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 struct btrfs_work work;
90 * extents on the btree inode are pretty simple, there's one extent
91 * that covers the entire device
93 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
94 size_t page_offset, u64 start, u64 len,
97 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
98 struct extent_map *em;
101 spin_lock(&em_tree->lock);
102 em = lookup_extent_mapping(em_tree, start, len);
105 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
106 spin_unlock(&em_tree->lock);
109 spin_unlock(&em_tree->lock);
111 em = alloc_extent_map(GFP_NOFS);
113 em = ERR_PTR(-ENOMEM);
119 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
121 spin_lock(&em_tree->lock);
122 ret = add_extent_mapping(em_tree, em);
123 if (ret == -EEXIST) {
124 u64 failed_start = em->start;
125 u64 failed_len = em->len;
127 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
128 em->start, em->len, em->block_start);
130 em = lookup_extent_mapping(em_tree, start, len);
132 printk("after failing, found %Lu %Lu %Lu\n",
133 em->start, em->len, em->block_start);
136 em = lookup_extent_mapping(em_tree, failed_start,
139 printk("double failure lookup gives us "
140 "%Lu %Lu -> %Lu\n", em->start,
141 em->len, em->block_start);
150 spin_unlock(&em_tree->lock);
158 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
160 return btrfs_crc32c(seed, data, len);
163 void btrfs_csum_final(u32 crc, char *result)
165 *(__le32 *)result = ~cpu_to_le32(crc);
169 * compute the csum for a btree block, and either verify it or write it
170 * into the csum field of the block.
172 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
175 char result[BTRFS_CRC32_SIZE];
177 unsigned long cur_len;
178 unsigned long offset = BTRFS_CSUM_SIZE;
179 char *map_token = NULL;
181 unsigned long map_start;
182 unsigned long map_len;
186 len = buf->len - offset;
188 err = map_private_extent_buffer(buf, offset, 32,
190 &map_start, &map_len, KM_USER0);
192 printk("failed to map extent buffer! %lu\n",
196 cur_len = min(len, map_len - (offset - map_start));
197 crc = btrfs_csum_data(root, kaddr + offset - map_start,
201 unmap_extent_buffer(buf, map_token, KM_USER0);
203 btrfs_csum_final(crc, result);
206 /* FIXME, this is not good */
207 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
210 memcpy(&found, result, BTRFS_CRC32_SIZE);
212 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
213 printk("btrfs: %s checksum verify failed on %llu "
214 "wanted %X found %X level %d\n",
215 root->fs_info->sb->s_id,
216 buf->start, val, found, btrfs_header_level(buf));
220 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
226 * we can't consider a given block up to date unless the transid of the
227 * block matches the transid in the parent node's pointer. This is how we
228 * detect blocks that either didn't get written at all or got written
229 * in the wrong place.
231 static int verify_parent_transid(struct extent_io_tree *io_tree,
232 struct extent_buffer *eb, u64 parent_transid)
236 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
239 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
240 if (extent_buffer_uptodate(io_tree, eb) &&
241 btrfs_header_generation(eb) == parent_transid) {
245 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
246 (unsigned long long)eb->start,
247 (unsigned long long)parent_transid,
248 (unsigned long long)btrfs_header_generation(eb));
250 clear_extent_buffer_uptodate(io_tree, eb);
252 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
258 * helper to read a given tree block, doing retries as required when
259 * the checksums don't match and we have alternate mirrors to try.
261 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
262 struct extent_buffer *eb,
263 u64 start, u64 parent_transid)
265 struct extent_io_tree *io_tree;
270 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
272 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
273 btree_get_extent, mirror_num);
275 !verify_parent_transid(io_tree, eb, parent_transid))
277 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
278 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
284 if (mirror_num > num_copies)
291 * checksum a dirty tree block before IO. This has extra checks to make
292 * sure we only fill in the checksum field in the first page of a multi-page block
294 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
296 struct extent_io_tree *tree;
297 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
301 struct extent_buffer *eb;
304 tree = &BTRFS_I(page->mapping->host)->io_tree;
306 if (page->private == EXTENT_PAGE_PRIVATE)
310 len = page->private >> 2;
314 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
315 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
316 btrfs_header_generation(eb));
318 found_start = btrfs_header_bytenr(eb);
319 if (found_start != start) {
320 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
321 start, found_start, len);
325 if (eb->first_page != page) {
326 printk("bad first page %lu %lu\n", eb->first_page->index,
331 if (!PageUptodate(page)) {
332 printk("csum not up to date page %lu\n", page->index);
336 found_level = btrfs_header_level(eb);
338 csum_tree_block(root, eb, 0);
340 free_extent_buffer(eb);
345 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
346 struct extent_state *state)
348 struct extent_io_tree *tree;
352 struct extent_buffer *eb;
353 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
356 tree = &BTRFS_I(page->mapping->host)->io_tree;
357 if (page->private == EXTENT_PAGE_PRIVATE)
361 len = page->private >> 2;
365 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
367 found_start = btrfs_header_bytenr(eb);
368 if (found_start != start) {
369 printk("bad tree block start %llu %llu\n",
370 (unsigned long long)found_start,
371 (unsigned long long)eb->start);
375 if (eb->first_page != page) {
376 printk("bad first page %lu %lu\n", eb->first_page->index,
382 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
383 (unsigned long)btrfs_header_fsid(eb),
385 printk("bad fsid on block %Lu\n", eb->start);
389 found_level = btrfs_header_level(eb);
391 ret = csum_tree_block(root, eb, 1);
395 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
396 end = eb->start + end - 1;
398 free_extent_buffer(eb);
403 static void end_workqueue_bio(struct bio *bio, int err)
405 struct end_io_wq *end_io_wq = bio->bi_private;
406 struct btrfs_fs_info *fs_info;
408 fs_info = end_io_wq->info;
409 end_io_wq->error = err;
410 end_io_wq->work.func = end_workqueue_fn;
411 end_io_wq->work.flags = 0;
412 if (bio->bi_rw & (1 << BIO_RW))
413 btrfs_queue_worker(&fs_info->endio_write_workers,
416 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
419 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
422 struct end_io_wq *end_io_wq;
423 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
427 end_io_wq->private = bio->bi_private;
428 end_io_wq->end_io = bio->bi_end_io;
429 end_io_wq->info = info;
430 end_io_wq->error = 0;
431 end_io_wq->bio = bio;
432 end_io_wq->metadata = metadata;
434 bio->bi_private = end_io_wq;
435 bio->bi_end_io = end_workqueue_bio;
439 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
441 unsigned long limit = min_t(unsigned long,
442 info->workers.max_workers,
443 info->fs_devices->open_devices);
447 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
449 return atomic_read(&info->nr_async_bios) >
450 btrfs_async_submit_limit(info);
453 static void run_one_async_submit(struct btrfs_work *work)
455 struct btrfs_fs_info *fs_info;
456 struct async_submit_bio *async;
459 async = container_of(work, struct async_submit_bio, work);
460 fs_info = BTRFS_I(async->inode)->root->fs_info;
462 limit = btrfs_async_submit_limit(fs_info);
463 limit = limit * 2 / 3;
465 atomic_dec(&fs_info->nr_async_submits);
467 if (atomic_read(&fs_info->nr_async_submits) < limit &&
468 waitqueue_active(&fs_info->async_submit_wait))
469 wake_up(&fs_info->async_submit_wait);
471 async->submit_bio_hook(async->inode, async->rw, async->bio,
476 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
477 int rw, struct bio *bio, int mirror_num,
478 extent_submit_bio_hook_t *submit_bio_hook)
480 struct async_submit_bio *async;
481 int limit = btrfs_async_submit_limit(fs_info);
483 async = kmalloc(sizeof(*async), GFP_NOFS);
487 async->inode = inode;
490 async->mirror_num = mirror_num;
491 async->submit_bio_hook = submit_bio_hook;
492 async->work.func = run_one_async_submit;
493 async->work.flags = 0;
495 while(atomic_read(&fs_info->async_submit_draining) &&
496 atomic_read(&fs_info->nr_async_submits)) {
497 wait_event(fs_info->async_submit_wait,
498 (atomic_read(&fs_info->nr_async_submits) == 0));
501 atomic_inc(&fs_info->nr_async_submits);
502 btrfs_queue_worker(&fs_info->workers, &async->work);
504 if (atomic_read(&fs_info->nr_async_submits) > limit) {
505 wait_event_timeout(fs_info->async_submit_wait,
506 (atomic_read(&fs_info->nr_async_submits) < limit),
509 wait_event_timeout(fs_info->async_submit_wait,
510 (atomic_read(&fs_info->nr_async_bios) < limit),
516 static int btree_csum_one_bio(struct bio *bio)
518 struct bio_vec *bvec = bio->bi_io_vec;
520 struct btrfs_root *root;
522 WARN_ON(bio->bi_vcnt <= 0);
523 while(bio_index < bio->bi_vcnt) {
524 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
525 csum_dirty_buffer(root, bvec->bv_page);
532 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
535 struct btrfs_root *root = BTRFS_I(inode)->root;
539 * when we're called for a write, we're already in the async
540 * submission context. Just jump into btrfs_map_bio
542 if (rw & (1 << BIO_RW)) {
543 btree_csum_one_bio(bio);
544 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
549 * called for a read, do the setup so that checksum validation
550 * can happen in the async kernel threads
552 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
555 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
558 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
562 * kthread helpers are used to submit writes so that checksumming
563 * can happen in parallel across all CPUs
565 if (!(rw & (1 << BIO_RW))) {
566 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
568 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
569 inode, rw, bio, mirror_num,
570 __btree_submit_bio_hook);
573 static int btree_writepage(struct page *page, struct writeback_control *wbc)
575 struct extent_io_tree *tree;
576 tree = &BTRFS_I(page->mapping->host)->io_tree;
578 if (current->flags & PF_MEMALLOC) {
579 redirty_page_for_writepage(wbc, page);
583 return extent_write_full_page(tree, page, btree_get_extent, wbc);
586 static int btree_writepages(struct address_space *mapping,
587 struct writeback_control *wbc)
589 struct extent_io_tree *tree;
590 tree = &BTRFS_I(mapping->host)->io_tree;
591 if (wbc->sync_mode == WB_SYNC_NONE) {
594 unsigned long thresh = 32 * 1024 * 1024;
596 if (wbc->for_kupdate)
599 num_dirty = count_range_bits(tree, &start, (u64)-1,
600 thresh, EXTENT_DIRTY);
601 if (num_dirty < thresh) {
605 return extent_writepages(tree, mapping, btree_get_extent, wbc);
608 int btree_readpage(struct file *file, struct page *page)
610 struct extent_io_tree *tree;
611 tree = &BTRFS_I(page->mapping->host)->io_tree;
612 return extent_read_full_page(tree, page, btree_get_extent);
615 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
617 struct extent_io_tree *tree;
618 struct extent_map_tree *map;
621 if (PageWriteback(page) || PageDirty(page))
624 tree = &BTRFS_I(page->mapping->host)->io_tree;
625 map = &BTRFS_I(page->mapping->host)->extent_tree;
627 ret = try_release_extent_state(map, tree, page, gfp_flags);
632 ret = try_release_extent_buffer(tree, page);
634 ClearPagePrivate(page);
635 set_page_private(page, 0);
636 page_cache_release(page);
642 static void btree_invalidatepage(struct page *page, unsigned long offset)
644 struct extent_io_tree *tree;
645 tree = &BTRFS_I(page->mapping->host)->io_tree;
646 extent_invalidatepage(tree, page, offset);
647 btree_releasepage(page, GFP_NOFS);
648 if (PagePrivate(page)) {
649 printk("warning page private not zero on page %Lu\n",
651 ClearPagePrivate(page);
652 set_page_private(page, 0);
653 page_cache_release(page);
658 static int btree_writepage(struct page *page, struct writeback_control *wbc)
660 struct buffer_head *bh;
661 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
662 struct buffer_head *head;
663 if (!page_has_buffers(page)) {
664 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
665 (1 << BH_Dirty)|(1 << BH_Uptodate));
667 head = page_buffers(page);
670 if (buffer_dirty(bh))
671 csum_tree_block(root, bh, 0);
672 bh = bh->b_this_page;
673 } while (bh != head);
674 return block_write_full_page(page, btree_get_block, wbc);
678 static struct address_space_operations btree_aops = {
679 .readpage = btree_readpage,
680 .writepage = btree_writepage,
681 .writepages = btree_writepages,
682 .releasepage = btree_releasepage,
683 .invalidatepage = btree_invalidatepage,
684 .sync_page = block_sync_page,
687 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
690 struct extent_buffer *buf = NULL;
691 struct inode *btree_inode = root->fs_info->btree_inode;
694 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
697 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
698 buf, 0, 0, btree_get_extent, 0);
699 free_extent_buffer(buf);
703 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
704 u64 bytenr, u32 blocksize)
706 struct inode *btree_inode = root->fs_info->btree_inode;
707 struct extent_buffer *eb;
708 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
709 bytenr, blocksize, GFP_NOFS);
713 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
714 u64 bytenr, u32 blocksize)
716 struct inode *btree_inode = root->fs_info->btree_inode;
717 struct extent_buffer *eb;
719 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
720 bytenr, blocksize, NULL, GFP_NOFS);
725 int btrfs_write_tree_block(struct extent_buffer *buf)
727 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
728 buf->start + buf->len - 1, WB_SYNC_ALL);
731 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
733 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
734 buf->start, buf->start + buf->len -1);
737 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
738 u32 blocksize, u64 parent_transid)
740 struct extent_buffer *buf = NULL;
741 struct inode *btree_inode = root->fs_info->btree_inode;
742 struct extent_io_tree *io_tree;
745 io_tree = &BTRFS_I(btree_inode)->io_tree;
747 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
751 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
754 buf->flags |= EXTENT_UPTODATE;
762 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
763 struct extent_buffer *buf)
765 struct inode *btree_inode = root->fs_info->btree_inode;
766 if (btrfs_header_generation(buf) ==
767 root->fs_info->running_transaction->transid) {
768 WARN_ON(!btrfs_tree_locked(buf));
769 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
775 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
776 u32 stripesize, struct btrfs_root *root,
777 struct btrfs_fs_info *fs_info,
782 root->commit_root = NULL;
783 root->ref_tree = NULL;
784 root->sectorsize = sectorsize;
785 root->nodesize = nodesize;
786 root->leafsize = leafsize;
787 root->stripesize = stripesize;
789 root->track_dirty = 0;
791 root->fs_info = fs_info;
792 root->objectid = objectid;
793 root->last_trans = 0;
794 root->highest_inode = 0;
795 root->last_inode_alloc = 0;
799 INIT_LIST_HEAD(&root->dirty_list);
800 INIT_LIST_HEAD(&root->orphan_list);
801 INIT_LIST_HEAD(&root->dead_list);
802 spin_lock_init(&root->node_lock);
803 spin_lock_init(&root->list_lock);
804 mutex_init(&root->objectid_mutex);
805 mutex_init(&root->log_mutex);
806 extent_io_tree_init(&root->dirty_log_pages,
807 fs_info->btree_inode->i_mapping, GFP_NOFS);
809 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
810 root->ref_tree = &root->ref_tree_struct;
812 memset(&root->root_key, 0, sizeof(root->root_key));
813 memset(&root->root_item, 0, sizeof(root->root_item));
814 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
815 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
816 root->defrag_trans_start = fs_info->generation;
817 init_completion(&root->kobj_unregister);
818 root->defrag_running = 0;
819 root->defrag_level = 0;
820 root->root_key.objectid = objectid;
824 static int find_and_setup_root(struct btrfs_root *tree_root,
825 struct btrfs_fs_info *fs_info,
827 struct btrfs_root *root)
832 __setup_root(tree_root->nodesize, tree_root->leafsize,
833 tree_root->sectorsize, tree_root->stripesize,
834 root, fs_info, objectid);
835 ret = btrfs_find_last_root(tree_root, objectid,
836 &root->root_item, &root->root_key);
839 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
840 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
846 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
847 struct btrfs_fs_info *fs_info)
849 struct extent_buffer *eb;
850 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
859 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
860 0, &start, &end, EXTENT_DIRTY);
864 clear_extent_dirty(&log_root_tree->dirty_log_pages,
865 start, end, GFP_NOFS);
867 eb = fs_info->log_root_tree->node;
869 WARN_ON(btrfs_header_level(eb) != 0);
870 WARN_ON(btrfs_header_nritems(eb) != 0);
872 ret = btrfs_free_reserved_extent(fs_info->tree_root,
876 free_extent_buffer(eb);
877 kfree(fs_info->log_root_tree);
878 fs_info->log_root_tree = NULL;
882 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
883 struct btrfs_fs_info *fs_info)
885 struct btrfs_root *root;
886 struct btrfs_root *tree_root = fs_info->tree_root;
888 root = kzalloc(sizeof(*root), GFP_NOFS);
892 __setup_root(tree_root->nodesize, tree_root->leafsize,
893 tree_root->sectorsize, tree_root->stripesize,
894 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
896 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
897 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
898 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
901 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
902 0, BTRFS_TREE_LOG_OBJECTID,
903 trans->transid, 0, 0, 0);
905 btrfs_set_header_nritems(root->node, 0);
906 btrfs_set_header_level(root->node, 0);
907 btrfs_set_header_bytenr(root->node, root->node->start);
908 btrfs_set_header_generation(root->node, trans->transid);
909 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
911 write_extent_buffer(root->node, root->fs_info->fsid,
912 (unsigned long)btrfs_header_fsid(root->node),
914 btrfs_mark_buffer_dirty(root->node);
915 btrfs_tree_unlock(root->node);
916 fs_info->log_root_tree = root;
920 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
921 struct btrfs_key *location)
923 struct btrfs_root *root;
924 struct btrfs_fs_info *fs_info = tree_root->fs_info;
925 struct btrfs_path *path;
926 struct extent_buffer *l;
931 root = kzalloc(sizeof(*root), GFP_NOFS);
933 return ERR_PTR(-ENOMEM);
934 if (location->offset == (u64)-1) {
935 ret = find_and_setup_root(tree_root, fs_info,
936 location->objectid, root);
944 __setup_root(tree_root->nodesize, tree_root->leafsize,
945 tree_root->sectorsize, tree_root->stripesize,
946 root, fs_info, location->objectid);
948 path = btrfs_alloc_path();
950 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
957 read_extent_buffer(l, &root->root_item,
958 btrfs_item_ptr_offset(l, path->slots[0]),
959 sizeof(root->root_item));
960 memcpy(&root->root_key, location, sizeof(*location));
963 btrfs_release_path(root, path);
964 btrfs_free_path(path);
969 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
970 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
974 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
976 ret = btrfs_find_highest_inode(root, &highest_inode);
978 root->highest_inode = highest_inode;
979 root->last_inode_alloc = highest_inode;
985 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
988 struct btrfs_root *root;
990 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
991 return fs_info->tree_root;
992 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
993 return fs_info->extent_root;
995 root = radix_tree_lookup(&fs_info->fs_roots_radix,
996 (unsigned long)root_objectid);
1000 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1001 struct btrfs_key *location)
1003 struct btrfs_root *root;
1006 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1007 return fs_info->tree_root;
1008 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1009 return fs_info->extent_root;
1010 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1011 return fs_info->chunk_root;
1012 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1013 return fs_info->dev_root;
1015 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1016 (unsigned long)location->objectid);
1020 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1023 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1024 (unsigned long)root->root_key.objectid,
1027 free_extent_buffer(root->node);
1029 return ERR_PTR(ret);
1031 ret = btrfs_find_dead_roots(fs_info->tree_root,
1032 root->root_key.objectid, root);
1038 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1039 struct btrfs_key *location,
1040 const char *name, int namelen)
1042 struct btrfs_root *root;
1045 root = btrfs_read_fs_root_no_name(fs_info, location);
1052 ret = btrfs_set_root_name(root, name, namelen);
1054 free_extent_buffer(root->node);
1056 return ERR_PTR(ret);
1059 ret = btrfs_sysfs_add_root(root);
1061 free_extent_buffer(root->node);
1064 return ERR_PTR(ret);
1070 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1071 struct btrfs_hasher *hasher;
1073 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1076 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1077 if (!hasher->hash_tfm) {
1081 spin_lock(&info->hash_lock);
1082 list_add(&hasher->list, &info->hashers);
1083 spin_unlock(&info->hash_lock);
1088 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1090 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1092 struct list_head *cur;
1093 struct btrfs_device *device;
1094 struct backing_dev_info *bdi;
1096 if ((bdi_bits & (1 << BDI_write_congested)) &&
1097 btrfs_congested_async(info, 0))
1100 list_for_each(cur, &info->fs_devices->devices) {
1101 device = list_entry(cur, struct btrfs_device, dev_list);
1104 bdi = blk_get_backing_dev_info(device->bdev);
1105 if (bdi && bdi_congested(bdi, bdi_bits)) {
1114 * this unplugs every device on the box, and it is only used when page
1117 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1119 struct list_head *cur;
1120 struct btrfs_device *device;
1121 struct btrfs_fs_info *info;
1123 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1124 list_for_each(cur, &info->fs_devices->devices) {
1125 device = list_entry(cur, struct btrfs_device, dev_list);
1126 bdi = blk_get_backing_dev_info(device->bdev);
1127 if (bdi->unplug_io_fn) {
1128 bdi->unplug_io_fn(bdi, page);
1133 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1135 struct inode *inode;
1136 struct extent_map_tree *em_tree;
1137 struct extent_map *em;
1138 struct address_space *mapping;
1141 /* the generic O_DIRECT read code does this */
1143 __unplug_io_fn(bdi, page);
1148 * page->mapping may change at any time. Get a consistent copy
1149 * and use that for everything below
1152 mapping = page->mapping;
1156 inode = mapping->host;
1157 offset = page_offset(page);
1159 em_tree = &BTRFS_I(inode)->extent_tree;
1160 spin_lock(&em_tree->lock);
1161 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1162 spin_unlock(&em_tree->lock);
1164 __unplug_io_fn(bdi, page);
1168 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1169 free_extent_map(em);
1170 __unplug_io_fn(bdi, page);
1173 offset = offset - em->start;
1174 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1175 em->block_start + offset, page);
1176 free_extent_map(em);
1179 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1182 bdi->ra_pages = default_backing_dev_info.ra_pages;
1184 bdi->capabilities = default_backing_dev_info.capabilities;
1185 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1186 bdi->unplug_io_data = info;
1187 bdi->congested_fn = btrfs_congested_fn;
1188 bdi->congested_data = info;
1192 static int bio_ready_for_csum(struct bio *bio)
1198 struct extent_io_tree *io_tree = NULL;
1199 struct btrfs_fs_info *info = NULL;
1200 struct bio_vec *bvec;
1204 bio_for_each_segment(bvec, bio, i) {
1205 page = bvec->bv_page;
1206 if (page->private == EXTENT_PAGE_PRIVATE) {
1207 length += bvec->bv_len;
1210 if (!page->private) {
1211 length += bvec->bv_len;
1214 length = bvec->bv_len;
1215 buf_len = page->private >> 2;
1216 start = page_offset(page) + bvec->bv_offset;
1217 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1218 info = BTRFS_I(page->mapping->host)->root->fs_info;
1220 /* are we fully contained in this bio? */
1221 if (buf_len <= length)
1224 ret = extent_range_uptodate(io_tree, start + length,
1225 start + buf_len - 1);
1232 * called by the kthread helper functions to finally call the bio end_io
1233 * functions. This is where read checksum verification actually happens
1235 static void end_workqueue_fn(struct btrfs_work *work)
1238 struct end_io_wq *end_io_wq;
1239 struct btrfs_fs_info *fs_info;
1242 end_io_wq = container_of(work, struct end_io_wq, work);
1243 bio = end_io_wq->bio;
1244 fs_info = end_io_wq->info;
1246 /* metadata bios are special because the whole tree block must
1247 * be checksummed at once. This makes sure the entire block is in
1248 * ram and up to date before trying to verify things. For
1249 * blocksize <= pagesize, it is basically a noop
1251 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1252 btrfs_queue_worker(&fs_info->endio_workers,
1256 error = end_io_wq->error;
1257 bio->bi_private = end_io_wq->private;
1258 bio->bi_end_io = end_io_wq->end_io;
1260 bio_endio(bio, error);
1263 static int cleaner_kthread(void *arg)
1265 struct btrfs_root *root = arg;
1269 if (root->fs_info->closing)
1272 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1273 mutex_lock(&root->fs_info->cleaner_mutex);
1274 btrfs_clean_old_snapshots(root);
1275 mutex_unlock(&root->fs_info->cleaner_mutex);
1277 if (freezing(current)) {
1281 if (root->fs_info->closing)
1283 set_current_state(TASK_INTERRUPTIBLE);
1285 __set_current_state(TASK_RUNNING);
1287 } while (!kthread_should_stop());
1291 static int transaction_kthread(void *arg)
1293 struct btrfs_root *root = arg;
1294 struct btrfs_trans_handle *trans;
1295 struct btrfs_transaction *cur;
1297 unsigned long delay;
1302 if (root->fs_info->closing)
1306 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1307 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1309 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1310 printk("btrfs: total reference cache size %Lu\n",
1311 root->fs_info->total_ref_cache_size);
1314 mutex_lock(&root->fs_info->trans_mutex);
1315 cur = root->fs_info->running_transaction;
1317 mutex_unlock(&root->fs_info->trans_mutex);
1321 now = get_seconds();
1322 if (now < cur->start_time || now - cur->start_time < 30) {
1323 mutex_unlock(&root->fs_info->trans_mutex);
1327 mutex_unlock(&root->fs_info->trans_mutex);
1328 trans = btrfs_start_transaction(root, 1);
1329 ret = btrfs_commit_transaction(trans, root);
1331 wake_up_process(root->fs_info->cleaner_kthread);
1332 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1334 if (freezing(current)) {
1337 if (root->fs_info->closing)
1339 set_current_state(TASK_INTERRUPTIBLE);
1340 schedule_timeout(delay);
1341 __set_current_state(TASK_RUNNING);
1343 } while (!kthread_should_stop());
1347 struct btrfs_root *open_ctree(struct super_block *sb,
1348 struct btrfs_fs_devices *fs_devices,
1356 struct buffer_head *bh;
1357 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1359 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1361 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1363 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1365 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1367 struct btrfs_root *log_tree_root;
1372 struct btrfs_super_block *disk_super;
1374 if (!extent_root || !tree_root || !fs_info ||
1375 !chunk_root || !dev_root) {
1379 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1380 INIT_LIST_HEAD(&fs_info->trans_list);
1381 INIT_LIST_HEAD(&fs_info->dead_roots);
1382 INIT_LIST_HEAD(&fs_info->hashers);
1383 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1384 spin_lock_init(&fs_info->hash_lock);
1385 spin_lock_init(&fs_info->delalloc_lock);
1386 spin_lock_init(&fs_info->new_trans_lock);
1387 spin_lock_init(&fs_info->ref_cache_lock);
1389 init_completion(&fs_info->kobj_unregister);
1390 fs_info->tree_root = tree_root;
1391 fs_info->extent_root = extent_root;
1392 fs_info->chunk_root = chunk_root;
1393 fs_info->dev_root = dev_root;
1394 fs_info->fs_devices = fs_devices;
1395 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1396 INIT_LIST_HEAD(&fs_info->space_info);
1397 btrfs_mapping_init(&fs_info->mapping_tree);
1398 atomic_set(&fs_info->nr_async_submits, 0);
1399 atomic_set(&fs_info->async_submit_draining, 0);
1400 atomic_set(&fs_info->nr_async_bios, 0);
1401 atomic_set(&fs_info->throttles, 0);
1402 atomic_set(&fs_info->throttle_gen, 0);
1404 fs_info->max_extent = (u64)-1;
1405 fs_info->max_inline = 8192 * 1024;
1406 setup_bdi(fs_info, &fs_info->bdi);
1407 fs_info->btree_inode = new_inode(sb);
1408 fs_info->btree_inode->i_ino = 1;
1409 fs_info->btree_inode->i_nlink = 1;
1410 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1412 INIT_LIST_HEAD(&fs_info->ordered_extents);
1413 spin_lock_init(&fs_info->ordered_extent_lock);
1415 sb->s_blocksize = 4096;
1416 sb->s_blocksize_bits = blksize_bits(4096);
1419 * we set the i_size on the btree inode to the max possible int.
1420 * the real end of the address space is determined by all of
1421 * the devices in the system
1423 fs_info->btree_inode->i_size = OFFSET_MAX;
1424 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1425 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1427 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1428 fs_info->btree_inode->i_mapping,
1430 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1433 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1435 spin_lock_init(&fs_info->block_group_cache_lock);
1436 fs_info->block_group_cache_tree.rb_node = NULL;
1438 extent_io_tree_init(&fs_info->pinned_extents,
1439 fs_info->btree_inode->i_mapping, GFP_NOFS);
1440 extent_io_tree_init(&fs_info->pending_del,
1441 fs_info->btree_inode->i_mapping, GFP_NOFS);
1442 extent_io_tree_init(&fs_info->extent_ins,
1443 fs_info->btree_inode->i_mapping, GFP_NOFS);
1444 fs_info->do_barriers = 1;
1446 extent_io_tree_init(&fs_info->reloc_mapping_tree,
1447 fs_info->btree_inode->i_mapping, GFP_NOFS);
1448 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1449 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1450 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1452 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1453 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1454 sizeof(struct btrfs_key));
1455 insert_inode_hash(fs_info->btree_inode);
1457 mutex_init(&fs_info->trans_mutex);
1458 mutex_init(&fs_info->tree_log_mutex);
1459 mutex_init(&fs_info->drop_mutex);
1460 mutex_init(&fs_info->alloc_mutex);
1461 mutex_init(&fs_info->chunk_mutex);
1462 mutex_init(&fs_info->transaction_kthread_mutex);
1463 mutex_init(&fs_info->cleaner_mutex);
1464 mutex_init(&fs_info->volume_mutex);
1465 mutex_init(&fs_info->tree_reloc_mutex);
1466 init_waitqueue_head(&fs_info->transaction_throttle);
1467 init_waitqueue_head(&fs_info->transaction_wait);
1468 init_waitqueue_head(&fs_info->async_submit_wait);
1469 init_waitqueue_head(&fs_info->tree_log_wait);
1470 atomic_set(&fs_info->tree_log_commit, 0);
1471 atomic_set(&fs_info->tree_log_writers, 0);
1472 fs_info->tree_log_transid = 0;
1475 ret = add_hasher(fs_info, "crc32c");
1477 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1482 __setup_root(4096, 4096, 4096, 4096, tree_root,
1483 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1486 bh = __bread(fs_devices->latest_bdev,
1487 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1491 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1494 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1496 disk_super = &fs_info->super_copy;
1497 if (!btrfs_super_root(disk_super))
1498 goto fail_sb_buffer;
1500 err = btrfs_parse_options(tree_root, options);
1502 goto fail_sb_buffer;
1505 * we need to start all the end_io workers up front because the
1506 * queue work function gets called at interrupt time, and so it
1507 * cannot dynamically grow.
1509 btrfs_init_workers(&fs_info->workers, "worker",
1510 fs_info->thread_pool_size);
1511 btrfs_init_workers(&fs_info->submit_workers, "submit",
1512 min_t(u64, fs_devices->num_devices,
1513 fs_info->thread_pool_size));
1515 /* a higher idle thresh on the submit workers makes it much more
1516 * likely that bios will be send down in a sane order to the
1519 fs_info->submit_workers.idle_thresh = 64;
1521 /* fs_info->workers is responsible for checksumming file data
1522 * blocks and metadata. Using a larger idle thresh allows each
1523 * worker thread to operate on things in roughly the order they
1524 * were sent by the writeback daemons, improving overall locality
1525 * of the IO going down the pipe.
1527 fs_info->workers.idle_thresh = 128;
1529 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1530 btrfs_init_workers(&fs_info->endio_workers, "endio",
1531 fs_info->thread_pool_size);
1532 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1533 fs_info->thread_pool_size);
1536 * endios are largely parallel and should have a very
1539 fs_info->endio_workers.idle_thresh = 4;
1540 fs_info->endio_write_workers.idle_thresh = 64;
1542 btrfs_start_workers(&fs_info->workers, 1);
1543 btrfs_start_workers(&fs_info->submit_workers, 1);
1544 btrfs_start_workers(&fs_info->fixup_workers, 1);
1545 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1546 btrfs_start_workers(&fs_info->endio_write_workers,
1547 fs_info->thread_pool_size);
1550 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1551 printk("Btrfs: wanted %llu devices, but found %llu\n",
1552 (unsigned long long)btrfs_super_num_devices(disk_super),
1553 (unsigned long long)fs_devices->open_devices);
1554 if (btrfs_test_opt(tree_root, DEGRADED))
1555 printk("continuing in degraded mode\n");
1557 goto fail_sb_buffer;
1561 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1563 nodesize = btrfs_super_nodesize(disk_super);
1564 leafsize = btrfs_super_leafsize(disk_super);
1565 sectorsize = btrfs_super_sectorsize(disk_super);
1566 stripesize = btrfs_super_stripesize(disk_super);
1567 tree_root->nodesize = nodesize;
1568 tree_root->leafsize = leafsize;
1569 tree_root->sectorsize = sectorsize;
1570 tree_root->stripesize = stripesize;
1572 sb->s_blocksize = sectorsize;
1573 sb->s_blocksize_bits = blksize_bits(sectorsize);
1575 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1576 sizeof(disk_super->magic))) {
1577 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1578 goto fail_sb_buffer;
1581 mutex_lock(&fs_info->chunk_mutex);
1582 ret = btrfs_read_sys_array(tree_root);
1583 mutex_unlock(&fs_info->chunk_mutex);
1585 printk("btrfs: failed to read the system array on %s\n",
1587 goto fail_sys_array;
1590 blocksize = btrfs_level_size(tree_root,
1591 btrfs_super_chunk_root_level(disk_super));
1593 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1594 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1596 chunk_root->node = read_tree_block(chunk_root,
1597 btrfs_super_chunk_root(disk_super),
1599 BUG_ON(!chunk_root->node);
1601 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1602 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1605 mutex_lock(&fs_info->chunk_mutex);
1606 ret = btrfs_read_chunk_tree(chunk_root);
1607 mutex_unlock(&fs_info->chunk_mutex);
1610 btrfs_close_extra_devices(fs_devices);
1612 blocksize = btrfs_level_size(tree_root,
1613 btrfs_super_root_level(disk_super));
1616 tree_root->node = read_tree_block(tree_root,
1617 btrfs_super_root(disk_super),
1619 if (!tree_root->node)
1620 goto fail_sb_buffer;
1623 ret = find_and_setup_root(tree_root, fs_info,
1624 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1626 goto fail_tree_root;
1627 extent_root->track_dirty = 1;
1629 ret = find_and_setup_root(tree_root, fs_info,
1630 BTRFS_DEV_TREE_OBJECTID, dev_root);
1631 dev_root->track_dirty = 1;
1634 goto fail_extent_root;
1636 btrfs_read_block_groups(extent_root);
1638 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1639 fs_info->data_alloc_profile = (u64)-1;
1640 fs_info->metadata_alloc_profile = (u64)-1;
1641 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1642 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1644 if (!fs_info->cleaner_kthread)
1645 goto fail_extent_root;
1647 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1649 "btrfs-transaction");
1650 if (!fs_info->transaction_kthread)
1653 if (btrfs_super_log_root(disk_super) != 0) {
1655 u64 bytenr = btrfs_super_log_root(disk_super);
1658 btrfs_level_size(tree_root,
1659 btrfs_super_log_root_level(disk_super));
1661 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1664 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1665 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1667 log_tree_root->node = read_tree_block(tree_root, bytenr,
1669 ret = btrfs_recover_log_trees(log_tree_root);
1673 ret = btrfs_cleanup_reloc_trees(tree_root);
1676 fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1680 kthread_stop(fs_info->cleaner_kthread);
1682 free_extent_buffer(extent_root->node);
1684 free_extent_buffer(tree_root->node);
1687 btrfs_stop_workers(&fs_info->fixup_workers);
1688 btrfs_stop_workers(&fs_info->workers);
1689 btrfs_stop_workers(&fs_info->endio_workers);
1690 btrfs_stop_workers(&fs_info->endio_write_workers);
1691 btrfs_stop_workers(&fs_info->submit_workers);
1693 iput(fs_info->btree_inode);
1695 btrfs_close_devices(fs_info->fs_devices);
1696 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1700 bdi_destroy(&fs_info->bdi);
1704 return ERR_PTR(err);
1707 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1709 char b[BDEVNAME_SIZE];
1712 set_buffer_uptodate(bh);
1714 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1715 printk(KERN_WARNING "lost page write due to "
1716 "I/O error on %s\n",
1717 bdevname(bh->b_bdev, b));
1719 /* note, we dont' set_buffer_write_io_error because we have
1720 * our own ways of dealing with the IO errors
1722 clear_buffer_uptodate(bh);
1728 int write_all_supers(struct btrfs_root *root)
1730 struct list_head *cur;
1731 struct list_head *head = &root->fs_info->fs_devices->devices;
1732 struct btrfs_device *dev;
1733 struct btrfs_super_block *sb;
1734 struct btrfs_dev_item *dev_item;
1735 struct buffer_head *bh;
1739 int total_errors = 0;
1743 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1744 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1746 sb = &root->fs_info->super_for_commit;
1747 dev_item = &sb->dev_item;
1748 list_for_each(cur, head) {
1749 dev = list_entry(cur, struct btrfs_device, dev_list);
1754 if (!dev->in_fs_metadata)
1757 btrfs_set_stack_device_type(dev_item, dev->type);
1758 btrfs_set_stack_device_id(dev_item, dev->devid);
1759 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1760 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1761 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1762 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1763 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1764 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1765 flags = btrfs_super_flags(sb);
1766 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1770 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1771 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1772 btrfs_csum_final(crc, sb->csum);
1774 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1775 BTRFS_SUPER_INFO_SIZE);
1777 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1778 dev->pending_io = bh;
1781 set_buffer_uptodate(bh);
1783 bh->b_end_io = btrfs_end_buffer_write_sync;
1785 if (do_barriers && dev->barriers) {
1786 ret = submit_bh(WRITE_BARRIER, bh);
1787 if (ret == -EOPNOTSUPP) {
1788 printk("btrfs: disabling barriers on dev %s\n",
1790 set_buffer_uptodate(bh);
1794 ret = submit_bh(WRITE, bh);
1797 ret = submit_bh(WRITE, bh);
1802 if (total_errors > max_errors) {
1803 printk("btrfs: %d errors while writing supers\n", total_errors);
1808 list_for_each(cur, head) {
1809 dev = list_entry(cur, struct btrfs_device, dev_list);
1812 if (!dev->in_fs_metadata)
1815 BUG_ON(!dev->pending_io);
1816 bh = dev->pending_io;
1818 if (!buffer_uptodate(dev->pending_io)) {
1819 if (do_barriers && dev->barriers) {
1820 printk("btrfs: disabling barriers on dev %s\n",
1822 set_buffer_uptodate(bh);
1826 ret = submit_bh(WRITE, bh);
1829 if (!buffer_uptodate(bh))
1836 dev->pending_io = NULL;
1839 if (total_errors > max_errors) {
1840 printk("btrfs: %d errors while writing supers\n", total_errors);
1846 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1851 ret = write_all_supers(root);
1855 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1857 radix_tree_delete(&fs_info->fs_roots_radix,
1858 (unsigned long)root->root_key.objectid);
1860 btrfs_sysfs_del_root(root);
1864 free_extent_buffer(root->node);
1865 if (root->commit_root)
1866 free_extent_buffer(root->commit_root);
1873 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1876 struct btrfs_root *gang[8];
1880 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1885 for (i = 0; i < ret; i++)
1886 btrfs_free_fs_root(fs_info, gang[i]);
1891 int close_ctree(struct btrfs_root *root)
1894 struct btrfs_trans_handle *trans;
1895 struct btrfs_fs_info *fs_info = root->fs_info;
1897 fs_info->closing = 1;
1900 kthread_stop(root->fs_info->transaction_kthread);
1901 kthread_stop(root->fs_info->cleaner_kthread);
1903 btrfs_clean_old_snapshots(root);
1904 trans = btrfs_start_transaction(root, 1);
1905 ret = btrfs_commit_transaction(trans, root);
1906 /* run commit again to drop the original snapshot */
1907 trans = btrfs_start_transaction(root, 1);
1908 btrfs_commit_transaction(trans, root);
1909 ret = btrfs_write_and_wait_transaction(NULL, root);
1912 write_ctree_super(NULL, root);
1914 if (fs_info->delalloc_bytes) {
1915 printk("btrfs: at unmount delalloc count %Lu\n",
1916 fs_info->delalloc_bytes);
1918 if (fs_info->total_ref_cache_size) {
1919 printk("btrfs: at umount reference cache size %Lu\n",
1920 fs_info->total_ref_cache_size);
1923 if (fs_info->extent_root->node)
1924 free_extent_buffer(fs_info->extent_root->node);
1926 if (fs_info->tree_root->node)
1927 free_extent_buffer(fs_info->tree_root->node);
1929 if (root->fs_info->chunk_root->node);
1930 free_extent_buffer(root->fs_info->chunk_root->node);
1932 if (root->fs_info->dev_root->node);
1933 free_extent_buffer(root->fs_info->dev_root->node);
1935 btrfs_free_block_groups(root->fs_info);
1936 fs_info->closing = 2;
1937 del_fs_roots(fs_info);
1939 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1941 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1943 btrfs_stop_workers(&fs_info->fixup_workers);
1944 btrfs_stop_workers(&fs_info->workers);
1945 btrfs_stop_workers(&fs_info->endio_workers);
1946 btrfs_stop_workers(&fs_info->endio_write_workers);
1947 btrfs_stop_workers(&fs_info->submit_workers);
1949 iput(fs_info->btree_inode);
1951 while(!list_empty(&fs_info->hashers)) {
1952 struct btrfs_hasher *hasher;
1953 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1955 list_del(&hasher->hashers);
1956 crypto_free_hash(&fs_info->hash_tfm);
1960 btrfs_close_devices(fs_info->fs_devices);
1961 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1963 bdi_destroy(&fs_info->bdi);
1965 kfree(fs_info->extent_root);
1966 kfree(fs_info->tree_root);
1967 kfree(fs_info->chunk_root);
1968 kfree(fs_info->dev_root);
1972 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1975 struct inode *btree_inode = buf->first_page->mapping->host;
1977 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1981 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1986 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1988 struct inode *btree_inode = buf->first_page->mapping->host;
1989 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1993 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1995 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1996 u64 transid = btrfs_header_generation(buf);
1997 struct inode *btree_inode = root->fs_info->btree_inode;
1999 WARN_ON(!btrfs_tree_locked(buf));
2000 if (transid != root->fs_info->generation) {
2001 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2002 (unsigned long long)buf->start,
2003 transid, root->fs_info->generation);
2006 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2009 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2012 * looks as though older kernels can get into trouble with
2013 * this code, they end up stuck in balance_dirty_pages forever
2015 struct extent_io_tree *tree;
2018 unsigned long thresh = 96 * 1024 * 1024;
2019 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2021 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2024 num_dirty = count_range_bits(tree, &start, (u64)-1,
2025 thresh, EXTENT_DIRTY);
2026 if (num_dirty > thresh) {
2027 balance_dirty_pages_ratelimited_nr(
2028 root->fs_info->btree_inode->i_mapping, 1);
2033 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2035 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2037 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2039 buf->flags |= EXTENT_UPTODATE;
2044 int btree_lock_page_hook(struct page *page)
2046 struct inode *inode = page->mapping->host;
2047 struct btrfs_root *root = BTRFS_I(inode)->root;
2048 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2049 struct extent_buffer *eb;
2051 u64 bytenr = page_offset(page);
2053 if (page->private == EXTENT_PAGE_PRIVATE)
2056 len = page->private >> 2;
2057 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2061 btrfs_tree_lock(eb);
2062 spin_lock(&root->fs_info->hash_lock);
2063 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2064 spin_unlock(&root->fs_info->hash_lock);
2065 btrfs_tree_unlock(eb);
2066 free_extent_buffer(eb);
2072 static struct extent_io_ops btree_extent_io_ops = {
2073 .write_cache_pages_lock_hook = btree_lock_page_hook,
2074 .readpage_end_io_hook = btree_readpage_end_io_hook,
2075 .submit_bio_hook = btree_submit_bio_hook,
2076 /* note we're sharing with inode.c for the merge bio hook */
2077 .merge_bio_hook = btrfs_merge_bio_hook,