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 * Please do not consolidate these warnings into a single if.
464 * It is useful to know what went wrong.
466 if (WARN_ON(found_start != start))
468 if (WARN_ON(!PageUptodate(page)))
471 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
472 offsetof(struct btrfs_header, fsid),
473 BTRFS_FSID_SIZE) == 0);
475 csum_tree_block(eb, result);
477 if (btrfs_header_level(eb))
478 ret = btrfs_check_node(eb);
480 ret = btrfs_check_leaf_full(eb);
483 btrfs_print_tree(eb, 0);
485 "block=%llu write time tree block corruption detected",
487 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
490 write_extent_buffer(eb, result, 0, fs_info->csum_size);
495 static int check_tree_block_fsid(struct extent_buffer *eb)
497 struct btrfs_fs_info *fs_info = eb->fs_info;
498 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
499 u8 fsid[BTRFS_FSID_SIZE];
502 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
505 * Checking the incompat flag is only valid for the current fs. For
506 * seed devices it's forbidden to have their uuid changed so reading
507 * ->fsid in this case is fine
509 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
510 metadata_uuid = fs_devices->metadata_uuid;
512 metadata_uuid = fs_devices->fsid;
514 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
517 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
518 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
524 /* Do basic extent buffer checks at read time */
525 static int validate_extent_buffer(struct extent_buffer *eb)
527 struct btrfs_fs_info *fs_info = eb->fs_info;
529 const u32 csum_size = fs_info->csum_size;
531 u8 result[BTRFS_CSUM_SIZE];
534 found_start = btrfs_header_bytenr(eb);
535 if (found_start != eb->start) {
536 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
537 eb->start, found_start);
541 if (check_tree_block_fsid(eb)) {
542 btrfs_err_rl(fs_info, "bad fsid on block %llu",
547 found_level = btrfs_header_level(eb);
548 if (found_level >= BTRFS_MAX_LEVEL) {
549 btrfs_err(fs_info, "bad tree block level %d on %llu",
550 (int)btrfs_header_level(eb), eb->start);
555 csum_tree_block(eb, result);
557 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
558 u8 val[BTRFS_CSUM_SIZE] = { 0 };
560 read_extent_buffer(eb, &val, 0, csum_size);
561 btrfs_warn_rl(fs_info,
562 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
563 fs_info->sb->s_id, eb->start,
564 CSUM_FMT_VALUE(csum_size, val),
565 CSUM_FMT_VALUE(csum_size, result),
566 btrfs_header_level(eb));
572 * If this is a leaf block and it is corrupt, set the corrupt bit so
573 * that we don't try and read the other copies of this block, just
576 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
577 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
581 if (found_level > 0 && btrfs_check_node(eb))
585 set_extent_buffer_uptodate(eb);
588 "block=%llu read time tree block corruption detected",
594 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
597 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
598 struct extent_buffer *eb;
603 * We don't allow bio merge for subpage metadata read, so we should
604 * only get one eb for each endio hook.
606 ASSERT(end == start + fs_info->nodesize - 1);
607 ASSERT(PagePrivate(page));
609 eb = find_extent_buffer(fs_info, start);
611 * When we are reading one tree block, eb must have been inserted into
612 * the radix tree. If not, something is wrong.
616 reads_done = atomic_dec_and_test(&eb->io_pages);
617 /* Subpage read must finish in page read */
620 eb->read_mirror = mirror;
621 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
625 ret = validate_extent_buffer(eb);
629 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
630 btree_readahead_hook(eb, ret);
632 set_extent_buffer_uptodate(eb);
634 free_extent_buffer(eb);
638 * end_bio_extent_readpage decrements io_pages in case of error,
639 * make sure it has something to decrement.
641 atomic_inc(&eb->io_pages);
642 clear_extent_buffer_uptodate(eb);
643 free_extent_buffer(eb);
647 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
648 struct page *page, u64 start, u64 end,
651 struct extent_buffer *eb;
655 ASSERT(page->private);
657 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
658 return validate_subpage_buffer(page, start, end, mirror);
660 eb = (struct extent_buffer *)page->private;
663 * The pending IO might have been the only thing that kept this buffer
664 * in memory. Make sure we have a ref for all this other checks
666 atomic_inc(&eb->refs);
668 reads_done = atomic_dec_and_test(&eb->io_pages);
672 eb->read_mirror = mirror;
673 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
677 ret = validate_extent_buffer(eb);
680 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
681 btree_readahead_hook(eb, ret);
685 * our io error hook is going to dec the io pages
686 * again, we have to make sure it has something
689 atomic_inc(&eb->io_pages);
690 clear_extent_buffer_uptodate(eb);
692 free_extent_buffer(eb);
697 static void end_workqueue_bio(struct bio *bio)
699 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
700 struct btrfs_fs_info *fs_info;
701 struct btrfs_workqueue *wq;
703 fs_info = end_io_wq->info;
704 end_io_wq->status = bio->bi_status;
706 if (bio_op(bio) == REQ_OP_WRITE) {
707 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
708 wq = fs_info->endio_meta_write_workers;
709 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
710 wq = fs_info->endio_freespace_worker;
711 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
712 wq = fs_info->endio_raid56_workers;
714 wq = fs_info->endio_write_workers;
716 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
717 wq = fs_info->endio_raid56_workers;
718 else if (end_io_wq->metadata)
719 wq = fs_info->endio_meta_workers;
721 wq = fs_info->endio_workers;
724 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
725 btrfs_queue_work(wq, &end_io_wq->work);
728 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
729 enum btrfs_wq_endio_type metadata)
731 struct btrfs_end_io_wq *end_io_wq;
733 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
735 return BLK_STS_RESOURCE;
737 end_io_wq->private = bio->bi_private;
738 end_io_wq->end_io = bio->bi_end_io;
739 end_io_wq->info = info;
740 end_io_wq->status = 0;
741 end_io_wq->bio = bio;
742 end_io_wq->metadata = metadata;
744 bio->bi_private = end_io_wq;
745 bio->bi_end_io = end_workqueue_bio;
749 static void run_one_async_start(struct btrfs_work *work)
751 struct async_submit_bio *async;
754 async = container_of(work, struct async_submit_bio, work);
755 ret = async->submit_bio_start(async->inode, async->bio,
756 async->dio_file_offset);
762 * In order to insert checksums into the metadata in large chunks, we wait
763 * until bio submission time. All the pages in the bio are checksummed and
764 * sums are attached onto the ordered extent record.
766 * At IO completion time the csums attached on the ordered extent record are
767 * inserted into the tree.
769 static void run_one_async_done(struct btrfs_work *work)
771 struct async_submit_bio *async;
775 async = container_of(work, struct async_submit_bio, work);
776 inode = async->inode;
778 /* If an error occurred we just want to clean up the bio and move on */
780 async->bio->bi_status = async->status;
781 bio_endio(async->bio);
786 * All of the bios that pass through here are from async helpers.
787 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
788 * This changes nothing when cgroups aren't in use.
790 async->bio->bi_opf |= REQ_CGROUP_PUNT;
791 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
793 async->bio->bi_status = ret;
794 bio_endio(async->bio);
798 static void run_one_async_free(struct btrfs_work *work)
800 struct async_submit_bio *async;
802 async = container_of(work, struct async_submit_bio, work);
806 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
807 int mirror_num, unsigned long bio_flags,
809 extent_submit_bio_start_t *submit_bio_start)
811 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
812 struct async_submit_bio *async;
814 async = kmalloc(sizeof(*async), GFP_NOFS);
816 return BLK_STS_RESOURCE;
818 async->inode = inode;
820 async->mirror_num = mirror_num;
821 async->submit_bio_start = submit_bio_start;
823 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
826 async->dio_file_offset = dio_file_offset;
830 if (op_is_sync(bio->bi_opf))
831 btrfs_set_work_high_priority(&async->work);
833 btrfs_queue_work(fs_info->workers, &async->work);
837 static blk_status_t btree_csum_one_bio(struct bio *bio)
839 struct bio_vec *bvec;
840 struct btrfs_root *root;
842 struct bvec_iter_all iter_all;
844 ASSERT(!bio_flagged(bio, BIO_CLONED));
845 bio_for_each_segment_all(bvec, bio, iter_all) {
846 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
847 ret = csum_dirty_buffer(root->fs_info, bvec);
852 return errno_to_blk_status(ret);
855 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
859 * when we're called for a write, we're already in the async
860 * submission context. Just jump into btrfs_map_bio
862 return btree_csum_one_bio(bio);
865 static int check_async_write(struct btrfs_fs_info *fs_info,
866 struct btrfs_inode *bi)
868 if (atomic_read(&bi->sync_writers))
870 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
875 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
876 int mirror_num, unsigned long bio_flags)
878 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
879 int async = check_async_write(fs_info, BTRFS_I(inode));
882 if (bio_op(bio) != REQ_OP_WRITE) {
884 * called for a read, do the setup so that checksum validation
885 * can happen in the async kernel threads
887 ret = btrfs_bio_wq_end_io(fs_info, bio,
888 BTRFS_WQ_ENDIO_METADATA);
891 ret = btrfs_map_bio(fs_info, bio, mirror_num);
893 ret = btree_csum_one_bio(bio);
896 ret = btrfs_map_bio(fs_info, bio, mirror_num);
899 * kthread helpers are used to submit writes so that
900 * checksumming can happen in parallel across all CPUs
902 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
903 0, btree_submit_bio_start);
911 bio->bi_status = ret;
916 #ifdef CONFIG_MIGRATION
917 static int btree_migratepage(struct address_space *mapping,
918 struct page *newpage, struct page *page,
919 enum migrate_mode mode)
922 * we can't safely write a btree page from here,
923 * we haven't done the locking hook
928 * Buffers may be managed in a filesystem specific way.
929 * We must have no buffers or drop them.
931 if (page_has_private(page) &&
932 !try_to_release_page(page, GFP_KERNEL))
934 return migrate_page(mapping, newpage, page, mode);
939 static int btree_writepages(struct address_space *mapping,
940 struct writeback_control *wbc)
942 struct btrfs_fs_info *fs_info;
945 if (wbc->sync_mode == WB_SYNC_NONE) {
947 if (wbc->for_kupdate)
950 fs_info = BTRFS_I(mapping->host)->root->fs_info;
951 /* this is a bit racy, but that's ok */
952 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
953 BTRFS_DIRTY_METADATA_THRESH,
954 fs_info->dirty_metadata_batch);
958 return btree_write_cache_pages(mapping, wbc);
961 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
963 if (PageWriteback(page) || PageDirty(page))
966 return try_release_extent_buffer(page);
969 static void btree_invalidatepage(struct page *page, unsigned int offset,
972 struct extent_io_tree *tree;
973 tree = &BTRFS_I(page->mapping->host)->io_tree;
974 extent_invalidatepage(tree, page, offset);
975 btree_releasepage(page, GFP_NOFS);
976 if (PagePrivate(page)) {
977 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
978 "page private not zero on page %llu",
979 (unsigned long long)page_offset(page));
980 detach_page_private(page);
984 static int btree_set_page_dirty(struct page *page)
987 struct extent_buffer *eb;
989 BUG_ON(!PagePrivate(page));
990 eb = (struct extent_buffer *)page->private;
992 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
993 BUG_ON(!atomic_read(&eb->refs));
994 btrfs_assert_tree_locked(eb);
996 return __set_page_dirty_nobuffers(page);
999 static const struct address_space_operations btree_aops = {
1000 .writepages = btree_writepages,
1001 .releasepage = btree_releasepage,
1002 .invalidatepage = btree_invalidatepage,
1003 #ifdef CONFIG_MIGRATION
1004 .migratepage = btree_migratepage,
1006 .set_page_dirty = btree_set_page_dirty,
1009 struct extent_buffer *btrfs_find_create_tree_block(
1010 struct btrfs_fs_info *fs_info,
1011 u64 bytenr, u64 owner_root,
1014 if (btrfs_is_testing(fs_info))
1015 return alloc_test_extent_buffer(fs_info, bytenr);
1016 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1020 * Read tree block at logical address @bytenr and do variant basic but critical
1023 * @owner_root: the objectid of the root owner for this block.
1024 * @parent_transid: expected transid of this tree block, skip check if 0
1025 * @level: expected level, mandatory check
1026 * @first_key: expected key in slot 0, skip check if NULL
1028 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1029 u64 owner_root, u64 parent_transid,
1030 int level, struct btrfs_key *first_key)
1032 struct extent_buffer *buf = NULL;
1035 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1039 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1042 free_extent_buffer_stale(buf);
1043 return ERR_PTR(ret);
1049 void btrfs_clean_tree_block(struct extent_buffer *buf)
1051 struct btrfs_fs_info *fs_info = buf->fs_info;
1052 if (btrfs_header_generation(buf) ==
1053 fs_info->running_transaction->transid) {
1054 btrfs_assert_tree_locked(buf);
1056 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1057 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1059 fs_info->dirty_metadata_batch);
1060 clear_extent_buffer_dirty(buf);
1065 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1068 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1069 root->fs_info = fs_info;
1071 root->commit_root = NULL;
1073 root->orphan_cleanup_state = 0;
1075 root->last_trans = 0;
1076 root->free_objectid = 0;
1077 root->nr_delalloc_inodes = 0;
1078 root->nr_ordered_extents = 0;
1079 root->inode_tree = RB_ROOT;
1080 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1081 root->block_rsv = NULL;
1083 INIT_LIST_HEAD(&root->dirty_list);
1084 INIT_LIST_HEAD(&root->root_list);
1085 INIT_LIST_HEAD(&root->delalloc_inodes);
1086 INIT_LIST_HEAD(&root->delalloc_root);
1087 INIT_LIST_HEAD(&root->ordered_extents);
1088 INIT_LIST_HEAD(&root->ordered_root);
1089 INIT_LIST_HEAD(&root->reloc_dirty_list);
1090 INIT_LIST_HEAD(&root->logged_list[0]);
1091 INIT_LIST_HEAD(&root->logged_list[1]);
1092 spin_lock_init(&root->inode_lock);
1093 spin_lock_init(&root->delalloc_lock);
1094 spin_lock_init(&root->ordered_extent_lock);
1095 spin_lock_init(&root->accounting_lock);
1096 spin_lock_init(&root->log_extents_lock[0]);
1097 spin_lock_init(&root->log_extents_lock[1]);
1098 spin_lock_init(&root->qgroup_meta_rsv_lock);
1099 mutex_init(&root->objectid_mutex);
1100 mutex_init(&root->log_mutex);
1101 mutex_init(&root->ordered_extent_mutex);
1102 mutex_init(&root->delalloc_mutex);
1103 init_waitqueue_head(&root->qgroup_flush_wait);
1104 init_waitqueue_head(&root->log_writer_wait);
1105 init_waitqueue_head(&root->log_commit_wait[0]);
1106 init_waitqueue_head(&root->log_commit_wait[1]);
1107 INIT_LIST_HEAD(&root->log_ctxs[0]);
1108 INIT_LIST_HEAD(&root->log_ctxs[1]);
1109 atomic_set(&root->log_commit[0], 0);
1110 atomic_set(&root->log_commit[1], 0);
1111 atomic_set(&root->log_writers, 0);
1112 atomic_set(&root->log_batch, 0);
1113 refcount_set(&root->refs, 1);
1114 atomic_set(&root->snapshot_force_cow, 0);
1115 atomic_set(&root->nr_swapfiles, 0);
1116 root->log_transid = 0;
1117 root->log_transid_committed = -1;
1118 root->last_log_commit = 0;
1120 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1121 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1122 extent_io_tree_init(fs_info, &root->log_csum_range,
1123 IO_TREE_LOG_CSUM_RANGE, NULL);
1126 memset(&root->root_key, 0, sizeof(root->root_key));
1127 memset(&root->root_item, 0, sizeof(root->root_item));
1128 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1129 root->root_key.objectid = objectid;
1132 spin_lock_init(&root->root_item_lock);
1133 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1134 #ifdef CONFIG_BTRFS_DEBUG
1135 INIT_LIST_HEAD(&root->leak_list);
1136 spin_lock(&fs_info->fs_roots_radix_lock);
1137 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1138 spin_unlock(&fs_info->fs_roots_radix_lock);
1142 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1143 u64 objectid, gfp_t flags)
1145 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1147 __setup_root(root, fs_info, objectid);
1151 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1152 /* Should only be used by the testing infrastructure */
1153 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1155 struct btrfs_root *root;
1158 return ERR_PTR(-EINVAL);
1160 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1162 return ERR_PTR(-ENOMEM);
1164 /* We don't use the stripesize in selftest, set it as sectorsize */
1165 root->alloc_bytenr = 0;
1171 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1174 struct btrfs_fs_info *fs_info = trans->fs_info;
1175 struct extent_buffer *leaf;
1176 struct btrfs_root *tree_root = fs_info->tree_root;
1177 struct btrfs_root *root;
1178 struct btrfs_key key;
1179 unsigned int nofs_flag;
1183 * We're holding a transaction handle, so use a NOFS memory allocation
1184 * context to avoid deadlock if reclaim happens.
1186 nofs_flag = memalloc_nofs_save();
1187 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1188 memalloc_nofs_restore(nofs_flag);
1190 return ERR_PTR(-ENOMEM);
1192 root->root_key.objectid = objectid;
1193 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1194 root->root_key.offset = 0;
1196 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1197 BTRFS_NESTING_NORMAL);
1199 ret = PTR_ERR(leaf);
1205 btrfs_mark_buffer_dirty(leaf);
1207 root->commit_root = btrfs_root_node(root);
1208 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1210 btrfs_set_root_flags(&root->root_item, 0);
1211 btrfs_set_root_limit(&root->root_item, 0);
1212 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1213 btrfs_set_root_generation(&root->root_item, trans->transid);
1214 btrfs_set_root_level(&root->root_item, 0);
1215 btrfs_set_root_refs(&root->root_item, 1);
1216 btrfs_set_root_used(&root->root_item, leaf->len);
1217 btrfs_set_root_last_snapshot(&root->root_item, 0);
1218 btrfs_set_root_dirid(&root->root_item, 0);
1219 if (is_fstree(objectid))
1220 generate_random_guid(root->root_item.uuid);
1222 export_guid(root->root_item.uuid, &guid_null);
1223 btrfs_set_root_drop_level(&root->root_item, 0);
1225 btrfs_tree_unlock(leaf);
1227 key.objectid = objectid;
1228 key.type = BTRFS_ROOT_ITEM_KEY;
1230 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1238 btrfs_tree_unlock(leaf);
1240 btrfs_put_root(root);
1242 return ERR_PTR(ret);
1245 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1246 struct btrfs_fs_info *fs_info)
1248 struct btrfs_root *root;
1249 struct extent_buffer *leaf;
1251 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1253 return ERR_PTR(-ENOMEM);
1255 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1256 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1257 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1260 * DON'T set SHAREABLE bit for log trees.
1262 * Log trees are not exposed to user space thus can't be snapshotted,
1263 * and they go away before a real commit is actually done.
1265 * They do store pointers to file data extents, and those reference
1266 * counts still get updated (along with back refs to the log tree).
1269 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1270 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1272 btrfs_put_root(root);
1273 return ERR_CAST(leaf);
1278 btrfs_mark_buffer_dirty(root->node);
1279 btrfs_tree_unlock(root->node);
1283 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1284 struct btrfs_fs_info *fs_info)
1286 struct btrfs_root *log_root;
1288 log_root = alloc_log_tree(trans, fs_info);
1289 if (IS_ERR(log_root))
1290 return PTR_ERR(log_root);
1291 WARN_ON(fs_info->log_root_tree);
1292 fs_info->log_root_tree = log_root;
1296 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1297 struct btrfs_root *root)
1299 struct btrfs_fs_info *fs_info = root->fs_info;
1300 struct btrfs_root *log_root;
1301 struct btrfs_inode_item *inode_item;
1303 log_root = alloc_log_tree(trans, fs_info);
1304 if (IS_ERR(log_root))
1305 return PTR_ERR(log_root);
1307 log_root->last_trans = trans->transid;
1308 log_root->root_key.offset = root->root_key.objectid;
1310 inode_item = &log_root->root_item.inode;
1311 btrfs_set_stack_inode_generation(inode_item, 1);
1312 btrfs_set_stack_inode_size(inode_item, 3);
1313 btrfs_set_stack_inode_nlink(inode_item, 1);
1314 btrfs_set_stack_inode_nbytes(inode_item,
1316 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1318 btrfs_set_root_node(&log_root->root_item, log_root->node);
1320 WARN_ON(root->log_root);
1321 root->log_root = log_root;
1322 root->log_transid = 0;
1323 root->log_transid_committed = -1;
1324 root->last_log_commit = 0;
1328 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1329 struct btrfs_path *path,
1330 struct btrfs_key *key)
1332 struct btrfs_root *root;
1333 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1338 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1340 return ERR_PTR(-ENOMEM);
1342 ret = btrfs_find_root(tree_root, key, path,
1343 &root->root_item, &root->root_key);
1350 generation = btrfs_root_generation(&root->root_item);
1351 level = btrfs_root_level(&root->root_item);
1352 root->node = read_tree_block(fs_info,
1353 btrfs_root_bytenr(&root->root_item),
1354 key->objectid, generation, level, NULL);
1355 if (IS_ERR(root->node)) {
1356 ret = PTR_ERR(root->node);
1359 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1363 root->commit_root = btrfs_root_node(root);
1366 btrfs_put_root(root);
1367 return ERR_PTR(ret);
1370 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1371 struct btrfs_key *key)
1373 struct btrfs_root *root;
1374 struct btrfs_path *path;
1376 path = btrfs_alloc_path();
1378 return ERR_PTR(-ENOMEM);
1379 root = read_tree_root_path(tree_root, path, key);
1380 btrfs_free_path(path);
1386 * Initialize subvolume root in-memory structure
1388 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1390 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1393 unsigned int nofs_flag;
1396 * We might be called under a transaction (e.g. indirect backref
1397 * resolution) which could deadlock if it triggers memory reclaim
1399 nofs_flag = memalloc_nofs_save();
1400 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1401 memalloc_nofs_restore(nofs_flag);
1405 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1406 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1407 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1408 btrfs_check_and_init_root_item(&root->root_item);
1412 * Don't assign anonymous block device to roots that are not exposed to
1413 * userspace, the id pool is limited to 1M
1415 if (is_fstree(root->root_key.objectid) &&
1416 btrfs_root_refs(&root->root_item) > 0) {
1418 ret = get_anon_bdev(&root->anon_dev);
1422 root->anon_dev = anon_dev;
1426 mutex_lock(&root->objectid_mutex);
1427 ret = btrfs_init_root_free_objectid(root);
1429 mutex_unlock(&root->objectid_mutex);
1433 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1435 mutex_unlock(&root->objectid_mutex);
1439 /* The caller is responsible to call btrfs_free_fs_root */
1443 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1446 struct btrfs_root *root;
1448 spin_lock(&fs_info->fs_roots_radix_lock);
1449 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1450 (unsigned long)root_id);
1452 root = btrfs_grab_root(root);
1453 spin_unlock(&fs_info->fs_roots_radix_lock);
1457 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1460 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1461 return btrfs_grab_root(fs_info->tree_root);
1462 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1463 return btrfs_grab_root(fs_info->extent_root);
1464 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1465 return btrfs_grab_root(fs_info->chunk_root);
1466 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1467 return btrfs_grab_root(fs_info->dev_root);
1468 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1469 return btrfs_grab_root(fs_info->csum_root);
1470 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1471 return btrfs_grab_root(fs_info->quota_root) ?
1472 fs_info->quota_root : ERR_PTR(-ENOENT);
1473 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1474 return btrfs_grab_root(fs_info->uuid_root) ?
1475 fs_info->uuid_root : ERR_PTR(-ENOENT);
1476 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1477 return btrfs_grab_root(fs_info->free_space_root) ?
1478 fs_info->free_space_root : ERR_PTR(-ENOENT);
1482 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1483 struct btrfs_root *root)
1487 ret = radix_tree_preload(GFP_NOFS);
1491 spin_lock(&fs_info->fs_roots_radix_lock);
1492 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1493 (unsigned long)root->root_key.objectid,
1496 btrfs_grab_root(root);
1497 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1499 spin_unlock(&fs_info->fs_roots_radix_lock);
1500 radix_tree_preload_end();
1505 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1507 #ifdef CONFIG_BTRFS_DEBUG
1508 struct btrfs_root *root;
1510 while (!list_empty(&fs_info->allocated_roots)) {
1511 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1513 root = list_first_entry(&fs_info->allocated_roots,
1514 struct btrfs_root, leak_list);
1515 btrfs_err(fs_info, "leaked root %s refcount %d",
1516 btrfs_root_name(&root->root_key, buf),
1517 refcount_read(&root->refs));
1518 while (refcount_read(&root->refs) > 1)
1519 btrfs_put_root(root);
1520 btrfs_put_root(root);
1525 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1527 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1528 percpu_counter_destroy(&fs_info->delalloc_bytes);
1529 percpu_counter_destroy(&fs_info->ordered_bytes);
1530 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1531 btrfs_free_csum_hash(fs_info);
1532 btrfs_free_stripe_hash_table(fs_info);
1533 btrfs_free_ref_cache(fs_info);
1534 kfree(fs_info->balance_ctl);
1535 kfree(fs_info->delayed_root);
1536 btrfs_put_root(fs_info->extent_root);
1537 btrfs_put_root(fs_info->tree_root);
1538 btrfs_put_root(fs_info->chunk_root);
1539 btrfs_put_root(fs_info->dev_root);
1540 btrfs_put_root(fs_info->csum_root);
1541 btrfs_put_root(fs_info->quota_root);
1542 btrfs_put_root(fs_info->uuid_root);
1543 btrfs_put_root(fs_info->free_space_root);
1544 btrfs_put_root(fs_info->fs_root);
1545 btrfs_put_root(fs_info->data_reloc_root);
1546 btrfs_check_leaked_roots(fs_info);
1547 btrfs_extent_buffer_leak_debug_check(fs_info);
1548 kfree(fs_info->super_copy);
1549 kfree(fs_info->super_for_commit);
1555 * Get an in-memory reference of a root structure.
1557 * For essential trees like root/extent tree, we grab it from fs_info directly.
1558 * For subvolume trees, we check the cached filesystem roots first. If not
1559 * found, then read it from disk and add it to cached fs roots.
1561 * Caller should release the root by calling btrfs_put_root() after the usage.
1563 * NOTE: Reloc and log trees can't be read by this function as they share the
1564 * same root objectid.
1566 * @objectid: root id
1567 * @anon_dev: preallocated anonymous block device number for new roots,
1568 * pass 0 for new allocation.
1569 * @check_ref: whether to check root item references, If true, return -ENOENT
1572 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1573 u64 objectid, dev_t anon_dev,
1576 struct btrfs_root *root;
1577 struct btrfs_path *path;
1578 struct btrfs_key key;
1581 root = btrfs_get_global_root(fs_info, objectid);
1585 root = btrfs_lookup_fs_root(fs_info, objectid);
1587 /* Shouldn't get preallocated anon_dev for cached roots */
1589 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1590 btrfs_put_root(root);
1591 return ERR_PTR(-ENOENT);
1596 key.objectid = objectid;
1597 key.type = BTRFS_ROOT_ITEM_KEY;
1598 key.offset = (u64)-1;
1599 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1603 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1608 ret = btrfs_init_fs_root(root, anon_dev);
1612 path = btrfs_alloc_path();
1617 key.objectid = BTRFS_ORPHAN_OBJECTID;
1618 key.type = BTRFS_ORPHAN_ITEM_KEY;
1619 key.offset = objectid;
1621 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1622 btrfs_free_path(path);
1626 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1628 ret = btrfs_insert_fs_root(fs_info, root);
1630 btrfs_put_root(root);
1637 btrfs_put_root(root);
1638 return ERR_PTR(ret);
1642 * Get in-memory reference of a root structure
1644 * @objectid: tree objectid
1645 * @check_ref: if set, verify that the tree exists and the item has at least
1648 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1649 u64 objectid, bool check_ref)
1651 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1655 * Get in-memory reference of a root structure, created as new, optionally pass
1656 * the anonymous block device id
1658 * @objectid: tree objectid
1659 * @anon_dev: if zero, allocate a new anonymous block device or use the
1662 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1663 u64 objectid, dev_t anon_dev)
1665 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1669 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1670 * @fs_info: the fs_info
1671 * @objectid: the objectid we need to lookup
1673 * This is exclusively used for backref walking, and exists specifically because
1674 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1675 * creation time, which means we may have to read the tree_root in order to look
1676 * up a fs root that is not in memory. If the root is not in memory we will
1677 * read the tree root commit root and look up the fs root from there. This is a
1678 * temporary root, it will not be inserted into the radix tree as it doesn't
1679 * have the most uptodate information, it'll simply be discarded once the
1680 * backref code is finished using the root.
1682 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1683 struct btrfs_path *path,
1686 struct btrfs_root *root;
1687 struct btrfs_key key;
1689 ASSERT(path->search_commit_root && path->skip_locking);
1692 * This can return -ENOENT if we ask for a root that doesn't exist, but
1693 * since this is called via the backref walking code we won't be looking
1694 * up a root that doesn't exist, unless there's corruption. So if root
1695 * != NULL just return it.
1697 root = btrfs_get_global_root(fs_info, objectid);
1701 root = btrfs_lookup_fs_root(fs_info, objectid);
1705 key.objectid = objectid;
1706 key.type = BTRFS_ROOT_ITEM_KEY;
1707 key.offset = (u64)-1;
1708 root = read_tree_root_path(fs_info->tree_root, path, &key);
1709 btrfs_release_path(path);
1715 * called by the kthread helper functions to finally call the bio end_io
1716 * functions. This is where read checksum verification actually happens
1718 static void end_workqueue_fn(struct btrfs_work *work)
1721 struct btrfs_end_io_wq *end_io_wq;
1723 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1724 bio = end_io_wq->bio;
1726 bio->bi_status = end_io_wq->status;
1727 bio->bi_private = end_io_wq->private;
1728 bio->bi_end_io = end_io_wq->end_io;
1730 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1733 static int cleaner_kthread(void *arg)
1735 struct btrfs_root *root = arg;
1736 struct btrfs_fs_info *fs_info = root->fs_info;
1742 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1744 /* Make the cleaner go to sleep early. */
1745 if (btrfs_need_cleaner_sleep(fs_info))
1749 * Do not do anything if we might cause open_ctree() to block
1750 * before we have finished mounting the filesystem.
1752 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1755 if (!mutex_trylock(&fs_info->cleaner_mutex))
1759 * Avoid the problem that we change the status of the fs
1760 * during the above check and trylock.
1762 if (btrfs_need_cleaner_sleep(fs_info)) {
1763 mutex_unlock(&fs_info->cleaner_mutex);
1767 btrfs_run_delayed_iputs(fs_info);
1769 again = btrfs_clean_one_deleted_snapshot(root);
1770 mutex_unlock(&fs_info->cleaner_mutex);
1773 * The defragger has dealt with the R/O remount and umount,
1774 * needn't do anything special here.
1776 btrfs_run_defrag_inodes(fs_info);
1779 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1780 * with relocation (btrfs_relocate_chunk) and relocation
1781 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1782 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1783 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1784 * unused block groups.
1786 btrfs_delete_unused_bgs(fs_info);
1788 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1789 if (kthread_should_park())
1791 if (kthread_should_stop())
1794 set_current_state(TASK_INTERRUPTIBLE);
1796 __set_current_state(TASK_RUNNING);
1801 static int transaction_kthread(void *arg)
1803 struct btrfs_root *root = arg;
1804 struct btrfs_fs_info *fs_info = root->fs_info;
1805 struct btrfs_trans_handle *trans;
1806 struct btrfs_transaction *cur;
1809 unsigned long delay;
1813 cannot_commit = false;
1814 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1815 mutex_lock(&fs_info->transaction_kthread_mutex);
1817 spin_lock(&fs_info->trans_lock);
1818 cur = fs_info->running_transaction;
1820 spin_unlock(&fs_info->trans_lock);
1824 delta = ktime_get_seconds() - cur->start_time;
1825 if (cur->state < TRANS_STATE_COMMIT_START &&
1826 delta < fs_info->commit_interval) {
1827 spin_unlock(&fs_info->trans_lock);
1828 delay -= msecs_to_jiffies((delta - 1) * 1000);
1830 msecs_to_jiffies(fs_info->commit_interval * 1000));
1833 transid = cur->transid;
1834 spin_unlock(&fs_info->trans_lock);
1836 /* If the file system is aborted, this will always fail. */
1837 trans = btrfs_attach_transaction(root);
1838 if (IS_ERR(trans)) {
1839 if (PTR_ERR(trans) != -ENOENT)
1840 cannot_commit = true;
1843 if (transid == trans->transid) {
1844 btrfs_commit_transaction(trans);
1846 btrfs_end_transaction(trans);
1849 wake_up_process(fs_info->cleaner_kthread);
1850 mutex_unlock(&fs_info->transaction_kthread_mutex);
1852 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1853 &fs_info->fs_state)))
1854 btrfs_cleanup_transaction(fs_info);
1855 if (!kthread_should_stop() &&
1856 (!btrfs_transaction_blocked(fs_info) ||
1858 schedule_timeout_interruptible(delay);
1859 } while (!kthread_should_stop());
1864 * This will find the highest generation in the array of root backups. The
1865 * index of the highest array is returned, or -EINVAL if we can't find
1868 * We check to make sure the array is valid by comparing the
1869 * generation of the latest root in the array with the generation
1870 * in the super block. If they don't match we pitch it.
1872 static int find_newest_super_backup(struct btrfs_fs_info *info)
1874 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1876 struct btrfs_root_backup *root_backup;
1879 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1880 root_backup = info->super_copy->super_roots + i;
1881 cur = btrfs_backup_tree_root_gen(root_backup);
1882 if (cur == newest_gen)
1890 * copy all the root pointers into the super backup array.
1891 * this will bump the backup pointer by one when it is
1894 static void backup_super_roots(struct btrfs_fs_info *info)
1896 const int next_backup = info->backup_root_index;
1897 struct btrfs_root_backup *root_backup;
1899 root_backup = info->super_for_commit->super_roots + next_backup;
1902 * make sure all of our padding and empty slots get zero filled
1903 * regardless of which ones we use today
1905 memset(root_backup, 0, sizeof(*root_backup));
1907 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1909 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1910 btrfs_set_backup_tree_root_gen(root_backup,
1911 btrfs_header_generation(info->tree_root->node));
1913 btrfs_set_backup_tree_root_level(root_backup,
1914 btrfs_header_level(info->tree_root->node));
1916 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1917 btrfs_set_backup_chunk_root_gen(root_backup,
1918 btrfs_header_generation(info->chunk_root->node));
1919 btrfs_set_backup_chunk_root_level(root_backup,
1920 btrfs_header_level(info->chunk_root->node));
1922 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1923 btrfs_set_backup_extent_root_gen(root_backup,
1924 btrfs_header_generation(info->extent_root->node));
1925 btrfs_set_backup_extent_root_level(root_backup,
1926 btrfs_header_level(info->extent_root->node));
1929 * we might commit during log recovery, which happens before we set
1930 * the fs_root. Make sure it is valid before we fill it in.
1932 if (info->fs_root && info->fs_root->node) {
1933 btrfs_set_backup_fs_root(root_backup,
1934 info->fs_root->node->start);
1935 btrfs_set_backup_fs_root_gen(root_backup,
1936 btrfs_header_generation(info->fs_root->node));
1937 btrfs_set_backup_fs_root_level(root_backup,
1938 btrfs_header_level(info->fs_root->node));
1941 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1942 btrfs_set_backup_dev_root_gen(root_backup,
1943 btrfs_header_generation(info->dev_root->node));
1944 btrfs_set_backup_dev_root_level(root_backup,
1945 btrfs_header_level(info->dev_root->node));
1947 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1948 btrfs_set_backup_csum_root_gen(root_backup,
1949 btrfs_header_generation(info->csum_root->node));
1950 btrfs_set_backup_csum_root_level(root_backup,
1951 btrfs_header_level(info->csum_root->node));
1953 btrfs_set_backup_total_bytes(root_backup,
1954 btrfs_super_total_bytes(info->super_copy));
1955 btrfs_set_backup_bytes_used(root_backup,
1956 btrfs_super_bytes_used(info->super_copy));
1957 btrfs_set_backup_num_devices(root_backup,
1958 btrfs_super_num_devices(info->super_copy));
1961 * if we don't copy this out to the super_copy, it won't get remembered
1962 * for the next commit
1964 memcpy(&info->super_copy->super_roots,
1965 &info->super_for_commit->super_roots,
1966 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1970 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1971 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1973 * fs_info - filesystem whose backup roots need to be read
1974 * priority - priority of backup root required
1976 * Returns backup root index on success and -EINVAL otherwise.
1978 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1980 int backup_index = find_newest_super_backup(fs_info);
1981 struct btrfs_super_block *super = fs_info->super_copy;
1982 struct btrfs_root_backup *root_backup;
1984 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1986 return backup_index;
1988 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1989 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1994 root_backup = super->super_roots + backup_index;
1996 btrfs_set_super_generation(super,
1997 btrfs_backup_tree_root_gen(root_backup));
1998 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1999 btrfs_set_super_root_level(super,
2000 btrfs_backup_tree_root_level(root_backup));
2001 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2004 * Fixme: the total bytes and num_devices need to match or we should
2007 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2008 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2010 return backup_index;
2013 /* helper to cleanup workers */
2014 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2016 btrfs_destroy_workqueue(fs_info->fixup_workers);
2017 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2018 btrfs_destroy_workqueue(fs_info->workers);
2019 btrfs_destroy_workqueue(fs_info->endio_workers);
2020 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2021 btrfs_destroy_workqueue(fs_info->rmw_workers);
2022 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2023 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2024 btrfs_destroy_workqueue(fs_info->delayed_workers);
2025 btrfs_destroy_workqueue(fs_info->caching_workers);
2026 btrfs_destroy_workqueue(fs_info->readahead_workers);
2027 btrfs_destroy_workqueue(fs_info->flush_workers);
2028 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2029 if (fs_info->discard_ctl.discard_workers)
2030 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2032 * Now that all other work queues are destroyed, we can safely destroy
2033 * the queues used for metadata I/O, since tasks from those other work
2034 * queues can do metadata I/O operations.
2036 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2037 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2040 static void free_root_extent_buffers(struct btrfs_root *root)
2043 free_extent_buffer(root->node);
2044 free_extent_buffer(root->commit_root);
2046 root->commit_root = NULL;
2050 /* helper to cleanup tree roots */
2051 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2053 free_root_extent_buffers(info->tree_root);
2055 free_root_extent_buffers(info->dev_root);
2056 free_root_extent_buffers(info->extent_root);
2057 free_root_extent_buffers(info->csum_root);
2058 free_root_extent_buffers(info->quota_root);
2059 free_root_extent_buffers(info->uuid_root);
2060 free_root_extent_buffers(info->fs_root);
2061 free_root_extent_buffers(info->data_reloc_root);
2062 if (free_chunk_root)
2063 free_root_extent_buffers(info->chunk_root);
2064 free_root_extent_buffers(info->free_space_root);
2067 void btrfs_put_root(struct btrfs_root *root)
2072 if (refcount_dec_and_test(&root->refs)) {
2073 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2074 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2076 free_anon_bdev(root->anon_dev);
2077 btrfs_drew_lock_destroy(&root->snapshot_lock);
2078 free_root_extent_buffers(root);
2079 #ifdef CONFIG_BTRFS_DEBUG
2080 spin_lock(&root->fs_info->fs_roots_radix_lock);
2081 list_del_init(&root->leak_list);
2082 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2088 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2091 struct btrfs_root *gang[8];
2094 while (!list_empty(&fs_info->dead_roots)) {
2095 gang[0] = list_entry(fs_info->dead_roots.next,
2096 struct btrfs_root, root_list);
2097 list_del(&gang[0]->root_list);
2099 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2100 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2101 btrfs_put_root(gang[0]);
2105 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2110 for (i = 0; i < ret; i++)
2111 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2115 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2117 mutex_init(&fs_info->scrub_lock);
2118 atomic_set(&fs_info->scrubs_running, 0);
2119 atomic_set(&fs_info->scrub_pause_req, 0);
2120 atomic_set(&fs_info->scrubs_paused, 0);
2121 atomic_set(&fs_info->scrub_cancel_req, 0);
2122 init_waitqueue_head(&fs_info->scrub_pause_wait);
2123 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2126 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2128 spin_lock_init(&fs_info->balance_lock);
2129 mutex_init(&fs_info->balance_mutex);
2130 atomic_set(&fs_info->balance_pause_req, 0);
2131 atomic_set(&fs_info->balance_cancel_req, 0);
2132 fs_info->balance_ctl = NULL;
2133 init_waitqueue_head(&fs_info->balance_wait_q);
2136 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2138 struct inode *inode = fs_info->btree_inode;
2140 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2141 set_nlink(inode, 1);
2143 * we set the i_size on the btree inode to the max possible int.
2144 * the real end of the address space is determined by all of
2145 * the devices in the system
2147 inode->i_size = OFFSET_MAX;
2148 inode->i_mapping->a_ops = &btree_aops;
2150 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2151 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2152 IO_TREE_BTREE_INODE_IO, inode);
2153 BTRFS_I(inode)->io_tree.track_uptodate = false;
2154 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2156 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2157 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2158 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2159 btrfs_insert_inode_hash(inode);
2162 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2164 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2165 init_rwsem(&fs_info->dev_replace.rwsem);
2166 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2169 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2171 spin_lock_init(&fs_info->qgroup_lock);
2172 mutex_init(&fs_info->qgroup_ioctl_lock);
2173 fs_info->qgroup_tree = RB_ROOT;
2174 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2175 fs_info->qgroup_seq = 1;
2176 fs_info->qgroup_ulist = NULL;
2177 fs_info->qgroup_rescan_running = false;
2178 mutex_init(&fs_info->qgroup_rescan_lock);
2181 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2182 struct btrfs_fs_devices *fs_devices)
2184 u32 max_active = fs_info->thread_pool_size;
2185 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2188 btrfs_alloc_workqueue(fs_info, "worker",
2189 flags | WQ_HIGHPRI, max_active, 16);
2191 fs_info->delalloc_workers =
2192 btrfs_alloc_workqueue(fs_info, "delalloc",
2193 flags, max_active, 2);
2195 fs_info->flush_workers =
2196 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2197 flags, max_active, 0);
2199 fs_info->caching_workers =
2200 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2202 fs_info->fixup_workers =
2203 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2206 * endios are largely parallel and should have a very
2209 fs_info->endio_workers =
2210 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2211 fs_info->endio_meta_workers =
2212 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2214 fs_info->endio_meta_write_workers =
2215 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2217 fs_info->endio_raid56_workers =
2218 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2220 fs_info->rmw_workers =
2221 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2222 fs_info->endio_write_workers =
2223 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2225 fs_info->endio_freespace_worker =
2226 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2228 fs_info->delayed_workers =
2229 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2231 fs_info->readahead_workers =
2232 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2234 fs_info->qgroup_rescan_workers =
2235 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2236 fs_info->discard_ctl.discard_workers =
2237 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2239 if (!(fs_info->workers && fs_info->delalloc_workers &&
2240 fs_info->flush_workers &&
2241 fs_info->endio_workers && fs_info->endio_meta_workers &&
2242 fs_info->endio_meta_write_workers &&
2243 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2244 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2245 fs_info->caching_workers && fs_info->readahead_workers &&
2246 fs_info->fixup_workers && fs_info->delayed_workers &&
2247 fs_info->qgroup_rescan_workers &&
2248 fs_info->discard_ctl.discard_workers)) {
2255 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2257 struct crypto_shash *csum_shash;
2258 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2260 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2262 if (IS_ERR(csum_shash)) {
2263 btrfs_err(fs_info, "error allocating %s hash for checksum",
2265 return PTR_ERR(csum_shash);
2268 fs_info->csum_shash = csum_shash;
2273 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2274 struct btrfs_fs_devices *fs_devices)
2277 struct btrfs_root *log_tree_root;
2278 struct btrfs_super_block *disk_super = fs_info->super_copy;
2279 u64 bytenr = btrfs_super_log_root(disk_super);
2280 int level = btrfs_super_log_root_level(disk_super);
2282 if (fs_devices->rw_devices == 0) {
2283 btrfs_warn(fs_info, "log replay required on RO media");
2287 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2292 log_tree_root->node = read_tree_block(fs_info, bytenr,
2293 BTRFS_TREE_LOG_OBJECTID,
2294 fs_info->generation + 1, level,
2296 if (IS_ERR(log_tree_root->node)) {
2297 btrfs_warn(fs_info, "failed to read log tree");
2298 ret = PTR_ERR(log_tree_root->node);
2299 log_tree_root->node = NULL;
2300 btrfs_put_root(log_tree_root);
2302 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2303 btrfs_err(fs_info, "failed to read log tree");
2304 btrfs_put_root(log_tree_root);
2307 /* returns with log_tree_root freed on success */
2308 ret = btrfs_recover_log_trees(log_tree_root);
2310 btrfs_handle_fs_error(fs_info, ret,
2311 "Failed to recover log tree");
2312 btrfs_put_root(log_tree_root);
2316 if (sb_rdonly(fs_info->sb)) {
2317 ret = btrfs_commit_super(fs_info);
2325 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2327 struct btrfs_root *tree_root = fs_info->tree_root;
2328 struct btrfs_root *root;
2329 struct btrfs_key location;
2332 BUG_ON(!fs_info->tree_root);
2334 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2335 location.type = BTRFS_ROOT_ITEM_KEY;
2336 location.offset = 0;
2338 root = btrfs_read_tree_root(tree_root, &location);
2340 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2341 ret = PTR_ERR(root);
2345 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2346 fs_info->extent_root = root;
2349 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2350 root = btrfs_read_tree_root(tree_root, &location);
2352 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2353 ret = PTR_ERR(root);
2357 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2358 fs_info->dev_root = root;
2359 btrfs_init_devices_late(fs_info);
2362 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2363 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2364 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2365 root = btrfs_read_tree_root(tree_root, &location);
2367 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2368 ret = PTR_ERR(root);
2372 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2373 fs_info->csum_root = root;
2378 * This tree can share blocks with some other fs tree during relocation
2379 * and we need a proper setup by btrfs_get_fs_root
2381 root = btrfs_get_fs_root(tree_root->fs_info,
2382 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2384 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2385 ret = PTR_ERR(root);
2389 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2390 fs_info->data_reloc_root = root;
2393 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2394 root = btrfs_read_tree_root(tree_root, &location);
2395 if (!IS_ERR(root)) {
2396 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2397 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2398 fs_info->quota_root = root;
2401 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2402 root = btrfs_read_tree_root(tree_root, &location);
2404 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2405 ret = PTR_ERR(root);
2410 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2411 fs_info->uuid_root = root;
2414 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2415 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2416 root = btrfs_read_tree_root(tree_root, &location);
2418 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2419 ret = PTR_ERR(root);
2423 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2424 fs_info->free_space_root = root;
2430 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2431 location.objectid, ret);
2436 * Real super block validation
2437 * NOTE: super csum type and incompat features will not be checked here.
2439 * @sb: super block to check
2440 * @mirror_num: the super block number to check its bytenr:
2441 * 0 the primary (1st) sb
2442 * 1, 2 2nd and 3rd backup copy
2443 * -1 skip bytenr check
2445 static int validate_super(struct btrfs_fs_info *fs_info,
2446 struct btrfs_super_block *sb, int mirror_num)
2448 u64 nodesize = btrfs_super_nodesize(sb);
2449 u64 sectorsize = btrfs_super_sectorsize(sb);
2452 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2453 btrfs_err(fs_info, "no valid FS found");
2456 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2457 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2458 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2461 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2462 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2463 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2466 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2467 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2468 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2471 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2472 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2473 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2478 * Check sectorsize and nodesize first, other check will need it.
2479 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2481 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2482 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2483 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2488 * For 4K page size, we only support 4K sector size.
2489 * For 64K page size, we support read-write for 64K sector size, and
2490 * read-only for 4K sector size.
2492 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2493 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2494 sectorsize != SZ_64K))) {
2496 "sectorsize %llu not yet supported for page size %lu",
2497 sectorsize, PAGE_SIZE);
2501 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2502 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2503 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2506 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2507 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2508 le32_to_cpu(sb->__unused_leafsize), nodesize);
2512 /* Root alignment check */
2513 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2514 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2515 btrfs_super_root(sb));
2518 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2519 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2520 btrfs_super_chunk_root(sb));
2523 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2524 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2525 btrfs_super_log_root(sb));
2529 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2530 BTRFS_FSID_SIZE) != 0) {
2532 "dev_item UUID does not match metadata fsid: %pU != %pU",
2533 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2538 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2541 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2542 btrfs_err(fs_info, "bytes_used is too small %llu",
2543 btrfs_super_bytes_used(sb));
2546 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2547 btrfs_err(fs_info, "invalid stripesize %u",
2548 btrfs_super_stripesize(sb));
2551 if (btrfs_super_num_devices(sb) > (1UL << 31))
2552 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2553 btrfs_super_num_devices(sb));
2554 if (btrfs_super_num_devices(sb) == 0) {
2555 btrfs_err(fs_info, "number of devices is 0");
2559 if (mirror_num >= 0 &&
2560 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2561 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2562 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2567 * Obvious sys_chunk_array corruptions, it must hold at least one key
2570 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2571 btrfs_err(fs_info, "system chunk array too big %u > %u",
2572 btrfs_super_sys_array_size(sb),
2573 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2576 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2577 + sizeof(struct btrfs_chunk)) {
2578 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2579 btrfs_super_sys_array_size(sb),
2580 sizeof(struct btrfs_disk_key)
2581 + sizeof(struct btrfs_chunk));
2586 * The generation is a global counter, we'll trust it more than the others
2587 * but it's still possible that it's the one that's wrong.
2589 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2591 "suspicious: generation < chunk_root_generation: %llu < %llu",
2592 btrfs_super_generation(sb),
2593 btrfs_super_chunk_root_generation(sb));
2594 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2595 && btrfs_super_cache_generation(sb) != (u64)-1)
2597 "suspicious: generation < cache_generation: %llu < %llu",
2598 btrfs_super_generation(sb),
2599 btrfs_super_cache_generation(sb));
2605 * Validation of super block at mount time.
2606 * Some checks already done early at mount time, like csum type and incompat
2607 * flags will be skipped.
2609 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2611 return validate_super(fs_info, fs_info->super_copy, 0);
2615 * Validation of super block at write time.
2616 * Some checks like bytenr check will be skipped as their values will be
2618 * Extra checks like csum type and incompat flags will be done here.
2620 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2621 struct btrfs_super_block *sb)
2625 ret = validate_super(fs_info, sb, -1);
2628 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2630 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2631 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2634 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2637 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2638 btrfs_super_incompat_flags(sb),
2639 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2645 "super block corruption detected before writing it to disk");
2649 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2651 int backup_index = find_newest_super_backup(fs_info);
2652 struct btrfs_super_block *sb = fs_info->super_copy;
2653 struct btrfs_root *tree_root = fs_info->tree_root;
2654 bool handle_error = false;
2658 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2663 if (!IS_ERR(tree_root->node))
2664 free_extent_buffer(tree_root->node);
2665 tree_root->node = NULL;
2667 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2670 free_root_pointers(fs_info, 0);
2673 * Don't use the log in recovery mode, it won't be
2676 btrfs_set_super_log_root(sb, 0);
2678 /* We can't trust the free space cache either */
2679 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2681 ret = read_backup_root(fs_info, i);
2686 generation = btrfs_super_generation(sb);
2687 level = btrfs_super_root_level(sb);
2688 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2689 BTRFS_ROOT_TREE_OBJECTID,
2690 generation, level, NULL);
2691 if (IS_ERR(tree_root->node)) {
2692 handle_error = true;
2693 ret = PTR_ERR(tree_root->node);
2694 tree_root->node = NULL;
2695 btrfs_warn(fs_info, "couldn't read tree root");
2698 } else if (!extent_buffer_uptodate(tree_root->node)) {
2699 handle_error = true;
2701 btrfs_warn(fs_info, "error while reading tree root");
2705 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2706 tree_root->commit_root = btrfs_root_node(tree_root);
2707 btrfs_set_root_refs(&tree_root->root_item, 1);
2710 * No need to hold btrfs_root::objectid_mutex since the fs
2711 * hasn't been fully initialised and we are the only user
2713 ret = btrfs_init_root_free_objectid(tree_root);
2715 handle_error = true;
2719 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2721 ret = btrfs_read_roots(fs_info);
2723 handle_error = true;
2727 /* All successful */
2728 fs_info->generation = generation;
2729 fs_info->last_trans_committed = generation;
2731 /* Always begin writing backup roots after the one being used */
2732 if (backup_index < 0) {
2733 fs_info->backup_root_index = 0;
2735 fs_info->backup_root_index = backup_index + 1;
2736 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2744 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2746 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2747 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2748 INIT_LIST_HEAD(&fs_info->trans_list);
2749 INIT_LIST_HEAD(&fs_info->dead_roots);
2750 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2751 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2752 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2753 spin_lock_init(&fs_info->delalloc_root_lock);
2754 spin_lock_init(&fs_info->trans_lock);
2755 spin_lock_init(&fs_info->fs_roots_radix_lock);
2756 spin_lock_init(&fs_info->delayed_iput_lock);
2757 spin_lock_init(&fs_info->defrag_inodes_lock);
2758 spin_lock_init(&fs_info->super_lock);
2759 spin_lock_init(&fs_info->buffer_lock);
2760 spin_lock_init(&fs_info->unused_bgs_lock);
2761 rwlock_init(&fs_info->tree_mod_log_lock);
2762 mutex_init(&fs_info->unused_bg_unpin_mutex);
2763 mutex_init(&fs_info->delete_unused_bgs_mutex);
2764 mutex_init(&fs_info->reloc_mutex);
2765 mutex_init(&fs_info->delalloc_root_mutex);
2766 seqlock_init(&fs_info->profiles_lock);
2768 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2769 INIT_LIST_HEAD(&fs_info->space_info);
2770 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2771 INIT_LIST_HEAD(&fs_info->unused_bgs);
2772 #ifdef CONFIG_BTRFS_DEBUG
2773 INIT_LIST_HEAD(&fs_info->allocated_roots);
2774 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2775 spin_lock_init(&fs_info->eb_leak_lock);
2777 extent_map_tree_init(&fs_info->mapping_tree);
2778 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2779 BTRFS_BLOCK_RSV_GLOBAL);
2780 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2781 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2782 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2783 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2784 BTRFS_BLOCK_RSV_DELOPS);
2785 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2786 BTRFS_BLOCK_RSV_DELREFS);
2788 atomic_set(&fs_info->async_delalloc_pages, 0);
2789 atomic_set(&fs_info->defrag_running, 0);
2790 atomic_set(&fs_info->reada_works_cnt, 0);
2791 atomic_set(&fs_info->nr_delayed_iputs, 0);
2792 atomic64_set(&fs_info->tree_mod_seq, 0);
2793 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2794 fs_info->metadata_ratio = 0;
2795 fs_info->defrag_inodes = RB_ROOT;
2796 atomic64_set(&fs_info->free_chunk_space, 0);
2797 fs_info->tree_mod_log = RB_ROOT;
2798 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2799 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2800 /* readahead state */
2801 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2802 spin_lock_init(&fs_info->reada_lock);
2803 btrfs_init_ref_verify(fs_info);
2805 fs_info->thread_pool_size = min_t(unsigned long,
2806 num_online_cpus() + 2, 8);
2808 INIT_LIST_HEAD(&fs_info->ordered_roots);
2809 spin_lock_init(&fs_info->ordered_root_lock);
2811 btrfs_init_scrub(fs_info);
2812 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2813 fs_info->check_integrity_print_mask = 0;
2815 btrfs_init_balance(fs_info);
2816 btrfs_init_async_reclaim_work(fs_info);
2818 spin_lock_init(&fs_info->block_group_cache_lock);
2819 fs_info->block_group_cache_tree = RB_ROOT;
2820 fs_info->first_logical_byte = (u64)-1;
2822 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2823 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2824 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2826 mutex_init(&fs_info->ordered_operations_mutex);
2827 mutex_init(&fs_info->tree_log_mutex);
2828 mutex_init(&fs_info->chunk_mutex);
2829 mutex_init(&fs_info->transaction_kthread_mutex);
2830 mutex_init(&fs_info->cleaner_mutex);
2831 mutex_init(&fs_info->ro_block_group_mutex);
2832 init_rwsem(&fs_info->commit_root_sem);
2833 init_rwsem(&fs_info->cleanup_work_sem);
2834 init_rwsem(&fs_info->subvol_sem);
2835 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2837 btrfs_init_dev_replace_locks(fs_info);
2838 btrfs_init_qgroup(fs_info);
2839 btrfs_discard_init(fs_info);
2841 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2842 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2844 init_waitqueue_head(&fs_info->transaction_throttle);
2845 init_waitqueue_head(&fs_info->transaction_wait);
2846 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2847 init_waitqueue_head(&fs_info->async_submit_wait);
2848 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2850 /* Usable values until the real ones are cached from the superblock */
2851 fs_info->nodesize = 4096;
2852 fs_info->sectorsize = 4096;
2853 fs_info->sectorsize_bits = ilog2(4096);
2854 fs_info->stripesize = 4096;
2856 spin_lock_init(&fs_info->swapfile_pins_lock);
2857 fs_info->swapfile_pins = RB_ROOT;
2859 fs_info->send_in_progress = 0;
2862 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2867 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2868 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2870 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2874 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2878 fs_info->dirty_metadata_batch = PAGE_SIZE *
2879 (1 + ilog2(nr_cpu_ids));
2881 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2885 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2890 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2892 if (!fs_info->delayed_root)
2894 btrfs_init_delayed_root(fs_info->delayed_root);
2897 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2899 return btrfs_alloc_stripe_hash_table(fs_info);
2902 static int btrfs_uuid_rescan_kthread(void *data)
2904 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2908 * 1st step is to iterate through the existing UUID tree and
2909 * to delete all entries that contain outdated data.
2910 * 2nd step is to add all missing entries to the UUID tree.
2912 ret = btrfs_uuid_tree_iterate(fs_info);
2915 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2917 up(&fs_info->uuid_tree_rescan_sem);
2920 return btrfs_uuid_scan_kthread(data);
2923 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2925 struct task_struct *task;
2927 down(&fs_info->uuid_tree_rescan_sem);
2928 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2930 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2931 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2932 up(&fs_info->uuid_tree_rescan_sem);
2933 return PTR_ERR(task);
2940 * Some options only have meaning at mount time and shouldn't persist across
2941 * remounts, or be displayed. Clear these at the end of mount and remount
2944 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
2946 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
2947 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
2951 * Mounting logic specific to read-write file systems. Shared by open_ctree
2952 * and btrfs_remount when remounting from read-only to read-write.
2954 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2957 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2958 bool clear_free_space_tree = false;
2960 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2961 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2962 clear_free_space_tree = true;
2963 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2964 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2965 btrfs_warn(fs_info, "free space tree is invalid");
2966 clear_free_space_tree = true;
2969 if (clear_free_space_tree) {
2970 btrfs_info(fs_info, "clearing free space tree");
2971 ret = btrfs_clear_free_space_tree(fs_info);
2974 "failed to clear free space tree: %d", ret);
2979 ret = btrfs_cleanup_fs_roots(fs_info);
2983 down_read(&fs_info->cleanup_work_sem);
2984 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2985 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2986 up_read(&fs_info->cleanup_work_sem);
2989 up_read(&fs_info->cleanup_work_sem);
2991 mutex_lock(&fs_info->cleaner_mutex);
2992 ret = btrfs_recover_relocation(fs_info->tree_root);
2993 mutex_unlock(&fs_info->cleaner_mutex);
2995 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
2999 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3000 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3001 btrfs_info(fs_info, "creating free space tree");
3002 ret = btrfs_create_free_space_tree(fs_info);
3005 "failed to create free space tree: %d", ret);
3010 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3011 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3016 ret = btrfs_resume_balance_async(fs_info);
3020 ret = btrfs_resume_dev_replace_async(fs_info);
3022 btrfs_warn(fs_info, "failed to resume dev_replace");
3026 btrfs_qgroup_rescan_resume(fs_info);
3028 if (!fs_info->uuid_root) {
3029 btrfs_info(fs_info, "creating UUID tree");
3030 ret = btrfs_create_uuid_tree(fs_info);
3033 "failed to create the UUID tree %d", ret);
3038 ret = btrfs_find_orphan_roots(fs_info);
3043 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3052 struct btrfs_super_block *disk_super;
3053 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3054 struct btrfs_root *tree_root;
3055 struct btrfs_root *chunk_root;
3060 ret = init_mount_fs_info(fs_info, sb);
3066 /* These need to be init'ed before we start creating inodes and such. */
3067 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3069 fs_info->tree_root = tree_root;
3070 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3072 fs_info->chunk_root = chunk_root;
3073 if (!tree_root || !chunk_root) {
3078 fs_info->btree_inode = new_inode(sb);
3079 if (!fs_info->btree_inode) {
3083 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3084 btrfs_init_btree_inode(fs_info);
3086 invalidate_bdev(fs_devices->latest_bdev);
3089 * Read super block and check the signature bytes only
3091 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
3092 if (IS_ERR(disk_super)) {
3093 err = PTR_ERR(disk_super);
3098 * Verify the type first, if that or the checksum value are
3099 * corrupted, we'll find out
3101 csum_type = btrfs_super_csum_type(disk_super);
3102 if (!btrfs_supported_super_csum(csum_type)) {
3103 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3106 btrfs_release_disk_super(disk_super);
3110 ret = btrfs_init_csum_hash(fs_info, csum_type);
3113 btrfs_release_disk_super(disk_super);
3118 * We want to check superblock checksum, the type is stored inside.
3119 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3121 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3122 btrfs_err(fs_info, "superblock checksum mismatch");
3124 btrfs_release_disk_super(disk_super);
3129 * super_copy is zeroed at allocation time and we never touch the
3130 * following bytes up to INFO_SIZE, the checksum is calculated from
3131 * the whole block of INFO_SIZE
3133 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3134 btrfs_release_disk_super(disk_super);
3136 disk_super = fs_info->super_copy;
3138 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
3141 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
3142 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
3143 fs_info->super_copy->metadata_uuid,
3147 features = btrfs_super_flags(disk_super);
3148 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3149 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3150 btrfs_set_super_flags(disk_super, features);
3152 "found metadata UUID change in progress flag, clearing");
3155 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3156 sizeof(*fs_info->super_for_commit));
3158 ret = btrfs_validate_mount_super(fs_info);
3160 btrfs_err(fs_info, "superblock contains fatal errors");
3165 if (!btrfs_super_root(disk_super))
3168 /* check FS state, whether FS is broken. */
3169 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3170 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3173 * In the long term, we'll store the compression type in the super
3174 * block, and it'll be used for per file compression control.
3176 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3178 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3184 features = btrfs_super_incompat_flags(disk_super) &
3185 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3188 "cannot mount because of unsupported optional features (%llx)",
3194 features = btrfs_super_incompat_flags(disk_super);
3195 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3196 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3197 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3198 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3199 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3201 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3202 btrfs_info(fs_info, "has skinny extents");
3204 fs_info->zoned = (features & BTRFS_FEATURE_INCOMPAT_ZONED);
3207 * flag our filesystem as having big metadata blocks if
3208 * they are bigger than the page size
3210 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3211 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3213 "flagging fs with big metadata feature");
3214 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3217 nodesize = btrfs_super_nodesize(disk_super);
3218 sectorsize = btrfs_super_sectorsize(disk_super);
3219 stripesize = sectorsize;
3220 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3221 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3223 /* Cache block sizes */
3224 fs_info->nodesize = nodesize;
3225 fs_info->sectorsize = sectorsize;
3226 fs_info->sectorsize_bits = ilog2(sectorsize);
3227 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3228 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3229 fs_info->stripesize = stripesize;
3232 * mixed block groups end up with duplicate but slightly offset
3233 * extent buffers for the same range. It leads to corruptions
3235 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3236 (sectorsize != nodesize)) {
3238 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3239 nodesize, sectorsize);
3244 * Needn't use the lock because there is no other task which will
3247 btrfs_set_super_incompat_flags(disk_super, features);
3249 features = btrfs_super_compat_ro_flags(disk_super) &
3250 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3251 if (!sb_rdonly(sb) && features) {
3253 "cannot mount read-write because of unsupported optional features (%llx)",
3259 /* For 4K sector size support, it's only read-only */
3260 if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) {
3261 if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) {
3263 "subpage sectorsize %u only supported read-only for page size %lu",
3264 sectorsize, PAGE_SIZE);
3270 ret = btrfs_init_workqueues(fs_info, fs_devices);
3273 goto fail_sb_buffer;
3276 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3277 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3279 sb->s_blocksize = sectorsize;
3280 sb->s_blocksize_bits = blksize_bits(sectorsize);
3281 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3283 mutex_lock(&fs_info->chunk_mutex);
3284 ret = btrfs_read_sys_array(fs_info);
3285 mutex_unlock(&fs_info->chunk_mutex);
3287 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3288 goto fail_sb_buffer;
3291 generation = btrfs_super_chunk_root_generation(disk_super);
3292 level = btrfs_super_chunk_root_level(disk_super);
3294 chunk_root->node = read_tree_block(fs_info,
3295 btrfs_super_chunk_root(disk_super),
3296 BTRFS_CHUNK_TREE_OBJECTID,
3297 generation, level, NULL);
3298 if (IS_ERR(chunk_root->node) ||
3299 !extent_buffer_uptodate(chunk_root->node)) {
3300 btrfs_err(fs_info, "failed to read chunk root");
3301 if (!IS_ERR(chunk_root->node))
3302 free_extent_buffer(chunk_root->node);
3303 chunk_root->node = NULL;
3304 goto fail_tree_roots;
3306 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3307 chunk_root->commit_root = btrfs_root_node(chunk_root);
3309 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3310 offsetof(struct btrfs_header, chunk_tree_uuid),
3313 ret = btrfs_read_chunk_tree(fs_info);
3315 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3316 goto fail_tree_roots;
3320 * At this point we know all the devices that make this filesystem,
3321 * including the seed devices but we don't know yet if the replace
3322 * target is required. So free devices that are not part of this
3323 * filesystem but skip the replace traget device which is checked
3324 * below in btrfs_init_dev_replace().
3326 btrfs_free_extra_devids(fs_devices);
3327 if (!fs_devices->latest_bdev) {
3328 btrfs_err(fs_info, "failed to read devices");
3329 goto fail_tree_roots;
3332 ret = init_tree_roots(fs_info);
3334 goto fail_tree_roots;
3337 * If we have a uuid root and we're not being told to rescan we need to
3338 * check the generation here so we can set the
3339 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3340 * transaction during a balance or the log replay without updating the
3341 * uuid generation, and then if we crash we would rescan the uuid tree,
3342 * even though it was perfectly fine.
3344 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3345 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3346 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3348 ret = btrfs_verify_dev_extents(fs_info);
3351 "failed to verify dev extents against chunks: %d",
3353 goto fail_block_groups;
3355 ret = btrfs_recover_balance(fs_info);
3357 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3358 goto fail_block_groups;
3361 ret = btrfs_init_dev_stats(fs_info);
3363 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3364 goto fail_block_groups;
3367 ret = btrfs_init_dev_replace(fs_info);
3369 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3370 goto fail_block_groups;
3373 ret = btrfs_check_zoned_mode(fs_info);
3375 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3377 goto fail_block_groups;
3380 ret = btrfs_sysfs_add_fsid(fs_devices);
3382 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3384 goto fail_block_groups;
3387 ret = btrfs_sysfs_add_mounted(fs_info);
3389 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3390 goto fail_fsdev_sysfs;
3393 ret = btrfs_init_space_info(fs_info);
3395 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3399 ret = btrfs_read_block_groups(fs_info);
3401 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3405 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3407 "writable mount is not allowed due to too many missing devices");
3411 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3413 if (IS_ERR(fs_info->cleaner_kthread))
3416 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3418 "btrfs-transaction");
3419 if (IS_ERR(fs_info->transaction_kthread))
3422 if (!btrfs_test_opt(fs_info, NOSSD) &&
3423 !fs_info->fs_devices->rotating) {
3424 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3428 * Mount does not set all options immediately, we can do it now and do
3429 * not have to wait for transaction commit
3431 btrfs_apply_pending_changes(fs_info);
3433 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3434 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3435 ret = btrfsic_mount(fs_info, fs_devices,
3436 btrfs_test_opt(fs_info,
3437 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3439 fs_info->check_integrity_print_mask);
3442 "failed to initialize integrity check module: %d",
3446 ret = btrfs_read_qgroup_config(fs_info);
3448 goto fail_trans_kthread;
3450 if (btrfs_build_ref_tree(fs_info))
3451 btrfs_err(fs_info, "couldn't build ref tree");
3453 /* do not make disk changes in broken FS or nologreplay is given */
3454 if (btrfs_super_log_root(disk_super) != 0 &&
3455 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3456 btrfs_info(fs_info, "start tree-log replay");
3457 ret = btrfs_replay_log(fs_info, fs_devices);
3464 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3465 if (IS_ERR(fs_info->fs_root)) {
3466 err = PTR_ERR(fs_info->fs_root);
3467 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3468 fs_info->fs_root = NULL;
3475 ret = btrfs_start_pre_rw_mount(fs_info);
3477 close_ctree(fs_info);
3480 btrfs_discard_resume(fs_info);
3482 if (fs_info->uuid_root &&
3483 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3484 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3485 btrfs_info(fs_info, "checking UUID tree");
3486 ret = btrfs_check_uuid_tree(fs_info);
3489 "failed to check the UUID tree: %d", ret);
3490 close_ctree(fs_info);
3495 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3498 btrfs_clear_oneshot_options(fs_info);
3502 btrfs_free_qgroup_config(fs_info);
3504 kthread_stop(fs_info->transaction_kthread);
3505 btrfs_cleanup_transaction(fs_info);
3506 btrfs_free_fs_roots(fs_info);
3508 kthread_stop(fs_info->cleaner_kthread);
3511 * make sure we're done with the btree inode before we stop our
3514 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3517 btrfs_sysfs_remove_mounted(fs_info);
3520 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3523 btrfs_put_block_group_cache(fs_info);
3526 if (fs_info->data_reloc_root)
3527 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3528 free_root_pointers(fs_info, true);
3529 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3532 btrfs_stop_all_workers(fs_info);
3533 btrfs_free_block_groups(fs_info);
3535 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3537 iput(fs_info->btree_inode);
3539 btrfs_close_devices(fs_info->fs_devices);
3542 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3544 static void btrfs_end_super_write(struct bio *bio)
3546 struct btrfs_device *device = bio->bi_private;
3547 struct bio_vec *bvec;
3548 struct bvec_iter_all iter_all;
3551 bio_for_each_segment_all(bvec, bio, iter_all) {
3552 page = bvec->bv_page;
3554 if (bio->bi_status) {
3555 btrfs_warn_rl_in_rcu(device->fs_info,
3556 "lost page write due to IO error on %s (%d)",
3557 rcu_str_deref(device->name),
3558 blk_status_to_errno(bio->bi_status));
3559 ClearPageUptodate(page);
3561 btrfs_dev_stat_inc_and_print(device,
3562 BTRFS_DEV_STAT_WRITE_ERRS);
3564 SetPageUptodate(page);
3574 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3577 struct btrfs_super_block *super;
3579 u64 bytenr, bytenr_orig;
3580 struct address_space *mapping = bdev->bd_inode->i_mapping;
3583 bytenr_orig = btrfs_sb_offset(copy_num);
3584 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3586 return ERR_PTR(-EINVAL);
3588 return ERR_PTR(ret);
3590 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3591 return ERR_PTR(-EINVAL);
3593 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3595 return ERR_CAST(page);
3597 super = page_address(page);
3598 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3599 btrfs_release_disk_super(super);
3600 return ERR_PTR(-ENODATA);
3603 if (btrfs_super_bytenr(super) != bytenr_orig) {
3604 btrfs_release_disk_super(super);
3605 return ERR_PTR(-EINVAL);
3612 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3614 struct btrfs_super_block *super, *latest = NULL;
3618 /* we would like to check all the supers, but that would make
3619 * a btrfs mount succeed after a mkfs from a different FS.
3620 * So, we need to add a special mount option to scan for
3621 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3623 for (i = 0; i < 1; i++) {
3624 super = btrfs_read_dev_one_super(bdev, i);
3628 if (!latest || btrfs_super_generation(super) > transid) {
3630 btrfs_release_disk_super(super);
3633 transid = btrfs_super_generation(super);
3641 * Write superblock @sb to the @device. Do not wait for completion, all the
3642 * pages we use for writing are locked.
3644 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3645 * the expected device size at commit time. Note that max_mirrors must be
3646 * same for write and wait phases.
3648 * Return number of errors when page is not found or submission fails.
3650 static int write_dev_supers(struct btrfs_device *device,
3651 struct btrfs_super_block *sb, int max_mirrors)
3653 struct btrfs_fs_info *fs_info = device->fs_info;
3654 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3655 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3659 u64 bytenr, bytenr_orig;
3661 if (max_mirrors == 0)
3662 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3664 shash->tfm = fs_info->csum_shash;
3666 for (i = 0; i < max_mirrors; i++) {
3669 struct btrfs_super_block *disk_super;
3671 bytenr_orig = btrfs_sb_offset(i);
3672 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3673 if (ret == -ENOENT) {
3675 } else if (ret < 0) {
3676 btrfs_err(device->fs_info,
3677 "couldn't get super block location for mirror %d",
3682 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3683 device->commit_total_bytes)
3686 btrfs_set_super_bytenr(sb, bytenr_orig);
3688 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3689 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3692 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3695 btrfs_err(device->fs_info,
3696 "couldn't get super block page for bytenr %llu",
3702 /* Bump the refcount for wait_dev_supers() */
3705 disk_super = page_address(page);
3706 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3709 * Directly use bios here instead of relying on the page cache
3710 * to do I/O, so we don't lose the ability to do integrity
3713 bio = bio_alloc(GFP_NOFS, 1);
3714 bio_set_dev(bio, device->bdev);
3715 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3716 bio->bi_private = device;
3717 bio->bi_end_io = btrfs_end_super_write;
3718 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3719 offset_in_page(bytenr));
3722 * We FUA only the first super block. The others we allow to
3723 * go down lazy and there's a short window where the on-disk
3724 * copies might still contain the older version.
3726 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3727 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3728 bio->bi_opf |= REQ_FUA;
3730 btrfsic_submit_bio(bio);
3731 btrfs_advance_sb_log(device, i);
3733 return errors < i ? 0 : -1;
3737 * Wait for write completion of superblocks done by write_dev_supers,
3738 * @max_mirrors same for write and wait phases.
3740 * Return number of errors when page is not found or not marked up to
3743 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3747 bool primary_failed = false;
3751 if (max_mirrors == 0)
3752 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3754 for (i = 0; i < max_mirrors; i++) {
3757 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3758 if (ret == -ENOENT) {
3760 } else if (ret < 0) {
3763 primary_failed = true;
3766 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3767 device->commit_total_bytes)
3770 page = find_get_page(device->bdev->bd_inode->i_mapping,
3771 bytenr >> PAGE_SHIFT);
3775 primary_failed = true;
3778 /* Page is submitted locked and unlocked once the IO completes */
3779 wait_on_page_locked(page);
3780 if (PageError(page)) {
3783 primary_failed = true;
3786 /* Drop our reference */
3789 /* Drop the reference from the writing run */
3793 /* log error, force error return */
3794 if (primary_failed) {
3795 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3800 return errors < i ? 0 : -1;
3804 * endio for the write_dev_flush, this will wake anyone waiting
3805 * for the barrier when it is done
3807 static void btrfs_end_empty_barrier(struct bio *bio)
3809 complete(bio->bi_private);
3813 * Submit a flush request to the device if it supports it. Error handling is
3814 * done in the waiting counterpart.
3816 static void write_dev_flush(struct btrfs_device *device)
3818 struct request_queue *q = bdev_get_queue(device->bdev);
3819 struct bio *bio = device->flush_bio;
3821 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3825 bio->bi_end_io = btrfs_end_empty_barrier;
3826 bio_set_dev(bio, device->bdev);
3827 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3828 init_completion(&device->flush_wait);
3829 bio->bi_private = &device->flush_wait;
3831 btrfsic_submit_bio(bio);
3832 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3836 * If the flush bio has been submitted by write_dev_flush, wait for it.
3838 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3840 struct bio *bio = device->flush_bio;
3842 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3845 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3846 wait_for_completion_io(&device->flush_wait);
3848 return bio->bi_status;
3851 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3853 if (!btrfs_check_rw_degradable(fs_info, NULL))
3859 * send an empty flush down to each device in parallel,
3860 * then wait for them
3862 static int barrier_all_devices(struct btrfs_fs_info *info)
3864 struct list_head *head;
3865 struct btrfs_device *dev;
3866 int errors_wait = 0;
3869 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3870 /* send down all the barriers */
3871 head = &info->fs_devices->devices;
3872 list_for_each_entry(dev, head, dev_list) {
3873 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3877 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3878 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3881 write_dev_flush(dev);
3882 dev->last_flush_error = BLK_STS_OK;
3885 /* wait for all the barriers */
3886 list_for_each_entry(dev, head, dev_list) {
3887 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3893 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3894 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3897 ret = wait_dev_flush(dev);
3899 dev->last_flush_error = ret;
3900 btrfs_dev_stat_inc_and_print(dev,
3901 BTRFS_DEV_STAT_FLUSH_ERRS);
3908 * At some point we need the status of all disks
3909 * to arrive at the volume status. So error checking
3910 * is being pushed to a separate loop.
3912 return check_barrier_error(info);
3917 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3920 int min_tolerated = INT_MAX;
3922 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3923 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3924 min_tolerated = min_t(int, min_tolerated,
3925 btrfs_raid_array[BTRFS_RAID_SINGLE].
3926 tolerated_failures);
3928 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3929 if (raid_type == BTRFS_RAID_SINGLE)
3931 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3933 min_tolerated = min_t(int, min_tolerated,
3934 btrfs_raid_array[raid_type].
3935 tolerated_failures);
3938 if (min_tolerated == INT_MAX) {
3939 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3943 return min_tolerated;
3946 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3948 struct list_head *head;
3949 struct btrfs_device *dev;
3950 struct btrfs_super_block *sb;
3951 struct btrfs_dev_item *dev_item;
3955 int total_errors = 0;
3958 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3961 * max_mirrors == 0 indicates we're from commit_transaction,
3962 * not from fsync where the tree roots in fs_info have not
3963 * been consistent on disk.
3965 if (max_mirrors == 0)
3966 backup_super_roots(fs_info);
3968 sb = fs_info->super_for_commit;
3969 dev_item = &sb->dev_item;
3971 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3972 head = &fs_info->fs_devices->devices;
3973 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3976 ret = barrier_all_devices(fs_info);
3979 &fs_info->fs_devices->device_list_mutex);
3980 btrfs_handle_fs_error(fs_info, ret,
3981 "errors while submitting device barriers.");
3986 list_for_each_entry(dev, head, dev_list) {
3991 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3992 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3995 btrfs_set_stack_device_generation(dev_item, 0);
3996 btrfs_set_stack_device_type(dev_item, dev->type);
3997 btrfs_set_stack_device_id(dev_item, dev->devid);
3998 btrfs_set_stack_device_total_bytes(dev_item,
3999 dev->commit_total_bytes);
4000 btrfs_set_stack_device_bytes_used(dev_item,
4001 dev->commit_bytes_used);
4002 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4003 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4004 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4005 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4006 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4009 flags = btrfs_super_flags(sb);
4010 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4012 ret = btrfs_validate_write_super(fs_info, sb);
4014 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4015 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4016 "unexpected superblock corruption detected");
4020 ret = write_dev_supers(dev, sb, max_mirrors);
4024 if (total_errors > max_errors) {
4025 btrfs_err(fs_info, "%d errors while writing supers",
4027 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4029 /* FUA is masked off if unsupported and can't be the reason */
4030 btrfs_handle_fs_error(fs_info, -EIO,
4031 "%d errors while writing supers",
4037 list_for_each_entry(dev, head, dev_list) {
4040 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4041 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4044 ret = wait_dev_supers(dev, max_mirrors);
4048 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4049 if (total_errors > max_errors) {
4050 btrfs_handle_fs_error(fs_info, -EIO,
4051 "%d errors while writing supers",
4058 /* Drop a fs root from the radix tree and free it. */
4059 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4060 struct btrfs_root *root)
4062 bool drop_ref = false;
4064 spin_lock(&fs_info->fs_roots_radix_lock);
4065 radix_tree_delete(&fs_info->fs_roots_radix,
4066 (unsigned long)root->root_key.objectid);
4067 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4069 spin_unlock(&fs_info->fs_roots_radix_lock);
4071 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4072 ASSERT(root->log_root == NULL);
4073 if (root->reloc_root) {
4074 btrfs_put_root(root->reloc_root);
4075 root->reloc_root = NULL;
4080 btrfs_put_root(root);
4083 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4085 u64 root_objectid = 0;
4086 struct btrfs_root *gang[8];
4089 unsigned int ret = 0;
4092 spin_lock(&fs_info->fs_roots_radix_lock);
4093 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4094 (void **)gang, root_objectid,
4097 spin_unlock(&fs_info->fs_roots_radix_lock);
4100 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4102 for (i = 0; i < ret; i++) {
4103 /* Avoid to grab roots in dead_roots */
4104 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4108 /* grab all the search result for later use */
4109 gang[i] = btrfs_grab_root(gang[i]);
4111 spin_unlock(&fs_info->fs_roots_radix_lock);
4113 for (i = 0; i < ret; i++) {
4116 root_objectid = gang[i]->root_key.objectid;
4117 err = btrfs_orphan_cleanup(gang[i]);
4120 btrfs_put_root(gang[i]);
4125 /* release the uncleaned roots due to error */
4126 for (; i < ret; i++) {
4128 btrfs_put_root(gang[i]);
4133 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4135 struct btrfs_root *root = fs_info->tree_root;
4136 struct btrfs_trans_handle *trans;
4138 mutex_lock(&fs_info->cleaner_mutex);
4139 btrfs_run_delayed_iputs(fs_info);
4140 mutex_unlock(&fs_info->cleaner_mutex);
4141 wake_up_process(fs_info->cleaner_kthread);
4143 /* wait until ongoing cleanup work done */
4144 down_write(&fs_info->cleanup_work_sem);
4145 up_write(&fs_info->cleanup_work_sem);
4147 trans = btrfs_join_transaction(root);
4149 return PTR_ERR(trans);
4150 return btrfs_commit_transaction(trans);
4153 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4157 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4159 * We don't want the cleaner to start new transactions, add more delayed
4160 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4161 * because that frees the task_struct, and the transaction kthread might
4162 * still try to wake up the cleaner.
4164 kthread_park(fs_info->cleaner_kthread);
4166 /* wait for the qgroup rescan worker to stop */
4167 btrfs_qgroup_wait_for_completion(fs_info, false);
4169 /* wait for the uuid_scan task to finish */
4170 down(&fs_info->uuid_tree_rescan_sem);
4171 /* avoid complains from lockdep et al., set sem back to initial state */
4172 up(&fs_info->uuid_tree_rescan_sem);
4174 /* pause restriper - we want to resume on mount */
4175 btrfs_pause_balance(fs_info);
4177 btrfs_dev_replace_suspend_for_unmount(fs_info);
4179 btrfs_scrub_cancel(fs_info);
4181 /* wait for any defraggers to finish */
4182 wait_event(fs_info->transaction_wait,
4183 (atomic_read(&fs_info->defrag_running) == 0));
4185 /* clear out the rbtree of defraggable inodes */
4186 btrfs_cleanup_defrag_inodes(fs_info);
4188 cancel_work_sync(&fs_info->async_reclaim_work);
4189 cancel_work_sync(&fs_info->async_data_reclaim_work);
4190 cancel_work_sync(&fs_info->preempt_reclaim_work);
4192 /* Cancel or finish ongoing discard work */
4193 btrfs_discard_cleanup(fs_info);
4195 if (!sb_rdonly(fs_info->sb)) {
4197 * The cleaner kthread is stopped, so do one final pass over
4198 * unused block groups.
4200 btrfs_delete_unused_bgs(fs_info);
4203 * There might be existing delayed inode workers still running
4204 * and holding an empty delayed inode item. We must wait for
4205 * them to complete first because they can create a transaction.
4206 * This happens when someone calls btrfs_balance_delayed_items()
4207 * and then a transaction commit runs the same delayed nodes
4208 * before any delayed worker has done something with the nodes.
4209 * We must wait for any worker here and not at transaction
4210 * commit time since that could cause a deadlock.
4211 * This is a very rare case.
4213 btrfs_flush_workqueue(fs_info->delayed_workers);
4215 ret = btrfs_commit_super(fs_info);
4217 btrfs_err(fs_info, "commit super ret %d", ret);
4220 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4221 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4222 btrfs_error_commit_super(fs_info);
4224 kthread_stop(fs_info->transaction_kthread);
4225 kthread_stop(fs_info->cleaner_kthread);
4227 ASSERT(list_empty(&fs_info->delayed_iputs));
4228 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4230 if (btrfs_check_quota_leak(fs_info)) {
4231 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4232 btrfs_err(fs_info, "qgroup reserved space leaked");
4235 btrfs_free_qgroup_config(fs_info);
4236 ASSERT(list_empty(&fs_info->delalloc_roots));
4238 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4239 btrfs_info(fs_info, "at unmount delalloc count %lld",
4240 percpu_counter_sum(&fs_info->delalloc_bytes));
4243 if (percpu_counter_sum(&fs_info->ordered_bytes))
4244 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4245 percpu_counter_sum(&fs_info->ordered_bytes));
4247 btrfs_sysfs_remove_mounted(fs_info);
4248 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4250 btrfs_put_block_group_cache(fs_info);
4253 * we must make sure there is not any read request to
4254 * submit after we stopping all workers.
4256 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4257 btrfs_stop_all_workers(fs_info);
4259 /* We shouldn't have any transaction open at this point */
4260 ASSERT(list_empty(&fs_info->trans_list));
4262 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4263 free_root_pointers(fs_info, true);
4264 btrfs_free_fs_roots(fs_info);
4267 * We must free the block groups after dropping the fs_roots as we could
4268 * have had an IO error and have left over tree log blocks that aren't
4269 * cleaned up until the fs roots are freed. This makes the block group
4270 * accounting appear to be wrong because there's pending reserved bytes,
4271 * so make sure we do the block group cleanup afterwards.
4273 btrfs_free_block_groups(fs_info);
4275 iput(fs_info->btree_inode);
4277 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4278 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4279 btrfsic_unmount(fs_info->fs_devices);
4282 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4283 btrfs_close_devices(fs_info->fs_devices);
4286 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4290 struct inode *btree_inode = buf->pages[0]->mapping->host;
4292 ret = extent_buffer_uptodate(buf);
4296 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4297 parent_transid, atomic);
4303 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4305 struct btrfs_fs_info *fs_info = buf->fs_info;
4306 u64 transid = btrfs_header_generation(buf);
4309 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4311 * This is a fast path so only do this check if we have sanity tests
4312 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4313 * outside of the sanity tests.
4315 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4318 btrfs_assert_tree_locked(buf);
4319 if (transid != fs_info->generation)
4320 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4321 buf->start, transid, fs_info->generation);
4322 was_dirty = set_extent_buffer_dirty(buf);
4324 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4326 fs_info->dirty_metadata_batch);
4327 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4329 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4330 * but item data not updated.
4331 * So here we should only check item pointers, not item data.
4333 if (btrfs_header_level(buf) == 0 &&
4334 btrfs_check_leaf_relaxed(buf)) {
4335 btrfs_print_leaf(buf);
4341 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4345 * looks as though older kernels can get into trouble with
4346 * this code, they end up stuck in balance_dirty_pages forever
4350 if (current->flags & PF_MEMALLOC)
4354 btrfs_balance_delayed_items(fs_info);
4356 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4357 BTRFS_DIRTY_METADATA_THRESH,
4358 fs_info->dirty_metadata_batch);
4360 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4364 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4366 __btrfs_btree_balance_dirty(fs_info, 1);
4369 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4371 __btrfs_btree_balance_dirty(fs_info, 0);
4374 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4375 struct btrfs_key *first_key)
4377 return btree_read_extent_buffer_pages(buf, parent_transid,
4381 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4383 /* cleanup FS via transaction */
4384 btrfs_cleanup_transaction(fs_info);
4386 mutex_lock(&fs_info->cleaner_mutex);
4387 btrfs_run_delayed_iputs(fs_info);
4388 mutex_unlock(&fs_info->cleaner_mutex);
4390 down_write(&fs_info->cleanup_work_sem);
4391 up_write(&fs_info->cleanup_work_sem);
4394 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4396 struct btrfs_root *gang[8];
4397 u64 root_objectid = 0;
4400 spin_lock(&fs_info->fs_roots_radix_lock);
4401 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4402 (void **)gang, root_objectid,
4403 ARRAY_SIZE(gang))) != 0) {
4406 for (i = 0; i < ret; i++)
4407 gang[i] = btrfs_grab_root(gang[i]);
4408 spin_unlock(&fs_info->fs_roots_radix_lock);
4410 for (i = 0; i < ret; i++) {
4413 root_objectid = gang[i]->root_key.objectid;
4414 btrfs_free_log(NULL, gang[i]);
4415 btrfs_put_root(gang[i]);
4418 spin_lock(&fs_info->fs_roots_radix_lock);
4420 spin_unlock(&fs_info->fs_roots_radix_lock);
4421 btrfs_free_log_root_tree(NULL, fs_info);
4424 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4426 struct btrfs_ordered_extent *ordered;
4428 spin_lock(&root->ordered_extent_lock);
4430 * This will just short circuit the ordered completion stuff which will
4431 * make sure the ordered extent gets properly cleaned up.
4433 list_for_each_entry(ordered, &root->ordered_extents,
4435 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4436 spin_unlock(&root->ordered_extent_lock);
4439 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4441 struct btrfs_root *root;
4442 struct list_head splice;
4444 INIT_LIST_HEAD(&splice);
4446 spin_lock(&fs_info->ordered_root_lock);
4447 list_splice_init(&fs_info->ordered_roots, &splice);
4448 while (!list_empty(&splice)) {
4449 root = list_first_entry(&splice, struct btrfs_root,
4451 list_move_tail(&root->ordered_root,
4452 &fs_info->ordered_roots);
4454 spin_unlock(&fs_info->ordered_root_lock);
4455 btrfs_destroy_ordered_extents(root);
4458 spin_lock(&fs_info->ordered_root_lock);
4460 spin_unlock(&fs_info->ordered_root_lock);
4463 * We need this here because if we've been flipped read-only we won't
4464 * get sync() from the umount, so we need to make sure any ordered
4465 * extents that haven't had their dirty pages IO start writeout yet
4466 * actually get run and error out properly.
4468 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4471 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4472 struct btrfs_fs_info *fs_info)
4474 struct rb_node *node;
4475 struct btrfs_delayed_ref_root *delayed_refs;
4476 struct btrfs_delayed_ref_node *ref;
4479 delayed_refs = &trans->delayed_refs;
4481 spin_lock(&delayed_refs->lock);
4482 if (atomic_read(&delayed_refs->num_entries) == 0) {
4483 spin_unlock(&delayed_refs->lock);
4484 btrfs_debug(fs_info, "delayed_refs has NO entry");
4488 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4489 struct btrfs_delayed_ref_head *head;
4491 bool pin_bytes = false;
4493 head = rb_entry(node, struct btrfs_delayed_ref_head,
4495 if (btrfs_delayed_ref_lock(delayed_refs, head))
4498 spin_lock(&head->lock);
4499 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4500 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4503 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4504 RB_CLEAR_NODE(&ref->ref_node);
4505 if (!list_empty(&ref->add_list))
4506 list_del(&ref->add_list);
4507 atomic_dec(&delayed_refs->num_entries);
4508 btrfs_put_delayed_ref(ref);
4510 if (head->must_insert_reserved)
4512 btrfs_free_delayed_extent_op(head->extent_op);
4513 btrfs_delete_ref_head(delayed_refs, head);
4514 spin_unlock(&head->lock);
4515 spin_unlock(&delayed_refs->lock);
4516 mutex_unlock(&head->mutex);
4519 struct btrfs_block_group *cache;
4521 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4524 spin_lock(&cache->space_info->lock);
4525 spin_lock(&cache->lock);
4526 cache->pinned += head->num_bytes;
4527 btrfs_space_info_update_bytes_pinned(fs_info,
4528 cache->space_info, head->num_bytes);
4529 cache->reserved -= head->num_bytes;
4530 cache->space_info->bytes_reserved -= head->num_bytes;
4531 spin_unlock(&cache->lock);
4532 spin_unlock(&cache->space_info->lock);
4533 percpu_counter_add_batch(
4534 &cache->space_info->total_bytes_pinned,
4535 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4537 btrfs_put_block_group(cache);
4539 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4540 head->bytenr + head->num_bytes - 1);
4542 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4543 btrfs_put_delayed_ref_head(head);
4545 spin_lock(&delayed_refs->lock);
4547 btrfs_qgroup_destroy_extent_records(trans);
4549 spin_unlock(&delayed_refs->lock);
4554 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4556 struct btrfs_inode *btrfs_inode;
4557 struct list_head splice;
4559 INIT_LIST_HEAD(&splice);
4561 spin_lock(&root->delalloc_lock);
4562 list_splice_init(&root->delalloc_inodes, &splice);
4564 while (!list_empty(&splice)) {
4565 struct inode *inode = NULL;
4566 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4568 __btrfs_del_delalloc_inode(root, btrfs_inode);
4569 spin_unlock(&root->delalloc_lock);
4572 * Make sure we get a live inode and that it'll not disappear
4575 inode = igrab(&btrfs_inode->vfs_inode);
4577 invalidate_inode_pages2(inode->i_mapping);
4580 spin_lock(&root->delalloc_lock);
4582 spin_unlock(&root->delalloc_lock);
4585 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4587 struct btrfs_root *root;
4588 struct list_head splice;
4590 INIT_LIST_HEAD(&splice);
4592 spin_lock(&fs_info->delalloc_root_lock);
4593 list_splice_init(&fs_info->delalloc_roots, &splice);
4594 while (!list_empty(&splice)) {
4595 root = list_first_entry(&splice, struct btrfs_root,
4597 root = btrfs_grab_root(root);
4599 spin_unlock(&fs_info->delalloc_root_lock);
4601 btrfs_destroy_delalloc_inodes(root);
4602 btrfs_put_root(root);
4604 spin_lock(&fs_info->delalloc_root_lock);
4606 spin_unlock(&fs_info->delalloc_root_lock);
4609 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4610 struct extent_io_tree *dirty_pages,
4614 struct extent_buffer *eb;
4619 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4624 clear_extent_bits(dirty_pages, start, end, mark);
4625 while (start <= end) {
4626 eb = find_extent_buffer(fs_info, start);
4627 start += fs_info->nodesize;
4630 wait_on_extent_buffer_writeback(eb);
4632 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4634 clear_extent_buffer_dirty(eb);
4635 free_extent_buffer_stale(eb);
4642 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4643 struct extent_io_tree *unpin)
4650 struct extent_state *cached_state = NULL;
4653 * The btrfs_finish_extent_commit() may get the same range as
4654 * ours between find_first_extent_bit and clear_extent_dirty.
4655 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4656 * the same extent range.
4658 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4659 ret = find_first_extent_bit(unpin, 0, &start, &end,
4660 EXTENT_DIRTY, &cached_state);
4662 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4666 clear_extent_dirty(unpin, start, end, &cached_state);
4667 free_extent_state(cached_state);
4668 btrfs_error_unpin_extent_range(fs_info, start, end);
4669 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4676 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4678 struct inode *inode;
4680 inode = cache->io_ctl.inode;
4682 invalidate_inode_pages2(inode->i_mapping);
4683 BTRFS_I(inode)->generation = 0;
4684 cache->io_ctl.inode = NULL;
4687 ASSERT(cache->io_ctl.pages == NULL);
4688 btrfs_put_block_group(cache);
4691 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4692 struct btrfs_fs_info *fs_info)
4694 struct btrfs_block_group *cache;
4696 spin_lock(&cur_trans->dirty_bgs_lock);
4697 while (!list_empty(&cur_trans->dirty_bgs)) {
4698 cache = list_first_entry(&cur_trans->dirty_bgs,
4699 struct btrfs_block_group,
4702 if (!list_empty(&cache->io_list)) {
4703 spin_unlock(&cur_trans->dirty_bgs_lock);
4704 list_del_init(&cache->io_list);
4705 btrfs_cleanup_bg_io(cache);
4706 spin_lock(&cur_trans->dirty_bgs_lock);
4709 list_del_init(&cache->dirty_list);
4710 spin_lock(&cache->lock);
4711 cache->disk_cache_state = BTRFS_DC_ERROR;
4712 spin_unlock(&cache->lock);
4714 spin_unlock(&cur_trans->dirty_bgs_lock);
4715 btrfs_put_block_group(cache);
4716 btrfs_delayed_refs_rsv_release(fs_info, 1);
4717 spin_lock(&cur_trans->dirty_bgs_lock);
4719 spin_unlock(&cur_trans->dirty_bgs_lock);
4722 * Refer to the definition of io_bgs member for details why it's safe
4723 * to use it without any locking
4725 while (!list_empty(&cur_trans->io_bgs)) {
4726 cache = list_first_entry(&cur_trans->io_bgs,
4727 struct btrfs_block_group,
4730 list_del_init(&cache->io_list);
4731 spin_lock(&cache->lock);
4732 cache->disk_cache_state = BTRFS_DC_ERROR;
4733 spin_unlock(&cache->lock);
4734 btrfs_cleanup_bg_io(cache);
4738 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4739 struct btrfs_fs_info *fs_info)
4741 struct btrfs_device *dev, *tmp;
4743 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4744 ASSERT(list_empty(&cur_trans->dirty_bgs));
4745 ASSERT(list_empty(&cur_trans->io_bgs));
4747 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4749 list_del_init(&dev->post_commit_list);
4752 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4754 cur_trans->state = TRANS_STATE_COMMIT_START;
4755 wake_up(&fs_info->transaction_blocked_wait);
4757 cur_trans->state = TRANS_STATE_UNBLOCKED;
4758 wake_up(&fs_info->transaction_wait);
4760 btrfs_destroy_delayed_inodes(fs_info);
4762 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4764 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4766 cur_trans->state =TRANS_STATE_COMPLETED;
4767 wake_up(&cur_trans->commit_wait);
4770 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4772 struct btrfs_transaction *t;
4774 mutex_lock(&fs_info->transaction_kthread_mutex);
4776 spin_lock(&fs_info->trans_lock);
4777 while (!list_empty(&fs_info->trans_list)) {
4778 t = list_first_entry(&fs_info->trans_list,
4779 struct btrfs_transaction, list);
4780 if (t->state >= TRANS_STATE_COMMIT_START) {
4781 refcount_inc(&t->use_count);
4782 spin_unlock(&fs_info->trans_lock);
4783 btrfs_wait_for_commit(fs_info, t->transid);
4784 btrfs_put_transaction(t);
4785 spin_lock(&fs_info->trans_lock);
4788 if (t == fs_info->running_transaction) {
4789 t->state = TRANS_STATE_COMMIT_DOING;
4790 spin_unlock(&fs_info->trans_lock);
4792 * We wait for 0 num_writers since we don't hold a trans
4793 * handle open currently for this transaction.
4795 wait_event(t->writer_wait,
4796 atomic_read(&t->num_writers) == 0);
4798 spin_unlock(&fs_info->trans_lock);
4800 btrfs_cleanup_one_transaction(t, fs_info);
4802 spin_lock(&fs_info->trans_lock);
4803 if (t == fs_info->running_transaction)
4804 fs_info->running_transaction = NULL;
4805 list_del_init(&t->list);
4806 spin_unlock(&fs_info->trans_lock);
4808 btrfs_put_transaction(t);
4809 trace_btrfs_transaction_commit(fs_info->tree_root);
4810 spin_lock(&fs_info->trans_lock);
4812 spin_unlock(&fs_info->trans_lock);
4813 btrfs_destroy_all_ordered_extents(fs_info);
4814 btrfs_destroy_delayed_inodes(fs_info);
4815 btrfs_assert_delayed_root_empty(fs_info);
4816 btrfs_destroy_all_delalloc_inodes(fs_info);
4817 btrfs_drop_all_logs(fs_info);
4818 mutex_unlock(&fs_info->transaction_kthread_mutex);
4823 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4825 struct btrfs_path *path;
4827 struct extent_buffer *l;
4828 struct btrfs_key search_key;
4829 struct btrfs_key found_key;
4832 path = btrfs_alloc_path();
4836 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4837 search_key.type = -1;
4838 search_key.offset = (u64)-1;
4839 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4842 BUG_ON(ret == 0); /* Corruption */
4843 if (path->slots[0] > 0) {
4844 slot = path->slots[0] - 1;
4846 btrfs_item_key_to_cpu(l, &found_key, slot);
4847 root->free_objectid = max_t(u64, found_key.objectid + 1,
4848 BTRFS_FIRST_FREE_OBJECTID);
4850 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4854 btrfs_free_path(path);
4858 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4861 mutex_lock(&root->objectid_mutex);
4863 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4864 btrfs_warn(root->fs_info,
4865 "the objectid of root %llu reaches its highest value",
4866 root->root_key.objectid);
4871 *objectid = root->free_objectid++;
4874 mutex_unlock(&root->objectid_mutex);