2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include "tree-checker.h"
55 #include <asm/cpufeature.h>
58 static const struct extent_io_ops btree_extent_io_ops;
59 static void end_workqueue_fn(struct btrfs_work *work);
60 static void free_fs_root(struct btrfs_root *root);
61 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
63 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_root *root);
66 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
67 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
68 struct extent_io_tree *dirty_pages,
70 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
71 struct extent_io_tree *pinned_extents);
72 static int btrfs_cleanup_transaction(struct btrfs_root *root);
73 static void btrfs_error_commit_super(struct btrfs_root *root);
76 * btrfs_end_io_wq structs are used to do processing in task context when an IO
77 * is complete. This is used during reads to verify checksums, and it is used
78 * by writes to insert metadata for new file extents after IO is complete.
80 struct btrfs_end_io_wq {
84 struct btrfs_fs_info *info;
86 enum btrfs_wq_endio_type metadata;
87 struct list_head list;
88 struct btrfs_work work;
91 static struct kmem_cache *btrfs_end_io_wq_cache;
93 int __init btrfs_end_io_wq_init(void)
95 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
96 sizeof(struct btrfs_end_io_wq),
98 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
100 if (!btrfs_end_io_wq_cache)
105 void btrfs_end_io_wq_exit(void)
107 if (btrfs_end_io_wq_cache)
108 kmem_cache_destroy(btrfs_end_io_wq_cache);
112 * async submit bios are used to offload expensive checksumming
113 * onto the worker threads. They checksum file and metadata bios
114 * just before they are sent down the IO stack.
116 struct async_submit_bio {
119 struct list_head list;
120 extent_submit_bio_hook_t *submit_bio_start;
121 extent_submit_bio_hook_t *submit_bio_done;
124 unsigned long bio_flags;
126 * bio_offset is optional, can be used if the pages in the bio
127 * can't tell us where in the file the bio should go
130 struct btrfs_work work;
135 * Lockdep class keys for extent_buffer->lock's in this root. For a given
136 * eb, the lockdep key is determined by the btrfs_root it belongs to and
137 * the level the eb occupies in the tree.
139 * Different roots are used for different purposes and may nest inside each
140 * other and they require separate keysets. As lockdep keys should be
141 * static, assign keysets according to the purpose of the root as indicated
142 * by btrfs_root->objectid. This ensures that all special purpose roots
143 * have separate keysets.
145 * Lock-nesting across peer nodes is always done with the immediate parent
146 * node locked thus preventing deadlock. As lockdep doesn't know this, use
147 * subclass to avoid triggering lockdep warning in such cases.
149 * The key is set by the readpage_end_io_hook after the buffer has passed
150 * csum validation but before the pages are unlocked. It is also set by
151 * btrfs_init_new_buffer on freshly allocated blocks.
153 * We also add a check to make sure the highest level of the tree is the
154 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
155 * needs update as well.
157 #ifdef CONFIG_DEBUG_LOCK_ALLOC
158 # if BTRFS_MAX_LEVEL != 8
162 static struct btrfs_lockdep_keyset {
163 u64 id; /* root objectid */
164 const char *name_stem; /* lock name stem */
165 char names[BTRFS_MAX_LEVEL + 1][20];
166 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
167 } btrfs_lockdep_keysets[] = {
168 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
169 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
170 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
171 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
172 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
173 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
174 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
175 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
176 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
177 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
178 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
179 { .id = 0, .name_stem = "tree" },
182 void __init btrfs_init_lockdep(void)
186 /* initialize lockdep class names */
187 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
188 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
190 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
191 snprintf(ks->names[j], sizeof(ks->names[j]),
192 "btrfs-%s-%02d", ks->name_stem, j);
196 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
199 struct btrfs_lockdep_keyset *ks;
201 BUG_ON(level >= ARRAY_SIZE(ks->keys));
203 /* find the matching keyset, id 0 is the default entry */
204 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
205 if (ks->id == objectid)
208 lockdep_set_class_and_name(&eb->lock,
209 &ks->keys[level], ks->names[level]);
215 * extents on the btree inode are pretty simple, there's one extent
216 * that covers the entire device
218 static struct extent_map *btree_get_extent(struct inode *inode,
219 struct page *page, size_t pg_offset, u64 start, u64 len,
222 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
223 struct extent_map *em;
226 read_lock(&em_tree->lock);
227 em = lookup_extent_mapping(em_tree, start, len);
230 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
231 read_unlock(&em_tree->lock);
234 read_unlock(&em_tree->lock);
236 em = alloc_extent_map();
238 em = ERR_PTR(-ENOMEM);
243 em->block_len = (u64)-1;
245 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
247 write_lock(&em_tree->lock);
248 ret = add_extent_mapping(em_tree, em, 0);
249 if (ret == -EEXIST) {
251 em = lookup_extent_mapping(em_tree, start, len);
258 write_unlock(&em_tree->lock);
264 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
266 return btrfs_crc32c(seed, data, len);
269 void btrfs_csum_final(u32 crc, char *result)
271 put_unaligned_le32(~crc, result);
275 * compute the csum for a btree block, and either verify it or write it
276 * into the csum field of the block.
278 static int csum_tree_block(struct btrfs_fs_info *fs_info,
279 struct extent_buffer *buf,
282 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
285 unsigned long cur_len;
286 unsigned long offset = BTRFS_CSUM_SIZE;
288 unsigned long map_start;
289 unsigned long map_len;
292 unsigned long inline_result;
294 len = buf->len - offset;
296 err = map_private_extent_buffer(buf, offset, 32,
297 &kaddr, &map_start, &map_len);
300 cur_len = min(len, map_len - (offset - map_start));
301 crc = btrfs_csum_data(kaddr + offset - map_start,
306 if (csum_size > sizeof(inline_result)) {
307 result = kzalloc(csum_size, GFP_NOFS);
311 result = (char *)&inline_result;
314 btrfs_csum_final(crc, result);
317 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
320 memcpy(&found, result, csum_size);
322 read_extent_buffer(buf, &val, 0, csum_size);
323 btrfs_warn_rl(fs_info,
324 "%s checksum verify failed on %llu wanted %X found %X "
326 fs_info->sb->s_id, buf->start,
327 val, found, btrfs_header_level(buf));
328 if (result != (char *)&inline_result)
333 write_extent_buffer(buf, result, 0, csum_size);
335 if (result != (char *)&inline_result)
341 * we can't consider a given block up to date unless the transid of the
342 * block matches the transid in the parent node's pointer. This is how we
343 * detect blocks that either didn't get written at all or got written
344 * in the wrong place.
346 static int verify_parent_transid(struct extent_io_tree *io_tree,
347 struct extent_buffer *eb, u64 parent_transid,
350 struct extent_state *cached_state = NULL;
352 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
354 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
361 btrfs_tree_read_lock(eb);
362 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
365 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
367 if (extent_buffer_uptodate(eb) &&
368 btrfs_header_generation(eb) == parent_transid) {
372 btrfs_err_rl(eb->fs_info,
373 "parent transid verify failed on %llu wanted %llu found %llu",
375 parent_transid, btrfs_header_generation(eb));
379 * Things reading via commit roots that don't have normal protection,
380 * like send, can have a really old block in cache that may point at a
381 * block that has been free'd and re-allocated. So don't clear uptodate
382 * if we find an eb that is under IO (dirty/writeback) because we could
383 * end up reading in the stale data and then writing it back out and
384 * making everybody very sad.
386 if (!extent_buffer_under_io(eb))
387 clear_extent_buffer_uptodate(eb);
389 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
390 &cached_state, GFP_NOFS);
392 btrfs_tree_read_unlock_blocking(eb);
397 * Return 0 if the superblock checksum type matches the checksum value of that
398 * algorithm. Pass the raw disk superblock data.
400 static int btrfs_check_super_csum(char *raw_disk_sb)
402 struct btrfs_super_block *disk_sb =
403 (struct btrfs_super_block *)raw_disk_sb;
404 u16 csum_type = btrfs_super_csum_type(disk_sb);
407 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
409 const int csum_size = sizeof(crc);
410 char result[csum_size];
413 * The super_block structure does not span the whole
414 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
415 * is filled with zeros and is included in the checkum.
417 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
418 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
419 btrfs_csum_final(crc, result);
421 if (memcmp(raw_disk_sb, result, csum_size))
425 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
426 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
435 * helper to read a given tree block, doing retries as required when
436 * the checksums don't match and we have alternate mirrors to try.
438 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
439 struct extent_buffer *eb,
440 u64 start, u64 parent_transid)
442 struct extent_io_tree *io_tree;
447 int failed_mirror = 0;
449 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
451 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
452 ret = read_extent_buffer_pages(io_tree, eb, start,
454 btree_get_extent, mirror_num);
456 if (!verify_parent_transid(io_tree, eb,
463 num_copies = btrfs_num_copies(root->fs_info,
468 if (!failed_mirror) {
470 failed_mirror = eb->read_mirror;
474 if (mirror_num == failed_mirror)
477 if (mirror_num > num_copies)
481 if (failed && !ret && failed_mirror)
482 repair_eb_io_failure(root, eb, failed_mirror);
488 * checksum a dirty tree block before IO. This has extra checks to make sure
489 * we only fill in the checksum field in the first page of a multi-page block
492 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
494 u64 start = page_offset(page);
496 struct extent_buffer *eb;
498 eb = (struct extent_buffer *)page->private;
499 if (page != eb->pages[0])
501 found_start = btrfs_header_bytenr(eb);
502 if (WARN_ON(found_start != start || !PageUptodate(page)))
504 csum_tree_block(fs_info, eb, 0);
508 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
509 struct extent_buffer *eb)
511 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
512 u8 fsid[BTRFS_UUID_SIZE];
515 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
517 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
521 fs_devices = fs_devices->seed;
526 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
527 u64 phy_offset, struct page *page,
528 u64 start, u64 end, int mirror)
532 struct extent_buffer *eb;
533 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
540 eb = (struct extent_buffer *)page->private;
542 /* the pending IO might have been the only thing that kept this buffer
543 * in memory. Make sure we have a ref for all this other checks
545 extent_buffer_get(eb);
547 reads_done = atomic_dec_and_test(&eb->io_pages);
551 eb->read_mirror = mirror;
552 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
557 found_start = btrfs_header_bytenr(eb);
558 if (found_start != eb->start) {
559 btrfs_err_rl(eb->fs_info, "bad tree block start %llu %llu",
560 found_start, eb->start);
564 if (check_tree_block_fsid(root->fs_info, eb)) {
565 btrfs_err_rl(eb->fs_info, "bad fsid on block %llu",
570 found_level = btrfs_header_level(eb);
571 if (found_level >= BTRFS_MAX_LEVEL) {
572 btrfs_err(root->fs_info, "bad tree block level %d",
573 (int)btrfs_header_level(eb));
578 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
581 ret = csum_tree_block(root->fs_info, eb, 1);
588 * If this is a leaf block and it is corrupt, set the corrupt bit so
589 * that we don't try and read the other copies of this block, just
592 if (found_level == 0 && btrfs_check_leaf_full(root, eb)) {
593 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
597 if (found_level > 0 && btrfs_check_node(root, eb))
601 set_extent_buffer_uptodate(eb);
604 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
605 btree_readahead_hook(root, eb, eb->start, ret);
609 * our io error hook is going to dec the io pages
610 * again, we have to make sure it has something
613 atomic_inc(&eb->io_pages);
614 clear_extent_buffer_uptodate(eb);
616 free_extent_buffer(eb);
621 static int btree_io_failed_hook(struct page *page, int failed_mirror)
623 struct extent_buffer *eb;
624 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
626 eb = (struct extent_buffer *)page->private;
627 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
628 eb->read_mirror = failed_mirror;
629 atomic_dec(&eb->io_pages);
630 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
631 btree_readahead_hook(root, eb, eb->start, -EIO);
632 return -EIO; /* we fixed nothing */
635 static void end_workqueue_bio(struct bio *bio)
637 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
638 struct btrfs_fs_info *fs_info;
639 struct btrfs_workqueue *wq;
640 btrfs_work_func_t func;
642 fs_info = end_io_wq->info;
643 end_io_wq->error = bio->bi_error;
645 if (bio->bi_rw & REQ_WRITE) {
646 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
647 wq = fs_info->endio_meta_write_workers;
648 func = btrfs_endio_meta_write_helper;
649 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
650 wq = fs_info->endio_freespace_worker;
651 func = btrfs_freespace_write_helper;
652 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
653 wq = fs_info->endio_raid56_workers;
654 func = btrfs_endio_raid56_helper;
656 wq = fs_info->endio_write_workers;
657 func = btrfs_endio_write_helper;
660 if (unlikely(end_io_wq->metadata ==
661 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
662 wq = fs_info->endio_repair_workers;
663 func = btrfs_endio_repair_helper;
664 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
665 wq = fs_info->endio_raid56_workers;
666 func = btrfs_endio_raid56_helper;
667 } else if (end_io_wq->metadata) {
668 wq = fs_info->endio_meta_workers;
669 func = btrfs_endio_meta_helper;
671 wq = fs_info->endio_workers;
672 func = btrfs_endio_helper;
676 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
677 btrfs_queue_work(wq, &end_io_wq->work);
680 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
681 enum btrfs_wq_endio_type metadata)
683 struct btrfs_end_io_wq *end_io_wq;
685 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
689 end_io_wq->private = bio->bi_private;
690 end_io_wq->end_io = bio->bi_end_io;
691 end_io_wq->info = info;
692 end_io_wq->error = 0;
693 end_io_wq->bio = bio;
694 end_io_wq->metadata = metadata;
696 bio->bi_private = end_io_wq;
697 bio->bi_end_io = end_workqueue_bio;
701 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
703 unsigned long limit = min_t(unsigned long,
704 info->thread_pool_size,
705 info->fs_devices->open_devices);
709 static void run_one_async_start(struct btrfs_work *work)
711 struct async_submit_bio *async;
714 async = container_of(work, struct async_submit_bio, work);
715 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
716 async->mirror_num, async->bio_flags,
722 static void run_one_async_done(struct btrfs_work *work)
724 struct btrfs_fs_info *fs_info;
725 struct async_submit_bio *async;
728 async = container_of(work, struct async_submit_bio, work);
729 fs_info = BTRFS_I(async->inode)->root->fs_info;
731 limit = btrfs_async_submit_limit(fs_info);
732 limit = limit * 2 / 3;
735 * atomic_dec_return implies a barrier for waitqueue_active
737 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
738 waitqueue_active(&fs_info->async_submit_wait))
739 wake_up(&fs_info->async_submit_wait);
741 /* If an error occured we just want to clean up the bio and move on */
743 async->bio->bi_error = async->error;
744 bio_endio(async->bio);
748 async->submit_bio_done(async->inode, async->rw, async->bio,
749 async->mirror_num, async->bio_flags,
753 static void run_one_async_free(struct btrfs_work *work)
755 struct async_submit_bio *async;
757 async = container_of(work, struct async_submit_bio, work);
761 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
762 int rw, struct bio *bio, int mirror_num,
763 unsigned long bio_flags,
765 extent_submit_bio_hook_t *submit_bio_start,
766 extent_submit_bio_hook_t *submit_bio_done)
768 struct async_submit_bio *async;
770 async = kmalloc(sizeof(*async), GFP_NOFS);
774 async->inode = inode;
777 async->mirror_num = mirror_num;
778 async->submit_bio_start = submit_bio_start;
779 async->submit_bio_done = submit_bio_done;
781 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
782 run_one_async_done, run_one_async_free);
784 async->bio_flags = bio_flags;
785 async->bio_offset = bio_offset;
789 atomic_inc(&fs_info->nr_async_submits);
792 btrfs_set_work_high_priority(&async->work);
794 btrfs_queue_work(fs_info->workers, &async->work);
796 while (atomic_read(&fs_info->async_submit_draining) &&
797 atomic_read(&fs_info->nr_async_submits)) {
798 wait_event(fs_info->async_submit_wait,
799 (atomic_read(&fs_info->nr_async_submits) == 0));
805 static int btree_csum_one_bio(struct bio *bio)
807 struct bio_vec *bvec;
808 struct btrfs_root *root;
811 bio_for_each_segment_all(bvec, bio, i) {
812 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
813 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
821 static int __btree_submit_bio_start(struct inode *inode, int rw,
822 struct bio *bio, int mirror_num,
823 unsigned long bio_flags,
827 * when we're called for a write, we're already in the async
828 * submission context. Just jump into btrfs_map_bio
830 return btree_csum_one_bio(bio);
833 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
834 int mirror_num, unsigned long bio_flags,
840 * when we're called for a write, we're already in the async
841 * submission context. Just jump into btrfs_map_bio
843 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
851 static int check_async_write(struct inode *inode, unsigned long bio_flags)
853 if (bio_flags & EXTENT_BIO_TREE_LOG)
856 if (static_cpu_has(X86_FEATURE_XMM4_2))
862 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
863 int mirror_num, unsigned long bio_flags,
866 int async = check_async_write(inode, bio_flags);
869 if (!(rw & REQ_WRITE)) {
871 * called for a read, do the setup so that checksum validation
872 * can happen in the async kernel threads
874 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
875 bio, BTRFS_WQ_ENDIO_METADATA);
878 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
881 ret = btree_csum_one_bio(bio);
884 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
888 * kthread helpers are used to submit writes so that
889 * checksumming can happen in parallel across all CPUs
891 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
892 inode, rw, bio, mirror_num, 0,
894 __btree_submit_bio_start,
895 __btree_submit_bio_done);
908 #ifdef CONFIG_MIGRATION
909 static int btree_migratepage(struct address_space *mapping,
910 struct page *newpage, struct page *page,
911 enum migrate_mode mode)
914 * we can't safely write a btree page from here,
915 * we haven't done the locking hook
920 * Buffers may be managed in a filesystem specific way.
921 * We must have no buffers or drop them.
923 if (page_has_private(page) &&
924 !try_to_release_page(page, GFP_KERNEL))
926 return migrate_page(mapping, newpage, page, mode);
931 static int btree_writepages(struct address_space *mapping,
932 struct writeback_control *wbc)
934 struct btrfs_fs_info *fs_info;
937 if (wbc->sync_mode == WB_SYNC_NONE) {
939 if (wbc->for_kupdate)
942 fs_info = BTRFS_I(mapping->host)->root->fs_info;
943 /* this is a bit racy, but that's ok */
944 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
945 BTRFS_DIRTY_METADATA_THRESH,
946 fs_info->dirty_metadata_batch);
950 return btree_write_cache_pages(mapping, wbc);
953 static int btree_readpage(struct file *file, struct page *page)
955 struct extent_io_tree *tree;
956 tree = &BTRFS_I(page->mapping->host)->io_tree;
957 return extent_read_full_page(tree, page, btree_get_extent, 0);
960 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
962 if (PageWriteback(page) || PageDirty(page))
965 return try_release_extent_buffer(page);
968 static void btree_invalidatepage(struct page *page, unsigned int offset,
971 struct extent_io_tree *tree;
972 tree = &BTRFS_I(page->mapping->host)->io_tree;
973 extent_invalidatepage(tree, page, offset);
974 btree_releasepage(page, GFP_NOFS);
975 if (PagePrivate(page)) {
976 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
977 "page private not zero on page %llu",
978 (unsigned long long)page_offset(page));
979 ClearPagePrivate(page);
980 set_page_private(page, 0);
981 page_cache_release(page);
985 static int btree_set_page_dirty(struct page *page)
988 struct extent_buffer *eb;
990 BUG_ON(!PagePrivate(page));
991 eb = (struct extent_buffer *)page->private;
993 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
994 BUG_ON(!atomic_read(&eb->refs));
995 btrfs_assert_tree_locked(eb);
997 return __set_page_dirty_nobuffers(page);
1000 static const struct address_space_operations btree_aops = {
1001 .readpage = btree_readpage,
1002 .writepages = btree_writepages,
1003 .releasepage = btree_releasepage,
1004 .invalidatepage = btree_invalidatepage,
1005 #ifdef CONFIG_MIGRATION
1006 .migratepage = btree_migratepage,
1008 .set_page_dirty = btree_set_page_dirty,
1011 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1013 struct extent_buffer *buf = NULL;
1014 struct inode *btree_inode = root->fs_info->btree_inode;
1016 buf = btrfs_find_create_tree_block(root, bytenr);
1019 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1020 buf, 0, WAIT_NONE, btree_get_extent, 0);
1021 free_extent_buffer(buf);
1024 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1025 int mirror_num, struct extent_buffer **eb)
1027 struct extent_buffer *buf = NULL;
1028 struct inode *btree_inode = root->fs_info->btree_inode;
1029 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1032 buf = btrfs_find_create_tree_block(root, bytenr);
1036 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1038 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1039 btree_get_extent, mirror_num);
1041 free_extent_buffer(buf);
1045 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1046 free_extent_buffer(buf);
1048 } else if (extent_buffer_uptodate(buf)) {
1051 free_extent_buffer(buf);
1056 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1059 return find_extent_buffer(fs_info, bytenr);
1062 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1065 if (btrfs_test_is_dummy_root(root))
1066 return alloc_test_extent_buffer(root->fs_info, bytenr);
1067 return alloc_extent_buffer(root->fs_info, bytenr);
1071 int btrfs_write_tree_block(struct extent_buffer *buf)
1073 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1074 buf->start + buf->len - 1);
1077 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1079 return filemap_fdatawait_range(buf->pages[0]->mapping,
1080 buf->start, buf->start + buf->len - 1);
1083 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1086 struct extent_buffer *buf = NULL;
1089 buf = btrfs_find_create_tree_block(root, bytenr);
1091 return ERR_PTR(-ENOMEM);
1093 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1095 free_extent_buffer(buf);
1096 return ERR_PTR(ret);
1102 void clean_tree_block(struct btrfs_trans_handle *trans,
1103 struct btrfs_fs_info *fs_info,
1104 struct extent_buffer *buf)
1106 if (btrfs_header_generation(buf) ==
1107 fs_info->running_transaction->transid) {
1108 btrfs_assert_tree_locked(buf);
1110 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1111 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1113 fs_info->dirty_metadata_batch);
1114 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1115 btrfs_set_lock_blocking(buf);
1116 clear_extent_buffer_dirty(buf);
1121 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1123 struct btrfs_subvolume_writers *writers;
1126 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1128 return ERR_PTR(-ENOMEM);
1130 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1133 return ERR_PTR(ret);
1136 init_waitqueue_head(&writers->wait);
1141 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1143 percpu_counter_destroy(&writers->counter);
1147 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1148 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1152 root->commit_root = NULL;
1153 root->sectorsize = sectorsize;
1154 root->nodesize = nodesize;
1155 root->stripesize = stripesize;
1157 root->orphan_cleanup_state = 0;
1159 root->objectid = objectid;
1160 root->last_trans = 0;
1161 root->highest_objectid = 0;
1162 root->nr_delalloc_inodes = 0;
1163 root->nr_ordered_extents = 0;
1165 root->inode_tree = RB_ROOT;
1166 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1167 root->block_rsv = NULL;
1168 root->orphan_block_rsv = NULL;
1170 INIT_LIST_HEAD(&root->dirty_list);
1171 INIT_LIST_HEAD(&root->root_list);
1172 INIT_LIST_HEAD(&root->delalloc_inodes);
1173 INIT_LIST_HEAD(&root->delalloc_root);
1174 INIT_LIST_HEAD(&root->ordered_extents);
1175 INIT_LIST_HEAD(&root->ordered_root);
1176 INIT_LIST_HEAD(&root->logged_list[0]);
1177 INIT_LIST_HEAD(&root->logged_list[1]);
1178 spin_lock_init(&root->orphan_lock);
1179 spin_lock_init(&root->inode_lock);
1180 spin_lock_init(&root->delalloc_lock);
1181 spin_lock_init(&root->ordered_extent_lock);
1182 spin_lock_init(&root->accounting_lock);
1183 spin_lock_init(&root->log_extents_lock[0]);
1184 spin_lock_init(&root->log_extents_lock[1]);
1185 mutex_init(&root->objectid_mutex);
1186 mutex_init(&root->log_mutex);
1187 mutex_init(&root->ordered_extent_mutex);
1188 mutex_init(&root->delalloc_mutex);
1189 init_waitqueue_head(&root->log_writer_wait);
1190 init_waitqueue_head(&root->log_commit_wait[0]);
1191 init_waitqueue_head(&root->log_commit_wait[1]);
1192 INIT_LIST_HEAD(&root->log_ctxs[0]);
1193 INIT_LIST_HEAD(&root->log_ctxs[1]);
1194 atomic_set(&root->log_commit[0], 0);
1195 atomic_set(&root->log_commit[1], 0);
1196 atomic_set(&root->log_writers, 0);
1197 atomic_set(&root->log_batch, 0);
1198 atomic_set(&root->orphan_inodes, 0);
1199 atomic_set(&root->refs, 1);
1200 atomic_set(&root->will_be_snapshoted, 0);
1201 atomic_set(&root->qgroup_meta_rsv, 0);
1202 root->log_transid = 0;
1203 root->log_transid_committed = -1;
1204 root->last_log_commit = 0;
1206 extent_io_tree_init(&root->dirty_log_pages,
1207 fs_info->btree_inode->i_mapping);
1209 memset(&root->root_key, 0, sizeof(root->root_key));
1210 memset(&root->root_item, 0, sizeof(root->root_item));
1211 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1213 root->defrag_trans_start = fs_info->generation;
1215 root->defrag_trans_start = 0;
1216 root->root_key.objectid = objectid;
1219 spin_lock_init(&root->root_item_lock);
1222 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1224 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1226 root->fs_info = fs_info;
1230 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1231 /* Should only be used by the testing infrastructure */
1232 struct btrfs_root *btrfs_alloc_dummy_root(void)
1234 struct btrfs_root *root;
1236 root = btrfs_alloc_root(NULL);
1238 return ERR_PTR(-ENOMEM);
1239 __setup_root(4096, 4096, 4096, root, NULL, 1);
1240 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1241 root->alloc_bytenr = 0;
1247 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1248 struct btrfs_fs_info *fs_info,
1251 struct extent_buffer *leaf;
1252 struct btrfs_root *tree_root = fs_info->tree_root;
1253 struct btrfs_root *root;
1254 struct btrfs_key key;
1258 root = btrfs_alloc_root(fs_info);
1260 return ERR_PTR(-ENOMEM);
1262 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1263 tree_root->stripesize, root, fs_info, objectid);
1264 root->root_key.objectid = objectid;
1265 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1266 root->root_key.offset = 0;
1268 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1270 ret = PTR_ERR(leaf);
1275 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1276 btrfs_set_header_bytenr(leaf, leaf->start);
1277 btrfs_set_header_generation(leaf, trans->transid);
1278 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1279 btrfs_set_header_owner(leaf, objectid);
1282 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1284 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1285 btrfs_header_chunk_tree_uuid(leaf),
1287 btrfs_mark_buffer_dirty(leaf);
1289 root->commit_root = btrfs_root_node(root);
1290 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1292 root->root_item.flags = 0;
1293 root->root_item.byte_limit = 0;
1294 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1295 btrfs_set_root_generation(&root->root_item, trans->transid);
1296 btrfs_set_root_level(&root->root_item, 0);
1297 btrfs_set_root_refs(&root->root_item, 1);
1298 btrfs_set_root_used(&root->root_item, leaf->len);
1299 btrfs_set_root_last_snapshot(&root->root_item, 0);
1300 btrfs_set_root_dirid(&root->root_item, 0);
1302 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1303 root->root_item.drop_level = 0;
1305 key.objectid = objectid;
1306 key.type = BTRFS_ROOT_ITEM_KEY;
1308 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1312 btrfs_tree_unlock(leaf);
1318 btrfs_tree_unlock(leaf);
1319 free_extent_buffer(root->commit_root);
1320 free_extent_buffer(leaf);
1324 return ERR_PTR(ret);
1327 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1328 struct btrfs_fs_info *fs_info)
1330 struct btrfs_root *root;
1331 struct btrfs_root *tree_root = fs_info->tree_root;
1332 struct extent_buffer *leaf;
1334 root = btrfs_alloc_root(fs_info);
1336 return ERR_PTR(-ENOMEM);
1338 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1339 tree_root->stripesize, root, fs_info,
1340 BTRFS_TREE_LOG_OBJECTID);
1342 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1343 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1344 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1347 * DON'T set REF_COWS for log trees
1349 * log trees do not get reference counted because they go away
1350 * before a real commit is actually done. They do store pointers
1351 * to file data extents, and those reference counts still get
1352 * updated (along with back refs to the log tree).
1355 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1359 return ERR_CAST(leaf);
1362 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1363 btrfs_set_header_bytenr(leaf, leaf->start);
1364 btrfs_set_header_generation(leaf, trans->transid);
1365 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1366 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1369 write_extent_buffer(root->node, root->fs_info->fsid,
1370 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1371 btrfs_mark_buffer_dirty(root->node);
1372 btrfs_tree_unlock(root->node);
1376 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1377 struct btrfs_fs_info *fs_info)
1379 struct btrfs_root *log_root;
1381 log_root = alloc_log_tree(trans, fs_info);
1382 if (IS_ERR(log_root))
1383 return PTR_ERR(log_root);
1384 WARN_ON(fs_info->log_root_tree);
1385 fs_info->log_root_tree = log_root;
1389 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1390 struct btrfs_root *root)
1392 struct btrfs_root *log_root;
1393 struct btrfs_inode_item *inode_item;
1395 log_root = alloc_log_tree(trans, root->fs_info);
1396 if (IS_ERR(log_root))
1397 return PTR_ERR(log_root);
1399 log_root->last_trans = trans->transid;
1400 log_root->root_key.offset = root->root_key.objectid;
1402 inode_item = &log_root->root_item.inode;
1403 btrfs_set_stack_inode_generation(inode_item, 1);
1404 btrfs_set_stack_inode_size(inode_item, 3);
1405 btrfs_set_stack_inode_nlink(inode_item, 1);
1406 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1407 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1409 btrfs_set_root_node(&log_root->root_item, log_root->node);
1411 WARN_ON(root->log_root);
1412 root->log_root = log_root;
1413 root->log_transid = 0;
1414 root->log_transid_committed = -1;
1415 root->last_log_commit = 0;
1419 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1420 struct btrfs_key *key)
1422 struct btrfs_root *root;
1423 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1424 struct btrfs_path *path;
1428 path = btrfs_alloc_path();
1430 return ERR_PTR(-ENOMEM);
1432 root = btrfs_alloc_root(fs_info);
1438 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1439 tree_root->stripesize, root, fs_info, key->objectid);
1441 ret = btrfs_find_root(tree_root, key, path,
1442 &root->root_item, &root->root_key);
1449 generation = btrfs_root_generation(&root->root_item);
1450 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1452 if (IS_ERR(root->node)) {
1453 ret = PTR_ERR(root->node);
1455 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1457 free_extent_buffer(root->node);
1460 root->commit_root = btrfs_root_node(root);
1462 btrfs_free_path(path);
1468 root = ERR_PTR(ret);
1472 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1473 struct btrfs_key *location)
1475 struct btrfs_root *root;
1477 root = btrfs_read_tree_root(tree_root, location);
1481 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1482 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1483 btrfs_check_and_init_root_item(&root->root_item);
1489 int btrfs_init_fs_root(struct btrfs_root *root)
1492 struct btrfs_subvolume_writers *writers;
1494 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1495 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1497 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1502 writers = btrfs_alloc_subvolume_writers();
1503 if (IS_ERR(writers)) {
1504 ret = PTR_ERR(writers);
1507 root->subv_writers = writers;
1509 btrfs_init_free_ino_ctl(root);
1510 spin_lock_init(&root->ino_cache_lock);
1511 init_waitqueue_head(&root->ino_cache_wait);
1513 ret = get_anon_bdev(&root->anon_dev);
1517 mutex_lock(&root->objectid_mutex);
1518 ret = btrfs_find_highest_objectid(root,
1519 &root->highest_objectid);
1521 mutex_unlock(&root->objectid_mutex);
1525 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1527 mutex_unlock(&root->objectid_mutex);
1532 free_anon_bdev(root->anon_dev);
1534 btrfs_free_subvolume_writers(root->subv_writers);
1536 kfree(root->free_ino_ctl);
1537 kfree(root->free_ino_pinned);
1541 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1544 struct btrfs_root *root;
1546 spin_lock(&fs_info->fs_roots_radix_lock);
1547 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1548 (unsigned long)root_id);
1549 spin_unlock(&fs_info->fs_roots_radix_lock);
1553 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1554 struct btrfs_root *root)
1558 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1562 spin_lock(&fs_info->fs_roots_radix_lock);
1563 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1564 (unsigned long)root->root_key.objectid,
1567 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1568 spin_unlock(&fs_info->fs_roots_radix_lock);
1569 radix_tree_preload_end();
1574 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1575 struct btrfs_key *location,
1578 struct btrfs_root *root;
1579 struct btrfs_path *path;
1580 struct btrfs_key key;
1583 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1584 return fs_info->tree_root;
1585 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1586 return fs_info->extent_root;
1587 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1588 return fs_info->chunk_root;
1589 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1590 return fs_info->dev_root;
1591 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1592 return fs_info->csum_root;
1593 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1594 return fs_info->quota_root ? fs_info->quota_root :
1596 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1597 return fs_info->uuid_root ? fs_info->uuid_root :
1600 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1602 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1603 return ERR_PTR(-ENOENT);
1607 root = btrfs_read_fs_root(fs_info->tree_root, location);
1611 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1616 ret = btrfs_init_fs_root(root);
1620 path = btrfs_alloc_path();
1625 key.objectid = BTRFS_ORPHAN_OBJECTID;
1626 key.type = BTRFS_ORPHAN_ITEM_KEY;
1627 key.offset = location->objectid;
1629 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1630 btrfs_free_path(path);
1634 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1636 ret = btrfs_insert_fs_root(fs_info, root);
1638 if (ret == -EEXIST) {
1647 return ERR_PTR(ret);
1650 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1652 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1654 struct btrfs_device *device;
1655 struct backing_dev_info *bdi;
1658 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1661 bdi = blk_get_backing_dev_info(device->bdev);
1662 if (bdi_congested(bdi, bdi_bits)) {
1671 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1675 err = bdi_setup_and_register(bdi, "btrfs");
1679 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1680 bdi->congested_fn = btrfs_congested_fn;
1681 bdi->congested_data = info;
1682 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1687 * called by the kthread helper functions to finally call the bio end_io
1688 * functions. This is where read checksum verification actually happens
1690 static void end_workqueue_fn(struct btrfs_work *work)
1693 struct btrfs_end_io_wq *end_io_wq;
1695 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1696 bio = end_io_wq->bio;
1698 bio->bi_error = end_io_wq->error;
1699 bio->bi_private = end_io_wq->private;
1700 bio->bi_end_io = end_io_wq->end_io;
1701 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1705 static int cleaner_kthread(void *arg)
1707 struct btrfs_root *root = arg;
1709 struct btrfs_trans_handle *trans;
1714 /* Make the cleaner go to sleep early. */
1715 if (btrfs_need_cleaner_sleep(root))
1718 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1722 * Avoid the problem that we change the status of the fs
1723 * during the above check and trylock.
1725 if (btrfs_need_cleaner_sleep(root)) {
1726 mutex_unlock(&root->fs_info->cleaner_mutex);
1730 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1731 btrfs_run_delayed_iputs(root);
1732 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1734 again = btrfs_clean_one_deleted_snapshot(root);
1735 mutex_unlock(&root->fs_info->cleaner_mutex);
1738 * The defragger has dealt with the R/O remount and umount,
1739 * needn't do anything special here.
1741 btrfs_run_defrag_inodes(root->fs_info);
1744 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1745 * with relocation (btrfs_relocate_chunk) and relocation
1746 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1747 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1748 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1749 * unused block groups.
1751 btrfs_delete_unused_bgs(root->fs_info);
1753 if (!try_to_freeze() && !again) {
1754 set_current_state(TASK_INTERRUPTIBLE);
1755 if (!kthread_should_stop())
1757 __set_current_state(TASK_RUNNING);
1759 } while (!kthread_should_stop());
1762 * Transaction kthread is stopped before us and wakes us up.
1763 * However we might have started a new transaction and COWed some
1764 * tree blocks when deleting unused block groups for example. So
1765 * make sure we commit the transaction we started to have a clean
1766 * shutdown when evicting the btree inode - if it has dirty pages
1767 * when we do the final iput() on it, eviction will trigger a
1768 * writeback for it which will fail with null pointer dereferences
1769 * since work queues and other resources were already released and
1770 * destroyed by the time the iput/eviction/writeback is made.
1772 trans = btrfs_attach_transaction(root);
1773 if (IS_ERR(trans)) {
1774 if (PTR_ERR(trans) != -ENOENT)
1775 btrfs_err(root->fs_info,
1776 "cleaner transaction attach returned %ld",
1781 ret = btrfs_commit_transaction(trans, root);
1783 btrfs_err(root->fs_info,
1784 "cleaner open transaction commit returned %d",
1791 static int transaction_kthread(void *arg)
1793 struct btrfs_root *root = arg;
1794 struct btrfs_trans_handle *trans;
1795 struct btrfs_transaction *cur;
1798 unsigned long delay;
1802 cannot_commit = false;
1803 delay = HZ * root->fs_info->commit_interval;
1804 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1806 spin_lock(&root->fs_info->trans_lock);
1807 cur = root->fs_info->running_transaction;
1809 spin_unlock(&root->fs_info->trans_lock);
1813 now = get_seconds();
1814 if (cur->state < TRANS_STATE_BLOCKED &&
1815 (now < cur->start_time ||
1816 now - cur->start_time < root->fs_info->commit_interval)) {
1817 spin_unlock(&root->fs_info->trans_lock);
1821 transid = cur->transid;
1822 spin_unlock(&root->fs_info->trans_lock);
1824 /* If the file system is aborted, this will always fail. */
1825 trans = btrfs_attach_transaction(root);
1826 if (IS_ERR(trans)) {
1827 if (PTR_ERR(trans) != -ENOENT)
1828 cannot_commit = true;
1831 if (transid == trans->transid) {
1832 btrfs_commit_transaction(trans, root);
1834 btrfs_end_transaction(trans, root);
1837 wake_up_process(root->fs_info->cleaner_kthread);
1838 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1840 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1841 &root->fs_info->fs_state)))
1842 btrfs_cleanup_transaction(root);
1843 if (!try_to_freeze()) {
1844 set_current_state(TASK_INTERRUPTIBLE);
1845 if (!kthread_should_stop() &&
1846 (!btrfs_transaction_blocked(root->fs_info) ||
1848 schedule_timeout(delay);
1849 __set_current_state(TASK_RUNNING);
1851 } while (!kthread_should_stop());
1856 * this will find the highest generation in the array of
1857 * root backups. The index of the highest array is returned,
1858 * or -1 if we can't find anything.
1860 * We check to make sure the array is valid by comparing the
1861 * generation of the latest root in the array with the generation
1862 * in the super block. If they don't match we pitch it.
1864 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1867 int newest_index = -1;
1868 struct btrfs_root_backup *root_backup;
1871 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1872 root_backup = info->super_copy->super_roots + i;
1873 cur = btrfs_backup_tree_root_gen(root_backup);
1874 if (cur == newest_gen)
1878 /* check to see if we actually wrapped around */
1879 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1880 root_backup = info->super_copy->super_roots;
1881 cur = btrfs_backup_tree_root_gen(root_backup);
1882 if (cur == newest_gen)
1885 return newest_index;
1890 * find the oldest backup so we know where to store new entries
1891 * in the backup array. This will set the backup_root_index
1892 * field in the fs_info struct
1894 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1897 int newest_index = -1;
1899 newest_index = find_newest_super_backup(info, newest_gen);
1900 /* if there was garbage in there, just move along */
1901 if (newest_index == -1) {
1902 info->backup_root_index = 0;
1904 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1909 * copy all the root pointers into the super backup array.
1910 * this will bump the backup pointer by one when it is
1913 static void backup_super_roots(struct btrfs_fs_info *info)
1916 struct btrfs_root_backup *root_backup;
1919 next_backup = info->backup_root_index;
1920 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1921 BTRFS_NUM_BACKUP_ROOTS;
1924 * just overwrite the last backup if we're at the same generation
1925 * this happens only at umount
1927 root_backup = info->super_for_commit->super_roots + last_backup;
1928 if (btrfs_backup_tree_root_gen(root_backup) ==
1929 btrfs_header_generation(info->tree_root->node))
1930 next_backup = last_backup;
1932 root_backup = info->super_for_commit->super_roots + next_backup;
1935 * make sure all of our padding and empty slots get zero filled
1936 * regardless of which ones we use today
1938 memset(root_backup, 0, sizeof(*root_backup));
1940 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1942 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1943 btrfs_set_backup_tree_root_gen(root_backup,
1944 btrfs_header_generation(info->tree_root->node));
1946 btrfs_set_backup_tree_root_level(root_backup,
1947 btrfs_header_level(info->tree_root->node));
1949 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1950 btrfs_set_backup_chunk_root_gen(root_backup,
1951 btrfs_header_generation(info->chunk_root->node));
1952 btrfs_set_backup_chunk_root_level(root_backup,
1953 btrfs_header_level(info->chunk_root->node));
1955 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1956 btrfs_set_backup_extent_root_gen(root_backup,
1957 btrfs_header_generation(info->extent_root->node));
1958 btrfs_set_backup_extent_root_level(root_backup,
1959 btrfs_header_level(info->extent_root->node));
1962 * we might commit during log recovery, which happens before we set
1963 * the fs_root. Make sure it is valid before we fill it in.
1965 if (info->fs_root && info->fs_root->node) {
1966 btrfs_set_backup_fs_root(root_backup,
1967 info->fs_root->node->start);
1968 btrfs_set_backup_fs_root_gen(root_backup,
1969 btrfs_header_generation(info->fs_root->node));
1970 btrfs_set_backup_fs_root_level(root_backup,
1971 btrfs_header_level(info->fs_root->node));
1974 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1975 btrfs_set_backup_dev_root_gen(root_backup,
1976 btrfs_header_generation(info->dev_root->node));
1977 btrfs_set_backup_dev_root_level(root_backup,
1978 btrfs_header_level(info->dev_root->node));
1980 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1981 btrfs_set_backup_csum_root_gen(root_backup,
1982 btrfs_header_generation(info->csum_root->node));
1983 btrfs_set_backup_csum_root_level(root_backup,
1984 btrfs_header_level(info->csum_root->node));
1986 btrfs_set_backup_total_bytes(root_backup,
1987 btrfs_super_total_bytes(info->super_copy));
1988 btrfs_set_backup_bytes_used(root_backup,
1989 btrfs_super_bytes_used(info->super_copy));
1990 btrfs_set_backup_num_devices(root_backup,
1991 btrfs_super_num_devices(info->super_copy));
1994 * if we don't copy this out to the super_copy, it won't get remembered
1995 * for the next commit
1997 memcpy(&info->super_copy->super_roots,
1998 &info->super_for_commit->super_roots,
1999 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2003 * this copies info out of the root backup array and back into
2004 * the in-memory super block. It is meant to help iterate through
2005 * the array, so you send it the number of backups you've already
2006 * tried and the last backup index you used.
2008 * this returns -1 when it has tried all the backups
2010 static noinline int next_root_backup(struct btrfs_fs_info *info,
2011 struct btrfs_super_block *super,
2012 int *num_backups_tried, int *backup_index)
2014 struct btrfs_root_backup *root_backup;
2015 int newest = *backup_index;
2017 if (*num_backups_tried == 0) {
2018 u64 gen = btrfs_super_generation(super);
2020 newest = find_newest_super_backup(info, gen);
2024 *backup_index = newest;
2025 *num_backups_tried = 1;
2026 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2027 /* we've tried all the backups, all done */
2030 /* jump to the next oldest backup */
2031 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2032 BTRFS_NUM_BACKUP_ROOTS;
2033 *backup_index = newest;
2034 *num_backups_tried += 1;
2036 root_backup = super->super_roots + newest;
2038 btrfs_set_super_generation(super,
2039 btrfs_backup_tree_root_gen(root_backup));
2040 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2041 btrfs_set_super_root_level(super,
2042 btrfs_backup_tree_root_level(root_backup));
2043 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2046 * fixme: the total bytes and num_devices need to match or we should
2049 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2050 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2054 /* helper to cleanup workers */
2055 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2057 btrfs_destroy_workqueue(fs_info->fixup_workers);
2058 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2059 btrfs_destroy_workqueue(fs_info->workers);
2060 btrfs_destroy_workqueue(fs_info->endio_workers);
2061 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2062 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2063 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2064 btrfs_destroy_workqueue(fs_info->rmw_workers);
2065 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2066 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2067 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2068 btrfs_destroy_workqueue(fs_info->submit_workers);
2069 btrfs_destroy_workqueue(fs_info->delayed_workers);
2070 btrfs_destroy_workqueue(fs_info->caching_workers);
2071 btrfs_destroy_workqueue(fs_info->readahead_workers);
2072 btrfs_destroy_workqueue(fs_info->flush_workers);
2073 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2074 btrfs_destroy_workqueue(fs_info->extent_workers);
2077 static void free_root_extent_buffers(struct btrfs_root *root)
2080 free_extent_buffer(root->node);
2081 free_extent_buffer(root->commit_root);
2083 root->commit_root = NULL;
2087 /* helper to cleanup tree roots */
2088 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2090 free_root_extent_buffers(info->tree_root);
2092 free_root_extent_buffers(info->dev_root);
2093 free_root_extent_buffers(info->extent_root);
2094 free_root_extent_buffers(info->csum_root);
2095 free_root_extent_buffers(info->quota_root);
2096 free_root_extent_buffers(info->uuid_root);
2098 free_root_extent_buffers(info->chunk_root);
2101 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2104 struct btrfs_root *gang[8];
2107 while (!list_empty(&fs_info->dead_roots)) {
2108 gang[0] = list_entry(fs_info->dead_roots.next,
2109 struct btrfs_root, root_list);
2110 list_del(&gang[0]->root_list);
2112 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2113 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2115 free_extent_buffer(gang[0]->node);
2116 free_extent_buffer(gang[0]->commit_root);
2117 btrfs_put_fs_root(gang[0]);
2122 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2127 for (i = 0; i < ret; i++)
2128 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2131 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2132 btrfs_free_log_root_tree(NULL, fs_info);
2133 btrfs_destroy_pinned_extent(fs_info->tree_root,
2134 fs_info->pinned_extents);
2138 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2140 mutex_init(&fs_info->scrub_lock);
2141 atomic_set(&fs_info->scrubs_running, 0);
2142 atomic_set(&fs_info->scrub_pause_req, 0);
2143 atomic_set(&fs_info->scrubs_paused, 0);
2144 atomic_set(&fs_info->scrub_cancel_req, 0);
2145 init_waitqueue_head(&fs_info->scrub_pause_wait);
2146 fs_info->scrub_workers_refcnt = 0;
2149 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2151 spin_lock_init(&fs_info->balance_lock);
2152 mutex_init(&fs_info->balance_mutex);
2153 atomic_set(&fs_info->balance_running, 0);
2154 atomic_set(&fs_info->balance_pause_req, 0);
2155 atomic_set(&fs_info->balance_cancel_req, 0);
2156 fs_info->balance_ctl = NULL;
2157 init_waitqueue_head(&fs_info->balance_wait_q);
2160 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2161 struct btrfs_root *tree_root)
2163 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2164 set_nlink(fs_info->btree_inode, 1);
2166 * we set the i_size on the btree inode to the max possible int.
2167 * the real end of the address space is determined by all of
2168 * the devices in the system
2170 fs_info->btree_inode->i_size = OFFSET_MAX;
2171 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2173 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2174 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2175 fs_info->btree_inode->i_mapping);
2176 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2177 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2179 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2181 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2182 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2183 sizeof(struct btrfs_key));
2184 set_bit(BTRFS_INODE_DUMMY,
2185 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2186 btrfs_insert_inode_hash(fs_info->btree_inode);
2189 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2191 fs_info->dev_replace.lock_owner = 0;
2192 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2193 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2194 mutex_init(&fs_info->dev_replace.lock_management_lock);
2195 mutex_init(&fs_info->dev_replace.lock);
2196 init_waitqueue_head(&fs_info->replace_wait);
2199 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2201 spin_lock_init(&fs_info->qgroup_lock);
2202 mutex_init(&fs_info->qgroup_ioctl_lock);
2203 fs_info->qgroup_tree = RB_ROOT;
2204 fs_info->qgroup_op_tree = RB_ROOT;
2205 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2206 fs_info->qgroup_seq = 1;
2207 fs_info->quota_enabled = 0;
2208 fs_info->pending_quota_state = 0;
2209 fs_info->qgroup_ulist = NULL;
2210 fs_info->qgroup_rescan_running = false;
2211 mutex_init(&fs_info->qgroup_rescan_lock);
2214 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2215 struct btrfs_fs_devices *fs_devices)
2217 int max_active = fs_info->thread_pool_size;
2218 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2221 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2224 fs_info->delalloc_workers =
2225 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2227 fs_info->flush_workers =
2228 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2230 fs_info->caching_workers =
2231 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2234 * a higher idle thresh on the submit workers makes it much more
2235 * likely that bios will be send down in a sane order to the
2238 fs_info->submit_workers =
2239 btrfs_alloc_workqueue("submit", flags,
2240 min_t(u64, fs_devices->num_devices,
2243 fs_info->fixup_workers =
2244 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2247 * endios are largely parallel and should have a very
2250 fs_info->endio_workers =
2251 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2252 fs_info->endio_meta_workers =
2253 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2254 fs_info->endio_meta_write_workers =
2255 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2256 fs_info->endio_raid56_workers =
2257 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2258 fs_info->endio_repair_workers =
2259 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2260 fs_info->rmw_workers =
2261 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2262 fs_info->endio_write_workers =
2263 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2264 fs_info->endio_freespace_worker =
2265 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2266 fs_info->delayed_workers =
2267 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2268 fs_info->readahead_workers =
2269 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2270 fs_info->qgroup_rescan_workers =
2271 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2272 fs_info->extent_workers =
2273 btrfs_alloc_workqueue("extent-refs", flags,
2274 min_t(u64, fs_devices->num_devices,
2277 if (!(fs_info->workers && fs_info->delalloc_workers &&
2278 fs_info->submit_workers && fs_info->flush_workers &&
2279 fs_info->endio_workers && fs_info->endio_meta_workers &&
2280 fs_info->endio_meta_write_workers &&
2281 fs_info->endio_repair_workers &&
2282 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2283 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2284 fs_info->caching_workers && fs_info->readahead_workers &&
2285 fs_info->fixup_workers && fs_info->delayed_workers &&
2286 fs_info->extent_workers &&
2287 fs_info->qgroup_rescan_workers)) {
2294 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2295 struct btrfs_fs_devices *fs_devices)
2298 struct btrfs_root *tree_root = fs_info->tree_root;
2299 struct btrfs_root *log_tree_root;
2300 struct btrfs_super_block *disk_super = fs_info->super_copy;
2301 u64 bytenr = btrfs_super_log_root(disk_super);
2303 if (fs_devices->rw_devices == 0) {
2304 btrfs_warn(fs_info, "log replay required on RO media");
2308 log_tree_root = btrfs_alloc_root(fs_info);
2312 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2313 tree_root->stripesize, log_tree_root, fs_info,
2314 BTRFS_TREE_LOG_OBJECTID);
2316 log_tree_root->node = read_tree_block(tree_root, bytenr,
2317 fs_info->generation + 1);
2318 if (IS_ERR(log_tree_root->node)) {
2319 btrfs_warn(fs_info, "failed to read log tree");
2320 ret = PTR_ERR(log_tree_root->node);
2321 kfree(log_tree_root);
2323 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2324 btrfs_err(fs_info, "failed to read log tree");
2325 free_extent_buffer(log_tree_root->node);
2326 kfree(log_tree_root);
2329 /* returns with log_tree_root freed on success */
2330 ret = btrfs_recover_log_trees(log_tree_root);
2332 btrfs_std_error(tree_root->fs_info, ret,
2333 "Failed to recover log tree");
2334 free_extent_buffer(log_tree_root->node);
2335 kfree(log_tree_root);
2339 if (fs_info->sb->s_flags & MS_RDONLY) {
2340 ret = btrfs_commit_super(tree_root);
2348 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2349 struct btrfs_root *tree_root)
2351 struct btrfs_root *root;
2352 struct btrfs_key location;
2355 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2356 location.type = BTRFS_ROOT_ITEM_KEY;
2357 location.offset = 0;
2359 root = btrfs_read_tree_root(tree_root, &location);
2361 return PTR_ERR(root);
2362 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2363 fs_info->extent_root = root;
2365 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2366 root = btrfs_read_tree_root(tree_root, &location);
2368 return PTR_ERR(root);
2369 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2370 fs_info->dev_root = root;
2371 btrfs_init_devices_late(fs_info);
2373 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2374 root = btrfs_read_tree_root(tree_root, &location);
2376 return PTR_ERR(root);
2377 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2378 fs_info->csum_root = root;
2380 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2381 root = btrfs_read_tree_root(tree_root, &location);
2382 if (!IS_ERR(root)) {
2383 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2384 fs_info->quota_enabled = 1;
2385 fs_info->pending_quota_state = 1;
2386 fs_info->quota_root = root;
2389 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2390 root = btrfs_read_tree_root(tree_root, &location);
2392 ret = PTR_ERR(root);
2396 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2397 fs_info->uuid_root = root;
2403 int open_ctree(struct super_block *sb,
2404 struct btrfs_fs_devices *fs_devices,
2412 struct btrfs_key location;
2413 struct buffer_head *bh;
2414 struct btrfs_super_block *disk_super;
2415 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2416 struct btrfs_root *tree_root;
2417 struct btrfs_root *chunk_root;
2420 int num_backups_tried = 0;
2421 int backup_index = 0;
2424 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2425 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2426 if (!tree_root || !chunk_root) {
2431 ret = init_srcu_struct(&fs_info->subvol_srcu);
2437 ret = setup_bdi(fs_info, &fs_info->bdi);
2443 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2448 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2449 (1 + ilog2(nr_cpu_ids));
2451 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2454 goto fail_dirty_metadata_bytes;
2457 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2460 goto fail_delalloc_bytes;
2463 fs_info->btree_inode = new_inode(sb);
2464 if (!fs_info->btree_inode) {
2466 goto fail_bio_counter;
2469 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2471 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2472 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2473 INIT_LIST_HEAD(&fs_info->trans_list);
2474 INIT_LIST_HEAD(&fs_info->dead_roots);
2475 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2476 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2477 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2478 spin_lock_init(&fs_info->delalloc_root_lock);
2479 spin_lock_init(&fs_info->trans_lock);
2480 spin_lock_init(&fs_info->fs_roots_radix_lock);
2481 spin_lock_init(&fs_info->delayed_iput_lock);
2482 spin_lock_init(&fs_info->defrag_inodes_lock);
2483 spin_lock_init(&fs_info->free_chunk_lock);
2484 spin_lock_init(&fs_info->tree_mod_seq_lock);
2485 spin_lock_init(&fs_info->super_lock);
2486 spin_lock_init(&fs_info->qgroup_op_lock);
2487 spin_lock_init(&fs_info->buffer_lock);
2488 spin_lock_init(&fs_info->unused_bgs_lock);
2489 rwlock_init(&fs_info->tree_mod_log_lock);
2490 mutex_init(&fs_info->unused_bg_unpin_mutex);
2491 mutex_init(&fs_info->delete_unused_bgs_mutex);
2492 mutex_init(&fs_info->reloc_mutex);
2493 mutex_init(&fs_info->delalloc_root_mutex);
2494 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2495 seqlock_init(&fs_info->profiles_lock);
2497 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2498 INIT_LIST_HEAD(&fs_info->space_info);
2499 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2500 INIT_LIST_HEAD(&fs_info->unused_bgs);
2501 btrfs_mapping_init(&fs_info->mapping_tree);
2502 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2503 BTRFS_BLOCK_RSV_GLOBAL);
2504 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2505 BTRFS_BLOCK_RSV_DELALLOC);
2506 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2507 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2508 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2509 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2510 BTRFS_BLOCK_RSV_DELOPS);
2511 atomic_set(&fs_info->nr_async_submits, 0);
2512 atomic_set(&fs_info->async_delalloc_pages, 0);
2513 atomic_set(&fs_info->async_submit_draining, 0);
2514 atomic_set(&fs_info->nr_async_bios, 0);
2515 atomic_set(&fs_info->defrag_running, 0);
2516 atomic_set(&fs_info->qgroup_op_seq, 0);
2517 atomic64_set(&fs_info->tree_mod_seq, 0);
2519 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2520 fs_info->metadata_ratio = 0;
2521 fs_info->defrag_inodes = RB_ROOT;
2522 fs_info->free_chunk_space = 0;
2523 fs_info->tree_mod_log = RB_ROOT;
2524 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2525 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2526 /* readahead state */
2527 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2528 spin_lock_init(&fs_info->reada_lock);
2530 fs_info->thread_pool_size = min_t(unsigned long,
2531 num_online_cpus() + 2, 8);
2533 INIT_LIST_HEAD(&fs_info->ordered_roots);
2534 spin_lock_init(&fs_info->ordered_root_lock);
2535 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2537 if (!fs_info->delayed_root) {
2541 btrfs_init_delayed_root(fs_info->delayed_root);
2543 btrfs_init_scrub(fs_info);
2544 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2545 fs_info->check_integrity_print_mask = 0;
2547 btrfs_init_balance(fs_info);
2548 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2550 sb->s_blocksize = 4096;
2551 sb->s_blocksize_bits = blksize_bits(4096);
2552 sb->s_bdi = &fs_info->bdi;
2554 btrfs_init_btree_inode(fs_info, tree_root);
2556 spin_lock_init(&fs_info->block_group_cache_lock);
2557 fs_info->block_group_cache_tree = RB_ROOT;
2558 fs_info->first_logical_byte = (u64)-1;
2560 extent_io_tree_init(&fs_info->freed_extents[0],
2561 fs_info->btree_inode->i_mapping);
2562 extent_io_tree_init(&fs_info->freed_extents[1],
2563 fs_info->btree_inode->i_mapping);
2564 fs_info->pinned_extents = &fs_info->freed_extents[0];
2565 fs_info->do_barriers = 1;
2568 mutex_init(&fs_info->ordered_operations_mutex);
2569 mutex_init(&fs_info->tree_log_mutex);
2570 mutex_init(&fs_info->chunk_mutex);
2571 mutex_init(&fs_info->transaction_kthread_mutex);
2572 mutex_init(&fs_info->cleaner_mutex);
2573 mutex_init(&fs_info->volume_mutex);
2574 mutex_init(&fs_info->ro_block_group_mutex);
2575 init_rwsem(&fs_info->commit_root_sem);
2576 init_rwsem(&fs_info->cleanup_work_sem);
2577 init_rwsem(&fs_info->subvol_sem);
2578 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2580 btrfs_init_dev_replace_locks(fs_info);
2581 btrfs_init_qgroup(fs_info);
2583 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2584 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2586 init_waitqueue_head(&fs_info->transaction_throttle);
2587 init_waitqueue_head(&fs_info->transaction_wait);
2588 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2589 init_waitqueue_head(&fs_info->async_submit_wait);
2591 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2593 ret = btrfs_alloc_stripe_hash_table(fs_info);
2599 __setup_root(4096, 4096, 4096, tree_root,
2600 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2602 invalidate_bdev(fs_devices->latest_bdev);
2605 * Read super block and check the signature bytes only
2607 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2614 * We want to check superblock checksum, the type is stored inside.
2615 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2617 if (btrfs_check_super_csum(bh->b_data)) {
2618 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2625 * super_copy is zeroed at allocation time and we never touch the
2626 * following bytes up to INFO_SIZE, the checksum is calculated from
2627 * the whole block of INFO_SIZE
2629 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2630 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2631 sizeof(*fs_info->super_for_commit));
2634 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2636 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2638 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2643 disk_super = fs_info->super_copy;
2644 if (!btrfs_super_root(disk_super))
2647 /* check FS state, whether FS is broken. */
2648 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2649 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2652 * run through our array of backup supers and setup
2653 * our ring pointer to the oldest one
2655 generation = btrfs_super_generation(disk_super);
2656 find_oldest_super_backup(fs_info, generation);
2659 * In the long term, we'll store the compression type in the super
2660 * block, and it'll be used for per file compression control.
2662 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2664 ret = btrfs_parse_options(tree_root, options);
2670 features = btrfs_super_incompat_flags(disk_super) &
2671 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2673 printk(KERN_ERR "BTRFS: couldn't mount because of "
2674 "unsupported optional features (%Lx).\n",
2681 * Leafsize and nodesize were always equal, this is only a sanity check.
2683 if (le32_to_cpu(disk_super->__unused_leafsize) !=
2684 btrfs_super_nodesize(disk_super)) {
2685 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2686 "blocksizes don't match. node %d leaf %d\n",
2687 btrfs_super_nodesize(disk_super),
2688 le32_to_cpu(disk_super->__unused_leafsize));
2692 if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2693 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2694 "blocksize (%d) was too large\n",
2695 btrfs_super_nodesize(disk_super));
2700 features = btrfs_super_incompat_flags(disk_super);
2701 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2702 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2703 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2705 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2706 printk(KERN_INFO "BTRFS: has skinny extents\n");
2709 * flag our filesystem as having big metadata blocks if
2710 * they are bigger than the page size
2712 if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2713 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2714 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2715 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2718 nodesize = btrfs_super_nodesize(disk_super);
2719 sectorsize = btrfs_super_sectorsize(disk_super);
2720 stripesize = btrfs_super_stripesize(disk_super);
2721 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2722 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2725 * mixed block groups end up with duplicate but slightly offset
2726 * extent buffers for the same range. It leads to corruptions
2728 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2729 (sectorsize != nodesize)) {
2730 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2731 "are not allowed for mixed block groups on %s\n",
2737 * Needn't use the lock because there is no other task which will
2740 btrfs_set_super_incompat_flags(disk_super, features);
2742 features = btrfs_super_compat_ro_flags(disk_super) &
2743 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2744 if (!(sb->s_flags & MS_RDONLY) && features) {
2745 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2746 "unsupported option features (%Lx).\n",
2752 max_active = fs_info->thread_pool_size;
2754 ret = btrfs_init_workqueues(fs_info, fs_devices);
2757 goto fail_sb_buffer;
2760 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2761 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2762 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2764 tree_root->nodesize = nodesize;
2765 tree_root->sectorsize = sectorsize;
2766 tree_root->stripesize = stripesize;
2768 sb->s_blocksize = sectorsize;
2769 sb->s_blocksize_bits = blksize_bits(sectorsize);
2771 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2772 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2773 goto fail_sb_buffer;
2776 if (sectorsize != PAGE_SIZE) {
2777 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2778 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2779 goto fail_sb_buffer;
2782 mutex_lock(&fs_info->chunk_mutex);
2783 ret = btrfs_read_sys_array(tree_root);
2784 mutex_unlock(&fs_info->chunk_mutex);
2786 printk(KERN_ERR "BTRFS: failed to read the system "
2787 "array on %s\n", sb->s_id);
2788 goto fail_sb_buffer;
2791 generation = btrfs_super_chunk_root_generation(disk_super);
2793 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2794 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2796 chunk_root->node = read_tree_block(chunk_root,
2797 btrfs_super_chunk_root(disk_super),
2799 if (IS_ERR(chunk_root->node) ||
2800 !extent_buffer_uptodate(chunk_root->node)) {
2801 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2803 if (!IS_ERR(chunk_root->node))
2804 free_extent_buffer(chunk_root->node);
2805 chunk_root->node = NULL;
2806 goto fail_tree_roots;
2808 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2809 chunk_root->commit_root = btrfs_root_node(chunk_root);
2811 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2812 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2814 ret = btrfs_read_chunk_tree(chunk_root);
2816 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2818 goto fail_tree_roots;
2822 * keep the device that is marked to be the target device for the
2823 * dev_replace procedure
2825 btrfs_close_extra_devices(fs_devices, 0);
2827 if (!fs_devices->latest_bdev) {
2828 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2830 goto fail_tree_roots;
2834 generation = btrfs_super_generation(disk_super);
2836 tree_root->node = read_tree_block(tree_root,
2837 btrfs_super_root(disk_super),
2839 if (IS_ERR(tree_root->node) ||
2840 !extent_buffer_uptodate(tree_root->node)) {
2841 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2843 if (!IS_ERR(tree_root->node))
2844 free_extent_buffer(tree_root->node);
2845 tree_root->node = NULL;
2846 goto recovery_tree_root;
2849 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2850 tree_root->commit_root = btrfs_root_node(tree_root);
2851 btrfs_set_root_refs(&tree_root->root_item, 1);
2853 mutex_lock(&tree_root->objectid_mutex);
2854 ret = btrfs_find_highest_objectid(tree_root,
2855 &tree_root->highest_objectid);
2857 mutex_unlock(&tree_root->objectid_mutex);
2858 goto recovery_tree_root;
2861 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2863 mutex_unlock(&tree_root->objectid_mutex);
2865 ret = btrfs_read_roots(fs_info, tree_root);
2867 goto recovery_tree_root;
2869 fs_info->generation = generation;
2870 fs_info->last_trans_committed = generation;
2872 ret = btrfs_recover_balance(fs_info);
2874 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2875 goto fail_block_groups;
2878 ret = btrfs_init_dev_stats(fs_info);
2880 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2882 goto fail_block_groups;
2885 ret = btrfs_init_dev_replace(fs_info);
2887 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2888 goto fail_block_groups;
2891 btrfs_close_extra_devices(fs_devices, 1);
2893 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2895 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2896 goto fail_block_groups;
2899 ret = btrfs_sysfs_add_device(fs_devices);
2901 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2902 goto fail_fsdev_sysfs;
2905 ret = btrfs_sysfs_add_mounted(fs_info);
2907 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2908 goto fail_fsdev_sysfs;
2911 ret = btrfs_init_space_info(fs_info);
2913 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2917 ret = btrfs_read_block_groups(fs_info->extent_root);
2919 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2922 fs_info->num_tolerated_disk_barrier_failures =
2923 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2924 if (fs_info->fs_devices->missing_devices >
2925 fs_info->num_tolerated_disk_barrier_failures &&
2926 !(sb->s_flags & MS_RDONLY)) {
2927 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2928 fs_info->fs_devices->missing_devices,
2929 fs_info->num_tolerated_disk_barrier_failures);
2933 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2935 if (IS_ERR(fs_info->cleaner_kthread))
2938 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2940 "btrfs-transaction");
2941 if (IS_ERR(fs_info->transaction_kthread))
2944 if (!btrfs_test_opt(tree_root, SSD) &&
2945 !btrfs_test_opt(tree_root, NOSSD) &&
2946 !fs_info->fs_devices->rotating) {
2947 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2949 btrfs_set_opt(fs_info->mount_opt, SSD);
2953 * Mount does not set all options immediatelly, we can do it now and do
2954 * not have to wait for transaction commit
2956 btrfs_apply_pending_changes(fs_info);
2958 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2959 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2960 ret = btrfsic_mount(tree_root, fs_devices,
2961 btrfs_test_opt(tree_root,
2962 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2964 fs_info->check_integrity_print_mask);
2966 printk(KERN_WARNING "BTRFS: failed to initialize"
2967 " integrity check module %s\n", sb->s_id);
2970 ret = btrfs_read_qgroup_config(fs_info);
2972 goto fail_trans_kthread;
2974 /* do not make disk changes in broken FS */
2975 if (btrfs_super_log_root(disk_super) != 0) {
2976 ret = btrfs_replay_log(fs_info, fs_devices);
2983 ret = btrfs_find_orphan_roots(tree_root);
2987 if (!(sb->s_flags & MS_RDONLY)) {
2988 ret = btrfs_cleanup_fs_roots(fs_info);
2992 mutex_lock(&fs_info->cleaner_mutex);
2993 ret = btrfs_recover_relocation(tree_root);
2994 mutex_unlock(&fs_info->cleaner_mutex);
2997 "BTRFS: failed to recover relocation\n");
3003 location.objectid = BTRFS_FS_TREE_OBJECTID;
3004 location.type = BTRFS_ROOT_ITEM_KEY;
3005 location.offset = 0;
3007 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3008 if (IS_ERR(fs_info->fs_root)) {
3009 err = PTR_ERR(fs_info->fs_root);
3013 if (sb->s_flags & MS_RDONLY)
3016 down_read(&fs_info->cleanup_work_sem);
3017 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3018 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3019 up_read(&fs_info->cleanup_work_sem);
3020 close_ctree(tree_root);
3023 up_read(&fs_info->cleanup_work_sem);
3025 ret = btrfs_resume_balance_async(fs_info);
3027 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3028 close_ctree(tree_root);
3032 ret = btrfs_resume_dev_replace_async(fs_info);
3034 pr_warn("BTRFS: failed to resume dev_replace\n");
3035 close_ctree(tree_root);
3039 btrfs_qgroup_rescan_resume(fs_info);
3041 if (!fs_info->uuid_root) {
3042 pr_info("BTRFS: creating UUID tree\n");
3043 ret = btrfs_create_uuid_tree(fs_info);
3045 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3047 close_ctree(tree_root);
3050 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3051 fs_info->generation !=
3052 btrfs_super_uuid_tree_generation(disk_super)) {
3053 pr_info("BTRFS: checking UUID tree\n");
3054 ret = btrfs_check_uuid_tree(fs_info);
3056 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3058 close_ctree(tree_root);
3062 fs_info->update_uuid_tree_gen = 1;
3070 btrfs_free_qgroup_config(fs_info);
3072 kthread_stop(fs_info->transaction_kthread);
3073 btrfs_cleanup_transaction(fs_info->tree_root);
3074 btrfs_free_fs_roots(fs_info);
3076 kthread_stop(fs_info->cleaner_kthread);
3079 * make sure we're done with the btree inode before we stop our
3082 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3085 btrfs_sysfs_remove_mounted(fs_info);
3088 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3091 btrfs_put_block_group_cache(fs_info);
3092 btrfs_free_block_groups(fs_info);
3095 free_root_pointers(fs_info, 1);
3096 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3099 btrfs_stop_all_workers(fs_info);
3102 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3104 iput(fs_info->btree_inode);
3106 percpu_counter_destroy(&fs_info->bio_counter);
3107 fail_delalloc_bytes:
3108 percpu_counter_destroy(&fs_info->delalloc_bytes);
3109 fail_dirty_metadata_bytes:
3110 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3112 bdi_destroy(&fs_info->bdi);
3114 cleanup_srcu_struct(&fs_info->subvol_srcu);
3116 btrfs_free_stripe_hash_table(fs_info);
3117 btrfs_close_devices(fs_info->fs_devices);
3121 if (!btrfs_test_opt(tree_root, RECOVERY))
3122 goto fail_tree_roots;
3124 free_root_pointers(fs_info, 0);
3126 /* don't use the log in recovery mode, it won't be valid */
3127 btrfs_set_super_log_root(disk_super, 0);
3129 /* we can't trust the free space cache either */
3130 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3132 ret = next_root_backup(fs_info, fs_info->super_copy,
3133 &num_backups_tried, &backup_index);
3135 goto fail_block_groups;
3136 goto retry_root_backup;
3139 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3142 set_buffer_uptodate(bh);
3144 struct btrfs_device *device = (struct btrfs_device *)
3147 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3148 "lost page write due to IO error on %s",
3149 rcu_str_deref(device->name));
3150 /* note, we dont' set_buffer_write_io_error because we have
3151 * our own ways of dealing with the IO errors
3153 clear_buffer_uptodate(bh);
3154 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3160 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3161 struct buffer_head **bh_ret)
3163 struct buffer_head *bh;
3164 struct btrfs_super_block *super;
3167 bytenr = btrfs_sb_offset(copy_num);
3168 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3171 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3173 * If we fail to read from the underlying devices, as of now
3174 * the best option we have is to mark it EIO.
3179 super = (struct btrfs_super_block *)bh->b_data;
3180 if (btrfs_super_bytenr(super) != bytenr ||
3181 btrfs_super_magic(super) != BTRFS_MAGIC) {
3191 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3193 struct buffer_head *bh;
3194 struct buffer_head *latest = NULL;
3195 struct btrfs_super_block *super;
3200 /* we would like to check all the supers, but that would make
3201 * a btrfs mount succeed after a mkfs from a different FS.
3202 * So, we need to add a special mount option to scan for
3203 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3205 for (i = 0; i < 1; i++) {
3206 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3210 super = (struct btrfs_super_block *)bh->b_data;
3212 if (!latest || btrfs_super_generation(super) > transid) {
3215 transid = btrfs_super_generation(super);
3222 return ERR_PTR(ret);
3228 * this should be called twice, once with wait == 0 and
3229 * once with wait == 1. When wait == 0 is done, all the buffer heads
3230 * we write are pinned.
3232 * They are released when wait == 1 is done.
3233 * max_mirrors must be the same for both runs, and it indicates how
3234 * many supers on this one device should be written.
3236 * max_mirrors == 0 means to write them all.
3238 static int write_dev_supers(struct btrfs_device *device,
3239 struct btrfs_super_block *sb,
3240 int do_barriers, int wait, int max_mirrors)
3242 struct buffer_head *bh;
3249 if (max_mirrors == 0)
3250 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3252 for (i = 0; i < max_mirrors; i++) {
3253 bytenr = btrfs_sb_offset(i);
3254 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3255 device->commit_total_bytes)
3259 bh = __find_get_block(device->bdev, bytenr / 4096,
3260 BTRFS_SUPER_INFO_SIZE);
3266 if (!buffer_uptodate(bh))
3269 /* drop our reference */
3272 /* drop the reference from the wait == 0 run */
3276 btrfs_set_super_bytenr(sb, bytenr);
3279 crc = btrfs_csum_data((char *)sb +
3280 BTRFS_CSUM_SIZE, crc,
3281 BTRFS_SUPER_INFO_SIZE -
3283 btrfs_csum_final(crc, sb->csum);
3286 * one reference for us, and we leave it for the
3289 bh = __getblk(device->bdev, bytenr / 4096,
3290 BTRFS_SUPER_INFO_SIZE);
3292 btrfs_err(device->dev_root->fs_info,
3293 "couldn't get super buffer head for bytenr %llu",
3299 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3301 /* one reference for submit_bh */
3304 set_buffer_uptodate(bh);
3306 bh->b_end_io = btrfs_end_buffer_write_sync;
3307 bh->b_private = device;
3311 * we fua the first super. The others we allow
3315 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3317 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3321 return errors < i ? 0 : -1;
3325 * endio for the write_dev_flush, this will wake anyone waiting
3326 * for the barrier when it is done
3328 static void btrfs_end_empty_barrier(struct bio *bio)
3330 if (bio->bi_private)
3331 complete(bio->bi_private);
3336 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3337 * sent down. With wait == 1, it waits for the previous flush.
3339 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3342 static int write_dev_flush(struct btrfs_device *device, int wait)
3347 if (device->nobarriers)
3351 bio = device->flush_bio;
3355 wait_for_completion(&device->flush_wait);
3357 if (bio->bi_error) {
3358 ret = bio->bi_error;
3359 btrfs_dev_stat_inc_and_print(device,
3360 BTRFS_DEV_STAT_FLUSH_ERRS);
3363 /* drop the reference from the wait == 0 run */
3365 device->flush_bio = NULL;
3371 * one reference for us, and we leave it for the
3374 device->flush_bio = NULL;
3375 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3379 bio->bi_end_io = btrfs_end_empty_barrier;
3380 bio->bi_bdev = device->bdev;
3381 init_completion(&device->flush_wait);
3382 bio->bi_private = &device->flush_wait;
3383 device->flush_bio = bio;
3386 btrfsic_submit_bio(WRITE_FLUSH, bio);
3392 * send an empty flush down to each device in parallel,
3393 * then wait for them
3395 static int barrier_all_devices(struct btrfs_fs_info *info)
3397 struct list_head *head;
3398 struct btrfs_device *dev;
3399 int errors_send = 0;
3400 int errors_wait = 0;
3403 /* send down all the barriers */
3404 head = &info->fs_devices->devices;
3405 list_for_each_entry_rcu(dev, head, dev_list) {
3412 if (!dev->in_fs_metadata || !dev->writeable)
3415 ret = write_dev_flush(dev, 0);
3420 /* wait for all the barriers */
3421 list_for_each_entry_rcu(dev, head, dev_list) {
3428 if (!dev->in_fs_metadata || !dev->writeable)
3431 ret = write_dev_flush(dev, 1);
3435 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3436 errors_wait > info->num_tolerated_disk_barrier_failures)
3441 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3444 int min_tolerated = INT_MAX;
3446 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3447 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3448 min_tolerated = min(min_tolerated,
3449 btrfs_raid_array[BTRFS_RAID_SINGLE].
3450 tolerated_failures);
3452 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3453 if (raid_type == BTRFS_RAID_SINGLE)
3455 if (!(flags & btrfs_raid_group[raid_type]))
3457 min_tolerated = min(min_tolerated,
3458 btrfs_raid_array[raid_type].
3459 tolerated_failures);
3462 if (min_tolerated == INT_MAX) {
3463 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3467 return min_tolerated;
3470 int btrfs_calc_num_tolerated_disk_barrier_failures(
3471 struct btrfs_fs_info *fs_info)
3473 struct btrfs_ioctl_space_info space;
3474 struct btrfs_space_info *sinfo;
3475 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3476 BTRFS_BLOCK_GROUP_SYSTEM,
3477 BTRFS_BLOCK_GROUP_METADATA,
3478 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3481 int num_tolerated_disk_barrier_failures =
3482 (int)fs_info->fs_devices->num_devices;
3484 for (i = 0; i < ARRAY_SIZE(types); i++) {
3485 struct btrfs_space_info *tmp;
3489 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3490 if (tmp->flags == types[i]) {
3500 down_read(&sinfo->groups_sem);
3501 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3504 if (list_empty(&sinfo->block_groups[c]))
3507 btrfs_get_block_group_info(&sinfo->block_groups[c],
3509 if (space.total_bytes == 0 || space.used_bytes == 0)
3511 flags = space.flags;
3513 num_tolerated_disk_barrier_failures = min(
3514 num_tolerated_disk_barrier_failures,
3515 btrfs_get_num_tolerated_disk_barrier_failures(
3518 up_read(&sinfo->groups_sem);
3521 return num_tolerated_disk_barrier_failures;
3524 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3526 struct list_head *head;
3527 struct btrfs_device *dev;
3528 struct btrfs_super_block *sb;
3529 struct btrfs_dev_item *dev_item;
3533 int total_errors = 0;
3536 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3537 backup_super_roots(root->fs_info);
3539 sb = root->fs_info->super_for_commit;
3540 dev_item = &sb->dev_item;
3542 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3543 head = &root->fs_info->fs_devices->devices;
3544 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3547 ret = barrier_all_devices(root->fs_info);
3550 &root->fs_info->fs_devices->device_list_mutex);
3551 btrfs_std_error(root->fs_info, ret,
3552 "errors while submitting device barriers.");
3557 list_for_each_entry_rcu(dev, head, dev_list) {
3562 if (!dev->in_fs_metadata || !dev->writeable)
3565 btrfs_set_stack_device_generation(dev_item, 0);
3566 btrfs_set_stack_device_type(dev_item, dev->type);
3567 btrfs_set_stack_device_id(dev_item, dev->devid);
3568 btrfs_set_stack_device_total_bytes(dev_item,
3569 dev->commit_total_bytes);
3570 btrfs_set_stack_device_bytes_used(dev_item,
3571 dev->commit_bytes_used);
3572 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3573 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3574 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3575 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3576 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3578 flags = btrfs_super_flags(sb);
3579 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3581 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3585 if (total_errors > max_errors) {
3586 btrfs_err(root->fs_info, "%d errors while writing supers",
3588 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3590 /* FUA is masked off if unsupported and can't be the reason */
3591 btrfs_std_error(root->fs_info, -EIO,
3592 "%d errors while writing supers", total_errors);
3597 list_for_each_entry_rcu(dev, head, dev_list) {
3600 if (!dev->in_fs_metadata || !dev->writeable)
3603 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3607 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3608 if (total_errors > max_errors) {
3609 btrfs_std_error(root->fs_info, -EIO,
3610 "%d errors while writing supers", total_errors);
3616 int write_ctree_super(struct btrfs_trans_handle *trans,
3617 struct btrfs_root *root, int max_mirrors)
3619 return write_all_supers(root, max_mirrors);
3622 /* Drop a fs root from the radix tree and free it. */
3623 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3624 struct btrfs_root *root)
3626 spin_lock(&fs_info->fs_roots_radix_lock);
3627 radix_tree_delete(&fs_info->fs_roots_radix,
3628 (unsigned long)root->root_key.objectid);
3629 spin_unlock(&fs_info->fs_roots_radix_lock);
3631 if (btrfs_root_refs(&root->root_item) == 0)
3632 synchronize_srcu(&fs_info->subvol_srcu);
3634 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3635 btrfs_free_log(NULL, root);
3637 if (root->free_ino_pinned)
3638 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3639 if (root->free_ino_ctl)
3640 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3644 static void free_fs_root(struct btrfs_root *root)
3646 iput(root->ino_cache_inode);
3647 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3648 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3649 root->orphan_block_rsv = NULL;
3651 free_anon_bdev(root->anon_dev);
3652 if (root->subv_writers)
3653 btrfs_free_subvolume_writers(root->subv_writers);
3654 free_extent_buffer(root->node);
3655 free_extent_buffer(root->commit_root);
3656 kfree(root->free_ino_ctl);
3657 kfree(root->free_ino_pinned);
3659 btrfs_put_fs_root(root);
3662 void btrfs_free_fs_root(struct btrfs_root *root)
3667 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3669 u64 root_objectid = 0;
3670 struct btrfs_root *gang[8];
3673 unsigned int ret = 0;
3677 index = srcu_read_lock(&fs_info->subvol_srcu);
3678 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3679 (void **)gang, root_objectid,
3682 srcu_read_unlock(&fs_info->subvol_srcu, index);
3685 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3687 for (i = 0; i < ret; i++) {
3688 /* Avoid to grab roots in dead_roots */
3689 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3693 /* grab all the search result for later use */
3694 gang[i] = btrfs_grab_fs_root(gang[i]);
3696 srcu_read_unlock(&fs_info->subvol_srcu, index);
3698 for (i = 0; i < ret; i++) {
3701 root_objectid = gang[i]->root_key.objectid;
3702 err = btrfs_orphan_cleanup(gang[i]);
3705 btrfs_put_fs_root(gang[i]);
3710 /* release the uncleaned roots due to error */
3711 for (; i < ret; i++) {
3713 btrfs_put_fs_root(gang[i]);
3718 int btrfs_commit_super(struct btrfs_root *root)
3720 struct btrfs_trans_handle *trans;
3722 mutex_lock(&root->fs_info->cleaner_mutex);
3723 btrfs_run_delayed_iputs(root);
3724 mutex_unlock(&root->fs_info->cleaner_mutex);
3725 wake_up_process(root->fs_info->cleaner_kthread);
3727 /* wait until ongoing cleanup work done */
3728 down_write(&root->fs_info->cleanup_work_sem);
3729 up_write(&root->fs_info->cleanup_work_sem);
3731 trans = btrfs_join_transaction(root);
3733 return PTR_ERR(trans);
3734 return btrfs_commit_transaction(trans, root);
3737 void close_ctree(struct btrfs_root *root)
3739 struct btrfs_fs_info *fs_info = root->fs_info;
3742 fs_info->closing = 1;
3745 /* wait for the qgroup rescan worker to stop */
3746 btrfs_qgroup_wait_for_completion(fs_info, false);
3748 /* wait for the uuid_scan task to finish */
3749 down(&fs_info->uuid_tree_rescan_sem);
3750 /* avoid complains from lockdep et al., set sem back to initial state */
3751 up(&fs_info->uuid_tree_rescan_sem);
3753 /* pause restriper - we want to resume on mount */
3754 btrfs_pause_balance(fs_info);
3756 btrfs_dev_replace_suspend_for_unmount(fs_info);
3758 btrfs_scrub_cancel(fs_info);
3760 /* wait for any defraggers to finish */
3761 wait_event(fs_info->transaction_wait,
3762 (atomic_read(&fs_info->defrag_running) == 0));
3764 /* clear out the rbtree of defraggable inodes */
3765 btrfs_cleanup_defrag_inodes(fs_info);
3767 cancel_work_sync(&fs_info->async_reclaim_work);
3769 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3771 * If the cleaner thread is stopped and there are
3772 * block groups queued for removal, the deletion will be
3773 * skipped when we quit the cleaner thread.
3775 btrfs_delete_unused_bgs(root->fs_info);
3777 ret = btrfs_commit_super(root);
3779 btrfs_err(fs_info, "commit super ret %d", ret);
3782 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3783 btrfs_error_commit_super(root);
3785 kthread_stop(fs_info->transaction_kthread);
3786 kthread_stop(fs_info->cleaner_kthread);
3788 fs_info->closing = 2;
3791 btrfs_free_qgroup_config(fs_info);
3793 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3794 btrfs_info(fs_info, "at unmount delalloc count %lld",
3795 percpu_counter_sum(&fs_info->delalloc_bytes));
3798 btrfs_sysfs_remove_mounted(fs_info);
3799 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3801 btrfs_free_fs_roots(fs_info);
3803 btrfs_put_block_group_cache(fs_info);
3805 btrfs_free_block_groups(fs_info);
3808 * we must make sure there is not any read request to
3809 * submit after we stopping all workers.
3811 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3812 btrfs_stop_all_workers(fs_info);
3815 free_root_pointers(fs_info, 1);
3817 iput(fs_info->btree_inode);
3819 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3820 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3821 btrfsic_unmount(root, fs_info->fs_devices);
3824 btrfs_close_devices(fs_info->fs_devices);
3825 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3827 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3828 percpu_counter_destroy(&fs_info->delalloc_bytes);
3829 percpu_counter_destroy(&fs_info->bio_counter);
3830 bdi_destroy(&fs_info->bdi);
3831 cleanup_srcu_struct(&fs_info->subvol_srcu);
3833 btrfs_free_stripe_hash_table(fs_info);
3835 __btrfs_free_block_rsv(root->orphan_block_rsv);
3836 root->orphan_block_rsv = NULL;
3839 while (!list_empty(&fs_info->pinned_chunks)) {
3840 struct extent_map *em;
3842 em = list_first_entry(&fs_info->pinned_chunks,
3843 struct extent_map, list);
3844 list_del_init(&em->list);
3845 free_extent_map(em);
3847 unlock_chunks(root);
3850 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3854 struct inode *btree_inode = buf->pages[0]->mapping->host;
3856 ret = extent_buffer_uptodate(buf);
3860 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3861 parent_transid, atomic);
3867 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3869 return set_extent_buffer_uptodate(buf);
3872 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3874 struct btrfs_root *root;
3875 u64 transid = btrfs_header_generation(buf);
3878 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3880 * This is a fast path so only do this check if we have sanity tests
3881 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3882 * outside of the sanity tests.
3884 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3887 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3888 btrfs_assert_tree_locked(buf);
3889 if (transid != root->fs_info->generation)
3890 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3891 "found %llu running %llu\n",
3892 buf->start, transid, root->fs_info->generation);
3893 was_dirty = set_extent_buffer_dirty(buf);
3895 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3897 root->fs_info->dirty_metadata_batch);
3898 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3900 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3901 * but item data not updated.
3902 * So here we should only check item pointers, not item data.
3904 if (btrfs_header_level(buf) == 0 &&
3905 btrfs_check_leaf_relaxed(root, buf)) {
3906 btrfs_print_leaf(root, buf);
3912 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3916 * looks as though older kernels can get into trouble with
3917 * this code, they end up stuck in balance_dirty_pages forever
3921 if (current->flags & PF_MEMALLOC)
3925 btrfs_balance_delayed_items(root);
3927 ret = __percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3928 BTRFS_DIRTY_METADATA_THRESH,
3929 root->fs_info->dirty_metadata_batch);
3931 balance_dirty_pages_ratelimited(
3932 root->fs_info->btree_inode->i_mapping);
3937 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3939 __btrfs_btree_balance_dirty(root, 1);
3942 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3944 __btrfs_btree_balance_dirty(root, 0);
3947 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3949 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3950 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3953 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3956 struct btrfs_super_block *sb = fs_info->super_copy;
3959 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3960 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3961 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3964 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3965 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3966 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3969 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3970 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3971 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3976 * The common minimum, we don't know if we can trust the nodesize/sectorsize
3977 * items yet, they'll be verified later. Issue just a warning.
3979 if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3980 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3981 btrfs_super_root(sb));
3982 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3983 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3984 btrfs_super_chunk_root(sb));
3985 if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3986 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3987 btrfs_super_log_root(sb));
3990 * Check the lower bound, the alignment and other constraints are
3993 if (btrfs_super_nodesize(sb) < 4096) {
3994 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3995 btrfs_super_nodesize(sb));
3998 if (btrfs_super_sectorsize(sb) < 4096) {
3999 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
4000 btrfs_super_sectorsize(sb));
4004 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4005 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4006 fs_info->fsid, sb->dev_item.fsid);
4011 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4014 if (btrfs_super_num_devices(sb) > (1UL << 31))
4015 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4016 btrfs_super_num_devices(sb));
4017 if (btrfs_super_num_devices(sb) == 0) {
4018 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4022 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4023 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4024 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4029 * Obvious sys_chunk_array corruptions, it must hold at least one key
4032 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4033 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4034 btrfs_super_sys_array_size(sb),
4035 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4038 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4039 + sizeof(struct btrfs_chunk)) {
4040 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4041 btrfs_super_sys_array_size(sb),
4042 sizeof(struct btrfs_disk_key)
4043 + sizeof(struct btrfs_chunk));
4048 * The generation is a global counter, we'll trust it more than the others
4049 * but it's still possible that it's the one that's wrong.
4051 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4053 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4054 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4055 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4056 && btrfs_super_cache_generation(sb) != (u64)-1)
4058 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4059 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4064 static void btrfs_error_commit_super(struct btrfs_root *root)
4066 mutex_lock(&root->fs_info->cleaner_mutex);
4067 btrfs_run_delayed_iputs(root);
4068 mutex_unlock(&root->fs_info->cleaner_mutex);
4070 down_write(&root->fs_info->cleanup_work_sem);
4071 up_write(&root->fs_info->cleanup_work_sem);
4073 /* cleanup FS via transaction */
4074 btrfs_cleanup_transaction(root);
4077 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4079 struct btrfs_ordered_extent *ordered;
4081 spin_lock(&root->ordered_extent_lock);
4083 * This will just short circuit the ordered completion stuff which will
4084 * make sure the ordered extent gets properly cleaned up.
4086 list_for_each_entry(ordered, &root->ordered_extents,
4088 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4089 spin_unlock(&root->ordered_extent_lock);
4092 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4094 struct btrfs_root *root;
4095 struct list_head splice;
4097 INIT_LIST_HEAD(&splice);
4099 spin_lock(&fs_info->ordered_root_lock);
4100 list_splice_init(&fs_info->ordered_roots, &splice);
4101 while (!list_empty(&splice)) {
4102 root = list_first_entry(&splice, struct btrfs_root,
4104 list_move_tail(&root->ordered_root,
4105 &fs_info->ordered_roots);
4107 spin_unlock(&fs_info->ordered_root_lock);
4108 btrfs_destroy_ordered_extents(root);
4111 spin_lock(&fs_info->ordered_root_lock);
4113 spin_unlock(&fs_info->ordered_root_lock);
4116 * We need this here because if we've been flipped read-only we won't
4117 * get sync() from the umount, so we need to make sure any ordered
4118 * extents that haven't had their dirty pages IO start writeout yet
4119 * actually get run and error out properly.
4121 btrfs_wait_ordered_roots(fs_info, -1);
4124 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4125 struct btrfs_root *root)
4127 struct rb_node *node;
4128 struct btrfs_delayed_ref_root *delayed_refs;
4129 struct btrfs_delayed_ref_node *ref;
4132 delayed_refs = &trans->delayed_refs;
4134 spin_lock(&delayed_refs->lock);
4135 if (atomic_read(&delayed_refs->num_entries) == 0) {
4136 spin_unlock(&delayed_refs->lock);
4137 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4141 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4142 struct btrfs_delayed_ref_head *head;
4143 struct btrfs_delayed_ref_node *tmp;
4144 bool pin_bytes = false;
4146 head = rb_entry(node, struct btrfs_delayed_ref_head,
4148 if (!mutex_trylock(&head->mutex)) {
4149 atomic_inc(&head->node.refs);
4150 spin_unlock(&delayed_refs->lock);
4152 mutex_lock(&head->mutex);
4153 mutex_unlock(&head->mutex);
4154 btrfs_put_delayed_ref(&head->node);
4155 spin_lock(&delayed_refs->lock);
4158 spin_lock(&head->lock);
4159 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4162 list_del(&ref->list);
4163 atomic_dec(&delayed_refs->num_entries);
4164 btrfs_put_delayed_ref(ref);
4166 if (head->must_insert_reserved)
4168 btrfs_free_delayed_extent_op(head->extent_op);
4169 delayed_refs->num_heads--;
4170 if (head->processing == 0)
4171 delayed_refs->num_heads_ready--;
4172 atomic_dec(&delayed_refs->num_entries);
4173 head->node.in_tree = 0;
4174 rb_erase(&head->href_node, &delayed_refs->href_root);
4175 spin_unlock(&head->lock);
4176 spin_unlock(&delayed_refs->lock);
4177 mutex_unlock(&head->mutex);
4180 btrfs_pin_extent(root, head->node.bytenr,
4181 head->node.num_bytes, 1);
4182 btrfs_put_delayed_ref(&head->node);
4184 spin_lock(&delayed_refs->lock);
4187 spin_unlock(&delayed_refs->lock);
4192 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4194 struct btrfs_inode *btrfs_inode;
4195 struct list_head splice;
4197 INIT_LIST_HEAD(&splice);
4199 spin_lock(&root->delalloc_lock);
4200 list_splice_init(&root->delalloc_inodes, &splice);
4202 while (!list_empty(&splice)) {
4203 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4206 list_del_init(&btrfs_inode->delalloc_inodes);
4207 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4208 &btrfs_inode->runtime_flags);
4209 spin_unlock(&root->delalloc_lock);
4211 btrfs_invalidate_inodes(btrfs_inode->root);
4213 spin_lock(&root->delalloc_lock);
4216 spin_unlock(&root->delalloc_lock);
4219 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4221 struct btrfs_root *root;
4222 struct list_head splice;
4224 INIT_LIST_HEAD(&splice);
4226 spin_lock(&fs_info->delalloc_root_lock);
4227 list_splice_init(&fs_info->delalloc_roots, &splice);
4228 while (!list_empty(&splice)) {
4229 root = list_first_entry(&splice, struct btrfs_root,
4231 list_del_init(&root->delalloc_root);
4232 root = btrfs_grab_fs_root(root);
4234 spin_unlock(&fs_info->delalloc_root_lock);
4236 btrfs_destroy_delalloc_inodes(root);
4237 btrfs_put_fs_root(root);
4239 spin_lock(&fs_info->delalloc_root_lock);
4241 spin_unlock(&fs_info->delalloc_root_lock);
4244 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4245 struct extent_io_tree *dirty_pages,
4249 struct extent_buffer *eb;
4254 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4259 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4260 while (start <= end) {
4261 eb = btrfs_find_tree_block(root->fs_info, start);
4262 start += root->nodesize;
4265 wait_on_extent_buffer_writeback(eb);
4267 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4269 clear_extent_buffer_dirty(eb);
4270 free_extent_buffer_stale(eb);
4277 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4278 struct extent_io_tree *pinned_extents)
4280 struct btrfs_fs_info *fs_info = root->fs_info;
4281 struct extent_io_tree *unpin;
4287 unpin = pinned_extents;
4291 * The btrfs_finish_extent_commit() may get the same range as
4292 * ours between find_first_extent_bit and clear_extent_dirty.
4293 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4294 * the same extent range.
4296 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4297 ret = find_first_extent_bit(unpin, 0, &start, &end,
4298 EXTENT_DIRTY, NULL);
4300 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4304 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4305 btrfs_error_unpin_extent_range(root, start, end);
4306 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4311 if (unpin == &fs_info->freed_extents[0])
4312 unpin = &fs_info->freed_extents[1];
4314 unpin = &fs_info->freed_extents[0];
4322 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4323 struct btrfs_root *root)
4325 btrfs_destroy_delayed_refs(cur_trans, root);
4327 cur_trans->state = TRANS_STATE_COMMIT_START;
4328 wake_up(&root->fs_info->transaction_blocked_wait);
4330 cur_trans->state = TRANS_STATE_UNBLOCKED;
4331 wake_up(&root->fs_info->transaction_wait);
4333 btrfs_destroy_delayed_inodes(root);
4334 btrfs_assert_delayed_root_empty(root);
4336 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4338 btrfs_destroy_pinned_extent(root,
4339 root->fs_info->pinned_extents);
4341 cur_trans->state =TRANS_STATE_COMPLETED;
4342 wake_up(&cur_trans->commit_wait);
4345 memset(cur_trans, 0, sizeof(*cur_trans));
4346 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4350 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4352 struct btrfs_transaction *t;
4354 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4356 spin_lock(&root->fs_info->trans_lock);
4357 while (!list_empty(&root->fs_info->trans_list)) {
4358 t = list_first_entry(&root->fs_info->trans_list,
4359 struct btrfs_transaction, list);
4360 if (t->state >= TRANS_STATE_COMMIT_START) {
4361 atomic_inc(&t->use_count);
4362 spin_unlock(&root->fs_info->trans_lock);
4363 btrfs_wait_for_commit(root, t->transid);
4364 btrfs_put_transaction(t);
4365 spin_lock(&root->fs_info->trans_lock);
4368 if (t == root->fs_info->running_transaction) {
4369 t->state = TRANS_STATE_COMMIT_DOING;
4370 spin_unlock(&root->fs_info->trans_lock);
4372 * We wait for 0 num_writers since we don't hold a trans
4373 * handle open currently for this transaction.
4375 wait_event(t->writer_wait,
4376 atomic_read(&t->num_writers) == 0);
4378 spin_unlock(&root->fs_info->trans_lock);
4380 btrfs_cleanup_one_transaction(t, root);
4382 spin_lock(&root->fs_info->trans_lock);
4383 if (t == root->fs_info->running_transaction)
4384 root->fs_info->running_transaction = NULL;
4385 list_del_init(&t->list);
4386 spin_unlock(&root->fs_info->trans_lock);
4388 btrfs_put_transaction(t);
4389 trace_btrfs_transaction_commit(root);
4390 spin_lock(&root->fs_info->trans_lock);
4392 spin_unlock(&root->fs_info->trans_lock);
4393 btrfs_destroy_all_ordered_extents(root->fs_info);
4394 btrfs_destroy_delayed_inodes(root);
4395 btrfs_assert_delayed_root_empty(root);
4396 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4397 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4398 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4403 static const struct extent_io_ops btree_extent_io_ops = {
4404 .readpage_end_io_hook = btree_readpage_end_io_hook,
4405 .readpage_io_failed_hook = btree_io_failed_hook,
4406 .submit_bio_hook = btree_submit_bio_hook,
4407 /* note we're sharing with inode.c for the merge bio hook */
4408 .merge_bio_hook = btrfs_merge_bio_hook,