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 root->log_transid = 0;
1204 root->log_transid_committed = -1;
1205 root->last_log_commit = 0;
1207 extent_io_tree_init(&root->dirty_log_pages, NULL);
1209 memset(&root->root_key, 0, sizeof(root->root_key));
1210 memset(&root->root_item, 0, sizeof(root->root_item));
1211 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1213 root->defrag_trans_start = fs_info->generation;
1215 root->defrag_trans_start = 0;
1216 root->root_key.objectid = objectid;
1219 spin_lock_init(&root->root_item_lock);
1222 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1225 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1227 root->fs_info = fs_info;
1231 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1232 /* Should only be used by the testing infrastructure */
1233 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1235 struct btrfs_root *root;
1238 return ERR_PTR(-EINVAL);
1240 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1242 return ERR_PTR(-ENOMEM);
1244 /* We don't use the stripesize in selftest, set it as sectorsize */
1245 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1246 root->alloc_bytenr = 0;
1252 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1253 struct btrfs_fs_info *fs_info,
1256 struct extent_buffer *leaf;
1257 struct btrfs_root *tree_root = fs_info->tree_root;
1258 struct btrfs_root *root;
1259 struct btrfs_key key;
1263 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1265 return ERR_PTR(-ENOMEM);
1267 __setup_root(root, fs_info, objectid);
1268 root->root_key.objectid = objectid;
1269 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1270 root->root_key.offset = 0;
1272 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1274 ret = PTR_ERR(leaf);
1279 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1280 btrfs_set_header_bytenr(leaf, leaf->start);
1281 btrfs_set_header_generation(leaf, trans->transid);
1282 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1283 btrfs_set_header_owner(leaf, objectid);
1286 write_extent_buffer_fsid(leaf, fs_info->fsid);
1287 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1288 btrfs_mark_buffer_dirty(leaf);
1290 root->commit_root = btrfs_root_node(root);
1291 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1293 root->root_item.flags = 0;
1294 root->root_item.byte_limit = 0;
1295 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1296 btrfs_set_root_generation(&root->root_item, trans->transid);
1297 btrfs_set_root_level(&root->root_item, 0);
1298 btrfs_set_root_refs(&root->root_item, 1);
1299 btrfs_set_root_used(&root->root_item, leaf->len);
1300 btrfs_set_root_last_snapshot(&root->root_item, 0);
1301 btrfs_set_root_dirid(&root->root_item, 0);
1303 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1304 root->root_item.drop_level = 0;
1306 key.objectid = objectid;
1307 key.type = BTRFS_ROOT_ITEM_KEY;
1309 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1313 btrfs_tree_unlock(leaf);
1319 btrfs_tree_unlock(leaf);
1320 free_extent_buffer(root->commit_root);
1321 free_extent_buffer(leaf);
1325 return ERR_PTR(ret);
1328 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1329 struct btrfs_fs_info *fs_info)
1331 struct btrfs_root *root;
1332 struct extent_buffer *leaf;
1334 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1336 return ERR_PTR(-ENOMEM);
1338 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1340 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1341 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1342 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1345 * DON'T set REF_COWS for log trees
1347 * log trees do not get reference counted because they go away
1348 * before a real commit is actually done. They do store pointers
1349 * to file data extents, and those reference counts still get
1350 * updated (along with back refs to the log tree).
1353 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1357 return ERR_CAST(leaf);
1360 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1361 btrfs_set_header_bytenr(leaf, leaf->start);
1362 btrfs_set_header_generation(leaf, trans->transid);
1363 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1364 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1367 write_extent_buffer_fsid(root->node, fs_info->fsid);
1368 btrfs_mark_buffer_dirty(root->node);
1369 btrfs_tree_unlock(root->node);
1373 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1374 struct btrfs_fs_info *fs_info)
1376 struct btrfs_root *log_root;
1378 log_root = alloc_log_tree(trans, fs_info);
1379 if (IS_ERR(log_root))
1380 return PTR_ERR(log_root);
1381 WARN_ON(fs_info->log_root_tree);
1382 fs_info->log_root_tree = log_root;
1386 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root)
1389 struct btrfs_fs_info *fs_info = root->fs_info;
1390 struct btrfs_root *log_root;
1391 struct btrfs_inode_item *inode_item;
1393 log_root = alloc_log_tree(trans, fs_info);
1394 if (IS_ERR(log_root))
1395 return PTR_ERR(log_root);
1397 log_root->last_trans = trans->transid;
1398 log_root->root_key.offset = root->root_key.objectid;
1400 inode_item = &log_root->root_item.inode;
1401 btrfs_set_stack_inode_generation(inode_item, 1);
1402 btrfs_set_stack_inode_size(inode_item, 3);
1403 btrfs_set_stack_inode_nlink(inode_item, 1);
1404 btrfs_set_stack_inode_nbytes(inode_item,
1406 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1408 btrfs_set_root_node(&log_root->root_item, log_root->node);
1410 WARN_ON(root->log_root);
1411 root->log_root = log_root;
1412 root->log_transid = 0;
1413 root->log_transid_committed = -1;
1414 root->last_log_commit = 0;
1418 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1419 struct btrfs_key *key)
1421 struct btrfs_root *root;
1422 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1423 struct btrfs_path *path;
1427 path = btrfs_alloc_path();
1429 return ERR_PTR(-ENOMEM);
1431 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1437 __setup_root(root, fs_info, key->objectid);
1439 ret = btrfs_find_root(tree_root, key, path,
1440 &root->root_item, &root->root_key);
1447 generation = btrfs_root_generation(&root->root_item);
1448 root->node = read_tree_block(fs_info,
1449 btrfs_root_bytenr(&root->root_item),
1451 if (IS_ERR(root->node)) {
1452 ret = PTR_ERR(root->node);
1454 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1456 free_extent_buffer(root->node);
1459 root->commit_root = btrfs_root_node(root);
1461 btrfs_free_path(path);
1467 root = ERR_PTR(ret);
1471 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1472 struct btrfs_key *location)
1474 struct btrfs_root *root;
1476 root = btrfs_read_tree_root(tree_root, location);
1480 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1481 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1482 btrfs_check_and_init_root_item(&root->root_item);
1488 int btrfs_init_fs_root(struct btrfs_root *root)
1491 struct btrfs_subvolume_writers *writers;
1493 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1494 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1496 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1501 writers = btrfs_alloc_subvolume_writers();
1502 if (IS_ERR(writers)) {
1503 ret = PTR_ERR(writers);
1506 root->subv_writers = writers;
1508 btrfs_init_free_ino_ctl(root);
1509 spin_lock_init(&root->ino_cache_lock);
1510 init_waitqueue_head(&root->ino_cache_wait);
1512 ret = get_anon_bdev(&root->anon_dev);
1516 mutex_lock(&root->objectid_mutex);
1517 ret = btrfs_find_highest_objectid(root,
1518 &root->highest_objectid);
1520 mutex_unlock(&root->objectid_mutex);
1524 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1526 mutex_unlock(&root->objectid_mutex);
1530 /* the caller is responsible to call free_fs_root */
1534 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1537 struct btrfs_root *root;
1539 spin_lock(&fs_info->fs_roots_radix_lock);
1540 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1541 (unsigned long)root_id);
1542 spin_unlock(&fs_info->fs_roots_radix_lock);
1546 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1547 struct btrfs_root *root)
1551 ret = radix_tree_preload(GFP_NOFS);
1555 spin_lock(&fs_info->fs_roots_radix_lock);
1556 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1557 (unsigned long)root->root_key.objectid,
1560 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1561 spin_unlock(&fs_info->fs_roots_radix_lock);
1562 radix_tree_preload_end();
1567 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1568 struct btrfs_key *location,
1571 struct btrfs_root *root;
1572 struct btrfs_path *path;
1573 struct btrfs_key key;
1576 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1577 return fs_info->tree_root;
1578 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1579 return fs_info->extent_root;
1580 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1581 return fs_info->chunk_root;
1582 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1583 return fs_info->dev_root;
1584 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1585 return fs_info->csum_root;
1586 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1587 return fs_info->quota_root ? fs_info->quota_root :
1589 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1590 return fs_info->uuid_root ? fs_info->uuid_root :
1592 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1593 return fs_info->free_space_root ? fs_info->free_space_root :
1596 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1598 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1599 return ERR_PTR(-ENOENT);
1603 root = btrfs_read_fs_root(fs_info->tree_root, location);
1607 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1612 ret = btrfs_init_fs_root(root);
1616 path = btrfs_alloc_path();
1621 key.objectid = BTRFS_ORPHAN_OBJECTID;
1622 key.type = BTRFS_ORPHAN_ITEM_KEY;
1623 key.offset = location->objectid;
1625 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1626 btrfs_free_path(path);
1630 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1632 ret = btrfs_insert_fs_root(fs_info, root);
1634 if (ret == -EEXIST) {
1643 return ERR_PTR(ret);
1646 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1648 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1650 struct btrfs_device *device;
1651 struct backing_dev_info *bdi;
1654 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1657 bdi = device->bdev->bd_bdi;
1658 if (bdi_congested(bdi, bdi_bits)) {
1668 * called by the kthread helper functions to finally call the bio end_io
1669 * functions. This is where read checksum verification actually happens
1671 static void end_workqueue_fn(struct btrfs_work *work)
1674 struct btrfs_end_io_wq *end_io_wq;
1676 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1677 bio = end_io_wq->bio;
1679 bio->bi_status = end_io_wq->status;
1680 bio->bi_private = end_io_wq->private;
1681 bio->bi_end_io = end_io_wq->end_io;
1682 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1686 static int cleaner_kthread(void *arg)
1688 struct btrfs_root *root = arg;
1689 struct btrfs_fs_info *fs_info = root->fs_info;
1691 struct btrfs_trans_handle *trans;
1696 /* Make the cleaner go to sleep early. */
1697 if (btrfs_need_cleaner_sleep(fs_info))
1701 * Do not do anything if we might cause open_ctree() to block
1702 * before we have finished mounting the filesystem.
1704 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1707 if (!mutex_trylock(&fs_info->cleaner_mutex))
1711 * Avoid the problem that we change the status of the fs
1712 * during the above check and trylock.
1714 if (btrfs_need_cleaner_sleep(fs_info)) {
1715 mutex_unlock(&fs_info->cleaner_mutex);
1719 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1720 btrfs_run_delayed_iputs(fs_info);
1721 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1723 again = btrfs_clean_one_deleted_snapshot(root);
1724 mutex_unlock(&fs_info->cleaner_mutex);
1727 * The defragger has dealt with the R/O remount and umount,
1728 * needn't do anything special here.
1730 btrfs_run_defrag_inodes(fs_info);
1733 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1734 * with relocation (btrfs_relocate_chunk) and relocation
1735 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1736 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1737 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1738 * unused block groups.
1740 btrfs_delete_unused_bgs(fs_info);
1743 set_current_state(TASK_INTERRUPTIBLE);
1744 if (!kthread_should_stop())
1746 __set_current_state(TASK_RUNNING);
1748 } while (!kthread_should_stop());
1751 * Transaction kthread is stopped before us and wakes us up.
1752 * However we might have started a new transaction and COWed some
1753 * tree blocks when deleting unused block groups for example. So
1754 * make sure we commit the transaction we started to have a clean
1755 * shutdown when evicting the btree inode - if it has dirty pages
1756 * when we do the final iput() on it, eviction will trigger a
1757 * writeback for it which will fail with null pointer dereferences
1758 * since work queues and other resources were already released and
1759 * destroyed by the time the iput/eviction/writeback is made.
1761 trans = btrfs_attach_transaction(root);
1762 if (IS_ERR(trans)) {
1763 if (PTR_ERR(trans) != -ENOENT)
1765 "cleaner transaction attach returned %ld",
1770 ret = btrfs_commit_transaction(trans);
1773 "cleaner open transaction commit returned %d",
1780 static int transaction_kthread(void *arg)
1782 struct btrfs_root *root = arg;
1783 struct btrfs_fs_info *fs_info = root->fs_info;
1784 struct btrfs_trans_handle *trans;
1785 struct btrfs_transaction *cur;
1788 unsigned long delay;
1792 cannot_commit = false;
1793 delay = HZ * fs_info->commit_interval;
1794 mutex_lock(&fs_info->transaction_kthread_mutex);
1796 spin_lock(&fs_info->trans_lock);
1797 cur = fs_info->running_transaction;
1799 spin_unlock(&fs_info->trans_lock);
1803 now = get_seconds();
1804 if (cur->state < TRANS_STATE_BLOCKED &&
1805 (now < cur->start_time ||
1806 now - cur->start_time < fs_info->commit_interval)) {
1807 spin_unlock(&fs_info->trans_lock);
1811 transid = cur->transid;
1812 spin_unlock(&fs_info->trans_lock);
1814 /* If the file system is aborted, this will always fail. */
1815 trans = btrfs_attach_transaction(root);
1816 if (IS_ERR(trans)) {
1817 if (PTR_ERR(trans) != -ENOENT)
1818 cannot_commit = true;
1821 if (transid == trans->transid) {
1822 btrfs_commit_transaction(trans);
1824 btrfs_end_transaction(trans);
1827 wake_up_process(fs_info->cleaner_kthread);
1828 mutex_unlock(&fs_info->transaction_kthread_mutex);
1830 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1831 &fs_info->fs_state)))
1832 btrfs_cleanup_transaction(fs_info);
1833 set_current_state(TASK_INTERRUPTIBLE);
1834 if (!kthread_should_stop() &&
1835 (!btrfs_transaction_blocked(fs_info) ||
1837 schedule_timeout(delay);
1838 __set_current_state(TASK_RUNNING);
1839 } while (!kthread_should_stop());
1844 * this will find the highest generation in the array of
1845 * root backups. The index of the highest array is returned,
1846 * or -1 if we can't find anything.
1848 * We check to make sure the array is valid by comparing the
1849 * generation of the latest root in the array with the generation
1850 * in the super block. If they don't match we pitch it.
1852 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1855 int newest_index = -1;
1856 struct btrfs_root_backup *root_backup;
1859 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1860 root_backup = info->super_copy->super_roots + i;
1861 cur = btrfs_backup_tree_root_gen(root_backup);
1862 if (cur == newest_gen)
1866 /* check to see if we actually wrapped around */
1867 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1868 root_backup = info->super_copy->super_roots;
1869 cur = btrfs_backup_tree_root_gen(root_backup);
1870 if (cur == newest_gen)
1873 return newest_index;
1878 * find the oldest backup so we know where to store new entries
1879 * in the backup array. This will set the backup_root_index
1880 * field in the fs_info struct
1882 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1885 int newest_index = -1;
1887 newest_index = find_newest_super_backup(info, newest_gen);
1888 /* if there was garbage in there, just move along */
1889 if (newest_index == -1) {
1890 info->backup_root_index = 0;
1892 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1897 * copy all the root pointers into the super backup array.
1898 * this will bump the backup pointer by one when it is
1901 static void backup_super_roots(struct btrfs_fs_info *info)
1904 struct btrfs_root_backup *root_backup;
1907 next_backup = info->backup_root_index;
1908 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1909 BTRFS_NUM_BACKUP_ROOTS;
1912 * just overwrite the last backup if we're at the same generation
1913 * this happens only at umount
1915 root_backup = info->super_for_commit->super_roots + last_backup;
1916 if (btrfs_backup_tree_root_gen(root_backup) ==
1917 btrfs_header_generation(info->tree_root->node))
1918 next_backup = last_backup;
1920 root_backup = info->super_for_commit->super_roots + next_backup;
1923 * make sure all of our padding and empty slots get zero filled
1924 * regardless of which ones we use today
1926 memset(root_backup, 0, sizeof(*root_backup));
1928 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1930 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1931 btrfs_set_backup_tree_root_gen(root_backup,
1932 btrfs_header_generation(info->tree_root->node));
1934 btrfs_set_backup_tree_root_level(root_backup,
1935 btrfs_header_level(info->tree_root->node));
1937 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1938 btrfs_set_backup_chunk_root_gen(root_backup,
1939 btrfs_header_generation(info->chunk_root->node));
1940 btrfs_set_backup_chunk_root_level(root_backup,
1941 btrfs_header_level(info->chunk_root->node));
1943 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1944 btrfs_set_backup_extent_root_gen(root_backup,
1945 btrfs_header_generation(info->extent_root->node));
1946 btrfs_set_backup_extent_root_level(root_backup,
1947 btrfs_header_level(info->extent_root->node));
1950 * we might commit during log recovery, which happens before we set
1951 * the fs_root. Make sure it is valid before we fill it in.
1953 if (info->fs_root && info->fs_root->node) {
1954 btrfs_set_backup_fs_root(root_backup,
1955 info->fs_root->node->start);
1956 btrfs_set_backup_fs_root_gen(root_backup,
1957 btrfs_header_generation(info->fs_root->node));
1958 btrfs_set_backup_fs_root_level(root_backup,
1959 btrfs_header_level(info->fs_root->node));
1962 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1963 btrfs_set_backup_dev_root_gen(root_backup,
1964 btrfs_header_generation(info->dev_root->node));
1965 btrfs_set_backup_dev_root_level(root_backup,
1966 btrfs_header_level(info->dev_root->node));
1968 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1969 btrfs_set_backup_csum_root_gen(root_backup,
1970 btrfs_header_generation(info->csum_root->node));
1971 btrfs_set_backup_csum_root_level(root_backup,
1972 btrfs_header_level(info->csum_root->node));
1974 btrfs_set_backup_total_bytes(root_backup,
1975 btrfs_super_total_bytes(info->super_copy));
1976 btrfs_set_backup_bytes_used(root_backup,
1977 btrfs_super_bytes_used(info->super_copy));
1978 btrfs_set_backup_num_devices(root_backup,
1979 btrfs_super_num_devices(info->super_copy));
1982 * if we don't copy this out to the super_copy, it won't get remembered
1983 * for the next commit
1985 memcpy(&info->super_copy->super_roots,
1986 &info->super_for_commit->super_roots,
1987 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1991 * this copies info out of the root backup array and back into
1992 * the in-memory super block. It is meant to help iterate through
1993 * the array, so you send it the number of backups you've already
1994 * tried and the last backup index you used.
1996 * this returns -1 when it has tried all the backups
1998 static noinline int next_root_backup(struct btrfs_fs_info *info,
1999 struct btrfs_super_block *super,
2000 int *num_backups_tried, int *backup_index)
2002 struct btrfs_root_backup *root_backup;
2003 int newest = *backup_index;
2005 if (*num_backups_tried == 0) {
2006 u64 gen = btrfs_super_generation(super);
2008 newest = find_newest_super_backup(info, gen);
2012 *backup_index = newest;
2013 *num_backups_tried = 1;
2014 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2015 /* we've tried all the backups, all done */
2018 /* jump to the next oldest backup */
2019 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2020 BTRFS_NUM_BACKUP_ROOTS;
2021 *backup_index = newest;
2022 *num_backups_tried += 1;
2024 root_backup = super->super_roots + newest;
2026 btrfs_set_super_generation(super,
2027 btrfs_backup_tree_root_gen(root_backup));
2028 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2029 btrfs_set_super_root_level(super,
2030 btrfs_backup_tree_root_level(root_backup));
2031 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2034 * fixme: the total bytes and num_devices need to match or we should
2037 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2038 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2042 /* helper to cleanup workers */
2043 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2045 btrfs_destroy_workqueue(fs_info->fixup_workers);
2046 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2047 btrfs_destroy_workqueue(fs_info->workers);
2048 btrfs_destroy_workqueue(fs_info->endio_workers);
2049 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2050 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2051 btrfs_destroy_workqueue(fs_info->rmw_workers);
2052 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2053 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2054 btrfs_destroy_workqueue(fs_info->submit_workers);
2055 btrfs_destroy_workqueue(fs_info->delayed_workers);
2056 btrfs_destroy_workqueue(fs_info->caching_workers);
2057 btrfs_destroy_workqueue(fs_info->readahead_workers);
2058 btrfs_destroy_workqueue(fs_info->flush_workers);
2059 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2060 btrfs_destroy_workqueue(fs_info->extent_workers);
2062 * Now that all other work queues are destroyed, we can safely destroy
2063 * the queues used for metadata I/O, since tasks from those other work
2064 * queues can do metadata I/O operations.
2066 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2067 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2070 static void free_root_extent_buffers(struct btrfs_root *root)
2073 free_extent_buffer(root->node);
2074 free_extent_buffer(root->commit_root);
2076 root->commit_root = NULL;
2080 /* helper to cleanup tree roots */
2081 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2083 free_root_extent_buffers(info->tree_root);
2085 free_root_extent_buffers(info->dev_root);
2086 free_root_extent_buffers(info->extent_root);
2087 free_root_extent_buffers(info->csum_root);
2088 free_root_extent_buffers(info->quota_root);
2089 free_root_extent_buffers(info->uuid_root);
2091 free_root_extent_buffers(info->chunk_root);
2092 free_root_extent_buffers(info->free_space_root);
2095 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2098 struct btrfs_root *gang[8];
2101 while (!list_empty(&fs_info->dead_roots)) {
2102 gang[0] = list_entry(fs_info->dead_roots.next,
2103 struct btrfs_root, root_list);
2104 list_del(&gang[0]->root_list);
2106 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2107 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2109 free_extent_buffer(gang[0]->node);
2110 free_extent_buffer(gang[0]->commit_root);
2111 btrfs_put_fs_root(gang[0]);
2116 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2121 for (i = 0; i < ret; i++)
2122 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2125 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2126 btrfs_free_log_root_tree(NULL, fs_info);
2127 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2131 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2133 mutex_init(&fs_info->scrub_lock);
2134 atomic_set(&fs_info->scrubs_running, 0);
2135 atomic_set(&fs_info->scrub_pause_req, 0);
2136 atomic_set(&fs_info->scrubs_paused, 0);
2137 atomic_set(&fs_info->scrub_cancel_req, 0);
2138 init_waitqueue_head(&fs_info->scrub_pause_wait);
2139 fs_info->scrub_workers_refcnt = 0;
2142 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2144 spin_lock_init(&fs_info->balance_lock);
2145 mutex_init(&fs_info->balance_mutex);
2146 atomic_set(&fs_info->balance_running, 0);
2147 atomic_set(&fs_info->balance_pause_req, 0);
2148 atomic_set(&fs_info->balance_cancel_req, 0);
2149 fs_info->balance_ctl = NULL;
2150 init_waitqueue_head(&fs_info->balance_wait_q);
2153 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2155 struct inode *inode = fs_info->btree_inode;
2157 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2158 set_nlink(inode, 1);
2160 * we set the i_size on the btree inode to the max possible int.
2161 * the real end of the address space is determined by all of
2162 * the devices in the system
2164 inode->i_size = OFFSET_MAX;
2165 inode->i_mapping->a_ops = &btree_aops;
2167 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2168 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2169 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2170 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2172 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2174 BTRFS_I(inode)->root = fs_info->tree_root;
2175 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2176 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2177 btrfs_insert_inode_hash(inode);
2180 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2182 fs_info->dev_replace.lock_owner = 0;
2183 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2184 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2185 rwlock_init(&fs_info->dev_replace.lock);
2186 atomic_set(&fs_info->dev_replace.read_locks, 0);
2187 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2188 init_waitqueue_head(&fs_info->replace_wait);
2189 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2192 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2194 spin_lock_init(&fs_info->qgroup_lock);
2195 mutex_init(&fs_info->qgroup_ioctl_lock);
2196 fs_info->qgroup_tree = RB_ROOT;
2197 fs_info->qgroup_op_tree = RB_ROOT;
2198 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2199 fs_info->qgroup_seq = 1;
2200 fs_info->qgroup_ulist = NULL;
2201 fs_info->qgroup_rescan_running = false;
2202 mutex_init(&fs_info->qgroup_rescan_lock);
2205 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2206 struct btrfs_fs_devices *fs_devices)
2208 int max_active = fs_info->thread_pool_size;
2209 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2212 btrfs_alloc_workqueue(fs_info, "worker",
2213 flags | WQ_HIGHPRI, max_active, 16);
2215 fs_info->delalloc_workers =
2216 btrfs_alloc_workqueue(fs_info, "delalloc",
2217 flags, max_active, 2);
2219 fs_info->flush_workers =
2220 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2221 flags, max_active, 0);
2223 fs_info->caching_workers =
2224 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2227 * a higher idle thresh on the submit workers makes it much more
2228 * likely that bios will be send down in a sane order to the
2231 fs_info->submit_workers =
2232 btrfs_alloc_workqueue(fs_info, "submit", flags,
2233 min_t(u64, fs_devices->num_devices,
2236 fs_info->fixup_workers =
2237 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2240 * endios are largely parallel and should have a very
2243 fs_info->endio_workers =
2244 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2245 fs_info->endio_meta_workers =
2246 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2248 fs_info->endio_meta_write_workers =
2249 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2251 fs_info->endio_raid56_workers =
2252 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2254 fs_info->endio_repair_workers =
2255 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2256 fs_info->rmw_workers =
2257 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2258 fs_info->endio_write_workers =
2259 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2261 fs_info->endio_freespace_worker =
2262 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2264 fs_info->delayed_workers =
2265 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2267 fs_info->readahead_workers =
2268 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2270 fs_info->qgroup_rescan_workers =
2271 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2272 fs_info->extent_workers =
2273 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2274 min_t(u64, fs_devices->num_devices,
2277 if (!(fs_info->workers && fs_info->delalloc_workers &&
2278 fs_info->submit_workers && fs_info->flush_workers &&
2279 fs_info->endio_workers && fs_info->endio_meta_workers &&
2280 fs_info->endio_meta_write_workers &&
2281 fs_info->endio_repair_workers &&
2282 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2283 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2284 fs_info->caching_workers && fs_info->readahead_workers &&
2285 fs_info->fixup_workers && fs_info->delayed_workers &&
2286 fs_info->extent_workers &&
2287 fs_info->qgroup_rescan_workers)) {
2294 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2295 struct btrfs_fs_devices *fs_devices)
2298 struct btrfs_root *log_tree_root;
2299 struct btrfs_super_block *disk_super = fs_info->super_copy;
2300 u64 bytenr = btrfs_super_log_root(disk_super);
2302 if (fs_devices->rw_devices == 0) {
2303 btrfs_warn(fs_info, "log replay required on RO media");
2307 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2311 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2313 log_tree_root->node = read_tree_block(fs_info, bytenr,
2314 fs_info->generation + 1);
2315 if (IS_ERR(log_tree_root->node)) {
2316 btrfs_warn(fs_info, "failed to read log tree");
2317 ret = PTR_ERR(log_tree_root->node);
2318 kfree(log_tree_root);
2320 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2321 btrfs_err(fs_info, "failed to read log tree");
2322 free_extent_buffer(log_tree_root->node);
2323 kfree(log_tree_root);
2326 /* returns with log_tree_root freed on success */
2327 ret = btrfs_recover_log_trees(log_tree_root);
2329 btrfs_handle_fs_error(fs_info, ret,
2330 "Failed to recover log tree");
2331 free_extent_buffer(log_tree_root->node);
2332 kfree(log_tree_root);
2336 if (sb_rdonly(fs_info->sb)) {
2337 ret = btrfs_commit_super(fs_info);
2345 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2347 struct btrfs_root *tree_root = fs_info->tree_root;
2348 struct btrfs_root *root;
2349 struct btrfs_key location;
2352 BUG_ON(!fs_info->tree_root);
2354 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2355 location.type = BTRFS_ROOT_ITEM_KEY;
2356 location.offset = 0;
2358 root = btrfs_read_tree_root(tree_root, &location);
2360 return PTR_ERR(root);
2361 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2362 fs_info->extent_root = root;
2364 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2365 root = btrfs_read_tree_root(tree_root, &location);
2367 return PTR_ERR(root);
2368 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2369 fs_info->dev_root = root;
2370 btrfs_init_devices_late(fs_info);
2372 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2373 root = btrfs_read_tree_root(tree_root, &location);
2375 return PTR_ERR(root);
2376 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2377 fs_info->csum_root = root;
2379 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2380 root = btrfs_read_tree_root(tree_root, &location);
2381 if (!IS_ERR(root)) {
2382 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2383 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2384 fs_info->quota_root = root;
2387 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2388 root = btrfs_read_tree_root(tree_root, &location);
2390 ret = PTR_ERR(root);
2394 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2395 fs_info->uuid_root = root;
2398 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2399 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2400 root = btrfs_read_tree_root(tree_root, &location);
2402 return PTR_ERR(root);
2403 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2404 fs_info->free_space_root = root;
2410 int open_ctree(struct super_block *sb,
2411 struct btrfs_fs_devices *fs_devices,
2419 struct btrfs_key location;
2420 struct buffer_head *bh;
2421 struct btrfs_super_block *disk_super;
2422 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2423 struct btrfs_root *tree_root;
2424 struct btrfs_root *chunk_root;
2427 int num_backups_tried = 0;
2428 int backup_index = 0;
2430 int clear_free_space_tree = 0;
2432 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2433 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2434 if (!tree_root || !chunk_root) {
2439 ret = init_srcu_struct(&fs_info->subvol_srcu);
2445 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2450 fs_info->dirty_metadata_batch = PAGE_SIZE *
2451 (1 + ilog2(nr_cpu_ids));
2453 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2456 goto fail_dirty_metadata_bytes;
2459 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2462 goto fail_delalloc_bytes;
2465 fs_info->btree_inode = new_inode(sb);
2466 if (!fs_info->btree_inode) {
2468 goto fail_bio_counter;
2471 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2473 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2474 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2475 INIT_LIST_HEAD(&fs_info->trans_list);
2476 INIT_LIST_HEAD(&fs_info->dead_roots);
2477 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2478 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2479 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2480 spin_lock_init(&fs_info->delalloc_root_lock);
2481 spin_lock_init(&fs_info->trans_lock);
2482 spin_lock_init(&fs_info->fs_roots_radix_lock);
2483 spin_lock_init(&fs_info->delayed_iput_lock);
2484 spin_lock_init(&fs_info->defrag_inodes_lock);
2485 spin_lock_init(&fs_info->tree_mod_seq_lock);
2486 spin_lock_init(&fs_info->super_lock);
2487 spin_lock_init(&fs_info->qgroup_op_lock);
2488 spin_lock_init(&fs_info->buffer_lock);
2489 spin_lock_init(&fs_info->unused_bgs_lock);
2490 rwlock_init(&fs_info->tree_mod_log_lock);
2491 mutex_init(&fs_info->unused_bg_unpin_mutex);
2492 mutex_init(&fs_info->delete_unused_bgs_mutex);
2493 mutex_init(&fs_info->reloc_mutex);
2494 mutex_init(&fs_info->delalloc_root_mutex);
2495 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2496 seqlock_init(&fs_info->profiles_lock);
2498 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2499 INIT_LIST_HEAD(&fs_info->space_info);
2500 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2501 INIT_LIST_HEAD(&fs_info->unused_bgs);
2502 btrfs_mapping_init(&fs_info->mapping_tree);
2503 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2504 BTRFS_BLOCK_RSV_GLOBAL);
2505 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2506 BTRFS_BLOCK_RSV_DELALLOC);
2507 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2508 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2509 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2510 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2511 BTRFS_BLOCK_RSV_DELOPS);
2512 atomic_set(&fs_info->nr_async_submits, 0);
2513 atomic_set(&fs_info->async_delalloc_pages, 0);
2514 atomic_set(&fs_info->async_submit_draining, 0);
2515 atomic_set(&fs_info->nr_async_bios, 0);
2516 atomic_set(&fs_info->defrag_running, 0);
2517 atomic_set(&fs_info->qgroup_op_seq, 0);
2518 atomic_set(&fs_info->reada_works_cnt, 0);
2519 atomic64_set(&fs_info->tree_mod_seq, 0);
2521 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2522 fs_info->metadata_ratio = 0;
2523 fs_info->defrag_inodes = RB_ROOT;
2524 atomic64_set(&fs_info->free_chunk_space, 0);
2525 fs_info->tree_mod_log = RB_ROOT;
2526 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2527 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2528 /* readahead state */
2529 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2530 spin_lock_init(&fs_info->reada_lock);
2532 fs_info->thread_pool_size = min_t(unsigned long,
2533 num_online_cpus() + 2, 8);
2535 INIT_LIST_HEAD(&fs_info->ordered_roots);
2536 spin_lock_init(&fs_info->ordered_root_lock);
2537 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2539 if (!fs_info->delayed_root) {
2543 btrfs_init_delayed_root(fs_info->delayed_root);
2545 btrfs_init_scrub(fs_info);
2546 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2547 fs_info->check_integrity_print_mask = 0;
2549 btrfs_init_balance(fs_info);
2550 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2552 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2553 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2555 btrfs_init_btree_inode(fs_info);
2557 spin_lock_init(&fs_info->block_group_cache_lock);
2558 fs_info->block_group_cache_tree = RB_ROOT;
2559 fs_info->first_logical_byte = (u64)-1;
2561 extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2562 extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2563 fs_info->pinned_extents = &fs_info->freed_extents[0];
2564 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2566 mutex_init(&fs_info->ordered_operations_mutex);
2567 mutex_init(&fs_info->tree_log_mutex);
2568 mutex_init(&fs_info->chunk_mutex);
2569 mutex_init(&fs_info->transaction_kthread_mutex);
2570 mutex_init(&fs_info->cleaner_mutex);
2571 mutex_init(&fs_info->volume_mutex);
2572 mutex_init(&fs_info->ro_block_group_mutex);
2573 init_rwsem(&fs_info->commit_root_sem);
2574 init_rwsem(&fs_info->cleanup_work_sem);
2575 init_rwsem(&fs_info->subvol_sem);
2576 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2578 btrfs_init_dev_replace_locks(fs_info);
2579 btrfs_init_qgroup(fs_info);
2581 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2582 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2584 init_waitqueue_head(&fs_info->transaction_throttle);
2585 init_waitqueue_head(&fs_info->transaction_wait);
2586 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2587 init_waitqueue_head(&fs_info->async_submit_wait);
2589 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2591 /* Usable values until the real ones are cached from the superblock */
2592 fs_info->nodesize = 4096;
2593 fs_info->sectorsize = 4096;
2594 fs_info->stripesize = 4096;
2596 ret = btrfs_alloc_stripe_hash_table(fs_info);
2602 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2604 invalidate_bdev(fs_devices->latest_bdev);
2607 * Read super block and check the signature bytes only
2609 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2616 * We want to check superblock checksum, the type is stored inside.
2617 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2619 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2620 btrfs_err(fs_info, "superblock checksum mismatch");
2627 * super_copy is zeroed at allocation time and we never touch the
2628 * following bytes up to INFO_SIZE, the checksum is calculated from
2629 * the whole block of INFO_SIZE
2631 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2632 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2633 sizeof(*fs_info->super_for_commit));
2636 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2638 ret = btrfs_check_super_valid(fs_info);
2640 btrfs_err(fs_info, "superblock contains fatal errors");
2645 disk_super = fs_info->super_copy;
2646 if (!btrfs_super_root(disk_super))
2649 /* check FS state, whether FS is broken. */
2650 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2651 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2654 * run through our array of backup supers and setup
2655 * our ring pointer to the oldest one
2657 generation = btrfs_super_generation(disk_super);
2658 find_oldest_super_backup(fs_info, generation);
2661 * In the long term, we'll store the compression type in the super
2662 * block, and it'll be used for per file compression control.
2664 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2666 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2672 features = btrfs_super_incompat_flags(disk_super) &
2673 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2676 "cannot mount because of unsupported optional features (%llx)",
2682 features = btrfs_super_incompat_flags(disk_super);
2683 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2684 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2685 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2686 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2687 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2689 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2690 btrfs_info(fs_info, "has skinny extents");
2693 * flag our filesystem as having big metadata blocks if
2694 * they are bigger than the page size
2696 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2697 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2699 "flagging fs with big metadata feature");
2700 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2703 nodesize = btrfs_super_nodesize(disk_super);
2704 sectorsize = btrfs_super_sectorsize(disk_super);
2705 stripesize = sectorsize;
2706 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2707 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2709 /* Cache block sizes */
2710 fs_info->nodesize = nodesize;
2711 fs_info->sectorsize = sectorsize;
2712 fs_info->stripesize = stripesize;
2715 * mixed block groups end up with duplicate but slightly offset
2716 * extent buffers for the same range. It leads to corruptions
2718 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2719 (sectorsize != nodesize)) {
2721 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2722 nodesize, sectorsize);
2727 * Needn't use the lock because there is no other task which will
2730 btrfs_set_super_incompat_flags(disk_super, features);
2732 features = btrfs_super_compat_ro_flags(disk_super) &
2733 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2734 if (!sb_rdonly(sb) && features) {
2736 "cannot mount read-write because of unsupported optional features (%llx)",
2742 max_active = fs_info->thread_pool_size;
2744 ret = btrfs_init_workqueues(fs_info, fs_devices);
2747 goto fail_sb_buffer;
2750 sb->s_bdi->congested_fn = btrfs_congested_fn;
2751 sb->s_bdi->congested_data = fs_info;
2752 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2753 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
2754 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2755 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2757 sb->s_blocksize = sectorsize;
2758 sb->s_blocksize_bits = blksize_bits(sectorsize);
2760 mutex_lock(&fs_info->chunk_mutex);
2761 ret = btrfs_read_sys_array(fs_info);
2762 mutex_unlock(&fs_info->chunk_mutex);
2764 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2765 goto fail_sb_buffer;
2768 generation = btrfs_super_chunk_root_generation(disk_super);
2770 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2772 chunk_root->node = read_tree_block(fs_info,
2773 btrfs_super_chunk_root(disk_super),
2775 if (IS_ERR(chunk_root->node) ||
2776 !extent_buffer_uptodate(chunk_root->node)) {
2777 btrfs_err(fs_info, "failed to read chunk root");
2778 if (!IS_ERR(chunk_root->node))
2779 free_extent_buffer(chunk_root->node);
2780 chunk_root->node = NULL;
2781 goto fail_tree_roots;
2783 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2784 chunk_root->commit_root = btrfs_root_node(chunk_root);
2786 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2787 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2789 ret = btrfs_read_chunk_tree(fs_info);
2791 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2792 goto fail_tree_roots;
2796 * keep the device that is marked to be the target device for the
2797 * dev_replace procedure
2799 btrfs_close_extra_devices(fs_devices, 0);
2801 if (!fs_devices->latest_bdev) {
2802 btrfs_err(fs_info, "failed to read devices");
2803 goto fail_tree_roots;
2807 generation = btrfs_super_generation(disk_super);
2809 tree_root->node = read_tree_block(fs_info,
2810 btrfs_super_root(disk_super),
2812 if (IS_ERR(tree_root->node) ||
2813 !extent_buffer_uptodate(tree_root->node)) {
2814 btrfs_warn(fs_info, "failed to read tree root");
2815 if (!IS_ERR(tree_root->node))
2816 free_extent_buffer(tree_root->node);
2817 tree_root->node = NULL;
2818 goto recovery_tree_root;
2821 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2822 tree_root->commit_root = btrfs_root_node(tree_root);
2823 btrfs_set_root_refs(&tree_root->root_item, 1);
2825 mutex_lock(&tree_root->objectid_mutex);
2826 ret = btrfs_find_highest_objectid(tree_root,
2827 &tree_root->highest_objectid);
2829 mutex_unlock(&tree_root->objectid_mutex);
2830 goto recovery_tree_root;
2833 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2835 mutex_unlock(&tree_root->objectid_mutex);
2837 ret = btrfs_read_roots(fs_info);
2839 goto recovery_tree_root;
2841 fs_info->generation = generation;
2842 fs_info->last_trans_committed = generation;
2844 ret = btrfs_recover_balance(fs_info);
2846 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2847 goto fail_block_groups;
2850 ret = btrfs_init_dev_stats(fs_info);
2852 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2853 goto fail_block_groups;
2856 ret = btrfs_init_dev_replace(fs_info);
2858 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2859 goto fail_block_groups;
2862 btrfs_close_extra_devices(fs_devices, 1);
2864 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2866 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2868 goto fail_block_groups;
2871 ret = btrfs_sysfs_add_device(fs_devices);
2873 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2875 goto fail_fsdev_sysfs;
2878 ret = btrfs_sysfs_add_mounted(fs_info);
2880 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2881 goto fail_fsdev_sysfs;
2884 ret = btrfs_init_space_info(fs_info);
2886 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2890 ret = btrfs_read_block_groups(fs_info);
2892 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2896 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info)) {
2898 "writeable mount is not allowed due to too many missing devices");
2902 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2904 if (IS_ERR(fs_info->cleaner_kthread))
2907 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2909 "btrfs-transaction");
2910 if (IS_ERR(fs_info->transaction_kthread))
2913 if (!btrfs_test_opt(fs_info, NOSSD) &&
2914 !fs_info->fs_devices->rotating) {
2915 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2919 * Mount does not set all options immediately, we can do it now and do
2920 * not have to wait for transaction commit
2922 btrfs_apply_pending_changes(fs_info);
2924 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2925 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2926 ret = btrfsic_mount(fs_info, fs_devices,
2927 btrfs_test_opt(fs_info,
2928 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2930 fs_info->check_integrity_print_mask);
2933 "failed to initialize integrity check module: %d",
2937 ret = btrfs_read_qgroup_config(fs_info);
2939 goto fail_trans_kthread;
2941 /* do not make disk changes in broken FS or nologreplay is given */
2942 if (btrfs_super_log_root(disk_super) != 0 &&
2943 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2944 ret = btrfs_replay_log(fs_info, fs_devices);
2951 ret = btrfs_find_orphan_roots(fs_info);
2955 if (!sb_rdonly(sb)) {
2956 ret = btrfs_cleanup_fs_roots(fs_info);
2960 mutex_lock(&fs_info->cleaner_mutex);
2961 ret = btrfs_recover_relocation(tree_root);
2962 mutex_unlock(&fs_info->cleaner_mutex);
2964 btrfs_warn(fs_info, "failed to recover relocation: %d",
2971 location.objectid = BTRFS_FS_TREE_OBJECTID;
2972 location.type = BTRFS_ROOT_ITEM_KEY;
2973 location.offset = 0;
2975 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2976 if (IS_ERR(fs_info->fs_root)) {
2977 err = PTR_ERR(fs_info->fs_root);
2984 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2985 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2986 clear_free_space_tree = 1;
2987 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2988 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2989 btrfs_warn(fs_info, "free space tree is invalid");
2990 clear_free_space_tree = 1;
2993 if (clear_free_space_tree) {
2994 btrfs_info(fs_info, "clearing free space tree");
2995 ret = btrfs_clear_free_space_tree(fs_info);
2998 "failed to clear free space tree: %d", ret);
2999 close_ctree(fs_info);
3004 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3005 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3006 btrfs_info(fs_info, "creating free space tree");
3007 ret = btrfs_create_free_space_tree(fs_info);
3010 "failed to create free space tree: %d", ret);
3011 close_ctree(fs_info);
3016 down_read(&fs_info->cleanup_work_sem);
3017 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3018 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3019 up_read(&fs_info->cleanup_work_sem);
3020 close_ctree(fs_info);
3023 up_read(&fs_info->cleanup_work_sem);
3025 ret = btrfs_resume_balance_async(fs_info);
3027 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3028 close_ctree(fs_info);
3032 ret = btrfs_resume_dev_replace_async(fs_info);
3034 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3035 close_ctree(fs_info);
3039 btrfs_qgroup_rescan_resume(fs_info);
3041 if (!fs_info->uuid_root) {
3042 btrfs_info(fs_info, "creating UUID tree");
3043 ret = btrfs_create_uuid_tree(fs_info);
3046 "failed to create the UUID tree: %d", ret);
3047 close_ctree(fs_info);
3050 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3051 fs_info->generation !=
3052 btrfs_super_uuid_tree_generation(disk_super)) {
3053 btrfs_info(fs_info, "checking UUID tree");
3054 ret = btrfs_check_uuid_tree(fs_info);
3057 "failed to check the UUID tree: %d", ret);
3058 close_ctree(fs_info);
3062 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3064 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3067 * backuproot only affect mount behavior, and if open_ctree succeeded,
3068 * no need to keep the flag
3070 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3075 btrfs_free_qgroup_config(fs_info);
3077 kthread_stop(fs_info->transaction_kthread);
3078 btrfs_cleanup_transaction(fs_info);
3079 btrfs_free_fs_roots(fs_info);
3081 kthread_stop(fs_info->cleaner_kthread);
3084 * make sure we're done with the btree inode before we stop our
3087 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3090 btrfs_sysfs_remove_mounted(fs_info);
3093 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3096 btrfs_put_block_group_cache(fs_info);
3099 free_root_pointers(fs_info, 1);
3100 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3103 btrfs_stop_all_workers(fs_info);
3104 btrfs_free_block_groups(fs_info);
3107 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3109 iput(fs_info->btree_inode);
3111 percpu_counter_destroy(&fs_info->bio_counter);
3112 fail_delalloc_bytes:
3113 percpu_counter_destroy(&fs_info->delalloc_bytes);
3114 fail_dirty_metadata_bytes:
3115 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3117 cleanup_srcu_struct(&fs_info->subvol_srcu);
3119 btrfs_free_stripe_hash_table(fs_info);
3120 btrfs_close_devices(fs_info->fs_devices);
3124 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3125 goto fail_tree_roots;
3127 free_root_pointers(fs_info, 0);
3129 /* don't use the log in recovery mode, it won't be valid */
3130 btrfs_set_super_log_root(disk_super, 0);
3132 /* we can't trust the free space cache either */
3133 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3135 ret = next_root_backup(fs_info, fs_info->super_copy,
3136 &num_backups_tried, &backup_index);
3138 goto fail_block_groups;
3139 goto retry_root_backup;
3142 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3145 set_buffer_uptodate(bh);
3147 struct btrfs_device *device = (struct btrfs_device *)
3150 btrfs_warn_rl_in_rcu(device->fs_info,
3151 "lost page write due to IO error on %s",
3152 rcu_str_deref(device->name));
3153 /* note, we don't set_buffer_write_io_error because we have
3154 * our own ways of dealing with the IO errors
3156 clear_buffer_uptodate(bh);
3157 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3163 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3164 struct buffer_head **bh_ret)
3166 struct buffer_head *bh;
3167 struct btrfs_super_block *super;
3170 bytenr = btrfs_sb_offset(copy_num);
3171 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3174 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3176 * If we fail to read from the underlying devices, as of now
3177 * the best option we have is to mark it EIO.
3182 super = (struct btrfs_super_block *)bh->b_data;
3183 if (btrfs_super_bytenr(super) != bytenr ||
3184 btrfs_super_magic(super) != BTRFS_MAGIC) {
3194 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3196 struct buffer_head *bh;
3197 struct buffer_head *latest = NULL;
3198 struct btrfs_super_block *super;
3203 /* we would like to check all the supers, but that would make
3204 * a btrfs mount succeed after a mkfs from a different FS.
3205 * So, we need to add a special mount option to scan for
3206 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3208 for (i = 0; i < 1; i++) {
3209 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3213 super = (struct btrfs_super_block *)bh->b_data;
3215 if (!latest || btrfs_super_generation(super) > transid) {
3218 transid = btrfs_super_generation(super);
3225 return ERR_PTR(ret);
3231 * Write superblock @sb to the @device. Do not wait for completion, all the
3232 * buffer heads we write are pinned.
3234 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3235 * the expected device size at commit time. Note that max_mirrors must be
3236 * same for write and wait phases.
3238 * Return number of errors when buffer head is not found or submission fails.
3240 static int write_dev_supers(struct btrfs_device *device,
3241 struct btrfs_super_block *sb, int max_mirrors)
3243 struct buffer_head *bh;
3251 if (max_mirrors == 0)
3252 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3254 for (i = 0; i < max_mirrors; i++) {
3255 bytenr = btrfs_sb_offset(i);
3256 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3257 device->commit_total_bytes)
3260 btrfs_set_super_bytenr(sb, bytenr);
3263 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3264 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3265 btrfs_csum_final(crc, sb->csum);
3267 /* One reference for us, and we leave it for the caller */
3268 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3269 BTRFS_SUPER_INFO_SIZE);
3271 btrfs_err(device->fs_info,
3272 "couldn't get super buffer head for bytenr %llu",
3278 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3280 /* one reference for submit_bh */
3283 set_buffer_uptodate(bh);
3285 bh->b_end_io = btrfs_end_buffer_write_sync;
3286 bh->b_private = device;
3289 * we fua the first super. The others we allow
3292 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3293 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3294 op_flags |= REQ_FUA;
3295 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3299 return errors < i ? 0 : -1;
3303 * Wait for write completion of superblocks done by write_dev_supers,
3304 * @max_mirrors same for write and wait phases.
3306 * Return number of errors when buffer head is not found or not marked up to
3309 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3311 struct buffer_head *bh;
3316 if (max_mirrors == 0)
3317 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3319 for (i = 0; i < max_mirrors; i++) {
3320 bytenr = btrfs_sb_offset(i);
3321 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3322 device->commit_total_bytes)
3325 bh = __find_get_block(device->bdev,
3326 bytenr / BTRFS_BDEV_BLOCKSIZE,
3327 BTRFS_SUPER_INFO_SIZE);
3333 if (!buffer_uptodate(bh))
3336 /* drop our reference */
3339 /* drop the reference from the writing run */
3343 return errors < i ? 0 : -1;
3347 * endio for the write_dev_flush, this will wake anyone waiting
3348 * for the barrier when it is done
3350 static void btrfs_end_empty_barrier(struct bio *bio)
3352 complete(bio->bi_private);
3356 * Submit a flush request to the device if it supports it. Error handling is
3357 * done in the waiting counterpart.
3359 static void write_dev_flush(struct btrfs_device *device)
3361 struct request_queue *q = bdev_get_queue(device->bdev);
3362 struct bio *bio = device->flush_bio;
3364 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3368 bio->bi_end_io = btrfs_end_empty_barrier;
3369 bio_set_dev(bio, device->bdev);
3370 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3371 init_completion(&device->flush_wait);
3372 bio->bi_private = &device->flush_wait;
3374 btrfsic_submit_bio(bio);
3375 device->flush_bio_sent = 1;
3379 * If the flush bio has been submitted by write_dev_flush, wait for it.
3381 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3383 struct bio *bio = device->flush_bio;
3385 if (!device->flush_bio_sent)
3388 device->flush_bio_sent = 0;
3389 wait_for_completion_io(&device->flush_wait);
3391 return bio->bi_status;
3394 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3396 if (!btrfs_check_rw_degradable(fs_info))
3402 * send an empty flush down to each device in parallel,
3403 * then wait for them
3405 static int barrier_all_devices(struct btrfs_fs_info *info)
3407 struct list_head *head;
3408 struct btrfs_device *dev;
3409 int errors_wait = 0;
3412 /* send down all the barriers */
3413 head = &info->fs_devices->devices;
3414 list_for_each_entry_rcu(dev, head, dev_list) {
3419 if (!dev->in_fs_metadata || !dev->writeable)
3422 write_dev_flush(dev);
3423 dev->last_flush_error = BLK_STS_OK;
3426 /* wait for all the barriers */
3427 list_for_each_entry_rcu(dev, head, dev_list) {
3434 if (!dev->in_fs_metadata || !dev->writeable)
3437 ret = wait_dev_flush(dev);
3439 dev->last_flush_error = ret;
3440 btrfs_dev_stat_inc_and_print(dev,
3441 BTRFS_DEV_STAT_FLUSH_ERRS);
3448 * At some point we need the status of all disks
3449 * to arrive at the volume status. So error checking
3450 * is being pushed to a separate loop.
3452 return check_barrier_error(info);
3457 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3460 int min_tolerated = INT_MAX;
3462 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3463 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3464 min_tolerated = min(min_tolerated,
3465 btrfs_raid_array[BTRFS_RAID_SINGLE].
3466 tolerated_failures);
3468 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3469 if (raid_type == BTRFS_RAID_SINGLE)
3471 if (!(flags & btrfs_raid_group[raid_type]))
3473 min_tolerated = min(min_tolerated,
3474 btrfs_raid_array[raid_type].
3475 tolerated_failures);
3478 if (min_tolerated == INT_MAX) {
3479 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3483 return min_tolerated;
3486 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3488 struct list_head *head;
3489 struct btrfs_device *dev;
3490 struct btrfs_super_block *sb;
3491 struct btrfs_dev_item *dev_item;
3495 int total_errors = 0;
3498 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3501 * max_mirrors == 0 indicates we're from commit_transaction,
3502 * not from fsync where the tree roots in fs_info have not
3503 * been consistent on disk.
3505 if (max_mirrors == 0)
3506 backup_super_roots(fs_info);
3508 sb = fs_info->super_for_commit;
3509 dev_item = &sb->dev_item;
3511 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3512 head = &fs_info->fs_devices->devices;
3513 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3516 ret = barrier_all_devices(fs_info);
3519 &fs_info->fs_devices->device_list_mutex);
3520 btrfs_handle_fs_error(fs_info, ret,
3521 "errors while submitting device barriers.");
3526 list_for_each_entry_rcu(dev, head, dev_list) {
3531 if (!dev->in_fs_metadata || !dev->writeable)
3534 btrfs_set_stack_device_generation(dev_item, 0);
3535 btrfs_set_stack_device_type(dev_item, dev->type);
3536 btrfs_set_stack_device_id(dev_item, dev->devid);
3537 btrfs_set_stack_device_total_bytes(dev_item,
3538 dev->commit_total_bytes);
3539 btrfs_set_stack_device_bytes_used(dev_item,
3540 dev->commit_bytes_used);
3541 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3542 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3543 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3544 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3545 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3547 flags = btrfs_super_flags(sb);
3548 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3550 ret = write_dev_supers(dev, sb, max_mirrors);
3554 if (total_errors > max_errors) {
3555 btrfs_err(fs_info, "%d errors while writing supers",
3557 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3559 /* FUA is masked off if unsupported and can't be the reason */
3560 btrfs_handle_fs_error(fs_info, -EIO,
3561 "%d errors while writing supers",
3567 list_for_each_entry_rcu(dev, head, dev_list) {
3570 if (!dev->in_fs_metadata || !dev->writeable)
3573 ret = wait_dev_supers(dev, max_mirrors);
3577 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3578 if (total_errors > max_errors) {
3579 btrfs_handle_fs_error(fs_info, -EIO,
3580 "%d errors while writing supers",
3587 /* Drop a fs root from the radix tree and free it. */
3588 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3589 struct btrfs_root *root)
3591 spin_lock(&fs_info->fs_roots_radix_lock);
3592 radix_tree_delete(&fs_info->fs_roots_radix,
3593 (unsigned long)root->root_key.objectid);
3594 spin_unlock(&fs_info->fs_roots_radix_lock);
3596 if (btrfs_root_refs(&root->root_item) == 0)
3597 synchronize_srcu(&fs_info->subvol_srcu);
3599 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3600 btrfs_free_log(NULL, root);
3601 if (root->reloc_root) {
3602 free_extent_buffer(root->reloc_root->node);
3603 free_extent_buffer(root->reloc_root->commit_root);
3604 btrfs_put_fs_root(root->reloc_root);
3605 root->reloc_root = NULL;
3609 if (root->free_ino_pinned)
3610 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3611 if (root->free_ino_ctl)
3612 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3616 static void free_fs_root(struct btrfs_root *root)
3618 iput(root->ino_cache_inode);
3619 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3620 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3621 root->orphan_block_rsv = NULL;
3623 free_anon_bdev(root->anon_dev);
3624 if (root->subv_writers)
3625 btrfs_free_subvolume_writers(root->subv_writers);
3626 free_extent_buffer(root->node);
3627 free_extent_buffer(root->commit_root);
3628 kfree(root->free_ino_ctl);
3629 kfree(root->free_ino_pinned);
3631 btrfs_put_fs_root(root);
3634 void btrfs_free_fs_root(struct btrfs_root *root)
3639 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3641 u64 root_objectid = 0;
3642 struct btrfs_root *gang[8];
3645 unsigned int ret = 0;
3649 index = srcu_read_lock(&fs_info->subvol_srcu);
3650 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3651 (void **)gang, root_objectid,
3654 srcu_read_unlock(&fs_info->subvol_srcu, index);
3657 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3659 for (i = 0; i < ret; i++) {
3660 /* Avoid to grab roots in dead_roots */
3661 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3665 /* grab all the search result for later use */
3666 gang[i] = btrfs_grab_fs_root(gang[i]);
3668 srcu_read_unlock(&fs_info->subvol_srcu, index);
3670 for (i = 0; i < ret; i++) {
3673 root_objectid = gang[i]->root_key.objectid;
3674 err = btrfs_orphan_cleanup(gang[i]);
3677 btrfs_put_fs_root(gang[i]);
3682 /* release the uncleaned roots due to error */
3683 for (; i < ret; i++) {
3685 btrfs_put_fs_root(gang[i]);
3690 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3692 struct btrfs_root *root = fs_info->tree_root;
3693 struct btrfs_trans_handle *trans;
3695 mutex_lock(&fs_info->cleaner_mutex);
3696 btrfs_run_delayed_iputs(fs_info);
3697 mutex_unlock(&fs_info->cleaner_mutex);
3698 wake_up_process(fs_info->cleaner_kthread);
3700 /* wait until ongoing cleanup work done */
3701 down_write(&fs_info->cleanup_work_sem);
3702 up_write(&fs_info->cleanup_work_sem);
3704 trans = btrfs_join_transaction(root);
3706 return PTR_ERR(trans);
3707 return btrfs_commit_transaction(trans);
3710 void close_ctree(struct btrfs_fs_info *fs_info)
3712 struct btrfs_root *root = fs_info->tree_root;
3715 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
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 * If the cleaner thread is stopped and there are
3744 * block groups queued for removal, the deletion will be
3745 * skipped when we quit the cleaner thread.
3747 btrfs_delete_unused_bgs(fs_info);
3749 ret = btrfs_commit_super(fs_info);
3751 btrfs_err(fs_info, "commit super ret %d", ret);
3754 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3755 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3756 btrfs_error_commit_super(fs_info);
3758 kthread_stop(fs_info->transaction_kthread);
3759 kthread_stop(fs_info->cleaner_kthread);
3761 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3763 btrfs_free_qgroup_config(fs_info);
3764 ASSERT(list_empty(&fs_info->delalloc_roots));
3766 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3767 btrfs_info(fs_info, "at unmount delalloc count %lld",
3768 percpu_counter_sum(&fs_info->delalloc_bytes));
3771 btrfs_sysfs_remove_mounted(fs_info);
3772 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3774 btrfs_free_fs_roots(fs_info);
3776 btrfs_put_block_group_cache(fs_info);
3779 * we must make sure there is not any read request to
3780 * submit after we stopping all workers.
3782 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3783 btrfs_stop_all_workers(fs_info);
3785 btrfs_free_block_groups(fs_info);
3787 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3788 free_root_pointers(fs_info, 1);
3790 iput(fs_info->btree_inode);
3792 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3793 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3794 btrfsic_unmount(fs_info->fs_devices);
3797 btrfs_close_devices(fs_info->fs_devices);
3798 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3800 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3801 percpu_counter_destroy(&fs_info->delalloc_bytes);
3802 percpu_counter_destroy(&fs_info->bio_counter);
3803 cleanup_srcu_struct(&fs_info->subvol_srcu);
3805 btrfs_free_stripe_hash_table(fs_info);
3807 __btrfs_free_block_rsv(root->orphan_block_rsv);
3808 root->orphan_block_rsv = NULL;
3810 while (!list_empty(&fs_info->pinned_chunks)) {
3811 struct extent_map *em;
3813 em = list_first_entry(&fs_info->pinned_chunks,
3814 struct extent_map, list);
3815 list_del_init(&em->list);
3816 free_extent_map(em);
3820 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3824 struct inode *btree_inode = buf->pages[0]->mapping->host;
3826 ret = extent_buffer_uptodate(buf);
3830 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3831 parent_transid, atomic);
3837 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3839 struct btrfs_fs_info *fs_info;
3840 struct btrfs_root *root;
3841 u64 transid = btrfs_header_generation(buf);
3844 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3846 * This is a fast path so only do this check if we have sanity tests
3847 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3848 * outside of the sanity tests.
3850 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3853 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3854 fs_info = root->fs_info;
3855 btrfs_assert_tree_locked(buf);
3856 if (transid != fs_info->generation)
3857 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3858 buf->start, transid, fs_info->generation);
3859 was_dirty = set_extent_buffer_dirty(buf);
3861 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3863 fs_info->dirty_metadata_batch);
3864 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3866 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3867 * but item data not updated.
3868 * So here we should only check item pointers, not item data.
3870 if (btrfs_header_level(buf) == 0 &&
3871 btrfs_check_leaf_relaxed(root, buf)) {
3872 btrfs_print_leaf(buf);
3878 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
3882 * looks as though older kernels can get into trouble with
3883 * this code, they end up stuck in balance_dirty_pages forever
3887 if (current->flags & PF_MEMALLOC)
3891 btrfs_balance_delayed_items(fs_info);
3893 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
3894 BTRFS_DIRTY_METADATA_THRESH,
3895 fs_info->dirty_metadata_batch);
3897 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
3901 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
3903 __btrfs_btree_balance_dirty(fs_info, 1);
3906 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
3908 __btrfs_btree_balance_dirty(fs_info, 0);
3911 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3913 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3914 struct btrfs_fs_info *fs_info = root->fs_info;
3916 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
3919 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
3921 struct btrfs_super_block *sb = fs_info->super_copy;
3922 u64 nodesize = btrfs_super_nodesize(sb);
3923 u64 sectorsize = btrfs_super_sectorsize(sb);
3926 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
3927 btrfs_err(fs_info, "no valid FS found");
3930 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
3931 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
3932 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
3935 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3936 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
3937 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3940 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3941 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
3942 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3945 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3946 btrfs_err(fs_info, "log_root level too big: %d >= %d",
3947 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3952 * Check sectorsize and nodesize first, other check will need it.
3953 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
3955 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
3956 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3957 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
3960 /* Only PAGE SIZE is supported yet */
3961 if (sectorsize != PAGE_SIZE) {
3963 "sectorsize %llu not supported yet, only support %lu",
3964 sectorsize, PAGE_SIZE);
3967 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
3968 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3969 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
3972 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
3973 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
3974 le32_to_cpu(sb->__unused_leafsize), nodesize);
3978 /* Root alignment check */
3979 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
3980 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
3981 btrfs_super_root(sb));
3984 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
3985 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
3986 btrfs_super_chunk_root(sb));
3989 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
3990 btrfs_warn(fs_info, "log_root block unaligned: %llu",
3991 btrfs_super_log_root(sb));
3995 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
3997 "dev_item UUID does not match fsid: %pU != %pU",
3998 fs_info->fsid, sb->dev_item.fsid);
4003 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4006 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4007 btrfs_err(fs_info, "bytes_used is too small %llu",
4008 btrfs_super_bytes_used(sb));
4011 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4012 btrfs_err(fs_info, "invalid stripesize %u",
4013 btrfs_super_stripesize(sb));
4016 if (btrfs_super_num_devices(sb) > (1UL << 31))
4017 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4018 btrfs_super_num_devices(sb));
4019 if (btrfs_super_num_devices(sb) == 0) {
4020 btrfs_err(fs_info, "number of devices is 0");
4024 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4025 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4026 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4031 * Obvious sys_chunk_array corruptions, it must hold at least one key
4034 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4035 btrfs_err(fs_info, "system chunk array too big %u > %u",
4036 btrfs_super_sys_array_size(sb),
4037 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4040 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4041 + sizeof(struct btrfs_chunk)) {
4042 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4043 btrfs_super_sys_array_size(sb),
4044 sizeof(struct btrfs_disk_key)
4045 + sizeof(struct btrfs_chunk));
4050 * The generation is a global counter, we'll trust it more than the others
4051 * but it's still possible that it's the one that's wrong.
4053 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4055 "suspicious: generation < chunk_root_generation: %llu < %llu",
4056 btrfs_super_generation(sb),
4057 btrfs_super_chunk_root_generation(sb));
4058 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4059 && btrfs_super_cache_generation(sb) != (u64)-1)
4061 "suspicious: generation < cache_generation: %llu < %llu",
4062 btrfs_super_generation(sb),
4063 btrfs_super_cache_generation(sb));
4068 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4070 /* cleanup FS via transaction */
4071 btrfs_cleanup_transaction(fs_info);
4073 mutex_lock(&fs_info->cleaner_mutex);
4074 btrfs_run_delayed_iputs(fs_info);
4075 mutex_unlock(&fs_info->cleaner_mutex);
4077 down_write(&fs_info->cleanup_work_sem);
4078 up_write(&fs_info->cleanup_work_sem);
4081 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4083 struct btrfs_ordered_extent *ordered;
4085 spin_lock(&root->ordered_extent_lock);
4087 * This will just short circuit the ordered completion stuff which will
4088 * make sure the ordered extent gets properly cleaned up.
4090 list_for_each_entry(ordered, &root->ordered_extents,
4092 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4093 spin_unlock(&root->ordered_extent_lock);
4096 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4098 struct btrfs_root *root;
4099 struct list_head splice;
4101 INIT_LIST_HEAD(&splice);
4103 spin_lock(&fs_info->ordered_root_lock);
4104 list_splice_init(&fs_info->ordered_roots, &splice);
4105 while (!list_empty(&splice)) {
4106 root = list_first_entry(&splice, struct btrfs_root,
4108 list_move_tail(&root->ordered_root,
4109 &fs_info->ordered_roots);
4111 spin_unlock(&fs_info->ordered_root_lock);
4112 btrfs_destroy_ordered_extents(root);
4115 spin_lock(&fs_info->ordered_root_lock);
4117 spin_unlock(&fs_info->ordered_root_lock);
4120 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4121 struct btrfs_fs_info *fs_info)
4123 struct rb_node *node;
4124 struct btrfs_delayed_ref_root *delayed_refs;
4125 struct btrfs_delayed_ref_node *ref;
4128 delayed_refs = &trans->delayed_refs;
4130 spin_lock(&delayed_refs->lock);
4131 if (atomic_read(&delayed_refs->num_entries) == 0) {
4132 spin_unlock(&delayed_refs->lock);
4133 btrfs_info(fs_info, "delayed_refs has NO entry");
4137 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4138 struct btrfs_delayed_ref_head *head;
4139 struct btrfs_delayed_ref_node *tmp;
4140 bool pin_bytes = false;
4142 head = rb_entry(node, struct btrfs_delayed_ref_head,
4144 if (!mutex_trylock(&head->mutex)) {
4145 refcount_inc(&head->node.refs);
4146 spin_unlock(&delayed_refs->lock);
4148 mutex_lock(&head->mutex);
4149 mutex_unlock(&head->mutex);
4150 btrfs_put_delayed_ref(&head->node);
4151 spin_lock(&delayed_refs->lock);
4154 spin_lock(&head->lock);
4155 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4158 list_del(&ref->list);
4159 if (!list_empty(&ref->add_list))
4160 list_del(&ref->add_list);
4161 atomic_dec(&delayed_refs->num_entries);
4162 btrfs_put_delayed_ref(ref);
4164 if (head->must_insert_reserved)
4166 btrfs_free_delayed_extent_op(head->extent_op);
4167 delayed_refs->num_heads--;
4168 if (head->processing == 0)
4169 delayed_refs->num_heads_ready--;
4170 atomic_dec(&delayed_refs->num_entries);
4171 head->node.in_tree = 0;
4172 rb_erase(&head->href_node, &delayed_refs->href_root);
4173 spin_unlock(&head->lock);
4174 spin_unlock(&delayed_refs->lock);
4175 mutex_unlock(&head->mutex);
4178 btrfs_pin_extent(fs_info, head->node.bytenr,
4179 head->node.num_bytes, 1);
4180 btrfs_put_delayed_ref(&head->node);
4182 spin_lock(&delayed_refs->lock);
4185 spin_unlock(&delayed_refs->lock);
4190 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4192 struct btrfs_inode *btrfs_inode;
4193 struct list_head splice;
4195 INIT_LIST_HEAD(&splice);
4197 spin_lock(&root->delalloc_lock);
4198 list_splice_init(&root->delalloc_inodes, &splice);
4200 while (!list_empty(&splice)) {
4201 struct inode *inode = NULL;
4202 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4204 __btrfs_del_delalloc_inode(root, btrfs_inode);
4205 spin_unlock(&root->delalloc_lock);
4208 * Make sure we get a live inode and that it'll not disappear
4211 inode = igrab(&btrfs_inode->vfs_inode);
4213 invalidate_inode_pages2(inode->i_mapping);
4216 spin_lock(&root->delalloc_lock);
4218 spin_unlock(&root->delalloc_lock);
4221 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4223 struct btrfs_root *root;
4224 struct list_head splice;
4226 INIT_LIST_HEAD(&splice);
4228 spin_lock(&fs_info->delalloc_root_lock);
4229 list_splice_init(&fs_info->delalloc_roots, &splice);
4230 while (!list_empty(&splice)) {
4231 root = list_first_entry(&splice, struct btrfs_root,
4233 root = btrfs_grab_fs_root(root);
4235 spin_unlock(&fs_info->delalloc_root_lock);
4237 btrfs_destroy_delalloc_inodes(root);
4238 btrfs_put_fs_root(root);
4240 spin_lock(&fs_info->delalloc_root_lock);
4242 spin_unlock(&fs_info->delalloc_root_lock);
4245 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4246 struct extent_io_tree *dirty_pages,
4250 struct extent_buffer *eb;
4255 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4260 clear_extent_bits(dirty_pages, start, end, mark);
4261 while (start <= end) {
4262 eb = find_extent_buffer(fs_info, start);
4263 start += fs_info->nodesize;
4266 wait_on_extent_buffer_writeback(eb);
4268 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4270 clear_extent_buffer_dirty(eb);
4271 free_extent_buffer_stale(eb);
4278 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4279 struct extent_io_tree *pinned_extents)
4281 struct extent_io_tree *unpin;
4287 unpin = pinned_extents;
4291 * The btrfs_finish_extent_commit() may get the same range as
4292 * ours between find_first_extent_bit and clear_extent_dirty.
4293 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4294 * the same extent range.
4296 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4297 ret = find_first_extent_bit(unpin, 0, &start, &end,
4298 EXTENT_DIRTY, NULL);
4300 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4304 clear_extent_dirty(unpin, start, end);
4305 btrfs_error_unpin_extent_range(fs_info, start, end);
4306 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4311 if (unpin == &fs_info->freed_extents[0])
4312 unpin = &fs_info->freed_extents[1];
4314 unpin = &fs_info->freed_extents[0];
4322 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4324 struct inode *inode;
4326 inode = cache->io_ctl.inode;
4328 invalidate_inode_pages2(inode->i_mapping);
4329 BTRFS_I(inode)->generation = 0;
4330 cache->io_ctl.inode = NULL;
4333 btrfs_put_block_group(cache);
4336 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4337 struct btrfs_fs_info *fs_info)
4339 struct btrfs_block_group_cache *cache;
4341 spin_lock(&cur_trans->dirty_bgs_lock);
4342 while (!list_empty(&cur_trans->dirty_bgs)) {
4343 cache = list_first_entry(&cur_trans->dirty_bgs,
4344 struct btrfs_block_group_cache,
4347 btrfs_err(fs_info, "orphan block group dirty_bgs list");
4348 spin_unlock(&cur_trans->dirty_bgs_lock);
4352 if (!list_empty(&cache->io_list)) {
4353 spin_unlock(&cur_trans->dirty_bgs_lock);
4354 list_del_init(&cache->io_list);
4355 btrfs_cleanup_bg_io(cache);
4356 spin_lock(&cur_trans->dirty_bgs_lock);
4359 list_del_init(&cache->dirty_list);
4360 spin_lock(&cache->lock);
4361 cache->disk_cache_state = BTRFS_DC_ERROR;
4362 spin_unlock(&cache->lock);
4364 spin_unlock(&cur_trans->dirty_bgs_lock);
4365 btrfs_put_block_group(cache);
4366 spin_lock(&cur_trans->dirty_bgs_lock);
4368 spin_unlock(&cur_trans->dirty_bgs_lock);
4370 while (!list_empty(&cur_trans->io_bgs)) {
4371 cache = list_first_entry(&cur_trans->io_bgs,
4372 struct btrfs_block_group_cache,
4375 btrfs_err(fs_info, "orphan block group on io_bgs list");
4379 list_del_init(&cache->io_list);
4380 spin_lock(&cache->lock);
4381 cache->disk_cache_state = BTRFS_DC_ERROR;
4382 spin_unlock(&cache->lock);
4383 btrfs_cleanup_bg_io(cache);
4387 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4388 struct btrfs_fs_info *fs_info)
4390 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4391 ASSERT(list_empty(&cur_trans->dirty_bgs));
4392 ASSERT(list_empty(&cur_trans->io_bgs));
4394 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4396 cur_trans->state = TRANS_STATE_COMMIT_START;
4397 wake_up(&fs_info->transaction_blocked_wait);
4399 cur_trans->state = TRANS_STATE_UNBLOCKED;
4400 wake_up(&fs_info->transaction_wait);
4402 btrfs_destroy_delayed_inodes(fs_info);
4403 btrfs_assert_delayed_root_empty(fs_info);
4405 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4407 btrfs_destroy_pinned_extent(fs_info,
4408 fs_info->pinned_extents);
4410 cur_trans->state =TRANS_STATE_COMPLETED;
4411 wake_up(&cur_trans->commit_wait);
4414 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4416 struct btrfs_transaction *t;
4418 mutex_lock(&fs_info->transaction_kthread_mutex);
4420 spin_lock(&fs_info->trans_lock);
4421 while (!list_empty(&fs_info->trans_list)) {
4422 t = list_first_entry(&fs_info->trans_list,
4423 struct btrfs_transaction, list);
4424 if (t->state >= TRANS_STATE_COMMIT_START) {
4425 refcount_inc(&t->use_count);
4426 spin_unlock(&fs_info->trans_lock);
4427 btrfs_wait_for_commit(fs_info, t->transid);
4428 btrfs_put_transaction(t);
4429 spin_lock(&fs_info->trans_lock);
4432 if (t == fs_info->running_transaction) {
4433 t->state = TRANS_STATE_COMMIT_DOING;
4434 spin_unlock(&fs_info->trans_lock);
4436 * We wait for 0 num_writers since we don't hold a trans
4437 * handle open currently for this transaction.
4439 wait_event(t->writer_wait,
4440 atomic_read(&t->num_writers) == 0);
4442 spin_unlock(&fs_info->trans_lock);
4444 btrfs_cleanup_one_transaction(t, fs_info);
4446 spin_lock(&fs_info->trans_lock);
4447 if (t == fs_info->running_transaction)
4448 fs_info->running_transaction = NULL;
4449 list_del_init(&t->list);
4450 spin_unlock(&fs_info->trans_lock);
4452 btrfs_put_transaction(t);
4453 trace_btrfs_transaction_commit(fs_info->tree_root);
4454 spin_lock(&fs_info->trans_lock);
4456 spin_unlock(&fs_info->trans_lock);
4457 btrfs_destroy_all_ordered_extents(fs_info);
4458 btrfs_destroy_delayed_inodes(fs_info);
4459 btrfs_assert_delayed_root_empty(fs_info);
4460 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4461 btrfs_destroy_all_delalloc_inodes(fs_info);
4462 mutex_unlock(&fs_info->transaction_kthread_mutex);
4467 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4469 struct inode *inode = private_data;
4470 return btrfs_sb(inode->i_sb);
4473 static const struct extent_io_ops btree_extent_io_ops = {
4474 /* mandatory callbacks */
4475 .submit_bio_hook = btree_submit_bio_hook,
4476 .readpage_end_io_hook = btree_readpage_end_io_hook,
4477 /* note we're sharing with inode.c for the merge bio hook */
4478 .merge_bio_hook = btrfs_merge_bio_hook,
4479 .readpage_io_failed_hook = btree_io_failed_hook,
4480 .set_range_writeback = btrfs_set_range_writeback,
4481 .tree_fs_info = btree_fs_info,
4483 /* optional callbacks */