2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/version.h>
21 #include <linux/blkdev.h>
22 #include <linux/scatterlist.h>
23 #include <linux/swap.h>
24 #include <linux/radix-tree.h>
25 #include <linux/writeback.h>
26 #include <linux/buffer_head.h> // for block_sync_page
27 #include <linux/workqueue.h>
28 #include <linux/kthread.h>
29 # include <linux/freezer.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
36 #include "print-tree.h"
37 #include "async-thread.h"
39 #include "ref-cache.h"
43 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
45 if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
46 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
47 (unsigned long long)extent_buffer_blocknr(buf),
48 (unsigned long long)btrfs_header_blocknr(buf));
55 static struct extent_io_ops btree_extent_io_ops;
56 static void end_workqueue_fn(struct btrfs_work *work);
59 * end_io_wq structs are used to do processing in task context when an IO is
60 * complete. This is used during reads to verify checksums, and it is used
61 * by writes to insert metadata for new file extents after IO is complete.
67 struct btrfs_fs_info *info;
70 struct list_head list;
71 struct btrfs_work work;
75 * async submit bios are used to offload expensive checksumming
76 * onto the worker threads. They checksum file and metadata bios
77 * just before they are sent down the IO stack.
79 struct async_submit_bio {
82 struct list_head list;
83 extent_submit_bio_hook_t *submit_bio_start;
84 extent_submit_bio_hook_t *submit_bio_done;
87 unsigned long bio_flags;
88 struct btrfs_work work;
92 * extents on the btree inode are pretty simple, there's one extent
93 * that covers the entire device
95 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
96 size_t page_offset, u64 start, u64 len,
99 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
100 struct extent_map *em;
103 spin_lock(&em_tree->lock);
104 em = lookup_extent_mapping(em_tree, start, len);
107 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
108 spin_unlock(&em_tree->lock);
111 spin_unlock(&em_tree->lock);
113 em = alloc_extent_map(GFP_NOFS);
115 em = ERR_PTR(-ENOMEM);
120 em->block_len = (u64)-1;
122 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
124 spin_lock(&em_tree->lock);
125 ret = add_extent_mapping(em_tree, em);
126 if (ret == -EEXIST) {
127 u64 failed_start = em->start;
128 u64 failed_len = em->len;
130 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
131 em->start, em->len, em->block_start);
133 em = lookup_extent_mapping(em_tree, start, len);
135 printk("after failing, found %Lu %Lu %Lu\n",
136 em->start, em->len, em->block_start);
139 em = lookup_extent_mapping(em_tree, failed_start,
142 printk("double failure lookup gives us "
143 "%Lu %Lu -> %Lu\n", em->start,
144 em->len, em->block_start);
153 spin_unlock(&em_tree->lock);
161 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
163 return btrfs_crc32c(seed, data, len);
166 void btrfs_csum_final(u32 crc, char *result)
168 *(__le32 *)result = ~cpu_to_le32(crc);
172 * compute the csum for a btree block, and either verify it or write it
173 * into the csum field of the block.
175 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
178 char result[BTRFS_CRC32_SIZE];
180 unsigned long cur_len;
181 unsigned long offset = BTRFS_CSUM_SIZE;
182 char *map_token = NULL;
184 unsigned long map_start;
185 unsigned long map_len;
189 len = buf->len - offset;
191 err = map_private_extent_buffer(buf, offset, 32,
193 &map_start, &map_len, KM_USER0);
195 printk("failed to map extent buffer! %lu\n",
199 cur_len = min(len, map_len - (offset - map_start));
200 crc = btrfs_csum_data(root, kaddr + offset - map_start,
204 unmap_extent_buffer(buf, map_token, KM_USER0);
206 btrfs_csum_final(crc, result);
209 /* FIXME, this is not good */
210 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
213 memcpy(&found, result, BTRFS_CRC32_SIZE);
215 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
216 printk("btrfs: %s checksum verify failed on %llu "
217 "wanted %X found %X level %d\n",
218 root->fs_info->sb->s_id,
219 buf->start, val, found, btrfs_header_level(buf));
223 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
229 * we can't consider a given block up to date unless the transid of the
230 * block matches the transid in the parent node's pointer. This is how we
231 * detect blocks that either didn't get written at all or got written
232 * in the wrong place.
234 static int verify_parent_transid(struct extent_io_tree *io_tree,
235 struct extent_buffer *eb, u64 parent_transid)
239 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
242 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
243 if (extent_buffer_uptodate(io_tree, eb) &&
244 btrfs_header_generation(eb) == parent_transid) {
248 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
249 (unsigned long long)eb->start,
250 (unsigned long long)parent_transid,
251 (unsigned long long)btrfs_header_generation(eb));
253 clear_extent_buffer_uptodate(io_tree, eb);
255 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
261 * helper to read a given tree block, doing retries as required when
262 * the checksums don't match and we have alternate mirrors to try.
264 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
265 struct extent_buffer *eb,
266 u64 start, u64 parent_transid)
268 struct extent_io_tree *io_tree;
273 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
275 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
276 btree_get_extent, mirror_num);
278 !verify_parent_transid(io_tree, eb, parent_transid))
280 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
281 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
287 if (mirror_num > num_copies)
294 * checksum a dirty tree block before IO. This has extra checks to make
295 * sure we only fill in the checksum field in the first page of a multi-page block
297 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
299 struct extent_io_tree *tree;
300 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
304 struct extent_buffer *eb;
307 tree = &BTRFS_I(page->mapping->host)->io_tree;
309 if (page->private == EXTENT_PAGE_PRIVATE)
313 len = page->private >> 2;
317 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
318 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
319 btrfs_header_generation(eb));
321 found_start = btrfs_header_bytenr(eb);
322 if (found_start != start) {
323 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
324 start, found_start, len);
328 if (eb->first_page != page) {
329 printk("bad first page %lu %lu\n", eb->first_page->index,
334 if (!PageUptodate(page)) {
335 printk("csum not up to date page %lu\n", page->index);
339 found_level = btrfs_header_level(eb);
341 csum_tree_block(root, eb, 0);
343 free_extent_buffer(eb);
348 static int check_tree_block_fsid(struct btrfs_root *root,
349 struct extent_buffer *eb)
351 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
352 u8 fsid[BTRFS_UUID_SIZE];
355 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
358 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
362 fs_devices = fs_devices->seed;
367 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
368 struct extent_state *state)
370 struct extent_io_tree *tree;
374 struct extent_buffer *eb;
375 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
378 tree = &BTRFS_I(page->mapping->host)->io_tree;
379 if (page->private == EXTENT_PAGE_PRIVATE)
383 len = page->private >> 2;
387 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
389 found_start = btrfs_header_bytenr(eb);
390 if (found_start != start) {
391 printk("bad tree block start %llu %llu\n",
392 (unsigned long long)found_start,
393 (unsigned long long)eb->start);
397 if (eb->first_page != page) {
398 printk("bad first page %lu %lu\n", eb->first_page->index,
404 if (check_tree_block_fsid(root, eb)) {
405 printk("bad fsid on block %Lu\n", eb->start);
409 found_level = btrfs_header_level(eb);
411 ret = csum_tree_block(root, eb, 1);
415 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
416 end = eb->start + end - 1;
418 free_extent_buffer(eb);
423 static void end_workqueue_bio(struct bio *bio, int err)
425 struct end_io_wq *end_io_wq = bio->bi_private;
426 struct btrfs_fs_info *fs_info;
428 fs_info = end_io_wq->info;
429 end_io_wq->error = err;
430 end_io_wq->work.func = end_workqueue_fn;
431 end_io_wq->work.flags = 0;
432 if (bio->bi_rw & (1 << BIO_RW))
433 btrfs_queue_worker(&fs_info->endio_write_workers,
436 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
439 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
442 struct end_io_wq *end_io_wq;
443 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
447 end_io_wq->private = bio->bi_private;
448 end_io_wq->end_io = bio->bi_end_io;
449 end_io_wq->info = info;
450 end_io_wq->error = 0;
451 end_io_wq->bio = bio;
452 end_io_wq->metadata = metadata;
454 bio->bi_private = end_io_wq;
455 bio->bi_end_io = end_workqueue_bio;
459 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
461 unsigned long limit = min_t(unsigned long,
462 info->workers.max_workers,
463 info->fs_devices->open_devices);
467 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
469 return atomic_read(&info->nr_async_bios) >
470 btrfs_async_submit_limit(info);
473 static void run_one_async_start(struct btrfs_work *work)
475 struct btrfs_fs_info *fs_info;
476 struct async_submit_bio *async;
478 async = container_of(work, struct async_submit_bio, work);
479 fs_info = BTRFS_I(async->inode)->root->fs_info;
480 async->submit_bio_start(async->inode, async->rw, async->bio,
481 async->mirror_num, async->bio_flags);
484 static void run_one_async_done(struct btrfs_work *work)
486 struct btrfs_fs_info *fs_info;
487 struct async_submit_bio *async;
490 async = container_of(work, struct async_submit_bio, work);
491 fs_info = BTRFS_I(async->inode)->root->fs_info;
493 limit = btrfs_async_submit_limit(fs_info);
494 limit = limit * 2 / 3;
496 atomic_dec(&fs_info->nr_async_submits);
498 if (atomic_read(&fs_info->nr_async_submits) < limit &&
499 waitqueue_active(&fs_info->async_submit_wait))
500 wake_up(&fs_info->async_submit_wait);
502 async->submit_bio_done(async->inode, async->rw, async->bio,
503 async->mirror_num, async->bio_flags);
506 static void run_one_async_free(struct btrfs_work *work)
508 struct async_submit_bio *async;
510 async = container_of(work, struct async_submit_bio, work);
514 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
515 int rw, struct bio *bio, int mirror_num,
516 unsigned long bio_flags,
517 extent_submit_bio_hook_t *submit_bio_start,
518 extent_submit_bio_hook_t *submit_bio_done)
520 struct async_submit_bio *async;
522 async = kmalloc(sizeof(*async), GFP_NOFS);
526 async->inode = inode;
529 async->mirror_num = mirror_num;
530 async->submit_bio_start = submit_bio_start;
531 async->submit_bio_done = submit_bio_done;
533 async->work.func = run_one_async_start;
534 async->work.ordered_func = run_one_async_done;
535 async->work.ordered_free = run_one_async_free;
537 async->work.flags = 0;
538 async->bio_flags = bio_flags;
540 atomic_inc(&fs_info->nr_async_submits);
541 btrfs_queue_worker(&fs_info->workers, &async->work);
543 int limit = btrfs_async_submit_limit(fs_info);
544 if (atomic_read(&fs_info->nr_async_submits) > limit) {
545 wait_event_timeout(fs_info->async_submit_wait,
546 (atomic_read(&fs_info->nr_async_submits) < limit),
549 wait_event_timeout(fs_info->async_submit_wait,
550 (atomic_read(&fs_info->nr_async_bios) < limit),
554 while(atomic_read(&fs_info->async_submit_draining) &&
555 atomic_read(&fs_info->nr_async_submits)) {
556 wait_event(fs_info->async_submit_wait,
557 (atomic_read(&fs_info->nr_async_submits) == 0));
563 static int btree_csum_one_bio(struct bio *bio)
565 struct bio_vec *bvec = bio->bi_io_vec;
567 struct btrfs_root *root;
569 WARN_ON(bio->bi_vcnt <= 0);
570 while(bio_index < bio->bi_vcnt) {
571 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
572 csum_dirty_buffer(root, bvec->bv_page);
579 static int __btree_submit_bio_start(struct inode *inode, int rw,
580 struct bio *bio, int mirror_num,
581 unsigned long bio_flags)
584 * when we're called for a write, we're already in the async
585 * submission context. Just jump into btrfs_map_bio
587 btree_csum_one_bio(bio);
591 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
592 int mirror_num, unsigned long bio_flags)
595 * when we're called for a write, we're already in the async
596 * submission context. Just jump into btrfs_map_bio
598 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
601 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
602 int mirror_num, unsigned long bio_flags)
605 * kthread helpers are used to submit writes so that checksumming
606 * can happen in parallel across all CPUs
608 if (!(rw & (1 << BIO_RW))) {
611 * called for a read, do the setup so that checksum validation
612 * can happen in the async kernel threads
614 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
618 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
621 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
622 inode, rw, bio, mirror_num, 0,
623 __btree_submit_bio_start,
624 __btree_submit_bio_done);
627 static int btree_writepage(struct page *page, struct writeback_control *wbc)
629 struct extent_io_tree *tree;
630 tree = &BTRFS_I(page->mapping->host)->io_tree;
632 if (current->flags & PF_MEMALLOC) {
633 redirty_page_for_writepage(wbc, page);
637 return extent_write_full_page(tree, page, btree_get_extent, wbc);
640 static int btree_writepages(struct address_space *mapping,
641 struct writeback_control *wbc)
643 struct extent_io_tree *tree;
644 tree = &BTRFS_I(mapping->host)->io_tree;
645 if (wbc->sync_mode == WB_SYNC_NONE) {
648 unsigned long thresh = 32 * 1024 * 1024;
650 if (wbc->for_kupdate)
653 num_dirty = count_range_bits(tree, &start, (u64)-1,
654 thresh, EXTENT_DIRTY);
655 if (num_dirty < thresh) {
659 return extent_writepages(tree, mapping, btree_get_extent, wbc);
662 int btree_readpage(struct file *file, struct page *page)
664 struct extent_io_tree *tree;
665 tree = &BTRFS_I(page->mapping->host)->io_tree;
666 return extent_read_full_page(tree, page, btree_get_extent);
669 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
671 struct extent_io_tree *tree;
672 struct extent_map_tree *map;
675 if (PageWriteback(page) || PageDirty(page))
678 tree = &BTRFS_I(page->mapping->host)->io_tree;
679 map = &BTRFS_I(page->mapping->host)->extent_tree;
681 ret = try_release_extent_state(map, tree, page, gfp_flags);
686 ret = try_release_extent_buffer(tree, page);
688 ClearPagePrivate(page);
689 set_page_private(page, 0);
690 page_cache_release(page);
696 static void btree_invalidatepage(struct page *page, unsigned long offset)
698 struct extent_io_tree *tree;
699 tree = &BTRFS_I(page->mapping->host)->io_tree;
700 extent_invalidatepage(tree, page, offset);
701 btree_releasepage(page, GFP_NOFS);
702 if (PagePrivate(page)) {
703 printk("warning page private not zero on page %Lu\n",
705 ClearPagePrivate(page);
706 set_page_private(page, 0);
707 page_cache_release(page);
712 static int btree_writepage(struct page *page, struct writeback_control *wbc)
714 struct buffer_head *bh;
715 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
716 struct buffer_head *head;
717 if (!page_has_buffers(page)) {
718 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
719 (1 << BH_Dirty)|(1 << BH_Uptodate));
721 head = page_buffers(page);
724 if (buffer_dirty(bh))
725 csum_tree_block(root, bh, 0);
726 bh = bh->b_this_page;
727 } while (bh != head);
728 return block_write_full_page(page, btree_get_block, wbc);
732 static struct address_space_operations btree_aops = {
733 .readpage = btree_readpage,
734 .writepage = btree_writepage,
735 .writepages = btree_writepages,
736 .releasepage = btree_releasepage,
737 .invalidatepage = btree_invalidatepage,
738 .sync_page = block_sync_page,
741 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
744 struct extent_buffer *buf = NULL;
745 struct inode *btree_inode = root->fs_info->btree_inode;
748 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
751 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
752 buf, 0, 0, btree_get_extent, 0);
753 free_extent_buffer(buf);
757 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
758 u64 bytenr, u32 blocksize)
760 struct inode *btree_inode = root->fs_info->btree_inode;
761 struct extent_buffer *eb;
762 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
763 bytenr, blocksize, GFP_NOFS);
767 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
768 u64 bytenr, u32 blocksize)
770 struct inode *btree_inode = root->fs_info->btree_inode;
771 struct extent_buffer *eb;
773 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
774 bytenr, blocksize, NULL, GFP_NOFS);
779 int btrfs_write_tree_block(struct extent_buffer *buf)
781 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
782 buf->start + buf->len - 1, WB_SYNC_ALL);
785 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
787 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
788 buf->start, buf->start + buf->len -1);
791 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
792 u32 blocksize, u64 parent_transid)
794 struct extent_buffer *buf = NULL;
795 struct inode *btree_inode = root->fs_info->btree_inode;
796 struct extent_io_tree *io_tree;
799 io_tree = &BTRFS_I(btree_inode)->io_tree;
801 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
805 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
808 buf->flags |= EXTENT_UPTODATE;
816 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
817 struct extent_buffer *buf)
819 struct inode *btree_inode = root->fs_info->btree_inode;
820 if (btrfs_header_generation(buf) ==
821 root->fs_info->running_transaction->transid) {
822 WARN_ON(!btrfs_tree_locked(buf));
823 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
829 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
830 u32 stripesize, struct btrfs_root *root,
831 struct btrfs_fs_info *fs_info,
835 root->commit_root = NULL;
836 root->ref_tree = NULL;
837 root->sectorsize = sectorsize;
838 root->nodesize = nodesize;
839 root->leafsize = leafsize;
840 root->stripesize = stripesize;
842 root->track_dirty = 0;
844 root->fs_info = fs_info;
845 root->objectid = objectid;
846 root->last_trans = 0;
847 root->highest_inode = 0;
848 root->last_inode_alloc = 0;
852 INIT_LIST_HEAD(&root->dirty_list);
853 INIT_LIST_HEAD(&root->orphan_list);
854 INIT_LIST_HEAD(&root->dead_list);
855 spin_lock_init(&root->node_lock);
856 spin_lock_init(&root->list_lock);
857 mutex_init(&root->objectid_mutex);
858 mutex_init(&root->log_mutex);
859 extent_io_tree_init(&root->dirty_log_pages,
860 fs_info->btree_inode->i_mapping, GFP_NOFS);
862 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
863 root->ref_tree = &root->ref_tree_struct;
865 memset(&root->root_key, 0, sizeof(root->root_key));
866 memset(&root->root_item, 0, sizeof(root->root_item));
867 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
868 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
869 root->defrag_trans_start = fs_info->generation;
870 init_completion(&root->kobj_unregister);
871 root->defrag_running = 0;
872 root->defrag_level = 0;
873 root->root_key.objectid = objectid;
874 root->anon_super.s_root = NULL;
875 root->anon_super.s_dev = 0;
876 INIT_LIST_HEAD(&root->anon_super.s_list);
877 INIT_LIST_HEAD(&root->anon_super.s_instances);
878 init_rwsem(&root->anon_super.s_umount);
883 static int find_and_setup_root(struct btrfs_root *tree_root,
884 struct btrfs_fs_info *fs_info,
886 struct btrfs_root *root)
892 __setup_root(tree_root->nodesize, tree_root->leafsize,
893 tree_root->sectorsize, tree_root->stripesize,
894 root, fs_info, objectid);
895 ret = btrfs_find_last_root(tree_root, objectid,
896 &root->root_item, &root->root_key);
899 generation = btrfs_root_generation(&root->root_item);
900 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
901 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
902 blocksize, generation);
907 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
908 struct btrfs_fs_info *fs_info)
910 struct extent_buffer *eb;
911 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
920 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
921 0, &start, &end, EXTENT_DIRTY);
925 clear_extent_dirty(&log_root_tree->dirty_log_pages,
926 start, end, GFP_NOFS);
928 eb = fs_info->log_root_tree->node;
930 WARN_ON(btrfs_header_level(eb) != 0);
931 WARN_ON(btrfs_header_nritems(eb) != 0);
933 ret = btrfs_free_reserved_extent(fs_info->tree_root,
937 free_extent_buffer(eb);
938 kfree(fs_info->log_root_tree);
939 fs_info->log_root_tree = NULL;
943 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
944 struct btrfs_fs_info *fs_info)
946 struct btrfs_root *root;
947 struct btrfs_root *tree_root = fs_info->tree_root;
949 root = kzalloc(sizeof(*root), GFP_NOFS);
953 __setup_root(tree_root->nodesize, tree_root->leafsize,
954 tree_root->sectorsize, tree_root->stripesize,
955 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
957 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
958 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
959 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
962 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
963 0, BTRFS_TREE_LOG_OBJECTID,
964 trans->transid, 0, 0, 0);
966 btrfs_set_header_nritems(root->node, 0);
967 btrfs_set_header_level(root->node, 0);
968 btrfs_set_header_bytenr(root->node, root->node->start);
969 btrfs_set_header_generation(root->node, trans->transid);
970 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
972 write_extent_buffer(root->node, root->fs_info->fsid,
973 (unsigned long)btrfs_header_fsid(root->node),
975 btrfs_mark_buffer_dirty(root->node);
976 btrfs_tree_unlock(root->node);
977 fs_info->log_root_tree = root;
981 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
982 struct btrfs_key *location)
984 struct btrfs_root *root;
985 struct btrfs_fs_info *fs_info = tree_root->fs_info;
986 struct btrfs_path *path;
987 struct extent_buffer *l;
993 root = kzalloc(sizeof(*root), GFP_NOFS);
995 return ERR_PTR(-ENOMEM);
996 if (location->offset == (u64)-1) {
997 ret = find_and_setup_root(tree_root, fs_info,
998 location->objectid, root);
1001 return ERR_PTR(ret);
1006 __setup_root(tree_root->nodesize, tree_root->leafsize,
1007 tree_root->sectorsize, tree_root->stripesize,
1008 root, fs_info, location->objectid);
1010 path = btrfs_alloc_path();
1012 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1019 read_extent_buffer(l, &root->root_item,
1020 btrfs_item_ptr_offset(l, path->slots[0]),
1021 sizeof(root->root_item));
1022 memcpy(&root->root_key, location, sizeof(*location));
1025 btrfs_release_path(root, path);
1026 btrfs_free_path(path);
1029 return ERR_PTR(ret);
1031 generation = btrfs_root_generation(&root->root_item);
1032 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1033 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1034 blocksize, generation);
1035 BUG_ON(!root->node);
1037 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1039 ret = btrfs_find_highest_inode(root, &highest_inode);
1041 root->highest_inode = highest_inode;
1042 root->last_inode_alloc = highest_inode;
1048 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1051 struct btrfs_root *root;
1053 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1054 return fs_info->tree_root;
1055 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1056 return fs_info->extent_root;
1058 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1059 (unsigned long)root_objectid);
1063 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1064 struct btrfs_key *location)
1066 struct btrfs_root *root;
1069 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1070 return fs_info->tree_root;
1071 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1072 return fs_info->extent_root;
1073 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1074 return fs_info->chunk_root;
1075 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1076 return fs_info->dev_root;
1078 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1079 (unsigned long)location->objectid);
1083 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1087 set_anon_super(&root->anon_super, NULL);
1089 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1090 (unsigned long)root->root_key.objectid,
1093 free_extent_buffer(root->node);
1095 return ERR_PTR(ret);
1097 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1098 ret = btrfs_find_dead_roots(fs_info->tree_root,
1099 root->root_key.objectid, root);
1101 btrfs_orphan_cleanup(root);
1106 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1107 struct btrfs_key *location,
1108 const char *name, int namelen)
1110 struct btrfs_root *root;
1113 root = btrfs_read_fs_root_no_name(fs_info, location);
1120 ret = btrfs_set_root_name(root, name, namelen);
1122 free_extent_buffer(root->node);
1124 return ERR_PTR(ret);
1127 ret = btrfs_sysfs_add_root(root);
1129 free_extent_buffer(root->node);
1132 return ERR_PTR(ret);
1139 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1140 struct btrfs_hasher *hasher;
1142 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1145 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1146 if (!hasher->hash_tfm) {
1150 spin_lock(&info->hash_lock);
1151 list_add(&hasher->list, &info->hashers);
1152 spin_unlock(&info->hash_lock);
1157 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1159 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1161 struct list_head *cur;
1162 struct btrfs_device *device;
1163 struct backing_dev_info *bdi;
1165 if ((bdi_bits & (1 << BDI_write_congested)) &&
1166 btrfs_congested_async(info, 0))
1169 list_for_each(cur, &info->fs_devices->devices) {
1170 device = list_entry(cur, struct btrfs_device, dev_list);
1173 bdi = blk_get_backing_dev_info(device->bdev);
1174 if (bdi && bdi_congested(bdi, bdi_bits)) {
1183 * this unplugs every device on the box, and it is only used when page
1186 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1188 struct list_head *cur;
1189 struct btrfs_device *device;
1190 struct btrfs_fs_info *info;
1192 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1193 list_for_each(cur, &info->fs_devices->devices) {
1194 device = list_entry(cur, struct btrfs_device, dev_list);
1195 bdi = blk_get_backing_dev_info(device->bdev);
1196 if (bdi->unplug_io_fn) {
1197 bdi->unplug_io_fn(bdi, page);
1202 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1204 struct inode *inode;
1205 struct extent_map_tree *em_tree;
1206 struct extent_map *em;
1207 struct address_space *mapping;
1210 /* the generic O_DIRECT read code does this */
1212 __unplug_io_fn(bdi, page);
1217 * page->mapping may change at any time. Get a consistent copy
1218 * and use that for everything below
1221 mapping = page->mapping;
1225 inode = mapping->host;
1228 * don't do the expensive searching for a small number of
1231 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1232 __unplug_io_fn(bdi, page);
1236 offset = page_offset(page);
1238 em_tree = &BTRFS_I(inode)->extent_tree;
1239 spin_lock(&em_tree->lock);
1240 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1241 spin_unlock(&em_tree->lock);
1243 __unplug_io_fn(bdi, page);
1247 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1248 free_extent_map(em);
1249 __unplug_io_fn(bdi, page);
1252 offset = offset - em->start;
1253 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1254 em->block_start + offset, page);
1255 free_extent_map(em);
1258 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1261 bdi->ra_pages = default_backing_dev_info.ra_pages;
1263 bdi->capabilities = default_backing_dev_info.capabilities;
1264 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1265 bdi->unplug_io_data = info;
1266 bdi->congested_fn = btrfs_congested_fn;
1267 bdi->congested_data = info;
1271 static int bio_ready_for_csum(struct bio *bio)
1277 struct extent_io_tree *io_tree = NULL;
1278 struct btrfs_fs_info *info = NULL;
1279 struct bio_vec *bvec;
1283 bio_for_each_segment(bvec, bio, i) {
1284 page = bvec->bv_page;
1285 if (page->private == EXTENT_PAGE_PRIVATE) {
1286 length += bvec->bv_len;
1289 if (!page->private) {
1290 length += bvec->bv_len;
1293 length = bvec->bv_len;
1294 buf_len = page->private >> 2;
1295 start = page_offset(page) + bvec->bv_offset;
1296 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1297 info = BTRFS_I(page->mapping->host)->root->fs_info;
1299 /* are we fully contained in this bio? */
1300 if (buf_len <= length)
1303 ret = extent_range_uptodate(io_tree, start + length,
1304 start + buf_len - 1);
1311 * called by the kthread helper functions to finally call the bio end_io
1312 * functions. This is where read checksum verification actually happens
1314 static void end_workqueue_fn(struct btrfs_work *work)
1317 struct end_io_wq *end_io_wq;
1318 struct btrfs_fs_info *fs_info;
1321 end_io_wq = container_of(work, struct end_io_wq, work);
1322 bio = end_io_wq->bio;
1323 fs_info = end_io_wq->info;
1325 /* metadata bios are special because the whole tree block must
1326 * be checksummed at once. This makes sure the entire block is in
1327 * ram and up to date before trying to verify things. For
1328 * blocksize <= pagesize, it is basically a noop
1330 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1331 btrfs_queue_worker(&fs_info->endio_workers,
1335 error = end_io_wq->error;
1336 bio->bi_private = end_io_wq->private;
1337 bio->bi_end_io = end_io_wq->end_io;
1339 bio_endio(bio, error);
1342 static int cleaner_kthread(void *arg)
1344 struct btrfs_root *root = arg;
1348 if (root->fs_info->closing)
1351 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1352 mutex_lock(&root->fs_info->cleaner_mutex);
1353 btrfs_clean_old_snapshots(root);
1354 mutex_unlock(&root->fs_info->cleaner_mutex);
1356 if (freezing(current)) {
1360 if (root->fs_info->closing)
1362 set_current_state(TASK_INTERRUPTIBLE);
1364 __set_current_state(TASK_RUNNING);
1366 } while (!kthread_should_stop());
1370 static int transaction_kthread(void *arg)
1372 struct btrfs_root *root = arg;
1373 struct btrfs_trans_handle *trans;
1374 struct btrfs_transaction *cur;
1376 unsigned long delay;
1381 if (root->fs_info->closing)
1385 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1386 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1388 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1389 printk("btrfs: total reference cache size %Lu\n",
1390 root->fs_info->total_ref_cache_size);
1393 mutex_lock(&root->fs_info->trans_mutex);
1394 cur = root->fs_info->running_transaction;
1396 mutex_unlock(&root->fs_info->trans_mutex);
1400 now = get_seconds();
1401 if (now < cur->start_time || now - cur->start_time < 30) {
1402 mutex_unlock(&root->fs_info->trans_mutex);
1406 mutex_unlock(&root->fs_info->trans_mutex);
1407 trans = btrfs_start_transaction(root, 1);
1408 ret = btrfs_commit_transaction(trans, root);
1410 wake_up_process(root->fs_info->cleaner_kthread);
1411 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1413 if (freezing(current)) {
1416 if (root->fs_info->closing)
1418 set_current_state(TASK_INTERRUPTIBLE);
1419 schedule_timeout(delay);
1420 __set_current_state(TASK_RUNNING);
1422 } while (!kthread_should_stop());
1426 struct btrfs_root *open_ctree(struct super_block *sb,
1427 struct btrfs_fs_devices *fs_devices,
1436 struct btrfs_key location;
1437 struct buffer_head *bh;
1438 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1440 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1442 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1444 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1446 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1448 struct btrfs_root *log_tree_root;
1453 struct btrfs_super_block *disk_super;
1455 if (!extent_root || !tree_root || !fs_info ||
1456 !chunk_root || !dev_root) {
1460 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1461 INIT_LIST_HEAD(&fs_info->trans_list);
1462 INIT_LIST_HEAD(&fs_info->dead_roots);
1463 INIT_LIST_HEAD(&fs_info->hashers);
1464 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1465 spin_lock_init(&fs_info->hash_lock);
1466 spin_lock_init(&fs_info->delalloc_lock);
1467 spin_lock_init(&fs_info->new_trans_lock);
1468 spin_lock_init(&fs_info->ref_cache_lock);
1470 init_completion(&fs_info->kobj_unregister);
1471 fs_info->tree_root = tree_root;
1472 fs_info->extent_root = extent_root;
1473 fs_info->chunk_root = chunk_root;
1474 fs_info->dev_root = dev_root;
1475 fs_info->fs_devices = fs_devices;
1476 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1477 INIT_LIST_HEAD(&fs_info->space_info);
1478 btrfs_mapping_init(&fs_info->mapping_tree);
1479 atomic_set(&fs_info->nr_async_submits, 0);
1480 atomic_set(&fs_info->async_delalloc_pages, 0);
1481 atomic_set(&fs_info->async_submit_draining, 0);
1482 atomic_set(&fs_info->nr_async_bios, 0);
1483 atomic_set(&fs_info->throttles, 0);
1484 atomic_set(&fs_info->throttle_gen, 0);
1486 fs_info->max_extent = (u64)-1;
1487 fs_info->max_inline = 8192 * 1024;
1488 setup_bdi(fs_info, &fs_info->bdi);
1489 fs_info->btree_inode = new_inode(sb);
1490 fs_info->btree_inode->i_ino = 1;
1491 fs_info->btree_inode->i_nlink = 1;
1493 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1495 INIT_LIST_HEAD(&fs_info->ordered_extents);
1496 spin_lock_init(&fs_info->ordered_extent_lock);
1498 sb->s_blocksize = 4096;
1499 sb->s_blocksize_bits = blksize_bits(4096);
1502 * we set the i_size on the btree inode to the max possible int.
1503 * the real end of the address space is determined by all of
1504 * the devices in the system
1506 fs_info->btree_inode->i_size = OFFSET_MAX;
1507 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1508 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1510 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1511 fs_info->btree_inode->i_mapping,
1513 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1516 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1518 spin_lock_init(&fs_info->block_group_cache_lock);
1519 fs_info->block_group_cache_tree.rb_node = NULL;
1521 extent_io_tree_init(&fs_info->pinned_extents,
1522 fs_info->btree_inode->i_mapping, GFP_NOFS);
1523 extent_io_tree_init(&fs_info->pending_del,
1524 fs_info->btree_inode->i_mapping, GFP_NOFS);
1525 extent_io_tree_init(&fs_info->extent_ins,
1526 fs_info->btree_inode->i_mapping, GFP_NOFS);
1527 fs_info->do_barriers = 1;
1529 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1530 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1531 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1533 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1534 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1535 sizeof(struct btrfs_key));
1536 insert_inode_hash(fs_info->btree_inode);
1538 mutex_init(&fs_info->trans_mutex);
1539 mutex_init(&fs_info->tree_log_mutex);
1540 mutex_init(&fs_info->drop_mutex);
1541 mutex_init(&fs_info->extent_ins_mutex);
1542 mutex_init(&fs_info->pinned_mutex);
1543 mutex_init(&fs_info->chunk_mutex);
1544 mutex_init(&fs_info->transaction_kthread_mutex);
1545 mutex_init(&fs_info->cleaner_mutex);
1546 mutex_init(&fs_info->volume_mutex);
1547 mutex_init(&fs_info->tree_reloc_mutex);
1548 init_waitqueue_head(&fs_info->transaction_throttle);
1549 init_waitqueue_head(&fs_info->transaction_wait);
1550 init_waitqueue_head(&fs_info->async_submit_wait);
1551 init_waitqueue_head(&fs_info->tree_log_wait);
1552 atomic_set(&fs_info->tree_log_commit, 0);
1553 atomic_set(&fs_info->tree_log_writers, 0);
1554 fs_info->tree_log_transid = 0;
1557 ret = add_hasher(fs_info, "crc32c");
1559 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1564 __setup_root(4096, 4096, 4096, 4096, tree_root,
1565 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1568 bh = __bread(fs_devices->latest_bdev,
1569 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1573 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1576 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1578 disk_super = &fs_info->super_copy;
1579 if (!btrfs_super_root(disk_super))
1580 goto fail_sb_buffer;
1582 ret = btrfs_parse_options(tree_root, options);
1585 goto fail_sb_buffer;
1589 * we need to start all the end_io workers up front because the
1590 * queue work function gets called at interrupt time, and so it
1591 * cannot dynamically grow.
1593 btrfs_init_workers(&fs_info->workers, "worker",
1594 fs_info->thread_pool_size);
1596 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1597 fs_info->thread_pool_size);
1599 btrfs_init_workers(&fs_info->submit_workers, "submit",
1600 min_t(u64, fs_devices->num_devices,
1601 fs_info->thread_pool_size));
1603 /* a higher idle thresh on the submit workers makes it much more
1604 * likely that bios will be send down in a sane order to the
1607 fs_info->submit_workers.idle_thresh = 64;
1609 fs_info->workers.idle_thresh = 16;
1610 fs_info->workers.ordered = 1;
1612 fs_info->delalloc_workers.idle_thresh = 2;
1613 fs_info->delalloc_workers.ordered = 1;
1615 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1616 btrfs_init_workers(&fs_info->endio_workers, "endio",
1617 fs_info->thread_pool_size);
1618 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1619 fs_info->thread_pool_size);
1622 * endios are largely parallel and should have a very
1625 fs_info->endio_workers.idle_thresh = 4;
1626 fs_info->endio_write_workers.idle_thresh = 64;
1628 btrfs_start_workers(&fs_info->workers, 1);
1629 btrfs_start_workers(&fs_info->submit_workers, 1);
1630 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1631 btrfs_start_workers(&fs_info->fixup_workers, 1);
1632 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1633 btrfs_start_workers(&fs_info->endio_write_workers,
1634 fs_info->thread_pool_size);
1636 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1637 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1638 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1640 nodesize = btrfs_super_nodesize(disk_super);
1641 leafsize = btrfs_super_leafsize(disk_super);
1642 sectorsize = btrfs_super_sectorsize(disk_super);
1643 stripesize = btrfs_super_stripesize(disk_super);
1644 tree_root->nodesize = nodesize;
1645 tree_root->leafsize = leafsize;
1646 tree_root->sectorsize = sectorsize;
1647 tree_root->stripesize = stripesize;
1649 sb->s_blocksize = sectorsize;
1650 sb->s_blocksize_bits = blksize_bits(sectorsize);
1652 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1653 sizeof(disk_super->magic))) {
1654 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1655 goto fail_sb_buffer;
1658 mutex_lock(&fs_info->chunk_mutex);
1659 ret = btrfs_read_sys_array(tree_root);
1660 mutex_unlock(&fs_info->chunk_mutex);
1662 printk("btrfs: failed to read the system array on %s\n",
1664 goto fail_sys_array;
1667 blocksize = btrfs_level_size(tree_root,
1668 btrfs_super_chunk_root_level(disk_super));
1669 generation = btrfs_super_chunk_root_generation(disk_super);
1671 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1672 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1674 chunk_root->node = read_tree_block(chunk_root,
1675 btrfs_super_chunk_root(disk_super),
1676 blocksize, generation);
1677 BUG_ON(!chunk_root->node);
1679 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1680 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1683 mutex_lock(&fs_info->chunk_mutex);
1684 ret = btrfs_read_chunk_tree(chunk_root);
1685 mutex_unlock(&fs_info->chunk_mutex);
1687 printk("btrfs: failed to read chunk tree on %s\n", sb->s_id);
1688 goto fail_chunk_root;
1691 btrfs_close_extra_devices(fs_devices);
1693 blocksize = btrfs_level_size(tree_root,
1694 btrfs_super_root_level(disk_super));
1695 generation = btrfs_super_generation(disk_super);
1697 tree_root->node = read_tree_block(tree_root,
1698 btrfs_super_root(disk_super),
1699 blocksize, generation);
1700 if (!tree_root->node)
1701 goto fail_chunk_root;
1704 ret = find_and_setup_root(tree_root, fs_info,
1705 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1707 goto fail_tree_root;
1708 extent_root->track_dirty = 1;
1710 ret = find_and_setup_root(tree_root, fs_info,
1711 BTRFS_DEV_TREE_OBJECTID, dev_root);
1712 dev_root->track_dirty = 1;
1715 goto fail_extent_root;
1717 btrfs_read_block_groups(extent_root);
1719 fs_info->generation = generation + 1;
1720 fs_info->last_trans_committed = generation;
1721 fs_info->data_alloc_profile = (u64)-1;
1722 fs_info->metadata_alloc_profile = (u64)-1;
1723 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1724 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1726 if (!fs_info->cleaner_kthread)
1727 goto fail_extent_root;
1729 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1731 "btrfs-transaction");
1732 if (!fs_info->transaction_kthread)
1735 if (btrfs_super_log_root(disk_super) != 0) {
1737 u64 bytenr = btrfs_super_log_root(disk_super);
1739 if (fs_devices->rw_devices == 0) {
1740 printk("Btrfs log replay required on RO media\n");
1742 goto fail_trans_kthread;
1745 btrfs_level_size(tree_root,
1746 btrfs_super_log_root_level(disk_super));
1748 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1751 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1752 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1754 log_tree_root->node = read_tree_block(tree_root, bytenr,
1757 ret = btrfs_recover_log_trees(log_tree_root);
1761 if (!(sb->s_flags & MS_RDONLY)) {
1762 ret = btrfs_cleanup_reloc_trees(tree_root);
1766 location.objectid = BTRFS_FS_TREE_OBJECTID;
1767 location.type = BTRFS_ROOT_ITEM_KEY;
1768 location.offset = (u64)-1;
1770 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1771 if (!fs_info->fs_root)
1772 goto fail_trans_kthread;
1776 kthread_stop(fs_info->transaction_kthread);
1778 kthread_stop(fs_info->cleaner_kthread);
1781 * make sure we're done with the btree inode before we stop our
1784 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1785 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1788 free_extent_buffer(extent_root->node);
1790 free_extent_buffer(tree_root->node);
1792 free_extent_buffer(chunk_root->node);
1794 free_extent_buffer(dev_root->node);
1796 btrfs_stop_workers(&fs_info->fixup_workers);
1797 btrfs_stop_workers(&fs_info->delalloc_workers);
1798 btrfs_stop_workers(&fs_info->workers);
1799 btrfs_stop_workers(&fs_info->endio_workers);
1800 btrfs_stop_workers(&fs_info->endio_write_workers);
1801 btrfs_stop_workers(&fs_info->submit_workers);
1803 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1804 iput(fs_info->btree_inode);
1806 btrfs_close_devices(fs_info->fs_devices);
1807 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1811 bdi_destroy(&fs_info->bdi);
1815 return ERR_PTR(err);
1818 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1820 char b[BDEVNAME_SIZE];
1823 set_buffer_uptodate(bh);
1825 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1826 printk(KERN_WARNING "lost page write due to "
1827 "I/O error on %s\n",
1828 bdevname(bh->b_bdev, b));
1830 /* note, we dont' set_buffer_write_io_error because we have
1831 * our own ways of dealing with the IO errors
1833 clear_buffer_uptodate(bh);
1839 int write_all_supers(struct btrfs_root *root)
1841 struct list_head *cur;
1842 struct list_head *head = &root->fs_info->fs_devices->devices;
1843 struct btrfs_device *dev;
1844 struct btrfs_super_block *sb;
1845 struct btrfs_dev_item *dev_item;
1846 struct buffer_head *bh;
1850 int total_errors = 0;
1854 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1855 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1857 sb = &root->fs_info->super_for_commit;
1858 dev_item = &sb->dev_item;
1859 list_for_each(cur, head) {
1860 dev = list_entry(cur, struct btrfs_device, dev_list);
1865 if (!dev->in_fs_metadata || !dev->writeable)
1868 btrfs_set_stack_device_generation(dev_item, 0);
1869 btrfs_set_stack_device_type(dev_item, dev->type);
1870 btrfs_set_stack_device_id(dev_item, dev->devid);
1871 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1872 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1873 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1874 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1875 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1876 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1877 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1878 flags = btrfs_super_flags(sb);
1879 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1883 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1884 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1885 btrfs_csum_final(crc, sb->csum);
1887 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1888 BTRFS_SUPER_INFO_SIZE);
1890 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1891 dev->pending_io = bh;
1894 set_buffer_uptodate(bh);
1896 bh->b_end_io = btrfs_end_buffer_write_sync;
1898 if (do_barriers && dev->barriers) {
1899 ret = submit_bh(WRITE_BARRIER, bh);
1900 if (ret == -EOPNOTSUPP) {
1901 printk("btrfs: disabling barriers on dev %s\n",
1903 set_buffer_uptodate(bh);
1907 ret = submit_bh(WRITE, bh);
1910 ret = submit_bh(WRITE, bh);
1915 if (total_errors > max_errors) {
1916 printk("btrfs: %d errors while writing supers\n", total_errors);
1921 list_for_each(cur, head) {
1922 dev = list_entry(cur, struct btrfs_device, dev_list);
1925 if (!dev->in_fs_metadata || !dev->writeable)
1928 BUG_ON(!dev->pending_io);
1929 bh = dev->pending_io;
1931 if (!buffer_uptodate(dev->pending_io)) {
1932 if (do_barriers && dev->barriers) {
1933 printk("btrfs: disabling barriers on dev %s\n",
1935 set_buffer_uptodate(bh);
1939 ret = submit_bh(WRITE, bh);
1942 if (!buffer_uptodate(bh))
1949 dev->pending_io = NULL;
1952 if (total_errors > max_errors) {
1953 printk("btrfs: %d errors while writing supers\n", total_errors);
1959 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1964 ret = write_all_supers(root);
1968 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1970 radix_tree_delete(&fs_info->fs_roots_radix,
1971 (unsigned long)root->root_key.objectid);
1972 if (root->anon_super.s_dev) {
1973 down_write(&root->anon_super.s_umount);
1974 kill_anon_super(&root->anon_super);
1978 btrfs_sysfs_del_root(root);
1981 free_extent_buffer(root->node);
1982 if (root->commit_root)
1983 free_extent_buffer(root->commit_root);
1990 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1993 struct btrfs_root *gang[8];
1997 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2002 for (i = 0; i < ret; i++)
2003 btrfs_free_fs_root(fs_info, gang[i]);
2008 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2010 u64 root_objectid = 0;
2011 struct btrfs_root *gang[8];
2016 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2017 (void **)gang, root_objectid,
2021 for (i = 0; i < ret; i++) {
2022 root_objectid = gang[i]->root_key.objectid;
2023 ret = btrfs_find_dead_roots(fs_info->tree_root,
2024 root_objectid, gang[i]);
2026 btrfs_orphan_cleanup(gang[i]);
2033 int btrfs_commit_super(struct btrfs_root *root)
2035 struct btrfs_trans_handle *trans;
2038 mutex_lock(&root->fs_info->cleaner_mutex);
2039 btrfs_clean_old_snapshots(root);
2040 mutex_unlock(&root->fs_info->cleaner_mutex);
2041 trans = btrfs_start_transaction(root, 1);
2042 ret = btrfs_commit_transaction(trans, root);
2044 /* run commit again to drop the original snapshot */
2045 trans = btrfs_start_transaction(root, 1);
2046 btrfs_commit_transaction(trans, root);
2047 ret = btrfs_write_and_wait_transaction(NULL, root);
2050 ret = write_ctree_super(NULL, root);
2054 int close_ctree(struct btrfs_root *root)
2056 struct btrfs_fs_info *fs_info = root->fs_info;
2059 fs_info->closing = 1;
2062 kthread_stop(root->fs_info->transaction_kthread);
2063 kthread_stop(root->fs_info->cleaner_kthread);
2065 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2066 ret = btrfs_commit_super(root);
2068 printk("btrfs: commit super returns %d\n", ret);
2072 if (fs_info->delalloc_bytes) {
2073 printk("btrfs: at unmount delalloc count %Lu\n",
2074 fs_info->delalloc_bytes);
2076 if (fs_info->total_ref_cache_size) {
2077 printk("btrfs: at umount reference cache size %Lu\n",
2078 fs_info->total_ref_cache_size);
2081 if (fs_info->extent_root->node)
2082 free_extent_buffer(fs_info->extent_root->node);
2084 if (fs_info->tree_root->node)
2085 free_extent_buffer(fs_info->tree_root->node);
2087 if (root->fs_info->chunk_root->node);
2088 free_extent_buffer(root->fs_info->chunk_root->node);
2090 if (root->fs_info->dev_root->node);
2091 free_extent_buffer(root->fs_info->dev_root->node);
2093 btrfs_free_block_groups(root->fs_info);
2095 del_fs_roots(fs_info);
2097 iput(fs_info->btree_inode);
2099 btrfs_stop_workers(&fs_info->fixup_workers);
2100 btrfs_stop_workers(&fs_info->delalloc_workers);
2101 btrfs_stop_workers(&fs_info->workers);
2102 btrfs_stop_workers(&fs_info->endio_workers);
2103 btrfs_stop_workers(&fs_info->endio_write_workers);
2104 btrfs_stop_workers(&fs_info->submit_workers);
2107 while(!list_empty(&fs_info->hashers)) {
2108 struct btrfs_hasher *hasher;
2109 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2111 list_del(&hasher->hashers);
2112 crypto_free_hash(&fs_info->hash_tfm);
2116 btrfs_close_devices(fs_info->fs_devices);
2117 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2119 bdi_destroy(&fs_info->bdi);
2121 kfree(fs_info->extent_root);
2122 kfree(fs_info->tree_root);
2123 kfree(fs_info->chunk_root);
2124 kfree(fs_info->dev_root);
2128 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2131 struct inode *btree_inode = buf->first_page->mapping->host;
2133 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2137 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2142 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2144 struct inode *btree_inode = buf->first_page->mapping->host;
2145 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2149 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2151 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2152 u64 transid = btrfs_header_generation(buf);
2153 struct inode *btree_inode = root->fs_info->btree_inode;
2155 WARN_ON(!btrfs_tree_locked(buf));
2156 if (transid != root->fs_info->generation) {
2157 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2158 (unsigned long long)buf->start,
2159 transid, root->fs_info->generation);
2162 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2165 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2168 * looks as though older kernels can get into trouble with
2169 * this code, they end up stuck in balance_dirty_pages forever
2171 struct extent_io_tree *tree;
2174 unsigned long thresh = 32 * 1024 * 1024;
2175 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2177 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2180 num_dirty = count_range_bits(tree, &start, (u64)-1,
2181 thresh, EXTENT_DIRTY);
2182 if (num_dirty > thresh) {
2183 balance_dirty_pages_ratelimited_nr(
2184 root->fs_info->btree_inode->i_mapping, 1);
2189 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2191 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2193 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2195 buf->flags |= EXTENT_UPTODATE;
2200 int btree_lock_page_hook(struct page *page)
2202 struct inode *inode = page->mapping->host;
2203 struct btrfs_root *root = BTRFS_I(inode)->root;
2204 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2205 struct extent_buffer *eb;
2207 u64 bytenr = page_offset(page);
2209 if (page->private == EXTENT_PAGE_PRIVATE)
2212 len = page->private >> 2;
2213 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2217 btrfs_tree_lock(eb);
2218 spin_lock(&root->fs_info->hash_lock);
2219 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2220 spin_unlock(&root->fs_info->hash_lock);
2221 btrfs_tree_unlock(eb);
2222 free_extent_buffer(eb);
2228 static struct extent_io_ops btree_extent_io_ops = {
2229 .write_cache_pages_lock_hook = btree_lock_page_hook,
2230 .readpage_end_io_hook = btree_readpage_end_io_hook,
2231 .submit_bio_hook = btree_submit_bio_hook,
2232 /* note we're sharing with inode.c for the merge bio hook */
2233 .merge_bio_hook = btrfs_merge_bio_hook,