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 # include <linux/freezer.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
36 #include "print-tree.h"
37 #include "async-thread.h"
39 #include "ref-cache.h"
43 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
45 if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
46 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
47 (unsigned long long)extent_buffer_blocknr(buf),
48 (unsigned long long)btrfs_header_blocknr(buf));
55 static struct extent_io_ops btree_extent_io_ops;
56 static void end_workqueue_fn(struct btrfs_work *work);
62 struct btrfs_fs_info *info;
65 struct list_head list;
66 struct btrfs_work work;
69 struct async_submit_bio {
72 struct list_head list;
73 extent_submit_bio_hook_t *submit_bio_hook;
76 struct btrfs_work work;
79 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
80 size_t page_offset, u64 start, u64 len,
83 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
84 struct extent_map *em;
87 spin_lock(&em_tree->lock);
88 em = lookup_extent_mapping(em_tree, start, len);
91 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
92 spin_unlock(&em_tree->lock);
95 spin_unlock(&em_tree->lock);
97 em = alloc_extent_map(GFP_NOFS);
99 em = ERR_PTR(-ENOMEM);
105 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
107 spin_lock(&em_tree->lock);
108 ret = add_extent_mapping(em_tree, em);
109 if (ret == -EEXIST) {
110 u64 failed_start = em->start;
111 u64 failed_len = em->len;
113 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
114 em->start, em->len, em->block_start);
116 em = lookup_extent_mapping(em_tree, start, len);
118 printk("after failing, found %Lu %Lu %Lu\n",
119 em->start, em->len, em->block_start);
122 em = lookup_extent_mapping(em_tree, failed_start,
125 printk("double failure lookup gives us "
126 "%Lu %Lu -> %Lu\n", em->start,
127 em->len, em->block_start);
136 spin_unlock(&em_tree->lock);
144 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
146 return btrfs_crc32c(seed, data, len);
149 void btrfs_csum_final(u32 crc, char *result)
151 *(__le32 *)result = ~cpu_to_le32(crc);
154 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
157 char result[BTRFS_CRC32_SIZE];
159 unsigned long cur_len;
160 unsigned long offset = BTRFS_CSUM_SIZE;
161 char *map_token = NULL;
163 unsigned long map_start;
164 unsigned long map_len;
168 len = buf->len - offset;
170 err = map_private_extent_buffer(buf, offset, 32,
172 &map_start, &map_len, KM_USER0);
174 printk("failed to map extent buffer! %lu\n",
178 cur_len = min(len, map_len - (offset - map_start));
179 crc = btrfs_csum_data(root, kaddr + offset - map_start,
183 unmap_extent_buffer(buf, map_token, KM_USER0);
185 btrfs_csum_final(crc, result);
188 /* FIXME, this is not good */
189 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
192 memcpy(&found, result, BTRFS_CRC32_SIZE);
194 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
195 printk("btrfs: %s checksum verify failed on %llu "
196 "wanted %X found %X level %d\n",
197 root->fs_info->sb->s_id,
198 buf->start, val, found, btrfs_header_level(buf));
202 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
207 static int verify_parent_transid(struct extent_io_tree *io_tree,
208 struct extent_buffer *eb, u64 parent_transid)
212 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
215 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
216 if (extent_buffer_uptodate(io_tree, eb) &&
217 btrfs_header_generation(eb) == parent_transid) {
221 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
222 (unsigned long long)eb->start,
223 (unsigned long long)parent_transid,
224 (unsigned long long)btrfs_header_generation(eb));
226 clear_extent_buffer_uptodate(io_tree, eb);
228 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
234 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
235 struct extent_buffer *eb,
236 u64 start, u64 parent_transid)
238 struct extent_io_tree *io_tree;
243 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
245 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
246 btree_get_extent, mirror_num);
248 !verify_parent_transid(io_tree, eb, parent_transid))
250 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
251 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
257 if (mirror_num > num_copies)
263 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
265 struct extent_io_tree *tree;
266 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
270 struct extent_buffer *eb;
273 tree = &BTRFS_I(page->mapping->host)->io_tree;
275 if (page->private == EXTENT_PAGE_PRIVATE)
279 len = page->private >> 2;
283 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
284 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
285 btrfs_header_generation(eb));
287 found_start = btrfs_header_bytenr(eb);
288 if (found_start != start) {
289 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
290 start, found_start, len);
294 if (eb->first_page != page) {
295 printk("bad first page %lu %lu\n", eb->first_page->index,
300 if (!PageUptodate(page)) {
301 printk("csum not up to date page %lu\n", page->index);
305 found_level = btrfs_header_level(eb);
307 csum_tree_block(root, eb, 0);
309 free_extent_buffer(eb);
314 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
315 struct extent_state *state)
317 struct extent_io_tree *tree;
321 struct extent_buffer *eb;
322 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
325 tree = &BTRFS_I(page->mapping->host)->io_tree;
326 if (page->private == EXTENT_PAGE_PRIVATE)
330 len = page->private >> 2;
334 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
336 found_start = btrfs_header_bytenr(eb);
337 if (found_start != start) {
338 printk("bad tree block start %llu %llu\n",
339 (unsigned long long)found_start,
340 (unsigned long long)eb->start);
344 if (eb->first_page != page) {
345 printk("bad first page %lu %lu\n", eb->first_page->index,
351 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
352 (unsigned long)btrfs_header_fsid(eb),
354 printk("bad fsid on block %Lu\n", eb->start);
358 found_level = btrfs_header_level(eb);
360 ret = csum_tree_block(root, eb, 1);
364 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
365 end = eb->start + end - 1;
367 free_extent_buffer(eb);
372 static void end_workqueue_bio(struct bio *bio, int err)
374 struct end_io_wq *end_io_wq = bio->bi_private;
375 struct btrfs_fs_info *fs_info;
377 fs_info = end_io_wq->info;
378 end_io_wq->error = err;
379 end_io_wq->work.func = end_workqueue_fn;
380 end_io_wq->work.flags = 0;
381 if (bio->bi_rw & (1 << BIO_RW))
382 btrfs_queue_worker(&fs_info->endio_write_workers,
385 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
388 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
391 struct end_io_wq *end_io_wq;
392 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
396 end_io_wq->private = bio->bi_private;
397 end_io_wq->end_io = bio->bi_end_io;
398 end_io_wq->info = info;
399 end_io_wq->error = 0;
400 end_io_wq->bio = bio;
401 end_io_wq->metadata = metadata;
403 bio->bi_private = end_io_wq;
404 bio->bi_end_io = end_workqueue_bio;
408 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
410 unsigned long limit = min_t(unsigned long,
411 info->workers.max_workers,
412 info->fs_devices->open_devices);
416 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
418 return atomic_read(&info->nr_async_bios) >
419 btrfs_async_submit_limit(info);
422 static void run_one_async_submit(struct btrfs_work *work)
424 struct btrfs_fs_info *fs_info;
425 struct async_submit_bio *async;
428 async = container_of(work, struct async_submit_bio, work);
429 fs_info = BTRFS_I(async->inode)->root->fs_info;
431 limit = btrfs_async_submit_limit(fs_info);
432 limit = limit * 2 / 3;
434 atomic_dec(&fs_info->nr_async_submits);
436 if (atomic_read(&fs_info->nr_async_submits) < limit &&
437 waitqueue_active(&fs_info->async_submit_wait))
438 wake_up(&fs_info->async_submit_wait);
440 async->submit_bio_hook(async->inode, async->rw, async->bio,
445 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
446 int rw, struct bio *bio, int mirror_num,
447 extent_submit_bio_hook_t *submit_bio_hook)
449 struct async_submit_bio *async;
450 int limit = btrfs_async_submit_limit(fs_info);
452 async = kmalloc(sizeof(*async), GFP_NOFS);
456 async->inode = inode;
459 async->mirror_num = mirror_num;
460 async->submit_bio_hook = submit_bio_hook;
461 async->work.func = run_one_async_submit;
462 async->work.flags = 0;
464 while(atomic_read(&fs_info->async_submit_draining) &&
465 atomic_read(&fs_info->nr_async_submits)) {
466 wait_event(fs_info->async_submit_wait,
467 (atomic_read(&fs_info->nr_async_submits) == 0));
470 atomic_inc(&fs_info->nr_async_submits);
471 btrfs_queue_worker(&fs_info->workers, &async->work);
473 if (atomic_read(&fs_info->nr_async_submits) > limit) {
474 wait_event_timeout(fs_info->async_submit_wait,
475 (atomic_read(&fs_info->nr_async_submits) < limit),
478 wait_event_timeout(fs_info->async_submit_wait,
479 (atomic_read(&fs_info->nr_async_bios) < limit),
485 static int btree_csum_one_bio(struct bio *bio)
487 struct bio_vec *bvec = bio->bi_io_vec;
489 struct btrfs_root *root;
491 WARN_ON(bio->bi_vcnt <= 0);
492 while(bio_index < bio->bi_vcnt) {
493 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
494 csum_dirty_buffer(root, bvec->bv_page);
501 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
504 struct btrfs_root *root = BTRFS_I(inode)->root;
508 * when we're called for a write, we're already in the async
509 * submission context. Just jump into btrfs_map_bio
511 if (rw & (1 << BIO_RW)) {
512 btree_csum_one_bio(bio);
513 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
518 * called for a read, do the setup so that checksum validation
519 * can happen in the async kernel threads
521 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
524 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
527 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
531 * kthread helpers are used to submit writes so that checksumming
532 * can happen in parallel across all CPUs
534 if (!(rw & (1 << BIO_RW))) {
535 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
537 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
538 inode, rw, bio, mirror_num,
539 __btree_submit_bio_hook);
542 static int btree_writepage(struct page *page, struct writeback_control *wbc)
544 struct extent_io_tree *tree;
545 tree = &BTRFS_I(page->mapping->host)->io_tree;
547 if (current->flags & PF_MEMALLOC) {
548 redirty_page_for_writepage(wbc, page);
552 return extent_write_full_page(tree, page, btree_get_extent, wbc);
555 static int btree_writepages(struct address_space *mapping,
556 struct writeback_control *wbc)
558 struct extent_io_tree *tree;
559 tree = &BTRFS_I(mapping->host)->io_tree;
560 if (wbc->sync_mode == WB_SYNC_NONE) {
563 unsigned long thresh = 32 * 1024 * 1024;
565 if (wbc->for_kupdate)
568 num_dirty = count_range_bits(tree, &start, (u64)-1,
569 thresh, EXTENT_DIRTY);
570 if (num_dirty < thresh) {
574 return extent_writepages(tree, mapping, btree_get_extent, wbc);
577 int btree_readpage(struct file *file, struct page *page)
579 struct extent_io_tree *tree;
580 tree = &BTRFS_I(page->mapping->host)->io_tree;
581 return extent_read_full_page(tree, page, btree_get_extent);
584 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
586 struct extent_io_tree *tree;
587 struct extent_map_tree *map;
590 if (PageWriteback(page) || PageDirty(page))
593 tree = &BTRFS_I(page->mapping->host)->io_tree;
594 map = &BTRFS_I(page->mapping->host)->extent_tree;
596 ret = try_release_extent_state(map, tree, page, gfp_flags);
601 ret = try_release_extent_buffer(tree, page);
603 ClearPagePrivate(page);
604 set_page_private(page, 0);
605 page_cache_release(page);
611 static void btree_invalidatepage(struct page *page, unsigned long offset)
613 struct extent_io_tree *tree;
614 tree = &BTRFS_I(page->mapping->host)->io_tree;
615 extent_invalidatepage(tree, page, offset);
616 btree_releasepage(page, GFP_NOFS);
617 if (PagePrivate(page)) {
618 printk("warning page private not zero on page %Lu\n",
620 ClearPagePrivate(page);
621 set_page_private(page, 0);
622 page_cache_release(page);
627 static int btree_writepage(struct page *page, struct writeback_control *wbc)
629 struct buffer_head *bh;
630 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
631 struct buffer_head *head;
632 if (!page_has_buffers(page)) {
633 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
634 (1 << BH_Dirty)|(1 << BH_Uptodate));
636 head = page_buffers(page);
639 if (buffer_dirty(bh))
640 csum_tree_block(root, bh, 0);
641 bh = bh->b_this_page;
642 } while (bh != head);
643 return block_write_full_page(page, btree_get_block, wbc);
647 static struct address_space_operations btree_aops = {
648 .readpage = btree_readpage,
649 .writepage = btree_writepage,
650 .writepages = btree_writepages,
651 .releasepage = btree_releasepage,
652 .invalidatepage = btree_invalidatepage,
653 .sync_page = block_sync_page,
656 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
659 struct extent_buffer *buf = NULL;
660 struct inode *btree_inode = root->fs_info->btree_inode;
663 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
666 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
667 buf, 0, 0, btree_get_extent, 0);
668 free_extent_buffer(buf);
672 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
673 u64 bytenr, u32 blocksize)
675 struct inode *btree_inode = root->fs_info->btree_inode;
676 struct extent_buffer *eb;
677 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
678 bytenr, blocksize, GFP_NOFS);
682 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
683 u64 bytenr, u32 blocksize)
685 struct inode *btree_inode = root->fs_info->btree_inode;
686 struct extent_buffer *eb;
688 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
689 bytenr, blocksize, NULL, GFP_NOFS);
694 int btrfs_write_tree_block(struct extent_buffer *buf)
696 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
697 buf->start + buf->len - 1, WB_SYNC_ALL);
700 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
702 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
703 buf->start, buf->start + buf->len -1);
706 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
707 u32 blocksize, u64 parent_transid)
709 struct extent_buffer *buf = NULL;
710 struct inode *btree_inode = root->fs_info->btree_inode;
711 struct extent_io_tree *io_tree;
714 io_tree = &BTRFS_I(btree_inode)->io_tree;
716 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
720 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
723 buf->flags |= EXTENT_UPTODATE;
731 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
732 struct extent_buffer *buf)
734 struct inode *btree_inode = root->fs_info->btree_inode;
735 if (btrfs_header_generation(buf) ==
736 root->fs_info->running_transaction->transid) {
737 WARN_ON(!btrfs_tree_locked(buf));
738 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
744 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
745 u32 stripesize, struct btrfs_root *root,
746 struct btrfs_fs_info *fs_info,
751 root->commit_root = NULL;
752 root->ref_tree = NULL;
753 root->sectorsize = sectorsize;
754 root->nodesize = nodesize;
755 root->leafsize = leafsize;
756 root->stripesize = stripesize;
758 root->track_dirty = 0;
760 root->fs_info = fs_info;
761 root->objectid = objectid;
762 root->last_trans = 0;
763 root->highest_inode = 0;
764 root->last_inode_alloc = 0;
768 INIT_LIST_HEAD(&root->dirty_list);
769 INIT_LIST_HEAD(&root->orphan_list);
770 INIT_LIST_HEAD(&root->dead_list);
771 spin_lock_init(&root->node_lock);
772 spin_lock_init(&root->list_lock);
773 mutex_init(&root->objectid_mutex);
774 mutex_init(&root->log_mutex);
775 extent_io_tree_init(&root->dirty_log_pages,
776 fs_info->btree_inode->i_mapping, GFP_NOFS);
778 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
779 root->ref_tree = &root->ref_tree_struct;
781 memset(&root->root_key, 0, sizeof(root->root_key));
782 memset(&root->root_item, 0, sizeof(root->root_item));
783 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
784 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
785 root->defrag_trans_start = fs_info->generation;
786 init_completion(&root->kobj_unregister);
787 root->defrag_running = 0;
788 root->defrag_level = 0;
789 root->root_key.objectid = objectid;
793 static int find_and_setup_root(struct btrfs_root *tree_root,
794 struct btrfs_fs_info *fs_info,
796 struct btrfs_root *root)
801 __setup_root(tree_root->nodesize, tree_root->leafsize,
802 tree_root->sectorsize, tree_root->stripesize,
803 root, fs_info, objectid);
804 ret = btrfs_find_last_root(tree_root, objectid,
805 &root->root_item, &root->root_key);
808 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
809 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
815 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
816 struct btrfs_fs_info *fs_info)
818 struct extent_buffer *eb;
819 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
828 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
829 0, &start, &end, EXTENT_DIRTY);
833 clear_extent_dirty(&log_root_tree->dirty_log_pages,
834 start, end, GFP_NOFS);
836 eb = fs_info->log_root_tree->node;
838 WARN_ON(btrfs_header_level(eb) != 0);
839 WARN_ON(btrfs_header_nritems(eb) != 0);
841 ret = btrfs_free_reserved_extent(fs_info->tree_root,
845 free_extent_buffer(eb);
846 kfree(fs_info->log_root_tree);
847 fs_info->log_root_tree = NULL;
851 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
852 struct btrfs_fs_info *fs_info)
854 struct btrfs_root *root;
855 struct btrfs_root *tree_root = fs_info->tree_root;
857 root = kzalloc(sizeof(*root), GFP_NOFS);
861 __setup_root(tree_root->nodesize, tree_root->leafsize,
862 tree_root->sectorsize, tree_root->stripesize,
863 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
865 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
866 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
867 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
870 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
871 0, BTRFS_TREE_LOG_OBJECTID,
872 trans->transid, 0, 0, 0);
874 btrfs_set_header_nritems(root->node, 0);
875 btrfs_set_header_level(root->node, 0);
876 btrfs_set_header_bytenr(root->node, root->node->start);
877 btrfs_set_header_generation(root->node, trans->transid);
878 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
880 write_extent_buffer(root->node, root->fs_info->fsid,
881 (unsigned long)btrfs_header_fsid(root->node),
883 btrfs_mark_buffer_dirty(root->node);
884 btrfs_tree_unlock(root->node);
885 fs_info->log_root_tree = root;
889 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
890 struct btrfs_key *location)
892 struct btrfs_root *root;
893 struct btrfs_fs_info *fs_info = tree_root->fs_info;
894 struct btrfs_path *path;
895 struct extent_buffer *l;
900 root = kzalloc(sizeof(*root), GFP_NOFS);
902 return ERR_PTR(-ENOMEM);
903 if (location->offset == (u64)-1) {
904 ret = find_and_setup_root(tree_root, fs_info,
905 location->objectid, root);
913 __setup_root(tree_root->nodesize, tree_root->leafsize,
914 tree_root->sectorsize, tree_root->stripesize,
915 root, fs_info, location->objectid);
917 path = btrfs_alloc_path();
919 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
926 read_extent_buffer(l, &root->root_item,
927 btrfs_item_ptr_offset(l, path->slots[0]),
928 sizeof(root->root_item));
929 memcpy(&root->root_key, location, sizeof(*location));
932 btrfs_release_path(root, path);
933 btrfs_free_path(path);
938 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
939 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
943 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
945 ret = btrfs_find_highest_inode(root, &highest_inode);
947 root->highest_inode = highest_inode;
948 root->last_inode_alloc = highest_inode;
954 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
957 struct btrfs_root *root;
959 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
960 return fs_info->tree_root;
961 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
962 return fs_info->extent_root;
964 root = radix_tree_lookup(&fs_info->fs_roots_radix,
965 (unsigned long)root_objectid);
969 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
970 struct btrfs_key *location)
972 struct btrfs_root *root;
975 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
976 return fs_info->tree_root;
977 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
978 return fs_info->extent_root;
979 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
980 return fs_info->chunk_root;
981 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
982 return fs_info->dev_root;
984 root = radix_tree_lookup(&fs_info->fs_roots_radix,
985 (unsigned long)location->objectid);
989 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
992 ret = radix_tree_insert(&fs_info->fs_roots_radix,
993 (unsigned long)root->root_key.objectid,
996 free_extent_buffer(root->node);
1000 ret = btrfs_find_dead_roots(fs_info->tree_root,
1001 root->root_key.objectid, root);
1007 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1008 struct btrfs_key *location,
1009 const char *name, int namelen)
1011 struct btrfs_root *root;
1014 root = btrfs_read_fs_root_no_name(fs_info, location);
1021 ret = btrfs_set_root_name(root, name, namelen);
1023 free_extent_buffer(root->node);
1025 return ERR_PTR(ret);
1028 ret = btrfs_sysfs_add_root(root);
1030 free_extent_buffer(root->node);
1033 return ERR_PTR(ret);
1039 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1040 struct btrfs_hasher *hasher;
1042 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1045 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1046 if (!hasher->hash_tfm) {
1050 spin_lock(&info->hash_lock);
1051 list_add(&hasher->list, &info->hashers);
1052 spin_unlock(&info->hash_lock);
1057 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1059 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1061 struct list_head *cur;
1062 struct btrfs_device *device;
1063 struct backing_dev_info *bdi;
1065 if ((bdi_bits & (1 << BDI_write_congested)) &&
1066 btrfs_congested_async(info, 0))
1069 list_for_each(cur, &info->fs_devices->devices) {
1070 device = list_entry(cur, struct btrfs_device, dev_list);
1073 bdi = blk_get_backing_dev_info(device->bdev);
1074 if (bdi && bdi_congested(bdi, bdi_bits)) {
1083 * this unplugs every device on the box, and it is only used when page
1086 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1088 struct list_head *cur;
1089 struct btrfs_device *device;
1090 struct btrfs_fs_info *info;
1092 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1093 list_for_each(cur, &info->fs_devices->devices) {
1094 device = list_entry(cur, struct btrfs_device, dev_list);
1095 bdi = blk_get_backing_dev_info(device->bdev);
1096 if (bdi->unplug_io_fn) {
1097 bdi->unplug_io_fn(bdi, page);
1102 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1104 struct inode *inode;
1105 struct extent_map_tree *em_tree;
1106 struct extent_map *em;
1107 struct address_space *mapping;
1110 /* the generic O_DIRECT read code does this */
1112 __unplug_io_fn(bdi, page);
1117 * page->mapping may change at any time. Get a consistent copy
1118 * and use that for everything below
1121 mapping = page->mapping;
1125 inode = mapping->host;
1126 offset = page_offset(page);
1128 em_tree = &BTRFS_I(inode)->extent_tree;
1129 spin_lock(&em_tree->lock);
1130 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1131 spin_unlock(&em_tree->lock);
1133 __unplug_io_fn(bdi, page);
1137 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1138 free_extent_map(em);
1139 __unplug_io_fn(bdi, page);
1142 offset = offset - em->start;
1143 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1144 em->block_start + offset, page);
1145 free_extent_map(em);
1148 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1151 bdi->ra_pages = default_backing_dev_info.ra_pages;
1153 bdi->capabilities = default_backing_dev_info.capabilities;
1154 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1155 bdi->unplug_io_data = info;
1156 bdi->congested_fn = btrfs_congested_fn;
1157 bdi->congested_data = info;
1161 static int bio_ready_for_csum(struct bio *bio)
1167 struct extent_io_tree *io_tree = NULL;
1168 struct btrfs_fs_info *info = NULL;
1169 struct bio_vec *bvec;
1173 bio_for_each_segment(bvec, bio, i) {
1174 page = bvec->bv_page;
1175 if (page->private == EXTENT_PAGE_PRIVATE) {
1176 length += bvec->bv_len;
1179 if (!page->private) {
1180 length += bvec->bv_len;
1183 length = bvec->bv_len;
1184 buf_len = page->private >> 2;
1185 start = page_offset(page) + bvec->bv_offset;
1186 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1187 info = BTRFS_I(page->mapping->host)->root->fs_info;
1189 /* are we fully contained in this bio? */
1190 if (buf_len <= length)
1193 ret = extent_range_uptodate(io_tree, start + length,
1194 start + buf_len - 1);
1201 * called by the kthread helper functions to finally call the bio end_io
1202 * functions. This is where read checksum verification actually happens
1204 static void end_workqueue_fn(struct btrfs_work *work)
1207 struct end_io_wq *end_io_wq;
1208 struct btrfs_fs_info *fs_info;
1211 end_io_wq = container_of(work, struct end_io_wq, work);
1212 bio = end_io_wq->bio;
1213 fs_info = end_io_wq->info;
1215 /* metadata bios are special because the whole tree block must
1216 * be checksummed at once. This makes sure the entire block is in
1217 * ram and up to date before trying to verify things. For
1218 * blocksize <= pagesize, it is basically a noop
1220 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1221 btrfs_queue_worker(&fs_info->endio_workers,
1225 error = end_io_wq->error;
1226 bio->bi_private = end_io_wq->private;
1227 bio->bi_end_io = end_io_wq->end_io;
1229 bio_endio(bio, error);
1232 static int cleaner_kthread(void *arg)
1234 struct btrfs_root *root = arg;
1238 if (root->fs_info->closing)
1241 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1242 mutex_lock(&root->fs_info->cleaner_mutex);
1243 btrfs_clean_old_snapshots(root);
1244 mutex_unlock(&root->fs_info->cleaner_mutex);
1246 if (freezing(current)) {
1250 if (root->fs_info->closing)
1252 set_current_state(TASK_INTERRUPTIBLE);
1254 __set_current_state(TASK_RUNNING);
1256 } while (!kthread_should_stop());
1260 static int transaction_kthread(void *arg)
1262 struct btrfs_root *root = arg;
1263 struct btrfs_trans_handle *trans;
1264 struct btrfs_transaction *cur;
1266 unsigned long delay;
1271 if (root->fs_info->closing)
1275 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1276 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1278 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1279 printk("btrfs: total reference cache size %Lu\n",
1280 root->fs_info->total_ref_cache_size);
1283 mutex_lock(&root->fs_info->trans_mutex);
1284 cur = root->fs_info->running_transaction;
1286 mutex_unlock(&root->fs_info->trans_mutex);
1290 now = get_seconds();
1291 if (now < cur->start_time || now - cur->start_time < 30) {
1292 mutex_unlock(&root->fs_info->trans_mutex);
1296 mutex_unlock(&root->fs_info->trans_mutex);
1297 trans = btrfs_start_transaction(root, 1);
1298 ret = btrfs_commit_transaction(trans, root);
1300 wake_up_process(root->fs_info->cleaner_kthread);
1301 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1303 if (freezing(current)) {
1306 if (root->fs_info->closing)
1308 set_current_state(TASK_INTERRUPTIBLE);
1309 schedule_timeout(delay);
1310 __set_current_state(TASK_RUNNING);
1312 } while (!kthread_should_stop());
1316 struct btrfs_root *open_ctree(struct super_block *sb,
1317 struct btrfs_fs_devices *fs_devices,
1325 struct buffer_head *bh;
1326 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1328 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1330 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1332 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1334 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1336 struct btrfs_root *log_tree_root;
1341 struct btrfs_super_block *disk_super;
1343 if (!extent_root || !tree_root || !fs_info) {
1347 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1348 INIT_LIST_HEAD(&fs_info->trans_list);
1349 INIT_LIST_HEAD(&fs_info->dead_roots);
1350 INIT_LIST_HEAD(&fs_info->hashers);
1351 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1352 spin_lock_init(&fs_info->hash_lock);
1353 spin_lock_init(&fs_info->delalloc_lock);
1354 spin_lock_init(&fs_info->new_trans_lock);
1355 spin_lock_init(&fs_info->ref_cache_lock);
1357 init_completion(&fs_info->kobj_unregister);
1358 fs_info->tree_root = tree_root;
1359 fs_info->extent_root = extent_root;
1360 fs_info->chunk_root = chunk_root;
1361 fs_info->dev_root = dev_root;
1362 fs_info->fs_devices = fs_devices;
1363 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1364 INIT_LIST_HEAD(&fs_info->space_info);
1365 btrfs_mapping_init(&fs_info->mapping_tree);
1366 atomic_set(&fs_info->nr_async_submits, 0);
1367 atomic_set(&fs_info->async_submit_draining, 0);
1368 atomic_set(&fs_info->nr_async_bios, 0);
1369 atomic_set(&fs_info->throttles, 0);
1370 atomic_set(&fs_info->throttle_gen, 0);
1372 fs_info->max_extent = (u64)-1;
1373 fs_info->max_inline = 8192 * 1024;
1374 setup_bdi(fs_info, &fs_info->bdi);
1375 fs_info->btree_inode = new_inode(sb);
1376 fs_info->btree_inode->i_ino = 1;
1377 fs_info->btree_inode->i_nlink = 1;
1378 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1380 INIT_LIST_HEAD(&fs_info->ordered_extents);
1381 spin_lock_init(&fs_info->ordered_extent_lock);
1383 sb->s_blocksize = 4096;
1384 sb->s_blocksize_bits = blksize_bits(4096);
1387 * we set the i_size on the btree inode to the max possible int.
1388 * the real end of the address space is determined by all of
1389 * the devices in the system
1391 fs_info->btree_inode->i_size = OFFSET_MAX;
1392 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1393 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1395 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1396 fs_info->btree_inode->i_mapping,
1398 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1401 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1403 spin_lock_init(&fs_info->block_group_cache_lock);
1404 fs_info->block_group_cache_tree.rb_node = NULL;
1406 extent_io_tree_init(&fs_info->pinned_extents,
1407 fs_info->btree_inode->i_mapping, GFP_NOFS);
1408 extent_io_tree_init(&fs_info->pending_del,
1409 fs_info->btree_inode->i_mapping, GFP_NOFS);
1410 extent_io_tree_init(&fs_info->extent_ins,
1411 fs_info->btree_inode->i_mapping, GFP_NOFS);
1412 fs_info->do_barriers = 1;
1414 extent_io_tree_init(&fs_info->reloc_mapping_tree,
1415 fs_info->btree_inode->i_mapping, GFP_NOFS);
1416 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1417 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1418 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1420 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1421 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1422 sizeof(struct btrfs_key));
1423 insert_inode_hash(fs_info->btree_inode);
1425 mutex_init(&fs_info->trans_mutex);
1426 mutex_init(&fs_info->tree_log_mutex);
1427 mutex_init(&fs_info->drop_mutex);
1428 mutex_init(&fs_info->alloc_mutex);
1429 mutex_init(&fs_info->chunk_mutex);
1430 mutex_init(&fs_info->transaction_kthread_mutex);
1431 mutex_init(&fs_info->cleaner_mutex);
1432 mutex_init(&fs_info->volume_mutex);
1433 mutex_init(&fs_info->tree_reloc_mutex);
1434 init_waitqueue_head(&fs_info->transaction_throttle);
1435 init_waitqueue_head(&fs_info->transaction_wait);
1436 init_waitqueue_head(&fs_info->async_submit_wait);
1437 init_waitqueue_head(&fs_info->tree_log_wait);
1438 atomic_set(&fs_info->tree_log_commit, 0);
1439 atomic_set(&fs_info->tree_log_writers, 0);
1440 fs_info->tree_log_transid = 0;
1443 ret = add_hasher(fs_info, "crc32c");
1445 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1450 __setup_root(4096, 4096, 4096, 4096, tree_root,
1451 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1454 bh = __bread(fs_devices->latest_bdev,
1455 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1459 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1462 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1464 disk_super = &fs_info->super_copy;
1465 if (!btrfs_super_root(disk_super))
1466 goto fail_sb_buffer;
1468 err = btrfs_parse_options(tree_root, options);
1470 goto fail_sb_buffer;
1473 * we need to start all the end_io workers up front because the
1474 * queue work function gets called at interrupt time, and so it
1475 * cannot dynamically grow.
1477 btrfs_init_workers(&fs_info->workers, "worker",
1478 fs_info->thread_pool_size);
1479 btrfs_init_workers(&fs_info->submit_workers, "submit",
1480 min_t(u64, fs_devices->num_devices,
1481 fs_info->thread_pool_size));
1483 /* a higher idle thresh on the submit workers makes it much more
1484 * likely that bios will be send down in a sane order to the
1487 fs_info->submit_workers.idle_thresh = 64;
1489 /* fs_info->workers is responsible for checksumming file data
1490 * blocks and metadata. Using a larger idle thresh allows each
1491 * worker thread to operate on things in roughly the order they
1492 * were sent by the writeback daemons, improving overall locality
1493 * of the IO going down the pipe.
1495 fs_info->workers.idle_thresh = 128;
1497 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1498 btrfs_init_workers(&fs_info->endio_workers, "endio",
1499 fs_info->thread_pool_size);
1500 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1501 fs_info->thread_pool_size);
1504 * endios are largely parallel and should have a very
1507 fs_info->endio_workers.idle_thresh = 4;
1508 fs_info->endio_write_workers.idle_thresh = 64;
1510 btrfs_start_workers(&fs_info->workers, 1);
1511 btrfs_start_workers(&fs_info->submit_workers, 1);
1512 btrfs_start_workers(&fs_info->fixup_workers, 1);
1513 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1514 btrfs_start_workers(&fs_info->endio_write_workers,
1515 fs_info->thread_pool_size);
1518 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1519 printk("Btrfs: wanted %llu devices, but found %llu\n",
1520 (unsigned long long)btrfs_super_num_devices(disk_super),
1521 (unsigned long long)fs_devices->open_devices);
1522 if (btrfs_test_opt(tree_root, DEGRADED))
1523 printk("continuing in degraded mode\n");
1525 goto fail_sb_buffer;
1529 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1531 nodesize = btrfs_super_nodesize(disk_super);
1532 leafsize = btrfs_super_leafsize(disk_super);
1533 sectorsize = btrfs_super_sectorsize(disk_super);
1534 stripesize = btrfs_super_stripesize(disk_super);
1535 tree_root->nodesize = nodesize;
1536 tree_root->leafsize = leafsize;
1537 tree_root->sectorsize = sectorsize;
1538 tree_root->stripesize = stripesize;
1540 sb->s_blocksize = sectorsize;
1541 sb->s_blocksize_bits = blksize_bits(sectorsize);
1543 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1544 sizeof(disk_super->magic))) {
1545 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1546 goto fail_sb_buffer;
1549 mutex_lock(&fs_info->chunk_mutex);
1550 ret = btrfs_read_sys_array(tree_root);
1551 mutex_unlock(&fs_info->chunk_mutex);
1553 printk("btrfs: failed to read the system array on %s\n",
1555 goto fail_sys_array;
1558 blocksize = btrfs_level_size(tree_root,
1559 btrfs_super_chunk_root_level(disk_super));
1561 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1562 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1564 chunk_root->node = read_tree_block(chunk_root,
1565 btrfs_super_chunk_root(disk_super),
1567 BUG_ON(!chunk_root->node);
1569 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1570 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1573 mutex_lock(&fs_info->chunk_mutex);
1574 ret = btrfs_read_chunk_tree(chunk_root);
1575 mutex_unlock(&fs_info->chunk_mutex);
1578 btrfs_close_extra_devices(fs_devices);
1580 blocksize = btrfs_level_size(tree_root,
1581 btrfs_super_root_level(disk_super));
1584 tree_root->node = read_tree_block(tree_root,
1585 btrfs_super_root(disk_super),
1587 if (!tree_root->node)
1588 goto fail_sb_buffer;
1591 ret = find_and_setup_root(tree_root, fs_info,
1592 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1594 goto fail_tree_root;
1595 extent_root->track_dirty = 1;
1597 ret = find_and_setup_root(tree_root, fs_info,
1598 BTRFS_DEV_TREE_OBJECTID, dev_root);
1599 dev_root->track_dirty = 1;
1602 goto fail_extent_root;
1604 btrfs_read_block_groups(extent_root);
1606 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1607 fs_info->data_alloc_profile = (u64)-1;
1608 fs_info->metadata_alloc_profile = (u64)-1;
1609 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1610 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1612 if (!fs_info->cleaner_kthread)
1613 goto fail_extent_root;
1615 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1617 "btrfs-transaction");
1618 if (!fs_info->transaction_kthread)
1621 if (btrfs_super_log_root(disk_super) != 0) {
1623 u64 bytenr = btrfs_super_log_root(disk_super);
1626 btrfs_level_size(tree_root,
1627 btrfs_super_log_root_level(disk_super));
1629 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1632 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1633 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1635 log_tree_root->node = read_tree_block(tree_root, bytenr,
1637 ret = btrfs_recover_log_trees(log_tree_root);
1641 ret = btrfs_cleanup_reloc_trees(tree_root);
1644 fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1648 kthread_stop(fs_info->cleaner_kthread);
1650 free_extent_buffer(extent_root->node);
1652 free_extent_buffer(tree_root->node);
1655 btrfs_stop_workers(&fs_info->fixup_workers);
1656 btrfs_stop_workers(&fs_info->workers);
1657 btrfs_stop_workers(&fs_info->endio_workers);
1658 btrfs_stop_workers(&fs_info->endio_write_workers);
1659 btrfs_stop_workers(&fs_info->submit_workers);
1661 iput(fs_info->btree_inode);
1663 btrfs_close_devices(fs_info->fs_devices);
1664 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1668 bdi_destroy(&fs_info->bdi);
1670 return ERR_PTR(err);
1673 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1675 char b[BDEVNAME_SIZE];
1678 set_buffer_uptodate(bh);
1680 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1681 printk(KERN_WARNING "lost page write due to "
1682 "I/O error on %s\n",
1683 bdevname(bh->b_bdev, b));
1685 /* note, we dont' set_buffer_write_io_error because we have
1686 * our own ways of dealing with the IO errors
1688 clear_buffer_uptodate(bh);
1694 int write_all_supers(struct btrfs_root *root)
1696 struct list_head *cur;
1697 struct list_head *head = &root->fs_info->fs_devices->devices;
1698 struct btrfs_device *dev;
1699 struct btrfs_super_block *sb;
1700 struct btrfs_dev_item *dev_item;
1701 struct buffer_head *bh;
1705 int total_errors = 0;
1709 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1710 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1712 sb = &root->fs_info->super_for_commit;
1713 dev_item = &sb->dev_item;
1714 list_for_each(cur, head) {
1715 dev = list_entry(cur, struct btrfs_device, dev_list);
1720 if (!dev->in_fs_metadata)
1723 btrfs_set_stack_device_type(dev_item, dev->type);
1724 btrfs_set_stack_device_id(dev_item, dev->devid);
1725 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1726 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1727 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1728 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1729 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1730 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1731 flags = btrfs_super_flags(sb);
1732 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1736 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1737 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1738 btrfs_csum_final(crc, sb->csum);
1740 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1741 BTRFS_SUPER_INFO_SIZE);
1743 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1744 dev->pending_io = bh;
1747 set_buffer_uptodate(bh);
1749 bh->b_end_io = btrfs_end_buffer_write_sync;
1751 if (do_barriers && dev->barriers) {
1752 ret = submit_bh(WRITE_BARRIER, bh);
1753 if (ret == -EOPNOTSUPP) {
1754 printk("btrfs: disabling barriers on dev %s\n",
1756 set_buffer_uptodate(bh);
1760 ret = submit_bh(WRITE, bh);
1763 ret = submit_bh(WRITE, bh);
1768 if (total_errors > max_errors) {
1769 printk("btrfs: %d errors while writing supers\n", total_errors);
1774 list_for_each(cur, head) {
1775 dev = list_entry(cur, struct btrfs_device, dev_list);
1778 if (!dev->in_fs_metadata)
1781 BUG_ON(!dev->pending_io);
1782 bh = dev->pending_io;
1784 if (!buffer_uptodate(dev->pending_io)) {
1785 if (do_barriers && dev->barriers) {
1786 printk("btrfs: disabling barriers on dev %s\n",
1788 set_buffer_uptodate(bh);
1792 ret = submit_bh(WRITE, bh);
1795 if (!buffer_uptodate(bh))
1802 dev->pending_io = NULL;
1805 if (total_errors > max_errors) {
1806 printk("btrfs: %d errors while writing supers\n", total_errors);
1812 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1817 ret = write_all_supers(root);
1821 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1823 radix_tree_delete(&fs_info->fs_roots_radix,
1824 (unsigned long)root->root_key.objectid);
1826 btrfs_sysfs_del_root(root);
1830 free_extent_buffer(root->node);
1831 if (root->commit_root)
1832 free_extent_buffer(root->commit_root);
1839 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1842 struct btrfs_root *gang[8];
1846 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1851 for (i = 0; i < ret; i++)
1852 btrfs_free_fs_root(fs_info, gang[i]);
1857 int close_ctree(struct btrfs_root *root)
1860 struct btrfs_trans_handle *trans;
1861 struct btrfs_fs_info *fs_info = root->fs_info;
1863 fs_info->closing = 1;
1866 kthread_stop(root->fs_info->transaction_kthread);
1867 kthread_stop(root->fs_info->cleaner_kthread);
1869 btrfs_clean_old_snapshots(root);
1870 trans = btrfs_start_transaction(root, 1);
1871 ret = btrfs_commit_transaction(trans, root);
1872 /* run commit again to drop the original snapshot */
1873 trans = btrfs_start_transaction(root, 1);
1874 btrfs_commit_transaction(trans, root);
1875 ret = btrfs_write_and_wait_transaction(NULL, root);
1878 write_ctree_super(NULL, root);
1880 if (fs_info->delalloc_bytes) {
1881 printk("btrfs: at unmount delalloc count %Lu\n",
1882 fs_info->delalloc_bytes);
1884 if (fs_info->total_ref_cache_size) {
1885 printk("btrfs: at umount reference cache size %Lu\n",
1886 fs_info->total_ref_cache_size);
1889 if (fs_info->extent_root->node)
1890 free_extent_buffer(fs_info->extent_root->node);
1892 if (fs_info->tree_root->node)
1893 free_extent_buffer(fs_info->tree_root->node);
1895 if (root->fs_info->chunk_root->node);
1896 free_extent_buffer(root->fs_info->chunk_root->node);
1898 if (root->fs_info->dev_root->node);
1899 free_extent_buffer(root->fs_info->dev_root->node);
1901 btrfs_free_block_groups(root->fs_info);
1902 fs_info->closing = 2;
1903 del_fs_roots(fs_info);
1905 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1907 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1909 btrfs_stop_workers(&fs_info->fixup_workers);
1910 btrfs_stop_workers(&fs_info->workers);
1911 btrfs_stop_workers(&fs_info->endio_workers);
1912 btrfs_stop_workers(&fs_info->endio_write_workers);
1913 btrfs_stop_workers(&fs_info->submit_workers);
1915 iput(fs_info->btree_inode);
1917 while(!list_empty(&fs_info->hashers)) {
1918 struct btrfs_hasher *hasher;
1919 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1921 list_del(&hasher->hashers);
1922 crypto_free_hash(&fs_info->hash_tfm);
1926 btrfs_close_devices(fs_info->fs_devices);
1927 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1929 bdi_destroy(&fs_info->bdi);
1931 kfree(fs_info->extent_root);
1932 kfree(fs_info->tree_root);
1933 kfree(fs_info->chunk_root);
1934 kfree(fs_info->dev_root);
1938 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1941 struct inode *btree_inode = buf->first_page->mapping->host;
1943 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1947 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1952 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1954 struct inode *btree_inode = buf->first_page->mapping->host;
1955 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1959 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1961 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1962 u64 transid = btrfs_header_generation(buf);
1963 struct inode *btree_inode = root->fs_info->btree_inode;
1965 WARN_ON(!btrfs_tree_locked(buf));
1966 if (transid != root->fs_info->generation) {
1967 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1968 (unsigned long long)buf->start,
1969 transid, root->fs_info->generation);
1972 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1975 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1978 * looks as though older kernels can get into trouble with
1979 * this code, they end up stuck in balance_dirty_pages forever
1981 struct extent_io_tree *tree;
1984 unsigned long thresh = 96 * 1024 * 1024;
1985 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1987 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
1990 num_dirty = count_range_bits(tree, &start, (u64)-1,
1991 thresh, EXTENT_DIRTY);
1992 if (num_dirty > thresh) {
1993 balance_dirty_pages_ratelimited_nr(
1994 root->fs_info->btree_inode->i_mapping, 1);
1999 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2001 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2003 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2005 buf->flags |= EXTENT_UPTODATE;
2010 int btree_lock_page_hook(struct page *page)
2012 struct inode *inode = page->mapping->host;
2013 struct btrfs_root *root = BTRFS_I(inode)->root;
2014 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2015 struct extent_buffer *eb;
2017 u64 bytenr = page_offset(page);
2019 if (page->private == EXTENT_PAGE_PRIVATE)
2022 len = page->private >> 2;
2023 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2027 btrfs_tree_lock(eb);
2028 spin_lock(&root->fs_info->hash_lock);
2029 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2030 spin_unlock(&root->fs_info->hash_lock);
2031 btrfs_tree_unlock(eb);
2032 free_extent_buffer(eb);
2038 static struct extent_io_ops btree_extent_io_ops = {
2039 .write_cache_pages_lock_hook = btree_lock_page_hook,
2040 .readpage_end_io_hook = btree_readpage_end_io_hook,
2041 .submit_bio_hook = btree_submit_bio_hook,
2042 /* note we're sharing with inode.c for the merge bio hook */
2043 .merge_bio_hook = btrfs_merge_bio_hook,