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 tree = &BTRFS_I(mapping->host)->io_tree;
950 if (wbc->sync_mode == WB_SYNC_NONE) {
951 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
953 unsigned long thresh = 32 * 1024 * 1024;
955 if (wbc->for_kupdate)
958 /* this is a bit racy, but that's ok */
959 num_dirty = root->fs_info->dirty_metadata_bytes;
960 if (num_dirty < thresh)
963 return btree_write_cache_pages(mapping, wbc);
966 static int btree_readpage(struct file *file, struct page *page)
968 struct extent_io_tree *tree;
969 tree = &BTRFS_I(page->mapping->host)->io_tree;
970 return extent_read_full_page(tree, page, btree_get_extent, 0);
973 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
975 if (PageWriteback(page) || PageDirty(page))
978 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
979 * slab allocation from alloc_extent_state down the callchain where
980 * it'd hit a BUG_ON as those flags are not allowed.
982 gfp_flags &= ~GFP_SLAB_BUG_MASK;
984 return try_release_extent_buffer(page, gfp_flags);
987 static void btree_invalidatepage(struct page *page, unsigned long offset)
989 struct extent_io_tree *tree;
990 tree = &BTRFS_I(page->mapping->host)->io_tree;
991 extent_invalidatepage(tree, page, offset);
992 btree_releasepage(page, GFP_NOFS);
993 if (PagePrivate(page)) {
994 printk(KERN_WARNING "btrfs warning page private not zero "
995 "on page %llu\n", (unsigned long long)page_offset(page));
996 ClearPagePrivate(page);
997 set_page_private(page, 0);
998 page_cache_release(page);
1002 static int btree_set_page_dirty(struct page *page)
1005 struct extent_buffer *eb;
1007 BUG_ON(!PagePrivate(page));
1008 eb = (struct extent_buffer *)page->private;
1010 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1011 BUG_ON(!atomic_read(&eb->refs));
1012 btrfs_assert_tree_locked(eb);
1014 return __set_page_dirty_nobuffers(page);
1017 static const struct address_space_operations btree_aops = {
1018 .readpage = btree_readpage,
1019 .writepages = btree_writepages,
1020 .releasepage = btree_releasepage,
1021 .invalidatepage = btree_invalidatepage,
1022 #ifdef CONFIG_MIGRATION
1023 .migratepage = btree_migratepage,
1025 .set_page_dirty = btree_set_page_dirty,
1028 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1031 struct extent_buffer *buf = NULL;
1032 struct inode *btree_inode = root->fs_info->btree_inode;
1035 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1038 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1039 buf, 0, WAIT_NONE, btree_get_extent, 0);
1040 free_extent_buffer(buf);
1044 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1045 int mirror_num, struct extent_buffer **eb)
1047 struct extent_buffer *buf = NULL;
1048 struct inode *btree_inode = root->fs_info->btree_inode;
1049 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1052 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1056 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1058 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1059 btree_get_extent, mirror_num);
1061 free_extent_buffer(buf);
1065 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1066 free_extent_buffer(buf);
1068 } else if (extent_buffer_uptodate(buf)) {
1071 free_extent_buffer(buf);
1076 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1077 u64 bytenr, u32 blocksize)
1079 struct inode *btree_inode = root->fs_info->btree_inode;
1080 struct extent_buffer *eb;
1081 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1086 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1087 u64 bytenr, u32 blocksize)
1089 struct inode *btree_inode = root->fs_info->btree_inode;
1090 struct extent_buffer *eb;
1092 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1098 int btrfs_write_tree_block(struct extent_buffer *buf)
1100 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1101 buf->start + buf->len - 1);
1104 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1106 return filemap_fdatawait_range(buf->pages[0]->mapping,
1107 buf->start, buf->start + buf->len - 1);
1110 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1111 u32 blocksize, u64 parent_transid)
1113 struct extent_buffer *buf = NULL;
1116 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1120 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1125 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1126 struct extent_buffer *buf)
1128 if (btrfs_header_generation(buf) ==
1129 root->fs_info->running_transaction->transid) {
1130 btrfs_assert_tree_locked(buf);
1132 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1133 spin_lock(&root->fs_info->delalloc_lock);
1134 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1135 root->fs_info->dirty_metadata_bytes -= buf->len;
1137 spin_unlock(&root->fs_info->delalloc_lock);
1138 btrfs_panic(root->fs_info, -EOVERFLOW,
1139 "Can't clear %lu bytes from "
1140 " dirty_mdatadata_bytes (%llu)",
1142 root->fs_info->dirty_metadata_bytes);
1144 spin_unlock(&root->fs_info->delalloc_lock);
1146 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1147 btrfs_set_lock_blocking(buf);
1148 clear_extent_buffer_dirty(buf);
1153 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1154 u32 stripesize, struct btrfs_root *root,
1155 struct btrfs_fs_info *fs_info,
1159 root->commit_root = NULL;
1160 root->sectorsize = sectorsize;
1161 root->nodesize = nodesize;
1162 root->leafsize = leafsize;
1163 root->stripesize = stripesize;
1165 root->track_dirty = 0;
1167 root->orphan_item_inserted = 0;
1168 root->orphan_cleanup_state = 0;
1170 root->objectid = objectid;
1171 root->last_trans = 0;
1172 root->highest_objectid = 0;
1174 root->inode_tree = RB_ROOT;
1175 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1176 root->block_rsv = NULL;
1177 root->orphan_block_rsv = NULL;
1179 INIT_LIST_HEAD(&root->dirty_list);
1180 INIT_LIST_HEAD(&root->root_list);
1181 INIT_LIST_HEAD(&root->logged_list[0]);
1182 INIT_LIST_HEAD(&root->logged_list[1]);
1183 spin_lock_init(&root->orphan_lock);
1184 spin_lock_init(&root->inode_lock);
1185 spin_lock_init(&root->accounting_lock);
1186 spin_lock_init(&root->log_extents_lock[0]);
1187 spin_lock_init(&root->log_extents_lock[1]);
1188 mutex_init(&root->objectid_mutex);
1189 mutex_init(&root->log_mutex);
1190 init_waitqueue_head(&root->log_writer_wait);
1191 init_waitqueue_head(&root->log_commit_wait[0]);
1192 init_waitqueue_head(&root->log_commit_wait[1]);
1193 atomic_set(&root->log_commit[0], 0);
1194 atomic_set(&root->log_commit[1], 0);
1195 atomic_set(&root->log_writers, 0);
1196 atomic_set(&root->log_batch, 0);
1197 atomic_set(&root->orphan_inodes, 0);
1198 root->log_transid = 0;
1199 root->last_log_commit = 0;
1200 extent_io_tree_init(&root->dirty_log_pages,
1201 fs_info->btree_inode->i_mapping);
1203 memset(&root->root_key, 0, sizeof(root->root_key));
1204 memset(&root->root_item, 0, sizeof(root->root_item));
1205 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1206 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1207 root->defrag_trans_start = fs_info->generation;
1208 init_completion(&root->kobj_unregister);
1209 root->defrag_running = 0;
1210 root->root_key.objectid = objectid;
1213 spin_lock_init(&root->root_item_lock);
1216 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1217 struct btrfs_fs_info *fs_info,
1219 struct btrfs_root *root)
1225 __setup_root(tree_root->nodesize, tree_root->leafsize,
1226 tree_root->sectorsize, tree_root->stripesize,
1227 root, fs_info, objectid);
1228 ret = btrfs_find_last_root(tree_root, objectid,
1229 &root->root_item, &root->root_key);
1235 generation = btrfs_root_generation(&root->root_item);
1236 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1237 root->commit_root = NULL;
1238 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1239 blocksize, generation);
1240 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1241 free_extent_buffer(root->node);
1245 root->commit_root = btrfs_root_node(root);
1249 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1251 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1253 root->fs_info = fs_info;
1257 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1258 struct btrfs_fs_info *fs_info,
1261 struct extent_buffer *leaf;
1262 struct btrfs_root *tree_root = fs_info->tree_root;
1263 struct btrfs_root *root;
1264 struct btrfs_key key;
1268 root = btrfs_alloc_root(fs_info);
1270 return ERR_PTR(-ENOMEM);
1272 __setup_root(tree_root->nodesize, tree_root->leafsize,
1273 tree_root->sectorsize, tree_root->stripesize,
1274 root, fs_info, objectid);
1275 root->root_key.objectid = objectid;
1276 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1277 root->root_key.offset = 0;
1279 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1280 0, objectid, NULL, 0, 0, 0);
1282 ret = PTR_ERR(leaf);
1286 bytenr = leaf->start;
1287 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1288 btrfs_set_header_bytenr(leaf, leaf->start);
1289 btrfs_set_header_generation(leaf, trans->transid);
1290 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1291 btrfs_set_header_owner(leaf, objectid);
1294 write_extent_buffer(leaf, fs_info->fsid,
1295 (unsigned long)btrfs_header_fsid(leaf),
1297 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1298 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1300 btrfs_mark_buffer_dirty(leaf);
1302 root->commit_root = btrfs_root_node(root);
1303 root->track_dirty = 1;
1306 root->root_item.flags = 0;
1307 root->root_item.byte_limit = 0;
1308 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1309 btrfs_set_root_generation(&root->root_item, trans->transid);
1310 btrfs_set_root_level(&root->root_item, 0);
1311 btrfs_set_root_refs(&root->root_item, 1);
1312 btrfs_set_root_used(&root->root_item, leaf->len);
1313 btrfs_set_root_last_snapshot(&root->root_item, 0);
1314 btrfs_set_root_dirid(&root->root_item, 0);
1315 root->root_item.drop_level = 0;
1317 key.objectid = objectid;
1318 key.type = BTRFS_ROOT_ITEM_KEY;
1320 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1324 btrfs_tree_unlock(leaf);
1328 return ERR_PTR(ret);
1333 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1334 struct btrfs_fs_info *fs_info)
1336 struct btrfs_root *root;
1337 struct btrfs_root *tree_root = fs_info->tree_root;
1338 struct extent_buffer *leaf;
1340 root = btrfs_alloc_root(fs_info);
1342 return ERR_PTR(-ENOMEM);
1344 __setup_root(tree_root->nodesize, tree_root->leafsize,
1345 tree_root->sectorsize, tree_root->stripesize,
1346 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1348 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1349 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1350 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1352 * log trees do not get reference counted because they go away
1353 * before a real commit is actually done. They do store pointers
1354 * to file data extents, and those reference counts still get
1355 * updated (along with back refs to the log tree).
1359 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1360 BTRFS_TREE_LOG_OBJECTID, NULL,
1364 return ERR_CAST(leaf);
1367 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1368 btrfs_set_header_bytenr(leaf, leaf->start);
1369 btrfs_set_header_generation(leaf, trans->transid);
1370 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1371 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1374 write_extent_buffer(root->node, root->fs_info->fsid,
1375 (unsigned long)btrfs_header_fsid(root->node),
1377 btrfs_mark_buffer_dirty(root->node);
1378 btrfs_tree_unlock(root->node);
1382 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1383 struct btrfs_fs_info *fs_info)
1385 struct btrfs_root *log_root;
1387 log_root = alloc_log_tree(trans, fs_info);
1388 if (IS_ERR(log_root))
1389 return PTR_ERR(log_root);
1390 WARN_ON(fs_info->log_root_tree);
1391 fs_info->log_root_tree = log_root;
1395 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1396 struct btrfs_root *root)
1398 struct btrfs_root *log_root;
1399 struct btrfs_inode_item *inode_item;
1401 log_root = alloc_log_tree(trans, root->fs_info);
1402 if (IS_ERR(log_root))
1403 return PTR_ERR(log_root);
1405 log_root->last_trans = trans->transid;
1406 log_root->root_key.offset = root->root_key.objectid;
1408 inode_item = &log_root->root_item.inode;
1409 inode_item->generation = cpu_to_le64(1);
1410 inode_item->size = cpu_to_le64(3);
1411 inode_item->nlink = cpu_to_le32(1);
1412 inode_item->nbytes = cpu_to_le64(root->leafsize);
1413 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1415 btrfs_set_root_node(&log_root->root_item, log_root->node);
1417 WARN_ON(root->log_root);
1418 root->log_root = log_root;
1419 root->log_transid = 0;
1420 root->last_log_commit = 0;
1424 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1425 struct btrfs_key *location)
1427 struct btrfs_root *root;
1428 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1429 struct btrfs_path *path;
1430 struct extent_buffer *l;
1436 root = btrfs_alloc_root(fs_info);
1438 return ERR_PTR(-ENOMEM);
1439 if (location->offset == (u64)-1) {
1440 ret = find_and_setup_root(tree_root, fs_info,
1441 location->objectid, root);
1444 return ERR_PTR(ret);
1449 __setup_root(tree_root->nodesize, tree_root->leafsize,
1450 tree_root->sectorsize, tree_root->stripesize,
1451 root, fs_info, location->objectid);
1453 path = btrfs_alloc_path();
1456 return ERR_PTR(-ENOMEM);
1458 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1461 slot = path->slots[0];
1462 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1463 memcpy(&root->root_key, location, sizeof(*location));
1465 btrfs_free_path(path);
1470 return ERR_PTR(ret);
1473 generation = btrfs_root_generation(&root->root_item);
1474 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1475 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1476 blocksize, generation);
1477 root->commit_root = btrfs_root_node(root);
1478 BUG_ON(!root->node); /* -ENOMEM */
1480 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1482 btrfs_check_and_init_root_item(&root->root_item);
1488 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1489 struct btrfs_key *location)
1491 struct btrfs_root *root;
1494 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1495 return fs_info->tree_root;
1496 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1497 return fs_info->extent_root;
1498 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1499 return fs_info->chunk_root;
1500 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1501 return fs_info->dev_root;
1502 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1503 return fs_info->csum_root;
1504 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1505 return fs_info->quota_root ? fs_info->quota_root :
1508 spin_lock(&fs_info->fs_roots_radix_lock);
1509 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1510 (unsigned long)location->objectid);
1511 spin_unlock(&fs_info->fs_roots_radix_lock);
1515 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1519 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1520 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1522 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1527 btrfs_init_free_ino_ctl(root);
1528 mutex_init(&root->fs_commit_mutex);
1529 spin_lock_init(&root->cache_lock);
1530 init_waitqueue_head(&root->cache_wait);
1532 ret = get_anon_bdev(&root->anon_dev);
1536 if (btrfs_root_refs(&root->root_item) == 0) {
1541 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1545 root->orphan_item_inserted = 1;
1547 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1551 spin_lock(&fs_info->fs_roots_radix_lock);
1552 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1553 (unsigned long)root->root_key.objectid,
1558 spin_unlock(&fs_info->fs_roots_radix_lock);
1559 radix_tree_preload_end();
1561 if (ret == -EEXIST) {
1568 ret = btrfs_find_dead_roots(fs_info->tree_root,
1569 root->root_key.objectid);
1574 return ERR_PTR(ret);
1577 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1579 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1581 struct btrfs_device *device;
1582 struct backing_dev_info *bdi;
1585 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1588 bdi = blk_get_backing_dev_info(device->bdev);
1589 if (bdi && bdi_congested(bdi, bdi_bits)) {
1599 * If this fails, caller must call bdi_destroy() to get rid of the
1602 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1606 bdi->capabilities = BDI_CAP_MAP_COPY;
1607 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1611 bdi->ra_pages = default_backing_dev_info.ra_pages;
1612 bdi->congested_fn = btrfs_congested_fn;
1613 bdi->congested_data = info;
1618 * called by the kthread helper functions to finally call the bio end_io
1619 * functions. This is where read checksum verification actually happens
1621 static void end_workqueue_fn(struct btrfs_work *work)
1624 struct end_io_wq *end_io_wq;
1625 struct btrfs_fs_info *fs_info;
1628 end_io_wq = container_of(work, struct end_io_wq, work);
1629 bio = end_io_wq->bio;
1630 fs_info = end_io_wq->info;
1632 error = end_io_wq->error;
1633 bio->bi_private = end_io_wq->private;
1634 bio->bi_end_io = end_io_wq->end_io;
1636 bio_endio(bio, error);
1639 static int cleaner_kthread(void *arg)
1641 struct btrfs_root *root = arg;
1644 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1645 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1646 btrfs_run_delayed_iputs(root);
1647 btrfs_clean_old_snapshots(root);
1648 mutex_unlock(&root->fs_info->cleaner_mutex);
1649 btrfs_run_defrag_inodes(root->fs_info);
1652 if (!try_to_freeze()) {
1653 set_current_state(TASK_INTERRUPTIBLE);
1654 if (!kthread_should_stop())
1656 __set_current_state(TASK_RUNNING);
1658 } while (!kthread_should_stop());
1662 static int transaction_kthread(void *arg)
1664 struct btrfs_root *root = arg;
1665 struct btrfs_trans_handle *trans;
1666 struct btrfs_transaction *cur;
1669 unsigned long delay;
1673 cannot_commit = false;
1675 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1677 spin_lock(&root->fs_info->trans_lock);
1678 cur = root->fs_info->running_transaction;
1680 spin_unlock(&root->fs_info->trans_lock);
1684 now = get_seconds();
1685 if (!cur->blocked &&
1686 (now < cur->start_time || now - cur->start_time < 30)) {
1687 spin_unlock(&root->fs_info->trans_lock);
1691 transid = cur->transid;
1692 spin_unlock(&root->fs_info->trans_lock);
1694 /* If the file system is aborted, this will always fail. */
1695 trans = btrfs_attach_transaction(root);
1696 if (IS_ERR(trans)) {
1697 if (PTR_ERR(trans) != -ENOENT)
1698 cannot_commit = true;
1701 if (transid == trans->transid) {
1702 btrfs_commit_transaction(trans, root);
1704 btrfs_end_transaction(trans, root);
1707 wake_up_process(root->fs_info->cleaner_kthread);
1708 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1710 if (!try_to_freeze()) {
1711 set_current_state(TASK_INTERRUPTIBLE);
1712 if (!kthread_should_stop() &&
1713 (!btrfs_transaction_blocked(root->fs_info) ||
1715 schedule_timeout(delay);
1716 __set_current_state(TASK_RUNNING);
1718 } while (!kthread_should_stop());
1723 * this will find the highest generation in the array of
1724 * root backups. The index of the highest array is returned,
1725 * or -1 if we can't find anything.
1727 * We check to make sure the array is valid by comparing the
1728 * generation of the latest root in the array with the generation
1729 * in the super block. If they don't match we pitch it.
1731 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1734 int newest_index = -1;
1735 struct btrfs_root_backup *root_backup;
1738 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1739 root_backup = info->super_copy->super_roots + i;
1740 cur = btrfs_backup_tree_root_gen(root_backup);
1741 if (cur == newest_gen)
1745 /* check to see if we actually wrapped around */
1746 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1747 root_backup = info->super_copy->super_roots;
1748 cur = btrfs_backup_tree_root_gen(root_backup);
1749 if (cur == newest_gen)
1752 return newest_index;
1757 * find the oldest backup so we know where to store new entries
1758 * in the backup array. This will set the backup_root_index
1759 * field in the fs_info struct
1761 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1764 int newest_index = -1;
1766 newest_index = find_newest_super_backup(info, newest_gen);
1767 /* if there was garbage in there, just move along */
1768 if (newest_index == -1) {
1769 info->backup_root_index = 0;
1771 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1776 * copy all the root pointers into the super backup array.
1777 * this will bump the backup pointer by one when it is
1780 static void backup_super_roots(struct btrfs_fs_info *info)
1783 struct btrfs_root_backup *root_backup;
1786 next_backup = info->backup_root_index;
1787 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1788 BTRFS_NUM_BACKUP_ROOTS;
1791 * just overwrite the last backup if we're at the same generation
1792 * this happens only at umount
1794 root_backup = info->super_for_commit->super_roots + last_backup;
1795 if (btrfs_backup_tree_root_gen(root_backup) ==
1796 btrfs_header_generation(info->tree_root->node))
1797 next_backup = last_backup;
1799 root_backup = info->super_for_commit->super_roots + next_backup;
1802 * make sure all of our padding and empty slots get zero filled
1803 * regardless of which ones we use today
1805 memset(root_backup, 0, sizeof(*root_backup));
1807 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1809 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1810 btrfs_set_backup_tree_root_gen(root_backup,
1811 btrfs_header_generation(info->tree_root->node));
1813 btrfs_set_backup_tree_root_level(root_backup,
1814 btrfs_header_level(info->tree_root->node));
1816 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1817 btrfs_set_backup_chunk_root_gen(root_backup,
1818 btrfs_header_generation(info->chunk_root->node));
1819 btrfs_set_backup_chunk_root_level(root_backup,
1820 btrfs_header_level(info->chunk_root->node));
1822 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1823 btrfs_set_backup_extent_root_gen(root_backup,
1824 btrfs_header_generation(info->extent_root->node));
1825 btrfs_set_backup_extent_root_level(root_backup,
1826 btrfs_header_level(info->extent_root->node));
1829 * we might commit during log recovery, which happens before we set
1830 * the fs_root. Make sure it is valid before we fill it in.
1832 if (info->fs_root && info->fs_root->node) {
1833 btrfs_set_backup_fs_root(root_backup,
1834 info->fs_root->node->start);
1835 btrfs_set_backup_fs_root_gen(root_backup,
1836 btrfs_header_generation(info->fs_root->node));
1837 btrfs_set_backup_fs_root_level(root_backup,
1838 btrfs_header_level(info->fs_root->node));
1841 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1842 btrfs_set_backup_dev_root_gen(root_backup,
1843 btrfs_header_generation(info->dev_root->node));
1844 btrfs_set_backup_dev_root_level(root_backup,
1845 btrfs_header_level(info->dev_root->node));
1847 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1848 btrfs_set_backup_csum_root_gen(root_backup,
1849 btrfs_header_generation(info->csum_root->node));
1850 btrfs_set_backup_csum_root_level(root_backup,
1851 btrfs_header_level(info->csum_root->node));
1853 btrfs_set_backup_total_bytes(root_backup,
1854 btrfs_super_total_bytes(info->super_copy));
1855 btrfs_set_backup_bytes_used(root_backup,
1856 btrfs_super_bytes_used(info->super_copy));
1857 btrfs_set_backup_num_devices(root_backup,
1858 btrfs_super_num_devices(info->super_copy));
1861 * if we don't copy this out to the super_copy, it won't get remembered
1862 * for the next commit
1864 memcpy(&info->super_copy->super_roots,
1865 &info->super_for_commit->super_roots,
1866 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1870 * this copies info out of the root backup array and back into
1871 * the in-memory super block. It is meant to help iterate through
1872 * the array, so you send it the number of backups you've already
1873 * tried and the last backup index you used.
1875 * this returns -1 when it has tried all the backups
1877 static noinline int next_root_backup(struct btrfs_fs_info *info,
1878 struct btrfs_super_block *super,
1879 int *num_backups_tried, int *backup_index)
1881 struct btrfs_root_backup *root_backup;
1882 int newest = *backup_index;
1884 if (*num_backups_tried == 0) {
1885 u64 gen = btrfs_super_generation(super);
1887 newest = find_newest_super_backup(info, gen);
1891 *backup_index = newest;
1892 *num_backups_tried = 1;
1893 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1894 /* we've tried all the backups, all done */
1897 /* jump to the next oldest backup */
1898 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1899 BTRFS_NUM_BACKUP_ROOTS;
1900 *backup_index = newest;
1901 *num_backups_tried += 1;
1903 root_backup = super->super_roots + newest;
1905 btrfs_set_super_generation(super,
1906 btrfs_backup_tree_root_gen(root_backup));
1907 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1908 btrfs_set_super_root_level(super,
1909 btrfs_backup_tree_root_level(root_backup));
1910 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1913 * fixme: the total bytes and num_devices need to match or we should
1916 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1917 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1921 /* helper to cleanup tree roots */
1922 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1924 free_extent_buffer(info->tree_root->node);
1925 free_extent_buffer(info->tree_root->commit_root);
1926 free_extent_buffer(info->dev_root->node);
1927 free_extent_buffer(info->dev_root->commit_root);
1928 free_extent_buffer(info->extent_root->node);
1929 free_extent_buffer(info->extent_root->commit_root);
1930 free_extent_buffer(info->csum_root->node);
1931 free_extent_buffer(info->csum_root->commit_root);
1932 if (info->quota_root) {
1933 free_extent_buffer(info->quota_root->node);
1934 free_extent_buffer(info->quota_root->commit_root);
1937 info->tree_root->node = NULL;
1938 info->tree_root->commit_root = NULL;
1939 info->dev_root->node = NULL;
1940 info->dev_root->commit_root = NULL;
1941 info->extent_root->node = NULL;
1942 info->extent_root->commit_root = NULL;
1943 info->csum_root->node = NULL;
1944 info->csum_root->commit_root = NULL;
1945 if (info->quota_root) {
1946 info->quota_root->node = NULL;
1947 info->quota_root->commit_root = NULL;
1951 free_extent_buffer(info->chunk_root->node);
1952 free_extent_buffer(info->chunk_root->commit_root);
1953 info->chunk_root->node = NULL;
1954 info->chunk_root->commit_root = NULL;
1959 int open_ctree(struct super_block *sb,
1960 struct btrfs_fs_devices *fs_devices,
1970 struct btrfs_key location;
1971 struct buffer_head *bh;
1972 struct btrfs_super_block *disk_super;
1973 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1974 struct btrfs_root *tree_root;
1975 struct btrfs_root *extent_root;
1976 struct btrfs_root *csum_root;
1977 struct btrfs_root *chunk_root;
1978 struct btrfs_root *dev_root;
1979 struct btrfs_root *quota_root;
1980 struct btrfs_root *log_tree_root;
1983 int num_backups_tried = 0;
1984 int backup_index = 0;
1986 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1987 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1988 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1989 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1990 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1991 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
1993 if (!tree_root || !extent_root || !csum_root ||
1994 !chunk_root || !dev_root || !quota_root) {
1999 ret = init_srcu_struct(&fs_info->subvol_srcu);
2005 ret = setup_bdi(fs_info, &fs_info->bdi);
2011 fs_info->btree_inode = new_inode(sb);
2012 if (!fs_info->btree_inode) {
2017 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2019 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2020 INIT_LIST_HEAD(&fs_info->trans_list);
2021 INIT_LIST_HEAD(&fs_info->dead_roots);
2022 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2023 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2024 INIT_LIST_HEAD(&fs_info->ordered_operations);
2025 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2026 spin_lock_init(&fs_info->delalloc_lock);
2027 spin_lock_init(&fs_info->trans_lock);
2028 spin_lock_init(&fs_info->fs_roots_radix_lock);
2029 spin_lock_init(&fs_info->delayed_iput_lock);
2030 spin_lock_init(&fs_info->defrag_inodes_lock);
2031 spin_lock_init(&fs_info->free_chunk_lock);
2032 spin_lock_init(&fs_info->tree_mod_seq_lock);
2033 rwlock_init(&fs_info->tree_mod_log_lock);
2034 mutex_init(&fs_info->reloc_mutex);
2036 init_completion(&fs_info->kobj_unregister);
2037 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2038 INIT_LIST_HEAD(&fs_info->space_info);
2039 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2040 btrfs_mapping_init(&fs_info->mapping_tree);
2041 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2042 BTRFS_BLOCK_RSV_GLOBAL);
2043 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2044 BTRFS_BLOCK_RSV_DELALLOC);
2045 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2046 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2047 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2048 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2049 BTRFS_BLOCK_RSV_DELOPS);
2050 atomic_set(&fs_info->nr_async_submits, 0);
2051 atomic_set(&fs_info->async_delalloc_pages, 0);
2052 atomic_set(&fs_info->async_submit_draining, 0);
2053 atomic_set(&fs_info->nr_async_bios, 0);
2054 atomic_set(&fs_info->defrag_running, 0);
2055 atomic_set(&fs_info->tree_mod_seq, 0);
2057 fs_info->max_inline = 8192 * 1024;
2058 fs_info->metadata_ratio = 0;
2059 fs_info->defrag_inodes = RB_ROOT;
2060 fs_info->trans_no_join = 0;
2061 fs_info->free_chunk_space = 0;
2062 fs_info->tree_mod_log = RB_ROOT;
2064 /* readahead state */
2065 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2066 spin_lock_init(&fs_info->reada_lock);
2068 fs_info->thread_pool_size = min_t(unsigned long,
2069 num_online_cpus() + 2, 8);
2071 INIT_LIST_HEAD(&fs_info->ordered_extents);
2072 spin_lock_init(&fs_info->ordered_extent_lock);
2073 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2075 if (!fs_info->delayed_root) {
2079 btrfs_init_delayed_root(fs_info->delayed_root);
2081 mutex_init(&fs_info->scrub_lock);
2082 atomic_set(&fs_info->scrubs_running, 0);
2083 atomic_set(&fs_info->scrub_pause_req, 0);
2084 atomic_set(&fs_info->scrubs_paused, 0);
2085 atomic_set(&fs_info->scrub_cancel_req, 0);
2086 init_waitqueue_head(&fs_info->scrub_pause_wait);
2087 init_rwsem(&fs_info->scrub_super_lock);
2088 fs_info->scrub_workers_refcnt = 0;
2089 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2090 fs_info->check_integrity_print_mask = 0;
2093 spin_lock_init(&fs_info->balance_lock);
2094 mutex_init(&fs_info->balance_mutex);
2095 atomic_set(&fs_info->balance_running, 0);
2096 atomic_set(&fs_info->balance_pause_req, 0);
2097 atomic_set(&fs_info->balance_cancel_req, 0);
2098 fs_info->balance_ctl = NULL;
2099 init_waitqueue_head(&fs_info->balance_wait_q);
2101 sb->s_blocksize = 4096;
2102 sb->s_blocksize_bits = blksize_bits(4096);
2103 sb->s_bdi = &fs_info->bdi;
2105 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2106 set_nlink(fs_info->btree_inode, 1);
2108 * we set the i_size on the btree inode to the max possible int.
2109 * the real end of the address space is determined by all of
2110 * the devices in the system
2112 fs_info->btree_inode->i_size = OFFSET_MAX;
2113 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2114 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2116 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2117 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2118 fs_info->btree_inode->i_mapping);
2119 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2120 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2122 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2124 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2125 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2126 sizeof(struct btrfs_key));
2127 set_bit(BTRFS_INODE_DUMMY,
2128 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2129 insert_inode_hash(fs_info->btree_inode);
2131 spin_lock_init(&fs_info->block_group_cache_lock);
2132 fs_info->block_group_cache_tree = RB_ROOT;
2133 fs_info->first_logical_byte = (u64)-1;
2135 extent_io_tree_init(&fs_info->freed_extents[0],
2136 fs_info->btree_inode->i_mapping);
2137 extent_io_tree_init(&fs_info->freed_extents[1],
2138 fs_info->btree_inode->i_mapping);
2139 fs_info->pinned_extents = &fs_info->freed_extents[0];
2140 fs_info->do_barriers = 1;
2143 mutex_init(&fs_info->ordered_operations_mutex);
2144 mutex_init(&fs_info->tree_log_mutex);
2145 mutex_init(&fs_info->chunk_mutex);
2146 mutex_init(&fs_info->transaction_kthread_mutex);
2147 mutex_init(&fs_info->cleaner_mutex);
2148 mutex_init(&fs_info->volume_mutex);
2149 init_rwsem(&fs_info->extent_commit_sem);
2150 init_rwsem(&fs_info->cleanup_work_sem);
2151 init_rwsem(&fs_info->subvol_sem);
2152 fs_info->dev_replace.lock_owner = 0;
2153 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2154 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2155 mutex_init(&fs_info->dev_replace.lock_management_lock);
2156 mutex_init(&fs_info->dev_replace.lock);
2158 spin_lock_init(&fs_info->qgroup_lock);
2159 fs_info->qgroup_tree = RB_ROOT;
2160 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2161 fs_info->qgroup_seq = 1;
2162 fs_info->quota_enabled = 0;
2163 fs_info->pending_quota_state = 0;
2165 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2166 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2168 init_waitqueue_head(&fs_info->transaction_throttle);
2169 init_waitqueue_head(&fs_info->transaction_wait);
2170 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2171 init_waitqueue_head(&fs_info->async_submit_wait);
2173 __setup_root(4096, 4096, 4096, 4096, tree_root,
2174 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2176 invalidate_bdev(fs_devices->latest_bdev);
2177 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2183 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2184 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2185 sizeof(*fs_info->super_for_commit));
2188 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2190 disk_super = fs_info->super_copy;
2191 if (!btrfs_super_root(disk_super))
2194 /* check FS state, whether FS is broken. */
2195 fs_info->fs_state |= btrfs_super_flags(disk_super);
2197 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2199 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2205 * run through our array of backup supers and setup
2206 * our ring pointer to the oldest one
2208 generation = btrfs_super_generation(disk_super);
2209 find_oldest_super_backup(fs_info, generation);
2212 * In the long term, we'll store the compression type in the super
2213 * block, and it'll be used for per file compression control.
2215 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2217 ret = btrfs_parse_options(tree_root, options);
2223 features = btrfs_super_incompat_flags(disk_super) &
2224 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2226 printk(KERN_ERR "BTRFS: couldn't mount because of "
2227 "unsupported optional features (%Lx).\n",
2228 (unsigned long long)features);
2233 if (btrfs_super_leafsize(disk_super) !=
2234 btrfs_super_nodesize(disk_super)) {
2235 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2236 "blocksizes don't match. node %d leaf %d\n",
2237 btrfs_super_nodesize(disk_super),
2238 btrfs_super_leafsize(disk_super));
2242 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2243 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2244 "blocksize (%d) was too large\n",
2245 btrfs_super_leafsize(disk_super));
2250 features = btrfs_super_incompat_flags(disk_super);
2251 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2252 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2253 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2256 * flag our filesystem as having big metadata blocks if
2257 * they are bigger than the page size
2259 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2260 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2261 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2262 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2265 nodesize = btrfs_super_nodesize(disk_super);
2266 leafsize = btrfs_super_leafsize(disk_super);
2267 sectorsize = btrfs_super_sectorsize(disk_super);
2268 stripesize = btrfs_super_stripesize(disk_super);
2271 * mixed block groups end up with duplicate but slightly offset
2272 * extent buffers for the same range. It leads to corruptions
2274 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2275 (sectorsize != leafsize)) {
2276 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2277 "are not allowed for mixed block groups on %s\n",
2282 btrfs_set_super_incompat_flags(disk_super, features);
2284 features = btrfs_super_compat_ro_flags(disk_super) &
2285 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2286 if (!(sb->s_flags & MS_RDONLY) && features) {
2287 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2288 "unsupported option features (%Lx).\n",
2289 (unsigned long long)features);
2294 btrfs_init_workers(&fs_info->generic_worker,
2295 "genwork", 1, NULL);
2297 btrfs_init_workers(&fs_info->workers, "worker",
2298 fs_info->thread_pool_size,
2299 &fs_info->generic_worker);
2301 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2302 fs_info->thread_pool_size,
2303 &fs_info->generic_worker);
2305 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2306 fs_info->thread_pool_size,
2307 &fs_info->generic_worker);
2309 btrfs_init_workers(&fs_info->submit_workers, "submit",
2310 min_t(u64, fs_devices->num_devices,
2311 fs_info->thread_pool_size),
2312 &fs_info->generic_worker);
2314 btrfs_init_workers(&fs_info->caching_workers, "cache",
2315 2, &fs_info->generic_worker);
2317 /* a higher idle thresh on the submit workers makes it much more
2318 * likely that bios will be send down in a sane order to the
2321 fs_info->submit_workers.idle_thresh = 64;
2323 fs_info->workers.idle_thresh = 16;
2324 fs_info->workers.ordered = 1;
2326 fs_info->delalloc_workers.idle_thresh = 2;
2327 fs_info->delalloc_workers.ordered = 1;
2329 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2330 &fs_info->generic_worker);
2331 btrfs_init_workers(&fs_info->endio_workers, "endio",
2332 fs_info->thread_pool_size,
2333 &fs_info->generic_worker);
2334 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2335 fs_info->thread_pool_size,
2336 &fs_info->generic_worker);
2337 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2338 "endio-meta-write", fs_info->thread_pool_size,
2339 &fs_info->generic_worker);
2340 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2341 fs_info->thread_pool_size,
2342 &fs_info->generic_worker);
2343 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2344 1, &fs_info->generic_worker);
2345 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2346 fs_info->thread_pool_size,
2347 &fs_info->generic_worker);
2348 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2349 fs_info->thread_pool_size,
2350 &fs_info->generic_worker);
2353 * endios are largely parallel and should have a very
2356 fs_info->endio_workers.idle_thresh = 4;
2357 fs_info->endio_meta_workers.idle_thresh = 4;
2359 fs_info->endio_write_workers.idle_thresh = 2;
2360 fs_info->endio_meta_write_workers.idle_thresh = 2;
2361 fs_info->readahead_workers.idle_thresh = 2;
2364 * btrfs_start_workers can really only fail because of ENOMEM so just
2365 * return -ENOMEM if any of these fail.
2367 ret = btrfs_start_workers(&fs_info->workers);
2368 ret |= btrfs_start_workers(&fs_info->generic_worker);
2369 ret |= btrfs_start_workers(&fs_info->submit_workers);
2370 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2371 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2372 ret |= btrfs_start_workers(&fs_info->endio_workers);
2373 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2374 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2375 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2376 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2377 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2378 ret |= btrfs_start_workers(&fs_info->caching_workers);
2379 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2380 ret |= btrfs_start_workers(&fs_info->flush_workers);
2383 goto fail_sb_buffer;
2386 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2387 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2388 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2390 tree_root->nodesize = nodesize;
2391 tree_root->leafsize = leafsize;
2392 tree_root->sectorsize = sectorsize;
2393 tree_root->stripesize = stripesize;
2395 sb->s_blocksize = sectorsize;
2396 sb->s_blocksize_bits = blksize_bits(sectorsize);
2398 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2399 sizeof(disk_super->magic))) {
2400 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2401 goto fail_sb_buffer;
2404 if (sectorsize != PAGE_SIZE) {
2405 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2406 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2407 goto fail_sb_buffer;
2410 mutex_lock(&fs_info->chunk_mutex);
2411 ret = btrfs_read_sys_array(tree_root);
2412 mutex_unlock(&fs_info->chunk_mutex);
2414 printk(KERN_WARNING "btrfs: failed to read the system "
2415 "array on %s\n", sb->s_id);
2416 goto fail_sb_buffer;
2419 blocksize = btrfs_level_size(tree_root,
2420 btrfs_super_chunk_root_level(disk_super));
2421 generation = btrfs_super_chunk_root_generation(disk_super);
2423 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2424 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2426 chunk_root->node = read_tree_block(chunk_root,
2427 btrfs_super_chunk_root(disk_super),
2428 blocksize, generation);
2429 BUG_ON(!chunk_root->node); /* -ENOMEM */
2430 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2431 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2433 goto fail_tree_roots;
2435 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2436 chunk_root->commit_root = btrfs_root_node(chunk_root);
2438 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2439 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2442 ret = btrfs_read_chunk_tree(chunk_root);
2444 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2446 goto fail_tree_roots;
2450 * keep the device that is marked to be the target device for the
2451 * dev_replace procedure
2453 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2455 if (!fs_devices->latest_bdev) {
2456 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2458 goto fail_tree_roots;
2462 blocksize = btrfs_level_size(tree_root,
2463 btrfs_super_root_level(disk_super));
2464 generation = btrfs_super_generation(disk_super);
2466 tree_root->node = read_tree_block(tree_root,
2467 btrfs_super_root(disk_super),
2468 blocksize, generation);
2469 if (!tree_root->node ||
2470 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2471 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2474 goto recovery_tree_root;
2477 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2478 tree_root->commit_root = btrfs_root_node(tree_root);
2480 ret = find_and_setup_root(tree_root, fs_info,
2481 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2483 goto recovery_tree_root;
2484 extent_root->track_dirty = 1;
2486 ret = find_and_setup_root(tree_root, fs_info,
2487 BTRFS_DEV_TREE_OBJECTID, dev_root);
2489 goto recovery_tree_root;
2490 dev_root->track_dirty = 1;
2492 ret = find_and_setup_root(tree_root, fs_info,
2493 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2495 goto recovery_tree_root;
2496 csum_root->track_dirty = 1;
2498 ret = find_and_setup_root(tree_root, fs_info,
2499 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2502 quota_root = fs_info->quota_root = NULL;
2504 quota_root->track_dirty = 1;
2505 fs_info->quota_enabled = 1;
2506 fs_info->pending_quota_state = 1;
2509 fs_info->generation = generation;
2510 fs_info->last_trans_committed = generation;
2512 ret = btrfs_recover_balance(fs_info);
2514 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2515 goto fail_block_groups;
2518 ret = btrfs_init_dev_stats(fs_info);
2520 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2522 goto fail_block_groups;
2525 ret = btrfs_init_dev_replace(fs_info);
2527 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2528 goto fail_block_groups;
2531 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2533 ret = btrfs_init_space_info(fs_info);
2535 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2536 goto fail_block_groups;
2539 ret = btrfs_read_block_groups(extent_root);
2541 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2542 goto fail_block_groups;
2544 fs_info->num_tolerated_disk_barrier_failures =
2545 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2546 if (fs_info->fs_devices->missing_devices >
2547 fs_info->num_tolerated_disk_barrier_failures &&
2548 !(sb->s_flags & MS_RDONLY)) {
2550 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2551 goto fail_block_groups;
2554 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2556 if (IS_ERR(fs_info->cleaner_kthread))
2557 goto fail_block_groups;
2559 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2561 "btrfs-transaction");
2562 if (IS_ERR(fs_info->transaction_kthread))
2565 if (!btrfs_test_opt(tree_root, SSD) &&
2566 !btrfs_test_opt(tree_root, NOSSD) &&
2567 !fs_info->fs_devices->rotating) {
2568 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2570 btrfs_set_opt(fs_info->mount_opt, SSD);
2573 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2574 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2575 ret = btrfsic_mount(tree_root, fs_devices,
2576 btrfs_test_opt(tree_root,
2577 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2579 fs_info->check_integrity_print_mask);
2581 printk(KERN_WARNING "btrfs: failed to initialize"
2582 " integrity check module %s\n", sb->s_id);
2585 ret = btrfs_read_qgroup_config(fs_info);
2587 goto fail_trans_kthread;
2589 /* do not make disk changes in broken FS */
2590 if (btrfs_super_log_root(disk_super) != 0) {
2591 u64 bytenr = btrfs_super_log_root(disk_super);
2593 if (fs_devices->rw_devices == 0) {
2594 printk(KERN_WARNING "Btrfs log replay required "
2600 btrfs_level_size(tree_root,
2601 btrfs_super_log_root_level(disk_super));
2603 log_tree_root = btrfs_alloc_root(fs_info);
2604 if (!log_tree_root) {
2609 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2610 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2612 log_tree_root->node = read_tree_block(tree_root, bytenr,
2615 /* returns with log_tree_root freed on success */
2616 ret = btrfs_recover_log_trees(log_tree_root);
2618 btrfs_error(tree_root->fs_info, ret,
2619 "Failed to recover log tree");
2620 free_extent_buffer(log_tree_root->node);
2621 kfree(log_tree_root);
2622 goto fail_trans_kthread;
2625 if (sb->s_flags & MS_RDONLY) {
2626 ret = btrfs_commit_super(tree_root);
2628 goto fail_trans_kthread;
2632 ret = btrfs_find_orphan_roots(tree_root);
2634 goto fail_trans_kthread;
2636 if (!(sb->s_flags & MS_RDONLY)) {
2637 ret = btrfs_cleanup_fs_roots(fs_info);
2639 goto fail_trans_kthread;
2641 ret = btrfs_recover_relocation(tree_root);
2644 "btrfs: failed to recover relocation\n");
2650 location.objectid = BTRFS_FS_TREE_OBJECTID;
2651 location.type = BTRFS_ROOT_ITEM_KEY;
2652 location.offset = (u64)-1;
2654 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2655 if (!fs_info->fs_root)
2657 if (IS_ERR(fs_info->fs_root)) {
2658 err = PTR_ERR(fs_info->fs_root);
2662 if (sb->s_flags & MS_RDONLY)
2665 down_read(&fs_info->cleanup_work_sem);
2666 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2667 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2668 up_read(&fs_info->cleanup_work_sem);
2669 close_ctree(tree_root);
2672 up_read(&fs_info->cleanup_work_sem);
2674 ret = btrfs_resume_balance_async(fs_info);
2676 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2677 close_ctree(tree_root);
2681 ret = btrfs_resume_dev_replace_async(fs_info);
2683 pr_warn("btrfs: failed to resume dev_replace\n");
2684 close_ctree(tree_root);
2691 btrfs_free_qgroup_config(fs_info);
2693 kthread_stop(fs_info->transaction_kthread);
2695 kthread_stop(fs_info->cleaner_kthread);
2698 * make sure we're done with the btree inode before we stop our
2701 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2702 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2705 btrfs_free_block_groups(fs_info);
2708 free_root_pointers(fs_info, 1);
2711 btrfs_stop_workers(&fs_info->generic_worker);
2712 btrfs_stop_workers(&fs_info->readahead_workers);
2713 btrfs_stop_workers(&fs_info->fixup_workers);
2714 btrfs_stop_workers(&fs_info->delalloc_workers);
2715 btrfs_stop_workers(&fs_info->workers);
2716 btrfs_stop_workers(&fs_info->endio_workers);
2717 btrfs_stop_workers(&fs_info->endio_meta_workers);
2718 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2719 btrfs_stop_workers(&fs_info->endio_write_workers);
2720 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2721 btrfs_stop_workers(&fs_info->submit_workers);
2722 btrfs_stop_workers(&fs_info->delayed_workers);
2723 btrfs_stop_workers(&fs_info->caching_workers);
2724 btrfs_stop_workers(&fs_info->flush_workers);
2727 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2729 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2730 iput(fs_info->btree_inode);
2732 bdi_destroy(&fs_info->bdi);
2734 cleanup_srcu_struct(&fs_info->subvol_srcu);
2736 btrfs_close_devices(fs_info->fs_devices);
2740 if (!btrfs_test_opt(tree_root, RECOVERY))
2741 goto fail_tree_roots;
2743 free_root_pointers(fs_info, 0);
2745 /* don't use the log in recovery mode, it won't be valid */
2746 btrfs_set_super_log_root(disk_super, 0);
2748 /* we can't trust the free space cache either */
2749 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2751 ret = next_root_backup(fs_info, fs_info->super_copy,
2752 &num_backups_tried, &backup_index);
2754 goto fail_block_groups;
2755 goto retry_root_backup;
2758 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2761 set_buffer_uptodate(bh);
2763 struct btrfs_device *device = (struct btrfs_device *)
2766 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2767 "I/O error on %s\n",
2768 rcu_str_deref(device->name));
2769 /* note, we dont' set_buffer_write_io_error because we have
2770 * our own ways of dealing with the IO errors
2772 clear_buffer_uptodate(bh);
2773 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2779 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2781 struct buffer_head *bh;
2782 struct buffer_head *latest = NULL;
2783 struct btrfs_super_block *super;
2788 /* we would like to check all the supers, but that would make
2789 * a btrfs mount succeed after a mkfs from a different FS.
2790 * So, we need to add a special mount option to scan for
2791 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2793 for (i = 0; i < 1; i++) {
2794 bytenr = btrfs_sb_offset(i);
2795 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2797 bh = __bread(bdev, bytenr / 4096, 4096);
2801 super = (struct btrfs_super_block *)bh->b_data;
2802 if (btrfs_super_bytenr(super) != bytenr ||
2803 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2804 sizeof(super->magic))) {
2809 if (!latest || btrfs_super_generation(super) > transid) {
2812 transid = btrfs_super_generation(super);
2821 * this should be called twice, once with wait == 0 and
2822 * once with wait == 1. When wait == 0 is done, all the buffer heads
2823 * we write are pinned.
2825 * They are released when wait == 1 is done.
2826 * max_mirrors must be the same for both runs, and it indicates how
2827 * many supers on this one device should be written.
2829 * max_mirrors == 0 means to write them all.
2831 static int write_dev_supers(struct btrfs_device *device,
2832 struct btrfs_super_block *sb,
2833 int do_barriers, int wait, int max_mirrors)
2835 struct buffer_head *bh;
2842 if (max_mirrors == 0)
2843 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2845 for (i = 0; i < max_mirrors; i++) {
2846 bytenr = btrfs_sb_offset(i);
2847 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2851 bh = __find_get_block(device->bdev, bytenr / 4096,
2852 BTRFS_SUPER_INFO_SIZE);
2855 if (!buffer_uptodate(bh))
2858 /* drop our reference */
2861 /* drop the reference from the wait == 0 run */
2865 btrfs_set_super_bytenr(sb, bytenr);
2868 crc = btrfs_csum_data(NULL, (char *)sb +
2869 BTRFS_CSUM_SIZE, crc,
2870 BTRFS_SUPER_INFO_SIZE -
2872 btrfs_csum_final(crc, sb->csum);
2875 * one reference for us, and we leave it for the
2878 bh = __getblk(device->bdev, bytenr / 4096,
2879 BTRFS_SUPER_INFO_SIZE);
2880 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2882 /* one reference for submit_bh */
2885 set_buffer_uptodate(bh);
2887 bh->b_end_io = btrfs_end_buffer_write_sync;
2888 bh->b_private = device;
2892 * we fua the first super. The others we allow
2895 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2899 return errors < i ? 0 : -1;
2903 * endio for the write_dev_flush, this will wake anyone waiting
2904 * for the barrier when it is done
2906 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2909 if (err == -EOPNOTSUPP)
2910 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2911 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2913 if (bio->bi_private)
2914 complete(bio->bi_private);
2919 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2920 * sent down. With wait == 1, it waits for the previous flush.
2922 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2925 static int write_dev_flush(struct btrfs_device *device, int wait)
2930 if (device->nobarriers)
2934 bio = device->flush_bio;
2938 wait_for_completion(&device->flush_wait);
2940 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2941 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2942 rcu_str_deref(device->name));
2943 device->nobarriers = 1;
2944 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
2946 btrfs_dev_stat_inc_and_print(device,
2947 BTRFS_DEV_STAT_FLUSH_ERRS);
2950 /* drop the reference from the wait == 0 run */
2952 device->flush_bio = NULL;
2958 * one reference for us, and we leave it for the
2961 device->flush_bio = NULL;
2962 bio = bio_alloc(GFP_NOFS, 0);
2966 bio->bi_end_io = btrfs_end_empty_barrier;
2967 bio->bi_bdev = device->bdev;
2968 init_completion(&device->flush_wait);
2969 bio->bi_private = &device->flush_wait;
2970 device->flush_bio = bio;
2973 btrfsic_submit_bio(WRITE_FLUSH, bio);
2979 * send an empty flush down to each device in parallel,
2980 * then wait for them
2982 static int barrier_all_devices(struct btrfs_fs_info *info)
2984 struct list_head *head;
2985 struct btrfs_device *dev;
2986 int errors_send = 0;
2987 int errors_wait = 0;
2990 /* send down all the barriers */
2991 head = &info->fs_devices->devices;
2992 list_for_each_entry_rcu(dev, head, dev_list) {
2997 if (!dev->in_fs_metadata || !dev->writeable)
3000 ret = write_dev_flush(dev, 0);
3005 /* wait for all the barriers */
3006 list_for_each_entry_rcu(dev, head, dev_list) {
3011 if (!dev->in_fs_metadata || !dev->writeable)
3014 ret = write_dev_flush(dev, 1);
3018 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3019 errors_wait > info->num_tolerated_disk_barrier_failures)
3024 int btrfs_calc_num_tolerated_disk_barrier_failures(
3025 struct btrfs_fs_info *fs_info)
3027 struct btrfs_ioctl_space_info space;
3028 struct btrfs_space_info *sinfo;
3029 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3030 BTRFS_BLOCK_GROUP_SYSTEM,
3031 BTRFS_BLOCK_GROUP_METADATA,
3032 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3036 int num_tolerated_disk_barrier_failures =
3037 (int)fs_info->fs_devices->num_devices;
3039 for (i = 0; i < num_types; i++) {
3040 struct btrfs_space_info *tmp;
3044 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3045 if (tmp->flags == types[i]) {
3055 down_read(&sinfo->groups_sem);
3056 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3057 if (!list_empty(&sinfo->block_groups[c])) {
3060 btrfs_get_block_group_info(
3061 &sinfo->block_groups[c], &space);
3062 if (space.total_bytes == 0 ||
3063 space.used_bytes == 0)
3065 flags = space.flags;
3068 * 0: if dup, single or RAID0 is configured for
3069 * any of metadata, system or data, else
3070 * 1: if RAID5 is configured, or if RAID1 or
3071 * RAID10 is configured and only two mirrors
3073 * 2: if RAID6 is configured, else
3074 * num_mirrors - 1: if RAID1 or RAID10 is
3075 * configured and more than
3076 * 2 mirrors are used.
3078 if (num_tolerated_disk_barrier_failures > 0 &&
3079 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3080 BTRFS_BLOCK_GROUP_RAID0)) ||
3081 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3083 num_tolerated_disk_barrier_failures = 0;
3084 else if (num_tolerated_disk_barrier_failures > 1
3086 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3087 BTRFS_BLOCK_GROUP_RAID10)))
3088 num_tolerated_disk_barrier_failures = 1;
3091 up_read(&sinfo->groups_sem);
3094 return num_tolerated_disk_barrier_failures;
3097 int write_all_supers(struct btrfs_root *root, int max_mirrors)
3099 struct list_head *head;
3100 struct btrfs_device *dev;
3101 struct btrfs_super_block *sb;
3102 struct btrfs_dev_item *dev_item;
3106 int total_errors = 0;
3109 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3110 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3111 backup_super_roots(root->fs_info);
3113 sb = root->fs_info->super_for_commit;
3114 dev_item = &sb->dev_item;
3116 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3117 head = &root->fs_info->fs_devices->devices;
3120 ret = barrier_all_devices(root->fs_info);
3123 &root->fs_info->fs_devices->device_list_mutex);
3124 btrfs_error(root->fs_info, ret,
3125 "errors while submitting device barriers.");
3130 list_for_each_entry_rcu(dev, head, dev_list) {
3135 if (!dev->in_fs_metadata || !dev->writeable)
3138 btrfs_set_stack_device_generation(dev_item, 0);
3139 btrfs_set_stack_device_type(dev_item, dev->type);
3140 btrfs_set_stack_device_id(dev_item, dev->devid);
3141 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3142 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3143 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3144 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3145 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3146 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3147 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3149 flags = btrfs_super_flags(sb);
3150 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3152 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3156 if (total_errors > max_errors) {
3157 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3160 /* This shouldn't happen. FUA is masked off if unsupported */
3165 list_for_each_entry_rcu(dev, head, dev_list) {
3168 if (!dev->in_fs_metadata || !dev->writeable)
3171 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3175 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3176 if (total_errors > max_errors) {
3177 btrfs_error(root->fs_info, -EIO,
3178 "%d errors while writing supers", total_errors);
3184 int write_ctree_super(struct btrfs_trans_handle *trans,
3185 struct btrfs_root *root, int max_mirrors)
3189 ret = write_all_supers(root, max_mirrors);
3193 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3195 spin_lock(&fs_info->fs_roots_radix_lock);
3196 radix_tree_delete(&fs_info->fs_roots_radix,
3197 (unsigned long)root->root_key.objectid);
3198 spin_unlock(&fs_info->fs_roots_radix_lock);
3200 if (btrfs_root_refs(&root->root_item) == 0)
3201 synchronize_srcu(&fs_info->subvol_srcu);
3203 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3204 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3208 static void free_fs_root(struct btrfs_root *root)
3210 iput(root->cache_inode);
3211 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3213 free_anon_bdev(root->anon_dev);
3214 free_extent_buffer(root->node);
3215 free_extent_buffer(root->commit_root);
3216 kfree(root->free_ino_ctl);
3217 kfree(root->free_ino_pinned);
3222 static void del_fs_roots(struct btrfs_fs_info *fs_info)
3225 struct btrfs_root *gang[8];
3228 while (!list_empty(&fs_info->dead_roots)) {
3229 gang[0] = list_entry(fs_info->dead_roots.next,
3230 struct btrfs_root, root_list);
3231 list_del(&gang[0]->root_list);
3233 if (gang[0]->in_radix) {
3234 btrfs_free_fs_root(fs_info, gang[0]);
3236 free_extent_buffer(gang[0]->node);
3237 free_extent_buffer(gang[0]->commit_root);
3243 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3248 for (i = 0; i < ret; i++)
3249 btrfs_free_fs_root(fs_info, gang[i]);
3253 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3255 u64 root_objectid = 0;
3256 struct btrfs_root *gang[8];
3261 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3262 (void **)gang, root_objectid,
3267 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3268 for (i = 0; i < ret; i++) {
3271 root_objectid = gang[i]->root_key.objectid;
3272 err = btrfs_orphan_cleanup(gang[i]);
3281 int btrfs_commit_super(struct btrfs_root *root)
3283 struct btrfs_trans_handle *trans;
3286 mutex_lock(&root->fs_info->cleaner_mutex);
3287 btrfs_run_delayed_iputs(root);
3288 btrfs_clean_old_snapshots(root);
3289 mutex_unlock(&root->fs_info->cleaner_mutex);
3291 /* wait until ongoing cleanup work done */
3292 down_write(&root->fs_info->cleanup_work_sem);
3293 up_write(&root->fs_info->cleanup_work_sem);
3295 trans = btrfs_join_transaction(root);
3297 return PTR_ERR(trans);
3298 ret = btrfs_commit_transaction(trans, root);
3301 /* run commit again to drop the original snapshot */
3302 trans = btrfs_join_transaction(root);
3304 return PTR_ERR(trans);
3305 ret = btrfs_commit_transaction(trans, root);
3308 ret = btrfs_write_and_wait_transaction(NULL, root);
3310 btrfs_error(root->fs_info, ret,
3311 "Failed to sync btree inode to disk.");
3315 ret = write_ctree_super(NULL, root, 0);
3319 int close_ctree(struct btrfs_root *root)
3321 struct btrfs_fs_info *fs_info = root->fs_info;
3324 fs_info->closing = 1;
3327 /* pause restriper - we want to resume on mount */
3328 btrfs_pause_balance(fs_info);
3330 btrfs_dev_replace_suspend_for_unmount(fs_info);
3332 btrfs_scrub_cancel(fs_info);
3334 /* wait for any defraggers to finish */
3335 wait_event(fs_info->transaction_wait,
3336 (atomic_read(&fs_info->defrag_running) == 0));
3338 /* clear out the rbtree of defraggable inodes */
3339 btrfs_cleanup_defrag_inodes(fs_info);
3341 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3342 ret = btrfs_commit_super(root);
3344 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3347 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3348 btrfs_error_commit_super(root);
3350 btrfs_put_block_group_cache(fs_info);
3352 kthread_stop(fs_info->transaction_kthread);
3353 kthread_stop(fs_info->cleaner_kthread);
3355 fs_info->closing = 2;
3358 btrfs_free_qgroup_config(root->fs_info);
3360 if (fs_info->delalloc_bytes) {
3361 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3362 (unsigned long long)fs_info->delalloc_bytes);
3365 free_extent_buffer(fs_info->extent_root->node);
3366 free_extent_buffer(fs_info->extent_root->commit_root);
3367 free_extent_buffer(fs_info->tree_root->node);
3368 free_extent_buffer(fs_info->tree_root->commit_root);
3369 free_extent_buffer(fs_info->chunk_root->node);
3370 free_extent_buffer(fs_info->chunk_root->commit_root);
3371 free_extent_buffer(fs_info->dev_root->node);
3372 free_extent_buffer(fs_info->dev_root->commit_root);
3373 free_extent_buffer(fs_info->csum_root->node);
3374 free_extent_buffer(fs_info->csum_root->commit_root);
3375 if (fs_info->quota_root) {
3376 free_extent_buffer(fs_info->quota_root->node);
3377 free_extent_buffer(fs_info->quota_root->commit_root);
3380 btrfs_free_block_groups(fs_info);
3382 del_fs_roots(fs_info);
3384 iput(fs_info->btree_inode);
3386 btrfs_stop_workers(&fs_info->generic_worker);
3387 btrfs_stop_workers(&fs_info->fixup_workers);
3388 btrfs_stop_workers(&fs_info->delalloc_workers);
3389 btrfs_stop_workers(&fs_info->workers);
3390 btrfs_stop_workers(&fs_info->endio_workers);
3391 btrfs_stop_workers(&fs_info->endio_meta_workers);
3392 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3393 btrfs_stop_workers(&fs_info->endio_write_workers);
3394 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3395 btrfs_stop_workers(&fs_info->submit_workers);
3396 btrfs_stop_workers(&fs_info->delayed_workers);
3397 btrfs_stop_workers(&fs_info->caching_workers);
3398 btrfs_stop_workers(&fs_info->readahead_workers);
3399 btrfs_stop_workers(&fs_info->flush_workers);
3401 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3402 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3403 btrfsic_unmount(root, fs_info->fs_devices);
3406 btrfs_close_devices(fs_info->fs_devices);
3407 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3409 bdi_destroy(&fs_info->bdi);
3410 cleanup_srcu_struct(&fs_info->subvol_srcu);
3415 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3419 struct inode *btree_inode = buf->pages[0]->mapping->host;
3421 ret = extent_buffer_uptodate(buf);
3425 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3426 parent_transid, atomic);
3432 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3434 return set_extent_buffer_uptodate(buf);
3437 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3439 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3440 u64 transid = btrfs_header_generation(buf);
3443 btrfs_assert_tree_locked(buf);
3444 if (transid != root->fs_info->generation)
3445 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3446 "found %llu running %llu\n",
3447 (unsigned long long)buf->start,
3448 (unsigned long long)transid,
3449 (unsigned long long)root->fs_info->generation);
3450 was_dirty = set_extent_buffer_dirty(buf);
3452 spin_lock(&root->fs_info->delalloc_lock);
3453 root->fs_info->dirty_metadata_bytes += buf->len;
3454 spin_unlock(&root->fs_info->delalloc_lock);
3458 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3462 * looks as though older kernels can get into trouble with
3463 * this code, they end up stuck in balance_dirty_pages forever
3466 unsigned long thresh = 32 * 1024 * 1024;
3468 if (current->flags & PF_MEMALLOC)
3472 btrfs_balance_delayed_items(root);
3474 num_dirty = root->fs_info->dirty_metadata_bytes;
3476 if (num_dirty > thresh) {
3477 balance_dirty_pages_ratelimited_nr(
3478 root->fs_info->btree_inode->i_mapping, 1);
3483 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3485 __btrfs_btree_balance_dirty(root, 1);
3488 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3490 __btrfs_btree_balance_dirty(root, 0);
3493 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3495 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3496 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3499 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3502 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3503 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3513 void btrfs_error_commit_super(struct btrfs_root *root)
3515 mutex_lock(&root->fs_info->cleaner_mutex);
3516 btrfs_run_delayed_iputs(root);
3517 mutex_unlock(&root->fs_info->cleaner_mutex);
3519 down_write(&root->fs_info->cleanup_work_sem);
3520 up_write(&root->fs_info->cleanup_work_sem);
3522 /* cleanup FS via transaction */
3523 btrfs_cleanup_transaction(root);
3526 static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3528 struct btrfs_inode *btrfs_inode;
3529 struct list_head splice;
3531 INIT_LIST_HEAD(&splice);
3533 mutex_lock(&root->fs_info->ordered_operations_mutex);
3534 spin_lock(&root->fs_info->ordered_extent_lock);
3536 list_splice_init(&root->fs_info->ordered_operations, &splice);
3537 while (!list_empty(&splice)) {
3538 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3539 ordered_operations);
3541 list_del_init(&btrfs_inode->ordered_operations);
3543 btrfs_invalidate_inodes(btrfs_inode->root);
3546 spin_unlock(&root->fs_info->ordered_extent_lock);
3547 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3550 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3552 struct list_head splice;
3553 struct btrfs_ordered_extent *ordered;
3554 struct inode *inode;
3556 INIT_LIST_HEAD(&splice);
3558 spin_lock(&root->fs_info->ordered_extent_lock);
3560 list_splice_init(&root->fs_info->ordered_extents, &splice);
3561 while (!list_empty(&splice)) {
3562 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3565 list_del_init(&ordered->root_extent_list);
3566 atomic_inc(&ordered->refs);
3568 /* the inode may be getting freed (in sys_unlink path). */
3569 inode = igrab(ordered->inode);
3571 spin_unlock(&root->fs_info->ordered_extent_lock);
3575 atomic_set(&ordered->refs, 1);
3576 btrfs_put_ordered_extent(ordered);
3578 spin_lock(&root->fs_info->ordered_extent_lock);
3581 spin_unlock(&root->fs_info->ordered_extent_lock);
3584 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3585 struct btrfs_root *root)
3587 struct rb_node *node;
3588 struct btrfs_delayed_ref_root *delayed_refs;
3589 struct btrfs_delayed_ref_node *ref;
3592 delayed_refs = &trans->delayed_refs;
3594 spin_lock(&delayed_refs->lock);
3595 if (delayed_refs->num_entries == 0) {
3596 spin_unlock(&delayed_refs->lock);
3597 printk(KERN_INFO "delayed_refs has NO entry\n");
3601 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3602 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3604 atomic_set(&ref->refs, 1);
3605 if (btrfs_delayed_ref_is_head(ref)) {
3606 struct btrfs_delayed_ref_head *head;
3608 head = btrfs_delayed_node_to_head(ref);
3609 if (!mutex_trylock(&head->mutex)) {
3610 atomic_inc(&ref->refs);
3611 spin_unlock(&delayed_refs->lock);
3613 /* Need to wait for the delayed ref to run */
3614 mutex_lock(&head->mutex);
3615 mutex_unlock(&head->mutex);
3616 btrfs_put_delayed_ref(ref);
3618 spin_lock(&delayed_refs->lock);
3622 btrfs_free_delayed_extent_op(head->extent_op);
3623 delayed_refs->num_heads--;
3624 if (list_empty(&head->cluster))
3625 delayed_refs->num_heads_ready--;
3626 list_del_init(&head->cluster);
3629 rb_erase(&ref->rb_node, &delayed_refs->root);
3630 delayed_refs->num_entries--;
3632 spin_unlock(&delayed_refs->lock);
3633 btrfs_put_delayed_ref(ref);
3636 spin_lock(&delayed_refs->lock);
3639 spin_unlock(&delayed_refs->lock);
3644 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3646 struct btrfs_pending_snapshot *snapshot;
3647 struct list_head splice;
3649 INIT_LIST_HEAD(&splice);
3651 list_splice_init(&t->pending_snapshots, &splice);
3653 while (!list_empty(&splice)) {
3654 snapshot = list_entry(splice.next,
3655 struct btrfs_pending_snapshot,
3658 list_del_init(&snapshot->list);
3664 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3666 struct btrfs_inode *btrfs_inode;
3667 struct list_head splice;
3669 INIT_LIST_HEAD(&splice);
3671 spin_lock(&root->fs_info->delalloc_lock);
3672 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3674 while (!list_empty(&splice)) {
3675 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3678 list_del_init(&btrfs_inode->delalloc_inodes);
3680 btrfs_invalidate_inodes(btrfs_inode->root);
3683 spin_unlock(&root->fs_info->delalloc_lock);
3686 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3687 struct extent_io_tree *dirty_pages,
3692 struct inode *btree_inode = root->fs_info->btree_inode;
3693 struct extent_buffer *eb;
3697 unsigned long index;
3700 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3705 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3706 while (start <= end) {
3707 index = start >> PAGE_CACHE_SHIFT;
3708 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3709 page = find_get_page(btree_inode->i_mapping, index);
3712 offset = page_offset(page);
3714 spin_lock(&dirty_pages->buffer_lock);
3715 eb = radix_tree_lookup(
3716 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3717 offset >> PAGE_CACHE_SHIFT);
3718 spin_unlock(&dirty_pages->buffer_lock);
3720 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3722 if (PageWriteback(page))
3723 end_page_writeback(page);
3726 if (PageDirty(page)) {
3727 clear_page_dirty_for_io(page);
3728 spin_lock_irq(&page->mapping->tree_lock);
3729 radix_tree_tag_clear(&page->mapping->page_tree,
3731 PAGECACHE_TAG_DIRTY);
3732 spin_unlock_irq(&page->mapping->tree_lock);
3736 page_cache_release(page);
3743 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3744 struct extent_io_tree *pinned_extents)
3746 struct extent_io_tree *unpin;
3752 unpin = pinned_extents;
3755 ret = find_first_extent_bit(unpin, 0, &start, &end,
3756 EXTENT_DIRTY, NULL);
3761 if (btrfs_test_opt(root, DISCARD))
3762 ret = btrfs_error_discard_extent(root, start,
3766 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3767 btrfs_error_unpin_extent_range(root, start, end);
3772 if (unpin == &root->fs_info->freed_extents[0])
3773 unpin = &root->fs_info->freed_extents[1];
3775 unpin = &root->fs_info->freed_extents[0];
3783 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3784 struct btrfs_root *root)
3786 btrfs_destroy_delayed_refs(cur_trans, root);
3787 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3788 cur_trans->dirty_pages.dirty_bytes);
3790 /* FIXME: cleanup wait for commit */
3791 cur_trans->in_commit = 1;
3792 cur_trans->blocked = 1;
3793 wake_up(&root->fs_info->transaction_blocked_wait);
3795 cur_trans->blocked = 0;
3796 wake_up(&root->fs_info->transaction_wait);
3798 cur_trans->commit_done = 1;
3799 wake_up(&cur_trans->commit_wait);
3801 btrfs_destroy_delayed_inodes(root);
3802 btrfs_assert_delayed_root_empty(root);
3804 btrfs_destroy_pending_snapshots(cur_trans);
3806 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3808 btrfs_destroy_pinned_extent(root,
3809 root->fs_info->pinned_extents);
3812 memset(cur_trans, 0, sizeof(*cur_trans));
3813 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3817 int btrfs_cleanup_transaction(struct btrfs_root *root)
3819 struct btrfs_transaction *t;
3822 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3824 spin_lock(&root->fs_info->trans_lock);
3825 list_splice_init(&root->fs_info->trans_list, &list);
3826 root->fs_info->trans_no_join = 1;
3827 spin_unlock(&root->fs_info->trans_lock);
3829 while (!list_empty(&list)) {
3830 t = list_entry(list.next, struct btrfs_transaction, list);
3834 btrfs_destroy_ordered_operations(root);
3836 btrfs_destroy_ordered_extents(root);
3838 btrfs_destroy_delayed_refs(t, root);
3840 btrfs_block_rsv_release(root,
3841 &root->fs_info->trans_block_rsv,
3842 t->dirty_pages.dirty_bytes);
3844 /* FIXME: cleanup wait for commit */
3848 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3849 wake_up(&root->fs_info->transaction_blocked_wait);
3853 if (waitqueue_active(&root->fs_info->transaction_wait))
3854 wake_up(&root->fs_info->transaction_wait);
3858 if (waitqueue_active(&t->commit_wait))
3859 wake_up(&t->commit_wait);
3861 btrfs_destroy_delayed_inodes(root);
3862 btrfs_assert_delayed_root_empty(root);
3864 btrfs_destroy_pending_snapshots(t);
3866 btrfs_destroy_delalloc_inodes(root);
3868 spin_lock(&root->fs_info->trans_lock);
3869 root->fs_info->running_transaction = NULL;
3870 spin_unlock(&root->fs_info->trans_lock);
3872 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3875 btrfs_destroy_pinned_extent(root,
3876 root->fs_info->pinned_extents);
3878 atomic_set(&t->use_count, 0);
3879 list_del_init(&t->list);
3880 memset(t, 0, sizeof(*t));
3881 kmem_cache_free(btrfs_transaction_cachep, t);
3884 spin_lock(&root->fs_info->trans_lock);
3885 root->fs_info->trans_no_join = 0;
3886 spin_unlock(&root->fs_info->trans_lock);
3887 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3892 static struct extent_io_ops btree_extent_io_ops = {
3893 .readpage_end_io_hook = btree_readpage_end_io_hook,
3894 .readpage_io_failed_hook = btree_io_failed_hook,
3895 .submit_bio_hook = btree_submit_bio_hook,
3896 /* note we're sharing with inode.c for the merge bio hook */
3897 .merge_bio_hook = btrfs_merge_bio_hook,