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 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
434 unsigned long limit = min_t(unsigned long,
435 info->workers.max_workers,
436 info->fs_devices->open_devices);
440 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
442 return atomic_read(&info->nr_async_bios) >
443 btrfs_async_submit_limit(info);
446 static void run_one_async_submit(struct btrfs_work *work)
448 struct btrfs_fs_info *fs_info;
449 struct async_submit_bio *async;
452 async = container_of(work, struct async_submit_bio, work);
453 fs_info = BTRFS_I(async->inode)->root->fs_info;
455 limit = btrfs_async_submit_limit(fs_info);
456 limit = limit * 2 / 3;
458 atomic_dec(&fs_info->nr_async_submits);
460 if (atomic_read(&fs_info->nr_async_submits) < limit &&
461 waitqueue_active(&fs_info->async_submit_wait))
462 wake_up(&fs_info->async_submit_wait);
464 async->submit_bio_hook(async->inode, async->rw, async->bio,
469 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
470 int rw, struct bio *bio, int mirror_num,
471 extent_submit_bio_hook_t *submit_bio_hook)
473 struct async_submit_bio *async;
474 int limit = btrfs_async_submit_limit(fs_info);
476 async = kmalloc(sizeof(*async), GFP_NOFS);
480 async->inode = inode;
483 async->mirror_num = mirror_num;
484 async->submit_bio_hook = submit_bio_hook;
485 async->work.func = run_one_async_submit;
486 async->work.flags = 0;
487 atomic_inc(&fs_info->nr_async_submits);
488 btrfs_queue_worker(&fs_info->workers, &async->work);
490 wait_event_timeout(fs_info->async_submit_wait,
491 (atomic_read(&fs_info->nr_async_submits) < limit),
496 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
499 struct btrfs_root *root = BTRFS_I(inode)->root;
503 offset = bio->bi_sector << 9;
506 * when we're called for a write, we're already in the async
507 * submission context. Just jump into btrfs_map_bio
509 if (rw & (1 << BIO_RW)) {
510 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
515 * called for a read, do the setup so that checksum validation
516 * can happen in the async kernel threads
518 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
521 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
524 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
528 * kthread helpers are used to submit writes so that checksumming
529 * can happen in parallel across all CPUs
531 if (!(rw & (1 << BIO_RW))) {
532 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
534 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
535 inode, rw, bio, mirror_num,
536 __btree_submit_bio_hook);
539 static int btree_writepage(struct page *page, struct writeback_control *wbc)
541 struct extent_io_tree *tree;
542 tree = &BTRFS_I(page->mapping->host)->io_tree;
544 if (current->flags & PF_MEMALLOC) {
545 redirty_page_for_writepage(wbc, page);
549 return extent_write_full_page(tree, page, btree_get_extent, wbc);
552 static int btree_writepages(struct address_space *mapping,
553 struct writeback_control *wbc)
555 struct extent_io_tree *tree;
556 tree = &BTRFS_I(mapping->host)->io_tree;
557 if (wbc->sync_mode == WB_SYNC_NONE) {
560 unsigned long thresh = 8 * 1024 * 1024;
562 if (wbc->for_kupdate)
565 num_dirty = count_range_bits(tree, &start, (u64)-1,
566 thresh, EXTENT_DIRTY);
567 if (num_dirty < thresh) {
571 return extent_writepages(tree, mapping, btree_get_extent, wbc);
574 int btree_readpage(struct file *file, struct page *page)
576 struct extent_io_tree *tree;
577 tree = &BTRFS_I(page->mapping->host)->io_tree;
578 return extent_read_full_page(tree, page, btree_get_extent);
581 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
583 struct extent_io_tree *tree;
584 struct extent_map_tree *map;
587 tree = &BTRFS_I(page->mapping->host)->io_tree;
588 map = &BTRFS_I(page->mapping->host)->extent_tree;
590 ret = try_release_extent_state(map, tree, page, gfp_flags);
595 ret = try_release_extent_buffer(tree, page);
597 ClearPagePrivate(page);
598 set_page_private(page, 0);
599 page_cache_release(page);
605 static void btree_invalidatepage(struct page *page, unsigned long offset)
607 struct extent_io_tree *tree;
608 tree = &BTRFS_I(page->mapping->host)->io_tree;
609 extent_invalidatepage(tree, page, offset);
610 btree_releasepage(page, GFP_NOFS);
611 if (PagePrivate(page)) {
612 printk("warning page private not zero on page %Lu\n",
614 ClearPagePrivate(page);
615 set_page_private(page, 0);
616 page_cache_release(page);
621 static int btree_writepage(struct page *page, struct writeback_control *wbc)
623 struct buffer_head *bh;
624 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
625 struct buffer_head *head;
626 if (!page_has_buffers(page)) {
627 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
628 (1 << BH_Dirty)|(1 << BH_Uptodate));
630 head = page_buffers(page);
633 if (buffer_dirty(bh))
634 csum_tree_block(root, bh, 0);
635 bh = bh->b_this_page;
636 } while (bh != head);
637 return block_write_full_page(page, btree_get_block, wbc);
641 static struct address_space_operations btree_aops = {
642 .readpage = btree_readpage,
643 .writepage = btree_writepage,
644 .writepages = btree_writepages,
645 .releasepage = btree_releasepage,
646 .invalidatepage = btree_invalidatepage,
647 .sync_page = block_sync_page,
650 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
653 struct extent_buffer *buf = NULL;
654 struct inode *btree_inode = root->fs_info->btree_inode;
657 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
660 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
661 buf, 0, 0, btree_get_extent, 0);
662 free_extent_buffer(buf);
666 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
667 u64 bytenr, u32 blocksize)
669 struct inode *btree_inode = root->fs_info->btree_inode;
670 struct extent_buffer *eb;
671 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
672 bytenr, blocksize, GFP_NOFS);
676 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
677 u64 bytenr, u32 blocksize)
679 struct inode *btree_inode = root->fs_info->btree_inode;
680 struct extent_buffer *eb;
682 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
683 bytenr, blocksize, NULL, GFP_NOFS);
688 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
689 u32 blocksize, u64 parent_transid)
691 struct extent_buffer *buf = NULL;
692 struct inode *btree_inode = root->fs_info->btree_inode;
693 struct extent_io_tree *io_tree;
696 io_tree = &BTRFS_I(btree_inode)->io_tree;
698 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
702 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
705 buf->flags |= EXTENT_UPTODATE;
711 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
712 struct extent_buffer *buf)
714 struct inode *btree_inode = root->fs_info->btree_inode;
715 if (btrfs_header_generation(buf) ==
716 root->fs_info->running_transaction->transid) {
717 WARN_ON(!btrfs_tree_locked(buf));
718 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
724 int wait_on_tree_block_writeback(struct btrfs_root *root,
725 struct extent_buffer *buf)
727 struct inode *btree_inode = root->fs_info->btree_inode;
728 wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
733 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
734 u32 stripesize, struct btrfs_root *root,
735 struct btrfs_fs_info *fs_info,
740 root->commit_root = NULL;
741 root->ref_tree = NULL;
742 root->sectorsize = sectorsize;
743 root->nodesize = nodesize;
744 root->leafsize = leafsize;
745 root->stripesize = stripesize;
747 root->track_dirty = 0;
749 root->fs_info = fs_info;
750 root->objectid = objectid;
751 root->last_trans = 0;
752 root->highest_inode = 0;
753 root->last_inode_alloc = 0;
757 INIT_LIST_HEAD(&root->dirty_list);
758 INIT_LIST_HEAD(&root->orphan_list);
759 INIT_LIST_HEAD(&root->dead_list);
760 spin_lock_init(&root->node_lock);
761 spin_lock_init(&root->list_lock);
762 mutex_init(&root->objectid_mutex);
764 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
765 root->ref_tree = &root->ref_tree_struct;
767 memset(&root->root_key, 0, sizeof(root->root_key));
768 memset(&root->root_item, 0, sizeof(root->root_item));
769 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
770 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
771 root->defrag_trans_start = fs_info->generation;
772 init_completion(&root->kobj_unregister);
773 root->defrag_running = 0;
774 root->defrag_level = 0;
775 root->root_key.objectid = objectid;
779 static int find_and_setup_root(struct btrfs_root *tree_root,
780 struct btrfs_fs_info *fs_info,
782 struct btrfs_root *root)
787 __setup_root(tree_root->nodesize, tree_root->leafsize,
788 tree_root->sectorsize, tree_root->stripesize,
789 root, fs_info, objectid);
790 ret = btrfs_find_last_root(tree_root, objectid,
791 &root->root_item, &root->root_key);
794 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
795 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
801 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
802 struct btrfs_key *location)
804 struct btrfs_root *root;
805 struct btrfs_root *tree_root = fs_info->tree_root;
806 struct btrfs_path *path;
807 struct extent_buffer *l;
812 root = kzalloc(sizeof(*root), GFP_NOFS);
814 return ERR_PTR(-ENOMEM);
815 if (location->offset == (u64)-1) {
816 ret = find_and_setup_root(tree_root, fs_info,
817 location->objectid, root);
825 __setup_root(tree_root->nodesize, tree_root->leafsize,
826 tree_root->sectorsize, tree_root->stripesize,
827 root, fs_info, location->objectid);
829 path = btrfs_alloc_path();
831 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
838 read_extent_buffer(l, &root->root_item,
839 btrfs_item_ptr_offset(l, path->slots[0]),
840 sizeof(root->root_item));
841 memcpy(&root->root_key, location, sizeof(*location));
844 btrfs_release_path(root, path);
845 btrfs_free_path(path);
850 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
851 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
856 ret = btrfs_find_highest_inode(root, &highest_inode);
858 root->highest_inode = highest_inode;
859 root->last_inode_alloc = highest_inode;
864 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
867 struct btrfs_root *root;
869 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
870 return fs_info->tree_root;
871 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
872 return fs_info->extent_root;
874 root = radix_tree_lookup(&fs_info->fs_roots_radix,
875 (unsigned long)root_objectid);
879 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
880 struct btrfs_key *location)
882 struct btrfs_root *root;
885 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
886 return fs_info->tree_root;
887 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
888 return fs_info->extent_root;
889 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
890 return fs_info->chunk_root;
891 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
892 return fs_info->dev_root;
894 root = radix_tree_lookup(&fs_info->fs_roots_radix,
895 (unsigned long)location->objectid);
899 root = btrfs_read_fs_root_no_radix(fs_info, location);
902 ret = radix_tree_insert(&fs_info->fs_roots_radix,
903 (unsigned long)root->root_key.objectid,
906 free_extent_buffer(root->node);
910 ret = btrfs_find_dead_roots(fs_info->tree_root,
911 root->root_key.objectid, root);
917 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
918 struct btrfs_key *location,
919 const char *name, int namelen)
921 struct btrfs_root *root;
924 root = btrfs_read_fs_root_no_name(fs_info, location);
931 ret = btrfs_set_root_name(root, name, namelen);
933 free_extent_buffer(root->node);
938 ret = btrfs_sysfs_add_root(root);
940 free_extent_buffer(root->node);
949 static int add_hasher(struct btrfs_fs_info *info, char *type) {
950 struct btrfs_hasher *hasher;
952 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
955 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
956 if (!hasher->hash_tfm) {
960 spin_lock(&info->hash_lock);
961 list_add(&hasher->list, &info->hashers);
962 spin_unlock(&info->hash_lock);
967 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
969 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
971 struct list_head *cur;
972 struct btrfs_device *device;
973 struct backing_dev_info *bdi;
975 if ((bdi_bits & (1 << BDI_write_congested)) &&
976 btrfs_congested_async(info, 0))
979 list_for_each(cur, &info->fs_devices->devices) {
980 device = list_entry(cur, struct btrfs_device, dev_list);
983 bdi = blk_get_backing_dev_info(device->bdev);
984 if (bdi && bdi_congested(bdi, bdi_bits)) {
993 * this unplugs every device on the box, and it is only used when page
996 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
998 struct list_head *cur;
999 struct btrfs_device *device;
1000 struct btrfs_fs_info *info;
1002 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1003 list_for_each(cur, &info->fs_devices->devices) {
1004 device = list_entry(cur, struct btrfs_device, dev_list);
1005 bdi = blk_get_backing_dev_info(device->bdev);
1006 if (bdi->unplug_io_fn) {
1007 bdi->unplug_io_fn(bdi, page);
1012 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1014 struct inode *inode;
1015 struct extent_map_tree *em_tree;
1016 struct extent_map *em;
1017 struct address_space *mapping;
1020 /* the generic O_DIRECT read code does this */
1022 __unplug_io_fn(bdi, page);
1027 * page->mapping may change at any time. Get a consistent copy
1028 * and use that for everything below
1031 mapping = page->mapping;
1035 inode = mapping->host;
1036 offset = page_offset(page);
1038 em_tree = &BTRFS_I(inode)->extent_tree;
1039 spin_lock(&em_tree->lock);
1040 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1041 spin_unlock(&em_tree->lock);
1043 __unplug_io_fn(bdi, page);
1047 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1048 free_extent_map(em);
1049 __unplug_io_fn(bdi, page);
1052 offset = offset - em->start;
1053 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1054 em->block_start + offset, page);
1055 free_extent_map(em);
1058 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1060 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1063 bdi->ra_pages = default_backing_dev_info.ra_pages;
1065 bdi->capabilities = default_backing_dev_info.capabilities;
1066 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1067 bdi->unplug_io_data = info;
1068 bdi->congested_fn = btrfs_congested_fn;
1069 bdi->congested_data = info;
1073 static int bio_ready_for_csum(struct bio *bio)
1079 struct extent_io_tree *io_tree = NULL;
1080 struct btrfs_fs_info *info = NULL;
1081 struct bio_vec *bvec;
1085 bio_for_each_segment(bvec, bio, i) {
1086 page = bvec->bv_page;
1087 if (page->private == EXTENT_PAGE_PRIVATE) {
1088 length += bvec->bv_len;
1091 if (!page->private) {
1092 length += bvec->bv_len;
1095 length = bvec->bv_len;
1096 buf_len = page->private >> 2;
1097 start = page_offset(page) + bvec->bv_offset;
1098 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1099 info = BTRFS_I(page->mapping->host)->root->fs_info;
1101 /* are we fully contained in this bio? */
1102 if (buf_len <= length)
1105 ret = extent_range_uptodate(io_tree, start + length,
1106 start + buf_len - 1);
1113 * called by the kthread helper functions to finally call the bio end_io
1114 * functions. This is where read checksum verification actually happens
1116 static void end_workqueue_fn(struct btrfs_work *work)
1119 struct end_io_wq *end_io_wq;
1120 struct btrfs_fs_info *fs_info;
1123 end_io_wq = container_of(work, struct end_io_wq, work);
1124 bio = end_io_wq->bio;
1125 fs_info = end_io_wq->info;
1127 /* metadata bios are special because the whole tree block must
1128 * be checksummed at once. This makes sure the entire block is in
1129 * ram and up to date before trying to verify things. For
1130 * blocksize <= pagesize, it is basically a noop
1132 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1133 btrfs_queue_worker(&fs_info->endio_workers,
1137 error = end_io_wq->error;
1138 bio->bi_private = end_io_wq->private;
1139 bio->bi_end_io = end_io_wq->end_io;
1141 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1142 bio_endio(bio, bio->bi_size, error);
1144 bio_endio(bio, error);
1148 static int cleaner_kthread(void *arg)
1150 struct btrfs_root *root = arg;
1154 if (root->fs_info->closing)
1157 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1158 mutex_lock(&root->fs_info->cleaner_mutex);
1159 btrfs_clean_old_snapshots(root);
1160 mutex_unlock(&root->fs_info->cleaner_mutex);
1162 if (freezing(current)) {
1166 if (root->fs_info->closing)
1168 set_current_state(TASK_INTERRUPTIBLE);
1170 __set_current_state(TASK_RUNNING);
1172 } while (!kthread_should_stop());
1176 static int transaction_kthread(void *arg)
1178 struct btrfs_root *root = arg;
1179 struct btrfs_trans_handle *trans;
1180 struct btrfs_transaction *cur;
1182 unsigned long delay;
1187 if (root->fs_info->closing)
1191 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1192 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1194 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1195 printk("btrfs: total reference cache size %Lu\n",
1196 root->fs_info->total_ref_cache_size);
1199 mutex_lock(&root->fs_info->trans_mutex);
1200 cur = root->fs_info->running_transaction;
1202 mutex_unlock(&root->fs_info->trans_mutex);
1206 now = get_seconds();
1207 if (now < cur->start_time || now - cur->start_time < 30) {
1208 mutex_unlock(&root->fs_info->trans_mutex);
1212 mutex_unlock(&root->fs_info->trans_mutex);
1213 trans = btrfs_start_transaction(root, 1);
1214 ret = btrfs_commit_transaction(trans, root);
1216 wake_up_process(root->fs_info->cleaner_kthread);
1217 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1219 if (freezing(current)) {
1222 if (root->fs_info->closing)
1224 set_current_state(TASK_INTERRUPTIBLE);
1225 schedule_timeout(delay);
1226 __set_current_state(TASK_RUNNING);
1228 } while (!kthread_should_stop());
1232 struct btrfs_root *open_ctree(struct super_block *sb,
1233 struct btrfs_fs_devices *fs_devices,
1241 struct buffer_head *bh;
1242 struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1244 struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1246 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1248 struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1250 struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1255 struct btrfs_super_block *disk_super;
1257 if (!extent_root || !tree_root || !fs_info) {
1261 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1262 INIT_LIST_HEAD(&fs_info->trans_list);
1263 INIT_LIST_HEAD(&fs_info->dead_roots);
1264 INIT_LIST_HEAD(&fs_info->hashers);
1265 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1266 spin_lock_init(&fs_info->hash_lock);
1267 spin_lock_init(&fs_info->delalloc_lock);
1268 spin_lock_init(&fs_info->new_trans_lock);
1269 spin_lock_init(&fs_info->ref_cache_lock);
1271 init_completion(&fs_info->kobj_unregister);
1272 fs_info->tree_root = tree_root;
1273 fs_info->extent_root = extent_root;
1274 fs_info->chunk_root = chunk_root;
1275 fs_info->dev_root = dev_root;
1276 fs_info->fs_devices = fs_devices;
1277 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1278 INIT_LIST_HEAD(&fs_info->space_info);
1279 btrfs_mapping_init(&fs_info->mapping_tree);
1280 atomic_set(&fs_info->nr_async_submits, 0);
1281 atomic_set(&fs_info->nr_async_bios, 0);
1282 atomic_set(&fs_info->throttles, 0);
1283 atomic_set(&fs_info->throttle_gen, 0);
1285 fs_info->max_extent = (u64)-1;
1286 fs_info->max_inline = 8192 * 1024;
1287 setup_bdi(fs_info, &fs_info->bdi);
1288 fs_info->btree_inode = new_inode(sb);
1289 fs_info->btree_inode->i_ino = 1;
1290 fs_info->btree_inode->i_nlink = 1;
1291 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1293 INIT_LIST_HEAD(&fs_info->ordered_extents);
1294 spin_lock_init(&fs_info->ordered_extent_lock);
1296 sb->s_blocksize = 4096;
1297 sb->s_blocksize_bits = blksize_bits(4096);
1300 * we set the i_size on the btree inode to the max possible int.
1301 * the real end of the address space is determined by all of
1302 * the devices in the system
1304 fs_info->btree_inode->i_size = OFFSET_MAX;
1305 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1306 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1308 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1309 fs_info->btree_inode->i_mapping,
1311 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1314 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1316 extent_io_tree_init(&fs_info->free_space_cache,
1317 fs_info->btree_inode->i_mapping, GFP_NOFS);
1318 extent_io_tree_init(&fs_info->block_group_cache,
1319 fs_info->btree_inode->i_mapping, GFP_NOFS);
1320 extent_io_tree_init(&fs_info->pinned_extents,
1321 fs_info->btree_inode->i_mapping, GFP_NOFS);
1322 extent_io_tree_init(&fs_info->pending_del,
1323 fs_info->btree_inode->i_mapping, GFP_NOFS);
1324 extent_io_tree_init(&fs_info->extent_ins,
1325 fs_info->btree_inode->i_mapping, GFP_NOFS);
1326 fs_info->do_barriers = 1;
1328 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1329 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1330 sizeof(struct btrfs_key));
1331 insert_inode_hash(fs_info->btree_inode);
1332 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1334 mutex_init(&fs_info->trans_mutex);
1335 mutex_init(&fs_info->drop_mutex);
1336 mutex_init(&fs_info->alloc_mutex);
1337 mutex_init(&fs_info->chunk_mutex);
1338 mutex_init(&fs_info->transaction_kthread_mutex);
1339 mutex_init(&fs_info->cleaner_mutex);
1340 mutex_init(&fs_info->volume_mutex);
1341 init_waitqueue_head(&fs_info->transaction_throttle);
1342 init_waitqueue_head(&fs_info->transaction_wait);
1343 init_waitqueue_head(&fs_info->async_submit_wait);
1346 ret = add_hasher(fs_info, "crc32c");
1348 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1353 __setup_root(4096, 4096, 4096, 4096, tree_root,
1354 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1357 bh = __bread(fs_devices->latest_bdev,
1358 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1362 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1365 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1367 disk_super = &fs_info->super_copy;
1368 if (!btrfs_super_root(disk_super))
1369 goto fail_sb_buffer;
1371 err = btrfs_parse_options(tree_root, options);
1373 goto fail_sb_buffer;
1376 * we need to start all the end_io workers up front because the
1377 * queue work function gets called at interrupt time, and so it
1378 * cannot dynamically grow.
1380 btrfs_init_workers(&fs_info->workers, "worker",
1381 fs_info->thread_pool_size);
1382 btrfs_init_workers(&fs_info->submit_workers, "submit",
1383 min_t(u64, fs_devices->num_devices,
1384 fs_info->thread_pool_size));
1386 /* a higher idle thresh on the submit workers makes it much more
1387 * likely that bios will be send down in a sane order to the
1390 fs_info->submit_workers.idle_thresh = 64;
1392 /* fs_info->workers is responsible for checksumming file data
1393 * blocks and metadata. Using a larger idle thresh allows each
1394 * worker thread to operate on things in roughly the order they
1395 * were sent by the writeback daemons, improving overall locality
1396 * of the IO going down the pipe.
1398 fs_info->workers.idle_thresh = 128;
1400 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1401 btrfs_init_workers(&fs_info->endio_workers, "endio",
1402 fs_info->thread_pool_size);
1403 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1404 fs_info->thread_pool_size);
1407 * endios are largely parallel and should have a very
1410 fs_info->endio_workers.idle_thresh = 4;
1411 fs_info->endio_write_workers.idle_thresh = 4;
1413 btrfs_start_workers(&fs_info->workers, 1);
1414 btrfs_start_workers(&fs_info->submit_workers, 1);
1415 btrfs_start_workers(&fs_info->fixup_workers, 1);
1416 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1417 btrfs_start_workers(&fs_info->endio_write_workers,
1418 fs_info->thread_pool_size);
1421 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1422 printk("Btrfs: wanted %llu devices, but found %llu\n",
1423 (unsigned long long)btrfs_super_num_devices(disk_super),
1424 (unsigned long long)fs_devices->open_devices);
1425 if (btrfs_test_opt(tree_root, DEGRADED))
1426 printk("continuing in degraded mode\n");
1428 goto fail_sb_buffer;
1432 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1434 nodesize = btrfs_super_nodesize(disk_super);
1435 leafsize = btrfs_super_leafsize(disk_super);
1436 sectorsize = btrfs_super_sectorsize(disk_super);
1437 stripesize = btrfs_super_stripesize(disk_super);
1438 tree_root->nodesize = nodesize;
1439 tree_root->leafsize = leafsize;
1440 tree_root->sectorsize = sectorsize;
1441 tree_root->stripesize = stripesize;
1443 sb->s_blocksize = sectorsize;
1444 sb->s_blocksize_bits = blksize_bits(sectorsize);
1446 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1447 sizeof(disk_super->magic))) {
1448 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1449 goto fail_sb_buffer;
1452 mutex_lock(&fs_info->chunk_mutex);
1453 ret = btrfs_read_sys_array(tree_root);
1454 mutex_unlock(&fs_info->chunk_mutex);
1456 printk("btrfs: failed to read the system array on %s\n",
1458 goto fail_sys_array;
1461 blocksize = btrfs_level_size(tree_root,
1462 btrfs_super_chunk_root_level(disk_super));
1464 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1465 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1467 chunk_root->node = read_tree_block(chunk_root,
1468 btrfs_super_chunk_root(disk_super),
1470 BUG_ON(!chunk_root->node);
1472 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1473 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1476 mutex_lock(&fs_info->chunk_mutex);
1477 ret = btrfs_read_chunk_tree(chunk_root);
1478 mutex_unlock(&fs_info->chunk_mutex);
1481 btrfs_close_extra_devices(fs_devices);
1483 blocksize = btrfs_level_size(tree_root,
1484 btrfs_super_root_level(disk_super));
1487 tree_root->node = read_tree_block(tree_root,
1488 btrfs_super_root(disk_super),
1490 if (!tree_root->node)
1491 goto fail_sb_buffer;
1494 ret = find_and_setup_root(tree_root, fs_info,
1495 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1497 goto fail_tree_root;
1498 extent_root->track_dirty = 1;
1500 ret = find_and_setup_root(tree_root, fs_info,
1501 BTRFS_DEV_TREE_OBJECTID, dev_root);
1502 dev_root->track_dirty = 1;
1505 goto fail_extent_root;
1507 btrfs_read_block_groups(extent_root);
1509 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1510 fs_info->data_alloc_profile = (u64)-1;
1511 fs_info->metadata_alloc_profile = (u64)-1;
1512 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1513 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1515 if (!fs_info->cleaner_kthread)
1516 goto fail_extent_root;
1518 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1520 "btrfs-transaction");
1521 if (!fs_info->transaction_kthread)
1528 kthread_stop(fs_info->cleaner_kthread);
1530 free_extent_buffer(extent_root->node);
1532 free_extent_buffer(tree_root->node);
1535 btrfs_stop_workers(&fs_info->fixup_workers);
1536 btrfs_stop_workers(&fs_info->workers);
1537 btrfs_stop_workers(&fs_info->endio_workers);
1538 btrfs_stop_workers(&fs_info->endio_write_workers);
1539 btrfs_stop_workers(&fs_info->submit_workers);
1541 iput(fs_info->btree_inode);
1543 btrfs_close_devices(fs_info->fs_devices);
1544 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1548 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1549 bdi_destroy(&fs_info->bdi);
1552 return ERR_PTR(err);
1555 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1557 char b[BDEVNAME_SIZE];
1560 set_buffer_uptodate(bh);
1562 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1563 printk(KERN_WARNING "lost page write due to "
1564 "I/O error on %s\n",
1565 bdevname(bh->b_bdev, b));
1567 /* note, we dont' set_buffer_write_io_error because we have
1568 * our own ways of dealing with the IO errors
1570 clear_buffer_uptodate(bh);
1576 int write_all_supers(struct btrfs_root *root)
1578 struct list_head *cur;
1579 struct list_head *head = &root->fs_info->fs_devices->devices;
1580 struct btrfs_device *dev;
1581 struct btrfs_super_block *sb;
1582 struct btrfs_dev_item *dev_item;
1583 struct buffer_head *bh;
1587 int total_errors = 0;
1591 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1592 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1594 sb = &root->fs_info->super_for_commit;
1595 dev_item = &sb->dev_item;
1596 list_for_each(cur, head) {
1597 dev = list_entry(cur, struct btrfs_device, dev_list);
1602 if (!dev->in_fs_metadata)
1605 btrfs_set_stack_device_type(dev_item, dev->type);
1606 btrfs_set_stack_device_id(dev_item, dev->devid);
1607 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1608 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1609 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1610 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1611 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1612 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1613 flags = btrfs_super_flags(sb);
1614 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1618 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1619 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1620 btrfs_csum_final(crc, sb->csum);
1622 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1623 BTRFS_SUPER_INFO_SIZE);
1625 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1626 dev->pending_io = bh;
1629 set_buffer_uptodate(bh);
1631 bh->b_end_io = btrfs_end_buffer_write_sync;
1633 if (do_barriers && dev->barriers) {
1634 ret = submit_bh(WRITE_BARRIER, bh);
1635 if (ret == -EOPNOTSUPP) {
1636 printk("btrfs: disabling barriers on dev %s\n",
1638 set_buffer_uptodate(bh);
1642 ret = submit_bh(WRITE, bh);
1645 ret = submit_bh(WRITE, bh);
1650 if (total_errors > max_errors) {
1651 printk("btrfs: %d errors while writing supers\n", total_errors);
1656 list_for_each(cur, head) {
1657 dev = list_entry(cur, struct btrfs_device, dev_list);
1660 if (!dev->in_fs_metadata)
1663 BUG_ON(!dev->pending_io);
1664 bh = dev->pending_io;
1666 if (!buffer_uptodate(dev->pending_io)) {
1667 if (do_barriers && dev->barriers) {
1668 printk("btrfs: disabling barriers on dev %s\n",
1670 set_buffer_uptodate(bh);
1674 ret = submit_bh(WRITE, bh);
1677 if (!buffer_uptodate(bh))
1684 dev->pending_io = NULL;
1687 if (total_errors > max_errors) {
1688 printk("btrfs: %d errors while writing supers\n", total_errors);
1694 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1699 ret = write_all_supers(root);
1703 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1705 radix_tree_delete(&fs_info->fs_roots_radix,
1706 (unsigned long)root->root_key.objectid);
1708 btrfs_sysfs_del_root(root);
1712 free_extent_buffer(root->node);
1713 if (root->commit_root)
1714 free_extent_buffer(root->commit_root);
1721 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1724 struct btrfs_root *gang[8];
1728 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1733 for (i = 0; i < ret; i++)
1734 btrfs_free_fs_root(fs_info, gang[i]);
1739 int close_ctree(struct btrfs_root *root)
1742 struct btrfs_trans_handle *trans;
1743 struct btrfs_fs_info *fs_info = root->fs_info;
1745 fs_info->closing = 1;
1748 kthread_stop(root->fs_info->transaction_kthread);
1749 kthread_stop(root->fs_info->cleaner_kthread);
1751 btrfs_clean_old_snapshots(root);
1752 trans = btrfs_start_transaction(root, 1);
1753 ret = btrfs_commit_transaction(trans, root);
1754 /* run commit again to drop the original snapshot */
1755 trans = btrfs_start_transaction(root, 1);
1756 btrfs_commit_transaction(trans, root);
1757 ret = btrfs_write_and_wait_transaction(NULL, root);
1760 write_ctree_super(NULL, root);
1762 if (fs_info->delalloc_bytes) {
1763 printk("btrfs: at unmount delalloc count %Lu\n",
1764 fs_info->delalloc_bytes);
1766 if (fs_info->total_ref_cache_size) {
1767 printk("btrfs: at umount reference cache size %Lu\n",
1768 fs_info->total_ref_cache_size);
1771 if (fs_info->extent_root->node)
1772 free_extent_buffer(fs_info->extent_root->node);
1774 if (fs_info->tree_root->node)
1775 free_extent_buffer(fs_info->tree_root->node);
1777 if (root->fs_info->chunk_root->node);
1778 free_extent_buffer(root->fs_info->chunk_root->node);
1780 if (root->fs_info->dev_root->node);
1781 free_extent_buffer(root->fs_info->dev_root->node);
1783 btrfs_free_block_groups(root->fs_info);
1784 fs_info->closing = 2;
1785 del_fs_roots(fs_info);
1787 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1789 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1791 btrfs_stop_workers(&fs_info->fixup_workers);
1792 btrfs_stop_workers(&fs_info->workers);
1793 btrfs_stop_workers(&fs_info->endio_workers);
1794 btrfs_stop_workers(&fs_info->endio_write_workers);
1795 btrfs_stop_workers(&fs_info->submit_workers);
1797 iput(fs_info->btree_inode);
1799 while(!list_empty(&fs_info->hashers)) {
1800 struct btrfs_hasher *hasher;
1801 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1803 list_del(&hasher->hashers);
1804 crypto_free_hash(&fs_info->hash_tfm);
1808 btrfs_close_devices(fs_info->fs_devices);
1809 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1811 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1812 bdi_destroy(&fs_info->bdi);
1815 kfree(fs_info->extent_root);
1816 kfree(fs_info->tree_root);
1817 kfree(fs_info->chunk_root);
1818 kfree(fs_info->dev_root);
1822 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1825 struct inode *btree_inode = buf->first_page->mapping->host;
1827 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1831 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1836 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1838 struct inode *btree_inode = buf->first_page->mapping->host;
1839 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1843 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1845 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1846 u64 transid = btrfs_header_generation(buf);
1847 struct inode *btree_inode = root->fs_info->btree_inode;
1849 WARN_ON(!btrfs_tree_locked(buf));
1850 if (transid != root->fs_info->generation) {
1851 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1852 (unsigned long long)buf->start,
1853 transid, root->fs_info->generation);
1856 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1859 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1862 * looks as though older kernels can get into trouble with
1863 * this code, they end up stuck in balance_dirty_pages forever
1865 struct extent_io_tree *tree;
1868 unsigned long thresh = 96 * 1024 * 1024;
1869 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1871 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
1874 num_dirty = count_range_bits(tree, &start, (u64)-1,
1875 thresh, EXTENT_DIRTY);
1876 if (num_dirty > thresh) {
1877 balance_dirty_pages_ratelimited_nr(
1878 root->fs_info->btree_inode->i_mapping, 1);
1883 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1885 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1887 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1889 buf->flags |= EXTENT_UPTODATE;
1894 static struct extent_io_ops btree_extent_io_ops = {
1895 .writepage_io_hook = btree_writepage_io_hook,
1896 .readpage_end_io_hook = btree_readpage_end_io_hook,
1897 .submit_bio_hook = btree_submit_bio_hook,
1898 /* note we're sharing with inode.c for the merge bio hook */
1899 .merge_bio_hook = btrfs_merge_bio_hook,