cache->lock);
4333 cache->disk_cache_state = BTRFS_DC_ERROR;
4334 spin_unlock(&cache->lock);
4336 spin_unlock(&cur_trans->dirty_bgs_lock);
4337 btrfs_put_block_group(cache);
4338 spin_lock(&cur_trans->dirty_bgs_lock);
4340 spin_unlock(&cur_trans->dirty_bgs_lock);
4342 while (!list_empty(&cur_trans->io_bgs)) {
4343 cache = list_first_entry(&cur_trans->io_bgs,
4344 struct btrfs_block_group_cache,
4347 btrfs_err(fs_info, "orphan block group on io_bgs list");
4351 list_del_init(&cache->io_list);
4352 spin_lock(&cache->lock);
4353 cache->disk_cache_state = BTRFS_DC_ERROR;
4354 spin_unlock(&cache->lock);
4355 btrfs_cleanup_bg_io(cache);
4359 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4360 struct btrfs_fs_info *fs_info)
4362 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4363 ASSERT(list_empty(&cur_trans->dirty_bgs));
4364 ASSERT(list_empty(&cur_trans->io_bgs));
4366 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4368 cur_trans->state = TRANS_STATE_COMMIT_START;
4369 wake_up(&fs_info->transaction_blocked_wait);
4371 cur_trans->state = TRANS_STATE_UNBLOCKED;
4372 wake_up(&fs_info->transaction_wait);
4374 btrfs_destroy_delayed_inodes(fs_info);
4375 btrfs_assert_delayed_root_empty(fs_info);
4377 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4379 btrfs_destroy_pinned_extent(fs_info,
4380 fs_info->pinned_extents);
4382 cur_trans->state =TRANS_STATE_COMPLETED;
4383 wake_up(&cur_trans->commit_wait);
4386 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4388 struct btrfs_transaction *t;
4390 mutex_lock(&fs_info->transaction_kthread_mutex);
4392 spin_lock(&fs_info->trans_lock);
4393 while (!list_empty(&fs_info->trans_list)) {
4394 t = list_first_entry(&fs_info->trans_list,
4395 struct btrfs_transaction, list);
4396 if (t->state >= TRANS_STATE_COMMIT_START) {
4397 refcount_inc(&t->use_count);
4398 spin_unlock(&fs_info->trans_lock);
4399 btrfs_wait_for_commit(fs_info, t->transid);
4400 btrfs_put_transaction(t);
4401 spin_lock(&fs_info->trans_lock);
4404 if (t == fs_info->running_transaction) {
4405 t->state = TRANS_STATE_COMMIT_DOING;
4406 spin_unlock(&fs_info->trans_lock);
4408 * We wait for 0 num_writers since we don't hold a trans
4409 * handle open currently for this transaction.
4411 wait_event(t->writer_wait,
4412 atomic_read(&t->num_writers) == 0);
4414 spin_unlock(&fs_info->trans_lock);
4416 btrfs_cleanup_one_transaction(t, fs_info);
4418 spin_lock(&fs_info->trans_lock);
4419 if (t == fs_info->running_transaction)
4420 fs_info->running_transaction = NULL;
4421 list_del_init(&t->list);
4422 spin_unlock(&fs_info->trans_lock);
4424 btrfs_put_transaction(t);
4425 trace_btrfs_transaction_commit(fs_info->tree_root);
4426 spin_lock(&fs_info->trans_lock);
4428 spin_unlock(&fs_info->trans_lock);
4429 btrfs_destroy_all_ordered_extents(fs_info);
4430 btrfs_destroy_delayed_inodes(fs_info);
4431 btrfs_assert_delayed_root_empty(fs_info);
4432 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4433 btrfs_destroy_all_delalloc_inodes(fs_info);
4434 mutex_unlock(&fs_info->transaction_kthread_mutex);
4439 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4441 struct inode *inode = private_data;
4442 return btrfs_sb(inode->i_sb);
4445 static const struct extent_io_ops btree_extent_io_ops = {
4446 /* mandatory callbacks */
4447 .submit_bio_hook = btree_submit_bio_hook,
4448 .readpage_end_io_hook = btree_readpage_end_io_hook,
4449 /* note we're sharing with inode.c for the merge bio hook */
4450 .merge_bio_hook = btrfs_merge_bio_hook,
4451 .readpage_io_failed_hook = btree_io_failed_hook,
4452 .set_range_writeback = btrfs_set_range_writeback,
4453 .tree_fs_info = btree_fs_info,
4455 /* optional callbacks */