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"
43 #include "ref-cache.h"
46 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
48 if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
49 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
50 (unsigned long long)extent_buffer_blocknr(buf),
51 (unsigned long long)btrfs_header_blocknr(buf));
58 static struct extent_io_ops btree_extent_io_ops;
59 static void end_workqueue_fn(struct btrfs_work *work);
65 struct btrfs_fs_info *info;
68 struct list_head list;
69 struct btrfs_work work;
72 struct async_submit_bio {
75 struct list_head list;
76 extent_submit_bio_hook_t *submit_bio_hook;
79 struct btrfs_work work;
82 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
83 size_t page_offset, u64 start, u64 len,
86 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
87 struct extent_map *em;
90 spin_lock(&em_tree->lock);
91 em = lookup_extent_mapping(em_tree, start, len);
94 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
95 spin_unlock(&em_tree->lock);
98 spin_unlock(&em_tree->lock);
100 em = alloc_extent_map(GFP_NOFS);
102 em = ERR_PTR(-ENOMEM);
108 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
110 spin_lock(&em_tree->lock);
111 ret = add_extent_mapping(em_tree, em);
112 if (ret == -EEXIST) {
113 u64 failed_start = em->start;
114 u64 failed_len = em->len;
116 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
117 em->start, em->len, em->block_start);
119 em = lookup_extent_mapping(em_tree, start, len);
121 printk("after failing, found %Lu %Lu %Lu\n",
122 em->start, em->len, em->block_start);
125 em = lookup_extent_mapping(em_tree, failed_start,
128 printk("double failure lookup gives us "
129 "%Lu %Lu -> %Lu\n", em->start,
130 em->len, em->block_start);
139 spin_unlock(&em_tree->lock);
147 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
149 return btrfs_crc32c(seed, data, len);
152 void btrfs_csum_final(u32 crc, char *result)
154 *(__le32 *)result = ~cpu_to_le32(crc);
157 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
160 char result[BTRFS_CRC32_SIZE];
162 unsigned long cur_len;
163 unsigned long offset = BTRFS_CSUM_SIZE;
164 char *map_token = NULL;
166 unsigned long map_start;
167 unsigned long map_len;
171 len = buf->len - offset;
173 err = map_private_extent_buffer(buf, offset, 32,
175 &map_start, &map_len, KM_USER0);
177 printk("failed to map extent buffer! %lu\n",
181 cur_len = min(len, map_len - (offset - map_start));
182 crc = btrfs_csum_data(root, kaddr + offset - map_start,
186 unmap_extent_buffer(buf, map_token, KM_USER0);
188 btrfs_csum_final(crc, result);
191 /* FIXME, this is not good */
192 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
195 memcpy(&found, result, BTRFS_CRC32_SIZE);
197 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
198 printk("btrfs: %s checksum verify failed on %llu "
199 "wanted %X found %X level %d\n",
200 root->fs_info->sb->s_id,
201 buf->start, val, found, btrfs_header_level(buf));
205 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
210 static int verify_parent_transid(struct extent_io_tree *io_tree,
211 struct extent_buffer *eb, u64 parent_transid)
215 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
218 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
219 if (extent_buffer_uptodate(io_tree, eb) &&
220 btrfs_header_generation(eb) == parent_transid) {
224 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
225 (unsigned long long)eb->start,
226 (unsigned long long)parent_transid,
227 (unsigned long long)btrfs_header_generation(eb));
229 clear_extent_buffer_uptodate(io_tree, eb);
231 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
237 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
238 struct extent_buffer *eb,
239 u64 start, u64 parent_transid)
241 struct extent_io_tree *io_tree;
246 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
248 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
249 btree_get_extent, mirror_num);
251 !verify_parent_transid(io_tree, eb, parent_transid))
254 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
260 if (mirror_num > num_copies)
266 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
268 struct extent_io_tree *tree;
269 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
273 struct extent_buffer *eb;
276 tree = &BTRFS_I(page->mapping->host)->io_tree;
278 if (page->private == EXTENT_PAGE_PRIVATE)
282 len = page->private >> 2;
286 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
287 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
288 btrfs_header_generation(eb));
290 found_start = btrfs_header_bytenr(eb);
291 if (found_start != start) {
292 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
293 start, found_start, len);
297 if (eb->first_page != page) {
298 printk("bad first page %lu %lu\n", eb->first_page->index,
303 if (!PageUptodate(page)) {
304 printk("csum not up to date page %lu\n", page->index);
308 found_level = btrfs_header_level(eb);
309 spin_lock(&root->fs_info->hash_lock);
310 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
311 spin_unlock(&root->fs_info->hash_lock);
312 csum_tree_block(root, eb, 0);
314 free_extent_buffer(eb);
319 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
321 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
323 csum_dirty_buffer(root, page);
327 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
328 struct extent_state *state)
330 struct extent_io_tree *tree;
334 struct extent_buffer *eb;
335 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
338 tree = &BTRFS_I(page->mapping->host)->io_tree;
339 if (page->private == EXTENT_PAGE_PRIVATE)
343 len = page->private >> 2;
347 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
349 found_start = btrfs_header_bytenr(eb);
350 if (found_start != start) {
354 if (eb->first_page != page) {
355 printk("bad first page %lu %lu\n", eb->first_page->index,
361 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
362 (unsigned long)btrfs_header_fsid(eb),
364 printk("bad fsid on block %Lu\n", eb->start);
368 found_level = btrfs_header_level(eb);
370 ret = csum_tree_block(root, eb, 1);
374 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
375 end = eb->start + end - 1;
377 free_extent_buffer(eb);
382 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
383 static void end_workqueue_bio(struct bio *bio, int err)
385 static int end_workqueue_bio(struct bio *bio,
386 unsigned int bytes_done, int err)
389 struct end_io_wq *end_io_wq = bio->bi_private;
390 struct btrfs_fs_info *fs_info;
392 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
397 fs_info = end_io_wq->info;
398 end_io_wq->error = err;
399 end_io_wq->work.func = end_workqueue_fn;
400 end_io_wq->work.flags = 0;
401 if (bio->bi_rw & (1 << BIO_RW))
402 btrfs_queue_worker(&fs_info->endio_write_workers,
405 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
407 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
412 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
415 struct end_io_wq *end_io_wq;
416 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
420 end_io_wq->private = bio->bi_private;
421 end_io_wq->end_io = bio->bi_end_io;
422 end_io_wq->info = info;
423 end_io_wq->error = 0;
424 end_io_wq->bio = bio;
425 end_io_wq->metadata = metadata;
427 bio->bi_private = end_io_wq;
428 bio->bi_end_io = end_workqueue_bio;
432 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
434 int limit = 256 * info->fs_devices->open_devices;
437 limit = (limit * 3) / 2;
438 if (atomic_read(&info->nr_async_submits) > limit)
441 return atomic_read(&info->nr_async_bios) > limit;
444 static void run_one_async_submit(struct btrfs_work *work)
446 struct btrfs_fs_info *fs_info;
447 struct async_submit_bio *async;
449 async = container_of(work, struct async_submit_bio, work);
450 fs_info = BTRFS_I(async->inode)->root->fs_info;
451 atomic_dec(&fs_info->nr_async_submits);
453 if ((async->bio->bi_rw & (1 << BIO_RW)) &&
454 !btrfs_congested_async(fs_info, 1)) {
455 clear_bdi_congested(&fs_info->bdi, WRITE);
457 async->submit_bio_hook(async->inode, async->rw, async->bio,
462 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
463 int rw, struct bio *bio, int mirror_num,
464 extent_submit_bio_hook_t *submit_bio_hook)
466 struct async_submit_bio *async;
468 async = kmalloc(sizeof(*async), GFP_NOFS);
472 async->inode = inode;
475 async->mirror_num = mirror_num;
476 async->submit_bio_hook = submit_bio_hook;
477 async->work.func = run_one_async_submit;
478 async->work.flags = 0;
479 atomic_inc(&fs_info->nr_async_submits);
480 btrfs_queue_worker(&fs_info->workers, &async->work);
484 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
487 struct btrfs_root *root = BTRFS_I(inode)->root;
491 offset = bio->bi_sector << 9;
494 * when we're called for a write, we're already in the async
495 * submission context. Just jump ingo btrfs_map_bio
497 if (rw & (1 << BIO_RW)) {
498 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
503 * called for a read, do the setup so that checksum validation
504 * can happen in the async kernel threads
506 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
509 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
512 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
516 * kthread helpers are used to submit writes so that checksumming
517 * can happen in parallel across all CPUs
519 if (!(rw & (1 << BIO_RW))) {
520 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
522 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
523 inode, rw, bio, mirror_num,
524 __btree_submit_bio_hook);
527 static int btree_writepage(struct page *page, struct writeback_control *wbc)
529 struct extent_io_tree *tree;
530 tree = &BTRFS_I(page->mapping->host)->io_tree;
531 return extent_write_full_page(tree, page, btree_get_extent, wbc);
534 static int btree_writepages(struct address_space *mapping,
535 struct writeback_control *wbc)
537 struct extent_io_tree *tree;
538 tree = &BTRFS_I(mapping->host)->io_tree;
539 if (wbc->sync_mode == WB_SYNC_NONE) {
542 unsigned long thresh = 96 * 1024 * 1024;
544 if (wbc->for_kupdate)
547 if (current_is_pdflush()) {
548 thresh = 96 * 1024 * 1024;
550 thresh = 8 * 1024 * 1024;
552 num_dirty = count_range_bits(tree, &start, (u64)-1,
553 thresh, EXTENT_DIRTY);
554 if (num_dirty < thresh) {
558 return extent_writepages(tree, mapping, btree_get_extent, wbc);
561 int btree_readpage(struct file *file, struct page *page)
563 struct extent_io_tree *tree;
564 tree = &BTRFS_I(page->mapping->host)->io_tree;
565 return extent_read_full_page(tree, page, btree_get_extent);
568 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
570 struct extent_io_tree *tree;
571 struct extent_map_tree *map;
574 tree = &BTRFS_I(page->mapping->host)->io_tree;
575 map = &BTRFS_I(page->mapping->host)->extent_tree;
577 ret = try_release_extent_state(map, tree, page, gfp_flags);
582 ret = try_release_extent_buffer(tree, page);
584 ClearPagePrivate(page);
585 set_page_private(page, 0);
586 page_cache_release(page);
592 static void btree_invalidatepage(struct page *page, unsigned long offset)
594 struct extent_io_tree *tree;
595 tree = &BTRFS_I(page->mapping->host)->io_tree;
596 extent_invalidatepage(tree, page, offset);
597 btree_releasepage(page, GFP_NOFS);
598 if (PagePrivate(page)) {
599 printk("warning page private not zero on page %Lu\n",
601 ClearPagePrivate(page);
602 set_page_private(page, 0);
603 page_cache_release(page);
608 static int btree_writepage(struct page *page, struct writeback_control *wbc)
610 struct buffer_head *bh;
611 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
612 struct buffer_head *head;
613 if (!page_has_buffers(page)) {
614 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
615 (1 << BH_Dirty)|(1 << BH_Uptodate));
617 head = page_buffers(page);
620 if (buffer_dirty(bh))
621 csum_tree_block(root, bh, 0);
622 bh = bh->b_this_page;
623 } while (bh != head);
624 return block_write_full_page(page, btree_get_block, wbc);
628 static struct address_space_operations btree_aops = {
629 .readpage = btree_readpage,
630 .writepage = btree_writepage,
631 .writepages = btree_writepages,
632 .releasepage = btree_releasepage,
633 .invalidatepage = btree_invalidatepage,
634 .sync_page = block_sync_page,
637 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
640 struct extent_buffer *buf = NULL;
641 struct inode *btree_inode = root->fs_info->btree_inode;
644 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
647 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
648 buf, 0, 0, btree_get_extent, 0);
649 free_extent_buffer(buf);
653 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
654 u64 bytenr, u32 blocksize)
656 struct inode *btree_inode = root->fs_info->btree_inode;
657 struct extent_buffer *eb;
658 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
659 bytenr, blocksize, GFP_NOFS);
663 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
664 u64 bytenr, u32 blocksize)
666 struct inode *btree_inode = root->fs_info->btree_inode;
667 struct extent_buffer *eb;
669 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
670 bytenr, blocksize, NULL, GFP_NOFS);
675 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
676 u32 blocksize, u64 parent_transid)
678 struct extent_buffer *buf = NULL;
679 struct inode *btree_inode = root->fs_info->btree_inode;
680 struct extent_io_tree *io_tree;
683 io_tree = &BTRFS_I(btree_inode)->io_tree;
685 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
689 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
692 buf->flags |= EXTENT_UPTODATE;
698 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
699 struct extent_buffer *buf)
701 struct inode *btree_inode = root->fs_info->btree_inode;
702 if (btrfs_header_generation(buf) ==
703 root->fs_info->running_transaction->transid) {
704 WARN_ON(!btrfs_tree_locked(buf));
705 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
711 int wait_on_tree_block_writeback(struct btrfs_root *root,
712 struct extent_buffer *buf)
714 struct inode *btree_inode = root->fs_info->btree_inode;
715 wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
720 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
721 u32 stripesize, struct btrfs_root *root,
722 struct btrfs_fs_info *fs_info,
727 root->commit_root = NULL;
728 root->ref_tree = NULL;
729 root->sectorsize = sectorsize;
730 root->nodesize = nodesize;
731 root->leafsize = leafsize;
732 root->stripesize = stripesize;
734 root->track_dirty = 0;
736 root->fs_info = fs_info;
737 root->objectid = objectid;
738 root->last_trans = 0;
739 root->highest_inode = 0;
740 root->last_inode_alloc = 0;
744 INIT_LIST_HEAD(&root->dirty_list);
745 INIT_LIST_HEAD(&root->orphan_list);
746 INIT_LIST_HEAD(&root->dead_list);
747 spin_lock_init(&root->node_lock);
748 spin_lock_init(&root->list_lock);
749 mutex_init(&root->objectid_mutex);
751 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
752 root->ref_tree = &root->ref_tree_struct;
754 memset(&root->root_key, 0, sizeof(root->root_key));
755 memset(&root->root_item, 0, sizeof(root->root_item));
756 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
757 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
758 root->defrag_trans_start = fs_info->generation;
759 init_completion(&root->kobj_unregister);
760 root->defrag_running = 0;
761 root->defrag_level = 0;
762 root->root_key.objectid = objectid;
766 static int find_and_setup_root(struct btrfs_root *tree_root,
767 struct btrfs_fs_info *fs_info,
769 struct btrfs_root *root)
774 __setup_root(tree_root->nodesize, tree_root->leafsize,
775 tree_root->sectorsize, tree_root->stripesize,
776 root, fs_info, objectid);
777 ret = btrfs_find_last_root(tree_root, objectid,
778 &root->root_item, &root->root_key);
781 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
782 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
788 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
789 struct btrfs_key *location)
791 struct btrfs_root *root;
792 struct btrfs_root *tree_root = fs_info->tree_root;
793 struct btrfs_path *path;
794 struct extent_buffer *l;
799 root = kzalloc(sizeof(*root), GFP_NOFS);
801 return ERR_PTR(-ENOMEM);
802 if (location->offset == (u64)-1) {
803 ret = find_and_setup_root(tree_root, fs_info,
804 location->objectid, root);
812 __setup_root(tree_root->nodesize, tree_root->leafsize,
813 tree_root->sectorsize, tree_root->stripesize,
814 root, fs_info, location->objectid);
816 path = btrfs_alloc_path();
818 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
825 read_extent_buffer(l, &root->root_item,
826 btrfs_item_ptr_offset(l, path->slots[0]),
827 sizeof(root->root_item));
828 memcpy(&root->root_key, location, sizeof(*location));
831 btrfs_release_path(root, path);
832 btrfs_free_path(path);
837 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
838 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
843 ret = btrfs_find_highest_inode(root, &highest_inode);
845 root->highest_inode = highest_inode;
846 root->last_inode_alloc = highest_inode;
851 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
854 struct btrfs_root *root;
856 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
857 return fs_info->tree_root;
858 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
859 return fs_info->extent_root;
861 root = radix_tree_lookup(&fs_info->fs_roots_radix,
862 (unsigned long)root_objectid);
866 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
867 struct btrfs_key *location)
869 struct btrfs_root *root;
872 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
873 return fs_info->tree_root;
874 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
875 return fs_info->extent_root;
876 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
877 return fs_info->chunk_root;
878 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
879 return fs_info->dev_root;
881 root = radix_tree_lookup(&fs_info->fs_roots_radix,
882 (unsigned long)location->objectid);
886 root = btrfs_read_fs_root_no_radix(fs_info, location);
889 ret = radix_tree_insert(&fs_info->fs_roots_radix,
890 (unsigned long)root->root_key.objectid,
893 free_extent_buffer(root->node);
897 ret = btrfs_find_dead_roots(fs_info->tree_root,
898 root->root_key.objectid, root);
904 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
905 struct btrfs_key *location,
906 const char *name, int namelen)
908 struct btrfs_root *root;
911 root = btrfs_read_fs_root_no_name(fs_info, location);
918 ret = btrfs_set_root_name(root, name, namelen);
920 free_extent_buffer(root->node);
925 ret = btrfs_sysfs_add_root(root);
927 free_extent_buffer(root->node);
936 static int add_hasher(struct btrfs_fs_info *info, char *type) {
937 struct btrfs_hasher *hasher;
939 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
942 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
943 if (!hasher->hash_tfm) {
947 spin_lock(&info->hash_lock);
948 list_add(&hasher->list, &info->hashers);
949 spin_unlock(&info->hash_lock);
954 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
956 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
958 struct list_head *cur;
959 struct btrfs_device *device;
960 struct backing_dev_info *bdi;
962 if ((bdi_bits & (1 << BDI_write_congested)) &&
963 btrfs_congested_async(info, 0))
966 list_for_each(cur, &info->fs_devices->devices) {
967 device = list_entry(cur, struct btrfs_device, dev_list);
970 bdi = blk_get_backing_dev_info(device->bdev);
971 if (bdi && bdi_congested(bdi, bdi_bits)) {
980 * this unplugs every device on the box, and it is only used when page
983 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
985 struct list_head *cur;
986 struct btrfs_device *device;
987 struct btrfs_fs_info *info;
989 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
990 list_for_each(cur, &info->fs_devices->devices) {
991 device = list_entry(cur, struct btrfs_device, dev_list);
992 bdi = blk_get_backing_dev_info(device->bdev);
993 if (bdi->unplug_io_fn) {
994 bdi->unplug_io_fn(bdi, page);
999 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1001 struct inode *inode;
1002 struct extent_map_tree *em_tree;
1003 struct extent_map *em;
1004 struct address_space *mapping;
1007 /* the generic O_DIRECT read code does this */
1009 __unplug_io_fn(bdi, page);
1014 * page->mapping may change at any time. Get a consistent copy
1015 * and use that for everything below
1018 mapping = page->mapping;
1022 inode = mapping->host;
1023 offset = page_offset(page);
1025 em_tree = &BTRFS_I(inode)->extent_tree;
1026 spin_lock(&em_tree->lock);
1027 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1028 spin_unlock(&em_tree->lock);
1030 __unplug_io_fn(bdi, page);
1034 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1035 free_extent_map(em);
1036 __unplug_io_fn(bdi, page);
1039 offset = offset - em->start;
1040 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1041 em->block_start + offset, page);
1042 free_extent_map(em);
1045 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1047 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1050 bdi->ra_pages = default_backing_dev_info.ra_pages;
1052 bdi->capabilities = default_backing_dev_info.capabilities;
1053 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1054 bdi->unplug_io_data = info;
1055 bdi->congested_fn = btrfs_congested_fn;
1056 bdi->congested_data = info;
1060 static int bio_ready_for_csum(struct bio *bio)
1066 struct extent_io_tree *io_tree = NULL;
1067 struct btrfs_fs_info *info = NULL;
1068 struct bio_vec *bvec;
1072 bio_for_each_segment(bvec, bio, i) {
1073 page = bvec->bv_page;
1074 if (page->private == EXTENT_PAGE_PRIVATE) {
1075 length += bvec->bv_len;
1078 if (!page->private) {
1079 length += bvec->bv_len;
1082 length = bvec->bv_len;
1083 buf_len = page->private >> 2;
1084 start = page_offset(page) + bvec->bv_offset;
1085 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1086 info = BTRFS_I(page->mapping->host)->root->fs_info;
1088 /* are we fully contained in this bio? */
1089 if (buf_len <= length)
1092 ret = extent_range_uptodate(io_tree, start + length,
1093 start + buf_len - 1);
1100 * called by the kthread helper functions to finally call the bio end_io
1101 * functions. This is where read checksum verification actually happens
1103 static void end_workqueue_fn(struct btrfs_work *work)
1106 struct end_io_wq *end_io_wq;
1107 struct btrfs_fs_info *fs_info;
1110 end_io_wq = container_of(work, struct end_io_wq, work);
1111 bio = end_io_wq->bio;
1112 fs_info = end_io_wq->info;
1114 /* metadata bios are special because the whole tree block must
1115 * be checksummed at once. This makes sure the entire block is in
1116 * ram and up to date before trying to verify things. For
1117 * blocksize <= pagesize, it is basically a noop
1119 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1120 btrfs_queue_worker(&fs_info->endio_workers,
1124 error = end_io_wq->error;
1125 bio->bi_private = end_io_wq->private;
1126 bio->bi_end_io = end_io_wq->end_io;
1128 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1129 bio_endio(bio, bio->bi_size, error);
1131 bio_endio(bio, error);
1135 static int cleaner_kthread(void *arg)
1137 struct btrfs_root *root = arg;
1141 if (root->fs_info->closing)
1144 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1145 mutex_lock(&root->fs_info->cleaner_mutex);
1146 btrfs_clean_old_snapshots(root);
1147 mutex_unlock(&root->fs_info->cleaner_mutex);
1149 if (freezing(current)) {
1153 if (root->fs_info->closing)
1155 set_current_state(TASK_INTERRUPTIBLE);
1157 __set_current_state(TASK_RUNNING);
1159 } while (!kthread_should_stop());
1163 static int transaction_kthread(void *arg)
1165 struct btrfs_root *root = arg;
1166 struct btrfs_trans_handle *trans;
1167 struct btrfs_transaction *cur;
1169 unsigned long delay;
1174 if (root->fs_info->closing)
1178 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1179 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1181 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1182 printk("btrfs: total reference cache size %Lu\n",
1183 root->fs_info->total_ref_cache_size);
1186 mutex_lock(&root->fs_info->trans_mutex);
1187 cur = root->fs_info->running_transaction;
1189 mutex_unlock(&root->fs_info->trans_mutex);
1193 now = get_seconds();
1194 if (now < cur->start_time || now - cur->start_time < 30) {
1195 mutex_unlock(&root->fs_info->trans_mutex);
1199 mutex_unlock(&root->fs_info->trans_mutex);
1200 trans = btrfs_start_transaction(root, 1);
1201 ret = btrfs_commit_transaction(trans, root);
1203 wake_up_process(root->fs_info->cleaner_kthread);
1204 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1206 if (freezing(current)) {
1209 if (root->fs_info->closing)
1211 set_current_state(TASK_INTERRUPTIBLE);
1212 schedule_timeout(delay);
1213 __set_current_state(TASK_RUNNING);
1215 } while (!kthread_should_stop());
1219 struct btrfs_root *open_ctree(struct super_block *sb,
1220 struct btrfs_fs_devices *fs_devices,
1228 struct buffer_head *bh;
1229 struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1231 struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1233 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1235 struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1237 struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1242 struct btrfs_super_block *disk_super;
1244 if (!extent_root || !tree_root || !fs_info) {
1248 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1249 INIT_LIST_HEAD(&fs_info->trans_list);
1250 INIT_LIST_HEAD(&fs_info->dead_roots);
1251 INIT_LIST_HEAD(&fs_info->hashers);
1252 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1253 spin_lock_init(&fs_info->hash_lock);
1254 spin_lock_init(&fs_info->delalloc_lock);
1255 spin_lock_init(&fs_info->new_trans_lock);
1256 spin_lock_init(&fs_info->ref_cache_lock);
1258 init_completion(&fs_info->kobj_unregister);
1259 fs_info->tree_root = tree_root;
1260 fs_info->extent_root = extent_root;
1261 fs_info->chunk_root = chunk_root;
1262 fs_info->dev_root = dev_root;
1263 fs_info->fs_devices = fs_devices;
1264 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1265 INIT_LIST_HEAD(&fs_info->space_info);
1266 btrfs_mapping_init(&fs_info->mapping_tree);
1267 atomic_set(&fs_info->nr_async_submits, 0);
1268 atomic_set(&fs_info->nr_async_bios, 0);
1269 atomic_set(&fs_info->throttles, 0);
1270 atomic_set(&fs_info->throttle_gen, 0);
1272 fs_info->max_extent = (u64)-1;
1273 fs_info->max_inline = 8192 * 1024;
1274 setup_bdi(fs_info, &fs_info->bdi);
1275 fs_info->btree_inode = new_inode(sb);
1276 fs_info->btree_inode->i_ino = 1;
1277 fs_info->btree_inode->i_nlink = 1;
1278 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1280 INIT_LIST_HEAD(&fs_info->ordered_extents);
1281 spin_lock_init(&fs_info->ordered_extent_lock);
1283 sb->s_blocksize = 4096;
1284 sb->s_blocksize_bits = blksize_bits(4096);
1287 * we set the i_size on the btree inode to the max possible int.
1288 * the real end of the address space is determined by all of
1289 * the devices in the system
1291 fs_info->btree_inode->i_size = OFFSET_MAX;
1292 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1293 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1295 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1296 fs_info->btree_inode->i_mapping,
1298 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1301 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1303 extent_io_tree_init(&fs_info->free_space_cache,
1304 fs_info->btree_inode->i_mapping, GFP_NOFS);
1305 extent_io_tree_init(&fs_info->block_group_cache,
1306 fs_info->btree_inode->i_mapping, GFP_NOFS);
1307 extent_io_tree_init(&fs_info->pinned_extents,
1308 fs_info->btree_inode->i_mapping, GFP_NOFS);
1309 extent_io_tree_init(&fs_info->pending_del,
1310 fs_info->btree_inode->i_mapping, GFP_NOFS);
1311 extent_io_tree_init(&fs_info->extent_ins,
1312 fs_info->btree_inode->i_mapping, GFP_NOFS);
1313 fs_info->do_barriers = 1;
1315 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1316 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1317 sizeof(struct btrfs_key));
1318 insert_inode_hash(fs_info->btree_inode);
1319 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1321 mutex_init(&fs_info->trans_mutex);
1322 mutex_init(&fs_info->drop_mutex);
1323 mutex_init(&fs_info->alloc_mutex);
1324 mutex_init(&fs_info->chunk_mutex);
1325 mutex_init(&fs_info->transaction_kthread_mutex);
1326 mutex_init(&fs_info->cleaner_mutex);
1327 mutex_init(&fs_info->volume_mutex);
1328 init_waitqueue_head(&fs_info->transaction_throttle);
1329 init_waitqueue_head(&fs_info->transaction_wait);
1332 ret = add_hasher(fs_info, "crc32c");
1334 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1339 __setup_root(4096, 4096, 4096, 4096, tree_root,
1340 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1343 bh = __bread(fs_devices->latest_bdev,
1344 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1348 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1351 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1353 disk_super = &fs_info->super_copy;
1354 if (!btrfs_super_root(disk_super))
1355 goto fail_sb_buffer;
1357 err = btrfs_parse_options(tree_root, options);
1359 goto fail_sb_buffer;
1362 * we need to start all the end_io workers up front because the
1363 * queue work function gets called at interrupt time, and so it
1364 * cannot dynamically grow.
1366 btrfs_init_workers(&fs_info->workers, fs_info->thread_pool_size);
1367 btrfs_init_workers(&fs_info->submit_workers,
1368 min_t(u64, fs_devices->num_devices,
1369 fs_info->thread_pool_size));
1371 /* a higher idle thresh on the submit workers makes it much more
1372 * likely that bios will be send down in a sane order to the
1375 fs_info->submit_workers.idle_thresh = 64;
1377 btrfs_init_workers(&fs_info->fixup_workers, 1);
1378 btrfs_init_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1379 btrfs_init_workers(&fs_info->endio_write_workers,
1380 fs_info->thread_pool_size);
1383 * endios are largely parallel and should have a very
1386 fs_info->endio_workers.idle_thresh = 4;
1387 fs_info->endio_write_workers.idle_thresh = 4;
1389 btrfs_start_workers(&fs_info->workers, 1);
1390 btrfs_start_workers(&fs_info->submit_workers, 1);
1391 btrfs_start_workers(&fs_info->fixup_workers, 1);
1392 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1393 btrfs_start_workers(&fs_info->endio_write_workers,
1394 fs_info->thread_pool_size);
1397 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1398 printk("Btrfs: wanted %llu devices, but found %llu\n",
1399 (unsigned long long)btrfs_super_num_devices(disk_super),
1400 (unsigned long long)fs_devices->open_devices);
1401 if (btrfs_test_opt(tree_root, DEGRADED))
1402 printk("continuing in degraded mode\n");
1404 goto fail_sb_buffer;
1408 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1410 nodesize = btrfs_super_nodesize(disk_super);
1411 leafsize = btrfs_super_leafsize(disk_super);
1412 sectorsize = btrfs_super_sectorsize(disk_super);
1413 stripesize = btrfs_super_stripesize(disk_super);
1414 tree_root->nodesize = nodesize;
1415 tree_root->leafsize = leafsize;
1416 tree_root->sectorsize = sectorsize;
1417 tree_root->stripesize = stripesize;
1419 sb->s_blocksize = sectorsize;
1420 sb->s_blocksize_bits = blksize_bits(sectorsize);
1422 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1423 sizeof(disk_super->magic))) {
1424 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1425 goto fail_sb_buffer;
1428 mutex_lock(&fs_info->chunk_mutex);
1429 ret = btrfs_read_sys_array(tree_root);
1430 mutex_unlock(&fs_info->chunk_mutex);
1432 printk("btrfs: failed to read the system array on %s\n",
1434 goto fail_sys_array;
1437 blocksize = btrfs_level_size(tree_root,
1438 btrfs_super_chunk_root_level(disk_super));
1440 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1441 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1443 chunk_root->node = read_tree_block(chunk_root,
1444 btrfs_super_chunk_root(disk_super),
1446 BUG_ON(!chunk_root->node);
1448 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1449 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1452 mutex_lock(&fs_info->chunk_mutex);
1453 ret = btrfs_read_chunk_tree(chunk_root);
1454 mutex_unlock(&fs_info->chunk_mutex);
1457 btrfs_close_extra_devices(fs_devices);
1459 blocksize = btrfs_level_size(tree_root,
1460 btrfs_super_root_level(disk_super));
1463 tree_root->node = read_tree_block(tree_root,
1464 btrfs_super_root(disk_super),
1466 if (!tree_root->node)
1467 goto fail_sb_buffer;
1470 ret = find_and_setup_root(tree_root, fs_info,
1471 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1473 goto fail_tree_root;
1474 extent_root->track_dirty = 1;
1476 ret = find_and_setup_root(tree_root, fs_info,
1477 BTRFS_DEV_TREE_OBJECTID, dev_root);
1478 dev_root->track_dirty = 1;
1481 goto fail_extent_root;
1483 btrfs_read_block_groups(extent_root);
1485 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1486 fs_info->data_alloc_profile = (u64)-1;
1487 fs_info->metadata_alloc_profile = (u64)-1;
1488 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1489 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1491 if (!fs_info->cleaner_kthread)
1492 goto fail_extent_root;
1494 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1496 "btrfs-transaction");
1497 if (!fs_info->transaction_kthread)
1504 kthread_stop(fs_info->cleaner_kthread);
1506 free_extent_buffer(extent_root->node);
1508 free_extent_buffer(tree_root->node);
1511 btrfs_stop_workers(&fs_info->fixup_workers);
1512 btrfs_stop_workers(&fs_info->workers);
1513 btrfs_stop_workers(&fs_info->endio_workers);
1514 btrfs_stop_workers(&fs_info->endio_write_workers);
1515 btrfs_stop_workers(&fs_info->submit_workers);
1517 iput(fs_info->btree_inode);
1519 btrfs_close_devices(fs_info->fs_devices);
1520 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1524 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1525 bdi_destroy(&fs_info->bdi);
1528 return ERR_PTR(err);
1531 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1533 char b[BDEVNAME_SIZE];
1536 set_buffer_uptodate(bh);
1538 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1539 printk(KERN_WARNING "lost page write due to "
1540 "I/O error on %s\n",
1541 bdevname(bh->b_bdev, b));
1543 /* note, we dont' set_buffer_write_io_error because we have
1544 * our own ways of dealing with the IO errors
1546 clear_buffer_uptodate(bh);
1552 int write_all_supers(struct btrfs_root *root)
1554 struct list_head *cur;
1555 struct list_head *head = &root->fs_info->fs_devices->devices;
1556 struct btrfs_device *dev;
1557 struct btrfs_super_block *sb;
1558 struct btrfs_dev_item *dev_item;
1559 struct buffer_head *bh;
1563 int total_errors = 0;
1567 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1568 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1570 sb = &root->fs_info->super_for_commit;
1571 dev_item = &sb->dev_item;
1572 list_for_each(cur, head) {
1573 dev = list_entry(cur, struct btrfs_device, dev_list);
1578 if (!dev->in_fs_metadata)
1581 btrfs_set_stack_device_type(dev_item, dev->type);
1582 btrfs_set_stack_device_id(dev_item, dev->devid);
1583 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1584 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1585 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1586 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1587 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1588 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1589 flags = btrfs_super_flags(sb);
1590 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1594 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1595 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1596 btrfs_csum_final(crc, sb->csum);
1598 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1599 BTRFS_SUPER_INFO_SIZE);
1601 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1602 dev->pending_io = bh;
1605 set_buffer_uptodate(bh);
1607 bh->b_end_io = btrfs_end_buffer_write_sync;
1609 if (do_barriers && dev->barriers) {
1610 ret = submit_bh(WRITE_BARRIER, bh);
1611 if (ret == -EOPNOTSUPP) {
1612 printk("btrfs: disabling barriers on dev %s\n",
1614 set_buffer_uptodate(bh);
1618 ret = submit_bh(WRITE, bh);
1621 ret = submit_bh(WRITE, bh);
1626 if (total_errors > max_errors) {
1627 printk("btrfs: %d errors while writing supers\n", total_errors);
1632 list_for_each(cur, head) {
1633 dev = list_entry(cur, struct btrfs_device, dev_list);
1636 if (!dev->in_fs_metadata)
1639 BUG_ON(!dev->pending_io);
1640 bh = dev->pending_io;
1642 if (!buffer_uptodate(dev->pending_io)) {
1643 if (do_barriers && dev->barriers) {
1644 printk("btrfs: disabling barriers on dev %s\n",
1646 set_buffer_uptodate(bh);
1650 ret = submit_bh(WRITE, bh);
1653 if (!buffer_uptodate(bh))
1660 dev->pending_io = NULL;
1663 if (total_errors > max_errors) {
1664 printk("btrfs: %d errors while writing supers\n", total_errors);
1670 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1675 ret = write_all_supers(root);
1679 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1681 radix_tree_delete(&fs_info->fs_roots_radix,
1682 (unsigned long)root->root_key.objectid);
1684 btrfs_sysfs_del_root(root);
1688 free_extent_buffer(root->node);
1689 if (root->commit_root)
1690 free_extent_buffer(root->commit_root);
1697 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1700 struct btrfs_root *gang[8];
1704 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1709 for (i = 0; i < ret; i++)
1710 btrfs_free_fs_root(fs_info, gang[i]);
1715 int close_ctree(struct btrfs_root *root)
1718 struct btrfs_trans_handle *trans;
1719 struct btrfs_fs_info *fs_info = root->fs_info;
1721 fs_info->closing = 1;
1724 kthread_stop(root->fs_info->transaction_kthread);
1725 kthread_stop(root->fs_info->cleaner_kthread);
1727 btrfs_clean_old_snapshots(root);
1728 trans = btrfs_start_transaction(root, 1);
1729 ret = btrfs_commit_transaction(trans, root);
1730 /* run commit again to drop the original snapshot */
1731 trans = btrfs_start_transaction(root, 1);
1732 btrfs_commit_transaction(trans, root);
1733 ret = btrfs_write_and_wait_transaction(NULL, root);
1736 write_ctree_super(NULL, root);
1738 if (fs_info->delalloc_bytes) {
1739 printk("btrfs: at unmount delalloc count %Lu\n",
1740 fs_info->delalloc_bytes);
1742 if (fs_info->total_ref_cache_size) {
1743 printk("btrfs: at umount reference cache size %Lu\n",
1744 fs_info->total_ref_cache_size);
1747 if (fs_info->extent_root->node)
1748 free_extent_buffer(fs_info->extent_root->node);
1750 if (fs_info->tree_root->node)
1751 free_extent_buffer(fs_info->tree_root->node);
1753 if (root->fs_info->chunk_root->node);
1754 free_extent_buffer(root->fs_info->chunk_root->node);
1756 if (root->fs_info->dev_root->node);
1757 free_extent_buffer(root->fs_info->dev_root->node);
1759 btrfs_free_block_groups(root->fs_info);
1760 fs_info->closing = 2;
1761 del_fs_roots(fs_info);
1763 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1765 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1767 btrfs_stop_workers(&fs_info->fixup_workers);
1768 btrfs_stop_workers(&fs_info->workers);
1769 btrfs_stop_workers(&fs_info->endio_workers);
1770 btrfs_stop_workers(&fs_info->endio_write_workers);
1771 btrfs_stop_workers(&fs_info->submit_workers);
1773 iput(fs_info->btree_inode);
1775 while(!list_empty(&fs_info->hashers)) {
1776 struct btrfs_hasher *hasher;
1777 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1779 list_del(&hasher->hashers);
1780 crypto_free_hash(&fs_info->hash_tfm);
1784 btrfs_close_devices(fs_info->fs_devices);
1785 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1787 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1788 bdi_destroy(&fs_info->bdi);
1791 kfree(fs_info->extent_root);
1792 kfree(fs_info->tree_root);
1793 kfree(fs_info->chunk_root);
1794 kfree(fs_info->dev_root);
1798 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1801 struct inode *btree_inode = buf->first_page->mapping->host;
1803 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1807 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1812 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1814 struct inode *btree_inode = buf->first_page->mapping->host;
1815 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1819 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1821 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1822 u64 transid = btrfs_header_generation(buf);
1823 struct inode *btree_inode = root->fs_info->btree_inode;
1825 WARN_ON(!btrfs_tree_locked(buf));
1826 if (transid != root->fs_info->generation) {
1827 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1828 (unsigned long long)buf->start,
1829 transid, root->fs_info->generation);
1832 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1835 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1838 * looks as though older kernels can get into trouble with
1839 * this code, they end up stuck in balance_dirty_pages forever
1841 struct extent_io_tree *tree;
1844 unsigned long thresh = 2 * 1024 * 1024;
1845 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1847 if (current_is_pdflush())
1850 num_dirty = count_range_bits(tree, &start, (u64)-1,
1851 thresh, EXTENT_DIRTY);
1852 if (num_dirty > thresh) {
1853 balance_dirty_pages_ratelimited_nr(
1854 root->fs_info->btree_inode->i_mapping, 1);
1859 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1861 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1863 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1865 buf->flags |= EXTENT_UPTODATE;
1870 static struct extent_io_ops btree_extent_io_ops = {
1871 .writepage_io_hook = btree_writepage_io_hook,
1872 .readpage_end_io_hook = btree_readpage_end_io_hook,
1873 .submit_bio_hook = btree_submit_bio_hook,
1874 /* note we're sharing with inode.c for the merge bio hook */
1875 .merge_bio_hook = btrfs_merge_bio_hook,