2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include "compression.h"
53 #include "tree-checker.h"
54 #include "ref-verify.h"
57 #include <asm/cpufeature.h>
60 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
61 BTRFS_HEADER_FLAG_RELOC |\
62 BTRFS_SUPER_FLAG_ERROR |\
63 BTRFS_SUPER_FLAG_SEEDING |\
64 BTRFS_SUPER_FLAG_METADUMP)
66 static const struct extent_io_ops btree_extent_io_ops;
67 static void end_workqueue_fn(struct btrfs_work *work);
68 static void free_fs_root(struct btrfs_root *root);
69 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
70 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
71 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
72 struct btrfs_fs_info *fs_info);
73 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
74 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
75 struct extent_io_tree *dirty_pages,
77 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
78 struct extent_io_tree *pinned_extents);
79 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
80 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
83 * btrfs_end_io_wq structs are used to do processing in task context when an IO
84 * is complete. This is used during reads to verify checksums, and it is used
85 * by writes to insert metadata for new file extents after IO is complete.
87 struct btrfs_end_io_wq {
91 struct btrfs_fs_info *info;
93 enum btrfs_wq_endio_type metadata;
94 struct btrfs_work work;
97 static struct kmem_cache *btrfs_end_io_wq_cache;
99 int __init btrfs_end_io_wq_init(void)
101 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq),
106 if (!btrfs_end_io_wq_cache)
111 void btrfs_end_io_wq_exit(void)
113 kmem_cache_destroy(btrfs_end_io_wq_cache);
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
121 struct async_submit_bio {
123 struct btrfs_fs_info *fs_info;
125 extent_submit_bio_hook_t *submit_bio_start;
126 extent_submit_bio_hook_t *submit_bio_done;
128 unsigned long bio_flags;
130 * bio_offset is optional, can be used if the pages in the bio
131 * can't tell us where in the file the bio should go
134 struct btrfs_work work;
139 * Lockdep class keys for extent_buffer->lock's in this root. For a given
140 * eb, the lockdep key is determined by the btrfs_root it belongs to and
141 * the level the eb occupies in the tree.
143 * Different roots are used for different purposes and may nest inside each
144 * other and they require separate keysets. As lockdep keys should be
145 * static, assign keysets according to the purpose of the root as indicated
146 * by btrfs_root->objectid. This ensures that all special purpose roots
147 * have separate keysets.
149 * Lock-nesting across peer nodes is always done with the immediate parent
150 * node locked thus preventing deadlock. As lockdep doesn't know this, use
151 * subclass to avoid triggering lockdep warning in such cases.
153 * The key is set by the readpage_end_io_hook after the buffer has passed
154 * csum validation but before the pages are unlocked. It is also set by
155 * btrfs_init_new_buffer on freshly allocated blocks.
157 * We also add a check to make sure the highest level of the tree is the
158 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
159 * needs update as well.
161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
162 # if BTRFS_MAX_LEVEL != 8
166 static struct btrfs_lockdep_keyset {
167 u64 id; /* root objectid */
168 const char *name_stem; /* lock name stem */
169 char names[BTRFS_MAX_LEVEL + 1][20];
170 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
171 } btrfs_lockdep_keysets[] = {
172 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
173 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
174 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
175 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
176 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
177 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
178 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
179 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
180 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
181 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
182 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
183 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
184 { .id = 0, .name_stem = "tree" },
187 void __init btrfs_init_lockdep(void)
191 /* initialize lockdep class names */
192 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
193 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
195 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
196 snprintf(ks->names[j], sizeof(ks->names[j]),
197 "btrfs-%s-%02d", ks->name_stem, j);
201 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
204 struct btrfs_lockdep_keyset *ks;
206 BUG_ON(level >= ARRAY_SIZE(ks->keys));
208 /* find the matching keyset, id 0 is the default entry */
209 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
210 if (ks->id == objectid)
213 lockdep_set_class_and_name(&eb->lock,
214 &ks->keys[level], ks->names[level]);
220 * extents on the btree inode are pretty simple, there's one extent
221 * that covers the entire device
223 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
224 struct page *page, size_t pg_offset, u64 start, u64 len,
227 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
228 struct extent_map_tree *em_tree = &inode->extent_tree;
229 struct extent_map *em;
232 read_lock(&em_tree->lock);
233 em = lookup_extent_mapping(em_tree, start, len);
235 em->bdev = fs_info->fs_devices->latest_bdev;
236 read_unlock(&em_tree->lock);
239 read_unlock(&em_tree->lock);
241 em = alloc_extent_map();
243 em = ERR_PTR(-ENOMEM);
248 em->block_len = (u64)-1;
250 em->bdev = fs_info->fs_devices->latest_bdev;
252 write_lock(&em_tree->lock);
253 ret = add_extent_mapping(em_tree, em, 0);
254 if (ret == -EEXIST) {
256 em = lookup_extent_mapping(em_tree, start, len);
263 write_unlock(&em_tree->lock);
269 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
271 return btrfs_crc32c(seed, data, len);
274 void btrfs_csum_final(u32 crc, u8 *result)
276 put_unaligned_le32(~crc, result);
280 * compute the csum for a btree block, and either verify it or write it
281 * into the csum field of the block.
283 static int csum_tree_block(struct btrfs_fs_info *fs_info,
284 struct extent_buffer *buf,
287 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
288 char result[BTRFS_CSUM_SIZE];
290 unsigned long cur_len;
291 unsigned long offset = BTRFS_CSUM_SIZE;
293 unsigned long map_start;
294 unsigned long map_len;
298 len = buf->len - offset;
300 err = map_private_extent_buffer(buf, offset, 32,
301 &kaddr, &map_start, &map_len);
304 cur_len = min(len, map_len - (offset - map_start));
305 crc = btrfs_csum_data(kaddr + offset - map_start,
310 memset(result, 0, BTRFS_CSUM_SIZE);
312 btrfs_csum_final(crc, result);
315 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
318 memcpy(&found, result, csum_size);
320 read_extent_buffer(buf, &val, 0, csum_size);
321 btrfs_warn_rl(fs_info,
322 "%s checksum verify failed on %llu wanted %X found %X level %d",
323 fs_info->sb->s_id, buf->start,
324 val, found, btrfs_header_level(buf));
328 write_extent_buffer(buf, result, 0, csum_size);
335 * we can't consider a given block up to date unless the transid of the
336 * block matches the transid in the parent node's pointer. This is how we
337 * detect blocks that either didn't get written at all or got written
338 * in the wrong place.
340 static int verify_parent_transid(struct extent_io_tree *io_tree,
341 struct extent_buffer *eb, u64 parent_transid,
344 struct extent_state *cached_state = NULL;
346 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
348 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
355 btrfs_tree_read_lock(eb);
356 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
359 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
361 if (extent_buffer_uptodate(eb) &&
362 btrfs_header_generation(eb) == parent_transid) {
366 btrfs_err_rl(eb->fs_info,
367 "parent transid verify failed on %llu wanted %llu found %llu",
369 parent_transid, btrfs_header_generation(eb));
373 * Things reading via commit roots that don't have normal protection,
374 * like send, can have a really old block in cache that may point at a
375 * block that has been freed and re-allocated. So don't clear uptodate
376 * if we find an eb that is under IO (dirty/writeback) because we could
377 * end up reading in the stale data and then writing it back out and
378 * making everybody very sad.
380 if (!extent_buffer_under_io(eb))
381 clear_extent_buffer_uptodate(eb);
383 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
384 &cached_state, GFP_NOFS);
386 btrfs_tree_read_unlock_blocking(eb);
391 * Return 0 if the superblock checksum type matches the checksum value of that
392 * algorithm. Pass the raw disk superblock data.
394 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
397 struct btrfs_super_block *disk_sb =
398 (struct btrfs_super_block *)raw_disk_sb;
399 u16 csum_type = btrfs_super_csum_type(disk_sb);
402 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
404 const int csum_size = sizeof(crc);
405 char result[csum_size];
408 * The super_block structure does not span the whole
409 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
410 * is filled with zeros and is included in the checksum.
412 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
413 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
414 btrfs_csum_final(crc, result);
416 if (memcmp(raw_disk_sb, result, csum_size))
420 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
421 btrfs_err(fs_info, "unsupported checksum algorithm %u",
430 * helper to read a given tree block, doing retries as required when
431 * the checksums don't match and we have alternate mirrors to try.
433 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
434 struct extent_buffer *eb,
437 struct extent_io_tree *io_tree;
442 int failed_mirror = 0;
444 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
445 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
447 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
450 if (!verify_parent_transid(io_tree, eb,
458 * This buffer's crc is fine, but its contents are corrupted, so
459 * there is no reason to read the other copies, they won't be
462 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
465 num_copies = btrfs_num_copies(fs_info,
470 if (!failed_mirror) {
472 failed_mirror = eb->read_mirror;
476 if (mirror_num == failed_mirror)
479 if (mirror_num > num_copies)
483 if (failed && !ret && failed_mirror)
484 repair_eb_io_failure(fs_info, eb, failed_mirror);
490 * checksum a dirty tree block before IO. This has extra checks to make sure
491 * we only fill in the checksum field in the first page of a multi-page block
494 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
496 u64 start = page_offset(page);
498 struct extent_buffer *eb;
500 eb = (struct extent_buffer *)page->private;
501 if (page != eb->pages[0])
504 found_start = btrfs_header_bytenr(eb);
506 * Please do not consolidate these warnings into a single if.
507 * It is useful to know what went wrong.
509 if (WARN_ON(found_start != start))
511 if (WARN_ON(!PageUptodate(page)))
514 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
515 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
517 return csum_tree_block(fs_info, eb, 0);
520 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
521 struct extent_buffer *eb)
523 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
524 u8 fsid[BTRFS_FSID_SIZE];
527 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
529 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
533 fs_devices = fs_devices->seed;
538 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
539 u64 phy_offset, struct page *page,
540 u64 start, u64 end, int mirror)
544 struct extent_buffer *eb;
545 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
546 struct btrfs_fs_info *fs_info = root->fs_info;
553 eb = (struct extent_buffer *)page->private;
555 /* the pending IO might have been the only thing that kept this buffer
556 * in memory. Make sure we have a ref for all this other checks
558 extent_buffer_get(eb);
560 reads_done = atomic_dec_and_test(&eb->io_pages);
564 eb->read_mirror = mirror;
565 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
570 found_start = btrfs_header_bytenr(eb);
571 if (found_start != eb->start) {
572 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
573 found_start, eb->start);
577 if (check_tree_block_fsid(fs_info, eb)) {
578 btrfs_err_rl(fs_info, "bad fsid on block %llu",
583 found_level = btrfs_header_level(eb);
584 if (found_level >= BTRFS_MAX_LEVEL) {
585 btrfs_err(fs_info, "bad tree block level %d",
586 (int)btrfs_header_level(eb));
591 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
594 ret = csum_tree_block(fs_info, eb, 1);
599 * If this is a leaf block and it is corrupt, set the corrupt bit so
600 * that we don't try and read the other copies of this block, just
603 if (found_level == 0 && btrfs_check_leaf_full(root, eb)) {
604 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
608 if (found_level > 0 && btrfs_check_node(root, eb))
612 set_extent_buffer_uptodate(eb);
615 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
616 btree_readahead_hook(eb, ret);
620 * our io error hook is going to dec the io pages
621 * again, we have to make sure it has something
624 atomic_inc(&eb->io_pages);
625 clear_extent_buffer_uptodate(eb);
627 free_extent_buffer(eb);
632 static int btree_io_failed_hook(struct page *page, int failed_mirror)
634 struct extent_buffer *eb;
636 eb = (struct extent_buffer *)page->private;
637 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
638 eb->read_mirror = failed_mirror;
639 atomic_dec(&eb->io_pages);
640 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
641 btree_readahead_hook(eb, -EIO);
642 return -EIO; /* we fixed nothing */
645 static void end_workqueue_bio(struct bio *bio)
647 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
648 struct btrfs_fs_info *fs_info;
649 struct btrfs_workqueue *wq;
650 btrfs_work_func_t func;
652 fs_info = end_io_wq->info;
653 end_io_wq->status = bio->bi_status;
655 if (bio_op(bio) == REQ_OP_WRITE) {
656 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
657 wq = fs_info->endio_meta_write_workers;
658 func = btrfs_endio_meta_write_helper;
659 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
660 wq = fs_info->endio_freespace_worker;
661 func = btrfs_freespace_write_helper;
662 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
663 wq = fs_info->endio_raid56_workers;
664 func = btrfs_endio_raid56_helper;
666 wq = fs_info->endio_write_workers;
667 func = btrfs_endio_write_helper;
670 if (unlikely(end_io_wq->metadata ==
671 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
672 wq = fs_info->endio_repair_workers;
673 func = btrfs_endio_repair_helper;
674 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
675 wq = fs_info->endio_raid56_workers;
676 func = btrfs_endio_raid56_helper;
677 } else if (end_io_wq->metadata) {
678 wq = fs_info->endio_meta_workers;
679 func = btrfs_endio_meta_helper;
681 wq = fs_info->endio_workers;
682 func = btrfs_endio_helper;
686 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
687 btrfs_queue_work(wq, &end_io_wq->work);
690 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
691 enum btrfs_wq_endio_type metadata)
693 struct btrfs_end_io_wq *end_io_wq;
695 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
697 return BLK_STS_RESOURCE;
699 end_io_wq->private = bio->bi_private;
700 end_io_wq->end_io = bio->bi_end_io;
701 end_io_wq->info = info;
702 end_io_wq->status = 0;
703 end_io_wq->bio = bio;
704 end_io_wq->metadata = metadata;
706 bio->bi_private = end_io_wq;
707 bio->bi_end_io = end_workqueue_bio;
711 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
713 unsigned long limit = min_t(unsigned long,
714 info->thread_pool_size,
715 info->fs_devices->open_devices);
719 static void run_one_async_start(struct btrfs_work *work)
721 struct async_submit_bio *async;
724 async = container_of(work, struct async_submit_bio, work);
725 ret = async->submit_bio_start(async->private_data, async->bio,
726 async->mirror_num, async->bio_flags,
732 static void run_one_async_done(struct btrfs_work *work)
734 struct async_submit_bio *async;
736 async = container_of(work, struct async_submit_bio, work);
738 /* If an error occurred we just want to clean up the bio and move on */
740 async->bio->bi_status = async->status;
741 bio_endio(async->bio);
745 async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
746 async->bio_flags, async->bio_offset);
749 static void run_one_async_free(struct btrfs_work *work)
751 struct async_submit_bio *async;
753 async = container_of(work, struct async_submit_bio, work);
757 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
758 int mirror_num, unsigned long bio_flags,
759 u64 bio_offset, void *private_data,
760 extent_submit_bio_hook_t *submit_bio_start,
761 extent_submit_bio_hook_t *submit_bio_done)
763 struct async_submit_bio *async;
765 async = kmalloc(sizeof(*async), GFP_NOFS);
767 return BLK_STS_RESOURCE;
769 async->private_data = private_data;
770 async->fs_info = fs_info;
772 async->mirror_num = mirror_num;
773 async->submit_bio_start = submit_bio_start;
774 async->submit_bio_done = submit_bio_done;
776 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
777 run_one_async_done, run_one_async_free);
779 async->bio_flags = bio_flags;
780 async->bio_offset = bio_offset;
784 if (op_is_sync(bio->bi_opf))
785 btrfs_set_work_high_priority(&async->work);
787 btrfs_queue_work(fs_info->workers, &async->work);
791 static blk_status_t btree_csum_one_bio(struct bio *bio)
793 struct bio_vec *bvec;
794 struct btrfs_root *root;
797 ASSERT(!bio_flagged(bio, BIO_CLONED));
798 bio_for_each_segment_all(bvec, bio, i) {
799 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
800 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
805 return errno_to_blk_status(ret);
808 static blk_status_t __btree_submit_bio_start(void *private_data, struct bio *bio,
809 int mirror_num, unsigned long bio_flags,
813 * when we're called for a write, we're already in the async
814 * submission context. Just jump into btrfs_map_bio
816 return btree_csum_one_bio(bio);
819 static blk_status_t __btree_submit_bio_done(void *private_data, struct bio *bio,
820 int mirror_num, unsigned long bio_flags,
823 struct inode *inode = private_data;
827 * when we're called for a write, we're already in the async
828 * submission context. Just jump into btrfs_map_bio
830 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
832 bio->bi_status = ret;
838 static int check_async_write(struct btrfs_inode *bi)
840 if (atomic_read(&bi->sync_writers))
843 if (static_cpu_has(X86_FEATURE_XMM4_2))
849 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
850 int mirror_num, unsigned long bio_flags,
853 struct inode *inode = private_data;
854 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
855 int async = check_async_write(BTRFS_I(inode));
858 if (bio_op(bio) != REQ_OP_WRITE) {
860 * called for a read, do the setup so that checksum validation
861 * can happen in the async kernel threads
863 ret = btrfs_bio_wq_end_io(fs_info, bio,
864 BTRFS_WQ_ENDIO_METADATA);
867 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
869 ret = btree_csum_one_bio(bio);
872 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
875 * kthread helpers are used to submit writes so that
876 * checksumming can happen in parallel across all CPUs
878 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
879 bio_offset, private_data,
880 __btree_submit_bio_start,
881 __btree_submit_bio_done);
889 bio->bi_status = ret;
894 #ifdef CONFIG_MIGRATION
895 static int btree_migratepage(struct address_space *mapping,
896 struct page *newpage, struct page *page,
897 enum migrate_mode mode)
900 * we can't safely write a btree page from here,
901 * we haven't done the locking hook
906 * Buffers may be managed in a filesystem specific way.
907 * We must have no buffers or drop them.
909 if (page_has_private(page) &&
910 !try_to_release_page(page, GFP_KERNEL))
912 return migrate_page(mapping, newpage, page, mode);
917 static int btree_writepages(struct address_space *mapping,
918 struct writeback_control *wbc)
920 struct btrfs_fs_info *fs_info;
923 if (wbc->sync_mode == WB_SYNC_NONE) {
925 if (wbc->for_kupdate)
928 fs_info = BTRFS_I(mapping->host)->root->fs_info;
929 /* this is a bit racy, but that's ok */
930 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
931 BTRFS_DIRTY_METADATA_THRESH);
935 return btree_write_cache_pages(mapping, wbc);
938 static int btree_readpage(struct file *file, struct page *page)
940 struct extent_io_tree *tree;
941 tree = &BTRFS_I(page->mapping->host)->io_tree;
942 return extent_read_full_page(tree, page, btree_get_extent, 0);
945 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
947 if (PageWriteback(page) || PageDirty(page))
950 return try_release_extent_buffer(page);
953 static void btree_invalidatepage(struct page *page, unsigned int offset,
956 struct extent_io_tree *tree;
957 tree = &BTRFS_I(page->mapping->host)->io_tree;
958 extent_invalidatepage(tree, page, offset);
959 btree_releasepage(page, GFP_NOFS);
960 if (PagePrivate(page)) {
961 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
962 "page private not zero on page %llu",
963 (unsigned long long)page_offset(page));
964 ClearPagePrivate(page);
965 set_page_private(page, 0);
970 static int btree_set_page_dirty(struct page *page)
973 struct extent_buffer *eb;
975 BUG_ON(!PagePrivate(page));
976 eb = (struct extent_buffer *)page->private;
978 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
979 BUG_ON(!atomic_read(&eb->refs));
980 btrfs_assert_tree_locked(eb);
982 return __set_page_dirty_nobuffers(page);
985 static const struct address_space_operations btree_aops = {
986 .readpage = btree_readpage,
987 .writepages = btree_writepages,
988 .releasepage = btree_releasepage,
989 .invalidatepage = btree_invalidatepage,
990 #ifdef CONFIG_MIGRATION
991 .migratepage = btree_migratepage,
993 .set_page_dirty = btree_set_page_dirty,
996 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
998 struct extent_buffer *buf = NULL;
999 struct inode *btree_inode = fs_info->btree_inode;
1001 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1004 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1006 free_extent_buffer(buf);
1009 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1010 int mirror_num, struct extent_buffer **eb)
1012 struct extent_buffer *buf = NULL;
1013 struct inode *btree_inode = fs_info->btree_inode;
1014 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1017 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1021 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1023 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1026 free_extent_buffer(buf);
1030 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1031 free_extent_buffer(buf);
1033 } else if (extent_buffer_uptodate(buf)) {
1036 free_extent_buffer(buf);
1041 struct extent_buffer *btrfs_find_create_tree_block(
1042 struct btrfs_fs_info *fs_info,
1045 if (btrfs_is_testing(fs_info))
1046 return alloc_test_extent_buffer(fs_info, bytenr);
1047 return alloc_extent_buffer(fs_info, bytenr);
1051 int btrfs_write_tree_block(struct extent_buffer *buf)
1053 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1054 buf->start + buf->len - 1);
1057 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1059 filemap_fdatawait_range(buf->pages[0]->mapping,
1060 buf->start, buf->start + buf->len - 1);
1063 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1066 struct extent_buffer *buf = NULL;
1069 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1073 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1075 free_extent_buffer(buf);
1076 return ERR_PTR(ret);
1082 void clean_tree_block(struct btrfs_fs_info *fs_info,
1083 struct extent_buffer *buf)
1085 if (btrfs_header_generation(buf) ==
1086 fs_info->running_transaction->transid) {
1087 btrfs_assert_tree_locked(buf);
1089 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1090 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1092 fs_info->dirty_metadata_batch);
1093 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1094 btrfs_set_lock_blocking(buf);
1095 clear_extent_buffer_dirty(buf);
1100 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1102 struct btrfs_subvolume_writers *writers;
1105 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1107 return ERR_PTR(-ENOMEM);
1109 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1112 return ERR_PTR(ret);
1115 init_waitqueue_head(&writers->wait);
1120 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1122 percpu_counter_destroy(&writers->counter);
1126 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1129 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1131 root->commit_root = NULL;
1133 root->orphan_cleanup_state = 0;
1135 root->objectid = objectid;
1136 root->last_trans = 0;
1137 root->highest_objectid = 0;
1138 root->nr_delalloc_inodes = 0;
1139 root->nr_ordered_extents = 0;
1141 root->inode_tree = RB_ROOT;
1142 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1143 root->block_rsv = NULL;
1144 root->orphan_block_rsv = NULL;
1146 INIT_LIST_HEAD(&root->dirty_list);
1147 INIT_LIST_HEAD(&root->root_list);
1148 INIT_LIST_HEAD(&root->delalloc_inodes);
1149 INIT_LIST_HEAD(&root->delalloc_root);
1150 INIT_LIST_HEAD(&root->ordered_extents);
1151 INIT_LIST_HEAD(&root->ordered_root);
1152 INIT_LIST_HEAD(&root->logged_list[0]);
1153 INIT_LIST_HEAD(&root->logged_list[1]);
1154 spin_lock_init(&root->orphan_lock);
1155 spin_lock_init(&root->inode_lock);
1156 spin_lock_init(&root->delalloc_lock);
1157 spin_lock_init(&root->ordered_extent_lock);
1158 spin_lock_init(&root->accounting_lock);
1159 spin_lock_init(&root->log_extents_lock[0]);
1160 spin_lock_init(&root->log_extents_lock[1]);
1161 mutex_init(&root->objectid_mutex);
1162 mutex_init(&root->log_mutex);
1163 mutex_init(&root->ordered_extent_mutex);
1164 mutex_init(&root->delalloc_mutex);
1165 init_waitqueue_head(&root->log_writer_wait);
1166 init_waitqueue_head(&root->log_commit_wait[0]);
1167 init_waitqueue_head(&root->log_commit_wait[1]);
1168 INIT_LIST_HEAD(&root->log_ctxs[0]);
1169 INIT_LIST_HEAD(&root->log_ctxs[1]);
1170 atomic_set(&root->log_commit[0], 0);
1171 atomic_set(&root->log_commit[1], 0);
1172 atomic_set(&root->log_writers, 0);
1173 atomic_set(&root->log_batch, 0);
1174 atomic_set(&root->orphan_inodes, 0);
1175 refcount_set(&root->refs, 1);
1176 atomic_set(&root->will_be_snapshotted, 0);
1177 atomic64_set(&root->qgroup_meta_rsv, 0);
1178 root->log_transid = 0;
1179 root->log_transid_committed = -1;
1180 root->last_log_commit = 0;
1182 extent_io_tree_init(&root->dirty_log_pages, NULL);
1184 memset(&root->root_key, 0, sizeof(root->root_key));
1185 memset(&root->root_item, 0, sizeof(root->root_item));
1186 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1188 root->defrag_trans_start = fs_info->generation;
1190 root->defrag_trans_start = 0;
1191 root->root_key.objectid = objectid;
1194 spin_lock_init(&root->root_item_lock);
1197 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1200 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1202 root->fs_info = fs_info;
1206 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1207 /* Should only be used by the testing infrastructure */
1208 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1210 struct btrfs_root *root;
1213 return ERR_PTR(-EINVAL);
1215 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1217 return ERR_PTR(-ENOMEM);
1219 /* We don't use the stripesize in selftest, set it as sectorsize */
1220 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1221 root->alloc_bytenr = 0;
1227 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1228 struct btrfs_fs_info *fs_info,
1231 struct extent_buffer *leaf;
1232 struct btrfs_root *tree_root = fs_info->tree_root;
1233 struct btrfs_root *root;
1234 struct btrfs_key key;
1236 uuid_le uuid = NULL_UUID_LE;
1238 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1240 return ERR_PTR(-ENOMEM);
1242 __setup_root(root, fs_info, objectid);
1243 root->root_key.objectid = objectid;
1244 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1245 root->root_key.offset = 0;
1247 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1249 ret = PTR_ERR(leaf);
1254 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1255 btrfs_set_header_bytenr(leaf, leaf->start);
1256 btrfs_set_header_generation(leaf, trans->transid);
1257 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1258 btrfs_set_header_owner(leaf, objectid);
1261 write_extent_buffer_fsid(leaf, fs_info->fsid);
1262 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1263 btrfs_mark_buffer_dirty(leaf);
1265 root->commit_root = btrfs_root_node(root);
1266 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1268 root->root_item.flags = 0;
1269 root->root_item.byte_limit = 0;
1270 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1271 btrfs_set_root_generation(&root->root_item, trans->transid);
1272 btrfs_set_root_level(&root->root_item, 0);
1273 btrfs_set_root_refs(&root->root_item, 1);
1274 btrfs_set_root_used(&root->root_item, leaf->len);
1275 btrfs_set_root_last_snapshot(&root->root_item, 0);
1276 btrfs_set_root_dirid(&root->root_item, 0);
1277 if (is_fstree(objectid))
1279 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1280 root->root_item.drop_level = 0;
1282 key.objectid = objectid;
1283 key.type = BTRFS_ROOT_ITEM_KEY;
1285 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1289 btrfs_tree_unlock(leaf);
1295 btrfs_tree_unlock(leaf);
1296 free_extent_buffer(root->commit_root);
1297 free_extent_buffer(leaf);
1301 return ERR_PTR(ret);
1304 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1305 struct btrfs_fs_info *fs_info)
1307 struct btrfs_root *root;
1308 struct extent_buffer *leaf;
1310 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1312 return ERR_PTR(-ENOMEM);
1314 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1316 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1317 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1318 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1321 * DON'T set REF_COWS for log trees
1323 * log trees do not get reference counted because they go away
1324 * before a real commit is actually done. They do store pointers
1325 * to file data extents, and those reference counts still get
1326 * updated (along with back refs to the log tree).
1329 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1333 return ERR_CAST(leaf);
1336 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1337 btrfs_set_header_bytenr(leaf, leaf->start);
1338 btrfs_set_header_generation(leaf, trans->transid);
1339 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1340 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1343 write_extent_buffer_fsid(root->node, fs_info->fsid);
1344 btrfs_mark_buffer_dirty(root->node);
1345 btrfs_tree_unlock(root->node);
1349 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1350 struct btrfs_fs_info *fs_info)
1352 struct btrfs_root *log_root;
1354 log_root = alloc_log_tree(trans, fs_info);
1355 if (IS_ERR(log_root))
1356 return PTR_ERR(log_root);
1357 WARN_ON(fs_info->log_root_tree);
1358 fs_info->log_root_tree = log_root;
1362 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root)
1365 struct btrfs_fs_info *fs_info = root->fs_info;
1366 struct btrfs_root *log_root;
1367 struct btrfs_inode_item *inode_item;
1369 log_root = alloc_log_tree(trans, fs_info);
1370 if (IS_ERR(log_root))
1371 return PTR_ERR(log_root);
1373 log_root->last_trans = trans->transid;
1374 log_root->root_key.offset = root->root_key.objectid;
1376 inode_item = &log_root->root_item.inode;
1377 btrfs_set_stack_inode_generation(inode_item, 1);
1378 btrfs_set_stack_inode_size(inode_item, 3);
1379 btrfs_set_stack_inode_nlink(inode_item, 1);
1380 btrfs_set_stack_inode_nbytes(inode_item,
1382 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1384 btrfs_set_root_node(&log_root->root_item, log_root->node);
1386 WARN_ON(root->log_root);
1387 root->log_root = log_root;
1388 root->log_transid = 0;
1389 root->log_transid_committed = -1;
1390 root->last_log_commit = 0;
1394 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1395 struct btrfs_key *key)
1397 struct btrfs_root *root;
1398 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1399 struct btrfs_path *path;
1403 path = btrfs_alloc_path();
1405 return ERR_PTR(-ENOMEM);
1407 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1413 __setup_root(root, fs_info, key->objectid);
1415 ret = btrfs_find_root(tree_root, key, path,
1416 &root->root_item, &root->root_key);
1423 generation = btrfs_root_generation(&root->root_item);
1424 root->node = read_tree_block(fs_info,
1425 btrfs_root_bytenr(&root->root_item),
1427 if (IS_ERR(root->node)) {
1428 ret = PTR_ERR(root->node);
1430 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1432 free_extent_buffer(root->node);
1435 root->commit_root = btrfs_root_node(root);
1437 btrfs_free_path(path);
1443 root = ERR_PTR(ret);
1447 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1448 struct btrfs_key *location)
1450 struct btrfs_root *root;
1452 root = btrfs_read_tree_root(tree_root, location);
1456 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1457 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1458 btrfs_check_and_init_root_item(&root->root_item);
1464 int btrfs_init_fs_root(struct btrfs_root *root)
1467 struct btrfs_subvolume_writers *writers;
1469 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1470 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1472 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1477 writers = btrfs_alloc_subvolume_writers();
1478 if (IS_ERR(writers)) {
1479 ret = PTR_ERR(writers);
1482 root->subv_writers = writers;
1484 btrfs_init_free_ino_ctl(root);
1485 spin_lock_init(&root->ino_cache_lock);
1486 init_waitqueue_head(&root->ino_cache_wait);
1488 ret = get_anon_bdev(&root->anon_dev);
1492 mutex_lock(&root->objectid_mutex);
1493 ret = btrfs_find_highest_objectid(root,
1494 &root->highest_objectid);
1496 mutex_unlock(&root->objectid_mutex);
1500 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1502 mutex_unlock(&root->objectid_mutex);
1506 /* the caller is responsible to call free_fs_root */
1510 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1513 struct btrfs_root *root;
1515 spin_lock(&fs_info->fs_roots_radix_lock);
1516 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1517 (unsigned long)root_id);
1518 spin_unlock(&fs_info->fs_roots_radix_lock);
1522 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1523 struct btrfs_root *root)
1527 ret = radix_tree_preload(GFP_NOFS);
1531 spin_lock(&fs_info->fs_roots_radix_lock);
1532 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1533 (unsigned long)root->root_key.objectid,
1536 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1537 spin_unlock(&fs_info->fs_roots_radix_lock);
1538 radix_tree_preload_end();
1543 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1544 struct btrfs_key *location,
1547 struct btrfs_root *root;
1548 struct btrfs_path *path;
1549 struct btrfs_key key;
1552 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1553 return fs_info->tree_root;
1554 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1555 return fs_info->extent_root;
1556 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1557 return fs_info->chunk_root;
1558 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1559 return fs_info->dev_root;
1560 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1561 return fs_info->csum_root;
1562 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1563 return fs_info->quota_root ? fs_info->quota_root :
1565 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1566 return fs_info->uuid_root ? fs_info->uuid_root :
1568 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1569 return fs_info->free_space_root ? fs_info->free_space_root :
1572 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1574 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1575 return ERR_PTR(-ENOENT);
1579 root = btrfs_read_fs_root(fs_info->tree_root, location);
1583 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1588 ret = btrfs_init_fs_root(root);
1592 path = btrfs_alloc_path();
1597 key.objectid = BTRFS_ORPHAN_OBJECTID;
1598 key.type = BTRFS_ORPHAN_ITEM_KEY;
1599 key.offset = location->objectid;
1601 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1602 btrfs_free_path(path);
1606 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1608 ret = btrfs_insert_fs_root(fs_info, root);
1610 if (ret == -EEXIST) {
1619 return ERR_PTR(ret);
1622 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1624 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1626 struct btrfs_device *device;
1627 struct backing_dev_info *bdi;
1630 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1633 bdi = device->bdev->bd_bdi;
1634 if (bdi_congested(bdi, bdi_bits)) {
1644 * called by the kthread helper functions to finally call the bio end_io
1645 * functions. This is where read checksum verification actually happens
1647 static void end_workqueue_fn(struct btrfs_work *work)
1650 struct btrfs_end_io_wq *end_io_wq;
1652 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1653 bio = end_io_wq->bio;
1655 bio->bi_status = end_io_wq->status;
1656 bio->bi_private = end_io_wq->private;
1657 bio->bi_end_io = end_io_wq->end_io;
1658 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1662 static int cleaner_kthread(void *arg)
1664 struct btrfs_root *root = arg;
1665 struct btrfs_fs_info *fs_info = root->fs_info;
1667 struct btrfs_trans_handle *trans;
1672 /* Make the cleaner go to sleep early. */
1673 if (btrfs_need_cleaner_sleep(fs_info))
1677 * Do not do anything if we might cause open_ctree() to block
1678 * before we have finished mounting the filesystem.
1680 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1683 if (!mutex_trylock(&fs_info->cleaner_mutex))
1687 * Avoid the problem that we change the status of the fs
1688 * during the above check and trylock.
1690 if (btrfs_need_cleaner_sleep(fs_info)) {
1691 mutex_unlock(&fs_info->cleaner_mutex);
1695 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1696 btrfs_run_delayed_iputs(fs_info);
1697 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1699 again = btrfs_clean_one_deleted_snapshot(root);
1700 mutex_unlock(&fs_info->cleaner_mutex);
1703 * The defragger has dealt with the R/O remount and umount,
1704 * needn't do anything special here.
1706 btrfs_run_defrag_inodes(fs_info);
1709 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1710 * with relocation (btrfs_relocate_chunk) and relocation
1711 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1712 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1713 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1714 * unused block groups.
1716 btrfs_delete_unused_bgs(fs_info);
1719 set_current_state(TASK_INTERRUPTIBLE);
1720 if (!kthread_should_stop())
1722 __set_current_state(TASK_RUNNING);
1724 } while (!kthread_should_stop());
1727 * Transaction kthread is stopped before us and wakes us up.
1728 * However we might have started a new transaction and COWed some
1729 * tree blocks when deleting unused block groups for example. So
1730 * make sure we commit the transaction we started to have a clean
1731 * shutdown when evicting the btree inode - if it has dirty pages
1732 * when we do the final iput() on it, eviction will trigger a
1733 * writeback for it which will fail with null pointer dereferences
1734 * since work queues and other resources were already released and
1735 * destroyed by the time the iput/eviction/writeback is made.
1737 trans = btrfs_attach_transaction(root);
1738 if (IS_ERR(trans)) {
1739 if (PTR_ERR(trans) != -ENOENT)
1741 "cleaner transaction attach returned %ld",
1746 ret = btrfs_commit_transaction(trans);
1749 "cleaner open transaction commit returned %d",
1756 static int transaction_kthread(void *arg)
1758 struct btrfs_root *root = arg;
1759 struct btrfs_fs_info *fs_info = root->fs_info;
1760 struct btrfs_trans_handle *trans;
1761 struct btrfs_transaction *cur;
1764 unsigned long delay;
1768 cannot_commit = false;
1769 delay = HZ * fs_info->commit_interval;
1770 mutex_lock(&fs_info->transaction_kthread_mutex);
1772 spin_lock(&fs_info->trans_lock);
1773 cur = fs_info->running_transaction;
1775 spin_unlock(&fs_info->trans_lock);
1779 now = get_seconds();
1780 if (cur->state < TRANS_STATE_BLOCKED &&
1781 (now < cur->start_time ||
1782 now - cur->start_time < fs_info->commit_interval)) {
1783 spin_unlock(&fs_info->trans_lock);
1787 transid = cur->transid;
1788 spin_unlock(&fs_info->trans_lock);
1790 /* If the file system is aborted, this will always fail. */
1791 trans = btrfs_attach_transaction(root);
1792 if (IS_ERR(trans)) {
1793 if (PTR_ERR(trans) != -ENOENT)
1794 cannot_commit = true;
1797 if (transid == trans->transid) {
1798 btrfs_commit_transaction(trans);
1800 btrfs_end_transaction(trans);
1803 wake_up_process(fs_info->cleaner_kthread);
1804 mutex_unlock(&fs_info->transaction_kthread_mutex);
1806 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1807 &fs_info->fs_state)))
1808 btrfs_cleanup_transaction(fs_info);
1809 set_current_state(TASK_INTERRUPTIBLE);
1810 if (!kthread_should_stop() &&
1811 (!btrfs_transaction_blocked(fs_info) ||
1813 schedule_timeout(delay);
1814 __set_current_state(TASK_RUNNING);
1815 } while (!kthread_should_stop());
1820 * this will find the highest generation in the array of
1821 * root backups. The index of the highest array is returned,
1822 * or -1 if we can't find anything.
1824 * We check to make sure the array is valid by comparing the
1825 * generation of the latest root in the array with the generation
1826 * in the super block. If they don't match we pitch it.
1828 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1831 int newest_index = -1;
1832 struct btrfs_root_backup *root_backup;
1835 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1836 root_backup = info->super_copy->super_roots + i;
1837 cur = btrfs_backup_tree_root_gen(root_backup);
1838 if (cur == newest_gen)
1842 /* check to see if we actually wrapped around */
1843 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1844 root_backup = info->super_copy->super_roots;
1845 cur = btrfs_backup_tree_root_gen(root_backup);
1846 if (cur == newest_gen)
1849 return newest_index;
1854 * find the oldest backup so we know where to store new entries
1855 * in the backup array. This will set the backup_root_index
1856 * field in the fs_info struct
1858 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1861 int newest_index = -1;
1863 newest_index = find_newest_super_backup(info, newest_gen);
1864 /* if there was garbage in there, just move along */
1865 if (newest_index == -1) {
1866 info->backup_root_index = 0;
1868 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1873 * copy all the root pointers into the super backup array.
1874 * this will bump the backup pointer by one when it is
1877 static void backup_super_roots(struct btrfs_fs_info *info)
1880 struct btrfs_root_backup *root_backup;
1883 next_backup = info->backup_root_index;
1884 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1885 BTRFS_NUM_BACKUP_ROOTS;
1888 * just overwrite the last backup if we're at the same generation
1889 * this happens only at umount
1891 root_backup = info->super_for_commit->super_roots + last_backup;
1892 if (btrfs_backup_tree_root_gen(root_backup) ==
1893 btrfs_header_generation(info->tree_root->node))
1894 next_backup = last_backup;
1896 root_backup = info->super_for_commit->super_roots + next_backup;
1899 * make sure all of our padding and empty slots get zero filled
1900 * regardless of which ones we use today
1902 memset(root_backup, 0, sizeof(*root_backup));
1904 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1906 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1907 btrfs_set_backup_tree_root_gen(root_backup,
1908 btrfs_header_generation(info->tree_root->node));
1910 btrfs_set_backup_tree_root_level(root_backup,
1911 btrfs_header_level(info->tree_root->node));
1913 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1914 btrfs_set_backup_chunk_root_gen(root_backup,
1915 btrfs_header_generation(info->chunk_root->node));
1916 btrfs_set_backup_chunk_root_level(root_backup,
1917 btrfs_header_level(info->chunk_root->node));
1919 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1920 btrfs_set_backup_extent_root_gen(root_backup,
1921 btrfs_header_generation(info->extent_root->node));
1922 btrfs_set_backup_extent_root_level(root_backup,
1923 btrfs_header_level(info->extent_root->node));
1926 * we might commit during log recovery, which happens before we set
1927 * the fs_root. Make sure it is valid before we fill it in.
1929 if (info->fs_root && info->fs_root->node) {
1930 btrfs_set_backup_fs_root(root_backup,
1931 info->fs_root->node->start);
1932 btrfs_set_backup_fs_root_gen(root_backup,
1933 btrfs_header_generation(info->fs_root->node));
1934 btrfs_set_backup_fs_root_level(root_backup,
1935 btrfs_header_level(info->fs_root->node));
1938 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1939 btrfs_set_backup_dev_root_gen(root_backup,
1940 btrfs_header_generation(info->dev_root->node));
1941 btrfs_set_backup_dev_root_level(root_backup,
1942 btrfs_header_level(info->dev_root->node));
1944 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1945 btrfs_set_backup_csum_root_gen(root_backup,
1946 btrfs_header_generation(info->csum_root->node));
1947 btrfs_set_backup_csum_root_level(root_backup,
1948 btrfs_header_level(info->csum_root->node));
1950 btrfs_set_backup_total_bytes(root_backup,
1951 btrfs_super_total_bytes(info->super_copy));
1952 btrfs_set_backup_bytes_used(root_backup,
1953 btrfs_super_bytes_used(info->super_copy));
1954 btrfs_set_backup_num_devices(root_backup,
1955 btrfs_super_num_devices(info->super_copy));
1958 * if we don't copy this out to the super_copy, it won't get remembered
1959 * for the next commit
1961 memcpy(&info->super_copy->super_roots,
1962 &info->super_for_commit->super_roots,
1963 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1967 * this copies info out of the root backup array and back into
1968 * the in-memory super block. It is meant to help iterate through
1969 * the array, so you send it the number of backups you've already
1970 * tried and the last backup index you used.
1972 * this returns -1 when it has tried all the backups
1974 static noinline int next_root_backup(struct btrfs_fs_info *info,
1975 struct btrfs_super_block *super,
1976 int *num_backups_tried, int *backup_index)
1978 struct btrfs_root_backup *root_backup;
1979 int newest = *backup_index;
1981 if (*num_backups_tried == 0) {
1982 u64 gen = btrfs_super_generation(super);
1984 newest = find_newest_super_backup(info, gen);
1988 *backup_index = newest;
1989 *num_backups_tried = 1;
1990 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1991 /* we've tried all the backups, all done */
1994 /* jump to the next oldest backup */
1995 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1996 BTRFS_NUM_BACKUP_ROOTS;
1997 *backup_index = newest;
1998 *num_backups_tried += 1;
2000 root_backup = super->super_roots + newest;
2002 btrfs_set_super_generation(super,
2003 btrfs_backup_tree_root_gen(root_backup));
2004 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2005 btrfs_set_super_root_level(super,
2006 btrfs_backup_tree_root_level(root_backup));
2007 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2010 * fixme: the total bytes and num_devices need to match or we should
2013 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2014 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2018 /* helper to cleanup workers */
2019 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2021 btrfs_destroy_workqueue(fs_info->fixup_workers);
2022 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2023 btrfs_destroy_workqueue(fs_info->workers);
2024 btrfs_destroy_workqueue(fs_info->endio_workers);
2025 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2026 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2027 btrfs_destroy_workqueue(fs_info->rmw_workers);
2028 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2029 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2030 btrfs_destroy_workqueue(fs_info->submit_workers);
2031 btrfs_destroy_workqueue(fs_info->delayed_workers);
2032 btrfs_destroy_workqueue(fs_info->caching_workers);
2033 btrfs_destroy_workqueue(fs_info->readahead_workers);
2034 btrfs_destroy_workqueue(fs_info->flush_workers);
2035 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2036 btrfs_destroy_workqueue(fs_info->extent_workers);
2038 * Now that all other work queues are destroyed, we can safely destroy
2039 * the queues used for metadata I/O, since tasks from those other work
2040 * queues can do metadata I/O operations.
2042 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2043 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2046 static void free_root_extent_buffers(struct btrfs_root *root)
2049 free_extent_buffer(root->node);
2050 free_extent_buffer(root->commit_root);
2052 root->commit_root = NULL;
2056 /* helper to cleanup tree roots */
2057 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2059 free_root_extent_buffers(info->tree_root);
2061 free_root_extent_buffers(info->dev_root);
2062 free_root_extent_buffers(info->extent_root);
2063 free_root_extent_buffers(info->csum_root);
2064 free_root_extent_buffers(info->quota_root);
2065 free_root_extent_buffers(info->uuid_root);
2067 free_root_extent_buffers(info->chunk_root);
2068 free_root_extent_buffers(info->free_space_root);
2071 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2074 struct btrfs_root *gang[8];
2077 while (!list_empty(&fs_info->dead_roots)) {
2078 gang[0] = list_entry(fs_info->dead_roots.next,
2079 struct btrfs_root, root_list);
2080 list_del(&gang[0]->root_list);
2082 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2083 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2085 free_extent_buffer(gang[0]->node);
2086 free_extent_buffer(gang[0]->commit_root);
2087 btrfs_put_fs_root(gang[0]);
2092 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2097 for (i = 0; i < ret; i++)
2098 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2101 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2102 btrfs_free_log_root_tree(NULL, fs_info);
2103 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2107 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2109 mutex_init(&fs_info->scrub_lock);
2110 atomic_set(&fs_info->scrubs_running, 0);
2111 atomic_set(&fs_info->scrub_pause_req, 0);
2112 atomic_set(&fs_info->scrubs_paused, 0);
2113 atomic_set(&fs_info->scrub_cancel_req, 0);
2114 init_waitqueue_head(&fs_info->scrub_pause_wait);
2115 fs_info->scrub_workers_refcnt = 0;
2118 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2120 spin_lock_init(&fs_info->balance_lock);
2121 mutex_init(&fs_info->balance_mutex);
2122 atomic_set(&fs_info->balance_running, 0);
2123 atomic_set(&fs_info->balance_pause_req, 0);
2124 atomic_set(&fs_info->balance_cancel_req, 0);
2125 fs_info->balance_ctl = NULL;
2126 init_waitqueue_head(&fs_info->balance_wait_q);
2129 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2131 struct inode *inode = fs_info->btree_inode;
2133 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2134 set_nlink(inode, 1);
2136 * we set the i_size on the btree inode to the max possible int.
2137 * the real end of the address space is determined by all of
2138 * the devices in the system
2140 inode->i_size = OFFSET_MAX;
2141 inode->i_mapping->a_ops = &btree_aops;
2143 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2144 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2145 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2146 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2148 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2150 BTRFS_I(inode)->root = fs_info->tree_root;
2151 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2152 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2153 btrfs_insert_inode_hash(inode);
2156 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2158 fs_info->dev_replace.lock_owner = 0;
2159 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2160 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2161 rwlock_init(&fs_info->dev_replace.lock);
2162 atomic_set(&fs_info->dev_replace.read_locks, 0);
2163 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2164 init_waitqueue_head(&fs_info->replace_wait);
2165 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2168 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2170 spin_lock_init(&fs_info->qgroup_lock);
2171 mutex_init(&fs_info->qgroup_ioctl_lock);
2172 fs_info->qgroup_tree = RB_ROOT;
2173 fs_info->qgroup_op_tree = RB_ROOT;
2174 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2175 fs_info->qgroup_seq = 1;
2176 fs_info->qgroup_ulist = NULL;
2177 fs_info->qgroup_rescan_running = false;
2178 mutex_init(&fs_info->qgroup_rescan_lock);
2181 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2182 struct btrfs_fs_devices *fs_devices)
2184 int max_active = fs_info->thread_pool_size;
2185 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2188 btrfs_alloc_workqueue(fs_info, "worker",
2189 flags | WQ_HIGHPRI, max_active, 16);
2191 fs_info->delalloc_workers =
2192 btrfs_alloc_workqueue(fs_info, "delalloc",
2193 flags, max_active, 2);
2195 fs_info->flush_workers =
2196 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2197 flags, max_active, 0);
2199 fs_info->caching_workers =
2200 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2203 * a higher idle thresh on the submit workers makes it much more
2204 * likely that bios will be send down in a sane order to the
2207 fs_info->submit_workers =
2208 btrfs_alloc_workqueue(fs_info, "submit", flags,
2209 min_t(u64, fs_devices->num_devices,
2212 fs_info->fixup_workers =
2213 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2216 * endios are largely parallel and should have a very
2219 fs_info->endio_workers =
2220 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2221 fs_info->endio_meta_workers =
2222 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2224 fs_info->endio_meta_write_workers =
2225 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2227 fs_info->endio_raid56_workers =
2228 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2230 fs_info->endio_repair_workers =
2231 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2232 fs_info->rmw_workers =
2233 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2234 fs_info->endio_write_workers =
2235 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2237 fs_info->endio_freespace_worker =
2238 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2240 fs_info->delayed_workers =
2241 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2243 fs_info->readahead_workers =
2244 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2246 fs_info->qgroup_rescan_workers =
2247 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2248 fs_info->extent_workers =
2249 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2250 min_t(u64, fs_devices->num_devices,
2253 if (!(fs_info->workers && fs_info->delalloc_workers &&
2254 fs_info->submit_workers && fs_info->flush_workers &&
2255 fs_info->endio_workers && fs_info->endio_meta_workers &&
2256 fs_info->endio_meta_write_workers &&
2257 fs_info->endio_repair_workers &&
2258 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2259 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2260 fs_info->caching_workers && fs_info->readahead_workers &&
2261 fs_info->fixup_workers && fs_info->delayed_workers &&
2262 fs_info->extent_workers &&
2263 fs_info->qgroup_rescan_workers)) {
2270 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2271 struct btrfs_fs_devices *fs_devices)
2274 struct btrfs_root *log_tree_root;
2275 struct btrfs_super_block *disk_super = fs_info->super_copy;
2276 u64 bytenr = btrfs_super_log_root(disk_super);
2278 if (fs_devices->rw_devices == 0) {
2279 btrfs_warn(fs_info, "log replay required on RO media");
2283 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2287 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2289 log_tree_root->node = read_tree_block(fs_info, bytenr,
2290 fs_info->generation + 1);
2291 if (IS_ERR(log_tree_root->node)) {
2292 btrfs_warn(fs_info, "failed to read log tree");
2293 ret = PTR_ERR(log_tree_root->node);
2294 kfree(log_tree_root);
2296 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2297 btrfs_err(fs_info, "failed to read log tree");
2298 free_extent_buffer(log_tree_root->node);
2299 kfree(log_tree_root);
2302 /* returns with log_tree_root freed on success */
2303 ret = btrfs_recover_log_trees(log_tree_root);
2305 btrfs_handle_fs_error(fs_info, ret,
2306 "Failed to recover log tree");
2307 free_extent_buffer(log_tree_root->node);
2308 kfree(log_tree_root);
2312 if (sb_rdonly(fs_info->sb)) {
2313 ret = btrfs_commit_super(fs_info);
2321 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2323 struct btrfs_root *tree_root = fs_info->tree_root;
2324 struct btrfs_root *root;
2325 struct btrfs_key location;
2328 BUG_ON(!fs_info->tree_root);
2330 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2331 location.type = BTRFS_ROOT_ITEM_KEY;
2332 location.offset = 0;
2334 root = btrfs_read_tree_root(tree_root, &location);
2336 return PTR_ERR(root);
2337 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2338 fs_info->extent_root = root;
2340 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2341 root = btrfs_read_tree_root(tree_root, &location);
2343 return PTR_ERR(root);
2344 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2345 fs_info->dev_root = root;
2346 btrfs_init_devices_late(fs_info);
2348 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2349 root = btrfs_read_tree_root(tree_root, &location);
2351 return PTR_ERR(root);
2352 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2353 fs_info->csum_root = root;
2355 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2356 root = btrfs_read_tree_root(tree_root, &location);
2357 if (!IS_ERR(root)) {
2358 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2359 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2360 fs_info->quota_root = root;
2363 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2364 root = btrfs_read_tree_root(tree_root, &location);
2366 ret = PTR_ERR(root);
2370 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2371 fs_info->uuid_root = root;
2374 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2375 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2376 root = btrfs_read_tree_root(tree_root, &location);
2378 return PTR_ERR(root);
2379 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2380 fs_info->free_space_root = root;
2386 int open_ctree(struct super_block *sb,
2387 struct btrfs_fs_devices *fs_devices,
2395 struct btrfs_key location;
2396 struct buffer_head *bh;
2397 struct btrfs_super_block *disk_super;
2398 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2399 struct btrfs_root *tree_root;
2400 struct btrfs_root *chunk_root;
2403 int num_backups_tried = 0;
2404 int backup_index = 0;
2406 int clear_free_space_tree = 0;
2408 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2409 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2410 if (!tree_root || !chunk_root) {
2415 ret = init_srcu_struct(&fs_info->subvol_srcu);
2421 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2426 fs_info->dirty_metadata_batch = PAGE_SIZE *
2427 (1 + ilog2(nr_cpu_ids));
2429 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2432 goto fail_dirty_metadata_bytes;
2435 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2438 goto fail_delalloc_bytes;
2441 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2442 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2443 INIT_LIST_HEAD(&fs_info->trans_list);
2444 INIT_LIST_HEAD(&fs_info->dead_roots);
2445 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2446 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2447 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2448 spin_lock_init(&fs_info->delalloc_root_lock);
2449 spin_lock_init(&fs_info->trans_lock);
2450 spin_lock_init(&fs_info->fs_roots_radix_lock);
2451 spin_lock_init(&fs_info->delayed_iput_lock);
2452 spin_lock_init(&fs_info->defrag_inodes_lock);
2453 spin_lock_init(&fs_info->tree_mod_seq_lock);
2454 spin_lock_init(&fs_info->super_lock);
2455 spin_lock_init(&fs_info->qgroup_op_lock);
2456 spin_lock_init(&fs_info->buffer_lock);
2457 spin_lock_init(&fs_info->unused_bgs_lock);
2458 rwlock_init(&fs_info->tree_mod_log_lock);
2459 mutex_init(&fs_info->unused_bg_unpin_mutex);
2460 mutex_init(&fs_info->delete_unused_bgs_mutex);
2461 mutex_init(&fs_info->reloc_mutex);
2462 mutex_init(&fs_info->delalloc_root_mutex);
2463 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2464 seqlock_init(&fs_info->profiles_lock);
2466 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2467 INIT_LIST_HEAD(&fs_info->space_info);
2468 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2469 INIT_LIST_HEAD(&fs_info->unused_bgs);
2470 btrfs_mapping_init(&fs_info->mapping_tree);
2471 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2472 BTRFS_BLOCK_RSV_GLOBAL);
2473 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2474 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2475 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2476 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2477 BTRFS_BLOCK_RSV_DELOPS);
2478 atomic_set(&fs_info->async_delalloc_pages, 0);
2479 atomic_set(&fs_info->defrag_running, 0);
2480 atomic_set(&fs_info->qgroup_op_seq, 0);
2481 atomic_set(&fs_info->reada_works_cnt, 0);
2482 atomic64_set(&fs_info->tree_mod_seq, 0);
2484 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2485 fs_info->metadata_ratio = 0;
2486 fs_info->defrag_inodes = RB_ROOT;
2487 atomic64_set(&fs_info->free_chunk_space, 0);
2488 fs_info->tree_mod_log = RB_ROOT;
2489 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2490 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2491 /* readahead state */
2492 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2493 spin_lock_init(&fs_info->reada_lock);
2494 btrfs_init_ref_verify(fs_info);
2496 fs_info->thread_pool_size = min_t(unsigned long,
2497 num_online_cpus() + 2, 8);
2499 INIT_LIST_HEAD(&fs_info->ordered_roots);
2500 spin_lock_init(&fs_info->ordered_root_lock);
2502 fs_info->btree_inode = new_inode(sb);
2503 if (!fs_info->btree_inode) {
2505 goto fail_bio_counter;
2507 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2509 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2511 if (!fs_info->delayed_root) {
2515 btrfs_init_delayed_root(fs_info->delayed_root);
2517 btrfs_init_scrub(fs_info);
2518 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2519 fs_info->check_integrity_print_mask = 0;
2521 btrfs_init_balance(fs_info);
2522 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2524 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2525 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2527 btrfs_init_btree_inode(fs_info);
2529 spin_lock_init(&fs_info->block_group_cache_lock);
2530 fs_info->block_group_cache_tree = RB_ROOT;
2531 fs_info->first_logical_byte = (u64)-1;
2533 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2534 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2535 fs_info->pinned_extents = &fs_info->freed_extents[0];
2536 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2538 mutex_init(&fs_info->ordered_operations_mutex);
2539 mutex_init(&fs_info->tree_log_mutex);
2540 mutex_init(&fs_info->chunk_mutex);
2541 mutex_init(&fs_info->transaction_kthread_mutex);
2542 mutex_init(&fs_info->cleaner_mutex);
2543 mutex_init(&fs_info->volume_mutex);
2544 mutex_init(&fs_info->ro_block_group_mutex);
2545 init_rwsem(&fs_info->commit_root_sem);
2546 init_rwsem(&fs_info->cleanup_work_sem);
2547 init_rwsem(&fs_info->subvol_sem);
2548 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2550 btrfs_init_dev_replace_locks(fs_info);
2551 btrfs_init_qgroup(fs_info);
2553 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2554 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2556 init_waitqueue_head(&fs_info->transaction_throttle);
2557 init_waitqueue_head(&fs_info->transaction_wait);
2558 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2559 init_waitqueue_head(&fs_info->async_submit_wait);
2561 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2563 /* Usable values until the real ones are cached from the superblock */
2564 fs_info->nodesize = 4096;
2565 fs_info->sectorsize = 4096;
2566 fs_info->stripesize = 4096;
2568 ret = btrfs_alloc_stripe_hash_table(fs_info);
2574 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2576 invalidate_bdev(fs_devices->latest_bdev);
2579 * Read super block and check the signature bytes only
2581 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2588 * We want to check superblock checksum, the type is stored inside.
2589 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2591 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2592 btrfs_err(fs_info, "superblock checksum mismatch");
2599 * super_copy is zeroed at allocation time and we never touch the
2600 * following bytes up to INFO_SIZE, the checksum is calculated from
2601 * the whole block of INFO_SIZE
2603 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2604 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2605 sizeof(*fs_info->super_for_commit));
2608 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2610 ret = btrfs_check_super_valid(fs_info);
2612 btrfs_err(fs_info, "superblock contains fatal errors");
2617 disk_super = fs_info->super_copy;
2618 if (!btrfs_super_root(disk_super))
2621 /* check FS state, whether FS is broken. */
2622 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2623 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2626 * run through our array of backup supers and setup
2627 * our ring pointer to the oldest one
2629 generation = btrfs_super_generation(disk_super);
2630 find_oldest_super_backup(fs_info, generation);
2633 * In the long term, we'll store the compression type in the super
2634 * block, and it'll be used for per file compression control.
2636 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2638 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2644 features = btrfs_super_incompat_flags(disk_super) &
2645 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2648 "cannot mount because of unsupported optional features (%llx)",
2654 features = btrfs_super_incompat_flags(disk_super);
2655 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2656 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2657 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2658 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2659 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2661 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2662 btrfs_info(fs_info, "has skinny extents");
2665 * flag our filesystem as having big metadata blocks if
2666 * they are bigger than the page size
2668 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2669 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2671 "flagging fs with big metadata feature");
2672 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2675 nodesize = btrfs_super_nodesize(disk_super);
2676 sectorsize = btrfs_super_sectorsize(disk_super);
2677 stripesize = sectorsize;
2678 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2679 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2681 /* Cache block sizes */
2682 fs_info->nodesize = nodesize;
2683 fs_info->sectorsize = sectorsize;
2684 fs_info->stripesize = stripesize;
2687 * mixed block groups end up with duplicate but slightly offset
2688 * extent buffers for the same range. It leads to corruptions
2690 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2691 (sectorsize != nodesize)) {
2693 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2694 nodesize, sectorsize);
2699 * Needn't use the lock because there is no other task which will
2702 btrfs_set_super_incompat_flags(disk_super, features);
2704 features = btrfs_super_compat_ro_flags(disk_super) &
2705 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2706 if (!sb_rdonly(sb) && features) {
2708 "cannot mount read-write because of unsupported optional features (%llx)",
2714 max_active = fs_info->thread_pool_size;
2716 ret = btrfs_init_workqueues(fs_info, fs_devices);
2719 goto fail_sb_buffer;
2722 sb->s_bdi->congested_fn = btrfs_congested_fn;
2723 sb->s_bdi->congested_data = fs_info;
2724 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2725 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2726 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2727 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2729 sb->s_blocksize = sectorsize;
2730 sb->s_blocksize_bits = blksize_bits(sectorsize);
2731 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2733 mutex_lock(&fs_info->chunk_mutex);
2734 ret = btrfs_read_sys_array(fs_info);
2735 mutex_unlock(&fs_info->chunk_mutex);
2737 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2738 goto fail_sb_buffer;
2741 generation = btrfs_super_chunk_root_generation(disk_super);
2743 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2745 chunk_root->node = read_tree_block(fs_info,
2746 btrfs_super_chunk_root(disk_super),
2748 if (IS_ERR(chunk_root->node) ||
2749 !extent_buffer_uptodate(chunk_root->node)) {
2750 btrfs_err(fs_info, "failed to read chunk root");
2751 if (!IS_ERR(chunk_root->node))
2752 free_extent_buffer(chunk_root->node);
2753 chunk_root->node = NULL;
2754 goto fail_tree_roots;
2756 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2757 chunk_root->commit_root = btrfs_root_node(chunk_root);
2759 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2760 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2762 ret = btrfs_read_chunk_tree(fs_info);
2764 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2765 goto fail_tree_roots;
2769 * keep the device that is marked to be the target device for the
2770 * dev_replace procedure
2772 btrfs_close_extra_devices(fs_devices, 0);
2774 if (!fs_devices->latest_bdev) {
2775 btrfs_err(fs_info, "failed to read devices");
2776 goto fail_tree_roots;
2780 generation = btrfs_super_generation(disk_super);
2782 tree_root->node = read_tree_block(fs_info,
2783 btrfs_super_root(disk_super),
2785 if (IS_ERR(tree_root->node) ||
2786 !extent_buffer_uptodate(tree_root->node)) {
2787 btrfs_warn(fs_info, "failed to read tree root");
2788 if (!IS_ERR(tree_root->node))
2789 free_extent_buffer(tree_root->node);
2790 tree_root->node = NULL;
2791 goto recovery_tree_root;
2794 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2795 tree_root->commit_root = btrfs_root_node(tree_root);
2796 btrfs_set_root_refs(&tree_root->root_item, 1);
2798 mutex_lock(&tree_root->objectid_mutex);
2799 ret = btrfs_find_highest_objectid(tree_root,
2800 &tree_root->highest_objectid);
2802 mutex_unlock(&tree_root->objectid_mutex);
2803 goto recovery_tree_root;
2806 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2808 mutex_unlock(&tree_root->objectid_mutex);
2810 ret = btrfs_read_roots(fs_info);
2812 goto recovery_tree_root;
2814 fs_info->generation = generation;
2815 fs_info->last_trans_committed = generation;
2817 ret = btrfs_recover_balance(fs_info);
2819 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2820 goto fail_block_groups;
2823 ret = btrfs_init_dev_stats(fs_info);
2825 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2826 goto fail_block_groups;
2829 ret = btrfs_init_dev_replace(fs_info);
2831 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2832 goto fail_block_groups;
2835 btrfs_close_extra_devices(fs_devices, 1);
2837 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2839 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2841 goto fail_block_groups;
2844 ret = btrfs_sysfs_add_device(fs_devices);
2846 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2848 goto fail_fsdev_sysfs;
2851 ret = btrfs_sysfs_add_mounted(fs_info);
2853 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2854 goto fail_fsdev_sysfs;
2857 ret = btrfs_init_space_info(fs_info);
2859 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2863 ret = btrfs_read_block_groups(fs_info);
2865 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2869 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info)) {
2871 "writeable mount is not allowed due to too many missing devices");
2875 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2877 if (IS_ERR(fs_info->cleaner_kthread))
2880 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2882 "btrfs-transaction");
2883 if (IS_ERR(fs_info->transaction_kthread))
2886 if (!btrfs_test_opt(fs_info, NOSSD) &&
2887 !fs_info->fs_devices->rotating) {
2888 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2892 * Mount does not set all options immediately, we can do it now and do
2893 * not have to wait for transaction commit
2895 btrfs_apply_pending_changes(fs_info);
2897 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2898 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2899 ret = btrfsic_mount(fs_info, fs_devices,
2900 btrfs_test_opt(fs_info,
2901 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2903 fs_info->check_integrity_print_mask);
2906 "failed to initialize integrity check module: %d",
2910 ret = btrfs_read_qgroup_config(fs_info);
2912 goto fail_trans_kthread;
2914 if (btrfs_build_ref_tree(fs_info))
2915 btrfs_err(fs_info, "couldn't build ref tree");
2917 /* do not make disk changes in broken FS or nologreplay is given */
2918 if (btrfs_super_log_root(disk_super) != 0 &&
2919 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2920 ret = btrfs_replay_log(fs_info, fs_devices);
2927 ret = btrfs_find_orphan_roots(fs_info);
2931 if (!sb_rdonly(sb)) {
2932 ret = btrfs_cleanup_fs_roots(fs_info);
2936 mutex_lock(&fs_info->cleaner_mutex);
2937 ret = btrfs_recover_relocation(tree_root);
2938 mutex_unlock(&fs_info->cleaner_mutex);
2940 btrfs_warn(fs_info, "failed to recover relocation: %d",
2947 location.objectid = BTRFS_FS_TREE_OBJECTID;
2948 location.type = BTRFS_ROOT_ITEM_KEY;
2949 location.offset = 0;
2951 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2952 if (IS_ERR(fs_info->fs_root)) {
2953 err = PTR_ERR(fs_info->fs_root);
2960 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2961 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2962 clear_free_space_tree = 1;
2963 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2964 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2965 btrfs_warn(fs_info, "free space tree is invalid");
2966 clear_free_space_tree = 1;
2969 if (clear_free_space_tree) {
2970 btrfs_info(fs_info, "clearing free space tree");
2971 ret = btrfs_clear_free_space_tree(fs_info);
2974 "failed to clear free space tree: %d", ret);
2975 close_ctree(fs_info);
2980 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
2981 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2982 btrfs_info(fs_info, "creating free space tree");
2983 ret = btrfs_create_free_space_tree(fs_info);
2986 "failed to create free space tree: %d", ret);
2987 close_ctree(fs_info);
2992 down_read(&fs_info->cleanup_work_sem);
2993 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2994 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2995 up_read(&fs_info->cleanup_work_sem);
2996 close_ctree(fs_info);
2999 up_read(&fs_info->cleanup_work_sem);
3001 ret = btrfs_resume_balance_async(fs_info);
3003 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3004 close_ctree(fs_info);
3008 ret = btrfs_resume_dev_replace_async(fs_info);
3010 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3011 close_ctree(fs_info);
3015 btrfs_qgroup_rescan_resume(fs_info);
3017 if (!fs_info->uuid_root) {
3018 btrfs_info(fs_info, "creating UUID tree");
3019 ret = btrfs_create_uuid_tree(fs_info);
3022 "failed to create the UUID tree: %d", ret);
3023 close_ctree(fs_info);
3026 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3027 fs_info->generation !=
3028 btrfs_super_uuid_tree_generation(disk_super)) {
3029 btrfs_info(fs_info, "checking UUID tree");
3030 ret = btrfs_check_uuid_tree(fs_info);
3033 "failed to check the UUID tree: %d", ret);
3034 close_ctree(fs_info);
3038 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3040 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3043 * backuproot only affect mount behavior, and if open_ctree succeeded,
3044 * no need to keep the flag
3046 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3051 btrfs_free_qgroup_config(fs_info);
3053 kthread_stop(fs_info->transaction_kthread);
3054 btrfs_cleanup_transaction(fs_info);
3055 btrfs_free_fs_roots(fs_info);
3057 kthread_stop(fs_info->cleaner_kthread);
3060 * make sure we're done with the btree inode before we stop our
3063 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3066 btrfs_sysfs_remove_mounted(fs_info);
3069 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3072 btrfs_put_block_group_cache(fs_info);
3075 free_root_pointers(fs_info, 1);
3076 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3079 btrfs_stop_all_workers(fs_info);
3080 btrfs_free_block_groups(fs_info);
3083 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3085 iput(fs_info->btree_inode);
3087 percpu_counter_destroy(&fs_info->bio_counter);
3088 fail_delalloc_bytes:
3089 percpu_counter_destroy(&fs_info->delalloc_bytes);
3090 fail_dirty_metadata_bytes:
3091 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3093 cleanup_srcu_struct(&fs_info->subvol_srcu);
3095 btrfs_free_stripe_hash_table(fs_info);
3096 btrfs_close_devices(fs_info->fs_devices);
3100 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3101 goto fail_tree_roots;
3103 free_root_pointers(fs_info, 0);
3105 /* don't use the log in recovery mode, it won't be valid */
3106 btrfs_set_super_log_root(disk_super, 0);
3108 /* we can't trust the free space cache either */
3109 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3111 ret = next_root_backup(fs_info, fs_info->super_copy,
3112 &num_backups_tried, &backup_index);
3114 goto fail_block_groups;
3115 goto retry_root_backup;
3118 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3121 set_buffer_uptodate(bh);
3123 struct btrfs_device *device = (struct btrfs_device *)
3126 btrfs_warn_rl_in_rcu(device->fs_info,
3127 "lost page write due to IO error on %s",
3128 rcu_str_deref(device->name));
3129 /* note, we don't set_buffer_write_io_error because we have
3130 * our own ways of dealing with the IO errors
3132 clear_buffer_uptodate(bh);
3133 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3139 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3140 struct buffer_head **bh_ret)
3142 struct buffer_head *bh;
3143 struct btrfs_super_block *super;
3146 bytenr = btrfs_sb_offset(copy_num);
3147 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3150 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3152 * If we fail to read from the underlying devices, as of now
3153 * the best option we have is to mark it EIO.
3158 super = (struct btrfs_super_block *)bh->b_data;
3159 if (btrfs_super_bytenr(super) != bytenr ||
3160 btrfs_super_magic(super) != BTRFS_MAGIC) {
3170 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3172 struct buffer_head *bh;
3173 struct buffer_head *latest = NULL;
3174 struct btrfs_super_block *super;
3179 /* we would like to check all the supers, but that would make
3180 * a btrfs mount succeed after a mkfs from a different FS.
3181 * So, we need to add a special mount option to scan for
3182 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3184 for (i = 0; i < 1; i++) {
3185 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3189 super = (struct btrfs_super_block *)bh->b_data;
3191 if (!latest || btrfs_super_generation(super) > transid) {
3194 transid = btrfs_super_generation(super);
3201 return ERR_PTR(ret);
3207 * Write superblock @sb to the @device. Do not wait for completion, all the
3208 * buffer heads we write are pinned.
3210 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3211 * the expected device size at commit time. Note that max_mirrors must be
3212 * same for write and wait phases.
3214 * Return number of errors when buffer head is not found or submission fails.
3216 static int write_dev_supers(struct btrfs_device *device,
3217 struct btrfs_super_block *sb, int max_mirrors)
3219 struct buffer_head *bh;
3227 if (max_mirrors == 0)
3228 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3230 for (i = 0; i < max_mirrors; i++) {
3231 bytenr = btrfs_sb_offset(i);
3232 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3233 device->commit_total_bytes)
3236 btrfs_set_super_bytenr(sb, bytenr);
3239 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3240 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3241 btrfs_csum_final(crc, sb->csum);
3243 /* One reference for us, and we leave it for the caller */
3244 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3245 BTRFS_SUPER_INFO_SIZE);
3247 btrfs_err(device->fs_info,
3248 "couldn't get super buffer head for bytenr %llu",
3254 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3256 /* one reference for submit_bh */
3259 set_buffer_uptodate(bh);
3261 bh->b_end_io = btrfs_end_buffer_write_sync;
3262 bh->b_private = device;
3265 * we fua the first super. The others we allow
3268 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3269 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3270 op_flags |= REQ_FUA;
3271 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3275 return errors < i ? 0 : -1;
3279 * Wait for write completion of superblocks done by write_dev_supers,
3280 * @max_mirrors same for write and wait phases.
3282 * Return number of errors when buffer head is not found or not marked up to
3285 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3287 struct buffer_head *bh;
3292 if (max_mirrors == 0)
3293 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3295 for (i = 0; i < max_mirrors; i++) {
3296 bytenr = btrfs_sb_offset(i);
3297 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3298 device->commit_total_bytes)
3301 bh = __find_get_block(device->bdev,
3302 bytenr / BTRFS_BDEV_BLOCKSIZE,
3303 BTRFS_SUPER_INFO_SIZE);
3309 if (!buffer_uptodate(bh))
3312 /* drop our reference */
3315 /* drop the reference from the writing run */
3319 return errors < i ? 0 : -1;
3323 * endio for the write_dev_flush, this will wake anyone waiting
3324 * for the barrier when it is done
3326 static void btrfs_end_empty_barrier(struct bio *bio)
3328 complete(bio->bi_private);
3332 * Submit a flush request to the device if it supports it. Error handling is
3333 * done in the waiting counterpart.
3335 static void write_dev_flush(struct btrfs_device *device)
3337 struct request_queue *q = bdev_get_queue(device->bdev);
3338 struct bio *bio = device->flush_bio;
3340 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3344 bio->bi_end_io = btrfs_end_empty_barrier;
3345 bio_set_dev(bio, device->bdev);
3346 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3347 init_completion(&device->flush_wait);
3348 bio->bi_private = &device->flush_wait;
3350 btrfsic_submit_bio(bio);
3351 device->flush_bio_sent = 1;
3355 * If the flush bio has been submitted by write_dev_flush, wait for it.
3357 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3359 struct bio *bio = device->flush_bio;
3361 if (!device->flush_bio_sent)
3364 device->flush_bio_sent = 0;
3365 wait_for_completion_io(&device->flush_wait);
3367 return bio->bi_status;
3370 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3372 if (!btrfs_check_rw_degradable(fs_info))
3378 * send an empty flush down to each device in parallel,
3379 * then wait for them
3381 static int barrier_all_devices(struct btrfs_fs_info *info)
3383 struct list_head *head;
3384 struct btrfs_device *dev;
3385 int errors_wait = 0;
3388 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3389 /* send down all the barriers */
3390 head = &info->fs_devices->devices;
3391 list_for_each_entry(dev, head, dev_list) {
3392 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3396 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3397 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3400 write_dev_flush(dev);
3401 dev->last_flush_error = BLK_STS_OK;
3404 /* wait for all the barriers */
3405 list_for_each_entry(dev, head, dev_list) {
3406 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3412 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3413 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3416 ret = wait_dev_flush(dev);
3418 dev->last_flush_error = ret;
3419 btrfs_dev_stat_inc_and_print(dev,
3420 BTRFS_DEV_STAT_FLUSH_ERRS);
3427 * At some point we need the status of all disks
3428 * to arrive at the volume status. So error checking
3429 * is being pushed to a separate loop.
3431 return check_barrier_error(info);
3436 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3439 int min_tolerated = INT_MAX;
3441 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3442 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3443 min_tolerated = min(min_tolerated,
3444 btrfs_raid_array[BTRFS_RAID_SINGLE].
3445 tolerated_failures);
3447 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3448 if (raid_type == BTRFS_RAID_SINGLE)
3450 if (!(flags & btrfs_raid_group[raid_type]))
3452 min_tolerated = min(min_tolerated,
3453 btrfs_raid_array[raid_type].
3454 tolerated_failures);
3457 if (min_tolerated == INT_MAX) {
3458 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3462 return min_tolerated;
3465 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3467 struct list_head *head;
3468 struct btrfs_device *dev;
3469 struct btrfs_super_block *sb;
3470 struct btrfs_dev_item *dev_item;
3474 int total_errors = 0;
3477 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3480 * max_mirrors == 0 indicates we're from commit_transaction,
3481 * not from fsync where the tree roots in fs_info have not
3482 * been consistent on disk.
3484 if (max_mirrors == 0)
3485 backup_super_roots(fs_info);
3487 sb = fs_info->super_for_commit;
3488 dev_item = &sb->dev_item;
3490 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3491 head = &fs_info->fs_devices->devices;
3492 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3495 ret = barrier_all_devices(fs_info);
3498 &fs_info->fs_devices->device_list_mutex);
3499 btrfs_handle_fs_error(fs_info, ret,
3500 "errors while submitting device barriers.");
3505 list_for_each_entry(dev, head, dev_list) {
3510 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3511 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3514 btrfs_set_stack_device_generation(dev_item, 0);
3515 btrfs_set_stack_device_type(dev_item, dev->type);
3516 btrfs_set_stack_device_id(dev_item, dev->devid);
3517 btrfs_set_stack_device_total_bytes(dev_item,
3518 dev->commit_total_bytes);
3519 btrfs_set_stack_device_bytes_used(dev_item,
3520 dev->commit_bytes_used);
3521 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3522 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3523 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3524 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3525 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3527 flags = btrfs_super_flags(sb);
3528 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3530 ret = write_dev_supers(dev, sb, max_mirrors);
3534 if (total_errors > max_errors) {
3535 btrfs_err(fs_info, "%d errors while writing supers",
3537 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3539 /* FUA is masked off if unsupported and can't be the reason */
3540 btrfs_handle_fs_error(fs_info, -EIO,
3541 "%d errors while writing supers",
3547 list_for_each_entry(dev, head, dev_list) {
3550 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3551 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3554 ret = wait_dev_supers(dev, max_mirrors);
3558 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3559 if (total_errors > max_errors) {
3560 btrfs_handle_fs_error(fs_info, -EIO,
3561 "%d errors while writing supers",
3568 /* Drop a fs root from the radix tree and free it. */
3569 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3570 struct btrfs_root *root)
3572 spin_lock(&fs_info->fs_roots_radix_lock);
3573 radix_tree_delete(&fs_info->fs_roots_radix,
3574 (unsigned long)root->root_key.objectid);
3575 spin_unlock(&fs_info->fs_roots_radix_lock);
3577 if (btrfs_root_refs(&root->root_item) == 0)
3578 synchronize_srcu(&fs_info->subvol_srcu);
3580 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3581 btrfs_free_log(NULL, root);
3582 if (root->reloc_root) {
3583 free_extent_buffer(root->reloc_root->node);
3584 free_extent_buffer(root->reloc_root->commit_root);
3585 btrfs_put_fs_root(root->reloc_root);
3586 root->reloc_root = NULL;
3590 if (root->free_ino_pinned)
3591 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3592 if (root->free_ino_ctl)
3593 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3597 static void free_fs_root(struct btrfs_root *root)
3599 iput(root->ino_cache_inode);
3600 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3601 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3602 root->orphan_block_rsv = NULL;
3604 free_anon_bdev(root->anon_dev);
3605 if (root->subv_writers)
3606 btrfs_free_subvolume_writers(root->subv_writers);
3607 free_extent_buffer(root->node);
3608 free_extent_buffer(root->commit_root);
3609 kfree(root->free_ino_ctl);
3610 kfree(root->free_ino_pinned);
3612 btrfs_put_fs_root(root);
3615 void btrfs_free_fs_root(struct btrfs_root *root)
3620 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3622 u64 root_objectid = 0;
3623 struct btrfs_root *gang[8];
3626 unsigned int ret = 0;
3630 index = srcu_read_lock(&fs_info->subvol_srcu);
3631 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3632 (void **)gang, root_objectid,
3635 srcu_read_unlock(&fs_info->subvol_srcu, index);
3638 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3640 for (i = 0; i < ret; i++) {
3641 /* Avoid to grab roots in dead_roots */
3642 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3646 /* grab all the search result for later use */
3647 gang[i] = btrfs_grab_fs_root(gang[i]);
3649 srcu_read_unlock(&fs_info->subvol_srcu, index);
3651 for (i = 0; i < ret; i++) {
3654 root_objectid = gang[i]->root_key.objectid;
3655 err = btrfs_orphan_cleanup(gang[i]);
3658 btrfs_put_fs_root(gang[i]);
3663 /* release the uncleaned roots due to error */
3664 for (; i < ret; i++) {
3666 btrfs_put_fs_root(gang[i]);
3671 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3673 struct btrfs_root *root = fs_info->tree_root;
3674 struct btrfs_trans_handle *trans;
3676 mutex_lock(&fs_info->cleaner_mutex);
3677 btrfs_run_delayed_iputs(fs_info);
3678 mutex_unlock(&fs_info->cleaner_mutex);
3679 wake_up_process(fs_info->cleaner_kthread);
3681 /* wait until ongoing cleanup work done */
3682 down_write(&fs_info->cleanup_work_sem);
3683 up_write(&fs_info->cleanup_work_sem);
3685 trans = btrfs_join_transaction(root);
3687 return PTR_ERR(trans);
3688 return btrfs_commit_transaction(trans);
3691 void close_ctree(struct btrfs_fs_info *fs_info)
3693 struct btrfs_root *root = fs_info->tree_root;
3696 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3698 /* wait for the qgroup rescan worker to stop */
3699 btrfs_qgroup_wait_for_completion(fs_info, false);
3701 /* wait for the uuid_scan task to finish */
3702 down(&fs_info->uuid_tree_rescan_sem);
3703 /* avoid complains from lockdep et al., set sem back to initial state */
3704 up(&fs_info->uuid_tree_rescan_sem);
3706 /* pause restriper - we want to resume on mount */
3707 btrfs_pause_balance(fs_info);
3709 btrfs_dev_replace_suspend_for_unmount(fs_info);
3711 btrfs_scrub_cancel(fs_info);
3713 /* wait for any defraggers to finish */
3714 wait_event(fs_info->transaction_wait,
3715 (atomic_read(&fs_info->defrag_running) == 0));
3717 /* clear out the rbtree of defraggable inodes */
3718 btrfs_cleanup_defrag_inodes(fs_info);
3720 cancel_work_sync(&fs_info->async_reclaim_work);
3722 if (!sb_rdonly(fs_info->sb)) {
3724 * If the cleaner thread is stopped and there are
3725 * block groups queued for removal, the deletion will be
3726 * skipped when we quit the cleaner thread.
3728 btrfs_delete_unused_bgs(fs_info);
3730 ret = btrfs_commit_super(fs_info);
3732 btrfs_err(fs_info, "commit super ret %d", ret);
3735 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3736 btrfs_error_commit_super(fs_info);
3738 kthread_stop(fs_info->transaction_kthread);
3739 kthread_stop(fs_info->cleaner_kthread);
3741 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3743 btrfs_free_qgroup_config(fs_info);
3745 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3746 btrfs_info(fs_info, "at unmount delalloc count %lld",
3747 percpu_counter_sum(&fs_info->delalloc_bytes));
3750 btrfs_sysfs_remove_mounted(fs_info);
3751 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3753 btrfs_free_fs_roots(fs_info);
3755 btrfs_put_block_group_cache(fs_info);
3758 * we must make sure there is not any read request to
3759 * submit after we stopping all workers.
3761 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3762 btrfs_stop_all_workers(fs_info);
3764 btrfs_free_block_groups(fs_info);
3766 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3767 free_root_pointers(fs_info, 1);
3769 iput(fs_info->btree_inode);
3771 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3772 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3773 btrfsic_unmount(fs_info->fs_devices);
3776 btrfs_close_devices(fs_info->fs_devices);
3777 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3779 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3780 percpu_counter_destroy(&fs_info->delalloc_bytes);
3781 percpu_counter_destroy(&fs_info->bio_counter);
3782 cleanup_srcu_struct(&fs_info->subvol_srcu);
3784 btrfs_free_stripe_hash_table(fs_info);
3785 btrfs_free_ref_cache(fs_info);
3787 __btrfs_free_block_rsv(root->orphan_block_rsv);
3788 root->orphan_block_rsv = NULL;
3790 while (!list_empty(&fs_info->pinned_chunks)) {
3791 struct extent_map *em;
3793 em = list_first_entry(&fs_info->pinned_chunks,
3794 struct extent_map, list);
3795 list_del_init(&em->list);
3796 free_extent_map(em);
3800 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3804 struct inode *btree_inode = buf->pages[0]->mapping->host;
3806 ret = extent_buffer_uptodate(buf);
3810 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3811 parent_transid, atomic);
3817 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3819 struct btrfs_fs_info *fs_info;
3820 struct btrfs_root *root;
3821 u64 transid = btrfs_header_generation(buf);
3824 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3826 * This is a fast path so only do this check if we have sanity tests
3827 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3828 * outside of the sanity tests.
3830 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3833 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3834 fs_info = root->fs_info;
3835 btrfs_assert_tree_locked(buf);
3836 if (transid != fs_info->generation)
3837 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3838 buf->start, transid, fs_info->generation);
3839 was_dirty = set_extent_buffer_dirty(buf);
3841 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3843 fs_info->dirty_metadata_batch);
3844 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3846 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3847 * but item data not updated.
3848 * So here we should only check item pointers, not item data.
3850 if (btrfs_header_level(buf) == 0 &&
3851 btrfs_check_leaf_relaxed(root, buf)) {
3852 btrfs_print_leaf(buf);
3858 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
3862 * looks as though older kernels can get into trouble with
3863 * this code, they end up stuck in balance_dirty_pages forever
3867 if (current->flags & PF_MEMALLOC)
3871 btrfs_balance_delayed_items(fs_info);
3873 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
3874 BTRFS_DIRTY_METADATA_THRESH);
3876 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
3880 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
3882 __btrfs_btree_balance_dirty(fs_info, 1);
3885 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
3887 __btrfs_btree_balance_dirty(fs_info, 0);
3890 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3892 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3893 struct btrfs_fs_info *fs_info = root->fs_info;
3895 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
3898 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
3900 struct btrfs_super_block *sb = fs_info->super_copy;
3901 u64 nodesize = btrfs_super_nodesize(sb);
3902 u64 sectorsize = btrfs_super_sectorsize(sb);
3905 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
3906 btrfs_err(fs_info, "no valid FS found");
3909 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
3910 btrfs_warn(fs_info, "unrecognized super flag: %llu",
3911 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
3912 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3913 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
3914 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3917 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3918 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
3919 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3922 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3923 btrfs_err(fs_info, "log_root level too big: %d >= %d",
3924 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3929 * Check sectorsize and nodesize first, other check will need it.
3930 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
3932 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
3933 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3934 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
3937 /* Only PAGE SIZE is supported yet */
3938 if (sectorsize != PAGE_SIZE) {
3940 "sectorsize %llu not supported yet, only support %lu",
3941 sectorsize, PAGE_SIZE);
3944 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
3945 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3946 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
3949 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
3950 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
3951 le32_to_cpu(sb->__unused_leafsize), nodesize);
3955 /* Root alignment check */
3956 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
3957 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
3958 btrfs_super_root(sb));
3961 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
3962 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
3963 btrfs_super_chunk_root(sb));
3966 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
3967 btrfs_warn(fs_info, "log_root block unaligned: %llu",
3968 btrfs_super_log_root(sb));
3972 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
3974 "dev_item UUID does not match fsid: %pU != %pU",
3975 fs_info->fsid, sb->dev_item.fsid);
3980 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3983 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
3984 btrfs_err(fs_info, "bytes_used is too small %llu",
3985 btrfs_super_bytes_used(sb));
3988 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
3989 btrfs_err(fs_info, "invalid stripesize %u",
3990 btrfs_super_stripesize(sb));
3993 if (btrfs_super_num_devices(sb) > (1UL << 31))
3994 btrfs_warn(fs_info, "suspicious number of devices: %llu",
3995 btrfs_super_num_devices(sb));
3996 if (btrfs_super_num_devices(sb) == 0) {
3997 btrfs_err(fs_info, "number of devices is 0");
4001 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4002 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4003 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4008 * Obvious sys_chunk_array corruptions, it must hold at least one key
4011 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4012 btrfs_err(fs_info, "system chunk array too big %u > %u",
4013 btrfs_super_sys_array_size(sb),
4014 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4017 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4018 + sizeof(struct btrfs_chunk)) {
4019 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4020 btrfs_super_sys_array_size(sb),
4021 sizeof(struct btrfs_disk_key)
4022 + sizeof(struct btrfs_chunk));
4027 * The generation is a global counter, we'll trust it more than the others
4028 * but it's still possible that it's the one that's wrong.
4030 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4032 "suspicious: generation < chunk_root_generation: %llu < %llu",
4033 btrfs_super_generation(sb),
4034 btrfs_super_chunk_root_generation(sb));
4035 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4036 && btrfs_super_cache_generation(sb) != (u64)-1)
4038 "suspicious: generation < cache_generation: %llu < %llu",
4039 btrfs_super_generation(sb),
4040 btrfs_super_cache_generation(sb));
4045 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4047 mutex_lock(&fs_info->cleaner_mutex);
4048 btrfs_run_delayed_iputs(fs_info);
4049 mutex_unlock(&fs_info->cleaner_mutex);
4051 down_write(&fs_info->cleanup_work_sem);
4052 up_write(&fs_info->cleanup_work_sem);
4054 /* cleanup FS via transaction */
4055 btrfs_cleanup_transaction(fs_info);
4058 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4060 struct btrfs_ordered_extent *ordered;
4062 spin_lock(&root->ordered_extent_lock);
4064 * This will just short circuit the ordered completion stuff which will
4065 * make sure the ordered extent gets properly cleaned up.
4067 list_for_each_entry(ordered, &root->ordered_extents,
4069 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4070 spin_unlock(&root->ordered_extent_lock);
4073 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4075 struct btrfs_root *root;
4076 struct list_head splice;
4078 INIT_LIST_HEAD(&splice);
4080 spin_lock(&fs_info->ordered_root_lock);
4081 list_splice_init(&fs_info->ordered_roots, &splice);
4082 while (!list_empty(&splice)) {
4083 root = list_first_entry(&splice, struct btrfs_root,
4085 list_move_tail(&root->ordered_root,
4086 &fs_info->ordered_roots);
4088 spin_unlock(&fs_info->ordered_root_lock);
4089 btrfs_destroy_ordered_extents(root);
4092 spin_lock(&fs_info->ordered_root_lock);
4094 spin_unlock(&fs_info->ordered_root_lock);
4097 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4098 struct btrfs_fs_info *fs_info)
4100 struct rb_node *node;
4101 struct btrfs_delayed_ref_root *delayed_refs;
4102 struct btrfs_delayed_ref_node *ref;
4105 delayed_refs = &trans->delayed_refs;
4107 spin_lock(&delayed_refs->lock);
4108 if (atomic_read(&delayed_refs->num_entries) == 0) {
4109 spin_unlock(&delayed_refs->lock);
4110 btrfs_info(fs_info, "delayed_refs has NO entry");
4114 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4115 struct btrfs_delayed_ref_head *head;
4117 bool pin_bytes = false;
4119 head = rb_entry(node, struct btrfs_delayed_ref_head,
4121 if (!mutex_trylock(&head->mutex)) {
4122 refcount_inc(&head->refs);
4123 spin_unlock(&delayed_refs->lock);
4125 mutex_lock(&head->mutex);
4126 mutex_unlock(&head->mutex);
4127 btrfs_put_delayed_ref_head(head);
4128 spin_lock(&delayed_refs->lock);
4131 spin_lock(&head->lock);
4132 while ((n = rb_first(&head->ref_tree)) != NULL) {
4133 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4136 rb_erase(&ref->ref_node, &head->ref_tree);
4137 RB_CLEAR_NODE(&ref->ref_node);
4138 if (!list_empty(&ref->add_list))
4139 list_del(&ref->add_list);
4140 atomic_dec(&delayed_refs->num_entries);
4141 btrfs_put_delayed_ref(ref);
4143 if (head->must_insert_reserved)
4145 btrfs_free_delayed_extent_op(head->extent_op);
4146 delayed_refs->num_heads--;
4147 if (head->processing == 0)
4148 delayed_refs->num_heads_ready--;
4149 atomic_dec(&delayed_refs->num_entries);
4150 rb_erase(&head->href_node, &delayed_refs->href_root);
4151 RB_CLEAR_NODE(&head->href_node);
4152 spin_unlock(&head->lock);
4153 spin_unlock(&delayed_refs->lock);
4154 mutex_unlock(&head->mutex);
4157 btrfs_pin_extent(fs_info, head->bytenr,
4158 head->num_bytes, 1);
4159 btrfs_put_delayed_ref_head(head);
4161 spin_lock(&delayed_refs->lock);
4164 spin_unlock(&delayed_refs->lock);
4169 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4171 struct btrfs_inode *btrfs_inode;
4172 struct list_head splice;
4174 INIT_LIST_HEAD(&splice);
4176 spin_lock(&root->delalloc_lock);
4177 list_splice_init(&root->delalloc_inodes, &splice);
4179 while (!list_empty(&splice)) {
4180 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4183 list_del_init(&btrfs_inode->delalloc_inodes);
4184 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4185 &btrfs_inode->runtime_flags);
4186 spin_unlock(&root->delalloc_lock);
4188 btrfs_invalidate_inodes(btrfs_inode->root);
4190 spin_lock(&root->delalloc_lock);
4193 spin_unlock(&root->delalloc_lock);
4196 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4198 struct btrfs_root *root;
4199 struct list_head splice;
4201 INIT_LIST_HEAD(&splice);
4203 spin_lock(&fs_info->delalloc_root_lock);
4204 list_splice_init(&fs_info->delalloc_roots, &splice);
4205 while (!list_empty(&splice)) {
4206 root = list_first_entry(&splice, struct btrfs_root,
4208 list_del_init(&root->delalloc_root);
4209 root = btrfs_grab_fs_root(root);
4211 spin_unlock(&fs_info->delalloc_root_lock);
4213 btrfs_destroy_delalloc_inodes(root);
4214 btrfs_put_fs_root(root);
4216 spin_lock(&fs_info->delalloc_root_lock);
4218 spin_unlock(&fs_info->delalloc_root_lock);
4221 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4222 struct extent_io_tree *dirty_pages,
4226 struct extent_buffer *eb;
4231 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4236 clear_extent_bits(dirty_pages, start, end, mark);
4237 while (start <= end) {
4238 eb = find_extent_buffer(fs_info, start);
4239 start += fs_info->nodesize;
4242 wait_on_extent_buffer_writeback(eb);
4244 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4246 clear_extent_buffer_dirty(eb);
4247 free_extent_buffer_stale(eb);
4254 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4255 struct extent_io_tree *pinned_extents)
4257 struct extent_io_tree *unpin;
4263 unpin = pinned_extents;
4266 ret = find_first_extent_bit(unpin, 0, &start, &end,
4267 EXTENT_DIRTY, NULL);
4271 clear_extent_dirty(unpin, start, end);
4272 btrfs_error_unpin_extent_range(fs_info, start, end);
4277 if (unpin == &fs_info->freed_extents[0])
4278 unpin = &fs_info->freed_extents[1];
4280 unpin = &fs_info->freed_extents[0];
4288 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4290 struct inode *inode;
4292 inode = cache->io_ctl.inode;
4294 invalidate_inode_pages2(inode->i_mapping);
4295 BTRFS_I(inode)->generation = 0;
4296 cache->io_ctl.inode = NULL;
4299 btrfs_put_block_group(cache);
4302 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4303 struct btrfs_fs_info *fs_info)
4305 struct btrfs_block_group_cache *cache;
4307 spin_lock(&cur_trans->dirty_bgs_lock);
4308 while (!list_empty(&cur_trans->dirty_bgs)) {
4309 cache = list_first_entry(&cur_trans->dirty_bgs,
4310 struct btrfs_block_group_cache,
4313 btrfs_err(fs_info, "orphan block group dirty_bgs list");
4314 spin_unlock(&cur_trans->dirty_bgs_lock);
4318 if (!list_empty(&cache->io_list)) {
4319 spin_unlock(&cur_trans->dirty_bgs_lock);
4320 list_del_init(&cache->io_list);
4321 btrfs_cleanup_bg_io(cache);
4322 spin_lock(&cur_trans->dirty_bgs_lock);
4325 list_del_init(&cache->dirty_list);
4326 spin_lock(&cache->lock);
4327 cache->disk_cache_state = BTRFS_DC_ERROR;
4328 spin_unlock(&cache->lock);
4330 spin_unlock(&cur_trans->dirty_bgs_lock);
4331 btrfs_put_block_group(cache);
4332 spin_lock(&cur_trans->dirty_bgs_lock);
4334 spin_unlock(&cur_trans->dirty_bgs_lock);
4336 while (!list_empty(&cur_trans->io_bgs)) {
4337 cache = list_first_entry(&cur_trans->io_bgs,
4338 struct btrfs_block_group_cache,
4341 btrfs_err(fs_info, "orphan block group on io_bgs list");
4345 list_del_init(&cache->io_list);
4346 spin_lock(&cache->lock);
4347 cache->disk_cache_state = BTRFS_DC_ERROR;
4348 spin_unlock(&cache->lock);
4349 btrfs_cleanup_bg_io(cache);
4353 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4354 struct btrfs_fs_info *fs_info)
4356 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4357 ASSERT(list_empty(&cur_trans->dirty_bgs));
4358 ASSERT(list_empty(&cur_trans->io_bgs));
4360 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4362 cur_trans->state = TRANS_STATE_COMMIT_START;
4363 wake_up(&fs_info->transaction_blocked_wait);
4365 cur_trans->state = TRANS_STATE_UNBLOCKED;
4366 wake_up(&fs_info->transaction_wait);
4368 btrfs_destroy_delayed_inodes(fs_info);
4369 btrfs_assert_delayed_root_empty(fs_info);
4371 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4373 btrfs_destroy_pinned_extent(fs_info,
4374 fs_info->pinned_extents);
4376 cur_trans->state =TRANS_STATE_COMPLETED;
4377 wake_up(&cur_trans->commit_wait);
4380 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4382 struct btrfs_transaction *t;
4384 mutex_lock(&fs_info->transaction_kthread_mutex);
4386 spin_lock(&fs_info->trans_lock);
4387 while (!list_empty(&fs_info->trans_list)) {
4388 t = list_first_entry(&fs_info->trans_list,
4389 struct btrfs_transaction, list);
4390 if (t->state >= TRANS_STATE_COMMIT_START) {
4391 refcount_inc(&t->use_count);
4392 spin_unlock(&fs_info->trans_lock);
4393 btrfs_wait_for_commit(fs_info, t->transid);
4394 btrfs_put_transaction(t);
4395 spin_lock(&fs_info->trans_lock);
4398 if (t == fs_info->running_transaction) {
4399 t->state = TRANS_STATE_COMMIT_DOING;
4400 spin_unlock(&fs_info->trans_lock);
4402 * We wait for 0 num_writers since we don't hold a trans
4403 * handle open currently for this transaction.
4405 wait_event(t->writer_wait,
4406 atomic_read(&t->num_writers) == 0);
4408 spin_unlock(&fs_info->trans_lock);
4410 btrfs_cleanup_one_transaction(t, fs_info);
4412 spin_lock(&fs_info->trans_lock);
4413 if (t == fs_info->running_transaction)
4414 fs_info->running_transaction = NULL;
4415 list_del_init(&t->list);
4416 spin_unlock(&fs_info->trans_lock);
4418 btrfs_put_transaction(t);
4419 trace_btrfs_transaction_commit(fs_info->tree_root);
4420 spin_lock(&fs_info->trans_lock);
4422 spin_unlock(&fs_info->trans_lock);
4423 btrfs_destroy_all_ordered_extents(fs_info);
4424 btrfs_destroy_delayed_inodes(fs_info);
4425 btrfs_assert_delayed_root_empty(fs_info);
4426 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4427 btrfs_destroy_all_delalloc_inodes(fs_info);
4428 mutex_unlock(&fs_info->transaction_kthread_mutex);
4433 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4435 struct inode *inode = private_data;
4436 return btrfs_sb(inode->i_sb);
4439 static const struct extent_io_ops btree_extent_io_ops = {
4440 /* mandatory callbacks */
4441 .submit_bio_hook = btree_submit_bio_hook,
4442 .readpage_end_io_hook = btree_readpage_end_io_hook,
4443 /* note we're sharing with inode.c for the merge bio hook */
4444 .merge_bio_hook = btrfs_merge_bio_hook,
4445 .readpage_io_failed_hook = btree_io_failed_hook,
4446 .set_range_writeback = btrfs_set_range_writeback,
4447 .tree_fs_info = btree_fs_info,
4449 /* optional callbacks */