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.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #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"
42 static struct extent_io_ops btree_extent_io_ops;
43 static void end_workqueue_fn(struct btrfs_work *work);
46 * end_io_wq structs are used to do processing in task context when an IO is
47 * complete. This is used during reads to verify checksums, and it is used
48 * by writes to insert metadata for new file extents after IO is complete.
54 struct btrfs_fs_info *info;
57 struct list_head list;
58 struct btrfs_work work;
62 * async submit bios are used to offload expensive checksumming
63 * onto the worker threads. They checksum file and metadata bios
64 * just before they are sent down the IO stack.
66 struct async_submit_bio {
69 struct list_head list;
70 extent_submit_bio_hook_t *submit_bio_start;
71 extent_submit_bio_hook_t *submit_bio_done;
74 unsigned long bio_flags;
75 struct btrfs_work work;
79 * extents on the btree inode are pretty simple, there's one extent
80 * that covers the entire device
82 static struct extent_map *btree_get_extent(struct inode *inode,
83 struct page *page, size_t page_offset, u64 start, u64 len,
86 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
87 struct extent_map *em;
90 spin_lock(&em_tree->lock);
91 em = lookup_extent_mapping(em_tree, start, len);
94 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
95 spin_unlock(&em_tree->lock);
98 spin_unlock(&em_tree->lock);
100 em = alloc_extent_map(GFP_NOFS);
102 em = ERR_PTR(-ENOMEM);
107 em->block_len = (u64)-1;
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;
118 em = lookup_extent_mapping(em_tree, start, len);
122 em = lookup_extent_mapping(em_tree, failed_start,
130 spin_unlock(&em_tree->lock);
138 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
140 return btrfs_crc32c(seed, data, len);
143 void btrfs_csum_final(u32 crc, char *result)
145 *(__le32 *)result = ~cpu_to_le32(crc);
149 * compute the csum for a btree block, and either verify it or write it
150 * into the csum field of the block.
152 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
156 btrfs_super_csum_size(&root->fs_info->super_copy);
159 unsigned long cur_len;
160 unsigned long offset = BTRFS_CSUM_SIZE;
161 char *map_token = NULL;
163 unsigned long map_start;
164 unsigned long map_len;
167 unsigned long inline_result;
169 len = buf->len - offset;
171 err = map_private_extent_buffer(buf, offset, 32,
173 &map_start, &map_len, KM_USER0);
176 cur_len = min(len, map_len - (offset - map_start));
177 crc = btrfs_csum_data(root, kaddr + offset - map_start,
181 unmap_extent_buffer(buf, map_token, KM_USER0);
183 if (csum_size > sizeof(inline_result)) {
184 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
188 result = (char *)&inline_result;
191 btrfs_csum_final(crc, result);
194 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
197 memcpy(&found, result, csum_size);
199 read_extent_buffer(buf, &val, 0, csum_size);
200 printk(KERN_INFO "btrfs: %s checksum verify failed "
201 "on %llu wanted %X found %X level %d\n",
202 root->fs_info->sb->s_id,
203 buf->start, val, found, btrfs_header_level(buf));
204 if (result != (char *)&inline_result)
209 write_extent_buffer(buf, result, 0, csum_size);
211 if (result != (char *)&inline_result)
217 * we can't consider a given block up to date unless the transid of the
218 * block matches the transid in the parent node's pointer. This is how we
219 * detect blocks that either didn't get written at all or got written
220 * in the wrong place.
222 static int verify_parent_transid(struct extent_io_tree *io_tree,
223 struct extent_buffer *eb, u64 parent_transid)
227 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
230 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
231 if (extent_buffer_uptodate(io_tree, eb) &&
232 btrfs_header_generation(eb) == parent_transid) {
236 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
237 (unsigned long long)eb->start,
238 (unsigned long long)parent_transid,
239 (unsigned long long)btrfs_header_generation(eb));
241 clear_extent_buffer_uptodate(io_tree, eb);
243 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
249 * helper to read a given tree block, doing retries as required when
250 * the checksums don't match and we have alternate mirrors to try.
252 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
253 struct extent_buffer *eb,
254 u64 start, u64 parent_transid)
256 struct extent_io_tree *io_tree;
261 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
263 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
264 btree_get_extent, mirror_num);
266 !verify_parent_transid(io_tree, eb, parent_transid))
269 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
275 if (mirror_num > num_copies)
282 * checksum a dirty tree block before IO. This has extra checks to make sure
283 * we only fill in the checksum field in the first page of a multi-page block
286 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
288 struct extent_io_tree *tree;
289 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
293 struct extent_buffer *eb;
296 tree = &BTRFS_I(page->mapping->host)->io_tree;
298 if (page->private == EXTENT_PAGE_PRIVATE)
302 len = page->private >> 2;
305 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
306 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
307 btrfs_header_generation(eb));
309 found_start = btrfs_header_bytenr(eb);
310 if (found_start != start) {
314 if (eb->first_page != page) {
318 if (!PageUptodate(page)) {
322 found_level = btrfs_header_level(eb);
324 csum_tree_block(root, eb, 0);
326 free_extent_buffer(eb);
331 static int check_tree_block_fsid(struct btrfs_root *root,
332 struct extent_buffer *eb)
334 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
335 u8 fsid[BTRFS_UUID_SIZE];
338 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
341 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
345 fs_devices = fs_devices->seed;
350 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
351 struct extent_state *state)
353 struct extent_io_tree *tree;
357 struct extent_buffer *eb;
358 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
361 tree = &BTRFS_I(page->mapping->host)->io_tree;
362 if (page->private == EXTENT_PAGE_PRIVATE)
367 len = page->private >> 2;
370 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
372 found_start = btrfs_header_bytenr(eb);
373 if (found_start != start) {
374 printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
375 (unsigned long long)found_start,
376 (unsigned long long)eb->start);
380 if (eb->first_page != page) {
381 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
382 eb->first_page->index, page->index);
387 if (check_tree_block_fsid(root, eb)) {
388 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
389 (unsigned long long)eb->start);
393 found_level = btrfs_header_level(eb);
395 ret = csum_tree_block(root, eb, 1);
399 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
400 end = eb->start + end - 1;
402 free_extent_buffer(eb);
407 static void end_workqueue_bio(struct bio *bio, int err)
409 struct end_io_wq *end_io_wq = bio->bi_private;
410 struct btrfs_fs_info *fs_info;
412 fs_info = end_io_wq->info;
413 end_io_wq->error = err;
414 end_io_wq->work.func = end_workqueue_fn;
415 end_io_wq->work.flags = 0;
417 if (bio->bi_rw & (1 << BIO_RW)) {
418 if (end_io_wq->metadata)
419 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
422 btrfs_queue_worker(&fs_info->endio_write_workers,
425 if (end_io_wq->metadata)
426 btrfs_queue_worker(&fs_info->endio_meta_workers,
429 btrfs_queue_worker(&fs_info->endio_workers,
434 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
437 struct end_io_wq *end_io_wq;
438 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
442 end_io_wq->private = bio->bi_private;
443 end_io_wq->end_io = bio->bi_end_io;
444 end_io_wq->info = info;
445 end_io_wq->error = 0;
446 end_io_wq->bio = bio;
447 end_io_wq->metadata = metadata;
449 bio->bi_private = end_io_wq;
450 bio->bi_end_io = end_workqueue_bio;
454 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
456 unsigned long limit = min_t(unsigned long,
457 info->workers.max_workers,
458 info->fs_devices->open_devices);
462 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
464 return atomic_read(&info->nr_async_bios) >
465 btrfs_async_submit_limit(info);
468 static void run_one_async_start(struct btrfs_work *work)
470 struct btrfs_fs_info *fs_info;
471 struct async_submit_bio *async;
473 async = container_of(work, struct async_submit_bio, work);
474 fs_info = BTRFS_I(async->inode)->root->fs_info;
475 async->submit_bio_start(async->inode, async->rw, async->bio,
476 async->mirror_num, async->bio_flags);
479 static void run_one_async_done(struct btrfs_work *work)
481 struct btrfs_fs_info *fs_info;
482 struct async_submit_bio *async;
485 async = container_of(work, struct async_submit_bio, work);
486 fs_info = BTRFS_I(async->inode)->root->fs_info;
488 limit = btrfs_async_submit_limit(fs_info);
489 limit = limit * 2 / 3;
491 atomic_dec(&fs_info->nr_async_submits);
493 if (atomic_read(&fs_info->nr_async_submits) < limit &&
494 waitqueue_active(&fs_info->async_submit_wait))
495 wake_up(&fs_info->async_submit_wait);
497 async->submit_bio_done(async->inode, async->rw, async->bio,
498 async->mirror_num, async->bio_flags);
501 static void run_one_async_free(struct btrfs_work *work)
503 struct async_submit_bio *async;
505 async = container_of(work, struct async_submit_bio, work);
509 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
510 int rw, struct bio *bio, int mirror_num,
511 unsigned long bio_flags,
512 extent_submit_bio_hook_t *submit_bio_start,
513 extent_submit_bio_hook_t *submit_bio_done)
515 struct async_submit_bio *async;
517 async = kmalloc(sizeof(*async), GFP_NOFS);
521 async->inode = inode;
524 async->mirror_num = mirror_num;
525 async->submit_bio_start = submit_bio_start;
526 async->submit_bio_done = submit_bio_done;
528 async->work.func = run_one_async_start;
529 async->work.ordered_func = run_one_async_done;
530 async->work.ordered_free = run_one_async_free;
532 async->work.flags = 0;
533 async->bio_flags = bio_flags;
535 atomic_inc(&fs_info->nr_async_submits);
536 btrfs_queue_worker(&fs_info->workers, &async->work);
538 int limit = btrfs_async_submit_limit(fs_info);
539 if (atomic_read(&fs_info->nr_async_submits) > limit) {
540 wait_event_timeout(fs_info->async_submit_wait,
541 (atomic_read(&fs_info->nr_async_submits) < limit),
544 wait_event_timeout(fs_info->async_submit_wait,
545 (atomic_read(&fs_info->nr_async_bios) < limit),
549 while (atomic_read(&fs_info->async_submit_draining) &&
550 atomic_read(&fs_info->nr_async_submits)) {
551 wait_event(fs_info->async_submit_wait,
552 (atomic_read(&fs_info->nr_async_submits) == 0));
558 static int btree_csum_one_bio(struct bio *bio)
560 struct bio_vec *bvec = bio->bi_io_vec;
562 struct btrfs_root *root;
564 WARN_ON(bio->bi_vcnt <= 0);
565 while (bio_index < bio->bi_vcnt) {
566 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
567 csum_dirty_buffer(root, bvec->bv_page);
574 static int __btree_submit_bio_start(struct inode *inode, int rw,
575 struct bio *bio, int mirror_num,
576 unsigned long bio_flags)
579 * when we're called for a write, we're already in the async
580 * submission context. Just jump into btrfs_map_bio
582 btree_csum_one_bio(bio);
586 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
587 int mirror_num, unsigned long bio_flags)
590 * when we're called for a write, we're already in the async
591 * submission context. Just jump into btrfs_map_bio
593 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
596 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
597 int mirror_num, unsigned long bio_flags)
601 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
605 if (!(rw & (1 << BIO_RW))) {
607 * called for a read, do the setup so that checksum validation
608 * can happen in the async kernel threads
610 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
614 * kthread helpers are used to submit writes so that checksumming
615 * can happen in parallel across all CPUs
617 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
618 inode, rw, bio, mirror_num, 0,
619 __btree_submit_bio_start,
620 __btree_submit_bio_done);
623 static int btree_writepage(struct page *page, struct writeback_control *wbc)
625 struct extent_io_tree *tree;
626 tree = &BTRFS_I(page->mapping->host)->io_tree;
628 if (current->flags & PF_MEMALLOC) {
629 redirty_page_for_writepage(wbc, page);
633 return extent_write_full_page(tree, page, btree_get_extent, wbc);
636 static int btree_writepages(struct address_space *mapping,
637 struct writeback_control *wbc)
639 struct extent_io_tree *tree;
640 tree = &BTRFS_I(mapping->host)->io_tree;
641 if (wbc->sync_mode == WB_SYNC_NONE) {
644 unsigned long thresh = 32 * 1024 * 1024;
646 if (wbc->for_kupdate)
649 num_dirty = count_range_bits(tree, &start, (u64)-1,
650 thresh, EXTENT_DIRTY);
651 if (num_dirty < thresh)
654 return extent_writepages(tree, mapping, btree_get_extent, wbc);
657 static int btree_readpage(struct file *file, struct page *page)
659 struct extent_io_tree *tree;
660 tree = &BTRFS_I(page->mapping->host)->io_tree;
661 return extent_read_full_page(tree, page, btree_get_extent);
664 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
666 struct extent_io_tree *tree;
667 struct extent_map_tree *map;
670 if (PageWriteback(page) || PageDirty(page))
673 tree = &BTRFS_I(page->mapping->host)->io_tree;
674 map = &BTRFS_I(page->mapping->host)->extent_tree;
676 ret = try_release_extent_state(map, tree, page, gfp_flags);
680 ret = try_release_extent_buffer(tree, page);
682 ClearPagePrivate(page);
683 set_page_private(page, 0);
684 page_cache_release(page);
690 static void btree_invalidatepage(struct page *page, unsigned long offset)
692 struct extent_io_tree *tree;
693 tree = &BTRFS_I(page->mapping->host)->io_tree;
694 extent_invalidatepage(tree, page, offset);
695 btree_releasepage(page, GFP_NOFS);
696 if (PagePrivate(page)) {
697 printk(KERN_WARNING "btrfs warning page private not zero "
698 "on page %llu\n", (unsigned long long)page_offset(page));
699 ClearPagePrivate(page);
700 set_page_private(page, 0);
701 page_cache_release(page);
706 static int btree_writepage(struct page *page, struct writeback_control *wbc)
708 struct buffer_head *bh;
709 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
710 struct buffer_head *head;
711 if (!page_has_buffers(page)) {
712 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
713 (1 << BH_Dirty)|(1 << BH_Uptodate));
715 head = page_buffers(page);
718 if (buffer_dirty(bh))
719 csum_tree_block(root, bh, 0);
720 bh = bh->b_this_page;
721 } while (bh != head);
722 return block_write_full_page(page, btree_get_block, wbc);
726 static struct address_space_operations btree_aops = {
727 .readpage = btree_readpage,
728 .writepage = btree_writepage,
729 .writepages = btree_writepages,
730 .releasepage = btree_releasepage,
731 .invalidatepage = btree_invalidatepage,
732 .sync_page = block_sync_page,
735 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
738 struct extent_buffer *buf = NULL;
739 struct inode *btree_inode = root->fs_info->btree_inode;
742 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
745 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
746 buf, 0, 0, btree_get_extent, 0);
747 free_extent_buffer(buf);
751 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
752 u64 bytenr, u32 blocksize)
754 struct inode *btree_inode = root->fs_info->btree_inode;
755 struct extent_buffer *eb;
756 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
757 bytenr, blocksize, GFP_NOFS);
761 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
762 u64 bytenr, u32 blocksize)
764 struct inode *btree_inode = root->fs_info->btree_inode;
765 struct extent_buffer *eb;
767 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
768 bytenr, blocksize, NULL, GFP_NOFS);
773 int btrfs_write_tree_block(struct extent_buffer *buf)
775 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
776 buf->start + buf->len - 1, WB_SYNC_ALL);
779 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
781 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
782 buf->start, buf->start + buf->len - 1);
785 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
786 u32 blocksize, u64 parent_transid)
788 struct extent_buffer *buf = NULL;
789 struct inode *btree_inode = root->fs_info->btree_inode;
790 struct extent_io_tree *io_tree;
793 io_tree = &BTRFS_I(btree_inode)->io_tree;
795 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
799 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
802 buf->flags |= EXTENT_UPTODATE;
809 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
810 struct extent_buffer *buf)
812 struct inode *btree_inode = root->fs_info->btree_inode;
813 if (btrfs_header_generation(buf) ==
814 root->fs_info->running_transaction->transid) {
815 WARN_ON(!btrfs_tree_locked(buf));
816 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
822 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
823 u32 stripesize, struct btrfs_root *root,
824 struct btrfs_fs_info *fs_info,
828 root->commit_root = NULL;
829 root->ref_tree = NULL;
830 root->sectorsize = sectorsize;
831 root->nodesize = nodesize;
832 root->leafsize = leafsize;
833 root->stripesize = stripesize;
835 root->track_dirty = 0;
837 root->fs_info = fs_info;
838 root->objectid = objectid;
839 root->last_trans = 0;
840 root->highest_inode = 0;
841 root->last_inode_alloc = 0;
845 INIT_LIST_HEAD(&root->dirty_list);
846 INIT_LIST_HEAD(&root->orphan_list);
847 INIT_LIST_HEAD(&root->dead_list);
848 spin_lock_init(&root->node_lock);
849 spin_lock_init(&root->list_lock);
850 mutex_init(&root->objectid_mutex);
851 mutex_init(&root->log_mutex);
852 extent_io_tree_init(&root->dirty_log_pages,
853 fs_info->btree_inode->i_mapping, GFP_NOFS);
855 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
856 root->ref_tree = &root->ref_tree_struct;
858 memset(&root->root_key, 0, sizeof(root->root_key));
859 memset(&root->root_item, 0, sizeof(root->root_item));
860 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
861 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
862 root->defrag_trans_start = fs_info->generation;
863 init_completion(&root->kobj_unregister);
864 root->defrag_running = 0;
865 root->defrag_level = 0;
866 root->root_key.objectid = objectid;
867 root->anon_super.s_root = NULL;
868 root->anon_super.s_dev = 0;
869 INIT_LIST_HEAD(&root->anon_super.s_list);
870 INIT_LIST_HEAD(&root->anon_super.s_instances);
871 init_rwsem(&root->anon_super.s_umount);
876 static int find_and_setup_root(struct btrfs_root *tree_root,
877 struct btrfs_fs_info *fs_info,
879 struct btrfs_root *root)
885 __setup_root(tree_root->nodesize, tree_root->leafsize,
886 tree_root->sectorsize, tree_root->stripesize,
887 root, fs_info, objectid);
888 ret = btrfs_find_last_root(tree_root, objectid,
889 &root->root_item, &root->root_key);
892 generation = btrfs_root_generation(&root->root_item);
893 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
894 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
895 blocksize, generation);
900 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
901 struct btrfs_fs_info *fs_info)
903 struct extent_buffer *eb;
904 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
913 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
914 0, &start, &end, EXTENT_DIRTY);
918 clear_extent_dirty(&log_root_tree->dirty_log_pages,
919 start, end, GFP_NOFS);
921 eb = fs_info->log_root_tree->node;
923 WARN_ON(btrfs_header_level(eb) != 0);
924 WARN_ON(btrfs_header_nritems(eb) != 0);
926 ret = btrfs_free_reserved_extent(fs_info->tree_root,
930 free_extent_buffer(eb);
931 kfree(fs_info->log_root_tree);
932 fs_info->log_root_tree = NULL;
936 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
937 struct btrfs_fs_info *fs_info)
939 struct btrfs_root *root;
940 struct btrfs_root *tree_root = fs_info->tree_root;
942 root = kzalloc(sizeof(*root), GFP_NOFS);
946 __setup_root(tree_root->nodesize, tree_root->leafsize,
947 tree_root->sectorsize, tree_root->stripesize,
948 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
950 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
951 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
952 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
955 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
956 0, BTRFS_TREE_LOG_OBJECTID,
957 trans->transid, 0, 0, 0);
959 btrfs_set_header_nritems(root->node, 0);
960 btrfs_set_header_level(root->node, 0);
961 btrfs_set_header_bytenr(root->node, root->node->start);
962 btrfs_set_header_generation(root->node, trans->transid);
963 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
965 write_extent_buffer(root->node, root->fs_info->fsid,
966 (unsigned long)btrfs_header_fsid(root->node),
968 btrfs_mark_buffer_dirty(root->node);
969 btrfs_tree_unlock(root->node);
970 fs_info->log_root_tree = root;
974 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
975 struct btrfs_key *location)
977 struct btrfs_root *root;
978 struct btrfs_fs_info *fs_info = tree_root->fs_info;
979 struct btrfs_path *path;
980 struct extent_buffer *l;
986 root = kzalloc(sizeof(*root), GFP_NOFS);
988 return ERR_PTR(-ENOMEM);
989 if (location->offset == (u64)-1) {
990 ret = find_and_setup_root(tree_root, fs_info,
991 location->objectid, root);
999 __setup_root(tree_root->nodesize, tree_root->leafsize,
1000 tree_root->sectorsize, tree_root->stripesize,
1001 root, fs_info, location->objectid);
1003 path = btrfs_alloc_path();
1005 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1012 read_extent_buffer(l, &root->root_item,
1013 btrfs_item_ptr_offset(l, path->slots[0]),
1014 sizeof(root->root_item));
1015 memcpy(&root->root_key, location, sizeof(*location));
1018 btrfs_release_path(root, path);
1019 btrfs_free_path(path);
1022 return ERR_PTR(ret);
1024 generation = btrfs_root_generation(&root->root_item);
1025 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1026 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1027 blocksize, generation);
1028 BUG_ON(!root->node);
1030 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1032 ret = btrfs_find_highest_inode(root, &highest_inode);
1034 root->highest_inode = highest_inode;
1035 root->last_inode_alloc = highest_inode;
1041 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1044 struct btrfs_root *root;
1046 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1047 return fs_info->tree_root;
1048 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1049 return fs_info->extent_root;
1051 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1052 (unsigned long)root_objectid);
1056 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1057 struct btrfs_key *location)
1059 struct btrfs_root *root;
1062 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1063 return fs_info->tree_root;
1064 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1065 return fs_info->extent_root;
1066 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1067 return fs_info->chunk_root;
1068 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1069 return fs_info->dev_root;
1070 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1071 return fs_info->csum_root;
1073 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1074 (unsigned long)location->objectid);
1078 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1082 set_anon_super(&root->anon_super, NULL);
1084 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1085 (unsigned long)root->root_key.objectid,
1088 free_extent_buffer(root->node);
1090 return ERR_PTR(ret);
1092 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1093 ret = btrfs_find_dead_roots(fs_info->tree_root,
1094 root->root_key.objectid, root);
1096 btrfs_orphan_cleanup(root);
1101 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1102 struct btrfs_key *location,
1103 const char *name, int namelen)
1105 struct btrfs_root *root;
1108 root = btrfs_read_fs_root_no_name(fs_info, location);
1115 ret = btrfs_set_root_name(root, name, namelen);
1117 free_extent_buffer(root->node);
1119 return ERR_PTR(ret);
1122 ret = btrfs_sysfs_add_root(root);
1124 free_extent_buffer(root->node);
1127 return ERR_PTR(ret);
1134 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1136 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1138 struct btrfs_device *device;
1139 struct backing_dev_info *bdi;
1141 if ((bdi_bits & (1 << BDI_write_congested)) &&
1142 btrfs_congested_async(info, 0))
1145 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1148 bdi = blk_get_backing_dev_info(device->bdev);
1149 if (bdi && bdi_congested(bdi, bdi_bits)) {
1158 * this unplugs every device on the box, and it is only used when page
1161 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1163 struct btrfs_device *device;
1164 struct btrfs_fs_info *info;
1166 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1167 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1171 bdi = blk_get_backing_dev_info(device->bdev);
1172 if (bdi->unplug_io_fn)
1173 bdi->unplug_io_fn(bdi, page);
1177 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1179 struct inode *inode;
1180 struct extent_map_tree *em_tree;
1181 struct extent_map *em;
1182 struct address_space *mapping;
1185 /* the generic O_DIRECT read code does this */
1187 __unplug_io_fn(bdi, page);
1192 * page->mapping may change at any time. Get a consistent copy
1193 * and use that for everything below
1196 mapping = page->mapping;
1200 inode = mapping->host;
1203 * don't do the expensive searching for a small number of
1206 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1207 __unplug_io_fn(bdi, page);
1211 offset = page_offset(page);
1213 em_tree = &BTRFS_I(inode)->extent_tree;
1214 spin_lock(&em_tree->lock);
1215 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1216 spin_unlock(&em_tree->lock);
1218 __unplug_io_fn(bdi, page);
1222 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1223 free_extent_map(em);
1224 __unplug_io_fn(bdi, page);
1227 offset = offset - em->start;
1228 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1229 em->block_start + offset, page);
1230 free_extent_map(em);
1233 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1236 bdi->ra_pages = default_backing_dev_info.ra_pages;
1238 bdi->capabilities = default_backing_dev_info.capabilities;
1239 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1240 bdi->unplug_io_data = info;
1241 bdi->congested_fn = btrfs_congested_fn;
1242 bdi->congested_data = info;
1246 static int bio_ready_for_csum(struct bio *bio)
1252 struct extent_io_tree *io_tree = NULL;
1253 struct btrfs_fs_info *info = NULL;
1254 struct bio_vec *bvec;
1258 bio_for_each_segment(bvec, bio, i) {
1259 page = bvec->bv_page;
1260 if (page->private == EXTENT_PAGE_PRIVATE) {
1261 length += bvec->bv_len;
1264 if (!page->private) {
1265 length += bvec->bv_len;
1268 length = bvec->bv_len;
1269 buf_len = page->private >> 2;
1270 start = page_offset(page) + bvec->bv_offset;
1271 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1272 info = BTRFS_I(page->mapping->host)->root->fs_info;
1274 /* are we fully contained in this bio? */
1275 if (buf_len <= length)
1278 ret = extent_range_uptodate(io_tree, start + length,
1279 start + buf_len - 1);
1286 * called by the kthread helper functions to finally call the bio end_io
1287 * functions. This is where read checksum verification actually happens
1289 static void end_workqueue_fn(struct btrfs_work *work)
1292 struct end_io_wq *end_io_wq;
1293 struct btrfs_fs_info *fs_info;
1296 end_io_wq = container_of(work, struct end_io_wq, work);
1297 bio = end_io_wq->bio;
1298 fs_info = end_io_wq->info;
1300 /* metadata bio reads are special because the whole tree block must
1301 * be checksummed at once. This makes sure the entire block is in
1302 * ram and up to date before trying to verify things. For
1303 * blocksize <= pagesize, it is basically a noop
1305 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1306 !bio_ready_for_csum(bio)) {
1307 btrfs_queue_worker(&fs_info->endio_meta_workers,
1311 error = end_io_wq->error;
1312 bio->bi_private = end_io_wq->private;
1313 bio->bi_end_io = end_io_wq->end_io;
1315 bio_endio(bio, error);
1318 static int cleaner_kthread(void *arg)
1320 struct btrfs_root *root = arg;
1324 if (root->fs_info->closing)
1327 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1328 mutex_lock(&root->fs_info->cleaner_mutex);
1329 btrfs_clean_old_snapshots(root);
1330 mutex_unlock(&root->fs_info->cleaner_mutex);
1332 if (freezing(current)) {
1336 if (root->fs_info->closing)
1338 set_current_state(TASK_INTERRUPTIBLE);
1340 __set_current_state(TASK_RUNNING);
1342 } while (!kthread_should_stop());
1346 static int transaction_kthread(void *arg)
1348 struct btrfs_root *root = arg;
1349 struct btrfs_trans_handle *trans;
1350 struct btrfs_transaction *cur;
1352 unsigned long delay;
1357 if (root->fs_info->closing)
1361 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1362 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1364 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1365 printk(KERN_INFO "btrfs: total reference cache "
1367 root->fs_info->total_ref_cache_size);
1370 mutex_lock(&root->fs_info->trans_mutex);
1371 cur = root->fs_info->running_transaction;
1373 mutex_unlock(&root->fs_info->trans_mutex);
1377 now = get_seconds();
1378 if (now < cur->start_time || now - cur->start_time < 30) {
1379 mutex_unlock(&root->fs_info->trans_mutex);
1383 mutex_unlock(&root->fs_info->trans_mutex);
1384 trans = btrfs_start_transaction(root, 1);
1385 ret = btrfs_commit_transaction(trans, root);
1387 wake_up_process(root->fs_info->cleaner_kthread);
1388 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1390 if (freezing(current)) {
1393 if (root->fs_info->closing)
1395 set_current_state(TASK_INTERRUPTIBLE);
1396 schedule_timeout(delay);
1397 __set_current_state(TASK_RUNNING);
1399 } while (!kthread_should_stop());
1403 struct btrfs_root *open_ctree(struct super_block *sb,
1404 struct btrfs_fs_devices *fs_devices,
1414 struct btrfs_key location;
1415 struct buffer_head *bh;
1416 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1418 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1420 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1422 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1424 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1426 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1428 struct btrfs_root *log_tree_root;
1433 struct btrfs_super_block *disk_super;
1435 if (!extent_root || !tree_root || !fs_info ||
1436 !chunk_root || !dev_root || !csum_root) {
1440 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1441 INIT_LIST_HEAD(&fs_info->trans_list);
1442 INIT_LIST_HEAD(&fs_info->dead_roots);
1443 INIT_LIST_HEAD(&fs_info->hashers);
1444 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1445 spin_lock_init(&fs_info->hash_lock);
1446 spin_lock_init(&fs_info->delalloc_lock);
1447 spin_lock_init(&fs_info->new_trans_lock);
1448 spin_lock_init(&fs_info->ref_cache_lock);
1450 init_completion(&fs_info->kobj_unregister);
1451 fs_info->tree_root = tree_root;
1452 fs_info->extent_root = extent_root;
1453 fs_info->csum_root = csum_root;
1454 fs_info->chunk_root = chunk_root;
1455 fs_info->dev_root = dev_root;
1456 fs_info->fs_devices = fs_devices;
1457 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1458 INIT_LIST_HEAD(&fs_info->space_info);
1459 btrfs_mapping_init(&fs_info->mapping_tree);
1460 atomic_set(&fs_info->nr_async_submits, 0);
1461 atomic_set(&fs_info->async_delalloc_pages, 0);
1462 atomic_set(&fs_info->async_submit_draining, 0);
1463 atomic_set(&fs_info->nr_async_bios, 0);
1464 atomic_set(&fs_info->throttles, 0);
1465 atomic_set(&fs_info->throttle_gen, 0);
1467 fs_info->max_extent = (u64)-1;
1468 fs_info->max_inline = 8192 * 1024;
1469 setup_bdi(fs_info, &fs_info->bdi);
1470 fs_info->btree_inode = new_inode(sb);
1471 fs_info->btree_inode->i_ino = 1;
1472 fs_info->btree_inode->i_nlink = 1;
1474 fs_info->thread_pool_size = min_t(unsigned long,
1475 num_online_cpus() + 2, 8);
1477 INIT_LIST_HEAD(&fs_info->ordered_extents);
1478 spin_lock_init(&fs_info->ordered_extent_lock);
1480 sb->s_blocksize = 4096;
1481 sb->s_blocksize_bits = blksize_bits(4096);
1484 * we set the i_size on the btree inode to the max possible int.
1485 * the real end of the address space is determined by all of
1486 * the devices in the system
1488 fs_info->btree_inode->i_size = OFFSET_MAX;
1489 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1490 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1492 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1493 fs_info->btree_inode->i_mapping,
1495 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1498 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1500 spin_lock_init(&fs_info->block_group_cache_lock);
1501 fs_info->block_group_cache_tree.rb_node = NULL;
1503 extent_io_tree_init(&fs_info->pinned_extents,
1504 fs_info->btree_inode->i_mapping, GFP_NOFS);
1505 extent_io_tree_init(&fs_info->pending_del,
1506 fs_info->btree_inode->i_mapping, GFP_NOFS);
1507 extent_io_tree_init(&fs_info->extent_ins,
1508 fs_info->btree_inode->i_mapping, GFP_NOFS);
1509 fs_info->do_barriers = 1;
1511 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1512 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1513 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1515 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1516 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1517 sizeof(struct btrfs_key));
1518 insert_inode_hash(fs_info->btree_inode);
1520 mutex_init(&fs_info->trans_mutex);
1521 mutex_init(&fs_info->tree_log_mutex);
1522 mutex_init(&fs_info->drop_mutex);
1523 mutex_init(&fs_info->extent_ins_mutex);
1524 mutex_init(&fs_info->pinned_mutex);
1525 mutex_init(&fs_info->chunk_mutex);
1526 mutex_init(&fs_info->transaction_kthread_mutex);
1527 mutex_init(&fs_info->cleaner_mutex);
1528 mutex_init(&fs_info->volume_mutex);
1529 mutex_init(&fs_info->tree_reloc_mutex);
1530 init_waitqueue_head(&fs_info->transaction_throttle);
1531 init_waitqueue_head(&fs_info->transaction_wait);
1532 init_waitqueue_head(&fs_info->async_submit_wait);
1533 init_waitqueue_head(&fs_info->tree_log_wait);
1534 atomic_set(&fs_info->tree_log_commit, 0);
1535 atomic_set(&fs_info->tree_log_writers, 0);
1536 fs_info->tree_log_transid = 0;
1538 __setup_root(4096, 4096, 4096, 4096, tree_root,
1539 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1542 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1546 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1547 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1548 sizeof(fs_info->super_for_commit));
1551 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1553 disk_super = &fs_info->super_copy;
1554 if (!btrfs_super_root(disk_super))
1557 ret = btrfs_parse_options(tree_root, options);
1563 features = btrfs_super_incompat_flags(disk_super) &
1564 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1566 printk(KERN_ERR "BTRFS: couldn't mount because of "
1567 "unsupported optional features (%Lx).\n",
1573 features = btrfs_super_compat_ro_flags(disk_super) &
1574 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1575 if (!(sb->s_flags & MS_RDONLY) && features) {
1576 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1577 "unsupported option features (%Lx).\n",
1584 * we need to start all the end_io workers up front because the
1585 * queue work function gets called at interrupt time, and so it
1586 * cannot dynamically grow.
1588 btrfs_init_workers(&fs_info->workers, "worker",
1589 fs_info->thread_pool_size);
1591 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1592 fs_info->thread_pool_size);
1594 btrfs_init_workers(&fs_info->submit_workers, "submit",
1595 min_t(u64, fs_devices->num_devices,
1596 fs_info->thread_pool_size));
1598 /* a higher idle thresh on the submit workers makes it much more
1599 * likely that bios will be send down in a sane order to the
1602 fs_info->submit_workers.idle_thresh = 64;
1604 fs_info->workers.idle_thresh = 16;
1605 fs_info->workers.ordered = 1;
1607 fs_info->delalloc_workers.idle_thresh = 2;
1608 fs_info->delalloc_workers.ordered = 1;
1610 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1611 btrfs_init_workers(&fs_info->endio_workers, "endio",
1612 fs_info->thread_pool_size);
1613 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1614 fs_info->thread_pool_size);
1615 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1616 "endio-meta-write", fs_info->thread_pool_size);
1617 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1618 fs_info->thread_pool_size);
1621 * endios are largely parallel and should have a very
1624 fs_info->endio_workers.idle_thresh = 4;
1625 fs_info->endio_write_workers.idle_thresh = 64;
1626 fs_info->endio_meta_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_meta_workers,
1634 fs_info->thread_pool_size);
1635 btrfs_start_workers(&fs_info->endio_meta_write_workers,
1636 fs_info->thread_pool_size);
1637 btrfs_start_workers(&fs_info->endio_write_workers,
1638 fs_info->thread_pool_size);
1640 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1641 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1642 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1644 nodesize = btrfs_super_nodesize(disk_super);
1645 leafsize = btrfs_super_leafsize(disk_super);
1646 sectorsize = btrfs_super_sectorsize(disk_super);
1647 stripesize = btrfs_super_stripesize(disk_super);
1648 tree_root->nodesize = nodesize;
1649 tree_root->leafsize = leafsize;
1650 tree_root->sectorsize = sectorsize;
1651 tree_root->stripesize = stripesize;
1653 sb->s_blocksize = sectorsize;
1654 sb->s_blocksize_bits = blksize_bits(sectorsize);
1656 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1657 sizeof(disk_super->magic))) {
1658 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1659 goto fail_sb_buffer;
1662 mutex_lock(&fs_info->chunk_mutex);
1663 ret = btrfs_read_sys_array(tree_root);
1664 mutex_unlock(&fs_info->chunk_mutex);
1666 printk(KERN_WARNING "btrfs: failed to read the system "
1667 "array on %s\n", sb->s_id);
1668 goto fail_sys_array;
1671 blocksize = btrfs_level_size(tree_root,
1672 btrfs_super_chunk_root_level(disk_super));
1673 generation = btrfs_super_chunk_root_generation(disk_super);
1675 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1676 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1678 chunk_root->node = read_tree_block(chunk_root,
1679 btrfs_super_chunk_root(disk_super),
1680 blocksize, generation);
1681 BUG_ON(!chunk_root->node);
1683 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1684 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1687 mutex_lock(&fs_info->chunk_mutex);
1688 ret = btrfs_read_chunk_tree(chunk_root);
1689 mutex_unlock(&fs_info->chunk_mutex);
1691 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1693 goto fail_chunk_root;
1696 btrfs_close_extra_devices(fs_devices);
1698 blocksize = btrfs_level_size(tree_root,
1699 btrfs_super_root_level(disk_super));
1700 generation = btrfs_super_generation(disk_super);
1702 tree_root->node = read_tree_block(tree_root,
1703 btrfs_super_root(disk_super),
1704 blocksize, generation);
1705 if (!tree_root->node)
1706 goto fail_chunk_root;
1709 ret = find_and_setup_root(tree_root, fs_info,
1710 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1712 goto fail_tree_root;
1713 extent_root->track_dirty = 1;
1715 ret = find_and_setup_root(tree_root, fs_info,
1716 BTRFS_DEV_TREE_OBJECTID, dev_root);
1717 dev_root->track_dirty = 1;
1720 goto fail_extent_root;
1722 ret = find_and_setup_root(tree_root, fs_info,
1723 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1725 goto fail_extent_root;
1727 csum_root->track_dirty = 1;
1729 btrfs_read_block_groups(extent_root);
1731 fs_info->generation = generation;
1732 fs_info->last_trans_committed = generation;
1733 fs_info->data_alloc_profile = (u64)-1;
1734 fs_info->metadata_alloc_profile = (u64)-1;
1735 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1736 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1738 if (IS_ERR(fs_info->cleaner_kthread))
1739 goto fail_csum_root;
1741 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1743 "btrfs-transaction");
1744 if (IS_ERR(fs_info->transaction_kthread))
1747 if (btrfs_super_log_root(disk_super) != 0) {
1748 u64 bytenr = btrfs_super_log_root(disk_super);
1750 if (fs_devices->rw_devices == 0) {
1751 printk(KERN_WARNING "Btrfs log replay required "
1754 goto fail_trans_kthread;
1757 btrfs_level_size(tree_root,
1758 btrfs_super_log_root_level(disk_super));
1760 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1763 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1764 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1766 log_tree_root->node = read_tree_block(tree_root, bytenr,
1769 ret = btrfs_recover_log_trees(log_tree_root);
1772 if (sb->s_flags & MS_RDONLY) {
1773 ret = btrfs_commit_super(tree_root);
1778 if (!(sb->s_flags & MS_RDONLY)) {
1779 ret = btrfs_cleanup_reloc_trees(tree_root);
1783 location.objectid = BTRFS_FS_TREE_OBJECTID;
1784 location.type = BTRFS_ROOT_ITEM_KEY;
1785 location.offset = (u64)-1;
1787 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1788 if (!fs_info->fs_root)
1789 goto fail_trans_kthread;
1793 kthread_stop(fs_info->transaction_kthread);
1795 kthread_stop(fs_info->cleaner_kthread);
1798 * make sure we're done with the btree inode before we stop our
1801 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1802 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1805 free_extent_buffer(csum_root->node);
1807 free_extent_buffer(extent_root->node);
1809 free_extent_buffer(tree_root->node);
1811 free_extent_buffer(chunk_root->node);
1813 free_extent_buffer(dev_root->node);
1815 btrfs_stop_workers(&fs_info->fixup_workers);
1816 btrfs_stop_workers(&fs_info->delalloc_workers);
1817 btrfs_stop_workers(&fs_info->workers);
1818 btrfs_stop_workers(&fs_info->endio_workers);
1819 btrfs_stop_workers(&fs_info->endio_meta_workers);
1820 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1821 btrfs_stop_workers(&fs_info->endio_write_workers);
1822 btrfs_stop_workers(&fs_info->submit_workers);
1824 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1825 iput(fs_info->btree_inode);
1827 btrfs_close_devices(fs_info->fs_devices);
1828 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1832 bdi_destroy(&fs_info->bdi);
1837 return ERR_PTR(err);
1840 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1842 char b[BDEVNAME_SIZE];
1845 set_buffer_uptodate(bh);
1847 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1848 printk(KERN_WARNING "lost page write due to "
1849 "I/O error on %s\n",
1850 bdevname(bh->b_bdev, b));
1852 /* note, we dont' set_buffer_write_io_error because we have
1853 * our own ways of dealing with the IO errors
1855 clear_buffer_uptodate(bh);
1861 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1863 struct buffer_head *bh;
1864 struct buffer_head *latest = NULL;
1865 struct btrfs_super_block *super;
1870 /* we would like to check all the supers, but that would make
1871 * a btrfs mount succeed after a mkfs from a different FS.
1872 * So, we need to add a special mount option to scan for
1873 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1875 for (i = 0; i < 1; i++) {
1876 bytenr = btrfs_sb_offset(i);
1877 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1879 bh = __bread(bdev, bytenr / 4096, 4096);
1883 super = (struct btrfs_super_block *)bh->b_data;
1884 if (btrfs_super_bytenr(super) != bytenr ||
1885 strncmp((char *)(&super->magic), BTRFS_MAGIC,
1886 sizeof(super->magic))) {
1891 if (!latest || btrfs_super_generation(super) > transid) {
1894 transid = btrfs_super_generation(super);
1902 static int write_dev_supers(struct btrfs_device *device,
1903 struct btrfs_super_block *sb,
1904 int do_barriers, int wait, int max_mirrors)
1906 struct buffer_head *bh;
1912 int last_barrier = 0;
1914 if (max_mirrors == 0)
1915 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
1917 /* make sure only the last submit_bh does a barrier */
1919 for (i = 0; i < max_mirrors; i++) {
1920 bytenr = btrfs_sb_offset(i);
1921 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1922 device->total_bytes)
1928 for (i = 0; i < max_mirrors; i++) {
1929 bytenr = btrfs_sb_offset(i);
1930 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1934 bh = __find_get_block(device->bdev, bytenr / 4096,
1935 BTRFS_SUPER_INFO_SIZE);
1939 if (buffer_uptodate(bh)) {
1944 btrfs_set_super_bytenr(sb, bytenr);
1947 crc = btrfs_csum_data(NULL, (char *)sb +
1948 BTRFS_CSUM_SIZE, crc,
1949 BTRFS_SUPER_INFO_SIZE -
1951 btrfs_csum_final(crc, sb->csum);
1953 bh = __getblk(device->bdev, bytenr / 4096,
1954 BTRFS_SUPER_INFO_SIZE);
1955 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1957 set_buffer_uptodate(bh);
1960 bh->b_end_io = btrfs_end_buffer_write_sync;
1963 if (i == last_barrier && do_barriers && device->barriers) {
1964 ret = submit_bh(WRITE_BARRIER, bh);
1965 if (ret == -EOPNOTSUPP) {
1966 printk("btrfs: disabling barriers on dev %s\n",
1968 set_buffer_uptodate(bh);
1969 device->barriers = 0;
1972 ret = submit_bh(WRITE, bh);
1975 ret = submit_bh(WRITE, bh);
1980 if (!buffer_uptodate(bh))
1988 return errors < i ? 0 : -1;
1991 int write_all_supers(struct btrfs_root *root, int max_mirrors)
1993 struct list_head *head = &root->fs_info->fs_devices->devices;
1994 struct btrfs_device *dev;
1995 struct btrfs_super_block *sb;
1996 struct btrfs_dev_item *dev_item;
2000 int total_errors = 0;
2003 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2004 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2006 sb = &root->fs_info->super_for_commit;
2007 dev_item = &sb->dev_item;
2008 list_for_each_entry(dev, head, dev_list) {
2013 if (!dev->in_fs_metadata || !dev->writeable)
2016 btrfs_set_stack_device_generation(dev_item, 0);
2017 btrfs_set_stack_device_type(dev_item, dev->type);
2018 btrfs_set_stack_device_id(dev_item, dev->devid);
2019 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2020 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2021 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2022 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2023 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2024 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2025 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2027 flags = btrfs_super_flags(sb);
2028 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2030 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2034 if (total_errors > max_errors) {
2035 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2041 list_for_each_entry(dev, head, dev_list) {
2044 if (!dev->in_fs_metadata || !dev->writeable)
2047 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2051 if (total_errors > max_errors) {
2052 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2059 int write_ctree_super(struct btrfs_trans_handle *trans,
2060 struct btrfs_root *root, int max_mirrors)
2064 ret = write_all_supers(root, max_mirrors);
2068 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2070 radix_tree_delete(&fs_info->fs_roots_radix,
2071 (unsigned long)root->root_key.objectid);
2072 if (root->anon_super.s_dev) {
2073 down_write(&root->anon_super.s_umount);
2074 kill_anon_super(&root->anon_super);
2077 free_extent_buffer(root->node);
2078 if (root->commit_root)
2079 free_extent_buffer(root->commit_root);
2085 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2088 struct btrfs_root *gang[8];
2092 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2097 for (i = 0; i < ret; i++)
2098 btrfs_free_fs_root(fs_info, gang[i]);
2103 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2105 u64 root_objectid = 0;
2106 struct btrfs_root *gang[8];
2111 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2112 (void **)gang, root_objectid,
2116 for (i = 0; i < ret; i++) {
2117 root_objectid = gang[i]->root_key.objectid;
2118 ret = btrfs_find_dead_roots(fs_info->tree_root,
2119 root_objectid, gang[i]);
2121 btrfs_orphan_cleanup(gang[i]);
2128 int btrfs_commit_super(struct btrfs_root *root)
2130 struct btrfs_trans_handle *trans;
2133 mutex_lock(&root->fs_info->cleaner_mutex);
2134 btrfs_clean_old_snapshots(root);
2135 mutex_unlock(&root->fs_info->cleaner_mutex);
2136 trans = btrfs_start_transaction(root, 1);
2137 ret = btrfs_commit_transaction(trans, root);
2139 /* run commit again to drop the original snapshot */
2140 trans = btrfs_start_transaction(root, 1);
2141 btrfs_commit_transaction(trans, root);
2142 ret = btrfs_write_and_wait_transaction(NULL, root);
2145 ret = write_ctree_super(NULL, root, 0);
2149 int close_ctree(struct btrfs_root *root)
2151 struct btrfs_fs_info *fs_info = root->fs_info;
2154 fs_info->closing = 1;
2157 kthread_stop(root->fs_info->transaction_kthread);
2158 kthread_stop(root->fs_info->cleaner_kthread);
2160 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2161 ret = btrfs_commit_super(root);
2163 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2166 if (fs_info->delalloc_bytes) {
2167 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2168 fs_info->delalloc_bytes);
2170 if (fs_info->total_ref_cache_size) {
2171 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2172 (unsigned long long)fs_info->total_ref_cache_size);
2175 if (fs_info->extent_root->node)
2176 free_extent_buffer(fs_info->extent_root->node);
2178 if (fs_info->tree_root->node)
2179 free_extent_buffer(fs_info->tree_root->node);
2181 if (root->fs_info->chunk_root->node)
2182 free_extent_buffer(root->fs_info->chunk_root->node);
2184 if (root->fs_info->dev_root->node)
2185 free_extent_buffer(root->fs_info->dev_root->node);
2187 if (root->fs_info->csum_root->node)
2188 free_extent_buffer(root->fs_info->csum_root->node);
2190 btrfs_free_block_groups(root->fs_info);
2192 del_fs_roots(fs_info);
2194 iput(fs_info->btree_inode);
2196 btrfs_stop_workers(&fs_info->fixup_workers);
2197 btrfs_stop_workers(&fs_info->delalloc_workers);
2198 btrfs_stop_workers(&fs_info->workers);
2199 btrfs_stop_workers(&fs_info->endio_workers);
2200 btrfs_stop_workers(&fs_info->endio_meta_workers);
2201 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2202 btrfs_stop_workers(&fs_info->endio_write_workers);
2203 btrfs_stop_workers(&fs_info->submit_workers);
2206 while (!list_empty(&fs_info->hashers)) {
2207 struct btrfs_hasher *hasher;
2208 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2210 list_del(&hasher->hashers);
2211 crypto_free_hash(&fs_info->hash_tfm);
2215 btrfs_close_devices(fs_info->fs_devices);
2216 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2218 bdi_destroy(&fs_info->bdi);
2220 kfree(fs_info->extent_root);
2221 kfree(fs_info->tree_root);
2222 kfree(fs_info->chunk_root);
2223 kfree(fs_info->dev_root);
2224 kfree(fs_info->csum_root);
2228 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2231 struct inode *btree_inode = buf->first_page->mapping->host;
2233 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2237 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2242 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2244 struct inode *btree_inode = buf->first_page->mapping->host;
2245 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2249 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2251 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2252 u64 transid = btrfs_header_generation(buf);
2253 struct inode *btree_inode = root->fs_info->btree_inode;
2255 WARN_ON(!btrfs_tree_locked(buf));
2256 if (transid != root->fs_info->generation) {
2257 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2258 "found %llu running %llu\n",
2259 (unsigned long long)buf->start,
2260 (unsigned long long)transid,
2261 (unsigned long long)root->fs_info->generation);
2264 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2267 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2270 * looks as though older kernels can get into trouble with
2271 * this code, they end up stuck in balance_dirty_pages forever
2273 struct extent_io_tree *tree;
2276 unsigned long thresh = 32 * 1024 * 1024;
2277 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2279 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2282 num_dirty = count_range_bits(tree, &start, (u64)-1,
2283 thresh, EXTENT_DIRTY);
2284 if (num_dirty > thresh) {
2285 balance_dirty_pages_ratelimited_nr(
2286 root->fs_info->btree_inode->i_mapping, 1);
2291 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2293 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2295 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2297 buf->flags |= EXTENT_UPTODATE;
2301 int btree_lock_page_hook(struct page *page)
2303 struct inode *inode = page->mapping->host;
2304 struct btrfs_root *root = BTRFS_I(inode)->root;
2305 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2306 struct extent_buffer *eb;
2308 u64 bytenr = page_offset(page);
2310 if (page->private == EXTENT_PAGE_PRIVATE)
2313 len = page->private >> 2;
2314 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2318 btrfs_tree_lock(eb);
2319 spin_lock(&root->fs_info->hash_lock);
2320 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2321 spin_unlock(&root->fs_info->hash_lock);
2322 btrfs_tree_unlock(eb);
2323 free_extent_buffer(eb);
2329 static struct extent_io_ops btree_extent_io_ops = {
2330 .write_cache_pages_lock_hook = btree_lock_page_hook,
2331 .readpage_end_io_hook = btree_readpage_end_io_hook,
2332 .submit_bio_hook = btree_submit_bio_hook,
2333 /* note we're sharing with inode.c for the merge bio hook */
2334 .merge_bio_hook = btrfs_merge_bio_hook,