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/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
51 #include <asm/cpufeature.h>
54 static struct extent_io_ops btree_extent_io_ops;
55 static void end_workqueue_fn(struct btrfs_work *work);
56 static void free_fs_root(struct btrfs_root *root);
57 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
59 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
60 struct btrfs_root *root);
61 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
62 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
63 struct btrfs_root *root);
64 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
67 struct extent_io_tree *dirty_pages,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
70 struct extent_io_tree *pinned_extents);
73 * end_io_wq structs are used to do processing in task context when an IO is
74 * complete. This is used during reads to verify checksums, and it is used
75 * by writes to insert metadata for new file extents after IO is complete.
81 struct btrfs_fs_info *info;
84 struct list_head list;
85 struct btrfs_work work;
89 * async submit bios are used to offload expensive checksumming
90 * onto the worker threads. They checksum file and metadata bios
91 * just before they are sent down the IO stack.
93 struct async_submit_bio {
96 struct list_head list;
97 extent_submit_bio_hook_t *submit_bio_start;
98 extent_submit_bio_hook_t *submit_bio_done;
101 unsigned long bio_flags;
103 * bio_offset is optional, can be used if the pages in the bio
104 * can't tell us where in the file the bio should go
107 struct btrfs_work work;
112 * Lockdep class keys for extent_buffer->lock's in this root. For a given
113 * eb, the lockdep key is determined by the btrfs_root it belongs to and
114 * the level the eb occupies in the tree.
116 * Different roots are used for different purposes and may nest inside each
117 * other and they require separate keysets. As lockdep keys should be
118 * static, assign keysets according to the purpose of the root as indicated
119 * by btrfs_root->objectid. This ensures that all special purpose roots
120 * have separate keysets.
122 * Lock-nesting across peer nodes is always done with the immediate parent
123 * node locked thus preventing deadlock. As lockdep doesn't know this, use
124 * subclass to avoid triggering lockdep warning in such cases.
126 * The key is set by the readpage_end_io_hook after the buffer has passed
127 * csum validation but before the pages are unlocked. It is also set by
128 * btrfs_init_new_buffer on freshly allocated blocks.
130 * We also add a check to make sure the highest level of the tree is the
131 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
132 * needs update as well.
134 #ifdef CONFIG_DEBUG_LOCK_ALLOC
135 # if BTRFS_MAX_LEVEL != 8
139 static struct btrfs_lockdep_keyset {
140 u64 id; /* root objectid */
141 const char *name_stem; /* lock name stem */
142 char names[BTRFS_MAX_LEVEL + 1][20];
143 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
144 } btrfs_lockdep_keysets[] = {
145 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
146 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
147 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
148 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
149 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
150 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
151 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
152 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
153 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
154 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
155 { .id = 0, .name_stem = "tree" },
158 void __init btrfs_init_lockdep(void)
162 /* initialize lockdep class names */
163 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
164 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
166 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
167 snprintf(ks->names[j], sizeof(ks->names[j]),
168 "btrfs-%s-%02d", ks->name_stem, j);
172 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
175 struct btrfs_lockdep_keyset *ks;
177 BUG_ON(level >= ARRAY_SIZE(ks->keys));
179 /* find the matching keyset, id 0 is the default entry */
180 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
181 if (ks->id == objectid)
184 lockdep_set_class_and_name(&eb->lock,
185 &ks->keys[level], ks->names[level]);
191 * extents on the btree inode are pretty simple, there's one extent
192 * that covers the entire device
194 static struct extent_map *btree_get_extent(struct inode *inode,
195 struct page *page, size_t pg_offset, u64 start, u64 len,
198 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
199 struct extent_map *em;
202 read_lock(&em_tree->lock);
203 em = lookup_extent_mapping(em_tree, start, len);
206 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
207 read_unlock(&em_tree->lock);
210 read_unlock(&em_tree->lock);
212 em = alloc_extent_map();
214 em = ERR_PTR(-ENOMEM);
219 em->block_len = (u64)-1;
221 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
223 write_lock(&em_tree->lock);
224 ret = add_extent_mapping(em_tree, em);
225 if (ret == -EEXIST) {
227 em = lookup_extent_mapping(em_tree, start, len);
234 write_unlock(&em_tree->lock);
240 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
242 return crc32c(seed, data, len);
245 void btrfs_csum_final(u32 crc, char *result)
247 put_unaligned_le32(~crc, result);
251 * compute the csum for a btree block, and either verify it or write it
252 * into the csum field of the block.
254 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
257 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
260 unsigned long cur_len;
261 unsigned long offset = BTRFS_CSUM_SIZE;
263 unsigned long map_start;
264 unsigned long map_len;
267 unsigned long inline_result;
269 len = buf->len - offset;
271 err = map_private_extent_buffer(buf, offset, 32,
272 &kaddr, &map_start, &map_len);
275 cur_len = min(len, map_len - (offset - map_start));
276 crc = btrfs_csum_data(root, kaddr + offset - map_start,
281 if (csum_size > sizeof(inline_result)) {
282 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
286 result = (char *)&inline_result;
289 btrfs_csum_final(crc, result);
292 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
295 memcpy(&found, result, csum_size);
297 read_extent_buffer(buf, &val, 0, csum_size);
298 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
299 "failed on %llu wanted %X found %X "
301 root->fs_info->sb->s_id,
302 (unsigned long long)buf->start, val, found,
303 btrfs_header_level(buf));
304 if (result != (char *)&inline_result)
309 write_extent_buffer(buf, result, 0, csum_size);
311 if (result != (char *)&inline_result)
317 * we can't consider a given block up to date unless the transid of the
318 * block matches the transid in the parent node's pointer. This is how we
319 * detect blocks that either didn't get written at all or got written
320 * in the wrong place.
322 static int verify_parent_transid(struct extent_io_tree *io_tree,
323 struct extent_buffer *eb, u64 parent_transid,
326 struct extent_state *cached_state = NULL;
329 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
335 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
337 if (extent_buffer_uptodate(eb) &&
338 btrfs_header_generation(eb) == parent_transid) {
342 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
344 (unsigned long long)eb->start,
345 (unsigned long long)parent_transid,
346 (unsigned long long)btrfs_header_generation(eb));
348 clear_extent_buffer_uptodate(eb);
350 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
351 &cached_state, GFP_NOFS);
356 * helper to read a given tree block, doing retries as required when
357 * the checksums don't match and we have alternate mirrors to try.
359 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
360 struct extent_buffer *eb,
361 u64 start, u64 parent_transid)
363 struct extent_io_tree *io_tree;
368 int failed_mirror = 0;
370 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
371 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
373 ret = read_extent_buffer_pages(io_tree, eb, start,
375 btree_get_extent, mirror_num);
377 if (!verify_parent_transid(io_tree, eb,
385 * This buffer's crc is fine, but its contents are corrupted, so
386 * there is no reason to read the other copies, they won't be
389 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
392 num_copies = btrfs_num_copies(root->fs_info,
397 if (!failed_mirror) {
399 failed_mirror = eb->read_mirror;
403 if (mirror_num == failed_mirror)
406 if (mirror_num > num_copies)
410 if (failed && !ret && failed_mirror)
411 repair_eb_io_failure(root, eb, failed_mirror);
417 * checksum a dirty tree block before IO. This has extra checks to make sure
418 * we only fill in the checksum field in the first page of a multi-page block
421 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
423 struct extent_io_tree *tree;
424 u64 start = page_offset(page);
426 struct extent_buffer *eb;
428 tree = &BTRFS_I(page->mapping->host)->io_tree;
430 eb = (struct extent_buffer *)page->private;
431 if (page != eb->pages[0])
433 found_start = btrfs_header_bytenr(eb);
434 if (found_start != start) {
438 if (!PageUptodate(page)) {
442 csum_tree_block(root, eb, 0);
446 static int check_tree_block_fsid(struct btrfs_root *root,
447 struct extent_buffer *eb)
449 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
450 u8 fsid[BTRFS_UUID_SIZE];
453 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
456 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
460 fs_devices = fs_devices->seed;
465 #define CORRUPT(reason, eb, root, slot) \
466 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
467 "root=%llu, slot=%d\n", reason, \
468 (unsigned long long)btrfs_header_bytenr(eb), \
469 (unsigned long long)root->objectid, slot)
471 static noinline int check_leaf(struct btrfs_root *root,
472 struct extent_buffer *leaf)
474 struct btrfs_key key;
475 struct btrfs_key leaf_key;
476 u32 nritems = btrfs_header_nritems(leaf);
482 /* Check the 0 item */
483 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
484 BTRFS_LEAF_DATA_SIZE(root)) {
485 CORRUPT("invalid item offset size pair", leaf, root, 0);
490 * Check to make sure each items keys are in the correct order and their
491 * offsets make sense. We only have to loop through nritems-1 because
492 * we check the current slot against the next slot, which verifies the
493 * next slot's offset+size makes sense and that the current's slot
496 for (slot = 0; slot < nritems - 1; slot++) {
497 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
498 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
500 /* Make sure the keys are in the right order */
501 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
502 CORRUPT("bad key order", leaf, root, slot);
507 * Make sure the offset and ends are right, remember that the
508 * item data starts at the end of the leaf and grows towards the
511 if (btrfs_item_offset_nr(leaf, slot) !=
512 btrfs_item_end_nr(leaf, slot + 1)) {
513 CORRUPT("slot offset bad", leaf, root, slot);
518 * Check to make sure that we don't point outside of the leaf,
519 * just incase all the items are consistent to eachother, but
520 * all point outside of the leaf.
522 if (btrfs_item_end_nr(leaf, slot) >
523 BTRFS_LEAF_DATA_SIZE(root)) {
524 CORRUPT("slot end outside of leaf", leaf, root, slot);
532 struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
533 struct page *page, int max_walk)
535 struct extent_buffer *eb;
536 u64 start = page_offset(page);
540 if (start < max_walk)
543 min_start = start - max_walk;
545 while (start >= min_start) {
546 eb = find_extent_buffer(tree, start, 0);
549 * we found an extent buffer and it contains our page
552 if (eb->start <= target &&
553 eb->start + eb->len > target)
556 /* we found an extent buffer that wasn't for us */
557 free_extent_buffer(eb);
562 start -= PAGE_CACHE_SIZE;
567 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
568 struct extent_state *state, int mirror)
570 struct extent_io_tree *tree;
573 struct extent_buffer *eb;
574 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
581 tree = &BTRFS_I(page->mapping->host)->io_tree;
582 eb = (struct extent_buffer *)page->private;
584 /* the pending IO might have been the only thing that kept this buffer
585 * in memory. Make sure we have a ref for all this other checks
587 extent_buffer_get(eb);
589 reads_done = atomic_dec_and_test(&eb->io_pages);
593 eb->read_mirror = mirror;
594 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
599 found_start = btrfs_header_bytenr(eb);
600 if (found_start != eb->start) {
601 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
603 (unsigned long long)found_start,
604 (unsigned long long)eb->start);
608 if (check_tree_block_fsid(root, eb)) {
609 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
610 (unsigned long long)eb->start);
614 found_level = btrfs_header_level(eb);
616 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
619 ret = csum_tree_block(root, eb, 1);
626 * If this is a leaf block and it is corrupt, set the corrupt bit so
627 * that we don't try and read the other copies of this block, just
630 if (found_level == 0 && check_leaf(root, eb)) {
631 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636 set_extent_buffer_uptodate(eb);
638 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
639 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
640 btree_readahead_hook(root, eb, eb->start, ret);
644 clear_extent_buffer_uptodate(eb);
645 free_extent_buffer(eb);
650 static int btree_io_failed_hook(struct page *page, int failed_mirror)
652 struct extent_buffer *eb;
653 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
655 eb = (struct extent_buffer *)page->private;
656 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
657 eb->read_mirror = failed_mirror;
658 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
659 btree_readahead_hook(root, eb, eb->start, -EIO);
660 return -EIO; /* we fixed nothing */
663 static void end_workqueue_bio(struct bio *bio, int err)
665 struct end_io_wq *end_io_wq = bio->bi_private;
666 struct btrfs_fs_info *fs_info;
668 fs_info = end_io_wq->info;
669 end_io_wq->error = err;
670 end_io_wq->work.func = end_workqueue_fn;
671 end_io_wq->work.flags = 0;
673 if (bio->bi_rw & REQ_WRITE) {
674 if (end_io_wq->metadata == 1)
675 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
677 else if (end_io_wq->metadata == 2)
678 btrfs_queue_worker(&fs_info->endio_freespace_worker,
681 btrfs_queue_worker(&fs_info->endio_write_workers,
684 if (end_io_wq->metadata)
685 btrfs_queue_worker(&fs_info->endio_meta_workers,
688 btrfs_queue_worker(&fs_info->endio_workers,
694 * For the metadata arg you want
697 * 1 - if normal metadta
698 * 2 - if writing to the free space cache area
700 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
703 struct end_io_wq *end_io_wq;
704 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
708 end_io_wq->private = bio->bi_private;
709 end_io_wq->end_io = bio->bi_end_io;
710 end_io_wq->info = info;
711 end_io_wq->error = 0;
712 end_io_wq->bio = bio;
713 end_io_wq->metadata = metadata;
715 bio->bi_private = end_io_wq;
716 bio->bi_end_io = end_workqueue_bio;
720 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
722 unsigned long limit = min_t(unsigned long,
723 info->workers.max_workers,
724 info->fs_devices->open_devices);
728 static void run_one_async_start(struct btrfs_work *work)
730 struct async_submit_bio *async;
733 async = container_of(work, struct async_submit_bio, work);
734 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
735 async->mirror_num, async->bio_flags,
741 static void run_one_async_done(struct btrfs_work *work)
743 struct btrfs_fs_info *fs_info;
744 struct async_submit_bio *async;
747 async = container_of(work, struct async_submit_bio, work);
748 fs_info = BTRFS_I(async->inode)->root->fs_info;
750 limit = btrfs_async_submit_limit(fs_info);
751 limit = limit * 2 / 3;
753 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
754 waitqueue_active(&fs_info->async_submit_wait))
755 wake_up(&fs_info->async_submit_wait);
757 /* If an error occured we just want to clean up the bio and move on */
759 bio_endio(async->bio, async->error);
763 async->submit_bio_done(async->inode, async->rw, async->bio,
764 async->mirror_num, async->bio_flags,
768 static void run_one_async_free(struct btrfs_work *work)
770 struct async_submit_bio *async;
772 async = container_of(work, struct async_submit_bio, work);
776 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
777 int rw, struct bio *bio, int mirror_num,
778 unsigned long bio_flags,
780 extent_submit_bio_hook_t *submit_bio_start,
781 extent_submit_bio_hook_t *submit_bio_done)
783 struct async_submit_bio *async;
785 async = kmalloc(sizeof(*async), GFP_NOFS);
789 async->inode = inode;
792 async->mirror_num = mirror_num;
793 async->submit_bio_start = submit_bio_start;
794 async->submit_bio_done = submit_bio_done;
796 async->work.func = run_one_async_start;
797 async->work.ordered_func = run_one_async_done;
798 async->work.ordered_free = run_one_async_free;
800 async->work.flags = 0;
801 async->bio_flags = bio_flags;
802 async->bio_offset = bio_offset;
806 atomic_inc(&fs_info->nr_async_submits);
809 btrfs_set_work_high_prio(&async->work);
811 btrfs_queue_worker(&fs_info->workers, &async->work);
813 while (atomic_read(&fs_info->async_submit_draining) &&
814 atomic_read(&fs_info->nr_async_submits)) {
815 wait_event(fs_info->async_submit_wait,
816 (atomic_read(&fs_info->nr_async_submits) == 0));
822 static int btree_csum_one_bio(struct bio *bio)
824 struct bio_vec *bvec = bio->bi_io_vec;
826 struct btrfs_root *root;
829 WARN_ON(bio->bi_vcnt <= 0);
830 while (bio_index < bio->bi_vcnt) {
831 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
832 ret = csum_dirty_buffer(root, bvec->bv_page);
841 static int __btree_submit_bio_start(struct inode *inode, int rw,
842 struct bio *bio, int mirror_num,
843 unsigned long bio_flags,
847 * when we're called for a write, we're already in the async
848 * submission context. Just jump into btrfs_map_bio
850 return btree_csum_one_bio(bio);
853 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
854 int mirror_num, unsigned long bio_flags,
860 * when we're called for a write, we're already in the async
861 * submission context. Just jump into btrfs_map_bio
863 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
869 static int check_async_write(struct inode *inode, unsigned long bio_flags)
871 if (bio_flags & EXTENT_BIO_TREE_LOG)
880 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
881 int mirror_num, unsigned long bio_flags,
884 int async = check_async_write(inode, bio_flags);
887 if (!(rw & REQ_WRITE)) {
889 * called for a read, do the setup so that checksum validation
890 * can happen in the async kernel threads
892 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
896 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
899 ret = btree_csum_one_bio(bio);
902 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
906 * kthread helpers are used to submit writes so that
907 * checksumming can happen in parallel across all CPUs
909 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
910 inode, rw, bio, mirror_num, 0,
912 __btree_submit_bio_start,
913 __btree_submit_bio_done);
923 #ifdef CONFIG_MIGRATION
924 static int btree_migratepage(struct address_space *mapping,
925 struct page *newpage, struct page *page,
926 enum migrate_mode mode)
929 * we can't safely write a btree page from here,
930 * we haven't done the locking hook
935 * Buffers may be managed in a filesystem specific way.
936 * We must have no buffers or drop them.
938 if (page_has_private(page) &&
939 !try_to_release_page(page, GFP_KERNEL))
941 return migrate_page(mapping, newpage, page, mode);
946 static int btree_writepages(struct address_space *mapping,
947 struct writeback_control *wbc)
949 struct extent_io_tree *tree;
950 struct btrfs_fs_info *fs_info;
953 tree = &BTRFS_I(mapping->host)->io_tree;
954 if (wbc->sync_mode == WB_SYNC_NONE) {
956 if (wbc->for_kupdate)
959 fs_info = BTRFS_I(mapping->host)->root->fs_info;
960 /* this is a bit racy, but that's ok */
961 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
962 BTRFS_DIRTY_METADATA_THRESH);
966 return btree_write_cache_pages(mapping, wbc);
969 static int btree_readpage(struct file *file, struct page *page)
971 struct extent_io_tree *tree;
972 tree = &BTRFS_I(page->mapping->host)->io_tree;
973 return extent_read_full_page(tree, page, btree_get_extent, 0);
976 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
978 if (PageWriteback(page) || PageDirty(page))
981 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
982 * slab allocation from alloc_extent_state down the callchain where
983 * it'd hit a BUG_ON as those flags are not allowed.
985 gfp_flags &= ~GFP_SLAB_BUG_MASK;
987 return try_release_extent_buffer(page, gfp_flags);
990 static void btree_invalidatepage(struct page *page, unsigned long offset)
992 struct extent_io_tree *tree;
993 tree = &BTRFS_I(page->mapping->host)->io_tree;
994 extent_invalidatepage(tree, page, offset);
995 btree_releasepage(page, GFP_NOFS);
996 if (PagePrivate(page)) {
997 printk(KERN_WARNING "btrfs warning page private not zero "
998 "on page %llu\n", (unsigned long long)page_offset(page));
999 ClearPagePrivate(page);
1000 set_page_private(page, 0);
1001 page_cache_release(page);
1005 static int btree_set_page_dirty(struct page *page)
1008 struct extent_buffer *eb;
1010 BUG_ON(!PagePrivate(page));
1011 eb = (struct extent_buffer *)page->private;
1013 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1014 BUG_ON(!atomic_read(&eb->refs));
1015 btrfs_assert_tree_locked(eb);
1017 return __set_page_dirty_nobuffers(page);
1020 static const struct address_space_operations btree_aops = {
1021 .readpage = btree_readpage,
1022 .writepages = btree_writepages,
1023 .releasepage = btree_releasepage,
1024 .invalidatepage = btree_invalidatepage,
1025 #ifdef CONFIG_MIGRATION
1026 .migratepage = btree_migratepage,
1028 .set_page_dirty = btree_set_page_dirty,
1031 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1034 struct extent_buffer *buf = NULL;
1035 struct inode *btree_inode = root->fs_info->btree_inode;
1038 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1041 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1042 buf, 0, WAIT_NONE, btree_get_extent, 0);
1043 free_extent_buffer(buf);
1047 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1048 int mirror_num, struct extent_buffer **eb)
1050 struct extent_buffer *buf = NULL;
1051 struct inode *btree_inode = root->fs_info->btree_inode;
1052 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1055 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1059 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1061 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1062 btree_get_extent, mirror_num);
1064 free_extent_buffer(buf);
1068 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1069 free_extent_buffer(buf);
1071 } else if (extent_buffer_uptodate(buf)) {
1074 free_extent_buffer(buf);
1079 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1080 u64 bytenr, u32 blocksize)
1082 struct inode *btree_inode = root->fs_info->btree_inode;
1083 struct extent_buffer *eb;
1084 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1089 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1090 u64 bytenr, u32 blocksize)
1092 struct inode *btree_inode = root->fs_info->btree_inode;
1093 struct extent_buffer *eb;
1095 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1101 int btrfs_write_tree_block(struct extent_buffer *buf)
1103 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1104 buf->start + buf->len - 1);
1107 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1109 return filemap_fdatawait_range(buf->pages[0]->mapping,
1110 buf->start, buf->start + buf->len - 1);
1113 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1114 u32 blocksize, u64 parent_transid)
1116 struct extent_buffer *buf = NULL;
1119 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1123 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1128 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1129 struct extent_buffer *buf)
1131 struct btrfs_fs_info *fs_info = root->fs_info;
1133 if (btrfs_header_generation(buf) ==
1134 fs_info->running_transaction->transid) {
1135 btrfs_assert_tree_locked(buf);
1137 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1138 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1140 fs_info->dirty_metadata_batch);
1141 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1142 btrfs_set_lock_blocking(buf);
1143 clear_extent_buffer_dirty(buf);
1148 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1149 u32 stripesize, struct btrfs_root *root,
1150 struct btrfs_fs_info *fs_info,
1154 root->commit_root = NULL;
1155 root->sectorsize = sectorsize;
1156 root->nodesize = nodesize;
1157 root->leafsize = leafsize;
1158 root->stripesize = stripesize;
1160 root->track_dirty = 0;
1162 root->orphan_item_inserted = 0;
1163 root->orphan_cleanup_state = 0;
1165 root->objectid = objectid;
1166 root->last_trans = 0;
1167 root->highest_objectid = 0;
1169 root->inode_tree = RB_ROOT;
1170 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1171 root->block_rsv = NULL;
1172 root->orphan_block_rsv = NULL;
1174 INIT_LIST_HEAD(&root->dirty_list);
1175 INIT_LIST_HEAD(&root->root_list);
1176 INIT_LIST_HEAD(&root->logged_list[0]);
1177 INIT_LIST_HEAD(&root->logged_list[1]);
1178 spin_lock_init(&root->orphan_lock);
1179 spin_lock_init(&root->inode_lock);
1180 spin_lock_init(&root->accounting_lock);
1181 spin_lock_init(&root->log_extents_lock[0]);
1182 spin_lock_init(&root->log_extents_lock[1]);
1183 mutex_init(&root->objectid_mutex);
1184 mutex_init(&root->log_mutex);
1185 init_waitqueue_head(&root->log_writer_wait);
1186 init_waitqueue_head(&root->log_commit_wait[0]);
1187 init_waitqueue_head(&root->log_commit_wait[1]);
1188 atomic_set(&root->log_commit[0], 0);
1189 atomic_set(&root->log_commit[1], 0);
1190 atomic_set(&root->log_writers, 0);
1191 atomic_set(&root->log_batch, 0);
1192 atomic_set(&root->orphan_inodes, 0);
1193 root->log_transid = 0;
1194 root->last_log_commit = 0;
1195 extent_io_tree_init(&root->dirty_log_pages,
1196 fs_info->btree_inode->i_mapping);
1198 memset(&root->root_key, 0, sizeof(root->root_key));
1199 memset(&root->root_item, 0, sizeof(root->root_item));
1200 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1201 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1202 root->defrag_trans_start = fs_info->generation;
1203 init_completion(&root->kobj_unregister);
1204 root->defrag_running = 0;
1205 root->root_key.objectid = objectid;
1208 spin_lock_init(&root->root_item_lock);
1211 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1212 struct btrfs_fs_info *fs_info,
1214 struct btrfs_root *root)
1220 __setup_root(tree_root->nodesize, tree_root->leafsize,
1221 tree_root->sectorsize, tree_root->stripesize,
1222 root, fs_info, objectid);
1223 ret = btrfs_find_last_root(tree_root, objectid,
1224 &root->root_item, &root->root_key);
1230 generation = btrfs_root_generation(&root->root_item);
1231 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1232 root->commit_root = NULL;
1233 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1234 blocksize, generation);
1235 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1236 free_extent_buffer(root->node);
1240 root->commit_root = btrfs_root_node(root);
1244 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1246 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1248 root->fs_info = fs_info;
1252 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1253 struct btrfs_fs_info *fs_info,
1256 struct extent_buffer *leaf;
1257 struct btrfs_root *tree_root = fs_info->tree_root;
1258 struct btrfs_root *root;
1259 struct btrfs_key key;
1263 root = btrfs_alloc_root(fs_info);
1265 return ERR_PTR(-ENOMEM);
1267 __setup_root(tree_root->nodesize, tree_root->leafsize,
1268 tree_root->sectorsize, tree_root->stripesize,
1269 root, fs_info, objectid);
1270 root->root_key.objectid = objectid;
1271 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1272 root->root_key.offset = 0;
1274 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1275 0, objectid, NULL, 0, 0, 0);
1277 ret = PTR_ERR(leaf);
1281 bytenr = leaf->start;
1282 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1283 btrfs_set_header_bytenr(leaf, leaf->start);
1284 btrfs_set_header_generation(leaf, trans->transid);
1285 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1286 btrfs_set_header_owner(leaf, objectid);
1289 write_extent_buffer(leaf, fs_info->fsid,
1290 (unsigned long)btrfs_header_fsid(leaf),
1292 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1293 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1295 btrfs_mark_buffer_dirty(leaf);
1297 root->commit_root = btrfs_root_node(root);
1298 root->track_dirty = 1;
1301 root->root_item.flags = 0;
1302 root->root_item.byte_limit = 0;
1303 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1304 btrfs_set_root_generation(&root->root_item, trans->transid);
1305 btrfs_set_root_level(&root->root_item, 0);
1306 btrfs_set_root_refs(&root->root_item, 1);
1307 btrfs_set_root_used(&root->root_item, leaf->len);
1308 btrfs_set_root_last_snapshot(&root->root_item, 0);
1309 btrfs_set_root_dirid(&root->root_item, 0);
1310 root->root_item.drop_level = 0;
1312 key.objectid = objectid;
1313 key.type = BTRFS_ROOT_ITEM_KEY;
1315 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1319 btrfs_tree_unlock(leaf);
1323 return ERR_PTR(ret);
1328 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1329 struct btrfs_fs_info *fs_info)
1331 struct btrfs_root *root;
1332 struct btrfs_root *tree_root = fs_info->tree_root;
1333 struct extent_buffer *leaf;
1335 root = btrfs_alloc_root(fs_info);
1337 return ERR_PTR(-ENOMEM);
1339 __setup_root(tree_root->nodesize, tree_root->leafsize,
1340 tree_root->sectorsize, tree_root->stripesize,
1341 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1343 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1344 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1345 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1347 * log trees do not get reference counted because they go away
1348 * before a real commit is actually done. They do store pointers
1349 * to file data extents, and those reference counts still get
1350 * updated (along with back refs to the log tree).
1354 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1355 BTRFS_TREE_LOG_OBJECTID, NULL,
1359 return ERR_CAST(leaf);
1362 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1363 btrfs_set_header_bytenr(leaf, leaf->start);
1364 btrfs_set_header_generation(leaf, trans->transid);
1365 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1366 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1369 write_extent_buffer(root->node, root->fs_info->fsid,
1370 (unsigned long)btrfs_header_fsid(root->node),
1372 btrfs_mark_buffer_dirty(root->node);
1373 btrfs_tree_unlock(root->node);
1377 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1378 struct btrfs_fs_info *fs_info)
1380 struct btrfs_root *log_root;
1382 log_root = alloc_log_tree(trans, fs_info);
1383 if (IS_ERR(log_root))
1384 return PTR_ERR(log_root);
1385 WARN_ON(fs_info->log_root_tree);
1386 fs_info->log_root_tree = log_root;
1390 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1391 struct btrfs_root *root)
1393 struct btrfs_root *log_root;
1394 struct btrfs_inode_item *inode_item;
1396 log_root = alloc_log_tree(trans, root->fs_info);
1397 if (IS_ERR(log_root))
1398 return PTR_ERR(log_root);
1400 log_root->last_trans = trans->transid;
1401 log_root->root_key.offset = root->root_key.objectid;
1403 inode_item = &log_root->root_item.inode;
1404 inode_item->generation = cpu_to_le64(1);
1405 inode_item->size = cpu_to_le64(3);
1406 inode_item->nlink = cpu_to_le32(1);
1407 inode_item->nbytes = cpu_to_le64(root->leafsize);
1408 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1410 btrfs_set_root_node(&log_root->root_item, log_root->node);
1412 WARN_ON(root->log_root);
1413 root->log_root = log_root;
1414 root->log_transid = 0;
1415 root->last_log_commit = 0;
1419 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1420 struct btrfs_key *location)
1422 struct btrfs_root *root;
1423 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1424 struct btrfs_path *path;
1425 struct extent_buffer *l;
1431 root = btrfs_alloc_root(fs_info);
1433 return ERR_PTR(-ENOMEM);
1434 if (location->offset == (u64)-1) {
1435 ret = find_and_setup_root(tree_root, fs_info,
1436 location->objectid, root);
1439 return ERR_PTR(ret);
1444 __setup_root(tree_root->nodesize, tree_root->leafsize,
1445 tree_root->sectorsize, tree_root->stripesize,
1446 root, fs_info, location->objectid);
1448 path = btrfs_alloc_path();
1451 return ERR_PTR(-ENOMEM);
1453 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1456 slot = path->slots[0];
1457 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1458 memcpy(&root->root_key, location, sizeof(*location));
1460 btrfs_free_path(path);
1465 return ERR_PTR(ret);
1468 generation = btrfs_root_generation(&root->root_item);
1469 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1470 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1471 blocksize, generation);
1472 root->commit_root = btrfs_root_node(root);
1473 BUG_ON(!root->node); /* -ENOMEM */
1475 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1477 btrfs_check_and_init_root_item(&root->root_item);
1483 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1484 struct btrfs_key *location)
1486 struct btrfs_root *root;
1489 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1490 return fs_info->tree_root;
1491 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1492 return fs_info->extent_root;
1493 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1494 return fs_info->chunk_root;
1495 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1496 return fs_info->dev_root;
1497 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1498 return fs_info->csum_root;
1499 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1500 return fs_info->quota_root ? fs_info->quota_root :
1503 spin_lock(&fs_info->fs_roots_radix_lock);
1504 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1505 (unsigned long)location->objectid);
1506 spin_unlock(&fs_info->fs_roots_radix_lock);
1510 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1514 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1515 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1517 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1522 btrfs_init_free_ino_ctl(root);
1523 mutex_init(&root->fs_commit_mutex);
1524 spin_lock_init(&root->cache_lock);
1525 init_waitqueue_head(&root->cache_wait);
1527 ret = get_anon_bdev(&root->anon_dev);
1531 if (btrfs_root_refs(&root->root_item) == 0) {
1536 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1540 root->orphan_item_inserted = 1;
1542 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1546 spin_lock(&fs_info->fs_roots_radix_lock);
1547 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1548 (unsigned long)root->root_key.objectid,
1553 spin_unlock(&fs_info->fs_roots_radix_lock);
1554 radix_tree_preload_end();
1556 if (ret == -EEXIST) {
1563 ret = btrfs_find_dead_roots(fs_info->tree_root,
1564 root->root_key.objectid);
1569 return ERR_PTR(ret);
1572 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1574 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1576 struct btrfs_device *device;
1577 struct backing_dev_info *bdi;
1580 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1583 bdi = blk_get_backing_dev_info(device->bdev);
1584 if (bdi && bdi_congested(bdi, bdi_bits)) {
1594 * If this fails, caller must call bdi_destroy() to get rid of the
1597 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1601 bdi->capabilities = BDI_CAP_MAP_COPY;
1602 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1606 bdi->ra_pages = default_backing_dev_info.ra_pages;
1607 bdi->congested_fn = btrfs_congested_fn;
1608 bdi->congested_data = info;
1613 * called by the kthread helper functions to finally call the bio end_io
1614 * functions. This is where read checksum verification actually happens
1616 static void end_workqueue_fn(struct btrfs_work *work)
1619 struct end_io_wq *end_io_wq;
1620 struct btrfs_fs_info *fs_info;
1623 end_io_wq = container_of(work, struct end_io_wq, work);
1624 bio = end_io_wq->bio;
1625 fs_info = end_io_wq->info;
1627 error = end_io_wq->error;
1628 bio->bi_private = end_io_wq->private;
1629 bio->bi_end_io = end_io_wq->end_io;
1631 bio_endio(bio, error);
1634 static int cleaner_kthread(void *arg)
1636 struct btrfs_root *root = arg;
1639 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1640 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1641 btrfs_run_delayed_iputs(root);
1642 btrfs_clean_old_snapshots(root);
1643 mutex_unlock(&root->fs_info->cleaner_mutex);
1644 btrfs_run_defrag_inodes(root->fs_info);
1647 if (!try_to_freeze()) {
1648 set_current_state(TASK_INTERRUPTIBLE);
1649 if (!kthread_should_stop())
1651 __set_current_state(TASK_RUNNING);
1653 } while (!kthread_should_stop());
1657 static int transaction_kthread(void *arg)
1659 struct btrfs_root *root = arg;
1660 struct btrfs_trans_handle *trans;
1661 struct btrfs_transaction *cur;
1664 unsigned long delay;
1668 cannot_commit = false;
1670 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1672 spin_lock(&root->fs_info->trans_lock);
1673 cur = root->fs_info->running_transaction;
1675 spin_unlock(&root->fs_info->trans_lock);
1679 now = get_seconds();
1680 if (!cur->blocked &&
1681 (now < cur->start_time || now - cur->start_time < 30)) {
1682 spin_unlock(&root->fs_info->trans_lock);
1686 transid = cur->transid;
1687 spin_unlock(&root->fs_info->trans_lock);
1689 /* If the file system is aborted, this will always fail. */
1690 trans = btrfs_attach_transaction(root);
1691 if (IS_ERR(trans)) {
1692 if (PTR_ERR(trans) != -ENOENT)
1693 cannot_commit = true;
1696 if (transid == trans->transid) {
1697 btrfs_commit_transaction(trans, root);
1699 btrfs_end_transaction(trans, root);
1702 wake_up_process(root->fs_info->cleaner_kthread);
1703 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1705 if (!try_to_freeze()) {
1706 set_current_state(TASK_INTERRUPTIBLE);
1707 if (!kthread_should_stop() &&
1708 (!btrfs_transaction_blocked(root->fs_info) ||
1710 schedule_timeout(delay);
1711 __set_current_state(TASK_RUNNING);
1713 } while (!kthread_should_stop());
1718 * this will find the highest generation in the array of
1719 * root backups. The index of the highest array is returned,
1720 * or -1 if we can't find anything.
1722 * We check to make sure the array is valid by comparing the
1723 * generation of the latest root in the array with the generation
1724 * in the super block. If they don't match we pitch it.
1726 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1729 int newest_index = -1;
1730 struct btrfs_root_backup *root_backup;
1733 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1734 root_backup = info->super_copy->super_roots + i;
1735 cur = btrfs_backup_tree_root_gen(root_backup);
1736 if (cur == newest_gen)
1740 /* check to see if we actually wrapped around */
1741 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1742 root_backup = info->super_copy->super_roots;
1743 cur = btrfs_backup_tree_root_gen(root_backup);
1744 if (cur == newest_gen)
1747 return newest_index;
1752 * find the oldest backup so we know where to store new entries
1753 * in the backup array. This will set the backup_root_index
1754 * field in the fs_info struct
1756 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1759 int newest_index = -1;
1761 newest_index = find_newest_super_backup(info, newest_gen);
1762 /* if there was garbage in there, just move along */
1763 if (newest_index == -1) {
1764 info->backup_root_index = 0;
1766 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1771 * copy all the root pointers into the super backup array.
1772 * this will bump the backup pointer by one when it is
1775 static void backup_super_roots(struct btrfs_fs_info *info)
1778 struct btrfs_root_backup *root_backup;
1781 next_backup = info->backup_root_index;
1782 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1783 BTRFS_NUM_BACKUP_ROOTS;
1786 * just overwrite the last backup if we're at the same generation
1787 * this happens only at umount
1789 root_backup = info->super_for_commit->super_roots + last_backup;
1790 if (btrfs_backup_tree_root_gen(root_backup) ==
1791 btrfs_header_generation(info->tree_root->node))
1792 next_backup = last_backup;
1794 root_backup = info->super_for_commit->super_roots + next_backup;
1797 * make sure all of our padding and empty slots get zero filled
1798 * regardless of which ones we use today
1800 memset(root_backup, 0, sizeof(*root_backup));
1802 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1804 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1805 btrfs_set_backup_tree_root_gen(root_backup,
1806 btrfs_header_generation(info->tree_root->node));
1808 btrfs_set_backup_tree_root_level(root_backup,
1809 btrfs_header_level(info->tree_root->node));
1811 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1812 btrfs_set_backup_chunk_root_gen(root_backup,
1813 btrfs_header_generation(info->chunk_root->node));
1814 btrfs_set_backup_chunk_root_level(root_backup,
1815 btrfs_header_level(info->chunk_root->node));
1817 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1818 btrfs_set_backup_extent_root_gen(root_backup,
1819 btrfs_header_generation(info->extent_root->node));
1820 btrfs_set_backup_extent_root_level(root_backup,
1821 btrfs_header_level(info->extent_root->node));
1824 * we might commit during log recovery, which happens before we set
1825 * the fs_root. Make sure it is valid before we fill it in.
1827 if (info->fs_root && info->fs_root->node) {
1828 btrfs_set_backup_fs_root(root_backup,
1829 info->fs_root->node->start);
1830 btrfs_set_backup_fs_root_gen(root_backup,
1831 btrfs_header_generation(info->fs_root->node));
1832 btrfs_set_backup_fs_root_level(root_backup,
1833 btrfs_header_level(info->fs_root->node));
1836 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1837 btrfs_set_backup_dev_root_gen(root_backup,
1838 btrfs_header_generation(info->dev_root->node));
1839 btrfs_set_backup_dev_root_level(root_backup,
1840 btrfs_header_level(info->dev_root->node));
1842 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1843 btrfs_set_backup_csum_root_gen(root_backup,
1844 btrfs_header_generation(info->csum_root->node));
1845 btrfs_set_backup_csum_root_level(root_backup,
1846 btrfs_header_level(info->csum_root->node));
1848 btrfs_set_backup_total_bytes(root_backup,
1849 btrfs_super_total_bytes(info->super_copy));
1850 btrfs_set_backup_bytes_used(root_backup,
1851 btrfs_super_bytes_used(info->super_copy));
1852 btrfs_set_backup_num_devices(root_backup,
1853 btrfs_super_num_devices(info->super_copy));
1856 * if we don't copy this out to the super_copy, it won't get remembered
1857 * for the next commit
1859 memcpy(&info->super_copy->super_roots,
1860 &info->super_for_commit->super_roots,
1861 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1865 * this copies info out of the root backup array and back into
1866 * the in-memory super block. It is meant to help iterate through
1867 * the array, so you send it the number of backups you've already
1868 * tried and the last backup index you used.
1870 * this returns -1 when it has tried all the backups
1872 static noinline int next_root_backup(struct btrfs_fs_info *info,
1873 struct btrfs_super_block *super,
1874 int *num_backups_tried, int *backup_index)
1876 struct btrfs_root_backup *root_backup;
1877 int newest = *backup_index;
1879 if (*num_backups_tried == 0) {
1880 u64 gen = btrfs_super_generation(super);
1882 newest = find_newest_super_backup(info, gen);
1886 *backup_index = newest;
1887 *num_backups_tried = 1;
1888 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1889 /* we've tried all the backups, all done */
1892 /* jump to the next oldest backup */
1893 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1894 BTRFS_NUM_BACKUP_ROOTS;
1895 *backup_index = newest;
1896 *num_backups_tried += 1;
1898 root_backup = super->super_roots + newest;
1900 btrfs_set_super_generation(super,
1901 btrfs_backup_tree_root_gen(root_backup));
1902 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1903 btrfs_set_super_root_level(super,
1904 btrfs_backup_tree_root_level(root_backup));
1905 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1908 * fixme: the total bytes and num_devices need to match or we should
1911 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1912 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1916 /* helper to cleanup tree roots */
1917 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1919 free_extent_buffer(info->tree_root->node);
1920 free_extent_buffer(info->tree_root->commit_root);
1921 free_extent_buffer(info->dev_root->node);
1922 free_extent_buffer(info->dev_root->commit_root);
1923 free_extent_buffer(info->extent_root->node);
1924 free_extent_buffer(info->extent_root->commit_root);
1925 free_extent_buffer(info->csum_root->node);
1926 free_extent_buffer(info->csum_root->commit_root);
1927 if (info->quota_root) {
1928 free_extent_buffer(info->quota_root->node);
1929 free_extent_buffer(info->quota_root->commit_root);
1932 info->tree_root->node = NULL;
1933 info->tree_root->commit_root = NULL;
1934 info->dev_root->node = NULL;
1935 info->dev_root->commit_root = NULL;
1936 info->extent_root->node = NULL;
1937 info->extent_root->commit_root = NULL;
1938 info->csum_root->node = NULL;
1939 info->csum_root->commit_root = NULL;
1940 if (info->quota_root) {
1941 info->quota_root->node = NULL;
1942 info->quota_root->commit_root = NULL;
1946 free_extent_buffer(info->chunk_root->node);
1947 free_extent_buffer(info->chunk_root->commit_root);
1948 info->chunk_root->node = NULL;
1949 info->chunk_root->commit_root = NULL;
1954 int open_ctree(struct super_block *sb,
1955 struct btrfs_fs_devices *fs_devices,
1965 struct btrfs_key location;
1966 struct buffer_head *bh;
1967 struct btrfs_super_block *disk_super;
1968 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1969 struct btrfs_root *tree_root;
1970 struct btrfs_root *extent_root;
1971 struct btrfs_root *csum_root;
1972 struct btrfs_root *chunk_root;
1973 struct btrfs_root *dev_root;
1974 struct btrfs_root *quota_root;
1975 struct btrfs_root *log_tree_root;
1978 int num_backups_tried = 0;
1979 int backup_index = 0;
1981 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1982 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1983 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1984 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1985 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1986 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
1988 if (!tree_root || !extent_root || !csum_root ||
1989 !chunk_root || !dev_root || !quota_root) {
1994 ret = init_srcu_struct(&fs_info->subvol_srcu);
2000 ret = setup_bdi(fs_info, &fs_info->bdi);
2006 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2011 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2012 (1 + ilog2(nr_cpu_ids));
2014 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2017 goto fail_dirty_metadata_bytes;
2020 fs_info->btree_inode = new_inode(sb);
2021 if (!fs_info->btree_inode) {
2023 goto fail_delalloc_bytes;
2026 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2028 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2029 INIT_LIST_HEAD(&fs_info->trans_list);
2030 INIT_LIST_HEAD(&fs_info->dead_roots);
2031 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2032 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2033 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2034 spin_lock_init(&fs_info->delalloc_lock);
2035 spin_lock_init(&fs_info->trans_lock);
2036 spin_lock_init(&fs_info->fs_roots_radix_lock);
2037 spin_lock_init(&fs_info->delayed_iput_lock);
2038 spin_lock_init(&fs_info->defrag_inodes_lock);
2039 spin_lock_init(&fs_info->free_chunk_lock);
2040 spin_lock_init(&fs_info->tree_mod_seq_lock);
2041 rwlock_init(&fs_info->tree_mod_log_lock);
2042 mutex_init(&fs_info->reloc_mutex);
2043 seqlock_init(&fs_info->profiles_lock);
2045 init_completion(&fs_info->kobj_unregister);
2046 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2047 INIT_LIST_HEAD(&fs_info->space_info);
2048 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2049 btrfs_mapping_init(&fs_info->mapping_tree);
2050 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2051 BTRFS_BLOCK_RSV_GLOBAL);
2052 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2053 BTRFS_BLOCK_RSV_DELALLOC);
2054 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2055 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2056 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2057 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2058 BTRFS_BLOCK_RSV_DELOPS);
2059 atomic_set(&fs_info->nr_async_submits, 0);
2060 atomic_set(&fs_info->async_delalloc_pages, 0);
2061 atomic_set(&fs_info->async_submit_draining, 0);
2062 atomic_set(&fs_info->nr_async_bios, 0);
2063 atomic_set(&fs_info->defrag_running, 0);
2064 atomic_set(&fs_info->tree_mod_seq, 0);
2066 fs_info->max_inline = 8192 * 1024;
2067 fs_info->metadata_ratio = 0;
2068 fs_info->defrag_inodes = RB_ROOT;
2069 fs_info->trans_no_join = 0;
2070 fs_info->free_chunk_space = 0;
2071 fs_info->tree_mod_log = RB_ROOT;
2073 /* readahead state */
2074 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2075 spin_lock_init(&fs_info->reada_lock);
2077 fs_info->thread_pool_size = min_t(unsigned long,
2078 num_online_cpus() + 2, 8);
2080 INIT_LIST_HEAD(&fs_info->ordered_extents);
2081 spin_lock_init(&fs_info->ordered_extent_lock);
2082 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2084 if (!fs_info->delayed_root) {
2088 btrfs_init_delayed_root(fs_info->delayed_root);
2090 mutex_init(&fs_info->scrub_lock);
2091 atomic_set(&fs_info->scrubs_running, 0);
2092 atomic_set(&fs_info->scrub_pause_req, 0);
2093 atomic_set(&fs_info->scrubs_paused, 0);
2094 atomic_set(&fs_info->scrub_cancel_req, 0);
2095 init_waitqueue_head(&fs_info->scrub_pause_wait);
2096 init_rwsem(&fs_info->scrub_super_lock);
2097 fs_info->scrub_workers_refcnt = 0;
2098 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2099 fs_info->check_integrity_print_mask = 0;
2102 spin_lock_init(&fs_info->balance_lock);
2103 mutex_init(&fs_info->balance_mutex);
2104 atomic_set(&fs_info->balance_running, 0);
2105 atomic_set(&fs_info->balance_pause_req, 0);
2106 atomic_set(&fs_info->balance_cancel_req, 0);
2107 fs_info->balance_ctl = NULL;
2108 init_waitqueue_head(&fs_info->balance_wait_q);
2110 sb->s_blocksize = 4096;
2111 sb->s_blocksize_bits = blksize_bits(4096);
2112 sb->s_bdi = &fs_info->bdi;
2114 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2115 set_nlink(fs_info->btree_inode, 1);
2117 * we set the i_size on the btree inode to the max possible int.
2118 * the real end of the address space is determined by all of
2119 * the devices in the system
2121 fs_info->btree_inode->i_size = OFFSET_MAX;
2122 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2123 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2125 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2126 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2127 fs_info->btree_inode->i_mapping);
2128 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2129 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2131 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2133 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2134 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2135 sizeof(struct btrfs_key));
2136 set_bit(BTRFS_INODE_DUMMY,
2137 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2138 insert_inode_hash(fs_info->btree_inode);
2140 spin_lock_init(&fs_info->block_group_cache_lock);
2141 fs_info->block_group_cache_tree = RB_ROOT;
2142 fs_info->first_logical_byte = (u64)-1;
2144 extent_io_tree_init(&fs_info->freed_extents[0],
2145 fs_info->btree_inode->i_mapping);
2146 extent_io_tree_init(&fs_info->freed_extents[1],
2147 fs_info->btree_inode->i_mapping);
2148 fs_info->pinned_extents = &fs_info->freed_extents[0];
2149 fs_info->do_barriers = 1;
2152 mutex_init(&fs_info->ordered_operations_mutex);
2153 mutex_init(&fs_info->tree_log_mutex);
2154 mutex_init(&fs_info->chunk_mutex);
2155 mutex_init(&fs_info->transaction_kthread_mutex);
2156 mutex_init(&fs_info->cleaner_mutex);
2157 mutex_init(&fs_info->volume_mutex);
2158 init_rwsem(&fs_info->extent_commit_sem);
2159 init_rwsem(&fs_info->cleanup_work_sem);
2160 init_rwsem(&fs_info->subvol_sem);
2161 fs_info->dev_replace.lock_owner = 0;
2162 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2163 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2164 mutex_init(&fs_info->dev_replace.lock_management_lock);
2165 mutex_init(&fs_info->dev_replace.lock);
2167 spin_lock_init(&fs_info->qgroup_lock);
2168 fs_info->qgroup_tree = RB_ROOT;
2169 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2170 fs_info->qgroup_seq = 1;
2171 fs_info->quota_enabled = 0;
2172 fs_info->pending_quota_state = 0;
2174 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2175 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2177 init_waitqueue_head(&fs_info->transaction_throttle);
2178 init_waitqueue_head(&fs_info->transaction_wait);
2179 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2180 init_waitqueue_head(&fs_info->async_submit_wait);
2182 __setup_root(4096, 4096, 4096, 4096, tree_root,
2183 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2185 invalidate_bdev(fs_devices->latest_bdev);
2186 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2192 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2193 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2194 sizeof(*fs_info->super_for_commit));
2197 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2199 disk_super = fs_info->super_copy;
2200 if (!btrfs_super_root(disk_super))
2203 /* check FS state, whether FS is broken. */
2204 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2205 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2207 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2209 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2215 * run through our array of backup supers and setup
2216 * our ring pointer to the oldest one
2218 generation = btrfs_super_generation(disk_super);
2219 find_oldest_super_backup(fs_info, generation);
2222 * In the long term, we'll store the compression type in the super
2223 * block, and it'll be used for per file compression control.
2225 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2227 ret = btrfs_parse_options(tree_root, options);
2233 features = btrfs_super_incompat_flags(disk_super) &
2234 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2236 printk(KERN_ERR "BTRFS: couldn't mount because of "
2237 "unsupported optional features (%Lx).\n",
2238 (unsigned long long)features);
2243 if (btrfs_super_leafsize(disk_super) !=
2244 btrfs_super_nodesize(disk_super)) {
2245 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2246 "blocksizes don't match. node %d leaf %d\n",
2247 btrfs_super_nodesize(disk_super),
2248 btrfs_super_leafsize(disk_super));
2252 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2253 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2254 "blocksize (%d) was too large\n",
2255 btrfs_super_leafsize(disk_super));
2260 features = btrfs_super_incompat_flags(disk_super);
2261 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2262 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2263 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2266 * flag our filesystem as having big metadata blocks if
2267 * they are bigger than the page size
2269 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2270 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2271 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2272 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2275 nodesize = btrfs_super_nodesize(disk_super);
2276 leafsize = btrfs_super_leafsize(disk_super);
2277 sectorsize = btrfs_super_sectorsize(disk_super);
2278 stripesize = btrfs_super_stripesize(disk_super);
2279 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2280 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2283 * mixed block groups end up with duplicate but slightly offset
2284 * extent buffers for the same range. It leads to corruptions
2286 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2287 (sectorsize != leafsize)) {
2288 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2289 "are not allowed for mixed block groups on %s\n",
2294 btrfs_set_super_incompat_flags(disk_super, features);
2296 features = btrfs_super_compat_ro_flags(disk_super) &
2297 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2298 if (!(sb->s_flags & MS_RDONLY) && features) {
2299 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2300 "unsupported option features (%Lx).\n",
2301 (unsigned long long)features);
2306 btrfs_init_workers(&fs_info->generic_worker,
2307 "genwork", 1, NULL);
2309 btrfs_init_workers(&fs_info->workers, "worker",
2310 fs_info->thread_pool_size,
2311 &fs_info->generic_worker);
2313 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2314 fs_info->thread_pool_size,
2315 &fs_info->generic_worker);
2317 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2318 fs_info->thread_pool_size,
2319 &fs_info->generic_worker);
2321 btrfs_init_workers(&fs_info->submit_workers, "submit",
2322 min_t(u64, fs_devices->num_devices,
2323 fs_info->thread_pool_size),
2324 &fs_info->generic_worker);
2326 btrfs_init_workers(&fs_info->caching_workers, "cache",
2327 2, &fs_info->generic_worker);
2329 /* a higher idle thresh on the submit workers makes it much more
2330 * likely that bios will be send down in a sane order to the
2333 fs_info->submit_workers.idle_thresh = 64;
2335 fs_info->workers.idle_thresh = 16;
2336 fs_info->workers.ordered = 1;
2338 fs_info->delalloc_workers.idle_thresh = 2;
2339 fs_info->delalloc_workers.ordered = 1;
2341 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2342 &fs_info->generic_worker);
2343 btrfs_init_workers(&fs_info->endio_workers, "endio",
2344 fs_info->thread_pool_size,
2345 &fs_info->generic_worker);
2346 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2347 fs_info->thread_pool_size,
2348 &fs_info->generic_worker);
2349 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2350 "endio-meta-write", fs_info->thread_pool_size,
2351 &fs_info->generic_worker);
2352 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2353 fs_info->thread_pool_size,
2354 &fs_info->generic_worker);
2355 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2356 1, &fs_info->generic_worker);
2357 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2358 fs_info->thread_pool_size,
2359 &fs_info->generic_worker);
2360 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2361 fs_info->thread_pool_size,
2362 &fs_info->generic_worker);
2365 * endios are largely parallel and should have a very
2368 fs_info->endio_workers.idle_thresh = 4;
2369 fs_info->endio_meta_workers.idle_thresh = 4;
2371 fs_info->endio_write_workers.idle_thresh = 2;
2372 fs_info->endio_meta_write_workers.idle_thresh = 2;
2373 fs_info->readahead_workers.idle_thresh = 2;
2376 * btrfs_start_workers can really only fail because of ENOMEM so just
2377 * return -ENOMEM if any of these fail.
2379 ret = btrfs_start_workers(&fs_info->workers);
2380 ret |= btrfs_start_workers(&fs_info->generic_worker);
2381 ret |= btrfs_start_workers(&fs_info->submit_workers);
2382 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2383 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2384 ret |= btrfs_start_workers(&fs_info->endio_workers);
2385 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2386 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2387 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2388 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2389 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2390 ret |= btrfs_start_workers(&fs_info->caching_workers);
2391 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2392 ret |= btrfs_start_workers(&fs_info->flush_workers);
2395 goto fail_sb_buffer;
2398 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2399 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2400 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2402 tree_root->nodesize = nodesize;
2403 tree_root->leafsize = leafsize;
2404 tree_root->sectorsize = sectorsize;
2405 tree_root->stripesize = stripesize;
2407 sb->s_blocksize = sectorsize;
2408 sb->s_blocksize_bits = blksize_bits(sectorsize);
2410 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2411 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2412 goto fail_sb_buffer;
2415 if (sectorsize != PAGE_SIZE) {
2416 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2417 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2418 goto fail_sb_buffer;
2421 mutex_lock(&fs_info->chunk_mutex);
2422 ret = btrfs_read_sys_array(tree_root);
2423 mutex_unlock(&fs_info->chunk_mutex);
2425 printk(KERN_WARNING "btrfs: failed to read the system "
2426 "array on %s\n", sb->s_id);
2427 goto fail_sb_buffer;
2430 blocksize = btrfs_level_size(tree_root,
2431 btrfs_super_chunk_root_level(disk_super));
2432 generation = btrfs_super_chunk_root_generation(disk_super);
2434 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2435 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2437 chunk_root->node = read_tree_block(chunk_root,
2438 btrfs_super_chunk_root(disk_super),
2439 blocksize, generation);
2440 BUG_ON(!chunk_root->node); /* -ENOMEM */
2441 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2442 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2444 goto fail_tree_roots;
2446 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2447 chunk_root->commit_root = btrfs_root_node(chunk_root);
2449 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2450 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2453 ret = btrfs_read_chunk_tree(chunk_root);
2455 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2457 goto fail_tree_roots;
2461 * keep the device that is marked to be the target device for the
2462 * dev_replace procedure
2464 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2466 if (!fs_devices->latest_bdev) {
2467 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2469 goto fail_tree_roots;
2473 blocksize = btrfs_level_size(tree_root,
2474 btrfs_super_root_level(disk_super));
2475 generation = btrfs_super_generation(disk_super);
2477 tree_root->node = read_tree_block(tree_root,
2478 btrfs_super_root(disk_super),
2479 blocksize, generation);
2480 if (!tree_root->node ||
2481 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2482 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2485 goto recovery_tree_root;
2488 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2489 tree_root->commit_root = btrfs_root_node(tree_root);
2491 ret = find_and_setup_root(tree_root, fs_info,
2492 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2494 goto recovery_tree_root;
2495 extent_root->track_dirty = 1;
2497 ret = find_and_setup_root(tree_root, fs_info,
2498 BTRFS_DEV_TREE_OBJECTID, dev_root);
2500 goto recovery_tree_root;
2501 dev_root->track_dirty = 1;
2503 ret = find_and_setup_root(tree_root, fs_info,
2504 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2506 goto recovery_tree_root;
2507 csum_root->track_dirty = 1;
2509 ret = find_and_setup_root(tree_root, fs_info,
2510 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2513 quota_root = fs_info->quota_root = NULL;
2515 quota_root->track_dirty = 1;
2516 fs_info->quota_enabled = 1;
2517 fs_info->pending_quota_state = 1;
2520 fs_info->generation = generation;
2521 fs_info->last_trans_committed = generation;
2523 ret = btrfs_recover_balance(fs_info);
2525 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2526 goto fail_block_groups;
2529 ret = btrfs_init_dev_stats(fs_info);
2531 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2533 goto fail_block_groups;
2536 ret = btrfs_init_dev_replace(fs_info);
2538 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2539 goto fail_block_groups;
2542 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2544 ret = btrfs_init_space_info(fs_info);
2546 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2547 goto fail_block_groups;
2550 ret = btrfs_read_block_groups(extent_root);
2552 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2553 goto fail_block_groups;
2555 fs_info->num_tolerated_disk_barrier_failures =
2556 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2557 if (fs_info->fs_devices->missing_devices >
2558 fs_info->num_tolerated_disk_barrier_failures &&
2559 !(sb->s_flags & MS_RDONLY)) {
2561 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2562 goto fail_block_groups;
2565 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2567 if (IS_ERR(fs_info->cleaner_kthread))
2568 goto fail_block_groups;
2570 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2572 "btrfs-transaction");
2573 if (IS_ERR(fs_info->transaction_kthread))
2576 if (!btrfs_test_opt(tree_root, SSD) &&
2577 !btrfs_test_opt(tree_root, NOSSD) &&
2578 !fs_info->fs_devices->rotating) {
2579 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2581 btrfs_set_opt(fs_info->mount_opt, SSD);
2584 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2585 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2586 ret = btrfsic_mount(tree_root, fs_devices,
2587 btrfs_test_opt(tree_root,
2588 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2590 fs_info->check_integrity_print_mask);
2592 printk(KERN_WARNING "btrfs: failed to initialize"
2593 " integrity check module %s\n", sb->s_id);
2596 ret = btrfs_read_qgroup_config(fs_info);
2598 goto fail_trans_kthread;
2600 /* do not make disk changes in broken FS */
2601 if (btrfs_super_log_root(disk_super) != 0) {
2602 u64 bytenr = btrfs_super_log_root(disk_super);
2604 if (fs_devices->rw_devices == 0) {
2605 printk(KERN_WARNING "Btrfs log replay required "
2611 btrfs_level_size(tree_root,
2612 btrfs_super_log_root_level(disk_super));
2614 log_tree_root = btrfs_alloc_root(fs_info);
2615 if (!log_tree_root) {
2620 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2621 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2623 log_tree_root->node = read_tree_block(tree_root, bytenr,
2626 /* returns with log_tree_root freed on success */
2627 ret = btrfs_recover_log_trees(log_tree_root);
2629 btrfs_error(tree_root->fs_info, ret,
2630 "Failed to recover log tree");
2631 free_extent_buffer(log_tree_root->node);
2632 kfree(log_tree_root);
2633 goto fail_trans_kthread;
2636 if (sb->s_flags & MS_RDONLY) {
2637 ret = btrfs_commit_super(tree_root);
2639 goto fail_trans_kthread;
2643 ret = btrfs_find_orphan_roots(tree_root);
2645 goto fail_trans_kthread;
2647 if (!(sb->s_flags & MS_RDONLY)) {
2648 ret = btrfs_cleanup_fs_roots(fs_info);
2650 goto fail_trans_kthread;
2652 ret = btrfs_recover_relocation(tree_root);
2655 "btrfs: failed to recover relocation\n");
2661 location.objectid = BTRFS_FS_TREE_OBJECTID;
2662 location.type = BTRFS_ROOT_ITEM_KEY;
2663 location.offset = (u64)-1;
2665 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2666 if (!fs_info->fs_root)
2668 if (IS_ERR(fs_info->fs_root)) {
2669 err = PTR_ERR(fs_info->fs_root);
2673 if (sb->s_flags & MS_RDONLY)
2676 down_read(&fs_info->cleanup_work_sem);
2677 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2678 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2679 up_read(&fs_info->cleanup_work_sem);
2680 close_ctree(tree_root);
2683 up_read(&fs_info->cleanup_work_sem);
2685 ret = btrfs_resume_balance_async(fs_info);
2687 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2688 close_ctree(tree_root);
2692 ret = btrfs_resume_dev_replace_async(fs_info);
2694 pr_warn("btrfs: failed to resume dev_replace\n");
2695 close_ctree(tree_root);
2702 btrfs_free_qgroup_config(fs_info);
2704 kthread_stop(fs_info->transaction_kthread);
2706 kthread_stop(fs_info->cleaner_kthread);
2709 * make sure we're done with the btree inode before we stop our
2712 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2715 btrfs_free_block_groups(fs_info);
2718 free_root_pointers(fs_info, 1);
2719 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2722 btrfs_stop_workers(&fs_info->generic_worker);
2723 btrfs_stop_workers(&fs_info->readahead_workers);
2724 btrfs_stop_workers(&fs_info->fixup_workers);
2725 btrfs_stop_workers(&fs_info->delalloc_workers);
2726 btrfs_stop_workers(&fs_info->workers);
2727 btrfs_stop_workers(&fs_info->endio_workers);
2728 btrfs_stop_workers(&fs_info->endio_meta_workers);
2729 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2730 btrfs_stop_workers(&fs_info->endio_write_workers);
2731 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2732 btrfs_stop_workers(&fs_info->submit_workers);
2733 btrfs_stop_workers(&fs_info->delayed_workers);
2734 btrfs_stop_workers(&fs_info->caching_workers);
2735 btrfs_stop_workers(&fs_info->flush_workers);
2738 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2740 iput(fs_info->btree_inode);
2741 fail_delalloc_bytes:
2742 percpu_counter_destroy(&fs_info->delalloc_bytes);
2743 fail_dirty_metadata_bytes:
2744 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2746 bdi_destroy(&fs_info->bdi);
2748 cleanup_srcu_struct(&fs_info->subvol_srcu);
2750 btrfs_close_devices(fs_info->fs_devices);
2754 if (!btrfs_test_opt(tree_root, RECOVERY))
2755 goto fail_tree_roots;
2757 free_root_pointers(fs_info, 0);
2759 /* don't use the log in recovery mode, it won't be valid */
2760 btrfs_set_super_log_root(disk_super, 0);
2762 /* we can't trust the free space cache either */
2763 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2765 ret = next_root_backup(fs_info, fs_info->super_copy,
2766 &num_backups_tried, &backup_index);
2768 goto fail_block_groups;
2769 goto retry_root_backup;
2772 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2775 set_buffer_uptodate(bh);
2777 struct btrfs_device *device = (struct btrfs_device *)
2780 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2781 "I/O error on %s\n",
2782 rcu_str_deref(device->name));
2783 /* note, we dont' set_buffer_write_io_error because we have
2784 * our own ways of dealing with the IO errors
2786 clear_buffer_uptodate(bh);
2787 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2793 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2795 struct buffer_head *bh;
2796 struct buffer_head *latest = NULL;
2797 struct btrfs_super_block *super;
2802 /* we would like to check all the supers, but that would make
2803 * a btrfs mount succeed after a mkfs from a different FS.
2804 * So, we need to add a special mount option to scan for
2805 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2807 for (i = 0; i < 1; i++) {
2808 bytenr = btrfs_sb_offset(i);
2809 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2811 bh = __bread(bdev, bytenr / 4096, 4096);
2815 super = (struct btrfs_super_block *)bh->b_data;
2816 if (btrfs_super_bytenr(super) != bytenr ||
2817 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2822 if (!latest || btrfs_super_generation(super) > transid) {
2825 transid = btrfs_super_generation(super);
2834 * this should be called twice, once with wait == 0 and
2835 * once with wait == 1. When wait == 0 is done, all the buffer heads
2836 * we write are pinned.
2838 * They are released when wait == 1 is done.
2839 * max_mirrors must be the same for both runs, and it indicates how
2840 * many supers on this one device should be written.
2842 * max_mirrors == 0 means to write them all.
2844 static int write_dev_supers(struct btrfs_device *device,
2845 struct btrfs_super_block *sb,
2846 int do_barriers, int wait, int max_mirrors)
2848 struct buffer_head *bh;
2855 if (max_mirrors == 0)
2856 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2858 for (i = 0; i < max_mirrors; i++) {
2859 bytenr = btrfs_sb_offset(i);
2860 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2864 bh = __find_get_block(device->bdev, bytenr / 4096,
2865 BTRFS_SUPER_INFO_SIZE);
2868 if (!buffer_uptodate(bh))
2871 /* drop our reference */
2874 /* drop the reference from the wait == 0 run */
2878 btrfs_set_super_bytenr(sb, bytenr);
2881 crc = btrfs_csum_data(NULL, (char *)sb +
2882 BTRFS_CSUM_SIZE, crc,
2883 BTRFS_SUPER_INFO_SIZE -
2885 btrfs_csum_final(crc, sb->csum);
2888 * one reference for us, and we leave it for the
2891 bh = __getblk(device->bdev, bytenr / 4096,
2892 BTRFS_SUPER_INFO_SIZE);
2893 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2895 /* one reference for submit_bh */
2898 set_buffer_uptodate(bh);
2900 bh->b_end_io = btrfs_end_buffer_write_sync;
2901 bh->b_private = device;
2905 * we fua the first super. The others we allow
2908 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2912 return errors < i ? 0 : -1;
2916 * endio for the write_dev_flush, this will wake anyone waiting
2917 * for the barrier when it is done
2919 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2922 if (err == -EOPNOTSUPP)
2923 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2924 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2926 if (bio->bi_private)
2927 complete(bio->bi_private);
2932 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2933 * sent down. With wait == 1, it waits for the previous flush.
2935 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2938 static int write_dev_flush(struct btrfs_device *device, int wait)
2943 if (device->nobarriers)
2947 bio = device->flush_bio;
2951 wait_for_completion(&device->flush_wait);
2953 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2954 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2955 rcu_str_deref(device->name));
2956 device->nobarriers = 1;
2957 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
2959 btrfs_dev_stat_inc_and_print(device,
2960 BTRFS_DEV_STAT_FLUSH_ERRS);
2963 /* drop the reference from the wait == 0 run */
2965 device->flush_bio = NULL;
2971 * one reference for us, and we leave it for the
2974 device->flush_bio = NULL;
2975 bio = bio_alloc(GFP_NOFS, 0);
2979 bio->bi_end_io = btrfs_end_empty_barrier;
2980 bio->bi_bdev = device->bdev;
2981 init_completion(&device->flush_wait);
2982 bio->bi_private = &device->flush_wait;
2983 device->flush_bio = bio;
2986 btrfsic_submit_bio(WRITE_FLUSH, bio);
2992 * send an empty flush down to each device in parallel,
2993 * then wait for them
2995 static int barrier_all_devices(struct btrfs_fs_info *info)
2997 struct list_head *head;
2998 struct btrfs_device *dev;
2999 int errors_send = 0;
3000 int errors_wait = 0;
3003 /* send down all the barriers */
3004 head = &info->fs_devices->devices;
3005 list_for_each_entry_rcu(dev, head, dev_list) {
3010 if (!dev->in_fs_metadata || !dev->writeable)
3013 ret = write_dev_flush(dev, 0);
3018 /* wait for all the barriers */
3019 list_for_each_entry_rcu(dev, head, dev_list) {
3024 if (!dev->in_fs_metadata || !dev->writeable)
3027 ret = write_dev_flush(dev, 1);
3031 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3032 errors_wait > info->num_tolerated_disk_barrier_failures)
3037 int btrfs_calc_num_tolerated_disk_barrier_failures(
3038 struct btrfs_fs_info *fs_info)
3040 struct btrfs_ioctl_space_info space;
3041 struct btrfs_space_info *sinfo;
3042 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3043 BTRFS_BLOCK_GROUP_SYSTEM,
3044 BTRFS_BLOCK_GROUP_METADATA,
3045 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3049 int num_tolerated_disk_barrier_failures =
3050 (int)fs_info->fs_devices->num_devices;
3052 for (i = 0; i < num_types; i++) {
3053 struct btrfs_space_info *tmp;
3057 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3058 if (tmp->flags == types[i]) {
3068 down_read(&sinfo->groups_sem);
3069 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3070 if (!list_empty(&sinfo->block_groups[c])) {
3073 btrfs_get_block_group_info(
3074 &sinfo->block_groups[c], &space);
3075 if (space.total_bytes == 0 ||
3076 space.used_bytes == 0)
3078 flags = space.flags;
3081 * 0: if dup, single or RAID0 is configured for
3082 * any of metadata, system or data, else
3083 * 1: if RAID5 is configured, or if RAID1 or
3084 * RAID10 is configured and only two mirrors
3086 * 2: if RAID6 is configured, else
3087 * num_mirrors - 1: if RAID1 or RAID10 is
3088 * configured and more than
3089 * 2 mirrors are used.
3091 if (num_tolerated_disk_barrier_failures > 0 &&
3092 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3093 BTRFS_BLOCK_GROUP_RAID0)) ||
3094 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3096 num_tolerated_disk_barrier_failures = 0;
3097 else if (num_tolerated_disk_barrier_failures > 1
3099 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3100 BTRFS_BLOCK_GROUP_RAID10)))
3101 num_tolerated_disk_barrier_failures = 1;
3104 up_read(&sinfo->groups_sem);
3107 return num_tolerated_disk_barrier_failures;
3110 int write_all_supers(struct btrfs_root *root, int max_mirrors)
3112 struct list_head *head;
3113 struct btrfs_device *dev;
3114 struct btrfs_super_block *sb;
3115 struct btrfs_dev_item *dev_item;
3119 int total_errors = 0;
3122 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3123 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3124 backup_super_roots(root->fs_info);
3126 sb = root->fs_info->super_for_commit;
3127 dev_item = &sb->dev_item;
3129 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3130 head = &root->fs_info->fs_devices->devices;
3133 ret = barrier_all_devices(root->fs_info);
3136 &root->fs_info->fs_devices->device_list_mutex);
3137 btrfs_error(root->fs_info, ret,
3138 "errors while submitting device barriers.");
3143 list_for_each_entry_rcu(dev, head, dev_list) {
3148 if (!dev->in_fs_metadata || !dev->writeable)
3151 btrfs_set_stack_device_generation(dev_item, 0);
3152 btrfs_set_stack_device_type(dev_item, dev->type);
3153 btrfs_set_stack_device_id(dev_item, dev->devid);
3154 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3155 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3156 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3157 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3158 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3159 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3160 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3162 flags = btrfs_super_flags(sb);
3163 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3165 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3169 if (total_errors > max_errors) {
3170 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3173 /* This shouldn't happen. FUA is masked off if unsupported */
3178 list_for_each_entry_rcu(dev, head, dev_list) {
3181 if (!dev->in_fs_metadata || !dev->writeable)
3184 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3188 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3189 if (total_errors > max_errors) {
3190 btrfs_error(root->fs_info, -EIO,
3191 "%d errors while writing supers", total_errors);
3197 int write_ctree_super(struct btrfs_trans_handle *trans,
3198 struct btrfs_root *root, int max_mirrors)
3202 ret = write_all_supers(root, max_mirrors);
3206 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3208 spin_lock(&fs_info->fs_roots_radix_lock);
3209 radix_tree_delete(&fs_info->fs_roots_radix,
3210 (unsigned long)root->root_key.objectid);
3211 spin_unlock(&fs_info->fs_roots_radix_lock);
3213 if (btrfs_root_refs(&root->root_item) == 0)
3214 synchronize_srcu(&fs_info->subvol_srcu);
3216 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3217 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3221 static void free_fs_root(struct btrfs_root *root)
3223 iput(root->cache_inode);
3224 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3226 free_anon_bdev(root->anon_dev);
3227 free_extent_buffer(root->node);
3228 free_extent_buffer(root->commit_root);
3229 kfree(root->free_ino_ctl);
3230 kfree(root->free_ino_pinned);
3235 static void del_fs_roots(struct btrfs_fs_info *fs_info)
3238 struct btrfs_root *gang[8];
3241 while (!list_empty(&fs_info->dead_roots)) {
3242 gang[0] = list_entry(fs_info->dead_roots.next,
3243 struct btrfs_root, root_list);
3244 list_del(&gang[0]->root_list);
3246 if (gang[0]->in_radix) {
3247 btrfs_free_fs_root(fs_info, gang[0]);
3249 free_extent_buffer(gang[0]->node);
3250 free_extent_buffer(gang[0]->commit_root);
3256 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3261 for (i = 0; i < ret; i++)
3262 btrfs_free_fs_root(fs_info, gang[i]);
3266 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3268 u64 root_objectid = 0;
3269 struct btrfs_root *gang[8];
3274 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3275 (void **)gang, root_objectid,
3280 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3281 for (i = 0; i < ret; i++) {
3284 root_objectid = gang[i]->root_key.objectid;
3285 err = btrfs_orphan_cleanup(gang[i]);
3294 int btrfs_commit_super(struct btrfs_root *root)
3296 struct btrfs_trans_handle *trans;
3299 mutex_lock(&root->fs_info->cleaner_mutex);
3300 btrfs_run_delayed_iputs(root);
3301 btrfs_clean_old_snapshots(root);
3302 mutex_unlock(&root->fs_info->cleaner_mutex);
3304 /* wait until ongoing cleanup work done */
3305 down_write(&root->fs_info->cleanup_work_sem);
3306 up_write(&root->fs_info->cleanup_work_sem);
3308 trans = btrfs_join_transaction(root);
3310 return PTR_ERR(trans);
3311 ret = btrfs_commit_transaction(trans, root);
3314 /* run commit again to drop the original snapshot */
3315 trans = btrfs_join_transaction(root);
3317 return PTR_ERR(trans);
3318 ret = btrfs_commit_transaction(trans, root);
3321 ret = btrfs_write_and_wait_transaction(NULL, root);
3323 btrfs_error(root->fs_info, ret,
3324 "Failed to sync btree inode to disk.");
3328 ret = write_ctree_super(NULL, root, 0);
3332 int close_ctree(struct btrfs_root *root)
3334 struct btrfs_fs_info *fs_info = root->fs_info;
3337 fs_info->closing = 1;
3340 /* pause restriper - we want to resume on mount */
3341 btrfs_pause_balance(fs_info);
3343 btrfs_dev_replace_suspend_for_unmount(fs_info);
3345 btrfs_scrub_cancel(fs_info);
3347 /* wait for any defraggers to finish */
3348 wait_event(fs_info->transaction_wait,
3349 (atomic_read(&fs_info->defrag_running) == 0));
3351 /* clear out the rbtree of defraggable inodes */
3352 btrfs_cleanup_defrag_inodes(fs_info);
3354 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3355 ret = btrfs_commit_super(root);
3357 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3360 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3361 btrfs_error_commit_super(root);
3363 btrfs_put_block_group_cache(fs_info);
3365 kthread_stop(fs_info->transaction_kthread);
3366 kthread_stop(fs_info->cleaner_kthread);
3368 fs_info->closing = 2;
3371 btrfs_free_qgroup_config(root->fs_info);
3373 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3374 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3375 percpu_counter_sum(&fs_info->delalloc_bytes));
3378 free_extent_buffer(fs_info->extent_root->node);
3379 free_extent_buffer(fs_info->extent_root->commit_root);
3380 free_extent_buffer(fs_info->tree_root->node);
3381 free_extent_buffer(fs_info->tree_root->commit_root);
3382 free_extent_buffer(fs_info->chunk_root->node);
3383 free_extent_buffer(fs_info->chunk_root->commit_root);
3384 free_extent_buffer(fs_info->dev_root->node);
3385 free_extent_buffer(fs_info->dev_root->commit_root);
3386 free_extent_buffer(fs_info->csum_root->node);
3387 free_extent_buffer(fs_info->csum_root->commit_root);
3388 if (fs_info->quota_root) {
3389 free_extent_buffer(fs_info->quota_root->node);
3390 free_extent_buffer(fs_info->quota_root->commit_root);
3393 btrfs_free_block_groups(fs_info);
3395 del_fs_roots(fs_info);
3397 iput(fs_info->btree_inode);
3399 btrfs_stop_workers(&fs_info->generic_worker);
3400 btrfs_stop_workers(&fs_info->fixup_workers);
3401 btrfs_stop_workers(&fs_info->delalloc_workers);
3402 btrfs_stop_workers(&fs_info->workers);
3403 btrfs_stop_workers(&fs_info->endio_workers);
3404 btrfs_stop_workers(&fs_info->endio_meta_workers);
3405 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3406 btrfs_stop_workers(&fs_info->endio_write_workers);
3407 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3408 btrfs_stop_workers(&fs_info->submit_workers);
3409 btrfs_stop_workers(&fs_info->delayed_workers);
3410 btrfs_stop_workers(&fs_info->caching_workers);
3411 btrfs_stop_workers(&fs_info->readahead_workers);
3412 btrfs_stop_workers(&fs_info->flush_workers);
3414 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3415 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3416 btrfsic_unmount(root, fs_info->fs_devices);
3419 btrfs_close_devices(fs_info->fs_devices);
3420 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3422 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3423 percpu_counter_destroy(&fs_info->delalloc_bytes);
3424 bdi_destroy(&fs_info->bdi);
3425 cleanup_srcu_struct(&fs_info->subvol_srcu);
3430 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3434 struct inode *btree_inode = buf->pages[0]->mapping->host;
3436 ret = extent_buffer_uptodate(buf);
3440 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3441 parent_transid, atomic);
3447 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3449 return set_extent_buffer_uptodate(buf);
3452 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3454 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3455 u64 transid = btrfs_header_generation(buf);
3458 btrfs_assert_tree_locked(buf);
3459 if (transid != root->fs_info->generation)
3460 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3461 "found %llu running %llu\n",
3462 (unsigned long long)buf->start,
3463 (unsigned long long)transid,
3464 (unsigned long long)root->fs_info->generation);
3465 was_dirty = set_extent_buffer_dirty(buf);
3467 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3469 root->fs_info->dirty_metadata_batch);
3472 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3476 * looks as though older kernels can get into trouble with
3477 * this code, they end up stuck in balance_dirty_pages forever
3481 if (current->flags & PF_MEMALLOC)
3485 btrfs_balance_delayed_items(root);
3487 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3488 BTRFS_DIRTY_METADATA_THRESH);
3490 balance_dirty_pages_ratelimited(
3491 root->fs_info->btree_inode->i_mapping);
3496 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3498 __btrfs_btree_balance_dirty(root, 1);
3501 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3503 __btrfs_btree_balance_dirty(root, 0);
3506 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3508 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3509 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3512 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3515 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3516 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3526 void btrfs_error_commit_super(struct btrfs_root *root)
3528 mutex_lock(&root->fs_info->cleaner_mutex);
3529 btrfs_run_delayed_iputs(root);
3530 mutex_unlock(&root->fs_info->cleaner_mutex);
3532 down_write(&root->fs_info->cleanup_work_sem);
3533 up_write(&root->fs_info->cleanup_work_sem);
3535 /* cleanup FS via transaction */
3536 btrfs_cleanup_transaction(root);
3539 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3540 struct btrfs_root *root)
3542 struct btrfs_inode *btrfs_inode;
3543 struct list_head splice;
3545 INIT_LIST_HEAD(&splice);
3547 mutex_lock(&root->fs_info->ordered_operations_mutex);
3548 spin_lock(&root->fs_info->ordered_extent_lock);
3550 list_splice_init(&t->ordered_operations, &splice);
3551 while (!list_empty(&splice)) {
3552 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3553 ordered_operations);
3555 list_del_init(&btrfs_inode->ordered_operations);
3557 btrfs_invalidate_inodes(btrfs_inode->root);
3560 spin_unlock(&root->fs_info->ordered_extent_lock);
3561 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3564 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3566 struct btrfs_ordered_extent *ordered;
3568 spin_lock(&root->fs_info->ordered_extent_lock);
3570 * This will just short circuit the ordered completion stuff which will
3571 * make sure the ordered extent gets properly cleaned up.
3573 list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3575 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3576 spin_unlock(&root->fs_info->ordered_extent_lock);
3579 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3580 struct btrfs_root *root)
3582 struct rb_node *node;
3583 struct btrfs_delayed_ref_root *delayed_refs;
3584 struct btrfs_delayed_ref_node *ref;
3587 delayed_refs = &trans->delayed_refs;
3589 spin_lock(&delayed_refs->lock);
3590 if (delayed_refs->num_entries == 0) {
3591 spin_unlock(&delayed_refs->lock);
3592 printk(KERN_INFO "delayed_refs has NO entry\n");
3596 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3597 struct btrfs_delayed_ref_head *head = NULL;
3599 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3600 atomic_set(&ref->refs, 1);
3601 if (btrfs_delayed_ref_is_head(ref)) {
3603 head = btrfs_delayed_node_to_head(ref);
3604 if (!mutex_trylock(&head->mutex)) {
3605 atomic_inc(&ref->refs);
3606 spin_unlock(&delayed_refs->lock);
3608 /* Need to wait for the delayed ref to run */
3609 mutex_lock(&head->mutex);
3610 mutex_unlock(&head->mutex);
3611 btrfs_put_delayed_ref(ref);
3613 spin_lock(&delayed_refs->lock);
3617 btrfs_free_delayed_extent_op(head->extent_op);
3618 delayed_refs->num_heads--;
3619 if (list_empty(&head->cluster))
3620 delayed_refs->num_heads_ready--;
3621 list_del_init(&head->cluster);
3625 rb_erase(&ref->rb_node, &delayed_refs->root);
3626 delayed_refs->num_entries--;
3628 mutex_unlock(&head->mutex);
3629 spin_unlock(&delayed_refs->lock);
3630 btrfs_put_delayed_ref(ref);
3633 spin_lock(&delayed_refs->lock);
3636 spin_unlock(&delayed_refs->lock);
3641 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3643 struct btrfs_pending_snapshot *snapshot;
3644 struct list_head splice;
3646 INIT_LIST_HEAD(&splice);
3648 list_splice_init(&t->pending_snapshots, &splice);
3650 while (!list_empty(&splice)) {
3651 snapshot = list_entry(splice.next,
3652 struct btrfs_pending_snapshot,
3655 list_del_init(&snapshot->list);
3661 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3663 struct btrfs_inode *btrfs_inode;
3664 struct list_head splice;
3666 INIT_LIST_HEAD(&splice);
3668 spin_lock(&root->fs_info->delalloc_lock);
3669 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3671 while (!list_empty(&splice)) {
3672 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3675 list_del_init(&btrfs_inode->delalloc_inodes);
3676 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3677 &btrfs_inode->runtime_flags);
3679 btrfs_invalidate_inodes(btrfs_inode->root);
3682 spin_unlock(&root->fs_info->delalloc_lock);
3685 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3686 struct extent_io_tree *dirty_pages,
3691 struct inode *btree_inode = root->fs_info->btree_inode;
3692 struct extent_buffer *eb;
3696 unsigned long index;
3699 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3704 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3705 while (start <= end) {
3706 index = start >> PAGE_CACHE_SHIFT;
3707 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3708 page = find_get_page(btree_inode->i_mapping, index);
3711 offset = page_offset(page);
3713 spin_lock(&dirty_pages->buffer_lock);
3714 eb = radix_tree_lookup(
3715 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3716 offset >> PAGE_CACHE_SHIFT);
3717 spin_unlock(&dirty_pages->buffer_lock);
3719 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3721 if (PageWriteback(page))
3722 end_page_writeback(page);
3725 if (PageDirty(page)) {
3726 clear_page_dirty_for_io(page);
3727 spin_lock_irq(&page->mapping->tree_lock);
3728 radix_tree_tag_clear(&page->mapping->page_tree,
3730 PAGECACHE_TAG_DIRTY);
3731 spin_unlock_irq(&page->mapping->tree_lock);
3735 page_cache_release(page);
3742 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3743 struct extent_io_tree *pinned_extents)
3745 struct extent_io_tree *unpin;
3751 unpin = pinned_extents;
3754 ret = find_first_extent_bit(unpin, 0, &start, &end,
3755 EXTENT_DIRTY, NULL);
3760 if (btrfs_test_opt(root, DISCARD))
3761 ret = btrfs_error_discard_extent(root, start,
3765 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3766 btrfs_error_unpin_extent_range(root, start, end);
3771 if (unpin == &root->fs_info->freed_extents[0])
3772 unpin = &root->fs_info->freed_extents[1];
3774 unpin = &root->fs_info->freed_extents[0];
3782 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3783 struct btrfs_root *root)
3785 btrfs_destroy_delayed_refs(cur_trans, root);
3786 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3787 cur_trans->dirty_pages.dirty_bytes);
3789 /* FIXME: cleanup wait for commit */
3790 cur_trans->in_commit = 1;
3791 cur_trans->blocked = 1;
3792 wake_up(&root->fs_info->transaction_blocked_wait);
3794 cur_trans->blocked = 0;
3795 wake_up(&root->fs_info->transaction_wait);
3797 cur_trans->commit_done = 1;
3798 wake_up(&cur_trans->commit_wait);
3800 btrfs_destroy_delayed_inodes(root);
3801 btrfs_assert_delayed_root_empty(root);
3803 btrfs_destroy_pending_snapshots(cur_trans);
3805 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3807 btrfs_destroy_pinned_extent(root,
3808 root->fs_info->pinned_extents);
3811 memset(cur_trans, 0, sizeof(*cur_trans));
3812 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3816 int btrfs_cleanup_transaction(struct btrfs_root *root)
3818 struct btrfs_transaction *t;
3821 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3823 spin_lock(&root->fs_info->trans_lock);
3824 list_splice_init(&root->fs_info->trans_list, &list);
3825 root->fs_info->trans_no_join = 1;
3826 spin_unlock(&root->fs_info->trans_lock);
3828 while (!list_empty(&list)) {
3829 t = list_entry(list.next, struct btrfs_transaction, list);
3831 btrfs_destroy_ordered_operations(t, root);
3833 btrfs_destroy_ordered_extents(root);
3835 btrfs_destroy_delayed_refs(t, root);
3837 btrfs_block_rsv_release(root,
3838 &root->fs_info->trans_block_rsv,
3839 t->dirty_pages.dirty_bytes);
3841 /* FIXME: cleanup wait for commit */
3845 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3846 wake_up(&root->fs_info->transaction_blocked_wait);
3850 if (waitqueue_active(&root->fs_info->transaction_wait))
3851 wake_up(&root->fs_info->transaction_wait);
3855 if (waitqueue_active(&t->commit_wait))
3856 wake_up(&t->commit_wait);
3858 btrfs_destroy_delayed_inodes(root);
3859 btrfs_assert_delayed_root_empty(root);
3861 btrfs_destroy_pending_snapshots(t);
3863 btrfs_destroy_delalloc_inodes(root);
3865 spin_lock(&root->fs_info->trans_lock);
3866 root->fs_info->running_transaction = NULL;
3867 spin_unlock(&root->fs_info->trans_lock);
3869 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3872 btrfs_destroy_pinned_extent(root,
3873 root->fs_info->pinned_extents);
3875 atomic_set(&t->use_count, 0);
3876 list_del_init(&t->list);
3877 memset(t, 0, sizeof(*t));
3878 kmem_cache_free(btrfs_transaction_cachep, t);
3881 spin_lock(&root->fs_info->trans_lock);
3882 root->fs_info->trans_no_join = 0;
3883 spin_unlock(&root->fs_info->trans_lock);
3884 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3889 static struct extent_io_ops btree_extent_io_ops = {
3890 .readpage_end_io_hook = btree_readpage_end_io_hook,
3891 .readpage_io_failed_hook = btree_io_failed_hook,
3892 .submit_bio_hook = btree_submit_bio_hook,
3893 /* note we're sharing with inode.c for the merge bio hook */
3894 .merge_bio_hook = btrfs_merge_bio_hook,