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 "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41 #include "block-group.h"
43 #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_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
177 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
178 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
179 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
180 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
181 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
182 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
183 { .id = 0, DEFINE_NAME("tree") },
189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
192 struct btrfs_lockdep_keyset *ks;
194 BUG_ON(level >= ARRAY_SIZE(ks->keys));
196 /* find the matching keyset, id 0 is the default entry */
197 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
198 if (ks->id == objectid)
201 lockdep_set_class_and_name(&eb->lock,
202 &ks->keys[level], ks->names[level]);
208 * Compute the csum of a btree block and store the result to provided buffer.
210 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
212 struct btrfs_fs_info *fs_info = buf->fs_info;
213 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
214 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
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]) + offset_in_page(buf->start);
222 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
223 first_page_part - 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);
256 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
258 if (extent_buffer_uptodate(eb) &&
259 btrfs_header_generation(eb) == parent_transid) {
263 btrfs_err_rl(eb->fs_info,
264 "parent transid verify failed on %llu wanted %llu found %llu",
266 parent_transid, btrfs_header_generation(eb));
270 * Things reading via commit roots that don't have normal protection,
271 * like send, can have a really old block in cache that may point at a
272 * block that has been freed and re-allocated. So don't clear uptodate
273 * if we find an eb that is under IO (dirty/writeback) because we could
274 * end up reading in the stale data and then writing it back out and
275 * making everybody very sad.
277 if (!extent_buffer_under_io(eb))
278 clear_extent_buffer_uptodate(eb);
280 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
283 btrfs_tree_read_unlock(eb);
287 static bool btrfs_supported_super_csum(u16 csum_type)
290 case BTRFS_CSUM_TYPE_CRC32:
291 case BTRFS_CSUM_TYPE_XXHASH:
292 case BTRFS_CSUM_TYPE_SHA256:
293 case BTRFS_CSUM_TYPE_BLAKE2:
301 * Return 0 if the superblock checksum type matches the checksum value of that
302 * algorithm. Pass the raw disk superblock data.
304 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
307 struct btrfs_super_block *disk_sb =
308 (struct btrfs_super_block *)raw_disk_sb;
309 char result[BTRFS_CSUM_SIZE];
310 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
312 shash->tfm = fs_info->csum_shash;
315 * The super_block structure does not span the whole
316 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
317 * filled with zeros and is included in the checksum.
319 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
320 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
322 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
328 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
329 struct btrfs_key *first_key, u64 parent_transid)
331 struct btrfs_fs_info *fs_info = eb->fs_info;
333 struct btrfs_key found_key;
336 found_level = btrfs_header_level(eb);
337 if (found_level != level) {
338 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
339 KERN_ERR "BTRFS: tree level check failed\n");
341 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
342 eb->start, level, found_level);
350 * For live tree block (new tree blocks in current transaction),
351 * we need proper lock context to avoid race, which is impossible here.
352 * So we only checks tree blocks which is read from disk, whose
353 * generation <= fs_info->last_trans_committed.
355 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
358 /* We have @first_key, so this @eb must have at least one item */
359 if (btrfs_header_nritems(eb) == 0) {
361 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
363 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
368 btrfs_node_key_to_cpu(eb, &found_key, 0);
370 btrfs_item_key_to_cpu(eb, &found_key, 0);
371 ret = btrfs_comp_cpu_keys(first_key, &found_key);
374 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
375 KERN_ERR "BTRFS: tree first key check failed\n");
377 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
378 eb->start, parent_transid, first_key->objectid,
379 first_key->type, first_key->offset,
380 found_key.objectid, found_key.type,
387 * helper to read a given tree block, doing retries as required when
388 * the checksums don't match and we have alternate mirrors to try.
390 * @parent_transid: expected transid, skip check if 0
391 * @level: expected level, mandatory check
392 * @first_key: expected key of first slot, skip check if NULL
394 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
395 u64 parent_transid, int level,
396 struct btrfs_key *first_key)
398 struct btrfs_fs_info *fs_info = eb->fs_info;
399 struct extent_io_tree *io_tree;
404 int failed_mirror = 0;
406 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
408 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
409 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
411 if (verify_parent_transid(io_tree, eb,
414 else if (btrfs_verify_level_key(eb, level,
415 first_key, parent_transid))
421 num_copies = btrfs_num_copies(fs_info,
426 if (!failed_mirror) {
428 failed_mirror = eb->read_mirror;
432 if (mirror_num == failed_mirror)
435 if (mirror_num > num_copies)
439 if (failed && !ret && failed_mirror)
440 btrfs_repair_eb_io_failure(eb, failed_mirror);
446 * Checksum a dirty tree block before IO. This has extra checks to make sure
447 * we only fill in the checksum field in the first page of a multi-page block.
448 * For subpage extent buffers we need bvec to also read the offset in the page.
450 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
452 struct page *page = bvec->bv_page;
453 u64 start = page_offset(page);
455 u8 result[BTRFS_CSUM_SIZE];
456 struct extent_buffer *eb;
459 eb = (struct extent_buffer *)page->private;
460 if (page != eb->pages[0])
463 found_start = btrfs_header_bytenr(eb);
465 * Please do not consolidate these warnings into a single if.
466 * It is useful to know what went wrong.
468 if (WARN_ON(found_start != start))
470 if (WARN_ON(!PageUptodate(page)))
473 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
474 offsetof(struct btrfs_header, fsid),
475 BTRFS_FSID_SIZE) == 0);
477 csum_tree_block(eb, result);
479 if (btrfs_header_level(eb))
480 ret = btrfs_check_node(eb);
482 ret = btrfs_check_leaf_full(eb);
485 btrfs_print_tree(eb, 0);
487 "block=%llu write time tree block corruption detected",
489 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
492 write_extent_buffer(eb, result, 0, fs_info->csum_size);
497 static int check_tree_block_fsid(struct extent_buffer *eb)
499 struct btrfs_fs_info *fs_info = eb->fs_info;
500 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
501 u8 fsid[BTRFS_FSID_SIZE];
504 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
507 * Checking the incompat flag is only valid for the current fs. For
508 * seed devices it's forbidden to have their uuid changed so reading
509 * ->fsid in this case is fine
511 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
512 metadata_uuid = fs_devices->metadata_uuid;
514 metadata_uuid = fs_devices->fsid;
516 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
519 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
520 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
526 /* Do basic extent buffer checks at read time */
527 static int validate_extent_buffer(struct extent_buffer *eb)
529 struct btrfs_fs_info *fs_info = eb->fs_info;
531 const u32 csum_size = fs_info->csum_size;
533 u8 result[BTRFS_CSUM_SIZE];
536 found_start = btrfs_header_bytenr(eb);
537 if (found_start != eb->start) {
538 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
539 eb->start, found_start);
543 if (check_tree_block_fsid(eb)) {
544 btrfs_err_rl(fs_info, "bad fsid on block %llu",
549 found_level = btrfs_header_level(eb);
550 if (found_level >= BTRFS_MAX_LEVEL) {
551 btrfs_err(fs_info, "bad tree block level %d on %llu",
552 (int)btrfs_header_level(eb), eb->start);
557 csum_tree_block(eb, result);
559 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
560 u8 val[BTRFS_CSUM_SIZE] = { 0 };
562 read_extent_buffer(eb, &val, 0, csum_size);
563 btrfs_warn_rl(fs_info,
564 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
565 fs_info->sb->s_id, eb->start,
566 CSUM_FMT_VALUE(csum_size, val),
567 CSUM_FMT_VALUE(csum_size, result),
568 btrfs_header_level(eb));
574 * If this is a leaf block and it is corrupt, set the corrupt bit so
575 * that we don't try and read the other copies of this block, just
578 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
579 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
583 if (found_level > 0 && btrfs_check_node(eb))
587 set_extent_buffer_uptodate(eb);
590 "block=%llu read time tree block corruption detected",
596 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
597 struct page *page, u64 start, u64 end,
600 struct extent_buffer *eb;
604 ASSERT(page->private);
605 eb = (struct extent_buffer *)page->private;
608 * The pending IO might have been the only thing that kept this buffer
609 * in memory. Make sure we have a ref for all this other checks
611 atomic_inc(&eb->refs);
613 reads_done = atomic_dec_and_test(&eb->io_pages);
617 eb->read_mirror = mirror;
618 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
622 ret = validate_extent_buffer(eb);
625 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
626 btree_readahead_hook(eb, ret);
630 * our io error hook is going to dec the io pages
631 * again, we have to make sure it has something
634 atomic_inc(&eb->io_pages);
635 clear_extent_buffer_uptodate(eb);
637 free_extent_buffer(eb);
642 static void end_workqueue_bio(struct bio *bio)
644 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
645 struct btrfs_fs_info *fs_info;
646 struct btrfs_workqueue *wq;
648 fs_info = end_io_wq->info;
649 end_io_wq->status = bio->bi_status;
651 if (bio_op(bio) == REQ_OP_WRITE) {
652 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
653 wq = fs_info->endio_meta_write_workers;
654 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
655 wq = fs_info->endio_freespace_worker;
656 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
657 wq = fs_info->endio_raid56_workers;
659 wq = fs_info->endio_write_workers;
661 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
662 wq = fs_info->endio_raid56_workers;
663 else if (end_io_wq->metadata)
664 wq = fs_info->endio_meta_workers;
666 wq = fs_info->endio_workers;
669 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
670 btrfs_queue_work(wq, &end_io_wq->work);
673 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
674 enum btrfs_wq_endio_type metadata)
676 struct btrfs_end_io_wq *end_io_wq;
678 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
680 return BLK_STS_RESOURCE;
682 end_io_wq->private = bio->bi_private;
683 end_io_wq->end_io = bio->bi_end_io;
684 end_io_wq->info = info;
685 end_io_wq->status = 0;
686 end_io_wq->bio = bio;
687 end_io_wq->metadata = metadata;
689 bio->bi_private = end_io_wq;
690 bio->bi_end_io = end_workqueue_bio;
694 static void run_one_async_start(struct btrfs_work *work)
696 struct async_submit_bio *async;
699 async = container_of(work, struct async_submit_bio, work);
700 ret = async->submit_bio_start(async->inode, async->bio, async->bio_offset);
706 * In order to insert checksums into the metadata in large chunks, we wait
707 * until bio submission time. All the pages in the bio are checksummed and
708 * sums are attached onto the ordered extent record.
710 * At IO completion time the csums attached on the ordered extent record are
711 * inserted into the tree.
713 static void run_one_async_done(struct btrfs_work *work)
715 struct async_submit_bio *async;
719 async = container_of(work, struct async_submit_bio, work);
720 inode = async->inode;
722 /* If an error occurred we just want to clean up the bio and move on */
724 async->bio->bi_status = async->status;
725 bio_endio(async->bio);
730 * All of the bios that pass through here are from async helpers.
731 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
732 * This changes nothing when cgroups aren't in use.
734 async->bio->bi_opf |= REQ_CGROUP_PUNT;
735 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
737 async->bio->bi_status = ret;
738 bio_endio(async->bio);
742 static void run_one_async_free(struct btrfs_work *work)
744 struct async_submit_bio *async;
746 async = container_of(work, struct async_submit_bio, work);
750 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
751 int mirror_num, unsigned long bio_flags,
753 extent_submit_bio_start_t *submit_bio_start)
755 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
756 struct async_submit_bio *async;
758 async = kmalloc(sizeof(*async), GFP_NOFS);
760 return BLK_STS_RESOURCE;
762 async->inode = inode;
764 async->mirror_num = mirror_num;
765 async->submit_bio_start = submit_bio_start;
767 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
770 async->bio_offset = bio_offset;
774 if (op_is_sync(bio->bi_opf))
775 btrfs_set_work_high_priority(&async->work);
777 btrfs_queue_work(fs_info->workers, &async->work);
781 static blk_status_t btree_csum_one_bio(struct bio *bio)
783 struct bio_vec *bvec;
784 struct btrfs_root *root;
786 struct bvec_iter_all iter_all;
788 ASSERT(!bio_flagged(bio, BIO_CLONED));
789 bio_for_each_segment_all(bvec, bio, iter_all) {
790 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
791 ret = csum_dirty_buffer(root->fs_info, bvec);
796 return errno_to_blk_status(ret);
799 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
803 * when we're called for a write, we're already in the async
804 * submission context. Just jump into btrfs_map_bio
806 return btree_csum_one_bio(bio);
809 static int check_async_write(struct btrfs_fs_info *fs_info,
810 struct btrfs_inode *bi)
812 if (atomic_read(&bi->sync_writers))
814 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
819 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
820 int mirror_num, unsigned long bio_flags)
822 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
823 int async = check_async_write(fs_info, BTRFS_I(inode));
826 if (bio_op(bio) != REQ_OP_WRITE) {
828 * called for a read, do the setup so that checksum validation
829 * can happen in the async kernel threads
831 ret = btrfs_bio_wq_end_io(fs_info, bio,
832 BTRFS_WQ_ENDIO_METADATA);
835 ret = btrfs_map_bio(fs_info, bio, mirror_num);
837 ret = btree_csum_one_bio(bio);
840 ret = btrfs_map_bio(fs_info, bio, mirror_num);
843 * kthread helpers are used to submit writes so that
844 * checksumming can happen in parallel across all CPUs
846 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
847 0, btree_submit_bio_start);
855 bio->bi_status = ret;
860 #ifdef CONFIG_MIGRATION
861 static int btree_migratepage(struct address_space *mapping,
862 struct page *newpage, struct page *page,
863 enum migrate_mode mode)
866 * we can't safely write a btree page from here,
867 * we haven't done the locking hook
872 * Buffers may be managed in a filesystem specific way.
873 * We must have no buffers or drop them.
875 if (page_has_private(page) &&
876 !try_to_release_page(page, GFP_KERNEL))
878 return migrate_page(mapping, newpage, page, mode);
883 static int btree_writepages(struct address_space *mapping,
884 struct writeback_control *wbc)
886 struct btrfs_fs_info *fs_info;
889 if (wbc->sync_mode == WB_SYNC_NONE) {
891 if (wbc->for_kupdate)
894 fs_info = BTRFS_I(mapping->host)->root->fs_info;
895 /* this is a bit racy, but that's ok */
896 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
897 BTRFS_DIRTY_METADATA_THRESH,
898 fs_info->dirty_metadata_batch);
902 return btree_write_cache_pages(mapping, wbc);
905 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
907 if (PageWriteback(page) || PageDirty(page))
910 return try_release_extent_buffer(page);
913 static void btree_invalidatepage(struct page *page, unsigned int offset,
916 struct extent_io_tree *tree;
917 tree = &BTRFS_I(page->mapping->host)->io_tree;
918 extent_invalidatepage(tree, page, offset);
919 btree_releasepage(page, GFP_NOFS);
920 if (PagePrivate(page)) {
921 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
922 "page private not zero on page %llu",
923 (unsigned long long)page_offset(page));
924 detach_page_private(page);
928 static int btree_set_page_dirty(struct page *page)
931 struct extent_buffer *eb;
933 BUG_ON(!PagePrivate(page));
934 eb = (struct extent_buffer *)page->private;
936 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
937 BUG_ON(!atomic_read(&eb->refs));
938 btrfs_assert_tree_locked(eb);
940 return __set_page_dirty_nobuffers(page);
943 static const struct address_space_operations btree_aops = {
944 .writepages = btree_writepages,
945 .releasepage = btree_releasepage,
946 .invalidatepage = btree_invalidatepage,
947 #ifdef CONFIG_MIGRATION
948 .migratepage = btree_migratepage,
950 .set_page_dirty = btree_set_page_dirty,
953 struct extent_buffer *btrfs_find_create_tree_block(
954 struct btrfs_fs_info *fs_info,
955 u64 bytenr, u64 owner_root,
958 if (btrfs_is_testing(fs_info))
959 return alloc_test_extent_buffer(fs_info, bytenr);
960 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
964 * Read tree block at logical address @bytenr and do variant basic but critical
967 * @owner_root: the objectid of the root owner for this block.
968 * @parent_transid: expected transid of this tree block, skip check if 0
969 * @level: expected level, mandatory check
970 * @first_key: expected key in slot 0, skip check if NULL
972 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
973 u64 owner_root, u64 parent_transid,
974 int level, struct btrfs_key *first_key)
976 struct extent_buffer *buf = NULL;
979 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
983 ret = btree_read_extent_buffer_pages(buf, parent_transid,
986 free_extent_buffer_stale(buf);
993 void btrfs_clean_tree_block(struct extent_buffer *buf)
995 struct btrfs_fs_info *fs_info = buf->fs_info;
996 if (btrfs_header_generation(buf) ==
997 fs_info->running_transaction->transid) {
998 btrfs_assert_tree_locked(buf);
1000 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1001 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1003 fs_info->dirty_metadata_batch);
1004 clear_extent_buffer_dirty(buf);
1009 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1012 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1013 root->fs_info = fs_info;
1015 root->commit_root = NULL;
1017 root->orphan_cleanup_state = 0;
1019 root->last_trans = 0;
1020 root->highest_objectid = 0;
1021 root->nr_delalloc_inodes = 0;
1022 root->nr_ordered_extents = 0;
1023 root->inode_tree = RB_ROOT;
1024 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1025 root->block_rsv = NULL;
1027 INIT_LIST_HEAD(&root->dirty_list);
1028 INIT_LIST_HEAD(&root->root_list);
1029 INIT_LIST_HEAD(&root->delalloc_inodes);
1030 INIT_LIST_HEAD(&root->delalloc_root);
1031 INIT_LIST_HEAD(&root->ordered_extents);
1032 INIT_LIST_HEAD(&root->ordered_root);
1033 INIT_LIST_HEAD(&root->reloc_dirty_list);
1034 INIT_LIST_HEAD(&root->logged_list[0]);
1035 INIT_LIST_HEAD(&root->logged_list[1]);
1036 spin_lock_init(&root->inode_lock);
1037 spin_lock_init(&root->delalloc_lock);
1038 spin_lock_init(&root->ordered_extent_lock);
1039 spin_lock_init(&root->accounting_lock);
1040 spin_lock_init(&root->log_extents_lock[0]);
1041 spin_lock_init(&root->log_extents_lock[1]);
1042 spin_lock_init(&root->qgroup_meta_rsv_lock);
1043 mutex_init(&root->objectid_mutex);
1044 mutex_init(&root->log_mutex);
1045 mutex_init(&root->ordered_extent_mutex);
1046 mutex_init(&root->delalloc_mutex);
1047 init_waitqueue_head(&root->qgroup_flush_wait);
1048 init_waitqueue_head(&root->log_writer_wait);
1049 init_waitqueue_head(&root->log_commit_wait[0]);
1050 init_waitqueue_head(&root->log_commit_wait[1]);
1051 INIT_LIST_HEAD(&root->log_ctxs[0]);
1052 INIT_LIST_HEAD(&root->log_ctxs[1]);
1053 atomic_set(&root->log_commit[0], 0);
1054 atomic_set(&root->log_commit[1], 0);
1055 atomic_set(&root->log_writers, 0);
1056 atomic_set(&root->log_batch, 0);
1057 refcount_set(&root->refs, 1);
1058 atomic_set(&root->snapshot_force_cow, 0);
1059 atomic_set(&root->nr_swapfiles, 0);
1060 root->log_transid = 0;
1061 root->log_transid_committed = -1;
1062 root->last_log_commit = 0;
1064 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1065 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1066 extent_io_tree_init(fs_info, &root->log_csum_range,
1067 IO_TREE_LOG_CSUM_RANGE, NULL);
1070 memset(&root->root_key, 0, sizeof(root->root_key));
1071 memset(&root->root_item, 0, sizeof(root->root_item));
1072 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1073 root->root_key.objectid = objectid;
1076 spin_lock_init(&root->root_item_lock);
1077 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1078 #ifdef CONFIG_BTRFS_DEBUG
1079 INIT_LIST_HEAD(&root->leak_list);
1080 spin_lock(&fs_info->fs_roots_radix_lock);
1081 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1082 spin_unlock(&fs_info->fs_roots_radix_lock);
1086 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1087 u64 objectid, gfp_t flags)
1089 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1091 __setup_root(root, fs_info, objectid);
1095 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1096 /* Should only be used by the testing infrastructure */
1097 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1099 struct btrfs_root *root;
1102 return ERR_PTR(-EINVAL);
1104 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1106 return ERR_PTR(-ENOMEM);
1108 /* We don't use the stripesize in selftest, set it as sectorsize */
1109 root->alloc_bytenr = 0;
1115 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1118 struct btrfs_fs_info *fs_info = trans->fs_info;
1119 struct extent_buffer *leaf;
1120 struct btrfs_root *tree_root = fs_info->tree_root;
1121 struct btrfs_root *root;
1122 struct btrfs_key key;
1123 unsigned int nofs_flag;
1127 * We're holding a transaction handle, so use a NOFS memory allocation
1128 * context to avoid deadlock if reclaim happens.
1130 nofs_flag = memalloc_nofs_save();
1131 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1132 memalloc_nofs_restore(nofs_flag);
1134 return ERR_PTR(-ENOMEM);
1136 root->root_key.objectid = objectid;
1137 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1138 root->root_key.offset = 0;
1140 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1141 BTRFS_NESTING_NORMAL);
1143 ret = PTR_ERR(leaf);
1149 btrfs_mark_buffer_dirty(leaf);
1151 root->commit_root = btrfs_root_node(root);
1152 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1154 btrfs_set_root_flags(&root->root_item, 0);
1155 btrfs_set_root_limit(&root->root_item, 0);
1156 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1157 btrfs_set_root_generation(&root->root_item, trans->transid);
1158 btrfs_set_root_level(&root->root_item, 0);
1159 btrfs_set_root_refs(&root->root_item, 1);
1160 btrfs_set_root_used(&root->root_item, leaf->len);
1161 btrfs_set_root_last_snapshot(&root->root_item, 0);
1162 btrfs_set_root_dirid(&root->root_item, 0);
1163 if (is_fstree(objectid))
1164 generate_random_guid(root->root_item.uuid);
1166 export_guid(root->root_item.uuid, &guid_null);
1167 btrfs_set_root_drop_level(&root->root_item, 0);
1169 key.objectid = objectid;
1170 key.type = BTRFS_ROOT_ITEM_KEY;
1172 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1176 btrfs_tree_unlock(leaf);
1182 btrfs_tree_unlock(leaf);
1183 btrfs_put_root(root);
1185 return ERR_PTR(ret);
1188 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1189 struct btrfs_fs_info *fs_info)
1191 struct btrfs_root *root;
1192 struct extent_buffer *leaf;
1194 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1196 return ERR_PTR(-ENOMEM);
1198 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1199 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1200 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1203 * DON'T set SHAREABLE bit for log trees.
1205 * Log trees are not exposed to user space thus can't be snapshotted,
1206 * and they go away before a real commit is actually done.
1208 * They do store pointers to file data extents, and those reference
1209 * counts still get updated (along with back refs to the log tree).
1212 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1213 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1215 btrfs_put_root(root);
1216 return ERR_CAST(leaf);
1221 btrfs_mark_buffer_dirty(root->node);
1222 btrfs_tree_unlock(root->node);
1226 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1227 struct btrfs_fs_info *fs_info)
1229 struct btrfs_root *log_root;
1231 log_root = alloc_log_tree(trans, fs_info);
1232 if (IS_ERR(log_root))
1233 return PTR_ERR(log_root);
1234 WARN_ON(fs_info->log_root_tree);
1235 fs_info->log_root_tree = log_root;
1239 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1240 struct btrfs_root *root)
1242 struct btrfs_fs_info *fs_info = root->fs_info;
1243 struct btrfs_root *log_root;
1244 struct btrfs_inode_item *inode_item;
1246 log_root = alloc_log_tree(trans, fs_info);
1247 if (IS_ERR(log_root))
1248 return PTR_ERR(log_root);
1250 log_root->last_trans = trans->transid;
1251 log_root->root_key.offset = root->root_key.objectid;
1253 inode_item = &log_root->root_item.inode;
1254 btrfs_set_stack_inode_generation(inode_item, 1);
1255 btrfs_set_stack_inode_size(inode_item, 3);
1256 btrfs_set_stack_inode_nlink(inode_item, 1);
1257 btrfs_set_stack_inode_nbytes(inode_item,
1259 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1261 btrfs_set_root_node(&log_root->root_item, log_root->node);
1263 WARN_ON(root->log_root);
1264 root->log_root = log_root;
1265 root->log_transid = 0;
1266 root->log_transid_committed = -1;
1267 root->last_log_commit = 0;
1271 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1272 struct btrfs_path *path,
1273 struct btrfs_key *key)
1275 struct btrfs_root *root;
1276 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1281 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1283 return ERR_PTR(-ENOMEM);
1285 ret = btrfs_find_root(tree_root, key, path,
1286 &root->root_item, &root->root_key);
1293 generation = btrfs_root_generation(&root->root_item);
1294 level = btrfs_root_level(&root->root_item);
1295 root->node = read_tree_block(fs_info,
1296 btrfs_root_bytenr(&root->root_item),
1297 key->objectid, generation, level, NULL);
1298 if (IS_ERR(root->node)) {
1299 ret = PTR_ERR(root->node);
1302 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1306 root->commit_root = btrfs_root_node(root);
1309 btrfs_put_root(root);
1310 return ERR_PTR(ret);
1313 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1314 struct btrfs_key *key)
1316 struct btrfs_root *root;
1317 struct btrfs_path *path;
1319 path = btrfs_alloc_path();
1321 return ERR_PTR(-ENOMEM);
1322 root = read_tree_root_path(tree_root, path, key);
1323 btrfs_free_path(path);
1329 * Initialize subvolume root in-memory structure
1331 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1333 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1336 unsigned int nofs_flag;
1339 * We might be called under a transaction (e.g. indirect backref
1340 * resolution) which could deadlock if it triggers memory reclaim
1342 nofs_flag = memalloc_nofs_save();
1343 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1344 memalloc_nofs_restore(nofs_flag);
1348 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1349 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1350 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1351 btrfs_check_and_init_root_item(&root->root_item);
1355 * Don't assign anonymous block device to roots that are not exposed to
1356 * userspace, the id pool is limited to 1M
1358 if (is_fstree(root->root_key.objectid) &&
1359 btrfs_root_refs(&root->root_item) > 0) {
1361 ret = get_anon_bdev(&root->anon_dev);
1365 root->anon_dev = anon_dev;
1369 mutex_lock(&root->objectid_mutex);
1370 ret = btrfs_find_highest_objectid(root,
1371 &root->highest_objectid);
1373 mutex_unlock(&root->objectid_mutex);
1377 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1379 mutex_unlock(&root->objectid_mutex);
1383 /* The caller is responsible to call btrfs_free_fs_root */
1387 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1390 struct btrfs_root *root;
1392 spin_lock(&fs_info->fs_roots_radix_lock);
1393 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1394 (unsigned long)root_id);
1396 root = btrfs_grab_root(root);
1397 spin_unlock(&fs_info->fs_roots_radix_lock);
1401 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1404 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1405 return btrfs_grab_root(fs_info->tree_root);
1406 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1407 return btrfs_grab_root(fs_info->extent_root);
1408 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1409 return btrfs_grab_root(fs_info->chunk_root);
1410 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1411 return btrfs_grab_root(fs_info->dev_root);
1412 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1413 return btrfs_grab_root(fs_info->csum_root);
1414 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1415 return btrfs_grab_root(fs_info->quota_root) ?
1416 fs_info->quota_root : ERR_PTR(-ENOENT);
1417 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1418 return btrfs_grab_root(fs_info->uuid_root) ?
1419 fs_info->uuid_root : ERR_PTR(-ENOENT);
1420 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1421 return btrfs_grab_root(fs_info->free_space_root) ?
1422 fs_info->free_space_root : ERR_PTR(-ENOENT);
1426 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1427 struct btrfs_root *root)
1431 ret = radix_tree_preload(GFP_NOFS);
1435 spin_lock(&fs_info->fs_roots_radix_lock);
1436 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1437 (unsigned long)root->root_key.objectid,
1440 btrfs_grab_root(root);
1441 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1443 spin_unlock(&fs_info->fs_roots_radix_lock);
1444 radix_tree_preload_end();
1449 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1451 #ifdef CONFIG_BTRFS_DEBUG
1452 struct btrfs_root *root;
1454 while (!list_empty(&fs_info->allocated_roots)) {
1455 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1457 root = list_first_entry(&fs_info->allocated_roots,
1458 struct btrfs_root, leak_list);
1459 btrfs_err(fs_info, "leaked root %s refcount %d",
1460 btrfs_root_name(root->root_key.objectid, buf),
1461 refcount_read(&root->refs));
1462 while (refcount_read(&root->refs) > 1)
1463 btrfs_put_root(root);
1464 btrfs_put_root(root);
1469 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1471 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1472 percpu_counter_destroy(&fs_info->delalloc_bytes);
1473 percpu_counter_destroy(&fs_info->dio_bytes);
1474 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1475 btrfs_free_csum_hash(fs_info);
1476 btrfs_free_stripe_hash_table(fs_info);
1477 btrfs_free_ref_cache(fs_info);
1478 kfree(fs_info->balance_ctl);
1479 kfree(fs_info->delayed_root);
1480 btrfs_put_root(fs_info->extent_root);
1481 btrfs_put_root(fs_info->tree_root);
1482 btrfs_put_root(fs_info->chunk_root);
1483 btrfs_put_root(fs_info->dev_root);
1484 btrfs_put_root(fs_info->csum_root);
1485 btrfs_put_root(fs_info->quota_root);
1486 btrfs_put_root(fs_info->uuid_root);
1487 btrfs_put_root(fs_info->free_space_root);
1488 btrfs_put_root(fs_info->fs_root);
1489 btrfs_put_root(fs_info->data_reloc_root);
1490 btrfs_check_leaked_roots(fs_info);
1491 btrfs_extent_buffer_leak_debug_check(fs_info);
1492 kfree(fs_info->super_copy);
1493 kfree(fs_info->super_for_commit);
1499 * Get an in-memory reference of a root structure.
1501 * For essential trees like root/extent tree, we grab it from fs_info directly.
1502 * For subvolume trees, we check the cached filesystem roots first. If not
1503 * found, then read it from disk and add it to cached fs roots.
1505 * Caller should release the root by calling btrfs_put_root() after the usage.
1507 * NOTE: Reloc and log trees can't be read by this function as they share the
1508 * same root objectid.
1510 * @objectid: root id
1511 * @anon_dev: preallocated anonymous block device number for new roots,
1512 * pass 0 for new allocation.
1513 * @check_ref: whether to check root item references, If true, return -ENOENT
1516 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1517 u64 objectid, dev_t anon_dev,
1520 struct btrfs_root *root;
1521 struct btrfs_path *path;
1522 struct btrfs_key key;
1525 root = btrfs_get_global_root(fs_info, objectid);
1529 root = btrfs_lookup_fs_root(fs_info, objectid);
1531 /* Shouldn't get preallocated anon_dev for cached roots */
1533 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1534 btrfs_put_root(root);
1535 return ERR_PTR(-ENOENT);
1540 key.objectid = objectid;
1541 key.type = BTRFS_ROOT_ITEM_KEY;
1542 key.offset = (u64)-1;
1543 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1547 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1552 ret = btrfs_init_fs_root(root, anon_dev);
1556 path = btrfs_alloc_path();
1561 key.objectid = BTRFS_ORPHAN_OBJECTID;
1562 key.type = BTRFS_ORPHAN_ITEM_KEY;
1563 key.offset = objectid;
1565 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1566 btrfs_free_path(path);
1570 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1572 ret = btrfs_insert_fs_root(fs_info, root);
1574 btrfs_put_root(root);
1581 btrfs_put_root(root);
1582 return ERR_PTR(ret);
1586 * Get in-memory reference of a root structure
1588 * @objectid: tree objectid
1589 * @check_ref: if set, verify that the tree exists and the item has at least
1592 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1593 u64 objectid, bool check_ref)
1595 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1599 * Get in-memory reference of a root structure, created as new, optionally pass
1600 * the anonymous block device id
1602 * @objectid: tree objectid
1603 * @anon_dev: if zero, allocate a new anonymous block device or use the
1606 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1607 u64 objectid, dev_t anon_dev)
1609 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1613 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1614 * @fs_info: the fs_info
1615 * @objectid: the objectid we need to lookup
1617 * This is exclusively used for backref walking, and exists specifically because
1618 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1619 * creation time, which means we may have to read the tree_root in order to look
1620 * up a fs root that is not in memory. If the root is not in memory we will
1621 * read the tree root commit root and look up the fs root from there. This is a
1622 * temporary root, it will not be inserted into the radix tree as it doesn't
1623 * have the most uptodate information, it'll simply be discarded once the
1624 * backref code is finished using the root.
1626 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1627 struct btrfs_path *path,
1630 struct btrfs_root *root;
1631 struct btrfs_key key;
1633 ASSERT(path->search_commit_root && path->skip_locking);
1636 * This can return -ENOENT if we ask for a root that doesn't exist, but
1637 * since this is called via the backref walking code we won't be looking
1638 * up a root that doesn't exist, unless there's corruption. So if root
1639 * != NULL just return it.
1641 root = btrfs_get_global_root(fs_info, objectid);
1645 root = btrfs_lookup_fs_root(fs_info, objectid);
1649 key.objectid = objectid;
1650 key.type = BTRFS_ROOT_ITEM_KEY;
1651 key.offset = (u64)-1;
1652 root = read_tree_root_path(fs_info->tree_root, path, &key);
1653 btrfs_release_path(path);
1659 * called by the kthread helper functions to finally call the bio end_io
1660 * functions. This is where read checksum verification actually happens
1662 static void end_workqueue_fn(struct btrfs_work *work)
1665 struct btrfs_end_io_wq *end_io_wq;
1667 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1668 bio = end_io_wq->bio;
1670 bio->bi_status = end_io_wq->status;
1671 bio->bi_private = end_io_wq->private;
1672 bio->bi_end_io = end_io_wq->end_io;
1674 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1677 static int cleaner_kthread(void *arg)
1679 struct btrfs_root *root = arg;
1680 struct btrfs_fs_info *fs_info = root->fs_info;
1686 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1688 /* Make the cleaner go to sleep early. */
1689 if (btrfs_need_cleaner_sleep(fs_info))
1693 * Do not do anything if we might cause open_ctree() to block
1694 * before we have finished mounting the filesystem.
1696 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1699 if (!mutex_trylock(&fs_info->cleaner_mutex))
1703 * Avoid the problem that we change the status of the fs
1704 * during the above check and trylock.
1706 if (btrfs_need_cleaner_sleep(fs_info)) {
1707 mutex_unlock(&fs_info->cleaner_mutex);
1711 btrfs_run_delayed_iputs(fs_info);
1713 again = btrfs_clean_one_deleted_snapshot(root);
1714 mutex_unlock(&fs_info->cleaner_mutex);
1717 * The defragger has dealt with the R/O remount and umount,
1718 * needn't do anything special here.
1720 btrfs_run_defrag_inodes(fs_info);
1723 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1724 * with relocation (btrfs_relocate_chunk) and relocation
1725 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1726 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1727 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1728 * unused block groups.
1730 btrfs_delete_unused_bgs(fs_info);
1732 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1733 if (kthread_should_park())
1735 if (kthread_should_stop())
1738 set_current_state(TASK_INTERRUPTIBLE);
1740 __set_current_state(TASK_RUNNING);
1745 static int transaction_kthread(void *arg)
1747 struct btrfs_root *root = arg;
1748 struct btrfs_fs_info *fs_info = root->fs_info;
1749 struct btrfs_trans_handle *trans;
1750 struct btrfs_transaction *cur;
1753 unsigned long delay;
1757 cannot_commit = false;
1758 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1759 mutex_lock(&fs_info->transaction_kthread_mutex);
1761 spin_lock(&fs_info->trans_lock);
1762 cur = fs_info->running_transaction;
1764 spin_unlock(&fs_info->trans_lock);
1768 delta = ktime_get_seconds() - cur->start_time;
1769 if (cur->state < TRANS_STATE_COMMIT_START &&
1770 delta < fs_info->commit_interval) {
1771 spin_unlock(&fs_info->trans_lock);
1772 delay -= msecs_to_jiffies((delta - 1) * 1000);
1774 msecs_to_jiffies(fs_info->commit_interval * 1000));
1777 transid = cur->transid;
1778 spin_unlock(&fs_info->trans_lock);
1780 /* If the file system is aborted, this will always fail. */
1781 trans = btrfs_attach_transaction(root);
1782 if (IS_ERR(trans)) {
1783 if (PTR_ERR(trans) != -ENOENT)
1784 cannot_commit = true;
1787 if (transid == trans->transid) {
1788 btrfs_commit_transaction(trans);
1790 btrfs_end_transaction(trans);
1793 wake_up_process(fs_info->cleaner_kthread);
1794 mutex_unlock(&fs_info->transaction_kthread_mutex);
1796 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1797 &fs_info->fs_state)))
1798 btrfs_cleanup_transaction(fs_info);
1799 if (!kthread_should_stop() &&
1800 (!btrfs_transaction_blocked(fs_info) ||
1802 schedule_timeout_interruptible(delay);
1803 } while (!kthread_should_stop());
1808 * This will find the highest generation in the array of root backups. The
1809 * index of the highest array is returned, or -EINVAL if we can't find
1812 * We check to make sure the array is valid by comparing the
1813 * generation of the latest root in the array with the generation
1814 * in the super block. If they don't match we pitch it.
1816 static int find_newest_super_backup(struct btrfs_fs_info *info)
1818 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1820 struct btrfs_root_backup *root_backup;
1823 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1824 root_backup = info->super_copy->super_roots + i;
1825 cur = btrfs_backup_tree_root_gen(root_backup);
1826 if (cur == newest_gen)
1834 * copy all the root pointers into the super backup array.
1835 * this will bump the backup pointer by one when it is
1838 static void backup_super_roots(struct btrfs_fs_info *info)
1840 const int next_backup = info->backup_root_index;
1841 struct btrfs_root_backup *root_backup;
1843 root_backup = info->super_for_commit->super_roots + next_backup;
1846 * make sure all of our padding and empty slots get zero filled
1847 * regardless of which ones we use today
1849 memset(root_backup, 0, sizeof(*root_backup));
1851 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1853 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1854 btrfs_set_backup_tree_root_gen(root_backup,
1855 btrfs_header_generation(info->tree_root->node));
1857 btrfs_set_backup_tree_root_level(root_backup,
1858 btrfs_header_level(info->tree_root->node));
1860 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1861 btrfs_set_backup_chunk_root_gen(root_backup,
1862 btrfs_header_generation(info->chunk_root->node));
1863 btrfs_set_backup_chunk_root_level(root_backup,
1864 btrfs_header_level(info->chunk_root->node));
1866 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1867 btrfs_set_backup_extent_root_gen(root_backup,
1868 btrfs_header_generation(info->extent_root->node));
1869 btrfs_set_backup_extent_root_level(root_backup,
1870 btrfs_header_level(info->extent_root->node));
1873 * we might commit during log recovery, which happens before we set
1874 * the fs_root. Make sure it is valid before we fill it in.
1876 if (info->fs_root && info->fs_root->node) {
1877 btrfs_set_backup_fs_root(root_backup,
1878 info->fs_root->node->start);
1879 btrfs_set_backup_fs_root_gen(root_backup,
1880 btrfs_header_generation(info->fs_root->node));
1881 btrfs_set_backup_fs_root_level(root_backup,
1882 btrfs_header_level(info->fs_root->node));
1885 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1886 btrfs_set_backup_dev_root_gen(root_backup,
1887 btrfs_header_generation(info->dev_root->node));
1888 btrfs_set_backup_dev_root_level(root_backup,
1889 btrfs_header_level(info->dev_root->node));
1891 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1892 btrfs_set_backup_csum_root_gen(root_backup,
1893 btrfs_header_generation(info->csum_root->node));
1894 btrfs_set_backup_csum_root_level(root_backup,
1895 btrfs_header_level(info->csum_root->node));
1897 btrfs_set_backup_total_bytes(root_backup,
1898 btrfs_super_total_bytes(info->super_copy));
1899 btrfs_set_backup_bytes_used(root_backup,
1900 btrfs_super_bytes_used(info->super_copy));
1901 btrfs_set_backup_num_devices(root_backup,
1902 btrfs_super_num_devices(info->super_copy));
1905 * if we don't copy this out to the super_copy, it won't get remembered
1906 * for the next commit
1908 memcpy(&info->super_copy->super_roots,
1909 &info->super_for_commit->super_roots,
1910 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1914 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1915 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1917 * fs_info - filesystem whose backup roots need to be read
1918 * priority - priority of backup root required
1920 * Returns backup root index on success and -EINVAL otherwise.
1922 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1924 int backup_index = find_newest_super_backup(fs_info);
1925 struct btrfs_super_block *super = fs_info->super_copy;
1926 struct btrfs_root_backup *root_backup;
1928 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1930 return backup_index;
1932 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1933 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1938 root_backup = super->super_roots + backup_index;
1940 btrfs_set_super_generation(super,
1941 btrfs_backup_tree_root_gen(root_backup));
1942 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1943 btrfs_set_super_root_level(super,
1944 btrfs_backup_tree_root_level(root_backup));
1945 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1948 * Fixme: the total bytes and num_devices need to match or we should
1951 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1952 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1954 return backup_index;
1957 /* helper to cleanup workers */
1958 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1960 btrfs_destroy_workqueue(fs_info->fixup_workers);
1961 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1962 btrfs_destroy_workqueue(fs_info->workers);
1963 btrfs_destroy_workqueue(fs_info->endio_workers);
1964 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1965 btrfs_destroy_workqueue(fs_info->rmw_workers);
1966 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1967 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1968 btrfs_destroy_workqueue(fs_info->delayed_workers);
1969 btrfs_destroy_workqueue(fs_info->caching_workers);
1970 btrfs_destroy_workqueue(fs_info->readahead_workers);
1971 btrfs_destroy_workqueue(fs_info->flush_workers);
1972 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1973 if (fs_info->discard_ctl.discard_workers)
1974 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1976 * Now that all other work queues are destroyed, we can safely destroy
1977 * the queues used for metadata I/O, since tasks from those other work
1978 * queues can do metadata I/O operations.
1980 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
1981 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
1984 static void free_root_extent_buffers(struct btrfs_root *root)
1987 free_extent_buffer(root->node);
1988 free_extent_buffer(root->commit_root);
1990 root->commit_root = NULL;
1994 /* helper to cleanup tree roots */
1995 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1997 free_root_extent_buffers(info->tree_root);
1999 free_root_extent_buffers(info->dev_root);
2000 free_root_extent_buffers(info->extent_root);
2001 free_root_extent_buffers(info->csum_root);
2002 free_root_extent_buffers(info->quota_root);
2003 free_root_extent_buffers(info->uuid_root);
2004 free_root_extent_buffers(info->fs_root);
2005 free_root_extent_buffers(info->data_reloc_root);
2006 if (free_chunk_root)
2007 free_root_extent_buffers(info->chunk_root);
2008 free_root_extent_buffers(info->free_space_root);
2011 void btrfs_put_root(struct btrfs_root *root)
2016 if (refcount_dec_and_test(&root->refs)) {
2017 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2018 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2020 free_anon_bdev(root->anon_dev);
2021 btrfs_drew_lock_destroy(&root->snapshot_lock);
2022 free_root_extent_buffers(root);
2023 #ifdef CONFIG_BTRFS_DEBUG
2024 spin_lock(&root->fs_info->fs_roots_radix_lock);
2025 list_del_init(&root->leak_list);
2026 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2032 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2035 struct btrfs_root *gang[8];
2038 while (!list_empty(&fs_info->dead_roots)) {
2039 gang[0] = list_entry(fs_info->dead_roots.next,
2040 struct btrfs_root, root_list);
2041 list_del(&gang[0]->root_list);
2043 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2044 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2045 btrfs_put_root(gang[0]);
2049 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2054 for (i = 0; i < ret; i++)
2055 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2059 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2061 mutex_init(&fs_info->scrub_lock);
2062 atomic_set(&fs_info->scrubs_running, 0);
2063 atomic_set(&fs_info->scrub_pause_req, 0);
2064 atomic_set(&fs_info->scrubs_paused, 0);
2065 atomic_set(&fs_info->scrub_cancel_req, 0);
2066 init_waitqueue_head(&fs_info->scrub_pause_wait);
2067 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2070 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2072 spin_lock_init(&fs_info->balance_lock);
2073 mutex_init(&fs_info->balance_mutex);
2074 atomic_set(&fs_info->balance_pause_req, 0);
2075 atomic_set(&fs_info->balance_cancel_req, 0);
2076 fs_info->balance_ctl = NULL;
2077 init_waitqueue_head(&fs_info->balance_wait_q);
2080 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2082 struct inode *inode = fs_info->btree_inode;
2084 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2085 set_nlink(inode, 1);
2087 * we set the i_size on the btree inode to the max possible int.
2088 * the real end of the address space is determined by all of
2089 * the devices in the system
2091 inode->i_size = OFFSET_MAX;
2092 inode->i_mapping->a_ops = &btree_aops;
2094 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2095 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2096 IO_TREE_BTREE_INODE_IO, inode);
2097 BTRFS_I(inode)->io_tree.track_uptodate = false;
2098 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2100 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2101 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2102 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2103 btrfs_insert_inode_hash(inode);
2106 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2108 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2109 init_rwsem(&fs_info->dev_replace.rwsem);
2110 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2113 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2115 spin_lock_init(&fs_info->qgroup_lock);
2116 mutex_init(&fs_info->qgroup_ioctl_lock);
2117 fs_info->qgroup_tree = RB_ROOT;
2118 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2119 fs_info->qgroup_seq = 1;
2120 fs_info->qgroup_ulist = NULL;
2121 fs_info->qgroup_rescan_running = false;
2122 mutex_init(&fs_info->qgroup_rescan_lock);
2125 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2126 struct btrfs_fs_devices *fs_devices)
2128 u32 max_active = fs_info->thread_pool_size;
2129 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2132 btrfs_alloc_workqueue(fs_info, "worker",
2133 flags | WQ_HIGHPRI, max_active, 16);
2135 fs_info->delalloc_workers =
2136 btrfs_alloc_workqueue(fs_info, "delalloc",
2137 flags, max_active, 2);
2139 fs_info->flush_workers =
2140 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2141 flags, max_active, 0);
2143 fs_info->caching_workers =
2144 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2146 fs_info->fixup_workers =
2147 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2150 * endios are largely parallel and should have a very
2153 fs_info->endio_workers =
2154 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2155 fs_info->endio_meta_workers =
2156 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2158 fs_info->endio_meta_write_workers =
2159 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2161 fs_info->endio_raid56_workers =
2162 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2164 fs_info->rmw_workers =
2165 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2166 fs_info->endio_write_workers =
2167 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2169 fs_info->endio_freespace_worker =
2170 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2172 fs_info->delayed_workers =
2173 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2175 fs_info->readahead_workers =
2176 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2178 fs_info->qgroup_rescan_workers =
2179 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2180 fs_info->discard_ctl.discard_workers =
2181 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2183 if (!(fs_info->workers && fs_info->delalloc_workers &&
2184 fs_info->flush_workers &&
2185 fs_info->endio_workers && fs_info->endio_meta_workers &&
2186 fs_info->endio_meta_write_workers &&
2187 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2188 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2189 fs_info->caching_workers && fs_info->readahead_workers &&
2190 fs_info->fixup_workers && fs_info->delayed_workers &&
2191 fs_info->qgroup_rescan_workers &&
2192 fs_info->discard_ctl.discard_workers)) {
2199 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2201 struct crypto_shash *csum_shash;
2202 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2204 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2206 if (IS_ERR(csum_shash)) {
2207 btrfs_err(fs_info, "error allocating %s hash for checksum",
2209 return PTR_ERR(csum_shash);
2212 fs_info->csum_shash = csum_shash;
2217 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2218 struct btrfs_fs_devices *fs_devices)
2221 struct btrfs_root *log_tree_root;
2222 struct btrfs_super_block *disk_super = fs_info->super_copy;
2223 u64 bytenr = btrfs_super_log_root(disk_super);
2224 int level = btrfs_super_log_root_level(disk_super);
2226 if (fs_devices->rw_devices == 0) {
2227 btrfs_warn(fs_info, "log replay required on RO media");
2231 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2236 log_tree_root->node = read_tree_block(fs_info, bytenr,
2237 BTRFS_TREE_LOG_OBJECTID,
2238 fs_info->generation + 1, level,
2240 if (IS_ERR(log_tree_root->node)) {
2241 btrfs_warn(fs_info, "failed to read log tree");
2242 ret = PTR_ERR(log_tree_root->node);
2243 log_tree_root->node = NULL;
2244 btrfs_put_root(log_tree_root);
2246 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2247 btrfs_err(fs_info, "failed to read log tree");
2248 btrfs_put_root(log_tree_root);
2251 /* returns with log_tree_root freed on success */
2252 ret = btrfs_recover_log_trees(log_tree_root);
2254 btrfs_handle_fs_error(fs_info, ret,
2255 "Failed to recover log tree");
2256 btrfs_put_root(log_tree_root);
2260 if (sb_rdonly(fs_info->sb)) {
2261 ret = btrfs_commit_super(fs_info);
2269 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2271 struct btrfs_root *tree_root = fs_info->tree_root;
2272 struct btrfs_root *root;
2273 struct btrfs_key location;
2276 BUG_ON(!fs_info->tree_root);
2278 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2279 location.type = BTRFS_ROOT_ITEM_KEY;
2280 location.offset = 0;
2282 root = btrfs_read_tree_root(tree_root, &location);
2284 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2285 ret = PTR_ERR(root);
2289 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2290 fs_info->extent_root = root;
2293 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2294 root = btrfs_read_tree_root(tree_root, &location);
2296 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2297 ret = PTR_ERR(root);
2301 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2302 fs_info->dev_root = root;
2303 btrfs_init_devices_late(fs_info);
2306 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2307 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2308 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
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->csum_root = root;
2322 * This tree can share blocks with some other fs tree during relocation
2323 * and we need a proper setup by btrfs_get_fs_root
2325 root = btrfs_get_fs_root(tree_root->fs_info,
2326 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2328 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2329 ret = PTR_ERR(root);
2333 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2334 fs_info->data_reloc_root = root;
2337 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2338 root = btrfs_read_tree_root(tree_root, &location);
2339 if (!IS_ERR(root)) {
2340 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2341 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2342 fs_info->quota_root = root;
2345 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2346 root = btrfs_read_tree_root(tree_root, &location);
2348 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2349 ret = PTR_ERR(root);
2354 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2355 fs_info->uuid_root = root;
2358 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2359 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2360 root = btrfs_read_tree_root(tree_root, &location);
2362 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2363 ret = PTR_ERR(root);
2367 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2368 fs_info->free_space_root = root;
2374 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2375 location.objectid, ret);
2380 * Real super block validation
2381 * NOTE: super csum type and incompat features will not be checked here.
2383 * @sb: super block to check
2384 * @mirror_num: the super block number to check its bytenr:
2385 * 0 the primary (1st) sb
2386 * 1, 2 2nd and 3rd backup copy
2387 * -1 skip bytenr check
2389 static int validate_super(struct btrfs_fs_info *fs_info,
2390 struct btrfs_super_block *sb, int mirror_num)
2392 u64 nodesize = btrfs_super_nodesize(sb);
2393 u64 sectorsize = btrfs_super_sectorsize(sb);
2396 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2397 btrfs_err(fs_info, "no valid FS found");
2400 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2401 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2402 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2405 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2406 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2407 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2410 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2411 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2412 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2415 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2416 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2417 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2422 * Check sectorsize and nodesize first, other check will need it.
2423 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2425 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2426 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2427 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2430 /* Only PAGE SIZE is supported yet */
2431 if (sectorsize != PAGE_SIZE) {
2433 "sectorsize %llu not supported yet, only support %lu",
2434 sectorsize, PAGE_SIZE);
2437 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2438 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2439 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2442 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2443 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2444 le32_to_cpu(sb->__unused_leafsize), nodesize);
2448 /* Root alignment check */
2449 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2450 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2451 btrfs_super_root(sb));
2454 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2455 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2456 btrfs_super_chunk_root(sb));
2459 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2460 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2461 btrfs_super_log_root(sb));
2465 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2466 BTRFS_FSID_SIZE) != 0) {
2468 "dev_item UUID does not match metadata fsid: %pU != %pU",
2469 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2474 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2477 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2478 btrfs_err(fs_info, "bytes_used is too small %llu",
2479 btrfs_super_bytes_used(sb));
2482 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2483 btrfs_err(fs_info, "invalid stripesize %u",
2484 btrfs_super_stripesize(sb));
2487 if (btrfs_super_num_devices(sb) > (1UL << 31))
2488 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2489 btrfs_super_num_devices(sb));
2490 if (btrfs_super_num_devices(sb) == 0) {
2491 btrfs_err(fs_info, "number of devices is 0");
2495 if (mirror_num >= 0 &&
2496 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2497 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2498 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2503 * Obvious sys_chunk_array corruptions, it must hold at least one key
2506 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2507 btrfs_err(fs_info, "system chunk array too big %u > %u",
2508 btrfs_super_sys_array_size(sb),
2509 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2512 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2513 + sizeof(struct btrfs_chunk)) {
2514 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2515 btrfs_super_sys_array_size(sb),
2516 sizeof(struct btrfs_disk_key)
2517 + sizeof(struct btrfs_chunk));
2522 * The generation is a global counter, we'll trust it more than the others
2523 * but it's still possible that it's the one that's wrong.
2525 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2527 "suspicious: generation < chunk_root_generation: %llu < %llu",
2528 btrfs_super_generation(sb),
2529 btrfs_super_chunk_root_generation(sb));
2530 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2531 && btrfs_super_cache_generation(sb) != (u64)-1)
2533 "suspicious: generation < cache_generation: %llu < %llu",
2534 btrfs_super_generation(sb),
2535 btrfs_super_cache_generation(sb));
2541 * Validation of super block at mount time.
2542 * Some checks already done early at mount time, like csum type and incompat
2543 * flags will be skipped.
2545 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2547 return validate_super(fs_info, fs_info->super_copy, 0);
2551 * Validation of super block at write time.
2552 * Some checks like bytenr check will be skipped as their values will be
2554 * Extra checks like csum type and incompat flags will be done here.
2556 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2557 struct btrfs_super_block *sb)
2561 ret = validate_super(fs_info, sb, -1);
2564 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2566 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2567 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2570 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2573 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2574 btrfs_super_incompat_flags(sb),
2575 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2581 "super block corruption detected before writing it to disk");
2585 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2587 int backup_index = find_newest_super_backup(fs_info);
2588 struct btrfs_super_block *sb = fs_info->super_copy;
2589 struct btrfs_root *tree_root = fs_info->tree_root;
2590 bool handle_error = false;
2594 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2599 if (!IS_ERR(tree_root->node))
2600 free_extent_buffer(tree_root->node);
2601 tree_root->node = NULL;
2603 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2606 free_root_pointers(fs_info, 0);
2609 * Don't use the log in recovery mode, it won't be
2612 btrfs_set_super_log_root(sb, 0);
2614 /* We can't trust the free space cache either */
2615 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2617 ret = read_backup_root(fs_info, i);
2622 generation = btrfs_super_generation(sb);
2623 level = btrfs_super_root_level(sb);
2624 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2625 BTRFS_ROOT_TREE_OBJECTID,
2626 generation, level, NULL);
2627 if (IS_ERR(tree_root->node)) {
2628 handle_error = true;
2629 ret = PTR_ERR(tree_root->node);
2630 tree_root->node = NULL;
2631 btrfs_warn(fs_info, "couldn't read tree root");
2634 } else if (!extent_buffer_uptodate(tree_root->node)) {
2635 handle_error = true;
2637 btrfs_warn(fs_info, "error while reading tree root");
2641 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2642 tree_root->commit_root = btrfs_root_node(tree_root);
2643 btrfs_set_root_refs(&tree_root->root_item, 1);
2646 * No need to hold btrfs_root::objectid_mutex since the fs
2647 * hasn't been fully initialised and we are the only user
2649 ret = btrfs_find_highest_objectid(tree_root,
2650 &tree_root->highest_objectid);
2652 handle_error = true;
2656 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2658 ret = btrfs_read_roots(fs_info);
2660 handle_error = true;
2664 /* All successful */
2665 fs_info->generation = generation;
2666 fs_info->last_trans_committed = generation;
2668 /* Always begin writing backup roots after the one being used */
2669 if (backup_index < 0) {
2670 fs_info->backup_root_index = 0;
2672 fs_info->backup_root_index = backup_index + 1;
2673 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2681 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2683 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2684 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2685 INIT_LIST_HEAD(&fs_info->trans_list);
2686 INIT_LIST_HEAD(&fs_info->dead_roots);
2687 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2688 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2689 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2690 spin_lock_init(&fs_info->delalloc_root_lock);
2691 spin_lock_init(&fs_info->trans_lock);
2692 spin_lock_init(&fs_info->fs_roots_radix_lock);
2693 spin_lock_init(&fs_info->delayed_iput_lock);
2694 spin_lock_init(&fs_info->defrag_inodes_lock);
2695 spin_lock_init(&fs_info->super_lock);
2696 spin_lock_init(&fs_info->buffer_lock);
2697 spin_lock_init(&fs_info->unused_bgs_lock);
2698 rwlock_init(&fs_info->tree_mod_log_lock);
2699 mutex_init(&fs_info->unused_bg_unpin_mutex);
2700 mutex_init(&fs_info->delete_unused_bgs_mutex);
2701 mutex_init(&fs_info->reloc_mutex);
2702 mutex_init(&fs_info->delalloc_root_mutex);
2703 seqlock_init(&fs_info->profiles_lock);
2705 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2706 INIT_LIST_HEAD(&fs_info->space_info);
2707 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2708 INIT_LIST_HEAD(&fs_info->unused_bgs);
2709 #ifdef CONFIG_BTRFS_DEBUG
2710 INIT_LIST_HEAD(&fs_info->allocated_roots);
2711 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2712 spin_lock_init(&fs_info->eb_leak_lock);
2714 extent_map_tree_init(&fs_info->mapping_tree);
2715 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2716 BTRFS_BLOCK_RSV_GLOBAL);
2717 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2718 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2719 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2720 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2721 BTRFS_BLOCK_RSV_DELOPS);
2722 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2723 BTRFS_BLOCK_RSV_DELREFS);
2725 atomic_set(&fs_info->async_delalloc_pages, 0);
2726 atomic_set(&fs_info->defrag_running, 0);
2727 atomic_set(&fs_info->reada_works_cnt, 0);
2728 atomic_set(&fs_info->nr_delayed_iputs, 0);
2729 atomic64_set(&fs_info->tree_mod_seq, 0);
2730 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2731 fs_info->metadata_ratio = 0;
2732 fs_info->defrag_inodes = RB_ROOT;
2733 atomic64_set(&fs_info->free_chunk_space, 0);
2734 fs_info->tree_mod_log = RB_ROOT;
2735 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2736 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2737 /* readahead state */
2738 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2739 spin_lock_init(&fs_info->reada_lock);
2740 btrfs_init_ref_verify(fs_info);
2742 fs_info->thread_pool_size = min_t(unsigned long,
2743 num_online_cpus() + 2, 8);
2745 INIT_LIST_HEAD(&fs_info->ordered_roots);
2746 spin_lock_init(&fs_info->ordered_root_lock);
2748 btrfs_init_scrub(fs_info);
2749 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2750 fs_info->check_integrity_print_mask = 0;
2752 btrfs_init_balance(fs_info);
2753 btrfs_init_async_reclaim_work(fs_info);
2755 spin_lock_init(&fs_info->block_group_cache_lock);
2756 fs_info->block_group_cache_tree = RB_ROOT;
2757 fs_info->first_logical_byte = (u64)-1;
2759 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2760 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2761 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2763 mutex_init(&fs_info->ordered_operations_mutex);
2764 mutex_init(&fs_info->tree_log_mutex);
2765 mutex_init(&fs_info->chunk_mutex);
2766 mutex_init(&fs_info->transaction_kthread_mutex);
2767 mutex_init(&fs_info->cleaner_mutex);
2768 mutex_init(&fs_info->ro_block_group_mutex);
2769 init_rwsem(&fs_info->commit_root_sem);
2770 init_rwsem(&fs_info->cleanup_work_sem);
2771 init_rwsem(&fs_info->subvol_sem);
2772 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2774 btrfs_init_dev_replace_locks(fs_info);
2775 btrfs_init_qgroup(fs_info);
2776 btrfs_discard_init(fs_info);
2778 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2779 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2781 init_waitqueue_head(&fs_info->transaction_throttle);
2782 init_waitqueue_head(&fs_info->transaction_wait);
2783 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2784 init_waitqueue_head(&fs_info->async_submit_wait);
2785 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2787 /* Usable values until the real ones are cached from the superblock */
2788 fs_info->nodesize = 4096;
2789 fs_info->sectorsize = 4096;
2790 fs_info->sectorsize_bits = ilog2(4096);
2791 fs_info->stripesize = 4096;
2793 spin_lock_init(&fs_info->swapfile_pins_lock);
2794 fs_info->swapfile_pins = RB_ROOT;
2796 fs_info->send_in_progress = 0;
2799 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2804 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2805 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2807 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2811 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2815 fs_info->dirty_metadata_batch = PAGE_SIZE *
2816 (1 + ilog2(nr_cpu_ids));
2818 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2822 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2827 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2829 if (!fs_info->delayed_root)
2831 btrfs_init_delayed_root(fs_info->delayed_root);
2833 return btrfs_alloc_stripe_hash_table(fs_info);
2836 static int btrfs_uuid_rescan_kthread(void *data)
2838 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2842 * 1st step is to iterate through the existing UUID tree and
2843 * to delete all entries that contain outdated data.
2844 * 2nd step is to add all missing entries to the UUID tree.
2846 ret = btrfs_uuid_tree_iterate(fs_info);
2849 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2851 up(&fs_info->uuid_tree_rescan_sem);
2854 return btrfs_uuid_scan_kthread(data);
2857 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2859 struct task_struct *task;
2861 down(&fs_info->uuid_tree_rescan_sem);
2862 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2864 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2865 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2866 up(&fs_info->uuid_tree_rescan_sem);
2867 return PTR_ERR(task);
2874 * Mounting logic specific to read-write file systems. Shared by open_ctree
2875 * and btrfs_remount when remounting from read-only to read-write.
2877 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2881 ret = btrfs_cleanup_fs_roots(fs_info);
2885 mutex_lock(&fs_info->cleaner_mutex);
2886 ret = btrfs_recover_relocation(fs_info->tree_root);
2887 mutex_unlock(&fs_info->cleaner_mutex);
2889 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
2893 ret = btrfs_resume_balance_async(fs_info);
2897 ret = btrfs_resume_dev_replace_async(fs_info);
2899 btrfs_warn(fs_info, "failed to resume dev_replace");
2903 btrfs_qgroup_rescan_resume(fs_info);
2905 if (!fs_info->uuid_root) {
2906 btrfs_info(fs_info, "creating UUID tree");
2907 ret = btrfs_create_uuid_tree(fs_info);
2910 "failed to create the UUID tree %d", ret);
2919 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2928 struct btrfs_super_block *disk_super;
2929 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2930 struct btrfs_root *tree_root;
2931 struct btrfs_root *chunk_root;
2934 int clear_free_space_tree = 0;
2937 ret = init_mount_fs_info(fs_info, sb);
2943 /* These need to be init'ed before we start creating inodes and such. */
2944 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2946 fs_info->tree_root = tree_root;
2947 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2949 fs_info->chunk_root = chunk_root;
2950 if (!tree_root || !chunk_root) {
2955 fs_info->btree_inode = new_inode(sb);
2956 if (!fs_info->btree_inode) {
2960 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2961 btrfs_init_btree_inode(fs_info);
2963 invalidate_bdev(fs_devices->latest_bdev);
2966 * Read super block and check the signature bytes only
2968 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2969 if (IS_ERR(disk_super)) {
2970 err = PTR_ERR(disk_super);
2975 * Verify the type first, if that or the checksum value are
2976 * corrupted, we'll find out
2978 csum_type = btrfs_super_csum_type(disk_super);
2979 if (!btrfs_supported_super_csum(csum_type)) {
2980 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2983 btrfs_release_disk_super(disk_super);
2987 ret = btrfs_init_csum_hash(fs_info, csum_type);
2990 btrfs_release_disk_super(disk_super);
2995 * We want to check superblock checksum, the type is stored inside.
2996 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2998 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2999 btrfs_err(fs_info, "superblock checksum mismatch");
3001 btrfs_release_disk_super(disk_super);
3006 * super_copy is zeroed at allocation time and we never touch the
3007 * following bytes up to INFO_SIZE, the checksum is calculated from
3008 * the whole block of INFO_SIZE
3010 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3011 btrfs_release_disk_super(disk_super);
3013 disk_super = fs_info->super_copy;
3015 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
3018 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
3019 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
3020 fs_info->super_copy->metadata_uuid,
3024 features = btrfs_super_flags(disk_super);
3025 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3026 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3027 btrfs_set_super_flags(disk_super, features);
3029 "found metadata UUID change in progress flag, clearing");
3032 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3033 sizeof(*fs_info->super_for_commit));
3035 ret = btrfs_validate_mount_super(fs_info);
3037 btrfs_err(fs_info, "superblock contains fatal errors");
3042 if (!btrfs_super_root(disk_super))
3045 /* check FS state, whether FS is broken. */
3046 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3047 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3050 * In the long term, we'll store the compression type in the super
3051 * block, and it'll be used for per file compression control.
3053 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3055 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3061 features = btrfs_super_incompat_flags(disk_super) &
3062 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3065 "cannot mount because of unsupported optional features (%llx)",
3071 features = btrfs_super_incompat_flags(disk_super);
3072 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3073 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3074 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3075 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3076 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3078 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3079 btrfs_info(fs_info, "has skinny extents");
3081 fs_info->zoned = (features & BTRFS_FEATURE_INCOMPAT_ZONED);
3084 * flag our filesystem as having big metadata blocks if
3085 * they are bigger than the page size
3087 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3088 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3090 "flagging fs with big metadata feature");
3091 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3094 nodesize = btrfs_super_nodesize(disk_super);
3095 sectorsize = btrfs_super_sectorsize(disk_super);
3096 stripesize = sectorsize;
3097 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3098 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3100 /* Cache block sizes */
3101 fs_info->nodesize = nodesize;
3102 fs_info->sectorsize = sectorsize;
3103 fs_info->sectorsize_bits = ilog2(sectorsize);
3104 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3105 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3106 fs_info->stripesize = stripesize;
3109 * mixed block groups end up with duplicate but slightly offset
3110 * extent buffers for the same range. It leads to corruptions
3112 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3113 (sectorsize != nodesize)) {
3115 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3116 nodesize, sectorsize);
3121 * Needn't use the lock because there is no other task which will
3124 btrfs_set_super_incompat_flags(disk_super, features);
3126 features = btrfs_super_compat_ro_flags(disk_super) &
3127 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3128 if (!sb_rdonly(sb) && features) {
3130 "cannot mount read-write because of unsupported optional features (%llx)",
3136 ret = btrfs_init_workqueues(fs_info, fs_devices);
3139 goto fail_sb_buffer;
3142 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3143 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3145 sb->s_blocksize = sectorsize;
3146 sb->s_blocksize_bits = blksize_bits(sectorsize);
3147 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3149 mutex_lock(&fs_info->chunk_mutex);
3150 ret = btrfs_read_sys_array(fs_info);
3151 mutex_unlock(&fs_info->chunk_mutex);
3153 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3154 goto fail_sb_buffer;
3157 generation = btrfs_super_chunk_root_generation(disk_super);
3158 level = btrfs_super_chunk_root_level(disk_super);
3160 chunk_root->node = read_tree_block(fs_info,
3161 btrfs_super_chunk_root(disk_super),
3162 BTRFS_CHUNK_TREE_OBJECTID,
3163 generation, level, NULL);
3164 if (IS_ERR(chunk_root->node) ||
3165 !extent_buffer_uptodate(chunk_root->node)) {
3166 btrfs_err(fs_info, "failed to read chunk root");
3167 if (!IS_ERR(chunk_root->node))
3168 free_extent_buffer(chunk_root->node);
3169 chunk_root->node = NULL;
3170 goto fail_tree_roots;
3172 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3173 chunk_root->commit_root = btrfs_root_node(chunk_root);
3175 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3176 offsetof(struct btrfs_header, chunk_tree_uuid),
3179 ret = btrfs_read_chunk_tree(fs_info);
3181 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3182 goto fail_tree_roots;
3186 * At this point we know all the devices that make this filesystem,
3187 * including the seed devices but we don't know yet if the replace
3188 * target is required. So free devices that are not part of this
3189 * filesystem but skip the replace traget device which is checked
3190 * below in btrfs_init_dev_replace().
3192 btrfs_free_extra_devids(fs_devices);
3193 if (!fs_devices->latest_bdev) {
3194 btrfs_err(fs_info, "failed to read devices");
3195 goto fail_tree_roots;
3198 ret = init_tree_roots(fs_info);
3200 goto fail_tree_roots;
3203 * If we have a uuid root and we're not being told to rescan we need to
3204 * check the generation here so we can set the
3205 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3206 * transaction during a balance or the log replay without updating the
3207 * uuid generation, and then if we crash we would rescan the uuid tree,
3208 * even though it was perfectly fine.
3210 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3211 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3212 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3214 ret = btrfs_verify_dev_extents(fs_info);
3217 "failed to verify dev extents against chunks: %d",
3219 goto fail_block_groups;
3221 ret = btrfs_recover_balance(fs_info);
3223 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3224 goto fail_block_groups;
3227 ret = btrfs_init_dev_stats(fs_info);
3229 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3230 goto fail_block_groups;
3233 ret = btrfs_init_dev_replace(fs_info);
3235 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3236 goto fail_block_groups;
3239 ret = btrfs_check_zoned_mode(fs_info);
3241 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3243 goto fail_block_groups;
3246 ret = btrfs_sysfs_add_fsid(fs_devices);
3248 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3250 goto fail_block_groups;
3253 ret = btrfs_sysfs_add_mounted(fs_info);
3255 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3256 goto fail_fsdev_sysfs;
3259 ret = btrfs_init_space_info(fs_info);
3261 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3265 ret = btrfs_read_block_groups(fs_info);
3267 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3271 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3273 "writable mount is not allowed due to too many missing devices");
3277 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3279 if (IS_ERR(fs_info->cleaner_kthread))
3282 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3284 "btrfs-transaction");
3285 if (IS_ERR(fs_info->transaction_kthread))
3288 if (!btrfs_test_opt(fs_info, NOSSD) &&
3289 !fs_info->fs_devices->rotating) {
3290 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3294 * Mount does not set all options immediately, we can do it now and do
3295 * not have to wait for transaction commit
3297 btrfs_apply_pending_changes(fs_info);
3299 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3300 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3301 ret = btrfsic_mount(fs_info, fs_devices,
3302 btrfs_test_opt(fs_info,
3303 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3305 fs_info->check_integrity_print_mask);
3308 "failed to initialize integrity check module: %d",
3312 ret = btrfs_read_qgroup_config(fs_info);
3314 goto fail_trans_kthread;
3316 if (btrfs_build_ref_tree(fs_info))
3317 btrfs_err(fs_info, "couldn't build ref tree");
3319 /* do not make disk changes in broken FS or nologreplay is given */
3320 if (btrfs_super_log_root(disk_super) != 0 &&
3321 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3322 btrfs_info(fs_info, "start tree-log replay");
3323 ret = btrfs_replay_log(fs_info, fs_devices);
3330 ret = btrfs_find_orphan_roots(fs_info);
3334 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3335 if (IS_ERR(fs_info->fs_root)) {
3336 err = PTR_ERR(fs_info->fs_root);
3337 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3338 fs_info->fs_root = NULL;
3345 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3346 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3347 clear_free_space_tree = 1;
3348 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3349 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3350 btrfs_warn(fs_info, "free space tree is invalid");
3351 clear_free_space_tree = 1;
3354 if (clear_free_space_tree) {
3355 btrfs_info(fs_info, "clearing free space tree");
3356 ret = btrfs_clear_free_space_tree(fs_info);
3359 "failed to clear free space tree: %d", ret);
3360 close_ctree(fs_info);
3365 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3366 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3367 btrfs_info(fs_info, "creating free space tree");
3368 ret = btrfs_create_free_space_tree(fs_info);
3371 "failed to create free space tree: %d", ret);
3372 close_ctree(fs_info);
3377 down_read(&fs_info->cleanup_work_sem);
3378 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3379 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3380 up_read(&fs_info->cleanup_work_sem);
3381 close_ctree(fs_info);
3384 up_read(&fs_info->cleanup_work_sem);
3386 ret = btrfs_start_pre_rw_mount(fs_info);
3388 close_ctree(fs_info);
3391 btrfs_discard_resume(fs_info);
3393 if (fs_info->uuid_root &&
3394 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3395 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3396 btrfs_info(fs_info, "checking UUID tree");
3397 ret = btrfs_check_uuid_tree(fs_info);
3400 "failed to check the UUID tree: %d", ret);
3401 close_ctree(fs_info);
3405 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3410 btrfs_free_qgroup_config(fs_info);
3412 kthread_stop(fs_info->transaction_kthread);
3413 btrfs_cleanup_transaction(fs_info);
3414 btrfs_free_fs_roots(fs_info);
3416 kthread_stop(fs_info->cleaner_kthread);
3419 * make sure we're done with the btree inode before we stop our
3422 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3425 btrfs_sysfs_remove_mounted(fs_info);
3428 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3431 btrfs_put_block_group_cache(fs_info);
3434 if (fs_info->data_reloc_root)
3435 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3436 free_root_pointers(fs_info, true);
3437 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3440 btrfs_stop_all_workers(fs_info);
3441 btrfs_free_block_groups(fs_info);
3443 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3445 iput(fs_info->btree_inode);
3447 btrfs_close_devices(fs_info->fs_devices);
3450 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3452 static void btrfs_end_super_write(struct bio *bio)
3454 struct btrfs_device *device = bio->bi_private;
3455 struct bio_vec *bvec;
3456 struct bvec_iter_all iter_all;
3459 bio_for_each_segment_all(bvec, bio, iter_all) {
3460 page = bvec->bv_page;
3462 if (bio->bi_status) {
3463 btrfs_warn_rl_in_rcu(device->fs_info,
3464 "lost page write due to IO error on %s (%d)",
3465 rcu_str_deref(device->name),
3466 blk_status_to_errno(bio->bi_status));
3467 ClearPageUptodate(page);
3469 btrfs_dev_stat_inc_and_print(device,
3470 BTRFS_DEV_STAT_WRITE_ERRS);
3472 SetPageUptodate(page);
3482 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3485 struct btrfs_super_block *super;
3487 u64 bytenr, bytenr_orig;
3488 struct address_space *mapping = bdev->bd_inode->i_mapping;
3491 bytenr_orig = btrfs_sb_offset(copy_num);
3492 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3494 return ERR_PTR(-EINVAL);
3496 return ERR_PTR(ret);
3498 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3499 return ERR_PTR(-EINVAL);
3501 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3503 return ERR_CAST(page);
3505 super = page_address(page);
3506 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3507 btrfs_release_disk_super(super);
3508 return ERR_PTR(-ENODATA);
3511 if (btrfs_super_bytenr(super) != bytenr_orig) {
3512 btrfs_release_disk_super(super);
3513 return ERR_PTR(-EINVAL);
3520 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3522 struct btrfs_super_block *super, *latest = NULL;
3526 /* we would like to check all the supers, but that would make
3527 * a btrfs mount succeed after a mkfs from a different FS.
3528 * So, we need to add a special mount option to scan for
3529 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3531 for (i = 0; i < 1; i++) {
3532 super = btrfs_read_dev_one_super(bdev, i);
3536 if (!latest || btrfs_super_generation(super) > transid) {
3538 btrfs_release_disk_super(super);
3541 transid = btrfs_super_generation(super);
3549 * Write superblock @sb to the @device. Do not wait for completion, all the
3550 * pages we use for writing are locked.
3552 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3553 * the expected device size at commit time. Note that max_mirrors must be
3554 * same for write and wait phases.
3556 * Return number of errors when page is not found or submission fails.
3558 static int write_dev_supers(struct btrfs_device *device,
3559 struct btrfs_super_block *sb, int max_mirrors)
3561 struct btrfs_fs_info *fs_info = device->fs_info;
3562 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3563 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3567 u64 bytenr, bytenr_orig;
3569 if (max_mirrors == 0)
3570 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3572 shash->tfm = fs_info->csum_shash;
3574 for (i = 0; i < max_mirrors; i++) {
3577 struct btrfs_super_block *disk_super;
3579 bytenr_orig = btrfs_sb_offset(i);
3580 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3581 if (ret == -ENOENT) {
3583 } else if (ret < 0) {
3584 btrfs_err(device->fs_info,
3585 "couldn't get super block location for mirror %d",
3590 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3591 device->commit_total_bytes)
3594 btrfs_set_super_bytenr(sb, bytenr_orig);
3596 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3597 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3600 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3603 btrfs_err(device->fs_info,
3604 "couldn't get super block page for bytenr %llu",
3610 /* Bump the refcount for wait_dev_supers() */
3613 disk_super = page_address(page);
3614 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3617 * Directly use bios here instead of relying on the page cache
3618 * to do I/O, so we don't lose the ability to do integrity
3621 bio = bio_alloc(GFP_NOFS, 1);
3622 bio_set_dev(bio, device->bdev);
3623 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3624 bio->bi_private = device;
3625 bio->bi_end_io = btrfs_end_super_write;
3626 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3627 offset_in_page(bytenr));
3630 * We FUA only the first super block. The others we allow to
3631 * go down lazy and there's a short window where the on-disk
3632 * copies might still contain the older version.
3634 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3635 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3636 bio->bi_opf |= REQ_FUA;
3638 btrfsic_submit_bio(bio);
3639 btrfs_advance_sb_log(device, i);
3641 return errors < i ? 0 : -1;
3645 * Wait for write completion of superblocks done by write_dev_supers,
3646 * @max_mirrors same for write and wait phases.
3648 * Return number of errors when page is not found or not marked up to
3651 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3655 bool primary_failed = false;
3659 if (max_mirrors == 0)
3660 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3662 for (i = 0; i < max_mirrors; i++) {
3665 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3666 if (ret == -ENOENT) {
3668 } else if (ret < 0) {
3671 primary_failed = true;
3674 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3675 device->commit_total_bytes)
3678 page = find_get_page(device->bdev->bd_inode->i_mapping,
3679 bytenr >> PAGE_SHIFT);
3683 primary_failed = true;
3686 /* Page is submitted locked and unlocked once the IO completes */
3687 wait_on_page_locked(page);
3688 if (PageError(page)) {
3691 primary_failed = true;
3694 /* Drop our reference */
3697 /* Drop the reference from the writing run */
3701 /* log error, force error return */
3702 if (primary_failed) {
3703 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3708 return errors < i ? 0 : -1;
3712 * endio for the write_dev_flush, this will wake anyone waiting
3713 * for the barrier when it is done
3715 static void btrfs_end_empty_barrier(struct bio *bio)
3717 complete(bio->bi_private);
3721 * Submit a flush request to the device if it supports it. Error handling is
3722 * done in the waiting counterpart.
3724 static void write_dev_flush(struct btrfs_device *device)
3726 struct request_queue *q = bdev_get_queue(device->bdev);
3727 struct bio *bio = device->flush_bio;
3729 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3733 bio->bi_end_io = btrfs_end_empty_barrier;
3734 bio_set_dev(bio, device->bdev);
3735 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3736 init_completion(&device->flush_wait);
3737 bio->bi_private = &device->flush_wait;
3739 btrfsic_submit_bio(bio);
3740 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3744 * If the flush bio has been submitted by write_dev_flush, wait for it.
3746 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3748 struct bio *bio = device->flush_bio;
3750 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3753 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3754 wait_for_completion_io(&device->flush_wait);
3756 return bio->bi_status;
3759 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3761 if (!btrfs_check_rw_degradable(fs_info, NULL))
3767 * send an empty flush down to each device in parallel,
3768 * then wait for them
3770 static int barrier_all_devices(struct btrfs_fs_info *info)
3772 struct list_head *head;
3773 struct btrfs_device *dev;
3774 int errors_wait = 0;
3777 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3778 /* send down all the barriers */
3779 head = &info->fs_devices->devices;
3780 list_for_each_entry(dev, head, dev_list) {
3781 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3785 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3786 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3789 write_dev_flush(dev);
3790 dev->last_flush_error = BLK_STS_OK;
3793 /* wait for all the barriers */
3794 list_for_each_entry(dev, head, dev_list) {
3795 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3801 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3802 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3805 ret = wait_dev_flush(dev);
3807 dev->last_flush_error = ret;
3808 btrfs_dev_stat_inc_and_print(dev,
3809 BTRFS_DEV_STAT_FLUSH_ERRS);
3816 * At some point we need the status of all disks
3817 * to arrive at the volume status. So error checking
3818 * is being pushed to a separate loop.
3820 return check_barrier_error(info);
3825 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3828 int min_tolerated = INT_MAX;
3830 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3831 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3832 min_tolerated = min_t(int, min_tolerated,
3833 btrfs_raid_array[BTRFS_RAID_SINGLE].
3834 tolerated_failures);
3836 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3837 if (raid_type == BTRFS_RAID_SINGLE)
3839 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3841 min_tolerated = min_t(int, min_tolerated,
3842 btrfs_raid_array[raid_type].
3843 tolerated_failures);
3846 if (min_tolerated == INT_MAX) {
3847 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3851 return min_tolerated;
3854 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3856 struct list_head *head;
3857 struct btrfs_device *dev;
3858 struct btrfs_super_block *sb;
3859 struct btrfs_dev_item *dev_item;
3863 int total_errors = 0;
3866 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3869 * max_mirrors == 0 indicates we're from commit_transaction,
3870 * not from fsync where the tree roots in fs_info have not
3871 * been consistent on disk.
3873 if (max_mirrors == 0)
3874 backup_super_roots(fs_info);
3876 sb = fs_info->super_for_commit;
3877 dev_item = &sb->dev_item;
3879 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3880 head = &fs_info->fs_devices->devices;
3881 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3884 ret = barrier_all_devices(fs_info);
3887 &fs_info->fs_devices->device_list_mutex);
3888 btrfs_handle_fs_error(fs_info, ret,
3889 "errors while submitting device barriers.");
3894 list_for_each_entry(dev, head, dev_list) {
3899 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3900 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3903 btrfs_set_stack_device_generation(dev_item, 0);
3904 btrfs_set_stack_device_type(dev_item, dev->type);
3905 btrfs_set_stack_device_id(dev_item, dev->devid);
3906 btrfs_set_stack_device_total_bytes(dev_item,
3907 dev->commit_total_bytes);
3908 btrfs_set_stack_device_bytes_used(dev_item,
3909 dev->commit_bytes_used);
3910 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3911 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3912 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3913 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3914 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3917 flags = btrfs_super_flags(sb);
3918 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3920 ret = btrfs_validate_write_super(fs_info, sb);
3922 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3923 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3924 "unexpected superblock corruption detected");
3928 ret = write_dev_supers(dev, sb, max_mirrors);
3932 if (total_errors > max_errors) {
3933 btrfs_err(fs_info, "%d errors while writing supers",
3935 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3937 /* FUA is masked off if unsupported and can't be the reason */
3938 btrfs_handle_fs_error(fs_info, -EIO,
3939 "%d errors while writing supers",
3945 list_for_each_entry(dev, head, dev_list) {
3948 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3949 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3952 ret = wait_dev_supers(dev, max_mirrors);
3956 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3957 if (total_errors > max_errors) {
3958 btrfs_handle_fs_error(fs_info, -EIO,
3959 "%d errors while writing supers",
3966 /* Drop a fs root from the radix tree and free it. */
3967 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3968 struct btrfs_root *root)
3970 bool drop_ref = false;
3972 spin_lock(&fs_info->fs_roots_radix_lock);
3973 radix_tree_delete(&fs_info->fs_roots_radix,
3974 (unsigned long)root->root_key.objectid);
3975 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3977 spin_unlock(&fs_info->fs_roots_radix_lock);
3979 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3980 ASSERT(root->log_root == NULL);
3981 if (root->reloc_root) {
3982 btrfs_put_root(root->reloc_root);
3983 root->reloc_root = NULL;
3988 btrfs_put_root(root);
3991 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3993 u64 root_objectid = 0;
3994 struct btrfs_root *gang[8];
3997 unsigned int ret = 0;
4000 spin_lock(&fs_info->fs_roots_radix_lock);
4001 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4002 (void **)gang, root_objectid,
4005 spin_unlock(&fs_info->fs_roots_radix_lock);
4008 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4010 for (i = 0; i < ret; i++) {
4011 /* Avoid to grab roots in dead_roots */
4012 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4016 /* grab all the search result for later use */
4017 gang[i] = btrfs_grab_root(gang[i]);
4019 spin_unlock(&fs_info->fs_roots_radix_lock);
4021 for (i = 0; i < ret; i++) {
4024 root_objectid = gang[i]->root_key.objectid;
4025 err = btrfs_orphan_cleanup(gang[i]);
4028 btrfs_put_root(gang[i]);
4033 /* release the uncleaned roots due to error */
4034 for (; i < ret; i++) {
4036 btrfs_put_root(gang[i]);
4041 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4043 struct btrfs_root *root = fs_info->tree_root;
4044 struct btrfs_trans_handle *trans;
4046 mutex_lock(&fs_info->cleaner_mutex);
4047 btrfs_run_delayed_iputs(fs_info);
4048 mutex_unlock(&fs_info->cleaner_mutex);
4049 wake_up_process(fs_info->cleaner_kthread);
4051 /* wait until ongoing cleanup work done */
4052 down_write(&fs_info->cleanup_work_sem);
4053 up_write(&fs_info->cleanup_work_sem);
4055 trans = btrfs_join_transaction(root);
4057 return PTR_ERR(trans);
4058 return btrfs_commit_transaction(trans);
4061 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4065 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4067 * We don't want the cleaner to start new transactions, add more delayed
4068 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4069 * because that frees the task_struct, and the transaction kthread might
4070 * still try to wake up the cleaner.
4072 kthread_park(fs_info->cleaner_kthread);
4074 /* wait for the qgroup rescan worker to stop */
4075 btrfs_qgroup_wait_for_completion(fs_info, false);
4077 /* wait for the uuid_scan task to finish */
4078 down(&fs_info->uuid_tree_rescan_sem);
4079 /* avoid complains from lockdep et al., set sem back to initial state */
4080 up(&fs_info->uuid_tree_rescan_sem);
4082 /* pause restriper - we want to resume on mount */
4083 btrfs_pause_balance(fs_info);
4085 btrfs_dev_replace_suspend_for_unmount(fs_info);
4087 btrfs_scrub_cancel(fs_info);
4089 /* wait for any defraggers to finish */
4090 wait_event(fs_info->transaction_wait,
4091 (atomic_read(&fs_info->defrag_running) == 0));
4093 /* clear out the rbtree of defraggable inodes */
4094 btrfs_cleanup_defrag_inodes(fs_info);
4096 cancel_work_sync(&fs_info->async_reclaim_work);
4097 cancel_work_sync(&fs_info->async_data_reclaim_work);
4099 /* Cancel or finish ongoing discard work */
4100 btrfs_discard_cleanup(fs_info);
4102 if (!sb_rdonly(fs_info->sb)) {
4104 * The cleaner kthread is stopped, so do one final pass over
4105 * unused block groups.
4107 btrfs_delete_unused_bgs(fs_info);
4110 * There might be existing delayed inode workers still running
4111 * and holding an empty delayed inode item. We must wait for
4112 * them to complete first because they can create a transaction.
4113 * This happens when someone calls btrfs_balance_delayed_items()
4114 * and then a transaction commit runs the same delayed nodes
4115 * before any delayed worker has done something with the nodes.
4116 * We must wait for any worker here and not at transaction
4117 * commit time since that could cause a deadlock.
4118 * This is a very rare case.
4120 btrfs_flush_workqueue(fs_info->delayed_workers);
4122 ret = btrfs_commit_super(fs_info);
4124 btrfs_err(fs_info, "commit super ret %d", ret);
4127 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4128 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4129 btrfs_error_commit_super(fs_info);
4131 kthread_stop(fs_info->transaction_kthread);
4132 kthread_stop(fs_info->cleaner_kthread);
4134 ASSERT(list_empty(&fs_info->delayed_iputs));
4135 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4137 if (btrfs_check_quota_leak(fs_info)) {
4138 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4139 btrfs_err(fs_info, "qgroup reserved space leaked");
4142 btrfs_free_qgroup_config(fs_info);
4143 ASSERT(list_empty(&fs_info->delalloc_roots));
4145 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4146 btrfs_info(fs_info, "at unmount delalloc count %lld",
4147 percpu_counter_sum(&fs_info->delalloc_bytes));
4150 if (percpu_counter_sum(&fs_info->dio_bytes))
4151 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4152 percpu_counter_sum(&fs_info->dio_bytes));
4154 btrfs_sysfs_remove_mounted(fs_info);
4155 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4157 btrfs_put_block_group_cache(fs_info);
4160 * we must make sure there is not any read request to
4161 * submit after we stopping all workers.
4163 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4164 btrfs_stop_all_workers(fs_info);
4166 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4167 free_root_pointers(fs_info, true);
4168 btrfs_free_fs_roots(fs_info);
4171 * We must free the block groups after dropping the fs_roots as we could
4172 * have had an IO error and have left over tree log blocks that aren't
4173 * cleaned up until the fs roots are freed. This makes the block group
4174 * accounting appear to be wrong because there's pending reserved bytes,
4175 * so make sure we do the block group cleanup afterwards.
4177 btrfs_free_block_groups(fs_info);
4179 iput(fs_info->btree_inode);
4181 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4182 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4183 btrfsic_unmount(fs_info->fs_devices);
4186 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4187 btrfs_close_devices(fs_info->fs_devices);
4190 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4194 struct inode *btree_inode = buf->pages[0]->mapping->host;
4196 ret = extent_buffer_uptodate(buf);
4200 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4201 parent_transid, atomic);
4207 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4209 struct btrfs_fs_info *fs_info = buf->fs_info;
4210 u64 transid = btrfs_header_generation(buf);
4213 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4215 * This is a fast path so only do this check if we have sanity tests
4216 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4217 * outside of the sanity tests.
4219 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4222 btrfs_assert_tree_locked(buf);
4223 if (transid != fs_info->generation)
4224 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4225 buf->start, transid, fs_info->generation);
4226 was_dirty = set_extent_buffer_dirty(buf);
4228 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4230 fs_info->dirty_metadata_batch);
4231 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4233 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4234 * but item data not updated.
4235 * So here we should only check item pointers, not item data.
4237 if (btrfs_header_level(buf) == 0 &&
4238 btrfs_check_leaf_relaxed(buf)) {
4239 btrfs_print_leaf(buf);
4245 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4249 * looks as though older kernels can get into trouble with
4250 * this code, they end up stuck in balance_dirty_pages forever
4254 if (current->flags & PF_MEMALLOC)
4258 btrfs_balance_delayed_items(fs_info);
4260 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4261 BTRFS_DIRTY_METADATA_THRESH,
4262 fs_info->dirty_metadata_batch);
4264 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4268 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4270 __btrfs_btree_balance_dirty(fs_info, 1);
4273 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4275 __btrfs_btree_balance_dirty(fs_info, 0);
4278 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4279 struct btrfs_key *first_key)
4281 return btree_read_extent_buffer_pages(buf, parent_transid,
4285 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4287 /* cleanup FS via transaction */
4288 btrfs_cleanup_transaction(fs_info);
4290 mutex_lock(&fs_info->cleaner_mutex);
4291 btrfs_run_delayed_iputs(fs_info);
4292 mutex_unlock(&fs_info->cleaner_mutex);
4294 down_write(&fs_info->cleanup_work_sem);
4295 up_write(&fs_info->cleanup_work_sem);
4298 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4300 struct btrfs_root *gang[8];
4301 u64 root_objectid = 0;
4304 spin_lock(&fs_info->fs_roots_radix_lock);
4305 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4306 (void **)gang, root_objectid,
4307 ARRAY_SIZE(gang))) != 0) {
4310 for (i = 0; i < ret; i++)
4311 gang[i] = btrfs_grab_root(gang[i]);
4312 spin_unlock(&fs_info->fs_roots_radix_lock);
4314 for (i = 0; i < ret; i++) {
4317 root_objectid = gang[i]->root_key.objectid;
4318 btrfs_free_log(NULL, gang[i]);
4319 btrfs_put_root(gang[i]);
4322 spin_lock(&fs_info->fs_roots_radix_lock);
4324 spin_unlock(&fs_info->fs_roots_radix_lock);
4325 btrfs_free_log_root_tree(NULL, fs_info);
4328 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4330 struct btrfs_ordered_extent *ordered;
4332 spin_lock(&root->ordered_extent_lock);
4334 * This will just short circuit the ordered completion stuff which will
4335 * make sure the ordered extent gets properly cleaned up.
4337 list_for_each_entry(ordered, &root->ordered_extents,
4339 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4340 spin_unlock(&root->ordered_extent_lock);
4343 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4345 struct btrfs_root *root;
4346 struct list_head splice;
4348 INIT_LIST_HEAD(&splice);
4350 spin_lock(&fs_info->ordered_root_lock);
4351 list_splice_init(&fs_info->ordered_roots, &splice);
4352 while (!list_empty(&splice)) {
4353 root = list_first_entry(&splice, struct btrfs_root,
4355 list_move_tail(&root->ordered_root,
4356 &fs_info->ordered_roots);
4358 spin_unlock(&fs_info->ordered_root_lock);
4359 btrfs_destroy_ordered_extents(root);
4362 spin_lock(&fs_info->ordered_root_lock);
4364 spin_unlock(&fs_info->ordered_root_lock);
4367 * We need this here because if we've been flipped read-only we won't
4368 * get sync() from the umount, so we need to make sure any ordered
4369 * extents that haven't had their dirty pages IO start writeout yet
4370 * actually get run and error out properly.
4372 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4375 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4376 struct btrfs_fs_info *fs_info)
4378 struct rb_node *node;
4379 struct btrfs_delayed_ref_root *delayed_refs;
4380 struct btrfs_delayed_ref_node *ref;
4383 delayed_refs = &trans->delayed_refs;
4385 spin_lock(&delayed_refs->lock);
4386 if (atomic_read(&delayed_refs->num_entries) == 0) {
4387 spin_unlock(&delayed_refs->lock);
4388 btrfs_debug(fs_info, "delayed_refs has NO entry");
4392 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4393 struct btrfs_delayed_ref_head *head;
4395 bool pin_bytes = false;
4397 head = rb_entry(node, struct btrfs_delayed_ref_head,
4399 if (btrfs_delayed_ref_lock(delayed_refs, head))
4402 spin_lock(&head->lock);
4403 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4404 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4407 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4408 RB_CLEAR_NODE(&ref->ref_node);
4409 if (!list_empty(&ref->add_list))
4410 list_del(&ref->add_list);
4411 atomic_dec(&delayed_refs->num_entries);
4412 btrfs_put_delayed_ref(ref);
4414 if (head->must_insert_reserved)
4416 btrfs_free_delayed_extent_op(head->extent_op);
4417 btrfs_delete_ref_head(delayed_refs, head);
4418 spin_unlock(&head->lock);
4419 spin_unlock(&delayed_refs->lock);
4420 mutex_unlock(&head->mutex);
4423 struct btrfs_block_group *cache;
4425 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4428 spin_lock(&cache->space_info->lock);
4429 spin_lock(&cache->lock);
4430 cache->pinned += head->num_bytes;
4431 btrfs_space_info_update_bytes_pinned(fs_info,
4432 cache->space_info, head->num_bytes);
4433 cache->reserved -= head->num_bytes;
4434 cache->space_info->bytes_reserved -= head->num_bytes;
4435 spin_unlock(&cache->lock);
4436 spin_unlock(&cache->space_info->lock);
4437 percpu_counter_add_batch(
4438 &cache->space_info->total_bytes_pinned,
4439 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4441 btrfs_put_block_group(cache);
4443 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4444 head->bytenr + head->num_bytes - 1);
4446 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4447 btrfs_put_delayed_ref_head(head);
4449 spin_lock(&delayed_refs->lock);
4451 btrfs_qgroup_destroy_extent_records(trans);
4453 spin_unlock(&delayed_refs->lock);
4458 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4460 struct btrfs_inode *btrfs_inode;
4461 struct list_head splice;
4463 INIT_LIST_HEAD(&splice);
4465 spin_lock(&root->delalloc_lock);
4466 list_splice_init(&root->delalloc_inodes, &splice);
4468 while (!list_empty(&splice)) {
4469 struct inode *inode = NULL;
4470 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4472 __btrfs_del_delalloc_inode(root, btrfs_inode);
4473 spin_unlock(&root->delalloc_lock);
4476 * Make sure we get a live inode and that it'll not disappear
4479 inode = igrab(&btrfs_inode->vfs_inode);
4481 invalidate_inode_pages2(inode->i_mapping);
4484 spin_lock(&root->delalloc_lock);
4486 spin_unlock(&root->delalloc_lock);
4489 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4491 struct btrfs_root *root;
4492 struct list_head splice;
4494 INIT_LIST_HEAD(&splice);
4496 spin_lock(&fs_info->delalloc_root_lock);
4497 list_splice_init(&fs_info->delalloc_roots, &splice);
4498 while (!list_empty(&splice)) {
4499 root = list_first_entry(&splice, struct btrfs_root,
4501 root = btrfs_grab_root(root);
4503 spin_unlock(&fs_info->delalloc_root_lock);
4505 btrfs_destroy_delalloc_inodes(root);
4506 btrfs_put_root(root);
4508 spin_lock(&fs_info->delalloc_root_lock);
4510 spin_unlock(&fs_info->delalloc_root_lock);
4513 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4514 struct extent_io_tree *dirty_pages,
4518 struct extent_buffer *eb;
4523 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4528 clear_extent_bits(dirty_pages, start, end, mark);
4529 while (start <= end) {
4530 eb = find_extent_buffer(fs_info, start);
4531 start += fs_info->nodesize;
4534 wait_on_extent_buffer_writeback(eb);
4536 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4538 clear_extent_buffer_dirty(eb);
4539 free_extent_buffer_stale(eb);
4546 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4547 struct extent_io_tree *unpin)
4554 struct extent_state *cached_state = NULL;
4557 * The btrfs_finish_extent_commit() may get the same range as
4558 * ours between find_first_extent_bit and clear_extent_dirty.
4559 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4560 * the same extent range.
4562 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4563 ret = find_first_extent_bit(unpin, 0, &start, &end,
4564 EXTENT_DIRTY, &cached_state);
4566 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4570 clear_extent_dirty(unpin, start, end, &cached_state);
4571 free_extent_state(cached_state);
4572 btrfs_error_unpin_extent_range(fs_info, start, end);
4573 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4580 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4582 struct inode *inode;
4584 inode = cache->io_ctl.inode;
4586 invalidate_inode_pages2(inode->i_mapping);
4587 BTRFS_I(inode)->generation = 0;
4588 cache->io_ctl.inode = NULL;
4591 ASSERT(cache->io_ctl.pages == NULL);
4592 btrfs_put_block_group(cache);
4595 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4596 struct btrfs_fs_info *fs_info)
4598 struct btrfs_block_group *cache;
4600 spin_lock(&cur_trans->dirty_bgs_lock);
4601 while (!list_empty(&cur_trans->dirty_bgs)) {
4602 cache = list_first_entry(&cur_trans->dirty_bgs,
4603 struct btrfs_block_group,
4606 if (!list_empty(&cache->io_list)) {
4607 spin_unlock(&cur_trans->dirty_bgs_lock);
4608 list_del_init(&cache->io_list);
4609 btrfs_cleanup_bg_io(cache);
4610 spin_lock(&cur_trans->dirty_bgs_lock);
4613 list_del_init(&cache->dirty_list);
4614 spin_lock(&cache->lock);
4615 cache->disk_cache_state = BTRFS_DC_ERROR;
4616 spin_unlock(&cache->lock);
4618 spin_unlock(&cur_trans->dirty_bgs_lock);
4619 btrfs_put_block_group(cache);
4620 btrfs_delayed_refs_rsv_release(fs_info, 1);
4621 spin_lock(&cur_trans->dirty_bgs_lock);
4623 spin_unlock(&cur_trans->dirty_bgs_lock);
4626 * Refer to the definition of io_bgs member for details why it's safe
4627 * to use it without any locking
4629 while (!list_empty(&cur_trans->io_bgs)) {
4630 cache = list_first_entry(&cur_trans->io_bgs,
4631 struct btrfs_block_group,
4634 list_del_init(&cache->io_list);
4635 spin_lock(&cache->lock);
4636 cache->disk_cache_state = BTRFS_DC_ERROR;
4637 spin_unlock(&cache->lock);
4638 btrfs_cleanup_bg_io(cache);
4642 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4643 struct btrfs_fs_info *fs_info)
4645 struct btrfs_device *dev, *tmp;
4647 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4648 ASSERT(list_empty(&cur_trans->dirty_bgs));
4649 ASSERT(list_empty(&cur_trans->io_bgs));
4651 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4653 list_del_init(&dev->post_commit_list);
4656 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4658 cur_trans->state = TRANS_STATE_COMMIT_START;
4659 wake_up(&fs_info->transaction_blocked_wait);
4661 cur_trans->state = TRANS_STATE_UNBLOCKED;
4662 wake_up(&fs_info->transaction_wait);
4664 btrfs_destroy_delayed_inodes(fs_info);
4666 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4668 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4670 cur_trans->state =TRANS_STATE_COMPLETED;
4671 wake_up(&cur_trans->commit_wait);
4674 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4676 struct btrfs_transaction *t;
4678 mutex_lock(&fs_info->transaction_kthread_mutex);
4680 spin_lock(&fs_info->trans_lock);
4681 while (!list_empty(&fs_info->trans_list)) {
4682 t = list_first_entry(&fs_info->trans_list,
4683 struct btrfs_transaction, list);
4684 if (t->state >= TRANS_STATE_COMMIT_START) {
4685 refcount_inc(&t->use_count);
4686 spin_unlock(&fs_info->trans_lock);
4687 btrfs_wait_for_commit(fs_info, t->transid);
4688 btrfs_put_transaction(t);
4689 spin_lock(&fs_info->trans_lock);
4692 if (t == fs_info->running_transaction) {
4693 t->state = TRANS_STATE_COMMIT_DOING;
4694 spin_unlock(&fs_info->trans_lock);
4696 * We wait for 0 num_writers since we don't hold a trans
4697 * handle open currently for this transaction.
4699 wait_event(t->writer_wait,
4700 atomic_read(&t->num_writers) == 0);
4702 spin_unlock(&fs_info->trans_lock);
4704 btrfs_cleanup_one_transaction(t, fs_info);
4706 spin_lock(&fs_info->trans_lock);
4707 if (t == fs_info->running_transaction)
4708 fs_info->running_transaction = NULL;
4709 list_del_init(&t->list);
4710 spin_unlock(&fs_info->trans_lock);
4712 btrfs_put_transaction(t);
4713 trace_btrfs_transaction_commit(fs_info->tree_root);
4714 spin_lock(&fs_info->trans_lock);
4716 spin_unlock(&fs_info->trans_lock);
4717 btrfs_destroy_all_ordered_extents(fs_info);
4718 btrfs_destroy_delayed_inodes(fs_info);
4719 btrfs_assert_delayed_root_empty(fs_info);
4720 btrfs_destroy_all_delalloc_inodes(fs_info);
4721 btrfs_drop_all_logs(fs_info);
4722 mutex_unlock(&fs_info->transaction_kthread_mutex);
4727 int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
4729 struct btrfs_path *path;
4731 struct extent_buffer *l;
4732 struct btrfs_key search_key;
4733 struct btrfs_key found_key;
4736 path = btrfs_alloc_path();
4740 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4741 search_key.type = -1;
4742 search_key.offset = (u64)-1;
4743 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4746 BUG_ON(ret == 0); /* Corruption */
4747 if (path->slots[0] > 0) {
4748 slot = path->slots[0] - 1;
4750 btrfs_item_key_to_cpu(l, &found_key, slot);
4751 *objectid = max_t(u64, found_key.objectid,
4752 BTRFS_FIRST_FREE_OBJECTID - 1);
4754 *objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
4758 btrfs_free_path(path);
4762 int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
4765 mutex_lock(&root->objectid_mutex);
4767 if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4768 btrfs_warn(root->fs_info,
4769 "the objectid of root %llu reaches its highest value",
4770 root->root_key.objectid);
4775 *objectid = ++root->highest_objectid;
4778 mutex_unlock(&root->objectid_mutex);