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 if (PageWriteback(page) || PageDirty(page))
598 tree = &BTRFS_I(page->mapping->host)->io_tree;
599 map = &BTRFS_I(page->mapping->host)->extent_tree;
601 ret = try_release_extent_state(map, tree, page, gfp_flags);
606 ret = try_release_extent_buffer(tree, page);
608 ClearPagePrivate(page);
609 set_page_private(page, 0);
610 page_cache_release(page);
616 static void btree_invalidatepage(struct page *page, unsigned long offset)
618 struct extent_io_tree *tree;
619 tree = &BTRFS_I(page->mapping->host)->io_tree;
620 extent_invalidatepage(tree, page, offset);
621 btree_releasepage(page, GFP_NOFS);
622 if (PagePrivate(page)) {
623 printk("warning page private not zero on page %Lu\n",
625 ClearPagePrivate(page);
626 set_page_private(page, 0);
627 page_cache_release(page);
632 static int btree_writepage(struct page *page, struct writeback_control *wbc)
634 struct buffer_head *bh;
635 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
636 struct buffer_head *head;
637 if (!page_has_buffers(page)) {
638 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
639 (1 << BH_Dirty)|(1 << BH_Uptodate));
641 head = page_buffers(page);
644 if (buffer_dirty(bh))
645 csum_tree_block(root, bh, 0);
646 bh = bh->b_this_page;
647 } while (bh != head);
648 return block_write_full_page(page, btree_get_block, wbc);
652 static struct address_space_operations btree_aops = {
653 .readpage = btree_readpage,
654 .writepage = btree_writepage,
655 .writepages = btree_writepages,
656 .releasepage = btree_releasepage,
657 .invalidatepage = btree_invalidatepage,
658 .sync_page = block_sync_page,
661 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
664 struct extent_buffer *buf = NULL;
665 struct inode *btree_inode = root->fs_info->btree_inode;
668 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
671 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
672 buf, 0, 0, btree_get_extent, 0);
673 free_extent_buffer(buf);
677 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
678 u64 bytenr, u32 blocksize)
680 struct inode *btree_inode = root->fs_info->btree_inode;
681 struct extent_buffer *eb;
682 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
683 bytenr, blocksize, GFP_NOFS);
687 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
688 u64 bytenr, u32 blocksize)
690 struct inode *btree_inode = root->fs_info->btree_inode;
691 struct extent_buffer *eb;
693 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
694 bytenr, blocksize, NULL, GFP_NOFS);
699 int btrfs_write_tree_block(struct extent_buffer *buf)
701 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
702 buf->start + buf->len - 1, WB_SYNC_NONE);
705 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
707 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
708 buf->start, buf->start + buf->len -1);
711 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
712 u32 blocksize, u64 parent_transid)
714 struct extent_buffer *buf = NULL;
715 struct inode *btree_inode = root->fs_info->btree_inode;
716 struct extent_io_tree *io_tree;
719 io_tree = &BTRFS_I(btree_inode)->io_tree;
721 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
725 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
728 buf->flags |= EXTENT_UPTODATE;
736 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
737 struct extent_buffer *buf)
739 struct inode *btree_inode = root->fs_info->btree_inode;
740 if (btrfs_header_generation(buf) ==
741 root->fs_info->running_transaction->transid) {
742 WARN_ON(!btrfs_tree_locked(buf));
743 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
749 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
750 u32 stripesize, struct btrfs_root *root,
751 struct btrfs_fs_info *fs_info,
756 root->commit_root = NULL;
757 root->ref_tree = NULL;
758 root->sectorsize = sectorsize;
759 root->nodesize = nodesize;
760 root->leafsize = leafsize;
761 root->stripesize = stripesize;
763 root->track_dirty = 0;
765 root->fs_info = fs_info;
766 root->objectid = objectid;
767 root->last_trans = 0;
768 root->highest_inode = 0;
769 root->last_inode_alloc = 0;
773 INIT_LIST_HEAD(&root->dirty_list);
774 INIT_LIST_HEAD(&root->orphan_list);
775 INIT_LIST_HEAD(&root->dead_list);
776 spin_lock_init(&root->node_lock);
777 spin_lock_init(&root->list_lock);
778 mutex_init(&root->objectid_mutex);
779 mutex_init(&root->log_mutex);
780 extent_io_tree_init(&root->dirty_log_pages,
781 fs_info->btree_inode->i_mapping, GFP_NOFS);
783 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
784 root->ref_tree = &root->ref_tree_struct;
786 memset(&root->root_key, 0, sizeof(root->root_key));
787 memset(&root->root_item, 0, sizeof(root->root_item));
788 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
789 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
790 root->defrag_trans_start = fs_info->generation;
791 init_completion(&root->kobj_unregister);
792 root->defrag_running = 0;
793 root->defrag_level = 0;
794 root->root_key.objectid = objectid;
798 static int find_and_setup_root(struct btrfs_root *tree_root,
799 struct btrfs_fs_info *fs_info,
801 struct btrfs_root *root)
806 __setup_root(tree_root->nodesize, tree_root->leafsize,
807 tree_root->sectorsize, tree_root->stripesize,
808 root, fs_info, objectid);
809 ret = btrfs_find_last_root(tree_root, objectid,
810 &root->root_item, &root->root_key);
813 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
814 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
820 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
821 struct btrfs_fs_info *fs_info)
823 struct extent_buffer *eb;
824 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
833 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
834 0, &start, &end, EXTENT_DIRTY);
838 clear_extent_dirty(&log_root_tree->dirty_log_pages,
839 start, end, GFP_NOFS);
841 eb = fs_info->log_root_tree->node;
843 WARN_ON(btrfs_header_level(eb) != 0);
844 WARN_ON(btrfs_header_nritems(eb) != 0);
846 ret = btrfs_free_reserved_extent(fs_info->tree_root,
850 free_extent_buffer(eb);
851 kfree(fs_info->log_root_tree);
852 fs_info->log_root_tree = NULL;
856 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
857 struct btrfs_fs_info *fs_info)
859 struct btrfs_root *root;
860 struct btrfs_root *tree_root = fs_info->tree_root;
862 root = kzalloc(sizeof(*root), GFP_NOFS);
866 __setup_root(tree_root->nodesize, tree_root->leafsize,
867 tree_root->sectorsize, tree_root->stripesize,
868 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
870 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
871 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
872 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
875 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
876 BTRFS_TREE_LOG_OBJECTID,
879 btrfs_set_header_nritems(root->node, 0);
880 btrfs_set_header_level(root->node, 0);
881 btrfs_set_header_bytenr(root->node, root->node->start);
882 btrfs_set_header_generation(root->node, trans->transid);
883 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
885 write_extent_buffer(root->node, root->fs_info->fsid,
886 (unsigned long)btrfs_header_fsid(root->node),
888 btrfs_mark_buffer_dirty(root->node);
889 btrfs_tree_unlock(root->node);
890 fs_info->log_root_tree = root;
894 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
895 struct btrfs_key *location)
897 struct btrfs_root *root;
898 struct btrfs_fs_info *fs_info = tree_root->fs_info;
899 struct btrfs_path *path;
900 struct extent_buffer *l;
905 root = kzalloc(sizeof(*root), GFP_NOFS);
907 return ERR_PTR(-ENOMEM);
908 if (location->offset == (u64)-1) {
909 ret = find_and_setup_root(tree_root, fs_info,
910 location->objectid, root);
918 __setup_root(tree_root->nodesize, tree_root->leafsize,
919 tree_root->sectorsize, tree_root->stripesize,
920 root, fs_info, location->objectid);
922 path = btrfs_alloc_path();
924 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
931 read_extent_buffer(l, &root->root_item,
932 btrfs_item_ptr_offset(l, path->slots[0]),
933 sizeof(root->root_item));
934 memcpy(&root->root_key, location, sizeof(*location));
937 btrfs_release_path(root, path);
938 btrfs_free_path(path);
943 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
944 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
948 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
950 ret = btrfs_find_highest_inode(root, &highest_inode);
952 root->highest_inode = highest_inode;
953 root->last_inode_alloc = highest_inode;
959 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
962 struct btrfs_root *root;
964 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
965 return fs_info->tree_root;
966 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
967 return fs_info->extent_root;
969 root = radix_tree_lookup(&fs_info->fs_roots_radix,
970 (unsigned long)root_objectid);
974 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
975 struct btrfs_key *location)
977 struct btrfs_root *root;
980 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
981 return fs_info->tree_root;
982 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
983 return fs_info->extent_root;
984 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
985 return fs_info->chunk_root;
986 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
987 return fs_info->dev_root;
989 root = radix_tree_lookup(&fs_info->fs_roots_radix,
990 (unsigned long)location->objectid);
994 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
997 ret = radix_tree_insert(&fs_info->fs_roots_radix,
998 (unsigned long)root->root_key.objectid,
1001 free_extent_buffer(root->node);
1003 return ERR_PTR(ret);
1005 ret = btrfs_find_dead_roots(fs_info->tree_root,
1006 root->root_key.objectid, root);
1012 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1013 struct btrfs_key *location,
1014 const char *name, int namelen)
1016 struct btrfs_root *root;
1019 root = btrfs_read_fs_root_no_name(fs_info, location);
1026 ret = btrfs_set_root_name(root, name, namelen);
1028 free_extent_buffer(root->node);
1030 return ERR_PTR(ret);
1033 ret = btrfs_sysfs_add_root(root);
1035 free_extent_buffer(root->node);
1038 return ERR_PTR(ret);
1044 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1045 struct btrfs_hasher *hasher;
1047 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1050 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1051 if (!hasher->hash_tfm) {
1055 spin_lock(&info->hash_lock);
1056 list_add(&hasher->list, &info->hashers);
1057 spin_unlock(&info->hash_lock);
1062 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1064 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1066 struct list_head *cur;
1067 struct btrfs_device *device;
1068 struct backing_dev_info *bdi;
1070 if ((bdi_bits & (1 << BDI_write_congested)) &&
1071 btrfs_congested_async(info, 0))
1074 list_for_each(cur, &info->fs_devices->devices) {
1075 device = list_entry(cur, struct btrfs_device, dev_list);
1078 bdi = blk_get_backing_dev_info(device->bdev);
1079 if (bdi && bdi_congested(bdi, bdi_bits)) {
1088 * this unplugs every device on the box, and it is only used when page
1091 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1093 struct list_head *cur;
1094 struct btrfs_device *device;
1095 struct btrfs_fs_info *info;
1097 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1098 list_for_each(cur, &info->fs_devices->devices) {
1099 device = list_entry(cur, struct btrfs_device, dev_list);
1100 bdi = blk_get_backing_dev_info(device->bdev);
1101 if (bdi->unplug_io_fn) {
1102 bdi->unplug_io_fn(bdi, page);
1107 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1109 struct inode *inode;
1110 struct extent_map_tree *em_tree;
1111 struct extent_map *em;
1112 struct address_space *mapping;
1115 /* the generic O_DIRECT read code does this */
1117 __unplug_io_fn(bdi, page);
1122 * page->mapping may change at any time. Get a consistent copy
1123 * and use that for everything below
1126 mapping = page->mapping;
1130 inode = mapping->host;
1131 offset = page_offset(page);
1133 em_tree = &BTRFS_I(inode)->extent_tree;
1134 spin_lock(&em_tree->lock);
1135 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1136 spin_unlock(&em_tree->lock);
1138 __unplug_io_fn(bdi, page);
1142 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1143 free_extent_map(em);
1144 __unplug_io_fn(bdi, page);
1147 offset = offset - em->start;
1148 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1149 em->block_start + offset, page);
1150 free_extent_map(em);
1153 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1155 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1158 bdi->ra_pages = default_backing_dev_info.ra_pages;
1160 bdi->capabilities = default_backing_dev_info.capabilities;
1161 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1162 bdi->unplug_io_data = info;
1163 bdi->congested_fn = btrfs_congested_fn;
1164 bdi->congested_data = info;
1168 static int bio_ready_for_csum(struct bio *bio)
1174 struct extent_io_tree *io_tree = NULL;
1175 struct btrfs_fs_info *info = NULL;
1176 struct bio_vec *bvec;
1180 bio_for_each_segment(bvec, bio, i) {
1181 page = bvec->bv_page;
1182 if (page->private == EXTENT_PAGE_PRIVATE) {
1183 length += bvec->bv_len;
1186 if (!page->private) {
1187 length += bvec->bv_len;
1190 length = bvec->bv_len;
1191 buf_len = page->private >> 2;
1192 start = page_offset(page) + bvec->bv_offset;
1193 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1194 info = BTRFS_I(page->mapping->host)->root->fs_info;
1196 /* are we fully contained in this bio? */
1197 if (buf_len <= length)
1200 ret = extent_range_uptodate(io_tree, start + length,
1201 start + buf_len - 1);
1208 * called by the kthread helper functions to finally call the bio end_io
1209 * functions. This is where read checksum verification actually happens
1211 static void end_workqueue_fn(struct btrfs_work *work)
1214 struct end_io_wq *end_io_wq;
1215 struct btrfs_fs_info *fs_info;
1218 end_io_wq = container_of(work, struct end_io_wq, work);
1219 bio = end_io_wq->bio;
1220 fs_info = end_io_wq->info;
1222 /* metadata bios are special because the whole tree block must
1223 * be checksummed at once. This makes sure the entire block is in
1224 * ram and up to date before trying to verify things. For
1225 * blocksize <= pagesize, it is basically a noop
1227 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1228 btrfs_queue_worker(&fs_info->endio_workers,
1232 error = end_io_wq->error;
1233 bio->bi_private = end_io_wq->private;
1234 bio->bi_end_io = end_io_wq->end_io;
1236 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1237 bio_endio(bio, bio->bi_size, error);
1239 bio_endio(bio, error);
1243 static int cleaner_kthread(void *arg)
1245 struct btrfs_root *root = arg;
1249 if (root->fs_info->closing)
1252 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1253 mutex_lock(&root->fs_info->cleaner_mutex);
1254 btrfs_clean_old_snapshots(root);
1255 mutex_unlock(&root->fs_info->cleaner_mutex);
1257 if (freezing(current)) {
1261 if (root->fs_info->closing)
1263 set_current_state(TASK_INTERRUPTIBLE);
1265 __set_current_state(TASK_RUNNING);
1267 } while (!kthread_should_stop());
1271 static int transaction_kthread(void *arg)
1273 struct btrfs_root *root = arg;
1274 struct btrfs_trans_handle *trans;
1275 struct btrfs_transaction *cur;
1277 unsigned long delay;
1282 if (root->fs_info->closing)
1286 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1287 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1289 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1290 printk("btrfs: total reference cache size %Lu\n",
1291 root->fs_info->total_ref_cache_size);
1294 mutex_lock(&root->fs_info->trans_mutex);
1295 cur = root->fs_info->running_transaction;
1297 mutex_unlock(&root->fs_info->trans_mutex);
1301 now = get_seconds();
1302 if (now < cur->start_time || now - cur->start_time < 30) {
1303 mutex_unlock(&root->fs_info->trans_mutex);
1307 mutex_unlock(&root->fs_info->trans_mutex);
1308 trans = btrfs_start_transaction(root, 1);
1309 ret = btrfs_commit_transaction(trans, root);
1311 wake_up_process(root->fs_info->cleaner_kthread);
1312 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1314 if (freezing(current)) {
1317 if (root->fs_info->closing)
1319 set_current_state(TASK_INTERRUPTIBLE);
1320 schedule_timeout(delay);
1321 __set_current_state(TASK_RUNNING);
1323 } while (!kthread_should_stop());
1327 struct btrfs_root *open_ctree(struct super_block *sb,
1328 struct btrfs_fs_devices *fs_devices,
1336 struct buffer_head *bh;
1337 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1339 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1341 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1343 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1345 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1347 struct btrfs_root *log_tree_root;
1352 struct btrfs_super_block *disk_super;
1354 if (!extent_root || !tree_root || !fs_info) {
1358 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1359 INIT_LIST_HEAD(&fs_info->trans_list);
1360 INIT_LIST_HEAD(&fs_info->dead_roots);
1361 INIT_LIST_HEAD(&fs_info->hashers);
1362 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1363 spin_lock_init(&fs_info->hash_lock);
1364 spin_lock_init(&fs_info->delalloc_lock);
1365 spin_lock_init(&fs_info->new_trans_lock);
1366 spin_lock_init(&fs_info->ref_cache_lock);
1368 init_completion(&fs_info->kobj_unregister);
1369 fs_info->tree_root = tree_root;
1370 fs_info->extent_root = extent_root;
1371 fs_info->chunk_root = chunk_root;
1372 fs_info->dev_root = dev_root;
1373 fs_info->fs_devices = fs_devices;
1374 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1375 INIT_LIST_HEAD(&fs_info->space_info);
1376 btrfs_mapping_init(&fs_info->mapping_tree);
1377 atomic_set(&fs_info->nr_async_submits, 0);
1378 atomic_set(&fs_info->nr_async_bios, 0);
1379 atomic_set(&fs_info->throttles, 0);
1380 atomic_set(&fs_info->throttle_gen, 0);
1382 fs_info->max_extent = (u64)-1;
1383 fs_info->max_inline = 8192 * 1024;
1384 setup_bdi(fs_info, &fs_info->bdi);
1385 fs_info->btree_inode = new_inode(sb);
1386 fs_info->btree_inode->i_ino = 1;
1387 fs_info->btree_inode->i_nlink = 1;
1388 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1390 INIT_LIST_HEAD(&fs_info->ordered_extents);
1391 spin_lock_init(&fs_info->ordered_extent_lock);
1393 sb->s_blocksize = 4096;
1394 sb->s_blocksize_bits = blksize_bits(4096);
1397 * we set the i_size on the btree inode to the max possible int.
1398 * the real end of the address space is determined by all of
1399 * the devices in the system
1401 fs_info->btree_inode->i_size = OFFSET_MAX;
1402 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1403 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1405 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1406 fs_info->btree_inode->i_mapping,
1408 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1411 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1413 extent_io_tree_init(&fs_info->free_space_cache,
1414 fs_info->btree_inode->i_mapping, GFP_NOFS);
1415 extent_io_tree_init(&fs_info->block_group_cache,
1416 fs_info->btree_inode->i_mapping, GFP_NOFS);
1417 extent_io_tree_init(&fs_info->pinned_extents,
1418 fs_info->btree_inode->i_mapping, GFP_NOFS);
1419 extent_io_tree_init(&fs_info->pending_del,
1420 fs_info->btree_inode->i_mapping, GFP_NOFS);
1421 extent_io_tree_init(&fs_info->extent_ins,
1422 fs_info->btree_inode->i_mapping, GFP_NOFS);
1423 fs_info->do_barriers = 1;
1425 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1426 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1427 sizeof(struct btrfs_key));
1428 insert_inode_hash(fs_info->btree_inode);
1430 mutex_init(&fs_info->trans_mutex);
1431 mutex_init(&fs_info->tree_log_mutex);
1432 mutex_init(&fs_info->drop_mutex);
1433 mutex_init(&fs_info->alloc_mutex);
1434 mutex_init(&fs_info->chunk_mutex);
1435 mutex_init(&fs_info->transaction_kthread_mutex);
1436 mutex_init(&fs_info->cleaner_mutex);
1437 mutex_init(&fs_info->volume_mutex);
1438 init_waitqueue_head(&fs_info->transaction_throttle);
1439 init_waitqueue_head(&fs_info->transaction_wait);
1440 init_waitqueue_head(&fs_info->async_submit_wait);
1441 init_waitqueue_head(&fs_info->tree_log_wait);
1442 atomic_set(&fs_info->tree_log_commit, 0);
1443 atomic_set(&fs_info->tree_log_writers, 0);
1444 fs_info->tree_log_transid = 0;
1447 ret = add_hasher(fs_info, "crc32c");
1449 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1454 __setup_root(4096, 4096, 4096, 4096, tree_root,
1455 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1458 bh = __bread(fs_devices->latest_bdev,
1459 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1463 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1466 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1468 disk_super = &fs_info->super_copy;
1469 if (!btrfs_super_root(disk_super))
1470 goto fail_sb_buffer;
1472 err = btrfs_parse_options(tree_root, options);
1474 goto fail_sb_buffer;
1477 * we need to start all the end_io workers up front because the
1478 * queue work function gets called at interrupt time, and so it
1479 * cannot dynamically grow.
1481 btrfs_init_workers(&fs_info->workers, "worker",
1482 fs_info->thread_pool_size);
1483 btrfs_init_workers(&fs_info->submit_workers, "submit",
1484 min_t(u64, fs_devices->num_devices,
1485 fs_info->thread_pool_size));
1487 /* a higher idle thresh on the submit workers makes it much more
1488 * likely that bios will be send down in a sane order to the
1491 fs_info->submit_workers.idle_thresh = 64;
1493 /* fs_info->workers is responsible for checksumming file data
1494 * blocks and metadata. Using a larger idle thresh allows each
1495 * worker thread to operate on things in roughly the order they
1496 * were sent by the writeback daemons, improving overall locality
1497 * of the IO going down the pipe.
1499 fs_info->workers.idle_thresh = 128;
1501 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1502 btrfs_init_workers(&fs_info->endio_workers, "endio",
1503 fs_info->thread_pool_size);
1504 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1505 fs_info->thread_pool_size);
1508 * endios are largely parallel and should have a very
1511 fs_info->endio_workers.idle_thresh = 4;
1512 fs_info->endio_write_workers.idle_thresh = 64;
1514 btrfs_start_workers(&fs_info->workers, 1);
1515 btrfs_start_workers(&fs_info->submit_workers, 1);
1516 btrfs_start_workers(&fs_info->fixup_workers, 1);
1517 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1518 btrfs_start_workers(&fs_info->endio_write_workers,
1519 fs_info->thread_pool_size);
1522 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1523 printk("Btrfs: wanted %llu devices, but found %llu\n",
1524 (unsigned long long)btrfs_super_num_devices(disk_super),
1525 (unsigned long long)fs_devices->open_devices);
1526 if (btrfs_test_opt(tree_root, DEGRADED))
1527 printk("continuing in degraded mode\n");
1529 goto fail_sb_buffer;
1533 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1535 nodesize = btrfs_super_nodesize(disk_super);
1536 leafsize = btrfs_super_leafsize(disk_super);
1537 sectorsize = btrfs_super_sectorsize(disk_super);
1538 stripesize = btrfs_super_stripesize(disk_super);
1539 tree_root->nodesize = nodesize;
1540 tree_root->leafsize = leafsize;
1541 tree_root->sectorsize = sectorsize;
1542 tree_root->stripesize = stripesize;
1544 sb->s_blocksize = sectorsize;
1545 sb->s_blocksize_bits = blksize_bits(sectorsize);
1547 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1548 sizeof(disk_super->magic))) {
1549 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1550 goto fail_sb_buffer;
1553 mutex_lock(&fs_info->chunk_mutex);
1554 ret = btrfs_read_sys_array(tree_root);
1555 mutex_unlock(&fs_info->chunk_mutex);
1557 printk("btrfs: failed to read the system array on %s\n",
1559 goto fail_sys_array;
1562 blocksize = btrfs_level_size(tree_root,
1563 btrfs_super_chunk_root_level(disk_super));
1565 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1566 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1568 chunk_root->node = read_tree_block(chunk_root,
1569 btrfs_super_chunk_root(disk_super),
1571 BUG_ON(!chunk_root->node);
1573 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1574 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1577 mutex_lock(&fs_info->chunk_mutex);
1578 ret = btrfs_read_chunk_tree(chunk_root);
1579 mutex_unlock(&fs_info->chunk_mutex);
1582 btrfs_close_extra_devices(fs_devices);
1584 blocksize = btrfs_level_size(tree_root,
1585 btrfs_super_root_level(disk_super));
1588 tree_root->node = read_tree_block(tree_root,
1589 btrfs_super_root(disk_super),
1591 if (!tree_root->node)
1592 goto fail_sb_buffer;
1595 ret = find_and_setup_root(tree_root, fs_info,
1596 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1598 goto fail_tree_root;
1599 extent_root->track_dirty = 1;
1601 ret = find_and_setup_root(tree_root, fs_info,
1602 BTRFS_DEV_TREE_OBJECTID, dev_root);
1603 dev_root->track_dirty = 1;
1606 goto fail_extent_root;
1608 btrfs_read_block_groups(extent_root);
1610 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1611 fs_info->data_alloc_profile = (u64)-1;
1612 fs_info->metadata_alloc_profile = (u64)-1;
1613 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1614 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1616 if (!fs_info->cleaner_kthread)
1617 goto fail_extent_root;
1619 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1621 "btrfs-transaction");
1622 if (!fs_info->transaction_kthread)
1625 if (btrfs_super_log_root(disk_super) != 0) {
1627 u64 bytenr = btrfs_super_log_root(disk_super);
1630 btrfs_level_size(tree_root,
1631 btrfs_super_log_root_level(disk_super));
1633 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1636 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1637 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1639 log_tree_root->node = read_tree_block(tree_root, bytenr,
1641 ret = btrfs_recover_log_trees(log_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 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1669 bdi_destroy(&fs_info->bdi);
1672 return ERR_PTR(err);
1675 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1677 char b[BDEVNAME_SIZE];
1680 set_buffer_uptodate(bh);
1682 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1683 printk(KERN_WARNING "lost page write due to "
1684 "I/O error on %s\n",
1685 bdevname(bh->b_bdev, b));
1687 /* note, we dont' set_buffer_write_io_error because we have
1688 * our own ways of dealing with the IO errors
1690 clear_buffer_uptodate(bh);
1696 int write_all_supers(struct btrfs_root *root)
1698 struct list_head *cur;
1699 struct list_head *head = &root->fs_info->fs_devices->devices;
1700 struct btrfs_device *dev;
1701 struct btrfs_super_block *sb;
1702 struct btrfs_dev_item *dev_item;
1703 struct buffer_head *bh;
1707 int total_errors = 0;
1711 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1712 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1714 sb = &root->fs_info->super_for_commit;
1715 dev_item = &sb->dev_item;
1716 list_for_each(cur, head) {
1717 dev = list_entry(cur, struct btrfs_device, dev_list);
1722 if (!dev->in_fs_metadata)
1725 btrfs_set_stack_device_type(dev_item, dev->type);
1726 btrfs_set_stack_device_id(dev_item, dev->devid);
1727 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1728 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1729 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1730 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1731 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1732 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1733 flags = btrfs_super_flags(sb);
1734 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1738 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1739 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1740 btrfs_csum_final(crc, sb->csum);
1742 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1743 BTRFS_SUPER_INFO_SIZE);
1745 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1746 dev->pending_io = bh;
1749 set_buffer_uptodate(bh);
1751 bh->b_end_io = btrfs_end_buffer_write_sync;
1753 if (do_barriers && dev->barriers) {
1754 ret = submit_bh(WRITE_BARRIER, bh);
1755 if (ret == -EOPNOTSUPP) {
1756 printk("btrfs: disabling barriers on dev %s\n",
1758 set_buffer_uptodate(bh);
1762 ret = submit_bh(WRITE, bh);
1765 ret = submit_bh(WRITE, bh);
1770 if (total_errors > max_errors) {
1771 printk("btrfs: %d errors while writing supers\n", total_errors);
1776 list_for_each(cur, head) {
1777 dev = list_entry(cur, struct btrfs_device, dev_list);
1780 if (!dev->in_fs_metadata)
1783 BUG_ON(!dev->pending_io);
1784 bh = dev->pending_io;
1786 if (!buffer_uptodate(dev->pending_io)) {
1787 if (do_barriers && dev->barriers) {
1788 printk("btrfs: disabling barriers on dev %s\n",
1790 set_buffer_uptodate(bh);
1794 ret = submit_bh(WRITE, bh);
1797 if (!buffer_uptodate(bh))
1804 dev->pending_io = NULL;
1807 if (total_errors > max_errors) {
1808 printk("btrfs: %d errors while writing supers\n", total_errors);
1814 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1819 ret = write_all_supers(root);
1823 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1825 radix_tree_delete(&fs_info->fs_roots_radix,
1826 (unsigned long)root->root_key.objectid);
1828 btrfs_sysfs_del_root(root);
1832 free_extent_buffer(root->node);
1833 if (root->commit_root)
1834 free_extent_buffer(root->commit_root);
1841 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1844 struct btrfs_root *gang[8];
1848 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1853 for (i = 0; i < ret; i++)
1854 btrfs_free_fs_root(fs_info, gang[i]);
1859 int close_ctree(struct btrfs_root *root)
1862 struct btrfs_trans_handle *trans;
1863 struct btrfs_fs_info *fs_info = root->fs_info;
1865 fs_info->closing = 1;
1868 kthread_stop(root->fs_info->transaction_kthread);
1869 kthread_stop(root->fs_info->cleaner_kthread);
1871 btrfs_clean_old_snapshots(root);
1872 trans = btrfs_start_transaction(root, 1);
1873 ret = btrfs_commit_transaction(trans, root);
1874 /* run commit again to drop the original snapshot */
1875 trans = btrfs_start_transaction(root, 1);
1876 btrfs_commit_transaction(trans, root);
1877 ret = btrfs_write_and_wait_transaction(NULL, root);
1880 write_ctree_super(NULL, root);
1882 if (fs_info->delalloc_bytes) {
1883 printk("btrfs: at unmount delalloc count %Lu\n",
1884 fs_info->delalloc_bytes);
1886 if (fs_info->total_ref_cache_size) {
1887 printk("btrfs: at umount reference cache size %Lu\n",
1888 fs_info->total_ref_cache_size);
1891 if (fs_info->extent_root->node)
1892 free_extent_buffer(fs_info->extent_root->node);
1894 if (fs_info->tree_root->node)
1895 free_extent_buffer(fs_info->tree_root->node);
1897 if (root->fs_info->chunk_root->node);
1898 free_extent_buffer(root->fs_info->chunk_root->node);
1900 if (root->fs_info->dev_root->node);
1901 free_extent_buffer(root->fs_info->dev_root->node);
1903 btrfs_free_block_groups(root->fs_info);
1904 fs_info->closing = 2;
1905 del_fs_roots(fs_info);
1907 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1909 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1911 btrfs_stop_workers(&fs_info->fixup_workers);
1912 btrfs_stop_workers(&fs_info->workers);
1913 btrfs_stop_workers(&fs_info->endio_workers);
1914 btrfs_stop_workers(&fs_info->endio_write_workers);
1915 btrfs_stop_workers(&fs_info->submit_workers);
1917 iput(fs_info->btree_inode);
1919 while(!list_empty(&fs_info->hashers)) {
1920 struct btrfs_hasher *hasher;
1921 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1923 list_del(&hasher->hashers);
1924 crypto_free_hash(&fs_info->hash_tfm);
1928 btrfs_close_devices(fs_info->fs_devices);
1929 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1931 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1932 bdi_destroy(&fs_info->bdi);
1935 kfree(fs_info->extent_root);
1936 kfree(fs_info->tree_root);
1937 kfree(fs_info->chunk_root);
1938 kfree(fs_info->dev_root);
1942 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1945 struct inode *btree_inode = buf->first_page->mapping->host;
1947 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1951 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1956 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1958 struct inode *btree_inode = buf->first_page->mapping->host;
1959 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1963 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1965 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1966 u64 transid = btrfs_header_generation(buf);
1967 struct inode *btree_inode = root->fs_info->btree_inode;
1969 WARN_ON(!btrfs_tree_locked(buf));
1970 if (transid != root->fs_info->generation) {
1971 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1972 (unsigned long long)buf->start,
1973 transid, root->fs_info->generation);
1976 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1979 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1982 * looks as though older kernels can get into trouble with
1983 * this code, they end up stuck in balance_dirty_pages forever
1985 struct extent_io_tree *tree;
1988 unsigned long thresh = 96 * 1024 * 1024;
1989 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1991 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
1994 num_dirty = count_range_bits(tree, &start, (u64)-1,
1995 thresh, EXTENT_DIRTY);
1996 if (num_dirty > thresh) {
1997 balance_dirty_pages_ratelimited_nr(
1998 root->fs_info->btree_inode->i_mapping, 1);
2003 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2005 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2007 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2009 buf->flags |= EXTENT_UPTODATE;
2014 int btree_lock_page_hook(struct page *page)
2016 struct inode *inode = page->mapping->host;
2017 struct btrfs_root *root = BTRFS_I(inode)->root;
2018 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2019 struct extent_buffer *eb;
2021 u64 bytenr = page_offset(page);
2023 if (page->private == EXTENT_PAGE_PRIVATE)
2026 len = page->private >> 2;
2027 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2031 btrfs_tree_lock(eb);
2032 spin_lock(&root->fs_info->hash_lock);
2033 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2034 spin_unlock(&root->fs_info->hash_lock);
2035 btrfs_tree_unlock(eb);
2036 free_extent_buffer(eb);
2042 static struct extent_io_ops btree_extent_io_ops = {
2043 .write_cache_pages_lock_hook = btree_lock_page_hook,
2044 .writepage_io_hook = btree_writepage_io_hook,
2045 .readpage_end_io_hook = btree_readpage_end_io_hook,
2046 .submit_bio_hook = btree_submit_bio_hook,
2047 /* note we're sharing with inode.c for the merge bio hook */
2048 .merge_bio_hook = btrfs_merge_bio_hook,