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 /* Optional parameter for submit_bio_start used by direct io */
119 struct btrfs_work work;
124 * Lockdep class keys for extent_buffer->lock's in this root. For a given
125 * eb, the lockdep key is determined by the btrfs_root it belongs to and
126 * the level the eb occupies in the tree.
128 * Different roots are used for different purposes and may nest inside each
129 * other and they require separate keysets. As lockdep keys should be
130 * static, assign keysets according to the purpose of the root as indicated
131 * by btrfs_root->root_key.objectid. This ensures that all special purpose
132 * roots have separate keysets.
134 * Lock-nesting across peer nodes is always done with the immediate parent
135 * node locked thus preventing deadlock. As lockdep doesn't know this, use
136 * subclass to avoid triggering lockdep warning in such cases.
138 * The key is set by the readpage_end_io_hook after the buffer has passed
139 * csum validation but before the pages are unlocked. It is also set by
140 * btrfs_init_new_buffer on freshly allocated blocks.
142 * We also add a check to make sure the highest level of the tree is the
143 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
144 * needs update as well.
146 #ifdef CONFIG_DEBUG_LOCK_ALLOC
147 # if BTRFS_MAX_LEVEL != 8
151 #define DEFINE_LEVEL(stem, level) \
152 .names[level] = "btrfs-" stem "-0" #level,
154 #define DEFINE_NAME(stem) \
155 DEFINE_LEVEL(stem, 0) \
156 DEFINE_LEVEL(stem, 1) \
157 DEFINE_LEVEL(stem, 2) \
158 DEFINE_LEVEL(stem, 3) \
159 DEFINE_LEVEL(stem, 4) \
160 DEFINE_LEVEL(stem, 5) \
161 DEFINE_LEVEL(stem, 6) \
162 DEFINE_LEVEL(stem, 7)
164 static struct btrfs_lockdep_keyset {
165 u64 id; /* root objectid */
166 /* Longest entry: btrfs-free-space-00 */
167 char names[BTRFS_MAX_LEVEL][20];
168 struct lock_class_key keys[BTRFS_MAX_LEVEL];
169 } btrfs_lockdep_keysets[] = {
170 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") },
171 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") },
172 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") },
173 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") },
174 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
175 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
176 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
177 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
178 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
179 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
180 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
181 { .id = 0, DEFINE_NAME("tree") },
187 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
190 struct btrfs_lockdep_keyset *ks;
192 BUG_ON(level >= ARRAY_SIZE(ks->keys));
194 /* find the matching keyset, id 0 is the default entry */
195 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
196 if (ks->id == objectid)
199 lockdep_set_class_and_name(&eb->lock,
200 &ks->keys[level], ks->names[level]);
206 * Compute the csum of a btree block and store the result to provided buffer.
208 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
210 struct btrfs_fs_info *fs_info = buf->fs_info;
211 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
212 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
213 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
217 shash->tfm = fs_info->csum_shash;
218 crypto_shash_init(shash);
219 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
220 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
221 first_page_part - BTRFS_CSUM_SIZE);
223 for (i = 1; i < num_pages; i++) {
224 kaddr = page_address(buf->pages[i]);
225 crypto_shash_update(shash, kaddr, PAGE_SIZE);
227 memset(result, 0, BTRFS_CSUM_SIZE);
228 crypto_shash_final(shash, result);
232 * we can't consider a given block up to date unless the transid of the
233 * block matches the transid in the parent node's pointer. This is how we
234 * detect blocks that either didn't get written at all or got written
235 * in the wrong place.
237 static int verify_parent_transid(struct extent_io_tree *io_tree,
238 struct extent_buffer *eb, u64 parent_transid,
241 struct extent_state *cached_state = NULL;
243 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
245 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
252 btrfs_tree_read_lock(eb);
254 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
256 if (extent_buffer_uptodate(eb) &&
257 btrfs_header_generation(eb) == parent_transid) {
261 btrfs_err_rl(eb->fs_info,
262 "parent transid verify failed on %llu wanted %llu found %llu",
264 parent_transid, btrfs_header_generation(eb));
268 * Things reading via commit roots that don't have normal protection,
269 * like send, can have a really old block in cache that may point at a
270 * block that has been freed and re-allocated. So don't clear uptodate
271 * if we find an eb that is under IO (dirty/writeback) because we could
272 * end up reading in the stale data and then writing it back out and
273 * making everybody very sad.
275 if (!extent_buffer_under_io(eb))
276 clear_extent_buffer_uptodate(eb);
278 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
281 btrfs_tree_read_unlock(eb);
285 static bool btrfs_supported_super_csum(u16 csum_type)
288 case BTRFS_CSUM_TYPE_CRC32:
289 case BTRFS_CSUM_TYPE_XXHASH:
290 case BTRFS_CSUM_TYPE_SHA256:
291 case BTRFS_CSUM_TYPE_BLAKE2:
299 * Return 0 if the superblock checksum type matches the checksum value of that
300 * algorithm. Pass the raw disk superblock data.
302 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
305 struct btrfs_super_block *disk_sb =
306 (struct btrfs_super_block *)raw_disk_sb;
307 char result[BTRFS_CSUM_SIZE];
308 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
310 shash->tfm = fs_info->csum_shash;
313 * The super_block structure does not span the whole
314 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
315 * filled with zeros and is included in the checksum.
317 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
318 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
320 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
326 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
327 struct btrfs_key *first_key, u64 parent_transid)
329 struct btrfs_fs_info *fs_info = eb->fs_info;
331 struct btrfs_key found_key;
334 found_level = btrfs_header_level(eb);
335 if (found_level != level) {
336 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
337 KERN_ERR "BTRFS: tree level check failed\n");
339 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
340 eb->start, level, found_level);
348 * For live tree block (new tree blocks in current transaction),
349 * we need proper lock context to avoid race, which is impossible here.
350 * So we only checks tree blocks which is read from disk, whose
351 * generation <= fs_info->last_trans_committed.
353 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
356 /* We have @first_key, so this @eb must have at least one item */
357 if (btrfs_header_nritems(eb) == 0) {
359 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
361 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
366 btrfs_node_key_to_cpu(eb, &found_key, 0);
368 btrfs_item_key_to_cpu(eb, &found_key, 0);
369 ret = btrfs_comp_cpu_keys(first_key, &found_key);
372 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
373 KERN_ERR "BTRFS: tree first key check failed\n");
375 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
376 eb->start, parent_transid, first_key->objectid,
377 first_key->type, first_key->offset,
378 found_key.objectid, found_key.type,
385 * helper to read a given tree block, doing retries as required when
386 * the checksums don't match and we have alternate mirrors to try.
388 * @parent_transid: expected transid, skip check if 0
389 * @level: expected level, mandatory check
390 * @first_key: expected key of first slot, skip check if NULL
392 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
393 u64 parent_transid, int level,
394 struct btrfs_key *first_key)
396 struct btrfs_fs_info *fs_info = eb->fs_info;
397 struct extent_io_tree *io_tree;
402 int failed_mirror = 0;
404 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
406 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
407 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
409 if (verify_parent_transid(io_tree, eb,
412 else if (btrfs_verify_level_key(eb, level,
413 first_key, parent_transid))
419 num_copies = btrfs_num_copies(fs_info,
424 if (!failed_mirror) {
426 failed_mirror = eb->read_mirror;
430 if (mirror_num == failed_mirror)
433 if (mirror_num > num_copies)
437 if (failed && !ret && failed_mirror)
438 btrfs_repair_eb_io_failure(eb, failed_mirror);
444 * Checksum a dirty tree block before IO. This has extra checks to make sure
445 * we only fill in the checksum field in the first page of a multi-page block.
446 * For subpage extent buffers we need bvec to also read the offset in the page.
448 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
450 struct page *page = bvec->bv_page;
451 u64 start = page_offset(page);
453 u8 result[BTRFS_CSUM_SIZE];
454 struct extent_buffer *eb;
457 eb = (struct extent_buffer *)page->private;
458 if (page != eb->pages[0])
461 found_start = btrfs_header_bytenr(eb);
463 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
464 WARN_ON(found_start != 0);
469 * Please do not consolidate these warnings into a single if.
470 * It is useful to know what went wrong.
472 if (WARN_ON(found_start != start))
474 if (WARN_ON(!PageUptodate(page)))
477 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
478 offsetof(struct btrfs_header, fsid),
479 BTRFS_FSID_SIZE) == 0);
481 csum_tree_block(eb, result);
483 if (btrfs_header_level(eb))
484 ret = btrfs_check_node(eb);
486 ret = btrfs_check_leaf_full(eb);
489 btrfs_print_tree(eb, 0);
491 "block=%llu write time tree block corruption detected",
493 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
496 write_extent_buffer(eb, result, 0, fs_info->csum_size);
501 static int check_tree_block_fsid(struct extent_buffer *eb)
503 struct btrfs_fs_info *fs_info = eb->fs_info;
504 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
505 u8 fsid[BTRFS_FSID_SIZE];
508 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
511 * Checking the incompat flag is only valid for the current fs. For
512 * seed devices it's forbidden to have their uuid changed so reading
513 * ->fsid in this case is fine
515 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
516 metadata_uuid = fs_devices->metadata_uuid;
518 metadata_uuid = fs_devices->fsid;
520 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
523 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
524 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
530 /* Do basic extent buffer checks at read time */
531 static int validate_extent_buffer(struct extent_buffer *eb)
533 struct btrfs_fs_info *fs_info = eb->fs_info;
535 const u32 csum_size = fs_info->csum_size;
537 u8 result[BTRFS_CSUM_SIZE];
540 found_start = btrfs_header_bytenr(eb);
541 if (found_start != eb->start) {
542 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
543 eb->start, found_start);
547 if (check_tree_block_fsid(eb)) {
548 btrfs_err_rl(fs_info, "bad fsid on block %llu",
553 found_level = btrfs_header_level(eb);
554 if (found_level >= BTRFS_MAX_LEVEL) {
555 btrfs_err(fs_info, "bad tree block level %d on %llu",
556 (int)btrfs_header_level(eb), eb->start);
561 csum_tree_block(eb, result);
563 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
564 u8 val[BTRFS_CSUM_SIZE] = { 0 };
566 read_extent_buffer(eb, &val, 0, csum_size);
567 btrfs_warn_rl(fs_info,
568 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
569 fs_info->sb->s_id, eb->start,
570 CSUM_FMT_VALUE(csum_size, val),
571 CSUM_FMT_VALUE(csum_size, result),
572 btrfs_header_level(eb));
578 * If this is a leaf block and it is corrupt, set the corrupt bit so
579 * that we don't try and read the other copies of this block, just
582 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
583 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
587 if (found_level > 0 && btrfs_check_node(eb))
591 set_extent_buffer_uptodate(eb);
594 "block=%llu read time tree block corruption detected",
600 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
603 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
604 struct extent_buffer *eb;
609 * We don't allow bio merge for subpage metadata read, so we should
610 * only get one eb for each endio hook.
612 ASSERT(end == start + fs_info->nodesize - 1);
613 ASSERT(PagePrivate(page));
615 eb = find_extent_buffer(fs_info, start);
617 * When we are reading one tree block, eb must have been inserted into
618 * the radix tree. If not, something is wrong.
622 reads_done = atomic_dec_and_test(&eb->io_pages);
623 /* Subpage read must finish in page read */
626 eb->read_mirror = mirror;
627 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
631 ret = validate_extent_buffer(eb);
635 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
636 btree_readahead_hook(eb, ret);
638 set_extent_buffer_uptodate(eb);
640 free_extent_buffer(eb);
644 * end_bio_extent_readpage decrements io_pages in case of error,
645 * make sure it has something to decrement.
647 atomic_inc(&eb->io_pages);
648 clear_extent_buffer_uptodate(eb);
649 free_extent_buffer(eb);
653 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
654 struct page *page, u64 start, u64 end,
657 struct extent_buffer *eb;
661 ASSERT(page->private);
663 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
664 return validate_subpage_buffer(page, start, end, mirror);
666 eb = (struct extent_buffer *)page->private;
669 * The pending IO might have been the only thing that kept this buffer
670 * in memory. Make sure we have a ref for all this other checks
672 atomic_inc(&eb->refs);
674 reads_done = atomic_dec_and_test(&eb->io_pages);
678 eb->read_mirror = mirror;
679 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
683 ret = validate_extent_buffer(eb);
686 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
687 btree_readahead_hook(eb, ret);
691 * our io error hook is going to dec the io pages
692 * again, we have to make sure it has something
695 atomic_inc(&eb->io_pages);
696 clear_extent_buffer_uptodate(eb);
698 free_extent_buffer(eb);
703 static void end_workqueue_bio(struct bio *bio)
705 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
706 struct btrfs_fs_info *fs_info;
707 struct btrfs_workqueue *wq;
709 fs_info = end_io_wq->info;
710 end_io_wq->status = bio->bi_status;
712 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
713 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
714 wq = fs_info->endio_meta_write_workers;
715 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
716 wq = fs_info->endio_freespace_worker;
717 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
718 wq = fs_info->endio_raid56_workers;
720 wq = fs_info->endio_write_workers;
722 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
723 wq = fs_info->endio_raid56_workers;
724 else if (end_io_wq->metadata)
725 wq = fs_info->endio_meta_workers;
727 wq = fs_info->endio_workers;
730 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
731 btrfs_queue_work(wq, &end_io_wq->work);
734 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
735 enum btrfs_wq_endio_type metadata)
737 struct btrfs_end_io_wq *end_io_wq;
739 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
741 return BLK_STS_RESOURCE;
743 end_io_wq->private = bio->bi_private;
744 end_io_wq->end_io = bio->bi_end_io;
745 end_io_wq->info = info;
746 end_io_wq->status = 0;
747 end_io_wq->bio = bio;
748 end_io_wq->metadata = metadata;
750 bio->bi_private = end_io_wq;
751 bio->bi_end_io = end_workqueue_bio;
755 static void run_one_async_start(struct btrfs_work *work)
757 struct async_submit_bio *async;
760 async = container_of(work, struct async_submit_bio, work);
761 ret = async->submit_bio_start(async->inode, async->bio,
762 async->dio_file_offset);
768 * In order to insert checksums into the metadata in large chunks, we wait
769 * until bio submission time. All the pages in the bio are checksummed and
770 * sums are attached onto the ordered extent record.
772 * At IO completion time the csums attached on the ordered extent record are
773 * inserted into the tree.
775 static void run_one_async_done(struct btrfs_work *work)
777 struct async_submit_bio *async;
781 async = container_of(work, struct async_submit_bio, work);
782 inode = async->inode;
784 /* If an error occurred we just want to clean up the bio and move on */
786 async->bio->bi_status = async->status;
787 bio_endio(async->bio);
792 * All of the bios that pass through here are from async helpers.
793 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
794 * This changes nothing when cgroups aren't in use.
796 async->bio->bi_opf |= REQ_CGROUP_PUNT;
797 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
799 async->bio->bi_status = ret;
800 bio_endio(async->bio);
804 static void run_one_async_free(struct btrfs_work *work)
806 struct async_submit_bio *async;
808 async = container_of(work, struct async_submit_bio, work);
812 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
813 int mirror_num, unsigned long bio_flags,
815 extent_submit_bio_start_t *submit_bio_start)
817 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
818 struct async_submit_bio *async;
820 async = kmalloc(sizeof(*async), GFP_NOFS);
822 return BLK_STS_RESOURCE;
824 async->inode = inode;
826 async->mirror_num = mirror_num;
827 async->submit_bio_start = submit_bio_start;
829 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
832 async->dio_file_offset = dio_file_offset;
836 if (op_is_sync(bio->bi_opf))
837 btrfs_set_work_high_priority(&async->work);
839 btrfs_queue_work(fs_info->workers, &async->work);
843 static blk_status_t btree_csum_one_bio(struct bio *bio)
845 struct bio_vec *bvec;
846 struct btrfs_root *root;
848 struct bvec_iter_all iter_all;
850 ASSERT(!bio_flagged(bio, BIO_CLONED));
851 bio_for_each_segment_all(bvec, bio, iter_all) {
852 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
853 ret = csum_dirty_buffer(root->fs_info, bvec);
858 return errno_to_blk_status(ret);
861 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
865 * when we're called for a write, we're already in the async
866 * submission context. Just jump into btrfs_map_bio
868 return btree_csum_one_bio(bio);
871 static int check_async_write(struct btrfs_fs_info *fs_info,
872 struct btrfs_inode *bi)
874 if (btrfs_is_zoned(fs_info))
876 if (atomic_read(&bi->sync_writers))
878 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
883 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
884 int mirror_num, unsigned long bio_flags)
886 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
887 int async = check_async_write(fs_info, BTRFS_I(inode));
890 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
892 * called for a read, do the setup so that checksum validation
893 * can happen in the async kernel threads
895 ret = btrfs_bio_wq_end_io(fs_info, bio,
896 BTRFS_WQ_ENDIO_METADATA);
899 ret = btrfs_map_bio(fs_info, bio, mirror_num);
901 ret = btree_csum_one_bio(bio);
904 ret = btrfs_map_bio(fs_info, bio, mirror_num);
907 * kthread helpers are used to submit writes so that
908 * checksumming can happen in parallel across all CPUs
910 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
911 0, btree_submit_bio_start);
919 bio->bi_status = ret;
924 #ifdef CONFIG_MIGRATION
925 static int btree_migratepage(struct address_space *mapping,
926 struct page *newpage, struct page *page,
927 enum migrate_mode mode)
930 * we can't safely write a btree page from here,
931 * we haven't done the locking hook
936 * Buffers may be managed in a filesystem specific way.
937 * We must have no buffers or drop them.
939 if (page_has_private(page) &&
940 !try_to_release_page(page, GFP_KERNEL))
942 return migrate_page(mapping, newpage, page, mode);
947 static int btree_writepages(struct address_space *mapping,
948 struct writeback_control *wbc)
950 struct btrfs_fs_info *fs_info;
953 if (wbc->sync_mode == WB_SYNC_NONE) {
955 if (wbc->for_kupdate)
958 fs_info = BTRFS_I(mapping->host)->root->fs_info;
959 /* this is a bit racy, but that's ok */
960 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
961 BTRFS_DIRTY_METADATA_THRESH,
962 fs_info->dirty_metadata_batch);
966 return btree_write_cache_pages(mapping, wbc);
969 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
971 if (PageWriteback(page) || PageDirty(page))
974 return try_release_extent_buffer(page);
977 static void btree_invalidatepage(struct page *page, unsigned int offset,
980 struct extent_io_tree *tree;
981 tree = &BTRFS_I(page->mapping->host)->io_tree;
982 extent_invalidatepage(tree, page, offset);
983 btree_releasepage(page, GFP_NOFS);
984 if (PagePrivate(page)) {
985 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
986 "page private not zero on page %llu",
987 (unsigned long long)page_offset(page));
988 detach_page_private(page);
992 static int btree_set_page_dirty(struct page *page)
995 struct extent_buffer *eb;
997 BUG_ON(!PagePrivate(page));
998 eb = (struct extent_buffer *)page->private;
1000 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1001 BUG_ON(!atomic_read(&eb->refs));
1002 btrfs_assert_tree_locked(eb);
1004 return __set_page_dirty_nobuffers(page);
1007 static const struct address_space_operations btree_aops = {
1008 .writepages = btree_writepages,
1009 .releasepage = btree_releasepage,
1010 .invalidatepage = btree_invalidatepage,
1011 #ifdef CONFIG_MIGRATION
1012 .migratepage = btree_migratepage,
1014 .set_page_dirty = btree_set_page_dirty,
1017 struct extent_buffer *btrfs_find_create_tree_block(
1018 struct btrfs_fs_info *fs_info,
1019 u64 bytenr, u64 owner_root,
1022 if (btrfs_is_testing(fs_info))
1023 return alloc_test_extent_buffer(fs_info, bytenr);
1024 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1028 * Read tree block at logical address @bytenr and do variant basic but critical
1031 * @owner_root: the objectid of the root owner for this block.
1032 * @parent_transid: expected transid of this tree block, skip check if 0
1033 * @level: expected level, mandatory check
1034 * @first_key: expected key in slot 0, skip check if NULL
1036 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1037 u64 owner_root, u64 parent_transid,
1038 int level, struct btrfs_key *first_key)
1040 struct extent_buffer *buf = NULL;
1043 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1047 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1050 free_extent_buffer_stale(buf);
1051 return ERR_PTR(ret);
1057 void btrfs_clean_tree_block(struct extent_buffer *buf)
1059 struct btrfs_fs_info *fs_info = buf->fs_info;
1060 if (btrfs_header_generation(buf) ==
1061 fs_info->running_transaction->transid) {
1062 btrfs_assert_tree_locked(buf);
1064 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1065 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1067 fs_info->dirty_metadata_batch);
1068 clear_extent_buffer_dirty(buf);
1073 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1076 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1077 root->fs_info = fs_info;
1079 root->commit_root = NULL;
1081 root->orphan_cleanup_state = 0;
1083 root->last_trans = 0;
1084 root->free_objectid = 0;
1085 root->nr_delalloc_inodes = 0;
1086 root->nr_ordered_extents = 0;
1087 root->inode_tree = RB_ROOT;
1088 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1089 root->block_rsv = NULL;
1091 INIT_LIST_HEAD(&root->dirty_list);
1092 INIT_LIST_HEAD(&root->root_list);
1093 INIT_LIST_HEAD(&root->delalloc_inodes);
1094 INIT_LIST_HEAD(&root->delalloc_root);
1095 INIT_LIST_HEAD(&root->ordered_extents);
1096 INIT_LIST_HEAD(&root->ordered_root);
1097 INIT_LIST_HEAD(&root->reloc_dirty_list);
1098 INIT_LIST_HEAD(&root->logged_list[0]);
1099 INIT_LIST_HEAD(&root->logged_list[1]);
1100 spin_lock_init(&root->inode_lock);
1101 spin_lock_init(&root->delalloc_lock);
1102 spin_lock_init(&root->ordered_extent_lock);
1103 spin_lock_init(&root->accounting_lock);
1104 spin_lock_init(&root->log_extents_lock[0]);
1105 spin_lock_init(&root->log_extents_lock[1]);
1106 spin_lock_init(&root->qgroup_meta_rsv_lock);
1107 mutex_init(&root->objectid_mutex);
1108 mutex_init(&root->log_mutex);
1109 mutex_init(&root->ordered_extent_mutex);
1110 mutex_init(&root->delalloc_mutex);
1111 init_waitqueue_head(&root->qgroup_flush_wait);
1112 init_waitqueue_head(&root->log_writer_wait);
1113 init_waitqueue_head(&root->log_commit_wait[0]);
1114 init_waitqueue_head(&root->log_commit_wait[1]);
1115 INIT_LIST_HEAD(&root->log_ctxs[0]);
1116 INIT_LIST_HEAD(&root->log_ctxs[1]);
1117 atomic_set(&root->log_commit[0], 0);
1118 atomic_set(&root->log_commit[1], 0);
1119 atomic_set(&root->log_writers, 0);
1120 atomic_set(&root->log_batch, 0);
1121 refcount_set(&root->refs, 1);
1122 atomic_set(&root->snapshot_force_cow, 0);
1123 atomic_set(&root->nr_swapfiles, 0);
1124 root->log_transid = 0;
1125 root->log_transid_committed = -1;
1126 root->last_log_commit = 0;
1128 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1129 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1130 extent_io_tree_init(fs_info, &root->log_csum_range,
1131 IO_TREE_LOG_CSUM_RANGE, NULL);
1134 memset(&root->root_key, 0, sizeof(root->root_key));
1135 memset(&root->root_item, 0, sizeof(root->root_item));
1136 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1137 root->root_key.objectid = objectid;
1140 spin_lock_init(&root->root_item_lock);
1141 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1142 #ifdef CONFIG_BTRFS_DEBUG
1143 INIT_LIST_HEAD(&root->leak_list);
1144 spin_lock(&fs_info->fs_roots_radix_lock);
1145 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1146 spin_unlock(&fs_info->fs_roots_radix_lock);
1150 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1151 u64 objectid, gfp_t flags)
1153 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1155 __setup_root(root, fs_info, objectid);
1159 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1160 /* Should only be used by the testing infrastructure */
1161 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1163 struct btrfs_root *root;
1166 return ERR_PTR(-EINVAL);
1168 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1170 return ERR_PTR(-ENOMEM);
1172 /* We don't use the stripesize in selftest, set it as sectorsize */
1173 root->alloc_bytenr = 0;
1179 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1182 struct btrfs_fs_info *fs_info = trans->fs_info;
1183 struct extent_buffer *leaf;
1184 struct btrfs_root *tree_root = fs_info->tree_root;
1185 struct btrfs_root *root;
1186 struct btrfs_key key;
1187 unsigned int nofs_flag;
1191 * We're holding a transaction handle, so use a NOFS memory allocation
1192 * context to avoid deadlock if reclaim happens.
1194 nofs_flag = memalloc_nofs_save();
1195 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1196 memalloc_nofs_restore(nofs_flag);
1198 return ERR_PTR(-ENOMEM);
1200 root->root_key.objectid = objectid;
1201 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1202 root->root_key.offset = 0;
1204 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1205 BTRFS_NESTING_NORMAL);
1207 ret = PTR_ERR(leaf);
1213 btrfs_mark_buffer_dirty(leaf);
1215 root->commit_root = btrfs_root_node(root);
1216 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1218 btrfs_set_root_flags(&root->root_item, 0);
1219 btrfs_set_root_limit(&root->root_item, 0);
1220 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1221 btrfs_set_root_generation(&root->root_item, trans->transid);
1222 btrfs_set_root_level(&root->root_item, 0);
1223 btrfs_set_root_refs(&root->root_item, 1);
1224 btrfs_set_root_used(&root->root_item, leaf->len);
1225 btrfs_set_root_last_snapshot(&root->root_item, 0);
1226 btrfs_set_root_dirid(&root->root_item, 0);
1227 if (is_fstree(objectid))
1228 generate_random_guid(root->root_item.uuid);
1230 export_guid(root->root_item.uuid, &guid_null);
1231 btrfs_set_root_drop_level(&root->root_item, 0);
1233 btrfs_tree_unlock(leaf);
1235 key.objectid = objectid;
1236 key.type = BTRFS_ROOT_ITEM_KEY;
1238 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1246 btrfs_tree_unlock(leaf);
1248 btrfs_put_root(root);
1250 return ERR_PTR(ret);
1253 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1254 struct btrfs_fs_info *fs_info)
1256 struct btrfs_root *root;
1258 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1260 return ERR_PTR(-ENOMEM);
1262 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1263 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1264 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1269 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1270 struct btrfs_root *root)
1272 struct extent_buffer *leaf;
1275 * DON'T set SHAREABLE bit for log trees.
1277 * Log trees are not exposed to user space thus can't be snapshotted,
1278 * and they go away before a real commit is actually done.
1280 * They do store pointers to file data extents, and those reference
1281 * counts still get updated (along with back refs to the log tree).
1284 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1285 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1287 return PTR_ERR(leaf);
1291 btrfs_mark_buffer_dirty(root->node);
1292 btrfs_tree_unlock(root->node);
1297 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1298 struct btrfs_fs_info *fs_info)
1300 struct btrfs_root *log_root;
1302 log_root = alloc_log_tree(trans, fs_info);
1303 if (IS_ERR(log_root))
1304 return PTR_ERR(log_root);
1306 if (!btrfs_is_zoned(fs_info)) {
1307 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1310 btrfs_put_root(log_root);
1315 WARN_ON(fs_info->log_root_tree);
1316 fs_info->log_root_tree = log_root;
1320 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1321 struct btrfs_root *root)
1323 struct btrfs_fs_info *fs_info = root->fs_info;
1324 struct btrfs_root *log_root;
1325 struct btrfs_inode_item *inode_item;
1328 log_root = alloc_log_tree(trans, fs_info);
1329 if (IS_ERR(log_root))
1330 return PTR_ERR(log_root);
1332 ret = btrfs_alloc_log_tree_node(trans, log_root);
1334 btrfs_put_root(log_root);
1338 log_root->last_trans = trans->transid;
1339 log_root->root_key.offset = root->root_key.objectid;
1341 inode_item = &log_root->root_item.inode;
1342 btrfs_set_stack_inode_generation(inode_item, 1);
1343 btrfs_set_stack_inode_size(inode_item, 3);
1344 btrfs_set_stack_inode_nlink(inode_item, 1);
1345 btrfs_set_stack_inode_nbytes(inode_item,
1347 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1349 btrfs_set_root_node(&log_root->root_item, log_root->node);
1351 WARN_ON(root->log_root);
1352 root->log_root = log_root;
1353 root->log_transid = 0;
1354 root->log_transid_committed = -1;
1355 root->last_log_commit = 0;
1359 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1360 struct btrfs_path *path,
1361 struct btrfs_key *key)
1363 struct btrfs_root *root;
1364 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1369 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1371 return ERR_PTR(-ENOMEM);
1373 ret = btrfs_find_root(tree_root, key, path,
1374 &root->root_item, &root->root_key);
1381 generation = btrfs_root_generation(&root->root_item);
1382 level = btrfs_root_level(&root->root_item);
1383 root->node = read_tree_block(fs_info,
1384 btrfs_root_bytenr(&root->root_item),
1385 key->objectid, generation, level, NULL);
1386 if (IS_ERR(root->node)) {
1387 ret = PTR_ERR(root->node);
1390 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1394 root->commit_root = btrfs_root_node(root);
1397 btrfs_put_root(root);
1398 return ERR_PTR(ret);
1401 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1402 struct btrfs_key *key)
1404 struct btrfs_root *root;
1405 struct btrfs_path *path;
1407 path = btrfs_alloc_path();
1409 return ERR_PTR(-ENOMEM);
1410 root = read_tree_root_path(tree_root, path, key);
1411 btrfs_free_path(path);
1417 * Initialize subvolume root in-memory structure
1419 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1421 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1424 unsigned int nofs_flag;
1427 * We might be called under a transaction (e.g. indirect backref
1428 * resolution) which could deadlock if it triggers memory reclaim
1430 nofs_flag = memalloc_nofs_save();
1431 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1432 memalloc_nofs_restore(nofs_flag);
1436 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1437 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1438 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1439 btrfs_check_and_init_root_item(&root->root_item);
1443 * Don't assign anonymous block device to roots that are not exposed to
1444 * userspace, the id pool is limited to 1M
1446 if (is_fstree(root->root_key.objectid) &&
1447 btrfs_root_refs(&root->root_item) > 0) {
1449 ret = get_anon_bdev(&root->anon_dev);
1453 root->anon_dev = anon_dev;
1457 mutex_lock(&root->objectid_mutex);
1458 ret = btrfs_init_root_free_objectid(root);
1460 mutex_unlock(&root->objectid_mutex);
1464 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1466 mutex_unlock(&root->objectid_mutex);
1470 /* The caller is responsible to call btrfs_free_fs_root */
1474 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1477 struct btrfs_root *root;
1479 spin_lock(&fs_info->fs_roots_radix_lock);
1480 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1481 (unsigned long)root_id);
1483 root = btrfs_grab_root(root);
1484 spin_unlock(&fs_info->fs_roots_radix_lock);
1488 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1491 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1492 return btrfs_grab_root(fs_info->tree_root);
1493 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1494 return btrfs_grab_root(fs_info->extent_root);
1495 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1496 return btrfs_grab_root(fs_info->chunk_root);
1497 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1498 return btrfs_grab_root(fs_info->dev_root);
1499 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1500 return btrfs_grab_root(fs_info->csum_root);
1501 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1502 return btrfs_grab_root(fs_info->quota_root) ?
1503 fs_info->quota_root : ERR_PTR(-ENOENT);
1504 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1505 return btrfs_grab_root(fs_info->uuid_root) ?
1506 fs_info->uuid_root : ERR_PTR(-ENOENT);
1507 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1508 return btrfs_grab_root(fs_info->free_space_root) ?
1509 fs_info->free_space_root : ERR_PTR(-ENOENT);
1513 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1514 struct btrfs_root *root)
1518 ret = radix_tree_preload(GFP_NOFS);
1522 spin_lock(&fs_info->fs_roots_radix_lock);
1523 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1524 (unsigned long)root->root_key.objectid,
1527 btrfs_grab_root(root);
1528 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1530 spin_unlock(&fs_info->fs_roots_radix_lock);
1531 radix_tree_preload_end();
1536 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1538 #ifdef CONFIG_BTRFS_DEBUG
1539 struct btrfs_root *root;
1541 while (!list_empty(&fs_info->allocated_roots)) {
1542 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1544 root = list_first_entry(&fs_info->allocated_roots,
1545 struct btrfs_root, leak_list);
1546 btrfs_err(fs_info, "leaked root %s refcount %d",
1547 btrfs_root_name(&root->root_key, buf),
1548 refcount_read(&root->refs));
1549 while (refcount_read(&root->refs) > 1)
1550 btrfs_put_root(root);
1551 btrfs_put_root(root);
1556 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1558 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1559 percpu_counter_destroy(&fs_info->delalloc_bytes);
1560 percpu_counter_destroy(&fs_info->ordered_bytes);
1561 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1562 btrfs_free_csum_hash(fs_info);
1563 btrfs_free_stripe_hash_table(fs_info);
1564 btrfs_free_ref_cache(fs_info);
1565 kfree(fs_info->balance_ctl);
1566 kfree(fs_info->delayed_root);
1567 btrfs_put_root(fs_info->extent_root);
1568 btrfs_put_root(fs_info->tree_root);
1569 btrfs_put_root(fs_info->chunk_root);
1570 btrfs_put_root(fs_info->dev_root);
1571 btrfs_put_root(fs_info->csum_root);
1572 btrfs_put_root(fs_info->quota_root);
1573 btrfs_put_root(fs_info->uuid_root);
1574 btrfs_put_root(fs_info->free_space_root);
1575 btrfs_put_root(fs_info->fs_root);
1576 btrfs_put_root(fs_info->data_reloc_root);
1577 btrfs_check_leaked_roots(fs_info);
1578 btrfs_extent_buffer_leak_debug_check(fs_info);
1579 kfree(fs_info->super_copy);
1580 kfree(fs_info->super_for_commit);
1586 * Get an in-memory reference of a root structure.
1588 * For essential trees like root/extent tree, we grab it from fs_info directly.
1589 * For subvolume trees, we check the cached filesystem roots first. If not
1590 * found, then read it from disk and add it to cached fs roots.
1592 * Caller should release the root by calling btrfs_put_root() after the usage.
1594 * NOTE: Reloc and log trees can't be read by this function as they share the
1595 * same root objectid.
1597 * @objectid: root id
1598 * @anon_dev: preallocated anonymous block device number for new roots,
1599 * pass 0 for new allocation.
1600 * @check_ref: whether to check root item references, If true, return -ENOENT
1603 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1604 u64 objectid, dev_t anon_dev,
1607 struct btrfs_root *root;
1608 struct btrfs_path *path;
1609 struct btrfs_key key;
1612 root = btrfs_get_global_root(fs_info, objectid);
1616 root = btrfs_lookup_fs_root(fs_info, objectid);
1618 /* Shouldn't get preallocated anon_dev for cached roots */
1620 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1621 btrfs_put_root(root);
1622 return ERR_PTR(-ENOENT);
1627 key.objectid = objectid;
1628 key.type = BTRFS_ROOT_ITEM_KEY;
1629 key.offset = (u64)-1;
1630 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1634 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1639 ret = btrfs_init_fs_root(root, anon_dev);
1643 path = btrfs_alloc_path();
1648 key.objectid = BTRFS_ORPHAN_OBJECTID;
1649 key.type = BTRFS_ORPHAN_ITEM_KEY;
1650 key.offset = objectid;
1652 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1653 btrfs_free_path(path);
1657 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1659 ret = btrfs_insert_fs_root(fs_info, root);
1661 btrfs_put_root(root);
1668 btrfs_put_root(root);
1669 return ERR_PTR(ret);
1673 * Get in-memory reference of a root structure
1675 * @objectid: tree objectid
1676 * @check_ref: if set, verify that the tree exists and the item has at least
1679 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1680 u64 objectid, bool check_ref)
1682 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1686 * Get in-memory reference of a root structure, created as new, optionally pass
1687 * the anonymous block device id
1689 * @objectid: tree objectid
1690 * @anon_dev: if zero, allocate a new anonymous block device or use the
1693 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1694 u64 objectid, dev_t anon_dev)
1696 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1700 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1701 * @fs_info: the fs_info
1702 * @objectid: the objectid we need to lookup
1704 * This is exclusively used for backref walking, and exists specifically because
1705 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1706 * creation time, which means we may have to read the tree_root in order to look
1707 * up a fs root that is not in memory. If the root is not in memory we will
1708 * read the tree root commit root and look up the fs root from there. This is a
1709 * temporary root, it will not be inserted into the radix tree as it doesn't
1710 * have the most uptodate information, it'll simply be discarded once the
1711 * backref code is finished using the root.
1713 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1714 struct btrfs_path *path,
1717 struct btrfs_root *root;
1718 struct btrfs_key key;
1720 ASSERT(path->search_commit_root && path->skip_locking);
1723 * This can return -ENOENT if we ask for a root that doesn't exist, but
1724 * since this is called via the backref walking code we won't be looking
1725 * up a root that doesn't exist, unless there's corruption. So if root
1726 * != NULL just return it.
1728 root = btrfs_get_global_root(fs_info, objectid);
1732 root = btrfs_lookup_fs_root(fs_info, objectid);
1736 key.objectid = objectid;
1737 key.type = BTRFS_ROOT_ITEM_KEY;
1738 key.offset = (u64)-1;
1739 root = read_tree_root_path(fs_info->tree_root, path, &key);
1740 btrfs_release_path(path);
1746 * called by the kthread helper functions to finally call the bio end_io
1747 * functions. This is where read checksum verification actually happens
1749 static void end_workqueue_fn(struct btrfs_work *work)
1752 struct btrfs_end_io_wq *end_io_wq;
1754 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1755 bio = end_io_wq->bio;
1757 bio->bi_status = end_io_wq->status;
1758 bio->bi_private = end_io_wq->private;
1759 bio->bi_end_io = end_io_wq->end_io;
1761 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1764 static int cleaner_kthread(void *arg)
1766 struct btrfs_root *root = arg;
1767 struct btrfs_fs_info *fs_info = root->fs_info;
1773 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1775 /* Make the cleaner go to sleep early. */
1776 if (btrfs_need_cleaner_sleep(fs_info))
1780 * Do not do anything if we might cause open_ctree() to block
1781 * before we have finished mounting the filesystem.
1783 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1786 if (!mutex_trylock(&fs_info->cleaner_mutex))
1790 * Avoid the problem that we change the status of the fs
1791 * during the above check and trylock.
1793 if (btrfs_need_cleaner_sleep(fs_info)) {
1794 mutex_unlock(&fs_info->cleaner_mutex);
1798 btrfs_run_delayed_iputs(fs_info);
1800 again = btrfs_clean_one_deleted_snapshot(root);
1801 mutex_unlock(&fs_info->cleaner_mutex);
1804 * The defragger has dealt with the R/O remount and umount,
1805 * needn't do anything special here.
1807 btrfs_run_defrag_inodes(fs_info);
1810 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1811 * with relocation (btrfs_relocate_chunk) and relocation
1812 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1813 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1814 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1815 * unused block groups.
1817 btrfs_delete_unused_bgs(fs_info);
1819 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1820 if (kthread_should_park())
1822 if (kthread_should_stop())
1825 set_current_state(TASK_INTERRUPTIBLE);
1827 __set_current_state(TASK_RUNNING);
1832 static int transaction_kthread(void *arg)
1834 struct btrfs_root *root = arg;
1835 struct btrfs_fs_info *fs_info = root->fs_info;
1836 struct btrfs_trans_handle *trans;
1837 struct btrfs_transaction *cur;
1840 unsigned long delay;
1844 cannot_commit = false;
1845 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1846 mutex_lock(&fs_info->transaction_kthread_mutex);
1848 spin_lock(&fs_info->trans_lock);
1849 cur = fs_info->running_transaction;
1851 spin_unlock(&fs_info->trans_lock);
1855 delta = ktime_get_seconds() - cur->start_time;
1856 if (cur->state < TRANS_STATE_COMMIT_START &&
1857 delta < fs_info->commit_interval) {
1858 spin_unlock(&fs_info->trans_lock);
1859 delay -= msecs_to_jiffies((delta - 1) * 1000);
1861 msecs_to_jiffies(fs_info->commit_interval * 1000));
1864 transid = cur->transid;
1865 spin_unlock(&fs_info->trans_lock);
1867 /* If the file system is aborted, this will always fail. */
1868 trans = btrfs_attach_transaction(root);
1869 if (IS_ERR(trans)) {
1870 if (PTR_ERR(trans) != -ENOENT)
1871 cannot_commit = true;
1874 if (transid == trans->transid) {
1875 btrfs_commit_transaction(trans);
1877 btrfs_end_transaction(trans);
1880 wake_up_process(fs_info->cleaner_kthread);
1881 mutex_unlock(&fs_info->transaction_kthread_mutex);
1883 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1884 &fs_info->fs_state)))
1885 btrfs_cleanup_transaction(fs_info);
1886 if (!kthread_should_stop() &&
1887 (!btrfs_transaction_blocked(fs_info) ||
1889 schedule_timeout_interruptible(delay);
1890 } while (!kthread_should_stop());
1895 * This will find the highest generation in the array of root backups. The
1896 * index of the highest array is returned, or -EINVAL if we can't find
1899 * We check to make sure the array is valid by comparing the
1900 * generation of the latest root in the array with the generation
1901 * in the super block. If they don't match we pitch it.
1903 static int find_newest_super_backup(struct btrfs_fs_info *info)
1905 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1907 struct btrfs_root_backup *root_backup;
1910 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1911 root_backup = info->super_copy->super_roots + i;
1912 cur = btrfs_backup_tree_root_gen(root_backup);
1913 if (cur == newest_gen)
1921 * copy all the root pointers into the super backup array.
1922 * this will bump the backup pointer by one when it is
1925 static void backup_super_roots(struct btrfs_fs_info *info)
1927 const int next_backup = info->backup_root_index;
1928 struct btrfs_root_backup *root_backup;
1930 root_backup = info->super_for_commit->super_roots + next_backup;
1933 * make sure all of our padding and empty slots get zero filled
1934 * regardless of which ones we use today
1936 memset(root_backup, 0, sizeof(*root_backup));
1938 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1940 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1941 btrfs_set_backup_tree_root_gen(root_backup,
1942 btrfs_header_generation(info->tree_root->node));
1944 btrfs_set_backup_tree_root_level(root_backup,
1945 btrfs_header_level(info->tree_root->node));
1947 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1948 btrfs_set_backup_chunk_root_gen(root_backup,
1949 btrfs_header_generation(info->chunk_root->node));
1950 btrfs_set_backup_chunk_root_level(root_backup,
1951 btrfs_header_level(info->chunk_root->node));
1953 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1954 btrfs_set_backup_extent_root_gen(root_backup,
1955 btrfs_header_generation(info->extent_root->node));
1956 btrfs_set_backup_extent_root_level(root_backup,
1957 btrfs_header_level(info->extent_root->node));
1960 * we might commit during log recovery, which happens before we set
1961 * the fs_root. Make sure it is valid before we fill it in.
1963 if (info->fs_root && info->fs_root->node) {
1964 btrfs_set_backup_fs_root(root_backup,
1965 info->fs_root->node->start);
1966 btrfs_set_backup_fs_root_gen(root_backup,
1967 btrfs_header_generation(info->fs_root->node));
1968 btrfs_set_backup_fs_root_level(root_backup,
1969 btrfs_header_level(info->fs_root->node));
1972 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1973 btrfs_set_backup_dev_root_gen(root_backup,
1974 btrfs_header_generation(info->dev_root->node));
1975 btrfs_set_backup_dev_root_level(root_backup,
1976 btrfs_header_level(info->dev_root->node));
1978 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1979 btrfs_set_backup_csum_root_gen(root_backup,
1980 btrfs_header_generation(info->csum_root->node));
1981 btrfs_set_backup_csum_root_level(root_backup,
1982 btrfs_header_level(info->csum_root->node));
1984 btrfs_set_backup_total_bytes(root_backup,
1985 btrfs_super_total_bytes(info->super_copy));
1986 btrfs_set_backup_bytes_used(root_backup,
1987 btrfs_super_bytes_used(info->super_copy));
1988 btrfs_set_backup_num_devices(root_backup,
1989 btrfs_super_num_devices(info->super_copy));
1992 * if we don't copy this out to the super_copy, it won't get remembered
1993 * for the next commit
1995 memcpy(&info->super_copy->super_roots,
1996 &info->super_for_commit->super_roots,
1997 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2001 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2002 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2004 * fs_info - filesystem whose backup roots need to be read
2005 * priority - priority of backup root required
2007 * Returns backup root index on success and -EINVAL otherwise.
2009 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2011 int backup_index = find_newest_super_backup(fs_info);
2012 struct btrfs_super_block *super = fs_info->super_copy;
2013 struct btrfs_root_backup *root_backup;
2015 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2017 return backup_index;
2019 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2020 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2025 root_backup = super->super_roots + backup_index;
2027 btrfs_set_super_generation(super,
2028 btrfs_backup_tree_root_gen(root_backup));
2029 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2030 btrfs_set_super_root_level(super,
2031 btrfs_backup_tree_root_level(root_backup));
2032 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2035 * Fixme: the total bytes and num_devices need to match or we should
2038 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2039 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2041 return backup_index;
2044 /* helper to cleanup workers */
2045 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2047 btrfs_destroy_workqueue(fs_info->fixup_workers);
2048 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2049 btrfs_destroy_workqueue(fs_info->workers);
2050 btrfs_destroy_workqueue(fs_info->endio_workers);
2051 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2052 btrfs_destroy_workqueue(fs_info->rmw_workers);
2053 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2054 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2055 btrfs_destroy_workqueue(fs_info->delayed_workers);
2056 btrfs_destroy_workqueue(fs_info->caching_workers);
2057 btrfs_destroy_workqueue(fs_info->readahead_workers);
2058 btrfs_destroy_workqueue(fs_info->flush_workers);
2059 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2060 if (fs_info->discard_ctl.discard_workers)
2061 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2063 * Now that all other work queues are destroyed, we can safely destroy
2064 * the queues used for metadata I/O, since tasks from those other work
2065 * queues can do metadata I/O operations.
2067 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2068 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2071 static void free_root_extent_buffers(struct btrfs_root *root)
2074 free_extent_buffer(root->node);
2075 free_extent_buffer(root->commit_root);
2077 root->commit_root = NULL;
2081 /* helper to cleanup tree roots */
2082 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2084 free_root_extent_buffers(info->tree_root);
2086 free_root_extent_buffers(info->dev_root);
2087 free_root_extent_buffers(info->extent_root);
2088 free_root_extent_buffers(info->csum_root);
2089 free_root_extent_buffers(info->quota_root);
2090 free_root_extent_buffers(info->uuid_root);
2091 free_root_extent_buffers(info->fs_root);
2092 free_root_extent_buffers(info->data_reloc_root);
2093 if (free_chunk_root)
2094 free_root_extent_buffers(info->chunk_root);
2095 free_root_extent_buffers(info->free_space_root);
2098 void btrfs_put_root(struct btrfs_root *root)
2103 if (refcount_dec_and_test(&root->refs)) {
2104 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2105 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2107 free_anon_bdev(root->anon_dev);
2108 btrfs_drew_lock_destroy(&root->snapshot_lock);
2109 free_root_extent_buffers(root);
2110 #ifdef CONFIG_BTRFS_DEBUG
2111 spin_lock(&root->fs_info->fs_roots_radix_lock);
2112 list_del_init(&root->leak_list);
2113 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2119 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2122 struct btrfs_root *gang[8];
2125 while (!list_empty(&fs_info->dead_roots)) {
2126 gang[0] = list_entry(fs_info->dead_roots.next,
2127 struct btrfs_root, root_list);
2128 list_del(&gang[0]->root_list);
2130 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2131 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2132 btrfs_put_root(gang[0]);
2136 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2141 for (i = 0; i < ret; i++)
2142 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2146 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2148 mutex_init(&fs_info->scrub_lock);
2149 atomic_set(&fs_info->scrubs_running, 0);
2150 atomic_set(&fs_info->scrub_pause_req, 0);
2151 atomic_set(&fs_info->scrubs_paused, 0);
2152 atomic_set(&fs_info->scrub_cancel_req, 0);
2153 init_waitqueue_head(&fs_info->scrub_pause_wait);
2154 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2157 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2159 spin_lock_init(&fs_info->balance_lock);
2160 mutex_init(&fs_info->balance_mutex);
2161 atomic_set(&fs_info->balance_pause_req, 0);
2162 atomic_set(&fs_info->balance_cancel_req, 0);
2163 fs_info->balance_ctl = NULL;
2164 init_waitqueue_head(&fs_info->balance_wait_q);
2167 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2169 struct inode *inode = fs_info->btree_inode;
2171 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2172 set_nlink(inode, 1);
2174 * we set the i_size on the btree inode to the max possible int.
2175 * the real end of the address space is determined by all of
2176 * the devices in the system
2178 inode->i_size = OFFSET_MAX;
2179 inode->i_mapping->a_ops = &btree_aops;
2181 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2182 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2183 IO_TREE_BTREE_INODE_IO, inode);
2184 BTRFS_I(inode)->io_tree.track_uptodate = false;
2185 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2187 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2188 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2189 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2190 btrfs_insert_inode_hash(inode);
2193 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2195 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2196 init_rwsem(&fs_info->dev_replace.rwsem);
2197 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2200 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2202 spin_lock_init(&fs_info->qgroup_lock);
2203 mutex_init(&fs_info->qgroup_ioctl_lock);
2204 fs_info->qgroup_tree = RB_ROOT;
2205 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2206 fs_info->qgroup_seq = 1;
2207 fs_info->qgroup_ulist = NULL;
2208 fs_info->qgroup_rescan_running = false;
2209 mutex_init(&fs_info->qgroup_rescan_lock);
2212 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2213 struct btrfs_fs_devices *fs_devices)
2215 u32 max_active = fs_info->thread_pool_size;
2216 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2219 btrfs_alloc_workqueue(fs_info, "worker",
2220 flags | WQ_HIGHPRI, max_active, 16);
2222 fs_info->delalloc_workers =
2223 btrfs_alloc_workqueue(fs_info, "delalloc",
2224 flags, max_active, 2);
2226 fs_info->flush_workers =
2227 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2228 flags, max_active, 0);
2230 fs_info->caching_workers =
2231 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2233 fs_info->fixup_workers =
2234 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2237 * endios are largely parallel and should have a very
2240 fs_info->endio_workers =
2241 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2242 fs_info->endio_meta_workers =
2243 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2245 fs_info->endio_meta_write_workers =
2246 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2248 fs_info->endio_raid56_workers =
2249 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2251 fs_info->rmw_workers =
2252 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2253 fs_info->endio_write_workers =
2254 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2256 fs_info->endio_freespace_worker =
2257 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2259 fs_info->delayed_workers =
2260 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2262 fs_info->readahead_workers =
2263 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2265 fs_info->qgroup_rescan_workers =
2266 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2267 fs_info->discard_ctl.discard_workers =
2268 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2270 if (!(fs_info->workers && fs_info->delalloc_workers &&
2271 fs_info->flush_workers &&
2272 fs_info->endio_workers && fs_info->endio_meta_workers &&
2273 fs_info->endio_meta_write_workers &&
2274 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2275 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2276 fs_info->caching_workers && fs_info->readahead_workers &&
2277 fs_info->fixup_workers && fs_info->delayed_workers &&
2278 fs_info->qgroup_rescan_workers &&
2279 fs_info->discard_ctl.discard_workers)) {
2286 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2288 struct crypto_shash *csum_shash;
2289 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2291 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2293 if (IS_ERR(csum_shash)) {
2294 btrfs_err(fs_info, "error allocating %s hash for checksum",
2296 return PTR_ERR(csum_shash);
2299 fs_info->csum_shash = csum_shash;
2304 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2305 struct btrfs_fs_devices *fs_devices)
2308 struct btrfs_root *log_tree_root;
2309 struct btrfs_super_block *disk_super = fs_info->super_copy;
2310 u64 bytenr = btrfs_super_log_root(disk_super);
2311 int level = btrfs_super_log_root_level(disk_super);
2313 if (fs_devices->rw_devices == 0) {
2314 btrfs_warn(fs_info, "log replay required on RO media");
2318 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2323 log_tree_root->node = read_tree_block(fs_info, bytenr,
2324 BTRFS_TREE_LOG_OBJECTID,
2325 fs_info->generation + 1, level,
2327 if (IS_ERR(log_tree_root->node)) {
2328 btrfs_warn(fs_info, "failed to read log tree");
2329 ret = PTR_ERR(log_tree_root->node);
2330 log_tree_root->node = NULL;
2331 btrfs_put_root(log_tree_root);
2333 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2334 btrfs_err(fs_info, "failed to read log tree");
2335 btrfs_put_root(log_tree_root);
2338 /* returns with log_tree_root freed on success */
2339 ret = btrfs_recover_log_trees(log_tree_root);
2341 btrfs_handle_fs_error(fs_info, ret,
2342 "Failed to recover log tree");
2343 btrfs_put_root(log_tree_root);
2347 if (sb_rdonly(fs_info->sb)) {
2348 ret = btrfs_commit_super(fs_info);
2356 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2358 struct btrfs_root *tree_root = fs_info->tree_root;
2359 struct btrfs_root *root;
2360 struct btrfs_key location;
2363 BUG_ON(!fs_info->tree_root);
2365 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2366 location.type = BTRFS_ROOT_ITEM_KEY;
2367 location.offset = 0;
2369 root = btrfs_read_tree_root(tree_root, &location);
2371 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2372 ret = PTR_ERR(root);
2376 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2377 fs_info->extent_root = root;
2380 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2381 root = btrfs_read_tree_root(tree_root, &location);
2383 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2384 ret = PTR_ERR(root);
2388 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2389 fs_info->dev_root = root;
2391 /* Initialize fs_info for all devices in any case */
2392 btrfs_init_devices_late(fs_info);
2394 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2395 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2396 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2397 root = btrfs_read_tree_root(tree_root, &location);
2399 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2400 ret = PTR_ERR(root);
2404 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2405 fs_info->csum_root = root;
2410 * This tree can share blocks with some other fs tree during relocation
2411 * and we need a proper setup by btrfs_get_fs_root
2413 root = btrfs_get_fs_root(tree_root->fs_info,
2414 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2416 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2417 ret = PTR_ERR(root);
2421 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2422 fs_info->data_reloc_root = root;
2425 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2426 root = btrfs_read_tree_root(tree_root, &location);
2427 if (!IS_ERR(root)) {
2428 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2429 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2430 fs_info->quota_root = root;
2433 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2434 root = btrfs_read_tree_root(tree_root, &location);
2436 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2437 ret = PTR_ERR(root);
2442 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2443 fs_info->uuid_root = root;
2446 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2447 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2448 root = btrfs_read_tree_root(tree_root, &location);
2450 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2451 ret = PTR_ERR(root);
2455 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2456 fs_info->free_space_root = root;
2462 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2463 location.objectid, ret);
2468 * Real super block validation
2469 * NOTE: super csum type and incompat features will not be checked here.
2471 * @sb: super block to check
2472 * @mirror_num: the super block number to check its bytenr:
2473 * 0 the primary (1st) sb
2474 * 1, 2 2nd and 3rd backup copy
2475 * -1 skip bytenr check
2477 static int validate_super(struct btrfs_fs_info *fs_info,
2478 struct btrfs_super_block *sb, int mirror_num)
2480 u64 nodesize = btrfs_super_nodesize(sb);
2481 u64 sectorsize = btrfs_super_sectorsize(sb);
2484 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2485 btrfs_err(fs_info, "no valid FS found");
2488 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2489 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2490 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2493 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2494 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2495 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2498 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2499 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2500 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2503 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2504 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2505 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2510 * Check sectorsize and nodesize first, other check will need it.
2511 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2513 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2514 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2515 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2520 * For 4K page size, we only support 4K sector size.
2521 * For 64K page size, we support read-write for 64K sector size, and
2522 * read-only for 4K sector size.
2524 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2525 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2526 sectorsize != SZ_64K))) {
2528 "sectorsize %llu not yet supported for page size %lu",
2529 sectorsize, PAGE_SIZE);
2533 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2534 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2535 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2538 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2539 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2540 le32_to_cpu(sb->__unused_leafsize), nodesize);
2544 /* Root alignment check */
2545 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2546 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2547 btrfs_super_root(sb));
2550 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2551 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2552 btrfs_super_chunk_root(sb));
2555 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2556 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2557 btrfs_super_log_root(sb));
2561 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2562 BTRFS_FSID_SIZE) != 0) {
2564 "dev_item UUID does not match metadata fsid: %pU != %pU",
2565 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2570 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2573 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2574 btrfs_err(fs_info, "bytes_used is too small %llu",
2575 btrfs_super_bytes_used(sb));
2578 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2579 btrfs_err(fs_info, "invalid stripesize %u",
2580 btrfs_super_stripesize(sb));
2583 if (btrfs_super_num_devices(sb) > (1UL << 31))
2584 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2585 btrfs_super_num_devices(sb));
2586 if (btrfs_super_num_devices(sb) == 0) {
2587 btrfs_err(fs_info, "number of devices is 0");
2591 if (mirror_num >= 0 &&
2592 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2593 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2594 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2599 * Obvious sys_chunk_array corruptions, it must hold at least one key
2602 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2603 btrfs_err(fs_info, "system chunk array too big %u > %u",
2604 btrfs_super_sys_array_size(sb),
2605 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2608 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2609 + sizeof(struct btrfs_chunk)) {
2610 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2611 btrfs_super_sys_array_size(sb),
2612 sizeof(struct btrfs_disk_key)
2613 + sizeof(struct btrfs_chunk));
2618 * The generation is a global counter, we'll trust it more than the others
2619 * but it's still possible that it's the one that's wrong.
2621 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2623 "suspicious: generation < chunk_root_generation: %llu < %llu",
2624 btrfs_super_generation(sb),
2625 btrfs_super_chunk_root_generation(sb));
2626 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2627 && btrfs_super_cache_generation(sb) != (u64)-1)
2629 "suspicious: generation < cache_generation: %llu < %llu",
2630 btrfs_super_generation(sb),
2631 btrfs_super_cache_generation(sb));
2637 * Validation of super block at mount time.
2638 * Some checks already done early at mount time, like csum type and incompat
2639 * flags will be skipped.
2641 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2643 return validate_super(fs_info, fs_info->super_copy, 0);
2647 * Validation of super block at write time.
2648 * Some checks like bytenr check will be skipped as their values will be
2650 * Extra checks like csum type and incompat flags will be done here.
2652 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2653 struct btrfs_super_block *sb)
2657 ret = validate_super(fs_info, sb, -1);
2660 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2662 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2663 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2666 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2669 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2670 btrfs_super_incompat_flags(sb),
2671 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2677 "super block corruption detected before writing it to disk");
2681 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2683 int backup_index = find_newest_super_backup(fs_info);
2684 struct btrfs_super_block *sb = fs_info->super_copy;
2685 struct btrfs_root *tree_root = fs_info->tree_root;
2686 bool handle_error = false;
2690 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2695 if (!IS_ERR(tree_root->node))
2696 free_extent_buffer(tree_root->node);
2697 tree_root->node = NULL;
2699 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2702 free_root_pointers(fs_info, 0);
2705 * Don't use the log in recovery mode, it won't be
2708 btrfs_set_super_log_root(sb, 0);
2710 /* We can't trust the free space cache either */
2711 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2713 ret = read_backup_root(fs_info, i);
2718 generation = btrfs_super_generation(sb);
2719 level = btrfs_super_root_level(sb);
2720 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2721 BTRFS_ROOT_TREE_OBJECTID,
2722 generation, level, NULL);
2723 if (IS_ERR(tree_root->node)) {
2724 handle_error = true;
2725 ret = PTR_ERR(tree_root->node);
2726 tree_root->node = NULL;
2727 btrfs_warn(fs_info, "couldn't read tree root");
2730 } else if (!extent_buffer_uptodate(tree_root->node)) {
2731 handle_error = true;
2733 btrfs_warn(fs_info, "error while reading tree root");
2737 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2738 tree_root->commit_root = btrfs_root_node(tree_root);
2739 btrfs_set_root_refs(&tree_root->root_item, 1);
2742 * No need to hold btrfs_root::objectid_mutex since the fs
2743 * hasn't been fully initialised and we are the only user
2745 ret = btrfs_init_root_free_objectid(tree_root);
2747 handle_error = true;
2751 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2753 ret = btrfs_read_roots(fs_info);
2755 handle_error = true;
2759 /* All successful */
2760 fs_info->generation = generation;
2761 fs_info->last_trans_committed = generation;
2763 /* Always begin writing backup roots after the one being used */
2764 if (backup_index < 0) {
2765 fs_info->backup_root_index = 0;
2767 fs_info->backup_root_index = backup_index + 1;
2768 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2776 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2778 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2779 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2780 INIT_LIST_HEAD(&fs_info->trans_list);
2781 INIT_LIST_HEAD(&fs_info->dead_roots);
2782 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2783 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2784 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2785 spin_lock_init(&fs_info->delalloc_root_lock);
2786 spin_lock_init(&fs_info->trans_lock);
2787 spin_lock_init(&fs_info->fs_roots_radix_lock);
2788 spin_lock_init(&fs_info->delayed_iput_lock);
2789 spin_lock_init(&fs_info->defrag_inodes_lock);
2790 spin_lock_init(&fs_info->super_lock);
2791 spin_lock_init(&fs_info->buffer_lock);
2792 spin_lock_init(&fs_info->unused_bgs_lock);
2793 spin_lock_init(&fs_info->treelog_bg_lock);
2794 rwlock_init(&fs_info->tree_mod_log_lock);
2795 mutex_init(&fs_info->unused_bg_unpin_mutex);
2796 mutex_init(&fs_info->delete_unused_bgs_mutex);
2797 mutex_init(&fs_info->reloc_mutex);
2798 mutex_init(&fs_info->delalloc_root_mutex);
2799 mutex_init(&fs_info->zoned_meta_io_lock);
2800 seqlock_init(&fs_info->profiles_lock);
2802 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2803 INIT_LIST_HEAD(&fs_info->space_info);
2804 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2805 INIT_LIST_HEAD(&fs_info->unused_bgs);
2806 #ifdef CONFIG_BTRFS_DEBUG
2807 INIT_LIST_HEAD(&fs_info->allocated_roots);
2808 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2809 spin_lock_init(&fs_info->eb_leak_lock);
2811 extent_map_tree_init(&fs_info->mapping_tree);
2812 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2813 BTRFS_BLOCK_RSV_GLOBAL);
2814 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2815 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2816 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2817 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2818 BTRFS_BLOCK_RSV_DELOPS);
2819 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2820 BTRFS_BLOCK_RSV_DELREFS);
2822 atomic_set(&fs_info->async_delalloc_pages, 0);
2823 atomic_set(&fs_info->defrag_running, 0);
2824 atomic_set(&fs_info->reada_works_cnt, 0);
2825 atomic_set(&fs_info->nr_delayed_iputs, 0);
2826 atomic64_set(&fs_info->tree_mod_seq, 0);
2827 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2828 fs_info->metadata_ratio = 0;
2829 fs_info->defrag_inodes = RB_ROOT;
2830 atomic64_set(&fs_info->free_chunk_space, 0);
2831 fs_info->tree_mod_log = RB_ROOT;
2832 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2833 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2834 /* readahead state */
2835 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2836 spin_lock_init(&fs_info->reada_lock);
2837 btrfs_init_ref_verify(fs_info);
2839 fs_info->thread_pool_size = min_t(unsigned long,
2840 num_online_cpus() + 2, 8);
2842 INIT_LIST_HEAD(&fs_info->ordered_roots);
2843 spin_lock_init(&fs_info->ordered_root_lock);
2845 btrfs_init_scrub(fs_info);
2846 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2847 fs_info->check_integrity_print_mask = 0;
2849 btrfs_init_balance(fs_info);
2850 btrfs_init_async_reclaim_work(fs_info);
2852 spin_lock_init(&fs_info->block_group_cache_lock);
2853 fs_info->block_group_cache_tree = RB_ROOT;
2854 fs_info->first_logical_byte = (u64)-1;
2856 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2857 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2858 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2860 mutex_init(&fs_info->ordered_operations_mutex);
2861 mutex_init(&fs_info->tree_log_mutex);
2862 mutex_init(&fs_info->chunk_mutex);
2863 mutex_init(&fs_info->transaction_kthread_mutex);
2864 mutex_init(&fs_info->cleaner_mutex);
2865 mutex_init(&fs_info->ro_block_group_mutex);
2866 init_rwsem(&fs_info->commit_root_sem);
2867 init_rwsem(&fs_info->cleanup_work_sem);
2868 init_rwsem(&fs_info->subvol_sem);
2869 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2871 btrfs_init_dev_replace_locks(fs_info);
2872 btrfs_init_qgroup(fs_info);
2873 btrfs_discard_init(fs_info);
2875 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2876 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2878 init_waitqueue_head(&fs_info->transaction_throttle);
2879 init_waitqueue_head(&fs_info->transaction_wait);
2880 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2881 init_waitqueue_head(&fs_info->async_submit_wait);
2882 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2884 /* Usable values until the real ones are cached from the superblock */
2885 fs_info->nodesize = 4096;
2886 fs_info->sectorsize = 4096;
2887 fs_info->sectorsize_bits = ilog2(4096);
2888 fs_info->stripesize = 4096;
2890 spin_lock_init(&fs_info->swapfile_pins_lock);
2891 fs_info->swapfile_pins = RB_ROOT;
2893 fs_info->send_in_progress = 0;
2896 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2901 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2902 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2904 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2908 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2912 fs_info->dirty_metadata_batch = PAGE_SIZE *
2913 (1 + ilog2(nr_cpu_ids));
2915 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2919 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2924 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2926 if (!fs_info->delayed_root)
2928 btrfs_init_delayed_root(fs_info->delayed_root);
2931 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2933 return btrfs_alloc_stripe_hash_table(fs_info);
2936 static int btrfs_uuid_rescan_kthread(void *data)
2938 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2942 * 1st step is to iterate through the existing UUID tree and
2943 * to delete all entries that contain outdated data.
2944 * 2nd step is to add all missing entries to the UUID tree.
2946 ret = btrfs_uuid_tree_iterate(fs_info);
2949 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2951 up(&fs_info->uuid_tree_rescan_sem);
2954 return btrfs_uuid_scan_kthread(data);
2957 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2959 struct task_struct *task;
2961 down(&fs_info->uuid_tree_rescan_sem);
2962 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2964 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2965 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2966 up(&fs_info->uuid_tree_rescan_sem);
2967 return PTR_ERR(task);
2974 * Some options only have meaning at mount time and shouldn't persist across
2975 * remounts, or be displayed. Clear these at the end of mount and remount
2978 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
2980 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
2981 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
2985 * Mounting logic specific to read-write file systems. Shared by open_ctree
2986 * and btrfs_remount when remounting from read-only to read-write.
2988 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2991 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2992 bool clear_free_space_tree = false;
2994 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2995 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2996 clear_free_space_tree = true;
2997 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2998 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2999 btrfs_warn(fs_info, "free space tree is invalid");
3000 clear_free_space_tree = true;
3003 if (clear_free_space_tree) {
3004 btrfs_info(fs_info, "clearing free space tree");
3005 ret = btrfs_clear_free_space_tree(fs_info);
3008 "failed to clear free space tree: %d", ret);
3014 * btrfs_find_orphan_roots() is responsible for finding all the dead
3015 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3016 * them into the fs_info->fs_roots_radix tree. This must be done before
3017 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3018 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3019 * item before the root's tree is deleted - this means that if we unmount
3020 * or crash before the deletion completes, on the next mount we will not
3021 * delete what remains of the tree because the orphan item does not
3022 * exists anymore, which is what tells us we have a pending deletion.
3024 ret = btrfs_find_orphan_roots(fs_info);
3028 ret = btrfs_cleanup_fs_roots(fs_info);
3032 down_read(&fs_info->cleanup_work_sem);
3033 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3034 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3035 up_read(&fs_info->cleanup_work_sem);
3038 up_read(&fs_info->cleanup_work_sem);
3040 mutex_lock(&fs_info->cleaner_mutex);
3041 ret = btrfs_recover_relocation(fs_info->tree_root);
3042 mutex_unlock(&fs_info->cleaner_mutex);
3044 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3048 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3049 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3050 btrfs_info(fs_info, "creating free space tree");
3051 ret = btrfs_create_free_space_tree(fs_info);
3054 "failed to create free space tree: %d", ret);
3059 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3060 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3065 ret = btrfs_resume_balance_async(fs_info);
3069 ret = btrfs_resume_dev_replace_async(fs_info);
3071 btrfs_warn(fs_info, "failed to resume dev_replace");
3075 btrfs_qgroup_rescan_resume(fs_info);
3077 if (!fs_info->uuid_root) {
3078 btrfs_info(fs_info, "creating UUID tree");
3079 ret = btrfs_create_uuid_tree(fs_info);
3082 "failed to create the UUID tree %d", ret);
3091 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3100 struct btrfs_super_block *disk_super;
3101 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3102 struct btrfs_root *tree_root;
3103 struct btrfs_root *chunk_root;
3108 ret = init_mount_fs_info(fs_info, sb);
3114 /* These need to be init'ed before we start creating inodes and such. */
3115 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3117 fs_info->tree_root = tree_root;
3118 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3120 fs_info->chunk_root = chunk_root;
3121 if (!tree_root || !chunk_root) {
3126 fs_info->btree_inode = new_inode(sb);
3127 if (!fs_info->btree_inode) {
3131 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3132 btrfs_init_btree_inode(fs_info);
3134 invalidate_bdev(fs_devices->latest_bdev);
3137 * Read super block and check the signature bytes only
3139 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
3140 if (IS_ERR(disk_super)) {
3141 err = PTR_ERR(disk_super);
3146 * Verify the type first, if that or the checksum value are
3147 * corrupted, we'll find out
3149 csum_type = btrfs_super_csum_type(disk_super);
3150 if (!btrfs_supported_super_csum(csum_type)) {
3151 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3154 btrfs_release_disk_super(disk_super);
3158 ret = btrfs_init_csum_hash(fs_info, csum_type);
3161 btrfs_release_disk_super(disk_super);
3166 * We want to check superblock checksum, the type is stored inside.
3167 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3169 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3170 btrfs_err(fs_info, "superblock checksum mismatch");
3172 btrfs_release_disk_super(disk_super);
3177 * super_copy is zeroed at allocation time and we never touch the
3178 * following bytes up to INFO_SIZE, the checksum is calculated from
3179 * the whole block of INFO_SIZE
3181 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3182 btrfs_release_disk_super(disk_super);
3184 disk_super = fs_info->super_copy;
3186 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
3189 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
3190 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
3191 fs_info->super_copy->metadata_uuid,
3195 features = btrfs_super_flags(disk_super);
3196 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3197 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3198 btrfs_set_super_flags(disk_super, features);
3200 "found metadata UUID change in progress flag, clearing");
3203 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3204 sizeof(*fs_info->super_for_commit));
3206 ret = btrfs_validate_mount_super(fs_info);
3208 btrfs_err(fs_info, "superblock contains fatal errors");
3213 if (!btrfs_super_root(disk_super))
3216 /* check FS state, whether FS is broken. */
3217 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3218 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3221 * In the long term, we'll store the compression type in the super
3222 * block, and it'll be used for per file compression control.
3224 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3226 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3232 features = btrfs_super_incompat_flags(disk_super) &
3233 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3236 "cannot mount because of unsupported optional features (%llx)",
3242 features = btrfs_super_incompat_flags(disk_super);
3243 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3244 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3245 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3246 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3247 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3249 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3250 btrfs_info(fs_info, "has skinny extents");
3253 * flag our filesystem as having big metadata blocks if
3254 * they are bigger than the page size
3256 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3257 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3259 "flagging fs with big metadata feature");
3260 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3263 nodesize = btrfs_super_nodesize(disk_super);
3264 sectorsize = btrfs_super_sectorsize(disk_super);
3265 stripesize = sectorsize;
3266 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3267 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3269 /* Cache block sizes */
3270 fs_info->nodesize = nodesize;
3271 fs_info->sectorsize = sectorsize;
3272 fs_info->sectorsize_bits = ilog2(sectorsize);
3273 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3274 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3275 fs_info->stripesize = stripesize;
3278 * mixed block groups end up with duplicate but slightly offset
3279 * extent buffers for the same range. It leads to corruptions
3281 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3282 (sectorsize != nodesize)) {
3284 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3285 nodesize, sectorsize);
3290 * Needn't use the lock because there is no other task which will
3293 btrfs_set_super_incompat_flags(disk_super, features);
3295 features = btrfs_super_compat_ro_flags(disk_super) &
3296 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3297 if (!sb_rdonly(sb) && features) {
3299 "cannot mount read-write because of unsupported optional features (%llx)",
3305 /* For 4K sector size support, it's only read-only */
3306 if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) {
3307 if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) {
3309 "subpage sectorsize %u only supported read-only for page size %lu",
3310 sectorsize, PAGE_SIZE);
3316 ret = btrfs_init_workqueues(fs_info, fs_devices);
3319 goto fail_sb_buffer;
3322 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3323 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3325 sb->s_blocksize = sectorsize;
3326 sb->s_blocksize_bits = blksize_bits(sectorsize);
3327 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3329 mutex_lock(&fs_info->chunk_mutex);
3330 ret = btrfs_read_sys_array(fs_info);
3331 mutex_unlock(&fs_info->chunk_mutex);
3333 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3334 goto fail_sb_buffer;
3337 generation = btrfs_super_chunk_root_generation(disk_super);
3338 level = btrfs_super_chunk_root_level(disk_super);
3340 chunk_root->node = read_tree_block(fs_info,
3341 btrfs_super_chunk_root(disk_super),
3342 BTRFS_CHUNK_TREE_OBJECTID,
3343 generation, level, NULL);
3344 if (IS_ERR(chunk_root->node) ||
3345 !extent_buffer_uptodate(chunk_root->node)) {
3346 btrfs_err(fs_info, "failed to read chunk root");
3347 if (!IS_ERR(chunk_root->node))
3348 free_extent_buffer(chunk_root->node);
3349 chunk_root->node = NULL;
3350 goto fail_tree_roots;
3352 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3353 chunk_root->commit_root = btrfs_root_node(chunk_root);
3355 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3356 offsetof(struct btrfs_header, chunk_tree_uuid),
3359 ret = btrfs_read_chunk_tree(fs_info);
3361 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3362 goto fail_tree_roots;
3366 * At this point we know all the devices that make this filesystem,
3367 * including the seed devices but we don't know yet if the replace
3368 * target is required. So free devices that are not part of this
3369 * filesystem but skip the replace traget device which is checked
3370 * below in btrfs_init_dev_replace().
3372 btrfs_free_extra_devids(fs_devices);
3373 if (!fs_devices->latest_bdev) {
3374 btrfs_err(fs_info, "failed to read devices");
3375 goto fail_tree_roots;
3378 ret = init_tree_roots(fs_info);
3380 goto fail_tree_roots;
3383 * Get zone type information of zoned block devices. This will also
3384 * handle emulation of a zoned filesystem if a regular device has the
3385 * zoned incompat feature flag set.
3387 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3390 "zoned: failed to read device zone info: %d",
3392 goto fail_block_groups;
3396 * If we have a uuid root and we're not being told to rescan we need to
3397 * check the generation here so we can set the
3398 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3399 * transaction during a balance or the log replay without updating the
3400 * uuid generation, and then if we crash we would rescan the uuid tree,
3401 * even though it was perfectly fine.
3403 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3404 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3405 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3407 ret = btrfs_verify_dev_extents(fs_info);
3410 "failed to verify dev extents against chunks: %d",
3412 goto fail_block_groups;
3414 ret = btrfs_recover_balance(fs_info);
3416 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3417 goto fail_block_groups;
3420 ret = btrfs_init_dev_stats(fs_info);
3422 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3423 goto fail_block_groups;
3426 ret = btrfs_init_dev_replace(fs_info);
3428 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3429 goto fail_block_groups;
3432 ret = btrfs_check_zoned_mode(fs_info);
3434 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3436 goto fail_block_groups;
3439 ret = btrfs_sysfs_add_fsid(fs_devices);
3441 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3443 goto fail_block_groups;
3446 ret = btrfs_sysfs_add_mounted(fs_info);
3448 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3449 goto fail_fsdev_sysfs;
3452 ret = btrfs_init_space_info(fs_info);
3454 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3458 ret = btrfs_read_block_groups(fs_info);
3460 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3464 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3466 "writable mount is not allowed due to too many missing devices");
3470 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3472 if (IS_ERR(fs_info->cleaner_kthread))
3475 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3477 "btrfs-transaction");
3478 if (IS_ERR(fs_info->transaction_kthread))
3481 if (!btrfs_test_opt(fs_info, NOSSD) &&
3482 !fs_info->fs_devices->rotating) {
3483 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3487 * Mount does not set all options immediately, we can do it now and do
3488 * not have to wait for transaction commit
3490 btrfs_apply_pending_changes(fs_info);
3492 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3493 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3494 ret = btrfsic_mount(fs_info, fs_devices,
3495 btrfs_test_opt(fs_info,
3496 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3498 fs_info->check_integrity_print_mask);
3501 "failed to initialize integrity check module: %d",
3505 ret = btrfs_read_qgroup_config(fs_info);
3507 goto fail_trans_kthread;
3509 if (btrfs_build_ref_tree(fs_info))
3510 btrfs_err(fs_info, "couldn't build ref tree");
3512 /* do not make disk changes in broken FS or nologreplay is given */
3513 if (btrfs_super_log_root(disk_super) != 0 &&
3514 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3515 btrfs_info(fs_info, "start tree-log replay");
3516 ret = btrfs_replay_log(fs_info, fs_devices);
3523 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3524 if (IS_ERR(fs_info->fs_root)) {
3525 err = PTR_ERR(fs_info->fs_root);
3526 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3527 fs_info->fs_root = NULL;
3534 ret = btrfs_start_pre_rw_mount(fs_info);
3536 close_ctree(fs_info);
3539 btrfs_discard_resume(fs_info);
3541 if (fs_info->uuid_root &&
3542 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3543 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3544 btrfs_info(fs_info, "checking UUID tree");
3545 ret = btrfs_check_uuid_tree(fs_info);
3548 "failed to check the UUID tree: %d", ret);
3549 close_ctree(fs_info);
3554 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3557 btrfs_clear_oneshot_options(fs_info);
3561 btrfs_free_qgroup_config(fs_info);
3563 kthread_stop(fs_info->transaction_kthread);
3564 btrfs_cleanup_transaction(fs_info);
3565 btrfs_free_fs_roots(fs_info);
3567 kthread_stop(fs_info->cleaner_kthread);
3570 * make sure we're done with the btree inode before we stop our
3573 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3576 btrfs_sysfs_remove_mounted(fs_info);
3579 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3582 btrfs_put_block_group_cache(fs_info);
3585 if (fs_info->data_reloc_root)
3586 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3587 free_root_pointers(fs_info, true);
3588 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3591 btrfs_stop_all_workers(fs_info);
3592 btrfs_free_block_groups(fs_info);
3594 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3596 iput(fs_info->btree_inode);
3598 btrfs_close_devices(fs_info->fs_devices);
3601 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3603 static void btrfs_end_super_write(struct bio *bio)
3605 struct btrfs_device *device = bio->bi_private;
3606 struct bio_vec *bvec;
3607 struct bvec_iter_all iter_all;
3610 bio_for_each_segment_all(bvec, bio, iter_all) {
3611 page = bvec->bv_page;
3613 if (bio->bi_status) {
3614 btrfs_warn_rl_in_rcu(device->fs_info,
3615 "lost page write due to IO error on %s (%d)",
3616 rcu_str_deref(device->name),
3617 blk_status_to_errno(bio->bi_status));
3618 ClearPageUptodate(page);
3620 btrfs_dev_stat_inc_and_print(device,
3621 BTRFS_DEV_STAT_WRITE_ERRS);
3623 SetPageUptodate(page);
3633 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3636 struct btrfs_super_block *super;
3638 u64 bytenr, bytenr_orig;
3639 struct address_space *mapping = bdev->bd_inode->i_mapping;
3642 bytenr_orig = btrfs_sb_offset(copy_num);
3643 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3645 return ERR_PTR(-EINVAL);
3647 return ERR_PTR(ret);
3649 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3650 return ERR_PTR(-EINVAL);
3652 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3654 return ERR_CAST(page);
3656 super = page_address(page);
3657 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3658 btrfs_release_disk_super(super);
3659 return ERR_PTR(-ENODATA);
3662 if (btrfs_super_bytenr(super) != bytenr_orig) {
3663 btrfs_release_disk_super(super);
3664 return ERR_PTR(-EINVAL);
3671 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3673 struct btrfs_super_block *super, *latest = NULL;
3677 /* we would like to check all the supers, but that would make
3678 * a btrfs mount succeed after a mkfs from a different FS.
3679 * So, we need to add a special mount option to scan for
3680 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3682 for (i = 0; i < 1; i++) {
3683 super = btrfs_read_dev_one_super(bdev, i);
3687 if (!latest || btrfs_super_generation(super) > transid) {
3689 btrfs_release_disk_super(super);
3692 transid = btrfs_super_generation(super);
3700 * Write superblock @sb to the @device. Do not wait for completion, all the
3701 * pages we use for writing are locked.
3703 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3704 * the expected device size at commit time. Note that max_mirrors must be
3705 * same for write and wait phases.
3707 * Return number of errors when page is not found or submission fails.
3709 static int write_dev_supers(struct btrfs_device *device,
3710 struct btrfs_super_block *sb, int max_mirrors)
3712 struct btrfs_fs_info *fs_info = device->fs_info;
3713 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3714 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3718 u64 bytenr, bytenr_orig;
3720 if (max_mirrors == 0)
3721 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3723 shash->tfm = fs_info->csum_shash;
3725 for (i = 0; i < max_mirrors; i++) {
3728 struct btrfs_super_block *disk_super;
3730 bytenr_orig = btrfs_sb_offset(i);
3731 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3732 if (ret == -ENOENT) {
3734 } else if (ret < 0) {
3735 btrfs_err(device->fs_info,
3736 "couldn't get super block location for mirror %d",
3741 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3742 device->commit_total_bytes)
3745 btrfs_set_super_bytenr(sb, bytenr_orig);
3747 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3748 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3751 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3754 btrfs_err(device->fs_info,
3755 "couldn't get super block page for bytenr %llu",
3761 /* Bump the refcount for wait_dev_supers() */
3764 disk_super = page_address(page);
3765 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3768 * Directly use bios here instead of relying on the page cache
3769 * to do I/O, so we don't lose the ability to do integrity
3772 bio = bio_alloc(GFP_NOFS, 1);
3773 bio_set_dev(bio, device->bdev);
3774 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3775 bio->bi_private = device;
3776 bio->bi_end_io = btrfs_end_super_write;
3777 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3778 offset_in_page(bytenr));
3781 * We FUA only the first super block. The others we allow to
3782 * go down lazy and there's a short window where the on-disk
3783 * copies might still contain the older version.
3785 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3786 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3787 bio->bi_opf |= REQ_FUA;
3789 btrfsic_submit_bio(bio);
3790 btrfs_advance_sb_log(device, i);
3792 return errors < i ? 0 : -1;
3796 * Wait for write completion of superblocks done by write_dev_supers,
3797 * @max_mirrors same for write and wait phases.
3799 * Return number of errors when page is not found or not marked up to
3802 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3806 bool primary_failed = false;
3810 if (max_mirrors == 0)
3811 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3813 for (i = 0; i < max_mirrors; i++) {
3816 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3817 if (ret == -ENOENT) {
3819 } else if (ret < 0) {
3822 primary_failed = true;
3825 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3826 device->commit_total_bytes)
3829 page = find_get_page(device->bdev->bd_inode->i_mapping,
3830 bytenr >> PAGE_SHIFT);
3834 primary_failed = true;
3837 /* Page is submitted locked and unlocked once the IO completes */
3838 wait_on_page_locked(page);
3839 if (PageError(page)) {
3842 primary_failed = true;
3845 /* Drop our reference */
3848 /* Drop the reference from the writing run */
3852 /* log error, force error return */
3853 if (primary_failed) {
3854 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3859 return errors < i ? 0 : -1;
3863 * endio for the write_dev_flush, this will wake anyone waiting
3864 * for the barrier when it is done
3866 static void btrfs_end_empty_barrier(struct bio *bio)
3868 complete(bio->bi_private);
3872 * Submit a flush request to the device if it supports it. Error handling is
3873 * done in the waiting counterpart.
3875 static void write_dev_flush(struct btrfs_device *device)
3877 struct request_queue *q = bdev_get_queue(device->bdev);
3878 struct bio *bio = device->flush_bio;
3880 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3884 bio->bi_end_io = btrfs_end_empty_barrier;
3885 bio_set_dev(bio, device->bdev);
3886 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3887 init_completion(&device->flush_wait);
3888 bio->bi_private = &device->flush_wait;
3890 btrfsic_submit_bio(bio);
3891 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3895 * If the flush bio has been submitted by write_dev_flush, wait for it.
3897 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3899 struct bio *bio = device->flush_bio;
3901 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3904 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3905 wait_for_completion_io(&device->flush_wait);
3907 return bio->bi_status;
3910 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3912 if (!btrfs_check_rw_degradable(fs_info, NULL))
3918 * send an empty flush down to each device in parallel,
3919 * then wait for them
3921 static int barrier_all_devices(struct btrfs_fs_info *info)
3923 struct list_head *head;
3924 struct btrfs_device *dev;
3925 int errors_wait = 0;
3928 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3929 /* send down all the barriers */
3930 head = &info->fs_devices->devices;
3931 list_for_each_entry(dev, head, dev_list) {
3932 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3936 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3937 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3940 write_dev_flush(dev);
3941 dev->last_flush_error = BLK_STS_OK;
3944 /* wait for all the barriers */
3945 list_for_each_entry(dev, head, dev_list) {
3946 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3952 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3953 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3956 ret = wait_dev_flush(dev);
3958 dev->last_flush_error = ret;
3959 btrfs_dev_stat_inc_and_print(dev,
3960 BTRFS_DEV_STAT_FLUSH_ERRS);
3967 * At some point we need the status of all disks
3968 * to arrive at the volume status. So error checking
3969 * is being pushed to a separate loop.
3971 return check_barrier_error(info);
3976 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3979 int min_tolerated = INT_MAX;
3981 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3982 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3983 min_tolerated = min_t(int, min_tolerated,
3984 btrfs_raid_array[BTRFS_RAID_SINGLE].
3985 tolerated_failures);
3987 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3988 if (raid_type == BTRFS_RAID_SINGLE)
3990 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3992 min_tolerated = min_t(int, min_tolerated,
3993 btrfs_raid_array[raid_type].
3994 tolerated_failures);
3997 if (min_tolerated == INT_MAX) {
3998 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4002 return min_tolerated;
4005 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4007 struct list_head *head;
4008 struct btrfs_device *dev;
4009 struct btrfs_super_block *sb;
4010 struct btrfs_dev_item *dev_item;
4014 int total_errors = 0;
4017 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4020 * max_mirrors == 0 indicates we're from commit_transaction,
4021 * not from fsync where the tree roots in fs_info have not
4022 * been consistent on disk.
4024 if (max_mirrors == 0)
4025 backup_super_roots(fs_info);
4027 sb = fs_info->super_for_commit;
4028 dev_item = &sb->dev_item;
4030 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4031 head = &fs_info->fs_devices->devices;
4032 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4035 ret = barrier_all_devices(fs_info);
4038 &fs_info->fs_devices->device_list_mutex);
4039 btrfs_handle_fs_error(fs_info, ret,
4040 "errors while submitting device barriers.");
4045 list_for_each_entry(dev, head, dev_list) {
4050 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4051 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4054 btrfs_set_stack_device_generation(dev_item, 0);
4055 btrfs_set_stack_device_type(dev_item, dev->type);
4056 btrfs_set_stack_device_id(dev_item, dev->devid);
4057 btrfs_set_stack_device_total_bytes(dev_item,
4058 dev->commit_total_bytes);
4059 btrfs_set_stack_device_bytes_used(dev_item,
4060 dev->commit_bytes_used);
4061 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4062 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4063 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4064 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4065 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4068 flags = btrfs_super_flags(sb);
4069 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4071 ret = btrfs_validate_write_super(fs_info, sb);
4073 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4074 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4075 "unexpected superblock corruption detected");
4079 ret = write_dev_supers(dev, sb, max_mirrors);
4083 if (total_errors > max_errors) {
4084 btrfs_err(fs_info, "%d errors while writing supers",
4086 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4088 /* FUA is masked off if unsupported and can't be the reason */
4089 btrfs_handle_fs_error(fs_info, -EIO,
4090 "%d errors while writing supers",
4096 list_for_each_entry(dev, head, dev_list) {
4099 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4100 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4103 ret = wait_dev_supers(dev, max_mirrors);
4107 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4108 if (total_errors > max_errors) {
4109 btrfs_handle_fs_error(fs_info, -EIO,
4110 "%d errors while writing supers",
4117 /* Drop a fs root from the radix tree and free it. */
4118 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4119 struct btrfs_root *root)
4121 bool drop_ref = false;
4123 spin_lock(&fs_info->fs_roots_radix_lock);
4124 radix_tree_delete(&fs_info->fs_roots_radix,
4125 (unsigned long)root->root_key.objectid);
4126 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4128 spin_unlock(&fs_info->fs_roots_radix_lock);
4130 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4131 ASSERT(root->log_root == NULL);
4132 if (root->reloc_root) {
4133 btrfs_put_root(root->reloc_root);
4134 root->reloc_root = NULL;
4139 btrfs_put_root(root);
4142 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4144 u64 root_objectid = 0;
4145 struct btrfs_root *gang[8];
4148 unsigned int ret = 0;
4151 spin_lock(&fs_info->fs_roots_radix_lock);
4152 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4153 (void **)gang, root_objectid,
4156 spin_unlock(&fs_info->fs_roots_radix_lock);
4159 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4161 for (i = 0; i < ret; i++) {
4162 /* Avoid to grab roots in dead_roots */
4163 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4167 /* grab all the search result for later use */
4168 gang[i] = btrfs_grab_root(gang[i]);
4170 spin_unlock(&fs_info->fs_roots_radix_lock);
4172 for (i = 0; i < ret; i++) {
4175 root_objectid = gang[i]->root_key.objectid;
4176 err = btrfs_orphan_cleanup(gang[i]);
4179 btrfs_put_root(gang[i]);
4184 /* release the uncleaned roots due to error */
4185 for (; i < ret; i++) {
4187 btrfs_put_root(gang[i]);
4192 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4194 struct btrfs_root *root = fs_info->tree_root;
4195 struct btrfs_trans_handle *trans;
4197 mutex_lock(&fs_info->cleaner_mutex);
4198 btrfs_run_delayed_iputs(fs_info);
4199 mutex_unlock(&fs_info->cleaner_mutex);
4200 wake_up_process(fs_info->cleaner_kthread);
4202 /* wait until ongoing cleanup work done */
4203 down_write(&fs_info->cleanup_work_sem);
4204 up_write(&fs_info->cleanup_work_sem);
4206 trans = btrfs_join_transaction(root);
4208 return PTR_ERR(trans);
4209 return btrfs_commit_transaction(trans);
4212 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4216 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4218 * We don't want the cleaner to start new transactions, add more delayed
4219 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4220 * because that frees the task_struct, and the transaction kthread might
4221 * still try to wake up the cleaner.
4223 kthread_park(fs_info->cleaner_kthread);
4225 /* wait for the qgroup rescan worker to stop */
4226 btrfs_qgroup_wait_for_completion(fs_info, false);
4228 /* wait for the uuid_scan task to finish */
4229 down(&fs_info->uuid_tree_rescan_sem);
4230 /* avoid complains from lockdep et al., set sem back to initial state */
4231 up(&fs_info->uuid_tree_rescan_sem);
4233 /* pause restriper - we want to resume on mount */
4234 btrfs_pause_balance(fs_info);
4236 btrfs_dev_replace_suspend_for_unmount(fs_info);
4238 btrfs_scrub_cancel(fs_info);
4240 /* wait for any defraggers to finish */
4241 wait_event(fs_info->transaction_wait,
4242 (atomic_read(&fs_info->defrag_running) == 0));
4244 /* clear out the rbtree of defraggable inodes */
4245 btrfs_cleanup_defrag_inodes(fs_info);
4247 cancel_work_sync(&fs_info->async_reclaim_work);
4248 cancel_work_sync(&fs_info->async_data_reclaim_work);
4249 cancel_work_sync(&fs_info->preempt_reclaim_work);
4251 /* Cancel or finish ongoing discard work */
4252 btrfs_discard_cleanup(fs_info);
4254 if (!sb_rdonly(fs_info->sb)) {
4256 * The cleaner kthread is stopped, so do one final pass over
4257 * unused block groups.
4259 btrfs_delete_unused_bgs(fs_info);
4262 * There might be existing delayed inode workers still running
4263 * and holding an empty delayed inode item. We must wait for
4264 * them to complete first because they can create a transaction.
4265 * This happens when someone calls btrfs_balance_delayed_items()
4266 * and then a transaction commit runs the same delayed nodes
4267 * before any delayed worker has done something with the nodes.
4268 * We must wait for any worker here and not at transaction
4269 * commit time since that could cause a deadlock.
4270 * This is a very rare case.
4272 btrfs_flush_workqueue(fs_info->delayed_workers);
4274 ret = btrfs_commit_super(fs_info);
4276 btrfs_err(fs_info, "commit super ret %d", ret);
4279 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4280 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4281 btrfs_error_commit_super(fs_info);
4283 kthread_stop(fs_info->transaction_kthread);
4284 kthread_stop(fs_info->cleaner_kthread);
4286 ASSERT(list_empty(&fs_info->delayed_iputs));
4287 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4289 if (btrfs_check_quota_leak(fs_info)) {
4290 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4291 btrfs_err(fs_info, "qgroup reserved space leaked");
4294 btrfs_free_qgroup_config(fs_info);
4295 ASSERT(list_empty(&fs_info->delalloc_roots));
4297 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4298 btrfs_info(fs_info, "at unmount delalloc count %lld",
4299 percpu_counter_sum(&fs_info->delalloc_bytes));
4302 if (percpu_counter_sum(&fs_info->ordered_bytes))
4303 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4304 percpu_counter_sum(&fs_info->ordered_bytes));
4306 btrfs_sysfs_remove_mounted(fs_info);
4307 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4309 btrfs_put_block_group_cache(fs_info);
4312 * we must make sure there is not any read request to
4313 * submit after we stopping all workers.
4315 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4316 btrfs_stop_all_workers(fs_info);
4318 /* We shouldn't have any transaction open at this point */
4319 ASSERT(list_empty(&fs_info->trans_list));
4321 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4322 free_root_pointers(fs_info, true);
4323 btrfs_free_fs_roots(fs_info);
4326 * We must free the block groups after dropping the fs_roots as we could
4327 * have had an IO error and have left over tree log blocks that aren't
4328 * cleaned up until the fs roots are freed. This makes the block group
4329 * accounting appear to be wrong because there's pending reserved bytes,
4330 * so make sure we do the block group cleanup afterwards.
4332 btrfs_free_block_groups(fs_info);
4334 iput(fs_info->btree_inode);
4336 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4337 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4338 btrfsic_unmount(fs_info->fs_devices);
4341 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4342 btrfs_close_devices(fs_info->fs_devices);
4345 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4349 struct inode *btree_inode = buf->pages[0]->mapping->host;
4351 ret = extent_buffer_uptodate(buf);
4355 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4356 parent_transid, atomic);
4362 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4364 struct btrfs_fs_info *fs_info = buf->fs_info;
4365 u64 transid = btrfs_header_generation(buf);
4368 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4370 * This is a fast path so only do this check if we have sanity tests
4371 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4372 * outside of the sanity tests.
4374 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4377 btrfs_assert_tree_locked(buf);
4378 if (transid != fs_info->generation)
4379 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4380 buf->start, transid, fs_info->generation);
4381 was_dirty = set_extent_buffer_dirty(buf);
4383 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4385 fs_info->dirty_metadata_batch);
4386 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4388 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4389 * but item data not updated.
4390 * So here we should only check item pointers, not item data.
4392 if (btrfs_header_level(buf) == 0 &&
4393 btrfs_check_leaf_relaxed(buf)) {
4394 btrfs_print_leaf(buf);
4400 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4404 * looks as though older kernels can get into trouble with
4405 * this code, they end up stuck in balance_dirty_pages forever
4409 if (current->flags & PF_MEMALLOC)
4413 btrfs_balance_delayed_items(fs_info);
4415 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4416 BTRFS_DIRTY_METADATA_THRESH,
4417 fs_info->dirty_metadata_batch);
4419 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4423 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4425 __btrfs_btree_balance_dirty(fs_info, 1);
4428 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4430 __btrfs_btree_balance_dirty(fs_info, 0);
4433 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4434 struct btrfs_key *first_key)
4436 return btree_read_extent_buffer_pages(buf, parent_transid,
4440 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4442 /* cleanup FS via transaction */
4443 btrfs_cleanup_transaction(fs_info);
4445 mutex_lock(&fs_info->cleaner_mutex);
4446 btrfs_run_delayed_iputs(fs_info);
4447 mutex_unlock(&fs_info->cleaner_mutex);
4449 down_write(&fs_info->cleanup_work_sem);
4450 up_write(&fs_info->cleanup_work_sem);
4453 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4455 struct btrfs_root *gang[8];
4456 u64 root_objectid = 0;
4459 spin_lock(&fs_info->fs_roots_radix_lock);
4460 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4461 (void **)gang, root_objectid,
4462 ARRAY_SIZE(gang))) != 0) {
4465 for (i = 0; i < ret; i++)
4466 gang[i] = btrfs_grab_root(gang[i]);
4467 spin_unlock(&fs_info->fs_roots_radix_lock);
4469 for (i = 0; i < ret; i++) {
4472 root_objectid = gang[i]->root_key.objectid;
4473 btrfs_free_log(NULL, gang[i]);
4474 btrfs_put_root(gang[i]);
4477 spin_lock(&fs_info->fs_roots_radix_lock);
4479 spin_unlock(&fs_info->fs_roots_radix_lock);
4480 btrfs_free_log_root_tree(NULL, fs_info);
4483 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4485 struct btrfs_ordered_extent *ordered;
4487 spin_lock(&root->ordered_extent_lock);
4489 * This will just short circuit the ordered completion stuff which will
4490 * make sure the ordered extent gets properly cleaned up.
4492 list_for_each_entry(ordered, &root->ordered_extents,
4494 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4495 spin_unlock(&root->ordered_extent_lock);
4498 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4500 struct btrfs_root *root;
4501 struct list_head splice;
4503 INIT_LIST_HEAD(&splice);
4505 spin_lock(&fs_info->ordered_root_lock);
4506 list_splice_init(&fs_info->ordered_roots, &splice);
4507 while (!list_empty(&splice)) {
4508 root = list_first_entry(&splice, struct btrfs_root,
4510 list_move_tail(&root->ordered_root,
4511 &fs_info->ordered_roots);
4513 spin_unlock(&fs_info->ordered_root_lock);
4514 btrfs_destroy_ordered_extents(root);
4517 spin_lock(&fs_info->ordered_root_lock);
4519 spin_unlock(&fs_info->ordered_root_lock);
4522 * We need this here because if we've been flipped read-only we won't
4523 * get sync() from the umount, so we need to make sure any ordered
4524 * extents that haven't had their dirty pages IO start writeout yet
4525 * actually get run and error out properly.
4527 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4530 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4531 struct btrfs_fs_info *fs_info)
4533 struct rb_node *node;
4534 struct btrfs_delayed_ref_root *delayed_refs;
4535 struct btrfs_delayed_ref_node *ref;
4538 delayed_refs = &trans->delayed_refs;
4540 spin_lock(&delayed_refs->lock);
4541 if (atomic_read(&delayed_refs->num_entries) == 0) {
4542 spin_unlock(&delayed_refs->lock);
4543 btrfs_debug(fs_info, "delayed_refs has NO entry");
4547 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4548 struct btrfs_delayed_ref_head *head;
4550 bool pin_bytes = false;
4552 head = rb_entry(node, struct btrfs_delayed_ref_head,
4554 if (btrfs_delayed_ref_lock(delayed_refs, head))
4557 spin_lock(&head->lock);
4558 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4559 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4562 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4563 RB_CLEAR_NODE(&ref->ref_node);
4564 if (!list_empty(&ref->add_list))
4565 list_del(&ref->add_list);
4566 atomic_dec(&delayed_refs->num_entries);
4567 btrfs_put_delayed_ref(ref);
4569 if (head->must_insert_reserved)
4571 btrfs_free_delayed_extent_op(head->extent_op);
4572 btrfs_delete_ref_head(delayed_refs, head);
4573 spin_unlock(&head->lock);
4574 spin_unlock(&delayed_refs->lock);
4575 mutex_unlock(&head->mutex);
4578 struct btrfs_block_group *cache;
4580 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4583 spin_lock(&cache->space_info->lock);
4584 spin_lock(&cache->lock);
4585 cache->pinned += head->num_bytes;
4586 btrfs_space_info_update_bytes_pinned(fs_info,
4587 cache->space_info, head->num_bytes);
4588 cache->reserved -= head->num_bytes;
4589 cache->space_info->bytes_reserved -= head->num_bytes;
4590 spin_unlock(&cache->lock);
4591 spin_unlock(&cache->space_info->lock);
4592 percpu_counter_add_batch(
4593 &cache->space_info->total_bytes_pinned,
4594 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4596 btrfs_put_block_group(cache);
4598 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4599 head->bytenr + head->num_bytes - 1);
4601 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4602 btrfs_put_delayed_ref_head(head);
4604 spin_lock(&delayed_refs->lock);
4606 btrfs_qgroup_destroy_extent_records(trans);
4608 spin_unlock(&delayed_refs->lock);
4613 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4615 struct btrfs_inode *btrfs_inode;
4616 struct list_head splice;
4618 INIT_LIST_HEAD(&splice);
4620 spin_lock(&root->delalloc_lock);
4621 list_splice_init(&root->delalloc_inodes, &splice);
4623 while (!list_empty(&splice)) {
4624 struct inode *inode = NULL;
4625 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4627 __btrfs_del_delalloc_inode(root, btrfs_inode);
4628 spin_unlock(&root->delalloc_lock);
4631 * Make sure we get a live inode and that it'll not disappear
4634 inode = igrab(&btrfs_inode->vfs_inode);
4636 invalidate_inode_pages2(inode->i_mapping);
4639 spin_lock(&root->delalloc_lock);
4641 spin_unlock(&root->delalloc_lock);
4644 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4646 struct btrfs_root *root;
4647 struct list_head splice;
4649 INIT_LIST_HEAD(&splice);
4651 spin_lock(&fs_info->delalloc_root_lock);
4652 list_splice_init(&fs_info->delalloc_roots, &splice);
4653 while (!list_empty(&splice)) {
4654 root = list_first_entry(&splice, struct btrfs_root,
4656 root = btrfs_grab_root(root);
4658 spin_unlock(&fs_info->delalloc_root_lock);
4660 btrfs_destroy_delalloc_inodes(root);
4661 btrfs_put_root(root);
4663 spin_lock(&fs_info->delalloc_root_lock);
4665 spin_unlock(&fs_info->delalloc_root_lock);
4668 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4669 struct extent_io_tree *dirty_pages,
4673 struct extent_buffer *eb;
4678 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4683 clear_extent_bits(dirty_pages, start, end, mark);
4684 while (start <= end) {
4685 eb = find_extent_buffer(fs_info, start);
4686 start += fs_info->nodesize;
4689 wait_on_extent_buffer_writeback(eb);
4691 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4693 clear_extent_buffer_dirty(eb);
4694 free_extent_buffer_stale(eb);
4701 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4702 struct extent_io_tree *unpin)
4709 struct extent_state *cached_state = NULL;
4712 * The btrfs_finish_extent_commit() may get the same range as
4713 * ours between find_first_extent_bit and clear_extent_dirty.
4714 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4715 * the same extent range.
4717 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4718 ret = find_first_extent_bit(unpin, 0, &start, &end,
4719 EXTENT_DIRTY, &cached_state);
4721 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4725 clear_extent_dirty(unpin, start, end, &cached_state);
4726 free_extent_state(cached_state);
4727 btrfs_error_unpin_extent_range(fs_info, start, end);
4728 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4735 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4737 struct inode *inode;
4739 inode = cache->io_ctl.inode;
4741 invalidate_inode_pages2(inode->i_mapping);
4742 BTRFS_I(inode)->generation = 0;
4743 cache->io_ctl.inode = NULL;
4746 ASSERT(cache->io_ctl.pages == NULL);
4747 btrfs_put_block_group(cache);
4750 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4751 struct btrfs_fs_info *fs_info)
4753 struct btrfs_block_group *cache;
4755 spin_lock(&cur_trans->dirty_bgs_lock);
4756 while (!list_empty(&cur_trans->dirty_bgs)) {
4757 cache = list_first_entry(&cur_trans->dirty_bgs,
4758 struct btrfs_block_group,
4761 if (!list_empty(&cache->io_list)) {
4762 spin_unlock(&cur_trans->dirty_bgs_lock);
4763 list_del_init(&cache->io_list);
4764 btrfs_cleanup_bg_io(cache);
4765 spin_lock(&cur_trans->dirty_bgs_lock);
4768 list_del_init(&cache->dirty_list);
4769 spin_lock(&cache->lock);
4770 cache->disk_cache_state = BTRFS_DC_ERROR;
4771 spin_unlock(&cache->lock);
4773 spin_unlock(&cur_trans->dirty_bgs_lock);
4774 btrfs_put_block_group(cache);
4775 btrfs_delayed_refs_rsv_release(fs_info, 1);
4776 spin_lock(&cur_trans->dirty_bgs_lock);
4778 spin_unlock(&cur_trans->dirty_bgs_lock);
4781 * Refer to the definition of io_bgs member for details why it's safe
4782 * to use it without any locking
4784 while (!list_empty(&cur_trans->io_bgs)) {
4785 cache = list_first_entry(&cur_trans->io_bgs,
4786 struct btrfs_block_group,
4789 list_del_init(&cache->io_list);
4790 spin_lock(&cache->lock);
4791 cache->disk_cache_state = BTRFS_DC_ERROR;
4792 spin_unlock(&cache->lock);
4793 btrfs_cleanup_bg_io(cache);
4797 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4798 struct btrfs_fs_info *fs_info)
4800 struct btrfs_device *dev, *tmp;
4802 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4803 ASSERT(list_empty(&cur_trans->dirty_bgs));
4804 ASSERT(list_empty(&cur_trans->io_bgs));
4806 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4808 list_del_init(&dev->post_commit_list);
4811 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4813 cur_trans->state = TRANS_STATE_COMMIT_START;
4814 wake_up(&fs_info->transaction_blocked_wait);
4816 cur_trans->state = TRANS_STATE_UNBLOCKED;
4817 wake_up(&fs_info->transaction_wait);
4819 btrfs_destroy_delayed_inodes(fs_info);
4821 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4823 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4825 btrfs_free_redirty_list(cur_trans);
4827 cur_trans->state =TRANS_STATE_COMPLETED;
4828 wake_up(&cur_trans->commit_wait);
4831 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4833 struct btrfs_transaction *t;
4835 mutex_lock(&fs_info->transaction_kthread_mutex);
4837 spin_lock(&fs_info->trans_lock);
4838 while (!list_empty(&fs_info->trans_list)) {
4839 t = list_first_entry(&fs_info->trans_list,
4840 struct btrfs_transaction, list);
4841 if (t->state >= TRANS_STATE_COMMIT_START) {
4842 refcount_inc(&t->use_count);
4843 spin_unlock(&fs_info->trans_lock);
4844 btrfs_wait_for_commit(fs_info, t->transid);
4845 btrfs_put_transaction(t);
4846 spin_lock(&fs_info->trans_lock);
4849 if (t == fs_info->running_transaction) {
4850 t->state = TRANS_STATE_COMMIT_DOING;
4851 spin_unlock(&fs_info->trans_lock);
4853 * We wait for 0 num_writers since we don't hold a trans
4854 * handle open currently for this transaction.
4856 wait_event(t->writer_wait,
4857 atomic_read(&t->num_writers) == 0);
4859 spin_unlock(&fs_info->trans_lock);
4861 btrfs_cleanup_one_transaction(t, fs_info);
4863 spin_lock(&fs_info->trans_lock);
4864 if (t == fs_info->running_transaction)
4865 fs_info->running_transaction = NULL;
4866 list_del_init(&t->list);
4867 spin_unlock(&fs_info->trans_lock);
4869 btrfs_put_transaction(t);
4870 trace_btrfs_transaction_commit(fs_info->tree_root);
4871 spin_lock(&fs_info->trans_lock);
4873 spin_unlock(&fs_info->trans_lock);
4874 btrfs_destroy_all_ordered_extents(fs_info);
4875 btrfs_destroy_delayed_inodes(fs_info);
4876 btrfs_assert_delayed_root_empty(fs_info);
4877 btrfs_destroy_all_delalloc_inodes(fs_info);
4878 btrfs_drop_all_logs(fs_info);
4879 mutex_unlock(&fs_info->transaction_kthread_mutex);
4884 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4886 struct btrfs_path *path;
4888 struct extent_buffer *l;
4889 struct btrfs_key search_key;
4890 struct btrfs_key found_key;
4893 path = btrfs_alloc_path();
4897 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4898 search_key.type = -1;
4899 search_key.offset = (u64)-1;
4900 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4903 BUG_ON(ret == 0); /* Corruption */
4904 if (path->slots[0] > 0) {
4905 slot = path->slots[0] - 1;
4907 btrfs_item_key_to_cpu(l, &found_key, slot);
4908 root->free_objectid = max_t(u64, found_key.objectid + 1,
4909 BTRFS_FIRST_FREE_OBJECTID);
4911 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4915 btrfs_free_path(path);
4919 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4922 mutex_lock(&root->objectid_mutex);
4924 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4925 btrfs_warn(root->fs_info,
4926 "the objectid of root %llu reaches its highest value",
4927 root->root_key.objectid);
4932 *objectid = root->free_objectid++;
4935 mutex_unlock(&root->objectid_mutex);