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"
47 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
49 if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
50 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
51 (unsigned long long)extent_buffer_blocknr(buf),
52 (unsigned long long)btrfs_header_blocknr(buf));
59 static struct extent_io_ops btree_extent_io_ops;
60 static void end_workqueue_fn(struct btrfs_work *work);
66 struct btrfs_fs_info *info;
69 struct list_head list;
70 struct btrfs_work work;
73 struct async_submit_bio {
76 struct list_head list;
77 extent_submit_bio_hook_t *submit_bio_hook;
80 struct btrfs_work work;
83 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
84 size_t page_offset, u64 start, u64 len,
87 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
88 struct extent_map *em;
91 spin_lock(&em_tree->lock);
92 em = lookup_extent_mapping(em_tree, start, len);
95 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
96 spin_unlock(&em_tree->lock);
99 spin_unlock(&em_tree->lock);
101 em = alloc_extent_map(GFP_NOFS);
103 em = ERR_PTR(-ENOMEM);
109 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
111 spin_lock(&em_tree->lock);
112 ret = add_extent_mapping(em_tree, em);
113 if (ret == -EEXIST) {
114 u64 failed_start = em->start;
115 u64 failed_len = em->len;
117 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
118 em->start, em->len, em->block_start);
120 em = lookup_extent_mapping(em_tree, start, len);
122 printk("after failing, found %Lu %Lu %Lu\n",
123 em->start, em->len, em->block_start);
126 em = lookup_extent_mapping(em_tree, failed_start,
129 printk("double failure lookup gives us "
130 "%Lu %Lu -> %Lu\n", em->start,
131 em->len, em->block_start);
140 spin_unlock(&em_tree->lock);
148 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
150 return btrfs_crc32c(seed, data, len);
153 void btrfs_csum_final(u32 crc, char *result)
155 *(__le32 *)result = ~cpu_to_le32(crc);
158 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
161 char result[BTRFS_CRC32_SIZE];
163 unsigned long cur_len;
164 unsigned long offset = BTRFS_CSUM_SIZE;
165 char *map_token = NULL;
167 unsigned long map_start;
168 unsigned long map_len;
172 len = buf->len - offset;
174 err = map_private_extent_buffer(buf, offset, 32,
176 &map_start, &map_len, KM_USER0);
178 printk("failed to map extent buffer! %lu\n",
182 cur_len = min(len, map_len - (offset - map_start));
183 crc = btrfs_csum_data(root, kaddr + offset - map_start,
187 unmap_extent_buffer(buf, map_token, KM_USER0);
189 btrfs_csum_final(crc, result);
192 /* FIXME, this is not good */
193 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
196 memcpy(&found, result, BTRFS_CRC32_SIZE);
198 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
199 printk("btrfs: %s checksum verify failed on %llu "
200 "wanted %X found %X level %d\n",
201 root->fs_info->sb->s_id,
202 buf->start, val, found, btrfs_header_level(buf));
206 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
211 static int verify_parent_transid(struct extent_io_tree *io_tree,
212 struct extent_buffer *eb, u64 parent_transid)
216 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
219 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
220 if (extent_buffer_uptodate(io_tree, eb) &&
221 btrfs_header_generation(eb) == parent_transid) {
225 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
226 (unsigned long long)eb->start,
227 (unsigned long long)parent_transid,
228 (unsigned long long)btrfs_header_generation(eb));
230 clear_extent_buffer_uptodate(io_tree, eb);
232 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
238 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
239 struct extent_buffer *eb,
240 u64 start, u64 parent_transid)
242 struct extent_io_tree *io_tree;
247 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
249 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
250 btree_get_extent, mirror_num);
252 !verify_parent_transid(io_tree, eb, parent_transid))
254 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
255 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
261 if (mirror_num > num_copies)
267 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
269 struct extent_io_tree *tree;
270 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
274 struct extent_buffer *eb;
277 tree = &BTRFS_I(page->mapping->host)->io_tree;
279 if (page->private == EXTENT_PAGE_PRIVATE)
283 len = page->private >> 2;
287 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
288 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
289 btrfs_header_generation(eb));
291 found_start = btrfs_header_bytenr(eb);
292 if (found_start != start) {
293 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
294 start, found_start, len);
298 if (eb->first_page != page) {
299 printk("bad first page %lu %lu\n", eb->first_page->index,
304 if (!PageUptodate(page)) {
305 printk("csum not up to date page %lu\n", page->index);
309 found_level = btrfs_header_level(eb);
311 csum_tree_block(root, eb, 0);
313 free_extent_buffer(eb);
318 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
320 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
322 csum_dirty_buffer(root, page);
326 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
327 struct extent_state *state)
329 struct extent_io_tree *tree;
333 struct extent_buffer *eb;
334 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
337 tree = &BTRFS_I(page->mapping->host)->io_tree;
338 if (page->private == EXTENT_PAGE_PRIVATE)
342 len = page->private >> 2;
346 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
348 found_start = btrfs_header_bytenr(eb);
349 if (found_start != start) {
350 printk("bad tree block start %llu %llu\n",
351 (unsigned long long)found_start,
352 (unsigned long long)eb->start);
356 if (eb->first_page != page) {
357 printk("bad first page %lu %lu\n", eb->first_page->index,
363 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
364 (unsigned long)btrfs_header_fsid(eb),
366 printk("bad fsid on block %Lu\n", eb->start);
370 found_level = btrfs_header_level(eb);
372 ret = csum_tree_block(root, eb, 1);
376 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
377 end = eb->start + end - 1;
379 free_extent_buffer(eb);
384 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
385 static void end_workqueue_bio(struct bio *bio, int err)
387 static int end_workqueue_bio(struct bio *bio,
388 unsigned int bytes_done, int err)
391 struct end_io_wq *end_io_wq = bio->bi_private;
392 struct btrfs_fs_info *fs_info;
394 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
399 fs_info = end_io_wq->info;
400 end_io_wq->error = err;
401 end_io_wq->work.func = end_workqueue_fn;
402 end_io_wq->work.flags = 0;
403 if (bio->bi_rw & (1 << BIO_RW))
404 btrfs_queue_worker(&fs_info->endio_write_workers,
407 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
409 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
414 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
417 struct end_io_wq *end_io_wq;
418 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
422 end_io_wq->private = bio->bi_private;
423 end_io_wq->end_io = bio->bi_end_io;
424 end_io_wq->info = info;
425 end_io_wq->error = 0;
426 end_io_wq->bio = bio;
427 end_io_wq->metadata = metadata;
429 bio->bi_private = end_io_wq;
430 bio->bi_end_io = end_workqueue_bio;
434 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
436 unsigned long limit = min_t(unsigned long,
437 info->workers.max_workers,
438 info->fs_devices->open_devices);
442 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
444 return atomic_read(&info->nr_async_bios) >
445 btrfs_async_submit_limit(info);
448 static void run_one_async_submit(struct btrfs_work *work)
450 struct btrfs_fs_info *fs_info;
451 struct async_submit_bio *async;
454 async = container_of(work, struct async_submit_bio, work);
455 fs_info = BTRFS_I(async->inode)->root->fs_info;
457 limit = btrfs_async_submit_limit(fs_info);
458 limit = limit * 2 / 3;
460 atomic_dec(&fs_info->nr_async_submits);
462 if (atomic_read(&fs_info->nr_async_submits) < limit &&
463 waitqueue_active(&fs_info->async_submit_wait))
464 wake_up(&fs_info->async_submit_wait);
466 async->submit_bio_hook(async->inode, async->rw, async->bio,
471 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
472 int rw, struct bio *bio, int mirror_num,
473 extent_submit_bio_hook_t *submit_bio_hook)
475 struct async_submit_bio *async;
476 int limit = btrfs_async_submit_limit(fs_info);
478 async = kmalloc(sizeof(*async), GFP_NOFS);
482 async->inode = inode;
485 async->mirror_num = mirror_num;
486 async->submit_bio_hook = submit_bio_hook;
487 async->work.func = run_one_async_submit;
488 async->work.flags = 0;
489 atomic_inc(&fs_info->nr_async_submits);
490 btrfs_queue_worker(&fs_info->workers, &async->work);
492 if (atomic_read(&fs_info->nr_async_submits) > limit) {
493 wait_event_timeout(fs_info->async_submit_wait,
494 (atomic_read(&fs_info->nr_async_submits) < limit),
497 wait_event_timeout(fs_info->async_submit_wait,
498 (atomic_read(&fs_info->nr_async_bios) < limit),
504 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
507 struct btrfs_root *root = BTRFS_I(inode)->root;
511 offset = bio->bi_sector << 9;
514 * when we're called for a write, we're already in the async
515 * submission context. Just jump into btrfs_map_bio
517 if (rw & (1 << BIO_RW)) {
518 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
523 * called for a read, do the setup so that checksum validation
524 * can happen in the async kernel threads
526 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
529 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
532 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
536 * kthread helpers are used to submit writes so that checksumming
537 * can happen in parallel across all CPUs
539 if (!(rw & (1 << BIO_RW))) {
540 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
542 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
543 inode, rw, bio, mirror_num,
544 __btree_submit_bio_hook);
547 static int btree_writepage(struct page *page, struct writeback_control *wbc)
549 struct extent_io_tree *tree;
550 tree = &BTRFS_I(page->mapping->host)->io_tree;
552 if (current->flags & PF_MEMALLOC) {
553 redirty_page_for_writepage(wbc, page);
557 return extent_write_full_page(tree, page, btree_get_extent, wbc);
560 static int btree_writepages(struct address_space *mapping,
561 struct writeback_control *wbc)
563 struct extent_io_tree *tree;
564 tree = &BTRFS_I(mapping->host)->io_tree;
565 if (wbc->sync_mode == WB_SYNC_NONE) {
568 unsigned long thresh = 8 * 1024 * 1024;
570 if (wbc->for_kupdate)
573 num_dirty = count_range_bits(tree, &start, (u64)-1,
574 thresh, EXTENT_DIRTY);
575 if (num_dirty < thresh) {
579 return extent_writepages(tree, mapping, btree_get_extent, wbc);
582 int btree_readpage(struct file *file, struct page *page)
584 struct extent_io_tree *tree;
585 tree = &BTRFS_I(page->mapping->host)->io_tree;
586 return extent_read_full_page(tree, page, btree_get_extent);
589 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
591 struct extent_io_tree *tree;
592 struct extent_map_tree *map;
595 tree = &BTRFS_I(page->mapping->host)->io_tree;
596 map = &BTRFS_I(page->mapping->host)->extent_tree;
598 ret = try_release_extent_state(map, tree, page, gfp_flags);
603 ret = try_release_extent_buffer(tree, page);
605 ClearPagePrivate(page);
606 set_page_private(page, 0);
607 page_cache_release(page);
613 static void btree_invalidatepage(struct page *page, unsigned long offset)
615 struct extent_io_tree *tree;
616 tree = &BTRFS_I(page->mapping->host)->io_tree;
617 extent_invalidatepage(tree, page, offset);
618 btree_releasepage(page, GFP_NOFS);
619 if (PagePrivate(page)) {
620 printk("warning page private not zero on page %Lu\n",
622 ClearPagePrivate(page);
623 set_page_private(page, 0);
624 page_cache_release(page);
629 static int btree_writepage(struct page *page, struct writeback_control *wbc)
631 struct buffer_head *bh;
632 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
633 struct buffer_head *head;
634 if (!page_has_buffers(page)) {
635 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
636 (1 << BH_Dirty)|(1 << BH_Uptodate));
638 head = page_buffers(page);
641 if (buffer_dirty(bh))
642 csum_tree_block(root, bh, 0);
643 bh = bh->b_this_page;
644 } while (bh != head);
645 return block_write_full_page(page, btree_get_block, wbc);
649 static struct address_space_operations btree_aops = {
650 .readpage = btree_readpage,
651 .writepage = btree_writepage,
652 .writepages = btree_writepages,
653 .releasepage = btree_releasepage,
654 .invalidatepage = btree_invalidatepage,
655 .sync_page = block_sync_page,
658 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
661 struct extent_buffer *buf = NULL;
662 struct inode *btree_inode = root->fs_info->btree_inode;
665 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
668 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
669 buf, 0, 0, btree_get_extent, 0);
670 free_extent_buffer(buf);
674 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
675 u64 bytenr, u32 blocksize)
677 struct inode *btree_inode = root->fs_info->btree_inode;
678 struct extent_buffer *eb;
679 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
680 bytenr, blocksize, GFP_NOFS);
684 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
685 u64 bytenr, u32 blocksize)
687 struct inode *btree_inode = root->fs_info->btree_inode;
688 struct extent_buffer *eb;
690 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
691 bytenr, blocksize, NULL, GFP_NOFS);
696 int btrfs_write_tree_block(struct extent_buffer *buf)
698 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
699 buf->start + buf->len - 1, WB_SYNC_NONE);
702 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
704 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
705 buf->start, buf->start + buf->len -1);
708 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
709 u32 blocksize, u64 parent_transid)
711 struct extent_buffer *buf = NULL;
712 struct inode *btree_inode = root->fs_info->btree_inode;
713 struct extent_io_tree *io_tree;
716 io_tree = &BTRFS_I(btree_inode)->io_tree;
718 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
722 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
725 buf->flags |= EXTENT_UPTODATE;
733 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
734 struct extent_buffer *buf)
736 struct inode *btree_inode = root->fs_info->btree_inode;
737 if (btrfs_header_generation(buf) ==
738 root->fs_info->running_transaction->transid) {
739 WARN_ON(!btrfs_tree_locked(buf));
740 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
746 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
747 u32 stripesize, struct btrfs_root *root,
748 struct btrfs_fs_info *fs_info,
753 root->commit_root = NULL;
754 root->ref_tree = NULL;
755 root->sectorsize = sectorsize;
756 root->nodesize = nodesize;
757 root->leafsize = leafsize;
758 root->stripesize = stripesize;
760 root->track_dirty = 0;
762 root->fs_info = fs_info;
763 root->objectid = objectid;
764 root->last_trans = 0;
765 root->highest_inode = 0;
766 root->last_inode_alloc = 0;
770 INIT_LIST_HEAD(&root->dirty_list);
771 INIT_LIST_HEAD(&root->orphan_list);
772 INIT_LIST_HEAD(&root->dead_list);
773 spin_lock_init(&root->node_lock);
774 spin_lock_init(&root->list_lock);
775 mutex_init(&root->objectid_mutex);
776 mutex_init(&root->log_mutex);
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;
821 if (!fs_info->log_root_tree)
824 eb = fs_info->log_root_tree->node;
826 WARN_ON(btrfs_header_level(eb) != 0);
827 WARN_ON(btrfs_header_nritems(eb) != 0);
829 ret = btrfs_free_extent(trans, fs_info->tree_root,
831 BTRFS_TREE_LOG_OBJECTID, 0, 0, 0, 1);
834 free_extent_buffer(eb);
835 kfree(fs_info->log_root_tree);
836 fs_info->log_root_tree = NULL;
840 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
841 struct btrfs_fs_info *fs_info)
843 struct btrfs_root *root;
844 struct btrfs_root *tree_root = fs_info->tree_root;
846 root = kzalloc(sizeof(*root), GFP_NOFS);
850 __setup_root(tree_root->nodesize, tree_root->leafsize,
851 tree_root->sectorsize, tree_root->stripesize,
852 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
854 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
855 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
856 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
859 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
860 BTRFS_TREE_LOG_OBJECTID,
863 btrfs_set_header_nritems(root->node, 0);
864 btrfs_set_header_level(root->node, 0);
865 btrfs_set_header_bytenr(root->node, root->node->start);
866 btrfs_set_header_generation(root->node, trans->transid);
867 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
869 write_extent_buffer(root->node, root->fs_info->fsid,
870 (unsigned long)btrfs_header_fsid(root->node),
872 btrfs_mark_buffer_dirty(root->node);
873 btrfs_tree_unlock(root->node);
874 fs_info->log_root_tree = root;
878 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
879 struct btrfs_key *location)
881 struct btrfs_root *root;
882 struct btrfs_fs_info *fs_info = tree_root->fs_info;
883 struct btrfs_path *path;
884 struct extent_buffer *l;
889 root = kzalloc(sizeof(*root), GFP_NOFS);
891 return ERR_PTR(-ENOMEM);
892 if (location->offset == (u64)-1) {
893 ret = find_and_setup_root(tree_root, fs_info,
894 location->objectid, root);
902 __setup_root(tree_root->nodesize, tree_root->leafsize,
903 tree_root->sectorsize, tree_root->stripesize,
904 root, fs_info, location->objectid);
906 path = btrfs_alloc_path();
908 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
915 read_extent_buffer(l, &root->root_item,
916 btrfs_item_ptr_offset(l, path->slots[0]),
917 sizeof(root->root_item));
918 memcpy(&root->root_key, location, sizeof(*location));
921 btrfs_release_path(root, path);
922 btrfs_free_path(path);
927 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
928 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
932 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
934 ret = btrfs_find_highest_inode(root, &highest_inode);
936 root->highest_inode = highest_inode;
937 root->last_inode_alloc = highest_inode;
943 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
946 struct btrfs_root *root;
948 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
949 return fs_info->tree_root;
950 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
951 return fs_info->extent_root;
953 root = radix_tree_lookup(&fs_info->fs_roots_radix,
954 (unsigned long)root_objectid);
958 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
959 struct btrfs_key *location)
961 struct btrfs_root *root;
964 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
965 return fs_info->tree_root;
966 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
967 return fs_info->extent_root;
968 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
969 return fs_info->chunk_root;
970 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
971 return fs_info->dev_root;
973 root = radix_tree_lookup(&fs_info->fs_roots_radix,
974 (unsigned long)location->objectid);
978 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
981 ret = radix_tree_insert(&fs_info->fs_roots_radix,
982 (unsigned long)root->root_key.objectid,
985 free_extent_buffer(root->node);
989 ret = btrfs_find_dead_roots(fs_info->tree_root,
990 root->root_key.objectid, root);
996 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
997 struct btrfs_key *location,
998 const char *name, int namelen)
1000 struct btrfs_root *root;
1003 root = btrfs_read_fs_root_no_name(fs_info, location);
1010 ret = btrfs_set_root_name(root, name, namelen);
1012 free_extent_buffer(root->node);
1014 return ERR_PTR(ret);
1017 ret = btrfs_sysfs_add_root(root);
1019 free_extent_buffer(root->node);
1022 return ERR_PTR(ret);
1028 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1029 struct btrfs_hasher *hasher;
1031 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1034 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1035 if (!hasher->hash_tfm) {
1039 spin_lock(&info->hash_lock);
1040 list_add(&hasher->list, &info->hashers);
1041 spin_unlock(&info->hash_lock);
1046 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1048 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1050 struct list_head *cur;
1051 struct btrfs_device *device;
1052 struct backing_dev_info *bdi;
1054 if ((bdi_bits & (1 << BDI_write_congested)) &&
1055 btrfs_congested_async(info, 0))
1058 list_for_each(cur, &info->fs_devices->devices) {
1059 device = list_entry(cur, struct btrfs_device, dev_list);
1062 bdi = blk_get_backing_dev_info(device->bdev);
1063 if (bdi && bdi_congested(bdi, bdi_bits)) {
1072 * this unplugs every device on the box, and it is only used when page
1075 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1077 struct list_head *cur;
1078 struct btrfs_device *device;
1079 struct btrfs_fs_info *info;
1081 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1082 list_for_each(cur, &info->fs_devices->devices) {
1083 device = list_entry(cur, struct btrfs_device, dev_list);
1084 bdi = blk_get_backing_dev_info(device->bdev);
1085 if (bdi->unplug_io_fn) {
1086 bdi->unplug_io_fn(bdi, page);
1091 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1093 struct inode *inode;
1094 struct extent_map_tree *em_tree;
1095 struct extent_map *em;
1096 struct address_space *mapping;
1099 /* the generic O_DIRECT read code does this */
1101 __unplug_io_fn(bdi, page);
1106 * page->mapping may change at any time. Get a consistent copy
1107 * and use that for everything below
1110 mapping = page->mapping;
1114 inode = mapping->host;
1115 offset = page_offset(page);
1117 em_tree = &BTRFS_I(inode)->extent_tree;
1118 spin_lock(&em_tree->lock);
1119 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1120 spin_unlock(&em_tree->lock);
1122 __unplug_io_fn(bdi, page);
1126 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1127 free_extent_map(em);
1128 __unplug_io_fn(bdi, page);
1131 offset = offset - em->start;
1132 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1133 em->block_start + offset, page);
1134 free_extent_map(em);
1137 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1139 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1142 bdi->ra_pages = default_backing_dev_info.ra_pages;
1144 bdi->capabilities = default_backing_dev_info.capabilities;
1145 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1146 bdi->unplug_io_data = info;
1147 bdi->congested_fn = btrfs_congested_fn;
1148 bdi->congested_data = info;
1152 static int bio_ready_for_csum(struct bio *bio)
1158 struct extent_io_tree *io_tree = NULL;
1159 struct btrfs_fs_info *info = NULL;
1160 struct bio_vec *bvec;
1164 bio_for_each_segment(bvec, bio, i) {
1165 page = bvec->bv_page;
1166 if (page->private == EXTENT_PAGE_PRIVATE) {
1167 length += bvec->bv_len;
1170 if (!page->private) {
1171 length += bvec->bv_len;
1174 length = bvec->bv_len;
1175 buf_len = page->private >> 2;
1176 start = page_offset(page) + bvec->bv_offset;
1177 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1178 info = BTRFS_I(page->mapping->host)->root->fs_info;
1180 /* are we fully contained in this bio? */
1181 if (buf_len <= length)
1184 ret = extent_range_uptodate(io_tree, start + length,
1185 start + buf_len - 1);
1192 * called by the kthread helper functions to finally call the bio end_io
1193 * functions. This is where read checksum verification actually happens
1195 static void end_workqueue_fn(struct btrfs_work *work)
1198 struct end_io_wq *end_io_wq;
1199 struct btrfs_fs_info *fs_info;
1202 end_io_wq = container_of(work, struct end_io_wq, work);
1203 bio = end_io_wq->bio;
1204 fs_info = end_io_wq->info;
1206 /* metadata bios are special because the whole tree block must
1207 * be checksummed at once. This makes sure the entire block is in
1208 * ram and up to date before trying to verify things. For
1209 * blocksize <= pagesize, it is basically a noop
1211 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1212 btrfs_queue_worker(&fs_info->endio_workers,
1216 error = end_io_wq->error;
1217 bio->bi_private = end_io_wq->private;
1218 bio->bi_end_io = end_io_wq->end_io;
1220 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1221 bio_endio(bio, bio->bi_size, error);
1223 bio_endio(bio, error);
1227 static int cleaner_kthread(void *arg)
1229 struct btrfs_root *root = arg;
1233 if (root->fs_info->closing)
1236 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1237 mutex_lock(&root->fs_info->cleaner_mutex);
1238 btrfs_clean_old_snapshots(root);
1239 mutex_unlock(&root->fs_info->cleaner_mutex);
1241 if (freezing(current)) {
1245 if (root->fs_info->closing)
1247 set_current_state(TASK_INTERRUPTIBLE);
1249 __set_current_state(TASK_RUNNING);
1251 } while (!kthread_should_stop());
1255 static int transaction_kthread(void *arg)
1257 struct btrfs_root *root = arg;
1258 struct btrfs_trans_handle *trans;
1259 struct btrfs_transaction *cur;
1261 unsigned long delay;
1266 if (root->fs_info->closing)
1270 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1271 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1273 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1274 printk("btrfs: total reference cache size %Lu\n",
1275 root->fs_info->total_ref_cache_size);
1278 mutex_lock(&root->fs_info->trans_mutex);
1279 cur = root->fs_info->running_transaction;
1281 mutex_unlock(&root->fs_info->trans_mutex);
1285 now = get_seconds();
1286 if (now < cur->start_time || now - cur->start_time < 30) {
1287 mutex_unlock(&root->fs_info->trans_mutex);
1291 mutex_unlock(&root->fs_info->trans_mutex);
1292 trans = btrfs_start_transaction(root, 1);
1293 ret = btrfs_commit_transaction(trans, root);
1295 wake_up_process(root->fs_info->cleaner_kthread);
1296 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1298 if (freezing(current)) {
1301 if (root->fs_info->closing)
1303 set_current_state(TASK_INTERRUPTIBLE);
1304 schedule_timeout(delay);
1305 __set_current_state(TASK_RUNNING);
1307 } while (!kthread_should_stop());
1311 struct btrfs_root *open_ctree(struct super_block *sb,
1312 struct btrfs_fs_devices *fs_devices,
1320 struct buffer_head *bh;
1321 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1323 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1325 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1327 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1329 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1331 struct btrfs_root *log_tree_root;
1336 struct btrfs_super_block *disk_super;
1338 if (!extent_root || !tree_root || !fs_info) {
1342 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1343 INIT_LIST_HEAD(&fs_info->trans_list);
1344 INIT_LIST_HEAD(&fs_info->dead_roots);
1345 INIT_LIST_HEAD(&fs_info->hashers);
1346 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1347 spin_lock_init(&fs_info->hash_lock);
1348 spin_lock_init(&fs_info->delalloc_lock);
1349 spin_lock_init(&fs_info->new_trans_lock);
1350 spin_lock_init(&fs_info->ref_cache_lock);
1352 init_completion(&fs_info->kobj_unregister);
1353 fs_info->tree_root = tree_root;
1354 fs_info->extent_root = extent_root;
1355 fs_info->chunk_root = chunk_root;
1356 fs_info->dev_root = dev_root;
1357 fs_info->fs_devices = fs_devices;
1358 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1359 INIT_LIST_HEAD(&fs_info->space_info);
1360 btrfs_mapping_init(&fs_info->mapping_tree);
1361 atomic_set(&fs_info->nr_async_submits, 0);
1362 atomic_set(&fs_info->nr_async_bios, 0);
1363 atomic_set(&fs_info->throttles, 0);
1364 atomic_set(&fs_info->throttle_gen, 0);
1366 fs_info->max_extent = (u64)-1;
1367 fs_info->max_inline = 8192 * 1024;
1368 setup_bdi(fs_info, &fs_info->bdi);
1369 fs_info->btree_inode = new_inode(sb);
1370 fs_info->btree_inode->i_ino = 1;
1371 fs_info->btree_inode->i_nlink = 1;
1372 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1374 INIT_LIST_HEAD(&fs_info->ordered_extents);
1375 spin_lock_init(&fs_info->ordered_extent_lock);
1377 sb->s_blocksize = 4096;
1378 sb->s_blocksize_bits = blksize_bits(4096);
1381 * we set the i_size on the btree inode to the max possible int.
1382 * the real end of the address space is determined by all of
1383 * the devices in the system
1385 fs_info->btree_inode->i_size = OFFSET_MAX;
1386 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1387 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1389 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1390 fs_info->btree_inode->i_mapping,
1392 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1395 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1397 extent_io_tree_init(&fs_info->free_space_cache,
1398 fs_info->btree_inode->i_mapping, GFP_NOFS);
1399 extent_io_tree_init(&fs_info->block_group_cache,
1400 fs_info->btree_inode->i_mapping, GFP_NOFS);
1401 extent_io_tree_init(&fs_info->pinned_extents,
1402 fs_info->btree_inode->i_mapping, GFP_NOFS);
1403 extent_io_tree_init(&fs_info->pending_del,
1404 fs_info->btree_inode->i_mapping, GFP_NOFS);
1405 extent_io_tree_init(&fs_info->extent_ins,
1406 fs_info->btree_inode->i_mapping, GFP_NOFS);
1407 fs_info->do_barriers = 1;
1409 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1410 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1411 sizeof(struct btrfs_key));
1412 insert_inode_hash(fs_info->btree_inode);
1413 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1415 mutex_init(&fs_info->trans_mutex);
1416 mutex_init(&fs_info->tree_log_mutex);
1417 mutex_init(&fs_info->drop_mutex);
1418 mutex_init(&fs_info->alloc_mutex);
1419 mutex_init(&fs_info->chunk_mutex);
1420 mutex_init(&fs_info->transaction_kthread_mutex);
1421 mutex_init(&fs_info->cleaner_mutex);
1422 mutex_init(&fs_info->volume_mutex);
1423 init_waitqueue_head(&fs_info->transaction_throttle);
1424 init_waitqueue_head(&fs_info->transaction_wait);
1425 init_waitqueue_head(&fs_info->async_submit_wait);
1426 init_waitqueue_head(&fs_info->tree_log_wait);
1427 atomic_set(&fs_info->tree_log_commit, 0);
1428 atomic_set(&fs_info->tree_log_writers, 0);
1429 fs_info->tree_log_transid = 0;
1432 ret = add_hasher(fs_info, "crc32c");
1434 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1439 __setup_root(4096, 4096, 4096, 4096, tree_root,
1440 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1443 bh = __bread(fs_devices->latest_bdev,
1444 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1448 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1451 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1453 disk_super = &fs_info->super_copy;
1454 if (!btrfs_super_root(disk_super))
1455 goto fail_sb_buffer;
1457 err = btrfs_parse_options(tree_root, options);
1459 goto fail_sb_buffer;
1462 * we need to start all the end_io workers up front because the
1463 * queue work function gets called at interrupt time, and so it
1464 * cannot dynamically grow.
1466 btrfs_init_workers(&fs_info->workers, "worker",
1467 fs_info->thread_pool_size);
1468 btrfs_init_workers(&fs_info->submit_workers, "submit",
1469 min_t(u64, fs_devices->num_devices,
1470 fs_info->thread_pool_size));
1472 /* a higher idle thresh on the submit workers makes it much more
1473 * likely that bios will be send down in a sane order to the
1476 fs_info->submit_workers.idle_thresh = 64;
1478 /* fs_info->workers is responsible for checksumming file data
1479 * blocks and metadata. Using a larger idle thresh allows each
1480 * worker thread to operate on things in roughly the order they
1481 * were sent by the writeback daemons, improving overall locality
1482 * of the IO going down the pipe.
1484 fs_info->workers.idle_thresh = 128;
1486 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1487 btrfs_init_workers(&fs_info->endio_workers, "endio",
1488 fs_info->thread_pool_size);
1489 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1490 fs_info->thread_pool_size);
1493 * endios are largely parallel and should have a very
1496 fs_info->endio_workers.idle_thresh = 4;
1497 fs_info->endio_write_workers.idle_thresh = 4;
1499 btrfs_start_workers(&fs_info->workers, 1);
1500 btrfs_start_workers(&fs_info->submit_workers, 1);
1501 btrfs_start_workers(&fs_info->fixup_workers, 1);
1502 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1503 btrfs_start_workers(&fs_info->endio_write_workers,
1504 fs_info->thread_pool_size);
1507 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1508 printk("Btrfs: wanted %llu devices, but found %llu\n",
1509 (unsigned long long)btrfs_super_num_devices(disk_super),
1510 (unsigned long long)fs_devices->open_devices);
1511 if (btrfs_test_opt(tree_root, DEGRADED))
1512 printk("continuing in degraded mode\n");
1514 goto fail_sb_buffer;
1518 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1520 nodesize = btrfs_super_nodesize(disk_super);
1521 leafsize = btrfs_super_leafsize(disk_super);
1522 sectorsize = btrfs_super_sectorsize(disk_super);
1523 stripesize = btrfs_super_stripesize(disk_super);
1524 tree_root->nodesize = nodesize;
1525 tree_root->leafsize = leafsize;
1526 tree_root->sectorsize = sectorsize;
1527 tree_root->stripesize = stripesize;
1529 sb->s_blocksize = sectorsize;
1530 sb->s_blocksize_bits = blksize_bits(sectorsize);
1532 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1533 sizeof(disk_super->magic))) {
1534 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1535 goto fail_sb_buffer;
1538 mutex_lock(&fs_info->chunk_mutex);
1539 ret = btrfs_read_sys_array(tree_root);
1540 mutex_unlock(&fs_info->chunk_mutex);
1542 printk("btrfs: failed to read the system array on %s\n",
1544 goto fail_sys_array;
1547 blocksize = btrfs_level_size(tree_root,
1548 btrfs_super_chunk_root_level(disk_super));
1550 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1551 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1553 chunk_root->node = read_tree_block(chunk_root,
1554 btrfs_super_chunk_root(disk_super),
1556 BUG_ON(!chunk_root->node);
1558 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1559 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1562 mutex_lock(&fs_info->chunk_mutex);
1563 ret = btrfs_read_chunk_tree(chunk_root);
1564 mutex_unlock(&fs_info->chunk_mutex);
1567 btrfs_close_extra_devices(fs_devices);
1569 blocksize = btrfs_level_size(tree_root,
1570 btrfs_super_root_level(disk_super));
1573 tree_root->node = read_tree_block(tree_root,
1574 btrfs_super_root(disk_super),
1576 if (!tree_root->node)
1577 goto fail_sb_buffer;
1580 ret = find_and_setup_root(tree_root, fs_info,
1581 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1583 goto fail_tree_root;
1584 extent_root->track_dirty = 1;
1586 ret = find_and_setup_root(tree_root, fs_info,
1587 BTRFS_DEV_TREE_OBJECTID, dev_root);
1588 dev_root->track_dirty = 1;
1591 goto fail_extent_root;
1593 btrfs_read_block_groups(extent_root);
1595 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1596 fs_info->data_alloc_profile = (u64)-1;
1597 fs_info->metadata_alloc_profile = (u64)-1;
1598 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1599 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1601 if (!fs_info->cleaner_kthread)
1602 goto fail_extent_root;
1604 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1606 "btrfs-transaction");
1607 if (!fs_info->transaction_kthread)
1610 if (btrfs_super_log_root(disk_super) != 0) {
1612 u64 bytenr = btrfs_super_log_root(disk_super);
1615 btrfs_level_size(tree_root,
1616 btrfs_super_log_root_level(disk_super));
1618 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1621 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1622 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1624 log_tree_root->node = read_tree_block(tree_root, bytenr,
1626 ret = btrfs_recover_log_trees(log_tree_root);
1629 fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1633 kthread_stop(fs_info->cleaner_kthread);
1635 free_extent_buffer(extent_root->node);
1637 free_extent_buffer(tree_root->node);
1640 btrfs_stop_workers(&fs_info->fixup_workers);
1641 btrfs_stop_workers(&fs_info->workers);
1642 btrfs_stop_workers(&fs_info->endio_workers);
1643 btrfs_stop_workers(&fs_info->endio_write_workers);
1644 btrfs_stop_workers(&fs_info->submit_workers);
1646 iput(fs_info->btree_inode);
1648 btrfs_close_devices(fs_info->fs_devices);
1649 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1653 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1654 bdi_destroy(&fs_info->bdi);
1657 return ERR_PTR(err);
1660 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1662 char b[BDEVNAME_SIZE];
1665 set_buffer_uptodate(bh);
1667 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1668 printk(KERN_WARNING "lost page write due to "
1669 "I/O error on %s\n",
1670 bdevname(bh->b_bdev, b));
1672 /* note, we dont' set_buffer_write_io_error because we have
1673 * our own ways of dealing with the IO errors
1675 clear_buffer_uptodate(bh);
1681 int write_all_supers(struct btrfs_root *root)
1683 struct list_head *cur;
1684 struct list_head *head = &root->fs_info->fs_devices->devices;
1685 struct btrfs_device *dev;
1686 struct btrfs_super_block *sb;
1687 struct btrfs_dev_item *dev_item;
1688 struct buffer_head *bh;
1692 int total_errors = 0;
1696 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1697 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1699 sb = &root->fs_info->super_for_commit;
1700 dev_item = &sb->dev_item;
1701 list_for_each(cur, head) {
1702 dev = list_entry(cur, struct btrfs_device, dev_list);
1707 if (!dev->in_fs_metadata)
1710 btrfs_set_stack_device_type(dev_item, dev->type);
1711 btrfs_set_stack_device_id(dev_item, dev->devid);
1712 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1713 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1714 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1715 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1716 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1717 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1718 flags = btrfs_super_flags(sb);
1719 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1723 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1724 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1725 btrfs_csum_final(crc, sb->csum);
1727 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1728 BTRFS_SUPER_INFO_SIZE);
1730 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1731 dev->pending_io = bh;
1734 set_buffer_uptodate(bh);
1736 bh->b_end_io = btrfs_end_buffer_write_sync;
1738 if (do_barriers && dev->barriers) {
1739 ret = submit_bh(WRITE_BARRIER, bh);
1740 if (ret == -EOPNOTSUPP) {
1741 printk("btrfs: disabling barriers on dev %s\n",
1743 set_buffer_uptodate(bh);
1747 ret = submit_bh(WRITE, bh);
1750 ret = submit_bh(WRITE, bh);
1755 if (total_errors > max_errors) {
1756 printk("btrfs: %d errors while writing supers\n", total_errors);
1761 list_for_each(cur, head) {
1762 dev = list_entry(cur, struct btrfs_device, dev_list);
1765 if (!dev->in_fs_metadata)
1768 BUG_ON(!dev->pending_io);
1769 bh = dev->pending_io;
1771 if (!buffer_uptodate(dev->pending_io)) {
1772 if (do_barriers && dev->barriers) {
1773 printk("btrfs: disabling barriers on dev %s\n",
1775 set_buffer_uptodate(bh);
1779 ret = submit_bh(WRITE, bh);
1782 if (!buffer_uptodate(bh))
1789 dev->pending_io = NULL;
1792 if (total_errors > max_errors) {
1793 printk("btrfs: %d errors while writing supers\n", total_errors);
1799 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1804 ret = write_all_supers(root);
1808 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1810 radix_tree_delete(&fs_info->fs_roots_radix,
1811 (unsigned long)root->root_key.objectid);
1813 btrfs_sysfs_del_root(root);
1817 free_extent_buffer(root->node);
1818 if (root->commit_root)
1819 free_extent_buffer(root->commit_root);
1826 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1829 struct btrfs_root *gang[8];
1833 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1838 for (i = 0; i < ret; i++)
1839 btrfs_free_fs_root(fs_info, gang[i]);
1844 int close_ctree(struct btrfs_root *root)
1847 struct btrfs_trans_handle *trans;
1848 struct btrfs_fs_info *fs_info = root->fs_info;
1850 fs_info->closing = 1;
1853 kthread_stop(root->fs_info->transaction_kthread);
1854 kthread_stop(root->fs_info->cleaner_kthread);
1856 btrfs_clean_old_snapshots(root);
1857 trans = btrfs_start_transaction(root, 1);
1858 ret = btrfs_commit_transaction(trans, root);
1859 /* run commit again to drop the original snapshot */
1860 trans = btrfs_start_transaction(root, 1);
1861 btrfs_commit_transaction(trans, root);
1862 ret = btrfs_write_and_wait_transaction(NULL, root);
1865 write_ctree_super(NULL, root);
1867 if (fs_info->delalloc_bytes) {
1868 printk("btrfs: at unmount delalloc count %Lu\n",
1869 fs_info->delalloc_bytes);
1871 if (fs_info->total_ref_cache_size) {
1872 printk("btrfs: at umount reference cache size %Lu\n",
1873 fs_info->total_ref_cache_size);
1876 if (fs_info->extent_root->node)
1877 free_extent_buffer(fs_info->extent_root->node);
1879 if (fs_info->tree_root->node)
1880 free_extent_buffer(fs_info->tree_root->node);
1882 if (root->fs_info->chunk_root->node);
1883 free_extent_buffer(root->fs_info->chunk_root->node);
1885 if (root->fs_info->dev_root->node);
1886 free_extent_buffer(root->fs_info->dev_root->node);
1888 btrfs_free_block_groups(root->fs_info);
1889 fs_info->closing = 2;
1890 del_fs_roots(fs_info);
1892 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1894 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1896 btrfs_stop_workers(&fs_info->fixup_workers);
1897 btrfs_stop_workers(&fs_info->workers);
1898 btrfs_stop_workers(&fs_info->endio_workers);
1899 btrfs_stop_workers(&fs_info->endio_write_workers);
1900 btrfs_stop_workers(&fs_info->submit_workers);
1902 iput(fs_info->btree_inode);
1904 while(!list_empty(&fs_info->hashers)) {
1905 struct btrfs_hasher *hasher;
1906 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1908 list_del(&hasher->hashers);
1909 crypto_free_hash(&fs_info->hash_tfm);
1913 btrfs_close_devices(fs_info->fs_devices);
1914 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1916 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1917 bdi_destroy(&fs_info->bdi);
1920 kfree(fs_info->extent_root);
1921 kfree(fs_info->tree_root);
1922 kfree(fs_info->chunk_root);
1923 kfree(fs_info->dev_root);
1927 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1930 struct inode *btree_inode = buf->first_page->mapping->host;
1932 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1936 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1941 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1943 struct inode *btree_inode = buf->first_page->mapping->host;
1944 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1948 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1950 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1951 u64 transid = btrfs_header_generation(buf);
1952 struct inode *btree_inode = root->fs_info->btree_inode;
1954 WARN_ON(!btrfs_tree_locked(buf));
1955 if (transid != root->fs_info->generation) {
1956 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1957 (unsigned long long)buf->start,
1958 transid, root->fs_info->generation);
1961 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1964 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1967 * looks as though older kernels can get into trouble with
1968 * this code, they end up stuck in balance_dirty_pages forever
1970 struct extent_io_tree *tree;
1973 unsigned long thresh = 96 * 1024 * 1024;
1974 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1976 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
1979 num_dirty = count_range_bits(tree, &start, (u64)-1,
1980 thresh, EXTENT_DIRTY);
1981 if (num_dirty > thresh) {
1982 balance_dirty_pages_ratelimited_nr(
1983 root->fs_info->btree_inode->i_mapping, 1);
1988 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1990 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1992 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1994 buf->flags |= EXTENT_UPTODATE;
1999 int btree_lock_page_hook(struct page *page)
2001 struct inode *inode = page->mapping->host;
2002 struct btrfs_root *root = BTRFS_I(inode)->root;
2003 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2004 struct extent_buffer *eb;
2006 u64 bytenr = page_offset(page);
2008 if (page->private == EXTENT_PAGE_PRIVATE)
2011 len = page->private >> 2;
2012 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2016 btrfs_tree_lock(eb);
2017 spin_lock(&root->fs_info->hash_lock);
2018 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2019 spin_unlock(&root->fs_info->hash_lock);
2020 btrfs_tree_unlock(eb);
2021 free_extent_buffer(eb);
2027 static struct extent_io_ops btree_extent_io_ops = {
2028 .write_cache_pages_lock_hook = btree_lock_page_hook,
2029 .writepage_io_hook = btree_writepage_io_hook,
2030 .readpage_end_io_hook = btree_readpage_end_io_hook,
2031 .submit_bio_hook = btree_submit_bio_hook,
2032 /* note we're sharing with inode.c for the merge bio hook */
2033 .merge_bio_hook = btrfs_merge_bio_hook,