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_root *root);
60 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
61 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
62 struct btrfs_root *root);
63 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
64 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
65 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
66 struct extent_io_tree *dirty_pages,
68 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
69 struct extent_io_tree *pinned_extents);
72 * end_io_wq structs are used to do processing in task context when an IO is
73 * complete. This is used during reads to verify checksums, and it is used
74 * by writes to insert metadata for new file extents after IO is complete.
80 struct btrfs_fs_info *info;
83 struct list_head list;
84 struct btrfs_work work;
88 * async submit bios are used to offload expensive checksumming
89 * onto the worker threads. They checksum file and metadata bios
90 * just before they are sent down the IO stack.
92 struct async_submit_bio {
95 struct list_head list;
96 extent_submit_bio_hook_t *submit_bio_start;
97 extent_submit_bio_hook_t *submit_bio_done;
100 unsigned long bio_flags;
102 * bio_offset is optional, can be used if the pages in the bio
103 * can't tell us where in the file the bio should go
106 struct btrfs_work work;
111 * Lockdep class keys for extent_buffer->lock's in this root. For a given
112 * eb, the lockdep key is determined by the btrfs_root it belongs to and
113 * the level the eb occupies in the tree.
115 * Different roots are used for different purposes and may nest inside each
116 * other and they require separate keysets. As lockdep keys should be
117 * static, assign keysets according to the purpose of the root as indicated
118 * by btrfs_root->objectid. This ensures that all special purpose roots
119 * have separate keysets.
121 * Lock-nesting across peer nodes is always done with the immediate parent
122 * node locked thus preventing deadlock. As lockdep doesn't know this, use
123 * subclass to avoid triggering lockdep warning in such cases.
125 * The key is set by the readpage_end_io_hook after the buffer has passed
126 * csum validation but before the pages are unlocked. It is also set by
127 * btrfs_init_new_buffer on freshly allocated blocks.
129 * We also add a check to make sure the highest level of the tree is the
130 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
131 * needs update as well.
133 #ifdef CONFIG_DEBUG_LOCK_ALLOC
134 # if BTRFS_MAX_LEVEL != 8
138 static struct btrfs_lockdep_keyset {
139 u64 id; /* root objectid */
140 const char *name_stem; /* lock name stem */
141 char names[BTRFS_MAX_LEVEL + 1][20];
142 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
143 } btrfs_lockdep_keysets[] = {
144 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
145 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
146 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
147 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
148 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
149 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
150 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
151 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
152 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
153 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
154 { .id = 0, .name_stem = "tree" },
157 void __init btrfs_init_lockdep(void)
161 /* initialize lockdep class names */
162 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
163 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
165 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
166 snprintf(ks->names[j], sizeof(ks->names[j]),
167 "btrfs-%s-%02d", ks->name_stem, j);
171 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
174 struct btrfs_lockdep_keyset *ks;
176 BUG_ON(level >= ARRAY_SIZE(ks->keys));
178 /* find the matching keyset, id 0 is the default entry */
179 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
180 if (ks->id == objectid)
183 lockdep_set_class_and_name(&eb->lock,
184 &ks->keys[level], ks->names[level]);
190 * extents on the btree inode are pretty simple, there's one extent
191 * that covers the entire device
193 static struct extent_map *btree_get_extent(struct inode *inode,
194 struct page *page, size_t pg_offset, u64 start, u64 len,
197 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
198 struct extent_map *em;
201 read_lock(&em_tree->lock);
202 em = lookup_extent_mapping(em_tree, start, len);
205 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
206 read_unlock(&em_tree->lock);
209 read_unlock(&em_tree->lock);
211 em = alloc_extent_map();
213 em = ERR_PTR(-ENOMEM);
218 em->block_len = (u64)-1;
220 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
222 write_lock(&em_tree->lock);
223 ret = add_extent_mapping(em_tree, em);
224 if (ret == -EEXIST) {
226 em = lookup_extent_mapping(em_tree, start, len);
233 write_unlock(&em_tree->lock);
239 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
241 return crc32c(seed, data, len);
244 void btrfs_csum_final(u32 crc, char *result)
246 put_unaligned_le32(~crc, result);
250 * compute the csum for a btree block, and either verify it or write it
251 * into the csum field of the block.
253 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
256 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
259 unsigned long cur_len;
260 unsigned long offset = BTRFS_CSUM_SIZE;
262 unsigned long map_start;
263 unsigned long map_len;
266 unsigned long inline_result;
268 len = buf->len - offset;
270 err = map_private_extent_buffer(buf, offset, 32,
271 &kaddr, &map_start, &map_len);
274 cur_len = min(len, map_len - (offset - map_start));
275 crc = btrfs_csum_data(root, kaddr + offset - map_start,
280 if (csum_size > sizeof(inline_result)) {
281 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
285 result = (char *)&inline_result;
288 btrfs_csum_final(crc, result);
291 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
294 memcpy(&found, result, csum_size);
296 read_extent_buffer(buf, &val, 0, csum_size);
297 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
298 "failed on %llu wanted %X found %X "
300 root->fs_info->sb->s_id,
301 (unsigned long long)buf->start, val, found,
302 btrfs_header_level(buf));
303 if (result != (char *)&inline_result)
308 write_extent_buffer(buf, result, 0, csum_size);
310 if (result != (char *)&inline_result)
316 * we can't consider a given block up to date unless the transid of the
317 * block matches the transid in the parent node's pointer. This is how we
318 * detect blocks that either didn't get written at all or got written
319 * in the wrong place.
321 static int verify_parent_transid(struct extent_io_tree *io_tree,
322 struct extent_buffer *eb, u64 parent_transid,
325 struct extent_state *cached_state = NULL;
328 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
336 if (extent_buffer_uptodate(eb) &&
337 btrfs_header_generation(eb) == parent_transid) {
341 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
343 (unsigned long long)eb->start,
344 (unsigned long long)parent_transid,
345 (unsigned long long)btrfs_header_generation(eb));
347 clear_extent_buffer_uptodate(eb);
349 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350 &cached_state, GFP_NOFS);
355 * helper to read a given tree block, doing retries as required when
356 * the checksums don't match and we have alternate mirrors to try.
358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359 struct extent_buffer *eb,
360 u64 start, u64 parent_transid)
362 struct extent_io_tree *io_tree;
367 int failed_mirror = 0;
369 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
370 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
372 ret = read_extent_buffer_pages(io_tree, eb, start,
374 btree_get_extent, mirror_num);
376 if (!verify_parent_transid(io_tree, eb,
384 * This buffer's crc is fine, but its contents are corrupted, so
385 * there is no reason to read the other copies, they won't be
388 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
391 num_copies = btrfs_num_copies(root->fs_info,
396 if (!failed_mirror) {
398 failed_mirror = eb->read_mirror;
402 if (mirror_num == failed_mirror)
405 if (mirror_num > num_copies)
409 if (failed && !ret && failed_mirror)
410 repair_eb_io_failure(root, eb, failed_mirror);
416 * checksum a dirty tree block before IO. This has extra checks to make sure
417 * we only fill in the checksum field in the first page of a multi-page block
420 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
422 struct extent_io_tree *tree;
423 u64 start = page_offset(page);
425 struct extent_buffer *eb;
427 tree = &BTRFS_I(page->mapping->host)->io_tree;
429 eb = (struct extent_buffer *)page->private;
430 if (page != eb->pages[0])
432 found_start = btrfs_header_bytenr(eb);
433 if (found_start != start) {
437 if (!PageUptodate(page)) {
441 csum_tree_block(root, eb, 0);
445 static int check_tree_block_fsid(struct btrfs_root *root,
446 struct extent_buffer *eb)
448 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
449 u8 fsid[BTRFS_UUID_SIZE];
452 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
455 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
459 fs_devices = fs_devices->seed;
464 #define CORRUPT(reason, eb, root, slot) \
465 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
466 "root=%llu, slot=%d\n", reason, \
467 (unsigned long long)btrfs_header_bytenr(eb), \
468 (unsigned long long)root->objectid, slot)
470 static noinline int check_leaf(struct btrfs_root *root,
471 struct extent_buffer *leaf)
473 struct btrfs_key key;
474 struct btrfs_key leaf_key;
475 u32 nritems = btrfs_header_nritems(leaf);
481 /* Check the 0 item */
482 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
483 BTRFS_LEAF_DATA_SIZE(root)) {
484 CORRUPT("invalid item offset size pair", leaf, root, 0);
489 * Check to make sure each items keys are in the correct order and their
490 * offsets make sense. We only have to loop through nritems-1 because
491 * we check the current slot against the next slot, which verifies the
492 * next slot's offset+size makes sense and that the current's slot
495 for (slot = 0; slot < nritems - 1; slot++) {
496 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
497 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
499 /* Make sure the keys are in the right order */
500 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
501 CORRUPT("bad key order", leaf, root, slot);
506 * Make sure the offset and ends are right, remember that the
507 * item data starts at the end of the leaf and grows towards the
510 if (btrfs_item_offset_nr(leaf, slot) !=
511 btrfs_item_end_nr(leaf, slot + 1)) {
512 CORRUPT("slot offset bad", leaf, root, slot);
517 * Check to make sure that we don't point outside of the leaf,
518 * just incase all the items are consistent to eachother, but
519 * all point outside of the leaf.
521 if (btrfs_item_end_nr(leaf, slot) >
522 BTRFS_LEAF_DATA_SIZE(root)) {
523 CORRUPT("slot end outside of leaf", leaf, root, slot);
531 struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
532 struct page *page, int max_walk)
534 struct extent_buffer *eb;
535 u64 start = page_offset(page);
539 if (start < max_walk)
542 min_start = start - max_walk;
544 while (start >= min_start) {
545 eb = find_extent_buffer(tree, start, 0);
548 * we found an extent buffer and it contains our page
551 if (eb->start <= target &&
552 eb->start + eb->len > target)
555 /* we found an extent buffer that wasn't for us */
556 free_extent_buffer(eb);
561 start -= PAGE_CACHE_SIZE;
566 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
567 struct extent_state *state, int mirror)
569 struct extent_io_tree *tree;
572 struct extent_buffer *eb;
573 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
580 tree = &BTRFS_I(page->mapping->host)->io_tree;
581 eb = (struct extent_buffer *)page->private;
583 /* the pending IO might have been the only thing that kept this buffer
584 * in memory. Make sure we have a ref for all this other checks
586 extent_buffer_get(eb);
588 reads_done = atomic_dec_and_test(&eb->io_pages);
592 eb->read_mirror = mirror;
593 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
598 found_start = btrfs_header_bytenr(eb);
599 if (found_start != eb->start) {
600 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
602 (unsigned long long)found_start,
603 (unsigned long long)eb->start);
607 if (check_tree_block_fsid(root, eb)) {
608 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
609 (unsigned long long)eb->start);
613 found_level = btrfs_header_level(eb);
615 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
618 ret = csum_tree_block(root, eb, 1);
625 * If this is a leaf block and it is corrupt, set the corrupt bit so
626 * that we don't try and read the other copies of this block, just
629 if (found_level == 0 && check_leaf(root, eb)) {
630 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
635 set_extent_buffer_uptodate(eb);
637 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
638 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
639 btree_readahead_hook(root, eb, eb->start, ret);
643 clear_extent_buffer_uptodate(eb);
644 free_extent_buffer(eb);
649 static int btree_io_failed_hook(struct page *page, int failed_mirror)
651 struct extent_buffer *eb;
652 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
654 eb = (struct extent_buffer *)page->private;
655 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
656 eb->read_mirror = failed_mirror;
657 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
658 btree_readahead_hook(root, eb, eb->start, -EIO);
659 return -EIO; /* we fixed nothing */
662 static void end_workqueue_bio(struct bio *bio, int err)
664 struct end_io_wq *end_io_wq = bio->bi_private;
665 struct btrfs_fs_info *fs_info;
667 fs_info = end_io_wq->info;
668 end_io_wq->error = err;
669 end_io_wq->work.func = end_workqueue_fn;
670 end_io_wq->work.flags = 0;
672 if (bio->bi_rw & REQ_WRITE) {
673 if (end_io_wq->metadata == 1)
674 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
676 else if (end_io_wq->metadata == 2)
677 btrfs_queue_worker(&fs_info->endio_freespace_worker,
680 btrfs_queue_worker(&fs_info->endio_write_workers,
683 if (end_io_wq->metadata)
684 btrfs_queue_worker(&fs_info->endio_meta_workers,
687 btrfs_queue_worker(&fs_info->endio_workers,
693 * For the metadata arg you want
696 * 1 - if normal metadta
697 * 2 - if writing to the free space cache area
699 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
702 struct end_io_wq *end_io_wq;
703 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
707 end_io_wq->private = bio->bi_private;
708 end_io_wq->end_io = bio->bi_end_io;
709 end_io_wq->info = info;
710 end_io_wq->error = 0;
711 end_io_wq->bio = bio;
712 end_io_wq->metadata = metadata;
714 bio->bi_private = end_io_wq;
715 bio->bi_end_io = end_workqueue_bio;
719 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
721 unsigned long limit = min_t(unsigned long,
722 info->workers.max_workers,
723 info->fs_devices->open_devices);
727 static void run_one_async_start(struct btrfs_work *work)
729 struct async_submit_bio *async;
732 async = container_of(work, struct async_submit_bio, work);
733 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
734 async->mirror_num, async->bio_flags,
740 static void run_one_async_done(struct btrfs_work *work)
742 struct btrfs_fs_info *fs_info;
743 struct async_submit_bio *async;
746 async = container_of(work, struct async_submit_bio, work);
747 fs_info = BTRFS_I(async->inode)->root->fs_info;
749 limit = btrfs_async_submit_limit(fs_info);
750 limit = limit * 2 / 3;
752 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
753 waitqueue_active(&fs_info->async_submit_wait))
754 wake_up(&fs_info->async_submit_wait);
756 /* If an error occured we just want to clean up the bio and move on */
758 bio_endio(async->bio, async->error);
762 async->submit_bio_done(async->inode, async->rw, async->bio,
763 async->mirror_num, async->bio_flags,
767 static void run_one_async_free(struct btrfs_work *work)
769 struct async_submit_bio *async;
771 async = container_of(work, struct async_submit_bio, work);
775 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
776 int rw, struct bio *bio, int mirror_num,
777 unsigned long bio_flags,
779 extent_submit_bio_hook_t *submit_bio_start,
780 extent_submit_bio_hook_t *submit_bio_done)
782 struct async_submit_bio *async;
784 async = kmalloc(sizeof(*async), GFP_NOFS);
788 async->inode = inode;
791 async->mirror_num = mirror_num;
792 async->submit_bio_start = submit_bio_start;
793 async->submit_bio_done = submit_bio_done;
795 async->work.func = run_one_async_start;
796 async->work.ordered_func = run_one_async_done;
797 async->work.ordered_free = run_one_async_free;
799 async->work.flags = 0;
800 async->bio_flags = bio_flags;
801 async->bio_offset = bio_offset;
805 atomic_inc(&fs_info->nr_async_submits);
808 btrfs_set_work_high_prio(&async->work);
810 btrfs_queue_worker(&fs_info->workers, &async->work);
812 while (atomic_read(&fs_info->async_submit_draining) &&
813 atomic_read(&fs_info->nr_async_submits)) {
814 wait_event(fs_info->async_submit_wait,
815 (atomic_read(&fs_info->nr_async_submits) == 0));
821 static int btree_csum_one_bio(struct bio *bio)
823 struct bio_vec *bvec = bio->bi_io_vec;
825 struct btrfs_root *root;
828 WARN_ON(bio->bi_vcnt <= 0);
829 while (bio_index < bio->bi_vcnt) {
830 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
831 ret = csum_dirty_buffer(root, bvec->bv_page);
840 static int __btree_submit_bio_start(struct inode *inode, int rw,
841 struct bio *bio, int mirror_num,
842 unsigned long bio_flags,
846 * when we're called for a write, we're already in the async
847 * submission context. Just jump into btrfs_map_bio
849 return btree_csum_one_bio(bio);
852 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
853 int mirror_num, unsigned long bio_flags,
859 * when we're called for a write, we're already in the async
860 * submission context. Just jump into btrfs_map_bio
862 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
868 static int check_async_write(struct inode *inode, unsigned long bio_flags)
870 if (bio_flags & EXTENT_BIO_TREE_LOG)
879 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
880 int mirror_num, unsigned long bio_flags,
883 int async = check_async_write(inode, bio_flags);
886 if (!(rw & REQ_WRITE)) {
888 * called for a read, do the setup so that checksum validation
889 * can happen in the async kernel threads
891 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
895 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
898 ret = btree_csum_one_bio(bio);
901 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
905 * kthread helpers are used to submit writes so that
906 * checksumming can happen in parallel across all CPUs
908 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
909 inode, rw, bio, mirror_num, 0,
911 __btree_submit_bio_start,
912 __btree_submit_bio_done);
922 #ifdef CONFIG_MIGRATION
923 static int btree_migratepage(struct address_space *mapping,
924 struct page *newpage, struct page *page,
925 enum migrate_mode mode)
928 * we can't safely write a btree page from here,
929 * we haven't done the locking hook
934 * Buffers may be managed in a filesystem specific way.
935 * We must have no buffers or drop them.
937 if (page_has_private(page) &&
938 !try_to_release_page(page, GFP_KERNEL))
940 return migrate_page(mapping, newpage, page, mode);
945 static int btree_writepages(struct address_space *mapping,
946 struct writeback_control *wbc)
948 struct extent_io_tree *tree;
949 struct btrfs_fs_info *fs_info;
952 tree = &BTRFS_I(mapping->host)->io_tree;
953 if (wbc->sync_mode == WB_SYNC_NONE) {
955 if (wbc->for_kupdate)
958 fs_info = BTRFS_I(mapping->host)->root->fs_info;
959 /* this is a bit racy, but that's ok */
960 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
961 BTRFS_DIRTY_METADATA_THRESH);
965 return btree_write_cache_pages(mapping, wbc);
968 static int btree_readpage(struct file *file, struct page *page)
970 struct extent_io_tree *tree;
971 tree = &BTRFS_I(page->mapping->host)->io_tree;
972 return extent_read_full_page(tree, page, btree_get_extent, 0);
975 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
977 if (PageWriteback(page) || PageDirty(page))
980 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
981 * slab allocation from alloc_extent_state down the callchain where
982 * it'd hit a BUG_ON as those flags are not allowed.
984 gfp_flags &= ~GFP_SLAB_BUG_MASK;
986 return try_release_extent_buffer(page, gfp_flags);
989 static void btree_invalidatepage(struct page *page, unsigned long offset)
991 struct extent_io_tree *tree;
992 tree = &BTRFS_I(page->mapping->host)->io_tree;
993 extent_invalidatepage(tree, page, offset);
994 btree_releasepage(page, GFP_NOFS);
995 if (PagePrivate(page)) {
996 printk(KERN_WARNING "btrfs warning page private not zero "
997 "on page %llu\n", (unsigned long long)page_offset(page));
998 ClearPagePrivate(page);
999 set_page_private(page, 0);
1000 page_cache_release(page);
1004 static int btree_set_page_dirty(struct page *page)
1007 struct extent_buffer *eb;
1009 BUG_ON(!PagePrivate(page));
1010 eb = (struct extent_buffer *)page->private;
1012 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1013 BUG_ON(!atomic_read(&eb->refs));
1014 btrfs_assert_tree_locked(eb);
1016 return __set_page_dirty_nobuffers(page);
1019 static const struct address_space_operations btree_aops = {
1020 .readpage = btree_readpage,
1021 .writepages = btree_writepages,
1022 .releasepage = btree_releasepage,
1023 .invalidatepage = btree_invalidatepage,
1024 #ifdef CONFIG_MIGRATION
1025 .migratepage = btree_migratepage,
1027 .set_page_dirty = btree_set_page_dirty,
1030 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1033 struct extent_buffer *buf = NULL;
1034 struct inode *btree_inode = root->fs_info->btree_inode;
1037 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1040 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1041 buf, 0, WAIT_NONE, btree_get_extent, 0);
1042 free_extent_buffer(buf);
1046 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1047 int mirror_num, struct extent_buffer **eb)
1049 struct extent_buffer *buf = NULL;
1050 struct inode *btree_inode = root->fs_info->btree_inode;
1051 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1054 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1058 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1060 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1061 btree_get_extent, mirror_num);
1063 free_extent_buffer(buf);
1067 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1068 free_extent_buffer(buf);
1070 } else if (extent_buffer_uptodate(buf)) {
1073 free_extent_buffer(buf);
1078 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1079 u64 bytenr, u32 blocksize)
1081 struct inode *btree_inode = root->fs_info->btree_inode;
1082 struct extent_buffer *eb;
1083 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1088 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1089 u64 bytenr, u32 blocksize)
1091 struct inode *btree_inode = root->fs_info->btree_inode;
1092 struct extent_buffer *eb;
1094 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1100 int btrfs_write_tree_block(struct extent_buffer *buf)
1102 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1103 buf->start + buf->len - 1);
1106 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1108 return filemap_fdatawait_range(buf->pages[0]->mapping,
1109 buf->start, buf->start + buf->len - 1);
1112 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1113 u32 blocksize, u64 parent_transid)
1115 struct extent_buffer *buf = NULL;
1118 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1122 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1127 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1128 struct extent_buffer *buf)
1130 struct btrfs_fs_info *fs_info = root->fs_info;
1132 if (btrfs_header_generation(buf) ==
1133 fs_info->running_transaction->transid) {
1134 btrfs_assert_tree_locked(buf);
1136 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1137 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1139 fs_info->dirty_metadata_batch);
1140 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1141 btrfs_set_lock_blocking(buf);
1142 clear_extent_buffer_dirty(buf);
1147 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1148 u32 stripesize, struct btrfs_root *root,
1149 struct btrfs_fs_info *fs_info,
1153 root->commit_root = NULL;
1154 root->sectorsize = sectorsize;
1155 root->nodesize = nodesize;
1156 root->leafsize = leafsize;
1157 root->stripesize = stripesize;
1159 root->track_dirty = 0;
1161 root->orphan_item_inserted = 0;
1162 root->orphan_cleanup_state = 0;
1164 root->objectid = objectid;
1165 root->last_trans = 0;
1166 root->highest_objectid = 0;
1168 root->inode_tree = RB_ROOT;
1169 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1170 root->block_rsv = NULL;
1171 root->orphan_block_rsv = NULL;
1173 INIT_LIST_HEAD(&root->dirty_list);
1174 INIT_LIST_HEAD(&root->root_list);
1175 INIT_LIST_HEAD(&root->logged_list[0]);
1176 INIT_LIST_HEAD(&root->logged_list[1]);
1177 spin_lock_init(&root->orphan_lock);
1178 spin_lock_init(&root->inode_lock);
1179 spin_lock_init(&root->accounting_lock);
1180 spin_lock_init(&root->log_extents_lock[0]);
1181 spin_lock_init(&root->log_extents_lock[1]);
1182 mutex_init(&root->objectid_mutex);
1183 mutex_init(&root->log_mutex);
1184 init_waitqueue_head(&root->log_writer_wait);
1185 init_waitqueue_head(&root->log_commit_wait[0]);
1186 init_waitqueue_head(&root->log_commit_wait[1]);
1187 atomic_set(&root->log_commit[0], 0);
1188 atomic_set(&root->log_commit[1], 0);
1189 atomic_set(&root->log_writers, 0);
1190 atomic_set(&root->log_batch, 0);
1191 atomic_set(&root->orphan_inodes, 0);
1192 root->log_transid = 0;
1193 root->last_log_commit = 0;
1194 extent_io_tree_init(&root->dirty_log_pages,
1195 fs_info->btree_inode->i_mapping);
1197 memset(&root->root_key, 0, sizeof(root->root_key));
1198 memset(&root->root_item, 0, sizeof(root->root_item));
1199 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1200 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1201 root->defrag_trans_start = fs_info->generation;
1202 init_completion(&root->kobj_unregister);
1203 root->defrag_running = 0;
1204 root->root_key.objectid = objectid;
1207 spin_lock_init(&root->root_item_lock);
1210 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1211 struct btrfs_fs_info *fs_info,
1213 struct btrfs_root *root)
1219 __setup_root(tree_root->nodesize, tree_root->leafsize,
1220 tree_root->sectorsize, tree_root->stripesize,
1221 root, fs_info, objectid);
1222 ret = btrfs_find_last_root(tree_root, objectid,
1223 &root->root_item, &root->root_key);
1229 generation = btrfs_root_generation(&root->root_item);
1230 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1231 root->commit_root = NULL;
1232 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1233 blocksize, generation);
1234 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1235 free_extent_buffer(root->node);
1239 root->commit_root = btrfs_root_node(root);
1243 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1245 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1247 root->fs_info = fs_info;
1251 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1252 struct btrfs_fs_info *fs_info,
1255 struct extent_buffer *leaf;
1256 struct btrfs_root *tree_root = fs_info->tree_root;
1257 struct btrfs_root *root;
1258 struct btrfs_key key;
1262 root = btrfs_alloc_root(fs_info);
1264 return ERR_PTR(-ENOMEM);
1266 __setup_root(tree_root->nodesize, tree_root->leafsize,
1267 tree_root->sectorsize, tree_root->stripesize,
1268 root, fs_info, objectid);
1269 root->root_key.objectid = objectid;
1270 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1271 root->root_key.offset = 0;
1273 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1274 0, objectid, NULL, 0, 0, 0);
1276 ret = PTR_ERR(leaf);
1280 bytenr = leaf->start;
1281 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1282 btrfs_set_header_bytenr(leaf, leaf->start);
1283 btrfs_set_header_generation(leaf, trans->transid);
1284 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1285 btrfs_set_header_owner(leaf, objectid);
1288 write_extent_buffer(leaf, fs_info->fsid,
1289 (unsigned long)btrfs_header_fsid(leaf),
1291 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1292 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1294 btrfs_mark_buffer_dirty(leaf);
1296 root->commit_root = btrfs_root_node(root);
1297 root->track_dirty = 1;
1300 root->root_item.flags = 0;
1301 root->root_item.byte_limit = 0;
1302 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1303 btrfs_set_root_generation(&root->root_item, trans->transid);
1304 btrfs_set_root_level(&root->root_item, 0);
1305 btrfs_set_root_refs(&root->root_item, 1);
1306 btrfs_set_root_used(&root->root_item, leaf->len);
1307 btrfs_set_root_last_snapshot(&root->root_item, 0);
1308 btrfs_set_root_dirid(&root->root_item, 0);
1309 root->root_item.drop_level = 0;
1311 key.objectid = objectid;
1312 key.type = BTRFS_ROOT_ITEM_KEY;
1314 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1318 btrfs_tree_unlock(leaf);
1322 return ERR_PTR(ret);
1327 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1328 struct btrfs_fs_info *fs_info)
1330 struct btrfs_root *root;
1331 struct btrfs_root *tree_root = fs_info->tree_root;
1332 struct extent_buffer *leaf;
1334 root = btrfs_alloc_root(fs_info);
1336 return ERR_PTR(-ENOMEM);
1338 __setup_root(tree_root->nodesize, tree_root->leafsize,
1339 tree_root->sectorsize, tree_root->stripesize,
1340 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1342 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1343 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1344 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1346 * log trees do not get reference counted because they go away
1347 * before a real commit is actually done. They do store pointers
1348 * to file data extents, and those reference counts still get
1349 * updated (along with back refs to the log tree).
1353 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1354 BTRFS_TREE_LOG_OBJECTID, NULL,
1358 return ERR_CAST(leaf);
1361 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1362 btrfs_set_header_bytenr(leaf, leaf->start);
1363 btrfs_set_header_generation(leaf, trans->transid);
1364 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1365 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1368 write_extent_buffer(root->node, root->fs_info->fsid,
1369 (unsigned long)btrfs_header_fsid(root->node),
1371 btrfs_mark_buffer_dirty(root->node);
1372 btrfs_tree_unlock(root->node);
1376 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1377 struct btrfs_fs_info *fs_info)
1379 struct btrfs_root *log_root;
1381 log_root = alloc_log_tree(trans, fs_info);
1382 if (IS_ERR(log_root))
1383 return PTR_ERR(log_root);
1384 WARN_ON(fs_info->log_root_tree);
1385 fs_info->log_root_tree = log_root;
1389 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1390 struct btrfs_root *root)
1392 struct btrfs_root *log_root;
1393 struct btrfs_inode_item *inode_item;
1395 log_root = alloc_log_tree(trans, root->fs_info);
1396 if (IS_ERR(log_root))
1397 return PTR_ERR(log_root);
1399 log_root->last_trans = trans->transid;
1400 log_root->root_key.offset = root->root_key.objectid;
1402 inode_item = &log_root->root_item.inode;
1403 inode_item->generation = cpu_to_le64(1);
1404 inode_item->size = cpu_to_le64(3);
1405 inode_item->nlink = cpu_to_le32(1);
1406 inode_item->nbytes = cpu_to_le64(root->leafsize);
1407 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1409 btrfs_set_root_node(&log_root->root_item, log_root->node);
1411 WARN_ON(root->log_root);
1412 root->log_root = log_root;
1413 root->log_transid = 0;
1414 root->last_log_commit = 0;
1418 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1419 struct btrfs_key *location)
1421 struct btrfs_root *root;
1422 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1423 struct btrfs_path *path;
1424 struct extent_buffer *l;
1430 root = btrfs_alloc_root(fs_info);
1432 return ERR_PTR(-ENOMEM);
1433 if (location->offset == (u64)-1) {
1434 ret = find_and_setup_root(tree_root, fs_info,
1435 location->objectid, root);
1438 return ERR_PTR(ret);
1443 __setup_root(tree_root->nodesize, tree_root->leafsize,
1444 tree_root->sectorsize, tree_root->stripesize,
1445 root, fs_info, location->objectid);
1447 path = btrfs_alloc_path();
1450 return ERR_PTR(-ENOMEM);
1452 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1455 slot = path->slots[0];
1456 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1457 memcpy(&root->root_key, location, sizeof(*location));
1459 btrfs_free_path(path);
1464 return ERR_PTR(ret);
1467 generation = btrfs_root_generation(&root->root_item);
1468 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1469 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1470 blocksize, generation);
1471 root->commit_root = btrfs_root_node(root);
1472 BUG_ON(!root->node); /* -ENOMEM */
1474 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1476 btrfs_check_and_init_root_item(&root->root_item);
1482 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1483 struct btrfs_key *location)
1485 struct btrfs_root *root;
1488 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1489 return fs_info->tree_root;
1490 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1491 return fs_info->extent_root;
1492 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1493 return fs_info->chunk_root;
1494 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1495 return fs_info->dev_root;
1496 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1497 return fs_info->csum_root;
1498 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1499 return fs_info->quota_root ? fs_info->quota_root :
1502 spin_lock(&fs_info->fs_roots_radix_lock);
1503 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1504 (unsigned long)location->objectid);
1505 spin_unlock(&fs_info->fs_roots_radix_lock);
1509 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1513 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1514 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1516 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1521 btrfs_init_free_ino_ctl(root);
1522 mutex_init(&root->fs_commit_mutex);
1523 spin_lock_init(&root->cache_lock);
1524 init_waitqueue_head(&root->cache_wait);
1526 ret = get_anon_bdev(&root->anon_dev);
1530 if (btrfs_root_refs(&root->root_item) == 0) {
1535 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1539 root->orphan_item_inserted = 1;
1541 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1545 spin_lock(&fs_info->fs_roots_radix_lock);
1546 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1547 (unsigned long)root->root_key.objectid,
1552 spin_unlock(&fs_info->fs_roots_radix_lock);
1553 radix_tree_preload_end();
1555 if (ret == -EEXIST) {
1562 ret = btrfs_find_dead_roots(fs_info->tree_root,
1563 root->root_key.objectid);
1568 return ERR_PTR(ret);
1571 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1573 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1575 struct btrfs_device *device;
1576 struct backing_dev_info *bdi;
1579 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1582 bdi = blk_get_backing_dev_info(device->bdev);
1583 if (bdi && bdi_congested(bdi, bdi_bits)) {
1593 * If this fails, caller must call bdi_destroy() to get rid of the
1596 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1600 bdi->capabilities = BDI_CAP_MAP_COPY;
1601 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1605 bdi->ra_pages = default_backing_dev_info.ra_pages;
1606 bdi->congested_fn = btrfs_congested_fn;
1607 bdi->congested_data = info;
1612 * called by the kthread helper functions to finally call the bio end_io
1613 * functions. This is where read checksum verification actually happens
1615 static void end_workqueue_fn(struct btrfs_work *work)
1618 struct end_io_wq *end_io_wq;
1619 struct btrfs_fs_info *fs_info;
1622 end_io_wq = container_of(work, struct end_io_wq, work);
1623 bio = end_io_wq->bio;
1624 fs_info = end_io_wq->info;
1626 error = end_io_wq->error;
1627 bio->bi_private = end_io_wq->private;
1628 bio->bi_end_io = end_io_wq->end_io;
1630 bio_endio(bio, error);
1633 static int cleaner_kthread(void *arg)
1635 struct btrfs_root *root = arg;
1638 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1639 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1640 btrfs_run_delayed_iputs(root);
1641 btrfs_clean_old_snapshots(root);
1642 mutex_unlock(&root->fs_info->cleaner_mutex);
1643 btrfs_run_defrag_inodes(root->fs_info);
1646 if (!try_to_freeze()) {
1647 set_current_state(TASK_INTERRUPTIBLE);
1648 if (!kthread_should_stop())
1650 __set_current_state(TASK_RUNNING);
1652 } while (!kthread_should_stop());
1656 static int transaction_kthread(void *arg)
1658 struct btrfs_root *root = arg;
1659 struct btrfs_trans_handle *trans;
1660 struct btrfs_transaction *cur;
1663 unsigned long delay;
1667 cannot_commit = false;
1669 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1671 spin_lock(&root->fs_info->trans_lock);
1672 cur = root->fs_info->running_transaction;
1674 spin_unlock(&root->fs_info->trans_lock);
1678 now = get_seconds();
1679 if (!cur->blocked &&
1680 (now < cur->start_time || now - cur->start_time < 30)) {
1681 spin_unlock(&root->fs_info->trans_lock);
1685 transid = cur->transid;
1686 spin_unlock(&root->fs_info->trans_lock);
1688 /* If the file system is aborted, this will always fail. */
1689 trans = btrfs_attach_transaction(root);
1690 if (IS_ERR(trans)) {
1691 if (PTR_ERR(trans) != -ENOENT)
1692 cannot_commit = true;
1695 if (transid == trans->transid) {
1696 btrfs_commit_transaction(trans, root);
1698 btrfs_end_transaction(trans, root);
1701 wake_up_process(root->fs_info->cleaner_kthread);
1702 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1704 if (!try_to_freeze()) {
1705 set_current_state(TASK_INTERRUPTIBLE);
1706 if (!kthread_should_stop() &&
1707 (!btrfs_transaction_blocked(root->fs_info) ||
1709 schedule_timeout(delay);
1710 __set_current_state(TASK_RUNNING);
1712 } while (!kthread_should_stop());
1717 * this will find the highest generation in the array of
1718 * root backups. The index of the highest array is returned,
1719 * or -1 if we can't find anything.
1721 * We check to make sure the array is valid by comparing the
1722 * generation of the latest root in the array with the generation
1723 * in the super block. If they don't match we pitch it.
1725 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1728 int newest_index = -1;
1729 struct btrfs_root_backup *root_backup;
1732 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1733 root_backup = info->super_copy->super_roots + i;
1734 cur = btrfs_backup_tree_root_gen(root_backup);
1735 if (cur == newest_gen)
1739 /* check to see if we actually wrapped around */
1740 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1741 root_backup = info->super_copy->super_roots;
1742 cur = btrfs_backup_tree_root_gen(root_backup);
1743 if (cur == newest_gen)
1746 return newest_index;
1751 * find the oldest backup so we know where to store new entries
1752 * in the backup array. This will set the backup_root_index
1753 * field in the fs_info struct
1755 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1758 int newest_index = -1;
1760 newest_index = find_newest_super_backup(info, newest_gen);
1761 /* if there was garbage in there, just move along */
1762 if (newest_index == -1) {
1763 info->backup_root_index = 0;
1765 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1770 * copy all the root pointers into the super backup array.
1771 * this will bump the backup pointer by one when it is
1774 static void backup_super_roots(struct btrfs_fs_info *info)
1777 struct btrfs_root_backup *root_backup;
1780 next_backup = info->backup_root_index;
1781 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1782 BTRFS_NUM_BACKUP_ROOTS;
1785 * just overwrite the last backup if we're at the same generation
1786 * this happens only at umount
1788 root_backup = info->super_for_commit->super_roots + last_backup;
1789 if (btrfs_backup_tree_root_gen(root_backup) ==
1790 btrfs_header_generation(info->tree_root->node))
1791 next_backup = last_backup;
1793 root_backup = info->super_for_commit->super_roots + next_backup;
1796 * make sure all of our padding and empty slots get zero filled
1797 * regardless of which ones we use today
1799 memset(root_backup, 0, sizeof(*root_backup));
1801 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1803 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1804 btrfs_set_backup_tree_root_gen(root_backup,
1805 btrfs_header_generation(info->tree_root->node));
1807 btrfs_set_backup_tree_root_level(root_backup,
1808 btrfs_header_level(info->tree_root->node));
1810 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1811 btrfs_set_backup_chunk_root_gen(root_backup,
1812 btrfs_header_generation(info->chunk_root->node));
1813 btrfs_set_backup_chunk_root_level(root_backup,
1814 btrfs_header_level(info->chunk_root->node));
1816 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1817 btrfs_set_backup_extent_root_gen(root_backup,
1818 btrfs_header_generation(info->extent_root->node));
1819 btrfs_set_backup_extent_root_level(root_backup,
1820 btrfs_header_level(info->extent_root->node));
1823 * we might commit during log recovery, which happens before we set
1824 * the fs_root. Make sure it is valid before we fill it in.
1826 if (info->fs_root && info->fs_root->node) {
1827 btrfs_set_backup_fs_root(root_backup,
1828 info->fs_root->node->start);
1829 btrfs_set_backup_fs_root_gen(root_backup,
1830 btrfs_header_generation(info->fs_root->node));
1831 btrfs_set_backup_fs_root_level(root_backup,
1832 btrfs_header_level(info->fs_root->node));
1835 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1836 btrfs_set_backup_dev_root_gen(root_backup,
1837 btrfs_header_generation(info->dev_root->node));
1838 btrfs_set_backup_dev_root_level(root_backup,
1839 btrfs_header_level(info->dev_root->node));
1841 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1842 btrfs_set_backup_csum_root_gen(root_backup,
1843 btrfs_header_generation(info->csum_root->node));
1844 btrfs_set_backup_csum_root_level(root_backup,
1845 btrfs_header_level(info->csum_root->node));
1847 btrfs_set_backup_total_bytes(root_backup,
1848 btrfs_super_total_bytes(info->super_copy));
1849 btrfs_set_backup_bytes_used(root_backup,
1850 btrfs_super_bytes_used(info->super_copy));
1851 btrfs_set_backup_num_devices(root_backup,
1852 btrfs_super_num_devices(info->super_copy));
1855 * if we don't copy this out to the super_copy, it won't get remembered
1856 * for the next commit
1858 memcpy(&info->super_copy->super_roots,
1859 &info->super_for_commit->super_roots,
1860 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1864 * this copies info out of the root backup array and back into
1865 * the in-memory super block. It is meant to help iterate through
1866 * the array, so you send it the number of backups you've already
1867 * tried and the last backup index you used.
1869 * this returns -1 when it has tried all the backups
1871 static noinline int next_root_backup(struct btrfs_fs_info *info,
1872 struct btrfs_super_block *super,
1873 int *num_backups_tried, int *backup_index)
1875 struct btrfs_root_backup *root_backup;
1876 int newest = *backup_index;
1878 if (*num_backups_tried == 0) {
1879 u64 gen = btrfs_super_generation(super);
1881 newest = find_newest_super_backup(info, gen);
1885 *backup_index = newest;
1886 *num_backups_tried = 1;
1887 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1888 /* we've tried all the backups, all done */
1891 /* jump to the next oldest backup */
1892 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1893 BTRFS_NUM_BACKUP_ROOTS;
1894 *backup_index = newest;
1895 *num_backups_tried += 1;
1897 root_backup = super->super_roots + newest;
1899 btrfs_set_super_generation(super,
1900 btrfs_backup_tree_root_gen(root_backup));
1901 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1902 btrfs_set_super_root_level(super,
1903 btrfs_backup_tree_root_level(root_backup));
1904 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1907 * fixme: the total bytes and num_devices need to match or we should
1910 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1911 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1915 /* helper to cleanup tree roots */
1916 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1918 free_extent_buffer(info->tree_root->node);
1919 free_extent_buffer(info->tree_root->commit_root);
1920 free_extent_buffer(info->dev_root->node);
1921 free_extent_buffer(info->dev_root->commit_root);
1922 free_extent_buffer(info->extent_root->node);
1923 free_extent_buffer(info->extent_root->commit_root);
1924 free_extent_buffer(info->csum_root->node);
1925 free_extent_buffer(info->csum_root->commit_root);
1926 if (info->quota_root) {
1927 free_extent_buffer(info->quota_root->node);
1928 free_extent_buffer(info->quota_root->commit_root);
1931 info->tree_root->node = NULL;
1932 info->tree_root->commit_root = NULL;
1933 info->dev_root->node = NULL;
1934 info->dev_root->commit_root = NULL;
1935 info->extent_root->node = NULL;
1936 info->extent_root->commit_root = NULL;
1937 info->csum_root->node = NULL;
1938 info->csum_root->commit_root = NULL;
1939 if (info->quota_root) {
1940 info->quota_root->node = NULL;
1941 info->quota_root->commit_root = NULL;
1945 free_extent_buffer(info->chunk_root->node);
1946 free_extent_buffer(info->chunk_root->commit_root);
1947 info->chunk_root->node = NULL;
1948 info->chunk_root->commit_root = NULL;
1953 int open_ctree(struct super_block *sb,
1954 struct btrfs_fs_devices *fs_devices,
1964 struct btrfs_key location;
1965 struct buffer_head *bh;
1966 struct btrfs_super_block *disk_super;
1967 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1968 struct btrfs_root *tree_root;
1969 struct btrfs_root *extent_root;
1970 struct btrfs_root *csum_root;
1971 struct btrfs_root *chunk_root;
1972 struct btrfs_root *dev_root;
1973 struct btrfs_root *quota_root;
1974 struct btrfs_root *log_tree_root;
1977 int num_backups_tried = 0;
1978 int backup_index = 0;
1980 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1981 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1982 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1983 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1984 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1985 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
1987 if (!tree_root || !extent_root || !csum_root ||
1988 !chunk_root || !dev_root || !quota_root) {
1993 ret = init_srcu_struct(&fs_info->subvol_srcu);
1999 ret = setup_bdi(fs_info, &fs_info->bdi);
2005 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2010 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2011 (1 + ilog2(nr_cpu_ids));
2013 fs_info->btree_inode = new_inode(sb);
2014 if (!fs_info->btree_inode) {
2016 goto fail_dirty_metadata_bytes;
2019 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2021 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2022 INIT_LIST_HEAD(&fs_info->trans_list);
2023 INIT_LIST_HEAD(&fs_info->dead_roots);
2024 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2025 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2026 INIT_LIST_HEAD(&fs_info->ordered_operations);
2027 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2028 spin_lock_init(&fs_info->delalloc_lock);
2029 spin_lock_init(&fs_info->trans_lock);
2030 spin_lock_init(&fs_info->fs_roots_radix_lock);
2031 spin_lock_init(&fs_info->delayed_iput_lock);
2032 spin_lock_init(&fs_info->defrag_inodes_lock);
2033 spin_lock_init(&fs_info->free_chunk_lock);
2034 spin_lock_init(&fs_info->tree_mod_seq_lock);
2035 rwlock_init(&fs_info->tree_mod_log_lock);
2036 mutex_init(&fs_info->reloc_mutex);
2038 init_completion(&fs_info->kobj_unregister);
2039 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2040 INIT_LIST_HEAD(&fs_info->space_info);
2041 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2042 btrfs_mapping_init(&fs_info->mapping_tree);
2043 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2044 BTRFS_BLOCK_RSV_GLOBAL);
2045 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2046 BTRFS_BLOCK_RSV_DELALLOC);
2047 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2048 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2049 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2050 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2051 BTRFS_BLOCK_RSV_DELOPS);
2052 atomic_set(&fs_info->nr_async_submits, 0);
2053 atomic_set(&fs_info->async_delalloc_pages, 0);
2054 atomic_set(&fs_info->async_submit_draining, 0);
2055 atomic_set(&fs_info->nr_async_bios, 0);
2056 atomic_set(&fs_info->defrag_running, 0);
2057 atomic_set(&fs_info->tree_mod_seq, 0);
2059 fs_info->max_inline = 8192 * 1024;
2060 fs_info->metadata_ratio = 0;
2061 fs_info->defrag_inodes = RB_ROOT;
2062 fs_info->trans_no_join = 0;
2063 fs_info->free_chunk_space = 0;
2064 fs_info->tree_mod_log = RB_ROOT;
2066 /* readahead state */
2067 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2068 spin_lock_init(&fs_info->reada_lock);
2070 fs_info->thread_pool_size = min_t(unsigned long,
2071 num_online_cpus() + 2, 8);
2073 INIT_LIST_HEAD(&fs_info->ordered_extents);
2074 spin_lock_init(&fs_info->ordered_extent_lock);
2075 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2077 if (!fs_info->delayed_root) {
2081 btrfs_init_delayed_root(fs_info->delayed_root);
2083 mutex_init(&fs_info->scrub_lock);
2084 atomic_set(&fs_info->scrubs_running, 0);
2085 atomic_set(&fs_info->scrub_pause_req, 0);
2086 atomic_set(&fs_info->scrubs_paused, 0);
2087 atomic_set(&fs_info->scrub_cancel_req, 0);
2088 init_waitqueue_head(&fs_info->scrub_pause_wait);
2089 init_rwsem(&fs_info->scrub_super_lock);
2090 fs_info->scrub_workers_refcnt = 0;
2091 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2092 fs_info->check_integrity_print_mask = 0;
2095 spin_lock_init(&fs_info->balance_lock);
2096 mutex_init(&fs_info->balance_mutex);
2097 atomic_set(&fs_info->balance_running, 0);
2098 atomic_set(&fs_info->balance_pause_req, 0);
2099 atomic_set(&fs_info->balance_cancel_req, 0);
2100 fs_info->balance_ctl = NULL;
2101 init_waitqueue_head(&fs_info->balance_wait_q);
2103 sb->s_blocksize = 4096;
2104 sb->s_blocksize_bits = blksize_bits(4096);
2105 sb->s_bdi = &fs_info->bdi;
2107 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2108 set_nlink(fs_info->btree_inode, 1);
2110 * we set the i_size on the btree inode to the max possible int.
2111 * the real end of the address space is determined by all of
2112 * the devices in the system
2114 fs_info->btree_inode->i_size = OFFSET_MAX;
2115 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2116 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2118 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2119 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2120 fs_info->btree_inode->i_mapping);
2121 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2122 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2124 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2126 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2127 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2128 sizeof(struct btrfs_key));
2129 set_bit(BTRFS_INODE_DUMMY,
2130 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2131 insert_inode_hash(fs_info->btree_inode);
2133 spin_lock_init(&fs_info->block_group_cache_lock);
2134 fs_info->block_group_cache_tree = RB_ROOT;
2135 fs_info->first_logical_byte = (u64)-1;
2137 extent_io_tree_init(&fs_info->freed_extents[0],
2138 fs_info->btree_inode->i_mapping);
2139 extent_io_tree_init(&fs_info->freed_extents[1],
2140 fs_info->btree_inode->i_mapping);
2141 fs_info->pinned_extents = &fs_info->freed_extents[0];
2142 fs_info->do_barriers = 1;
2145 mutex_init(&fs_info->ordered_operations_mutex);
2146 mutex_init(&fs_info->tree_log_mutex);
2147 mutex_init(&fs_info->chunk_mutex);
2148 mutex_init(&fs_info->transaction_kthread_mutex);
2149 mutex_init(&fs_info->cleaner_mutex);
2150 mutex_init(&fs_info->volume_mutex);
2151 init_rwsem(&fs_info->extent_commit_sem);
2152 init_rwsem(&fs_info->cleanup_work_sem);
2153 init_rwsem(&fs_info->subvol_sem);
2154 fs_info->dev_replace.lock_owner = 0;
2155 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2156 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2157 mutex_init(&fs_info->dev_replace.lock_management_lock);
2158 mutex_init(&fs_info->dev_replace.lock);
2160 spin_lock_init(&fs_info->qgroup_lock);
2161 fs_info->qgroup_tree = RB_ROOT;
2162 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2163 fs_info->qgroup_seq = 1;
2164 fs_info->quota_enabled = 0;
2165 fs_info->pending_quota_state = 0;
2167 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2168 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2170 init_waitqueue_head(&fs_info->transaction_throttle);
2171 init_waitqueue_head(&fs_info->transaction_wait);
2172 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2173 init_waitqueue_head(&fs_info->async_submit_wait);
2175 __setup_root(4096, 4096, 4096, 4096, tree_root,
2176 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2178 invalidate_bdev(fs_devices->latest_bdev);
2179 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2185 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2186 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2187 sizeof(*fs_info->super_for_commit));
2190 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2192 disk_super = fs_info->super_copy;
2193 if (!btrfs_super_root(disk_super))
2196 /* check FS state, whether FS is broken. */
2197 fs_info->fs_state |= btrfs_super_flags(disk_super);
2199 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2201 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2207 * run through our array of backup supers and setup
2208 * our ring pointer to the oldest one
2210 generation = btrfs_super_generation(disk_super);
2211 find_oldest_super_backup(fs_info, generation);
2214 * In the long term, we'll store the compression type in the super
2215 * block, and it'll be used for per file compression control.
2217 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2219 ret = btrfs_parse_options(tree_root, options);
2225 features = btrfs_super_incompat_flags(disk_super) &
2226 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2228 printk(KERN_ERR "BTRFS: couldn't mount because of "
2229 "unsupported optional features (%Lx).\n",
2230 (unsigned long long)features);
2235 if (btrfs_super_leafsize(disk_super) !=
2236 btrfs_super_nodesize(disk_super)) {
2237 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2238 "blocksizes don't match. node %d leaf %d\n",
2239 btrfs_super_nodesize(disk_super),
2240 btrfs_super_leafsize(disk_super));
2244 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2245 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2246 "blocksize (%d) was too large\n",
2247 btrfs_super_leafsize(disk_super));
2252 features = btrfs_super_incompat_flags(disk_super);
2253 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2254 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2255 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2258 * flag our filesystem as having big metadata blocks if
2259 * they are bigger than the page size
2261 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2262 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2263 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2264 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2267 nodesize = btrfs_super_nodesize(disk_super);
2268 leafsize = btrfs_super_leafsize(disk_super);
2269 sectorsize = btrfs_super_sectorsize(disk_super);
2270 stripesize = btrfs_super_stripesize(disk_super);
2271 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2274 * mixed block groups end up with duplicate but slightly offset
2275 * extent buffers for the same range. It leads to corruptions
2277 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2278 (sectorsize != leafsize)) {
2279 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2280 "are not allowed for mixed block groups on %s\n",
2285 btrfs_set_super_incompat_flags(disk_super, features);
2287 features = btrfs_super_compat_ro_flags(disk_super) &
2288 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2289 if (!(sb->s_flags & MS_RDONLY) && features) {
2290 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2291 "unsupported option features (%Lx).\n",
2292 (unsigned long long)features);
2297 btrfs_init_workers(&fs_info->generic_worker,
2298 "genwork", 1, NULL);
2300 btrfs_init_workers(&fs_info->workers, "worker",
2301 fs_info->thread_pool_size,
2302 &fs_info->generic_worker);
2304 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2305 fs_info->thread_pool_size,
2306 &fs_info->generic_worker);
2308 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2309 fs_info->thread_pool_size,
2310 &fs_info->generic_worker);
2312 btrfs_init_workers(&fs_info->submit_workers, "submit",
2313 min_t(u64, fs_devices->num_devices,
2314 fs_info->thread_pool_size),
2315 &fs_info->generic_worker);
2317 btrfs_init_workers(&fs_info->caching_workers, "cache",
2318 2, &fs_info->generic_worker);
2320 /* a higher idle thresh on the submit workers makes it much more
2321 * likely that bios will be send down in a sane order to the
2324 fs_info->submit_workers.idle_thresh = 64;
2326 fs_info->workers.idle_thresh = 16;
2327 fs_info->workers.ordered = 1;
2329 fs_info->delalloc_workers.idle_thresh = 2;
2330 fs_info->delalloc_workers.ordered = 1;
2332 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2333 &fs_info->generic_worker);
2334 btrfs_init_workers(&fs_info->endio_workers, "endio",
2335 fs_info->thread_pool_size,
2336 &fs_info->generic_worker);
2337 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2338 fs_info->thread_pool_size,
2339 &fs_info->generic_worker);
2340 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2341 "endio-meta-write", fs_info->thread_pool_size,
2342 &fs_info->generic_worker);
2343 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2344 fs_info->thread_pool_size,
2345 &fs_info->generic_worker);
2346 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2347 1, &fs_info->generic_worker);
2348 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2349 fs_info->thread_pool_size,
2350 &fs_info->generic_worker);
2351 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2352 fs_info->thread_pool_size,
2353 &fs_info->generic_worker);
2356 * endios are largely parallel and should have a very
2359 fs_info->endio_workers.idle_thresh = 4;
2360 fs_info->endio_meta_workers.idle_thresh = 4;
2362 fs_info->endio_write_workers.idle_thresh = 2;
2363 fs_info->endio_meta_write_workers.idle_thresh = 2;
2364 fs_info->readahead_workers.idle_thresh = 2;
2367 * btrfs_start_workers can really only fail because of ENOMEM so just
2368 * return -ENOMEM if any of these fail.
2370 ret = btrfs_start_workers(&fs_info->workers);
2371 ret |= btrfs_start_workers(&fs_info->generic_worker);
2372 ret |= btrfs_start_workers(&fs_info->submit_workers);
2373 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2374 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2375 ret |= btrfs_start_workers(&fs_info->endio_workers);
2376 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2377 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2378 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2379 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2380 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2381 ret |= btrfs_start_workers(&fs_info->caching_workers);
2382 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2383 ret |= btrfs_start_workers(&fs_info->flush_workers);
2386 goto fail_sb_buffer;
2389 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2390 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2391 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2393 tree_root->nodesize = nodesize;
2394 tree_root->leafsize = leafsize;
2395 tree_root->sectorsize = sectorsize;
2396 tree_root->stripesize = stripesize;
2398 sb->s_blocksize = sectorsize;
2399 sb->s_blocksize_bits = blksize_bits(sectorsize);
2401 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2402 sizeof(disk_super->magic))) {
2403 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2404 goto fail_sb_buffer;
2407 if (sectorsize != PAGE_SIZE) {
2408 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2409 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2410 goto fail_sb_buffer;
2413 mutex_lock(&fs_info->chunk_mutex);
2414 ret = btrfs_read_sys_array(tree_root);
2415 mutex_unlock(&fs_info->chunk_mutex);
2417 printk(KERN_WARNING "btrfs: failed to read the system "
2418 "array on %s\n", sb->s_id);
2419 goto fail_sb_buffer;
2422 blocksize = btrfs_level_size(tree_root,
2423 btrfs_super_chunk_root_level(disk_super));
2424 generation = btrfs_super_chunk_root_generation(disk_super);
2426 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2427 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2429 chunk_root->node = read_tree_block(chunk_root,
2430 btrfs_super_chunk_root(disk_super),
2431 blocksize, generation);
2432 BUG_ON(!chunk_root->node); /* -ENOMEM */
2433 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2434 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2436 goto fail_tree_roots;
2438 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2439 chunk_root->commit_root = btrfs_root_node(chunk_root);
2441 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2442 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2445 ret = btrfs_read_chunk_tree(chunk_root);
2447 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2449 goto fail_tree_roots;
2453 * keep the device that is marked to be the target device for the
2454 * dev_replace procedure
2456 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2458 if (!fs_devices->latest_bdev) {
2459 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2461 goto fail_tree_roots;
2465 blocksize = btrfs_level_size(tree_root,
2466 btrfs_super_root_level(disk_super));
2467 generation = btrfs_super_generation(disk_super);
2469 tree_root->node = read_tree_block(tree_root,
2470 btrfs_super_root(disk_super),
2471 blocksize, generation);
2472 if (!tree_root->node ||
2473 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2474 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2477 goto recovery_tree_root;
2480 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2481 tree_root->commit_root = btrfs_root_node(tree_root);
2483 ret = find_and_setup_root(tree_root, fs_info,
2484 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2486 goto recovery_tree_root;
2487 extent_root->track_dirty = 1;
2489 ret = find_and_setup_root(tree_root, fs_info,
2490 BTRFS_DEV_TREE_OBJECTID, dev_root);
2492 goto recovery_tree_root;
2493 dev_root->track_dirty = 1;
2495 ret = find_and_setup_root(tree_root, fs_info,
2496 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2498 goto recovery_tree_root;
2499 csum_root->track_dirty = 1;
2501 ret = find_and_setup_root(tree_root, fs_info,
2502 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2505 quota_root = fs_info->quota_root = NULL;
2507 quota_root->track_dirty = 1;
2508 fs_info->quota_enabled = 1;
2509 fs_info->pending_quota_state = 1;
2512 fs_info->generation = generation;
2513 fs_info->last_trans_committed = generation;
2515 ret = btrfs_recover_balance(fs_info);
2517 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2518 goto fail_block_groups;
2521 ret = btrfs_init_dev_stats(fs_info);
2523 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2525 goto fail_block_groups;
2528 ret = btrfs_init_dev_replace(fs_info);
2530 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2531 goto fail_block_groups;
2534 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2536 ret = btrfs_init_space_info(fs_info);
2538 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2539 goto fail_block_groups;
2542 ret = btrfs_read_block_groups(extent_root);
2544 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2545 goto fail_block_groups;
2547 fs_info->num_tolerated_disk_barrier_failures =
2548 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2549 if (fs_info->fs_devices->missing_devices >
2550 fs_info->num_tolerated_disk_barrier_failures &&
2551 !(sb->s_flags & MS_RDONLY)) {
2553 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2554 goto fail_block_groups;
2557 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2559 if (IS_ERR(fs_info->cleaner_kthread))
2560 goto fail_block_groups;
2562 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2564 "btrfs-transaction");
2565 if (IS_ERR(fs_info->transaction_kthread))
2568 if (!btrfs_test_opt(tree_root, SSD) &&
2569 !btrfs_test_opt(tree_root, NOSSD) &&
2570 !fs_info->fs_devices->rotating) {
2571 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2573 btrfs_set_opt(fs_info->mount_opt, SSD);
2576 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2577 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2578 ret = btrfsic_mount(tree_root, fs_devices,
2579 btrfs_test_opt(tree_root,
2580 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2582 fs_info->check_integrity_print_mask);
2584 printk(KERN_WARNING "btrfs: failed to initialize"
2585 " integrity check module %s\n", sb->s_id);
2588 ret = btrfs_read_qgroup_config(fs_info);
2590 goto fail_trans_kthread;
2592 /* do not make disk changes in broken FS */
2593 if (btrfs_super_log_root(disk_super) != 0) {
2594 u64 bytenr = btrfs_super_log_root(disk_super);
2596 if (fs_devices->rw_devices == 0) {
2597 printk(KERN_WARNING "Btrfs log replay required "
2603 btrfs_level_size(tree_root,
2604 btrfs_super_log_root_level(disk_super));
2606 log_tree_root = btrfs_alloc_root(fs_info);
2607 if (!log_tree_root) {
2612 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2613 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2615 log_tree_root->node = read_tree_block(tree_root, bytenr,
2618 /* returns with log_tree_root freed on success */
2619 ret = btrfs_recover_log_trees(log_tree_root);
2621 btrfs_error(tree_root->fs_info, ret,
2622 "Failed to recover log tree");
2623 free_extent_buffer(log_tree_root->node);
2624 kfree(log_tree_root);
2625 goto fail_trans_kthread;
2628 if (sb->s_flags & MS_RDONLY) {
2629 ret = btrfs_commit_super(tree_root);
2631 goto fail_trans_kthread;
2635 ret = btrfs_find_orphan_roots(tree_root);
2637 goto fail_trans_kthread;
2639 if (!(sb->s_flags & MS_RDONLY)) {
2640 ret = btrfs_cleanup_fs_roots(fs_info);
2642 goto fail_trans_kthread;
2644 ret = btrfs_recover_relocation(tree_root);
2647 "btrfs: failed to recover relocation\n");
2653 location.objectid = BTRFS_FS_TREE_OBJECTID;
2654 location.type = BTRFS_ROOT_ITEM_KEY;
2655 location.offset = (u64)-1;
2657 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2658 if (!fs_info->fs_root)
2660 if (IS_ERR(fs_info->fs_root)) {
2661 err = PTR_ERR(fs_info->fs_root);
2665 if (sb->s_flags & MS_RDONLY)
2668 down_read(&fs_info->cleanup_work_sem);
2669 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2670 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2671 up_read(&fs_info->cleanup_work_sem);
2672 close_ctree(tree_root);
2675 up_read(&fs_info->cleanup_work_sem);
2677 ret = btrfs_resume_balance_async(fs_info);
2679 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2680 close_ctree(tree_root);
2684 ret = btrfs_resume_dev_replace_async(fs_info);
2686 pr_warn("btrfs: failed to resume dev_replace\n");
2687 close_ctree(tree_root);
2694 btrfs_free_qgroup_config(fs_info);
2696 kthread_stop(fs_info->transaction_kthread);
2698 kthread_stop(fs_info->cleaner_kthread);
2701 * make sure we're done with the btree inode before we stop our
2704 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2705 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2708 btrfs_free_block_groups(fs_info);
2711 free_root_pointers(fs_info, 1);
2714 btrfs_stop_workers(&fs_info->generic_worker);
2715 btrfs_stop_workers(&fs_info->readahead_workers);
2716 btrfs_stop_workers(&fs_info->fixup_workers);
2717 btrfs_stop_workers(&fs_info->delalloc_workers);
2718 btrfs_stop_workers(&fs_info->workers);
2719 btrfs_stop_workers(&fs_info->endio_workers);
2720 btrfs_stop_workers(&fs_info->endio_meta_workers);
2721 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2722 btrfs_stop_workers(&fs_info->endio_write_workers);
2723 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2724 btrfs_stop_workers(&fs_info->submit_workers);
2725 btrfs_stop_workers(&fs_info->delayed_workers);
2726 btrfs_stop_workers(&fs_info->caching_workers);
2727 btrfs_stop_workers(&fs_info->flush_workers);
2730 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2732 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2733 iput(fs_info->btree_inode);
2734 fail_dirty_metadata_bytes:
2735 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2737 bdi_destroy(&fs_info->bdi);
2739 cleanup_srcu_struct(&fs_info->subvol_srcu);
2741 btrfs_close_devices(fs_info->fs_devices);
2745 if (!btrfs_test_opt(tree_root, RECOVERY))
2746 goto fail_tree_roots;
2748 free_root_pointers(fs_info, 0);
2750 /* don't use the log in recovery mode, it won't be valid */
2751 btrfs_set_super_log_root(disk_super, 0);
2753 /* we can't trust the free space cache either */
2754 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2756 ret = next_root_backup(fs_info, fs_info->super_copy,
2757 &num_backups_tried, &backup_index);
2759 goto fail_block_groups;
2760 goto retry_root_backup;
2763 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2766 set_buffer_uptodate(bh);
2768 struct btrfs_device *device = (struct btrfs_device *)
2771 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2772 "I/O error on %s\n",
2773 rcu_str_deref(device->name));
2774 /* note, we dont' set_buffer_write_io_error because we have
2775 * our own ways of dealing with the IO errors
2777 clear_buffer_uptodate(bh);
2778 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2784 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2786 struct buffer_head *bh;
2787 struct buffer_head *latest = NULL;
2788 struct btrfs_super_block *super;
2793 /* we would like to check all the supers, but that would make
2794 * a btrfs mount succeed after a mkfs from a different FS.
2795 * So, we need to add a special mount option to scan for
2796 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2798 for (i = 0; i < 1; i++) {
2799 bytenr = btrfs_sb_offset(i);
2800 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2802 bh = __bread(bdev, bytenr / 4096, 4096);
2806 super = (struct btrfs_super_block *)bh->b_data;
2807 if (btrfs_super_bytenr(super) != bytenr ||
2808 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2809 sizeof(super->magic))) {
2814 if (!latest || btrfs_super_generation(super) > transid) {
2817 transid = btrfs_super_generation(super);
2826 * this should be called twice, once with wait == 0 and
2827 * once with wait == 1. When wait == 0 is done, all the buffer heads
2828 * we write are pinned.
2830 * They are released when wait == 1 is done.
2831 * max_mirrors must be the same for both runs, and it indicates how
2832 * many supers on this one device should be written.
2834 * max_mirrors == 0 means to write them all.
2836 static int write_dev_supers(struct btrfs_device *device,
2837 struct btrfs_super_block *sb,
2838 int do_barriers, int wait, int max_mirrors)
2840 struct buffer_head *bh;
2847 if (max_mirrors == 0)
2848 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2850 for (i = 0; i < max_mirrors; i++) {
2851 bytenr = btrfs_sb_offset(i);
2852 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2856 bh = __find_get_block(device->bdev, bytenr / 4096,
2857 BTRFS_SUPER_INFO_SIZE);
2860 if (!buffer_uptodate(bh))
2863 /* drop our reference */
2866 /* drop the reference from the wait == 0 run */
2870 btrfs_set_super_bytenr(sb, bytenr);
2873 crc = btrfs_csum_data(NULL, (char *)sb +
2874 BTRFS_CSUM_SIZE, crc,
2875 BTRFS_SUPER_INFO_SIZE -
2877 btrfs_csum_final(crc, sb->csum);
2880 * one reference for us, and we leave it for the
2883 bh = __getblk(device->bdev, bytenr / 4096,
2884 BTRFS_SUPER_INFO_SIZE);
2885 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2887 /* one reference for submit_bh */
2890 set_buffer_uptodate(bh);
2892 bh->b_end_io = btrfs_end_buffer_write_sync;
2893 bh->b_private = device;
2897 * we fua the first super. The others we allow
2900 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2904 return errors < i ? 0 : -1;
2908 * endio for the write_dev_flush, this will wake anyone waiting
2909 * for the barrier when it is done
2911 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2914 if (err == -EOPNOTSUPP)
2915 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2916 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2918 if (bio->bi_private)
2919 complete(bio->bi_private);
2924 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2925 * sent down. With wait == 1, it waits for the previous flush.
2927 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2930 static int write_dev_flush(struct btrfs_device *device, int wait)
2935 if (device->nobarriers)
2939 bio = device->flush_bio;
2943 wait_for_completion(&device->flush_wait);
2945 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2946 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2947 rcu_str_deref(device->name));
2948 device->nobarriers = 1;
2949 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
2951 btrfs_dev_stat_inc_and_print(device,
2952 BTRFS_DEV_STAT_FLUSH_ERRS);
2955 /* drop the reference from the wait == 0 run */
2957 device->flush_bio = NULL;
2963 * one reference for us, and we leave it for the
2966 device->flush_bio = NULL;
2967 bio = bio_alloc(GFP_NOFS, 0);
2971 bio->bi_end_io = btrfs_end_empty_barrier;
2972 bio->bi_bdev = device->bdev;
2973 init_completion(&device->flush_wait);
2974 bio->bi_private = &device->flush_wait;
2975 device->flush_bio = bio;
2978 btrfsic_submit_bio(WRITE_FLUSH, bio);
2984 * send an empty flush down to each device in parallel,
2985 * then wait for them
2987 static int barrier_all_devices(struct btrfs_fs_info *info)
2989 struct list_head *head;
2990 struct btrfs_device *dev;
2991 int errors_send = 0;
2992 int errors_wait = 0;
2995 /* send down all the barriers */
2996 head = &info->fs_devices->devices;
2997 list_for_each_entry_rcu(dev, head, dev_list) {
3002 if (!dev->in_fs_metadata || !dev->writeable)
3005 ret = write_dev_flush(dev, 0);
3010 /* wait for all the barriers */
3011 list_for_each_entry_rcu(dev, head, dev_list) {
3016 if (!dev->in_fs_metadata || !dev->writeable)
3019 ret = write_dev_flush(dev, 1);
3023 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3024 errors_wait > info->num_tolerated_disk_barrier_failures)
3029 int btrfs_calc_num_tolerated_disk_barrier_failures(
3030 struct btrfs_fs_info *fs_info)
3032 struct btrfs_ioctl_space_info space;
3033 struct btrfs_space_info *sinfo;
3034 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3035 BTRFS_BLOCK_GROUP_SYSTEM,
3036 BTRFS_BLOCK_GROUP_METADATA,
3037 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3041 int num_tolerated_disk_barrier_failures =
3042 (int)fs_info->fs_devices->num_devices;
3044 for (i = 0; i < num_types; i++) {
3045 struct btrfs_space_info *tmp;
3049 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3050 if (tmp->flags == types[i]) {
3060 down_read(&sinfo->groups_sem);
3061 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3062 if (!list_empty(&sinfo->block_groups[c])) {
3065 btrfs_get_block_group_info(
3066 &sinfo->block_groups[c], &space);
3067 if (space.total_bytes == 0 ||
3068 space.used_bytes == 0)
3070 flags = space.flags;
3073 * 0: if dup, single or RAID0 is configured for
3074 * any of metadata, system or data, else
3075 * 1: if RAID5 is configured, or if RAID1 or
3076 * RAID10 is configured and only two mirrors
3078 * 2: if RAID6 is configured, else
3079 * num_mirrors - 1: if RAID1 or RAID10 is
3080 * configured and more than
3081 * 2 mirrors are used.
3083 if (num_tolerated_disk_barrier_failures > 0 &&
3084 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3085 BTRFS_BLOCK_GROUP_RAID0)) ||
3086 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3088 num_tolerated_disk_barrier_failures = 0;
3089 else if (num_tolerated_disk_barrier_failures > 1
3091 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3092 BTRFS_BLOCK_GROUP_RAID10)))
3093 num_tolerated_disk_barrier_failures = 1;
3096 up_read(&sinfo->groups_sem);
3099 return num_tolerated_disk_barrier_failures;
3102 int write_all_supers(struct btrfs_root *root, int max_mirrors)
3104 struct list_head *head;
3105 struct btrfs_device *dev;
3106 struct btrfs_super_block *sb;
3107 struct btrfs_dev_item *dev_item;
3111 int total_errors = 0;
3114 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3115 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3116 backup_super_roots(root->fs_info);
3118 sb = root->fs_info->super_for_commit;
3119 dev_item = &sb->dev_item;
3121 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3122 head = &root->fs_info->fs_devices->devices;
3125 ret = barrier_all_devices(root->fs_info);
3128 &root->fs_info->fs_devices->device_list_mutex);
3129 btrfs_error(root->fs_info, ret,
3130 "errors while submitting device barriers.");
3135 list_for_each_entry_rcu(dev, head, dev_list) {
3140 if (!dev->in_fs_metadata || !dev->writeable)
3143 btrfs_set_stack_device_generation(dev_item, 0);
3144 btrfs_set_stack_device_type(dev_item, dev->type);
3145 btrfs_set_stack_device_id(dev_item, dev->devid);
3146 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3147 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3148 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3149 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3150 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3151 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3152 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3154 flags = btrfs_super_flags(sb);
3155 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3157 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3161 if (total_errors > max_errors) {
3162 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3165 /* This shouldn't happen. FUA is masked off if unsupported */
3170 list_for_each_entry_rcu(dev, head, dev_list) {
3173 if (!dev->in_fs_metadata || !dev->writeable)
3176 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3180 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3181 if (total_errors > max_errors) {
3182 btrfs_error(root->fs_info, -EIO,
3183 "%d errors while writing supers", total_errors);
3189 int write_ctree_super(struct btrfs_trans_handle *trans,
3190 struct btrfs_root *root, int max_mirrors)
3194 ret = write_all_supers(root, max_mirrors);
3198 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3200 spin_lock(&fs_info->fs_roots_radix_lock);
3201 radix_tree_delete(&fs_info->fs_roots_radix,
3202 (unsigned long)root->root_key.objectid);
3203 spin_unlock(&fs_info->fs_roots_radix_lock);
3205 if (btrfs_root_refs(&root->root_item) == 0)
3206 synchronize_srcu(&fs_info->subvol_srcu);
3208 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3209 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3213 static void free_fs_root(struct btrfs_root *root)
3215 iput(root->cache_inode);
3216 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3218 free_anon_bdev(root->anon_dev);
3219 free_extent_buffer(root->node);
3220 free_extent_buffer(root->commit_root);
3221 kfree(root->free_ino_ctl);
3222 kfree(root->free_ino_pinned);
3227 static void del_fs_roots(struct btrfs_fs_info *fs_info)
3230 struct btrfs_root *gang[8];
3233 while (!list_empty(&fs_info->dead_roots)) {
3234 gang[0] = list_entry(fs_info->dead_roots.next,
3235 struct btrfs_root, root_list);
3236 list_del(&gang[0]->root_list);
3238 if (gang[0]->in_radix) {
3239 btrfs_free_fs_root(fs_info, gang[0]);
3241 free_extent_buffer(gang[0]->node);
3242 free_extent_buffer(gang[0]->commit_root);
3248 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3253 for (i = 0; i < ret; i++)
3254 btrfs_free_fs_root(fs_info, gang[i]);
3258 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3260 u64 root_objectid = 0;
3261 struct btrfs_root *gang[8];
3266 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3267 (void **)gang, root_objectid,
3272 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3273 for (i = 0; i < ret; i++) {
3276 root_objectid = gang[i]->root_key.objectid;
3277 err = btrfs_orphan_cleanup(gang[i]);
3286 int btrfs_commit_super(struct btrfs_root *root)
3288 struct btrfs_trans_handle *trans;
3291 mutex_lock(&root->fs_info->cleaner_mutex);
3292 btrfs_run_delayed_iputs(root);
3293 btrfs_clean_old_snapshots(root);
3294 mutex_unlock(&root->fs_info->cleaner_mutex);
3296 /* wait until ongoing cleanup work done */
3297 down_write(&root->fs_info->cleanup_work_sem);
3298 up_write(&root->fs_info->cleanup_work_sem);
3300 trans = btrfs_join_transaction(root);
3302 return PTR_ERR(trans);
3303 ret = btrfs_commit_transaction(trans, root);
3306 /* run commit again to drop the original snapshot */
3307 trans = btrfs_join_transaction(root);
3309 return PTR_ERR(trans);
3310 ret = btrfs_commit_transaction(trans, root);
3313 ret = btrfs_write_and_wait_transaction(NULL, root);
3315 btrfs_error(root->fs_info, ret,
3316 "Failed to sync btree inode to disk.");
3320 ret = write_ctree_super(NULL, root, 0);
3324 int close_ctree(struct btrfs_root *root)
3326 struct btrfs_fs_info *fs_info = root->fs_info;
3329 fs_info->closing = 1;
3332 /* pause restriper - we want to resume on mount */
3333 btrfs_pause_balance(fs_info);
3335 btrfs_dev_replace_suspend_for_unmount(fs_info);
3337 btrfs_scrub_cancel(fs_info);
3339 /* wait for any defraggers to finish */
3340 wait_event(fs_info->transaction_wait,
3341 (atomic_read(&fs_info->defrag_running) == 0));
3343 /* clear out the rbtree of defraggable inodes */
3344 btrfs_cleanup_defrag_inodes(fs_info);
3346 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3347 ret = btrfs_commit_super(root);
3349 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3352 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3353 btrfs_error_commit_super(root);
3355 btrfs_put_block_group_cache(fs_info);
3357 kthread_stop(fs_info->transaction_kthread);
3358 kthread_stop(fs_info->cleaner_kthread);
3360 fs_info->closing = 2;
3363 btrfs_free_qgroup_config(root->fs_info);
3365 if (fs_info->delalloc_bytes) {
3366 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3367 (unsigned long long)fs_info->delalloc_bytes);
3370 free_extent_buffer(fs_info->extent_root->node);
3371 free_extent_buffer(fs_info->extent_root->commit_root);
3372 free_extent_buffer(fs_info->tree_root->node);
3373 free_extent_buffer(fs_info->tree_root->commit_root);
3374 free_extent_buffer(fs_info->chunk_root->node);
3375 free_extent_buffer(fs_info->chunk_root->commit_root);
3376 free_extent_buffer(fs_info->dev_root->node);
3377 free_extent_buffer(fs_info->dev_root->commit_root);
3378 free_extent_buffer(fs_info->csum_root->node);
3379 free_extent_buffer(fs_info->csum_root->commit_root);
3380 if (fs_info->quota_root) {
3381 free_extent_buffer(fs_info->quota_root->node);
3382 free_extent_buffer(fs_info->quota_root->commit_root);
3385 btrfs_free_block_groups(fs_info);
3387 del_fs_roots(fs_info);
3389 iput(fs_info->btree_inode);
3391 btrfs_stop_workers(&fs_info->generic_worker);
3392 btrfs_stop_workers(&fs_info->fixup_workers);
3393 btrfs_stop_workers(&fs_info->delalloc_workers);
3394 btrfs_stop_workers(&fs_info->workers);
3395 btrfs_stop_workers(&fs_info->endio_workers);
3396 btrfs_stop_workers(&fs_info->endio_meta_workers);
3397 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3398 btrfs_stop_workers(&fs_info->endio_write_workers);
3399 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3400 btrfs_stop_workers(&fs_info->submit_workers);
3401 btrfs_stop_workers(&fs_info->delayed_workers);
3402 btrfs_stop_workers(&fs_info->caching_workers);
3403 btrfs_stop_workers(&fs_info->readahead_workers);
3404 btrfs_stop_workers(&fs_info->flush_workers);
3406 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3407 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3408 btrfsic_unmount(root, fs_info->fs_devices);
3411 btrfs_close_devices(fs_info->fs_devices);
3412 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3414 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3415 bdi_destroy(&fs_info->bdi);
3416 cleanup_srcu_struct(&fs_info->subvol_srcu);
3421 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3425 struct inode *btree_inode = buf->pages[0]->mapping->host;
3427 ret = extent_buffer_uptodate(buf);
3431 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3432 parent_transid, atomic);
3438 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3440 return set_extent_buffer_uptodate(buf);
3443 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3445 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3446 u64 transid = btrfs_header_generation(buf);
3449 btrfs_assert_tree_locked(buf);
3450 if (transid != root->fs_info->generation)
3451 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3452 "found %llu running %llu\n",
3453 (unsigned long long)buf->start,
3454 (unsigned long long)transid,
3455 (unsigned long long)root->fs_info->generation);
3456 was_dirty = set_extent_buffer_dirty(buf);
3458 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3460 root->fs_info->dirty_metadata_batch);
3463 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3467 * looks as though older kernels can get into trouble with
3468 * this code, they end up stuck in balance_dirty_pages forever
3472 if (current->flags & PF_MEMALLOC)
3476 btrfs_balance_delayed_items(root);
3478 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3479 BTRFS_DIRTY_METADATA_THRESH);
3481 balance_dirty_pages_ratelimited_nr(
3482 root->fs_info->btree_inode->i_mapping, 1);
3487 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3489 __btrfs_btree_balance_dirty(root, 1);
3492 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3494 __btrfs_btree_balance_dirty(root, 0);
3497 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3499 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3500 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3503 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3506 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3507 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3517 void btrfs_error_commit_super(struct btrfs_root *root)
3519 mutex_lock(&root->fs_info->cleaner_mutex);
3520 btrfs_run_delayed_iputs(root);
3521 mutex_unlock(&root->fs_info->cleaner_mutex);
3523 down_write(&root->fs_info->cleanup_work_sem);
3524 up_write(&root->fs_info->cleanup_work_sem);
3526 /* cleanup FS via transaction */
3527 btrfs_cleanup_transaction(root);
3530 static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3532 struct btrfs_inode *btrfs_inode;
3533 struct list_head splice;
3535 INIT_LIST_HEAD(&splice);
3537 mutex_lock(&root->fs_info->ordered_operations_mutex);
3538 spin_lock(&root->fs_info->ordered_extent_lock);
3540 list_splice_init(&root->fs_info->ordered_operations, &splice);
3541 while (!list_empty(&splice)) {
3542 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3543 ordered_operations);
3545 list_del_init(&btrfs_inode->ordered_operations);
3547 btrfs_invalidate_inodes(btrfs_inode->root);
3550 spin_unlock(&root->fs_info->ordered_extent_lock);
3551 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3554 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3556 struct list_head splice;
3557 struct btrfs_ordered_extent *ordered;
3558 struct inode *inode;
3560 INIT_LIST_HEAD(&splice);
3562 spin_lock(&root->fs_info->ordered_extent_lock);
3564 list_splice_init(&root->fs_info->ordered_extents, &splice);
3565 while (!list_empty(&splice)) {
3566 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3569 list_del_init(&ordered->root_extent_list);
3570 atomic_inc(&ordered->refs);
3572 /* the inode may be getting freed (in sys_unlink path). */
3573 inode = igrab(ordered->inode);
3575 spin_unlock(&root->fs_info->ordered_extent_lock);
3579 atomic_set(&ordered->refs, 1);
3580 btrfs_put_ordered_extent(ordered);
3582 spin_lock(&root->fs_info->ordered_extent_lock);
3585 spin_unlock(&root->fs_info->ordered_extent_lock);
3588 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3589 struct btrfs_root *root)
3591 struct rb_node *node;
3592 struct btrfs_delayed_ref_root *delayed_refs;
3593 struct btrfs_delayed_ref_node *ref;
3596 delayed_refs = &trans->delayed_refs;
3598 spin_lock(&delayed_refs->lock);
3599 if (delayed_refs->num_entries == 0) {
3600 spin_unlock(&delayed_refs->lock);
3601 printk(KERN_INFO "delayed_refs has NO entry\n");
3605 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3606 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3608 atomic_set(&ref->refs, 1);
3609 if (btrfs_delayed_ref_is_head(ref)) {
3610 struct btrfs_delayed_ref_head *head;
3612 head = btrfs_delayed_node_to_head(ref);
3613 if (!mutex_trylock(&head->mutex)) {
3614 atomic_inc(&ref->refs);
3615 spin_unlock(&delayed_refs->lock);
3617 /* Need to wait for the delayed ref to run */
3618 mutex_lock(&head->mutex);
3619 mutex_unlock(&head->mutex);
3620 btrfs_put_delayed_ref(ref);
3622 spin_lock(&delayed_refs->lock);
3626 btrfs_free_delayed_extent_op(head->extent_op);
3627 delayed_refs->num_heads--;
3628 if (list_empty(&head->cluster))
3629 delayed_refs->num_heads_ready--;
3630 list_del_init(&head->cluster);
3633 rb_erase(&ref->rb_node, &delayed_refs->root);
3634 delayed_refs->num_entries--;
3636 spin_unlock(&delayed_refs->lock);
3637 btrfs_put_delayed_ref(ref);
3640 spin_lock(&delayed_refs->lock);
3643 spin_unlock(&delayed_refs->lock);
3648 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3650 struct btrfs_pending_snapshot *snapshot;
3651 struct list_head splice;
3653 INIT_LIST_HEAD(&splice);
3655 list_splice_init(&t->pending_snapshots, &splice);
3657 while (!list_empty(&splice)) {
3658 snapshot = list_entry(splice.next,
3659 struct btrfs_pending_snapshot,
3662 list_del_init(&snapshot->list);
3668 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3670 struct btrfs_inode *btrfs_inode;
3671 struct list_head splice;
3673 INIT_LIST_HEAD(&splice);
3675 spin_lock(&root->fs_info->delalloc_lock);
3676 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3678 while (!list_empty(&splice)) {
3679 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3682 list_del_init(&btrfs_inode->delalloc_inodes);
3684 btrfs_invalidate_inodes(btrfs_inode->root);
3687 spin_unlock(&root->fs_info->delalloc_lock);
3690 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3691 struct extent_io_tree *dirty_pages,
3696 struct inode *btree_inode = root->fs_info->btree_inode;
3697 struct extent_buffer *eb;
3701 unsigned long index;
3704 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3709 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3710 while (start <= end) {
3711 index = start >> PAGE_CACHE_SHIFT;
3712 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3713 page = find_get_page(btree_inode->i_mapping, index);
3716 offset = page_offset(page);
3718 spin_lock(&dirty_pages->buffer_lock);
3719 eb = radix_tree_lookup(
3720 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3721 offset >> PAGE_CACHE_SHIFT);
3722 spin_unlock(&dirty_pages->buffer_lock);
3724 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3726 if (PageWriteback(page))
3727 end_page_writeback(page);
3730 if (PageDirty(page)) {
3731 clear_page_dirty_for_io(page);
3732 spin_lock_irq(&page->mapping->tree_lock);
3733 radix_tree_tag_clear(&page->mapping->page_tree,
3735 PAGECACHE_TAG_DIRTY);
3736 spin_unlock_irq(&page->mapping->tree_lock);
3740 page_cache_release(page);
3747 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3748 struct extent_io_tree *pinned_extents)
3750 struct extent_io_tree *unpin;
3756 unpin = pinned_extents;
3759 ret = find_first_extent_bit(unpin, 0, &start, &end,
3760 EXTENT_DIRTY, NULL);
3765 if (btrfs_test_opt(root, DISCARD))
3766 ret = btrfs_error_discard_extent(root, start,
3770 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3771 btrfs_error_unpin_extent_range(root, start, end);
3776 if (unpin == &root->fs_info->freed_extents[0])
3777 unpin = &root->fs_info->freed_extents[1];
3779 unpin = &root->fs_info->freed_extents[0];
3787 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3788 struct btrfs_root *root)
3790 btrfs_destroy_delayed_refs(cur_trans, root);
3791 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3792 cur_trans->dirty_pages.dirty_bytes);
3794 /* FIXME: cleanup wait for commit */
3795 cur_trans->in_commit = 1;
3796 cur_trans->blocked = 1;
3797 wake_up(&root->fs_info->transaction_blocked_wait);
3799 cur_trans->blocked = 0;
3800 wake_up(&root->fs_info->transaction_wait);
3802 cur_trans->commit_done = 1;
3803 wake_up(&cur_trans->commit_wait);
3805 btrfs_destroy_delayed_inodes(root);
3806 btrfs_assert_delayed_root_empty(root);
3808 btrfs_destroy_pending_snapshots(cur_trans);
3810 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3812 btrfs_destroy_pinned_extent(root,
3813 root->fs_info->pinned_extents);
3816 memset(cur_trans, 0, sizeof(*cur_trans));
3817 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3821 int btrfs_cleanup_transaction(struct btrfs_root *root)
3823 struct btrfs_transaction *t;
3826 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3828 spin_lock(&root->fs_info->trans_lock);
3829 list_splice_init(&root->fs_info->trans_list, &list);
3830 root->fs_info->trans_no_join = 1;
3831 spin_unlock(&root->fs_info->trans_lock);
3833 while (!list_empty(&list)) {
3834 t = list_entry(list.next, struct btrfs_transaction, list);
3838 btrfs_destroy_ordered_operations(root);
3840 btrfs_destroy_ordered_extents(root);
3842 btrfs_destroy_delayed_refs(t, root);
3844 btrfs_block_rsv_release(root,
3845 &root->fs_info->trans_block_rsv,
3846 t->dirty_pages.dirty_bytes);
3848 /* FIXME: cleanup wait for commit */
3852 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3853 wake_up(&root->fs_info->transaction_blocked_wait);
3857 if (waitqueue_active(&root->fs_info->transaction_wait))
3858 wake_up(&root->fs_info->transaction_wait);
3862 if (waitqueue_active(&t->commit_wait))
3863 wake_up(&t->commit_wait);
3865 btrfs_destroy_delayed_inodes(root);
3866 btrfs_assert_delayed_root_empty(root);
3868 btrfs_destroy_pending_snapshots(t);
3870 btrfs_destroy_delalloc_inodes(root);
3872 spin_lock(&root->fs_info->trans_lock);
3873 root->fs_info->running_transaction = NULL;
3874 spin_unlock(&root->fs_info->trans_lock);
3876 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3879 btrfs_destroy_pinned_extent(root,
3880 root->fs_info->pinned_extents);
3882 atomic_set(&t->use_count, 0);
3883 list_del_init(&t->list);
3884 memset(t, 0, sizeof(*t));
3885 kmem_cache_free(btrfs_transaction_cachep, t);
3888 spin_lock(&root->fs_info->trans_lock);
3889 root->fs_info->trans_no_join = 0;
3890 spin_unlock(&root->fs_info->trans_lock);
3891 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3896 static struct extent_io_ops btree_extent_io_ops = {
3897 .readpage_end_io_hook = btree_readpage_end_io_hook,
3898 .readpage_io_failed_hook = btree_io_failed_hook,
3899 .submit_bio_hook = btree_submit_bio_hook,
3900 /* note we're sharing with inode.c for the merge bio hook */
3901 .merge_bio_hook = btrfs_merge_bio_hook,