2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
48 static struct extent_io_ops btree_extent_io_ops;
49 static void end_workqueue_fn(struct btrfs_work *work);
50 static void free_fs_root(struct btrfs_root *root);
51 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
53 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
54 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_root *root);
57 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
58 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
60 struct extent_io_tree *dirty_pages,
62 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
63 struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_root *root);
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;
105 * Lockdep class keys for extent_buffer->lock's in this root. For a given
106 * eb, the lockdep key is determined by the btrfs_root it belongs to and
107 * the level the eb occupies in the tree.
109 * Different roots are used for different purposes and may nest inside each
110 * other and they require separate keysets. As lockdep keys should be
111 * static, assign keysets according to the purpose of the root as indicated
112 * by btrfs_root->objectid. This ensures that all special purpose roots
113 * have separate keysets.
115 * Lock-nesting across peer nodes is always done with the immediate parent
116 * node locked thus preventing deadlock. As lockdep doesn't know this, use
117 * subclass to avoid triggering lockdep warning in such cases.
119 * The key is set by the readpage_end_io_hook after the buffer has passed
120 * csum validation but before the pages are unlocked. It is also set by
121 * btrfs_init_new_buffer on freshly allocated blocks.
123 * We also add a check to make sure the highest level of the tree is the
124 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
125 * needs update as well.
127 #ifdef CONFIG_DEBUG_LOCK_ALLOC
128 # if BTRFS_MAX_LEVEL != 8
132 static struct btrfs_lockdep_keyset {
133 u64 id; /* root objectid */
134 const char *name_stem; /* lock name stem */
135 char names[BTRFS_MAX_LEVEL + 1][20];
136 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
137 } btrfs_lockdep_keysets[] = {
138 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
139 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
140 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
141 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
142 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
143 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
144 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
145 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
146 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
147 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
148 { .id = 0, .name_stem = "tree" },
151 void __init btrfs_init_lockdep(void)
155 /* initialize lockdep class names */
156 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
157 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
159 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
160 snprintf(ks->names[j], sizeof(ks->names[j]),
161 "btrfs-%s-%02d", ks->name_stem, j);
165 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
168 struct btrfs_lockdep_keyset *ks;
170 BUG_ON(level >= ARRAY_SIZE(ks->keys));
172 /* find the matching keyset, id 0 is the default entry */
173 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
174 if (ks->id == objectid)
177 lockdep_set_class_and_name(&eb->lock,
178 &ks->keys[level], ks->names[level]);
184 * extents on the btree inode are pretty simple, there's one extent
185 * that covers the entire device
187 static struct extent_map *btree_get_extent(struct inode *inode,
188 struct page *page, size_t pg_offset, u64 start, u64 len,
191 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
192 struct extent_map *em;
195 read_lock(&em_tree->lock);
196 em = lookup_extent_mapping(em_tree, start, len);
199 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
200 read_unlock(&em_tree->lock);
203 read_unlock(&em_tree->lock);
205 em = alloc_extent_map();
207 em = ERR_PTR(-ENOMEM);
212 em->block_len = (u64)-1;
214 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
216 write_lock(&em_tree->lock);
217 ret = add_extent_mapping(em_tree, em);
218 if (ret == -EEXIST) {
219 u64 failed_start = em->start;
220 u64 failed_len = em->len;
223 em = lookup_extent_mapping(em_tree, start, len);
227 em = lookup_extent_mapping(em_tree, failed_start,
235 write_unlock(&em_tree->lock);
243 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
245 return crc32c(seed, data, len);
248 void btrfs_csum_final(u32 crc, char *result)
250 put_unaligned_le32(~crc, result);
254 * compute the csum for a btree block, and either verify it or write it
255 * into the csum field of the block.
257 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263 unsigned long cur_len;
264 unsigned long offset = BTRFS_CSUM_SIZE;
266 unsigned long map_start;
267 unsigned long map_len;
270 unsigned long inline_result;
272 len = buf->len - offset;
274 err = map_private_extent_buffer(buf, offset, 32,
275 &kaddr, &map_start, &map_len);
278 cur_len = min(len, map_len - (offset - map_start));
279 crc = btrfs_csum_data(root, kaddr + offset - map_start,
284 if (csum_size > sizeof(inline_result)) {
285 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 result = (char *)&inline_result;
292 btrfs_csum_final(crc, result);
295 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 memcpy(&found, result, csum_size);
300 read_extent_buffer(buf, &val, 0, csum_size);
301 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
302 "failed on %llu wanted %X found %X "
304 root->fs_info->sb->s_id,
305 (unsigned long long)buf->start, val, found,
306 btrfs_header_level(buf));
307 if (result != (char *)&inline_result)
312 write_extent_buffer(buf, result, 0, csum_size);
314 if (result != (char *)&inline_result)
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
325 static int verify_parent_transid(struct extent_io_tree *io_tree,
326 struct extent_buffer *eb, u64 parent_transid)
328 struct extent_state *cached_state = NULL;
331 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
335 0, &cached_state, GFP_NOFS);
336 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
337 btrfs_header_generation(eb) == parent_transid) {
341 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
343 (unsigned long long)eb->start,
344 (unsigned long long)parent_transid,
345 (unsigned long long)btrfs_header_generation(eb));
347 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
349 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350 &cached_state, GFP_NOFS);
355 * helper to read a given tree block, doing retries as required when
356 * the checksums don't match and we have alternate mirrors to try.
358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359 struct extent_buffer *eb,
360 u64 start, u64 parent_transid)
362 struct extent_io_tree *io_tree;
367 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
368 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
370 ret = read_extent_buffer_pages(io_tree, eb, start,
372 btree_get_extent, mirror_num);
374 !verify_parent_transid(io_tree, eb, parent_transid))
378 * This buffer's crc is fine, but its contents are corrupted, so
379 * there is no reason to read the other copies, they won't be
382 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
385 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
391 if (mirror_num > num_copies)
398 * checksum a dirty tree block before IO. This has extra checks to make sure
399 * we only fill in the checksum field in the first page of a multi-page block
402 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
404 struct extent_io_tree *tree;
405 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
408 struct extent_buffer *eb;
411 tree = &BTRFS_I(page->mapping->host)->io_tree;
413 if (page->private == EXTENT_PAGE_PRIVATE) {
417 if (!page->private) {
421 len = page->private >> 2;
424 eb = alloc_extent_buffer(tree, start, len, page);
429 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
430 btrfs_header_generation(eb));
432 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
434 found_start = btrfs_header_bytenr(eb);
435 if (found_start != start) {
439 if (eb->first_page != page) {
443 if (!PageUptodate(page)) {
447 csum_tree_block(root, eb, 0);
449 free_extent_buffer(eb);
454 static int check_tree_block_fsid(struct btrfs_root *root,
455 struct extent_buffer *eb)
457 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
458 u8 fsid[BTRFS_UUID_SIZE];
461 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
464 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
468 fs_devices = fs_devices->seed;
473 #define CORRUPT(reason, eb, root, slot) \
474 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
475 "root=%llu, slot=%d\n", reason, \
476 (unsigned long long)btrfs_header_bytenr(eb), \
477 (unsigned long long)root->objectid, slot)
479 static noinline int check_leaf(struct btrfs_root *root,
480 struct extent_buffer *leaf)
482 struct btrfs_key key;
483 struct btrfs_key leaf_key;
484 u32 nritems = btrfs_header_nritems(leaf);
490 /* Check the 0 item */
491 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
492 BTRFS_LEAF_DATA_SIZE(root)) {
493 CORRUPT("invalid item offset size pair", leaf, root, 0);
498 * Check to make sure each items keys are in the correct order and their
499 * offsets make sense. We only have to loop through nritems-1 because
500 * we check the current slot against the next slot, which verifies the
501 * next slot's offset+size makes sense and that the current's slot
504 for (slot = 0; slot < nritems - 1; slot++) {
505 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
506 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
508 /* Make sure the keys are in the right order */
509 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
510 CORRUPT("bad key order", leaf, root, slot);
515 * Make sure the offset and ends are right, remember that the
516 * item data starts at the end of the leaf and grows towards the
519 if (btrfs_item_offset_nr(leaf, slot) !=
520 btrfs_item_end_nr(leaf, slot + 1)) {
521 CORRUPT("slot offset bad", leaf, root, slot);
526 * Check to make sure that we don't point outside of the leaf,
527 * just incase all the items are consistent to eachother, but
528 * all point outside of the leaf.
530 if (btrfs_item_end_nr(leaf, slot) >
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("slot end outside of leaf", leaf, root, slot);
540 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
541 struct extent_state *state)
543 struct extent_io_tree *tree;
547 struct extent_buffer *eb;
548 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
551 tree = &BTRFS_I(page->mapping->host)->io_tree;
552 if (page->private == EXTENT_PAGE_PRIVATE)
557 len = page->private >> 2;
560 eb = alloc_extent_buffer(tree, start, len, page);
566 found_start = btrfs_header_bytenr(eb);
567 if (found_start != start) {
568 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
570 (unsigned long long)found_start,
571 (unsigned long long)eb->start);
575 if (eb->first_page != page) {
576 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
577 eb->first_page->index, page->index);
582 if (check_tree_block_fsid(root, eb)) {
583 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
584 (unsigned long long)eb->start);
588 found_level = btrfs_header_level(eb);
590 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
593 ret = csum_tree_block(root, eb, 1);
600 * If this is a leaf block and it is corrupt, set the corrupt bit so
601 * that we don't try and read the other copies of this block, just
604 if (found_level == 0 && check_leaf(root, eb)) {
605 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
609 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
610 end = eb->start + end - 1;
612 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
613 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
614 btree_readahead_hook(root, eb, eb->start, ret);
617 free_extent_buffer(eb);
622 static int btree_io_failed_hook(struct bio *failed_bio,
623 struct page *page, u64 start, u64 end,
624 int mirror_num, struct extent_state *state)
626 struct extent_io_tree *tree;
628 struct extent_buffer *eb;
629 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
631 tree = &BTRFS_I(page->mapping->host)->io_tree;
632 if (page->private == EXTENT_PAGE_PRIVATE)
637 len = page->private >> 2;
640 eb = alloc_extent_buffer(tree, start, len, page);
644 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
645 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
646 btree_readahead_hook(root, eb, eb->start, -EIO);
648 free_extent_buffer(eb);
651 return -EIO; /* we fixed nothing */
654 static void end_workqueue_bio(struct bio *bio, int err)
656 struct end_io_wq *end_io_wq = bio->bi_private;
657 struct btrfs_fs_info *fs_info;
659 fs_info = end_io_wq->info;
660 end_io_wq->error = err;
661 end_io_wq->work.func = end_workqueue_fn;
662 end_io_wq->work.flags = 0;
664 if (bio->bi_rw & REQ_WRITE) {
665 if (end_io_wq->metadata == 1)
666 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
668 else if (end_io_wq->metadata == 2)
669 btrfs_queue_worker(&fs_info->endio_freespace_worker,
672 btrfs_queue_worker(&fs_info->endio_write_workers,
675 if (end_io_wq->metadata)
676 btrfs_queue_worker(&fs_info->endio_meta_workers,
679 btrfs_queue_worker(&fs_info->endio_workers,
685 * For the metadata arg you want
688 * 1 - if normal metadta
689 * 2 - if writing to the free space cache area
691 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
694 struct end_io_wq *end_io_wq;
695 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
699 end_io_wq->private = bio->bi_private;
700 end_io_wq->end_io = bio->bi_end_io;
701 end_io_wq->info = info;
702 end_io_wq->error = 0;
703 end_io_wq->bio = bio;
704 end_io_wq->metadata = metadata;
706 bio->bi_private = end_io_wq;
707 bio->bi_end_io = end_workqueue_bio;
711 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
713 unsigned long limit = min_t(unsigned long,
714 info->workers.max_workers,
715 info->fs_devices->open_devices);
719 static void run_one_async_start(struct btrfs_work *work)
721 struct async_submit_bio *async;
723 async = container_of(work, struct async_submit_bio, work);
724 async->submit_bio_start(async->inode, async->rw, async->bio,
725 async->mirror_num, async->bio_flags,
729 static void run_one_async_done(struct btrfs_work *work)
731 struct btrfs_fs_info *fs_info;
732 struct async_submit_bio *async;
735 async = container_of(work, struct async_submit_bio, work);
736 fs_info = BTRFS_I(async->inode)->root->fs_info;
738 limit = btrfs_async_submit_limit(fs_info);
739 limit = limit * 2 / 3;
741 atomic_dec(&fs_info->nr_async_submits);
743 if (atomic_read(&fs_info->nr_async_submits) < limit &&
744 waitqueue_active(&fs_info->async_submit_wait))
745 wake_up(&fs_info->async_submit_wait);
747 async->submit_bio_done(async->inode, async->rw, async->bio,
748 async->mirror_num, async->bio_flags,
752 static void run_one_async_free(struct btrfs_work *work)
754 struct async_submit_bio *async;
756 async = container_of(work, struct async_submit_bio, work);
760 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
761 int rw, struct bio *bio, int mirror_num,
762 unsigned long bio_flags,
764 extent_submit_bio_hook_t *submit_bio_start,
765 extent_submit_bio_hook_t *submit_bio_done)
767 struct async_submit_bio *async;
769 async = kmalloc(sizeof(*async), GFP_NOFS);
773 async->inode = inode;
776 async->mirror_num = mirror_num;
777 async->submit_bio_start = submit_bio_start;
778 async->submit_bio_done = submit_bio_done;
780 async->work.func = run_one_async_start;
781 async->work.ordered_func = run_one_async_done;
782 async->work.ordered_free = run_one_async_free;
784 async->work.flags = 0;
785 async->bio_flags = bio_flags;
786 async->bio_offset = bio_offset;
788 atomic_inc(&fs_info->nr_async_submits);
791 btrfs_set_work_high_prio(&async->work);
793 btrfs_queue_worker(&fs_info->workers, &async->work);
795 while (atomic_read(&fs_info->async_submit_draining) &&
796 atomic_read(&fs_info->nr_async_submits)) {
797 wait_event(fs_info->async_submit_wait,
798 (atomic_read(&fs_info->nr_async_submits) == 0));
804 static int btree_csum_one_bio(struct bio *bio)
806 struct bio_vec *bvec = bio->bi_io_vec;
808 struct btrfs_root *root;
810 WARN_ON(bio->bi_vcnt <= 0);
811 while (bio_index < bio->bi_vcnt) {
812 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
813 csum_dirty_buffer(root, bvec->bv_page);
820 static int __btree_submit_bio_start(struct inode *inode, int rw,
821 struct bio *bio, int mirror_num,
822 unsigned long bio_flags,
826 * when we're called for a write, we're already in the async
827 * submission context. Just jump into btrfs_map_bio
829 btree_csum_one_bio(bio);
833 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
834 int mirror_num, unsigned long bio_flags,
838 * when we're called for a write, we're already in the async
839 * submission context. Just jump into btrfs_map_bio
841 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
844 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
845 int mirror_num, unsigned long bio_flags,
850 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
854 if (!(rw & REQ_WRITE)) {
856 * called for a read, do the setup so that checksum validation
857 * can happen in the async kernel threads
859 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
864 * kthread helpers are used to submit writes so that checksumming
865 * can happen in parallel across all CPUs
867 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
868 inode, rw, bio, mirror_num, 0,
870 __btree_submit_bio_start,
871 __btree_submit_bio_done);
874 #ifdef CONFIG_MIGRATION
875 static int btree_migratepage(struct address_space *mapping,
876 struct page *newpage, struct page *page)
879 * we can't safely write a btree page from here,
880 * we haven't done the locking hook
885 * Buffers may be managed in a filesystem specific way.
886 * We must have no buffers or drop them.
888 if (page_has_private(page) &&
889 !try_to_release_page(page, GFP_KERNEL))
891 return migrate_page(mapping, newpage, page);
895 static int btree_writepage(struct page *page, struct writeback_control *wbc)
897 struct extent_io_tree *tree;
898 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
899 struct extent_buffer *eb;
902 tree = &BTRFS_I(page->mapping->host)->io_tree;
903 if (!(current->flags & PF_MEMALLOC)) {
904 return extent_write_full_page(tree, page,
905 btree_get_extent, wbc);
908 redirty_page_for_writepage(wbc, page);
909 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
912 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
914 spin_lock(&root->fs_info->delalloc_lock);
915 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
916 spin_unlock(&root->fs_info->delalloc_lock);
918 free_extent_buffer(eb);
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 extent_writepages(tree, mapping, btree_get_extent, 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 struct extent_io_tree *tree;
955 struct extent_map_tree *map;
958 if (PageWriteback(page) || PageDirty(page))
961 tree = &BTRFS_I(page->mapping->host)->io_tree;
962 map = &BTRFS_I(page->mapping->host)->extent_tree;
965 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
966 * slab allocation from alloc_extent_state down the callchain where
967 * it'd hit a BUG_ON as those flags are not allowed.
969 gfp_flags &= ~GFP_SLAB_BUG_MASK;
971 ret = try_release_extent_state(map, tree, page, gfp_flags);
975 ret = try_release_extent_buffer(tree, page);
977 ClearPagePrivate(page);
978 set_page_private(page, 0);
979 page_cache_release(page);
985 static void btree_invalidatepage(struct page *page, unsigned long offset)
987 struct extent_io_tree *tree;
988 tree = &BTRFS_I(page->mapping->host)->io_tree;
989 extent_invalidatepage(tree, page, offset);
990 btree_releasepage(page, GFP_NOFS);
991 if (PagePrivate(page)) {
992 printk(KERN_WARNING "btrfs warning page private not zero "
993 "on page %llu\n", (unsigned long long)page_offset(page));
994 ClearPagePrivate(page);
995 set_page_private(page, 0);
996 page_cache_release(page);
1000 static const struct address_space_operations btree_aops = {
1001 .readpage = btree_readpage,
1002 .writepage = btree_writepage,
1003 .writepages = btree_writepages,
1004 .releasepage = btree_releasepage,
1005 .invalidatepage = btree_invalidatepage,
1006 #ifdef CONFIG_MIGRATION
1007 .migratepage = btree_migratepage,
1011 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1014 struct extent_buffer *buf = NULL;
1015 struct inode *btree_inode = root->fs_info->btree_inode;
1018 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1021 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1022 buf, 0, WAIT_NONE, btree_get_extent, 0);
1023 free_extent_buffer(buf);
1027 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1028 int mirror_num, struct extent_buffer **eb)
1030 struct extent_buffer *buf = NULL;
1031 struct inode *btree_inode = root->fs_info->btree_inode;
1032 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1035 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1039 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1041 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1042 btree_get_extent, mirror_num);
1044 free_extent_buffer(buf);
1048 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1049 free_extent_buffer(buf);
1051 } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1054 free_extent_buffer(buf);
1059 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1060 u64 bytenr, u32 blocksize)
1062 struct inode *btree_inode = root->fs_info->btree_inode;
1063 struct extent_buffer *eb;
1064 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1069 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1070 u64 bytenr, u32 blocksize)
1072 struct inode *btree_inode = root->fs_info->btree_inode;
1073 struct extent_buffer *eb;
1075 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1076 bytenr, blocksize, NULL);
1081 int btrfs_write_tree_block(struct extent_buffer *buf)
1083 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1084 buf->start + buf->len - 1);
1087 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1089 return filemap_fdatawait_range(buf->first_page->mapping,
1090 buf->start, buf->start + buf->len - 1);
1093 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1094 u32 blocksize, u64 parent_transid)
1096 struct extent_buffer *buf = NULL;
1099 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1103 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1106 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1111 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1112 struct extent_buffer *buf)
1114 struct inode *btree_inode = root->fs_info->btree_inode;
1115 if (btrfs_header_generation(buf) ==
1116 root->fs_info->running_transaction->transid) {
1117 btrfs_assert_tree_locked(buf);
1119 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1120 spin_lock(&root->fs_info->delalloc_lock);
1121 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1122 root->fs_info->dirty_metadata_bytes -= buf->len;
1125 spin_unlock(&root->fs_info->delalloc_lock);
1128 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1129 btrfs_set_lock_blocking(buf);
1130 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1136 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1137 u32 stripesize, struct btrfs_root *root,
1138 struct btrfs_fs_info *fs_info,
1142 root->commit_root = NULL;
1143 root->sectorsize = sectorsize;
1144 root->nodesize = nodesize;
1145 root->leafsize = leafsize;
1146 root->stripesize = stripesize;
1148 root->track_dirty = 0;
1150 root->orphan_item_inserted = 0;
1151 root->orphan_cleanup_state = 0;
1153 root->fs_info = fs_info;
1154 root->objectid = objectid;
1155 root->last_trans = 0;
1156 root->highest_objectid = 0;
1158 root->inode_tree = RB_ROOT;
1159 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1160 root->block_rsv = NULL;
1161 root->orphan_block_rsv = NULL;
1163 INIT_LIST_HEAD(&root->dirty_list);
1164 INIT_LIST_HEAD(&root->orphan_list);
1165 INIT_LIST_HEAD(&root->root_list);
1166 spin_lock_init(&root->orphan_lock);
1167 spin_lock_init(&root->inode_lock);
1168 spin_lock_init(&root->accounting_lock);
1169 mutex_init(&root->objectid_mutex);
1170 mutex_init(&root->log_mutex);
1171 init_waitqueue_head(&root->log_writer_wait);
1172 init_waitqueue_head(&root->log_commit_wait[0]);
1173 init_waitqueue_head(&root->log_commit_wait[1]);
1174 atomic_set(&root->log_commit[0], 0);
1175 atomic_set(&root->log_commit[1], 0);
1176 atomic_set(&root->log_writers, 0);
1177 root->log_batch = 0;
1178 root->log_transid = 0;
1179 root->last_log_commit = 0;
1180 extent_io_tree_init(&root->dirty_log_pages,
1181 fs_info->btree_inode->i_mapping);
1183 memset(&root->root_key, 0, sizeof(root->root_key));
1184 memset(&root->root_item, 0, sizeof(root->root_item));
1185 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1186 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1187 root->defrag_trans_start = fs_info->generation;
1188 init_completion(&root->kobj_unregister);
1189 root->defrag_running = 0;
1190 root->root_key.objectid = objectid;
1195 static int find_and_setup_root(struct btrfs_root *tree_root,
1196 struct btrfs_fs_info *fs_info,
1198 struct btrfs_root *root)
1204 __setup_root(tree_root->nodesize, tree_root->leafsize,
1205 tree_root->sectorsize, tree_root->stripesize,
1206 root, fs_info, objectid);
1207 ret = btrfs_find_last_root(tree_root, objectid,
1208 &root->root_item, &root->root_key);
1213 generation = btrfs_root_generation(&root->root_item);
1214 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1215 root->commit_root = NULL;
1216 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1217 blocksize, generation);
1218 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1219 free_extent_buffer(root->node);
1223 root->commit_root = btrfs_root_node(root);
1227 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1228 struct btrfs_fs_info *fs_info)
1230 struct btrfs_root *root;
1231 struct btrfs_root *tree_root = fs_info->tree_root;
1232 struct extent_buffer *leaf;
1234 root = kzalloc(sizeof(*root), GFP_NOFS);
1236 return ERR_PTR(-ENOMEM);
1238 __setup_root(tree_root->nodesize, tree_root->leafsize,
1239 tree_root->sectorsize, tree_root->stripesize,
1240 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1242 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1243 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1244 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1246 * log trees do not get reference counted because they go away
1247 * before a real commit is actually done. They do store pointers
1248 * to file data extents, and those reference counts still get
1249 * updated (along with back refs to the log tree).
1253 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1254 BTRFS_TREE_LOG_OBJECTID, NULL,
1258 return ERR_CAST(leaf);
1261 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1262 btrfs_set_header_bytenr(leaf, leaf->start);
1263 btrfs_set_header_generation(leaf, trans->transid);
1264 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1265 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1268 write_extent_buffer(root->node, root->fs_info->fsid,
1269 (unsigned long)btrfs_header_fsid(root->node),
1271 btrfs_mark_buffer_dirty(root->node);
1272 btrfs_tree_unlock(root->node);
1276 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1277 struct btrfs_fs_info *fs_info)
1279 struct btrfs_root *log_root;
1281 log_root = alloc_log_tree(trans, fs_info);
1282 if (IS_ERR(log_root))
1283 return PTR_ERR(log_root);
1284 WARN_ON(fs_info->log_root_tree);
1285 fs_info->log_root_tree = log_root;
1289 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1290 struct btrfs_root *root)
1292 struct btrfs_root *log_root;
1293 struct btrfs_inode_item *inode_item;
1295 log_root = alloc_log_tree(trans, root->fs_info);
1296 if (IS_ERR(log_root))
1297 return PTR_ERR(log_root);
1299 log_root->last_trans = trans->transid;
1300 log_root->root_key.offset = root->root_key.objectid;
1302 inode_item = &log_root->root_item.inode;
1303 inode_item->generation = cpu_to_le64(1);
1304 inode_item->size = cpu_to_le64(3);
1305 inode_item->nlink = cpu_to_le32(1);
1306 inode_item->nbytes = cpu_to_le64(root->leafsize);
1307 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1309 btrfs_set_root_node(&log_root->root_item, log_root->node);
1311 WARN_ON(root->log_root);
1312 root->log_root = log_root;
1313 root->log_transid = 0;
1314 root->last_log_commit = 0;
1318 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1319 struct btrfs_key *location)
1321 struct btrfs_root *root;
1322 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1323 struct btrfs_path *path;
1324 struct extent_buffer *l;
1329 root = kzalloc(sizeof(*root), GFP_NOFS);
1331 return ERR_PTR(-ENOMEM);
1332 if (location->offset == (u64)-1) {
1333 ret = find_and_setup_root(tree_root, fs_info,
1334 location->objectid, root);
1337 return ERR_PTR(ret);
1342 __setup_root(tree_root->nodesize, tree_root->leafsize,
1343 tree_root->sectorsize, tree_root->stripesize,
1344 root, fs_info, location->objectid);
1346 path = btrfs_alloc_path();
1349 return ERR_PTR(-ENOMEM);
1351 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1354 read_extent_buffer(l, &root->root_item,
1355 btrfs_item_ptr_offset(l, path->slots[0]),
1356 sizeof(root->root_item));
1357 memcpy(&root->root_key, location, sizeof(*location));
1359 btrfs_free_path(path);
1364 return ERR_PTR(ret);
1367 generation = btrfs_root_generation(&root->root_item);
1368 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1369 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1370 blocksize, generation);
1371 root->commit_root = btrfs_root_node(root);
1372 BUG_ON(!root->node);
1374 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1376 btrfs_check_and_init_root_item(&root->root_item);
1382 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1383 struct btrfs_key *location)
1385 struct btrfs_root *root;
1388 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1389 return fs_info->tree_root;
1390 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1391 return fs_info->extent_root;
1392 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1393 return fs_info->chunk_root;
1394 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1395 return fs_info->dev_root;
1396 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1397 return fs_info->csum_root;
1399 spin_lock(&fs_info->fs_roots_radix_lock);
1400 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1401 (unsigned long)location->objectid);
1402 spin_unlock(&fs_info->fs_roots_radix_lock);
1406 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1410 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1411 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1413 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1418 btrfs_init_free_ino_ctl(root);
1419 mutex_init(&root->fs_commit_mutex);
1420 spin_lock_init(&root->cache_lock);
1421 init_waitqueue_head(&root->cache_wait);
1423 ret = get_anon_bdev(&root->anon_dev);
1427 if (btrfs_root_refs(&root->root_item) == 0) {
1432 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1436 root->orphan_item_inserted = 1;
1438 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1442 spin_lock(&fs_info->fs_roots_radix_lock);
1443 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1444 (unsigned long)root->root_key.objectid,
1449 spin_unlock(&fs_info->fs_roots_radix_lock);
1450 radix_tree_preload_end();
1452 if (ret == -EEXIST) {
1459 ret = btrfs_find_dead_roots(fs_info->tree_root,
1460 root->root_key.objectid);
1465 return ERR_PTR(ret);
1468 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1470 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1472 struct btrfs_device *device;
1473 struct backing_dev_info *bdi;
1476 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1479 bdi = blk_get_backing_dev_info(device->bdev);
1480 if (bdi && bdi_congested(bdi, bdi_bits)) {
1490 * If this fails, caller must call bdi_destroy() to get rid of the
1493 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1497 bdi->capabilities = BDI_CAP_MAP_COPY;
1498 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1502 bdi->ra_pages = default_backing_dev_info.ra_pages;
1503 bdi->congested_fn = btrfs_congested_fn;
1504 bdi->congested_data = info;
1508 static int bio_ready_for_csum(struct bio *bio)
1514 struct extent_io_tree *io_tree = NULL;
1515 struct bio_vec *bvec;
1519 bio_for_each_segment(bvec, bio, i) {
1520 page = bvec->bv_page;
1521 if (page->private == EXTENT_PAGE_PRIVATE) {
1522 length += bvec->bv_len;
1525 if (!page->private) {
1526 length += bvec->bv_len;
1529 length = bvec->bv_len;
1530 buf_len = page->private >> 2;
1531 start = page_offset(page) + bvec->bv_offset;
1532 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1534 /* are we fully contained in this bio? */
1535 if (buf_len <= length)
1538 ret = extent_range_uptodate(io_tree, start + length,
1539 start + buf_len - 1);
1544 * called by the kthread helper functions to finally call the bio end_io
1545 * functions. This is where read checksum verification actually happens
1547 static void end_workqueue_fn(struct btrfs_work *work)
1550 struct end_io_wq *end_io_wq;
1551 struct btrfs_fs_info *fs_info;
1554 end_io_wq = container_of(work, struct end_io_wq, work);
1555 bio = end_io_wq->bio;
1556 fs_info = end_io_wq->info;
1558 /* metadata bio reads are special because the whole tree block must
1559 * be checksummed at once. This makes sure the entire block is in
1560 * ram and up to date before trying to verify things. For
1561 * blocksize <= pagesize, it is basically a noop
1563 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1564 !bio_ready_for_csum(bio)) {
1565 btrfs_queue_worker(&fs_info->endio_meta_workers,
1569 error = end_io_wq->error;
1570 bio->bi_private = end_io_wq->private;
1571 bio->bi_end_io = end_io_wq->end_io;
1573 bio_endio(bio, error);
1576 static int cleaner_kthread(void *arg)
1578 struct btrfs_root *root = arg;
1581 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1583 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1584 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1585 btrfs_run_delayed_iputs(root);
1586 btrfs_clean_old_snapshots(root);
1587 mutex_unlock(&root->fs_info->cleaner_mutex);
1588 btrfs_run_defrag_inodes(root->fs_info);
1591 if (freezing(current)) {
1594 set_current_state(TASK_INTERRUPTIBLE);
1595 if (!kthread_should_stop())
1597 __set_current_state(TASK_RUNNING);
1599 } while (!kthread_should_stop());
1603 static int transaction_kthread(void *arg)
1605 struct btrfs_root *root = arg;
1606 struct btrfs_trans_handle *trans;
1607 struct btrfs_transaction *cur;
1610 unsigned long delay;
1615 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1616 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1618 spin_lock(&root->fs_info->trans_lock);
1619 cur = root->fs_info->running_transaction;
1621 spin_unlock(&root->fs_info->trans_lock);
1625 now = get_seconds();
1626 if (!cur->blocked &&
1627 (now < cur->start_time || now - cur->start_time < 30)) {
1628 spin_unlock(&root->fs_info->trans_lock);
1632 transid = cur->transid;
1633 spin_unlock(&root->fs_info->trans_lock);
1635 trans = btrfs_join_transaction(root);
1636 BUG_ON(IS_ERR(trans));
1637 if (transid == trans->transid) {
1638 ret = btrfs_commit_transaction(trans, root);
1641 btrfs_end_transaction(trans, root);
1644 wake_up_process(root->fs_info->cleaner_kthread);
1645 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1647 if (freezing(current)) {
1650 set_current_state(TASK_INTERRUPTIBLE);
1651 if (!kthread_should_stop() &&
1652 !btrfs_transaction_blocked(root->fs_info))
1653 schedule_timeout(delay);
1654 __set_current_state(TASK_RUNNING);
1656 } while (!kthread_should_stop());
1661 * this will find the highest generation in the array of
1662 * root backups. The index of the highest array is returned,
1663 * or -1 if we can't find anything.
1665 * We check to make sure the array is valid by comparing the
1666 * generation of the latest root in the array with the generation
1667 * in the super block. If they don't match we pitch it.
1669 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1672 int newest_index = -1;
1673 struct btrfs_root_backup *root_backup;
1676 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1677 root_backup = info->super_copy->super_roots + i;
1678 cur = btrfs_backup_tree_root_gen(root_backup);
1679 if (cur == newest_gen)
1683 /* check to see if we actually wrapped around */
1684 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1685 root_backup = info->super_copy->super_roots;
1686 cur = btrfs_backup_tree_root_gen(root_backup);
1687 if (cur == newest_gen)
1690 return newest_index;
1695 * find the oldest backup so we know where to store new entries
1696 * in the backup array. This will set the backup_root_index
1697 * field in the fs_info struct
1699 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1702 int newest_index = -1;
1704 newest_index = find_newest_super_backup(info, newest_gen);
1705 /* if there was garbage in there, just move along */
1706 if (newest_index == -1) {
1707 info->backup_root_index = 0;
1709 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1714 * copy all the root pointers into the super backup array.
1715 * this will bump the backup pointer by one when it is
1718 static void backup_super_roots(struct btrfs_fs_info *info)
1721 struct btrfs_root_backup *root_backup;
1724 next_backup = info->backup_root_index;
1725 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1726 BTRFS_NUM_BACKUP_ROOTS;
1729 * just overwrite the last backup if we're at the same generation
1730 * this happens only at umount
1732 root_backup = info->super_for_commit->super_roots + last_backup;
1733 if (btrfs_backup_tree_root_gen(root_backup) ==
1734 btrfs_header_generation(info->tree_root->node))
1735 next_backup = last_backup;
1737 root_backup = info->super_for_commit->super_roots + next_backup;
1740 * make sure all of our padding and empty slots get zero filled
1741 * regardless of which ones we use today
1743 memset(root_backup, 0, sizeof(*root_backup));
1745 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1747 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1748 btrfs_set_backup_tree_root_gen(root_backup,
1749 btrfs_header_generation(info->tree_root->node));
1751 btrfs_set_backup_tree_root_level(root_backup,
1752 btrfs_header_level(info->tree_root->node));
1754 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1755 btrfs_set_backup_chunk_root_gen(root_backup,
1756 btrfs_header_generation(info->chunk_root->node));
1757 btrfs_set_backup_chunk_root_level(root_backup,
1758 btrfs_header_level(info->chunk_root->node));
1760 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1761 btrfs_set_backup_extent_root_gen(root_backup,
1762 btrfs_header_generation(info->extent_root->node));
1763 btrfs_set_backup_extent_root_level(root_backup,
1764 btrfs_header_level(info->extent_root->node));
1767 * we might commit during log recovery, which happens before we set
1768 * the fs_root. Make sure it is valid before we fill it in.
1770 if (info->fs_root && info->fs_root->node) {
1771 btrfs_set_backup_fs_root(root_backup,
1772 info->fs_root->node->start);
1773 btrfs_set_backup_fs_root_gen(root_backup,
1774 btrfs_header_generation(info->fs_root->node));
1775 btrfs_set_backup_fs_root_level(root_backup,
1776 btrfs_header_level(info->fs_root->node));
1779 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1780 btrfs_set_backup_dev_root_gen(root_backup,
1781 btrfs_header_generation(info->dev_root->node));
1782 btrfs_set_backup_dev_root_level(root_backup,
1783 btrfs_header_level(info->dev_root->node));
1785 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1786 btrfs_set_backup_csum_root_gen(root_backup,
1787 btrfs_header_generation(info->csum_root->node));
1788 btrfs_set_backup_csum_root_level(root_backup,
1789 btrfs_header_level(info->csum_root->node));
1791 btrfs_set_backup_total_bytes(root_backup,
1792 btrfs_super_total_bytes(info->super_copy));
1793 btrfs_set_backup_bytes_used(root_backup,
1794 btrfs_super_bytes_used(info->super_copy));
1795 btrfs_set_backup_num_devices(root_backup,
1796 btrfs_super_num_devices(info->super_copy));
1799 * if we don't copy this out to the super_copy, it won't get remembered
1800 * for the next commit
1802 memcpy(&info->super_copy->super_roots,
1803 &info->super_for_commit->super_roots,
1804 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1808 * this copies info out of the root backup array and back into
1809 * the in-memory super block. It is meant to help iterate through
1810 * the array, so you send it the number of backups you've already
1811 * tried and the last backup index you used.
1813 * this returns -1 when it has tried all the backups
1815 static noinline int next_root_backup(struct btrfs_fs_info *info,
1816 struct btrfs_super_block *super,
1817 int *num_backups_tried, int *backup_index)
1819 struct btrfs_root_backup *root_backup;
1820 int newest = *backup_index;
1822 if (*num_backups_tried == 0) {
1823 u64 gen = btrfs_super_generation(super);
1825 newest = find_newest_super_backup(info, gen);
1829 *backup_index = newest;
1830 *num_backups_tried = 1;
1831 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1832 /* we've tried all the backups, all done */
1835 /* jump to the next oldest backup */
1836 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1837 BTRFS_NUM_BACKUP_ROOTS;
1838 *backup_index = newest;
1839 *num_backups_tried += 1;
1841 root_backup = super->super_roots + newest;
1843 btrfs_set_super_generation(super,
1844 btrfs_backup_tree_root_gen(root_backup));
1845 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1846 btrfs_set_super_root_level(super,
1847 btrfs_backup_tree_root_level(root_backup));
1848 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1851 * fixme: the total bytes and num_devices need to match or we should
1854 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1855 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1859 /* helper to cleanup tree roots */
1860 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1862 free_extent_buffer(info->tree_root->node);
1863 free_extent_buffer(info->tree_root->commit_root);
1864 free_extent_buffer(info->dev_root->node);
1865 free_extent_buffer(info->dev_root->commit_root);
1866 free_extent_buffer(info->extent_root->node);
1867 free_extent_buffer(info->extent_root->commit_root);
1868 free_extent_buffer(info->csum_root->node);
1869 free_extent_buffer(info->csum_root->commit_root);
1871 info->tree_root->node = NULL;
1872 info->tree_root->commit_root = NULL;
1873 info->dev_root->node = NULL;
1874 info->dev_root->commit_root = NULL;
1875 info->extent_root->node = NULL;
1876 info->extent_root->commit_root = NULL;
1877 info->csum_root->node = NULL;
1878 info->csum_root->commit_root = NULL;
1881 free_extent_buffer(info->chunk_root->node);
1882 free_extent_buffer(info->chunk_root->commit_root);
1883 info->chunk_root->node = NULL;
1884 info->chunk_root->commit_root = NULL;
1889 struct btrfs_root *open_ctree(struct super_block *sb,
1890 struct btrfs_fs_devices *fs_devices,
1900 struct btrfs_key location;
1901 struct buffer_head *bh;
1902 struct btrfs_super_block *disk_super;
1903 struct btrfs_root *tree_root = btrfs_sb(sb);
1904 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1905 struct btrfs_root *extent_root;
1906 struct btrfs_root *csum_root;
1907 struct btrfs_root *chunk_root;
1908 struct btrfs_root *dev_root;
1909 struct btrfs_root *log_tree_root;
1912 int num_backups_tried = 0;
1913 int backup_index = 0;
1915 extent_root = fs_info->extent_root =
1916 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1917 csum_root = fs_info->csum_root =
1918 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1919 chunk_root = fs_info->chunk_root =
1920 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1921 dev_root = fs_info->dev_root =
1922 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1924 if (!extent_root || !csum_root || !chunk_root || !dev_root) {
1929 ret = init_srcu_struct(&fs_info->subvol_srcu);
1935 ret = setup_bdi(fs_info, &fs_info->bdi);
1941 fs_info->btree_inode = new_inode(sb);
1942 if (!fs_info->btree_inode) {
1947 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1949 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1950 INIT_LIST_HEAD(&fs_info->trans_list);
1951 INIT_LIST_HEAD(&fs_info->dead_roots);
1952 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1953 INIT_LIST_HEAD(&fs_info->hashers);
1954 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1955 INIT_LIST_HEAD(&fs_info->ordered_operations);
1956 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1957 spin_lock_init(&fs_info->delalloc_lock);
1958 spin_lock_init(&fs_info->trans_lock);
1959 spin_lock_init(&fs_info->ref_cache_lock);
1960 spin_lock_init(&fs_info->fs_roots_radix_lock);
1961 spin_lock_init(&fs_info->delayed_iput_lock);
1962 spin_lock_init(&fs_info->defrag_inodes_lock);
1963 spin_lock_init(&fs_info->free_chunk_lock);
1964 mutex_init(&fs_info->reloc_mutex);
1966 init_completion(&fs_info->kobj_unregister);
1967 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1968 INIT_LIST_HEAD(&fs_info->space_info);
1969 btrfs_mapping_init(&fs_info->mapping_tree);
1970 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1971 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1972 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1973 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1974 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1975 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1976 atomic_set(&fs_info->nr_async_submits, 0);
1977 atomic_set(&fs_info->async_delalloc_pages, 0);
1978 atomic_set(&fs_info->async_submit_draining, 0);
1979 atomic_set(&fs_info->nr_async_bios, 0);
1980 atomic_set(&fs_info->defrag_running, 0);
1982 fs_info->max_inline = 8192 * 1024;
1983 fs_info->metadata_ratio = 0;
1984 fs_info->defrag_inodes = RB_ROOT;
1985 fs_info->trans_no_join = 0;
1986 fs_info->free_chunk_space = 0;
1988 /* readahead state */
1989 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1990 spin_lock_init(&fs_info->reada_lock);
1992 fs_info->thread_pool_size = min_t(unsigned long,
1993 num_online_cpus() + 2, 8);
1995 INIT_LIST_HEAD(&fs_info->ordered_extents);
1996 spin_lock_init(&fs_info->ordered_extent_lock);
1997 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1999 if (!fs_info->delayed_root) {
2003 btrfs_init_delayed_root(fs_info->delayed_root);
2005 mutex_init(&fs_info->scrub_lock);
2006 atomic_set(&fs_info->scrubs_running, 0);
2007 atomic_set(&fs_info->scrub_pause_req, 0);
2008 atomic_set(&fs_info->scrubs_paused, 0);
2009 atomic_set(&fs_info->scrub_cancel_req, 0);
2010 init_waitqueue_head(&fs_info->scrub_pause_wait);
2011 init_rwsem(&fs_info->scrub_super_lock);
2012 fs_info->scrub_workers_refcnt = 0;
2013 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2014 fs_info->check_integrity_print_mask = 0;
2017 spin_lock_init(&fs_info->balance_lock);
2018 mutex_init(&fs_info->balance_mutex);
2019 atomic_set(&fs_info->balance_running, 0);
2020 atomic_set(&fs_info->balance_pause_req, 0);
2021 atomic_set(&fs_info->balance_cancel_req, 0);
2022 fs_info->balance_ctl = NULL;
2023 init_waitqueue_head(&fs_info->balance_wait_q);
2025 sb->s_blocksize = 4096;
2026 sb->s_blocksize_bits = blksize_bits(4096);
2027 sb->s_bdi = &fs_info->bdi;
2029 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2030 set_nlink(fs_info->btree_inode, 1);
2032 * we set the i_size on the btree inode to the max possible int.
2033 * the real end of the address space is determined by all of
2034 * the devices in the system
2036 fs_info->btree_inode->i_size = OFFSET_MAX;
2037 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2038 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2040 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2041 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2042 fs_info->btree_inode->i_mapping);
2043 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2045 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2047 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2048 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2049 sizeof(struct btrfs_key));
2050 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2051 insert_inode_hash(fs_info->btree_inode);
2053 spin_lock_init(&fs_info->block_group_cache_lock);
2054 fs_info->block_group_cache_tree = RB_ROOT;
2056 extent_io_tree_init(&fs_info->freed_extents[0],
2057 fs_info->btree_inode->i_mapping);
2058 extent_io_tree_init(&fs_info->freed_extents[1],
2059 fs_info->btree_inode->i_mapping);
2060 fs_info->pinned_extents = &fs_info->freed_extents[0];
2061 fs_info->do_barriers = 1;
2064 mutex_init(&fs_info->ordered_operations_mutex);
2065 mutex_init(&fs_info->tree_log_mutex);
2066 mutex_init(&fs_info->chunk_mutex);
2067 mutex_init(&fs_info->transaction_kthread_mutex);
2068 mutex_init(&fs_info->cleaner_mutex);
2069 mutex_init(&fs_info->volume_mutex);
2070 init_rwsem(&fs_info->extent_commit_sem);
2071 init_rwsem(&fs_info->cleanup_work_sem);
2072 init_rwsem(&fs_info->subvol_sem);
2074 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2075 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2077 init_waitqueue_head(&fs_info->transaction_throttle);
2078 init_waitqueue_head(&fs_info->transaction_wait);
2079 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2080 init_waitqueue_head(&fs_info->async_submit_wait);
2082 __setup_root(4096, 4096, 4096, 4096, tree_root,
2083 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2085 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2091 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2092 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2093 sizeof(*fs_info->super_for_commit));
2096 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2098 disk_super = fs_info->super_copy;
2099 if (!btrfs_super_root(disk_super))
2102 /* check FS state, whether FS is broken. */
2103 fs_info->fs_state |= btrfs_super_flags(disk_super);
2105 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2108 * run through our array of backup supers and setup
2109 * our ring pointer to the oldest one
2111 generation = btrfs_super_generation(disk_super);
2112 find_oldest_super_backup(fs_info, generation);
2115 * In the long term, we'll store the compression type in the super
2116 * block, and it'll be used for per file compression control.
2118 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2120 ret = btrfs_parse_options(tree_root, options);
2126 features = btrfs_super_incompat_flags(disk_super) &
2127 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2129 printk(KERN_ERR "BTRFS: couldn't mount because of "
2130 "unsupported optional features (%Lx).\n",
2131 (unsigned long long)features);
2136 features = btrfs_super_incompat_flags(disk_super);
2137 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2138 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2139 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2140 btrfs_set_super_incompat_flags(disk_super, features);
2142 features = btrfs_super_compat_ro_flags(disk_super) &
2143 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2144 if (!(sb->s_flags & MS_RDONLY) && features) {
2145 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2146 "unsupported option features (%Lx).\n",
2147 (unsigned long long)features);
2152 btrfs_init_workers(&fs_info->generic_worker,
2153 "genwork", 1, NULL);
2155 btrfs_init_workers(&fs_info->workers, "worker",
2156 fs_info->thread_pool_size,
2157 &fs_info->generic_worker);
2159 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2160 fs_info->thread_pool_size,
2161 &fs_info->generic_worker);
2163 btrfs_init_workers(&fs_info->submit_workers, "submit",
2164 min_t(u64, fs_devices->num_devices,
2165 fs_info->thread_pool_size),
2166 &fs_info->generic_worker);
2168 btrfs_init_workers(&fs_info->caching_workers, "cache",
2169 2, &fs_info->generic_worker);
2171 /* a higher idle thresh on the submit workers makes it much more
2172 * likely that bios will be send down in a sane order to the
2175 fs_info->submit_workers.idle_thresh = 64;
2177 fs_info->workers.idle_thresh = 16;
2178 fs_info->workers.ordered = 1;
2180 fs_info->delalloc_workers.idle_thresh = 2;
2181 fs_info->delalloc_workers.ordered = 1;
2183 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2184 &fs_info->generic_worker);
2185 btrfs_init_workers(&fs_info->endio_workers, "endio",
2186 fs_info->thread_pool_size,
2187 &fs_info->generic_worker);
2188 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2189 fs_info->thread_pool_size,
2190 &fs_info->generic_worker);
2191 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2192 "endio-meta-write", fs_info->thread_pool_size,
2193 &fs_info->generic_worker);
2194 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2195 fs_info->thread_pool_size,
2196 &fs_info->generic_worker);
2197 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2198 1, &fs_info->generic_worker);
2199 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2200 fs_info->thread_pool_size,
2201 &fs_info->generic_worker);
2202 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2203 fs_info->thread_pool_size,
2204 &fs_info->generic_worker);
2207 * endios are largely parallel and should have a very
2210 fs_info->endio_workers.idle_thresh = 4;
2211 fs_info->endio_meta_workers.idle_thresh = 4;
2213 fs_info->endio_write_workers.idle_thresh = 2;
2214 fs_info->endio_meta_write_workers.idle_thresh = 2;
2215 fs_info->readahead_workers.idle_thresh = 2;
2218 * btrfs_start_workers can really only fail because of ENOMEM so just
2219 * return -ENOMEM if any of these fail.
2221 ret = btrfs_start_workers(&fs_info->workers);
2222 ret |= btrfs_start_workers(&fs_info->generic_worker);
2223 ret |= btrfs_start_workers(&fs_info->submit_workers);
2224 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2225 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2226 ret |= btrfs_start_workers(&fs_info->endio_workers);
2227 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2228 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2229 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2230 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2231 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2232 ret |= btrfs_start_workers(&fs_info->caching_workers);
2233 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2236 goto fail_sb_buffer;
2239 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2240 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2241 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2243 nodesize = btrfs_super_nodesize(disk_super);
2244 leafsize = btrfs_super_leafsize(disk_super);
2245 sectorsize = btrfs_super_sectorsize(disk_super);
2246 stripesize = btrfs_super_stripesize(disk_super);
2247 tree_root->nodesize = nodesize;
2248 tree_root->leafsize = leafsize;
2249 tree_root->sectorsize = sectorsize;
2250 tree_root->stripesize = stripesize;
2252 sb->s_blocksize = sectorsize;
2253 sb->s_blocksize_bits = blksize_bits(sectorsize);
2255 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2256 sizeof(disk_super->magic))) {
2257 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2258 goto fail_sb_buffer;
2261 if (sectorsize < PAGE_SIZE) {
2262 printk(KERN_WARNING "btrfs: Incompatible sector size "
2263 "found on %s\n", sb->s_id);
2264 goto fail_sb_buffer;
2267 mutex_lock(&fs_info->chunk_mutex);
2268 ret = btrfs_read_sys_array(tree_root);
2269 mutex_unlock(&fs_info->chunk_mutex);
2271 printk(KERN_WARNING "btrfs: failed to read the system "
2272 "array on %s\n", sb->s_id);
2273 goto fail_sb_buffer;
2276 blocksize = btrfs_level_size(tree_root,
2277 btrfs_super_chunk_root_level(disk_super));
2278 generation = btrfs_super_chunk_root_generation(disk_super);
2280 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2281 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2283 chunk_root->node = read_tree_block(chunk_root,
2284 btrfs_super_chunk_root(disk_super),
2285 blocksize, generation);
2286 BUG_ON(!chunk_root->node);
2287 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2288 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2290 goto fail_tree_roots;
2292 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2293 chunk_root->commit_root = btrfs_root_node(chunk_root);
2295 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2296 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2299 ret = btrfs_read_chunk_tree(chunk_root);
2301 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2303 goto fail_tree_roots;
2306 btrfs_close_extra_devices(fs_devices);
2309 blocksize = btrfs_level_size(tree_root,
2310 btrfs_super_root_level(disk_super));
2311 generation = btrfs_super_generation(disk_super);
2313 tree_root->node = read_tree_block(tree_root,
2314 btrfs_super_root(disk_super),
2315 blocksize, generation);
2316 if (!tree_root->node ||
2317 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2318 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2321 goto recovery_tree_root;
2324 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2325 tree_root->commit_root = btrfs_root_node(tree_root);
2327 ret = find_and_setup_root(tree_root, fs_info,
2328 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2330 goto recovery_tree_root;
2331 extent_root->track_dirty = 1;
2333 ret = find_and_setup_root(tree_root, fs_info,
2334 BTRFS_DEV_TREE_OBJECTID, dev_root);
2336 goto recovery_tree_root;
2337 dev_root->track_dirty = 1;
2339 ret = find_and_setup_root(tree_root, fs_info,
2340 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2342 goto recovery_tree_root;
2344 csum_root->track_dirty = 1;
2346 fs_info->generation = generation;
2347 fs_info->last_trans_committed = generation;
2349 ret = btrfs_init_space_info(fs_info);
2351 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2352 goto fail_block_groups;
2355 ret = btrfs_read_block_groups(extent_root);
2357 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2358 goto fail_block_groups;
2361 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2363 if (IS_ERR(fs_info->cleaner_kthread))
2364 goto fail_block_groups;
2366 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2368 "btrfs-transaction");
2369 if (IS_ERR(fs_info->transaction_kthread))
2372 if (!btrfs_test_opt(tree_root, SSD) &&
2373 !btrfs_test_opt(tree_root, NOSSD) &&
2374 !fs_info->fs_devices->rotating) {
2375 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2377 btrfs_set_opt(fs_info->mount_opt, SSD);
2380 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2381 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2382 ret = btrfsic_mount(tree_root, fs_devices,
2383 btrfs_test_opt(tree_root,
2384 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2386 fs_info->check_integrity_print_mask);
2388 printk(KERN_WARNING "btrfs: failed to initialize"
2389 " integrity check module %s\n", sb->s_id);
2393 /* do not make disk changes in broken FS */
2394 if (btrfs_super_log_root(disk_super) != 0 &&
2395 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2396 u64 bytenr = btrfs_super_log_root(disk_super);
2398 if (fs_devices->rw_devices == 0) {
2399 printk(KERN_WARNING "Btrfs log replay required "
2402 goto fail_trans_kthread;
2405 btrfs_level_size(tree_root,
2406 btrfs_super_log_root_level(disk_super));
2408 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2409 if (!log_tree_root) {
2411 goto fail_trans_kthread;
2414 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2415 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2417 log_tree_root->node = read_tree_block(tree_root, bytenr,
2420 ret = btrfs_recover_log_trees(log_tree_root);
2423 if (sb->s_flags & MS_RDONLY) {
2424 ret = btrfs_commit_super(tree_root);
2429 ret = btrfs_find_orphan_roots(tree_root);
2432 if (!(sb->s_flags & MS_RDONLY)) {
2433 ret = btrfs_cleanup_fs_roots(fs_info);
2436 ret = btrfs_recover_relocation(tree_root);
2439 "btrfs: failed to recover relocation\n");
2441 goto fail_trans_kthread;
2445 location.objectid = BTRFS_FS_TREE_OBJECTID;
2446 location.type = BTRFS_ROOT_ITEM_KEY;
2447 location.offset = (u64)-1;
2449 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2450 if (!fs_info->fs_root)
2451 goto fail_trans_kthread;
2452 if (IS_ERR(fs_info->fs_root)) {
2453 err = PTR_ERR(fs_info->fs_root);
2454 goto fail_trans_kthread;
2457 if (!(sb->s_flags & MS_RDONLY)) {
2458 down_read(&fs_info->cleanup_work_sem);
2459 err = btrfs_orphan_cleanup(fs_info->fs_root);
2461 err = btrfs_orphan_cleanup(fs_info->tree_root);
2462 up_read(&fs_info->cleanup_work_sem);
2465 err = btrfs_recover_balance(fs_info->tree_root);
2468 close_ctree(tree_root);
2469 return ERR_PTR(err);
2476 kthread_stop(fs_info->transaction_kthread);
2478 kthread_stop(fs_info->cleaner_kthread);
2481 * make sure we're done with the btree inode before we stop our
2484 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2485 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2488 btrfs_free_block_groups(fs_info);
2491 free_root_pointers(fs_info, 1);
2494 btrfs_stop_workers(&fs_info->generic_worker);
2495 btrfs_stop_workers(&fs_info->readahead_workers);
2496 btrfs_stop_workers(&fs_info->fixup_workers);
2497 btrfs_stop_workers(&fs_info->delalloc_workers);
2498 btrfs_stop_workers(&fs_info->workers);
2499 btrfs_stop_workers(&fs_info->endio_workers);
2500 btrfs_stop_workers(&fs_info->endio_meta_workers);
2501 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2502 btrfs_stop_workers(&fs_info->endio_write_workers);
2503 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2504 btrfs_stop_workers(&fs_info->submit_workers);
2505 btrfs_stop_workers(&fs_info->delayed_workers);
2506 btrfs_stop_workers(&fs_info->caching_workers);
2509 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2511 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2512 iput(fs_info->btree_inode);
2514 bdi_destroy(&fs_info->bdi);
2516 cleanup_srcu_struct(&fs_info->subvol_srcu);
2518 btrfs_close_devices(fs_info->fs_devices);
2519 free_fs_info(fs_info);
2520 return ERR_PTR(err);
2523 if (!btrfs_test_opt(tree_root, RECOVERY))
2524 goto fail_tree_roots;
2526 free_root_pointers(fs_info, 0);
2528 /* don't use the log in recovery mode, it won't be valid */
2529 btrfs_set_super_log_root(disk_super, 0);
2531 /* we can't trust the free space cache either */
2532 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2534 ret = next_root_backup(fs_info, fs_info->super_copy,
2535 &num_backups_tried, &backup_index);
2537 goto fail_block_groups;
2538 goto retry_root_backup;
2541 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2543 char b[BDEVNAME_SIZE];
2546 set_buffer_uptodate(bh);
2548 printk_ratelimited(KERN_WARNING "lost page write due to "
2549 "I/O error on %s\n",
2550 bdevname(bh->b_bdev, b));
2551 /* note, we dont' set_buffer_write_io_error because we have
2552 * our own ways of dealing with the IO errors
2554 clear_buffer_uptodate(bh);
2560 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2562 struct buffer_head *bh;
2563 struct buffer_head *latest = NULL;
2564 struct btrfs_super_block *super;
2569 /* we would like to check all the supers, but that would make
2570 * a btrfs mount succeed after a mkfs from a different FS.
2571 * So, we need to add a special mount option to scan for
2572 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2574 for (i = 0; i < 1; i++) {
2575 bytenr = btrfs_sb_offset(i);
2576 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2578 bh = __bread(bdev, bytenr / 4096, 4096);
2582 super = (struct btrfs_super_block *)bh->b_data;
2583 if (btrfs_super_bytenr(super) != bytenr ||
2584 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2585 sizeof(super->magic))) {
2590 if (!latest || btrfs_super_generation(super) > transid) {
2593 transid = btrfs_super_generation(super);
2602 * this should be called twice, once with wait == 0 and
2603 * once with wait == 1. When wait == 0 is done, all the buffer heads
2604 * we write are pinned.
2606 * They are released when wait == 1 is done.
2607 * max_mirrors must be the same for both runs, and it indicates how
2608 * many supers on this one device should be written.
2610 * max_mirrors == 0 means to write them all.
2612 static int write_dev_supers(struct btrfs_device *device,
2613 struct btrfs_super_block *sb,
2614 int do_barriers, int wait, int max_mirrors)
2616 struct buffer_head *bh;
2623 if (max_mirrors == 0)
2624 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2626 for (i = 0; i < max_mirrors; i++) {
2627 bytenr = btrfs_sb_offset(i);
2628 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2632 bh = __find_get_block(device->bdev, bytenr / 4096,
2633 BTRFS_SUPER_INFO_SIZE);
2636 if (!buffer_uptodate(bh))
2639 /* drop our reference */
2642 /* drop the reference from the wait == 0 run */
2646 btrfs_set_super_bytenr(sb, bytenr);
2649 crc = btrfs_csum_data(NULL, (char *)sb +
2650 BTRFS_CSUM_SIZE, crc,
2651 BTRFS_SUPER_INFO_SIZE -
2653 btrfs_csum_final(crc, sb->csum);
2656 * one reference for us, and we leave it for the
2659 bh = __getblk(device->bdev, bytenr / 4096,
2660 BTRFS_SUPER_INFO_SIZE);
2661 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2663 /* one reference for submit_bh */
2666 set_buffer_uptodate(bh);
2668 bh->b_end_io = btrfs_end_buffer_write_sync;
2672 * we fua the first super. The others we allow
2675 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2679 return errors < i ? 0 : -1;
2683 * endio for the write_dev_flush, this will wake anyone waiting
2684 * for the barrier when it is done
2686 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2689 if (err == -EOPNOTSUPP)
2690 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2691 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2693 if (bio->bi_private)
2694 complete(bio->bi_private);
2699 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2700 * sent down. With wait == 1, it waits for the previous flush.
2702 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2705 static int write_dev_flush(struct btrfs_device *device, int wait)
2710 if (device->nobarriers)
2714 bio = device->flush_bio;
2718 wait_for_completion(&device->flush_wait);
2720 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2721 printk("btrfs: disabling barriers on dev %s\n",
2723 device->nobarriers = 1;
2725 if (!bio_flagged(bio, BIO_UPTODATE)) {
2729 /* drop the reference from the wait == 0 run */
2731 device->flush_bio = NULL;
2737 * one reference for us, and we leave it for the
2740 device->flush_bio = NULL;;
2741 bio = bio_alloc(GFP_NOFS, 0);
2745 bio->bi_end_io = btrfs_end_empty_barrier;
2746 bio->bi_bdev = device->bdev;
2747 init_completion(&device->flush_wait);
2748 bio->bi_private = &device->flush_wait;
2749 device->flush_bio = bio;
2752 btrfsic_submit_bio(WRITE_FLUSH, bio);
2758 * send an empty flush down to each device in parallel,
2759 * then wait for them
2761 static int barrier_all_devices(struct btrfs_fs_info *info)
2763 struct list_head *head;
2764 struct btrfs_device *dev;
2768 /* send down all the barriers */
2769 head = &info->fs_devices->devices;
2770 list_for_each_entry_rcu(dev, head, dev_list) {
2775 if (!dev->in_fs_metadata || !dev->writeable)
2778 ret = write_dev_flush(dev, 0);
2783 /* wait for all the barriers */
2784 list_for_each_entry_rcu(dev, head, dev_list) {
2789 if (!dev->in_fs_metadata || !dev->writeable)
2792 ret = write_dev_flush(dev, 1);
2801 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2803 struct list_head *head;
2804 struct btrfs_device *dev;
2805 struct btrfs_super_block *sb;
2806 struct btrfs_dev_item *dev_item;
2810 int total_errors = 0;
2813 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2814 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2815 backup_super_roots(root->fs_info);
2817 sb = root->fs_info->super_for_commit;
2818 dev_item = &sb->dev_item;
2820 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2821 head = &root->fs_info->fs_devices->devices;
2824 barrier_all_devices(root->fs_info);
2826 list_for_each_entry_rcu(dev, head, dev_list) {
2831 if (!dev->in_fs_metadata || !dev->writeable)
2834 btrfs_set_stack_device_generation(dev_item, 0);
2835 btrfs_set_stack_device_type(dev_item, dev->type);
2836 btrfs_set_stack_device_id(dev_item, dev->devid);
2837 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2838 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2839 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2840 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2841 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2842 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2843 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2845 flags = btrfs_super_flags(sb);
2846 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2848 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2852 if (total_errors > max_errors) {
2853 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2859 list_for_each_entry_rcu(dev, head, dev_list) {
2862 if (!dev->in_fs_metadata || !dev->writeable)
2865 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2869 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2870 if (total_errors > max_errors) {
2871 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2878 int write_ctree_super(struct btrfs_trans_handle *trans,
2879 struct btrfs_root *root, int max_mirrors)
2883 ret = write_all_supers(root, max_mirrors);
2887 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2889 spin_lock(&fs_info->fs_roots_radix_lock);
2890 radix_tree_delete(&fs_info->fs_roots_radix,
2891 (unsigned long)root->root_key.objectid);
2892 spin_unlock(&fs_info->fs_roots_radix_lock);
2894 if (btrfs_root_refs(&root->root_item) == 0)
2895 synchronize_srcu(&fs_info->subvol_srcu);
2897 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2898 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2903 static void free_fs_root(struct btrfs_root *root)
2905 iput(root->cache_inode);
2906 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2908 free_anon_bdev(root->anon_dev);
2909 free_extent_buffer(root->node);
2910 free_extent_buffer(root->commit_root);
2911 kfree(root->free_ino_ctl);
2912 kfree(root->free_ino_pinned);
2917 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2920 struct btrfs_root *gang[8];
2923 while (!list_empty(&fs_info->dead_roots)) {
2924 gang[0] = list_entry(fs_info->dead_roots.next,
2925 struct btrfs_root, root_list);
2926 list_del(&gang[0]->root_list);
2928 if (gang[0]->in_radix) {
2929 btrfs_free_fs_root(fs_info, gang[0]);
2931 free_extent_buffer(gang[0]->node);
2932 free_extent_buffer(gang[0]->commit_root);
2938 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2943 for (i = 0; i < ret; i++)
2944 btrfs_free_fs_root(fs_info, gang[i]);
2949 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2951 u64 root_objectid = 0;
2952 struct btrfs_root *gang[8];
2957 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2958 (void **)gang, root_objectid,
2963 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2964 for (i = 0; i < ret; i++) {
2967 root_objectid = gang[i]->root_key.objectid;
2968 err = btrfs_orphan_cleanup(gang[i]);
2977 int btrfs_commit_super(struct btrfs_root *root)
2979 struct btrfs_trans_handle *trans;
2982 mutex_lock(&root->fs_info->cleaner_mutex);
2983 btrfs_run_delayed_iputs(root);
2984 btrfs_clean_old_snapshots(root);
2985 mutex_unlock(&root->fs_info->cleaner_mutex);
2987 /* wait until ongoing cleanup work done */
2988 down_write(&root->fs_info->cleanup_work_sem);
2989 up_write(&root->fs_info->cleanup_work_sem);
2991 trans = btrfs_join_transaction(root);
2993 return PTR_ERR(trans);
2994 ret = btrfs_commit_transaction(trans, root);
2996 /* run commit again to drop the original snapshot */
2997 trans = btrfs_join_transaction(root);
2999 return PTR_ERR(trans);
3000 btrfs_commit_transaction(trans, root);
3001 ret = btrfs_write_and_wait_transaction(NULL, root);
3004 ret = write_ctree_super(NULL, root, 0);
3008 int close_ctree(struct btrfs_root *root)
3010 struct btrfs_fs_info *fs_info = root->fs_info;
3013 fs_info->closing = 1;
3016 /* pause restriper - we want to resume on mount */
3017 btrfs_pause_balance(root->fs_info);
3019 btrfs_scrub_cancel(root);
3021 /* wait for any defraggers to finish */
3022 wait_event(fs_info->transaction_wait,
3023 (atomic_read(&fs_info->defrag_running) == 0));
3025 /* clear out the rbtree of defraggable inodes */
3026 btrfs_run_defrag_inodes(root->fs_info);
3029 * Here come 2 situations when btrfs is broken to flip readonly:
3031 * 1. when btrfs flips readonly somewhere else before
3032 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3033 * and btrfs will skip to write sb directly to keep
3034 * ERROR state on disk.
3036 * 2. when btrfs flips readonly just in btrfs_commit_super,
3037 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3038 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3039 * btrfs will cleanup all FS resources first and write sb then.
3041 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3042 ret = btrfs_commit_super(root);
3044 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3047 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3048 ret = btrfs_error_commit_super(root);
3050 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3053 btrfs_put_block_group_cache(fs_info);
3055 kthread_stop(root->fs_info->transaction_kthread);
3056 kthread_stop(root->fs_info->cleaner_kthread);
3058 fs_info->closing = 2;
3061 if (fs_info->delalloc_bytes) {
3062 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3063 (unsigned long long)fs_info->delalloc_bytes);
3065 if (fs_info->total_ref_cache_size) {
3066 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3067 (unsigned long long)fs_info->total_ref_cache_size);
3070 free_extent_buffer(fs_info->extent_root->node);
3071 free_extent_buffer(fs_info->extent_root->commit_root);
3072 free_extent_buffer(fs_info->tree_root->node);
3073 free_extent_buffer(fs_info->tree_root->commit_root);
3074 free_extent_buffer(root->fs_info->chunk_root->node);
3075 free_extent_buffer(root->fs_info->chunk_root->commit_root);
3076 free_extent_buffer(root->fs_info->dev_root->node);
3077 free_extent_buffer(root->fs_info->dev_root->commit_root);
3078 free_extent_buffer(root->fs_info->csum_root->node);
3079 free_extent_buffer(root->fs_info->csum_root->commit_root);
3081 btrfs_free_block_groups(root->fs_info);
3083 del_fs_roots(fs_info);
3085 iput(fs_info->btree_inode);
3087 btrfs_stop_workers(&fs_info->generic_worker);
3088 btrfs_stop_workers(&fs_info->fixup_workers);
3089 btrfs_stop_workers(&fs_info->delalloc_workers);
3090 btrfs_stop_workers(&fs_info->workers);
3091 btrfs_stop_workers(&fs_info->endio_workers);
3092 btrfs_stop_workers(&fs_info->endio_meta_workers);
3093 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3094 btrfs_stop_workers(&fs_info->endio_write_workers);
3095 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3096 btrfs_stop_workers(&fs_info->submit_workers);
3097 btrfs_stop_workers(&fs_info->delayed_workers);
3098 btrfs_stop_workers(&fs_info->caching_workers);
3099 btrfs_stop_workers(&fs_info->readahead_workers);
3101 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3102 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3103 btrfsic_unmount(root, fs_info->fs_devices);
3106 btrfs_close_devices(fs_info->fs_devices);
3107 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3109 bdi_destroy(&fs_info->bdi);
3110 cleanup_srcu_struct(&fs_info->subvol_srcu);
3112 free_fs_info(fs_info);
3117 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3120 struct inode *btree_inode = buf->first_page->mapping->host;
3122 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3127 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3132 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3134 struct inode *btree_inode = buf->first_page->mapping->host;
3135 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3139 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3141 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3142 u64 transid = btrfs_header_generation(buf);
3143 struct inode *btree_inode = root->fs_info->btree_inode;
3146 btrfs_assert_tree_locked(buf);
3147 if (transid != root->fs_info->generation) {
3148 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3149 "found %llu running %llu\n",
3150 (unsigned long long)buf->start,
3151 (unsigned long long)transid,
3152 (unsigned long long)root->fs_info->generation);
3155 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3158 spin_lock(&root->fs_info->delalloc_lock);
3159 root->fs_info->dirty_metadata_bytes += buf->len;
3160 spin_unlock(&root->fs_info->delalloc_lock);
3164 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3167 * looks as though older kernels can get into trouble with
3168 * this code, they end up stuck in balance_dirty_pages forever
3171 unsigned long thresh = 32 * 1024 * 1024;
3173 if (current->flags & PF_MEMALLOC)
3176 btrfs_balance_delayed_items(root);
3178 num_dirty = root->fs_info->dirty_metadata_bytes;
3180 if (num_dirty > thresh) {
3181 balance_dirty_pages_ratelimited_nr(
3182 root->fs_info->btree_inode->i_mapping, 1);
3187 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3190 * looks as though older kernels can get into trouble with
3191 * this code, they end up stuck in balance_dirty_pages forever
3194 unsigned long thresh = 32 * 1024 * 1024;
3196 if (current->flags & PF_MEMALLOC)
3199 num_dirty = root->fs_info->dirty_metadata_bytes;
3201 if (num_dirty > thresh) {
3202 balance_dirty_pages_ratelimited_nr(
3203 root->fs_info->btree_inode->i_mapping, 1);
3208 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3210 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3212 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3214 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3218 static int btree_lock_page_hook(struct page *page, void *data,
3219 void (*flush_fn)(void *))
3221 struct inode *inode = page->mapping->host;
3222 struct btrfs_root *root = BTRFS_I(inode)->root;
3223 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3224 struct extent_buffer *eb;
3226 u64 bytenr = page_offset(page);
3228 if (page->private == EXTENT_PAGE_PRIVATE)
3231 len = page->private >> 2;
3232 eb = find_extent_buffer(io_tree, bytenr, len);
3236 if (!btrfs_try_tree_write_lock(eb)) {
3238 btrfs_tree_lock(eb);
3240 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3242 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3243 spin_lock(&root->fs_info->delalloc_lock);
3244 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3245 root->fs_info->dirty_metadata_bytes -= eb->len;
3248 spin_unlock(&root->fs_info->delalloc_lock);
3251 btrfs_tree_unlock(eb);
3252 free_extent_buffer(eb);
3254 if (!trylock_page(page)) {
3261 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3267 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3268 printk(KERN_WARNING "warning: mount fs with errors, "
3269 "running btrfsck is recommended\n");
3272 int btrfs_error_commit_super(struct btrfs_root *root)
3276 mutex_lock(&root->fs_info->cleaner_mutex);
3277 btrfs_run_delayed_iputs(root);
3278 mutex_unlock(&root->fs_info->cleaner_mutex);
3280 down_write(&root->fs_info->cleanup_work_sem);
3281 up_write(&root->fs_info->cleanup_work_sem);
3283 /* cleanup FS via transaction */
3284 btrfs_cleanup_transaction(root);
3286 ret = write_ctree_super(NULL, root, 0);
3291 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3293 struct btrfs_inode *btrfs_inode;
3294 struct list_head splice;
3296 INIT_LIST_HEAD(&splice);
3298 mutex_lock(&root->fs_info->ordered_operations_mutex);
3299 spin_lock(&root->fs_info->ordered_extent_lock);
3301 list_splice_init(&root->fs_info->ordered_operations, &splice);
3302 while (!list_empty(&splice)) {
3303 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3304 ordered_operations);
3306 list_del_init(&btrfs_inode->ordered_operations);
3308 btrfs_invalidate_inodes(btrfs_inode->root);
3311 spin_unlock(&root->fs_info->ordered_extent_lock);
3312 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3317 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3319 struct list_head splice;
3320 struct btrfs_ordered_extent *ordered;
3321 struct inode *inode;
3323 INIT_LIST_HEAD(&splice);
3325 spin_lock(&root->fs_info->ordered_extent_lock);
3327 list_splice_init(&root->fs_info->ordered_extents, &splice);
3328 while (!list_empty(&splice)) {
3329 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3332 list_del_init(&ordered->root_extent_list);
3333 atomic_inc(&ordered->refs);
3335 /* the inode may be getting freed (in sys_unlink path). */
3336 inode = igrab(ordered->inode);
3338 spin_unlock(&root->fs_info->ordered_extent_lock);
3342 atomic_set(&ordered->refs, 1);
3343 btrfs_put_ordered_extent(ordered);
3345 spin_lock(&root->fs_info->ordered_extent_lock);
3348 spin_unlock(&root->fs_info->ordered_extent_lock);
3353 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3354 struct btrfs_root *root)
3356 struct rb_node *node;
3357 struct btrfs_delayed_ref_root *delayed_refs;
3358 struct btrfs_delayed_ref_node *ref;
3361 delayed_refs = &trans->delayed_refs;
3363 spin_lock(&delayed_refs->lock);
3364 if (delayed_refs->num_entries == 0) {
3365 spin_unlock(&delayed_refs->lock);
3366 printk(KERN_INFO "delayed_refs has NO entry\n");
3370 node = rb_first(&delayed_refs->root);
3372 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3373 node = rb_next(node);
3376 rb_erase(&ref->rb_node, &delayed_refs->root);
3377 delayed_refs->num_entries--;
3379 atomic_set(&ref->refs, 1);
3380 if (btrfs_delayed_ref_is_head(ref)) {
3381 struct btrfs_delayed_ref_head *head;
3383 head = btrfs_delayed_node_to_head(ref);
3384 mutex_lock(&head->mutex);
3385 kfree(head->extent_op);
3386 delayed_refs->num_heads--;
3387 if (list_empty(&head->cluster))
3388 delayed_refs->num_heads_ready--;
3389 list_del_init(&head->cluster);
3390 mutex_unlock(&head->mutex);
3393 spin_unlock(&delayed_refs->lock);
3394 btrfs_put_delayed_ref(ref);
3397 spin_lock(&delayed_refs->lock);
3400 spin_unlock(&delayed_refs->lock);
3405 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3407 struct btrfs_pending_snapshot *snapshot;
3408 struct list_head splice;
3410 INIT_LIST_HEAD(&splice);
3412 list_splice_init(&t->pending_snapshots, &splice);
3414 while (!list_empty(&splice)) {
3415 snapshot = list_entry(splice.next,
3416 struct btrfs_pending_snapshot,
3419 list_del_init(&snapshot->list);
3427 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3429 struct btrfs_inode *btrfs_inode;
3430 struct list_head splice;
3432 INIT_LIST_HEAD(&splice);
3434 spin_lock(&root->fs_info->delalloc_lock);
3435 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3437 while (!list_empty(&splice)) {
3438 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3441 list_del_init(&btrfs_inode->delalloc_inodes);
3443 btrfs_invalidate_inodes(btrfs_inode->root);
3446 spin_unlock(&root->fs_info->delalloc_lock);
3451 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3452 struct extent_io_tree *dirty_pages,
3457 struct inode *btree_inode = root->fs_info->btree_inode;
3458 struct extent_buffer *eb;
3462 unsigned long index;
3465 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3470 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3471 while (start <= end) {
3472 index = start >> PAGE_CACHE_SHIFT;
3473 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3474 page = find_get_page(btree_inode->i_mapping, index);
3477 offset = page_offset(page);
3479 spin_lock(&dirty_pages->buffer_lock);
3480 eb = radix_tree_lookup(
3481 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3482 offset >> PAGE_CACHE_SHIFT);
3483 spin_unlock(&dirty_pages->buffer_lock);
3485 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3487 atomic_set(&eb->refs, 1);
3489 if (PageWriteback(page))
3490 end_page_writeback(page);
3493 if (PageDirty(page)) {
3494 clear_page_dirty_for_io(page);
3495 spin_lock_irq(&page->mapping->tree_lock);
3496 radix_tree_tag_clear(&page->mapping->page_tree,
3498 PAGECACHE_TAG_DIRTY);
3499 spin_unlock_irq(&page->mapping->tree_lock);
3502 page->mapping->a_ops->invalidatepage(page, 0);
3510 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3511 struct extent_io_tree *pinned_extents)
3513 struct extent_io_tree *unpin;
3518 unpin = pinned_extents;
3520 ret = find_first_extent_bit(unpin, 0, &start, &end,
3526 if (btrfs_test_opt(root, DISCARD))
3527 ret = btrfs_error_discard_extent(root, start,
3531 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3532 btrfs_error_unpin_extent_range(root, start, end);
3539 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3541 struct btrfs_transaction *t;
3546 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3548 spin_lock(&root->fs_info->trans_lock);
3549 list_splice_init(&root->fs_info->trans_list, &list);
3550 root->fs_info->trans_no_join = 1;
3551 spin_unlock(&root->fs_info->trans_lock);
3553 while (!list_empty(&list)) {
3554 t = list_entry(list.next, struct btrfs_transaction, list);
3558 btrfs_destroy_ordered_operations(root);
3560 btrfs_destroy_ordered_extents(root);
3562 btrfs_destroy_delayed_refs(t, root);
3564 btrfs_block_rsv_release(root,
3565 &root->fs_info->trans_block_rsv,
3566 t->dirty_pages.dirty_bytes);
3568 /* FIXME: cleanup wait for commit */
3571 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3572 wake_up(&root->fs_info->transaction_blocked_wait);
3575 if (waitqueue_active(&root->fs_info->transaction_wait))
3576 wake_up(&root->fs_info->transaction_wait);
3579 if (waitqueue_active(&t->commit_wait))
3580 wake_up(&t->commit_wait);
3582 btrfs_destroy_pending_snapshots(t);
3584 btrfs_destroy_delalloc_inodes(root);
3586 spin_lock(&root->fs_info->trans_lock);
3587 root->fs_info->running_transaction = NULL;
3588 spin_unlock(&root->fs_info->trans_lock);
3590 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3593 btrfs_destroy_pinned_extent(root,
3594 root->fs_info->pinned_extents);
3596 atomic_set(&t->use_count, 0);
3597 list_del_init(&t->list);
3598 memset(t, 0, sizeof(*t));
3599 kmem_cache_free(btrfs_transaction_cachep, t);
3602 spin_lock(&root->fs_info->trans_lock);
3603 root->fs_info->trans_no_join = 0;
3604 spin_unlock(&root->fs_info->trans_lock);
3605 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3610 static struct extent_io_ops btree_extent_io_ops = {
3611 .write_cache_pages_lock_hook = btree_lock_page_hook,
3612 .readpage_end_io_hook = btree_readpage_end_io_hook,
3613 .readpage_io_failed_hook = btree_io_failed_hook,
3614 .submit_bio_hook = btree_submit_bio_hook,
3615 /* note we're sharing with inode.c for the merge bio hook */
3616 .merge_bio_hook = btrfs_merge_bio_hook,