2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
49 static struct extent_io_ops btree_extent_io_ops;
50 static void end_workqueue_fn(struct btrfs_work *work);
51 static void free_fs_root(struct btrfs_root *root);
52 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
54 static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_root *root);
58 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
61 struct extent_io_tree *dirty_pages,
63 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
64 struct extent_io_tree *pinned_extents);
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
75 struct btrfs_fs_info *info;
78 struct list_head list;
79 struct btrfs_work work;
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
87 struct async_submit_bio {
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
95 unsigned long bio_flags;
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
101 struct btrfs_work work;
106 * Lockdep class keys for extent_buffer->lock's in this root. For a given
107 * eb, the lockdep key is determined by the btrfs_root it belongs to and
108 * the level the eb occupies in the tree.
110 * Different roots are used for different purposes and may nest inside each
111 * other and they require separate keysets. As lockdep keys should be
112 * static, assign keysets according to the purpose of the root as indicated
113 * by btrfs_root->objectid. This ensures that all special purpose roots
114 * have separate keysets.
116 * Lock-nesting across peer nodes is always done with the immediate parent
117 * node locked thus preventing deadlock. As lockdep doesn't know this, use
118 * subclass to avoid triggering lockdep warning in such cases.
120 * The key is set by the readpage_end_io_hook after the buffer has passed
121 * csum validation but before the pages are unlocked. It is also set by
122 * btrfs_init_new_buffer on freshly allocated blocks.
124 * We also add a check to make sure the highest level of the tree is the
125 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
126 * needs update as well.
128 #ifdef CONFIG_DEBUG_LOCK_ALLOC
129 # if BTRFS_MAX_LEVEL != 8
133 static struct btrfs_lockdep_keyset {
134 u64 id; /* root objectid */
135 const char *name_stem; /* lock name stem */
136 char names[BTRFS_MAX_LEVEL + 1][20];
137 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
138 } btrfs_lockdep_keysets[] = {
139 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
140 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
141 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
142 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
143 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
144 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
145 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
146 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
147 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
148 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
149 { .id = 0, .name_stem = "tree" },
152 void __init btrfs_init_lockdep(void)
156 /* initialize lockdep class names */
157 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
158 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
160 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
161 snprintf(ks->names[j], sizeof(ks->names[j]),
162 "btrfs-%s-%02d", ks->name_stem, j);
166 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
169 struct btrfs_lockdep_keyset *ks;
171 BUG_ON(level >= ARRAY_SIZE(ks->keys));
173 /* find the matching keyset, id 0 is the default entry */
174 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
175 if (ks->id == objectid)
178 lockdep_set_class_and_name(&eb->lock,
179 &ks->keys[level], ks->names[level]);
185 * extents on the btree inode are pretty simple, there's one extent
186 * that covers the entire device
188 static struct extent_map *btree_get_extent(struct inode *inode,
189 struct page *page, size_t pg_offset, u64 start, u64 len,
192 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
193 struct extent_map *em;
196 read_lock(&em_tree->lock);
197 em = lookup_extent_mapping(em_tree, start, len);
200 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
201 read_unlock(&em_tree->lock);
204 read_unlock(&em_tree->lock);
206 em = alloc_extent_map();
208 em = ERR_PTR(-ENOMEM);
213 em->block_len = (u64)-1;
215 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
217 write_lock(&em_tree->lock);
218 ret = add_extent_mapping(em_tree, em);
219 if (ret == -EEXIST) {
220 u64 failed_start = em->start;
221 u64 failed_len = em->len;
224 em = lookup_extent_mapping(em_tree, start, len);
228 em = lookup_extent_mapping(em_tree, failed_start,
236 write_unlock(&em_tree->lock);
244 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
246 return crc32c(seed, data, len);
249 void btrfs_csum_final(u32 crc, char *result)
251 put_unaligned_le32(~crc, result);
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
258 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
267 unsigned long map_start;
268 unsigned long map_len;
271 unsigned long inline_result;
273 len = buf->len - offset;
275 err = map_private_extent_buffer(buf, offset, 32,
276 &kaddr, &map_start, &map_len);
279 cur_len = min(len, map_len - (offset - map_start));
280 crc = btrfs_csum_data(root, kaddr + offset - map_start,
285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
290 result = (char *)&inline_result;
293 btrfs_csum_final(crc, result);
296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
299 memcpy(&found, result, csum_size);
301 read_extent_buffer(buf, &val, 0, csum_size);
302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303 "failed on %llu wanted %X found %X "
305 root->fs_info->sb->s_id,
306 (unsigned long long)buf->start, val, found,
307 btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
313 write_extent_buffer(buf, result, 0, csum_size);
315 if (result != (char *)&inline_result)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
330 struct extent_state *cached_state = NULL;
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 (unsigned long long)eb->start,
349 (unsigned long long)parent_transid,
350 (unsigned long long)btrfs_header_generation(eb));
352 clear_extent_buffer_uptodate(eb);
354 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 &cached_state, GFP_NOFS);
360 * helper to read a given tree block, doing retries as required when
361 * the checksums don't match and we have alternate mirrors to try.
363 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
364 struct extent_buffer *eb,
365 u64 start, u64 parent_transid)
367 struct extent_io_tree *io_tree;
372 int failed_mirror = 0;
374 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
375 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
377 ret = read_extent_buffer_pages(io_tree, eb, start,
379 btree_get_extent, mirror_num);
380 if (!ret && !verify_parent_transid(io_tree, eb,
385 * This buffer's crc is fine, but its contents are corrupted, so
386 * there is no reason to read the other copies, they won't be
389 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
392 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
397 if (!failed_mirror) {
399 failed_mirror = eb->read_mirror;
403 if (mirror_num == failed_mirror)
406 if (mirror_num > num_copies)
411 repair_eb_io_failure(root, eb, failed_mirror);
417 * checksum a dirty tree block before IO. This has extra checks to make sure
418 * we only fill in the checksum field in the first page of a multi-page block
421 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
423 struct extent_io_tree *tree;
424 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
426 struct extent_buffer *eb;
428 tree = &BTRFS_I(page->mapping->host)->io_tree;
430 eb = (struct extent_buffer *)page->private;
431 if (page != eb->pages[0])
433 found_start = btrfs_header_bytenr(eb);
434 if (found_start != start) {
438 if (eb->pages[0] != page) {
442 if (!PageUptodate(page)) {
446 csum_tree_block(root, eb, 0);
450 static int check_tree_block_fsid(struct btrfs_root *root,
451 struct extent_buffer *eb)
453 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
454 u8 fsid[BTRFS_UUID_SIZE];
457 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
460 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
464 fs_devices = fs_devices->seed;
469 #define CORRUPT(reason, eb, root, slot) \
470 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
471 "root=%llu, slot=%d\n", reason, \
472 (unsigned long long)btrfs_header_bytenr(eb), \
473 (unsigned long long)root->objectid, slot)
475 static noinline int check_leaf(struct btrfs_root *root,
476 struct extent_buffer *leaf)
478 struct btrfs_key key;
479 struct btrfs_key leaf_key;
480 u32 nritems = btrfs_header_nritems(leaf);
486 /* Check the 0 item */
487 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
488 BTRFS_LEAF_DATA_SIZE(root)) {
489 CORRUPT("invalid item offset size pair", leaf, root, 0);
494 * Check to make sure each items keys are in the correct order and their
495 * offsets make sense. We only have to loop through nritems-1 because
496 * we check the current slot against the next slot, which verifies the
497 * next slot's offset+size makes sense and that the current's slot
500 for (slot = 0; slot < nritems - 1; slot++) {
501 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
502 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
504 /* Make sure the keys are in the right order */
505 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
506 CORRUPT("bad key order", leaf, root, slot);
511 * Make sure the offset and ends are right, remember that the
512 * item data starts at the end of the leaf and grows towards the
515 if (btrfs_item_offset_nr(leaf, slot) !=
516 btrfs_item_end_nr(leaf, slot + 1)) {
517 CORRUPT("slot offset bad", leaf, root, slot);
522 * Check to make sure that we don't point outside of the leaf,
523 * just incase all the items are consistent to eachother, but
524 * all point outside of the leaf.
526 if (btrfs_item_end_nr(leaf, slot) >
527 BTRFS_LEAF_DATA_SIZE(root)) {
528 CORRUPT("slot end outside of leaf", leaf, root, slot);
536 struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
537 struct page *page, int max_walk)
539 struct extent_buffer *eb;
540 u64 start = page_offset(page);
544 if (start < max_walk)
547 min_start = start - max_walk;
549 while (start >= min_start) {
550 eb = find_extent_buffer(tree, start, 0);
553 * we found an extent buffer and it contains our page
556 if (eb->start <= target &&
557 eb->start + eb->len > target)
560 /* we found an extent buffer that wasn't for us */
561 free_extent_buffer(eb);
566 start -= PAGE_CACHE_SIZE;
571 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
572 struct extent_state *state, int mirror)
574 struct extent_io_tree *tree;
577 struct extent_buffer *eb;
578 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
585 tree = &BTRFS_I(page->mapping->host)->io_tree;
586 eb = (struct extent_buffer *)page->private;
588 /* the pending IO might have been the only thing that kept this buffer
589 * in memory. Make sure we have a ref for all this other checks
591 extent_buffer_get(eb);
593 reads_done = atomic_dec_and_test(&eb->io_pages);
597 eb->read_mirror = mirror;
598 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
603 found_start = btrfs_header_bytenr(eb);
604 if (found_start != eb->start) {
605 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
607 (unsigned long long)found_start,
608 (unsigned long long)eb->start);
612 if (check_tree_block_fsid(root, eb)) {
613 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
614 (unsigned long long)eb->start);
618 found_level = btrfs_header_level(eb);
620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
623 ret = csum_tree_block(root, eb, 1);
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
634 if (found_level == 0 && check_leaf(root, eb)) {
635 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
640 set_extent_buffer_uptodate(eb);
642 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
643 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
644 btree_readahead_hook(root, eb, eb->start, ret);
648 clear_extent_buffer_uptodate(eb);
649 free_extent_buffer(eb);
654 static int btree_io_failed_hook(struct page *page, int failed_mirror)
656 struct extent_buffer *eb;
657 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
659 eb = (struct extent_buffer *)page->private;
660 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
661 eb->read_mirror = failed_mirror;
662 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
663 btree_readahead_hook(root, eb, eb->start, -EIO);
664 return -EIO; /* we fixed nothing */
667 static void end_workqueue_bio(struct bio *bio, int err)
669 struct end_io_wq *end_io_wq = bio->bi_private;
670 struct btrfs_fs_info *fs_info;
672 fs_info = end_io_wq->info;
673 end_io_wq->error = err;
674 end_io_wq->work.func = end_workqueue_fn;
675 end_io_wq->work.flags = 0;
677 if (bio->bi_rw & REQ_WRITE) {
678 if (end_io_wq->metadata == 1)
679 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
681 else if (end_io_wq->metadata == 2)
682 btrfs_queue_worker(&fs_info->endio_freespace_worker,
685 btrfs_queue_worker(&fs_info->endio_write_workers,
688 if (end_io_wq->metadata)
689 btrfs_queue_worker(&fs_info->endio_meta_workers,
692 btrfs_queue_worker(&fs_info->endio_workers,
698 * For the metadata arg you want
701 * 1 - if normal metadta
702 * 2 - if writing to the free space cache area
704 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
707 struct end_io_wq *end_io_wq;
708 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
712 end_io_wq->private = bio->bi_private;
713 end_io_wq->end_io = bio->bi_end_io;
714 end_io_wq->info = info;
715 end_io_wq->error = 0;
716 end_io_wq->bio = bio;
717 end_io_wq->metadata = metadata;
719 bio->bi_private = end_io_wq;
720 bio->bi_end_io = end_workqueue_bio;
724 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
726 unsigned long limit = min_t(unsigned long,
727 info->workers.max_workers,
728 info->fs_devices->open_devices);
732 static void run_one_async_start(struct btrfs_work *work)
734 struct async_submit_bio *async;
737 async = container_of(work, struct async_submit_bio, work);
738 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
739 async->mirror_num, async->bio_flags,
745 static void run_one_async_done(struct btrfs_work *work)
747 struct btrfs_fs_info *fs_info;
748 struct async_submit_bio *async;
751 async = container_of(work, struct async_submit_bio, work);
752 fs_info = BTRFS_I(async->inode)->root->fs_info;
754 limit = btrfs_async_submit_limit(fs_info);
755 limit = limit * 2 / 3;
757 atomic_dec(&fs_info->nr_async_submits);
759 if (atomic_read(&fs_info->nr_async_submits) < limit &&
760 waitqueue_active(&fs_info->async_submit_wait))
761 wake_up(&fs_info->async_submit_wait);
763 /* If an error occured we just want to clean up the bio and move on */
765 bio_endio(async->bio, async->error);
769 async->submit_bio_done(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
774 static void run_one_async_free(struct btrfs_work *work)
776 struct async_submit_bio *async;
778 async = container_of(work, struct async_submit_bio, work);
782 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
783 int rw, struct bio *bio, int mirror_num,
784 unsigned long bio_flags,
786 extent_submit_bio_hook_t *submit_bio_start,
787 extent_submit_bio_hook_t *submit_bio_done)
789 struct async_submit_bio *async;
791 async = kmalloc(sizeof(*async), GFP_NOFS);
795 async->inode = inode;
798 async->mirror_num = mirror_num;
799 async->submit_bio_start = submit_bio_start;
800 async->submit_bio_done = submit_bio_done;
802 async->work.func = run_one_async_start;
803 async->work.ordered_func = run_one_async_done;
804 async->work.ordered_free = run_one_async_free;
806 async->work.flags = 0;
807 async->bio_flags = bio_flags;
808 async->bio_offset = bio_offset;
812 atomic_inc(&fs_info->nr_async_submits);
815 btrfs_set_work_high_prio(&async->work);
817 btrfs_queue_worker(&fs_info->workers, &async->work);
819 while (atomic_read(&fs_info->async_submit_draining) &&
820 atomic_read(&fs_info->nr_async_submits)) {
821 wait_event(fs_info->async_submit_wait,
822 (atomic_read(&fs_info->nr_async_submits) == 0));
828 static int btree_csum_one_bio(struct bio *bio)
830 struct bio_vec *bvec = bio->bi_io_vec;
832 struct btrfs_root *root;
835 WARN_ON(bio->bi_vcnt <= 0);
836 while (bio_index < bio->bi_vcnt) {
837 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
838 ret = csum_dirty_buffer(root, bvec->bv_page);
847 static int __btree_submit_bio_start(struct inode *inode, int rw,
848 struct bio *bio, int mirror_num,
849 unsigned long bio_flags,
853 * when we're called for a write, we're already in the async
854 * submission context. Just jump into btrfs_map_bio
856 return btree_csum_one_bio(bio);
859 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
860 int mirror_num, unsigned long bio_flags,
864 * when we're called for a write, we're already in the async
865 * submission context. Just jump into btrfs_map_bio
867 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
870 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
871 int mirror_num, unsigned long bio_flags,
876 if (!(rw & REQ_WRITE)) {
879 * called for a read, do the setup so that checksum validation
880 * can happen in the async kernel threads
882 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
886 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
891 * kthread helpers are used to submit writes so that checksumming
892 * can happen in parallel across all CPUs
894 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
895 inode, rw, bio, mirror_num, 0,
897 __btree_submit_bio_start,
898 __btree_submit_bio_done);
901 #ifdef CONFIG_MIGRATION
902 static int btree_migratepage(struct address_space *mapping,
903 struct page *newpage, struct page *page,
904 enum migrate_mode mode)
907 * we can't safely write a btree page from here,
908 * we haven't done the locking hook
913 * Buffers may be managed in a filesystem specific way.
914 * We must have no buffers or drop them.
916 if (page_has_private(page) &&
917 !try_to_release_page(page, GFP_KERNEL))
919 return migrate_page(mapping, newpage, page, mode);
924 static int btree_writepages(struct address_space *mapping,
925 struct writeback_control *wbc)
927 struct extent_io_tree *tree;
928 tree = &BTRFS_I(mapping->host)->io_tree;
929 if (wbc->sync_mode == WB_SYNC_NONE) {
930 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
932 unsigned long thresh = 32 * 1024 * 1024;
934 if (wbc->for_kupdate)
937 /* this is a bit racy, but that's ok */
938 num_dirty = root->fs_info->dirty_metadata_bytes;
939 if (num_dirty < thresh)
942 return btree_write_cache_pages(mapping, wbc);
945 static int btree_readpage(struct file *file, struct page *page)
947 struct extent_io_tree *tree;
948 tree = &BTRFS_I(page->mapping->host)->io_tree;
949 return extent_read_full_page(tree, page, btree_get_extent, 0);
952 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
954 if (PageWriteback(page) || PageDirty(page))
957 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
958 * slab allocation from alloc_extent_state down the callchain where
959 * it'd hit a BUG_ON as those flags are not allowed.
961 gfp_flags &= ~GFP_SLAB_BUG_MASK;
963 return try_release_extent_buffer(page, gfp_flags);
966 static void btree_invalidatepage(struct page *page, unsigned long offset)
968 struct extent_io_tree *tree;
969 tree = &BTRFS_I(page->mapping->host)->io_tree;
970 extent_invalidatepage(tree, page, offset);
971 btree_releasepage(page, GFP_NOFS);
972 if (PagePrivate(page)) {
973 printk(KERN_WARNING "btrfs warning page private not zero "
974 "on page %llu\n", (unsigned long long)page_offset(page));
975 ClearPagePrivate(page);
976 set_page_private(page, 0);
977 page_cache_release(page);
981 static int btree_set_page_dirty(struct page *page)
983 struct extent_buffer *eb;
985 BUG_ON(!PagePrivate(page));
986 eb = (struct extent_buffer *)page->private;
988 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
989 BUG_ON(!atomic_read(&eb->refs));
990 btrfs_assert_tree_locked(eb);
991 return __set_page_dirty_nobuffers(page);
994 static const struct address_space_operations btree_aops = {
995 .readpage = btree_readpage,
996 .writepages = btree_writepages,
997 .releasepage = btree_releasepage,
998 .invalidatepage = btree_invalidatepage,
999 #ifdef CONFIG_MIGRATION
1000 .migratepage = btree_migratepage,
1002 .set_page_dirty = btree_set_page_dirty,
1005 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1008 struct extent_buffer *buf = NULL;
1009 struct inode *btree_inode = root->fs_info->btree_inode;
1012 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1015 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016 buf, 0, WAIT_NONE, btree_get_extent, 0);
1017 free_extent_buffer(buf);
1021 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1022 int mirror_num, struct extent_buffer **eb)
1024 struct extent_buffer *buf = NULL;
1025 struct inode *btree_inode = root->fs_info->btree_inode;
1026 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1029 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1033 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1035 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1036 btree_get_extent, mirror_num);
1038 free_extent_buffer(buf);
1042 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1043 free_extent_buffer(buf);
1045 } else if (extent_buffer_uptodate(buf)) {
1048 free_extent_buffer(buf);
1053 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1054 u64 bytenr, u32 blocksize)
1056 struct inode *btree_inode = root->fs_info->btree_inode;
1057 struct extent_buffer *eb;
1058 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1063 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1064 u64 bytenr, u32 blocksize)
1066 struct inode *btree_inode = root->fs_info->btree_inode;
1067 struct extent_buffer *eb;
1069 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1075 int btrfs_write_tree_block(struct extent_buffer *buf)
1077 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1078 buf->start + buf->len - 1);
1081 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1083 return filemap_fdatawait_range(buf->pages[0]->mapping,
1084 buf->start, buf->start + buf->len - 1);
1087 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1088 u32 blocksize, u64 parent_transid)
1090 struct extent_buffer *buf = NULL;
1093 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1097 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1102 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1103 struct extent_buffer *buf)
1105 if (btrfs_header_generation(buf) ==
1106 root->fs_info->running_transaction->transid) {
1107 btrfs_assert_tree_locked(buf);
1109 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1110 spin_lock(&root->fs_info->delalloc_lock);
1111 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1112 root->fs_info->dirty_metadata_bytes -= buf->len;
1114 spin_unlock(&root->fs_info->delalloc_lock);
1115 btrfs_panic(root->fs_info, -EOVERFLOW,
1116 "Can't clear %lu bytes from "
1117 " dirty_mdatadata_bytes (%lu)",
1119 root->fs_info->dirty_metadata_bytes);
1121 spin_unlock(&root->fs_info->delalloc_lock);
1124 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1125 btrfs_set_lock_blocking(buf);
1126 clear_extent_buffer_dirty(buf);
1130 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1131 u32 stripesize, struct btrfs_root *root,
1132 struct btrfs_fs_info *fs_info,
1136 root->commit_root = NULL;
1137 root->sectorsize = sectorsize;
1138 root->nodesize = nodesize;
1139 root->leafsize = leafsize;
1140 root->stripesize = stripesize;
1142 root->track_dirty = 0;
1144 root->orphan_item_inserted = 0;
1145 root->orphan_cleanup_state = 0;
1147 root->objectid = objectid;
1148 root->last_trans = 0;
1149 root->highest_objectid = 0;
1151 root->inode_tree = RB_ROOT;
1152 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153 root->block_rsv = NULL;
1154 root->orphan_block_rsv = NULL;
1156 INIT_LIST_HEAD(&root->dirty_list);
1157 INIT_LIST_HEAD(&root->root_list);
1158 spin_lock_init(&root->orphan_lock);
1159 spin_lock_init(&root->inode_lock);
1160 spin_lock_init(&root->accounting_lock);
1161 mutex_init(&root->objectid_mutex);
1162 mutex_init(&root->log_mutex);
1163 init_waitqueue_head(&root->log_writer_wait);
1164 init_waitqueue_head(&root->log_commit_wait[0]);
1165 init_waitqueue_head(&root->log_commit_wait[1]);
1166 atomic_set(&root->log_commit[0], 0);
1167 atomic_set(&root->log_commit[1], 0);
1168 atomic_set(&root->log_writers, 0);
1169 atomic_set(&root->orphan_inodes, 0);
1170 root->log_batch = 0;
1171 root->log_transid = 0;
1172 root->last_log_commit = 0;
1173 extent_io_tree_init(&root->dirty_log_pages,
1174 fs_info->btree_inode->i_mapping);
1176 memset(&root->root_key, 0, sizeof(root->root_key));
1177 memset(&root->root_item, 0, sizeof(root->root_item));
1178 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1179 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1180 root->defrag_trans_start = fs_info->generation;
1181 init_completion(&root->kobj_unregister);
1182 root->defrag_running = 0;
1183 root->root_key.objectid = objectid;
1187 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1188 struct btrfs_fs_info *fs_info,
1190 struct btrfs_root *root)
1196 __setup_root(tree_root->nodesize, tree_root->leafsize,
1197 tree_root->sectorsize, tree_root->stripesize,
1198 root, fs_info, objectid);
1199 ret = btrfs_find_last_root(tree_root, objectid,
1200 &root->root_item, &root->root_key);
1206 generation = btrfs_root_generation(&root->root_item);
1207 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1208 root->commit_root = NULL;
1209 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1210 blocksize, generation);
1211 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1212 free_extent_buffer(root->node);
1216 root->commit_root = btrfs_root_node(root);
1220 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1222 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1224 root->fs_info = fs_info;
1228 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1229 struct btrfs_fs_info *fs_info,
1232 struct extent_buffer *leaf;
1233 struct btrfs_root *tree_root = fs_info->tree_root;
1234 struct btrfs_root *root;
1235 struct btrfs_key key;
1239 root = btrfs_alloc_root(fs_info);
1241 return ERR_PTR(-ENOMEM);
1243 __setup_root(tree_root->nodesize, tree_root->leafsize,
1244 tree_root->sectorsize, tree_root->stripesize,
1245 root, fs_info, objectid);
1246 root->root_key.objectid = objectid;
1247 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1248 root->root_key.offset = 0;
1250 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1251 0, objectid, NULL, 0, 0, 0);
1253 ret = PTR_ERR(leaf);
1257 bytenr = leaf->start;
1258 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1259 btrfs_set_header_bytenr(leaf, leaf->start);
1260 btrfs_set_header_generation(leaf, trans->transid);
1261 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1262 btrfs_set_header_owner(leaf, objectid);
1265 write_extent_buffer(leaf, fs_info->fsid,
1266 (unsigned long)btrfs_header_fsid(leaf),
1268 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1269 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1271 btrfs_mark_buffer_dirty(leaf);
1273 root->commit_root = btrfs_root_node(root);
1274 root->track_dirty = 1;
1277 root->root_item.flags = 0;
1278 root->root_item.byte_limit = 0;
1279 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1280 btrfs_set_root_generation(&root->root_item, trans->transid);
1281 btrfs_set_root_level(&root->root_item, 0);
1282 btrfs_set_root_refs(&root->root_item, 1);
1283 btrfs_set_root_used(&root->root_item, leaf->len);
1284 btrfs_set_root_last_snapshot(&root->root_item, 0);
1285 btrfs_set_root_dirid(&root->root_item, 0);
1286 root->root_item.drop_level = 0;
1288 key.objectid = objectid;
1289 key.type = BTRFS_ROOT_ITEM_KEY;
1291 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1295 btrfs_tree_unlock(leaf);
1299 return ERR_PTR(ret);
1304 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1305 struct btrfs_fs_info *fs_info)
1307 struct btrfs_root *root;
1308 struct btrfs_root *tree_root = fs_info->tree_root;
1309 struct extent_buffer *leaf;
1311 root = btrfs_alloc_root(fs_info);
1313 return ERR_PTR(-ENOMEM);
1315 __setup_root(tree_root->nodesize, tree_root->leafsize,
1316 tree_root->sectorsize, tree_root->stripesize,
1317 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1319 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1320 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1321 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1323 * log trees do not get reference counted because they go away
1324 * before a real commit is actually done. They do store pointers
1325 * to file data extents, and those reference counts still get
1326 * updated (along with back refs to the log tree).
1330 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1331 BTRFS_TREE_LOG_OBJECTID, NULL,
1335 return ERR_CAST(leaf);
1338 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1339 btrfs_set_header_bytenr(leaf, leaf->start);
1340 btrfs_set_header_generation(leaf, trans->transid);
1341 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1342 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1345 write_extent_buffer(root->node, root->fs_info->fsid,
1346 (unsigned long)btrfs_header_fsid(root->node),
1348 btrfs_mark_buffer_dirty(root->node);
1349 btrfs_tree_unlock(root->node);
1353 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1354 struct btrfs_fs_info *fs_info)
1356 struct btrfs_root *log_root;
1358 log_root = alloc_log_tree(trans, fs_info);
1359 if (IS_ERR(log_root))
1360 return PTR_ERR(log_root);
1361 WARN_ON(fs_info->log_root_tree);
1362 fs_info->log_root_tree = log_root;
1366 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1367 struct btrfs_root *root)
1369 struct btrfs_root *log_root;
1370 struct btrfs_inode_item *inode_item;
1372 log_root = alloc_log_tree(trans, root->fs_info);
1373 if (IS_ERR(log_root))
1374 return PTR_ERR(log_root);
1376 log_root->last_trans = trans->transid;
1377 log_root->root_key.offset = root->root_key.objectid;
1379 inode_item = &log_root->root_item.inode;
1380 inode_item->generation = cpu_to_le64(1);
1381 inode_item->size = cpu_to_le64(3);
1382 inode_item->nlink = cpu_to_le32(1);
1383 inode_item->nbytes = cpu_to_le64(root->leafsize);
1384 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1386 btrfs_set_root_node(&log_root->root_item, log_root->node);
1388 WARN_ON(root->log_root);
1389 root->log_root = log_root;
1390 root->log_transid = 0;
1391 root->last_log_commit = 0;
1395 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1396 struct btrfs_key *location)
1398 struct btrfs_root *root;
1399 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1400 struct btrfs_path *path;
1401 struct extent_buffer *l;
1406 root = btrfs_alloc_root(fs_info);
1408 return ERR_PTR(-ENOMEM);
1409 if (location->offset == (u64)-1) {
1410 ret = find_and_setup_root(tree_root, fs_info,
1411 location->objectid, root);
1414 return ERR_PTR(ret);
1419 __setup_root(tree_root->nodesize, tree_root->leafsize,
1420 tree_root->sectorsize, tree_root->stripesize,
1421 root, fs_info, location->objectid);
1423 path = btrfs_alloc_path();
1426 return ERR_PTR(-ENOMEM);
1428 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1431 read_extent_buffer(l, &root->root_item,
1432 btrfs_item_ptr_offset(l, path->slots[0]),
1433 sizeof(root->root_item));
1434 memcpy(&root->root_key, location, sizeof(*location));
1436 btrfs_free_path(path);
1441 return ERR_PTR(ret);
1444 generation = btrfs_root_generation(&root->root_item);
1445 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1446 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1447 blocksize, generation);
1448 root->commit_root = btrfs_root_node(root);
1449 BUG_ON(!root->node); /* -ENOMEM */
1451 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1453 btrfs_check_and_init_root_item(&root->root_item);
1459 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1460 struct btrfs_key *location)
1462 struct btrfs_root *root;
1465 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1466 return fs_info->tree_root;
1467 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1468 return fs_info->extent_root;
1469 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1470 return fs_info->chunk_root;
1471 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1472 return fs_info->dev_root;
1473 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1474 return fs_info->csum_root;
1476 spin_lock(&fs_info->fs_roots_radix_lock);
1477 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1478 (unsigned long)location->objectid);
1479 spin_unlock(&fs_info->fs_roots_radix_lock);
1483 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1487 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1488 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1490 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1495 btrfs_init_free_ino_ctl(root);
1496 mutex_init(&root->fs_commit_mutex);
1497 spin_lock_init(&root->cache_lock);
1498 init_waitqueue_head(&root->cache_wait);
1500 ret = get_anon_bdev(&root->anon_dev);
1504 if (btrfs_root_refs(&root->root_item) == 0) {
1509 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1513 root->orphan_item_inserted = 1;
1515 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1519 spin_lock(&fs_info->fs_roots_radix_lock);
1520 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1521 (unsigned long)root->root_key.objectid,
1526 spin_unlock(&fs_info->fs_roots_radix_lock);
1527 radix_tree_preload_end();
1529 if (ret == -EEXIST) {
1536 ret = btrfs_find_dead_roots(fs_info->tree_root,
1537 root->root_key.objectid);
1542 return ERR_PTR(ret);
1545 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1547 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1549 struct btrfs_device *device;
1550 struct backing_dev_info *bdi;
1553 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1556 bdi = blk_get_backing_dev_info(device->bdev);
1557 if (bdi && bdi_congested(bdi, bdi_bits)) {
1567 * If this fails, caller must call bdi_destroy() to get rid of the
1570 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1574 bdi->capabilities = BDI_CAP_MAP_COPY;
1575 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1579 bdi->ra_pages = default_backing_dev_info.ra_pages;
1580 bdi->congested_fn = btrfs_congested_fn;
1581 bdi->congested_data = info;
1586 * called by the kthread helper functions to finally call the bio end_io
1587 * functions. This is where read checksum verification actually happens
1589 static void end_workqueue_fn(struct btrfs_work *work)
1592 struct end_io_wq *end_io_wq;
1593 struct btrfs_fs_info *fs_info;
1596 end_io_wq = container_of(work, struct end_io_wq, work);
1597 bio = end_io_wq->bio;
1598 fs_info = end_io_wq->info;
1600 error = end_io_wq->error;
1601 bio->bi_private = end_io_wq->private;
1602 bio->bi_end_io = end_io_wq->end_io;
1604 bio_endio(bio, error);
1607 static int cleaner_kthread(void *arg)
1609 struct btrfs_root *root = arg;
1612 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1614 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1615 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1616 btrfs_run_delayed_iputs(root);
1617 btrfs_clean_old_snapshots(root);
1618 mutex_unlock(&root->fs_info->cleaner_mutex);
1619 btrfs_run_defrag_inodes(root->fs_info);
1622 if (!try_to_freeze()) {
1623 set_current_state(TASK_INTERRUPTIBLE);
1624 if (!kthread_should_stop())
1626 __set_current_state(TASK_RUNNING);
1628 } while (!kthread_should_stop());
1632 static int transaction_kthread(void *arg)
1634 struct btrfs_root *root = arg;
1635 struct btrfs_trans_handle *trans;
1636 struct btrfs_transaction *cur;
1639 unsigned long delay;
1643 cannot_commit = false;
1645 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1646 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1648 spin_lock(&root->fs_info->trans_lock);
1649 cur = root->fs_info->running_transaction;
1651 spin_unlock(&root->fs_info->trans_lock);
1655 now = get_seconds();
1656 if (!cur->blocked &&
1657 (now < cur->start_time || now - cur->start_time < 30)) {
1658 spin_unlock(&root->fs_info->trans_lock);
1662 transid = cur->transid;
1663 spin_unlock(&root->fs_info->trans_lock);
1665 /* If the file system is aborted, this will always fail. */
1666 trans = btrfs_join_transaction(root);
1667 if (IS_ERR(trans)) {
1668 cannot_commit = true;
1671 if (transid == trans->transid) {
1672 btrfs_commit_transaction(trans, root);
1674 btrfs_end_transaction(trans, root);
1677 wake_up_process(root->fs_info->cleaner_kthread);
1678 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1680 if (!try_to_freeze()) {
1681 set_current_state(TASK_INTERRUPTIBLE);
1682 if (!kthread_should_stop() &&
1683 (!btrfs_transaction_blocked(root->fs_info) ||
1685 schedule_timeout(delay);
1686 __set_current_state(TASK_RUNNING);
1688 } while (!kthread_should_stop());
1693 * this will find the highest generation in the array of
1694 * root backups. The index of the highest array is returned,
1695 * or -1 if we can't find anything.
1697 * We check to make sure the array is valid by comparing the
1698 * generation of the latest root in the array with the generation
1699 * in the super block. If they don't match we pitch it.
1701 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1704 int newest_index = -1;
1705 struct btrfs_root_backup *root_backup;
1708 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1709 root_backup = info->super_copy->super_roots + i;
1710 cur = btrfs_backup_tree_root_gen(root_backup);
1711 if (cur == newest_gen)
1715 /* check to see if we actually wrapped around */
1716 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1717 root_backup = info->super_copy->super_roots;
1718 cur = btrfs_backup_tree_root_gen(root_backup);
1719 if (cur == newest_gen)
1722 return newest_index;
1727 * find the oldest backup so we know where to store new entries
1728 * in the backup array. This will set the backup_root_index
1729 * field in the fs_info struct
1731 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1734 int newest_index = -1;
1736 newest_index = find_newest_super_backup(info, newest_gen);
1737 /* if there was garbage in there, just move along */
1738 if (newest_index == -1) {
1739 info->backup_root_index = 0;
1741 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1746 * copy all the root pointers into the super backup array.
1747 * this will bump the backup pointer by one when it is
1750 static void backup_super_roots(struct btrfs_fs_info *info)
1753 struct btrfs_root_backup *root_backup;
1756 next_backup = info->backup_root_index;
1757 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1758 BTRFS_NUM_BACKUP_ROOTS;
1761 * just overwrite the last backup if we're at the same generation
1762 * this happens only at umount
1764 root_backup = info->super_for_commit->super_roots + last_backup;
1765 if (btrfs_backup_tree_root_gen(root_backup) ==
1766 btrfs_header_generation(info->tree_root->node))
1767 next_backup = last_backup;
1769 root_backup = info->super_for_commit->super_roots + next_backup;
1772 * make sure all of our padding and empty slots get zero filled
1773 * regardless of which ones we use today
1775 memset(root_backup, 0, sizeof(*root_backup));
1777 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1779 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1780 btrfs_set_backup_tree_root_gen(root_backup,
1781 btrfs_header_generation(info->tree_root->node));
1783 btrfs_set_backup_tree_root_level(root_backup,
1784 btrfs_header_level(info->tree_root->node));
1786 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1787 btrfs_set_backup_chunk_root_gen(root_backup,
1788 btrfs_header_generation(info->chunk_root->node));
1789 btrfs_set_backup_chunk_root_level(root_backup,
1790 btrfs_header_level(info->chunk_root->node));
1792 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1793 btrfs_set_backup_extent_root_gen(root_backup,
1794 btrfs_header_generation(info->extent_root->node));
1795 btrfs_set_backup_extent_root_level(root_backup,
1796 btrfs_header_level(info->extent_root->node));
1799 * we might commit during log recovery, which happens before we set
1800 * the fs_root. Make sure it is valid before we fill it in.
1802 if (info->fs_root && info->fs_root->node) {
1803 btrfs_set_backup_fs_root(root_backup,
1804 info->fs_root->node->start);
1805 btrfs_set_backup_fs_root_gen(root_backup,
1806 btrfs_header_generation(info->fs_root->node));
1807 btrfs_set_backup_fs_root_level(root_backup,
1808 btrfs_header_level(info->fs_root->node));
1811 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1812 btrfs_set_backup_dev_root_gen(root_backup,
1813 btrfs_header_generation(info->dev_root->node));
1814 btrfs_set_backup_dev_root_level(root_backup,
1815 btrfs_header_level(info->dev_root->node));
1817 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1818 btrfs_set_backup_csum_root_gen(root_backup,
1819 btrfs_header_generation(info->csum_root->node));
1820 btrfs_set_backup_csum_root_level(root_backup,
1821 btrfs_header_level(info->csum_root->node));
1823 btrfs_set_backup_total_bytes(root_backup,
1824 btrfs_super_total_bytes(info->super_copy));
1825 btrfs_set_backup_bytes_used(root_backup,
1826 btrfs_super_bytes_used(info->super_copy));
1827 btrfs_set_backup_num_devices(root_backup,
1828 btrfs_super_num_devices(info->super_copy));
1831 * if we don't copy this out to the super_copy, it won't get remembered
1832 * for the next commit
1834 memcpy(&info->super_copy->super_roots,
1835 &info->super_for_commit->super_roots,
1836 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1840 * this copies info out of the root backup array and back into
1841 * the in-memory super block. It is meant to help iterate through
1842 * the array, so you send it the number of backups you've already
1843 * tried and the last backup index you used.
1845 * this returns -1 when it has tried all the backups
1847 static noinline int next_root_backup(struct btrfs_fs_info *info,
1848 struct btrfs_super_block *super,
1849 int *num_backups_tried, int *backup_index)
1851 struct btrfs_root_backup *root_backup;
1852 int newest = *backup_index;
1854 if (*num_backups_tried == 0) {
1855 u64 gen = btrfs_super_generation(super);
1857 newest = find_newest_super_backup(info, gen);
1861 *backup_index = newest;
1862 *num_backups_tried = 1;
1863 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1864 /* we've tried all the backups, all done */
1867 /* jump to the next oldest backup */
1868 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1869 BTRFS_NUM_BACKUP_ROOTS;
1870 *backup_index = newest;
1871 *num_backups_tried += 1;
1873 root_backup = super->super_roots + newest;
1875 btrfs_set_super_generation(super,
1876 btrfs_backup_tree_root_gen(root_backup));
1877 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1878 btrfs_set_super_root_level(super,
1879 btrfs_backup_tree_root_level(root_backup));
1880 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1883 * fixme: the total bytes and num_devices need to match or we should
1886 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1887 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1891 /* helper to cleanup tree roots */
1892 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1894 free_extent_buffer(info->tree_root->node);
1895 free_extent_buffer(info->tree_root->commit_root);
1896 free_extent_buffer(info->dev_root->node);
1897 free_extent_buffer(info->dev_root->commit_root);
1898 free_extent_buffer(info->extent_root->node);
1899 free_extent_buffer(info->extent_root->commit_root);
1900 free_extent_buffer(info->csum_root->node);
1901 free_extent_buffer(info->csum_root->commit_root);
1903 info->tree_root->node = NULL;
1904 info->tree_root->commit_root = NULL;
1905 info->dev_root->node = NULL;
1906 info->dev_root->commit_root = NULL;
1907 info->extent_root->node = NULL;
1908 info->extent_root->commit_root = NULL;
1909 info->csum_root->node = NULL;
1910 info->csum_root->commit_root = NULL;
1913 free_extent_buffer(info->chunk_root->node);
1914 free_extent_buffer(info->chunk_root->commit_root);
1915 info->chunk_root->node = NULL;
1916 info->chunk_root->commit_root = NULL;
1921 int open_ctree(struct super_block *sb,
1922 struct btrfs_fs_devices *fs_devices,
1932 struct btrfs_key location;
1933 struct buffer_head *bh;
1934 struct btrfs_super_block *disk_super;
1935 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1936 struct btrfs_root *tree_root;
1937 struct btrfs_root *extent_root;
1938 struct btrfs_root *csum_root;
1939 struct btrfs_root *chunk_root;
1940 struct btrfs_root *dev_root;
1941 struct btrfs_root *log_tree_root;
1944 int num_backups_tried = 0;
1945 int backup_index = 0;
1947 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1948 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1949 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1950 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1951 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1953 if (!tree_root || !extent_root || !csum_root ||
1954 !chunk_root || !dev_root) {
1959 ret = init_srcu_struct(&fs_info->subvol_srcu);
1965 ret = setup_bdi(fs_info, &fs_info->bdi);
1971 fs_info->btree_inode = new_inode(sb);
1972 if (!fs_info->btree_inode) {
1977 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1979 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1980 INIT_LIST_HEAD(&fs_info->trans_list);
1981 INIT_LIST_HEAD(&fs_info->dead_roots);
1982 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1983 INIT_LIST_HEAD(&fs_info->hashers);
1984 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1985 INIT_LIST_HEAD(&fs_info->ordered_operations);
1986 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1987 spin_lock_init(&fs_info->delalloc_lock);
1988 spin_lock_init(&fs_info->trans_lock);
1989 spin_lock_init(&fs_info->ref_cache_lock);
1990 spin_lock_init(&fs_info->fs_roots_radix_lock);
1991 spin_lock_init(&fs_info->delayed_iput_lock);
1992 spin_lock_init(&fs_info->defrag_inodes_lock);
1993 spin_lock_init(&fs_info->free_chunk_lock);
1994 spin_lock_init(&fs_info->tree_mod_seq_lock);
1995 rwlock_init(&fs_info->tree_mod_log_lock);
1996 mutex_init(&fs_info->reloc_mutex);
1998 init_completion(&fs_info->kobj_unregister);
1999 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2000 INIT_LIST_HEAD(&fs_info->space_info);
2001 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
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);
2014 atomic_set(&fs_info->tree_mod_seq, 0);
2016 fs_info->max_inline = 8192 * 1024;
2017 fs_info->metadata_ratio = 0;
2018 fs_info->defrag_inodes = RB_ROOT;
2019 fs_info->trans_no_join = 0;
2020 fs_info->free_chunk_space = 0;
2021 fs_info->tree_mod_log = RB_ROOT;
2023 init_waitqueue_head(&fs_info->tree_mod_seq_wait);
2025 /* readahead state */
2026 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2027 spin_lock_init(&fs_info->reada_lock);
2029 fs_info->thread_pool_size = min_t(unsigned long,
2030 num_online_cpus() + 2, 8);
2032 INIT_LIST_HEAD(&fs_info->ordered_extents);
2033 spin_lock_init(&fs_info->ordered_extent_lock);
2034 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2036 if (!fs_info->delayed_root) {
2040 btrfs_init_delayed_root(fs_info->delayed_root);
2042 mutex_init(&fs_info->scrub_lock);
2043 atomic_set(&fs_info->scrubs_running, 0);
2044 atomic_set(&fs_info->scrub_pause_req, 0);
2045 atomic_set(&fs_info->scrubs_paused, 0);
2046 atomic_set(&fs_info->scrub_cancel_req, 0);
2047 init_waitqueue_head(&fs_info->scrub_pause_wait);
2048 init_rwsem(&fs_info->scrub_super_lock);
2049 fs_info->scrub_workers_refcnt = 0;
2050 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2051 fs_info->check_integrity_print_mask = 0;
2054 spin_lock_init(&fs_info->balance_lock);
2055 mutex_init(&fs_info->balance_mutex);
2056 atomic_set(&fs_info->balance_running, 0);
2057 atomic_set(&fs_info->balance_pause_req, 0);
2058 atomic_set(&fs_info->balance_cancel_req, 0);
2059 fs_info->balance_ctl = NULL;
2060 init_waitqueue_head(&fs_info->balance_wait_q);
2062 sb->s_blocksize = 4096;
2063 sb->s_blocksize_bits = blksize_bits(4096);
2064 sb->s_bdi = &fs_info->bdi;
2066 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2067 set_nlink(fs_info->btree_inode, 1);
2069 * we set the i_size on the btree inode to the max possible int.
2070 * the real end of the address space is determined by all of
2071 * the devices in the system
2073 fs_info->btree_inode->i_size = OFFSET_MAX;
2074 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2075 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2077 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2078 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2079 fs_info->btree_inode->i_mapping);
2080 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2081 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2083 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2085 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2086 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2087 sizeof(struct btrfs_key));
2088 set_bit(BTRFS_INODE_DUMMY,
2089 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2090 insert_inode_hash(fs_info->btree_inode);
2092 spin_lock_init(&fs_info->block_group_cache_lock);
2093 fs_info->block_group_cache_tree = RB_ROOT;
2095 extent_io_tree_init(&fs_info->freed_extents[0],
2096 fs_info->btree_inode->i_mapping);
2097 extent_io_tree_init(&fs_info->freed_extents[1],
2098 fs_info->btree_inode->i_mapping);
2099 fs_info->pinned_extents = &fs_info->freed_extents[0];
2100 fs_info->do_barriers = 1;
2103 mutex_init(&fs_info->ordered_operations_mutex);
2104 mutex_init(&fs_info->tree_log_mutex);
2105 mutex_init(&fs_info->chunk_mutex);
2106 mutex_init(&fs_info->transaction_kthread_mutex);
2107 mutex_init(&fs_info->cleaner_mutex);
2108 mutex_init(&fs_info->volume_mutex);
2109 init_rwsem(&fs_info->extent_commit_sem);
2110 init_rwsem(&fs_info->cleanup_work_sem);
2111 init_rwsem(&fs_info->subvol_sem);
2113 spin_lock_init(&fs_info->qgroup_lock);
2114 fs_info->qgroup_tree = RB_ROOT;
2115 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2116 fs_info->qgroup_seq = 1;
2117 fs_info->quota_enabled = 0;
2118 fs_info->pending_quota_state = 0;
2120 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2121 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2123 init_waitqueue_head(&fs_info->transaction_throttle);
2124 init_waitqueue_head(&fs_info->transaction_wait);
2125 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2126 init_waitqueue_head(&fs_info->async_submit_wait);
2128 __setup_root(4096, 4096, 4096, 4096, tree_root,
2129 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2131 invalidate_bdev(fs_devices->latest_bdev);
2132 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2138 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2139 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2140 sizeof(*fs_info->super_for_commit));
2143 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2145 disk_super = fs_info->super_copy;
2146 if (!btrfs_super_root(disk_super))
2149 /* check FS state, whether FS is broken. */
2150 fs_info->fs_state |= btrfs_super_flags(disk_super);
2152 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2154 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2160 * run through our array of backup supers and setup
2161 * our ring pointer to the oldest one
2163 generation = btrfs_super_generation(disk_super);
2164 find_oldest_super_backup(fs_info, generation);
2167 * In the long term, we'll store the compression type in the super
2168 * block, and it'll be used for per file compression control.
2170 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2172 ret = btrfs_parse_options(tree_root, options);
2178 features = btrfs_super_incompat_flags(disk_super) &
2179 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2181 printk(KERN_ERR "BTRFS: couldn't mount because of "
2182 "unsupported optional features (%Lx).\n",
2183 (unsigned long long)features);
2188 if (btrfs_super_leafsize(disk_super) !=
2189 btrfs_super_nodesize(disk_super)) {
2190 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2191 "blocksizes don't match. node %d leaf %d\n",
2192 btrfs_super_nodesize(disk_super),
2193 btrfs_super_leafsize(disk_super));
2197 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2198 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2199 "blocksize (%d) was too large\n",
2200 btrfs_super_leafsize(disk_super));
2205 features = btrfs_super_incompat_flags(disk_super);
2206 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2207 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2208 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2211 * flag our filesystem as having big metadata blocks if
2212 * they are bigger than the page size
2214 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2215 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2216 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2217 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2220 nodesize = btrfs_super_nodesize(disk_super);
2221 leafsize = btrfs_super_leafsize(disk_super);
2222 sectorsize = btrfs_super_sectorsize(disk_super);
2223 stripesize = btrfs_super_stripesize(disk_super);
2226 * mixed block groups end up with duplicate but slightly offset
2227 * extent buffers for the same range. It leads to corruptions
2229 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2230 (sectorsize != leafsize)) {
2231 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2232 "are not allowed for mixed block groups on %s\n",
2237 btrfs_set_super_incompat_flags(disk_super, features);
2239 features = btrfs_super_compat_ro_flags(disk_super) &
2240 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2241 if (!(sb->s_flags & MS_RDONLY) && features) {
2242 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2243 "unsupported option features (%Lx).\n",
2244 (unsigned long long)features);
2249 btrfs_init_workers(&fs_info->generic_worker,
2250 "genwork", 1, NULL);
2252 btrfs_init_workers(&fs_info->workers, "worker",
2253 fs_info->thread_pool_size,
2254 &fs_info->generic_worker);
2256 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2257 fs_info->thread_pool_size,
2258 &fs_info->generic_worker);
2260 btrfs_init_workers(&fs_info->submit_workers, "submit",
2261 min_t(u64, fs_devices->num_devices,
2262 fs_info->thread_pool_size),
2263 &fs_info->generic_worker);
2265 btrfs_init_workers(&fs_info->caching_workers, "cache",
2266 2, &fs_info->generic_worker);
2268 /* a higher idle thresh on the submit workers makes it much more
2269 * likely that bios will be send down in a sane order to the
2272 fs_info->submit_workers.idle_thresh = 64;
2274 fs_info->workers.idle_thresh = 16;
2275 fs_info->workers.ordered = 1;
2277 fs_info->delalloc_workers.idle_thresh = 2;
2278 fs_info->delalloc_workers.ordered = 1;
2280 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2281 &fs_info->generic_worker);
2282 btrfs_init_workers(&fs_info->endio_workers, "endio",
2283 fs_info->thread_pool_size,
2284 &fs_info->generic_worker);
2285 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2286 fs_info->thread_pool_size,
2287 &fs_info->generic_worker);
2288 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2289 "endio-meta-write", fs_info->thread_pool_size,
2290 &fs_info->generic_worker);
2291 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2292 fs_info->thread_pool_size,
2293 &fs_info->generic_worker);
2294 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2295 1, &fs_info->generic_worker);
2296 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2297 fs_info->thread_pool_size,
2298 &fs_info->generic_worker);
2299 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2300 fs_info->thread_pool_size,
2301 &fs_info->generic_worker);
2304 * endios are largely parallel and should have a very
2307 fs_info->endio_workers.idle_thresh = 4;
2308 fs_info->endio_meta_workers.idle_thresh = 4;
2310 fs_info->endio_write_workers.idle_thresh = 2;
2311 fs_info->endio_meta_write_workers.idle_thresh = 2;
2312 fs_info->readahead_workers.idle_thresh = 2;
2315 * btrfs_start_workers can really only fail because of ENOMEM so just
2316 * return -ENOMEM if any of these fail.
2318 ret = btrfs_start_workers(&fs_info->workers);
2319 ret |= btrfs_start_workers(&fs_info->generic_worker);
2320 ret |= btrfs_start_workers(&fs_info->submit_workers);
2321 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2322 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2323 ret |= btrfs_start_workers(&fs_info->endio_workers);
2324 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2325 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2326 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2327 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2328 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2329 ret |= btrfs_start_workers(&fs_info->caching_workers);
2330 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2333 goto fail_sb_buffer;
2336 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2337 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2338 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2340 tree_root->nodesize = nodesize;
2341 tree_root->leafsize = leafsize;
2342 tree_root->sectorsize = sectorsize;
2343 tree_root->stripesize = stripesize;
2345 sb->s_blocksize = sectorsize;
2346 sb->s_blocksize_bits = blksize_bits(sectorsize);
2348 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2349 sizeof(disk_super->magic))) {
2350 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2351 goto fail_sb_buffer;
2354 if (sectorsize != PAGE_SIZE) {
2355 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2356 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2357 goto fail_sb_buffer;
2360 mutex_lock(&fs_info->chunk_mutex);
2361 ret = btrfs_read_sys_array(tree_root);
2362 mutex_unlock(&fs_info->chunk_mutex);
2364 printk(KERN_WARNING "btrfs: failed to read the system "
2365 "array on %s\n", sb->s_id);
2366 goto fail_sb_buffer;
2369 blocksize = btrfs_level_size(tree_root,
2370 btrfs_super_chunk_root_level(disk_super));
2371 generation = btrfs_super_chunk_root_generation(disk_super);
2373 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2374 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2376 chunk_root->node = read_tree_block(chunk_root,
2377 btrfs_super_chunk_root(disk_super),
2378 blocksize, generation);
2379 BUG_ON(!chunk_root->node); /* -ENOMEM */
2380 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2381 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2383 goto fail_tree_roots;
2385 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2386 chunk_root->commit_root = btrfs_root_node(chunk_root);
2388 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2389 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2392 ret = btrfs_read_chunk_tree(chunk_root);
2394 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2396 goto fail_tree_roots;
2399 btrfs_close_extra_devices(fs_devices);
2401 if (!fs_devices->latest_bdev) {
2402 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2404 goto fail_tree_roots;
2408 blocksize = btrfs_level_size(tree_root,
2409 btrfs_super_root_level(disk_super));
2410 generation = btrfs_super_generation(disk_super);
2412 tree_root->node = read_tree_block(tree_root,
2413 btrfs_super_root(disk_super),
2414 blocksize, generation);
2415 if (!tree_root->node ||
2416 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2417 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2420 goto recovery_tree_root;
2423 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2424 tree_root->commit_root = btrfs_root_node(tree_root);
2426 ret = find_and_setup_root(tree_root, fs_info,
2427 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2429 goto recovery_tree_root;
2430 extent_root->track_dirty = 1;
2432 ret = find_and_setup_root(tree_root, fs_info,
2433 BTRFS_DEV_TREE_OBJECTID, dev_root);
2435 goto recovery_tree_root;
2436 dev_root->track_dirty = 1;
2438 ret = find_and_setup_root(tree_root, fs_info,
2439 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2441 goto recovery_tree_root;
2442 csum_root->track_dirty = 1;
2444 fs_info->generation = generation;
2445 fs_info->last_trans_committed = generation;
2447 ret = btrfs_recover_balance(fs_info);
2449 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2450 goto fail_block_groups;
2453 ret = btrfs_init_dev_stats(fs_info);
2455 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2457 goto fail_block_groups;
2460 ret = btrfs_init_space_info(fs_info);
2462 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2463 goto fail_block_groups;
2466 ret = btrfs_read_block_groups(extent_root);
2468 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2469 goto fail_block_groups;
2472 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2474 if (IS_ERR(fs_info->cleaner_kthread))
2475 goto fail_block_groups;
2477 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2479 "btrfs-transaction");
2480 if (IS_ERR(fs_info->transaction_kthread))
2483 if (!btrfs_test_opt(tree_root, SSD) &&
2484 !btrfs_test_opt(tree_root, NOSSD) &&
2485 !fs_info->fs_devices->rotating) {
2486 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2488 btrfs_set_opt(fs_info->mount_opt, SSD);
2491 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2492 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2493 ret = btrfsic_mount(tree_root, fs_devices,
2494 btrfs_test_opt(tree_root,
2495 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2497 fs_info->check_integrity_print_mask);
2499 printk(KERN_WARNING "btrfs: failed to initialize"
2500 " integrity check module %s\n", sb->s_id);
2504 /* do not make disk changes in broken FS */
2505 if (btrfs_super_log_root(disk_super) != 0 &&
2506 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2507 u64 bytenr = btrfs_super_log_root(disk_super);
2509 if (fs_devices->rw_devices == 0) {
2510 printk(KERN_WARNING "Btrfs log replay required "
2513 goto fail_trans_kthread;
2516 btrfs_level_size(tree_root,
2517 btrfs_super_log_root_level(disk_super));
2519 log_tree_root = btrfs_alloc_root(fs_info);
2520 if (!log_tree_root) {
2522 goto fail_trans_kthread;
2525 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2526 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2528 log_tree_root->node = read_tree_block(tree_root, bytenr,
2531 /* returns with log_tree_root freed on success */
2532 ret = btrfs_recover_log_trees(log_tree_root);
2534 btrfs_error(tree_root->fs_info, ret,
2535 "Failed to recover log tree");
2536 free_extent_buffer(log_tree_root->node);
2537 kfree(log_tree_root);
2538 goto fail_trans_kthread;
2541 if (sb->s_flags & MS_RDONLY) {
2542 ret = btrfs_commit_super(tree_root);
2544 goto fail_trans_kthread;
2548 ret = btrfs_find_orphan_roots(tree_root);
2550 goto fail_trans_kthread;
2552 if (!(sb->s_flags & MS_RDONLY)) {
2553 ret = btrfs_cleanup_fs_roots(fs_info);
2557 ret = btrfs_recover_relocation(tree_root);
2560 "btrfs: failed to recover relocation\n");
2562 goto fail_trans_kthread;
2566 location.objectid = BTRFS_FS_TREE_OBJECTID;
2567 location.type = BTRFS_ROOT_ITEM_KEY;
2568 location.offset = (u64)-1;
2570 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2571 if (!fs_info->fs_root)
2572 goto fail_trans_kthread;
2573 if (IS_ERR(fs_info->fs_root)) {
2574 err = PTR_ERR(fs_info->fs_root);
2575 goto fail_trans_kthread;
2578 if (sb->s_flags & MS_RDONLY)
2581 down_read(&fs_info->cleanup_work_sem);
2582 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2583 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2584 up_read(&fs_info->cleanup_work_sem);
2585 close_ctree(tree_root);
2588 up_read(&fs_info->cleanup_work_sem);
2590 ret = btrfs_resume_balance_async(fs_info);
2592 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2593 close_ctree(tree_root);
2600 kthread_stop(fs_info->transaction_kthread);
2602 kthread_stop(fs_info->cleaner_kthread);
2605 * make sure we're done with the btree inode before we stop our
2608 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2609 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2612 btrfs_free_block_groups(fs_info);
2615 free_root_pointers(fs_info, 1);
2618 btrfs_stop_workers(&fs_info->generic_worker);
2619 btrfs_stop_workers(&fs_info->readahead_workers);
2620 btrfs_stop_workers(&fs_info->fixup_workers);
2621 btrfs_stop_workers(&fs_info->delalloc_workers);
2622 btrfs_stop_workers(&fs_info->workers);
2623 btrfs_stop_workers(&fs_info->endio_workers);
2624 btrfs_stop_workers(&fs_info->endio_meta_workers);
2625 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2626 btrfs_stop_workers(&fs_info->endio_write_workers);
2627 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2628 btrfs_stop_workers(&fs_info->submit_workers);
2629 btrfs_stop_workers(&fs_info->delayed_workers);
2630 btrfs_stop_workers(&fs_info->caching_workers);
2633 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2635 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2636 iput(fs_info->btree_inode);
2638 bdi_destroy(&fs_info->bdi);
2640 cleanup_srcu_struct(&fs_info->subvol_srcu);
2642 btrfs_close_devices(fs_info->fs_devices);
2646 if (!btrfs_test_opt(tree_root, RECOVERY))
2647 goto fail_tree_roots;
2649 free_root_pointers(fs_info, 0);
2651 /* don't use the log in recovery mode, it won't be valid */
2652 btrfs_set_super_log_root(disk_super, 0);
2654 /* we can't trust the free space cache either */
2655 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2657 ret = next_root_backup(fs_info, fs_info->super_copy,
2658 &num_backups_tried, &backup_index);
2660 goto fail_block_groups;
2661 goto retry_root_backup;
2664 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2667 set_buffer_uptodate(bh);
2669 struct btrfs_device *device = (struct btrfs_device *)
2672 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2673 "I/O error on %s\n",
2674 rcu_str_deref(device->name));
2675 /* note, we dont' set_buffer_write_io_error because we have
2676 * our own ways of dealing with the IO errors
2678 clear_buffer_uptodate(bh);
2679 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2685 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2687 struct buffer_head *bh;
2688 struct buffer_head *latest = NULL;
2689 struct btrfs_super_block *super;
2694 /* we would like to check all the supers, but that would make
2695 * a btrfs mount succeed after a mkfs from a different FS.
2696 * So, we need to add a special mount option to scan for
2697 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2699 for (i = 0; i < 1; i++) {
2700 bytenr = btrfs_sb_offset(i);
2701 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2703 bh = __bread(bdev, bytenr / 4096, 4096);
2707 super = (struct btrfs_super_block *)bh->b_data;
2708 if (btrfs_super_bytenr(super) != bytenr ||
2709 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2710 sizeof(super->magic))) {
2715 if (!latest || btrfs_super_generation(super) > transid) {
2718 transid = btrfs_super_generation(super);
2727 * this should be called twice, once with wait == 0 and
2728 * once with wait == 1. When wait == 0 is done, all the buffer heads
2729 * we write are pinned.
2731 * They are released when wait == 1 is done.
2732 * max_mirrors must be the same for both runs, and it indicates how
2733 * many supers on this one device should be written.
2735 * max_mirrors == 0 means to write them all.
2737 static int write_dev_supers(struct btrfs_device *device,
2738 struct btrfs_super_block *sb,
2739 int do_barriers, int wait, int max_mirrors)
2741 struct buffer_head *bh;
2748 if (max_mirrors == 0)
2749 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2751 for (i = 0; i < max_mirrors; i++) {
2752 bytenr = btrfs_sb_offset(i);
2753 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2757 bh = __find_get_block(device->bdev, bytenr / 4096,
2758 BTRFS_SUPER_INFO_SIZE);
2761 if (!buffer_uptodate(bh))
2764 /* drop our reference */
2767 /* drop the reference from the wait == 0 run */
2771 btrfs_set_super_bytenr(sb, bytenr);
2774 crc = btrfs_csum_data(NULL, (char *)sb +
2775 BTRFS_CSUM_SIZE, crc,
2776 BTRFS_SUPER_INFO_SIZE -
2778 btrfs_csum_final(crc, sb->csum);
2781 * one reference for us, and we leave it for the
2784 bh = __getblk(device->bdev, bytenr / 4096,
2785 BTRFS_SUPER_INFO_SIZE);
2786 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2788 /* one reference for submit_bh */
2791 set_buffer_uptodate(bh);
2793 bh->b_end_io = btrfs_end_buffer_write_sync;
2794 bh->b_private = device;
2798 * we fua the first super. The others we allow
2801 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2805 return errors < i ? 0 : -1;
2809 * endio for the write_dev_flush, this will wake anyone waiting
2810 * for the barrier when it is done
2812 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2815 if (err == -EOPNOTSUPP)
2816 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2817 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2819 if (bio->bi_private)
2820 complete(bio->bi_private);
2825 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2826 * sent down. With wait == 1, it waits for the previous flush.
2828 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2831 static int write_dev_flush(struct btrfs_device *device, int wait)
2836 if (device->nobarriers)
2840 bio = device->flush_bio;
2844 wait_for_completion(&device->flush_wait);
2846 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2847 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2848 rcu_str_deref(device->name));
2849 device->nobarriers = 1;
2851 if (!bio_flagged(bio, BIO_UPTODATE)) {
2853 if (!bio_flagged(bio, BIO_EOPNOTSUPP))
2854 btrfs_dev_stat_inc_and_print(device,
2855 BTRFS_DEV_STAT_FLUSH_ERRS);
2858 /* drop the reference from the wait == 0 run */
2860 device->flush_bio = NULL;
2866 * one reference for us, and we leave it for the
2869 device->flush_bio = NULL;;
2870 bio = bio_alloc(GFP_NOFS, 0);
2874 bio->bi_end_io = btrfs_end_empty_barrier;
2875 bio->bi_bdev = device->bdev;
2876 init_completion(&device->flush_wait);
2877 bio->bi_private = &device->flush_wait;
2878 device->flush_bio = bio;
2881 btrfsic_submit_bio(WRITE_FLUSH, bio);
2887 * send an empty flush down to each device in parallel,
2888 * then wait for them
2890 static int barrier_all_devices(struct btrfs_fs_info *info)
2892 struct list_head *head;
2893 struct btrfs_device *dev;
2897 /* send down all the barriers */
2898 head = &info->fs_devices->devices;
2899 list_for_each_entry_rcu(dev, head, dev_list) {
2904 if (!dev->in_fs_metadata || !dev->writeable)
2907 ret = write_dev_flush(dev, 0);
2912 /* wait for all the barriers */
2913 list_for_each_entry_rcu(dev, head, dev_list) {
2918 if (!dev->in_fs_metadata || !dev->writeable)
2921 ret = write_dev_flush(dev, 1);
2930 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2932 struct list_head *head;
2933 struct btrfs_device *dev;
2934 struct btrfs_super_block *sb;
2935 struct btrfs_dev_item *dev_item;
2939 int total_errors = 0;
2942 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2943 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2944 backup_super_roots(root->fs_info);
2946 sb = root->fs_info->super_for_commit;
2947 dev_item = &sb->dev_item;
2949 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2950 head = &root->fs_info->fs_devices->devices;
2953 barrier_all_devices(root->fs_info);
2955 list_for_each_entry_rcu(dev, head, dev_list) {
2960 if (!dev->in_fs_metadata || !dev->writeable)
2963 btrfs_set_stack_device_generation(dev_item, 0);
2964 btrfs_set_stack_device_type(dev_item, dev->type);
2965 btrfs_set_stack_device_id(dev_item, dev->devid);
2966 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2967 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2968 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2969 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2970 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2971 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2972 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2974 flags = btrfs_super_flags(sb);
2975 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2977 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2981 if (total_errors > max_errors) {
2982 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2985 /* This shouldn't happen. FUA is masked off if unsupported */
2990 list_for_each_entry_rcu(dev, head, dev_list) {
2993 if (!dev->in_fs_metadata || !dev->writeable)
2996 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3000 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3001 if (total_errors > max_errors) {
3002 btrfs_error(root->fs_info, -EIO,
3003 "%d errors while writing supers", total_errors);
3009 int write_ctree_super(struct btrfs_trans_handle *trans,
3010 struct btrfs_root *root, int max_mirrors)
3014 ret = write_all_supers(root, max_mirrors);
3018 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3020 spin_lock(&fs_info->fs_roots_radix_lock);
3021 radix_tree_delete(&fs_info->fs_roots_radix,
3022 (unsigned long)root->root_key.objectid);
3023 spin_unlock(&fs_info->fs_roots_radix_lock);
3025 if (btrfs_root_refs(&root->root_item) == 0)
3026 synchronize_srcu(&fs_info->subvol_srcu);
3028 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3029 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3033 static void free_fs_root(struct btrfs_root *root)
3035 iput(root->cache_inode);
3036 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3038 free_anon_bdev(root->anon_dev);
3039 free_extent_buffer(root->node);
3040 free_extent_buffer(root->commit_root);
3041 kfree(root->free_ino_ctl);
3042 kfree(root->free_ino_pinned);
3047 static void del_fs_roots(struct btrfs_fs_info *fs_info)
3050 struct btrfs_root *gang[8];
3053 while (!list_empty(&fs_info->dead_roots)) {
3054 gang[0] = list_entry(fs_info->dead_roots.next,
3055 struct btrfs_root, root_list);
3056 list_del(&gang[0]->root_list);
3058 if (gang[0]->in_radix) {
3059 btrfs_free_fs_root(fs_info, gang[0]);
3061 free_extent_buffer(gang[0]->node);
3062 free_extent_buffer(gang[0]->commit_root);
3068 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3073 for (i = 0; i < ret; i++)
3074 btrfs_free_fs_root(fs_info, gang[i]);
3078 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3080 u64 root_objectid = 0;
3081 struct btrfs_root *gang[8];
3086 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3087 (void **)gang, root_objectid,
3092 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3093 for (i = 0; i < ret; i++) {
3096 root_objectid = gang[i]->root_key.objectid;
3097 err = btrfs_orphan_cleanup(gang[i]);
3106 int btrfs_commit_super(struct btrfs_root *root)
3108 struct btrfs_trans_handle *trans;
3111 mutex_lock(&root->fs_info->cleaner_mutex);
3112 btrfs_run_delayed_iputs(root);
3113 btrfs_clean_old_snapshots(root);
3114 mutex_unlock(&root->fs_info->cleaner_mutex);
3116 /* wait until ongoing cleanup work done */
3117 down_write(&root->fs_info->cleanup_work_sem);
3118 up_write(&root->fs_info->cleanup_work_sem);
3120 trans = btrfs_join_transaction(root);
3122 return PTR_ERR(trans);
3123 ret = btrfs_commit_transaction(trans, root);
3126 /* run commit again to drop the original snapshot */
3127 trans = btrfs_join_transaction(root);
3129 return PTR_ERR(trans);
3130 ret = btrfs_commit_transaction(trans, root);
3133 ret = btrfs_write_and_wait_transaction(NULL, root);
3135 btrfs_error(root->fs_info, ret,
3136 "Failed to sync btree inode to disk.");
3140 ret = write_ctree_super(NULL, root, 0);
3144 int close_ctree(struct btrfs_root *root)
3146 struct btrfs_fs_info *fs_info = root->fs_info;
3149 fs_info->closing = 1;
3152 /* pause restriper - we want to resume on mount */
3153 btrfs_pause_balance(root->fs_info);
3155 btrfs_scrub_cancel(root);
3157 /* wait for any defraggers to finish */
3158 wait_event(fs_info->transaction_wait,
3159 (atomic_read(&fs_info->defrag_running) == 0));
3161 /* clear out the rbtree of defraggable inodes */
3162 btrfs_run_defrag_inodes(fs_info);
3165 * Here come 2 situations when btrfs is broken to flip readonly:
3167 * 1. when btrfs flips readonly somewhere else before
3168 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3169 * and btrfs will skip to write sb directly to keep
3170 * ERROR state on disk.
3172 * 2. when btrfs flips readonly just in btrfs_commit_super,
3173 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3174 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3175 * btrfs will cleanup all FS resources first and write sb then.
3177 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3178 ret = btrfs_commit_super(root);
3180 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3183 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3184 ret = btrfs_error_commit_super(root);
3186 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3189 btrfs_put_block_group_cache(fs_info);
3191 kthread_stop(fs_info->transaction_kthread);
3192 kthread_stop(fs_info->cleaner_kthread);
3194 fs_info->closing = 2;
3197 if (fs_info->delalloc_bytes) {
3198 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3199 (unsigned long long)fs_info->delalloc_bytes);
3201 if (fs_info->total_ref_cache_size) {
3202 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3203 (unsigned long long)fs_info->total_ref_cache_size);
3206 free_extent_buffer(fs_info->extent_root->node);
3207 free_extent_buffer(fs_info->extent_root->commit_root);
3208 free_extent_buffer(fs_info->tree_root->node);
3209 free_extent_buffer(fs_info->tree_root->commit_root);
3210 free_extent_buffer(fs_info->chunk_root->node);
3211 free_extent_buffer(fs_info->chunk_root->commit_root);
3212 free_extent_buffer(fs_info->dev_root->node);
3213 free_extent_buffer(fs_info->dev_root->commit_root);
3214 free_extent_buffer(fs_info->csum_root->node);
3215 free_extent_buffer(fs_info->csum_root->commit_root);
3217 btrfs_free_block_groups(fs_info);
3219 del_fs_roots(fs_info);
3221 iput(fs_info->btree_inode);
3223 btrfs_stop_workers(&fs_info->generic_worker);
3224 btrfs_stop_workers(&fs_info->fixup_workers);
3225 btrfs_stop_workers(&fs_info->delalloc_workers);
3226 btrfs_stop_workers(&fs_info->workers);
3227 btrfs_stop_workers(&fs_info->endio_workers);
3228 btrfs_stop_workers(&fs_info->endio_meta_workers);
3229 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3230 btrfs_stop_workers(&fs_info->endio_write_workers);
3231 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3232 btrfs_stop_workers(&fs_info->submit_workers);
3233 btrfs_stop_workers(&fs_info->delayed_workers);
3234 btrfs_stop_workers(&fs_info->caching_workers);
3235 btrfs_stop_workers(&fs_info->readahead_workers);
3237 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3238 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3239 btrfsic_unmount(root, fs_info->fs_devices);
3242 btrfs_close_devices(fs_info->fs_devices);
3243 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3245 bdi_destroy(&fs_info->bdi);
3246 cleanup_srcu_struct(&fs_info->subvol_srcu);
3251 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3255 struct inode *btree_inode = buf->pages[0]->mapping->host;
3257 ret = extent_buffer_uptodate(buf);
3261 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3262 parent_transid, atomic);
3268 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3270 return set_extent_buffer_uptodate(buf);
3273 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3275 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3276 u64 transid = btrfs_header_generation(buf);
3279 btrfs_assert_tree_locked(buf);
3280 if (transid != root->fs_info->generation) {
3281 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3282 "found %llu running %llu\n",
3283 (unsigned long long)buf->start,
3284 (unsigned long long)transid,
3285 (unsigned long long)root->fs_info->generation);
3288 was_dirty = set_extent_buffer_dirty(buf);
3290 spin_lock(&root->fs_info->delalloc_lock);
3291 root->fs_info->dirty_metadata_bytes += buf->len;
3292 spin_unlock(&root->fs_info->delalloc_lock);
3296 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3299 * looks as though older kernels can get into trouble with
3300 * this code, they end up stuck in balance_dirty_pages forever
3303 unsigned long thresh = 32 * 1024 * 1024;
3305 if (current->flags & PF_MEMALLOC)
3308 btrfs_balance_delayed_items(root);
3310 num_dirty = root->fs_info->dirty_metadata_bytes;
3312 if (num_dirty > thresh) {
3313 balance_dirty_pages_ratelimited_nr(
3314 root->fs_info->btree_inode->i_mapping, 1);
3319 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3322 * looks as though older kernels can get into trouble with
3323 * this code, they end up stuck in balance_dirty_pages forever
3326 unsigned long thresh = 32 * 1024 * 1024;
3328 if (current->flags & PF_MEMALLOC)
3331 num_dirty = root->fs_info->dirty_metadata_bytes;
3333 if (num_dirty > thresh) {
3334 balance_dirty_pages_ratelimited_nr(
3335 root->fs_info->btree_inode->i_mapping, 1);
3340 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3342 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3343 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3346 int btree_lock_page_hook(struct page *page, void *data,
3347 void (*flush_fn)(void *))
3349 struct inode *inode = page->mapping->host;
3350 struct btrfs_root *root = BTRFS_I(inode)->root;
3351 struct extent_buffer *eb;
3354 * We culled this eb but the page is still hanging out on the mapping,
3357 if (!PagePrivate(page))
3360 eb = (struct extent_buffer *)page->private;
3365 if (page != eb->pages[0])
3368 if (!btrfs_try_tree_write_lock(eb)) {
3370 btrfs_tree_lock(eb);
3372 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3374 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3375 spin_lock(&root->fs_info->delalloc_lock);
3376 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3377 root->fs_info->dirty_metadata_bytes -= eb->len;
3380 spin_unlock(&root->fs_info->delalloc_lock);
3383 btrfs_tree_unlock(eb);
3385 if (!trylock_page(page)) {
3392 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3395 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3396 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3403 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3404 printk(KERN_WARNING "warning: mount fs with errors, "
3405 "running btrfsck is recommended\n");
3411 int btrfs_error_commit_super(struct btrfs_root *root)
3415 mutex_lock(&root->fs_info->cleaner_mutex);
3416 btrfs_run_delayed_iputs(root);
3417 mutex_unlock(&root->fs_info->cleaner_mutex);
3419 down_write(&root->fs_info->cleanup_work_sem);
3420 up_write(&root->fs_info->cleanup_work_sem);
3422 /* cleanup FS via transaction */
3423 btrfs_cleanup_transaction(root);
3425 ret = write_ctree_super(NULL, root, 0);
3430 static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3432 struct btrfs_inode *btrfs_inode;
3433 struct list_head splice;
3435 INIT_LIST_HEAD(&splice);
3437 mutex_lock(&root->fs_info->ordered_operations_mutex);
3438 spin_lock(&root->fs_info->ordered_extent_lock);
3440 list_splice_init(&root->fs_info->ordered_operations, &splice);
3441 while (!list_empty(&splice)) {
3442 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3443 ordered_operations);
3445 list_del_init(&btrfs_inode->ordered_operations);
3447 btrfs_invalidate_inodes(btrfs_inode->root);
3450 spin_unlock(&root->fs_info->ordered_extent_lock);
3451 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3454 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3456 struct list_head splice;
3457 struct btrfs_ordered_extent *ordered;
3458 struct inode *inode;
3460 INIT_LIST_HEAD(&splice);
3462 spin_lock(&root->fs_info->ordered_extent_lock);
3464 list_splice_init(&root->fs_info->ordered_extents, &splice);
3465 while (!list_empty(&splice)) {
3466 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3469 list_del_init(&ordered->root_extent_list);
3470 atomic_inc(&ordered->refs);
3472 /* the inode may be getting freed (in sys_unlink path). */
3473 inode = igrab(ordered->inode);
3475 spin_unlock(&root->fs_info->ordered_extent_lock);
3479 atomic_set(&ordered->refs, 1);
3480 btrfs_put_ordered_extent(ordered);
3482 spin_lock(&root->fs_info->ordered_extent_lock);
3485 spin_unlock(&root->fs_info->ordered_extent_lock);
3488 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3489 struct btrfs_root *root)
3491 struct rb_node *node;
3492 struct btrfs_delayed_ref_root *delayed_refs;
3493 struct btrfs_delayed_ref_node *ref;
3496 delayed_refs = &trans->delayed_refs;
3498 spin_lock(&delayed_refs->lock);
3499 if (delayed_refs->num_entries == 0) {
3500 spin_unlock(&delayed_refs->lock);
3501 printk(KERN_INFO "delayed_refs has NO entry\n");
3505 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3506 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3508 atomic_set(&ref->refs, 1);
3509 if (btrfs_delayed_ref_is_head(ref)) {
3510 struct btrfs_delayed_ref_head *head;
3512 head = btrfs_delayed_node_to_head(ref);
3513 if (!mutex_trylock(&head->mutex)) {
3514 atomic_inc(&ref->refs);
3515 spin_unlock(&delayed_refs->lock);
3517 /* Need to wait for the delayed ref to run */
3518 mutex_lock(&head->mutex);
3519 mutex_unlock(&head->mutex);
3520 btrfs_put_delayed_ref(ref);
3522 spin_lock(&delayed_refs->lock);
3526 kfree(head->extent_op);
3527 delayed_refs->num_heads--;
3528 if (list_empty(&head->cluster))
3529 delayed_refs->num_heads_ready--;
3530 list_del_init(&head->cluster);
3533 rb_erase(&ref->rb_node, &delayed_refs->root);
3534 delayed_refs->num_entries--;
3536 spin_unlock(&delayed_refs->lock);
3537 btrfs_put_delayed_ref(ref);
3540 spin_lock(&delayed_refs->lock);
3543 spin_unlock(&delayed_refs->lock);
3548 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3550 struct btrfs_pending_snapshot *snapshot;
3551 struct list_head splice;
3553 INIT_LIST_HEAD(&splice);
3555 list_splice_init(&t->pending_snapshots, &splice);
3557 while (!list_empty(&splice)) {
3558 snapshot = list_entry(splice.next,
3559 struct btrfs_pending_snapshot,
3562 list_del_init(&snapshot->list);
3568 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3570 struct btrfs_inode *btrfs_inode;
3571 struct list_head splice;
3573 INIT_LIST_HEAD(&splice);
3575 spin_lock(&root->fs_info->delalloc_lock);
3576 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3578 while (!list_empty(&splice)) {
3579 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3582 list_del_init(&btrfs_inode->delalloc_inodes);
3584 btrfs_invalidate_inodes(btrfs_inode->root);
3587 spin_unlock(&root->fs_info->delalloc_lock);
3590 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3591 struct extent_io_tree *dirty_pages,
3596 struct inode *btree_inode = root->fs_info->btree_inode;
3597 struct extent_buffer *eb;
3601 unsigned long index;
3604 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3609 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3610 while (start <= end) {
3611 index = start >> PAGE_CACHE_SHIFT;
3612 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3613 page = find_get_page(btree_inode->i_mapping, index);
3616 offset = page_offset(page);
3618 spin_lock(&dirty_pages->buffer_lock);
3619 eb = radix_tree_lookup(
3620 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3621 offset >> PAGE_CACHE_SHIFT);
3622 spin_unlock(&dirty_pages->buffer_lock);
3624 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3626 if (PageWriteback(page))
3627 end_page_writeback(page);
3630 if (PageDirty(page)) {
3631 clear_page_dirty_for_io(page);
3632 spin_lock_irq(&page->mapping->tree_lock);
3633 radix_tree_tag_clear(&page->mapping->page_tree,
3635 PAGECACHE_TAG_DIRTY);
3636 spin_unlock_irq(&page->mapping->tree_lock);
3640 page_cache_release(page);
3647 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3648 struct extent_io_tree *pinned_extents)
3650 struct extent_io_tree *unpin;
3656 unpin = pinned_extents;
3659 ret = find_first_extent_bit(unpin, 0, &start, &end,
3665 if (btrfs_test_opt(root, DISCARD))
3666 ret = btrfs_error_discard_extent(root, start,
3670 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3671 btrfs_error_unpin_extent_range(root, start, end);
3676 if (unpin == &root->fs_info->freed_extents[0])
3677 unpin = &root->fs_info->freed_extents[1];
3679 unpin = &root->fs_info->freed_extents[0];
3687 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3688 struct btrfs_root *root)
3690 btrfs_destroy_delayed_refs(cur_trans, root);
3691 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3692 cur_trans->dirty_pages.dirty_bytes);
3694 /* FIXME: cleanup wait for commit */
3695 cur_trans->in_commit = 1;
3696 cur_trans->blocked = 1;
3697 wake_up(&root->fs_info->transaction_blocked_wait);
3699 cur_trans->blocked = 0;
3700 wake_up(&root->fs_info->transaction_wait);
3702 cur_trans->commit_done = 1;
3703 wake_up(&cur_trans->commit_wait);
3705 btrfs_destroy_delayed_inodes(root);
3706 btrfs_assert_delayed_root_empty(root);
3708 btrfs_destroy_pending_snapshots(cur_trans);
3710 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3712 btrfs_destroy_pinned_extent(root,
3713 root->fs_info->pinned_extents);
3716 memset(cur_trans, 0, sizeof(*cur_trans));
3717 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3721 int btrfs_cleanup_transaction(struct btrfs_root *root)
3723 struct btrfs_transaction *t;
3726 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3728 spin_lock(&root->fs_info->trans_lock);
3729 list_splice_init(&root->fs_info->trans_list, &list);
3730 root->fs_info->trans_no_join = 1;
3731 spin_unlock(&root->fs_info->trans_lock);
3733 while (!list_empty(&list)) {
3734 t = list_entry(list.next, struct btrfs_transaction, list);
3738 btrfs_destroy_ordered_operations(root);
3740 btrfs_destroy_ordered_extents(root);
3742 btrfs_destroy_delayed_refs(t, root);
3744 btrfs_block_rsv_release(root,
3745 &root->fs_info->trans_block_rsv,
3746 t->dirty_pages.dirty_bytes);
3748 /* FIXME: cleanup wait for commit */
3751 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3752 wake_up(&root->fs_info->transaction_blocked_wait);
3755 if (waitqueue_active(&root->fs_info->transaction_wait))
3756 wake_up(&root->fs_info->transaction_wait);
3759 if (waitqueue_active(&t->commit_wait))
3760 wake_up(&t->commit_wait);
3762 btrfs_destroy_delayed_inodes(root);
3763 btrfs_assert_delayed_root_empty(root);
3765 btrfs_destroy_pending_snapshots(t);
3767 btrfs_destroy_delalloc_inodes(root);
3769 spin_lock(&root->fs_info->trans_lock);
3770 root->fs_info->running_transaction = NULL;
3771 spin_unlock(&root->fs_info->trans_lock);
3773 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3776 btrfs_destroy_pinned_extent(root,
3777 root->fs_info->pinned_extents);
3779 atomic_set(&t->use_count, 0);
3780 list_del_init(&t->list);
3781 memset(t, 0, sizeof(*t));
3782 kmem_cache_free(btrfs_transaction_cachep, t);
3785 spin_lock(&root->fs_info->trans_lock);
3786 root->fs_info->trans_no_join = 0;
3787 spin_unlock(&root->fs_info->trans_lock);
3788 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3793 static struct extent_io_ops btree_extent_io_ops = {
3794 .write_cache_pages_lock_hook = btree_lock_page_hook,
3795 .readpage_end_io_hook = btree_readpage_end_io_hook,
3796 .readpage_io_failed_hook = btree_io_failed_hook,
3797 .submit_bio_hook = btree_submit_bio_hook,
3798 /* note we're sharing with inode.c for the merge bio hook */
3799 .merge_bio_hook = btrfs_merge_bio_hook,