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.
19 #include <linux/version.h>
21 #include <linux/blkdev.h>
22 #include <linux/scatterlist.h>
23 #include <linux/swap.h>
24 #include <linux/radix-tree.h>
25 #include <linux/writeback.h>
26 #include <linux/buffer_head.h> // for block_sync_page
27 #include <linux/workqueue.h>
28 #include <linux/kthread.h>
29 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
30 # include <linux/freezer.h>
32 # include <linux/sched.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
45 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
47 if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
48 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
49 (unsigned long long)extent_buffer_blocknr(buf),
50 (unsigned long long)btrfs_header_blocknr(buf));
57 static struct extent_io_ops btree_extent_io_ops;
58 static void end_workqueue_fn(struct btrfs_work *work);
64 struct btrfs_fs_info *info;
67 struct list_head list;
68 struct btrfs_work work;
71 struct async_submit_bio {
74 struct list_head list;
75 extent_submit_bio_hook_t *submit_bio_hook;
78 struct btrfs_work work;
81 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
82 size_t page_offset, u64 start, u64 len,
85 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
86 struct extent_map *em;
89 spin_lock(&em_tree->lock);
90 em = lookup_extent_mapping(em_tree, start, len);
93 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
94 spin_unlock(&em_tree->lock);
97 spin_unlock(&em_tree->lock);
99 em = alloc_extent_map(GFP_NOFS);
101 em = ERR_PTR(-ENOMEM);
107 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
109 spin_lock(&em_tree->lock);
110 ret = add_extent_mapping(em_tree, em);
111 if (ret == -EEXIST) {
112 u64 failed_start = em->start;
113 u64 failed_len = em->len;
115 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
116 em->start, em->len, em->block_start);
118 em = lookup_extent_mapping(em_tree, start, len);
120 printk("after failing, found %Lu %Lu %Lu\n",
121 em->start, em->len, em->block_start);
124 em = lookup_extent_mapping(em_tree, failed_start,
127 printk("double failure lookup gives us "
128 "%Lu %Lu -> %Lu\n", em->start,
129 em->len, em->block_start);
138 spin_unlock(&em_tree->lock);
146 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
148 return btrfs_crc32c(seed, data, len);
151 void btrfs_csum_final(u32 crc, char *result)
153 *(__le32 *)result = ~cpu_to_le32(crc);
156 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
159 char result[BTRFS_CRC32_SIZE];
161 unsigned long cur_len;
162 unsigned long offset = BTRFS_CSUM_SIZE;
163 char *map_token = NULL;
165 unsigned long map_start;
166 unsigned long map_len;
170 len = buf->len - offset;
172 err = map_private_extent_buffer(buf, offset, 32,
174 &map_start, &map_len, KM_USER0);
176 printk("failed to map extent buffer! %lu\n",
180 cur_len = min(len, map_len - (offset - map_start));
181 crc = btrfs_csum_data(root, kaddr + offset - map_start,
185 unmap_extent_buffer(buf, map_token, KM_USER0);
187 btrfs_csum_final(crc, result);
190 int from_this_trans = 0;
192 if (root->fs_info->running_transaction &&
193 btrfs_header_generation(buf) ==
194 root->fs_info->running_transaction->transid)
197 /* FIXME, this is not good */
198 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
201 memcpy(&found, result, BTRFS_CRC32_SIZE);
203 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
204 printk("btrfs: %s checksum verify failed on %llu "
205 "wanted %X found %X from_this_trans %d "
207 root->fs_info->sb->s_id,
208 buf->start, val, found, from_this_trans,
209 btrfs_header_level(buf));
213 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
218 static int verify_parent_transid(struct extent_io_tree *io_tree,
219 struct extent_buffer *eb, u64 parent_transid)
223 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
226 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
227 if (extent_buffer_uptodate(io_tree, eb) &&
228 btrfs_header_generation(eb) == parent_transid) {
232 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
233 (unsigned long long)eb->start,
234 (unsigned long long)parent_transid,
235 (unsigned long long)btrfs_header_generation(eb));
238 clear_extent_buffer_uptodate(io_tree, eb);
239 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
245 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
246 struct extent_buffer *eb,
247 u64 start, u64 parent_transid)
249 struct extent_io_tree *io_tree;
254 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
256 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
257 btree_get_extent, mirror_num);
259 !verify_parent_transid(io_tree, eb, parent_transid))
262 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
268 if (mirror_num > num_copies)
274 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
276 struct extent_io_tree *tree;
277 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
281 struct extent_buffer *eb;
284 tree = &BTRFS_I(page->mapping->host)->io_tree;
286 if (page->private == EXTENT_PAGE_PRIVATE)
290 len = page->private >> 2;
294 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
295 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
296 btrfs_header_generation(eb));
298 found_start = btrfs_header_bytenr(eb);
299 if (found_start != start) {
300 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
301 start, found_start, len);
305 if (eb->first_page != page) {
306 printk("bad first page %lu %lu\n", eb->first_page->index,
311 if (!PageUptodate(page)) {
312 printk("csum not up to date page %lu\n", page->index);
316 found_level = btrfs_header_level(eb);
317 spin_lock(&root->fs_info->hash_lock);
318 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
319 spin_unlock(&root->fs_info->hash_lock);
320 csum_tree_block(root, eb, 0);
322 free_extent_buffer(eb);
327 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
329 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
331 csum_dirty_buffer(root, page);
335 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
336 struct extent_state *state)
338 struct extent_io_tree *tree;
342 struct extent_buffer *eb;
343 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
346 tree = &BTRFS_I(page->mapping->host)->io_tree;
347 if (page->private == EXTENT_PAGE_PRIVATE)
351 len = page->private >> 2;
355 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
357 found_start = btrfs_header_bytenr(eb);
358 if (found_start != start) {
362 if (eb->first_page != page) {
363 printk("bad first page %lu %lu\n", eb->first_page->index,
369 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
370 (unsigned long)btrfs_header_fsid(eb),
372 printk("bad fsid on block %Lu\n", eb->start);
376 found_level = btrfs_header_level(eb);
378 ret = csum_tree_block(root, eb, 1);
382 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
383 end = eb->start + end - 1;
385 free_extent_buffer(eb);
390 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
391 static void end_workqueue_bio(struct bio *bio, int err)
393 static int end_workqueue_bio(struct bio *bio,
394 unsigned int bytes_done, int err)
397 struct end_io_wq *end_io_wq = bio->bi_private;
398 struct btrfs_fs_info *fs_info;
400 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
405 fs_info = end_io_wq->info;
406 end_io_wq->error = err;
407 end_io_wq->work.func = end_workqueue_fn;
408 end_io_wq->work.flags = 0;
409 if (bio->bi_rw & (1 << BIO_RW))
410 btrfs_queue_worker(&fs_info->endio_write_workers,
413 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
415 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
420 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
423 struct end_io_wq *end_io_wq;
424 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
428 end_io_wq->private = bio->bi_private;
429 end_io_wq->end_io = bio->bi_end_io;
430 end_io_wq->info = info;
431 end_io_wq->error = 0;
432 end_io_wq->bio = bio;
433 end_io_wq->metadata = metadata;
435 bio->bi_private = end_io_wq;
436 bio->bi_end_io = end_workqueue_bio;
440 static void run_one_async_submit(struct btrfs_work *work)
442 struct btrfs_fs_info *fs_info;
443 struct async_submit_bio *async;
445 async = container_of(work, struct async_submit_bio, work);
446 fs_info = BTRFS_I(async->inode)->root->fs_info;
447 atomic_dec(&fs_info->nr_async_submits);
448 async->submit_bio_hook(async->inode, async->rw, async->bio,
453 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
454 int rw, struct bio *bio, int mirror_num,
455 extent_submit_bio_hook_t *submit_bio_hook)
457 struct async_submit_bio *async;
459 async = kmalloc(sizeof(*async), GFP_NOFS);
463 async->inode = inode;
466 async->mirror_num = mirror_num;
467 async->submit_bio_hook = submit_bio_hook;
468 async->work.func = run_one_async_submit;
469 async->work.flags = 0;
470 atomic_inc(&fs_info->nr_async_submits);
471 btrfs_queue_worker(&fs_info->workers, &async->work);
475 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
478 struct btrfs_root *root = BTRFS_I(inode)->root;
482 offset = bio->bi_sector << 9;
485 * when we're called for a write, we're already in the async
486 * submission context. Just jump ingo btrfs_map_bio
488 if (rw & (1 << BIO_RW)) {
489 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
494 * called for a read, do the setup so that checksum validation
495 * can happen in the async kernel threads
497 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
500 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
503 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
507 * kthread helpers are used to submit writes so that checksumming
508 * can happen in parallel across all CPUs
510 if (!(rw & (1 << BIO_RW))) {
511 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
513 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
514 inode, rw, bio, mirror_num,
515 __btree_submit_bio_hook);
518 static int btree_writepage(struct page *page, struct writeback_control *wbc)
520 struct extent_io_tree *tree;
521 tree = &BTRFS_I(page->mapping->host)->io_tree;
522 return extent_write_full_page(tree, page, btree_get_extent, wbc);
525 static int btree_writepages(struct address_space *mapping,
526 struct writeback_control *wbc)
528 struct extent_io_tree *tree;
529 tree = &BTRFS_I(mapping->host)->io_tree;
530 if (wbc->sync_mode == WB_SYNC_NONE) {
533 unsigned long thresh = 96 * 1024 * 1024;
535 if (wbc->for_kupdate)
538 if (current_is_pdflush()) {
539 thresh = 96 * 1024 * 1024;
541 thresh = 8 * 1024 * 1024;
543 num_dirty = count_range_bits(tree, &start, (u64)-1,
544 thresh, EXTENT_DIRTY);
545 if (num_dirty < thresh) {
549 return extent_writepages(tree, mapping, btree_get_extent, wbc);
552 int btree_readpage(struct file *file, struct page *page)
554 struct extent_io_tree *tree;
555 tree = &BTRFS_I(page->mapping->host)->io_tree;
556 return extent_read_full_page(tree, page, btree_get_extent);
559 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
561 struct extent_io_tree *tree;
562 struct extent_map_tree *map;
565 tree = &BTRFS_I(page->mapping->host)->io_tree;
566 map = &BTRFS_I(page->mapping->host)->extent_tree;
568 ret = try_release_extent_state(map, tree, page, gfp_flags);
573 ret = try_release_extent_buffer(tree, page);
575 ClearPagePrivate(page);
576 set_page_private(page, 0);
577 page_cache_release(page);
583 static void btree_invalidatepage(struct page *page, unsigned long offset)
585 struct extent_io_tree *tree;
586 tree = &BTRFS_I(page->mapping->host)->io_tree;
587 extent_invalidatepage(tree, page, offset);
588 btree_releasepage(page, GFP_NOFS);
589 if (PagePrivate(page)) {
590 printk("warning page private not zero on page %Lu\n",
592 ClearPagePrivate(page);
593 set_page_private(page, 0);
594 page_cache_release(page);
599 static int btree_writepage(struct page *page, struct writeback_control *wbc)
601 struct buffer_head *bh;
602 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
603 struct buffer_head *head;
604 if (!page_has_buffers(page)) {
605 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
606 (1 << BH_Dirty)|(1 << BH_Uptodate));
608 head = page_buffers(page);
611 if (buffer_dirty(bh))
612 csum_tree_block(root, bh, 0);
613 bh = bh->b_this_page;
614 } while (bh != head);
615 return block_write_full_page(page, btree_get_block, wbc);
619 static struct address_space_operations btree_aops = {
620 .readpage = btree_readpage,
621 .writepage = btree_writepage,
622 .writepages = btree_writepages,
623 .releasepage = btree_releasepage,
624 .invalidatepage = btree_invalidatepage,
625 .sync_page = block_sync_page,
628 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
631 struct extent_buffer *buf = NULL;
632 struct inode *btree_inode = root->fs_info->btree_inode;
635 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
638 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
639 buf, 0, 0, btree_get_extent, 0);
640 free_extent_buffer(buf);
644 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
645 u64 bytenr, u32 blocksize)
647 struct inode *btree_inode = root->fs_info->btree_inode;
648 struct extent_buffer *eb;
649 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
650 bytenr, blocksize, GFP_NOFS);
654 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
655 u64 bytenr, u32 blocksize)
657 struct inode *btree_inode = root->fs_info->btree_inode;
658 struct extent_buffer *eb;
660 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
661 bytenr, blocksize, NULL, GFP_NOFS);
666 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
667 u32 blocksize, u64 parent_transid)
669 struct extent_buffer *buf = NULL;
670 struct inode *btree_inode = root->fs_info->btree_inode;
671 struct extent_io_tree *io_tree;
674 io_tree = &BTRFS_I(btree_inode)->io_tree;
676 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
680 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
683 buf->flags |= EXTENT_UPTODATE;
689 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
690 struct extent_buffer *buf)
692 struct inode *btree_inode = root->fs_info->btree_inode;
693 if (btrfs_header_generation(buf) ==
694 root->fs_info->running_transaction->transid) {
695 WARN_ON(!btrfs_tree_locked(buf));
696 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
702 int wait_on_tree_block_writeback(struct btrfs_root *root,
703 struct extent_buffer *buf)
705 struct inode *btree_inode = root->fs_info->btree_inode;
706 wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
711 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
712 u32 stripesize, struct btrfs_root *root,
713 struct btrfs_fs_info *fs_info,
718 root->commit_root = NULL;
719 root->sectorsize = sectorsize;
720 root->nodesize = nodesize;
721 root->leafsize = leafsize;
722 root->stripesize = stripesize;
724 root->track_dirty = 0;
726 root->fs_info = fs_info;
727 root->objectid = objectid;
728 root->last_trans = 0;
729 root->highest_inode = 0;
730 root->last_inode_alloc = 0;
734 INIT_LIST_HEAD(&root->dirty_list);
735 INIT_LIST_HEAD(&root->orphan_list);
736 spin_lock_init(&root->node_lock);
737 spin_lock_init(&root->orphan_lock);
738 mutex_init(&root->objectid_mutex);
739 memset(&root->root_key, 0, sizeof(root->root_key));
740 memset(&root->root_item, 0, sizeof(root->root_item));
741 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
742 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
743 root->defrag_trans_start = fs_info->generation;
744 init_completion(&root->kobj_unregister);
745 root->defrag_running = 0;
746 root->defrag_level = 0;
747 root->root_key.objectid = objectid;
751 static int find_and_setup_root(struct btrfs_root *tree_root,
752 struct btrfs_fs_info *fs_info,
754 struct btrfs_root *root)
759 __setup_root(tree_root->nodesize, tree_root->leafsize,
760 tree_root->sectorsize, tree_root->stripesize,
761 root, fs_info, objectid);
762 ret = btrfs_find_last_root(tree_root, objectid,
763 &root->root_item, &root->root_key);
766 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
767 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
773 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
774 struct btrfs_key *location)
776 struct btrfs_root *root;
777 struct btrfs_root *tree_root = fs_info->tree_root;
778 struct btrfs_path *path;
779 struct extent_buffer *l;
784 root = kzalloc(sizeof(*root), GFP_NOFS);
786 return ERR_PTR(-ENOMEM);
787 if (location->offset == (u64)-1) {
788 ret = find_and_setup_root(tree_root, fs_info,
789 location->objectid, root);
797 __setup_root(tree_root->nodesize, tree_root->leafsize,
798 tree_root->sectorsize, tree_root->stripesize,
799 root, fs_info, location->objectid);
801 path = btrfs_alloc_path();
803 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
810 read_extent_buffer(l, &root->root_item,
811 btrfs_item_ptr_offset(l, path->slots[0]),
812 sizeof(root->root_item));
813 memcpy(&root->root_key, location, sizeof(*location));
816 btrfs_release_path(root, path);
817 btrfs_free_path(path);
822 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
823 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
828 ret = btrfs_find_highest_inode(root, &highest_inode);
830 root->highest_inode = highest_inode;
831 root->last_inode_alloc = highest_inode;
836 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
839 struct btrfs_root *root;
841 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
842 return fs_info->tree_root;
843 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
844 return fs_info->extent_root;
846 root = radix_tree_lookup(&fs_info->fs_roots_radix,
847 (unsigned long)root_objectid);
851 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
852 struct btrfs_key *location)
854 struct btrfs_root *root;
857 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
858 return fs_info->tree_root;
859 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
860 return fs_info->extent_root;
861 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
862 return fs_info->chunk_root;
863 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
864 return fs_info->dev_root;
866 root = radix_tree_lookup(&fs_info->fs_roots_radix,
867 (unsigned long)location->objectid);
871 root = btrfs_read_fs_root_no_radix(fs_info, location);
874 ret = radix_tree_insert(&fs_info->fs_roots_radix,
875 (unsigned long)root->root_key.objectid,
878 free_extent_buffer(root->node);
882 ret = btrfs_find_dead_roots(fs_info->tree_root,
883 root->root_key.objectid, root);
889 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
890 struct btrfs_key *location,
891 const char *name, int namelen)
893 struct btrfs_root *root;
896 root = btrfs_read_fs_root_no_name(fs_info, location);
903 ret = btrfs_set_root_name(root, name, namelen);
905 free_extent_buffer(root->node);
910 ret = btrfs_sysfs_add_root(root);
912 free_extent_buffer(root->node);
921 static int add_hasher(struct btrfs_fs_info *info, char *type) {
922 struct btrfs_hasher *hasher;
924 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
927 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
928 if (!hasher->hash_tfm) {
932 spin_lock(&info->hash_lock);
933 list_add(&hasher->list, &info->hashers);
934 spin_unlock(&info->hash_lock);
939 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
941 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
943 int limit = 256 * info->fs_devices->open_devices;
944 struct list_head *cur;
945 struct btrfs_device *device;
946 struct backing_dev_info *bdi;
948 if ((bdi_bits & (1 << BDI_write_congested)) &&
949 atomic_read(&info->nr_async_submits) > limit) {
953 list_for_each(cur, &info->fs_devices->devices) {
954 device = list_entry(cur, struct btrfs_device, dev_list);
957 bdi = blk_get_backing_dev_info(device->bdev);
958 if (bdi && bdi_congested(bdi, bdi_bits)) {
967 * this unplugs every device on the box, and it is only used when page
970 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
972 struct list_head *cur;
973 struct btrfs_device *device;
974 struct btrfs_fs_info *info;
976 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
977 list_for_each(cur, &info->fs_devices->devices) {
978 device = list_entry(cur, struct btrfs_device, dev_list);
979 bdi = blk_get_backing_dev_info(device->bdev);
980 if (bdi->unplug_io_fn) {
981 bdi->unplug_io_fn(bdi, page);
986 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
989 struct extent_map_tree *em_tree;
990 struct extent_map *em;
991 struct address_space *mapping;
994 /* the generic O_DIRECT read code does this */
996 __unplug_io_fn(bdi, page);
1001 * page->mapping may change at any time. Get a consistent copy
1002 * and use that for everything below
1005 mapping = page->mapping;
1009 inode = mapping->host;
1010 offset = page_offset(page);
1012 em_tree = &BTRFS_I(inode)->extent_tree;
1013 spin_lock(&em_tree->lock);
1014 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1015 spin_unlock(&em_tree->lock);
1017 __unplug_io_fn(bdi, page);
1021 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1022 free_extent_map(em);
1023 __unplug_io_fn(bdi, page);
1026 offset = offset - em->start;
1027 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1028 em->block_start + offset, page);
1029 free_extent_map(em);
1032 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1034 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1037 bdi->ra_pages = default_backing_dev_info.ra_pages;
1039 bdi->capabilities = default_backing_dev_info.capabilities;
1040 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1041 bdi->unplug_io_data = info;
1042 bdi->congested_fn = btrfs_congested_fn;
1043 bdi->congested_data = info;
1047 static int bio_ready_for_csum(struct bio *bio)
1053 struct extent_io_tree *io_tree = NULL;
1054 struct btrfs_fs_info *info = NULL;
1055 struct bio_vec *bvec;
1059 bio_for_each_segment(bvec, bio, i) {
1060 page = bvec->bv_page;
1061 if (page->private == EXTENT_PAGE_PRIVATE) {
1062 length += bvec->bv_len;
1065 if (!page->private) {
1066 length += bvec->bv_len;
1069 length = bvec->bv_len;
1070 buf_len = page->private >> 2;
1071 start = page_offset(page) + bvec->bv_offset;
1072 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1073 info = BTRFS_I(page->mapping->host)->root->fs_info;
1075 /* are we fully contained in this bio? */
1076 if (buf_len <= length)
1079 ret = extent_range_uptodate(io_tree, start + length,
1080 start + buf_len - 1);
1087 * called by the kthread helper functions to finally call the bio end_io
1088 * functions. This is where read checksum verification actually happens
1090 static void end_workqueue_fn(struct btrfs_work *work)
1093 struct end_io_wq *end_io_wq;
1094 struct btrfs_fs_info *fs_info;
1097 end_io_wq = container_of(work, struct end_io_wq, work);
1098 bio = end_io_wq->bio;
1099 fs_info = end_io_wq->info;
1101 /* metadata bios are special because the whole tree block must
1102 * be checksummed at once. This makes sure the entire block is in
1103 * ram and up to date before trying to verify things. For
1104 * blocksize <= pagesize, it is basically a noop
1106 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1107 btrfs_queue_worker(&fs_info->endio_workers,
1111 error = end_io_wq->error;
1112 bio->bi_private = end_io_wq->private;
1113 bio->bi_end_io = end_io_wq->end_io;
1115 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1116 bio_endio(bio, bio->bi_size, error);
1118 bio_endio(bio, error);
1122 static int cleaner_kthread(void *arg)
1124 struct btrfs_root *root = arg;
1128 if (root->fs_info->closing)
1131 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1132 mutex_lock(&root->fs_info->cleaner_mutex);
1133 btrfs_clean_old_snapshots(root);
1134 mutex_unlock(&root->fs_info->cleaner_mutex);
1136 if (freezing(current)) {
1140 if (root->fs_info->closing)
1142 set_current_state(TASK_INTERRUPTIBLE);
1144 __set_current_state(TASK_RUNNING);
1146 } while (!kthread_should_stop());
1150 static int transaction_kthread(void *arg)
1152 struct btrfs_root *root = arg;
1153 struct btrfs_trans_handle *trans;
1154 struct btrfs_transaction *cur;
1156 unsigned long delay;
1161 if (root->fs_info->closing)
1165 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1166 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1168 mutex_lock(&root->fs_info->trans_mutex);
1169 cur = root->fs_info->running_transaction;
1171 mutex_unlock(&root->fs_info->trans_mutex);
1174 now = get_seconds();
1175 if (now < cur->start_time || now - cur->start_time < 30) {
1176 mutex_unlock(&root->fs_info->trans_mutex);
1180 mutex_unlock(&root->fs_info->trans_mutex);
1181 trans = btrfs_start_transaction(root, 1);
1182 ret = btrfs_commit_transaction(trans, root);
1184 wake_up_process(root->fs_info->cleaner_kthread);
1185 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1187 if (freezing(current)) {
1190 if (root->fs_info->closing)
1192 set_current_state(TASK_INTERRUPTIBLE);
1193 schedule_timeout(delay);
1194 __set_current_state(TASK_RUNNING);
1196 } while (!kthread_should_stop());
1200 struct btrfs_root *open_ctree(struct super_block *sb,
1201 struct btrfs_fs_devices *fs_devices,
1209 struct buffer_head *bh;
1210 struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1212 struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1214 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1216 struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1218 struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1223 struct btrfs_super_block *disk_super;
1225 if (!extent_root || !tree_root || !fs_info) {
1229 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1230 INIT_LIST_HEAD(&fs_info->trans_list);
1231 INIT_LIST_HEAD(&fs_info->dead_roots);
1232 INIT_LIST_HEAD(&fs_info->hashers);
1233 spin_lock_init(&fs_info->hash_lock);
1234 spin_lock_init(&fs_info->delalloc_lock);
1235 spin_lock_init(&fs_info->new_trans_lock);
1237 init_completion(&fs_info->kobj_unregister);
1238 fs_info->tree_root = tree_root;
1239 fs_info->extent_root = extent_root;
1240 fs_info->chunk_root = chunk_root;
1241 fs_info->dev_root = dev_root;
1242 fs_info->fs_devices = fs_devices;
1243 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1244 INIT_LIST_HEAD(&fs_info->space_info);
1245 btrfs_mapping_init(&fs_info->mapping_tree);
1246 atomic_set(&fs_info->nr_async_submits, 0);
1247 atomic_set(&fs_info->throttles, 0);
1249 fs_info->max_extent = (u64)-1;
1250 fs_info->max_inline = 8192 * 1024;
1251 setup_bdi(fs_info, &fs_info->bdi);
1252 fs_info->btree_inode = new_inode(sb);
1253 fs_info->btree_inode->i_ino = 1;
1254 fs_info->btree_inode->i_nlink = 1;
1255 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1257 INIT_LIST_HEAD(&fs_info->ordered_extents);
1258 spin_lock_init(&fs_info->ordered_extent_lock);
1260 sb->s_blocksize = 4096;
1261 sb->s_blocksize_bits = blksize_bits(4096);
1264 * we set the i_size on the btree inode to the max possible int.
1265 * the real end of the address space is determined by all of
1266 * the devices in the system
1268 fs_info->btree_inode->i_size = OFFSET_MAX;
1269 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1270 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1272 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1273 fs_info->btree_inode->i_mapping,
1275 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1278 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1280 extent_io_tree_init(&fs_info->free_space_cache,
1281 fs_info->btree_inode->i_mapping, GFP_NOFS);
1282 extent_io_tree_init(&fs_info->block_group_cache,
1283 fs_info->btree_inode->i_mapping, GFP_NOFS);
1284 extent_io_tree_init(&fs_info->pinned_extents,
1285 fs_info->btree_inode->i_mapping, GFP_NOFS);
1286 extent_io_tree_init(&fs_info->pending_del,
1287 fs_info->btree_inode->i_mapping, GFP_NOFS);
1288 extent_io_tree_init(&fs_info->extent_ins,
1289 fs_info->btree_inode->i_mapping, GFP_NOFS);
1290 fs_info->do_barriers = 1;
1292 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1293 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1294 sizeof(struct btrfs_key));
1295 insert_inode_hash(fs_info->btree_inode);
1296 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1298 mutex_init(&fs_info->trans_mutex);
1299 mutex_init(&fs_info->drop_mutex);
1300 mutex_init(&fs_info->alloc_mutex);
1301 mutex_init(&fs_info->chunk_mutex);
1302 mutex_init(&fs_info->transaction_kthread_mutex);
1303 mutex_init(&fs_info->cleaner_mutex);
1304 mutex_init(&fs_info->volume_mutex);
1305 init_waitqueue_head(&fs_info->transaction_throttle);
1306 init_waitqueue_head(&fs_info->transaction_wait);
1309 ret = add_hasher(fs_info, "crc32c");
1311 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1316 __setup_root(4096, 4096, 4096, 4096, tree_root,
1317 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1320 bh = __bread(fs_devices->latest_bdev,
1321 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1325 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1328 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1330 disk_super = &fs_info->super_copy;
1331 if (!btrfs_super_root(disk_super))
1332 goto fail_sb_buffer;
1334 err = btrfs_parse_options(tree_root, options);
1336 goto fail_sb_buffer;
1339 * we need to start all the end_io workers up front because the
1340 * queue work function gets called at interrupt time, and so it
1341 * cannot dynamically grow.
1343 btrfs_init_workers(&fs_info->workers, fs_info->thread_pool_size);
1344 btrfs_init_workers(&fs_info->submit_workers, fs_info->thread_pool_size);
1345 btrfs_init_workers(&fs_info->fixup_workers, 1);
1346 btrfs_init_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1347 btrfs_init_workers(&fs_info->endio_write_workers,
1348 fs_info->thread_pool_size);
1349 btrfs_start_workers(&fs_info->workers, 1);
1350 btrfs_start_workers(&fs_info->submit_workers, 1);
1351 btrfs_start_workers(&fs_info->fixup_workers, 1);
1352 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1353 btrfs_start_workers(&fs_info->endio_write_workers,
1354 fs_info->thread_pool_size);
1357 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1358 printk("Btrfs: wanted %llu devices, but found %llu\n",
1359 (unsigned long long)btrfs_super_num_devices(disk_super),
1360 (unsigned long long)fs_devices->open_devices);
1361 if (btrfs_test_opt(tree_root, DEGRADED))
1362 printk("continuing in degraded mode\n");
1364 goto fail_sb_buffer;
1368 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1370 nodesize = btrfs_super_nodesize(disk_super);
1371 leafsize = btrfs_super_leafsize(disk_super);
1372 sectorsize = btrfs_super_sectorsize(disk_super);
1373 stripesize = btrfs_super_stripesize(disk_super);
1374 tree_root->nodesize = nodesize;
1375 tree_root->leafsize = leafsize;
1376 tree_root->sectorsize = sectorsize;
1377 tree_root->stripesize = stripesize;
1379 sb->s_blocksize = sectorsize;
1380 sb->s_blocksize_bits = blksize_bits(sectorsize);
1382 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1383 sizeof(disk_super->magic))) {
1384 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1385 goto fail_sb_buffer;
1388 mutex_lock(&fs_info->chunk_mutex);
1389 ret = btrfs_read_sys_array(tree_root);
1390 mutex_unlock(&fs_info->chunk_mutex);
1392 printk("btrfs: failed to read the system array on %s\n",
1394 goto fail_sys_array;
1397 blocksize = btrfs_level_size(tree_root,
1398 btrfs_super_chunk_root_level(disk_super));
1400 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1401 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1403 chunk_root->node = read_tree_block(chunk_root,
1404 btrfs_super_chunk_root(disk_super),
1406 BUG_ON(!chunk_root->node);
1408 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1409 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1412 mutex_lock(&fs_info->chunk_mutex);
1413 ret = btrfs_read_chunk_tree(chunk_root);
1414 mutex_unlock(&fs_info->chunk_mutex);
1417 btrfs_close_extra_devices(fs_devices);
1419 blocksize = btrfs_level_size(tree_root,
1420 btrfs_super_root_level(disk_super));
1423 tree_root->node = read_tree_block(tree_root,
1424 btrfs_super_root(disk_super),
1426 if (!tree_root->node)
1427 goto fail_sb_buffer;
1430 ret = find_and_setup_root(tree_root, fs_info,
1431 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1433 goto fail_tree_root;
1434 extent_root->track_dirty = 1;
1436 ret = find_and_setup_root(tree_root, fs_info,
1437 BTRFS_DEV_TREE_OBJECTID, dev_root);
1438 dev_root->track_dirty = 1;
1441 goto fail_extent_root;
1443 btrfs_read_block_groups(extent_root);
1445 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1446 fs_info->data_alloc_profile = (u64)-1;
1447 fs_info->metadata_alloc_profile = (u64)-1;
1448 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1449 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1451 if (!fs_info->cleaner_kthread)
1452 goto fail_extent_root;
1454 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1456 "btrfs-transaction");
1457 if (!fs_info->transaction_kthread)
1464 kthread_stop(fs_info->cleaner_kthread);
1466 free_extent_buffer(extent_root->node);
1468 free_extent_buffer(tree_root->node);
1471 btrfs_stop_workers(&fs_info->fixup_workers);
1472 btrfs_stop_workers(&fs_info->workers);
1473 btrfs_stop_workers(&fs_info->endio_workers);
1474 btrfs_stop_workers(&fs_info->endio_write_workers);
1475 btrfs_stop_workers(&fs_info->submit_workers);
1477 iput(fs_info->btree_inode);
1479 btrfs_close_devices(fs_info->fs_devices);
1480 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1484 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1485 bdi_destroy(&fs_info->bdi);
1488 return ERR_PTR(err);
1491 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1493 char b[BDEVNAME_SIZE];
1496 set_buffer_uptodate(bh);
1498 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1499 printk(KERN_WARNING "lost page write due to "
1500 "I/O error on %s\n",
1501 bdevname(bh->b_bdev, b));
1503 /* note, we dont' set_buffer_write_io_error because we have
1504 * our own ways of dealing with the IO errors
1506 clear_buffer_uptodate(bh);
1512 int write_all_supers(struct btrfs_root *root)
1514 struct list_head *cur;
1515 struct list_head *head = &root->fs_info->fs_devices->devices;
1516 struct btrfs_device *dev;
1517 struct btrfs_super_block *sb;
1518 struct btrfs_dev_item *dev_item;
1519 struct buffer_head *bh;
1523 int total_errors = 0;
1527 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1528 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1530 sb = &root->fs_info->super_for_commit;
1531 dev_item = &sb->dev_item;
1532 list_for_each(cur, head) {
1533 dev = list_entry(cur, struct btrfs_device, dev_list);
1538 if (!dev->in_fs_metadata)
1541 btrfs_set_stack_device_type(dev_item, dev->type);
1542 btrfs_set_stack_device_id(dev_item, dev->devid);
1543 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1544 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1545 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1546 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1547 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1548 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1549 flags = btrfs_super_flags(sb);
1550 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1554 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1555 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1556 btrfs_csum_final(crc, sb->csum);
1558 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1559 BTRFS_SUPER_INFO_SIZE);
1561 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1562 dev->pending_io = bh;
1565 set_buffer_uptodate(bh);
1567 bh->b_end_io = btrfs_end_buffer_write_sync;
1569 if (do_barriers && dev->barriers) {
1570 ret = submit_bh(WRITE_BARRIER, bh);
1571 if (ret == -EOPNOTSUPP) {
1572 printk("btrfs: disabling barriers on dev %s\n",
1574 set_buffer_uptodate(bh);
1578 ret = submit_bh(WRITE, bh);
1581 ret = submit_bh(WRITE, bh);
1586 if (total_errors > max_errors) {
1587 printk("btrfs: %d errors while writing supers\n", total_errors);
1592 list_for_each(cur, head) {
1593 dev = list_entry(cur, struct btrfs_device, dev_list);
1596 if (!dev->in_fs_metadata)
1599 BUG_ON(!dev->pending_io);
1600 bh = dev->pending_io;
1602 if (!buffer_uptodate(dev->pending_io)) {
1603 if (do_barriers && dev->barriers) {
1604 printk("btrfs: disabling barriers on dev %s\n",
1606 set_buffer_uptodate(bh);
1610 ret = submit_bh(WRITE, bh);
1613 if (!buffer_uptodate(bh))
1620 dev->pending_io = NULL;
1623 if (total_errors > max_errors) {
1624 printk("btrfs: %d errors while writing supers\n", total_errors);
1630 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1635 ret = write_all_supers(root);
1639 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1641 radix_tree_delete(&fs_info->fs_roots_radix,
1642 (unsigned long)root->root_key.objectid);
1644 btrfs_sysfs_del_root(root);
1648 free_extent_buffer(root->node);
1649 if (root->commit_root)
1650 free_extent_buffer(root->commit_root);
1657 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1660 struct btrfs_root *gang[8];
1664 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1669 for (i = 0; i < ret; i++)
1670 btrfs_free_fs_root(fs_info, gang[i]);
1675 int close_ctree(struct btrfs_root *root)
1678 struct btrfs_trans_handle *trans;
1679 struct btrfs_fs_info *fs_info = root->fs_info;
1681 fs_info->closing = 1;
1684 kthread_stop(root->fs_info->transaction_kthread);
1685 kthread_stop(root->fs_info->cleaner_kthread);
1687 btrfs_clean_old_snapshots(root);
1688 trans = btrfs_start_transaction(root, 1);
1689 ret = btrfs_commit_transaction(trans, root);
1690 /* run commit again to drop the original snapshot */
1691 trans = btrfs_start_transaction(root, 1);
1692 btrfs_commit_transaction(trans, root);
1693 ret = btrfs_write_and_wait_transaction(NULL, root);
1696 write_ctree_super(NULL, root);
1698 if (fs_info->delalloc_bytes) {
1699 printk("btrfs: at unmount delalloc count %Lu\n",
1700 fs_info->delalloc_bytes);
1702 if (fs_info->extent_root->node)
1703 free_extent_buffer(fs_info->extent_root->node);
1705 if (fs_info->tree_root->node)
1706 free_extent_buffer(fs_info->tree_root->node);
1708 if (root->fs_info->chunk_root->node);
1709 free_extent_buffer(root->fs_info->chunk_root->node);
1711 if (root->fs_info->dev_root->node);
1712 free_extent_buffer(root->fs_info->dev_root->node);
1714 btrfs_free_block_groups(root->fs_info);
1715 del_fs_roots(fs_info);
1717 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1719 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1721 btrfs_stop_workers(&fs_info->fixup_workers);
1722 btrfs_stop_workers(&fs_info->workers);
1723 btrfs_stop_workers(&fs_info->endio_workers);
1724 btrfs_stop_workers(&fs_info->endio_write_workers);
1725 btrfs_stop_workers(&fs_info->submit_workers);
1727 iput(fs_info->btree_inode);
1729 while(!list_empty(&fs_info->hashers)) {
1730 struct btrfs_hasher *hasher;
1731 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1733 list_del(&hasher->hashers);
1734 crypto_free_hash(&fs_info->hash_tfm);
1738 btrfs_close_devices(fs_info->fs_devices);
1739 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1741 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1742 bdi_destroy(&fs_info->bdi);
1745 kfree(fs_info->extent_root);
1746 kfree(fs_info->tree_root);
1747 kfree(fs_info->chunk_root);
1748 kfree(fs_info->dev_root);
1752 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1755 struct inode *btree_inode = buf->first_page->mapping->host;
1757 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1761 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1766 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1768 struct inode *btree_inode = buf->first_page->mapping->host;
1769 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1773 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1775 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1776 u64 transid = btrfs_header_generation(buf);
1777 struct inode *btree_inode = root->fs_info->btree_inode;
1779 WARN_ON(!btrfs_tree_locked(buf));
1780 if (transid != root->fs_info->generation) {
1781 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1782 (unsigned long long)buf->start,
1783 transid, root->fs_info->generation);
1786 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1789 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1792 * looks as though older kernels can get into trouble with
1793 * this code, they end up stuck in balance_dirty_pages forever
1795 struct extent_io_tree *tree;
1798 unsigned long thresh = 16 * 1024 * 1024;
1799 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1801 if (current_is_pdflush())
1804 num_dirty = count_range_bits(tree, &start, (u64)-1,
1805 thresh, EXTENT_DIRTY);
1806 if (num_dirty > thresh) {
1807 balance_dirty_pages_ratelimited_nr(
1808 root->fs_info->btree_inode->i_mapping, 1);
1813 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1815 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1817 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1819 buf->flags |= EXTENT_UPTODATE;
1824 static struct extent_io_ops btree_extent_io_ops = {
1825 .writepage_io_hook = btree_writepage_io_hook,
1826 .readpage_end_io_hook = btree_readpage_end_io_hook,
1827 .submit_bio_hook = btree_submit_bio_hook,
1828 /* note we're sharing with inode.c for the merge bio hook */
1829 .merge_bio_hook = btrfs_merge_bio_hook,