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"
42 #include "async-thread.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
55 #include <asm/cpufeature.h>
58 static 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_operations(struct btrfs_transaction *t,
64 struct btrfs_root *root);
65 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
66 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
67 struct btrfs_root *root);
68 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
69 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
72 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
74 static int btrfs_cleanup_transaction(struct btrfs_root *root);
75 static void btrfs_error_commit_super(struct btrfs_root *root);
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_fs_info *info;
89 struct list_head list;
90 struct btrfs_work work;
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
98 struct async_submit_bio {
101 struct list_head list;
102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
106 unsigned long bio_flags;
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
112 struct btrfs_work work;
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
140 # if BTRFS_MAX_LEVEL != 8
144 static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149 } btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
160 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
161 { .id = 0, .name_stem = "tree" },
164 void __init btrfs_init_lockdep(void)
168 /* initialize lockdep class names */
169 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
170 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
172 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
173 snprintf(ks->names[j], sizeof(ks->names[j]),
174 "btrfs-%s-%02d", ks->name_stem, j);
178 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
181 struct btrfs_lockdep_keyset *ks;
183 BUG_ON(level >= ARRAY_SIZE(ks->keys));
185 /* find the matching keyset, id 0 is the default entry */
186 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
187 if (ks->id == objectid)
190 lockdep_set_class_and_name(&eb->lock,
191 &ks->keys[level], ks->names[level]);
197 * extents on the btree inode are pretty simple, there's one extent
198 * that covers the entire device
200 static struct extent_map *btree_get_extent(struct inode *inode,
201 struct page *page, size_t pg_offset, u64 start, u64 len,
204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
205 struct extent_map *em;
208 read_lock(&em_tree->lock);
209 em = lookup_extent_mapping(em_tree, start, len);
212 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
213 read_unlock(&em_tree->lock);
216 read_unlock(&em_tree->lock);
218 em = alloc_extent_map();
220 em = ERR_PTR(-ENOMEM);
225 em->block_len = (u64)-1;
227 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
229 write_lock(&em_tree->lock);
230 ret = add_extent_mapping(em_tree, em, 0);
231 if (ret == -EEXIST) {
233 em = lookup_extent_mapping(em_tree, start, len);
240 write_unlock(&em_tree->lock);
246 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
248 return btrfs_crc32c(seed, data, len);
251 void btrfs_csum_final(u32 crc, char *result)
253 put_unaligned_le32(~crc, result);
257 * compute the csum for a btree block, and either verify it or write it
258 * into the csum field of the block.
260 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
263 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
266 unsigned long cur_len;
267 unsigned long offset = BTRFS_CSUM_SIZE;
269 unsigned long map_start;
270 unsigned long map_len;
273 unsigned long inline_result;
275 len = buf->len - offset;
277 err = map_private_extent_buffer(buf, offset, 32,
278 &kaddr, &map_start, &map_len);
281 cur_len = min(len, map_len - (offset - map_start));
282 crc = btrfs_csum_data(kaddr + offset - map_start,
287 if (csum_size > sizeof(inline_result)) {
288 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
292 result = (char *)&inline_result;
295 btrfs_csum_final(crc, result);
298 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
301 memcpy(&found, result, csum_size);
303 read_extent_buffer(buf, &val, 0, csum_size);
304 printk_ratelimited(KERN_INFO
305 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
307 root->fs_info->sb->s_id, buf->start,
308 val, found, btrfs_header_level(buf));
309 if (result != (char *)&inline_result)
314 write_extent_buffer(buf, result, 0, csum_size);
316 if (result != (char *)&inline_result)
322 * we can't consider a given block up to date unless the transid of the
323 * block matches the transid in the parent node's pointer. This is how we
324 * detect blocks that either didn't get written at all or got written
325 * in the wrong place.
327 static int verify_parent_transid(struct extent_io_tree *io_tree,
328 struct extent_buffer *eb, u64 parent_transid,
331 struct extent_state *cached_state = NULL;
333 bool need_lock = (current->journal_info ==
334 (void *)BTRFS_SEND_TRANS_STUB);
336 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
343 btrfs_tree_read_lock(eb);
344 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
347 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
349 if (extent_buffer_uptodate(eb) &&
350 btrfs_header_generation(eb) == parent_transid) {
354 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
356 eb->start, parent_transid, btrfs_header_generation(eb));
360 * Things reading via commit roots that don't have normal protection,
361 * like send, can have a really old block in cache that may point at a
362 * block that has been free'd and re-allocated. So don't clear uptodate
363 * if we find an eb that is under IO (dirty/writeback) because we could
364 * end up reading in the stale data and then writing it back out and
365 * making everybody very sad.
367 if (!extent_buffer_under_io(eb))
368 clear_extent_buffer_uptodate(eb);
370 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
371 &cached_state, GFP_NOFS);
372 btrfs_tree_read_unlock_blocking(eb);
377 * Return 0 if the superblock checksum type matches the checksum value of that
378 * algorithm. Pass the raw disk superblock data.
380 static int btrfs_check_super_csum(char *raw_disk_sb)
382 struct btrfs_super_block *disk_sb =
383 (struct btrfs_super_block *)raw_disk_sb;
384 u16 csum_type = btrfs_super_csum_type(disk_sb);
387 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
389 const int csum_size = sizeof(crc);
390 char result[csum_size];
393 * The super_block structure does not span the whole
394 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
395 * is filled with zeros and is included in the checkum.
397 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
398 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
399 btrfs_csum_final(crc, result);
401 if (memcmp(raw_disk_sb, result, csum_size))
404 if (ret && btrfs_super_generation(disk_sb) < 10) {
406 "BTRFS: super block crcs don't match, older mkfs detected\n");
411 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
412 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
421 * helper to read a given tree block, doing retries as required when
422 * the checksums don't match and we have alternate mirrors to try.
424 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
425 struct extent_buffer *eb,
426 u64 start, u64 parent_transid)
428 struct extent_io_tree *io_tree;
433 int failed_mirror = 0;
435 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
436 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
438 ret = read_extent_buffer_pages(io_tree, eb, start,
440 btree_get_extent, mirror_num);
442 if (!verify_parent_transid(io_tree, eb,
450 * This buffer's crc is fine, but its contents are corrupted, so
451 * there is no reason to read the other copies, they won't be
454 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
457 num_copies = btrfs_num_copies(root->fs_info,
462 if (!failed_mirror) {
464 failed_mirror = eb->read_mirror;
468 if (mirror_num == failed_mirror)
471 if (mirror_num > num_copies)
475 if (failed && !ret && failed_mirror)
476 repair_eb_io_failure(root, eb, failed_mirror);
482 * checksum a dirty tree block before IO. This has extra checks to make sure
483 * we only fill in the checksum field in the first page of a multi-page block
486 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
488 u64 start = page_offset(page);
490 struct extent_buffer *eb;
492 eb = (struct extent_buffer *)page->private;
493 if (page != eb->pages[0])
495 found_start = btrfs_header_bytenr(eb);
496 if (WARN_ON(found_start != start || !PageUptodate(page)))
498 csum_tree_block(root, eb, 0);
502 static int check_tree_block_fsid(struct btrfs_root *root,
503 struct extent_buffer *eb)
505 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
506 u8 fsid[BTRFS_UUID_SIZE];
509 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
511 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
515 fs_devices = fs_devices->seed;
520 #define CORRUPT(reason, eb, root, slot) \
521 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
522 "root=%llu, slot=%d", reason, \
523 btrfs_header_bytenr(eb), root->objectid, slot)
525 static noinline int check_leaf(struct btrfs_root *root,
526 struct extent_buffer *leaf)
528 struct btrfs_key key;
529 struct btrfs_key leaf_key;
530 u32 nritems = btrfs_header_nritems(leaf);
536 /* Check the 0 item */
537 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
538 BTRFS_LEAF_DATA_SIZE(root)) {
539 CORRUPT("invalid item offset size pair", leaf, root, 0);
544 * Check to make sure each items keys are in the correct order and their
545 * offsets make sense. We only have to loop through nritems-1 because
546 * we check the current slot against the next slot, which verifies the
547 * next slot's offset+size makes sense and that the current's slot
550 for (slot = 0; slot < nritems - 1; slot++) {
551 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
552 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
554 /* Make sure the keys are in the right order */
555 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
556 CORRUPT("bad key order", leaf, root, slot);
561 * Make sure the offset and ends are right, remember that the
562 * item data starts at the end of the leaf and grows towards the
565 if (btrfs_item_offset_nr(leaf, slot) !=
566 btrfs_item_end_nr(leaf, slot + 1)) {
567 CORRUPT("slot offset bad", leaf, root, slot);
572 * Check to make sure that we don't point outside of the leaf,
573 * just incase all the items are consistent to eachother, but
574 * all point outside of the leaf.
576 if (btrfs_item_end_nr(leaf, slot) >
577 BTRFS_LEAF_DATA_SIZE(root)) {
578 CORRUPT("slot end outside of leaf", leaf, root, slot);
586 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
587 u64 phy_offset, struct page *page,
588 u64 start, u64 end, int mirror)
592 struct extent_buffer *eb;
593 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
600 eb = (struct extent_buffer *)page->private;
602 /* the pending IO might have been the only thing that kept this buffer
603 * in memory. Make sure we have a ref for all this other checks
605 extent_buffer_get(eb);
607 reads_done = atomic_dec_and_test(&eb->io_pages);
611 eb->read_mirror = mirror;
612 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
617 found_start = btrfs_header_bytenr(eb);
618 if (found_start != eb->start) {
619 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
621 found_start, eb->start);
625 if (check_tree_block_fsid(root, eb)) {
626 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
631 found_level = btrfs_header_level(eb);
632 if (found_level >= BTRFS_MAX_LEVEL) {
633 btrfs_info(root->fs_info, "bad tree block level %d",
634 (int)btrfs_header_level(eb));
639 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
642 ret = csum_tree_block(root, eb, 1);
649 * If this is a leaf block and it is corrupt, set the corrupt bit so
650 * that we don't try and read the other copies of this block, just
653 if (found_level == 0 && check_leaf(root, eb)) {
654 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
659 set_extent_buffer_uptodate(eb);
662 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
663 btree_readahead_hook(root, eb, eb->start, ret);
667 * our io error hook is going to dec the io pages
668 * again, we have to make sure it has something
671 atomic_inc(&eb->io_pages);
672 clear_extent_buffer_uptodate(eb);
674 free_extent_buffer(eb);
679 static int btree_io_failed_hook(struct page *page, int failed_mirror)
681 struct extent_buffer *eb;
682 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
684 eb = (struct extent_buffer *)page->private;
685 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
686 eb->read_mirror = failed_mirror;
687 atomic_dec(&eb->io_pages);
688 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
689 btree_readahead_hook(root, eb, eb->start, -EIO);
690 return -EIO; /* we fixed nothing */
693 static void end_workqueue_bio(struct bio *bio, int err)
695 struct end_io_wq *end_io_wq = bio->bi_private;
696 struct btrfs_fs_info *fs_info;
698 fs_info = end_io_wq->info;
699 end_io_wq->error = err;
700 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
702 if (bio->bi_rw & REQ_WRITE) {
703 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
704 btrfs_queue_work(fs_info->endio_meta_write_workers,
706 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
707 btrfs_queue_work(fs_info->endio_freespace_worker,
709 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
710 btrfs_queue_work(fs_info->endio_raid56_workers,
713 btrfs_queue_work(fs_info->endio_write_workers,
716 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
717 btrfs_queue_work(fs_info->endio_raid56_workers,
719 else if (end_io_wq->metadata)
720 btrfs_queue_work(fs_info->endio_meta_workers,
723 btrfs_queue_work(fs_info->endio_workers,
729 * For the metadata arg you want
732 * 1 - if normal metadta
733 * 2 - if writing to the free space cache area
734 * 3 - raid parity work
736 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
739 struct end_io_wq *end_io_wq;
740 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
744 end_io_wq->private = bio->bi_private;
745 end_io_wq->end_io = bio->bi_end_io;
746 end_io_wq->info = info;
747 end_io_wq->error = 0;
748 end_io_wq->bio = bio;
749 end_io_wq->metadata = metadata;
751 bio->bi_private = end_io_wq;
752 bio->bi_end_io = end_workqueue_bio;
756 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
758 unsigned long limit = min_t(unsigned long,
759 info->thread_pool_size,
760 info->fs_devices->open_devices);
764 static void run_one_async_start(struct btrfs_work *work)
766 struct async_submit_bio *async;
769 async = container_of(work, struct async_submit_bio, work);
770 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
771 async->mirror_num, async->bio_flags,
777 static void run_one_async_done(struct btrfs_work *work)
779 struct btrfs_fs_info *fs_info;
780 struct async_submit_bio *async;
783 async = container_of(work, struct async_submit_bio, work);
784 fs_info = BTRFS_I(async->inode)->root->fs_info;
786 limit = btrfs_async_submit_limit(fs_info);
787 limit = limit * 2 / 3;
789 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
790 waitqueue_active(&fs_info->async_submit_wait))
791 wake_up(&fs_info->async_submit_wait);
793 /* If an error occured we just want to clean up the bio and move on */
795 bio_endio(async->bio, async->error);
799 async->submit_bio_done(async->inode, async->rw, async->bio,
800 async->mirror_num, async->bio_flags,
804 static void run_one_async_free(struct btrfs_work *work)
806 struct async_submit_bio *async;
808 async = container_of(work, struct async_submit_bio, work);
812 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
813 int rw, struct bio *bio, int mirror_num,
814 unsigned long bio_flags,
816 extent_submit_bio_hook_t *submit_bio_start,
817 extent_submit_bio_hook_t *submit_bio_done)
819 struct async_submit_bio *async;
821 async = kmalloc(sizeof(*async), GFP_NOFS);
825 async->inode = inode;
828 async->mirror_num = mirror_num;
829 async->submit_bio_start = submit_bio_start;
830 async->submit_bio_done = submit_bio_done;
832 btrfs_init_work(&async->work, run_one_async_start,
833 run_one_async_done, run_one_async_free);
835 async->bio_flags = bio_flags;
836 async->bio_offset = bio_offset;
840 atomic_inc(&fs_info->nr_async_submits);
843 btrfs_set_work_high_priority(&async->work);
845 btrfs_queue_work(fs_info->workers, &async->work);
847 while (atomic_read(&fs_info->async_submit_draining) &&
848 atomic_read(&fs_info->nr_async_submits)) {
849 wait_event(fs_info->async_submit_wait,
850 (atomic_read(&fs_info->nr_async_submits) == 0));
856 static int btree_csum_one_bio(struct bio *bio)
858 struct bio_vec *bvec;
859 struct btrfs_root *root;
862 bio_for_each_segment_all(bvec, bio, i) {
863 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
864 ret = csum_dirty_buffer(root, bvec->bv_page);
872 static int __btree_submit_bio_start(struct inode *inode, int rw,
873 struct bio *bio, int mirror_num,
874 unsigned long bio_flags,
878 * when we're called for a write, we're already in the async
879 * submission context. Just jump into btrfs_map_bio
881 return btree_csum_one_bio(bio);
884 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
885 int mirror_num, unsigned long bio_flags,
891 * when we're called for a write, we're already in the async
892 * submission context. Just jump into btrfs_map_bio
894 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
900 static int check_async_write(struct inode *inode, unsigned long bio_flags)
902 if (bio_flags & EXTENT_BIO_TREE_LOG)
911 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
912 int mirror_num, unsigned long bio_flags,
915 int async = check_async_write(inode, bio_flags);
918 if (!(rw & REQ_WRITE)) {
920 * called for a read, do the setup so that checksum validation
921 * can happen in the async kernel threads
923 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
927 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
930 ret = btree_csum_one_bio(bio);
933 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
937 * kthread helpers are used to submit writes so that
938 * checksumming can happen in parallel across all CPUs
940 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
941 inode, rw, bio, mirror_num, 0,
943 __btree_submit_bio_start,
944 __btree_submit_bio_done);
954 #ifdef CONFIG_MIGRATION
955 static int btree_migratepage(struct address_space *mapping,
956 struct page *newpage, struct page *page,
957 enum migrate_mode mode)
960 * we can't safely write a btree page from here,
961 * we haven't done the locking hook
966 * Buffers may be managed in a filesystem specific way.
967 * We must have no buffers or drop them.
969 if (page_has_private(page) &&
970 !try_to_release_page(page, GFP_KERNEL))
972 return migrate_page(mapping, newpage, page, mode);
977 static int btree_writepages(struct address_space *mapping,
978 struct writeback_control *wbc)
980 struct btrfs_fs_info *fs_info;
983 if (wbc->sync_mode == WB_SYNC_NONE) {
985 if (wbc->for_kupdate)
988 fs_info = BTRFS_I(mapping->host)->root->fs_info;
989 /* this is a bit racy, but that's ok */
990 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
991 BTRFS_DIRTY_METADATA_THRESH);
995 return btree_write_cache_pages(mapping, wbc);
998 static int btree_readpage(struct file *file, struct page *page)
1000 struct extent_io_tree *tree;
1001 tree = &BTRFS_I(page->mapping->host)->io_tree;
1002 return extent_read_full_page(tree, page, btree_get_extent, 0);
1005 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1007 if (PageWriteback(page) || PageDirty(page))
1010 return try_release_extent_buffer(page);
1013 static void btree_invalidatepage(struct page *page, unsigned int offset,
1014 unsigned int length)
1016 struct extent_io_tree *tree;
1017 tree = &BTRFS_I(page->mapping->host)->io_tree;
1018 extent_invalidatepage(tree, page, offset);
1019 btree_releasepage(page, GFP_NOFS);
1020 if (PagePrivate(page)) {
1021 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1022 "page private not zero on page %llu",
1023 (unsigned long long)page_offset(page));
1024 ClearPagePrivate(page);
1025 set_page_private(page, 0);
1026 page_cache_release(page);
1030 static int btree_set_page_dirty(struct page *page)
1033 struct extent_buffer *eb;
1035 BUG_ON(!PagePrivate(page));
1036 eb = (struct extent_buffer *)page->private;
1038 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1039 BUG_ON(!atomic_read(&eb->refs));
1040 btrfs_assert_tree_locked(eb);
1042 return __set_page_dirty_nobuffers(page);
1045 static const struct address_space_operations btree_aops = {
1046 .readpage = btree_readpage,
1047 .writepages = btree_writepages,
1048 .releasepage = btree_releasepage,
1049 .invalidatepage = btree_invalidatepage,
1050 #ifdef CONFIG_MIGRATION
1051 .migratepage = btree_migratepage,
1053 .set_page_dirty = btree_set_page_dirty,
1056 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1059 struct extent_buffer *buf = NULL;
1060 struct inode *btree_inode = root->fs_info->btree_inode;
1063 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1066 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1067 buf, 0, WAIT_NONE, btree_get_extent, 0);
1068 free_extent_buffer(buf);
1072 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1073 int mirror_num, struct extent_buffer **eb)
1075 struct extent_buffer *buf = NULL;
1076 struct inode *btree_inode = root->fs_info->btree_inode;
1077 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1080 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1084 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1086 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1087 btree_get_extent, mirror_num);
1089 free_extent_buffer(buf);
1093 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1094 free_extent_buffer(buf);
1096 } else if (extent_buffer_uptodate(buf)) {
1099 free_extent_buffer(buf);
1104 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1105 u64 bytenr, u32 blocksize)
1107 return find_extent_buffer(root->fs_info, bytenr);
1110 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1111 u64 bytenr, u32 blocksize)
1113 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1117 int btrfs_write_tree_block(struct extent_buffer *buf)
1119 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1120 buf->start + buf->len - 1);
1123 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1125 return filemap_fdatawait_range(buf->pages[0]->mapping,
1126 buf->start, buf->start + buf->len - 1);
1129 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1130 u32 blocksize, u64 parent_transid)
1132 struct extent_buffer *buf = NULL;
1135 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1139 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1141 free_extent_buffer(buf);
1148 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1149 struct extent_buffer *buf)
1151 struct btrfs_fs_info *fs_info = root->fs_info;
1153 if (btrfs_header_generation(buf) ==
1154 fs_info->running_transaction->transid) {
1155 btrfs_assert_tree_locked(buf);
1157 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1158 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1160 fs_info->dirty_metadata_batch);
1161 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1162 btrfs_set_lock_blocking(buf);
1163 clear_extent_buffer_dirty(buf);
1168 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1170 struct btrfs_subvolume_writers *writers;
1173 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1175 return ERR_PTR(-ENOMEM);
1177 ret = percpu_counter_init(&writers->counter, 0);
1180 return ERR_PTR(ret);
1183 init_waitqueue_head(&writers->wait);
1188 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1190 percpu_counter_destroy(&writers->counter);
1194 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1195 u32 stripesize, struct btrfs_root *root,
1196 struct btrfs_fs_info *fs_info,
1200 root->commit_root = NULL;
1201 root->sectorsize = sectorsize;
1202 root->nodesize = nodesize;
1203 root->leafsize = leafsize;
1204 root->stripesize = stripesize;
1206 root->orphan_cleanup_state = 0;
1208 root->objectid = objectid;
1209 root->last_trans = 0;
1210 root->highest_objectid = 0;
1211 root->nr_delalloc_inodes = 0;
1212 root->nr_ordered_extents = 0;
1214 root->inode_tree = RB_ROOT;
1215 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1216 root->block_rsv = NULL;
1217 root->orphan_block_rsv = NULL;
1219 INIT_LIST_HEAD(&root->dirty_list);
1220 INIT_LIST_HEAD(&root->root_list);
1221 INIT_LIST_HEAD(&root->delalloc_inodes);
1222 INIT_LIST_HEAD(&root->delalloc_root);
1223 INIT_LIST_HEAD(&root->ordered_extents);
1224 INIT_LIST_HEAD(&root->ordered_root);
1225 INIT_LIST_HEAD(&root->logged_list[0]);
1226 INIT_LIST_HEAD(&root->logged_list[1]);
1227 spin_lock_init(&root->orphan_lock);
1228 spin_lock_init(&root->inode_lock);
1229 spin_lock_init(&root->delalloc_lock);
1230 spin_lock_init(&root->ordered_extent_lock);
1231 spin_lock_init(&root->accounting_lock);
1232 spin_lock_init(&root->log_extents_lock[0]);
1233 spin_lock_init(&root->log_extents_lock[1]);
1234 mutex_init(&root->objectid_mutex);
1235 mutex_init(&root->log_mutex);
1236 mutex_init(&root->ordered_extent_mutex);
1237 mutex_init(&root->delalloc_mutex);
1238 init_waitqueue_head(&root->log_writer_wait);
1239 init_waitqueue_head(&root->log_commit_wait[0]);
1240 init_waitqueue_head(&root->log_commit_wait[1]);
1241 INIT_LIST_HEAD(&root->log_ctxs[0]);
1242 INIT_LIST_HEAD(&root->log_ctxs[1]);
1243 atomic_set(&root->log_commit[0], 0);
1244 atomic_set(&root->log_commit[1], 0);
1245 atomic_set(&root->log_writers, 0);
1246 atomic_set(&root->log_batch, 0);
1247 atomic_set(&root->orphan_inodes, 0);
1248 atomic_set(&root->refs, 1);
1249 atomic_set(&root->will_be_snapshoted, 0);
1250 root->log_transid = 0;
1251 root->log_transid_committed = -1;
1252 root->last_log_commit = 0;
1254 extent_io_tree_init(&root->dirty_log_pages,
1255 fs_info->btree_inode->i_mapping);
1257 memset(&root->root_key, 0, sizeof(root->root_key));
1258 memset(&root->root_item, 0, sizeof(root->root_item));
1259 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1260 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1262 root->defrag_trans_start = fs_info->generation;
1264 root->defrag_trans_start = 0;
1265 init_completion(&root->kobj_unregister);
1266 root->root_key.objectid = objectid;
1269 spin_lock_init(&root->root_item_lock);
1272 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1274 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1276 root->fs_info = fs_info;
1280 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1281 /* Should only be used by the testing infrastructure */
1282 struct btrfs_root *btrfs_alloc_dummy_root(void)
1284 struct btrfs_root *root;
1286 root = btrfs_alloc_root(NULL);
1288 return ERR_PTR(-ENOMEM);
1289 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1290 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1296 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1297 struct btrfs_fs_info *fs_info,
1300 struct extent_buffer *leaf;
1301 struct btrfs_root *tree_root = fs_info->tree_root;
1302 struct btrfs_root *root;
1303 struct btrfs_key key;
1307 root = btrfs_alloc_root(fs_info);
1309 return ERR_PTR(-ENOMEM);
1311 __setup_root(tree_root->nodesize, tree_root->leafsize,
1312 tree_root->sectorsize, tree_root->stripesize,
1313 root, fs_info, objectid);
1314 root->root_key.objectid = objectid;
1315 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1316 root->root_key.offset = 0;
1318 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1319 0, objectid, NULL, 0, 0, 0);
1321 ret = PTR_ERR(leaf);
1326 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1327 btrfs_set_header_bytenr(leaf, leaf->start);
1328 btrfs_set_header_generation(leaf, trans->transid);
1329 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1330 btrfs_set_header_owner(leaf, objectid);
1333 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1335 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1336 btrfs_header_chunk_tree_uuid(leaf),
1338 btrfs_mark_buffer_dirty(leaf);
1340 root->commit_root = btrfs_root_node(root);
1341 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1343 root->root_item.flags = 0;
1344 root->root_item.byte_limit = 0;
1345 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1346 btrfs_set_root_generation(&root->root_item, trans->transid);
1347 btrfs_set_root_level(&root->root_item, 0);
1348 btrfs_set_root_refs(&root->root_item, 1);
1349 btrfs_set_root_used(&root->root_item, leaf->len);
1350 btrfs_set_root_last_snapshot(&root->root_item, 0);
1351 btrfs_set_root_dirid(&root->root_item, 0);
1353 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1354 root->root_item.drop_level = 0;
1356 key.objectid = objectid;
1357 key.type = BTRFS_ROOT_ITEM_KEY;
1359 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1363 btrfs_tree_unlock(leaf);
1369 btrfs_tree_unlock(leaf);
1370 free_extent_buffer(root->commit_root);
1371 free_extent_buffer(leaf);
1375 return ERR_PTR(ret);
1378 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1379 struct btrfs_fs_info *fs_info)
1381 struct btrfs_root *root;
1382 struct btrfs_root *tree_root = fs_info->tree_root;
1383 struct extent_buffer *leaf;
1385 root = btrfs_alloc_root(fs_info);
1387 return ERR_PTR(-ENOMEM);
1389 __setup_root(tree_root->nodesize, tree_root->leafsize,
1390 tree_root->sectorsize, tree_root->stripesize,
1391 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1393 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1394 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1395 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1398 * DON'T set REF_COWS for log trees
1400 * log trees do not get reference counted because they go away
1401 * before a real commit is actually done. They do store pointers
1402 * to file data extents, and those reference counts still get
1403 * updated (along with back refs to the log tree).
1406 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1407 BTRFS_TREE_LOG_OBJECTID, NULL,
1411 return ERR_CAST(leaf);
1414 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1415 btrfs_set_header_bytenr(leaf, leaf->start);
1416 btrfs_set_header_generation(leaf, trans->transid);
1417 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1418 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1421 write_extent_buffer(root->node, root->fs_info->fsid,
1422 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1423 btrfs_mark_buffer_dirty(root->node);
1424 btrfs_tree_unlock(root->node);
1428 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1429 struct btrfs_fs_info *fs_info)
1431 struct btrfs_root *log_root;
1433 log_root = alloc_log_tree(trans, fs_info);
1434 if (IS_ERR(log_root))
1435 return PTR_ERR(log_root);
1436 WARN_ON(fs_info->log_root_tree);
1437 fs_info->log_root_tree = log_root;
1441 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1442 struct btrfs_root *root)
1444 struct btrfs_root *log_root;
1445 struct btrfs_inode_item *inode_item;
1447 log_root = alloc_log_tree(trans, root->fs_info);
1448 if (IS_ERR(log_root))
1449 return PTR_ERR(log_root);
1451 log_root->last_trans = trans->transid;
1452 log_root->root_key.offset = root->root_key.objectid;
1454 inode_item = &log_root->root_item.inode;
1455 btrfs_set_stack_inode_generation(inode_item, 1);
1456 btrfs_set_stack_inode_size(inode_item, 3);
1457 btrfs_set_stack_inode_nlink(inode_item, 1);
1458 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1459 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1461 btrfs_set_root_node(&log_root->root_item, log_root->node);
1463 WARN_ON(root->log_root);
1464 root->log_root = log_root;
1465 root->log_transid = 0;
1466 root->log_transid_committed = -1;
1467 root->last_log_commit = 0;
1471 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1472 struct btrfs_key *key)
1474 struct btrfs_root *root;
1475 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1476 struct btrfs_path *path;
1481 path = btrfs_alloc_path();
1483 return ERR_PTR(-ENOMEM);
1485 root = btrfs_alloc_root(fs_info);
1491 __setup_root(tree_root->nodesize, tree_root->leafsize,
1492 tree_root->sectorsize, tree_root->stripesize,
1493 root, fs_info, key->objectid);
1495 ret = btrfs_find_root(tree_root, key, path,
1496 &root->root_item, &root->root_key);
1503 generation = btrfs_root_generation(&root->root_item);
1504 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1505 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1506 blocksize, generation);
1510 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1514 root->commit_root = btrfs_root_node(root);
1516 btrfs_free_path(path);
1520 free_extent_buffer(root->node);
1524 root = ERR_PTR(ret);
1528 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1529 struct btrfs_key *location)
1531 struct btrfs_root *root;
1533 root = btrfs_read_tree_root(tree_root, location);
1537 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1538 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1539 btrfs_check_and_init_root_item(&root->root_item);
1545 int btrfs_init_fs_root(struct btrfs_root *root)
1548 struct btrfs_subvolume_writers *writers;
1550 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1551 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1553 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1558 writers = btrfs_alloc_subvolume_writers();
1559 if (IS_ERR(writers)) {
1560 ret = PTR_ERR(writers);
1563 root->subv_writers = writers;
1565 btrfs_init_free_ino_ctl(root);
1566 spin_lock_init(&root->cache_lock);
1567 init_waitqueue_head(&root->cache_wait);
1569 ret = get_anon_bdev(&root->anon_dev);
1575 btrfs_free_subvolume_writers(root->subv_writers);
1577 kfree(root->free_ino_ctl);
1578 kfree(root->free_ino_pinned);
1582 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1585 struct btrfs_root *root;
1587 spin_lock(&fs_info->fs_roots_radix_lock);
1588 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1589 (unsigned long)root_id);
1590 spin_unlock(&fs_info->fs_roots_radix_lock);
1594 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1595 struct btrfs_root *root)
1599 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1603 spin_lock(&fs_info->fs_roots_radix_lock);
1604 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1605 (unsigned long)root->root_key.objectid,
1608 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1609 spin_unlock(&fs_info->fs_roots_radix_lock);
1610 radix_tree_preload_end();
1615 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1616 struct btrfs_key *location,
1619 struct btrfs_root *root;
1622 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1623 return fs_info->tree_root;
1624 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1625 return fs_info->extent_root;
1626 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1627 return fs_info->chunk_root;
1628 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1629 return fs_info->dev_root;
1630 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1631 return fs_info->csum_root;
1632 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1633 return fs_info->quota_root ? fs_info->quota_root :
1635 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1636 return fs_info->uuid_root ? fs_info->uuid_root :
1639 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1641 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1642 return ERR_PTR(-ENOENT);
1646 root = btrfs_read_fs_root(fs_info->tree_root, location);
1650 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1655 ret = btrfs_init_fs_root(root);
1659 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1660 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1664 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1666 ret = btrfs_insert_fs_root(fs_info, root);
1668 if (ret == -EEXIST) {
1677 return ERR_PTR(ret);
1680 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1682 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1684 struct btrfs_device *device;
1685 struct backing_dev_info *bdi;
1688 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1691 bdi = blk_get_backing_dev_info(device->bdev);
1692 if (bdi && bdi_congested(bdi, bdi_bits)) {
1702 * If this fails, caller must call bdi_destroy() to get rid of the
1705 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1709 bdi->capabilities = BDI_CAP_MAP_COPY;
1710 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1714 bdi->ra_pages = default_backing_dev_info.ra_pages;
1715 bdi->congested_fn = btrfs_congested_fn;
1716 bdi->congested_data = info;
1721 * called by the kthread helper functions to finally call the bio end_io
1722 * functions. This is where read checksum verification actually happens
1724 static void end_workqueue_fn(struct btrfs_work *work)
1727 struct end_io_wq *end_io_wq;
1730 end_io_wq = container_of(work, struct end_io_wq, work);
1731 bio = end_io_wq->bio;
1733 error = end_io_wq->error;
1734 bio->bi_private = end_io_wq->private;
1735 bio->bi_end_io = end_io_wq->end_io;
1737 bio_endio_nodec(bio, error);
1740 static int cleaner_kthread(void *arg)
1742 struct btrfs_root *root = arg;
1748 /* Make the cleaner go to sleep early. */
1749 if (btrfs_need_cleaner_sleep(root))
1752 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1756 * Avoid the problem that we change the status of the fs
1757 * during the above check and trylock.
1759 if (btrfs_need_cleaner_sleep(root)) {
1760 mutex_unlock(&root->fs_info->cleaner_mutex);
1764 btrfs_run_delayed_iputs(root);
1765 again = btrfs_clean_one_deleted_snapshot(root);
1766 mutex_unlock(&root->fs_info->cleaner_mutex);
1769 * The defragger has dealt with the R/O remount and umount,
1770 * needn't do anything special here.
1772 btrfs_run_defrag_inodes(root->fs_info);
1774 if (!try_to_freeze() && !again) {
1775 set_current_state(TASK_INTERRUPTIBLE);
1776 if (!kthread_should_stop())
1778 __set_current_state(TASK_RUNNING);
1780 } while (!kthread_should_stop());
1784 static int transaction_kthread(void *arg)
1786 struct btrfs_root *root = arg;
1787 struct btrfs_trans_handle *trans;
1788 struct btrfs_transaction *cur;
1791 unsigned long delay;
1795 cannot_commit = false;
1796 delay = HZ * root->fs_info->commit_interval;
1797 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1799 spin_lock(&root->fs_info->trans_lock);
1800 cur = root->fs_info->running_transaction;
1802 spin_unlock(&root->fs_info->trans_lock);
1806 now = get_seconds();
1807 if (cur->state < TRANS_STATE_BLOCKED &&
1808 (now < cur->start_time ||
1809 now - cur->start_time < root->fs_info->commit_interval)) {
1810 spin_unlock(&root->fs_info->trans_lock);
1814 transid = cur->transid;
1815 spin_unlock(&root->fs_info->trans_lock);
1817 /* If the file system is aborted, this will always fail. */
1818 trans = btrfs_attach_transaction(root);
1819 if (IS_ERR(trans)) {
1820 if (PTR_ERR(trans) != -ENOENT)
1821 cannot_commit = true;
1824 if (transid == trans->transid) {
1825 btrfs_commit_transaction(trans, root);
1827 btrfs_end_transaction(trans, root);
1830 wake_up_process(root->fs_info->cleaner_kthread);
1831 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1833 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1834 &root->fs_info->fs_state)))
1835 btrfs_cleanup_transaction(root);
1836 if (!try_to_freeze()) {
1837 set_current_state(TASK_INTERRUPTIBLE);
1838 if (!kthread_should_stop() &&
1839 (!btrfs_transaction_blocked(root->fs_info) ||
1841 schedule_timeout(delay);
1842 __set_current_state(TASK_RUNNING);
1844 } while (!kthread_should_stop());
1849 * this will find the highest generation in the array of
1850 * root backups. The index of the highest array is returned,
1851 * or -1 if we can't find anything.
1853 * We check to make sure the array is valid by comparing the
1854 * generation of the latest root in the array with the generation
1855 * in the super block. If they don't match we pitch it.
1857 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1860 int newest_index = -1;
1861 struct btrfs_root_backup *root_backup;
1864 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1865 root_backup = info->super_copy->super_roots + i;
1866 cur = btrfs_backup_tree_root_gen(root_backup);
1867 if (cur == newest_gen)
1871 /* check to see if we actually wrapped around */
1872 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1873 root_backup = info->super_copy->super_roots;
1874 cur = btrfs_backup_tree_root_gen(root_backup);
1875 if (cur == newest_gen)
1878 return newest_index;
1883 * find the oldest backup so we know where to store new entries
1884 * in the backup array. This will set the backup_root_index
1885 * field in the fs_info struct
1887 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1890 int newest_index = -1;
1892 newest_index = find_newest_super_backup(info, newest_gen);
1893 /* if there was garbage in there, just move along */
1894 if (newest_index == -1) {
1895 info->backup_root_index = 0;
1897 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1902 * copy all the root pointers into the super backup array.
1903 * this will bump the backup pointer by one when it is
1906 static void backup_super_roots(struct btrfs_fs_info *info)
1909 struct btrfs_root_backup *root_backup;
1912 next_backup = info->backup_root_index;
1913 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1914 BTRFS_NUM_BACKUP_ROOTS;
1917 * just overwrite the last backup if we're at the same generation
1918 * this happens only at umount
1920 root_backup = info->super_for_commit->super_roots + last_backup;
1921 if (btrfs_backup_tree_root_gen(root_backup) ==
1922 btrfs_header_generation(info->tree_root->node))
1923 next_backup = last_backup;
1925 root_backup = info->super_for_commit->super_roots + next_backup;
1928 * make sure all of our padding and empty slots get zero filled
1929 * regardless of which ones we use today
1931 memset(root_backup, 0, sizeof(*root_backup));
1933 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1935 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1936 btrfs_set_backup_tree_root_gen(root_backup,
1937 btrfs_header_generation(info->tree_root->node));
1939 btrfs_set_backup_tree_root_level(root_backup,
1940 btrfs_header_level(info->tree_root->node));
1942 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1943 btrfs_set_backup_chunk_root_gen(root_backup,
1944 btrfs_header_generation(info->chunk_root->node));
1945 btrfs_set_backup_chunk_root_level(root_backup,
1946 btrfs_header_level(info->chunk_root->node));
1948 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1949 btrfs_set_backup_extent_root_gen(root_backup,
1950 btrfs_header_generation(info->extent_root->node));
1951 btrfs_set_backup_extent_root_level(root_backup,
1952 btrfs_header_level(info->extent_root->node));
1955 * we might commit during log recovery, which happens before we set
1956 * the fs_root. Make sure it is valid before we fill it in.
1958 if (info->fs_root && info->fs_root->node) {
1959 btrfs_set_backup_fs_root(root_backup,
1960 info->fs_root->node->start);
1961 btrfs_set_backup_fs_root_gen(root_backup,
1962 btrfs_header_generation(info->fs_root->node));
1963 btrfs_set_backup_fs_root_level(root_backup,
1964 btrfs_header_level(info->fs_root->node));
1967 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1968 btrfs_set_backup_dev_root_gen(root_backup,
1969 btrfs_header_generation(info->dev_root->node));
1970 btrfs_set_backup_dev_root_level(root_backup,
1971 btrfs_header_level(info->dev_root->node));
1973 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1974 btrfs_set_backup_csum_root_gen(root_backup,
1975 btrfs_header_generation(info->csum_root->node));
1976 btrfs_set_backup_csum_root_level(root_backup,
1977 btrfs_header_level(info->csum_root->node));
1979 btrfs_set_backup_total_bytes(root_backup,
1980 btrfs_super_total_bytes(info->super_copy));
1981 btrfs_set_backup_bytes_used(root_backup,
1982 btrfs_super_bytes_used(info->super_copy));
1983 btrfs_set_backup_num_devices(root_backup,
1984 btrfs_super_num_devices(info->super_copy));
1987 * if we don't copy this out to the super_copy, it won't get remembered
1988 * for the next commit
1990 memcpy(&info->super_copy->super_roots,
1991 &info->super_for_commit->super_roots,
1992 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1996 * this copies info out of the root backup array and back into
1997 * the in-memory super block. It is meant to help iterate through
1998 * the array, so you send it the number of backups you've already
1999 * tried and the last backup index you used.
2001 * this returns -1 when it has tried all the backups
2003 static noinline int next_root_backup(struct btrfs_fs_info *info,
2004 struct btrfs_super_block *super,
2005 int *num_backups_tried, int *backup_index)
2007 struct btrfs_root_backup *root_backup;
2008 int newest = *backup_index;
2010 if (*num_backups_tried == 0) {
2011 u64 gen = btrfs_super_generation(super);
2013 newest = find_newest_super_backup(info, gen);
2017 *backup_index = newest;
2018 *num_backups_tried = 1;
2019 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2020 /* we've tried all the backups, all done */
2023 /* jump to the next oldest backup */
2024 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2025 BTRFS_NUM_BACKUP_ROOTS;
2026 *backup_index = newest;
2027 *num_backups_tried += 1;
2029 root_backup = super->super_roots + newest;
2031 btrfs_set_super_generation(super,
2032 btrfs_backup_tree_root_gen(root_backup));
2033 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2034 btrfs_set_super_root_level(super,
2035 btrfs_backup_tree_root_level(root_backup));
2036 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2039 * fixme: the total bytes and num_devices need to match or we should
2042 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2043 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2047 /* helper to cleanup workers */
2048 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2050 btrfs_destroy_workqueue(fs_info->fixup_workers);
2051 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2052 btrfs_destroy_workqueue(fs_info->workers);
2053 btrfs_destroy_workqueue(fs_info->endio_workers);
2054 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2055 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2056 btrfs_destroy_workqueue(fs_info->rmw_workers);
2057 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2058 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2059 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2060 btrfs_destroy_workqueue(fs_info->submit_workers);
2061 btrfs_destroy_workqueue(fs_info->delayed_workers);
2062 btrfs_destroy_workqueue(fs_info->caching_workers);
2063 btrfs_destroy_workqueue(fs_info->readahead_workers);
2064 btrfs_destroy_workqueue(fs_info->flush_workers);
2065 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2068 static void free_root_extent_buffers(struct btrfs_root *root)
2071 free_extent_buffer(root->node);
2072 free_extent_buffer(root->commit_root);
2074 root->commit_root = NULL;
2078 /* helper to cleanup tree roots */
2079 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2081 free_root_extent_buffers(info->tree_root);
2083 free_root_extent_buffers(info->dev_root);
2084 free_root_extent_buffers(info->extent_root);
2085 free_root_extent_buffers(info->csum_root);
2086 free_root_extent_buffers(info->quota_root);
2087 free_root_extent_buffers(info->uuid_root);
2089 free_root_extent_buffers(info->chunk_root);
2092 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2095 struct btrfs_root *gang[8];
2098 while (!list_empty(&fs_info->dead_roots)) {
2099 gang[0] = list_entry(fs_info->dead_roots.next,
2100 struct btrfs_root, root_list);
2101 list_del(&gang[0]->root_list);
2103 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2104 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2106 free_extent_buffer(gang[0]->node);
2107 free_extent_buffer(gang[0]->commit_root);
2108 btrfs_put_fs_root(gang[0]);
2113 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2118 for (i = 0; i < ret; i++)
2119 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2122 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2123 btrfs_free_log_root_tree(NULL, fs_info);
2124 btrfs_destroy_pinned_extent(fs_info->tree_root,
2125 fs_info->pinned_extents);
2129 int open_ctree(struct super_block *sb,
2130 struct btrfs_fs_devices *fs_devices,
2140 struct btrfs_key location;
2141 struct buffer_head *bh;
2142 struct btrfs_super_block *disk_super;
2143 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2144 struct btrfs_root *tree_root;
2145 struct btrfs_root *extent_root;
2146 struct btrfs_root *csum_root;
2147 struct btrfs_root *chunk_root;
2148 struct btrfs_root *dev_root;
2149 struct btrfs_root *quota_root;
2150 struct btrfs_root *uuid_root;
2151 struct btrfs_root *log_tree_root;
2154 int num_backups_tried = 0;
2155 int backup_index = 0;
2157 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2158 bool create_uuid_tree;
2159 bool check_uuid_tree;
2161 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2162 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2163 if (!tree_root || !chunk_root) {
2168 ret = init_srcu_struct(&fs_info->subvol_srcu);
2174 ret = setup_bdi(fs_info, &fs_info->bdi);
2180 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2185 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2186 (1 + ilog2(nr_cpu_ids));
2188 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2191 goto fail_dirty_metadata_bytes;
2194 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2197 goto fail_delalloc_bytes;
2200 fs_info->btree_inode = new_inode(sb);
2201 if (!fs_info->btree_inode) {
2203 goto fail_bio_counter;
2206 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2208 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2209 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2210 INIT_LIST_HEAD(&fs_info->trans_list);
2211 INIT_LIST_HEAD(&fs_info->dead_roots);
2212 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2213 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2214 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2215 spin_lock_init(&fs_info->delalloc_root_lock);
2216 spin_lock_init(&fs_info->trans_lock);
2217 spin_lock_init(&fs_info->fs_roots_radix_lock);
2218 spin_lock_init(&fs_info->delayed_iput_lock);
2219 spin_lock_init(&fs_info->defrag_inodes_lock);
2220 spin_lock_init(&fs_info->free_chunk_lock);
2221 spin_lock_init(&fs_info->tree_mod_seq_lock);
2222 spin_lock_init(&fs_info->super_lock);
2223 spin_lock_init(&fs_info->qgroup_op_lock);
2224 spin_lock_init(&fs_info->buffer_lock);
2225 rwlock_init(&fs_info->tree_mod_log_lock);
2226 mutex_init(&fs_info->reloc_mutex);
2227 mutex_init(&fs_info->delalloc_root_mutex);
2228 seqlock_init(&fs_info->profiles_lock);
2230 init_completion(&fs_info->kobj_unregister);
2231 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2232 INIT_LIST_HEAD(&fs_info->space_info);
2233 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2234 btrfs_mapping_init(&fs_info->mapping_tree);
2235 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2236 BTRFS_BLOCK_RSV_GLOBAL);
2237 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2238 BTRFS_BLOCK_RSV_DELALLOC);
2239 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2240 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2241 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2242 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2243 BTRFS_BLOCK_RSV_DELOPS);
2244 atomic_set(&fs_info->nr_async_submits, 0);
2245 atomic_set(&fs_info->async_delalloc_pages, 0);
2246 atomic_set(&fs_info->async_submit_draining, 0);
2247 atomic_set(&fs_info->nr_async_bios, 0);
2248 atomic_set(&fs_info->defrag_running, 0);
2249 atomic_set(&fs_info->qgroup_op_seq, 0);
2250 atomic64_set(&fs_info->tree_mod_seq, 0);
2252 fs_info->max_inline = 8192 * 1024;
2253 fs_info->metadata_ratio = 0;
2254 fs_info->defrag_inodes = RB_ROOT;
2255 fs_info->free_chunk_space = 0;
2256 fs_info->tree_mod_log = RB_ROOT;
2257 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2258 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2259 /* readahead state */
2260 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2261 spin_lock_init(&fs_info->reada_lock);
2263 fs_info->thread_pool_size = min_t(unsigned long,
2264 num_online_cpus() + 2, 8);
2266 INIT_LIST_HEAD(&fs_info->ordered_roots);
2267 spin_lock_init(&fs_info->ordered_root_lock);
2268 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2270 if (!fs_info->delayed_root) {
2274 btrfs_init_delayed_root(fs_info->delayed_root);
2276 mutex_init(&fs_info->scrub_lock);
2277 atomic_set(&fs_info->scrubs_running, 0);
2278 atomic_set(&fs_info->scrub_pause_req, 0);
2279 atomic_set(&fs_info->scrubs_paused, 0);
2280 atomic_set(&fs_info->scrub_cancel_req, 0);
2281 init_waitqueue_head(&fs_info->replace_wait);
2282 init_waitqueue_head(&fs_info->scrub_pause_wait);
2283 fs_info->scrub_workers_refcnt = 0;
2284 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2285 fs_info->check_integrity_print_mask = 0;
2288 spin_lock_init(&fs_info->balance_lock);
2289 mutex_init(&fs_info->balance_mutex);
2290 atomic_set(&fs_info->balance_running, 0);
2291 atomic_set(&fs_info->balance_pause_req, 0);
2292 atomic_set(&fs_info->balance_cancel_req, 0);
2293 fs_info->balance_ctl = NULL;
2294 init_waitqueue_head(&fs_info->balance_wait_q);
2295 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2297 sb->s_blocksize = 4096;
2298 sb->s_blocksize_bits = blksize_bits(4096);
2299 sb->s_bdi = &fs_info->bdi;
2301 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2302 set_nlink(fs_info->btree_inode, 1);
2304 * we set the i_size on the btree inode to the max possible int.
2305 * the real end of the address space is determined by all of
2306 * the devices in the system
2308 fs_info->btree_inode->i_size = OFFSET_MAX;
2309 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2310 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2312 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2313 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2314 fs_info->btree_inode->i_mapping);
2315 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2316 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2318 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2320 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2321 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2322 sizeof(struct btrfs_key));
2323 set_bit(BTRFS_INODE_DUMMY,
2324 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2325 btrfs_insert_inode_hash(fs_info->btree_inode);
2327 spin_lock_init(&fs_info->block_group_cache_lock);
2328 fs_info->block_group_cache_tree = RB_ROOT;
2329 fs_info->first_logical_byte = (u64)-1;
2331 extent_io_tree_init(&fs_info->freed_extents[0],
2332 fs_info->btree_inode->i_mapping);
2333 extent_io_tree_init(&fs_info->freed_extents[1],
2334 fs_info->btree_inode->i_mapping);
2335 fs_info->pinned_extents = &fs_info->freed_extents[0];
2336 fs_info->do_barriers = 1;
2339 mutex_init(&fs_info->ordered_operations_mutex);
2340 mutex_init(&fs_info->ordered_extent_flush_mutex);
2341 mutex_init(&fs_info->tree_log_mutex);
2342 mutex_init(&fs_info->chunk_mutex);
2343 mutex_init(&fs_info->transaction_kthread_mutex);
2344 mutex_init(&fs_info->cleaner_mutex);
2345 mutex_init(&fs_info->volume_mutex);
2346 init_rwsem(&fs_info->commit_root_sem);
2347 init_rwsem(&fs_info->cleanup_work_sem);
2348 init_rwsem(&fs_info->subvol_sem);
2349 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2350 fs_info->dev_replace.lock_owner = 0;
2351 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2352 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2353 mutex_init(&fs_info->dev_replace.lock_management_lock);
2354 mutex_init(&fs_info->dev_replace.lock);
2356 spin_lock_init(&fs_info->qgroup_lock);
2357 mutex_init(&fs_info->qgroup_ioctl_lock);
2358 fs_info->qgroup_tree = RB_ROOT;
2359 fs_info->qgroup_op_tree = RB_ROOT;
2360 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2361 fs_info->qgroup_seq = 1;
2362 fs_info->quota_enabled = 0;
2363 fs_info->pending_quota_state = 0;
2364 fs_info->qgroup_ulist = NULL;
2365 mutex_init(&fs_info->qgroup_rescan_lock);
2367 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2368 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2370 init_waitqueue_head(&fs_info->transaction_throttle);
2371 init_waitqueue_head(&fs_info->transaction_wait);
2372 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2373 init_waitqueue_head(&fs_info->async_submit_wait);
2375 ret = btrfs_alloc_stripe_hash_table(fs_info);
2381 __setup_root(4096, 4096, 4096, 4096, tree_root,
2382 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2384 invalidate_bdev(fs_devices->latest_bdev);
2387 * Read super block and check the signature bytes only
2389 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2396 * We want to check superblock checksum, the type is stored inside.
2397 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2399 if (btrfs_check_super_csum(bh->b_data)) {
2400 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2406 * super_copy is zeroed at allocation time and we never touch the
2407 * following bytes up to INFO_SIZE, the checksum is calculated from
2408 * the whole block of INFO_SIZE
2410 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2411 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2412 sizeof(*fs_info->super_for_commit));
2415 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2417 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2419 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2424 disk_super = fs_info->super_copy;
2425 if (!btrfs_super_root(disk_super))
2428 /* check FS state, whether FS is broken. */
2429 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2430 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2433 * run through our array of backup supers and setup
2434 * our ring pointer to the oldest one
2436 generation = btrfs_super_generation(disk_super);
2437 find_oldest_super_backup(fs_info, generation);
2440 * In the long term, we'll store the compression type in the super
2441 * block, and it'll be used for per file compression control.
2443 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2445 ret = btrfs_parse_options(tree_root, options);
2451 features = btrfs_super_incompat_flags(disk_super) &
2452 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2454 printk(KERN_ERR "BTRFS: couldn't mount because of "
2455 "unsupported optional features (%Lx).\n",
2461 if (btrfs_super_leafsize(disk_super) !=
2462 btrfs_super_nodesize(disk_super)) {
2463 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2464 "blocksizes don't match. node %d leaf %d\n",
2465 btrfs_super_nodesize(disk_super),
2466 btrfs_super_leafsize(disk_super));
2470 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2471 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2472 "blocksize (%d) was too large\n",
2473 btrfs_super_leafsize(disk_super));
2478 features = btrfs_super_incompat_flags(disk_super);
2479 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2480 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2481 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2483 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2484 printk(KERN_ERR "BTRFS: has skinny extents\n");
2487 * flag our filesystem as having big metadata blocks if
2488 * they are bigger than the page size
2490 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2491 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2492 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2493 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2496 nodesize = btrfs_super_nodesize(disk_super);
2497 leafsize = btrfs_super_leafsize(disk_super);
2498 sectorsize = btrfs_super_sectorsize(disk_super);
2499 stripesize = btrfs_super_stripesize(disk_super);
2500 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2501 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2504 * mixed block groups end up with duplicate but slightly offset
2505 * extent buffers for the same range. It leads to corruptions
2507 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2508 (sectorsize != leafsize)) {
2509 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2510 "are not allowed for mixed block groups on %s\n",
2516 * Needn't use the lock because there is no other task which will
2519 btrfs_set_super_incompat_flags(disk_super, features);
2521 features = btrfs_super_compat_ro_flags(disk_super) &
2522 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2523 if (!(sb->s_flags & MS_RDONLY) && features) {
2524 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2525 "unsupported option features (%Lx).\n",
2531 max_active = fs_info->thread_pool_size;
2534 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2537 fs_info->delalloc_workers =
2538 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2540 fs_info->flush_workers =
2541 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2543 fs_info->caching_workers =
2544 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2547 * a higher idle thresh on the submit workers makes it much more
2548 * likely that bios will be send down in a sane order to the
2551 fs_info->submit_workers =
2552 btrfs_alloc_workqueue("submit", flags,
2553 min_t(u64, fs_devices->num_devices,
2556 fs_info->fixup_workers =
2557 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2560 * endios are largely parallel and should have a very
2563 fs_info->endio_workers =
2564 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2565 fs_info->endio_meta_workers =
2566 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2567 fs_info->endio_meta_write_workers =
2568 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2569 fs_info->endio_raid56_workers =
2570 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2571 fs_info->rmw_workers =
2572 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2573 fs_info->endio_write_workers =
2574 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2575 fs_info->endio_freespace_worker =
2576 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2577 fs_info->delayed_workers =
2578 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2579 fs_info->readahead_workers =
2580 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2581 fs_info->qgroup_rescan_workers =
2582 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2584 if (!(fs_info->workers && fs_info->delalloc_workers &&
2585 fs_info->submit_workers && fs_info->flush_workers &&
2586 fs_info->endio_workers && fs_info->endio_meta_workers &&
2587 fs_info->endio_meta_write_workers &&
2588 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2589 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2590 fs_info->caching_workers && fs_info->readahead_workers &&
2591 fs_info->fixup_workers && fs_info->delayed_workers &&
2592 fs_info->qgroup_rescan_workers)) {
2594 goto fail_sb_buffer;
2597 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2598 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2599 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2601 tree_root->nodesize = nodesize;
2602 tree_root->leafsize = leafsize;
2603 tree_root->sectorsize = sectorsize;
2604 tree_root->stripesize = stripesize;
2606 sb->s_blocksize = sectorsize;
2607 sb->s_blocksize_bits = blksize_bits(sectorsize);
2609 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2610 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2611 goto fail_sb_buffer;
2614 if (sectorsize != PAGE_SIZE) {
2615 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2616 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2617 goto fail_sb_buffer;
2620 mutex_lock(&fs_info->chunk_mutex);
2621 ret = btrfs_read_sys_array(tree_root);
2622 mutex_unlock(&fs_info->chunk_mutex);
2624 printk(KERN_WARNING "BTRFS: failed to read the system "
2625 "array on %s\n", sb->s_id);
2626 goto fail_sb_buffer;
2629 blocksize = btrfs_level_size(tree_root,
2630 btrfs_super_chunk_root_level(disk_super));
2631 generation = btrfs_super_chunk_root_generation(disk_super);
2633 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2634 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2636 chunk_root->node = read_tree_block(chunk_root,
2637 btrfs_super_chunk_root(disk_super),
2638 blocksize, generation);
2639 if (!chunk_root->node ||
2640 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2641 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2643 goto fail_tree_roots;
2645 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2646 chunk_root->commit_root = btrfs_root_node(chunk_root);
2648 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2649 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2651 ret = btrfs_read_chunk_tree(chunk_root);
2653 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2655 goto fail_tree_roots;
2659 * keep the device that is marked to be the target device for the
2660 * dev_replace procedure
2662 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2664 if (!fs_devices->latest_bdev) {
2665 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2667 goto fail_tree_roots;
2671 blocksize = btrfs_level_size(tree_root,
2672 btrfs_super_root_level(disk_super));
2673 generation = btrfs_super_generation(disk_super);
2675 tree_root->node = read_tree_block(tree_root,
2676 btrfs_super_root(disk_super),
2677 blocksize, generation);
2678 if (!tree_root->node ||
2679 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2680 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2683 goto recovery_tree_root;
2686 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2687 tree_root->commit_root = btrfs_root_node(tree_root);
2688 btrfs_set_root_refs(&tree_root->root_item, 1);
2690 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2691 location.type = BTRFS_ROOT_ITEM_KEY;
2692 location.offset = 0;
2694 extent_root = btrfs_read_tree_root(tree_root, &location);
2695 if (IS_ERR(extent_root)) {
2696 ret = PTR_ERR(extent_root);
2697 goto recovery_tree_root;
2699 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2700 fs_info->extent_root = extent_root;
2702 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2703 dev_root = btrfs_read_tree_root(tree_root, &location);
2704 if (IS_ERR(dev_root)) {
2705 ret = PTR_ERR(dev_root);
2706 goto recovery_tree_root;
2708 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2709 fs_info->dev_root = dev_root;
2710 btrfs_init_devices_late(fs_info);
2712 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2713 csum_root = btrfs_read_tree_root(tree_root, &location);
2714 if (IS_ERR(csum_root)) {
2715 ret = PTR_ERR(csum_root);
2716 goto recovery_tree_root;
2718 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2719 fs_info->csum_root = csum_root;
2721 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2722 quota_root = btrfs_read_tree_root(tree_root, &location);
2723 if (!IS_ERR(quota_root)) {
2724 set_bit(BTRFS_ROOT_TRACK_DIRTY, "a_root->state);
2725 fs_info->quota_enabled = 1;
2726 fs_info->pending_quota_state = 1;
2727 fs_info->quota_root = quota_root;
2730 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2731 uuid_root = btrfs_read_tree_root(tree_root, &location);
2732 if (IS_ERR(uuid_root)) {
2733 ret = PTR_ERR(uuid_root);
2735 goto recovery_tree_root;
2736 create_uuid_tree = true;
2737 check_uuid_tree = false;
2739 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2740 fs_info->uuid_root = uuid_root;
2741 create_uuid_tree = false;
2743 generation != btrfs_super_uuid_tree_generation(disk_super);
2746 fs_info->generation = generation;
2747 fs_info->last_trans_committed = generation;
2749 ret = btrfs_recover_balance(fs_info);
2751 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2752 goto fail_block_groups;
2755 ret = btrfs_init_dev_stats(fs_info);
2757 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2759 goto fail_block_groups;
2762 ret = btrfs_init_dev_replace(fs_info);
2764 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2765 goto fail_block_groups;
2768 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2770 ret = btrfs_sysfs_add_one(fs_info);
2772 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2773 goto fail_block_groups;
2776 ret = btrfs_init_space_info(fs_info);
2778 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2782 ret = btrfs_read_block_groups(extent_root);
2784 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2787 fs_info->num_tolerated_disk_barrier_failures =
2788 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2789 if (fs_info->fs_devices->missing_devices >
2790 fs_info->num_tolerated_disk_barrier_failures &&
2791 !(sb->s_flags & MS_RDONLY)) {
2792 printk(KERN_WARNING "BTRFS: "
2793 "too many missing devices, writeable mount is not allowed\n");
2797 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2799 if (IS_ERR(fs_info->cleaner_kthread))
2802 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2804 "btrfs-transaction");
2805 if (IS_ERR(fs_info->transaction_kthread))
2808 if (!btrfs_test_opt(tree_root, SSD) &&
2809 !btrfs_test_opt(tree_root, NOSSD) &&
2810 !fs_info->fs_devices->rotating) {
2811 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2813 btrfs_set_opt(fs_info->mount_opt, SSD);
2816 /* Set the real inode map cache flag */
2817 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2818 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2820 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2821 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2822 ret = btrfsic_mount(tree_root, fs_devices,
2823 btrfs_test_opt(tree_root,
2824 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2826 fs_info->check_integrity_print_mask);
2828 printk(KERN_WARNING "BTRFS: failed to initialize"
2829 " integrity check module %s\n", sb->s_id);
2832 ret = btrfs_read_qgroup_config(fs_info);
2834 goto fail_trans_kthread;
2836 /* do not make disk changes in broken FS */
2837 if (btrfs_super_log_root(disk_super) != 0) {
2838 u64 bytenr = btrfs_super_log_root(disk_super);
2840 if (fs_devices->rw_devices == 0) {
2841 printk(KERN_WARNING "BTRFS: log replay required "
2847 btrfs_level_size(tree_root,
2848 btrfs_super_log_root_level(disk_super));
2850 log_tree_root = btrfs_alloc_root(fs_info);
2851 if (!log_tree_root) {
2856 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2857 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2859 log_tree_root->node = read_tree_block(tree_root, bytenr,
2862 if (!log_tree_root->node ||
2863 !extent_buffer_uptodate(log_tree_root->node)) {
2864 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2865 free_extent_buffer(log_tree_root->node);
2866 kfree(log_tree_root);
2869 /* returns with log_tree_root freed on success */
2870 ret = btrfs_recover_log_trees(log_tree_root);
2872 btrfs_error(tree_root->fs_info, ret,
2873 "Failed to recover log tree");
2874 free_extent_buffer(log_tree_root->node);
2875 kfree(log_tree_root);
2879 if (sb->s_flags & MS_RDONLY) {
2880 ret = btrfs_commit_super(tree_root);
2886 ret = btrfs_find_orphan_roots(tree_root);
2890 if (!(sb->s_flags & MS_RDONLY)) {
2891 ret = btrfs_cleanup_fs_roots(fs_info);
2895 ret = btrfs_recover_relocation(tree_root);
2898 "BTRFS: failed to recover relocation\n");
2904 location.objectid = BTRFS_FS_TREE_OBJECTID;
2905 location.type = BTRFS_ROOT_ITEM_KEY;
2906 location.offset = 0;
2908 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2909 if (IS_ERR(fs_info->fs_root)) {
2910 err = PTR_ERR(fs_info->fs_root);
2914 if (sb->s_flags & MS_RDONLY)
2917 down_read(&fs_info->cleanup_work_sem);
2918 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2919 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2920 up_read(&fs_info->cleanup_work_sem);
2921 close_ctree(tree_root);
2924 up_read(&fs_info->cleanup_work_sem);
2926 ret = btrfs_resume_balance_async(fs_info);
2928 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2929 close_ctree(tree_root);
2933 ret = btrfs_resume_dev_replace_async(fs_info);
2935 pr_warn("BTRFS: failed to resume dev_replace\n");
2936 close_ctree(tree_root);
2940 btrfs_qgroup_rescan_resume(fs_info);
2942 if (create_uuid_tree) {
2943 pr_info("BTRFS: creating UUID tree\n");
2944 ret = btrfs_create_uuid_tree(fs_info);
2946 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2948 close_ctree(tree_root);
2951 } else if (check_uuid_tree ||
2952 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2953 pr_info("BTRFS: checking UUID tree\n");
2954 ret = btrfs_check_uuid_tree(fs_info);
2956 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2958 close_ctree(tree_root);
2962 fs_info->update_uuid_tree_gen = 1;
2968 btrfs_free_qgroup_config(fs_info);
2970 kthread_stop(fs_info->transaction_kthread);
2971 btrfs_cleanup_transaction(fs_info->tree_root);
2972 del_fs_roots(fs_info);
2974 kthread_stop(fs_info->cleaner_kthread);
2977 * make sure we're done with the btree inode before we stop our
2980 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2983 btrfs_sysfs_remove_one(fs_info);
2986 btrfs_put_block_group_cache(fs_info);
2987 btrfs_free_block_groups(fs_info);
2990 free_root_pointers(fs_info, 1);
2991 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2994 btrfs_stop_all_workers(fs_info);
2997 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2999 iput(fs_info->btree_inode);
3001 percpu_counter_destroy(&fs_info->bio_counter);
3002 fail_delalloc_bytes:
3003 percpu_counter_destroy(&fs_info->delalloc_bytes);
3004 fail_dirty_metadata_bytes:
3005 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3007 bdi_destroy(&fs_info->bdi);
3009 cleanup_srcu_struct(&fs_info->subvol_srcu);
3011 btrfs_free_stripe_hash_table(fs_info);
3012 btrfs_close_devices(fs_info->fs_devices);
3016 if (!btrfs_test_opt(tree_root, RECOVERY))
3017 goto fail_tree_roots;
3019 free_root_pointers(fs_info, 0);
3021 /* don't use the log in recovery mode, it won't be valid */
3022 btrfs_set_super_log_root(disk_super, 0);
3024 /* we can't trust the free space cache either */
3025 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3027 ret = next_root_backup(fs_info, fs_info->super_copy,
3028 &num_backups_tried, &backup_index);
3030 goto fail_block_groups;
3031 goto retry_root_backup;
3034 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3037 set_buffer_uptodate(bh);
3039 struct btrfs_device *device = (struct btrfs_device *)
3042 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3043 "I/O error on %s\n",
3044 rcu_str_deref(device->name));
3045 /* note, we dont' set_buffer_write_io_error because we have
3046 * our own ways of dealing with the IO errors
3048 clear_buffer_uptodate(bh);
3049 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3055 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3057 struct buffer_head *bh;
3058 struct buffer_head *latest = NULL;
3059 struct btrfs_super_block *super;
3064 /* we would like to check all the supers, but that would make
3065 * a btrfs mount succeed after a mkfs from a different FS.
3066 * So, we need to add a special mount option to scan for
3067 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3069 for (i = 0; i < 1; i++) {
3070 bytenr = btrfs_sb_offset(i);
3071 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3072 i_size_read(bdev->bd_inode))
3074 bh = __bread(bdev, bytenr / 4096,
3075 BTRFS_SUPER_INFO_SIZE);
3079 super = (struct btrfs_super_block *)bh->b_data;
3080 if (btrfs_super_bytenr(super) != bytenr ||
3081 btrfs_super_magic(super) != BTRFS_MAGIC) {
3086 if (!latest || btrfs_super_generation(super) > transid) {
3089 transid = btrfs_super_generation(super);
3098 * this should be called twice, once with wait == 0 and
3099 * once with wait == 1. When wait == 0 is done, all the buffer heads
3100 * we write are pinned.
3102 * They are released when wait == 1 is done.
3103 * max_mirrors must be the same for both runs, and it indicates how
3104 * many supers on this one device should be written.
3106 * max_mirrors == 0 means to write them all.
3108 static int write_dev_supers(struct btrfs_device *device,
3109 struct btrfs_super_block *sb,
3110 int do_barriers, int wait, int max_mirrors)
3112 struct buffer_head *bh;
3119 if (max_mirrors == 0)
3120 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3122 for (i = 0; i < max_mirrors; i++) {
3123 bytenr = btrfs_sb_offset(i);
3124 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3128 bh = __find_get_block(device->bdev, bytenr / 4096,
3129 BTRFS_SUPER_INFO_SIZE);
3135 if (!buffer_uptodate(bh))
3138 /* drop our reference */
3141 /* drop the reference from the wait == 0 run */
3145 btrfs_set_super_bytenr(sb, bytenr);
3148 crc = btrfs_csum_data((char *)sb +
3149 BTRFS_CSUM_SIZE, crc,
3150 BTRFS_SUPER_INFO_SIZE -
3152 btrfs_csum_final(crc, sb->csum);
3155 * one reference for us, and we leave it for the
3158 bh = __getblk(device->bdev, bytenr / 4096,
3159 BTRFS_SUPER_INFO_SIZE);
3161 printk(KERN_ERR "BTRFS: couldn't get super "
3162 "buffer head for bytenr %Lu\n", bytenr);
3167 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3169 /* one reference for submit_bh */
3172 set_buffer_uptodate(bh);
3174 bh->b_end_io = btrfs_end_buffer_write_sync;
3175 bh->b_private = device;
3179 * we fua the first super. The others we allow
3183 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3185 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3189 return errors < i ? 0 : -1;
3193 * endio for the write_dev_flush, this will wake anyone waiting
3194 * for the barrier when it is done
3196 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3199 if (err == -EOPNOTSUPP)
3200 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3201 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3203 if (bio->bi_private)
3204 complete(bio->bi_private);
3209 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3210 * sent down. With wait == 1, it waits for the previous flush.
3212 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3215 static int write_dev_flush(struct btrfs_device *device, int wait)
3220 if (device->nobarriers)
3224 bio = device->flush_bio;
3228 wait_for_completion(&device->flush_wait);
3230 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3231 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3232 rcu_str_deref(device->name));
3233 device->nobarriers = 1;
3234 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3236 btrfs_dev_stat_inc_and_print(device,
3237 BTRFS_DEV_STAT_FLUSH_ERRS);
3240 /* drop the reference from the wait == 0 run */
3242 device->flush_bio = NULL;
3248 * one reference for us, and we leave it for the
3251 device->flush_bio = NULL;
3252 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3256 bio->bi_end_io = btrfs_end_empty_barrier;
3257 bio->bi_bdev = device->bdev;
3258 init_completion(&device->flush_wait);
3259 bio->bi_private = &device->flush_wait;
3260 device->flush_bio = bio;
3263 btrfsic_submit_bio(WRITE_FLUSH, bio);
3269 * send an empty flush down to each device in parallel,
3270 * then wait for them
3272 static int barrier_all_devices(struct btrfs_fs_info *info)
3274 struct list_head *head;
3275 struct btrfs_device *dev;
3276 int errors_send = 0;
3277 int errors_wait = 0;
3280 /* send down all the barriers */
3281 head = &info->fs_devices->devices;
3282 list_for_each_entry_rcu(dev, head, dev_list) {
3289 if (!dev->in_fs_metadata || !dev->writeable)
3292 ret = write_dev_flush(dev, 0);
3297 /* wait for all the barriers */
3298 list_for_each_entry_rcu(dev, head, dev_list) {
3305 if (!dev->in_fs_metadata || !dev->writeable)
3308 ret = write_dev_flush(dev, 1);
3312 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3313 errors_wait > info->num_tolerated_disk_barrier_failures)
3318 int btrfs_calc_num_tolerated_disk_barrier_failures(
3319 struct btrfs_fs_info *fs_info)
3321 struct btrfs_ioctl_space_info space;
3322 struct btrfs_space_info *sinfo;
3323 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3324 BTRFS_BLOCK_GROUP_SYSTEM,
3325 BTRFS_BLOCK_GROUP_METADATA,
3326 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3330 int num_tolerated_disk_barrier_failures =
3331 (int)fs_info->fs_devices->num_devices;
3333 for (i = 0; i < num_types; i++) {
3334 struct btrfs_space_info *tmp;
3338 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3339 if (tmp->flags == types[i]) {
3349 down_read(&sinfo->groups_sem);
3350 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3351 if (!list_empty(&sinfo->block_groups[c])) {
3354 btrfs_get_block_group_info(
3355 &sinfo->block_groups[c], &space);
3356 if (space.total_bytes == 0 ||
3357 space.used_bytes == 0)
3359 flags = space.flags;
3362 * 0: if dup, single or RAID0 is configured for
3363 * any of metadata, system or data, else
3364 * 1: if RAID5 is configured, or if RAID1 or
3365 * RAID10 is configured and only two mirrors
3367 * 2: if RAID6 is configured, else
3368 * num_mirrors - 1: if RAID1 or RAID10 is
3369 * configured and more than
3370 * 2 mirrors are used.
3372 if (num_tolerated_disk_barrier_failures > 0 &&
3373 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3374 BTRFS_BLOCK_GROUP_RAID0)) ||
3375 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3377 num_tolerated_disk_barrier_failures = 0;
3378 else if (num_tolerated_disk_barrier_failures > 1) {
3379 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3380 BTRFS_BLOCK_GROUP_RAID5 |
3381 BTRFS_BLOCK_GROUP_RAID10)) {
3382 num_tolerated_disk_barrier_failures = 1;
3384 BTRFS_BLOCK_GROUP_RAID6) {
3385 num_tolerated_disk_barrier_failures = 2;
3390 up_read(&sinfo->groups_sem);
3393 return num_tolerated_disk_barrier_failures;
3396 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3398 struct list_head *head;
3399 struct btrfs_device *dev;
3400 struct btrfs_super_block *sb;
3401 struct btrfs_dev_item *dev_item;
3405 int total_errors = 0;
3408 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3409 backup_super_roots(root->fs_info);
3411 sb = root->fs_info->super_for_commit;
3412 dev_item = &sb->dev_item;
3414 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3415 head = &root->fs_info->fs_devices->devices;
3416 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3419 ret = barrier_all_devices(root->fs_info);
3422 &root->fs_info->fs_devices->device_list_mutex);
3423 btrfs_error(root->fs_info, ret,
3424 "errors while submitting device barriers.");
3429 list_for_each_entry_rcu(dev, head, dev_list) {
3434 if (!dev->in_fs_metadata || !dev->writeable)
3437 btrfs_set_stack_device_generation(dev_item, 0);
3438 btrfs_set_stack_device_type(dev_item, dev->type);
3439 btrfs_set_stack_device_id(dev_item, dev->devid);
3440 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3441 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3442 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3443 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3444 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3445 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3446 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3448 flags = btrfs_super_flags(sb);
3449 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3451 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3455 if (total_errors > max_errors) {
3456 btrfs_err(root->fs_info, "%d errors while writing supers",
3458 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3460 /* FUA is masked off if unsupported and can't be the reason */
3461 btrfs_error(root->fs_info, -EIO,
3462 "%d errors while writing supers", total_errors);
3467 list_for_each_entry_rcu(dev, head, dev_list) {
3470 if (!dev->in_fs_metadata || !dev->writeable)
3473 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3477 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3478 if (total_errors > max_errors) {
3479 btrfs_error(root->fs_info, -EIO,
3480 "%d errors while writing supers", total_errors);
3486 int write_ctree_super(struct btrfs_trans_handle *trans,
3487 struct btrfs_root *root, int max_mirrors)
3489 return write_all_supers(root, max_mirrors);
3492 /* Drop a fs root from the radix tree and free it. */
3493 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3494 struct btrfs_root *root)
3496 spin_lock(&fs_info->fs_roots_radix_lock);
3497 radix_tree_delete(&fs_info->fs_roots_radix,
3498 (unsigned long)root->root_key.objectid);
3499 spin_unlock(&fs_info->fs_roots_radix_lock);
3501 if (btrfs_root_refs(&root->root_item) == 0)
3502 synchronize_srcu(&fs_info->subvol_srcu);
3504 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3505 btrfs_free_log(NULL, root);
3507 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3508 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3512 static void free_fs_root(struct btrfs_root *root)
3514 iput(root->cache_inode);
3515 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3516 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3517 root->orphan_block_rsv = NULL;
3519 free_anon_bdev(root->anon_dev);
3520 if (root->subv_writers)
3521 btrfs_free_subvolume_writers(root->subv_writers);
3522 free_extent_buffer(root->node);
3523 free_extent_buffer(root->commit_root);
3524 kfree(root->free_ino_ctl);
3525 kfree(root->free_ino_pinned);
3527 btrfs_put_fs_root(root);
3530 void btrfs_free_fs_root(struct btrfs_root *root)
3535 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3537 u64 root_objectid = 0;
3538 struct btrfs_root *gang[8];
3541 unsigned int ret = 0;
3545 index = srcu_read_lock(&fs_info->subvol_srcu);
3546 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3547 (void **)gang, root_objectid,
3550 srcu_read_unlock(&fs_info->subvol_srcu, index);
3553 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3555 for (i = 0; i < ret; i++) {
3556 /* Avoid to grab roots in dead_roots */
3557 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3561 /* grab all the search result for later use */
3562 gang[i] = btrfs_grab_fs_root(gang[i]);
3564 srcu_read_unlock(&fs_info->subvol_srcu, index);
3566 for (i = 0; i < ret; i++) {
3569 root_objectid = gang[i]->root_key.objectid;
3570 err = btrfs_orphan_cleanup(gang[i]);
3573 btrfs_put_fs_root(gang[i]);
3578 /* release the uncleaned roots due to error */
3579 for (; i < ret; i++) {
3581 btrfs_put_fs_root(gang[i]);
3586 int btrfs_commit_super(struct btrfs_root *root)
3588 struct btrfs_trans_handle *trans;
3590 mutex_lock(&root->fs_info->cleaner_mutex);
3591 btrfs_run_delayed_iputs(root);
3592 mutex_unlock(&root->fs_info->cleaner_mutex);
3593 wake_up_process(root->fs_info->cleaner_kthread);
3595 /* wait until ongoing cleanup work done */
3596 down_write(&root->fs_info->cleanup_work_sem);
3597 up_write(&root->fs_info->cleanup_work_sem);
3599 trans = btrfs_join_transaction(root);
3601 return PTR_ERR(trans);
3602 return btrfs_commit_transaction(trans, root);
3605 int close_ctree(struct btrfs_root *root)
3607 struct btrfs_fs_info *fs_info = root->fs_info;
3610 fs_info->closing = 1;
3613 /* wait for the uuid_scan task to finish */
3614 down(&fs_info->uuid_tree_rescan_sem);
3615 /* avoid complains from lockdep et al., set sem back to initial state */
3616 up(&fs_info->uuid_tree_rescan_sem);
3618 /* pause restriper - we want to resume on mount */
3619 btrfs_pause_balance(fs_info);
3621 btrfs_dev_replace_suspend_for_unmount(fs_info);
3623 btrfs_scrub_cancel(fs_info);
3625 /* wait for any defraggers to finish */
3626 wait_event(fs_info->transaction_wait,
3627 (atomic_read(&fs_info->defrag_running) == 0));
3629 /* clear out the rbtree of defraggable inodes */
3630 btrfs_cleanup_defrag_inodes(fs_info);
3632 cancel_work_sync(&fs_info->async_reclaim_work);
3634 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3635 ret = btrfs_commit_super(root);
3637 btrfs_err(root->fs_info, "commit super ret %d", ret);
3640 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3641 btrfs_error_commit_super(root);
3643 kthread_stop(fs_info->transaction_kthread);
3644 kthread_stop(fs_info->cleaner_kthread);
3646 fs_info->closing = 2;
3649 btrfs_free_qgroup_config(root->fs_info);
3651 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3652 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3653 percpu_counter_sum(&fs_info->delalloc_bytes));
3656 btrfs_sysfs_remove_one(fs_info);
3658 del_fs_roots(fs_info);
3660 btrfs_put_block_group_cache(fs_info);
3662 btrfs_free_block_groups(fs_info);
3665 * we must make sure there is not any read request to
3666 * submit after we stopping all workers.
3668 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3669 btrfs_stop_all_workers(fs_info);
3671 free_root_pointers(fs_info, 1);
3673 iput(fs_info->btree_inode);
3675 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3676 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3677 btrfsic_unmount(root, fs_info->fs_devices);
3680 btrfs_close_devices(fs_info->fs_devices);
3681 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3683 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3684 percpu_counter_destroy(&fs_info->delalloc_bytes);
3685 percpu_counter_destroy(&fs_info->bio_counter);
3686 bdi_destroy(&fs_info->bdi);
3687 cleanup_srcu_struct(&fs_info->subvol_srcu);
3689 btrfs_free_stripe_hash_table(fs_info);
3691 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3692 root->orphan_block_rsv = NULL;
3697 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3701 struct inode *btree_inode = buf->pages[0]->mapping->host;
3703 ret = extent_buffer_uptodate(buf);
3707 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3708 parent_transid, atomic);
3714 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3716 return set_extent_buffer_uptodate(buf);
3719 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3721 struct btrfs_root *root;
3722 u64 transid = btrfs_header_generation(buf);
3725 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3727 * This is a fast path so only do this check if we have sanity tests
3728 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3729 * outside of the sanity tests.
3731 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3734 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3735 btrfs_assert_tree_locked(buf);
3736 if (transid != root->fs_info->generation)
3737 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3738 "found %llu running %llu\n",
3739 buf->start, transid, root->fs_info->generation);
3740 was_dirty = set_extent_buffer_dirty(buf);
3742 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3744 root->fs_info->dirty_metadata_batch);
3745 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3746 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3747 btrfs_print_leaf(root, buf);
3753 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3757 * looks as though older kernels can get into trouble with
3758 * this code, they end up stuck in balance_dirty_pages forever
3762 if (current->flags & PF_MEMALLOC)
3766 btrfs_balance_delayed_items(root);
3768 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3769 BTRFS_DIRTY_METADATA_THRESH);
3771 balance_dirty_pages_ratelimited(
3772 root->fs_info->btree_inode->i_mapping);
3777 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3779 __btrfs_btree_balance_dirty(root, 1);
3782 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3784 __btrfs_btree_balance_dirty(root, 0);
3787 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3789 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3790 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3793 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3797 * Placeholder for checks
3802 static void btrfs_error_commit_super(struct btrfs_root *root)
3804 mutex_lock(&root->fs_info->cleaner_mutex);
3805 btrfs_run_delayed_iputs(root);
3806 mutex_unlock(&root->fs_info->cleaner_mutex);
3808 down_write(&root->fs_info->cleanup_work_sem);
3809 up_write(&root->fs_info->cleanup_work_sem);
3811 /* cleanup FS via transaction */
3812 btrfs_cleanup_transaction(root);
3815 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3816 struct btrfs_root *root)
3818 struct btrfs_inode *btrfs_inode;
3819 struct list_head splice;
3821 INIT_LIST_HEAD(&splice);
3823 mutex_lock(&root->fs_info->ordered_operations_mutex);
3824 spin_lock(&root->fs_info->ordered_root_lock);
3826 list_splice_init(&t->ordered_operations, &splice);
3827 while (!list_empty(&splice)) {
3828 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3829 ordered_operations);
3831 list_del_init(&btrfs_inode->ordered_operations);
3832 spin_unlock(&root->fs_info->ordered_root_lock);
3834 btrfs_invalidate_inodes(btrfs_inode->root);
3836 spin_lock(&root->fs_info->ordered_root_lock);
3839 spin_unlock(&root->fs_info->ordered_root_lock);
3840 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3843 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3845 struct btrfs_ordered_extent *ordered;
3847 spin_lock(&root->ordered_extent_lock);
3849 * This will just short circuit the ordered completion stuff which will
3850 * make sure the ordered extent gets properly cleaned up.
3852 list_for_each_entry(ordered, &root->ordered_extents,
3854 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3855 spin_unlock(&root->ordered_extent_lock);
3858 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3860 struct btrfs_root *root;
3861 struct list_head splice;
3863 INIT_LIST_HEAD(&splice);
3865 spin_lock(&fs_info->ordered_root_lock);
3866 list_splice_init(&fs_info->ordered_roots, &splice);
3867 while (!list_empty(&splice)) {
3868 root = list_first_entry(&splice, struct btrfs_root,
3870 list_move_tail(&root->ordered_root,
3871 &fs_info->ordered_roots);
3873 spin_unlock(&fs_info->ordered_root_lock);
3874 btrfs_destroy_ordered_extents(root);
3877 spin_lock(&fs_info->ordered_root_lock);
3879 spin_unlock(&fs_info->ordered_root_lock);
3882 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3883 struct btrfs_root *root)
3885 struct rb_node *node;
3886 struct btrfs_delayed_ref_root *delayed_refs;
3887 struct btrfs_delayed_ref_node *ref;
3890 delayed_refs = &trans->delayed_refs;
3892 spin_lock(&delayed_refs->lock);
3893 if (atomic_read(&delayed_refs->num_entries) == 0) {
3894 spin_unlock(&delayed_refs->lock);
3895 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3899 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3900 struct btrfs_delayed_ref_head *head;
3901 bool pin_bytes = false;
3903 head = rb_entry(node, struct btrfs_delayed_ref_head,
3905 if (!mutex_trylock(&head->mutex)) {
3906 atomic_inc(&head->node.refs);
3907 spin_unlock(&delayed_refs->lock);
3909 mutex_lock(&head->mutex);
3910 mutex_unlock(&head->mutex);
3911 btrfs_put_delayed_ref(&head->node);
3912 spin_lock(&delayed_refs->lock);
3915 spin_lock(&head->lock);
3916 while ((node = rb_first(&head->ref_root)) != NULL) {
3917 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3920 rb_erase(&ref->rb_node, &head->ref_root);
3921 atomic_dec(&delayed_refs->num_entries);
3922 btrfs_put_delayed_ref(ref);
3924 if (head->must_insert_reserved)
3926 btrfs_free_delayed_extent_op(head->extent_op);
3927 delayed_refs->num_heads--;
3928 if (head->processing == 0)
3929 delayed_refs->num_heads_ready--;
3930 atomic_dec(&delayed_refs->num_entries);
3931 head->node.in_tree = 0;
3932 rb_erase(&head->href_node, &delayed_refs->href_root);
3933 spin_unlock(&head->lock);
3934 spin_unlock(&delayed_refs->lock);
3935 mutex_unlock(&head->mutex);
3938 btrfs_pin_extent(root, head->node.bytenr,
3939 head->node.num_bytes, 1);
3940 btrfs_put_delayed_ref(&head->node);
3942 spin_lock(&delayed_refs->lock);
3945 spin_unlock(&delayed_refs->lock);
3950 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3952 struct btrfs_inode *btrfs_inode;
3953 struct list_head splice;
3955 INIT_LIST_HEAD(&splice);
3957 spin_lock(&root->delalloc_lock);
3958 list_splice_init(&root->delalloc_inodes, &splice);
3960 while (!list_empty(&splice)) {
3961 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3964 list_del_init(&btrfs_inode->delalloc_inodes);
3965 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3966 &btrfs_inode->runtime_flags);
3967 spin_unlock(&root->delalloc_lock);
3969 btrfs_invalidate_inodes(btrfs_inode->root);
3971 spin_lock(&root->delalloc_lock);
3974 spin_unlock(&root->delalloc_lock);
3977 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3979 struct btrfs_root *root;
3980 struct list_head splice;
3982 INIT_LIST_HEAD(&splice);
3984 spin_lock(&fs_info->delalloc_root_lock);
3985 list_splice_init(&fs_info->delalloc_roots, &splice);
3986 while (!list_empty(&splice)) {
3987 root = list_first_entry(&splice, struct btrfs_root,
3989 list_del_init(&root->delalloc_root);
3990 root = btrfs_grab_fs_root(root);
3992 spin_unlock(&fs_info->delalloc_root_lock);
3994 btrfs_destroy_delalloc_inodes(root);
3995 btrfs_put_fs_root(root);
3997 spin_lock(&fs_info->delalloc_root_lock);
3999 spin_unlock(&fs_info->delalloc_root_lock);
4002 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4003 struct extent_io_tree *dirty_pages,
4007 struct extent_buffer *eb;
4012 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4017 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4018 while (start <= end) {
4019 eb = btrfs_find_tree_block(root, start,
4021 start += root->leafsize;
4024 wait_on_extent_buffer_writeback(eb);
4026 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4028 clear_extent_buffer_dirty(eb);
4029 free_extent_buffer_stale(eb);
4036 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4037 struct extent_io_tree *pinned_extents)
4039 struct extent_io_tree *unpin;
4045 unpin = pinned_extents;
4048 ret = find_first_extent_bit(unpin, 0, &start, &end,
4049 EXTENT_DIRTY, NULL);
4054 if (btrfs_test_opt(root, DISCARD))
4055 ret = btrfs_error_discard_extent(root, start,
4059 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4060 btrfs_error_unpin_extent_range(root, start, end);
4065 if (unpin == &root->fs_info->freed_extents[0])
4066 unpin = &root->fs_info->freed_extents[1];
4068 unpin = &root->fs_info->freed_extents[0];
4076 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4077 struct btrfs_root *root)
4079 btrfs_destroy_ordered_operations(cur_trans, root);
4081 btrfs_destroy_delayed_refs(cur_trans, root);
4083 cur_trans->state = TRANS_STATE_COMMIT_START;
4084 wake_up(&root->fs_info->transaction_blocked_wait);
4086 cur_trans->state = TRANS_STATE_UNBLOCKED;
4087 wake_up(&root->fs_info->transaction_wait);
4089 btrfs_destroy_delayed_inodes(root);
4090 btrfs_assert_delayed_root_empty(root);
4092 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4094 btrfs_destroy_pinned_extent(root,
4095 root->fs_info->pinned_extents);
4097 cur_trans->state =TRANS_STATE_COMPLETED;
4098 wake_up(&cur_trans->commit_wait);
4101 memset(cur_trans, 0, sizeof(*cur_trans));
4102 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4106 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4108 struct btrfs_transaction *t;
4110 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4112 spin_lock(&root->fs_info->trans_lock);
4113 while (!list_empty(&root->fs_info->trans_list)) {
4114 t = list_first_entry(&root->fs_info->trans_list,
4115 struct btrfs_transaction, list);
4116 if (t->state >= TRANS_STATE_COMMIT_START) {
4117 atomic_inc(&t->use_count);
4118 spin_unlock(&root->fs_info->trans_lock);
4119 btrfs_wait_for_commit(root, t->transid);
4120 btrfs_put_transaction(t);
4121 spin_lock(&root->fs_info->trans_lock);
4124 if (t == root->fs_info->running_transaction) {
4125 t->state = TRANS_STATE_COMMIT_DOING;
4126 spin_unlock(&root->fs_info->trans_lock);
4128 * We wait for 0 num_writers since we don't hold a trans
4129 * handle open currently for this transaction.
4131 wait_event(t->writer_wait,
4132 atomic_read(&t->num_writers) == 0);
4134 spin_unlock(&root->fs_info->trans_lock);
4136 btrfs_cleanup_one_transaction(t, root);
4138 spin_lock(&root->fs_info->trans_lock);
4139 if (t == root->fs_info->running_transaction)
4140 root->fs_info->running_transaction = NULL;
4141 list_del_init(&t->list);
4142 spin_unlock(&root->fs_info->trans_lock);
4144 btrfs_put_transaction(t);
4145 trace_btrfs_transaction_commit(root);
4146 spin_lock(&root->fs_info->trans_lock);
4148 spin_unlock(&root->fs_info->trans_lock);
4149 btrfs_destroy_all_ordered_extents(root->fs_info);
4150 btrfs_destroy_delayed_inodes(root);
4151 btrfs_assert_delayed_root_empty(root);
4152 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4153 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4154 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4159 static struct extent_io_ops btree_extent_io_ops = {
4160 .readpage_end_io_hook = btree_readpage_end_io_hook,
4161 .readpage_io_failed_hook = btree_io_failed_hook,
4162 .submit_bio_hook = btree_submit_bio_hook,
4163 /* note we're sharing with inode.c for the merge bio hook */
4164 .merge_bio_hook = btrfs_merge_bio_hook,