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_hook;
86 unsigned long bio_flags;
87 struct btrfs_work work;
91 * extents on the btree inode are pretty simple, there's one extent
92 * that covers the entire device
94 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
95 size_t page_offset, u64 start, u64 len,
98 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
99 struct extent_map *em;
102 spin_lock(&em_tree->lock);
103 em = lookup_extent_mapping(em_tree, start, len);
106 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
107 spin_unlock(&em_tree->lock);
110 spin_unlock(&em_tree->lock);
112 em = alloc_extent_map(GFP_NOFS);
114 em = ERR_PTR(-ENOMEM);
119 em->block_len = (u64)-1;
121 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
123 spin_lock(&em_tree->lock);
124 ret = add_extent_mapping(em_tree, em);
125 if (ret == -EEXIST) {
126 u64 failed_start = em->start;
127 u64 failed_len = em->len;
129 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
130 em->start, em->len, em->block_start);
132 em = lookup_extent_mapping(em_tree, start, len);
134 printk("after failing, found %Lu %Lu %Lu\n",
135 em->start, em->len, em->block_start);
138 em = lookup_extent_mapping(em_tree, failed_start,
141 printk("double failure lookup gives us "
142 "%Lu %Lu -> %Lu\n", em->start,
143 em->len, em->block_start);
152 spin_unlock(&em_tree->lock);
160 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
162 return btrfs_crc32c(seed, data, len);
165 void btrfs_csum_final(u32 crc, char *result)
167 *(__le32 *)result = ~cpu_to_le32(crc);
171 * compute the csum for a btree block, and either verify it or write it
172 * into the csum field of the block.
174 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
177 char result[BTRFS_CRC32_SIZE];
179 unsigned long cur_len;
180 unsigned long offset = BTRFS_CSUM_SIZE;
181 char *map_token = NULL;
183 unsigned long map_start;
184 unsigned long map_len;
188 len = buf->len - offset;
190 err = map_private_extent_buffer(buf, offset, 32,
192 &map_start, &map_len, KM_USER0);
194 printk("failed to map extent buffer! %lu\n",
198 cur_len = min(len, map_len - (offset - map_start));
199 crc = btrfs_csum_data(root, kaddr + offset - map_start,
203 unmap_extent_buffer(buf, map_token, KM_USER0);
205 btrfs_csum_final(crc, result);
208 /* FIXME, this is not good */
209 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
212 memcpy(&found, result, BTRFS_CRC32_SIZE);
214 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
215 printk("btrfs: %s checksum verify failed on %llu "
216 "wanted %X found %X level %d\n",
217 root->fs_info->sb->s_id,
218 buf->start, val, found, btrfs_header_level(buf));
222 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
228 * we can't consider a given block up to date unless the transid of the
229 * block matches the transid in the parent node's pointer. This is how we
230 * detect blocks that either didn't get written at all or got written
231 * in the wrong place.
233 static int verify_parent_transid(struct extent_io_tree *io_tree,
234 struct extent_buffer *eb, u64 parent_transid)
238 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
241 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
242 if (extent_buffer_uptodate(io_tree, eb) &&
243 btrfs_header_generation(eb) == parent_transid) {
247 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
248 (unsigned long long)eb->start,
249 (unsigned long long)parent_transid,
250 (unsigned long long)btrfs_header_generation(eb));
252 clear_extent_buffer_uptodate(io_tree, eb);
254 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
260 * helper to read a given tree block, doing retries as required when
261 * the checksums don't match and we have alternate mirrors to try.
263 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
264 struct extent_buffer *eb,
265 u64 start, u64 parent_transid)
267 struct extent_io_tree *io_tree;
272 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
274 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
275 btree_get_extent, mirror_num);
277 !verify_parent_transid(io_tree, eb, parent_transid))
279 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
280 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
286 if (mirror_num > num_copies)
293 * checksum a dirty tree block before IO. This has extra checks to make
294 * sure we only fill in the checksum field in the first page of a multi-page block
296 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
298 struct extent_io_tree *tree;
299 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
303 struct extent_buffer *eb;
306 tree = &BTRFS_I(page->mapping->host)->io_tree;
308 if (page->private == EXTENT_PAGE_PRIVATE)
312 len = page->private >> 2;
316 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
317 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
318 btrfs_header_generation(eb));
320 found_start = btrfs_header_bytenr(eb);
321 if (found_start != start) {
322 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
323 start, found_start, len);
327 if (eb->first_page != page) {
328 printk("bad first page %lu %lu\n", eb->first_page->index,
333 if (!PageUptodate(page)) {
334 printk("csum not up to date page %lu\n", page->index);
338 found_level = btrfs_header_level(eb);
340 csum_tree_block(root, eb, 0);
342 free_extent_buffer(eb);
347 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
348 struct extent_state *state)
350 struct extent_io_tree *tree;
354 struct extent_buffer *eb;
355 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
358 tree = &BTRFS_I(page->mapping->host)->io_tree;
359 if (page->private == EXTENT_PAGE_PRIVATE)
363 len = page->private >> 2;
367 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
369 found_start = btrfs_header_bytenr(eb);
370 if (found_start != start) {
371 printk("bad tree block start %llu %llu\n",
372 (unsigned long long)found_start,
373 (unsigned long long)eb->start);
377 if (eb->first_page != page) {
378 printk("bad first page %lu %lu\n", eb->first_page->index,
384 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
385 (unsigned long)btrfs_header_fsid(eb),
387 printk("bad fsid on block %Lu\n", eb->start);
391 found_level = btrfs_header_level(eb);
393 ret = csum_tree_block(root, eb, 1);
397 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
398 end = eb->start + end - 1;
400 free_extent_buffer(eb);
405 static void end_workqueue_bio(struct bio *bio, int err)
407 struct end_io_wq *end_io_wq = bio->bi_private;
408 struct btrfs_fs_info *fs_info;
410 fs_info = end_io_wq->info;
411 end_io_wq->error = err;
412 end_io_wq->work.func = end_workqueue_fn;
413 end_io_wq->work.flags = 0;
414 if (bio->bi_rw & (1 << BIO_RW))
415 btrfs_queue_worker(&fs_info->endio_write_workers,
418 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
421 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
424 struct end_io_wq *end_io_wq;
425 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
429 end_io_wq->private = bio->bi_private;
430 end_io_wq->end_io = bio->bi_end_io;
431 end_io_wq->info = info;
432 end_io_wq->error = 0;
433 end_io_wq->bio = bio;
434 end_io_wq->metadata = metadata;
436 bio->bi_private = end_io_wq;
437 bio->bi_end_io = end_workqueue_bio;
441 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
443 unsigned long limit = min_t(unsigned long,
444 info->workers.max_workers,
445 info->fs_devices->open_devices);
449 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
451 return atomic_read(&info->nr_async_bios) >
452 btrfs_async_submit_limit(info);
455 static void run_one_async_submit(struct btrfs_work *work)
457 struct btrfs_fs_info *fs_info;
458 struct async_submit_bio *async;
461 async = container_of(work, struct async_submit_bio, work);
462 fs_info = BTRFS_I(async->inode)->root->fs_info;
464 limit = btrfs_async_submit_limit(fs_info);
465 limit = limit * 2 / 3;
467 atomic_dec(&fs_info->nr_async_submits);
469 if (atomic_read(&fs_info->nr_async_submits) < limit &&
470 waitqueue_active(&fs_info->async_submit_wait))
471 wake_up(&fs_info->async_submit_wait);
473 async->submit_bio_hook(async->inode, async->rw, async->bio,
474 async->mirror_num, async->bio_flags);
478 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
479 int rw, struct bio *bio, int mirror_num,
480 unsigned long bio_flags,
481 extent_submit_bio_hook_t *submit_bio_hook)
483 struct async_submit_bio *async;
484 int limit = btrfs_async_submit_limit(fs_info);
486 async = kmalloc(sizeof(*async), GFP_NOFS);
490 async->inode = inode;
493 async->mirror_num = mirror_num;
494 async->submit_bio_hook = submit_bio_hook;
495 async->work.func = run_one_async_submit;
496 async->work.flags = 0;
497 async->bio_flags = bio_flags;
499 while(atomic_read(&fs_info->async_submit_draining) &&
500 atomic_read(&fs_info->nr_async_submits)) {
501 wait_event(fs_info->async_submit_wait,
502 (atomic_read(&fs_info->nr_async_submits) == 0));
505 atomic_inc(&fs_info->nr_async_submits);
506 btrfs_queue_worker(&fs_info->workers, &async->work);
508 if (atomic_read(&fs_info->nr_async_submits) > limit) {
509 wait_event_timeout(fs_info->async_submit_wait,
510 (atomic_read(&fs_info->nr_async_submits) < limit),
513 wait_event_timeout(fs_info->async_submit_wait,
514 (atomic_read(&fs_info->nr_async_bios) < limit),
520 static int btree_csum_one_bio(struct bio *bio)
522 struct bio_vec *bvec = bio->bi_io_vec;
524 struct btrfs_root *root;
526 WARN_ON(bio->bi_vcnt <= 0);
527 while(bio_index < bio->bi_vcnt) {
528 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
529 csum_dirty_buffer(root, bvec->bv_page);
536 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
537 int mirror_num, unsigned long bio_flags)
539 struct btrfs_root *root = BTRFS_I(inode)->root;
543 * when we're called for a write, we're already in the async
544 * submission context. Just jump into btrfs_map_bio
546 if (rw & (1 << BIO_RW)) {
547 btree_csum_one_bio(bio);
548 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
553 * called for a read, do the setup so that checksum validation
554 * can happen in the async kernel threads
556 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
559 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
562 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
563 int mirror_num, unsigned long bio_flags)
566 * kthread helpers are used to submit writes so that checksumming
567 * can happen in parallel across all CPUs
569 if (!(rw & (1 << BIO_RW))) {
570 return __btree_submit_bio_hook(inode, rw, bio, mirror_num, 0);
572 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
573 inode, rw, bio, mirror_num, 0,
574 __btree_submit_bio_hook);
577 static int btree_writepage(struct page *page, struct writeback_control *wbc)
579 struct extent_io_tree *tree;
580 tree = &BTRFS_I(page->mapping->host)->io_tree;
582 if (current->flags & PF_MEMALLOC) {
583 redirty_page_for_writepage(wbc, page);
587 return extent_write_full_page(tree, page, btree_get_extent, wbc);
590 static int btree_writepages(struct address_space *mapping,
591 struct writeback_control *wbc)
593 struct extent_io_tree *tree;
594 tree = &BTRFS_I(mapping->host)->io_tree;
595 if (wbc->sync_mode == WB_SYNC_NONE) {
598 unsigned long thresh = 32 * 1024 * 1024;
600 if (wbc->for_kupdate)
603 num_dirty = count_range_bits(tree, &start, (u64)-1,
604 thresh, EXTENT_DIRTY);
605 if (num_dirty < thresh) {
609 return extent_writepages(tree, mapping, btree_get_extent, wbc);
612 int btree_readpage(struct file *file, struct page *page)
614 struct extent_io_tree *tree;
615 tree = &BTRFS_I(page->mapping->host)->io_tree;
616 return extent_read_full_page(tree, page, btree_get_extent);
619 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
621 struct extent_io_tree *tree;
622 struct extent_map_tree *map;
625 if (PageWriteback(page) || PageDirty(page))
628 tree = &BTRFS_I(page->mapping->host)->io_tree;
629 map = &BTRFS_I(page->mapping->host)->extent_tree;
631 ret = try_release_extent_state(map, tree, page, gfp_flags);
636 ret = try_release_extent_buffer(tree, page);
638 ClearPagePrivate(page);
639 set_page_private(page, 0);
640 page_cache_release(page);
646 static void btree_invalidatepage(struct page *page, unsigned long offset)
648 struct extent_io_tree *tree;
649 tree = &BTRFS_I(page->mapping->host)->io_tree;
650 extent_invalidatepage(tree, page, offset);
651 btree_releasepage(page, GFP_NOFS);
652 if (PagePrivate(page)) {
653 printk("warning page private not zero on page %Lu\n",
655 ClearPagePrivate(page);
656 set_page_private(page, 0);
657 page_cache_release(page);
662 static int btree_writepage(struct page *page, struct writeback_control *wbc)
664 struct buffer_head *bh;
665 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
666 struct buffer_head *head;
667 if (!page_has_buffers(page)) {
668 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
669 (1 << BH_Dirty)|(1 << BH_Uptodate));
671 head = page_buffers(page);
674 if (buffer_dirty(bh))
675 csum_tree_block(root, bh, 0);
676 bh = bh->b_this_page;
677 } while (bh != head);
678 return block_write_full_page(page, btree_get_block, wbc);
682 static struct address_space_operations btree_aops = {
683 .readpage = btree_readpage,
684 .writepage = btree_writepage,
685 .writepages = btree_writepages,
686 .releasepage = btree_releasepage,
687 .invalidatepage = btree_invalidatepage,
688 .sync_page = block_sync_page,
691 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
694 struct extent_buffer *buf = NULL;
695 struct inode *btree_inode = root->fs_info->btree_inode;
698 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
701 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
702 buf, 0, 0, btree_get_extent, 0);
703 free_extent_buffer(buf);
707 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
708 u64 bytenr, u32 blocksize)
710 struct inode *btree_inode = root->fs_info->btree_inode;
711 struct extent_buffer *eb;
712 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
713 bytenr, blocksize, GFP_NOFS);
717 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
718 u64 bytenr, u32 blocksize)
720 struct inode *btree_inode = root->fs_info->btree_inode;
721 struct extent_buffer *eb;
723 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
724 bytenr, blocksize, NULL, GFP_NOFS);
729 int btrfs_write_tree_block(struct extent_buffer *buf)
731 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
732 buf->start + buf->len - 1, WB_SYNC_ALL);
735 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
737 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
738 buf->start, buf->start + buf->len -1);
741 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
742 u32 blocksize, u64 parent_transid)
744 struct extent_buffer *buf = NULL;
745 struct inode *btree_inode = root->fs_info->btree_inode;
746 struct extent_io_tree *io_tree;
749 io_tree = &BTRFS_I(btree_inode)->io_tree;
751 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
755 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
758 buf->flags |= EXTENT_UPTODATE;
766 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
767 struct extent_buffer *buf)
769 struct inode *btree_inode = root->fs_info->btree_inode;
770 if (btrfs_header_generation(buf) ==
771 root->fs_info->running_transaction->transid) {
772 WARN_ON(!btrfs_tree_locked(buf));
773 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
779 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
780 u32 stripesize, struct btrfs_root *root,
781 struct btrfs_fs_info *fs_info,
786 root->commit_root = NULL;
787 root->ref_tree = NULL;
788 root->sectorsize = sectorsize;
789 root->nodesize = nodesize;
790 root->leafsize = leafsize;
791 root->stripesize = stripesize;
793 root->track_dirty = 0;
795 root->fs_info = fs_info;
796 root->objectid = objectid;
797 root->last_trans = 0;
798 root->highest_inode = 0;
799 root->last_inode_alloc = 0;
803 INIT_LIST_HEAD(&root->dirty_list);
804 INIT_LIST_HEAD(&root->orphan_list);
805 INIT_LIST_HEAD(&root->dead_list);
806 spin_lock_init(&root->node_lock);
807 spin_lock_init(&root->list_lock);
808 mutex_init(&root->objectid_mutex);
809 mutex_init(&root->log_mutex);
810 extent_io_tree_init(&root->dirty_log_pages,
811 fs_info->btree_inode->i_mapping, GFP_NOFS);
813 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
814 root->ref_tree = &root->ref_tree_struct;
816 memset(&root->root_key, 0, sizeof(root->root_key));
817 memset(&root->root_item, 0, sizeof(root->root_item));
818 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
819 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
820 root->defrag_trans_start = fs_info->generation;
821 init_completion(&root->kobj_unregister);
822 root->defrag_running = 0;
823 root->defrag_level = 0;
824 root->root_key.objectid = objectid;
828 static int find_and_setup_root(struct btrfs_root *tree_root,
829 struct btrfs_fs_info *fs_info,
831 struct btrfs_root *root)
837 __setup_root(tree_root->nodesize, tree_root->leafsize,
838 tree_root->sectorsize, tree_root->stripesize,
839 root, fs_info, objectid);
840 ret = btrfs_find_last_root(tree_root, objectid,
841 &root->root_item, &root->root_key);
844 generation = btrfs_root_generation(&root->root_item);
845 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
846 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
847 blocksize, generation);
852 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
853 struct btrfs_fs_info *fs_info)
855 struct extent_buffer *eb;
856 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
865 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
866 0, &start, &end, EXTENT_DIRTY);
870 clear_extent_dirty(&log_root_tree->dirty_log_pages,
871 start, end, GFP_NOFS);
873 eb = fs_info->log_root_tree->node;
875 WARN_ON(btrfs_header_level(eb) != 0);
876 WARN_ON(btrfs_header_nritems(eb) != 0);
878 ret = btrfs_free_reserved_extent(fs_info->tree_root,
882 free_extent_buffer(eb);
883 kfree(fs_info->log_root_tree);
884 fs_info->log_root_tree = NULL;
888 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
889 struct btrfs_fs_info *fs_info)
891 struct btrfs_root *root;
892 struct btrfs_root *tree_root = fs_info->tree_root;
894 root = kzalloc(sizeof(*root), GFP_NOFS);
898 __setup_root(tree_root->nodesize, tree_root->leafsize,
899 tree_root->sectorsize, tree_root->stripesize,
900 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
902 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
903 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
904 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
907 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
908 0, BTRFS_TREE_LOG_OBJECTID,
909 trans->transid, 0, 0, 0);
911 btrfs_set_header_nritems(root->node, 0);
912 btrfs_set_header_level(root->node, 0);
913 btrfs_set_header_bytenr(root->node, root->node->start);
914 btrfs_set_header_generation(root->node, trans->transid);
915 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
917 write_extent_buffer(root->node, root->fs_info->fsid,
918 (unsigned long)btrfs_header_fsid(root->node),
920 btrfs_mark_buffer_dirty(root->node);
921 btrfs_tree_unlock(root->node);
922 fs_info->log_root_tree = root;
926 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
927 struct btrfs_key *location)
929 struct btrfs_root *root;
930 struct btrfs_fs_info *fs_info = tree_root->fs_info;
931 struct btrfs_path *path;
932 struct extent_buffer *l;
938 root = kzalloc(sizeof(*root), GFP_NOFS);
940 return ERR_PTR(-ENOMEM);
941 if (location->offset == (u64)-1) {
942 ret = find_and_setup_root(tree_root, fs_info,
943 location->objectid, root);
951 __setup_root(tree_root->nodesize, tree_root->leafsize,
952 tree_root->sectorsize, tree_root->stripesize,
953 root, fs_info, location->objectid);
955 path = btrfs_alloc_path();
957 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
964 read_extent_buffer(l, &root->root_item,
965 btrfs_item_ptr_offset(l, path->slots[0]),
966 sizeof(root->root_item));
967 memcpy(&root->root_key, location, sizeof(*location));
970 btrfs_release_path(root, path);
971 btrfs_free_path(path);
976 generation = btrfs_root_generation(&root->root_item);
977 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
978 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
979 blocksize, generation);
982 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
984 ret = btrfs_find_highest_inode(root, &highest_inode);
986 root->highest_inode = highest_inode;
987 root->last_inode_alloc = highest_inode;
993 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
996 struct btrfs_root *root;
998 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
999 return fs_info->tree_root;
1000 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1001 return fs_info->extent_root;
1003 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1004 (unsigned long)root_objectid);
1008 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1009 struct btrfs_key *location)
1011 struct btrfs_root *root;
1014 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1015 return fs_info->tree_root;
1016 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1017 return fs_info->extent_root;
1018 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1019 return fs_info->chunk_root;
1020 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1021 return fs_info->dev_root;
1023 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1024 (unsigned long)location->objectid);
1028 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1031 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1032 (unsigned long)root->root_key.objectid,
1035 free_extent_buffer(root->node);
1037 return ERR_PTR(ret);
1039 ret = btrfs_find_dead_roots(fs_info->tree_root,
1040 root->root_key.objectid, root);
1046 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1047 struct btrfs_key *location,
1048 const char *name, int namelen)
1050 struct btrfs_root *root;
1053 root = btrfs_read_fs_root_no_name(fs_info, location);
1060 ret = btrfs_set_root_name(root, name, namelen);
1062 free_extent_buffer(root->node);
1064 return ERR_PTR(ret);
1067 ret = btrfs_sysfs_add_root(root);
1069 free_extent_buffer(root->node);
1072 return ERR_PTR(ret);
1078 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1079 struct btrfs_hasher *hasher;
1081 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1084 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1085 if (!hasher->hash_tfm) {
1089 spin_lock(&info->hash_lock);
1090 list_add(&hasher->list, &info->hashers);
1091 spin_unlock(&info->hash_lock);
1096 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1098 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1100 struct list_head *cur;
1101 struct btrfs_device *device;
1102 struct backing_dev_info *bdi;
1104 if ((bdi_bits & (1 << BDI_write_congested)) &&
1105 btrfs_congested_async(info, 0))
1108 list_for_each(cur, &info->fs_devices->devices) {
1109 device = list_entry(cur, struct btrfs_device, dev_list);
1112 bdi = blk_get_backing_dev_info(device->bdev);
1113 if (bdi && bdi_congested(bdi, bdi_bits)) {
1122 * this unplugs every device on the box, and it is only used when page
1125 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1127 struct list_head *cur;
1128 struct btrfs_device *device;
1129 struct btrfs_fs_info *info;
1131 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1132 list_for_each(cur, &info->fs_devices->devices) {
1133 device = list_entry(cur, struct btrfs_device, dev_list);
1134 bdi = blk_get_backing_dev_info(device->bdev);
1135 if (bdi->unplug_io_fn) {
1136 bdi->unplug_io_fn(bdi, page);
1141 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1143 struct inode *inode;
1144 struct extent_map_tree *em_tree;
1145 struct extent_map *em;
1146 struct address_space *mapping;
1149 /* the generic O_DIRECT read code does this */
1151 __unplug_io_fn(bdi, page);
1156 * page->mapping may change at any time. Get a consistent copy
1157 * and use that for everything below
1160 mapping = page->mapping;
1164 inode = mapping->host;
1165 offset = page_offset(page);
1167 em_tree = &BTRFS_I(inode)->extent_tree;
1168 spin_lock(&em_tree->lock);
1169 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1170 spin_unlock(&em_tree->lock);
1172 __unplug_io_fn(bdi, page);
1176 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1177 free_extent_map(em);
1178 __unplug_io_fn(bdi, page);
1181 offset = offset - em->start;
1182 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1183 em->block_start + offset, page);
1184 free_extent_map(em);
1187 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1190 bdi->ra_pages = default_backing_dev_info.ra_pages;
1192 bdi->capabilities = default_backing_dev_info.capabilities;
1193 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1194 bdi->unplug_io_data = info;
1195 bdi->congested_fn = btrfs_congested_fn;
1196 bdi->congested_data = info;
1200 static int bio_ready_for_csum(struct bio *bio)
1206 struct extent_io_tree *io_tree = NULL;
1207 struct btrfs_fs_info *info = NULL;
1208 struct bio_vec *bvec;
1212 bio_for_each_segment(bvec, bio, i) {
1213 page = bvec->bv_page;
1214 if (page->private == EXTENT_PAGE_PRIVATE) {
1215 length += bvec->bv_len;
1218 if (!page->private) {
1219 length += bvec->bv_len;
1222 length = bvec->bv_len;
1223 buf_len = page->private >> 2;
1224 start = page_offset(page) + bvec->bv_offset;
1225 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1226 info = BTRFS_I(page->mapping->host)->root->fs_info;
1228 /* are we fully contained in this bio? */
1229 if (buf_len <= length)
1232 ret = extent_range_uptodate(io_tree, start + length,
1233 start + buf_len - 1);
1240 * called by the kthread helper functions to finally call the bio end_io
1241 * functions. This is where read checksum verification actually happens
1243 static void end_workqueue_fn(struct btrfs_work *work)
1246 struct end_io_wq *end_io_wq;
1247 struct btrfs_fs_info *fs_info;
1250 end_io_wq = container_of(work, struct end_io_wq, work);
1251 bio = end_io_wq->bio;
1252 fs_info = end_io_wq->info;
1254 /* metadata bios are special because the whole tree block must
1255 * be checksummed at once. This makes sure the entire block is in
1256 * ram and up to date before trying to verify things. For
1257 * blocksize <= pagesize, it is basically a noop
1259 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1260 btrfs_queue_worker(&fs_info->endio_workers,
1264 error = end_io_wq->error;
1265 bio->bi_private = end_io_wq->private;
1266 bio->bi_end_io = end_io_wq->end_io;
1268 bio_endio(bio, error);
1271 static int cleaner_kthread(void *arg)
1273 struct btrfs_root *root = arg;
1277 if (root->fs_info->closing)
1280 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1281 mutex_lock(&root->fs_info->cleaner_mutex);
1282 btrfs_clean_old_snapshots(root);
1283 mutex_unlock(&root->fs_info->cleaner_mutex);
1285 if (freezing(current)) {
1289 if (root->fs_info->closing)
1291 set_current_state(TASK_INTERRUPTIBLE);
1293 __set_current_state(TASK_RUNNING);
1295 } while (!kthread_should_stop());
1299 static int transaction_kthread(void *arg)
1301 struct btrfs_root *root = arg;
1302 struct btrfs_trans_handle *trans;
1303 struct btrfs_transaction *cur;
1305 unsigned long delay;
1310 if (root->fs_info->closing)
1314 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1315 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1317 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1318 printk("btrfs: total reference cache size %Lu\n",
1319 root->fs_info->total_ref_cache_size);
1322 mutex_lock(&root->fs_info->trans_mutex);
1323 cur = root->fs_info->running_transaction;
1325 mutex_unlock(&root->fs_info->trans_mutex);
1329 now = get_seconds();
1330 if (now < cur->start_time || now - cur->start_time < 30) {
1331 mutex_unlock(&root->fs_info->trans_mutex);
1335 mutex_unlock(&root->fs_info->trans_mutex);
1336 trans = btrfs_start_transaction(root, 1);
1337 ret = btrfs_commit_transaction(trans, root);
1339 wake_up_process(root->fs_info->cleaner_kthread);
1340 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1342 if (freezing(current)) {
1345 if (root->fs_info->closing)
1347 set_current_state(TASK_INTERRUPTIBLE);
1348 schedule_timeout(delay);
1349 __set_current_state(TASK_RUNNING);
1351 } while (!kthread_should_stop());
1355 struct btrfs_root *open_ctree(struct super_block *sb,
1356 struct btrfs_fs_devices *fs_devices,
1365 struct buffer_head *bh;
1366 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1368 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1370 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1372 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1374 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1376 struct btrfs_root *log_tree_root;
1381 struct btrfs_super_block *disk_super;
1383 if (!extent_root || !tree_root || !fs_info ||
1384 !chunk_root || !dev_root) {
1388 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1389 INIT_LIST_HEAD(&fs_info->trans_list);
1390 INIT_LIST_HEAD(&fs_info->dead_roots);
1391 INIT_LIST_HEAD(&fs_info->hashers);
1392 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1393 spin_lock_init(&fs_info->hash_lock);
1394 spin_lock_init(&fs_info->delalloc_lock);
1395 spin_lock_init(&fs_info->new_trans_lock);
1396 spin_lock_init(&fs_info->ref_cache_lock);
1398 init_completion(&fs_info->kobj_unregister);
1399 fs_info->tree_root = tree_root;
1400 fs_info->extent_root = extent_root;
1401 fs_info->chunk_root = chunk_root;
1402 fs_info->dev_root = dev_root;
1403 fs_info->fs_devices = fs_devices;
1404 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1405 INIT_LIST_HEAD(&fs_info->space_info);
1406 btrfs_mapping_init(&fs_info->mapping_tree);
1407 atomic_set(&fs_info->nr_async_submits, 0);
1408 atomic_set(&fs_info->async_submit_draining, 0);
1409 atomic_set(&fs_info->nr_async_bios, 0);
1410 atomic_set(&fs_info->throttles, 0);
1411 atomic_set(&fs_info->throttle_gen, 0);
1413 fs_info->max_extent = (u64)-1;
1414 fs_info->max_inline = 8192 * 1024;
1415 setup_bdi(fs_info, &fs_info->bdi);
1416 fs_info->btree_inode = new_inode(sb);
1417 fs_info->btree_inode->i_ino = 1;
1418 fs_info->btree_inode->i_nlink = 1;
1420 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1422 INIT_LIST_HEAD(&fs_info->ordered_extents);
1423 spin_lock_init(&fs_info->ordered_extent_lock);
1425 sb->s_blocksize = 4096;
1426 sb->s_blocksize_bits = blksize_bits(4096);
1429 * we set the i_size on the btree inode to the max possible int.
1430 * the real end of the address space is determined by all of
1431 * the devices in the system
1433 fs_info->btree_inode->i_size = OFFSET_MAX;
1434 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1435 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1437 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1438 fs_info->btree_inode->i_mapping,
1440 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1443 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1445 spin_lock_init(&fs_info->block_group_cache_lock);
1446 fs_info->block_group_cache_tree.rb_node = NULL;
1448 extent_io_tree_init(&fs_info->pinned_extents,
1449 fs_info->btree_inode->i_mapping, GFP_NOFS);
1450 extent_io_tree_init(&fs_info->pending_del,
1451 fs_info->btree_inode->i_mapping, GFP_NOFS);
1452 extent_io_tree_init(&fs_info->extent_ins,
1453 fs_info->btree_inode->i_mapping, GFP_NOFS);
1454 fs_info->do_barriers = 1;
1456 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1457 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1458 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1460 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1461 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1462 sizeof(struct btrfs_key));
1463 insert_inode_hash(fs_info->btree_inode);
1465 mutex_init(&fs_info->trans_mutex);
1466 mutex_init(&fs_info->tree_log_mutex);
1467 mutex_init(&fs_info->drop_mutex);
1468 mutex_init(&fs_info->extent_ins_mutex);
1469 mutex_init(&fs_info->pinned_mutex);
1470 mutex_init(&fs_info->chunk_mutex);
1471 mutex_init(&fs_info->transaction_kthread_mutex);
1472 mutex_init(&fs_info->cleaner_mutex);
1473 mutex_init(&fs_info->volume_mutex);
1474 mutex_init(&fs_info->tree_reloc_mutex);
1475 init_waitqueue_head(&fs_info->transaction_throttle);
1476 init_waitqueue_head(&fs_info->transaction_wait);
1477 init_waitqueue_head(&fs_info->async_submit_wait);
1478 init_waitqueue_head(&fs_info->tree_log_wait);
1479 atomic_set(&fs_info->tree_log_commit, 0);
1480 atomic_set(&fs_info->tree_log_writers, 0);
1481 fs_info->tree_log_transid = 0;
1484 ret = add_hasher(fs_info, "crc32c");
1486 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1491 __setup_root(4096, 4096, 4096, 4096, tree_root,
1492 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1495 bh = __bread(fs_devices->latest_bdev,
1496 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1500 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1503 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1505 disk_super = &fs_info->super_copy;
1506 if (!btrfs_super_root(disk_super))
1507 goto fail_sb_buffer;
1509 err = btrfs_parse_options(tree_root, options);
1511 goto fail_sb_buffer;
1514 * we need to start all the end_io workers up front because the
1515 * queue work function gets called at interrupt time, and so it
1516 * cannot dynamically grow.
1518 btrfs_init_workers(&fs_info->workers, "worker",
1519 fs_info->thread_pool_size);
1521 btrfs_init_workers(&fs_info->submit_workers, "submit",
1522 min_t(u64, fs_devices->num_devices,
1523 fs_info->thread_pool_size));
1525 /* a higher idle thresh on the submit workers makes it much more
1526 * likely that bios will be send down in a sane order to the
1529 fs_info->submit_workers.idle_thresh = 64;
1531 /* fs_info->workers is responsible for checksumming file data
1532 * blocks and metadata. Using a larger idle thresh allows each
1533 * worker thread to operate on things in roughly the order they
1534 * were sent by the writeback daemons, improving overall locality
1535 * of the IO going down the pipe.
1537 fs_info->workers.idle_thresh = 128;
1539 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1540 btrfs_init_workers(&fs_info->endio_workers, "endio",
1541 fs_info->thread_pool_size);
1542 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1543 fs_info->thread_pool_size);
1546 * endios are largely parallel and should have a very
1549 fs_info->endio_workers.idle_thresh = 4;
1550 fs_info->endio_write_workers.idle_thresh = 64;
1552 btrfs_start_workers(&fs_info->workers, 1);
1553 btrfs_start_workers(&fs_info->submit_workers, 1);
1554 btrfs_start_workers(&fs_info->fixup_workers, 1);
1555 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1556 btrfs_start_workers(&fs_info->endio_write_workers,
1557 fs_info->thread_pool_size);
1560 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1561 printk("Btrfs: wanted %llu devices, but found %llu\n",
1562 (unsigned long long)btrfs_super_num_devices(disk_super),
1563 (unsigned long long)fs_devices->open_devices);
1564 if (btrfs_test_opt(tree_root, DEGRADED))
1565 printk("continuing in degraded mode\n");
1567 goto fail_sb_buffer;
1571 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1572 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1573 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1575 nodesize = btrfs_super_nodesize(disk_super);
1576 leafsize = btrfs_super_leafsize(disk_super);
1577 sectorsize = btrfs_super_sectorsize(disk_super);
1578 stripesize = btrfs_super_stripesize(disk_super);
1579 tree_root->nodesize = nodesize;
1580 tree_root->leafsize = leafsize;
1581 tree_root->sectorsize = sectorsize;
1582 tree_root->stripesize = stripesize;
1584 sb->s_blocksize = sectorsize;
1585 sb->s_blocksize_bits = blksize_bits(sectorsize);
1587 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1588 sizeof(disk_super->magic))) {
1589 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1590 goto fail_sb_buffer;
1593 mutex_lock(&fs_info->chunk_mutex);
1594 ret = btrfs_read_sys_array(tree_root);
1595 mutex_unlock(&fs_info->chunk_mutex);
1597 printk("btrfs: failed to read the system array on %s\n",
1599 goto fail_sys_array;
1602 blocksize = btrfs_level_size(tree_root,
1603 btrfs_super_chunk_root_level(disk_super));
1604 generation = btrfs_super_chunk_root_generation(disk_super);
1606 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1607 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1609 chunk_root->node = read_tree_block(chunk_root,
1610 btrfs_super_chunk_root(disk_super),
1611 blocksize, generation);
1612 BUG_ON(!chunk_root->node);
1614 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1615 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1618 mutex_lock(&fs_info->chunk_mutex);
1619 ret = btrfs_read_chunk_tree(chunk_root);
1620 mutex_unlock(&fs_info->chunk_mutex);
1623 btrfs_close_extra_devices(fs_devices);
1625 blocksize = btrfs_level_size(tree_root,
1626 btrfs_super_root_level(disk_super));
1627 generation = btrfs_super_generation(disk_super);
1629 tree_root->node = read_tree_block(tree_root,
1630 btrfs_super_root(disk_super),
1631 blocksize, generation);
1632 if (!tree_root->node)
1633 goto fail_sb_buffer;
1636 ret = find_and_setup_root(tree_root, fs_info,
1637 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1639 goto fail_tree_root;
1640 extent_root->track_dirty = 1;
1642 ret = find_and_setup_root(tree_root, fs_info,
1643 BTRFS_DEV_TREE_OBJECTID, dev_root);
1644 dev_root->track_dirty = 1;
1647 goto fail_extent_root;
1649 btrfs_read_block_groups(extent_root);
1651 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1652 fs_info->data_alloc_profile = (u64)-1;
1653 fs_info->metadata_alloc_profile = (u64)-1;
1654 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1655 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1657 if (!fs_info->cleaner_kthread)
1658 goto fail_extent_root;
1660 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1662 "btrfs-transaction");
1663 if (!fs_info->transaction_kthread)
1666 if (btrfs_super_log_root(disk_super) != 0) {
1668 u64 bytenr = btrfs_super_log_root(disk_super);
1671 btrfs_level_size(tree_root,
1672 btrfs_super_log_root_level(disk_super));
1674 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1677 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1678 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1680 log_tree_root->node = read_tree_block(tree_root, bytenr,
1683 ret = btrfs_recover_log_trees(log_tree_root);
1686 fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1688 ret = btrfs_cleanup_reloc_trees(tree_root);
1694 kthread_stop(fs_info->cleaner_kthread);
1696 free_extent_buffer(extent_root->node);
1698 free_extent_buffer(tree_root->node);
1701 btrfs_stop_workers(&fs_info->fixup_workers);
1702 btrfs_stop_workers(&fs_info->workers);
1703 btrfs_stop_workers(&fs_info->endio_workers);
1704 btrfs_stop_workers(&fs_info->endio_write_workers);
1705 btrfs_stop_workers(&fs_info->submit_workers);
1707 iput(fs_info->btree_inode);
1709 btrfs_close_devices(fs_info->fs_devices);
1710 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1714 bdi_destroy(&fs_info->bdi);
1718 return ERR_PTR(err);
1721 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1723 char b[BDEVNAME_SIZE];
1726 set_buffer_uptodate(bh);
1728 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1729 printk(KERN_WARNING "lost page write due to "
1730 "I/O error on %s\n",
1731 bdevname(bh->b_bdev, b));
1733 /* note, we dont' set_buffer_write_io_error because we have
1734 * our own ways of dealing with the IO errors
1736 clear_buffer_uptodate(bh);
1742 int write_all_supers(struct btrfs_root *root)
1744 struct list_head *cur;
1745 struct list_head *head = &root->fs_info->fs_devices->devices;
1746 struct btrfs_device *dev;
1747 struct btrfs_super_block *sb;
1748 struct btrfs_dev_item *dev_item;
1749 struct buffer_head *bh;
1753 int total_errors = 0;
1757 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1758 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1760 sb = &root->fs_info->super_for_commit;
1761 dev_item = &sb->dev_item;
1762 list_for_each(cur, head) {
1763 dev = list_entry(cur, struct btrfs_device, dev_list);
1768 if (!dev->in_fs_metadata)
1771 btrfs_set_stack_device_type(dev_item, dev->type);
1772 btrfs_set_stack_device_id(dev_item, dev->devid);
1773 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1774 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1775 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1776 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1777 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1778 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1779 flags = btrfs_super_flags(sb);
1780 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1784 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1785 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1786 btrfs_csum_final(crc, sb->csum);
1788 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1789 BTRFS_SUPER_INFO_SIZE);
1791 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1792 dev->pending_io = bh;
1795 set_buffer_uptodate(bh);
1797 bh->b_end_io = btrfs_end_buffer_write_sync;
1799 if (do_barriers && dev->barriers) {
1800 ret = submit_bh(WRITE_BARRIER, bh);
1801 if (ret == -EOPNOTSUPP) {
1802 printk("btrfs: disabling barriers on dev %s\n",
1804 set_buffer_uptodate(bh);
1808 ret = submit_bh(WRITE, bh);
1811 ret = submit_bh(WRITE, bh);
1816 if (total_errors > max_errors) {
1817 printk("btrfs: %d errors while writing supers\n", total_errors);
1822 list_for_each(cur, head) {
1823 dev = list_entry(cur, struct btrfs_device, dev_list);
1826 if (!dev->in_fs_metadata)
1829 BUG_ON(!dev->pending_io);
1830 bh = dev->pending_io;
1832 if (!buffer_uptodate(dev->pending_io)) {
1833 if (do_barriers && dev->barriers) {
1834 printk("btrfs: disabling barriers on dev %s\n",
1836 set_buffer_uptodate(bh);
1840 ret = submit_bh(WRITE, bh);
1843 if (!buffer_uptodate(bh))
1850 dev->pending_io = NULL;
1853 if (total_errors > max_errors) {
1854 printk("btrfs: %d errors while writing supers\n", total_errors);
1860 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1865 ret = write_all_supers(root);
1869 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1871 radix_tree_delete(&fs_info->fs_roots_radix,
1872 (unsigned long)root->root_key.objectid);
1874 btrfs_sysfs_del_root(root);
1878 free_extent_buffer(root->node);
1879 if (root->commit_root)
1880 free_extent_buffer(root->commit_root);
1887 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1890 struct btrfs_root *gang[8];
1894 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1899 for (i = 0; i < ret; i++)
1900 btrfs_free_fs_root(fs_info, gang[i]);
1905 int close_ctree(struct btrfs_root *root)
1908 struct btrfs_trans_handle *trans;
1909 struct btrfs_fs_info *fs_info = root->fs_info;
1911 fs_info->closing = 1;
1914 kthread_stop(root->fs_info->transaction_kthread);
1915 kthread_stop(root->fs_info->cleaner_kthread);
1917 btrfs_clean_old_snapshots(root);
1918 trans = btrfs_start_transaction(root, 1);
1919 ret = btrfs_commit_transaction(trans, root);
1920 /* run commit again to drop the original snapshot */
1921 trans = btrfs_start_transaction(root, 1);
1922 btrfs_commit_transaction(trans, root);
1923 ret = btrfs_write_and_wait_transaction(NULL, root);
1926 write_ctree_super(NULL, root);
1928 if (fs_info->delalloc_bytes) {
1929 printk("btrfs: at unmount delalloc count %Lu\n",
1930 fs_info->delalloc_bytes);
1932 if (fs_info->total_ref_cache_size) {
1933 printk("btrfs: at umount reference cache size %Lu\n",
1934 fs_info->total_ref_cache_size);
1937 if (fs_info->extent_root->node)
1938 free_extent_buffer(fs_info->extent_root->node);
1940 if (fs_info->tree_root->node)
1941 free_extent_buffer(fs_info->tree_root->node);
1943 if (root->fs_info->chunk_root->node);
1944 free_extent_buffer(root->fs_info->chunk_root->node);
1946 if (root->fs_info->dev_root->node);
1947 free_extent_buffer(root->fs_info->dev_root->node);
1949 btrfs_free_block_groups(root->fs_info);
1950 fs_info->closing = 2;
1951 del_fs_roots(fs_info);
1953 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1955 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1957 btrfs_stop_workers(&fs_info->fixup_workers);
1958 btrfs_stop_workers(&fs_info->workers);
1959 btrfs_stop_workers(&fs_info->endio_workers);
1960 btrfs_stop_workers(&fs_info->endio_write_workers);
1961 btrfs_stop_workers(&fs_info->submit_workers);
1963 iput(fs_info->btree_inode);
1965 while(!list_empty(&fs_info->hashers)) {
1966 struct btrfs_hasher *hasher;
1967 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1969 list_del(&hasher->hashers);
1970 crypto_free_hash(&fs_info->hash_tfm);
1974 btrfs_close_devices(fs_info->fs_devices);
1975 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1977 bdi_destroy(&fs_info->bdi);
1979 kfree(fs_info->extent_root);
1980 kfree(fs_info->tree_root);
1981 kfree(fs_info->chunk_root);
1982 kfree(fs_info->dev_root);
1986 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1989 struct inode *btree_inode = buf->first_page->mapping->host;
1991 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1995 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2000 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2002 struct inode *btree_inode = buf->first_page->mapping->host;
2003 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2007 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2009 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2010 u64 transid = btrfs_header_generation(buf);
2011 struct inode *btree_inode = root->fs_info->btree_inode;
2013 WARN_ON(!btrfs_tree_locked(buf));
2014 if (transid != root->fs_info->generation) {
2015 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2016 (unsigned long long)buf->start,
2017 transid, root->fs_info->generation);
2020 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2023 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2026 * looks as though older kernels can get into trouble with
2027 * this code, they end up stuck in balance_dirty_pages forever
2029 struct extent_io_tree *tree;
2032 unsigned long thresh = 96 * 1024 * 1024;
2033 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2035 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2038 num_dirty = count_range_bits(tree, &start, (u64)-1,
2039 thresh, EXTENT_DIRTY);
2040 if (num_dirty > thresh) {
2041 balance_dirty_pages_ratelimited_nr(
2042 root->fs_info->btree_inode->i_mapping, 1);
2047 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2049 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2051 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2053 buf->flags |= EXTENT_UPTODATE;
2058 int btree_lock_page_hook(struct page *page)
2060 struct inode *inode = page->mapping->host;
2061 struct btrfs_root *root = BTRFS_I(inode)->root;
2062 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2063 struct extent_buffer *eb;
2065 u64 bytenr = page_offset(page);
2067 if (page->private == EXTENT_PAGE_PRIVATE)
2070 len = page->private >> 2;
2071 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2075 btrfs_tree_lock(eb);
2076 spin_lock(&root->fs_info->hash_lock);
2077 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2078 spin_unlock(&root->fs_info->hash_lock);
2079 btrfs_tree_unlock(eb);
2080 free_extent_buffer(eb);
2086 static struct extent_io_ops btree_extent_io_ops = {
2087 .write_cache_pages_lock_hook = btree_lock_page_hook,
2088 .readpage_end_io_hook = btree_readpage_end_io_hook,
2089 .submit_bio_hook = btree_submit_bio_hook,
2090 /* note we're sharing with inode.c for the merge bio hook */
2091 .merge_bio_hook = btrfs_merge_bio_hook,