1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "inode-map.h"
33 #include "check-integrity.h"
34 #include "rcu-string.h"
35 #include "dev-replace.h"
39 #include "compression.h"
40 #include "tree-checker.h"
41 #include "ref-verify.h"
42 #include "block-group.h"
44 #include "space-info.h"
46 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
53 static void end_workqueue_fn(struct btrfs_work *work);
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
68 * is complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
71 struct btrfs_end_io_wq {
75 struct btrfs_fs_info *info;
77 enum btrfs_wq_endio_type metadata;
78 struct btrfs_work work;
81 static struct kmem_cache *btrfs_end_io_wq_cache;
83 int __init btrfs_end_io_wq_init(void)
85 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 sizeof(struct btrfs_end_io_wq),
90 if (!btrfs_end_io_wq_cache)
95 void __cold btrfs_end_io_wq_exit(void)
97 kmem_cache_destroy(btrfs_end_io_wq_cache);
100 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102 if (fs_info->csum_shash)
103 crypto_free_shash(fs_info->csum_shash);
107 * async submit bios are used to offload expensive checksumming
108 * onto the worker threads. They checksum file and metadata bios
109 * just before they are sent down the IO stack.
111 struct async_submit_bio {
114 extent_submit_bio_start_t *submit_bio_start;
117 * bio_offset is optional, can be used if the pages in the bio
118 * can't tell us where in the file the bio should go
121 struct btrfs_work work;
126 * Lockdep class keys for extent_buffer->lock's in this root. For a given
127 * eb, the lockdep key is determined by the btrfs_root it belongs to and
128 * the level the eb occupies in the tree.
130 * Different roots are used for different purposes and may nest inside each
131 * other and they require separate keysets. As lockdep keys should be
132 * static, assign keysets according to the purpose of the root as indicated
133 * by btrfs_root->root_key.objectid. This ensures that all special purpose
134 * roots have separate keysets.
136 * Lock-nesting across peer nodes is always done with the immediate parent
137 * node locked thus preventing deadlock. As lockdep doesn't know this, use
138 * subclass to avoid triggering lockdep warning in such cases.
140 * The key is set by the readpage_end_io_hook after the buffer has passed
141 * csum validation but before the pages are unlocked. It is also set by
142 * btrfs_init_new_buffer on freshly allocated blocks.
144 * We also add a check to make sure the highest level of the tree is the
145 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
146 * needs update as well.
148 #ifdef CONFIG_DEBUG_LOCK_ALLOC
149 # if BTRFS_MAX_LEVEL != 8
153 #define DEFINE_LEVEL(stem, level) \
154 .names[level] = "btrfs-" stem "-0" #level,
156 #define DEFINE_NAME(stem) \
157 DEFINE_LEVEL(stem, 0) \
158 DEFINE_LEVEL(stem, 1) \
159 DEFINE_LEVEL(stem, 2) \
160 DEFINE_LEVEL(stem, 3) \
161 DEFINE_LEVEL(stem, 4) \
162 DEFINE_LEVEL(stem, 5) \
163 DEFINE_LEVEL(stem, 6) \
164 DEFINE_LEVEL(stem, 7)
166 static struct btrfs_lockdep_keyset {
167 u64 id; /* root objectid */
168 /* Longest entry: btrfs-free-space-00 */
169 char names[BTRFS_MAX_LEVEL][20];
170 struct lock_class_key keys[BTRFS_MAX_LEVEL];
171 } btrfs_lockdep_keysets[] = {
172 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") },
173 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") },
174 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") },
175 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") },
176 { .id = BTRFS_FS_TREE_OBJECTID, DEFINE_NAME("fs") },
177 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
178 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
179 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
180 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
181 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
182 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
183 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
184 { .id = 0, DEFINE_NAME("tree") },
190 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
193 struct btrfs_lockdep_keyset *ks;
195 BUG_ON(level >= ARRAY_SIZE(ks->keys));
197 /* find the matching keyset, id 0 is the default entry */
198 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
199 if (ks->id == objectid)
202 lockdep_set_class_and_name(&eb->lock,
203 &ks->keys[level], ks->names[level]);
209 * Compute the csum of a btree block and store the result to provided buffer.
211 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
213 struct btrfs_fs_info *fs_info = buf->fs_info;
214 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
215 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
219 shash->tfm = fs_info->csum_shash;
220 crypto_shash_init(shash);
221 kaddr = page_address(buf->pages[0]);
222 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
223 PAGE_SIZE - BTRFS_CSUM_SIZE);
225 for (i = 1; i < num_pages; i++) {
226 kaddr = page_address(buf->pages[i]);
227 crypto_shash_update(shash, kaddr, PAGE_SIZE);
229 memset(result, 0, BTRFS_CSUM_SIZE);
230 crypto_shash_final(shash, result);
234 * we can't consider a given block up to date unless the transid of the
235 * block matches the transid in the parent node's pointer. This is how we
236 * detect blocks that either didn't get written at all or got written
237 * in the wrong place.
239 static int verify_parent_transid(struct extent_io_tree *io_tree,
240 struct extent_buffer *eb, u64 parent_transid,
243 struct extent_state *cached_state = NULL;
245 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
247 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
254 btrfs_tree_read_lock(eb);
255 btrfs_set_lock_blocking_read(eb);
258 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
260 if (extent_buffer_uptodate(eb) &&
261 btrfs_header_generation(eb) == parent_transid) {
265 btrfs_err_rl(eb->fs_info,
266 "parent transid verify failed on %llu wanted %llu found %llu",
268 parent_transid, btrfs_header_generation(eb));
272 * Things reading via commit roots that don't have normal protection,
273 * like send, can have a really old block in cache that may point at a
274 * block that has been freed and re-allocated. So don't clear uptodate
275 * if we find an eb that is under IO (dirty/writeback) because we could
276 * end up reading in the stale data and then writing it back out and
277 * making everybody very sad.
279 if (!extent_buffer_under_io(eb))
280 clear_extent_buffer_uptodate(eb);
282 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
285 btrfs_tree_read_unlock_blocking(eb);
289 static bool btrfs_supported_super_csum(u16 csum_type)
292 case BTRFS_CSUM_TYPE_CRC32:
293 case BTRFS_CSUM_TYPE_XXHASH:
294 case BTRFS_CSUM_TYPE_SHA256:
295 case BTRFS_CSUM_TYPE_BLAKE2:
303 * Return 0 if the superblock checksum type matches the checksum value of that
304 * algorithm. Pass the raw disk superblock data.
306 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
309 struct btrfs_super_block *disk_sb =
310 (struct btrfs_super_block *)raw_disk_sb;
311 char result[BTRFS_CSUM_SIZE];
312 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
314 shash->tfm = fs_info->csum_shash;
317 * The super_block structure does not span the whole
318 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
319 * filled with zeros and is included in the checksum.
321 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
322 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
324 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
330 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
331 struct btrfs_key *first_key, u64 parent_transid)
333 struct btrfs_fs_info *fs_info = eb->fs_info;
335 struct btrfs_key found_key;
338 found_level = btrfs_header_level(eb);
339 if (found_level != level) {
340 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
341 KERN_ERR "BTRFS: tree level check failed\n");
343 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
344 eb->start, level, found_level);
352 * For live tree block (new tree blocks in current transaction),
353 * we need proper lock context to avoid race, which is impossible here.
354 * So we only checks tree blocks which is read from disk, whose
355 * generation <= fs_info->last_trans_committed.
357 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
360 /* We have @first_key, so this @eb must have at least one item */
361 if (btrfs_header_nritems(eb) == 0) {
363 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
365 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
370 btrfs_node_key_to_cpu(eb, &found_key, 0);
372 btrfs_item_key_to_cpu(eb, &found_key, 0);
373 ret = btrfs_comp_cpu_keys(first_key, &found_key);
376 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
377 KERN_ERR "BTRFS: tree first key check failed\n");
379 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
380 eb->start, parent_transid, first_key->objectid,
381 first_key->type, first_key->offset,
382 found_key.objectid, found_key.type,
389 * helper to read a given tree block, doing retries as required when
390 * the checksums don't match and we have alternate mirrors to try.
392 * @parent_transid: expected transid, skip check if 0
393 * @level: expected level, mandatory check
394 * @first_key: expected key of first slot, skip check if NULL
396 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
397 u64 parent_transid, int level,
398 struct btrfs_key *first_key)
400 struct btrfs_fs_info *fs_info = eb->fs_info;
401 struct extent_io_tree *io_tree;
406 int failed_mirror = 0;
408 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
410 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
411 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
413 if (verify_parent_transid(io_tree, eb,
416 else if (btrfs_verify_level_key(eb, level,
417 first_key, parent_transid))
423 num_copies = btrfs_num_copies(fs_info,
428 if (!failed_mirror) {
430 failed_mirror = eb->read_mirror;
434 if (mirror_num == failed_mirror)
437 if (mirror_num > num_copies)
441 if (failed && !ret && failed_mirror)
442 btrfs_repair_eb_io_failure(eb, failed_mirror);
448 * checksum a dirty tree block before IO. This has extra checks to make sure
449 * we only fill in the checksum field in the first page of a multi-page block
452 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
454 u64 start = page_offset(page);
456 u8 result[BTRFS_CSUM_SIZE];
457 struct extent_buffer *eb;
460 eb = (struct extent_buffer *)page->private;
461 if (page != eb->pages[0])
464 found_start = btrfs_header_bytenr(eb);
466 * Please do not consolidate these warnings into a single if.
467 * It is useful to know what went wrong.
469 if (WARN_ON(found_start != start))
471 if (WARN_ON(!PageUptodate(page)))
474 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
475 offsetof(struct btrfs_header, fsid),
476 BTRFS_FSID_SIZE) == 0);
478 csum_tree_block(eb, result);
480 if (btrfs_header_level(eb))
481 ret = btrfs_check_node(eb);
483 ret = btrfs_check_leaf_full(eb);
486 btrfs_print_tree(eb, 0);
488 "block=%llu write time tree block corruption detected",
490 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
493 write_extent_buffer(eb, result, 0, fs_info->csum_size);
498 static int check_tree_block_fsid(struct extent_buffer *eb)
500 struct btrfs_fs_info *fs_info = eb->fs_info;
501 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
502 u8 fsid[BTRFS_FSID_SIZE];
505 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
508 * Checking the incompat flag is only valid for the current fs. For
509 * seed devices it's forbidden to have their uuid changed so reading
510 * ->fsid in this case is fine
512 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
513 metadata_uuid = fs_devices->metadata_uuid;
515 metadata_uuid = fs_devices->fsid;
517 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
520 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
521 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
527 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, u64 phy_offset,
528 struct page *page, u64 start, u64 end,
533 struct extent_buffer *eb;
534 struct btrfs_fs_info *fs_info;
537 u8 result[BTRFS_CSUM_SIZE];
543 eb = (struct extent_buffer *)page->private;
544 fs_info = eb->fs_info;
545 csum_size = fs_info->csum_size;
547 /* the pending IO might have been the only thing that kept this buffer
548 * in memory. Make sure we have a ref for all this other checks
550 atomic_inc(&eb->refs);
552 reads_done = atomic_dec_and_test(&eb->io_pages);
556 eb->read_mirror = mirror;
557 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
562 found_start = btrfs_header_bytenr(eb);
563 if (found_start != eb->start) {
564 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
565 eb->start, found_start);
569 if (check_tree_block_fsid(eb)) {
570 btrfs_err_rl(fs_info, "bad fsid on block %llu",
575 found_level = btrfs_header_level(eb);
576 if (found_level >= BTRFS_MAX_LEVEL) {
577 btrfs_err(fs_info, "bad tree block level %d on %llu",
578 (int)btrfs_header_level(eb), eb->start);
583 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
586 csum_tree_block(eb, result);
588 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
589 u8 val[BTRFS_CSUM_SIZE] = { 0 };
591 read_extent_buffer(eb, &val, 0, csum_size);
592 btrfs_warn_rl(fs_info,
593 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
594 fs_info->sb->s_id, eb->start,
595 CSUM_FMT_VALUE(csum_size, val),
596 CSUM_FMT_VALUE(csum_size, result),
597 btrfs_header_level(eb));
603 * If this is a leaf block and it is corrupt, set the corrupt bit so
604 * that we don't try and read the other copies of this block, just
607 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
608 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
612 if (found_level > 0 && btrfs_check_node(eb))
616 set_extent_buffer_uptodate(eb);
619 "block=%llu read time tree block corruption detected",
623 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
624 btree_readahead_hook(eb, ret);
628 * our io error hook is going to dec the io pages
629 * again, we have to make sure it has something
632 atomic_inc(&eb->io_pages);
633 clear_extent_buffer_uptodate(eb);
635 free_extent_buffer(eb);
640 static void end_workqueue_bio(struct bio *bio)
642 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
643 struct btrfs_fs_info *fs_info;
644 struct btrfs_workqueue *wq;
646 fs_info = end_io_wq->info;
647 end_io_wq->status = bio->bi_status;
649 if (bio_op(bio) == REQ_OP_WRITE) {
650 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
651 wq = fs_info->endio_meta_write_workers;
652 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
653 wq = fs_info->endio_freespace_worker;
654 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
655 wq = fs_info->endio_raid56_workers;
657 wq = fs_info->endio_write_workers;
659 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
660 wq = fs_info->endio_raid56_workers;
661 else if (end_io_wq->metadata)
662 wq = fs_info->endio_meta_workers;
664 wq = fs_info->endio_workers;
667 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
668 btrfs_queue_work(wq, &end_io_wq->work);
671 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
672 enum btrfs_wq_endio_type metadata)
674 struct btrfs_end_io_wq *end_io_wq;
676 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
678 return BLK_STS_RESOURCE;
680 end_io_wq->private = bio->bi_private;
681 end_io_wq->end_io = bio->bi_end_io;
682 end_io_wq->info = info;
683 end_io_wq->status = 0;
684 end_io_wq->bio = bio;
685 end_io_wq->metadata = metadata;
687 bio->bi_private = end_io_wq;
688 bio->bi_end_io = end_workqueue_bio;
692 static void run_one_async_start(struct btrfs_work *work)
694 struct async_submit_bio *async;
697 async = container_of(work, struct async_submit_bio, work);
698 ret = async->submit_bio_start(async->inode, async->bio, async->bio_offset);
704 * In order to insert checksums into the metadata in large chunks, we wait
705 * until bio submission time. All the pages in the bio are checksummed and
706 * sums are attached onto the ordered extent record.
708 * At IO completion time the csums attached on the ordered extent record are
709 * inserted into the tree.
711 static void run_one_async_done(struct btrfs_work *work)
713 struct async_submit_bio *async;
717 async = container_of(work, struct async_submit_bio, work);
718 inode = async->inode;
720 /* If an error occurred we just want to clean up the bio and move on */
722 async->bio->bi_status = async->status;
723 bio_endio(async->bio);
728 * All of the bios that pass through here are from async helpers.
729 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
730 * This changes nothing when cgroups aren't in use.
732 async->bio->bi_opf |= REQ_CGROUP_PUNT;
733 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
735 async->bio->bi_status = ret;
736 bio_endio(async->bio);
740 static void run_one_async_free(struct btrfs_work *work)
742 struct async_submit_bio *async;
744 async = container_of(work, struct async_submit_bio, work);
748 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
749 int mirror_num, unsigned long bio_flags,
751 extent_submit_bio_start_t *submit_bio_start)
753 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
754 struct async_submit_bio *async;
756 async = kmalloc(sizeof(*async), GFP_NOFS);
758 return BLK_STS_RESOURCE;
760 async->inode = inode;
762 async->mirror_num = mirror_num;
763 async->submit_bio_start = submit_bio_start;
765 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
768 async->bio_offset = bio_offset;
772 if (op_is_sync(bio->bi_opf))
773 btrfs_set_work_high_priority(&async->work);
775 btrfs_queue_work(fs_info->workers, &async->work);
779 static blk_status_t btree_csum_one_bio(struct bio *bio)
781 struct bio_vec *bvec;
782 struct btrfs_root *root;
784 struct bvec_iter_all iter_all;
786 ASSERT(!bio_flagged(bio, BIO_CLONED));
787 bio_for_each_segment_all(bvec, bio, iter_all) {
788 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
789 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
794 return errno_to_blk_status(ret);
797 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
801 * when we're called for a write, we're already in the async
802 * submission context. Just jump into btrfs_map_bio
804 return btree_csum_one_bio(bio);
807 static int check_async_write(struct btrfs_fs_info *fs_info,
808 struct btrfs_inode *bi)
810 if (atomic_read(&bi->sync_writers))
812 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
817 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
818 int mirror_num, unsigned long bio_flags)
820 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
821 int async = check_async_write(fs_info, BTRFS_I(inode));
824 if (bio_op(bio) != REQ_OP_WRITE) {
826 * called for a read, do the setup so that checksum validation
827 * can happen in the async kernel threads
829 ret = btrfs_bio_wq_end_io(fs_info, bio,
830 BTRFS_WQ_ENDIO_METADATA);
833 ret = btrfs_map_bio(fs_info, bio, mirror_num);
835 ret = btree_csum_one_bio(bio);
838 ret = btrfs_map_bio(fs_info, bio, mirror_num);
841 * kthread helpers are used to submit writes so that
842 * checksumming can happen in parallel across all CPUs
844 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
845 0, btree_submit_bio_start);
853 bio->bi_status = ret;
858 #ifdef CONFIG_MIGRATION
859 static int btree_migratepage(struct address_space *mapping,
860 struct page *newpage, struct page *page,
861 enum migrate_mode mode)
864 * we can't safely write a btree page from here,
865 * we haven't done the locking hook
870 * Buffers may be managed in a filesystem specific way.
871 * We must have no buffers or drop them.
873 if (page_has_private(page) &&
874 !try_to_release_page(page, GFP_KERNEL))
876 return migrate_page(mapping, newpage, page, mode);
881 static int btree_writepages(struct address_space *mapping,
882 struct writeback_control *wbc)
884 struct btrfs_fs_info *fs_info;
887 if (wbc->sync_mode == WB_SYNC_NONE) {
889 if (wbc->for_kupdate)
892 fs_info = BTRFS_I(mapping->host)->root->fs_info;
893 /* this is a bit racy, but that's ok */
894 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
895 BTRFS_DIRTY_METADATA_THRESH,
896 fs_info->dirty_metadata_batch);
900 return btree_write_cache_pages(mapping, wbc);
903 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
905 if (PageWriteback(page) || PageDirty(page))
908 return try_release_extent_buffer(page);
911 static void btree_invalidatepage(struct page *page, unsigned int offset,
914 struct extent_io_tree *tree;
915 tree = &BTRFS_I(page->mapping->host)->io_tree;
916 extent_invalidatepage(tree, page, offset);
917 btree_releasepage(page, GFP_NOFS);
918 if (PagePrivate(page)) {
919 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
920 "page private not zero on page %llu",
921 (unsigned long long)page_offset(page));
922 detach_page_private(page);
926 static int btree_set_page_dirty(struct page *page)
929 struct extent_buffer *eb;
931 BUG_ON(!PagePrivate(page));
932 eb = (struct extent_buffer *)page->private;
934 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
935 BUG_ON(!atomic_read(&eb->refs));
936 btrfs_assert_tree_locked(eb);
938 return __set_page_dirty_nobuffers(page);
941 static const struct address_space_operations btree_aops = {
942 .writepages = btree_writepages,
943 .releasepage = btree_releasepage,
944 .invalidatepage = btree_invalidatepage,
945 #ifdef CONFIG_MIGRATION
946 .migratepage = btree_migratepage,
948 .set_page_dirty = btree_set_page_dirty,
951 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
953 struct extent_buffer *buf = NULL;
956 buf = btrfs_find_create_tree_block(fs_info, bytenr);
960 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
962 free_extent_buffer_stale(buf);
964 free_extent_buffer(buf);
967 struct extent_buffer *btrfs_find_create_tree_block(
968 struct btrfs_fs_info *fs_info,
971 if (btrfs_is_testing(fs_info))
972 return alloc_test_extent_buffer(fs_info, bytenr);
973 return alloc_extent_buffer(fs_info, bytenr);
977 * Read tree block at logical address @bytenr and do variant basic but critical
980 * @parent_transid: expected transid of this tree block, skip check if 0
981 * @level: expected level, mandatory check
982 * @first_key: expected key in slot 0, skip check if NULL
984 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
985 u64 parent_transid, int level,
986 struct btrfs_key *first_key)
988 struct extent_buffer *buf = NULL;
991 buf = btrfs_find_create_tree_block(fs_info, bytenr);
995 ret = btree_read_extent_buffer_pages(buf, parent_transid,
998 free_extent_buffer_stale(buf);
1005 void btrfs_clean_tree_block(struct extent_buffer *buf)
1007 struct btrfs_fs_info *fs_info = buf->fs_info;
1008 if (btrfs_header_generation(buf) ==
1009 fs_info->running_transaction->transid) {
1010 btrfs_assert_tree_locked(buf);
1012 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1013 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1015 fs_info->dirty_metadata_batch);
1016 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1017 btrfs_set_lock_blocking_write(buf);
1018 clear_extent_buffer_dirty(buf);
1023 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1026 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1027 root->fs_info = fs_info;
1029 root->commit_root = NULL;
1031 root->orphan_cleanup_state = 0;
1033 root->last_trans = 0;
1034 root->highest_objectid = 0;
1035 root->nr_delalloc_inodes = 0;
1036 root->nr_ordered_extents = 0;
1037 root->inode_tree = RB_ROOT;
1038 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1039 root->block_rsv = NULL;
1041 INIT_LIST_HEAD(&root->dirty_list);
1042 INIT_LIST_HEAD(&root->root_list);
1043 INIT_LIST_HEAD(&root->delalloc_inodes);
1044 INIT_LIST_HEAD(&root->delalloc_root);
1045 INIT_LIST_HEAD(&root->ordered_extents);
1046 INIT_LIST_HEAD(&root->ordered_root);
1047 INIT_LIST_HEAD(&root->reloc_dirty_list);
1048 INIT_LIST_HEAD(&root->logged_list[0]);
1049 INIT_LIST_HEAD(&root->logged_list[1]);
1050 spin_lock_init(&root->inode_lock);
1051 spin_lock_init(&root->delalloc_lock);
1052 spin_lock_init(&root->ordered_extent_lock);
1053 spin_lock_init(&root->accounting_lock);
1054 spin_lock_init(&root->log_extents_lock[0]);
1055 spin_lock_init(&root->log_extents_lock[1]);
1056 spin_lock_init(&root->qgroup_meta_rsv_lock);
1057 mutex_init(&root->objectid_mutex);
1058 mutex_init(&root->log_mutex);
1059 mutex_init(&root->ordered_extent_mutex);
1060 mutex_init(&root->delalloc_mutex);
1061 init_waitqueue_head(&root->qgroup_flush_wait);
1062 init_waitqueue_head(&root->log_writer_wait);
1063 init_waitqueue_head(&root->log_commit_wait[0]);
1064 init_waitqueue_head(&root->log_commit_wait[1]);
1065 INIT_LIST_HEAD(&root->log_ctxs[0]);
1066 INIT_LIST_HEAD(&root->log_ctxs[1]);
1067 atomic_set(&root->log_commit[0], 0);
1068 atomic_set(&root->log_commit[1], 0);
1069 atomic_set(&root->log_writers, 0);
1070 atomic_set(&root->log_batch, 0);
1071 refcount_set(&root->refs, 1);
1072 atomic_set(&root->snapshot_force_cow, 0);
1073 atomic_set(&root->nr_swapfiles, 0);
1074 root->log_transid = 0;
1075 root->log_transid_committed = -1;
1076 root->last_log_commit = 0;
1078 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1079 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1080 extent_io_tree_init(fs_info, &root->log_csum_range,
1081 IO_TREE_LOG_CSUM_RANGE, NULL);
1084 memset(&root->root_key, 0, sizeof(root->root_key));
1085 memset(&root->root_item, 0, sizeof(root->root_item));
1086 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1087 root->root_key.objectid = objectid;
1090 spin_lock_init(&root->root_item_lock);
1091 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1092 #ifdef CONFIG_BTRFS_DEBUG
1093 INIT_LIST_HEAD(&root->leak_list);
1094 spin_lock(&fs_info->fs_roots_radix_lock);
1095 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1096 spin_unlock(&fs_info->fs_roots_radix_lock);
1100 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1101 u64 objectid, gfp_t flags)
1103 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1105 __setup_root(root, fs_info, objectid);
1109 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1110 /* Should only be used by the testing infrastructure */
1111 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1113 struct btrfs_root *root;
1116 return ERR_PTR(-EINVAL);
1118 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1120 return ERR_PTR(-ENOMEM);
1122 /* We don't use the stripesize in selftest, set it as sectorsize */
1123 root->alloc_bytenr = 0;
1129 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1132 struct btrfs_fs_info *fs_info = trans->fs_info;
1133 struct extent_buffer *leaf;
1134 struct btrfs_root *tree_root = fs_info->tree_root;
1135 struct btrfs_root *root;
1136 struct btrfs_key key;
1137 unsigned int nofs_flag;
1141 * We're holding a transaction handle, so use a NOFS memory allocation
1142 * context to avoid deadlock if reclaim happens.
1144 nofs_flag = memalloc_nofs_save();
1145 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1146 memalloc_nofs_restore(nofs_flag);
1148 return ERR_PTR(-ENOMEM);
1150 root->root_key.objectid = objectid;
1151 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1152 root->root_key.offset = 0;
1154 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1155 BTRFS_NESTING_NORMAL);
1157 ret = PTR_ERR(leaf);
1163 btrfs_mark_buffer_dirty(leaf);
1165 root->commit_root = btrfs_root_node(root);
1166 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1168 btrfs_set_root_flags(&root->root_item, 0);
1169 btrfs_set_root_limit(&root->root_item, 0);
1170 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1171 btrfs_set_root_generation(&root->root_item, trans->transid);
1172 btrfs_set_root_level(&root->root_item, 0);
1173 btrfs_set_root_refs(&root->root_item, 1);
1174 btrfs_set_root_used(&root->root_item, leaf->len);
1175 btrfs_set_root_last_snapshot(&root->root_item, 0);
1176 btrfs_set_root_dirid(&root->root_item, 0);
1177 if (is_fstree(objectid))
1178 generate_random_guid(root->root_item.uuid);
1180 export_guid(root->root_item.uuid, &guid_null);
1181 btrfs_set_root_drop_level(&root->root_item, 0);
1183 key.objectid = objectid;
1184 key.type = BTRFS_ROOT_ITEM_KEY;
1186 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1190 btrfs_tree_unlock(leaf);
1196 btrfs_tree_unlock(leaf);
1197 btrfs_put_root(root);
1199 return ERR_PTR(ret);
1202 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1203 struct btrfs_fs_info *fs_info)
1205 struct btrfs_root *root;
1206 struct extent_buffer *leaf;
1208 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1210 return ERR_PTR(-ENOMEM);
1212 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1213 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1214 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1217 * DON'T set SHAREABLE bit for log trees.
1219 * Log trees are not exposed to user space thus can't be snapshotted,
1220 * and they go away before a real commit is actually done.
1222 * They do store pointers to file data extents, and those reference
1223 * counts still get updated (along with back refs to the log tree).
1226 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1227 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1229 btrfs_put_root(root);
1230 return ERR_CAST(leaf);
1235 btrfs_mark_buffer_dirty(root->node);
1236 btrfs_tree_unlock(root->node);
1240 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1241 struct btrfs_fs_info *fs_info)
1243 struct btrfs_root *log_root;
1245 log_root = alloc_log_tree(trans, fs_info);
1246 if (IS_ERR(log_root))
1247 return PTR_ERR(log_root);
1248 WARN_ON(fs_info->log_root_tree);
1249 fs_info->log_root_tree = log_root;
1253 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1254 struct btrfs_root *root)
1256 struct btrfs_fs_info *fs_info = root->fs_info;
1257 struct btrfs_root *log_root;
1258 struct btrfs_inode_item *inode_item;
1260 log_root = alloc_log_tree(trans, fs_info);
1261 if (IS_ERR(log_root))
1262 return PTR_ERR(log_root);
1264 log_root->last_trans = trans->transid;
1265 log_root->root_key.offset = root->root_key.objectid;
1267 inode_item = &log_root->root_item.inode;
1268 btrfs_set_stack_inode_generation(inode_item, 1);
1269 btrfs_set_stack_inode_size(inode_item, 3);
1270 btrfs_set_stack_inode_nlink(inode_item, 1);
1271 btrfs_set_stack_inode_nbytes(inode_item,
1273 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1275 btrfs_set_root_node(&log_root->root_item, log_root->node);
1277 WARN_ON(root->log_root);
1278 root->log_root = log_root;
1279 root->log_transid = 0;
1280 root->log_transid_committed = -1;
1281 root->last_log_commit = 0;
1285 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1286 struct btrfs_path *path,
1287 struct btrfs_key *key)
1289 struct btrfs_root *root;
1290 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1295 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1297 return ERR_PTR(-ENOMEM);
1299 ret = btrfs_find_root(tree_root, key, path,
1300 &root->root_item, &root->root_key);
1307 generation = btrfs_root_generation(&root->root_item);
1308 level = btrfs_root_level(&root->root_item);
1309 root->node = read_tree_block(fs_info,
1310 btrfs_root_bytenr(&root->root_item),
1311 generation, level, NULL);
1312 if (IS_ERR(root->node)) {
1313 ret = PTR_ERR(root->node);
1316 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1320 root->commit_root = btrfs_root_node(root);
1323 btrfs_put_root(root);
1324 return ERR_PTR(ret);
1327 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1328 struct btrfs_key *key)
1330 struct btrfs_root *root;
1331 struct btrfs_path *path;
1333 path = btrfs_alloc_path();
1335 return ERR_PTR(-ENOMEM);
1336 root = read_tree_root_path(tree_root, path, key);
1337 btrfs_free_path(path);
1343 * Initialize subvolume root in-memory structure
1345 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1347 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1350 unsigned int nofs_flag;
1352 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1353 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1355 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1361 * We might be called under a transaction (e.g. indirect backref
1362 * resolution) which could deadlock if it triggers memory reclaim
1364 nofs_flag = memalloc_nofs_save();
1365 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1366 memalloc_nofs_restore(nofs_flag);
1370 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1371 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1372 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1373 btrfs_check_and_init_root_item(&root->root_item);
1376 btrfs_init_free_ino_ctl(root);
1377 spin_lock_init(&root->ino_cache_lock);
1378 init_waitqueue_head(&root->ino_cache_wait);
1381 * Don't assign anonymous block device to roots that are not exposed to
1382 * userspace, the id pool is limited to 1M
1384 if (is_fstree(root->root_key.objectid) &&
1385 btrfs_root_refs(&root->root_item) > 0) {
1387 ret = get_anon_bdev(&root->anon_dev);
1391 root->anon_dev = anon_dev;
1395 mutex_lock(&root->objectid_mutex);
1396 ret = btrfs_find_highest_objectid(root,
1397 &root->highest_objectid);
1399 mutex_unlock(&root->objectid_mutex);
1403 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1405 mutex_unlock(&root->objectid_mutex);
1409 /* The caller is responsible to call btrfs_free_fs_root */
1413 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1416 struct btrfs_root *root;
1418 spin_lock(&fs_info->fs_roots_radix_lock);
1419 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1420 (unsigned long)root_id);
1422 root = btrfs_grab_root(root);
1423 spin_unlock(&fs_info->fs_roots_radix_lock);
1427 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1430 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1431 return btrfs_grab_root(fs_info->tree_root);
1432 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1433 return btrfs_grab_root(fs_info->extent_root);
1434 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1435 return btrfs_grab_root(fs_info->chunk_root);
1436 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1437 return btrfs_grab_root(fs_info->dev_root);
1438 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1439 return btrfs_grab_root(fs_info->csum_root);
1440 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1441 return btrfs_grab_root(fs_info->quota_root) ?
1442 fs_info->quota_root : ERR_PTR(-ENOENT);
1443 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1444 return btrfs_grab_root(fs_info->uuid_root) ?
1445 fs_info->uuid_root : ERR_PTR(-ENOENT);
1446 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1447 return btrfs_grab_root(fs_info->free_space_root) ?
1448 fs_info->free_space_root : ERR_PTR(-ENOENT);
1452 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1453 struct btrfs_root *root)
1457 ret = radix_tree_preload(GFP_NOFS);
1461 spin_lock(&fs_info->fs_roots_radix_lock);
1462 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1463 (unsigned long)root->root_key.objectid,
1466 btrfs_grab_root(root);
1467 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1469 spin_unlock(&fs_info->fs_roots_radix_lock);
1470 radix_tree_preload_end();
1475 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1477 #ifdef CONFIG_BTRFS_DEBUG
1478 struct btrfs_root *root;
1480 while (!list_empty(&fs_info->allocated_roots)) {
1481 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1483 root = list_first_entry(&fs_info->allocated_roots,
1484 struct btrfs_root, leak_list);
1485 btrfs_err(fs_info, "leaked root %s refcount %d",
1486 btrfs_root_name(root->root_key.objectid, buf),
1487 refcount_read(&root->refs));
1488 while (refcount_read(&root->refs) > 1)
1489 btrfs_put_root(root);
1490 btrfs_put_root(root);
1495 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1497 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1498 percpu_counter_destroy(&fs_info->delalloc_bytes);
1499 percpu_counter_destroy(&fs_info->dio_bytes);
1500 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1501 btrfs_free_csum_hash(fs_info);
1502 btrfs_free_stripe_hash_table(fs_info);
1503 btrfs_free_ref_cache(fs_info);
1504 kfree(fs_info->balance_ctl);
1505 kfree(fs_info->delayed_root);
1506 btrfs_put_root(fs_info->extent_root);
1507 btrfs_put_root(fs_info->tree_root);
1508 btrfs_put_root(fs_info->chunk_root);
1509 btrfs_put_root(fs_info->dev_root);
1510 btrfs_put_root(fs_info->csum_root);
1511 btrfs_put_root(fs_info->quota_root);
1512 btrfs_put_root(fs_info->uuid_root);
1513 btrfs_put_root(fs_info->free_space_root);
1514 btrfs_put_root(fs_info->fs_root);
1515 btrfs_put_root(fs_info->data_reloc_root);
1516 btrfs_check_leaked_roots(fs_info);
1517 btrfs_extent_buffer_leak_debug_check(fs_info);
1518 kfree(fs_info->super_copy);
1519 kfree(fs_info->super_for_commit);
1525 * Get an in-memory reference of a root structure.
1527 * For essential trees like root/extent tree, we grab it from fs_info directly.
1528 * For subvolume trees, we check the cached filesystem roots first. If not
1529 * found, then read it from disk and add it to cached fs roots.
1531 * Caller should release the root by calling btrfs_put_root() after the usage.
1533 * NOTE: Reloc and log trees can't be read by this function as they share the
1534 * same root objectid.
1536 * @objectid: root id
1537 * @anon_dev: preallocated anonymous block device number for new roots,
1538 * pass 0 for new allocation.
1539 * @check_ref: whether to check root item references, If true, return -ENOENT
1542 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1543 u64 objectid, dev_t anon_dev,
1546 struct btrfs_root *root;
1547 struct btrfs_path *path;
1548 struct btrfs_key key;
1551 root = btrfs_get_global_root(fs_info, objectid);
1555 root = btrfs_lookup_fs_root(fs_info, objectid);
1557 /* Shouldn't get preallocated anon_dev for cached roots */
1559 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1560 btrfs_put_root(root);
1561 return ERR_PTR(-ENOENT);
1566 key.objectid = objectid;
1567 key.type = BTRFS_ROOT_ITEM_KEY;
1568 key.offset = (u64)-1;
1569 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1573 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1578 ret = btrfs_init_fs_root(root, anon_dev);
1582 path = btrfs_alloc_path();
1587 key.objectid = BTRFS_ORPHAN_OBJECTID;
1588 key.type = BTRFS_ORPHAN_ITEM_KEY;
1589 key.offset = objectid;
1591 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1592 btrfs_free_path(path);
1596 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1598 ret = btrfs_insert_fs_root(fs_info, root);
1600 btrfs_put_root(root);
1607 btrfs_put_root(root);
1608 return ERR_PTR(ret);
1612 * Get in-memory reference of a root structure
1614 * @objectid: tree objectid
1615 * @check_ref: if set, verify that the tree exists and the item has at least
1618 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1619 u64 objectid, bool check_ref)
1621 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1625 * Get in-memory reference of a root structure, created as new, optionally pass
1626 * the anonymous block device id
1628 * @objectid: tree objectid
1629 * @anon_dev: if zero, allocate a new anonymous block device or use the
1632 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1633 u64 objectid, dev_t anon_dev)
1635 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1639 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1640 * @fs_info: the fs_info
1641 * @objectid: the objectid we need to lookup
1643 * This is exclusively used for backref walking, and exists specifically because
1644 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1645 * creation time, which means we may have to read the tree_root in order to look
1646 * up a fs root that is not in memory. If the root is not in memory we will
1647 * read the tree root commit root and look up the fs root from there. This is a
1648 * temporary root, it will not be inserted into the radix tree as it doesn't
1649 * have the most uptodate information, it'll simply be discarded once the
1650 * backref code is finished using the root.
1652 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1653 struct btrfs_path *path,
1656 struct btrfs_root *root;
1657 struct btrfs_key key;
1659 ASSERT(path->search_commit_root && path->skip_locking);
1662 * This can return -ENOENT if we ask for a root that doesn't exist, but
1663 * since this is called via the backref walking code we won't be looking
1664 * up a root that doesn't exist, unless there's corruption. So if root
1665 * != NULL just return it.
1667 root = btrfs_get_global_root(fs_info, objectid);
1671 root = btrfs_lookup_fs_root(fs_info, objectid);
1675 key.objectid = objectid;
1676 key.type = BTRFS_ROOT_ITEM_KEY;
1677 key.offset = (u64)-1;
1678 root = read_tree_root_path(fs_info->tree_root, path, &key);
1679 btrfs_release_path(path);
1685 * called by the kthread helper functions to finally call the bio end_io
1686 * functions. This is where read checksum verification actually happens
1688 static void end_workqueue_fn(struct btrfs_work *work)
1691 struct btrfs_end_io_wq *end_io_wq;
1693 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1694 bio = end_io_wq->bio;
1696 bio->bi_status = end_io_wq->status;
1697 bio->bi_private = end_io_wq->private;
1698 bio->bi_end_io = end_io_wq->end_io;
1700 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1703 static int cleaner_kthread(void *arg)
1705 struct btrfs_root *root = arg;
1706 struct btrfs_fs_info *fs_info = root->fs_info;
1712 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1714 /* Make the cleaner go to sleep early. */
1715 if (btrfs_need_cleaner_sleep(fs_info))
1719 * Do not do anything if we might cause open_ctree() to block
1720 * before we have finished mounting the filesystem.
1722 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1725 if (!mutex_trylock(&fs_info->cleaner_mutex))
1729 * Avoid the problem that we change the status of the fs
1730 * during the above check and trylock.
1732 if (btrfs_need_cleaner_sleep(fs_info)) {
1733 mutex_unlock(&fs_info->cleaner_mutex);
1737 btrfs_run_delayed_iputs(fs_info);
1739 again = btrfs_clean_one_deleted_snapshot(root);
1740 mutex_unlock(&fs_info->cleaner_mutex);
1743 * The defragger has dealt with the R/O remount and umount,
1744 * needn't do anything special here.
1746 btrfs_run_defrag_inodes(fs_info);
1749 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1750 * with relocation (btrfs_relocate_chunk) and relocation
1751 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1752 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1753 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1754 * unused block groups.
1756 btrfs_delete_unused_bgs(fs_info);
1758 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1759 if (kthread_should_park())
1761 if (kthread_should_stop())
1764 set_current_state(TASK_INTERRUPTIBLE);
1766 __set_current_state(TASK_RUNNING);
1771 static int transaction_kthread(void *arg)
1773 struct btrfs_root *root = arg;
1774 struct btrfs_fs_info *fs_info = root->fs_info;
1775 struct btrfs_trans_handle *trans;
1776 struct btrfs_transaction *cur;
1779 unsigned long delay;
1783 cannot_commit = false;
1784 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1785 mutex_lock(&fs_info->transaction_kthread_mutex);
1787 spin_lock(&fs_info->trans_lock);
1788 cur = fs_info->running_transaction;
1790 spin_unlock(&fs_info->trans_lock);
1794 delta = ktime_get_seconds() - cur->start_time;
1795 if (cur->state < TRANS_STATE_COMMIT_START &&
1796 delta < fs_info->commit_interval) {
1797 spin_unlock(&fs_info->trans_lock);
1798 delay -= msecs_to_jiffies((delta - 1) * 1000);
1800 msecs_to_jiffies(fs_info->commit_interval * 1000));
1803 transid = cur->transid;
1804 spin_unlock(&fs_info->trans_lock);
1806 /* If the file system is aborted, this will always fail. */
1807 trans = btrfs_attach_transaction(root);
1808 if (IS_ERR(trans)) {
1809 if (PTR_ERR(trans) != -ENOENT)
1810 cannot_commit = true;
1813 if (transid == trans->transid) {
1814 btrfs_commit_transaction(trans);
1816 btrfs_end_transaction(trans);
1819 wake_up_process(fs_info->cleaner_kthread);
1820 mutex_unlock(&fs_info->transaction_kthread_mutex);
1822 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1823 &fs_info->fs_state)))
1824 btrfs_cleanup_transaction(fs_info);
1825 if (!kthread_should_stop() &&
1826 (!btrfs_transaction_blocked(fs_info) ||
1828 schedule_timeout_interruptible(delay);
1829 } while (!kthread_should_stop());
1834 * This will find the highest generation in the array of root backups. The
1835 * index of the highest array is returned, or -EINVAL if we can't find
1838 * We check to make sure the array is valid by comparing the
1839 * generation of the latest root in the array with the generation
1840 * in the super block. If they don't match we pitch it.
1842 static int find_newest_super_backup(struct btrfs_fs_info *info)
1844 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1846 struct btrfs_root_backup *root_backup;
1849 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1850 root_backup = info->super_copy->super_roots + i;
1851 cur = btrfs_backup_tree_root_gen(root_backup);
1852 if (cur == newest_gen)
1860 * copy all the root pointers into the super backup array.
1861 * this will bump the backup pointer by one when it is
1864 static void backup_super_roots(struct btrfs_fs_info *info)
1866 const int next_backup = info->backup_root_index;
1867 struct btrfs_root_backup *root_backup;
1869 root_backup = info->super_for_commit->super_roots + next_backup;
1872 * make sure all of our padding and empty slots get zero filled
1873 * regardless of which ones we use today
1875 memset(root_backup, 0, sizeof(*root_backup));
1877 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1879 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1880 btrfs_set_backup_tree_root_gen(root_backup,
1881 btrfs_header_generation(info->tree_root->node));
1883 btrfs_set_backup_tree_root_level(root_backup,
1884 btrfs_header_level(info->tree_root->node));
1886 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1887 btrfs_set_backup_chunk_root_gen(root_backup,
1888 btrfs_header_generation(info->chunk_root->node));
1889 btrfs_set_backup_chunk_root_level(root_backup,
1890 btrfs_header_level(info->chunk_root->node));
1892 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1893 btrfs_set_backup_extent_root_gen(root_backup,
1894 btrfs_header_generation(info->extent_root->node));
1895 btrfs_set_backup_extent_root_level(root_backup,
1896 btrfs_header_level(info->extent_root->node));
1899 * we might commit during log recovery, which happens before we set
1900 * the fs_root. Make sure it is valid before we fill it in.
1902 if (info->fs_root && info->fs_root->node) {
1903 btrfs_set_backup_fs_root(root_backup,
1904 info->fs_root->node->start);
1905 btrfs_set_backup_fs_root_gen(root_backup,
1906 btrfs_header_generation(info->fs_root->node));
1907 btrfs_set_backup_fs_root_level(root_backup,
1908 btrfs_header_level(info->fs_root->node));
1911 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1912 btrfs_set_backup_dev_root_gen(root_backup,
1913 btrfs_header_generation(info->dev_root->node));
1914 btrfs_set_backup_dev_root_level(root_backup,
1915 btrfs_header_level(info->dev_root->node));
1917 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1918 btrfs_set_backup_csum_root_gen(root_backup,
1919 btrfs_header_generation(info->csum_root->node));
1920 btrfs_set_backup_csum_root_level(root_backup,
1921 btrfs_header_level(info->csum_root->node));
1923 btrfs_set_backup_total_bytes(root_backup,
1924 btrfs_super_total_bytes(info->super_copy));
1925 btrfs_set_backup_bytes_used(root_backup,
1926 btrfs_super_bytes_used(info->super_copy));
1927 btrfs_set_backup_num_devices(root_backup,
1928 btrfs_super_num_devices(info->super_copy));
1931 * if we don't copy this out to the super_copy, it won't get remembered
1932 * for the next commit
1934 memcpy(&info->super_copy->super_roots,
1935 &info->super_for_commit->super_roots,
1936 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1940 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1941 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1943 * fs_info - filesystem whose backup roots need to be read
1944 * priority - priority of backup root required
1946 * Returns backup root index on success and -EINVAL otherwise.
1948 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1950 int backup_index = find_newest_super_backup(fs_info);
1951 struct btrfs_super_block *super = fs_info->super_copy;
1952 struct btrfs_root_backup *root_backup;
1954 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1956 return backup_index;
1958 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1959 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1964 root_backup = super->super_roots + backup_index;
1966 btrfs_set_super_generation(super,
1967 btrfs_backup_tree_root_gen(root_backup));
1968 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1969 btrfs_set_super_root_level(super,
1970 btrfs_backup_tree_root_level(root_backup));
1971 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1974 * Fixme: the total bytes and num_devices need to match or we should
1977 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1978 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1980 return backup_index;
1983 /* helper to cleanup workers */
1984 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1986 btrfs_destroy_workqueue(fs_info->fixup_workers);
1987 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1988 btrfs_destroy_workqueue(fs_info->workers);
1989 btrfs_destroy_workqueue(fs_info->endio_workers);
1990 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1991 btrfs_destroy_workqueue(fs_info->rmw_workers);
1992 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1993 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1994 btrfs_destroy_workqueue(fs_info->delayed_workers);
1995 btrfs_destroy_workqueue(fs_info->caching_workers);
1996 btrfs_destroy_workqueue(fs_info->readahead_workers);
1997 btrfs_destroy_workqueue(fs_info->flush_workers);
1998 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1999 if (fs_info->discard_ctl.discard_workers)
2000 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2002 * Now that all other work queues are destroyed, we can safely destroy
2003 * the queues used for metadata I/O, since tasks from those other work
2004 * queues can do metadata I/O operations.
2006 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2007 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2010 static void free_root_extent_buffers(struct btrfs_root *root)
2013 free_extent_buffer(root->node);
2014 free_extent_buffer(root->commit_root);
2016 root->commit_root = NULL;
2020 /* helper to cleanup tree roots */
2021 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2023 free_root_extent_buffers(info->tree_root);
2025 free_root_extent_buffers(info->dev_root);
2026 free_root_extent_buffers(info->extent_root);
2027 free_root_extent_buffers(info->csum_root);
2028 free_root_extent_buffers(info->quota_root);
2029 free_root_extent_buffers(info->uuid_root);
2030 free_root_extent_buffers(info->fs_root);
2031 free_root_extent_buffers(info->data_reloc_root);
2032 if (free_chunk_root)
2033 free_root_extent_buffers(info->chunk_root);
2034 free_root_extent_buffers(info->free_space_root);
2037 void btrfs_put_root(struct btrfs_root *root)
2042 if (refcount_dec_and_test(&root->refs)) {
2043 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2044 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2046 free_anon_bdev(root->anon_dev);
2047 btrfs_drew_lock_destroy(&root->snapshot_lock);
2048 free_root_extent_buffers(root);
2049 kfree(root->free_ino_ctl);
2050 kfree(root->free_ino_pinned);
2051 #ifdef CONFIG_BTRFS_DEBUG
2052 spin_lock(&root->fs_info->fs_roots_radix_lock);
2053 list_del_init(&root->leak_list);
2054 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2060 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2063 struct btrfs_root *gang[8];
2066 while (!list_empty(&fs_info->dead_roots)) {
2067 gang[0] = list_entry(fs_info->dead_roots.next,
2068 struct btrfs_root, root_list);
2069 list_del(&gang[0]->root_list);
2071 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2072 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2073 btrfs_put_root(gang[0]);
2077 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2082 for (i = 0; i < ret; i++)
2083 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2087 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2089 mutex_init(&fs_info->scrub_lock);
2090 atomic_set(&fs_info->scrubs_running, 0);
2091 atomic_set(&fs_info->scrub_pause_req, 0);
2092 atomic_set(&fs_info->scrubs_paused, 0);
2093 atomic_set(&fs_info->scrub_cancel_req, 0);
2094 init_waitqueue_head(&fs_info->scrub_pause_wait);
2095 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2098 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2100 spin_lock_init(&fs_info->balance_lock);
2101 mutex_init(&fs_info->balance_mutex);
2102 atomic_set(&fs_info->balance_pause_req, 0);
2103 atomic_set(&fs_info->balance_cancel_req, 0);
2104 fs_info->balance_ctl = NULL;
2105 init_waitqueue_head(&fs_info->balance_wait_q);
2108 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2110 struct inode *inode = fs_info->btree_inode;
2112 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2113 set_nlink(inode, 1);
2115 * we set the i_size on the btree inode to the max possible int.
2116 * the real end of the address space is determined by all of
2117 * the devices in the system
2119 inode->i_size = OFFSET_MAX;
2120 inode->i_mapping->a_ops = &btree_aops;
2122 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2123 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2124 IO_TREE_BTREE_INODE_IO, inode);
2125 BTRFS_I(inode)->io_tree.track_uptodate = false;
2126 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2128 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2129 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2130 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2131 btrfs_insert_inode_hash(inode);
2134 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2136 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2137 init_rwsem(&fs_info->dev_replace.rwsem);
2138 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2141 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2143 spin_lock_init(&fs_info->qgroup_lock);
2144 mutex_init(&fs_info->qgroup_ioctl_lock);
2145 fs_info->qgroup_tree = RB_ROOT;
2146 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2147 fs_info->qgroup_seq = 1;
2148 fs_info->qgroup_ulist = NULL;
2149 fs_info->qgroup_rescan_running = false;
2150 mutex_init(&fs_info->qgroup_rescan_lock);
2153 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2154 struct btrfs_fs_devices *fs_devices)
2156 u32 max_active = fs_info->thread_pool_size;
2157 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2160 btrfs_alloc_workqueue(fs_info, "worker",
2161 flags | WQ_HIGHPRI, max_active, 16);
2163 fs_info->delalloc_workers =
2164 btrfs_alloc_workqueue(fs_info, "delalloc",
2165 flags, max_active, 2);
2167 fs_info->flush_workers =
2168 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2169 flags, max_active, 0);
2171 fs_info->caching_workers =
2172 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2174 fs_info->fixup_workers =
2175 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2178 * endios are largely parallel and should have a very
2181 fs_info->endio_workers =
2182 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2183 fs_info->endio_meta_workers =
2184 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2186 fs_info->endio_meta_write_workers =
2187 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2189 fs_info->endio_raid56_workers =
2190 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2192 fs_info->rmw_workers =
2193 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2194 fs_info->endio_write_workers =
2195 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2197 fs_info->endio_freespace_worker =
2198 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2200 fs_info->delayed_workers =
2201 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2203 fs_info->readahead_workers =
2204 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2206 fs_info->qgroup_rescan_workers =
2207 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2208 fs_info->discard_ctl.discard_workers =
2209 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2211 if (!(fs_info->workers && fs_info->delalloc_workers &&
2212 fs_info->flush_workers &&
2213 fs_info->endio_workers && fs_info->endio_meta_workers &&
2214 fs_info->endio_meta_write_workers &&
2215 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2216 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2217 fs_info->caching_workers && fs_info->readahead_workers &&
2218 fs_info->fixup_workers && fs_info->delayed_workers &&
2219 fs_info->qgroup_rescan_workers &&
2220 fs_info->discard_ctl.discard_workers)) {
2227 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2229 struct crypto_shash *csum_shash;
2230 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2232 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2234 if (IS_ERR(csum_shash)) {
2235 btrfs_err(fs_info, "error allocating %s hash for checksum",
2237 return PTR_ERR(csum_shash);
2240 fs_info->csum_shash = csum_shash;
2245 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2246 struct btrfs_fs_devices *fs_devices)
2249 struct btrfs_root *log_tree_root;
2250 struct btrfs_super_block *disk_super = fs_info->super_copy;
2251 u64 bytenr = btrfs_super_log_root(disk_super);
2252 int level = btrfs_super_log_root_level(disk_super);
2254 if (fs_devices->rw_devices == 0) {
2255 btrfs_warn(fs_info, "log replay required on RO media");
2259 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2264 log_tree_root->node = read_tree_block(fs_info, bytenr,
2265 fs_info->generation + 1,
2267 if (IS_ERR(log_tree_root->node)) {
2268 btrfs_warn(fs_info, "failed to read log tree");
2269 ret = PTR_ERR(log_tree_root->node);
2270 log_tree_root->node = NULL;
2271 btrfs_put_root(log_tree_root);
2273 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2274 btrfs_err(fs_info, "failed to read log tree");
2275 btrfs_put_root(log_tree_root);
2278 /* returns with log_tree_root freed on success */
2279 ret = btrfs_recover_log_trees(log_tree_root);
2281 btrfs_handle_fs_error(fs_info, ret,
2282 "Failed to recover log tree");
2283 btrfs_put_root(log_tree_root);
2287 if (sb_rdonly(fs_info->sb)) {
2288 ret = btrfs_commit_super(fs_info);
2296 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2298 struct btrfs_root *tree_root = fs_info->tree_root;
2299 struct btrfs_root *root;
2300 struct btrfs_key location;
2303 BUG_ON(!fs_info->tree_root);
2305 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2306 location.type = BTRFS_ROOT_ITEM_KEY;
2307 location.offset = 0;
2309 root = btrfs_read_tree_root(tree_root, &location);
2311 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2312 ret = PTR_ERR(root);
2316 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2317 fs_info->extent_root = root;
2320 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2321 root = btrfs_read_tree_root(tree_root, &location);
2323 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2324 ret = PTR_ERR(root);
2328 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2329 fs_info->dev_root = root;
2330 btrfs_init_devices_late(fs_info);
2333 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2334 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2335 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2336 root = btrfs_read_tree_root(tree_root, &location);
2338 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2339 ret = PTR_ERR(root);
2343 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2344 fs_info->csum_root = root;
2349 * This tree can share blocks with some other fs tree during relocation
2350 * and we need a proper setup by btrfs_get_fs_root
2352 root = btrfs_get_fs_root(tree_root->fs_info,
2353 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2355 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2356 ret = PTR_ERR(root);
2360 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2361 fs_info->data_reloc_root = root;
2364 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2365 root = btrfs_read_tree_root(tree_root, &location);
2366 if (!IS_ERR(root)) {
2367 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2368 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2369 fs_info->quota_root = root;
2372 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2373 root = btrfs_read_tree_root(tree_root, &location);
2375 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2376 ret = PTR_ERR(root);
2381 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2382 fs_info->uuid_root = root;
2385 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2386 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2387 root = btrfs_read_tree_root(tree_root, &location);
2389 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2390 ret = PTR_ERR(root);
2394 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2395 fs_info->free_space_root = root;
2401 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2402 location.objectid, ret);
2407 * Real super block validation
2408 * NOTE: super csum type and incompat features will not be checked here.
2410 * @sb: super block to check
2411 * @mirror_num: the super block number to check its bytenr:
2412 * 0 the primary (1st) sb
2413 * 1, 2 2nd and 3rd backup copy
2414 * -1 skip bytenr check
2416 static int validate_super(struct btrfs_fs_info *fs_info,
2417 struct btrfs_super_block *sb, int mirror_num)
2419 u64 nodesize = btrfs_super_nodesize(sb);
2420 u64 sectorsize = btrfs_super_sectorsize(sb);
2423 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2424 btrfs_err(fs_info, "no valid FS found");
2427 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2428 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2429 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2432 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2433 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2434 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2437 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2438 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2439 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2442 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2443 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2444 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2449 * Check sectorsize and nodesize first, other check will need it.
2450 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2452 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2453 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2454 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2457 /* Only PAGE SIZE is supported yet */
2458 if (sectorsize != PAGE_SIZE) {
2460 "sectorsize %llu not supported yet, only support %lu",
2461 sectorsize, PAGE_SIZE);
2464 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2465 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2466 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2469 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2470 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2471 le32_to_cpu(sb->__unused_leafsize), nodesize);
2475 /* Root alignment check */
2476 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2477 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2478 btrfs_super_root(sb));
2481 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2482 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2483 btrfs_super_chunk_root(sb));
2486 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2487 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2488 btrfs_super_log_root(sb));
2492 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2493 BTRFS_FSID_SIZE) != 0) {
2495 "dev_item UUID does not match metadata fsid: %pU != %pU",
2496 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2501 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2504 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2505 btrfs_err(fs_info, "bytes_used is too small %llu",
2506 btrfs_super_bytes_used(sb));
2509 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2510 btrfs_err(fs_info, "invalid stripesize %u",
2511 btrfs_super_stripesize(sb));
2514 if (btrfs_super_num_devices(sb) > (1UL << 31))
2515 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2516 btrfs_super_num_devices(sb));
2517 if (btrfs_super_num_devices(sb) == 0) {
2518 btrfs_err(fs_info, "number of devices is 0");
2522 if (mirror_num >= 0 &&
2523 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2524 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2525 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2530 * Obvious sys_chunk_array corruptions, it must hold at least one key
2533 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2534 btrfs_err(fs_info, "system chunk array too big %u > %u",
2535 btrfs_super_sys_array_size(sb),
2536 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2539 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2540 + sizeof(struct btrfs_chunk)) {
2541 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2542 btrfs_super_sys_array_size(sb),
2543 sizeof(struct btrfs_disk_key)
2544 + sizeof(struct btrfs_chunk));
2549 * The generation is a global counter, we'll trust it more than the others
2550 * but it's still possible that it's the one that's wrong.
2552 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2554 "suspicious: generation < chunk_root_generation: %llu < %llu",
2555 btrfs_super_generation(sb),
2556 btrfs_super_chunk_root_generation(sb));
2557 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2558 && btrfs_super_cache_generation(sb) != (u64)-1)
2560 "suspicious: generation < cache_generation: %llu < %llu",
2561 btrfs_super_generation(sb),
2562 btrfs_super_cache_generation(sb));
2568 * Validation of super block at mount time.
2569 * Some checks already done early at mount time, like csum type and incompat
2570 * flags will be skipped.
2572 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2574 return validate_super(fs_info, fs_info->super_copy, 0);
2578 * Validation of super block at write time.
2579 * Some checks like bytenr check will be skipped as their values will be
2581 * Extra checks like csum type and incompat flags will be done here.
2583 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2584 struct btrfs_super_block *sb)
2588 ret = validate_super(fs_info, sb, -1);
2591 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2593 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2594 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2597 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2600 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2601 btrfs_super_incompat_flags(sb),
2602 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2608 "super block corruption detected before writing it to disk");
2612 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2614 int backup_index = find_newest_super_backup(fs_info);
2615 struct btrfs_super_block *sb = fs_info->super_copy;
2616 struct btrfs_root *tree_root = fs_info->tree_root;
2617 bool handle_error = false;
2621 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2626 if (!IS_ERR(tree_root->node))
2627 free_extent_buffer(tree_root->node);
2628 tree_root->node = NULL;
2630 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2633 free_root_pointers(fs_info, 0);
2636 * Don't use the log in recovery mode, it won't be
2639 btrfs_set_super_log_root(sb, 0);
2641 /* We can't trust the free space cache either */
2642 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2644 ret = read_backup_root(fs_info, i);
2649 generation = btrfs_super_generation(sb);
2650 level = btrfs_super_root_level(sb);
2651 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2652 generation, level, NULL);
2653 if (IS_ERR(tree_root->node)) {
2654 handle_error = true;
2655 ret = PTR_ERR(tree_root->node);
2656 tree_root->node = NULL;
2657 btrfs_warn(fs_info, "couldn't read tree root");
2660 } else if (!extent_buffer_uptodate(tree_root->node)) {
2661 handle_error = true;
2663 btrfs_warn(fs_info, "error while reading tree root");
2667 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2668 tree_root->commit_root = btrfs_root_node(tree_root);
2669 btrfs_set_root_refs(&tree_root->root_item, 1);
2672 * No need to hold btrfs_root::objectid_mutex since the fs
2673 * hasn't been fully initialised and we are the only user
2675 ret = btrfs_find_highest_objectid(tree_root,
2676 &tree_root->highest_objectid);
2678 handle_error = true;
2682 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2684 ret = btrfs_read_roots(fs_info);
2686 handle_error = true;
2690 /* All successful */
2691 fs_info->generation = generation;
2692 fs_info->last_trans_committed = generation;
2694 /* Always begin writing backup roots after the one being used */
2695 if (backup_index < 0) {
2696 fs_info->backup_root_index = 0;
2698 fs_info->backup_root_index = backup_index + 1;
2699 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2707 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2709 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2710 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2711 INIT_LIST_HEAD(&fs_info->trans_list);
2712 INIT_LIST_HEAD(&fs_info->dead_roots);
2713 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2714 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2715 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2716 spin_lock_init(&fs_info->delalloc_root_lock);
2717 spin_lock_init(&fs_info->trans_lock);
2718 spin_lock_init(&fs_info->fs_roots_radix_lock);
2719 spin_lock_init(&fs_info->delayed_iput_lock);
2720 spin_lock_init(&fs_info->defrag_inodes_lock);
2721 spin_lock_init(&fs_info->super_lock);
2722 spin_lock_init(&fs_info->buffer_lock);
2723 spin_lock_init(&fs_info->unused_bgs_lock);
2724 rwlock_init(&fs_info->tree_mod_log_lock);
2725 mutex_init(&fs_info->unused_bg_unpin_mutex);
2726 mutex_init(&fs_info->delete_unused_bgs_mutex);
2727 mutex_init(&fs_info->reloc_mutex);
2728 mutex_init(&fs_info->delalloc_root_mutex);
2729 seqlock_init(&fs_info->profiles_lock);
2731 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2732 INIT_LIST_HEAD(&fs_info->space_info);
2733 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2734 INIT_LIST_HEAD(&fs_info->unused_bgs);
2735 #ifdef CONFIG_BTRFS_DEBUG
2736 INIT_LIST_HEAD(&fs_info->allocated_roots);
2737 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2738 spin_lock_init(&fs_info->eb_leak_lock);
2740 extent_map_tree_init(&fs_info->mapping_tree);
2741 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2742 BTRFS_BLOCK_RSV_GLOBAL);
2743 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2744 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2745 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2746 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2747 BTRFS_BLOCK_RSV_DELOPS);
2748 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2749 BTRFS_BLOCK_RSV_DELREFS);
2751 atomic_set(&fs_info->async_delalloc_pages, 0);
2752 atomic_set(&fs_info->defrag_running, 0);
2753 atomic_set(&fs_info->reada_works_cnt, 0);
2754 atomic_set(&fs_info->nr_delayed_iputs, 0);
2755 atomic64_set(&fs_info->tree_mod_seq, 0);
2756 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2757 fs_info->metadata_ratio = 0;
2758 fs_info->defrag_inodes = RB_ROOT;
2759 atomic64_set(&fs_info->free_chunk_space, 0);
2760 fs_info->tree_mod_log = RB_ROOT;
2761 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2762 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2763 /* readahead state */
2764 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2765 spin_lock_init(&fs_info->reada_lock);
2766 btrfs_init_ref_verify(fs_info);
2768 fs_info->thread_pool_size = min_t(unsigned long,
2769 num_online_cpus() + 2, 8);
2771 INIT_LIST_HEAD(&fs_info->ordered_roots);
2772 spin_lock_init(&fs_info->ordered_root_lock);
2774 btrfs_init_scrub(fs_info);
2775 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2776 fs_info->check_integrity_print_mask = 0;
2778 btrfs_init_balance(fs_info);
2779 btrfs_init_async_reclaim_work(fs_info);
2781 spin_lock_init(&fs_info->block_group_cache_lock);
2782 fs_info->block_group_cache_tree = RB_ROOT;
2783 fs_info->first_logical_byte = (u64)-1;
2785 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2786 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2787 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2789 mutex_init(&fs_info->ordered_operations_mutex);
2790 mutex_init(&fs_info->tree_log_mutex);
2791 mutex_init(&fs_info->chunk_mutex);
2792 mutex_init(&fs_info->transaction_kthread_mutex);
2793 mutex_init(&fs_info->cleaner_mutex);
2794 mutex_init(&fs_info->ro_block_group_mutex);
2795 init_rwsem(&fs_info->commit_root_sem);
2796 init_rwsem(&fs_info->cleanup_work_sem);
2797 init_rwsem(&fs_info->subvol_sem);
2798 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2800 btrfs_init_dev_replace_locks(fs_info);
2801 btrfs_init_qgroup(fs_info);
2802 btrfs_discard_init(fs_info);
2804 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2805 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2807 init_waitqueue_head(&fs_info->transaction_throttle);
2808 init_waitqueue_head(&fs_info->transaction_wait);
2809 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2810 init_waitqueue_head(&fs_info->async_submit_wait);
2811 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2813 /* Usable values until the real ones are cached from the superblock */
2814 fs_info->nodesize = 4096;
2815 fs_info->sectorsize = 4096;
2816 fs_info->sectorsize_bits = ilog2(4096);
2817 fs_info->stripesize = 4096;
2819 spin_lock_init(&fs_info->swapfile_pins_lock);
2820 fs_info->swapfile_pins = RB_ROOT;
2822 fs_info->send_in_progress = 0;
2825 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2830 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2831 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2833 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2837 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2841 fs_info->dirty_metadata_batch = PAGE_SIZE *
2842 (1 + ilog2(nr_cpu_ids));
2844 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2848 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2853 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2855 if (!fs_info->delayed_root)
2857 btrfs_init_delayed_root(fs_info->delayed_root);
2859 return btrfs_alloc_stripe_hash_table(fs_info);
2862 static int btrfs_uuid_rescan_kthread(void *data)
2864 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2868 * 1st step is to iterate through the existing UUID tree and
2869 * to delete all entries that contain outdated data.
2870 * 2nd step is to add all missing entries to the UUID tree.
2872 ret = btrfs_uuid_tree_iterate(fs_info);
2875 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2877 up(&fs_info->uuid_tree_rescan_sem);
2880 return btrfs_uuid_scan_kthread(data);
2883 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2885 struct task_struct *task;
2887 down(&fs_info->uuid_tree_rescan_sem);
2888 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2890 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2891 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2892 up(&fs_info->uuid_tree_rescan_sem);
2893 return PTR_ERR(task);
2899 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2908 struct btrfs_super_block *disk_super;
2909 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2910 struct btrfs_root *tree_root;
2911 struct btrfs_root *chunk_root;
2914 int clear_free_space_tree = 0;
2917 ret = init_mount_fs_info(fs_info, sb);
2923 /* These need to be init'ed before we start creating inodes and such. */
2924 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2926 fs_info->tree_root = tree_root;
2927 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2929 fs_info->chunk_root = chunk_root;
2930 if (!tree_root || !chunk_root) {
2935 fs_info->btree_inode = new_inode(sb);
2936 if (!fs_info->btree_inode) {
2940 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2941 btrfs_init_btree_inode(fs_info);
2943 invalidate_bdev(fs_devices->latest_bdev);
2946 * Read super block and check the signature bytes only
2948 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2949 if (IS_ERR(disk_super)) {
2950 err = PTR_ERR(disk_super);
2955 * Verify the type first, if that or the checksum value are
2956 * corrupted, we'll find out
2958 csum_type = btrfs_super_csum_type(disk_super);
2959 if (!btrfs_supported_super_csum(csum_type)) {
2960 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2963 btrfs_release_disk_super(disk_super);
2967 ret = btrfs_init_csum_hash(fs_info, csum_type);
2970 btrfs_release_disk_super(disk_super);
2975 * We want to check superblock checksum, the type is stored inside.
2976 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2978 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2979 btrfs_err(fs_info, "superblock checksum mismatch");
2981 btrfs_release_disk_super(disk_super);
2986 * super_copy is zeroed at allocation time and we never touch the
2987 * following bytes up to INFO_SIZE, the checksum is calculated from
2988 * the whole block of INFO_SIZE
2990 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2991 btrfs_release_disk_super(disk_super);
2993 disk_super = fs_info->super_copy;
2995 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2998 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2999 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
3000 fs_info->super_copy->metadata_uuid,
3004 features = btrfs_super_flags(disk_super);
3005 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3006 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3007 btrfs_set_super_flags(disk_super, features);
3009 "found metadata UUID change in progress flag, clearing");
3012 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3013 sizeof(*fs_info->super_for_commit));
3015 ret = btrfs_validate_mount_super(fs_info);
3017 btrfs_err(fs_info, "superblock contains fatal errors");
3022 if (!btrfs_super_root(disk_super))
3025 /* check FS state, whether FS is broken. */
3026 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3027 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3030 * In the long term, we'll store the compression type in the super
3031 * block, and it'll be used for per file compression control.
3033 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3035 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3041 features = btrfs_super_incompat_flags(disk_super) &
3042 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3045 "cannot mount because of unsupported optional features (%llx)",
3051 features = btrfs_super_incompat_flags(disk_super);
3052 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3053 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3054 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3055 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3056 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3058 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3059 btrfs_info(fs_info, "has skinny extents");
3062 * flag our filesystem as having big metadata blocks if
3063 * they are bigger than the page size
3065 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3066 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3068 "flagging fs with big metadata feature");
3069 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3072 nodesize = btrfs_super_nodesize(disk_super);
3073 sectorsize = btrfs_super_sectorsize(disk_super);
3074 stripesize = sectorsize;
3075 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3076 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3078 /* Cache block sizes */
3079 fs_info->nodesize = nodesize;
3080 fs_info->sectorsize = sectorsize;
3081 fs_info->sectorsize_bits = ilog2(sectorsize);
3082 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3083 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3084 fs_info->stripesize = stripesize;
3087 * mixed block groups end up with duplicate but slightly offset
3088 * extent buffers for the same range. It leads to corruptions
3090 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3091 (sectorsize != nodesize)) {
3093 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3094 nodesize, sectorsize);
3099 * Needn't use the lock because there is no other task which will
3102 btrfs_set_super_incompat_flags(disk_super, features);
3104 features = btrfs_super_compat_ro_flags(disk_super) &
3105 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3106 if (!sb_rdonly(sb) && features) {
3108 "cannot mount read-write because of unsupported optional features (%llx)",
3114 ret = btrfs_init_workqueues(fs_info, fs_devices);
3117 goto fail_sb_buffer;
3120 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3121 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3123 sb->s_blocksize = sectorsize;
3124 sb->s_blocksize_bits = blksize_bits(sectorsize);
3125 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3127 mutex_lock(&fs_info->chunk_mutex);
3128 ret = btrfs_read_sys_array(fs_info);
3129 mutex_unlock(&fs_info->chunk_mutex);
3131 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3132 goto fail_sb_buffer;
3135 generation = btrfs_super_chunk_root_generation(disk_super);
3136 level = btrfs_super_chunk_root_level(disk_super);
3138 chunk_root->node = read_tree_block(fs_info,
3139 btrfs_super_chunk_root(disk_super),
3140 generation, level, NULL);
3141 if (IS_ERR(chunk_root->node) ||
3142 !extent_buffer_uptodate(chunk_root->node)) {
3143 btrfs_err(fs_info, "failed to read chunk root");
3144 if (!IS_ERR(chunk_root->node))
3145 free_extent_buffer(chunk_root->node);
3146 chunk_root->node = NULL;
3147 goto fail_tree_roots;
3149 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3150 chunk_root->commit_root = btrfs_root_node(chunk_root);
3152 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3153 offsetof(struct btrfs_header, chunk_tree_uuid),
3156 ret = btrfs_read_chunk_tree(fs_info);
3158 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3159 goto fail_tree_roots;
3163 * Keep the devid that is marked to be the target device for the
3164 * device replace procedure
3166 btrfs_free_extra_devids(fs_devices, 0);
3168 if (!fs_devices->latest_bdev) {
3169 btrfs_err(fs_info, "failed to read devices");
3170 goto fail_tree_roots;
3173 ret = init_tree_roots(fs_info);
3175 goto fail_tree_roots;
3178 * If we have a uuid root and we're not being told to rescan we need to
3179 * check the generation here so we can set the
3180 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3181 * transaction during a balance or the log replay without updating the
3182 * uuid generation, and then if we crash we would rescan the uuid tree,
3183 * even though it was perfectly fine.
3185 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3186 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3187 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3189 ret = btrfs_verify_dev_extents(fs_info);
3192 "failed to verify dev extents against chunks: %d",
3194 goto fail_block_groups;
3196 ret = btrfs_recover_balance(fs_info);
3198 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3199 goto fail_block_groups;
3202 ret = btrfs_init_dev_stats(fs_info);
3204 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3205 goto fail_block_groups;
3208 ret = btrfs_init_dev_replace(fs_info);
3210 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3211 goto fail_block_groups;
3214 btrfs_free_extra_devids(fs_devices, 1);
3216 ret = btrfs_sysfs_add_fsid(fs_devices);
3218 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3220 goto fail_block_groups;
3223 ret = btrfs_sysfs_add_mounted(fs_info);
3225 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3226 goto fail_fsdev_sysfs;
3229 ret = btrfs_init_space_info(fs_info);
3231 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3235 ret = btrfs_read_block_groups(fs_info);
3237 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3241 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3243 "writable mount is not allowed due to too many missing devices");
3247 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3249 if (IS_ERR(fs_info->cleaner_kthread))
3252 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3254 "btrfs-transaction");
3255 if (IS_ERR(fs_info->transaction_kthread))
3258 if (!btrfs_test_opt(fs_info, NOSSD) &&
3259 !fs_info->fs_devices->rotating) {
3260 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3264 * Mount does not set all options immediately, we can do it now and do
3265 * not have to wait for transaction commit
3267 btrfs_apply_pending_changes(fs_info);
3269 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3270 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3271 ret = btrfsic_mount(fs_info, fs_devices,
3272 btrfs_test_opt(fs_info,
3273 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3275 fs_info->check_integrity_print_mask);
3278 "failed to initialize integrity check module: %d",
3282 ret = btrfs_read_qgroup_config(fs_info);
3284 goto fail_trans_kthread;
3286 if (btrfs_build_ref_tree(fs_info))
3287 btrfs_err(fs_info, "couldn't build ref tree");
3289 /* do not make disk changes in broken FS or nologreplay is given */
3290 if (btrfs_super_log_root(disk_super) != 0 &&
3291 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3292 btrfs_info(fs_info, "start tree-log replay");
3293 ret = btrfs_replay_log(fs_info, fs_devices);
3300 ret = btrfs_find_orphan_roots(fs_info);
3304 if (!sb_rdonly(sb)) {
3305 ret = btrfs_cleanup_fs_roots(fs_info);
3309 mutex_lock(&fs_info->cleaner_mutex);
3310 ret = btrfs_recover_relocation(tree_root);
3311 mutex_unlock(&fs_info->cleaner_mutex);
3313 btrfs_warn(fs_info, "failed to recover relocation: %d",
3320 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3321 if (IS_ERR(fs_info->fs_root)) {
3322 err = PTR_ERR(fs_info->fs_root);
3323 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3324 fs_info->fs_root = NULL;
3331 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3332 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3333 clear_free_space_tree = 1;
3334 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3335 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3336 btrfs_warn(fs_info, "free space tree is invalid");
3337 clear_free_space_tree = 1;
3340 if (clear_free_space_tree) {
3341 btrfs_info(fs_info, "clearing free space tree");
3342 ret = btrfs_clear_free_space_tree(fs_info);
3345 "failed to clear free space tree: %d", ret);
3346 close_ctree(fs_info);
3351 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3352 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3353 btrfs_info(fs_info, "creating free space tree");
3354 ret = btrfs_create_free_space_tree(fs_info);
3357 "failed to create free space tree: %d", ret);
3358 close_ctree(fs_info);
3363 down_read(&fs_info->cleanup_work_sem);
3364 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3365 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3366 up_read(&fs_info->cleanup_work_sem);
3367 close_ctree(fs_info);
3370 up_read(&fs_info->cleanup_work_sem);
3372 ret = btrfs_resume_balance_async(fs_info);
3374 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3375 close_ctree(fs_info);
3379 ret = btrfs_resume_dev_replace_async(fs_info);
3381 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3382 close_ctree(fs_info);
3386 btrfs_qgroup_rescan_resume(fs_info);
3387 btrfs_discard_resume(fs_info);
3389 if (!fs_info->uuid_root) {
3390 btrfs_info(fs_info, "creating UUID tree");
3391 ret = btrfs_create_uuid_tree(fs_info);
3394 "failed to create the UUID tree: %d", ret);
3395 close_ctree(fs_info);
3398 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3399 fs_info->generation !=
3400 btrfs_super_uuid_tree_generation(disk_super)) {
3401 btrfs_info(fs_info, "checking UUID tree");
3402 ret = btrfs_check_uuid_tree(fs_info);
3405 "failed to check the UUID tree: %d", ret);
3406 close_ctree(fs_info);
3410 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3415 btrfs_free_qgroup_config(fs_info);
3417 kthread_stop(fs_info->transaction_kthread);
3418 btrfs_cleanup_transaction(fs_info);
3419 btrfs_free_fs_roots(fs_info);
3421 kthread_stop(fs_info->cleaner_kthread);
3424 * make sure we're done with the btree inode before we stop our
3427 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3430 btrfs_sysfs_remove_mounted(fs_info);
3433 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3436 btrfs_put_block_group_cache(fs_info);
3439 if (fs_info->data_reloc_root)
3440 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3441 free_root_pointers(fs_info, true);
3442 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3445 btrfs_stop_all_workers(fs_info);
3446 btrfs_free_block_groups(fs_info);
3448 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3450 iput(fs_info->btree_inode);
3452 btrfs_close_devices(fs_info->fs_devices);
3455 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3457 static void btrfs_end_super_write(struct bio *bio)
3459 struct btrfs_device *device = bio->bi_private;
3460 struct bio_vec *bvec;
3461 struct bvec_iter_all iter_all;
3464 bio_for_each_segment_all(bvec, bio, iter_all) {
3465 page = bvec->bv_page;
3467 if (bio->bi_status) {
3468 btrfs_warn_rl_in_rcu(device->fs_info,
3469 "lost page write due to IO error on %s (%d)",
3470 rcu_str_deref(device->name),
3471 blk_status_to_errno(bio->bi_status));
3472 ClearPageUptodate(page);
3474 btrfs_dev_stat_inc_and_print(device,
3475 BTRFS_DEV_STAT_WRITE_ERRS);
3477 SetPageUptodate(page);
3487 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3490 struct btrfs_super_block *super;
3493 struct address_space *mapping = bdev->bd_inode->i_mapping;
3495 bytenr = btrfs_sb_offset(copy_num);
3496 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3497 return ERR_PTR(-EINVAL);
3499 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3501 return ERR_CAST(page);
3503 super = page_address(page);
3504 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3505 btrfs_release_disk_super(super);
3506 return ERR_PTR(-ENODATA);
3509 if (btrfs_super_bytenr(super) != bytenr) {
3510 btrfs_release_disk_super(super);
3511 return ERR_PTR(-EINVAL);
3518 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3520 struct btrfs_super_block *super, *latest = NULL;
3524 /* we would like to check all the supers, but that would make
3525 * a btrfs mount succeed after a mkfs from a different FS.
3526 * So, we need to add a special mount option to scan for
3527 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3529 for (i = 0; i < 1; i++) {
3530 super = btrfs_read_dev_one_super(bdev, i);
3534 if (!latest || btrfs_super_generation(super) > transid) {
3536 btrfs_release_disk_super(super);
3539 transid = btrfs_super_generation(super);
3547 * Write superblock @sb to the @device. Do not wait for completion, all the
3548 * pages we use for writing are locked.
3550 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3551 * the expected device size at commit time. Note that max_mirrors must be
3552 * same for write and wait phases.
3554 * Return number of errors when page is not found or submission fails.
3556 static int write_dev_supers(struct btrfs_device *device,
3557 struct btrfs_super_block *sb, int max_mirrors)
3559 struct btrfs_fs_info *fs_info = device->fs_info;
3560 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3561 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3566 if (max_mirrors == 0)
3567 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3569 shash->tfm = fs_info->csum_shash;
3571 for (i = 0; i < max_mirrors; i++) {
3574 struct btrfs_super_block *disk_super;
3576 bytenr = btrfs_sb_offset(i);
3577 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3578 device->commit_total_bytes)
3581 btrfs_set_super_bytenr(sb, bytenr);
3583 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3584 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3587 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3590 btrfs_err(device->fs_info,
3591 "couldn't get super block page for bytenr %llu",
3597 /* Bump the refcount for wait_dev_supers() */
3600 disk_super = page_address(page);
3601 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3604 * Directly use bios here instead of relying on the page cache
3605 * to do I/O, so we don't lose the ability to do integrity
3608 bio = bio_alloc(GFP_NOFS, 1);
3609 bio_set_dev(bio, device->bdev);
3610 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3611 bio->bi_private = device;
3612 bio->bi_end_io = btrfs_end_super_write;
3613 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3614 offset_in_page(bytenr));
3617 * We FUA only the first super block. The others we allow to
3618 * go down lazy and there's a short window where the on-disk
3619 * copies might still contain the older version.
3621 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3622 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3623 bio->bi_opf |= REQ_FUA;
3625 btrfsic_submit_bio(bio);
3627 return errors < i ? 0 : -1;
3631 * Wait for write completion of superblocks done by write_dev_supers,
3632 * @max_mirrors same for write and wait phases.
3634 * Return number of errors when page is not found or not marked up to
3637 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3641 bool primary_failed = false;
3644 if (max_mirrors == 0)
3645 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3647 for (i = 0; i < max_mirrors; i++) {
3650 bytenr = btrfs_sb_offset(i);
3651 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3652 device->commit_total_bytes)
3655 page = find_get_page(device->bdev->bd_inode->i_mapping,
3656 bytenr >> PAGE_SHIFT);
3660 primary_failed = true;
3663 /* Page is submitted locked and unlocked once the IO completes */
3664 wait_on_page_locked(page);
3665 if (PageError(page)) {
3668 primary_failed = true;
3671 /* Drop our reference */
3674 /* Drop the reference from the writing run */
3678 /* log error, force error return */
3679 if (primary_failed) {
3680 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3685 return errors < i ? 0 : -1;
3689 * endio for the write_dev_flush, this will wake anyone waiting
3690 * for the barrier when it is done
3692 static void btrfs_end_empty_barrier(struct bio *bio)
3694 complete(bio->bi_private);
3698 * Submit a flush request to the device if it supports it. Error handling is
3699 * done in the waiting counterpart.
3701 static void write_dev_flush(struct btrfs_device *device)
3703 struct request_queue *q = bdev_get_queue(device->bdev);
3704 struct bio *bio = device->flush_bio;
3706 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3710 bio->bi_end_io = btrfs_end_empty_barrier;
3711 bio_set_dev(bio, device->bdev);
3712 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3713 init_completion(&device->flush_wait);
3714 bio->bi_private = &device->flush_wait;
3716 btrfsic_submit_bio(bio);
3717 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3721 * If the flush bio has been submitted by write_dev_flush, wait for it.
3723 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3725 struct bio *bio = device->flush_bio;
3727 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3730 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3731 wait_for_completion_io(&device->flush_wait);
3733 return bio->bi_status;
3736 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3738 if (!btrfs_check_rw_degradable(fs_info, NULL))
3744 * send an empty flush down to each device in parallel,
3745 * then wait for them
3747 static int barrier_all_devices(struct btrfs_fs_info *info)
3749 struct list_head *head;
3750 struct btrfs_device *dev;
3751 int errors_wait = 0;
3754 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3755 /* send down all the barriers */
3756 head = &info->fs_devices->devices;
3757 list_for_each_entry(dev, head, dev_list) {
3758 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3762 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3763 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3766 write_dev_flush(dev);
3767 dev->last_flush_error = BLK_STS_OK;
3770 /* wait for all the barriers */
3771 list_for_each_entry(dev, head, dev_list) {
3772 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3778 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3779 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3782 ret = wait_dev_flush(dev);
3784 dev->last_flush_error = ret;
3785 btrfs_dev_stat_inc_and_print(dev,
3786 BTRFS_DEV_STAT_FLUSH_ERRS);
3793 * At some point we need the status of all disks
3794 * to arrive at the volume status. So error checking
3795 * is being pushed to a separate loop.
3797 return check_barrier_error(info);
3802 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3805 int min_tolerated = INT_MAX;
3807 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3808 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3809 min_tolerated = min_t(int, min_tolerated,
3810 btrfs_raid_array[BTRFS_RAID_SINGLE].
3811 tolerated_failures);
3813 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3814 if (raid_type == BTRFS_RAID_SINGLE)
3816 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3818 min_tolerated = min_t(int, min_tolerated,
3819 btrfs_raid_array[raid_type].
3820 tolerated_failures);
3823 if (min_tolerated == INT_MAX) {
3824 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3828 return min_tolerated;
3831 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3833 struct list_head *head;
3834 struct btrfs_device *dev;
3835 struct btrfs_super_block *sb;
3836 struct btrfs_dev_item *dev_item;
3840 int total_errors = 0;
3843 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3846 * max_mirrors == 0 indicates we're from commit_transaction,
3847 * not from fsync where the tree roots in fs_info have not
3848 * been consistent on disk.
3850 if (max_mirrors == 0)
3851 backup_super_roots(fs_info);
3853 sb = fs_info->super_for_commit;
3854 dev_item = &sb->dev_item;
3856 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3857 head = &fs_info->fs_devices->devices;
3858 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3861 ret = barrier_all_devices(fs_info);
3864 &fs_info->fs_devices->device_list_mutex);
3865 btrfs_handle_fs_error(fs_info, ret,
3866 "errors while submitting device barriers.");
3871 list_for_each_entry(dev, head, dev_list) {
3876 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3877 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3880 btrfs_set_stack_device_generation(dev_item, 0);
3881 btrfs_set_stack_device_type(dev_item, dev->type);
3882 btrfs_set_stack_device_id(dev_item, dev->devid);
3883 btrfs_set_stack_device_total_bytes(dev_item,
3884 dev->commit_total_bytes);
3885 btrfs_set_stack_device_bytes_used(dev_item,
3886 dev->commit_bytes_used);
3887 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3888 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3889 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3890 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3891 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3894 flags = btrfs_super_flags(sb);
3895 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3897 ret = btrfs_validate_write_super(fs_info, sb);
3899 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3900 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3901 "unexpected superblock corruption detected");
3905 ret = write_dev_supers(dev, sb, max_mirrors);
3909 if (total_errors > max_errors) {
3910 btrfs_err(fs_info, "%d errors while writing supers",
3912 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3914 /* FUA is masked off if unsupported and can't be the reason */
3915 btrfs_handle_fs_error(fs_info, -EIO,
3916 "%d errors while writing supers",
3922 list_for_each_entry(dev, head, dev_list) {
3925 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3926 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3929 ret = wait_dev_supers(dev, max_mirrors);
3933 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3934 if (total_errors > max_errors) {
3935 btrfs_handle_fs_error(fs_info, -EIO,
3936 "%d errors while writing supers",
3943 /* Drop a fs root from the radix tree and free it. */
3944 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3945 struct btrfs_root *root)
3947 bool drop_ref = false;
3949 spin_lock(&fs_info->fs_roots_radix_lock);
3950 radix_tree_delete(&fs_info->fs_roots_radix,
3951 (unsigned long)root->root_key.objectid);
3952 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3954 spin_unlock(&fs_info->fs_roots_radix_lock);
3956 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3957 ASSERT(root->log_root == NULL);
3958 if (root->reloc_root) {
3959 btrfs_put_root(root->reloc_root);
3960 root->reloc_root = NULL;
3964 if (root->free_ino_pinned)
3965 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3966 if (root->free_ino_ctl)
3967 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3968 if (root->ino_cache_inode) {
3969 iput(root->ino_cache_inode);
3970 root->ino_cache_inode = NULL;
3973 btrfs_put_root(root);
3976 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3978 u64 root_objectid = 0;
3979 struct btrfs_root *gang[8];
3982 unsigned int ret = 0;
3985 spin_lock(&fs_info->fs_roots_radix_lock);
3986 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3987 (void **)gang, root_objectid,
3990 spin_unlock(&fs_info->fs_roots_radix_lock);
3993 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3995 for (i = 0; i < ret; i++) {
3996 /* Avoid to grab roots in dead_roots */
3997 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4001 /* grab all the search result for later use */
4002 gang[i] = btrfs_grab_root(gang[i]);
4004 spin_unlock(&fs_info->fs_roots_radix_lock);
4006 for (i = 0; i < ret; i++) {
4009 root_objectid = gang[i]->root_key.objectid;
4010 err = btrfs_orphan_cleanup(gang[i]);
4013 btrfs_put_root(gang[i]);
4018 /* release the uncleaned roots due to error */
4019 for (; i < ret; i++) {
4021 btrfs_put_root(gang[i]);
4026 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4028 struct btrfs_root *root = fs_info->tree_root;
4029 struct btrfs_trans_handle *trans;
4031 mutex_lock(&fs_info->cleaner_mutex);
4032 btrfs_run_delayed_iputs(fs_info);
4033 mutex_unlock(&fs_info->cleaner_mutex);
4034 wake_up_process(fs_info->cleaner_kthread);
4036 /* wait until ongoing cleanup work done */
4037 down_write(&fs_info->cleanup_work_sem);
4038 up_write(&fs_info->cleanup_work_sem);
4040 trans = btrfs_join_transaction(root);
4042 return PTR_ERR(trans);
4043 return btrfs_commit_transaction(trans);
4046 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4050 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4052 * We don't want the cleaner to start new transactions, add more delayed
4053 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4054 * because that frees the task_struct, and the transaction kthread might
4055 * still try to wake up the cleaner.
4057 kthread_park(fs_info->cleaner_kthread);
4059 /* wait for the qgroup rescan worker to stop */
4060 btrfs_qgroup_wait_for_completion(fs_info, false);
4062 /* wait for the uuid_scan task to finish */
4063 down(&fs_info->uuid_tree_rescan_sem);
4064 /* avoid complains from lockdep et al., set sem back to initial state */
4065 up(&fs_info->uuid_tree_rescan_sem);
4067 /* pause restriper - we want to resume on mount */
4068 btrfs_pause_balance(fs_info);
4070 btrfs_dev_replace_suspend_for_unmount(fs_info);
4072 btrfs_scrub_cancel(fs_info);
4074 /* wait for any defraggers to finish */
4075 wait_event(fs_info->transaction_wait,
4076 (atomic_read(&fs_info->defrag_running) == 0));
4078 /* clear out the rbtree of defraggable inodes */
4079 btrfs_cleanup_defrag_inodes(fs_info);
4081 cancel_work_sync(&fs_info->async_reclaim_work);
4082 cancel_work_sync(&fs_info->async_data_reclaim_work);
4084 /* Cancel or finish ongoing discard work */
4085 btrfs_discard_cleanup(fs_info);
4087 if (!sb_rdonly(fs_info->sb)) {
4089 * The cleaner kthread is stopped, so do one final pass over
4090 * unused block groups.
4092 btrfs_delete_unused_bgs(fs_info);
4095 * There might be existing delayed inode workers still running
4096 * and holding an empty delayed inode item. We must wait for
4097 * them to complete first because they can create a transaction.
4098 * This happens when someone calls btrfs_balance_delayed_items()
4099 * and then a transaction commit runs the same delayed nodes
4100 * before any delayed worker has done something with the nodes.
4101 * We must wait for any worker here and not at transaction
4102 * commit time since that could cause a deadlock.
4103 * This is a very rare case.
4105 btrfs_flush_workqueue(fs_info->delayed_workers);
4107 ret = btrfs_commit_super(fs_info);
4109 btrfs_err(fs_info, "commit super ret %d", ret);
4112 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4113 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4114 btrfs_error_commit_super(fs_info);
4116 kthread_stop(fs_info->transaction_kthread);
4117 kthread_stop(fs_info->cleaner_kthread);
4119 ASSERT(list_empty(&fs_info->delayed_iputs));
4120 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4122 if (btrfs_check_quota_leak(fs_info)) {
4123 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4124 btrfs_err(fs_info, "qgroup reserved space leaked");
4127 btrfs_free_qgroup_config(fs_info);
4128 ASSERT(list_empty(&fs_info->delalloc_roots));
4130 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4131 btrfs_info(fs_info, "at unmount delalloc count %lld",
4132 percpu_counter_sum(&fs_info->delalloc_bytes));
4135 if (percpu_counter_sum(&fs_info->dio_bytes))
4136 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4137 percpu_counter_sum(&fs_info->dio_bytes));
4139 btrfs_sysfs_remove_mounted(fs_info);
4140 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4142 btrfs_put_block_group_cache(fs_info);
4145 * we must make sure there is not any read request to
4146 * submit after we stopping all workers.
4148 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4149 btrfs_stop_all_workers(fs_info);
4151 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4152 free_root_pointers(fs_info, true);
4153 btrfs_free_fs_roots(fs_info);
4156 * We must free the block groups after dropping the fs_roots as we could
4157 * have had an IO error and have left over tree log blocks that aren't
4158 * cleaned up until the fs roots are freed. This makes the block group
4159 * accounting appear to be wrong because there's pending reserved bytes,
4160 * so make sure we do the block group cleanup afterwards.
4162 btrfs_free_block_groups(fs_info);
4164 iput(fs_info->btree_inode);
4166 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4167 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4168 btrfsic_unmount(fs_info->fs_devices);
4171 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4172 btrfs_close_devices(fs_info->fs_devices);
4175 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4179 struct inode *btree_inode = buf->pages[0]->mapping->host;
4181 ret = extent_buffer_uptodate(buf);
4185 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4186 parent_transid, atomic);
4192 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4194 struct btrfs_fs_info *fs_info;
4195 struct btrfs_root *root;
4196 u64 transid = btrfs_header_generation(buf);
4199 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4201 * This is a fast path so only do this check if we have sanity tests
4202 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4203 * outside of the sanity tests.
4205 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4208 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4209 fs_info = root->fs_info;
4210 btrfs_assert_tree_locked(buf);
4211 if (transid != fs_info->generation)
4212 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4213 buf->start, transid, fs_info->generation);
4214 was_dirty = set_extent_buffer_dirty(buf);
4216 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4218 fs_info->dirty_metadata_batch);
4219 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4221 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4222 * but item data not updated.
4223 * So here we should only check item pointers, not item data.
4225 if (btrfs_header_level(buf) == 0 &&
4226 btrfs_check_leaf_relaxed(buf)) {
4227 btrfs_print_leaf(buf);
4233 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4237 * looks as though older kernels can get into trouble with
4238 * this code, they end up stuck in balance_dirty_pages forever
4242 if (current->flags & PF_MEMALLOC)
4246 btrfs_balance_delayed_items(fs_info);
4248 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4249 BTRFS_DIRTY_METADATA_THRESH,
4250 fs_info->dirty_metadata_batch);
4252 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4256 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4258 __btrfs_btree_balance_dirty(fs_info, 1);
4261 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4263 __btrfs_btree_balance_dirty(fs_info, 0);
4266 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4267 struct btrfs_key *first_key)
4269 return btree_read_extent_buffer_pages(buf, parent_transid,
4273 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4275 /* cleanup FS via transaction */
4276 btrfs_cleanup_transaction(fs_info);
4278 mutex_lock(&fs_info->cleaner_mutex);
4279 btrfs_run_delayed_iputs(fs_info);
4280 mutex_unlock(&fs_info->cleaner_mutex);
4282 down_write(&fs_info->cleanup_work_sem);
4283 up_write(&fs_info->cleanup_work_sem);
4286 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4288 struct btrfs_root *gang[8];
4289 u64 root_objectid = 0;
4292 spin_lock(&fs_info->fs_roots_radix_lock);
4293 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4294 (void **)gang, root_objectid,
4295 ARRAY_SIZE(gang))) != 0) {
4298 for (i = 0; i < ret; i++)
4299 gang[i] = btrfs_grab_root(gang[i]);
4300 spin_unlock(&fs_info->fs_roots_radix_lock);
4302 for (i = 0; i < ret; i++) {
4305 root_objectid = gang[i]->root_key.objectid;
4306 btrfs_free_log(NULL, gang[i]);
4307 btrfs_put_root(gang[i]);
4310 spin_lock(&fs_info->fs_roots_radix_lock);
4312 spin_unlock(&fs_info->fs_roots_radix_lock);
4313 btrfs_free_log_root_tree(NULL, fs_info);
4316 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4318 struct btrfs_ordered_extent *ordered;
4320 spin_lock(&root->ordered_extent_lock);
4322 * This will just short circuit the ordered completion stuff which will
4323 * make sure the ordered extent gets properly cleaned up.
4325 list_for_each_entry(ordered, &root->ordered_extents,
4327 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4328 spin_unlock(&root->ordered_extent_lock);
4331 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4333 struct btrfs_root *root;
4334 struct list_head splice;
4336 INIT_LIST_HEAD(&splice);
4338 spin_lock(&fs_info->ordered_root_lock);
4339 list_splice_init(&fs_info->ordered_roots, &splice);
4340 while (!list_empty(&splice)) {
4341 root = list_first_entry(&splice, struct btrfs_root,
4343 list_move_tail(&root->ordered_root,
4344 &fs_info->ordered_roots);
4346 spin_unlock(&fs_info->ordered_root_lock);
4347 btrfs_destroy_ordered_extents(root);
4350 spin_lock(&fs_info->ordered_root_lock);
4352 spin_unlock(&fs_info->ordered_root_lock);
4355 * We need this here because if we've been flipped read-only we won't
4356 * get sync() from the umount, so we need to make sure any ordered
4357 * extents that haven't had their dirty pages IO start writeout yet
4358 * actually get run and error out properly.
4360 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4363 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4364 struct btrfs_fs_info *fs_info)
4366 struct rb_node *node;
4367 struct btrfs_delayed_ref_root *delayed_refs;
4368 struct btrfs_delayed_ref_node *ref;
4371 delayed_refs = &trans->delayed_refs;
4373 spin_lock(&delayed_refs->lock);
4374 if (atomic_read(&delayed_refs->num_entries) == 0) {
4375 spin_unlock(&delayed_refs->lock);
4376 btrfs_debug(fs_info, "delayed_refs has NO entry");
4380 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4381 struct btrfs_delayed_ref_head *head;
4383 bool pin_bytes = false;
4385 head = rb_entry(node, struct btrfs_delayed_ref_head,
4387 if (btrfs_delayed_ref_lock(delayed_refs, head))
4390 spin_lock(&head->lock);
4391 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4392 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4395 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4396 RB_CLEAR_NODE(&ref->ref_node);
4397 if (!list_empty(&ref->add_list))
4398 list_del(&ref->add_list);
4399 atomic_dec(&delayed_refs->num_entries);
4400 btrfs_put_delayed_ref(ref);
4402 if (head->must_insert_reserved)
4404 btrfs_free_delayed_extent_op(head->extent_op);
4405 btrfs_delete_ref_head(delayed_refs, head);
4406 spin_unlock(&head->lock);
4407 spin_unlock(&delayed_refs->lock);
4408 mutex_unlock(&head->mutex);
4411 struct btrfs_block_group *cache;
4413 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4416 spin_lock(&cache->space_info->lock);
4417 spin_lock(&cache->lock);
4418 cache->pinned += head->num_bytes;
4419 btrfs_space_info_update_bytes_pinned(fs_info,
4420 cache->space_info, head->num_bytes);
4421 cache->reserved -= head->num_bytes;
4422 cache->space_info->bytes_reserved -= head->num_bytes;
4423 spin_unlock(&cache->lock);
4424 spin_unlock(&cache->space_info->lock);
4425 percpu_counter_add_batch(
4426 &cache->space_info->total_bytes_pinned,
4427 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4429 btrfs_put_block_group(cache);
4431 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4432 head->bytenr + head->num_bytes - 1);
4434 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4435 btrfs_put_delayed_ref_head(head);
4437 spin_lock(&delayed_refs->lock);
4439 btrfs_qgroup_destroy_extent_records(trans);
4441 spin_unlock(&delayed_refs->lock);
4446 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4448 struct btrfs_inode *btrfs_inode;
4449 struct list_head splice;
4451 INIT_LIST_HEAD(&splice);
4453 spin_lock(&root->delalloc_lock);
4454 list_splice_init(&root->delalloc_inodes, &splice);
4456 while (!list_empty(&splice)) {
4457 struct inode *inode = NULL;
4458 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4460 __btrfs_del_delalloc_inode(root, btrfs_inode);
4461 spin_unlock(&root->delalloc_lock);
4464 * Make sure we get a live inode and that it'll not disappear
4467 inode = igrab(&btrfs_inode->vfs_inode);
4469 invalidate_inode_pages2(inode->i_mapping);
4472 spin_lock(&root->delalloc_lock);
4474 spin_unlock(&root->delalloc_lock);
4477 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4479 struct btrfs_root *root;
4480 struct list_head splice;
4482 INIT_LIST_HEAD(&splice);
4484 spin_lock(&fs_info->delalloc_root_lock);
4485 list_splice_init(&fs_info->delalloc_roots, &splice);
4486 while (!list_empty(&splice)) {
4487 root = list_first_entry(&splice, struct btrfs_root,
4489 root = btrfs_grab_root(root);
4491 spin_unlock(&fs_info->delalloc_root_lock);
4493 btrfs_destroy_delalloc_inodes(root);
4494 btrfs_put_root(root);
4496 spin_lock(&fs_info->delalloc_root_lock);
4498 spin_unlock(&fs_info->delalloc_root_lock);
4501 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4502 struct extent_io_tree *dirty_pages,
4506 struct extent_buffer *eb;
4511 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4516 clear_extent_bits(dirty_pages, start, end, mark);
4517 while (start <= end) {
4518 eb = find_extent_buffer(fs_info, start);
4519 start += fs_info->nodesize;
4522 wait_on_extent_buffer_writeback(eb);
4524 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4526 clear_extent_buffer_dirty(eb);
4527 free_extent_buffer_stale(eb);
4534 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4535 struct extent_io_tree *unpin)
4542 struct extent_state *cached_state = NULL;
4545 * The btrfs_finish_extent_commit() may get the same range as
4546 * ours between find_first_extent_bit and clear_extent_dirty.
4547 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4548 * the same extent range.
4550 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4551 ret = find_first_extent_bit(unpin, 0, &start, &end,
4552 EXTENT_DIRTY, &cached_state);
4554 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4558 clear_extent_dirty(unpin, start, end, &cached_state);
4559 free_extent_state(cached_state);
4560 btrfs_error_unpin_extent_range(fs_info, start, end);
4561 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4568 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4570 struct inode *inode;
4572 inode = cache->io_ctl.inode;
4574 invalidate_inode_pages2(inode->i_mapping);
4575 BTRFS_I(inode)->generation = 0;
4576 cache->io_ctl.inode = NULL;
4579 ASSERT(cache->io_ctl.pages == NULL);
4580 btrfs_put_block_group(cache);
4583 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4584 struct btrfs_fs_info *fs_info)
4586 struct btrfs_block_group *cache;
4588 spin_lock(&cur_trans->dirty_bgs_lock);
4589 while (!list_empty(&cur_trans->dirty_bgs)) {
4590 cache = list_first_entry(&cur_trans->dirty_bgs,
4591 struct btrfs_block_group,
4594 if (!list_empty(&cache->io_list)) {
4595 spin_unlock(&cur_trans->dirty_bgs_lock);
4596 list_del_init(&cache->io_list);
4597 btrfs_cleanup_bg_io(cache);
4598 spin_lock(&cur_trans->dirty_bgs_lock);
4601 list_del_init(&cache->dirty_list);
4602 spin_lock(&cache->lock);
4603 cache->disk_cache_state = BTRFS_DC_ERROR;
4604 spin_unlock(&cache->lock);
4606 spin_unlock(&cur_trans->dirty_bgs_lock);
4607 btrfs_put_block_group(cache);
4608 btrfs_delayed_refs_rsv_release(fs_info, 1);
4609 spin_lock(&cur_trans->dirty_bgs_lock);
4611 spin_unlock(&cur_trans->dirty_bgs_lock);
4614 * Refer to the definition of io_bgs member for details why it's safe
4615 * to use it without any locking
4617 while (!list_empty(&cur_trans->io_bgs)) {
4618 cache = list_first_entry(&cur_trans->io_bgs,
4619 struct btrfs_block_group,
4622 list_del_init(&cache->io_list);
4623 spin_lock(&cache->lock);
4624 cache->disk_cache_state = BTRFS_DC_ERROR;
4625 spin_unlock(&cache->lock);
4626 btrfs_cleanup_bg_io(cache);
4630 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4631 struct btrfs_fs_info *fs_info)
4633 struct btrfs_device *dev, *tmp;
4635 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4636 ASSERT(list_empty(&cur_trans->dirty_bgs));
4637 ASSERT(list_empty(&cur_trans->io_bgs));
4639 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4641 list_del_init(&dev->post_commit_list);
4644 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4646 cur_trans->state = TRANS_STATE_COMMIT_START;
4647 wake_up(&fs_info->transaction_blocked_wait);
4649 cur_trans->state = TRANS_STATE_UNBLOCKED;
4650 wake_up(&fs_info->transaction_wait);
4652 btrfs_destroy_delayed_inodes(fs_info);
4654 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4656 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4658 cur_trans->state =TRANS_STATE_COMPLETED;
4659 wake_up(&cur_trans->commit_wait);
4662 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4664 struct btrfs_transaction *t;
4666 mutex_lock(&fs_info->transaction_kthread_mutex);
4668 spin_lock(&fs_info->trans_lock);
4669 while (!list_empty(&fs_info->trans_list)) {
4670 t = list_first_entry(&fs_info->trans_list,
4671 struct btrfs_transaction, list);
4672 if (t->state >= TRANS_STATE_COMMIT_START) {
4673 refcount_inc(&t->use_count);
4674 spin_unlock(&fs_info->trans_lock);
4675 btrfs_wait_for_commit(fs_info, t->transid);
4676 btrfs_put_transaction(t);
4677 spin_lock(&fs_info->trans_lock);
4680 if (t == fs_info->running_transaction) {
4681 t->state = TRANS_STATE_COMMIT_DOING;
4682 spin_unlock(&fs_info->trans_lock);
4684 * We wait for 0 num_writers since we don't hold a trans
4685 * handle open currently for this transaction.
4687 wait_event(t->writer_wait,
4688 atomic_read(&t->num_writers) == 0);
4690 spin_unlock(&fs_info->trans_lock);
4692 btrfs_cleanup_one_transaction(t, fs_info);
4694 spin_lock(&fs_info->trans_lock);
4695 if (t == fs_info->running_transaction)
4696 fs_info->running_transaction = NULL;
4697 list_del_init(&t->list);
4698 spin_unlock(&fs_info->trans_lock);
4700 btrfs_put_transaction(t);
4701 trace_btrfs_transaction_commit(fs_info->tree_root);
4702 spin_lock(&fs_info->trans_lock);
4704 spin_unlock(&fs_info->trans_lock);
4705 btrfs_destroy_all_ordered_extents(fs_info);
4706 btrfs_destroy_delayed_inodes(fs_info);
4707 btrfs_assert_delayed_root_empty(fs_info);
4708 btrfs_destroy_all_delalloc_inodes(fs_info);
4709 btrfs_drop_all_logs(fs_info);
4710 mutex_unlock(&fs_info->transaction_kthread_mutex);