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"
56 #include <asm/cpufeature.h>
59 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
60 BTRFS_HEADER_FLAG_RELOC |\
61 BTRFS_SUPER_FLAG_ERROR |\
62 BTRFS_SUPER_FLAG_SEEDING |\
63 BTRFS_SUPER_FLAG_METADUMP |\
64 BTRFS_SUPER_FLAG_METADUMP_V2)
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 static 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);
290 unsigned long cur_len;
291 unsigned long offset = BTRFS_CSUM_SIZE;
293 unsigned long map_start;
294 unsigned long map_len;
297 unsigned long inline_result;
299 len = buf->len - offset;
301 err = map_private_extent_buffer(buf, offset, 32,
302 &kaddr, &map_start, &map_len);
305 cur_len = min(len, map_len - (offset - map_start));
306 crc = btrfs_csum_data(kaddr + offset - map_start,
311 if (csum_size > sizeof(inline_result)) {
312 result = kzalloc(csum_size, GFP_NOFS);
316 result = (char *)&inline_result;
319 btrfs_csum_final(crc, result);
322 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
325 memcpy(&found, result, csum_size);
327 read_extent_buffer(buf, &val, 0, csum_size);
328 btrfs_warn_rl(fs_info,
329 "%s checksum verify failed on %llu wanted %X found %X level %d",
330 fs_info->sb->s_id, buf->start,
331 val, found, btrfs_header_level(buf));
332 if (result != (char *)&inline_result)
337 write_extent_buffer(buf, result, 0, csum_size);
339 if (result != (char *)&inline_result)
345 * we can't consider a given block up to date unless the transid of the
346 * block matches the transid in the parent node's pointer. This is how we
347 * detect blocks that either didn't get written at all or got written
348 * in the wrong place.
350 static int verify_parent_transid(struct extent_io_tree *io_tree,
351 struct extent_buffer *eb, u64 parent_transid,
354 struct extent_state *cached_state = NULL;
356 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
358 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
365 btrfs_tree_read_lock(eb);
366 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
369 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
371 if (extent_buffer_uptodate(eb) &&
372 btrfs_header_generation(eb) == parent_transid) {
376 btrfs_err_rl(eb->fs_info,
377 "parent transid verify failed on %llu wanted %llu found %llu",
379 parent_transid, btrfs_header_generation(eb));
383 * Things reading via commit roots that don't have normal protection,
384 * like send, can have a really old block in cache that may point at a
385 * block that has been freed and re-allocated. So don't clear uptodate
386 * if we find an eb that is under IO (dirty/writeback) because we could
387 * end up reading in the stale data and then writing it back out and
388 * making everybody very sad.
390 if (!extent_buffer_under_io(eb))
391 clear_extent_buffer_uptodate(eb);
393 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
394 &cached_state, GFP_NOFS);
396 btrfs_tree_read_unlock_blocking(eb);
401 * Return 0 if the superblock checksum type matches the checksum value of that
402 * algorithm. Pass the raw disk superblock data.
404 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
407 struct btrfs_super_block *disk_sb =
408 (struct btrfs_super_block *)raw_disk_sb;
409 u16 csum_type = btrfs_super_csum_type(disk_sb);
412 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
414 const int csum_size = sizeof(crc);
415 char result[csum_size];
418 * The super_block structure does not span the whole
419 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
420 * is filled with zeros and is included in the checksum.
422 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
423 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
424 btrfs_csum_final(crc, result);
426 if (memcmp(raw_disk_sb, result, csum_size))
430 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
431 btrfs_err(fs_info, "unsupported checksum algorithm %u",
440 * helper to read a given tree block, doing retries as required when
441 * the checksums don't match and we have alternate mirrors to try.
443 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
444 struct extent_buffer *eb,
447 struct extent_io_tree *io_tree;
452 int failed_mirror = 0;
454 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
456 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
457 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
458 btree_get_extent, mirror_num);
460 if (!verify_parent_transid(io_tree, eb,
467 num_copies = btrfs_num_copies(fs_info,
472 if (!failed_mirror) {
474 failed_mirror = eb->read_mirror;
478 if (mirror_num == failed_mirror)
481 if (mirror_num > num_copies)
485 if (failed && !ret && failed_mirror)
486 repair_eb_io_failure(fs_info, eb, failed_mirror);
492 * checksum a dirty tree block before IO. This has extra checks to make sure
493 * we only fill in the checksum field in the first page of a multi-page block
496 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
498 u64 start = page_offset(page);
500 struct extent_buffer *eb;
502 eb = (struct extent_buffer *)page->private;
503 if (page != eb->pages[0])
506 found_start = btrfs_header_bytenr(eb);
508 * Please do not consolidate these warnings into a single if.
509 * It is useful to know what went wrong.
511 if (WARN_ON(found_start != start))
513 if (WARN_ON(!PageUptodate(page)))
516 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
517 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
519 return csum_tree_block(fs_info, eb, 0);
522 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
523 struct extent_buffer *eb)
525 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
526 u8 fsid[BTRFS_FSID_SIZE];
529 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
531 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
535 fs_devices = fs_devices->seed;
540 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
541 u64 phy_offset, struct page *page,
542 u64 start, u64 end, int mirror)
546 struct extent_buffer *eb;
547 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
548 struct btrfs_fs_info *fs_info = root->fs_info;
555 eb = (struct extent_buffer *)page->private;
557 /* the pending IO might have been the only thing that kept this buffer
558 * in memory. Make sure we have a ref for all this other checks
560 extent_buffer_get(eb);
562 reads_done = atomic_dec_and_test(&eb->io_pages);
566 eb->read_mirror = mirror;
567 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
572 found_start = btrfs_header_bytenr(eb);
573 if (found_start != eb->start) {
574 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
575 found_start, eb->start);
579 if (check_tree_block_fsid(fs_info, eb)) {
580 btrfs_err_rl(fs_info, "bad fsid on block %llu",
585 found_level = btrfs_header_level(eb);
586 if (found_level >= BTRFS_MAX_LEVEL) {
587 btrfs_err(fs_info, "bad tree block level %d",
588 (int)btrfs_header_level(eb));
593 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
596 ret = csum_tree_block(fs_info, eb, 1);
601 * If this is a leaf block and it is corrupt, set the corrupt bit so
602 * that we don't try and read the other copies of this block, just
605 if (found_level == 0 && btrfs_check_leaf_full(root, eb)) {
606 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
610 if (found_level > 0 && btrfs_check_node(root, eb))
614 set_extent_buffer_uptodate(eb);
617 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
618 btree_readahead_hook(eb, ret);
622 * our io error hook is going to dec the io pages
623 * again, we have to make sure it has something
626 atomic_inc(&eb->io_pages);
627 clear_extent_buffer_uptodate(eb);
629 free_extent_buffer(eb);
634 static int btree_io_failed_hook(struct page *page, int failed_mirror)
636 struct extent_buffer *eb;
638 eb = (struct extent_buffer *)page->private;
639 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
640 eb->read_mirror = failed_mirror;
641 atomic_dec(&eb->io_pages);
642 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
643 btree_readahead_hook(eb, -EIO);
644 return -EIO; /* we fixed nothing */
647 static void end_workqueue_bio(struct bio *bio)
649 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
650 struct btrfs_fs_info *fs_info;
651 struct btrfs_workqueue *wq;
652 btrfs_work_func_t func;
654 fs_info = end_io_wq->info;
655 end_io_wq->status = bio->bi_status;
657 if (bio_op(bio) == REQ_OP_WRITE) {
658 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
659 wq = fs_info->endio_meta_write_workers;
660 func = btrfs_endio_meta_write_helper;
661 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
662 wq = fs_info->endio_freespace_worker;
663 func = btrfs_freespace_write_helper;
664 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
665 wq = fs_info->endio_raid56_workers;
666 func = btrfs_endio_raid56_helper;
668 wq = fs_info->endio_write_workers;
669 func = btrfs_endio_write_helper;
672 if (unlikely(end_io_wq->metadata ==
673 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
674 wq = fs_info->endio_repair_workers;
675 func = btrfs_endio_repair_helper;
676 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
677 wq = fs_info->endio_raid56_workers;
678 func = btrfs_endio_raid56_helper;
679 } else if (end_io_wq->metadata) {
680 wq = fs_info->endio_meta_workers;
681 func = btrfs_endio_meta_helper;
683 wq = fs_info->endio_workers;
684 func = btrfs_endio_helper;
688 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
689 btrfs_queue_work(wq, &end_io_wq->work);
692 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
693 enum btrfs_wq_endio_type metadata)
695 struct btrfs_end_io_wq *end_io_wq;
697 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
699 return BLK_STS_RESOURCE;
701 end_io_wq->private = bio->bi_private;
702 end_io_wq->end_io = bio->bi_end_io;
703 end_io_wq->info = info;
704 end_io_wq->status = 0;
705 end_io_wq->bio = bio;
706 end_io_wq->metadata = metadata;
708 bio->bi_private = end_io_wq;
709 bio->bi_end_io = end_workqueue_bio;
713 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
715 unsigned long limit = min_t(unsigned long,
716 info->thread_pool_size,
717 info->fs_devices->open_devices);
721 static void run_one_async_start(struct btrfs_work *work)
723 struct async_submit_bio *async;
726 async = container_of(work, struct async_submit_bio, work);
727 ret = async->submit_bio_start(async->private_data, async->bio,
728 async->mirror_num, async->bio_flags,
734 static void run_one_async_done(struct btrfs_work *work)
736 struct btrfs_fs_info *fs_info;
737 struct async_submit_bio *async;
740 async = container_of(work, struct async_submit_bio, work);
741 fs_info = async->fs_info;
743 limit = btrfs_async_submit_limit(fs_info);
744 limit = limit * 2 / 3;
747 * atomic_dec_return implies a barrier for waitqueue_active
749 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
750 waitqueue_active(&fs_info->async_submit_wait))
751 wake_up(&fs_info->async_submit_wait);
753 /* If an error occurred we just want to clean up the bio and move on */
755 async->bio->bi_status = async->status;
756 bio_endio(async->bio);
760 async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
761 async->bio_flags, async->bio_offset);
764 static void run_one_async_free(struct btrfs_work *work)
766 struct async_submit_bio *async;
768 async = container_of(work, struct async_submit_bio, work);
772 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
773 int mirror_num, unsigned long bio_flags,
774 u64 bio_offset, void *private_data,
775 extent_submit_bio_hook_t *submit_bio_start,
776 extent_submit_bio_hook_t *submit_bio_done)
778 struct async_submit_bio *async;
780 async = kmalloc(sizeof(*async), GFP_NOFS);
782 return BLK_STS_RESOURCE;
784 async->private_data = private_data;
785 async->fs_info = fs_info;
787 async->mirror_num = mirror_num;
788 async->submit_bio_start = submit_bio_start;
789 async->submit_bio_done = submit_bio_done;
791 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
792 run_one_async_done, run_one_async_free);
794 async->bio_flags = bio_flags;
795 async->bio_offset = bio_offset;
799 atomic_inc(&fs_info->nr_async_submits);
801 if (op_is_sync(bio->bi_opf))
802 btrfs_set_work_high_priority(&async->work);
804 btrfs_queue_work(fs_info->workers, &async->work);
806 while (atomic_read(&fs_info->async_submit_draining) &&
807 atomic_read(&fs_info->nr_async_submits)) {
808 wait_event(fs_info->async_submit_wait,
809 (atomic_read(&fs_info->nr_async_submits) == 0));
815 static blk_status_t btree_csum_one_bio(struct bio *bio)
817 struct bio_vec *bvec;
818 struct btrfs_root *root;
821 ASSERT(!bio_flagged(bio, BIO_CLONED));
822 bio_for_each_segment_all(bvec, bio, i) {
823 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
824 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
829 return errno_to_blk_status(ret);
832 static blk_status_t __btree_submit_bio_start(void *private_data, struct bio *bio,
833 int mirror_num, unsigned long bio_flags,
837 * when we're called for a write, we're already in the async
838 * submission context. Just jump into btrfs_map_bio
840 return btree_csum_one_bio(bio);
843 static blk_status_t __btree_submit_bio_done(void *private_data, struct bio *bio,
844 int mirror_num, unsigned long bio_flags,
847 struct inode *inode = private_data;
851 * when we're called for a write, we're already in the async
852 * submission context. Just jump into btrfs_map_bio
854 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
856 bio->bi_status = ret;
862 static int check_async_write(unsigned long bio_flags)
864 if (bio_flags & EXTENT_BIO_TREE_LOG)
867 if (static_cpu_has(X86_FEATURE_XMM4_2))
873 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
874 int mirror_num, unsigned long bio_flags,
877 struct inode *inode = private_data;
878 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
879 int async = check_async_write(bio_flags);
882 if (bio_op(bio) != REQ_OP_WRITE) {
884 * called for a read, do the setup so that checksum validation
885 * can happen in the async kernel threads
887 ret = btrfs_bio_wq_end_io(fs_info, bio,
888 BTRFS_WQ_ENDIO_METADATA);
891 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
893 ret = btree_csum_one_bio(bio);
896 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
899 * kthread helpers are used to submit writes so that
900 * checksumming can happen in parallel across all CPUs
902 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
903 bio_offset, private_data,
904 __btree_submit_bio_start,
905 __btree_submit_bio_done);
913 bio->bi_status = ret;
918 #ifdef CONFIG_MIGRATION
919 static int btree_migratepage(struct address_space *mapping,
920 struct page *newpage, struct page *page,
921 enum migrate_mode mode)
924 * we can't safely write a btree page from here,
925 * we haven't done the locking hook
930 * Buffers may be managed in a filesystem specific way.
931 * We must have no buffers or drop them.
933 if (page_has_private(page) &&
934 !try_to_release_page(page, GFP_KERNEL))
936 return migrate_page(mapping, newpage, page, mode);
941 static int btree_writepages(struct address_space *mapping,
942 struct writeback_control *wbc)
944 struct btrfs_fs_info *fs_info;
947 if (wbc->sync_mode == WB_SYNC_NONE) {
949 if (wbc->for_kupdate)
952 fs_info = BTRFS_I(mapping->host)->root->fs_info;
953 /* this is a bit racy, but that's ok */
954 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
955 BTRFS_DIRTY_METADATA_THRESH,
956 fs_info->dirty_metadata_batch);
960 return btree_write_cache_pages(mapping, wbc);
963 static int btree_readpage(struct file *file, struct page *page)
965 struct extent_io_tree *tree;
966 tree = &BTRFS_I(page->mapping->host)->io_tree;
967 return extent_read_full_page(tree, page, btree_get_extent, 0);
970 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
972 if (PageWriteback(page) || PageDirty(page))
975 return try_release_extent_buffer(page);
978 static void btree_invalidatepage(struct page *page, unsigned int offset,
981 struct extent_io_tree *tree;
982 tree = &BTRFS_I(page->mapping->host)->io_tree;
983 extent_invalidatepage(tree, page, offset);
984 btree_releasepage(page, GFP_NOFS);
985 if (PagePrivate(page)) {
986 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
987 "page private not zero on page %llu",
988 (unsigned long long)page_offset(page));
989 ClearPagePrivate(page);
990 set_page_private(page, 0);
995 static int btree_set_page_dirty(struct page *page)
998 struct extent_buffer *eb;
1000 BUG_ON(!PagePrivate(page));
1001 eb = (struct extent_buffer *)page->private;
1003 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1004 BUG_ON(!atomic_read(&eb->refs));
1005 btrfs_assert_tree_locked(eb);
1007 return __set_page_dirty_nobuffers(page);
1010 static const struct address_space_operations btree_aops = {
1011 .readpage = btree_readpage,
1012 .writepages = btree_writepages,
1013 .releasepage = btree_releasepage,
1014 .invalidatepage = btree_invalidatepage,
1015 #ifdef CONFIG_MIGRATION
1016 .migratepage = btree_migratepage,
1018 .set_page_dirty = btree_set_page_dirty,
1021 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1023 struct extent_buffer *buf = NULL;
1024 struct inode *btree_inode = fs_info->btree_inode;
1026 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1029 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1030 buf, WAIT_NONE, btree_get_extent, 0);
1031 free_extent_buffer(buf);
1034 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1035 int mirror_num, struct extent_buffer **eb)
1037 struct extent_buffer *buf = NULL;
1038 struct inode *btree_inode = fs_info->btree_inode;
1039 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1042 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1046 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1048 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1049 btree_get_extent, mirror_num);
1051 free_extent_buffer(buf);
1055 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1056 free_extent_buffer(buf);
1058 } else if (extent_buffer_uptodate(buf)) {
1061 free_extent_buffer(buf);
1066 struct extent_buffer *btrfs_find_create_tree_block(
1067 struct btrfs_fs_info *fs_info,
1070 if (btrfs_is_testing(fs_info))
1071 return alloc_test_extent_buffer(fs_info, bytenr);
1072 return alloc_extent_buffer(fs_info, bytenr);
1076 int btrfs_write_tree_block(struct extent_buffer *buf)
1078 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1079 buf->start + buf->len - 1);
1082 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1084 filemap_fdatawait_range(buf->pages[0]->mapping,
1085 buf->start, buf->start + buf->len - 1);
1088 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1091 struct extent_buffer *buf = NULL;
1094 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1098 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1100 free_extent_buffer(buf);
1101 return ERR_PTR(ret);
1107 void clean_tree_block(struct btrfs_fs_info *fs_info,
1108 struct extent_buffer *buf)
1110 if (btrfs_header_generation(buf) ==
1111 fs_info->running_transaction->transid) {
1112 btrfs_assert_tree_locked(buf);
1114 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1115 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1117 fs_info->dirty_metadata_batch);
1118 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1119 btrfs_set_lock_blocking(buf);
1120 clear_extent_buffer_dirty(buf);
1125 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1127 struct btrfs_subvolume_writers *writers;
1130 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1132 return ERR_PTR(-ENOMEM);
1134 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1137 return ERR_PTR(ret);
1140 init_waitqueue_head(&writers->wait);
1145 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1147 percpu_counter_destroy(&writers->counter);
1151 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1154 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1156 root->commit_root = NULL;
1158 root->orphan_cleanup_state = 0;
1160 root->objectid = objectid;
1161 root->last_trans = 0;
1162 root->highest_objectid = 0;
1163 root->nr_delalloc_inodes = 0;
1164 root->nr_ordered_extents = 0;
1166 root->inode_tree = RB_ROOT;
1167 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1168 root->block_rsv = NULL;
1169 root->orphan_block_rsv = NULL;
1171 INIT_LIST_HEAD(&root->dirty_list);
1172 INIT_LIST_HEAD(&root->root_list);
1173 INIT_LIST_HEAD(&root->delalloc_inodes);
1174 INIT_LIST_HEAD(&root->delalloc_root);
1175 INIT_LIST_HEAD(&root->ordered_extents);
1176 INIT_LIST_HEAD(&root->ordered_root);
1177 INIT_LIST_HEAD(&root->logged_list[0]);
1178 INIT_LIST_HEAD(&root->logged_list[1]);
1179 spin_lock_init(&root->orphan_lock);
1180 spin_lock_init(&root->inode_lock);
1181 spin_lock_init(&root->delalloc_lock);
1182 spin_lock_init(&root->ordered_extent_lock);
1183 spin_lock_init(&root->accounting_lock);
1184 spin_lock_init(&root->log_extents_lock[0]);
1185 spin_lock_init(&root->log_extents_lock[1]);
1186 mutex_init(&root->objectid_mutex);
1187 mutex_init(&root->log_mutex);
1188 mutex_init(&root->ordered_extent_mutex);
1189 mutex_init(&root->delalloc_mutex);
1190 init_waitqueue_head(&root->log_writer_wait);
1191 init_waitqueue_head(&root->log_commit_wait[0]);
1192 init_waitqueue_head(&root->log_commit_wait[1]);
1193 INIT_LIST_HEAD(&root->log_ctxs[0]);
1194 INIT_LIST_HEAD(&root->log_ctxs[1]);
1195 atomic_set(&root->log_commit[0], 0);
1196 atomic_set(&root->log_commit[1], 0);
1197 atomic_set(&root->log_writers, 0);
1198 atomic_set(&root->log_batch, 0);
1199 atomic_set(&root->orphan_inodes, 0);
1200 refcount_set(&root->refs, 1);
1201 atomic_set(&root->will_be_snapshotted, 0);
1202 atomic64_set(&root->qgroup_meta_rsv, 0);
1203 atomic_set(&root->snapshot_force_cow, 0);
1204 root->log_transid = 0;
1205 root->log_transid_committed = -1;
1206 root->last_log_commit = 0;
1208 extent_io_tree_init(&root->dirty_log_pages, NULL);
1210 memset(&root->root_key, 0, sizeof(root->root_key));
1211 memset(&root->root_item, 0, sizeof(root->root_item));
1212 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1214 root->defrag_trans_start = fs_info->generation;
1216 root->defrag_trans_start = 0;
1217 root->root_key.objectid = objectid;
1220 spin_lock_init(&root->root_item_lock);
1223 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1226 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1228 root->fs_info = fs_info;
1232 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1233 /* Should only be used by the testing infrastructure */
1234 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1236 struct btrfs_root *root;
1239 return ERR_PTR(-EINVAL);
1241 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1243 return ERR_PTR(-ENOMEM);
1245 /* We don't use the stripesize in selftest, set it as sectorsize */
1246 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1247 root->alloc_bytenr = 0;
1253 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1254 struct btrfs_fs_info *fs_info,
1257 struct extent_buffer *leaf;
1258 struct btrfs_root *tree_root = fs_info->tree_root;
1259 struct btrfs_root *root;
1260 struct btrfs_key key;
1264 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1266 return ERR_PTR(-ENOMEM);
1268 __setup_root(root, fs_info, objectid);
1269 root->root_key.objectid = objectid;
1270 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1271 root->root_key.offset = 0;
1273 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1275 ret = PTR_ERR(leaf);
1280 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1281 btrfs_set_header_bytenr(leaf, leaf->start);
1282 btrfs_set_header_generation(leaf, trans->transid);
1283 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1284 btrfs_set_header_owner(leaf, objectid);
1287 write_extent_buffer_fsid(leaf, fs_info->fsid);
1288 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1289 btrfs_mark_buffer_dirty(leaf);
1291 root->commit_root = btrfs_root_node(root);
1292 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1294 root->root_item.flags = 0;
1295 root->root_item.byte_limit = 0;
1296 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1297 btrfs_set_root_generation(&root->root_item, trans->transid);
1298 btrfs_set_root_level(&root->root_item, 0);
1299 btrfs_set_root_refs(&root->root_item, 1);
1300 btrfs_set_root_used(&root->root_item, leaf->len);
1301 btrfs_set_root_last_snapshot(&root->root_item, 0);
1302 btrfs_set_root_dirid(&root->root_item, 0);
1304 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1305 root->root_item.drop_level = 0;
1307 key.objectid = objectid;
1308 key.type = BTRFS_ROOT_ITEM_KEY;
1310 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1314 btrfs_tree_unlock(leaf);
1320 btrfs_tree_unlock(leaf);
1321 free_extent_buffer(root->commit_root);
1322 free_extent_buffer(leaf);
1326 return ERR_PTR(ret);
1329 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1330 struct btrfs_fs_info *fs_info)
1332 struct btrfs_root *root;
1333 struct extent_buffer *leaf;
1335 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1337 return ERR_PTR(-ENOMEM);
1339 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1341 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1342 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1343 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1346 * DON'T set REF_COWS for log trees
1348 * log trees do not get reference counted because they go away
1349 * before a real commit is actually done. They do store pointers
1350 * to file data extents, and those reference counts still get
1351 * updated (along with back refs to the log tree).
1354 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1358 return ERR_CAST(leaf);
1361 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1362 btrfs_set_header_bytenr(leaf, leaf->start);
1363 btrfs_set_header_generation(leaf, trans->transid);
1364 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1365 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1368 write_extent_buffer_fsid(root->node, fs_info->fsid);
1369 btrfs_mark_buffer_dirty(root->node);
1370 btrfs_tree_unlock(root->node);
1374 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1375 struct btrfs_fs_info *fs_info)
1377 struct btrfs_root *log_root;
1379 log_root = alloc_log_tree(trans, fs_info);
1380 if (IS_ERR(log_root))
1381 return PTR_ERR(log_root);
1382 WARN_ON(fs_info->log_root_tree);
1383 fs_info->log_root_tree = log_root;
1387 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1388 struct btrfs_root *root)
1390 struct btrfs_fs_info *fs_info = root->fs_info;
1391 struct btrfs_root *log_root;
1392 struct btrfs_inode_item *inode_item;
1394 log_root = alloc_log_tree(trans, fs_info);
1395 if (IS_ERR(log_root))
1396 return PTR_ERR(log_root);
1398 log_root->last_trans = trans->transid;
1399 log_root->root_key.offset = root->root_key.objectid;
1401 inode_item = &log_root->root_item.inode;
1402 btrfs_set_stack_inode_generation(inode_item, 1);
1403 btrfs_set_stack_inode_size(inode_item, 3);
1404 btrfs_set_stack_inode_nlink(inode_item, 1);
1405 btrfs_set_stack_inode_nbytes(inode_item,
1407 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1409 btrfs_set_root_node(&log_root->root_item, log_root->node);
1411 WARN_ON(root->log_root);
1412 root->log_root = log_root;
1413 root->log_transid = 0;
1414 root->log_transid_committed = -1;
1415 root->last_log_commit = 0;
1419 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1420 struct btrfs_key *key)
1422 struct btrfs_root *root;
1423 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1424 struct btrfs_path *path;
1428 path = btrfs_alloc_path();
1430 return ERR_PTR(-ENOMEM);
1432 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1438 __setup_root(root, fs_info, key->objectid);
1440 ret = btrfs_find_root(tree_root, key, path,
1441 &root->root_item, &root->root_key);
1448 generation = btrfs_root_generation(&root->root_item);
1449 root->node = read_tree_block(fs_info,
1450 btrfs_root_bytenr(&root->root_item),
1452 if (IS_ERR(root->node)) {
1453 ret = PTR_ERR(root->node);
1455 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1457 free_extent_buffer(root->node);
1460 root->commit_root = btrfs_root_node(root);
1462 btrfs_free_path(path);
1468 root = ERR_PTR(ret);
1472 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1473 struct btrfs_key *location)
1475 struct btrfs_root *root;
1477 root = btrfs_read_tree_root(tree_root, location);
1481 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1482 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1483 btrfs_check_and_init_root_item(&root->root_item);
1489 int btrfs_init_fs_root(struct btrfs_root *root)
1492 struct btrfs_subvolume_writers *writers;
1494 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1495 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1497 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1502 writers = btrfs_alloc_subvolume_writers();
1503 if (IS_ERR(writers)) {
1504 ret = PTR_ERR(writers);
1507 root->subv_writers = writers;
1509 btrfs_init_free_ino_ctl(root);
1510 spin_lock_init(&root->ino_cache_lock);
1511 init_waitqueue_head(&root->ino_cache_wait);
1514 * Don't assign anonymous block device to roots that are not exposed to
1515 * userspace, the id pool is limited to 1M
1517 if (is_fstree(root->root_key.objectid) &&
1518 btrfs_root_refs(&root->root_item) > 0) {
1519 ret = get_anon_bdev(&root->anon_dev);
1524 mutex_lock(&root->objectid_mutex);
1525 ret = btrfs_find_highest_objectid(root,
1526 &root->highest_objectid);
1528 mutex_unlock(&root->objectid_mutex);
1532 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1534 mutex_unlock(&root->objectid_mutex);
1538 /* the caller is responsible to call free_fs_root */
1542 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1545 struct btrfs_root *root;
1547 spin_lock(&fs_info->fs_roots_radix_lock);
1548 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1549 (unsigned long)root_id);
1550 spin_unlock(&fs_info->fs_roots_radix_lock);
1554 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1555 struct btrfs_root *root)
1559 ret = radix_tree_preload(GFP_NOFS);
1563 spin_lock(&fs_info->fs_roots_radix_lock);
1564 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1565 (unsigned long)root->root_key.objectid,
1568 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1569 spin_unlock(&fs_info->fs_roots_radix_lock);
1570 radix_tree_preload_end();
1575 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1576 struct btrfs_key *location,
1579 struct btrfs_root *root;
1580 struct btrfs_path *path;
1581 struct btrfs_key key;
1584 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1585 return fs_info->tree_root;
1586 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1587 return fs_info->extent_root;
1588 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1589 return fs_info->chunk_root;
1590 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1591 return fs_info->dev_root;
1592 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1593 return fs_info->csum_root;
1594 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1595 return fs_info->quota_root ? fs_info->quota_root :
1597 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1598 return fs_info->uuid_root ? fs_info->uuid_root :
1600 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1601 return fs_info->free_space_root ? fs_info->free_space_root :
1604 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1606 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1607 return ERR_PTR(-ENOENT);
1611 root = btrfs_read_fs_root(fs_info->tree_root, location);
1615 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1620 ret = btrfs_init_fs_root(root);
1624 path = btrfs_alloc_path();
1629 key.objectid = BTRFS_ORPHAN_OBJECTID;
1630 key.type = BTRFS_ORPHAN_ITEM_KEY;
1631 key.offset = location->objectid;
1633 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1634 btrfs_free_path(path);
1638 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1640 ret = btrfs_insert_fs_root(fs_info, root);
1642 if (ret == -EEXIST) {
1651 return ERR_PTR(ret);
1654 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1656 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1658 struct btrfs_device *device;
1659 struct backing_dev_info *bdi;
1662 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1665 bdi = device->bdev->bd_bdi;
1666 if (bdi_congested(bdi, bdi_bits)) {
1676 * called by the kthread helper functions to finally call the bio end_io
1677 * functions. This is where read checksum verification actually happens
1679 static void end_workqueue_fn(struct btrfs_work *work)
1682 struct btrfs_end_io_wq *end_io_wq;
1684 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1685 bio = end_io_wq->bio;
1687 bio->bi_status = end_io_wq->status;
1688 bio->bi_private = end_io_wq->private;
1689 bio->bi_end_io = end_io_wq->end_io;
1691 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1694 static int cleaner_kthread(void *arg)
1696 struct btrfs_root *root = arg;
1697 struct btrfs_fs_info *fs_info = root->fs_info;
1703 /* Make the cleaner go to sleep early. */
1704 if (btrfs_need_cleaner_sleep(fs_info))
1708 * Do not do anything if we might cause open_ctree() to block
1709 * before we have finished mounting the filesystem.
1711 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1714 if (!mutex_trylock(&fs_info->cleaner_mutex))
1718 * Avoid the problem that we change the status of the fs
1719 * during the above check and trylock.
1721 if (btrfs_need_cleaner_sleep(fs_info)) {
1722 mutex_unlock(&fs_info->cleaner_mutex);
1726 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1727 btrfs_run_delayed_iputs(fs_info);
1728 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1730 again = btrfs_clean_one_deleted_snapshot(root);
1731 mutex_unlock(&fs_info->cleaner_mutex);
1734 * The defragger has dealt with the R/O remount and umount,
1735 * needn't do anything special here.
1737 btrfs_run_defrag_inodes(fs_info);
1740 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1741 * with relocation (btrfs_relocate_chunk) and relocation
1742 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1743 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1744 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1745 * unused block groups.
1747 btrfs_delete_unused_bgs(fs_info);
1749 if (kthread_should_park())
1751 if (kthread_should_stop())
1754 set_current_state(TASK_INTERRUPTIBLE);
1756 __set_current_state(TASK_RUNNING);
1761 static int transaction_kthread(void *arg)
1763 struct btrfs_root *root = arg;
1764 struct btrfs_fs_info *fs_info = root->fs_info;
1765 struct btrfs_trans_handle *trans;
1766 struct btrfs_transaction *cur;
1769 unsigned long delay;
1773 cannot_commit = false;
1774 delay = HZ * fs_info->commit_interval;
1775 mutex_lock(&fs_info->transaction_kthread_mutex);
1777 spin_lock(&fs_info->trans_lock);
1778 cur = fs_info->running_transaction;
1780 spin_unlock(&fs_info->trans_lock);
1784 now = get_seconds();
1785 if (cur->state < TRANS_STATE_BLOCKED &&
1786 (now < cur->start_time ||
1787 now - cur->start_time < fs_info->commit_interval)) {
1788 spin_unlock(&fs_info->trans_lock);
1792 transid = cur->transid;
1793 spin_unlock(&fs_info->trans_lock);
1795 /* If the file system is aborted, this will always fail. */
1796 trans = btrfs_attach_transaction(root);
1797 if (IS_ERR(trans)) {
1798 if (PTR_ERR(trans) != -ENOENT)
1799 cannot_commit = true;
1802 if (transid == trans->transid) {
1803 btrfs_commit_transaction(trans);
1805 btrfs_end_transaction(trans);
1808 wake_up_process(fs_info->cleaner_kthread);
1809 mutex_unlock(&fs_info->transaction_kthread_mutex);
1811 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1812 &fs_info->fs_state)))
1813 btrfs_cleanup_transaction(fs_info);
1814 set_current_state(TASK_INTERRUPTIBLE);
1815 if (!kthread_should_stop() &&
1816 (!btrfs_transaction_blocked(fs_info) ||
1818 schedule_timeout(delay);
1819 __set_current_state(TASK_RUNNING);
1820 } while (!kthread_should_stop());
1825 * this will find the highest generation in the array of
1826 * root backups. The index of the highest array is returned,
1827 * or -1 if we can't find anything.
1829 * We check to make sure the array is valid by comparing the
1830 * generation of the latest root in the array with the generation
1831 * in the super block. If they don't match we pitch it.
1833 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1836 int newest_index = -1;
1837 struct btrfs_root_backup *root_backup;
1840 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1841 root_backup = info->super_copy->super_roots + i;
1842 cur = btrfs_backup_tree_root_gen(root_backup);
1843 if (cur == newest_gen)
1847 /* check to see if we actually wrapped around */
1848 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1849 root_backup = info->super_copy->super_roots;
1850 cur = btrfs_backup_tree_root_gen(root_backup);
1851 if (cur == newest_gen)
1854 return newest_index;
1859 * find the oldest backup so we know where to store new entries
1860 * in the backup array. This will set the backup_root_index
1861 * field in the fs_info struct
1863 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1866 int newest_index = -1;
1868 newest_index = find_newest_super_backup(info, newest_gen);
1869 /* if there was garbage in there, just move along */
1870 if (newest_index == -1) {
1871 info->backup_root_index = 0;
1873 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1878 * copy all the root pointers into the super backup array.
1879 * this will bump the backup pointer by one when it is
1882 static void backup_super_roots(struct btrfs_fs_info *info)
1885 struct btrfs_root_backup *root_backup;
1888 next_backup = info->backup_root_index;
1889 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1890 BTRFS_NUM_BACKUP_ROOTS;
1893 * just overwrite the last backup if we're at the same generation
1894 * this happens only at umount
1896 root_backup = info->super_for_commit->super_roots + last_backup;
1897 if (btrfs_backup_tree_root_gen(root_backup) ==
1898 btrfs_header_generation(info->tree_root->node))
1899 next_backup = last_backup;
1901 root_backup = info->super_for_commit->super_roots + next_backup;
1904 * make sure all of our padding and empty slots get zero filled
1905 * regardless of which ones we use today
1907 memset(root_backup, 0, sizeof(*root_backup));
1909 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1911 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1912 btrfs_set_backup_tree_root_gen(root_backup,
1913 btrfs_header_generation(info->tree_root->node));
1915 btrfs_set_backup_tree_root_level(root_backup,
1916 btrfs_header_level(info->tree_root->node));
1918 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1919 btrfs_set_backup_chunk_root_gen(root_backup,
1920 btrfs_header_generation(info->chunk_root->node));
1921 btrfs_set_backup_chunk_root_level(root_backup,
1922 btrfs_header_level(info->chunk_root->node));
1924 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1925 btrfs_set_backup_extent_root_gen(root_backup,
1926 btrfs_header_generation(info->extent_root->node));
1927 btrfs_set_backup_extent_root_level(root_backup,
1928 btrfs_header_level(info->extent_root->node));
1931 * we might commit during log recovery, which happens before we set
1932 * the fs_root. Make sure it is valid before we fill it in.
1934 if (info->fs_root && info->fs_root->node) {
1935 btrfs_set_backup_fs_root(root_backup,
1936 info->fs_root->node->start);
1937 btrfs_set_backup_fs_root_gen(root_backup,
1938 btrfs_header_generation(info->fs_root->node));
1939 btrfs_set_backup_fs_root_level(root_backup,
1940 btrfs_header_level(info->fs_root->node));
1943 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1944 btrfs_set_backup_dev_root_gen(root_backup,
1945 btrfs_header_generation(info->dev_root->node));
1946 btrfs_set_backup_dev_root_level(root_backup,
1947 btrfs_header_level(info->dev_root->node));
1949 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1950 btrfs_set_backup_csum_root_gen(root_backup,
1951 btrfs_header_generation(info->csum_root->node));
1952 btrfs_set_backup_csum_root_level(root_backup,
1953 btrfs_header_level(info->csum_root->node));
1955 btrfs_set_backup_total_bytes(root_backup,
1956 btrfs_super_total_bytes(info->super_copy));
1957 btrfs_set_backup_bytes_used(root_backup,
1958 btrfs_super_bytes_used(info->super_copy));
1959 btrfs_set_backup_num_devices(root_backup,
1960 btrfs_super_num_devices(info->super_copy));
1963 * if we don't copy this out to the super_copy, it won't get remembered
1964 * for the next commit
1966 memcpy(&info->super_copy->super_roots,
1967 &info->super_for_commit->super_roots,
1968 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1972 * this copies info out of the root backup array and back into
1973 * the in-memory super block. It is meant to help iterate through
1974 * the array, so you send it the number of backups you've already
1975 * tried and the last backup index you used.
1977 * this returns -1 when it has tried all the backups
1979 static noinline int next_root_backup(struct btrfs_fs_info *info,
1980 struct btrfs_super_block *super,
1981 int *num_backups_tried, int *backup_index)
1983 struct btrfs_root_backup *root_backup;
1984 int newest = *backup_index;
1986 if (*num_backups_tried == 0) {
1987 u64 gen = btrfs_super_generation(super);
1989 newest = find_newest_super_backup(info, gen);
1993 *backup_index = newest;
1994 *num_backups_tried = 1;
1995 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1996 /* we've tried all the backups, all done */
1999 /* jump to the next oldest backup */
2000 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2001 BTRFS_NUM_BACKUP_ROOTS;
2002 *backup_index = newest;
2003 *num_backups_tried += 1;
2005 root_backup = super->super_roots + newest;
2007 btrfs_set_super_generation(super,
2008 btrfs_backup_tree_root_gen(root_backup));
2009 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2010 btrfs_set_super_root_level(super,
2011 btrfs_backup_tree_root_level(root_backup));
2012 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2015 * fixme: the total bytes and num_devices need to match or we should
2018 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2019 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2023 /* helper to cleanup workers */
2024 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2026 btrfs_destroy_workqueue(fs_info->fixup_workers);
2027 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2028 btrfs_destroy_workqueue(fs_info->workers);
2029 btrfs_destroy_workqueue(fs_info->endio_workers);
2030 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2031 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2032 btrfs_destroy_workqueue(fs_info->rmw_workers);
2033 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2034 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2035 btrfs_destroy_workqueue(fs_info->submit_workers);
2036 btrfs_destroy_workqueue(fs_info->delayed_workers);
2037 btrfs_destroy_workqueue(fs_info->caching_workers);
2038 btrfs_destroy_workqueue(fs_info->readahead_workers);
2039 btrfs_destroy_workqueue(fs_info->flush_workers);
2040 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2041 btrfs_destroy_workqueue(fs_info->extent_workers);
2043 * Now that all other work queues are destroyed, we can safely destroy
2044 * the queues used for metadata I/O, since tasks from those other work
2045 * queues can do metadata I/O operations.
2047 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2048 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2051 static void free_root_extent_buffers(struct btrfs_root *root)
2054 free_extent_buffer(root->node);
2055 free_extent_buffer(root->commit_root);
2057 root->commit_root = NULL;
2061 /* helper to cleanup tree roots */
2062 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2064 free_root_extent_buffers(info->tree_root);
2066 free_root_extent_buffers(info->dev_root);
2067 free_root_extent_buffers(info->extent_root);
2068 free_root_extent_buffers(info->csum_root);
2069 free_root_extent_buffers(info->quota_root);
2070 free_root_extent_buffers(info->uuid_root);
2071 if (free_chunk_root)
2072 free_root_extent_buffers(info->chunk_root);
2073 free_root_extent_buffers(info->free_space_root);
2076 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2079 struct btrfs_root *gang[8];
2082 while (!list_empty(&fs_info->dead_roots)) {
2083 gang[0] = list_entry(fs_info->dead_roots.next,
2084 struct btrfs_root, root_list);
2085 list_del(&gang[0]->root_list);
2087 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2088 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2090 free_extent_buffer(gang[0]->node);
2091 free_extent_buffer(gang[0]->commit_root);
2092 btrfs_put_fs_root(gang[0]);
2097 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2102 for (i = 0; i < ret; i++)
2103 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2106 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2107 btrfs_free_log_root_tree(NULL, fs_info);
2108 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2112 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2114 mutex_init(&fs_info->scrub_lock);
2115 atomic_set(&fs_info->scrubs_running, 0);
2116 atomic_set(&fs_info->scrub_pause_req, 0);
2117 atomic_set(&fs_info->scrubs_paused, 0);
2118 atomic_set(&fs_info->scrub_cancel_req, 0);
2119 init_waitqueue_head(&fs_info->scrub_pause_wait);
2120 fs_info->scrub_workers_refcnt = 0;
2123 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2125 spin_lock_init(&fs_info->balance_lock);
2126 mutex_init(&fs_info->balance_mutex);
2127 atomic_set(&fs_info->balance_running, 0);
2128 atomic_set(&fs_info->balance_pause_req, 0);
2129 atomic_set(&fs_info->balance_cancel_req, 0);
2130 fs_info->balance_ctl = NULL;
2131 init_waitqueue_head(&fs_info->balance_wait_q);
2134 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2136 struct inode *inode = fs_info->btree_inode;
2138 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2139 set_nlink(inode, 1);
2141 * we set the i_size on the btree inode to the max possible int.
2142 * the real end of the address space is determined by all of
2143 * the devices in the system
2145 inode->i_size = OFFSET_MAX;
2146 inode->i_mapping->a_ops = &btree_aops;
2148 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2149 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2150 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2151 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2153 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2155 BTRFS_I(inode)->root = fs_info->tree_root;
2156 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2157 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2158 btrfs_insert_inode_hash(inode);
2161 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2163 fs_info->dev_replace.lock_owner = 0;
2164 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2165 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2166 rwlock_init(&fs_info->dev_replace.lock);
2167 atomic_set(&fs_info->dev_replace.read_locks, 0);
2168 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2169 init_waitqueue_head(&fs_info->replace_wait);
2170 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2173 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2175 spin_lock_init(&fs_info->qgroup_lock);
2176 mutex_init(&fs_info->qgroup_ioctl_lock);
2177 fs_info->qgroup_tree = RB_ROOT;
2178 fs_info->qgroup_op_tree = RB_ROOT;
2179 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2180 fs_info->qgroup_seq = 1;
2181 fs_info->qgroup_ulist = NULL;
2182 fs_info->qgroup_rescan_running = false;
2183 mutex_init(&fs_info->qgroup_rescan_lock);
2186 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2187 struct btrfs_fs_devices *fs_devices)
2189 int max_active = fs_info->thread_pool_size;
2190 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2193 btrfs_alloc_workqueue(fs_info, "worker",
2194 flags | WQ_HIGHPRI, max_active, 16);
2196 fs_info->delalloc_workers =
2197 btrfs_alloc_workqueue(fs_info, "delalloc",
2198 flags, max_active, 2);
2200 fs_info->flush_workers =
2201 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2202 flags, max_active, 0);
2204 fs_info->caching_workers =
2205 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2208 * a higher idle thresh on the submit workers makes it much more
2209 * likely that bios will be send down in a sane order to the
2212 fs_info->submit_workers =
2213 btrfs_alloc_workqueue(fs_info, "submit", flags,
2214 min_t(u64, fs_devices->num_devices,
2217 fs_info->fixup_workers =
2218 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2221 * endios are largely parallel and should have a very
2224 fs_info->endio_workers =
2225 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2226 fs_info->endio_meta_workers =
2227 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2229 fs_info->endio_meta_write_workers =
2230 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2232 fs_info->endio_raid56_workers =
2233 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2235 fs_info->endio_repair_workers =
2236 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2237 fs_info->rmw_workers =
2238 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2239 fs_info->endio_write_workers =
2240 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2242 fs_info->endio_freespace_worker =
2243 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2245 fs_info->delayed_workers =
2246 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2248 fs_info->readahead_workers =
2249 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2251 fs_info->qgroup_rescan_workers =
2252 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2253 fs_info->extent_workers =
2254 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2255 min_t(u64, fs_devices->num_devices,
2258 if (!(fs_info->workers && fs_info->delalloc_workers &&
2259 fs_info->submit_workers && fs_info->flush_workers &&
2260 fs_info->endio_workers && fs_info->endio_meta_workers &&
2261 fs_info->endio_meta_write_workers &&
2262 fs_info->endio_repair_workers &&
2263 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2264 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2265 fs_info->caching_workers && fs_info->readahead_workers &&
2266 fs_info->fixup_workers && fs_info->delayed_workers &&
2267 fs_info->extent_workers &&
2268 fs_info->qgroup_rescan_workers)) {
2275 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2276 struct btrfs_fs_devices *fs_devices)
2279 struct btrfs_root *log_tree_root;
2280 struct btrfs_super_block *disk_super = fs_info->super_copy;
2281 u64 bytenr = btrfs_super_log_root(disk_super);
2283 if (fs_devices->rw_devices == 0) {
2284 btrfs_warn(fs_info, "log replay required on RO media");
2288 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2292 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2294 log_tree_root->node = read_tree_block(fs_info, bytenr,
2295 fs_info->generation + 1);
2296 if (IS_ERR(log_tree_root->node)) {
2297 btrfs_warn(fs_info, "failed to read log tree");
2298 ret = PTR_ERR(log_tree_root->node);
2299 kfree(log_tree_root);
2301 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2302 btrfs_err(fs_info, "failed to read log tree");
2303 free_extent_buffer(log_tree_root->node);
2304 kfree(log_tree_root);
2307 /* returns with log_tree_root freed on success */
2308 ret = btrfs_recover_log_trees(log_tree_root);
2310 btrfs_handle_fs_error(fs_info, ret,
2311 "Failed to recover log tree");
2312 free_extent_buffer(log_tree_root->node);
2313 kfree(log_tree_root);
2317 if (sb_rdonly(fs_info->sb)) {
2318 ret = btrfs_commit_super(fs_info);
2326 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2328 struct btrfs_root *tree_root = fs_info->tree_root;
2329 struct btrfs_root *root;
2330 struct btrfs_key location;
2333 BUG_ON(!fs_info->tree_root);
2335 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2336 location.type = BTRFS_ROOT_ITEM_KEY;
2337 location.offset = 0;
2339 root = btrfs_read_tree_root(tree_root, &location);
2341 return PTR_ERR(root);
2342 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2343 fs_info->extent_root = root;
2345 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2346 root = btrfs_read_tree_root(tree_root, &location);
2348 return PTR_ERR(root);
2349 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2350 fs_info->dev_root = root;
2351 btrfs_init_devices_late(fs_info);
2353 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2354 root = btrfs_read_tree_root(tree_root, &location);
2356 return PTR_ERR(root);
2357 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2358 fs_info->csum_root = root;
2360 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2361 root = btrfs_read_tree_root(tree_root, &location);
2362 if (!IS_ERR(root)) {
2363 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2364 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2365 fs_info->quota_root = root;
2368 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2369 root = btrfs_read_tree_root(tree_root, &location);
2371 ret = PTR_ERR(root);
2375 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2376 fs_info->uuid_root = root;
2379 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2380 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2381 root = btrfs_read_tree_root(tree_root, &location);
2383 return PTR_ERR(root);
2384 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2385 fs_info->free_space_root = root;
2391 int open_ctree(struct super_block *sb,
2392 struct btrfs_fs_devices *fs_devices,
2400 struct btrfs_key location;
2401 struct buffer_head *bh;
2402 struct btrfs_super_block *disk_super;
2403 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2404 struct btrfs_root *tree_root;
2405 struct btrfs_root *chunk_root;
2408 int num_backups_tried = 0;
2409 int backup_index = 0;
2411 int clear_free_space_tree = 0;
2413 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2414 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2415 if (!tree_root || !chunk_root) {
2420 ret = init_srcu_struct(&fs_info->subvol_srcu);
2426 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2431 fs_info->dirty_metadata_batch = PAGE_SIZE *
2432 (1 + ilog2(nr_cpu_ids));
2434 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2437 goto fail_dirty_metadata_bytes;
2440 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2443 goto fail_delalloc_bytes;
2446 fs_info->btree_inode = new_inode(sb);
2447 if (!fs_info->btree_inode) {
2449 goto fail_bio_counter;
2452 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2454 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2455 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2456 INIT_LIST_HEAD(&fs_info->trans_list);
2457 INIT_LIST_HEAD(&fs_info->dead_roots);
2458 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2459 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2460 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2461 spin_lock_init(&fs_info->delalloc_root_lock);
2462 spin_lock_init(&fs_info->trans_lock);
2463 spin_lock_init(&fs_info->fs_roots_radix_lock);
2464 spin_lock_init(&fs_info->delayed_iput_lock);
2465 spin_lock_init(&fs_info->defrag_inodes_lock);
2466 spin_lock_init(&fs_info->super_lock);
2467 spin_lock_init(&fs_info->qgroup_op_lock);
2468 spin_lock_init(&fs_info->buffer_lock);
2469 spin_lock_init(&fs_info->unused_bgs_lock);
2470 rwlock_init(&fs_info->tree_mod_log_lock);
2471 mutex_init(&fs_info->unused_bg_unpin_mutex);
2472 mutex_init(&fs_info->delete_unused_bgs_mutex);
2473 mutex_init(&fs_info->reloc_mutex);
2474 mutex_init(&fs_info->delalloc_root_mutex);
2475 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2476 seqlock_init(&fs_info->profiles_lock);
2478 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2479 INIT_LIST_HEAD(&fs_info->space_info);
2480 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2481 INIT_LIST_HEAD(&fs_info->unused_bgs);
2482 btrfs_mapping_init(&fs_info->mapping_tree);
2483 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2484 BTRFS_BLOCK_RSV_GLOBAL);
2485 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2486 BTRFS_BLOCK_RSV_DELALLOC);
2487 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2488 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2489 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2490 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2491 BTRFS_BLOCK_RSV_DELOPS);
2492 atomic_set(&fs_info->nr_async_submits, 0);
2493 atomic_set(&fs_info->async_delalloc_pages, 0);
2494 atomic_set(&fs_info->async_submit_draining, 0);
2495 atomic_set(&fs_info->nr_async_bios, 0);
2496 atomic_set(&fs_info->defrag_running, 0);
2497 atomic_set(&fs_info->qgroup_op_seq, 0);
2498 atomic_set(&fs_info->reada_works_cnt, 0);
2499 atomic64_set(&fs_info->tree_mod_seq, 0);
2501 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2502 fs_info->metadata_ratio = 0;
2503 fs_info->defrag_inodes = RB_ROOT;
2504 atomic64_set(&fs_info->free_chunk_space, 0);
2505 fs_info->tree_mod_log = RB_ROOT;
2506 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2507 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2508 /* readahead state */
2509 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2510 spin_lock_init(&fs_info->reada_lock);
2512 fs_info->thread_pool_size = min_t(unsigned long,
2513 num_online_cpus() + 2, 8);
2515 INIT_LIST_HEAD(&fs_info->ordered_roots);
2516 spin_lock_init(&fs_info->ordered_root_lock);
2517 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2519 if (!fs_info->delayed_root) {
2523 btrfs_init_delayed_root(fs_info->delayed_root);
2525 btrfs_init_scrub(fs_info);
2526 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2527 fs_info->check_integrity_print_mask = 0;
2529 btrfs_init_balance(fs_info);
2530 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2532 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2533 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2535 btrfs_init_btree_inode(fs_info);
2537 spin_lock_init(&fs_info->block_group_cache_lock);
2538 fs_info->block_group_cache_tree = RB_ROOT;
2539 fs_info->first_logical_byte = (u64)-1;
2541 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2542 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2543 fs_info->pinned_extents = &fs_info->freed_extents[0];
2544 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2546 mutex_init(&fs_info->ordered_operations_mutex);
2547 mutex_init(&fs_info->tree_log_mutex);
2548 mutex_init(&fs_info->chunk_mutex);
2549 mutex_init(&fs_info->transaction_kthread_mutex);
2550 mutex_init(&fs_info->cleaner_mutex);
2551 mutex_init(&fs_info->volume_mutex);
2552 mutex_init(&fs_info->ro_block_group_mutex);
2553 init_rwsem(&fs_info->commit_root_sem);
2554 init_rwsem(&fs_info->cleanup_work_sem);
2555 init_rwsem(&fs_info->subvol_sem);
2556 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2558 btrfs_init_dev_replace_locks(fs_info);
2559 btrfs_init_qgroup(fs_info);
2561 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2562 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2564 init_waitqueue_head(&fs_info->transaction_throttle);
2565 init_waitqueue_head(&fs_info->transaction_wait);
2566 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2567 init_waitqueue_head(&fs_info->async_submit_wait);
2569 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2571 /* Usable values until the real ones are cached from the superblock */
2572 fs_info->nodesize = 4096;
2573 fs_info->sectorsize = 4096;
2574 fs_info->stripesize = 4096;
2576 ret = btrfs_alloc_stripe_hash_table(fs_info);
2582 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2584 invalidate_bdev(fs_devices->latest_bdev);
2587 * Read super block and check the signature bytes only
2589 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2596 * We want to check superblock checksum, the type is stored inside.
2597 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2599 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2600 btrfs_err(fs_info, "superblock checksum mismatch");
2607 * super_copy is zeroed at allocation time and we never touch the
2608 * following bytes up to INFO_SIZE, the checksum is calculated from
2609 * the whole block of INFO_SIZE
2611 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2612 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2613 sizeof(*fs_info->super_for_commit));
2616 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2618 ret = btrfs_check_super_valid(fs_info);
2620 btrfs_err(fs_info, "superblock contains fatal errors");
2625 disk_super = fs_info->super_copy;
2626 if (!btrfs_super_root(disk_super))
2629 /* check FS state, whether FS is broken. */
2630 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2631 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2634 * run through our array of backup supers and setup
2635 * our ring pointer to the oldest one
2637 generation = btrfs_super_generation(disk_super);
2638 find_oldest_super_backup(fs_info, generation);
2641 * In the long term, we'll store the compression type in the super
2642 * block, and it'll be used for per file compression control.
2644 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2646 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2652 features = btrfs_super_incompat_flags(disk_super) &
2653 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2656 "cannot mount because of unsupported optional features (0x%llx)",
2662 features = btrfs_super_incompat_flags(disk_super);
2663 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2664 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2665 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2666 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2667 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2669 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2670 btrfs_info(fs_info, "has skinny extents");
2673 * flag our filesystem as having big metadata blocks if
2674 * they are bigger than the page size
2676 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2677 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2679 "flagging fs with big metadata feature");
2680 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2683 nodesize = btrfs_super_nodesize(disk_super);
2684 sectorsize = btrfs_super_sectorsize(disk_super);
2685 stripesize = sectorsize;
2686 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2687 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2689 /* Cache block sizes */
2690 fs_info->nodesize = nodesize;
2691 fs_info->sectorsize = sectorsize;
2692 fs_info->stripesize = stripesize;
2695 * mixed block groups end up with duplicate but slightly offset
2696 * extent buffers for the same range. It leads to corruptions
2698 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2699 (sectorsize != nodesize)) {
2701 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2702 nodesize, sectorsize);
2707 * Needn't use the lock because there is no other task which will
2710 btrfs_set_super_incompat_flags(disk_super, features);
2712 features = btrfs_super_compat_ro_flags(disk_super) &
2713 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2714 if (!sb_rdonly(sb) && features) {
2716 "cannot mount read-write because of unsupported optional features (0x%llx)",
2722 max_active = fs_info->thread_pool_size;
2724 ret = btrfs_init_workqueues(fs_info, fs_devices);
2727 goto fail_sb_buffer;
2730 sb->s_bdi->congested_fn = btrfs_congested_fn;
2731 sb->s_bdi->congested_data = fs_info;
2732 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2733 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
2734 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2735 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2737 sb->s_blocksize = sectorsize;
2738 sb->s_blocksize_bits = blksize_bits(sectorsize);
2740 mutex_lock(&fs_info->chunk_mutex);
2741 ret = btrfs_read_sys_array(fs_info);
2742 mutex_unlock(&fs_info->chunk_mutex);
2744 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2745 goto fail_sb_buffer;
2748 generation = btrfs_super_chunk_root_generation(disk_super);
2750 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2752 chunk_root->node = read_tree_block(fs_info,
2753 btrfs_super_chunk_root(disk_super),
2755 if (IS_ERR(chunk_root->node) ||
2756 !extent_buffer_uptodate(chunk_root->node)) {
2757 btrfs_err(fs_info, "failed to read chunk root");
2758 if (!IS_ERR(chunk_root->node))
2759 free_extent_buffer(chunk_root->node);
2760 chunk_root->node = NULL;
2761 goto fail_tree_roots;
2763 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2764 chunk_root->commit_root = btrfs_root_node(chunk_root);
2766 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2767 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2769 ret = btrfs_read_chunk_tree(fs_info);
2771 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2772 goto fail_tree_roots;
2776 * keep the device that is marked to be the target device for the
2777 * dev_replace procedure
2779 btrfs_close_extra_devices(fs_devices, 0);
2781 if (!fs_devices->latest_bdev) {
2782 btrfs_err(fs_info, "failed to read devices");
2783 goto fail_tree_roots;
2787 generation = btrfs_super_generation(disk_super);
2789 tree_root->node = read_tree_block(fs_info,
2790 btrfs_super_root(disk_super),
2792 if (IS_ERR(tree_root->node) ||
2793 !extent_buffer_uptodate(tree_root->node)) {
2794 btrfs_warn(fs_info, "failed to read tree root");
2795 if (!IS_ERR(tree_root->node))
2796 free_extent_buffer(tree_root->node);
2797 tree_root->node = NULL;
2798 goto recovery_tree_root;
2801 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2802 tree_root->commit_root = btrfs_root_node(tree_root);
2803 btrfs_set_root_refs(&tree_root->root_item, 1);
2805 mutex_lock(&tree_root->objectid_mutex);
2806 ret = btrfs_find_highest_objectid(tree_root,
2807 &tree_root->highest_objectid);
2809 mutex_unlock(&tree_root->objectid_mutex);
2810 goto recovery_tree_root;
2813 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2815 mutex_unlock(&tree_root->objectid_mutex);
2817 ret = btrfs_read_roots(fs_info);
2819 goto recovery_tree_root;
2821 fs_info->generation = generation;
2822 fs_info->last_trans_committed = generation;
2824 ret = btrfs_recover_balance(fs_info);
2826 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2827 goto fail_block_groups;
2830 ret = btrfs_init_dev_stats(fs_info);
2832 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2833 goto fail_block_groups;
2836 ret = btrfs_init_dev_replace(fs_info);
2838 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2839 goto fail_block_groups;
2842 btrfs_close_extra_devices(fs_devices, 1);
2844 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2846 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2848 goto fail_block_groups;
2851 ret = btrfs_sysfs_add_device(fs_devices);
2853 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2855 goto fail_fsdev_sysfs;
2858 ret = btrfs_sysfs_add_mounted(fs_info);
2860 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2861 goto fail_fsdev_sysfs;
2864 ret = btrfs_init_space_info(fs_info);
2866 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2870 ret = btrfs_read_block_groups(fs_info);
2872 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2876 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info)) {
2878 "writeable mount is not allowed due to too many missing devices");
2882 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2884 if (IS_ERR(fs_info->cleaner_kthread))
2887 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2889 "btrfs-transaction");
2890 if (IS_ERR(fs_info->transaction_kthread))
2893 if (!btrfs_test_opt(fs_info, NOSSD) &&
2894 !fs_info->fs_devices->rotating) {
2895 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2899 * Mount does not set all options immediately, we can do it now and do
2900 * not have to wait for transaction commit
2902 btrfs_apply_pending_changes(fs_info);
2904 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2905 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2906 ret = btrfsic_mount(fs_info, fs_devices,
2907 btrfs_test_opt(fs_info,
2908 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2910 fs_info->check_integrity_print_mask);
2913 "failed to initialize integrity check module: %d",
2917 ret = btrfs_read_qgroup_config(fs_info);
2919 goto fail_trans_kthread;
2921 /* do not make disk changes in broken FS or nologreplay is given */
2922 if (btrfs_super_log_root(disk_super) != 0 &&
2923 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2924 btrfs_info(fs_info, "start tree-log replay");
2925 ret = btrfs_replay_log(fs_info, fs_devices);
2932 ret = btrfs_find_orphan_roots(fs_info);
2936 if (!sb_rdonly(sb)) {
2937 ret = btrfs_cleanup_fs_roots(fs_info);
2941 mutex_lock(&fs_info->cleaner_mutex);
2942 ret = btrfs_recover_relocation(tree_root);
2943 mutex_unlock(&fs_info->cleaner_mutex);
2945 btrfs_warn(fs_info, "failed to recover relocation: %d",
2952 location.objectid = BTRFS_FS_TREE_OBJECTID;
2953 location.type = BTRFS_ROOT_ITEM_KEY;
2954 location.offset = 0;
2956 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2957 if (IS_ERR(fs_info->fs_root)) {
2958 err = PTR_ERR(fs_info->fs_root);
2965 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2966 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2967 clear_free_space_tree = 1;
2968 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2969 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2970 btrfs_warn(fs_info, "free space tree is invalid");
2971 clear_free_space_tree = 1;
2974 if (clear_free_space_tree) {
2975 btrfs_info(fs_info, "clearing free space tree");
2976 ret = btrfs_clear_free_space_tree(fs_info);
2979 "failed to clear free space tree: %d", ret);
2980 close_ctree(fs_info);
2985 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
2986 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2987 btrfs_info(fs_info, "creating free space tree");
2988 ret = btrfs_create_free_space_tree(fs_info);
2991 "failed to create free space tree: %d", ret);
2992 close_ctree(fs_info);
2997 down_read(&fs_info->cleanup_work_sem);
2998 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2999 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3000 up_read(&fs_info->cleanup_work_sem);
3001 close_ctree(fs_info);
3004 up_read(&fs_info->cleanup_work_sem);
3006 ret = btrfs_resume_balance_async(fs_info);
3008 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3009 close_ctree(fs_info);
3013 ret = btrfs_resume_dev_replace_async(fs_info);
3015 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3016 close_ctree(fs_info);
3020 btrfs_qgroup_rescan_resume(fs_info);
3022 if (!fs_info->uuid_root) {
3023 btrfs_info(fs_info, "creating UUID tree");
3024 ret = btrfs_create_uuid_tree(fs_info);
3027 "failed to create the UUID tree: %d", ret);
3028 close_ctree(fs_info);
3031 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3032 fs_info->generation !=
3033 btrfs_super_uuid_tree_generation(disk_super)) {
3034 btrfs_info(fs_info, "checking UUID tree");
3035 ret = btrfs_check_uuid_tree(fs_info);
3038 "failed to check the UUID tree: %d", ret);
3039 close_ctree(fs_info);
3043 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3045 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3048 * backuproot only affect mount behavior, and if open_ctree succeeded,
3049 * no need to keep the flag
3051 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3056 btrfs_free_qgroup_config(fs_info);
3058 kthread_stop(fs_info->transaction_kthread);
3059 btrfs_cleanup_transaction(fs_info);
3060 btrfs_free_fs_roots(fs_info);
3062 kthread_stop(fs_info->cleaner_kthread);
3065 * make sure we're done with the btree inode before we stop our
3068 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3071 btrfs_sysfs_remove_mounted(fs_info);
3074 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3077 btrfs_put_block_group_cache(fs_info);
3080 free_root_pointers(fs_info, true);
3081 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3084 btrfs_stop_all_workers(fs_info);
3085 btrfs_free_block_groups(fs_info);
3088 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3090 iput(fs_info->btree_inode);
3092 percpu_counter_destroy(&fs_info->bio_counter);
3093 fail_delalloc_bytes:
3094 percpu_counter_destroy(&fs_info->delalloc_bytes);
3095 fail_dirty_metadata_bytes:
3096 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3098 cleanup_srcu_struct(&fs_info->subvol_srcu);
3100 btrfs_free_stripe_hash_table(fs_info);
3101 btrfs_close_devices(fs_info->fs_devices);
3105 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3106 goto fail_tree_roots;
3108 free_root_pointers(fs_info, false);
3110 /* don't use the log in recovery mode, it won't be valid */
3111 btrfs_set_super_log_root(disk_super, 0);
3113 /* we can't trust the free space cache either */
3114 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3116 ret = next_root_backup(fs_info, fs_info->super_copy,
3117 &num_backups_tried, &backup_index);
3119 goto fail_block_groups;
3120 goto retry_root_backup;
3123 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3126 set_buffer_uptodate(bh);
3128 struct btrfs_device *device = (struct btrfs_device *)
3131 btrfs_warn_rl_in_rcu(device->fs_info,
3132 "lost page write due to IO error on %s",
3133 rcu_str_deref(device->name));
3134 /* note, we don't set_buffer_write_io_error because we have
3135 * our own ways of dealing with the IO errors
3137 clear_buffer_uptodate(bh);
3138 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3144 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3145 struct buffer_head **bh_ret)
3147 struct buffer_head *bh;
3148 struct btrfs_super_block *super;
3151 bytenr = btrfs_sb_offset(copy_num);
3152 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3155 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3157 * If we fail to read from the underlying devices, as of now
3158 * the best option we have is to mark it EIO.
3163 super = (struct btrfs_super_block *)bh->b_data;
3164 if (btrfs_super_bytenr(super) != bytenr ||
3165 btrfs_super_magic(super) != BTRFS_MAGIC) {
3175 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3177 struct buffer_head *bh;
3178 struct buffer_head *latest = NULL;
3179 struct btrfs_super_block *super;
3184 /* we would like to check all the supers, but that would make
3185 * a btrfs mount succeed after a mkfs from a different FS.
3186 * So, we need to add a special mount option to scan for
3187 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3189 for (i = 0; i < 1; i++) {
3190 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3194 super = (struct btrfs_super_block *)bh->b_data;
3196 if (!latest || btrfs_super_generation(super) > transid) {
3199 transid = btrfs_super_generation(super);
3206 return ERR_PTR(ret);
3212 * Write superblock @sb to the @device. Do not wait for completion, all the
3213 * buffer heads we write are pinned.
3215 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3216 * the expected device size at commit time. Note that max_mirrors must be
3217 * same for write and wait phases.
3219 * Return number of errors when buffer head is not found or submission fails.
3221 static int write_dev_supers(struct btrfs_device *device,
3222 struct btrfs_super_block *sb, int max_mirrors)
3224 struct buffer_head *bh;
3232 if (max_mirrors == 0)
3233 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3235 for (i = 0; i < max_mirrors; i++) {
3236 bytenr = btrfs_sb_offset(i);
3237 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3238 device->commit_total_bytes)
3241 btrfs_set_super_bytenr(sb, bytenr);
3244 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3245 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3246 btrfs_csum_final(crc, sb->csum);
3248 /* One reference for us, and we leave it for the caller */
3249 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3250 BTRFS_SUPER_INFO_SIZE);
3252 btrfs_err(device->fs_info,
3253 "couldn't get super buffer head for bytenr %llu",
3259 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3261 /* one reference for submit_bh */
3264 set_buffer_uptodate(bh);
3266 bh->b_end_io = btrfs_end_buffer_write_sync;
3267 bh->b_private = device;
3270 * we fua the first super. The others we allow
3273 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3274 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3275 op_flags |= REQ_FUA;
3276 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3280 return errors < i ? 0 : -1;
3284 * Wait for write completion of superblocks done by write_dev_supers,
3285 * @max_mirrors same for write and wait phases.
3287 * Return number of errors when buffer head is not found or not marked up to
3290 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3292 struct buffer_head *bh;
3297 if (max_mirrors == 0)
3298 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3300 for (i = 0; i < max_mirrors; i++) {
3301 bytenr = btrfs_sb_offset(i);
3302 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3303 device->commit_total_bytes)
3306 bh = __find_get_block(device->bdev,
3307 bytenr / BTRFS_BDEV_BLOCKSIZE,
3308 BTRFS_SUPER_INFO_SIZE);
3314 if (!buffer_uptodate(bh))
3317 /* drop our reference */
3320 /* drop the reference from the writing run */
3324 return errors < i ? 0 : -1;
3328 * endio for the write_dev_flush, this will wake anyone waiting
3329 * for the barrier when it is done
3331 static void btrfs_end_empty_barrier(struct bio *bio)
3333 complete(bio->bi_private);
3337 * Submit a flush request to the device if it supports it. Error handling is
3338 * done in the waiting counterpart.
3340 static void write_dev_flush(struct btrfs_device *device)
3342 struct bio *bio = device->flush_bio;
3344 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3346 * When a disk has write caching disabled, we skip submission of a bio
3347 * with flush and sync requests before writing the superblock, since
3348 * it's not needed. However when the integrity checker is enabled, this
3349 * results in reports that there are metadata blocks referred by a
3350 * superblock that were not properly flushed. So don't skip the bio
3351 * submission only when the integrity checker is enabled for the sake
3352 * of simplicity, since this is a debug tool and not meant for use in
3355 struct request_queue *q = bdev_get_queue(device->bdev);
3356 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3361 bio->bi_end_io = btrfs_end_empty_barrier;
3362 bio_set_dev(bio, device->bdev);
3363 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3364 init_completion(&device->flush_wait);
3365 bio->bi_private = &device->flush_wait;
3367 btrfsic_submit_bio(bio);
3368 device->flush_bio_sent = 1;
3372 * If the flush bio has been submitted by write_dev_flush, wait for it.
3374 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3376 struct bio *bio = device->flush_bio;
3378 if (!device->flush_bio_sent)
3381 device->flush_bio_sent = 0;
3382 wait_for_completion_io(&device->flush_wait);
3384 return bio->bi_status;
3387 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3389 if (!btrfs_check_rw_degradable(fs_info))
3395 * send an empty flush down to each device in parallel,
3396 * then wait for them
3398 static int barrier_all_devices(struct btrfs_fs_info *info)
3400 struct list_head *head;
3401 struct btrfs_device *dev;
3402 int errors_wait = 0;
3405 /* send down all the barriers */
3406 head = &info->fs_devices->devices;
3407 list_for_each_entry_rcu(dev, head, dev_list) {
3412 if (!dev->in_fs_metadata || !dev->writeable)
3415 write_dev_flush(dev);
3416 dev->last_flush_error = BLK_STS_OK;
3419 /* wait for all the barriers */
3420 list_for_each_entry_rcu(dev, head, dev_list) {
3427 if (!dev->in_fs_metadata || !dev->writeable)
3430 ret = wait_dev_flush(dev);
3432 dev->last_flush_error = ret;
3433 btrfs_dev_stat_inc_and_print(dev,
3434 BTRFS_DEV_STAT_FLUSH_ERRS);
3441 * At some point we need the status of all disks
3442 * to arrive at the volume status. So error checking
3443 * is being pushed to a separate loop.
3445 return check_barrier_error(info);
3450 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3453 int min_tolerated = INT_MAX;
3455 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3456 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3457 min_tolerated = min(min_tolerated,
3458 btrfs_raid_array[BTRFS_RAID_SINGLE].
3459 tolerated_failures);
3461 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3462 if (raid_type == BTRFS_RAID_SINGLE)
3464 if (!(flags & btrfs_raid_group[raid_type]))
3466 min_tolerated = min(min_tolerated,
3467 btrfs_raid_array[raid_type].
3468 tolerated_failures);
3471 if (min_tolerated == INT_MAX) {
3472 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3476 return min_tolerated;
3479 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3481 struct list_head *head;
3482 struct btrfs_device *dev;
3483 struct btrfs_super_block *sb;
3484 struct btrfs_dev_item *dev_item;
3488 int total_errors = 0;
3491 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3494 * max_mirrors == 0 indicates we're from commit_transaction,
3495 * not from fsync where the tree roots in fs_info have not
3496 * been consistent on disk.
3498 if (max_mirrors == 0)
3499 backup_super_roots(fs_info);
3501 sb = fs_info->super_for_commit;
3502 dev_item = &sb->dev_item;
3504 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3505 head = &fs_info->fs_devices->devices;
3506 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3509 ret = barrier_all_devices(fs_info);
3512 &fs_info->fs_devices->device_list_mutex);
3513 btrfs_handle_fs_error(fs_info, ret,
3514 "errors while submitting device barriers.");
3519 list_for_each_entry_rcu(dev, head, dev_list) {
3524 if (!dev->in_fs_metadata || !dev->writeable)
3527 btrfs_set_stack_device_generation(dev_item, 0);
3528 btrfs_set_stack_device_type(dev_item, dev->type);
3529 btrfs_set_stack_device_id(dev_item, dev->devid);
3530 btrfs_set_stack_device_total_bytes(dev_item,
3531 dev->commit_total_bytes);
3532 btrfs_set_stack_device_bytes_used(dev_item,
3533 dev->commit_bytes_used);
3534 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3535 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3536 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3537 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3538 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3540 flags = btrfs_super_flags(sb);
3541 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3543 ret = write_dev_supers(dev, sb, max_mirrors);
3547 if (total_errors > max_errors) {
3548 btrfs_err(fs_info, "%d errors while writing supers",
3550 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3552 /* FUA is masked off if unsupported and can't be the reason */
3553 btrfs_handle_fs_error(fs_info, -EIO,
3554 "%d errors while writing supers",
3560 list_for_each_entry_rcu(dev, head, dev_list) {
3563 if (!dev->in_fs_metadata || !dev->writeable)
3566 ret = wait_dev_supers(dev, max_mirrors);
3570 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3571 if (total_errors > max_errors) {
3572 btrfs_handle_fs_error(fs_info, -EIO,
3573 "%d errors while writing supers",
3580 /* Drop a fs root from the radix tree and free it. */
3581 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3582 struct btrfs_root *root)
3584 spin_lock(&fs_info->fs_roots_radix_lock);
3585 radix_tree_delete(&fs_info->fs_roots_radix,
3586 (unsigned long)root->root_key.objectid);
3587 spin_unlock(&fs_info->fs_roots_radix_lock);
3589 if (btrfs_root_refs(&root->root_item) == 0)
3590 synchronize_srcu(&fs_info->subvol_srcu);
3592 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3593 btrfs_free_log(NULL, root);
3594 if (root->reloc_root) {
3595 free_extent_buffer(root->reloc_root->node);
3596 free_extent_buffer(root->reloc_root->commit_root);
3597 btrfs_put_fs_root(root->reloc_root);
3598 root->reloc_root = NULL;
3602 if (root->free_ino_pinned)
3603 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3604 if (root->free_ino_ctl)
3605 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3609 static void free_fs_root(struct btrfs_root *root)
3611 iput(root->ino_cache_inode);
3612 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3613 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3614 root->orphan_block_rsv = NULL;
3616 free_anon_bdev(root->anon_dev);
3617 if (root->subv_writers)
3618 btrfs_free_subvolume_writers(root->subv_writers);
3619 free_extent_buffer(root->node);
3620 free_extent_buffer(root->commit_root);
3621 kfree(root->free_ino_ctl);
3622 kfree(root->free_ino_pinned);
3624 btrfs_put_fs_root(root);
3627 void btrfs_free_fs_root(struct btrfs_root *root)
3632 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3634 u64 root_objectid = 0;
3635 struct btrfs_root *gang[8];
3638 unsigned int ret = 0;
3642 index = srcu_read_lock(&fs_info->subvol_srcu);
3643 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3644 (void **)gang, root_objectid,
3647 srcu_read_unlock(&fs_info->subvol_srcu, index);
3650 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3652 for (i = 0; i < ret; i++) {
3653 /* Avoid to grab roots in dead_roots */
3654 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3658 /* grab all the search result for later use */
3659 gang[i] = btrfs_grab_fs_root(gang[i]);
3661 srcu_read_unlock(&fs_info->subvol_srcu, index);
3663 for (i = 0; i < ret; i++) {
3666 root_objectid = gang[i]->root_key.objectid;
3667 err = btrfs_orphan_cleanup(gang[i]);
3670 btrfs_put_fs_root(gang[i]);
3675 /* release the uncleaned roots due to error */
3676 for (; i < ret; i++) {
3678 btrfs_put_fs_root(gang[i]);
3683 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3685 struct btrfs_root *root = fs_info->tree_root;
3686 struct btrfs_trans_handle *trans;
3688 mutex_lock(&fs_info->cleaner_mutex);
3689 btrfs_run_delayed_iputs(fs_info);
3690 mutex_unlock(&fs_info->cleaner_mutex);
3691 wake_up_process(fs_info->cleaner_kthread);
3693 /* wait until ongoing cleanup work done */
3694 down_write(&fs_info->cleanup_work_sem);
3695 up_write(&fs_info->cleanup_work_sem);
3697 trans = btrfs_join_transaction(root);
3699 return PTR_ERR(trans);
3700 return btrfs_commit_transaction(trans);
3703 void close_ctree(struct btrfs_fs_info *fs_info)
3705 struct btrfs_root *root = fs_info->tree_root;
3708 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3710 * We don't want the cleaner to start new transactions, add more delayed
3711 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3712 * because that frees the task_struct, and the transaction kthread might
3713 * still try to wake up the cleaner.
3715 kthread_park(fs_info->cleaner_kthread);
3717 /* wait for the qgroup rescan worker to stop */
3718 btrfs_qgroup_wait_for_completion(fs_info, false);
3720 /* wait for the uuid_scan task to finish */
3721 down(&fs_info->uuid_tree_rescan_sem);
3722 /* avoid complains from lockdep et al., set sem back to initial state */
3723 up(&fs_info->uuid_tree_rescan_sem);
3725 /* pause restriper - we want to resume on mount */
3726 btrfs_pause_balance(fs_info);
3728 btrfs_dev_replace_suspend_for_unmount(fs_info);
3730 btrfs_scrub_cancel(fs_info);
3732 /* wait for any defraggers to finish */
3733 wait_event(fs_info->transaction_wait,
3734 (atomic_read(&fs_info->defrag_running) == 0));
3736 /* clear out the rbtree of defraggable inodes */
3737 btrfs_cleanup_defrag_inodes(fs_info);
3739 cancel_work_sync(&fs_info->async_reclaim_work);
3741 if (!sb_rdonly(fs_info->sb)) {
3743 * The cleaner kthread is stopped, so do one final pass over
3744 * unused block groups.
3746 btrfs_delete_unused_bgs(fs_info);
3749 * There might be existing delayed inode workers still running
3750 * and holding an empty delayed inode item. We must wait for
3751 * them to complete first because they can create a transaction.
3752 * This happens when someone calls btrfs_balance_delayed_items()
3753 * and then a transaction commit runs the same delayed nodes
3754 * before any delayed worker has done something with the nodes.
3755 * We must wait for any worker here and not at transaction
3756 * commit time since that could cause a deadlock.
3757 * This is a very rare case.
3759 btrfs_flush_workqueue(fs_info->delayed_workers);
3761 ret = btrfs_commit_super(fs_info);
3763 btrfs_err(fs_info, "commit super ret %d", ret);
3766 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3767 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3768 btrfs_error_commit_super(fs_info);
3770 kthread_stop(fs_info->transaction_kthread);
3771 kthread_stop(fs_info->cleaner_kthread);
3773 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3775 btrfs_free_qgroup_config(fs_info);
3776 ASSERT(list_empty(&fs_info->delalloc_roots));
3778 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3779 btrfs_info(fs_info, "at unmount delalloc count %lld",
3780 percpu_counter_sum(&fs_info->delalloc_bytes));
3783 btrfs_sysfs_remove_mounted(fs_info);
3784 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3786 btrfs_free_fs_roots(fs_info);
3788 btrfs_put_block_group_cache(fs_info);
3791 * we must make sure there is not any read request to
3792 * submit after we stopping all workers.
3794 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3795 btrfs_stop_all_workers(fs_info);
3797 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3798 free_root_pointers(fs_info, true);
3801 * We must free the block groups after dropping the fs_roots as we could
3802 * have had an IO error and have left over tree log blocks that aren't
3803 * cleaned up until the fs roots are freed. This makes the block group
3804 * accounting appear to be wrong because there's pending reserved bytes,
3805 * so make sure we do the block group cleanup afterwards.
3807 btrfs_free_block_groups(fs_info);
3809 iput(fs_info->btree_inode);
3811 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3812 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3813 btrfsic_unmount(fs_info->fs_devices);
3816 btrfs_close_devices(fs_info->fs_devices);
3817 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3819 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3820 percpu_counter_destroy(&fs_info->delalloc_bytes);
3821 percpu_counter_destroy(&fs_info->bio_counter);
3822 cleanup_srcu_struct(&fs_info->subvol_srcu);
3824 btrfs_free_stripe_hash_table(fs_info);
3826 __btrfs_free_block_rsv(root->orphan_block_rsv);
3827 root->orphan_block_rsv = NULL;
3829 while (!list_empty(&fs_info->pinned_chunks)) {
3830 struct extent_map *em;
3832 em = list_first_entry(&fs_info->pinned_chunks,
3833 struct extent_map, list);
3834 list_del_init(&em->list);
3835 free_extent_map(em);
3839 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3843 struct inode *btree_inode = buf->pages[0]->mapping->host;
3845 ret = extent_buffer_uptodate(buf);
3849 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3850 parent_transid, atomic);
3856 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3858 struct btrfs_fs_info *fs_info;
3859 struct btrfs_root *root;
3860 u64 transid = btrfs_header_generation(buf);
3863 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3865 * This is a fast path so only do this check if we have sanity tests
3866 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3867 * outside of the sanity tests.
3869 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3872 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3873 fs_info = root->fs_info;
3874 btrfs_assert_tree_locked(buf);
3875 if (transid != fs_info->generation)
3876 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3877 buf->start, transid, fs_info->generation);
3878 was_dirty = set_extent_buffer_dirty(buf);
3880 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3882 fs_info->dirty_metadata_batch);
3883 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3885 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3886 * but item data not updated.
3887 * So here we should only check item pointers, not item data.
3889 if (btrfs_header_level(buf) == 0 &&
3890 btrfs_check_leaf_relaxed(root, buf)) {
3891 btrfs_print_leaf(buf);
3897 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
3901 * looks as though older kernels can get into trouble with
3902 * this code, they end up stuck in balance_dirty_pages forever
3906 if (current->flags & PF_MEMALLOC)
3910 btrfs_balance_delayed_items(fs_info);
3912 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
3913 BTRFS_DIRTY_METADATA_THRESH,
3914 fs_info->dirty_metadata_batch);
3916 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
3920 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
3922 __btrfs_btree_balance_dirty(fs_info, 1);
3925 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
3927 __btrfs_btree_balance_dirty(fs_info, 0);
3930 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3932 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3933 struct btrfs_fs_info *fs_info = root->fs_info;
3935 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
3938 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
3940 struct btrfs_super_block *sb = fs_info->super_copy;
3941 u64 nodesize = btrfs_super_nodesize(sb);
3942 u64 sectorsize = btrfs_super_sectorsize(sb);
3945 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
3946 btrfs_err(fs_info, "no valid FS found");
3949 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
3950 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
3951 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
3954 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3955 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
3956 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3959 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3960 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
3961 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3964 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3965 btrfs_err(fs_info, "log_root level too big: %d >= %d",
3966 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3971 * Check sectorsize and nodesize first, other check will need it.
3972 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
3974 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
3975 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3976 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
3979 /* Only PAGE SIZE is supported yet */
3980 if (sectorsize != PAGE_SIZE) {
3982 "sectorsize %llu not supported yet, only support %lu",
3983 sectorsize, PAGE_SIZE);
3986 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
3987 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3988 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
3991 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
3992 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
3993 le32_to_cpu(sb->__unused_leafsize), nodesize);
3997 /* Root alignment check */
3998 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
3999 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4000 btrfs_super_root(sb));
4003 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4004 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4005 btrfs_super_chunk_root(sb));
4008 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4009 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4010 btrfs_super_log_root(sb));
4014 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
4016 "dev_item UUID does not match fsid: %pU != %pU",
4017 fs_info->fsid, sb->dev_item.fsid);
4022 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4025 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4026 btrfs_err(fs_info, "bytes_used is too small %llu",
4027 btrfs_super_bytes_used(sb));
4030 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4031 btrfs_err(fs_info, "invalid stripesize %u",
4032 btrfs_super_stripesize(sb));
4035 if (btrfs_super_num_devices(sb) > (1UL << 31))
4036 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4037 btrfs_super_num_devices(sb));
4038 if (btrfs_super_num_devices(sb) == 0) {
4039 btrfs_err(fs_info, "number of devices is 0");
4043 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4044 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4045 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4050 * Obvious sys_chunk_array corruptions, it must hold at least one key
4053 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4054 btrfs_err(fs_info, "system chunk array too big %u > %u",
4055 btrfs_super_sys_array_size(sb),
4056 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4059 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4060 + sizeof(struct btrfs_chunk)) {
4061 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4062 btrfs_super_sys_array_size(sb),
4063 sizeof(struct btrfs_disk_key)
4064 + sizeof(struct btrfs_chunk));
4069 * The generation is a global counter, we'll trust it more than the others
4070 * but it's still possible that it's the one that's wrong.
4072 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4074 "suspicious: generation < chunk_root_generation: %llu < %llu",
4075 btrfs_super_generation(sb),
4076 btrfs_super_chunk_root_generation(sb));
4077 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4078 && btrfs_super_cache_generation(sb) != (u64)-1)
4080 "suspicious: generation < cache_generation: %llu < %llu",
4081 btrfs_super_generation(sb),
4082 btrfs_super_cache_generation(sb));
4087 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4089 /* cleanup FS via transaction */
4090 btrfs_cleanup_transaction(fs_info);
4092 mutex_lock(&fs_info->cleaner_mutex);
4093 btrfs_run_delayed_iputs(fs_info);
4094 mutex_unlock(&fs_info->cleaner_mutex);
4096 down_write(&fs_info->cleanup_work_sem);
4097 up_write(&fs_info->cleanup_work_sem);
4100 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4102 struct btrfs_ordered_extent *ordered;
4104 spin_lock(&root->ordered_extent_lock);
4106 * This will just short circuit the ordered completion stuff which will
4107 * make sure the ordered extent gets properly cleaned up.
4109 list_for_each_entry(ordered, &root->ordered_extents,
4111 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4112 spin_unlock(&root->ordered_extent_lock);
4115 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4117 struct btrfs_root *root;
4118 struct list_head splice;
4120 INIT_LIST_HEAD(&splice);
4122 spin_lock(&fs_info->ordered_root_lock);
4123 list_splice_init(&fs_info->ordered_roots, &splice);
4124 while (!list_empty(&splice)) {
4125 root = list_first_entry(&splice, struct btrfs_root,
4127 list_move_tail(&root->ordered_root,
4128 &fs_info->ordered_roots);
4130 spin_unlock(&fs_info->ordered_root_lock);
4131 btrfs_destroy_ordered_extents(root);
4134 spin_lock(&fs_info->ordered_root_lock);
4136 spin_unlock(&fs_info->ordered_root_lock);
4139 * We need this here because if we've been flipped read-only we won't
4140 * get sync() from the umount, so we need to make sure any ordered
4141 * extents that haven't had their dirty pages IO start writeout yet
4142 * actually get run and error out properly.
4144 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4147 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4148 struct btrfs_fs_info *fs_info)
4150 struct rb_node *node;
4151 struct btrfs_delayed_ref_root *delayed_refs;
4152 struct btrfs_delayed_ref_node *ref;
4155 delayed_refs = &trans->delayed_refs;
4157 spin_lock(&delayed_refs->lock);
4158 if (atomic_read(&delayed_refs->num_entries) == 0) {
4159 spin_unlock(&delayed_refs->lock);
4160 btrfs_info(fs_info, "delayed_refs has NO entry");
4164 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4165 struct btrfs_delayed_ref_head *head;
4166 struct btrfs_delayed_ref_node *tmp;
4167 bool pin_bytes = false;
4169 head = rb_entry(node, struct btrfs_delayed_ref_head,
4171 if (!mutex_trylock(&head->mutex)) {
4172 refcount_inc(&head->node.refs);
4173 spin_unlock(&delayed_refs->lock);
4175 mutex_lock(&head->mutex);
4176 mutex_unlock(&head->mutex);
4177 btrfs_put_delayed_ref(&head->node);
4178 spin_lock(&delayed_refs->lock);
4181 spin_lock(&head->lock);
4182 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4185 list_del(&ref->list);
4186 if (!list_empty(&ref->add_list))
4187 list_del(&ref->add_list);
4188 atomic_dec(&delayed_refs->num_entries);
4189 btrfs_put_delayed_ref(ref);
4191 if (head->must_insert_reserved)
4193 btrfs_free_delayed_extent_op(head->extent_op);
4194 delayed_refs->num_heads--;
4195 if (head->processing == 0)
4196 delayed_refs->num_heads_ready--;
4197 atomic_dec(&delayed_refs->num_entries);
4198 head->node.in_tree = 0;
4199 rb_erase(&head->href_node, &delayed_refs->href_root);
4200 spin_unlock(&head->lock);
4201 spin_unlock(&delayed_refs->lock);
4202 mutex_unlock(&head->mutex);
4205 btrfs_pin_extent(fs_info, head->node.bytenr,
4206 head->node.num_bytes, 1);
4207 btrfs_put_delayed_ref(&head->node);
4209 spin_lock(&delayed_refs->lock);
4212 spin_unlock(&delayed_refs->lock);
4217 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4219 struct btrfs_inode *btrfs_inode;
4220 struct list_head splice;
4222 INIT_LIST_HEAD(&splice);
4224 spin_lock(&root->delalloc_lock);
4225 list_splice_init(&root->delalloc_inodes, &splice);
4227 while (!list_empty(&splice)) {
4228 struct inode *inode = NULL;
4229 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4231 __btrfs_del_delalloc_inode(root, btrfs_inode);
4232 spin_unlock(&root->delalloc_lock);
4235 * Make sure we get a live inode and that it'll not disappear
4238 inode = igrab(&btrfs_inode->vfs_inode);
4240 invalidate_inode_pages2(inode->i_mapping);
4243 spin_lock(&root->delalloc_lock);
4245 spin_unlock(&root->delalloc_lock);
4248 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4250 struct btrfs_root *root;
4251 struct list_head splice;
4253 INIT_LIST_HEAD(&splice);
4255 spin_lock(&fs_info->delalloc_root_lock);
4256 list_splice_init(&fs_info->delalloc_roots, &splice);
4257 while (!list_empty(&splice)) {
4258 root = list_first_entry(&splice, struct btrfs_root,
4260 root = btrfs_grab_fs_root(root);
4262 spin_unlock(&fs_info->delalloc_root_lock);
4264 btrfs_destroy_delalloc_inodes(root);
4265 btrfs_put_fs_root(root);
4267 spin_lock(&fs_info->delalloc_root_lock);
4269 spin_unlock(&fs_info->delalloc_root_lock);
4272 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4273 struct extent_io_tree *dirty_pages,
4277 struct extent_buffer *eb;
4282 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4287 clear_extent_bits(dirty_pages, start, end, mark);
4288 while (start <= end) {
4289 eb = find_extent_buffer(fs_info, start);
4290 start += fs_info->nodesize;
4293 wait_on_extent_buffer_writeback(eb);
4295 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4297 clear_extent_buffer_dirty(eb);
4298 free_extent_buffer_stale(eb);
4305 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4306 struct extent_io_tree *pinned_extents)
4308 struct extent_io_tree *unpin;
4314 unpin = pinned_extents;
4318 * The btrfs_finish_extent_commit() may get the same range as
4319 * ours between find_first_extent_bit and clear_extent_dirty.
4320 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4321 * the same extent range.
4323 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4324 ret = find_first_extent_bit(unpin, 0, &start, &end,
4325 EXTENT_DIRTY, NULL);
4327 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4331 clear_extent_dirty(unpin, start, end);
4332 btrfs_error_unpin_extent_range(fs_info, start, end);
4333 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4338 if (unpin == &fs_info->freed_extents[0])
4339 unpin = &fs_info->freed_extents[1];
4341 unpin = &fs_info->freed_extents[0];
4349 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4351 struct inode *inode;
4353 inode = cache->io_ctl.inode;
4355 invalidate_inode_pages2(inode->i_mapping);
4356 BTRFS_I(inode)->generation = 0;
4357 cache->io_ctl.inode = NULL;
4360 ASSERT(cache->io_ctl.pages == NULL);
4361 btrfs_put_block_group(cache);
4364 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4365 struct btrfs_fs_info *fs_info)
4367 struct btrfs_block_group_cache *cache;
4369 spin_lock(&cur_trans->dirty_bgs_lock);
4370 while (!list_empty(&cur_trans->dirty_bgs)) {
4371 cache = list_first_entry(&cur_trans->dirty_bgs,
4372 struct btrfs_block_group_cache,
4375 btrfs_err(fs_info, "orphan block group dirty_bgs list");
4376 spin_unlock(&cur_trans->dirty_bgs_lock);
4380 if (!list_empty(&cache->io_list)) {
4381 spin_unlock(&cur_trans->dirty_bgs_lock);
4382 list_del_init(&cache->io_list);
4383 btrfs_cleanup_bg_io(cache);
4384 spin_lock(&cur_trans->dirty_bgs_lock);
4387 list_del_init(&cache->dirty_list);
4388 spin_lock(&cache->lock);
4389 cache->disk_cache_state = BTRFS_DC_ERROR;
4390 spin_unlock(&cache->lock);
4392 spin_unlock(&cur_trans->dirty_bgs_lock);
4393 btrfs_put_block_group(cache);
4394 spin_lock(&cur_trans->dirty_bgs_lock);
4396 spin_unlock(&cur_trans->dirty_bgs_lock);
4398 while (!list_empty(&cur_trans->io_bgs)) {
4399 cache = list_first_entry(&cur_trans->io_bgs,
4400 struct btrfs_block_group_cache,
4403 btrfs_err(fs_info, "orphan block group on io_bgs list");
4407 list_del_init(&cache->io_list);
4408 spin_lock(&cache->lock);
4409 cache->disk_cache_state = BTRFS_DC_ERROR;
4410 spin_unlock(&cache->lock);
4411 btrfs_cleanup_bg_io(cache);
4415 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4416 struct btrfs_fs_info *fs_info)
4418 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4419 ASSERT(list_empty(&cur_trans->dirty_bgs));
4420 ASSERT(list_empty(&cur_trans->io_bgs));
4422 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4424 cur_trans->state = TRANS_STATE_COMMIT_START;
4425 wake_up(&fs_info->transaction_blocked_wait);
4427 cur_trans->state = TRANS_STATE_UNBLOCKED;
4428 wake_up(&fs_info->transaction_wait);
4430 btrfs_destroy_delayed_inodes(fs_info);
4432 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4434 btrfs_destroy_pinned_extent(fs_info,
4435 fs_info->pinned_extents);
4437 cur_trans->state =TRANS_STATE_COMPLETED;
4438 wake_up(&cur_trans->commit_wait);
4441 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4443 struct btrfs_transaction *t;
4445 mutex_lock(&fs_info->transaction_kthread_mutex);
4447 spin_lock(&fs_info->trans_lock);
4448 while (!list_empty(&fs_info->trans_list)) {
4449 t = list_first_entry(&fs_info->trans_list,
4450 struct btrfs_transaction, list);
4451 if (t->state >= TRANS_STATE_COMMIT_START) {
4452 refcount_inc(&t->use_count);
4453 spin_unlock(&fs_info->trans_lock);
4454 btrfs_wait_for_commit(fs_info, t->transid);
4455 btrfs_put_transaction(t);
4456 spin_lock(&fs_info->trans_lock);
4459 if (t == fs_info->running_transaction) {
4460 t->state = TRANS_STATE_COMMIT_DOING;
4461 spin_unlock(&fs_info->trans_lock);
4463 * We wait for 0 num_writers since we don't hold a trans
4464 * handle open currently for this transaction.
4466 wait_event(t->writer_wait,
4467 atomic_read(&t->num_writers) == 0);
4469 spin_unlock(&fs_info->trans_lock);
4471 btrfs_cleanup_one_transaction(t, fs_info);
4473 spin_lock(&fs_info->trans_lock);
4474 if (t == fs_info->running_transaction)
4475 fs_info->running_transaction = NULL;
4476 list_del_init(&t->list);
4477 spin_unlock(&fs_info->trans_lock);
4479 btrfs_put_transaction(t);
4480 trace_btrfs_transaction_commit(fs_info->tree_root);
4481 spin_lock(&fs_info->trans_lock);
4483 spin_unlock(&fs_info->trans_lock);
4484 btrfs_destroy_all_ordered_extents(fs_info);
4485 btrfs_destroy_delayed_inodes(fs_info);
4486 btrfs_assert_delayed_root_empty(fs_info);
4487 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4488 btrfs_destroy_all_delalloc_inodes(fs_info);
4489 mutex_unlock(&fs_info->transaction_kthread_mutex);
4494 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4496 struct inode *inode = private_data;
4497 return btrfs_sb(inode->i_sb);
4500 static const struct extent_io_ops btree_extent_io_ops = {
4501 /* mandatory callbacks */
4502 .submit_bio_hook = btree_submit_bio_hook,
4503 .readpage_end_io_hook = btree_readpage_end_io_hook,
4504 /* note we're sharing with inode.c for the merge bio hook */
4505 .merge_bio_hook = btrfs_merge_bio_hook,
4506 .readpage_io_failed_hook = btree_io_failed_hook,
4507 .set_range_writeback = btrfs_set_range_writeback,
4508 .tree_fs_info = btree_fs_info,
4510 /* optional callbacks */