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"
48 static struct extent_io_ops btree_extent_io_ops;
49 static void end_workqueue_fn(struct btrfs_work *work);
50 static void free_fs_root(struct btrfs_root *root);
51 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
53 static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_root *root);
57 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
60 struct extent_io_tree *dirty_pages,
62 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
63 struct extent_io_tree *pinned_extents);
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
74 struct btrfs_fs_info *info;
77 struct list_head list;
78 struct btrfs_work work;
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
86 struct async_submit_bio {
89 struct list_head list;
90 extent_submit_bio_hook_t *submit_bio_start;
91 extent_submit_bio_hook_t *submit_bio_done;
94 unsigned long bio_flags;
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
100 struct btrfs_work work;
105 * Lockdep class keys for extent_buffer->lock's in this root. For a given
106 * eb, the lockdep key is determined by the btrfs_root it belongs to and
107 * the level the eb occupies in the tree.
109 * Different roots are used for different purposes and may nest inside each
110 * other and they require separate keysets. As lockdep keys should be
111 * static, assign keysets according to the purpose of the root as indicated
112 * by btrfs_root->objectid. This ensures that all special purpose roots
113 * have separate keysets.
115 * Lock-nesting across peer nodes is always done with the immediate parent
116 * node locked thus preventing deadlock. As lockdep doesn't know this, use
117 * subclass to avoid triggering lockdep warning in such cases.
119 * The key is set by the readpage_end_io_hook after the buffer has passed
120 * csum validation but before the pages are unlocked. It is also set by
121 * btrfs_init_new_buffer on freshly allocated blocks.
123 * We also add a check to make sure the highest level of the tree is the
124 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
125 * needs update as well.
127 #ifdef CONFIG_DEBUG_LOCK_ALLOC
128 # if BTRFS_MAX_LEVEL != 8
132 static struct btrfs_lockdep_keyset {
133 u64 id; /* root objectid */
134 const char *name_stem; /* lock name stem */
135 char names[BTRFS_MAX_LEVEL + 1][20];
136 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
137 } btrfs_lockdep_keysets[] = {
138 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
139 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
140 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
141 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
142 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
143 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
144 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
145 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
146 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
147 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
148 { .id = 0, .name_stem = "tree" },
151 void __init btrfs_init_lockdep(void)
155 /* initialize lockdep class names */
156 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
157 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
159 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
160 snprintf(ks->names[j], sizeof(ks->names[j]),
161 "btrfs-%s-%02d", ks->name_stem, j);
165 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
168 struct btrfs_lockdep_keyset *ks;
170 BUG_ON(level >= ARRAY_SIZE(ks->keys));
172 /* find the matching keyset, id 0 is the default entry */
173 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
174 if (ks->id == objectid)
177 lockdep_set_class_and_name(&eb->lock,
178 &ks->keys[level], ks->names[level]);
184 * extents on the btree inode are pretty simple, there's one extent
185 * that covers the entire device
187 static struct extent_map *btree_get_extent(struct inode *inode,
188 struct page *page, size_t pg_offset, u64 start, u64 len,
191 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
192 struct extent_map *em;
195 read_lock(&em_tree->lock);
196 em = lookup_extent_mapping(em_tree, start, len);
199 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
200 read_unlock(&em_tree->lock);
203 read_unlock(&em_tree->lock);
205 em = alloc_extent_map();
207 em = ERR_PTR(-ENOMEM);
212 em->block_len = (u64)-1;
214 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
216 write_lock(&em_tree->lock);
217 ret = add_extent_mapping(em_tree, em);
218 if (ret == -EEXIST) {
219 u64 failed_start = em->start;
220 u64 failed_len = em->len;
223 em = lookup_extent_mapping(em_tree, start, len);
227 em = lookup_extent_mapping(em_tree, failed_start,
235 write_unlock(&em_tree->lock);
243 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
245 return crc32c(seed, data, len);
248 void btrfs_csum_final(u32 crc, char *result)
250 put_unaligned_le32(~crc, result);
254 * compute the csum for a btree block, and either verify it or write it
255 * into the csum field of the block.
257 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263 unsigned long cur_len;
264 unsigned long offset = BTRFS_CSUM_SIZE;
266 unsigned long map_start;
267 unsigned long map_len;
270 unsigned long inline_result;
272 len = buf->len - offset;
274 err = map_private_extent_buffer(buf, offset, 32,
275 &kaddr, &map_start, &map_len);
278 cur_len = min(len, map_len - (offset - map_start));
279 crc = btrfs_csum_data(root, kaddr + offset - map_start,
284 if (csum_size > sizeof(inline_result)) {
285 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 result = (char *)&inline_result;
292 btrfs_csum_final(crc, result);
295 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 memcpy(&found, result, csum_size);
300 read_extent_buffer(buf, &val, 0, csum_size);
301 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
302 "failed on %llu wanted %X found %X "
304 root->fs_info->sb->s_id,
305 (unsigned long long)buf->start, val, found,
306 btrfs_header_level(buf));
307 if (result != (char *)&inline_result)
312 write_extent_buffer(buf, result, 0, csum_size);
314 if (result != (char *)&inline_result)
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
325 static int verify_parent_transid(struct extent_io_tree *io_tree,
326 struct extent_buffer *eb, u64 parent_transid)
328 struct extent_state *cached_state = NULL;
331 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(io_tree, eb, cached_state) &&
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(io_tree, eb, &cached_state);
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 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
368 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
370 ret = read_extent_buffer_pages(io_tree, eb, start,
372 btree_get_extent, mirror_num);
374 !verify_parent_transid(io_tree, eb, parent_transid))
378 * This buffer's crc is fine, but its contents are corrupted, so
379 * there is no reason to read the other copies, they won't be
382 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
385 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
391 if (mirror_num > num_copies)
398 * checksum a dirty tree block before IO. This has extra checks to make sure
399 * we only fill in the checksum field in the first page of a multi-page block
402 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
404 struct extent_io_tree *tree;
405 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
408 struct extent_buffer *eb;
411 tree = &BTRFS_I(page->mapping->host)->io_tree;
413 if (page->private == EXTENT_PAGE_PRIVATE) {
417 if (!page->private) {
421 len = page->private >> 2;
424 eb = alloc_extent_buffer(tree, start, len, page);
430 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
431 btrfs_header_generation(eb));
433 btrfs_printk(root->fs_info, KERN_WARNING
434 "Failed to checksum dirty buffer @ %llu[%lu]\n",
438 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
441 found_start = btrfs_header_bytenr(eb);
442 if (found_start != start) {
446 if (eb->first_page != page) {
450 if (!PageUptodate(page)) {
454 csum_tree_block(root, eb, 0);
457 free_extent_buffer(eb);
462 static int check_tree_block_fsid(struct btrfs_root *root,
463 struct extent_buffer *eb)
465 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
466 u8 fsid[BTRFS_UUID_SIZE];
469 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
472 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
476 fs_devices = fs_devices->seed;
481 #define CORRUPT(reason, eb, root, slot) \
482 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
483 "root=%llu, slot=%d\n", reason, \
484 (unsigned long long)btrfs_header_bytenr(eb), \
485 (unsigned long long)root->objectid, slot)
487 static noinline int check_leaf(struct btrfs_root *root,
488 struct extent_buffer *leaf)
490 struct btrfs_key key;
491 struct btrfs_key leaf_key;
492 u32 nritems = btrfs_header_nritems(leaf);
498 /* Check the 0 item */
499 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
500 BTRFS_LEAF_DATA_SIZE(root)) {
501 CORRUPT("invalid item offset size pair", leaf, root, 0);
506 * Check to make sure each items keys are in the correct order and their
507 * offsets make sense. We only have to loop through nritems-1 because
508 * we check the current slot against the next slot, which verifies the
509 * next slot's offset+size makes sense and that the current's slot
512 for (slot = 0; slot < nritems - 1; slot++) {
513 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
514 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
516 /* Make sure the keys are in the right order */
517 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
518 CORRUPT("bad key order", leaf, root, slot);
523 * Make sure the offset and ends are right, remember that the
524 * item data starts at the end of the leaf and grows towards the
527 if (btrfs_item_offset_nr(leaf, slot) !=
528 btrfs_item_end_nr(leaf, slot + 1)) {
529 CORRUPT("slot offset bad", leaf, root, slot);
534 * Check to make sure that we don't point outside of the leaf,
535 * just incase all the items are consistent to eachother, but
536 * all point outside of the leaf.
538 if (btrfs_item_end_nr(leaf, slot) >
539 BTRFS_LEAF_DATA_SIZE(root)) {
540 CORRUPT("slot end outside of leaf", leaf, root, slot);
548 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
549 struct extent_state *state)
551 struct extent_io_tree *tree;
555 struct extent_buffer *eb;
556 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
559 tree = &BTRFS_I(page->mapping->host)->io_tree;
560 if (page->private == EXTENT_PAGE_PRIVATE)
565 len = page->private >> 2;
568 eb = alloc_extent_buffer(tree, start, len, page);
574 found_start = btrfs_header_bytenr(eb);
575 if (found_start != start) {
576 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
578 (unsigned long long)found_start,
579 (unsigned long long)eb->start);
583 if (eb->first_page != page) {
584 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
585 eb->first_page->index, page->index);
590 if (check_tree_block_fsid(root, eb)) {
591 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
592 (unsigned long long)eb->start);
596 found_level = btrfs_header_level(eb);
598 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
601 ret = csum_tree_block(root, eb, 1);
608 * If this is a leaf block and it is corrupt, set the corrupt bit so
609 * that we don't try and read the other copies of this block, just
612 if (found_level == 0 && check_leaf(root, eb)) {
613 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
617 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
618 end = eb->start + end - 1;
620 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
621 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
622 btree_readahead_hook(root, eb, eb->start, ret);
625 free_extent_buffer(eb);
630 static int btree_io_failed_hook(struct bio *failed_bio,
631 struct page *page, u64 start, u64 end,
632 int mirror_num, struct extent_state *state)
634 struct extent_io_tree *tree;
636 struct extent_buffer *eb;
637 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
639 tree = &BTRFS_I(page->mapping->host)->io_tree;
640 if (page->private == EXTENT_PAGE_PRIVATE)
645 len = page->private >> 2;
648 eb = alloc_extent_buffer(tree, start, len, page);
652 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
653 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
654 btree_readahead_hook(root, eb, eb->start, -EIO);
656 free_extent_buffer(eb);
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 atomic_dec(&fs_info->nr_async_submits);
754 if (atomic_read(&fs_info->nr_async_submits) < limit &&
755 waitqueue_active(&fs_info->async_submit_wait))
756 wake_up(&fs_info->async_submit_wait);
758 /* If an error occured we just want to clean up the bio and move on */
760 bio_endio(async->bio, async->error);
764 async->submit_bio_done(async->inode, async->rw, async->bio,
765 async->mirror_num, async->bio_flags,
769 static void run_one_async_free(struct btrfs_work *work)
771 struct async_submit_bio *async;
773 async = container_of(work, struct async_submit_bio, work);
777 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
778 int rw, struct bio *bio, int mirror_num,
779 unsigned long bio_flags,
781 extent_submit_bio_hook_t *submit_bio_start,
782 extent_submit_bio_hook_t *submit_bio_done)
784 struct async_submit_bio *async;
786 async = kmalloc(sizeof(*async), GFP_NOFS);
790 async->inode = inode;
793 async->mirror_num = mirror_num;
794 async->submit_bio_start = submit_bio_start;
795 async->submit_bio_done = submit_bio_done;
797 async->work.func = run_one_async_start;
798 async->work.ordered_func = run_one_async_done;
799 async->work.ordered_free = run_one_async_free;
801 async->work.flags = 0;
802 async->bio_flags = bio_flags;
803 async->bio_offset = bio_offset;
807 atomic_inc(&fs_info->nr_async_submits);
810 btrfs_set_work_high_prio(&async->work);
812 btrfs_queue_worker(&fs_info->workers, &async->work);
814 while (atomic_read(&fs_info->async_submit_draining) &&
815 atomic_read(&fs_info->nr_async_submits)) {
816 wait_event(fs_info->async_submit_wait,
817 (atomic_read(&fs_info->nr_async_submits) == 0));
823 static int btree_csum_one_bio(struct bio *bio)
825 struct bio_vec *bvec = bio->bi_io_vec;
827 struct btrfs_root *root;
830 WARN_ON(bio->bi_vcnt <= 0);
831 while (bio_index < bio->bi_vcnt) {
832 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
833 ret = csum_dirty_buffer(root, bvec->bv_page);
842 static int __btree_submit_bio_start(struct inode *inode, int rw,
843 struct bio *bio, int mirror_num,
844 unsigned long bio_flags,
848 * when we're called for a write, we're already in the async
849 * submission context. Just jump into btrfs_map_bio
851 return btree_csum_one_bio(bio);
854 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
855 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 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
865 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
866 int mirror_num, unsigned long bio_flags,
871 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, bio, 1);
875 if (!(rw & REQ_WRITE)) {
877 * called for a read, do the setup so that checksum validation
878 * can happen in the async kernel threads
880 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
885 * kthread helpers are used to submit writes so that checksumming
886 * can happen in parallel across all CPUs
888 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
889 inode, rw, bio, mirror_num, 0,
891 __btree_submit_bio_start,
892 __btree_submit_bio_done);
895 #ifdef CONFIG_MIGRATION
896 static int btree_migratepage(struct address_space *mapping,
897 struct page *newpage, struct page *page,
898 enum migrate_mode mode)
901 * we can't safely write a btree page from here,
902 * we haven't done the locking hook
907 * Buffers may be managed in a filesystem specific way.
908 * We must have no buffers or drop them.
910 if (page_has_private(page) &&
911 !try_to_release_page(page, GFP_KERNEL))
913 return migrate_page(mapping, newpage, page, mode);
917 static int btree_writepage(struct page *page, struct writeback_control *wbc)
919 struct extent_io_tree *tree;
920 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
921 struct extent_buffer *eb;
924 tree = &BTRFS_I(page->mapping->host)->io_tree;
925 if (!(current->flags & PF_MEMALLOC)) {
926 return extent_write_full_page(tree, page,
927 btree_get_extent, wbc);
930 redirty_page_for_writepage(wbc, page);
931 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
934 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
936 spin_lock(&root->fs_info->delalloc_lock);
937 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
938 spin_unlock(&root->fs_info->delalloc_lock);
940 free_extent_buffer(eb);
946 static int btree_writepages(struct address_space *mapping,
947 struct writeback_control *wbc)
949 struct extent_io_tree *tree;
950 tree = &BTRFS_I(mapping->host)->io_tree;
951 if (wbc->sync_mode == WB_SYNC_NONE) {
952 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
954 unsigned long thresh = 32 * 1024 * 1024;
956 if (wbc->for_kupdate)
959 /* this is a bit racy, but that's ok */
960 num_dirty = root->fs_info->dirty_metadata_bytes;
961 if (num_dirty < thresh)
964 return extent_writepages(tree, mapping, btree_get_extent, wbc);
967 static int btree_readpage(struct file *file, struct page *page)
969 struct extent_io_tree *tree;
970 tree = &BTRFS_I(page->mapping->host)->io_tree;
971 return extent_read_full_page(tree, page, btree_get_extent, 0);
974 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
976 struct extent_io_tree *tree;
977 struct extent_map_tree *map;
980 if (PageWriteback(page) || PageDirty(page))
983 tree = &BTRFS_I(page->mapping->host)->io_tree;
984 map = &BTRFS_I(page->mapping->host)->extent_tree;
987 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
988 * slab allocation from alloc_extent_state down the callchain where
989 * it'd hit a BUG_ON as those flags are not allowed.
991 gfp_flags &= ~GFP_SLAB_BUG_MASK;
993 ret = try_release_extent_state(map, tree, page, gfp_flags);
997 ret = try_release_extent_buffer(tree, page);
999 ClearPagePrivate(page);
1000 set_page_private(page, 0);
1001 page_cache_release(page);
1007 static void btree_invalidatepage(struct page *page, unsigned long offset)
1009 struct extent_io_tree *tree;
1010 tree = &BTRFS_I(page->mapping->host)->io_tree;
1011 extent_invalidatepage(tree, page, offset);
1012 btree_releasepage(page, GFP_NOFS);
1013 if (PagePrivate(page)) {
1014 printk(KERN_WARNING "btrfs warning page private not zero "
1015 "on page %llu\n", (unsigned long long)page_offset(page));
1016 ClearPagePrivate(page);
1017 set_page_private(page, 0);
1018 page_cache_release(page);
1022 static const struct address_space_operations btree_aops = {
1023 .readpage = btree_readpage,
1024 .writepage = btree_writepage,
1025 .writepages = btree_writepages,
1026 .releasepage = btree_releasepage,
1027 .invalidatepage = btree_invalidatepage,
1028 #ifdef CONFIG_MIGRATION
1029 .migratepage = btree_migratepage,
1033 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1036 struct extent_buffer *buf = NULL;
1037 struct inode *btree_inode = root->fs_info->btree_inode;
1040 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1043 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1044 buf, 0, WAIT_NONE, btree_get_extent, 0);
1045 free_extent_buffer(buf);
1049 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1050 int mirror_num, struct extent_buffer **eb)
1052 struct extent_buffer *buf = NULL;
1053 struct inode *btree_inode = root->fs_info->btree_inode;
1054 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1057 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1061 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1063 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1064 btree_get_extent, mirror_num);
1066 free_extent_buffer(buf);
1070 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1071 free_extent_buffer(buf);
1073 } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1076 free_extent_buffer(buf);
1081 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1082 u64 bytenr, u32 blocksize)
1084 struct inode *btree_inode = root->fs_info->btree_inode;
1085 struct extent_buffer *eb;
1086 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1091 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1092 u64 bytenr, u32 blocksize)
1094 struct inode *btree_inode = root->fs_info->btree_inode;
1095 struct extent_buffer *eb;
1097 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1098 bytenr, blocksize, NULL);
1103 int btrfs_write_tree_block(struct extent_buffer *buf)
1105 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1106 buf->start + buf->len - 1);
1109 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1111 return filemap_fdatawait_range(buf->first_page->mapping,
1112 buf->start, buf->start + buf->len - 1);
1115 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1116 u32 blocksize, u64 parent_transid)
1118 struct extent_buffer *buf = NULL;
1121 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1125 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1128 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1133 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1134 struct extent_buffer *buf)
1136 struct inode *btree_inode = root->fs_info->btree_inode;
1137 if (btrfs_header_generation(buf) ==
1138 root->fs_info->running_transaction->transid) {
1139 btrfs_assert_tree_locked(buf);
1141 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1142 spin_lock(&root->fs_info->delalloc_lock);
1143 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1144 root->fs_info->dirty_metadata_bytes -= buf->len;
1146 spin_unlock(&root->fs_info->delalloc_lock);
1147 btrfs_panic(root->fs_info, -EOVERFLOW,
1148 "Can't clear %lu bytes from "
1149 " dirty_mdatadata_bytes (%lu)",
1151 root->fs_info->dirty_metadata_bytes);
1153 spin_unlock(&root->fs_info->delalloc_lock);
1156 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1157 btrfs_set_lock_blocking(buf);
1158 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1163 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1164 u32 stripesize, struct btrfs_root *root,
1165 struct btrfs_fs_info *fs_info,
1169 root->commit_root = NULL;
1170 root->sectorsize = sectorsize;
1171 root->nodesize = nodesize;
1172 root->leafsize = leafsize;
1173 root->stripesize = stripesize;
1175 root->track_dirty = 0;
1177 root->orphan_item_inserted = 0;
1178 root->orphan_cleanup_state = 0;
1180 root->objectid = objectid;
1181 root->last_trans = 0;
1182 root->highest_objectid = 0;
1184 root->inode_tree = RB_ROOT;
1185 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1186 root->block_rsv = NULL;
1187 root->orphan_block_rsv = NULL;
1189 INIT_LIST_HEAD(&root->dirty_list);
1190 INIT_LIST_HEAD(&root->orphan_list);
1191 INIT_LIST_HEAD(&root->root_list);
1192 spin_lock_init(&root->orphan_lock);
1193 spin_lock_init(&root->inode_lock);
1194 spin_lock_init(&root->accounting_lock);
1195 mutex_init(&root->objectid_mutex);
1196 mutex_init(&root->log_mutex);
1197 init_waitqueue_head(&root->log_writer_wait);
1198 init_waitqueue_head(&root->log_commit_wait[0]);
1199 init_waitqueue_head(&root->log_commit_wait[1]);
1200 atomic_set(&root->log_commit[0], 0);
1201 atomic_set(&root->log_commit[1], 0);
1202 atomic_set(&root->log_writers, 0);
1203 root->log_batch = 0;
1204 root->log_transid = 0;
1205 root->last_log_commit = 0;
1206 extent_io_tree_init(&root->dirty_log_pages,
1207 fs_info->btree_inode->i_mapping);
1209 memset(&root->root_key, 0, sizeof(root->root_key));
1210 memset(&root->root_item, 0, sizeof(root->root_item));
1211 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1212 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1213 root->defrag_trans_start = fs_info->generation;
1214 init_completion(&root->kobj_unregister);
1215 root->defrag_running = 0;
1216 root->root_key.objectid = objectid;
1220 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1221 struct btrfs_fs_info *fs_info,
1223 struct btrfs_root *root)
1229 __setup_root(tree_root->nodesize, tree_root->leafsize,
1230 tree_root->sectorsize, tree_root->stripesize,
1231 root, fs_info, objectid);
1232 ret = btrfs_find_last_root(tree_root, objectid,
1233 &root->root_item, &root->root_key);
1239 generation = btrfs_root_generation(&root->root_item);
1240 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1241 root->commit_root = NULL;
1242 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1243 blocksize, generation);
1244 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1245 free_extent_buffer(root->node);
1249 root->commit_root = btrfs_root_node(root);
1253 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1255 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1257 root->fs_info = fs_info;
1261 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1262 struct btrfs_fs_info *fs_info)
1264 struct btrfs_root *root;
1265 struct btrfs_root *tree_root = fs_info->tree_root;
1266 struct extent_buffer *leaf;
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, BTRFS_TREE_LOG_OBJECTID);
1276 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1277 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1278 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1280 * log trees do not get reference counted because they go away
1281 * before a real commit is actually done. They do store pointers
1282 * to file data extents, and those reference counts still get
1283 * updated (along with back refs to the log tree).
1287 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1288 BTRFS_TREE_LOG_OBJECTID, NULL,
1292 return ERR_CAST(leaf);
1295 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1296 btrfs_set_header_bytenr(leaf, leaf->start);
1297 btrfs_set_header_generation(leaf, trans->transid);
1298 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1299 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1302 write_extent_buffer(root->node, root->fs_info->fsid,
1303 (unsigned long)btrfs_header_fsid(root->node),
1305 btrfs_mark_buffer_dirty(root->node);
1306 btrfs_tree_unlock(root->node);
1310 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1311 struct btrfs_fs_info *fs_info)
1313 struct btrfs_root *log_root;
1315 log_root = alloc_log_tree(trans, fs_info);
1316 if (IS_ERR(log_root))
1317 return PTR_ERR(log_root);
1318 WARN_ON(fs_info->log_root_tree);
1319 fs_info->log_root_tree = log_root;
1323 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1324 struct btrfs_root *root)
1326 struct btrfs_root *log_root;
1327 struct btrfs_inode_item *inode_item;
1329 log_root = alloc_log_tree(trans, root->fs_info);
1330 if (IS_ERR(log_root))
1331 return PTR_ERR(log_root);
1333 log_root->last_trans = trans->transid;
1334 log_root->root_key.offset = root->root_key.objectid;
1336 inode_item = &log_root->root_item.inode;
1337 inode_item->generation = cpu_to_le64(1);
1338 inode_item->size = cpu_to_le64(3);
1339 inode_item->nlink = cpu_to_le32(1);
1340 inode_item->nbytes = cpu_to_le64(root->leafsize);
1341 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1343 btrfs_set_root_node(&log_root->root_item, log_root->node);
1345 WARN_ON(root->log_root);
1346 root->log_root = log_root;
1347 root->log_transid = 0;
1348 root->last_log_commit = 0;
1352 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1353 struct btrfs_key *location)
1355 struct btrfs_root *root;
1356 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1357 struct btrfs_path *path;
1358 struct extent_buffer *l;
1363 root = btrfs_alloc_root(fs_info);
1365 return ERR_PTR(-ENOMEM);
1366 if (location->offset == (u64)-1) {
1367 ret = find_and_setup_root(tree_root, fs_info,
1368 location->objectid, root);
1371 return ERR_PTR(ret);
1376 __setup_root(tree_root->nodesize, tree_root->leafsize,
1377 tree_root->sectorsize, tree_root->stripesize,
1378 root, fs_info, location->objectid);
1380 path = btrfs_alloc_path();
1383 return ERR_PTR(-ENOMEM);
1385 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1388 read_extent_buffer(l, &root->root_item,
1389 btrfs_item_ptr_offset(l, path->slots[0]),
1390 sizeof(root->root_item));
1391 memcpy(&root->root_key, location, sizeof(*location));
1393 btrfs_free_path(path);
1398 return ERR_PTR(ret);
1401 generation = btrfs_root_generation(&root->root_item);
1402 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1403 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1404 blocksize, generation);
1405 root->commit_root = btrfs_root_node(root);
1406 BUG_ON(!root->node); /* -ENOMEM */
1408 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1410 btrfs_check_and_init_root_item(&root->root_item);
1416 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1417 struct btrfs_key *location)
1419 struct btrfs_root *root;
1422 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1423 return fs_info->tree_root;
1424 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1425 return fs_info->extent_root;
1426 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1427 return fs_info->chunk_root;
1428 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1429 return fs_info->dev_root;
1430 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1431 return fs_info->csum_root;
1433 spin_lock(&fs_info->fs_roots_radix_lock);
1434 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1435 (unsigned long)location->objectid);
1436 spin_unlock(&fs_info->fs_roots_radix_lock);
1440 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1444 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1445 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1447 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1452 btrfs_init_free_ino_ctl(root);
1453 mutex_init(&root->fs_commit_mutex);
1454 spin_lock_init(&root->cache_lock);
1455 init_waitqueue_head(&root->cache_wait);
1457 ret = get_anon_bdev(&root->anon_dev);
1461 if (btrfs_root_refs(&root->root_item) == 0) {
1466 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1470 root->orphan_item_inserted = 1;
1472 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1476 spin_lock(&fs_info->fs_roots_radix_lock);
1477 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1478 (unsigned long)root->root_key.objectid,
1483 spin_unlock(&fs_info->fs_roots_radix_lock);
1484 radix_tree_preload_end();
1486 if (ret == -EEXIST) {
1493 ret = btrfs_find_dead_roots(fs_info->tree_root,
1494 root->root_key.objectid);
1499 return ERR_PTR(ret);
1502 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1504 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1506 struct btrfs_device *device;
1507 struct backing_dev_info *bdi;
1510 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1513 bdi = blk_get_backing_dev_info(device->bdev);
1514 if (bdi && bdi_congested(bdi, bdi_bits)) {
1524 * If this fails, caller must call bdi_destroy() to get rid of the
1527 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1531 bdi->capabilities = BDI_CAP_MAP_COPY;
1532 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1536 bdi->ra_pages = default_backing_dev_info.ra_pages;
1537 bdi->congested_fn = btrfs_congested_fn;
1538 bdi->congested_data = info;
1542 static int bio_ready_for_csum(struct bio *bio)
1548 struct extent_io_tree *io_tree = NULL;
1549 struct bio_vec *bvec;
1553 bio_for_each_segment(bvec, bio, i) {
1554 page = bvec->bv_page;
1555 if (page->private == EXTENT_PAGE_PRIVATE) {
1556 length += bvec->bv_len;
1559 if (!page->private) {
1560 length += bvec->bv_len;
1563 length = bvec->bv_len;
1564 buf_len = page->private >> 2;
1565 start = page_offset(page) + bvec->bv_offset;
1566 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1568 /* are we fully contained in this bio? */
1569 if (buf_len <= length)
1572 ret = extent_range_uptodate(io_tree, start + length,
1573 start + buf_len - 1);
1578 * called by the kthread helper functions to finally call the bio end_io
1579 * functions. This is where read checksum verification actually happens
1581 static void end_workqueue_fn(struct btrfs_work *work)
1584 struct end_io_wq *end_io_wq;
1585 struct btrfs_fs_info *fs_info;
1588 end_io_wq = container_of(work, struct end_io_wq, work);
1589 bio = end_io_wq->bio;
1590 fs_info = end_io_wq->info;
1592 /* metadata bio reads are special because the whole tree block must
1593 * be checksummed at once. This makes sure the entire block is in
1594 * ram and up to date before trying to verify things. For
1595 * blocksize <= pagesize, it is basically a noop
1597 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1598 !bio_ready_for_csum(bio)) {
1599 btrfs_queue_worker(&fs_info->endio_meta_workers,
1603 error = end_io_wq->error;
1604 bio->bi_private = end_io_wq->private;
1605 bio->bi_end_io = end_io_wq->end_io;
1607 bio_endio(bio, error);
1610 static int cleaner_kthread(void *arg)
1612 struct btrfs_root *root = arg;
1615 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1617 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1618 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1619 btrfs_run_delayed_iputs(root);
1620 btrfs_clean_old_snapshots(root);
1621 mutex_unlock(&root->fs_info->cleaner_mutex);
1622 btrfs_run_defrag_inodes(root->fs_info);
1625 if (!try_to_freeze()) {
1626 set_current_state(TASK_INTERRUPTIBLE);
1627 if (!kthread_should_stop())
1629 __set_current_state(TASK_RUNNING);
1631 } while (!kthread_should_stop());
1635 static int transaction_kthread(void *arg)
1637 struct btrfs_root *root = arg;
1638 struct btrfs_trans_handle *trans;
1639 struct btrfs_transaction *cur;
1642 unsigned long delay;
1646 cannot_commit = false;
1648 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1649 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1651 spin_lock(&root->fs_info->trans_lock);
1652 cur = root->fs_info->running_transaction;
1654 spin_unlock(&root->fs_info->trans_lock);
1658 now = get_seconds();
1659 if (!cur->blocked &&
1660 (now < cur->start_time || now - cur->start_time < 30)) {
1661 spin_unlock(&root->fs_info->trans_lock);
1665 transid = cur->transid;
1666 spin_unlock(&root->fs_info->trans_lock);
1668 /* If the file system is aborted, this will always fail. */
1669 trans = btrfs_join_transaction(root);
1670 if (IS_ERR(trans)) {
1671 cannot_commit = true;
1674 if (transid == trans->transid) {
1675 btrfs_commit_transaction(trans, root);
1677 btrfs_end_transaction(trans, root);
1680 wake_up_process(root->fs_info->cleaner_kthread);
1681 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1683 if (!try_to_freeze()) {
1684 set_current_state(TASK_INTERRUPTIBLE);
1685 if (!kthread_should_stop() &&
1686 (!btrfs_transaction_blocked(root->fs_info) ||
1688 schedule_timeout(delay);
1689 __set_current_state(TASK_RUNNING);
1691 } while (!kthread_should_stop());
1696 * this will find the highest generation in the array of
1697 * root backups. The index of the highest array is returned,
1698 * or -1 if we can't find anything.
1700 * We check to make sure the array is valid by comparing the
1701 * generation of the latest root in the array with the generation
1702 * in the super block. If they don't match we pitch it.
1704 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1707 int newest_index = -1;
1708 struct btrfs_root_backup *root_backup;
1711 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1712 root_backup = info->super_copy->super_roots + i;
1713 cur = btrfs_backup_tree_root_gen(root_backup);
1714 if (cur == newest_gen)
1718 /* check to see if we actually wrapped around */
1719 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1720 root_backup = info->super_copy->super_roots;
1721 cur = btrfs_backup_tree_root_gen(root_backup);
1722 if (cur == newest_gen)
1725 return newest_index;
1730 * find the oldest backup so we know where to store new entries
1731 * in the backup array. This will set the backup_root_index
1732 * field in the fs_info struct
1734 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1737 int newest_index = -1;
1739 newest_index = find_newest_super_backup(info, newest_gen);
1740 /* if there was garbage in there, just move along */
1741 if (newest_index == -1) {
1742 info->backup_root_index = 0;
1744 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1749 * copy all the root pointers into the super backup array.
1750 * this will bump the backup pointer by one when it is
1753 static void backup_super_roots(struct btrfs_fs_info *info)
1756 struct btrfs_root_backup *root_backup;
1759 next_backup = info->backup_root_index;
1760 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1761 BTRFS_NUM_BACKUP_ROOTS;
1764 * just overwrite the last backup if we're at the same generation
1765 * this happens only at umount
1767 root_backup = info->super_for_commit->super_roots + last_backup;
1768 if (btrfs_backup_tree_root_gen(root_backup) ==
1769 btrfs_header_generation(info->tree_root->node))
1770 next_backup = last_backup;
1772 root_backup = info->super_for_commit->super_roots + next_backup;
1775 * make sure all of our padding and empty slots get zero filled
1776 * regardless of which ones we use today
1778 memset(root_backup, 0, sizeof(*root_backup));
1780 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1782 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1783 btrfs_set_backup_tree_root_gen(root_backup,
1784 btrfs_header_generation(info->tree_root->node));
1786 btrfs_set_backup_tree_root_level(root_backup,
1787 btrfs_header_level(info->tree_root->node));
1789 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1790 btrfs_set_backup_chunk_root_gen(root_backup,
1791 btrfs_header_generation(info->chunk_root->node));
1792 btrfs_set_backup_chunk_root_level(root_backup,
1793 btrfs_header_level(info->chunk_root->node));
1795 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1796 btrfs_set_backup_extent_root_gen(root_backup,
1797 btrfs_header_generation(info->extent_root->node));
1798 btrfs_set_backup_extent_root_level(root_backup,
1799 btrfs_header_level(info->extent_root->node));
1802 * we might commit during log recovery, which happens before we set
1803 * the fs_root. Make sure it is valid before we fill it in.
1805 if (info->fs_root && info->fs_root->node) {
1806 btrfs_set_backup_fs_root(root_backup,
1807 info->fs_root->node->start);
1808 btrfs_set_backup_fs_root_gen(root_backup,
1809 btrfs_header_generation(info->fs_root->node));
1810 btrfs_set_backup_fs_root_level(root_backup,
1811 btrfs_header_level(info->fs_root->node));
1814 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1815 btrfs_set_backup_dev_root_gen(root_backup,
1816 btrfs_header_generation(info->dev_root->node));
1817 btrfs_set_backup_dev_root_level(root_backup,
1818 btrfs_header_level(info->dev_root->node));
1820 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1821 btrfs_set_backup_csum_root_gen(root_backup,
1822 btrfs_header_generation(info->csum_root->node));
1823 btrfs_set_backup_csum_root_level(root_backup,
1824 btrfs_header_level(info->csum_root->node));
1826 btrfs_set_backup_total_bytes(root_backup,
1827 btrfs_super_total_bytes(info->super_copy));
1828 btrfs_set_backup_bytes_used(root_backup,
1829 btrfs_super_bytes_used(info->super_copy));
1830 btrfs_set_backup_num_devices(root_backup,
1831 btrfs_super_num_devices(info->super_copy));
1834 * if we don't copy this out to the super_copy, it won't get remembered
1835 * for the next commit
1837 memcpy(&info->super_copy->super_roots,
1838 &info->super_for_commit->super_roots,
1839 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1843 * this copies info out of the root backup array and back into
1844 * the in-memory super block. It is meant to help iterate through
1845 * the array, so you send it the number of backups you've already
1846 * tried and the last backup index you used.
1848 * this returns -1 when it has tried all the backups
1850 static noinline int next_root_backup(struct btrfs_fs_info *info,
1851 struct btrfs_super_block *super,
1852 int *num_backups_tried, int *backup_index)
1854 struct btrfs_root_backup *root_backup;
1855 int newest = *backup_index;
1857 if (*num_backups_tried == 0) {
1858 u64 gen = btrfs_super_generation(super);
1860 newest = find_newest_super_backup(info, gen);
1864 *backup_index = newest;
1865 *num_backups_tried = 1;
1866 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1867 /* we've tried all the backups, all done */
1870 /* jump to the next oldest backup */
1871 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1872 BTRFS_NUM_BACKUP_ROOTS;
1873 *backup_index = newest;
1874 *num_backups_tried += 1;
1876 root_backup = super->super_roots + newest;
1878 btrfs_set_super_generation(super,
1879 btrfs_backup_tree_root_gen(root_backup));
1880 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1881 btrfs_set_super_root_level(super,
1882 btrfs_backup_tree_root_level(root_backup));
1883 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1886 * fixme: the total bytes and num_devices need to match or we should
1889 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1890 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1894 /* helper to cleanup tree roots */
1895 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1897 free_extent_buffer(info->tree_root->node);
1898 free_extent_buffer(info->tree_root->commit_root);
1899 free_extent_buffer(info->dev_root->node);
1900 free_extent_buffer(info->dev_root->commit_root);
1901 free_extent_buffer(info->extent_root->node);
1902 free_extent_buffer(info->extent_root->commit_root);
1903 free_extent_buffer(info->csum_root->node);
1904 free_extent_buffer(info->csum_root->commit_root);
1906 info->tree_root->node = NULL;
1907 info->tree_root->commit_root = NULL;
1908 info->dev_root->node = NULL;
1909 info->dev_root->commit_root = NULL;
1910 info->extent_root->node = NULL;
1911 info->extent_root->commit_root = NULL;
1912 info->csum_root->node = NULL;
1913 info->csum_root->commit_root = NULL;
1916 free_extent_buffer(info->chunk_root->node);
1917 free_extent_buffer(info->chunk_root->commit_root);
1918 info->chunk_root->node = NULL;
1919 info->chunk_root->commit_root = NULL;
1924 int open_ctree(struct super_block *sb,
1925 struct btrfs_fs_devices *fs_devices,
1935 struct btrfs_key location;
1936 struct buffer_head *bh;
1937 struct btrfs_super_block *disk_super;
1938 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1939 struct btrfs_root *tree_root;
1940 struct btrfs_root *extent_root;
1941 struct btrfs_root *csum_root;
1942 struct btrfs_root *chunk_root;
1943 struct btrfs_root *dev_root;
1944 struct btrfs_root *log_tree_root;
1947 int num_backups_tried = 0;
1948 int backup_index = 0;
1950 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1951 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1952 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1953 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1954 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1956 if (!tree_root || !extent_root || !csum_root ||
1957 !chunk_root || !dev_root) {
1962 ret = init_srcu_struct(&fs_info->subvol_srcu);
1968 ret = setup_bdi(fs_info, &fs_info->bdi);
1974 fs_info->btree_inode = new_inode(sb);
1975 if (!fs_info->btree_inode) {
1980 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1982 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1983 INIT_LIST_HEAD(&fs_info->trans_list);
1984 INIT_LIST_HEAD(&fs_info->dead_roots);
1985 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1986 INIT_LIST_HEAD(&fs_info->hashers);
1987 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1988 INIT_LIST_HEAD(&fs_info->ordered_operations);
1989 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1990 spin_lock_init(&fs_info->delalloc_lock);
1991 spin_lock_init(&fs_info->trans_lock);
1992 spin_lock_init(&fs_info->ref_cache_lock);
1993 spin_lock_init(&fs_info->fs_roots_radix_lock);
1994 spin_lock_init(&fs_info->delayed_iput_lock);
1995 spin_lock_init(&fs_info->defrag_inodes_lock);
1996 spin_lock_init(&fs_info->free_chunk_lock);
1997 mutex_init(&fs_info->reloc_mutex);
1999 init_completion(&fs_info->kobj_unregister);
2000 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2001 INIT_LIST_HEAD(&fs_info->space_info);
2002 btrfs_mapping_init(&fs_info->mapping_tree);
2003 btrfs_init_block_rsv(&fs_info->global_block_rsv);
2004 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
2005 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
2006 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
2007 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
2008 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
2009 atomic_set(&fs_info->nr_async_submits, 0);
2010 atomic_set(&fs_info->async_delalloc_pages, 0);
2011 atomic_set(&fs_info->async_submit_draining, 0);
2012 atomic_set(&fs_info->nr_async_bios, 0);
2013 atomic_set(&fs_info->defrag_running, 0);
2015 fs_info->max_inline = 8192 * 1024;
2016 fs_info->metadata_ratio = 0;
2017 fs_info->defrag_inodes = RB_ROOT;
2018 fs_info->trans_no_join = 0;
2019 fs_info->free_chunk_space = 0;
2021 /* readahead state */
2022 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2023 spin_lock_init(&fs_info->reada_lock);
2025 fs_info->thread_pool_size = min_t(unsigned long,
2026 num_online_cpus() + 2, 8);
2028 INIT_LIST_HEAD(&fs_info->ordered_extents);
2029 spin_lock_init(&fs_info->ordered_extent_lock);
2030 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2032 if (!fs_info->delayed_root) {
2036 btrfs_init_delayed_root(fs_info->delayed_root);
2038 mutex_init(&fs_info->scrub_lock);
2039 atomic_set(&fs_info->scrubs_running, 0);
2040 atomic_set(&fs_info->scrub_pause_req, 0);
2041 atomic_set(&fs_info->scrubs_paused, 0);
2042 atomic_set(&fs_info->scrub_cancel_req, 0);
2043 init_waitqueue_head(&fs_info->scrub_pause_wait);
2044 init_rwsem(&fs_info->scrub_super_lock);
2045 fs_info->scrub_workers_refcnt = 0;
2046 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2047 fs_info->check_integrity_print_mask = 0;
2050 spin_lock_init(&fs_info->balance_lock);
2051 mutex_init(&fs_info->balance_mutex);
2052 atomic_set(&fs_info->balance_running, 0);
2053 atomic_set(&fs_info->balance_pause_req, 0);
2054 atomic_set(&fs_info->balance_cancel_req, 0);
2055 fs_info->balance_ctl = NULL;
2056 init_waitqueue_head(&fs_info->balance_wait_q);
2058 sb->s_blocksize = 4096;
2059 sb->s_blocksize_bits = blksize_bits(4096);
2060 sb->s_bdi = &fs_info->bdi;
2062 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2063 set_nlink(fs_info->btree_inode, 1);
2065 * we set the i_size on the btree inode to the max possible int.
2066 * the real end of the address space is determined by all of
2067 * the devices in the system
2069 fs_info->btree_inode->i_size = OFFSET_MAX;
2070 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2071 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2073 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2074 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2075 fs_info->btree_inode->i_mapping);
2076 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2078 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2080 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2081 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2082 sizeof(struct btrfs_key));
2083 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2084 insert_inode_hash(fs_info->btree_inode);
2086 spin_lock_init(&fs_info->block_group_cache_lock);
2087 fs_info->block_group_cache_tree = RB_ROOT;
2089 extent_io_tree_init(&fs_info->freed_extents[0],
2090 fs_info->btree_inode->i_mapping);
2091 extent_io_tree_init(&fs_info->freed_extents[1],
2092 fs_info->btree_inode->i_mapping);
2093 fs_info->pinned_extents = &fs_info->freed_extents[0];
2094 fs_info->do_barriers = 1;
2097 mutex_init(&fs_info->ordered_operations_mutex);
2098 mutex_init(&fs_info->tree_log_mutex);
2099 mutex_init(&fs_info->chunk_mutex);
2100 mutex_init(&fs_info->transaction_kthread_mutex);
2101 mutex_init(&fs_info->cleaner_mutex);
2102 mutex_init(&fs_info->volume_mutex);
2103 init_rwsem(&fs_info->extent_commit_sem);
2104 init_rwsem(&fs_info->cleanup_work_sem);
2105 init_rwsem(&fs_info->subvol_sem);
2107 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2108 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2110 init_waitqueue_head(&fs_info->transaction_throttle);
2111 init_waitqueue_head(&fs_info->transaction_wait);
2112 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2113 init_waitqueue_head(&fs_info->async_submit_wait);
2115 __setup_root(4096, 4096, 4096, 4096, tree_root,
2116 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2118 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2124 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2125 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2126 sizeof(*fs_info->super_for_commit));
2129 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2131 disk_super = fs_info->super_copy;
2132 if (!btrfs_super_root(disk_super))
2135 /* check FS state, whether FS is broken. */
2136 fs_info->fs_state |= btrfs_super_flags(disk_super);
2138 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2141 * run through our array of backup supers and setup
2142 * our ring pointer to the oldest one
2144 generation = btrfs_super_generation(disk_super);
2145 find_oldest_super_backup(fs_info, generation);
2148 * In the long term, we'll store the compression type in the super
2149 * block, and it'll be used for per file compression control.
2151 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2153 ret = btrfs_parse_options(tree_root, options);
2159 features = btrfs_super_incompat_flags(disk_super) &
2160 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2162 printk(KERN_ERR "BTRFS: couldn't mount because of "
2163 "unsupported optional features (%Lx).\n",
2164 (unsigned long long)features);
2169 features = btrfs_super_incompat_flags(disk_super);
2170 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2171 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2172 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2173 btrfs_set_super_incompat_flags(disk_super, features);
2175 features = btrfs_super_compat_ro_flags(disk_super) &
2176 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2177 if (!(sb->s_flags & MS_RDONLY) && features) {
2178 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2179 "unsupported option features (%Lx).\n",
2180 (unsigned long long)features);
2185 btrfs_init_workers(&fs_info->generic_worker,
2186 "genwork", 1, NULL);
2188 btrfs_init_workers(&fs_info->workers, "worker",
2189 fs_info->thread_pool_size,
2190 &fs_info->generic_worker);
2192 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2193 fs_info->thread_pool_size,
2194 &fs_info->generic_worker);
2196 btrfs_init_workers(&fs_info->submit_workers, "submit",
2197 min_t(u64, fs_devices->num_devices,
2198 fs_info->thread_pool_size),
2199 &fs_info->generic_worker);
2201 btrfs_init_workers(&fs_info->caching_workers, "cache",
2202 2, &fs_info->generic_worker);
2204 /* a higher idle thresh on the submit workers makes it much more
2205 * likely that bios will be send down in a sane order to the
2208 fs_info->submit_workers.idle_thresh = 64;
2210 fs_info->workers.idle_thresh = 16;
2211 fs_info->workers.ordered = 1;
2213 fs_info->delalloc_workers.idle_thresh = 2;
2214 fs_info->delalloc_workers.ordered = 1;
2216 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2217 &fs_info->generic_worker);
2218 btrfs_init_workers(&fs_info->endio_workers, "endio",
2219 fs_info->thread_pool_size,
2220 &fs_info->generic_worker);
2221 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2222 fs_info->thread_pool_size,
2223 &fs_info->generic_worker);
2224 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2225 "endio-meta-write", fs_info->thread_pool_size,
2226 &fs_info->generic_worker);
2227 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2228 fs_info->thread_pool_size,
2229 &fs_info->generic_worker);
2230 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2231 1, &fs_info->generic_worker);
2232 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2233 fs_info->thread_pool_size,
2234 &fs_info->generic_worker);
2235 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2236 fs_info->thread_pool_size,
2237 &fs_info->generic_worker);
2240 * endios are largely parallel and should have a very
2243 fs_info->endio_workers.idle_thresh = 4;
2244 fs_info->endio_meta_workers.idle_thresh = 4;
2246 fs_info->endio_write_workers.idle_thresh = 2;
2247 fs_info->endio_meta_write_workers.idle_thresh = 2;
2248 fs_info->readahead_workers.idle_thresh = 2;
2251 * btrfs_start_workers can really only fail because of ENOMEM so just
2252 * return -ENOMEM if any of these fail.
2254 ret = btrfs_start_workers(&fs_info->workers);
2255 ret |= btrfs_start_workers(&fs_info->generic_worker);
2256 ret |= btrfs_start_workers(&fs_info->submit_workers);
2257 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2258 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2259 ret |= btrfs_start_workers(&fs_info->endio_workers);
2260 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2261 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2262 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2263 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2264 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2265 ret |= btrfs_start_workers(&fs_info->caching_workers);
2266 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2269 goto fail_sb_buffer;
2272 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2273 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2274 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2276 nodesize = btrfs_super_nodesize(disk_super);
2277 leafsize = btrfs_super_leafsize(disk_super);
2278 sectorsize = btrfs_super_sectorsize(disk_super);
2279 stripesize = btrfs_super_stripesize(disk_super);
2280 tree_root->nodesize = nodesize;
2281 tree_root->leafsize = leafsize;
2282 tree_root->sectorsize = sectorsize;
2283 tree_root->stripesize = stripesize;
2285 sb->s_blocksize = sectorsize;
2286 sb->s_blocksize_bits = blksize_bits(sectorsize);
2288 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2289 sizeof(disk_super->magic))) {
2290 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2291 goto fail_sb_buffer;
2294 if (sectorsize < PAGE_SIZE) {
2295 printk(KERN_WARNING "btrfs: Incompatible sector size "
2296 "found on %s\n", sb->s_id);
2297 goto fail_sb_buffer;
2300 mutex_lock(&fs_info->chunk_mutex);
2301 ret = btrfs_read_sys_array(tree_root);
2302 mutex_unlock(&fs_info->chunk_mutex);
2304 printk(KERN_WARNING "btrfs: failed to read the system "
2305 "array on %s\n", sb->s_id);
2306 goto fail_sb_buffer;
2309 blocksize = btrfs_level_size(tree_root,
2310 btrfs_super_chunk_root_level(disk_super));
2311 generation = btrfs_super_chunk_root_generation(disk_super);
2313 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2314 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2316 chunk_root->node = read_tree_block(chunk_root,
2317 btrfs_super_chunk_root(disk_super),
2318 blocksize, generation);
2319 BUG_ON(!chunk_root->node); /* -ENOMEM */
2320 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2321 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2323 goto fail_tree_roots;
2325 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2326 chunk_root->commit_root = btrfs_root_node(chunk_root);
2328 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2329 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2332 ret = btrfs_read_chunk_tree(chunk_root);
2334 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2336 goto fail_tree_roots;
2339 btrfs_close_extra_devices(fs_devices);
2341 if (!fs_devices->latest_bdev) {
2342 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2344 goto fail_tree_roots;
2348 blocksize = btrfs_level_size(tree_root,
2349 btrfs_super_root_level(disk_super));
2350 generation = btrfs_super_generation(disk_super);
2352 tree_root->node = read_tree_block(tree_root,
2353 btrfs_super_root(disk_super),
2354 blocksize, generation);
2355 if (!tree_root->node ||
2356 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2357 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2360 goto recovery_tree_root;
2363 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2364 tree_root->commit_root = btrfs_root_node(tree_root);
2366 ret = find_and_setup_root(tree_root, fs_info,
2367 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2369 goto recovery_tree_root;
2370 extent_root->track_dirty = 1;
2372 ret = find_and_setup_root(tree_root, fs_info,
2373 BTRFS_DEV_TREE_OBJECTID, dev_root);
2375 goto recovery_tree_root;
2376 dev_root->track_dirty = 1;
2378 ret = find_and_setup_root(tree_root, fs_info,
2379 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2381 goto recovery_tree_root;
2383 csum_root->track_dirty = 1;
2385 fs_info->generation = generation;
2386 fs_info->last_trans_committed = generation;
2388 ret = btrfs_init_space_info(fs_info);
2390 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2391 goto fail_block_groups;
2394 ret = btrfs_read_block_groups(extent_root);
2396 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2397 goto fail_block_groups;
2400 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2402 if (IS_ERR(fs_info->cleaner_kthread))
2403 goto fail_block_groups;
2405 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2407 "btrfs-transaction");
2408 if (IS_ERR(fs_info->transaction_kthread))
2411 if (!btrfs_test_opt(tree_root, SSD) &&
2412 !btrfs_test_opt(tree_root, NOSSD) &&
2413 !fs_info->fs_devices->rotating) {
2414 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2416 btrfs_set_opt(fs_info->mount_opt, SSD);
2419 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2420 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2421 ret = btrfsic_mount(tree_root, fs_devices,
2422 btrfs_test_opt(tree_root,
2423 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2425 fs_info->check_integrity_print_mask);
2427 printk(KERN_WARNING "btrfs: failed to initialize"
2428 " integrity check module %s\n", sb->s_id);
2432 /* do not make disk changes in broken FS */
2433 if (btrfs_super_log_root(disk_super) != 0 &&
2434 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2435 u64 bytenr = btrfs_super_log_root(disk_super);
2437 if (fs_devices->rw_devices == 0) {
2438 printk(KERN_WARNING "Btrfs log replay required "
2441 goto fail_trans_kthread;
2444 btrfs_level_size(tree_root,
2445 btrfs_super_log_root_level(disk_super));
2447 log_tree_root = btrfs_alloc_root(fs_info);
2448 if (!log_tree_root) {
2450 goto fail_trans_kthread;
2453 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2454 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2456 log_tree_root->node = read_tree_block(tree_root, bytenr,
2459 /* returns with log_tree_root freed on success */
2460 ret = btrfs_recover_log_trees(log_tree_root);
2462 btrfs_error(tree_root->fs_info, ret,
2463 "Failed to recover log tree");
2464 free_extent_buffer(log_tree_root->node);
2465 kfree(log_tree_root);
2466 goto fail_trans_kthread;
2469 if (sb->s_flags & MS_RDONLY) {
2470 ret = btrfs_commit_super(tree_root);
2472 goto fail_trans_kthread;
2476 ret = btrfs_find_orphan_roots(tree_root);
2478 goto fail_trans_kthread;
2480 if (!(sb->s_flags & MS_RDONLY)) {
2481 ret = btrfs_cleanup_fs_roots(fs_info);
2485 ret = btrfs_recover_relocation(tree_root);
2488 "btrfs: failed to recover relocation\n");
2490 goto fail_trans_kthread;
2494 location.objectid = BTRFS_FS_TREE_OBJECTID;
2495 location.type = BTRFS_ROOT_ITEM_KEY;
2496 location.offset = (u64)-1;
2498 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2499 if (!fs_info->fs_root)
2500 goto fail_trans_kthread;
2501 if (IS_ERR(fs_info->fs_root)) {
2502 err = PTR_ERR(fs_info->fs_root);
2503 goto fail_trans_kthread;
2506 if (!(sb->s_flags & MS_RDONLY)) {
2507 down_read(&fs_info->cleanup_work_sem);
2508 err = btrfs_orphan_cleanup(fs_info->fs_root);
2510 err = btrfs_orphan_cleanup(fs_info->tree_root);
2511 up_read(&fs_info->cleanup_work_sem);
2514 err = btrfs_recover_balance(fs_info->tree_root);
2517 close_ctree(tree_root);
2525 kthread_stop(fs_info->transaction_kthread);
2527 kthread_stop(fs_info->cleaner_kthread);
2530 * make sure we're done with the btree inode before we stop our
2533 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2534 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2537 btrfs_free_block_groups(fs_info);
2540 free_root_pointers(fs_info, 1);
2543 btrfs_stop_workers(&fs_info->generic_worker);
2544 btrfs_stop_workers(&fs_info->readahead_workers);
2545 btrfs_stop_workers(&fs_info->fixup_workers);
2546 btrfs_stop_workers(&fs_info->delalloc_workers);
2547 btrfs_stop_workers(&fs_info->workers);
2548 btrfs_stop_workers(&fs_info->endio_workers);
2549 btrfs_stop_workers(&fs_info->endio_meta_workers);
2550 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2551 btrfs_stop_workers(&fs_info->endio_write_workers);
2552 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2553 btrfs_stop_workers(&fs_info->submit_workers);
2554 btrfs_stop_workers(&fs_info->delayed_workers);
2555 btrfs_stop_workers(&fs_info->caching_workers);
2558 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2560 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2561 iput(fs_info->btree_inode);
2563 bdi_destroy(&fs_info->bdi);
2565 cleanup_srcu_struct(&fs_info->subvol_srcu);
2567 btrfs_close_devices(fs_info->fs_devices);
2571 if (!btrfs_test_opt(tree_root, RECOVERY))
2572 goto fail_tree_roots;
2574 free_root_pointers(fs_info, 0);
2576 /* don't use the log in recovery mode, it won't be valid */
2577 btrfs_set_super_log_root(disk_super, 0);
2579 /* we can't trust the free space cache either */
2580 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2582 ret = next_root_backup(fs_info, fs_info->super_copy,
2583 &num_backups_tried, &backup_index);
2585 goto fail_block_groups;
2586 goto retry_root_backup;
2589 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2591 char b[BDEVNAME_SIZE];
2594 set_buffer_uptodate(bh);
2596 printk_ratelimited(KERN_WARNING "lost page write due to "
2597 "I/O error on %s\n",
2598 bdevname(bh->b_bdev, b));
2599 /* note, we dont' set_buffer_write_io_error because we have
2600 * our own ways of dealing with the IO errors
2602 clear_buffer_uptodate(bh);
2608 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2610 struct buffer_head *bh;
2611 struct buffer_head *latest = NULL;
2612 struct btrfs_super_block *super;
2617 /* we would like to check all the supers, but that would make
2618 * a btrfs mount succeed after a mkfs from a different FS.
2619 * So, we need to add a special mount option to scan for
2620 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2622 for (i = 0; i < 1; i++) {
2623 bytenr = btrfs_sb_offset(i);
2624 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2626 bh = __bread(bdev, bytenr / 4096, 4096);
2630 super = (struct btrfs_super_block *)bh->b_data;
2631 if (btrfs_super_bytenr(super) != bytenr ||
2632 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2633 sizeof(super->magic))) {
2638 if (!latest || btrfs_super_generation(super) > transid) {
2641 transid = btrfs_super_generation(super);
2650 * this should be called twice, once with wait == 0 and
2651 * once with wait == 1. When wait == 0 is done, all the buffer heads
2652 * we write are pinned.
2654 * They are released when wait == 1 is done.
2655 * max_mirrors must be the same for both runs, and it indicates how
2656 * many supers on this one device should be written.
2658 * max_mirrors == 0 means to write them all.
2660 static int write_dev_supers(struct btrfs_device *device,
2661 struct btrfs_super_block *sb,
2662 int do_barriers, int wait, int max_mirrors)
2664 struct buffer_head *bh;
2671 if (max_mirrors == 0)
2672 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2674 for (i = 0; i < max_mirrors; i++) {
2675 bytenr = btrfs_sb_offset(i);
2676 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2680 bh = __find_get_block(device->bdev, bytenr / 4096,
2681 BTRFS_SUPER_INFO_SIZE);
2684 if (!buffer_uptodate(bh))
2687 /* drop our reference */
2690 /* drop the reference from the wait == 0 run */
2694 btrfs_set_super_bytenr(sb, bytenr);
2697 crc = btrfs_csum_data(NULL, (char *)sb +
2698 BTRFS_CSUM_SIZE, crc,
2699 BTRFS_SUPER_INFO_SIZE -
2701 btrfs_csum_final(crc, sb->csum);
2704 * one reference for us, and we leave it for the
2707 bh = __getblk(device->bdev, bytenr / 4096,
2708 BTRFS_SUPER_INFO_SIZE);
2709 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2711 /* one reference for submit_bh */
2714 set_buffer_uptodate(bh);
2716 bh->b_end_io = btrfs_end_buffer_write_sync;
2720 * we fua the first super. The others we allow
2723 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2727 return errors < i ? 0 : -1;
2731 * endio for the write_dev_flush, this will wake anyone waiting
2732 * for the barrier when it is done
2734 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2737 if (err == -EOPNOTSUPP)
2738 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2739 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2741 if (bio->bi_private)
2742 complete(bio->bi_private);
2747 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2748 * sent down. With wait == 1, it waits for the previous flush.
2750 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2753 static int write_dev_flush(struct btrfs_device *device, int wait)
2758 if (device->nobarriers)
2762 bio = device->flush_bio;
2766 wait_for_completion(&device->flush_wait);
2768 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2769 printk("btrfs: disabling barriers on dev %s\n",
2771 device->nobarriers = 1;
2773 if (!bio_flagged(bio, BIO_UPTODATE)) {
2777 /* drop the reference from the wait == 0 run */
2779 device->flush_bio = NULL;
2785 * one reference for us, and we leave it for the
2788 device->flush_bio = NULL;;
2789 bio = bio_alloc(GFP_NOFS, 0);
2793 bio->bi_end_io = btrfs_end_empty_barrier;
2794 bio->bi_bdev = device->bdev;
2795 init_completion(&device->flush_wait);
2796 bio->bi_private = &device->flush_wait;
2797 device->flush_bio = bio;
2800 btrfsic_submit_bio(WRITE_FLUSH, bio);
2806 * send an empty flush down to each device in parallel,
2807 * then wait for them
2809 static int barrier_all_devices(struct btrfs_fs_info *info)
2811 struct list_head *head;
2812 struct btrfs_device *dev;
2816 /* send down all the barriers */
2817 head = &info->fs_devices->devices;
2818 list_for_each_entry_rcu(dev, head, dev_list) {
2823 if (!dev->in_fs_metadata || !dev->writeable)
2826 ret = write_dev_flush(dev, 0);
2831 /* wait for all the barriers */
2832 list_for_each_entry_rcu(dev, head, dev_list) {
2837 if (!dev->in_fs_metadata || !dev->writeable)
2840 ret = write_dev_flush(dev, 1);
2849 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2851 struct list_head *head;
2852 struct btrfs_device *dev;
2853 struct btrfs_super_block *sb;
2854 struct btrfs_dev_item *dev_item;
2858 int total_errors = 0;
2861 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2862 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2863 backup_super_roots(root->fs_info);
2865 sb = root->fs_info->super_for_commit;
2866 dev_item = &sb->dev_item;
2868 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2869 head = &root->fs_info->fs_devices->devices;
2872 barrier_all_devices(root->fs_info);
2874 list_for_each_entry_rcu(dev, head, dev_list) {
2879 if (!dev->in_fs_metadata || !dev->writeable)
2882 btrfs_set_stack_device_generation(dev_item, 0);
2883 btrfs_set_stack_device_type(dev_item, dev->type);
2884 btrfs_set_stack_device_id(dev_item, dev->devid);
2885 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2886 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2887 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2888 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2889 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2890 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2891 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2893 flags = btrfs_super_flags(sb);
2894 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2896 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2900 if (total_errors > max_errors) {
2901 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2904 /* This shouldn't happen. FUA is masked off if unsupported */
2909 list_for_each_entry_rcu(dev, head, dev_list) {
2912 if (!dev->in_fs_metadata || !dev->writeable)
2915 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2919 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2920 if (total_errors > max_errors) {
2921 btrfs_error(root->fs_info, -EIO,
2922 "%d errors while writing supers", total_errors);
2928 int write_ctree_super(struct btrfs_trans_handle *trans,
2929 struct btrfs_root *root, int max_mirrors)
2933 ret = write_all_supers(root, max_mirrors);
2937 /* Kill all outstanding I/O */
2938 void btrfs_abort_devices(struct btrfs_root *root)
2940 struct list_head *head;
2941 struct btrfs_device *dev;
2942 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2943 head = &root->fs_info->fs_devices->devices;
2944 list_for_each_entry_rcu(dev, head, dev_list) {
2945 blk_abort_queue(dev->bdev->bd_disk->queue);
2947 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2950 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2952 spin_lock(&fs_info->fs_roots_radix_lock);
2953 radix_tree_delete(&fs_info->fs_roots_radix,
2954 (unsigned long)root->root_key.objectid);
2955 spin_unlock(&fs_info->fs_roots_radix_lock);
2957 if (btrfs_root_refs(&root->root_item) == 0)
2958 synchronize_srcu(&fs_info->subvol_srcu);
2960 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2961 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2965 static void free_fs_root(struct btrfs_root *root)
2967 iput(root->cache_inode);
2968 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2970 free_anon_bdev(root->anon_dev);
2971 free_extent_buffer(root->node);
2972 free_extent_buffer(root->commit_root);
2973 kfree(root->free_ino_ctl);
2974 kfree(root->free_ino_pinned);
2979 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2982 struct btrfs_root *gang[8];
2985 while (!list_empty(&fs_info->dead_roots)) {
2986 gang[0] = list_entry(fs_info->dead_roots.next,
2987 struct btrfs_root, root_list);
2988 list_del(&gang[0]->root_list);
2990 if (gang[0]->in_radix) {
2991 btrfs_free_fs_root(fs_info, gang[0]);
2993 free_extent_buffer(gang[0]->node);
2994 free_extent_buffer(gang[0]->commit_root);
3000 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3005 for (i = 0; i < ret; i++)
3006 btrfs_free_fs_root(fs_info, gang[i]);
3010 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3012 u64 root_objectid = 0;
3013 struct btrfs_root *gang[8];
3018 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3019 (void **)gang, root_objectid,
3024 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3025 for (i = 0; i < ret; i++) {
3028 root_objectid = gang[i]->root_key.objectid;
3029 err = btrfs_orphan_cleanup(gang[i]);
3038 int btrfs_commit_super(struct btrfs_root *root)
3040 struct btrfs_trans_handle *trans;
3043 mutex_lock(&root->fs_info->cleaner_mutex);
3044 btrfs_run_delayed_iputs(root);
3045 btrfs_clean_old_snapshots(root);
3046 mutex_unlock(&root->fs_info->cleaner_mutex);
3048 /* wait until ongoing cleanup work done */
3049 down_write(&root->fs_info->cleanup_work_sem);
3050 up_write(&root->fs_info->cleanup_work_sem);
3052 trans = btrfs_join_transaction(root);
3054 return PTR_ERR(trans);
3055 ret = btrfs_commit_transaction(trans, root);
3058 /* run commit again to drop the original snapshot */
3059 trans = btrfs_join_transaction(root);
3061 return PTR_ERR(trans);
3062 ret = btrfs_commit_transaction(trans, root);
3065 ret = btrfs_write_and_wait_transaction(NULL, root);
3067 btrfs_error(root->fs_info, ret,
3068 "Failed to sync btree inode to disk.");
3072 ret = write_ctree_super(NULL, root, 0);
3076 int close_ctree(struct btrfs_root *root)
3078 struct btrfs_fs_info *fs_info = root->fs_info;
3081 fs_info->closing = 1;
3084 /* pause restriper - we want to resume on mount */
3085 btrfs_pause_balance(root->fs_info);
3087 btrfs_scrub_cancel(root);
3089 /* wait for any defraggers to finish */
3090 wait_event(fs_info->transaction_wait,
3091 (atomic_read(&fs_info->defrag_running) == 0));
3093 /* clear out the rbtree of defraggable inodes */
3094 btrfs_run_defrag_inodes(fs_info);
3097 * Here come 2 situations when btrfs is broken to flip readonly:
3099 * 1. when btrfs flips readonly somewhere else before
3100 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3101 * and btrfs will skip to write sb directly to keep
3102 * ERROR state on disk.
3104 * 2. when btrfs flips readonly just in btrfs_commit_super,
3105 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3106 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3107 * btrfs will cleanup all FS resources first and write sb then.
3109 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3110 ret = btrfs_commit_super(root);
3112 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3115 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3116 ret = btrfs_error_commit_super(root);
3118 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3121 btrfs_put_block_group_cache(fs_info);
3123 kthread_stop(fs_info->transaction_kthread);
3124 kthread_stop(fs_info->cleaner_kthread);
3126 fs_info->closing = 2;
3129 if (fs_info->delalloc_bytes) {
3130 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3131 (unsigned long long)fs_info->delalloc_bytes);
3133 if (fs_info->total_ref_cache_size) {
3134 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3135 (unsigned long long)fs_info->total_ref_cache_size);
3138 free_extent_buffer(fs_info->extent_root->node);
3139 free_extent_buffer(fs_info->extent_root->commit_root);
3140 free_extent_buffer(fs_info->tree_root->node);
3141 free_extent_buffer(fs_info->tree_root->commit_root);
3142 free_extent_buffer(fs_info->chunk_root->node);
3143 free_extent_buffer(fs_info->chunk_root->commit_root);
3144 free_extent_buffer(fs_info->dev_root->node);
3145 free_extent_buffer(fs_info->dev_root->commit_root);
3146 free_extent_buffer(fs_info->csum_root->node);
3147 free_extent_buffer(fs_info->csum_root->commit_root);
3149 btrfs_free_block_groups(fs_info);
3151 del_fs_roots(fs_info);
3153 iput(fs_info->btree_inode);
3155 btrfs_stop_workers(&fs_info->generic_worker);
3156 btrfs_stop_workers(&fs_info->fixup_workers);
3157 btrfs_stop_workers(&fs_info->delalloc_workers);
3158 btrfs_stop_workers(&fs_info->workers);
3159 btrfs_stop_workers(&fs_info->endio_workers);
3160 btrfs_stop_workers(&fs_info->endio_meta_workers);
3161 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3162 btrfs_stop_workers(&fs_info->endio_write_workers);
3163 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3164 btrfs_stop_workers(&fs_info->submit_workers);
3165 btrfs_stop_workers(&fs_info->delayed_workers);
3166 btrfs_stop_workers(&fs_info->caching_workers);
3167 btrfs_stop_workers(&fs_info->readahead_workers);
3169 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3170 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3171 btrfsic_unmount(root, fs_info->fs_devices);
3174 btrfs_close_devices(fs_info->fs_devices);
3175 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3177 bdi_destroy(&fs_info->bdi);
3178 cleanup_srcu_struct(&fs_info->subvol_srcu);
3183 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3186 struct inode *btree_inode = buf->first_page->mapping->host;
3188 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3193 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3198 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3200 struct inode *btree_inode = buf->first_page->mapping->host;
3201 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3205 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3207 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3208 u64 transid = btrfs_header_generation(buf);
3209 struct inode *btree_inode = root->fs_info->btree_inode;
3212 btrfs_assert_tree_locked(buf);
3213 if (transid != root->fs_info->generation) {
3214 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3215 "found %llu running %llu\n",
3216 (unsigned long long)buf->start,
3217 (unsigned long long)transid,
3218 (unsigned long long)root->fs_info->generation);
3221 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3224 spin_lock(&root->fs_info->delalloc_lock);
3225 root->fs_info->dirty_metadata_bytes += buf->len;
3226 spin_unlock(&root->fs_info->delalloc_lock);
3230 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3233 * looks as though older kernels can get into trouble with
3234 * this code, they end up stuck in balance_dirty_pages forever
3237 unsigned long thresh = 32 * 1024 * 1024;
3239 if (current->flags & PF_MEMALLOC)
3242 btrfs_balance_delayed_items(root);
3244 num_dirty = root->fs_info->dirty_metadata_bytes;
3246 if (num_dirty > thresh) {
3247 balance_dirty_pages_ratelimited_nr(
3248 root->fs_info->btree_inode->i_mapping, 1);
3253 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3256 * looks as though older kernels can get into trouble with
3257 * this code, they end up stuck in balance_dirty_pages forever
3260 unsigned long thresh = 32 * 1024 * 1024;
3262 if (current->flags & PF_MEMALLOC)
3265 num_dirty = root->fs_info->dirty_metadata_bytes;
3267 if (num_dirty > thresh) {
3268 balance_dirty_pages_ratelimited_nr(
3269 root->fs_info->btree_inode->i_mapping, 1);
3274 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3276 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3278 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3280 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3284 static int btree_lock_page_hook(struct page *page, void *data,
3285 void (*flush_fn)(void *))
3287 struct inode *inode = page->mapping->host;
3288 struct btrfs_root *root = BTRFS_I(inode)->root;
3289 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3290 struct extent_buffer *eb;
3292 u64 bytenr = page_offset(page);
3294 if (page->private == EXTENT_PAGE_PRIVATE)
3297 len = page->private >> 2;
3298 eb = find_extent_buffer(io_tree, bytenr, len);
3302 if (!btrfs_try_tree_write_lock(eb)) {
3304 btrfs_tree_lock(eb);
3306 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3308 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3309 spin_lock(&root->fs_info->delalloc_lock);
3310 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3311 root->fs_info->dirty_metadata_bytes -= eb->len;
3314 spin_unlock(&root->fs_info->delalloc_lock);
3317 btrfs_tree_unlock(eb);
3318 free_extent_buffer(eb);
3320 if (!trylock_page(page)) {
3327 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3333 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3334 printk(KERN_WARNING "warning: mount fs with errors, "
3335 "running btrfsck is recommended\n");
3338 int btrfs_error_commit_super(struct btrfs_root *root)
3342 mutex_lock(&root->fs_info->cleaner_mutex);
3343 btrfs_run_delayed_iputs(root);
3344 mutex_unlock(&root->fs_info->cleaner_mutex);
3346 down_write(&root->fs_info->cleanup_work_sem);
3347 up_write(&root->fs_info->cleanup_work_sem);
3349 /* cleanup FS via transaction */
3350 btrfs_cleanup_transaction(root);
3352 ret = write_ctree_super(NULL, root, 0);
3357 static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3359 struct btrfs_inode *btrfs_inode;
3360 struct list_head splice;
3362 INIT_LIST_HEAD(&splice);
3364 mutex_lock(&root->fs_info->ordered_operations_mutex);
3365 spin_lock(&root->fs_info->ordered_extent_lock);
3367 list_splice_init(&root->fs_info->ordered_operations, &splice);
3368 while (!list_empty(&splice)) {
3369 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3370 ordered_operations);
3372 list_del_init(&btrfs_inode->ordered_operations);
3374 btrfs_invalidate_inodes(btrfs_inode->root);
3377 spin_unlock(&root->fs_info->ordered_extent_lock);
3378 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3381 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3383 struct list_head splice;
3384 struct btrfs_ordered_extent *ordered;
3385 struct inode *inode;
3387 INIT_LIST_HEAD(&splice);
3389 spin_lock(&root->fs_info->ordered_extent_lock);
3391 list_splice_init(&root->fs_info->ordered_extents, &splice);
3392 while (!list_empty(&splice)) {
3393 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3396 list_del_init(&ordered->root_extent_list);
3397 atomic_inc(&ordered->refs);
3399 /* the inode may be getting freed (in sys_unlink path). */
3400 inode = igrab(ordered->inode);
3402 spin_unlock(&root->fs_info->ordered_extent_lock);
3406 atomic_set(&ordered->refs, 1);
3407 btrfs_put_ordered_extent(ordered);
3409 spin_lock(&root->fs_info->ordered_extent_lock);
3412 spin_unlock(&root->fs_info->ordered_extent_lock);
3415 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3416 struct btrfs_root *root)
3418 struct rb_node *node;
3419 struct btrfs_delayed_ref_root *delayed_refs;
3420 struct btrfs_delayed_ref_node *ref;
3423 delayed_refs = &trans->delayed_refs;
3426 spin_lock(&delayed_refs->lock);
3427 if (delayed_refs->num_entries == 0) {
3428 spin_unlock(&delayed_refs->lock);
3429 printk(KERN_INFO "delayed_refs has NO entry\n");
3433 node = rb_first(&delayed_refs->root);
3435 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3436 node = rb_next(node);
3439 rb_erase(&ref->rb_node, &delayed_refs->root);
3440 delayed_refs->num_entries--;
3442 atomic_set(&ref->refs, 1);
3443 if (btrfs_delayed_ref_is_head(ref)) {
3444 struct btrfs_delayed_ref_head *head;
3446 head = btrfs_delayed_node_to_head(ref);
3447 spin_unlock(&delayed_refs->lock);
3448 mutex_lock(&head->mutex);
3449 kfree(head->extent_op);
3450 delayed_refs->num_heads--;
3451 if (list_empty(&head->cluster))
3452 delayed_refs->num_heads_ready--;
3453 list_del_init(&head->cluster);
3454 mutex_unlock(&head->mutex);
3455 btrfs_put_delayed_ref(ref);
3458 spin_unlock(&delayed_refs->lock);
3459 btrfs_put_delayed_ref(ref);
3462 spin_lock(&delayed_refs->lock);
3465 spin_unlock(&delayed_refs->lock);
3470 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3472 struct btrfs_pending_snapshot *snapshot;
3473 struct list_head splice;
3475 INIT_LIST_HEAD(&splice);
3477 list_splice_init(&t->pending_snapshots, &splice);
3479 while (!list_empty(&splice)) {
3480 snapshot = list_entry(splice.next,
3481 struct btrfs_pending_snapshot,
3484 list_del_init(&snapshot->list);
3490 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3492 struct btrfs_inode *btrfs_inode;
3493 struct list_head splice;
3495 INIT_LIST_HEAD(&splice);
3497 spin_lock(&root->fs_info->delalloc_lock);
3498 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3500 while (!list_empty(&splice)) {
3501 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3504 list_del_init(&btrfs_inode->delalloc_inodes);
3506 btrfs_invalidate_inodes(btrfs_inode->root);
3509 spin_unlock(&root->fs_info->delalloc_lock);
3512 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3513 struct extent_io_tree *dirty_pages,
3518 struct inode *btree_inode = root->fs_info->btree_inode;
3519 struct extent_buffer *eb;
3523 unsigned long index;
3526 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3531 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3532 while (start <= end) {
3533 index = start >> PAGE_CACHE_SHIFT;
3534 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3535 page = find_get_page(btree_inode->i_mapping, index);
3538 offset = page_offset(page);
3540 spin_lock(&dirty_pages->buffer_lock);
3541 eb = radix_tree_lookup(
3542 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3543 offset >> PAGE_CACHE_SHIFT);
3544 spin_unlock(&dirty_pages->buffer_lock);
3546 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3548 atomic_set(&eb->refs, 1);
3550 if (PageWriteback(page))
3551 end_page_writeback(page);
3554 if (PageDirty(page)) {
3555 clear_page_dirty_for_io(page);
3556 spin_lock_irq(&page->mapping->tree_lock);
3557 radix_tree_tag_clear(&page->mapping->page_tree,
3559 PAGECACHE_TAG_DIRTY);
3560 spin_unlock_irq(&page->mapping->tree_lock);
3563 page->mapping->a_ops->invalidatepage(page, 0);
3571 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3572 struct extent_io_tree *pinned_extents)
3574 struct extent_io_tree *unpin;
3579 unpin = pinned_extents;
3581 ret = find_first_extent_bit(unpin, 0, &start, &end,
3587 if (btrfs_test_opt(root, DISCARD))
3588 ret = btrfs_error_discard_extent(root, start,
3592 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3593 btrfs_error_unpin_extent_range(root, start, end);
3600 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3601 struct btrfs_root *root)
3603 btrfs_destroy_delayed_refs(cur_trans, root);
3604 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3605 cur_trans->dirty_pages.dirty_bytes);
3607 /* FIXME: cleanup wait for commit */
3608 cur_trans->in_commit = 1;
3609 cur_trans->blocked = 1;
3610 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3611 wake_up(&root->fs_info->transaction_blocked_wait);
3613 cur_trans->blocked = 0;
3614 if (waitqueue_active(&root->fs_info->transaction_wait))
3615 wake_up(&root->fs_info->transaction_wait);
3617 cur_trans->commit_done = 1;
3618 if (waitqueue_active(&cur_trans->commit_wait))
3619 wake_up(&cur_trans->commit_wait);
3621 btrfs_destroy_pending_snapshots(cur_trans);
3623 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3627 memset(cur_trans, 0, sizeof(*cur_trans));
3628 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3632 int btrfs_cleanup_transaction(struct btrfs_root *root)
3634 struct btrfs_transaction *t;
3637 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3639 spin_lock(&root->fs_info->trans_lock);
3640 list_splice_init(&root->fs_info->trans_list, &list);
3641 root->fs_info->trans_no_join = 1;
3642 spin_unlock(&root->fs_info->trans_lock);
3644 while (!list_empty(&list)) {
3645 t = list_entry(list.next, struct btrfs_transaction, list);
3649 btrfs_destroy_ordered_operations(root);
3651 btrfs_destroy_ordered_extents(root);
3653 btrfs_destroy_delayed_refs(t, root);
3655 btrfs_block_rsv_release(root,
3656 &root->fs_info->trans_block_rsv,
3657 t->dirty_pages.dirty_bytes);
3659 /* FIXME: cleanup wait for commit */
3662 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3663 wake_up(&root->fs_info->transaction_blocked_wait);
3666 if (waitqueue_active(&root->fs_info->transaction_wait))
3667 wake_up(&root->fs_info->transaction_wait);
3670 if (waitqueue_active(&t->commit_wait))
3671 wake_up(&t->commit_wait);
3673 btrfs_destroy_pending_snapshots(t);
3675 btrfs_destroy_delalloc_inodes(root);
3677 spin_lock(&root->fs_info->trans_lock);
3678 root->fs_info->running_transaction = NULL;
3679 spin_unlock(&root->fs_info->trans_lock);
3681 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3684 btrfs_destroy_pinned_extent(root,
3685 root->fs_info->pinned_extents);
3687 atomic_set(&t->use_count, 0);
3688 list_del_init(&t->list);
3689 memset(t, 0, sizeof(*t));
3690 kmem_cache_free(btrfs_transaction_cachep, t);
3693 spin_lock(&root->fs_info->trans_lock);
3694 root->fs_info->trans_no_join = 0;
3695 spin_unlock(&root->fs_info->trans_lock);
3696 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3701 static int btree_writepage_io_failed_hook(struct bio *bio, struct page *page,
3703 struct extent_state *state)
3705 struct super_block *sb = page->mapping->host->i_sb;
3706 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3707 btrfs_error(fs_info, -EIO,
3708 "Error occured while writing out btree at %llu", start);
3712 static struct extent_io_ops btree_extent_io_ops = {
3713 .write_cache_pages_lock_hook = btree_lock_page_hook,
3714 .readpage_end_io_hook = btree_readpage_end_io_hook,
3715 .readpage_io_failed_hook = btree_io_failed_hook,
3716 .submit_bio_hook = btree_submit_bio_hook,
3717 /* note we're sharing with inode.c for the merge bio hook */
3718 .merge_bio_hook = btrfs_merge_bio_hook,
3719 .writepage_io_failed_hook = btree_writepage_io_failed_hook,