2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/version.h>
21 #include <linux/blkdev.h>
22 #include <linux/scatterlist.h>
23 #include <linux/swap.h>
24 #include <linux/radix-tree.h>
25 #include <linux/writeback.h>
26 #include <linux/buffer_head.h> // for block_sync_page
27 #include <linux/workqueue.h>
28 #include <linux/kthread.h>
29 # include <linux/freezer.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
36 #include "print-tree.h"
37 #include "async-thread.h"
39 #include "ref-cache.h"
43 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
45 if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
46 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
47 (unsigned long long)extent_buffer_blocknr(buf),
48 (unsigned long long)btrfs_header_blocknr(buf));
55 static struct extent_io_ops btree_extent_io_ops;
56 static void end_workqueue_fn(struct btrfs_work *work);
59 * end_io_wq structs are used to do processing in task context when an IO is
60 * complete. This is used during reads to verify checksums, and it is used
61 * by writes to insert metadata for new file extents after IO is complete.
67 struct btrfs_fs_info *info;
70 struct list_head list;
71 struct btrfs_work work;
75 * async submit bios are used to offload expensive checksumming
76 * onto the worker threads. They checksum file and metadata bios
77 * just before they are sent down the IO stack.
79 struct async_submit_bio {
82 struct list_head list;
83 extent_submit_bio_hook_t *submit_bio_start;
84 extent_submit_bio_hook_t *submit_bio_done;
87 unsigned long bio_flags;
88 struct btrfs_work work;
92 * extents on the btree inode are pretty simple, there's one extent
93 * that covers the entire device
95 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
96 size_t page_offset, u64 start, u64 len,
99 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
100 struct extent_map *em;
103 spin_lock(&em_tree->lock);
104 em = lookup_extent_mapping(em_tree, start, len);
107 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
108 spin_unlock(&em_tree->lock);
111 spin_unlock(&em_tree->lock);
113 em = alloc_extent_map(GFP_NOFS);
115 em = ERR_PTR(-ENOMEM);
120 em->block_len = (u64)-1;
122 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
124 spin_lock(&em_tree->lock);
125 ret = add_extent_mapping(em_tree, em);
126 if (ret == -EEXIST) {
127 u64 failed_start = em->start;
128 u64 failed_len = em->len;
130 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
131 em->start, em->len, em->block_start);
133 em = lookup_extent_mapping(em_tree, start, len);
135 printk("after failing, found %Lu %Lu %Lu\n",
136 em->start, em->len, em->block_start);
139 em = lookup_extent_mapping(em_tree, failed_start,
142 printk("double failure lookup gives us "
143 "%Lu %Lu -> %Lu\n", em->start,
144 em->len, em->block_start);
153 spin_unlock(&em_tree->lock);
161 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
163 return btrfs_crc32c(seed, data, len);
166 void btrfs_csum_final(u32 crc, char *result)
168 *(__le32 *)result = ~cpu_to_le32(crc);
172 * compute the csum for a btree block, and either verify it or write it
173 * into the csum field of the block.
175 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
178 char result[BTRFS_CRC32_SIZE];
180 unsigned long cur_len;
181 unsigned long offset = BTRFS_CSUM_SIZE;
182 char *map_token = NULL;
184 unsigned long map_start;
185 unsigned long map_len;
189 len = buf->len - offset;
191 err = map_private_extent_buffer(buf, offset, 32,
193 &map_start, &map_len, KM_USER0);
195 printk("failed to map extent buffer! %lu\n",
199 cur_len = min(len, map_len - (offset - map_start));
200 crc = btrfs_csum_data(root, kaddr + offset - map_start,
204 unmap_extent_buffer(buf, map_token, KM_USER0);
206 btrfs_csum_final(crc, result);
209 /* FIXME, this is not good */
210 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
213 memcpy(&found, result, BTRFS_CRC32_SIZE);
215 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
216 printk("btrfs: %s checksum verify failed on %llu "
217 "wanted %X found %X level %d\n",
218 root->fs_info->sb->s_id,
219 buf->start, val, found, btrfs_header_level(buf));
223 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
229 * we can't consider a given block up to date unless the transid of the
230 * block matches the transid in the parent node's pointer. This is how we
231 * detect blocks that either didn't get written at all or got written
232 * in the wrong place.
234 static int verify_parent_transid(struct extent_io_tree *io_tree,
235 struct extent_buffer *eb, u64 parent_transid)
239 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
242 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
243 if (extent_buffer_uptodate(io_tree, eb) &&
244 btrfs_header_generation(eb) == parent_transid) {
248 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
249 (unsigned long long)eb->start,
250 (unsigned long long)parent_transid,
251 (unsigned long long)btrfs_header_generation(eb));
253 clear_extent_buffer_uptodate(io_tree, eb);
255 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
261 * helper to read a given tree block, doing retries as required when
262 * the checksums don't match and we have alternate mirrors to try.
264 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
265 struct extent_buffer *eb,
266 u64 start, u64 parent_transid)
268 struct extent_io_tree *io_tree;
273 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
275 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
276 btree_get_extent, mirror_num);
278 !verify_parent_transid(io_tree, eb, parent_transid))
280 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
281 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
287 if (mirror_num > num_copies)
294 * checksum a dirty tree block before IO. This has extra checks to make
295 * sure we only fill in the checksum field in the first page of a multi-page block
297 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
299 struct extent_io_tree *tree;
300 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
304 struct extent_buffer *eb;
307 tree = &BTRFS_I(page->mapping->host)->io_tree;
309 if (page->private == EXTENT_PAGE_PRIVATE)
313 len = page->private >> 2;
317 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
318 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
319 btrfs_header_generation(eb));
321 found_start = btrfs_header_bytenr(eb);
322 if (found_start != start) {
323 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
324 start, found_start, len);
328 if (eb->first_page != page) {
329 printk("bad first page %lu %lu\n", eb->first_page->index,
334 if (!PageUptodate(page)) {
335 printk("csum not up to date page %lu\n", page->index);
339 found_level = btrfs_header_level(eb);
341 csum_tree_block(root, eb, 0);
343 free_extent_buffer(eb);
348 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
349 struct extent_state *state)
351 struct extent_io_tree *tree;
355 struct extent_buffer *eb;
356 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
359 tree = &BTRFS_I(page->mapping->host)->io_tree;
360 if (page->private == EXTENT_PAGE_PRIVATE)
364 len = page->private >> 2;
368 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
370 found_start = btrfs_header_bytenr(eb);
371 if (found_start != start) {
372 printk("bad tree block start %llu %llu\n",
373 (unsigned long long)found_start,
374 (unsigned long long)eb->start);
378 if (eb->first_page != page) {
379 printk("bad first page %lu %lu\n", eb->first_page->index,
385 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
386 (unsigned long)btrfs_header_fsid(eb),
388 printk("bad fsid on block %Lu\n", eb->start);
392 found_level = btrfs_header_level(eb);
394 ret = csum_tree_block(root, eb, 1);
398 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
399 end = eb->start + end - 1;
401 free_extent_buffer(eb);
406 static void end_workqueue_bio(struct bio *bio, int err)
408 struct end_io_wq *end_io_wq = bio->bi_private;
409 struct btrfs_fs_info *fs_info;
411 fs_info = end_io_wq->info;
412 end_io_wq->error = err;
413 end_io_wq->work.func = end_workqueue_fn;
414 end_io_wq->work.flags = 0;
415 if (bio->bi_rw & (1 << BIO_RW))
416 btrfs_queue_worker(&fs_info->endio_write_workers,
419 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
422 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
425 struct end_io_wq *end_io_wq;
426 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
430 end_io_wq->private = bio->bi_private;
431 end_io_wq->end_io = bio->bi_end_io;
432 end_io_wq->info = info;
433 end_io_wq->error = 0;
434 end_io_wq->bio = bio;
435 end_io_wq->metadata = metadata;
437 bio->bi_private = end_io_wq;
438 bio->bi_end_io = end_workqueue_bio;
442 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
444 unsigned long limit = min_t(unsigned long,
445 info->workers.max_workers,
446 info->fs_devices->open_devices);
450 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
452 return atomic_read(&info->nr_async_bios) >
453 btrfs_async_submit_limit(info);
456 static void run_one_async_start(struct btrfs_work *work)
458 struct btrfs_fs_info *fs_info;
459 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;
463 async->submit_bio_start(async->inode, async->rw, async->bio,
464 async->mirror_num, async->bio_flags);
467 static void run_one_async_done(struct btrfs_work *work)
469 struct btrfs_fs_info *fs_info;
470 struct async_submit_bio *async;
473 async = container_of(work, struct async_submit_bio, work);
474 fs_info = BTRFS_I(async->inode)->root->fs_info;
476 limit = btrfs_async_submit_limit(fs_info);
477 limit = limit * 2 / 3;
479 atomic_dec(&fs_info->nr_async_submits);
481 if (atomic_read(&fs_info->nr_async_submits) < limit &&
482 waitqueue_active(&fs_info->async_submit_wait))
483 wake_up(&fs_info->async_submit_wait);
485 async->submit_bio_done(async->inode, async->rw, async->bio,
486 async->mirror_num, async->bio_flags);
489 static void run_one_async_free(struct btrfs_work *work)
491 struct async_submit_bio *async;
493 async = container_of(work, struct async_submit_bio, work);
497 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
498 int rw, struct bio *bio, int mirror_num,
499 unsigned long bio_flags,
500 extent_submit_bio_hook_t *submit_bio_start,
501 extent_submit_bio_hook_t *submit_bio_done)
503 struct async_submit_bio *async;
504 int limit = btrfs_async_submit_limit(fs_info);
506 async = kmalloc(sizeof(*async), GFP_NOFS);
510 async->inode = inode;
513 async->mirror_num = mirror_num;
514 async->submit_bio_start = submit_bio_start;
515 async->submit_bio_done = submit_bio_done;
517 async->work.func = run_one_async_start;
518 async->work.ordered_func = run_one_async_done;
519 async->work.ordered_free = run_one_async_free;
521 async->work.flags = 0;
522 async->bio_flags = bio_flags;
524 while(atomic_read(&fs_info->async_submit_draining) &&
525 atomic_read(&fs_info->nr_async_submits)) {
526 wait_event(fs_info->async_submit_wait,
527 (atomic_read(&fs_info->nr_async_submits) == 0));
530 atomic_inc(&fs_info->nr_async_submits);
531 btrfs_queue_worker(&fs_info->workers, &async->work);
533 if (atomic_read(&fs_info->nr_async_submits) > limit) {
534 wait_event_timeout(fs_info->async_submit_wait,
535 (atomic_read(&fs_info->nr_async_submits) < limit),
538 wait_event_timeout(fs_info->async_submit_wait,
539 (atomic_read(&fs_info->nr_async_bios) < limit),
545 static int btree_csum_one_bio(struct bio *bio)
547 struct bio_vec *bvec = bio->bi_io_vec;
549 struct btrfs_root *root;
551 WARN_ON(bio->bi_vcnt <= 0);
552 while(bio_index < bio->bi_vcnt) {
553 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
554 csum_dirty_buffer(root, bvec->bv_page);
561 static int __btree_submit_bio_start(struct inode *inode, int rw,
562 struct bio *bio, int mirror_num,
563 unsigned long bio_flags)
566 * when we're called for a write, we're already in the async
567 * submission context. Just jump into btrfs_map_bio
569 btree_csum_one_bio(bio);
573 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
574 int mirror_num, unsigned long bio_flags)
577 * when we're called for a write, we're already in the async
578 * submission context. Just jump into btrfs_map_bio
580 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
583 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
584 int mirror_num, unsigned long bio_flags)
587 * kthread helpers are used to submit writes so that checksumming
588 * can happen in parallel across all CPUs
590 if (!(rw & (1 << BIO_RW))) {
593 * called for a read, do the setup so that checksum validation
594 * can happen in the async kernel threads
596 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
600 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
603 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
604 inode, rw, bio, mirror_num, 0,
605 __btree_submit_bio_start,
606 __btree_submit_bio_done);
609 static int btree_writepage(struct page *page, struct writeback_control *wbc)
611 struct extent_io_tree *tree;
612 tree = &BTRFS_I(page->mapping->host)->io_tree;
614 if (current->flags & PF_MEMALLOC) {
615 redirty_page_for_writepage(wbc, page);
619 return extent_write_full_page(tree, page, btree_get_extent, wbc);
622 static int btree_writepages(struct address_space *mapping,
623 struct writeback_control *wbc)
625 struct extent_io_tree *tree;
626 tree = &BTRFS_I(mapping->host)->io_tree;
627 if (wbc->sync_mode == WB_SYNC_NONE) {
630 unsigned long thresh = 32 * 1024 * 1024;
632 if (wbc->for_kupdate)
635 num_dirty = count_range_bits(tree, &start, (u64)-1,
636 thresh, EXTENT_DIRTY);
637 if (num_dirty < thresh) {
641 return extent_writepages(tree, mapping, btree_get_extent, wbc);
644 int btree_readpage(struct file *file, struct page *page)
646 struct extent_io_tree *tree;
647 tree = &BTRFS_I(page->mapping->host)->io_tree;
648 return extent_read_full_page(tree, page, btree_get_extent);
651 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
653 struct extent_io_tree *tree;
654 struct extent_map_tree *map;
657 if (PageWriteback(page) || PageDirty(page))
660 tree = &BTRFS_I(page->mapping->host)->io_tree;
661 map = &BTRFS_I(page->mapping->host)->extent_tree;
663 ret = try_release_extent_state(map, tree, page, gfp_flags);
668 ret = try_release_extent_buffer(tree, page);
670 ClearPagePrivate(page);
671 set_page_private(page, 0);
672 page_cache_release(page);
678 static void btree_invalidatepage(struct page *page, unsigned long offset)
680 struct extent_io_tree *tree;
681 tree = &BTRFS_I(page->mapping->host)->io_tree;
682 extent_invalidatepage(tree, page, offset);
683 btree_releasepage(page, GFP_NOFS);
684 if (PagePrivate(page)) {
685 printk("warning page private not zero on page %Lu\n",
687 ClearPagePrivate(page);
688 set_page_private(page, 0);
689 page_cache_release(page);
694 static int btree_writepage(struct page *page, struct writeback_control *wbc)
696 struct buffer_head *bh;
697 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
698 struct buffer_head *head;
699 if (!page_has_buffers(page)) {
700 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
701 (1 << BH_Dirty)|(1 << BH_Uptodate));
703 head = page_buffers(page);
706 if (buffer_dirty(bh))
707 csum_tree_block(root, bh, 0);
708 bh = bh->b_this_page;
709 } while (bh != head);
710 return block_write_full_page(page, btree_get_block, wbc);
714 static struct address_space_operations btree_aops = {
715 .readpage = btree_readpage,
716 .writepage = btree_writepage,
717 .writepages = btree_writepages,
718 .releasepage = btree_releasepage,
719 .invalidatepage = btree_invalidatepage,
720 .sync_page = block_sync_page,
723 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
726 struct extent_buffer *buf = NULL;
727 struct inode *btree_inode = root->fs_info->btree_inode;
730 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
733 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
734 buf, 0, 0, btree_get_extent, 0);
735 free_extent_buffer(buf);
739 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
740 u64 bytenr, u32 blocksize)
742 struct inode *btree_inode = root->fs_info->btree_inode;
743 struct extent_buffer *eb;
744 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
745 bytenr, blocksize, GFP_NOFS);
749 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
750 u64 bytenr, u32 blocksize)
752 struct inode *btree_inode = root->fs_info->btree_inode;
753 struct extent_buffer *eb;
755 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
756 bytenr, blocksize, NULL, GFP_NOFS);
761 int btrfs_write_tree_block(struct extent_buffer *buf)
763 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
764 buf->start + buf->len - 1, WB_SYNC_ALL);
767 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
769 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
770 buf->start, buf->start + buf->len -1);
773 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
774 u32 blocksize, u64 parent_transid)
776 struct extent_buffer *buf = NULL;
777 struct inode *btree_inode = root->fs_info->btree_inode;
778 struct extent_io_tree *io_tree;
781 io_tree = &BTRFS_I(btree_inode)->io_tree;
783 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
787 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
790 buf->flags |= EXTENT_UPTODATE;
798 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
799 struct extent_buffer *buf)
801 struct inode *btree_inode = root->fs_info->btree_inode;
802 if (btrfs_header_generation(buf) ==
803 root->fs_info->running_transaction->transid) {
804 WARN_ON(!btrfs_tree_locked(buf));
805 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
811 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
812 u32 stripesize, struct btrfs_root *root,
813 struct btrfs_fs_info *fs_info,
818 root->commit_root = NULL;
819 root->ref_tree = NULL;
820 root->sectorsize = sectorsize;
821 root->nodesize = nodesize;
822 root->leafsize = leafsize;
823 root->stripesize = stripesize;
825 root->track_dirty = 0;
827 root->fs_info = fs_info;
828 root->objectid = objectid;
829 root->last_trans = 0;
830 root->highest_inode = 0;
831 root->last_inode_alloc = 0;
835 INIT_LIST_HEAD(&root->dirty_list);
836 INIT_LIST_HEAD(&root->orphan_list);
837 INIT_LIST_HEAD(&root->dead_list);
838 spin_lock_init(&root->node_lock);
839 spin_lock_init(&root->list_lock);
840 mutex_init(&root->objectid_mutex);
841 mutex_init(&root->log_mutex);
842 extent_io_tree_init(&root->dirty_log_pages,
843 fs_info->btree_inode->i_mapping, GFP_NOFS);
845 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
846 root->ref_tree = &root->ref_tree_struct;
848 memset(&root->root_key, 0, sizeof(root->root_key));
849 memset(&root->root_item, 0, sizeof(root->root_item));
850 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
851 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
852 root->defrag_trans_start = fs_info->generation;
853 init_completion(&root->kobj_unregister);
854 root->defrag_running = 0;
855 root->defrag_level = 0;
856 root->root_key.objectid = objectid;
860 static int find_and_setup_root(struct btrfs_root *tree_root,
861 struct btrfs_fs_info *fs_info,
863 struct btrfs_root *root)
869 __setup_root(tree_root->nodesize, tree_root->leafsize,
870 tree_root->sectorsize, tree_root->stripesize,
871 root, fs_info, objectid);
872 ret = btrfs_find_last_root(tree_root, objectid,
873 &root->root_item, &root->root_key);
876 generation = btrfs_root_generation(&root->root_item);
877 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
878 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
879 blocksize, generation);
884 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
885 struct btrfs_fs_info *fs_info)
887 struct extent_buffer *eb;
888 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
897 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
898 0, &start, &end, EXTENT_DIRTY);
902 clear_extent_dirty(&log_root_tree->dirty_log_pages,
903 start, end, GFP_NOFS);
905 eb = fs_info->log_root_tree->node;
907 WARN_ON(btrfs_header_level(eb) != 0);
908 WARN_ON(btrfs_header_nritems(eb) != 0);
910 ret = btrfs_free_reserved_extent(fs_info->tree_root,
914 free_extent_buffer(eb);
915 kfree(fs_info->log_root_tree);
916 fs_info->log_root_tree = NULL;
920 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
921 struct btrfs_fs_info *fs_info)
923 struct btrfs_root *root;
924 struct btrfs_root *tree_root = fs_info->tree_root;
926 root = kzalloc(sizeof(*root), GFP_NOFS);
930 __setup_root(tree_root->nodesize, tree_root->leafsize,
931 tree_root->sectorsize, tree_root->stripesize,
932 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
934 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
935 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
936 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
939 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
940 0, BTRFS_TREE_LOG_OBJECTID,
941 trans->transid, 0, 0, 0);
943 btrfs_set_header_nritems(root->node, 0);
944 btrfs_set_header_level(root->node, 0);
945 btrfs_set_header_bytenr(root->node, root->node->start);
946 btrfs_set_header_generation(root->node, trans->transid);
947 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
949 write_extent_buffer(root->node, root->fs_info->fsid,
950 (unsigned long)btrfs_header_fsid(root->node),
952 btrfs_mark_buffer_dirty(root->node);
953 btrfs_tree_unlock(root->node);
954 fs_info->log_root_tree = root;
958 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
959 struct btrfs_key *location)
961 struct btrfs_root *root;
962 struct btrfs_fs_info *fs_info = tree_root->fs_info;
963 struct btrfs_path *path;
964 struct extent_buffer *l;
970 root = kzalloc(sizeof(*root), GFP_NOFS);
972 return ERR_PTR(-ENOMEM);
973 if (location->offset == (u64)-1) {
974 ret = find_and_setup_root(tree_root, fs_info,
975 location->objectid, root);
983 __setup_root(tree_root->nodesize, tree_root->leafsize,
984 tree_root->sectorsize, tree_root->stripesize,
985 root, fs_info, location->objectid);
987 path = btrfs_alloc_path();
989 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
996 read_extent_buffer(l, &root->root_item,
997 btrfs_item_ptr_offset(l, path->slots[0]),
998 sizeof(root->root_item));
999 memcpy(&root->root_key, location, sizeof(*location));
1002 btrfs_release_path(root, path);
1003 btrfs_free_path(path);
1006 return ERR_PTR(ret);
1008 generation = btrfs_root_generation(&root->root_item);
1009 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1010 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1011 blocksize, generation);
1012 BUG_ON(!root->node);
1014 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1016 ret = btrfs_find_highest_inode(root, &highest_inode);
1018 root->highest_inode = highest_inode;
1019 root->last_inode_alloc = highest_inode;
1025 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1028 struct btrfs_root *root;
1030 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1031 return fs_info->tree_root;
1032 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1033 return fs_info->extent_root;
1035 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1036 (unsigned long)root_objectid);
1040 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1041 struct btrfs_key *location)
1043 struct btrfs_root *root;
1046 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1047 return fs_info->tree_root;
1048 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1049 return fs_info->extent_root;
1050 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1051 return fs_info->chunk_root;
1052 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1053 return fs_info->dev_root;
1055 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1056 (unsigned long)location->objectid);
1060 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1063 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1064 (unsigned long)root->root_key.objectid,
1067 free_extent_buffer(root->node);
1069 return ERR_PTR(ret);
1071 ret = btrfs_find_dead_roots(fs_info->tree_root,
1072 root->root_key.objectid, root);
1078 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1079 struct btrfs_key *location,
1080 const char *name, int namelen)
1082 struct btrfs_root *root;
1085 root = btrfs_read_fs_root_no_name(fs_info, location);
1092 ret = btrfs_set_root_name(root, name, namelen);
1094 free_extent_buffer(root->node);
1096 return ERR_PTR(ret);
1099 ret = btrfs_sysfs_add_root(root);
1101 free_extent_buffer(root->node);
1104 return ERR_PTR(ret);
1110 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1111 struct btrfs_hasher *hasher;
1113 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1116 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1117 if (!hasher->hash_tfm) {
1121 spin_lock(&info->hash_lock);
1122 list_add(&hasher->list, &info->hashers);
1123 spin_unlock(&info->hash_lock);
1128 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1130 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1132 struct list_head *cur;
1133 struct btrfs_device *device;
1134 struct backing_dev_info *bdi;
1136 if ((bdi_bits & (1 << BDI_write_congested)) &&
1137 btrfs_congested_async(info, 0))
1140 list_for_each(cur, &info->fs_devices->devices) {
1141 device = list_entry(cur, struct btrfs_device, dev_list);
1144 bdi = blk_get_backing_dev_info(device->bdev);
1145 if (bdi && bdi_congested(bdi, bdi_bits)) {
1154 * this unplugs every device on the box, and it is only used when page
1157 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1159 struct list_head *cur;
1160 struct btrfs_device *device;
1161 struct btrfs_fs_info *info;
1163 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1164 list_for_each(cur, &info->fs_devices->devices) {
1165 device = list_entry(cur, struct btrfs_device, dev_list);
1166 bdi = blk_get_backing_dev_info(device->bdev);
1167 if (bdi->unplug_io_fn) {
1168 bdi->unplug_io_fn(bdi, page);
1173 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1175 struct inode *inode;
1176 struct extent_map_tree *em_tree;
1177 struct extent_map *em;
1178 struct address_space *mapping;
1181 /* the generic O_DIRECT read code does this */
1183 __unplug_io_fn(bdi, page);
1188 * page->mapping may change at any time. Get a consistent copy
1189 * and use that for everything below
1192 mapping = page->mapping;
1196 inode = mapping->host;
1197 offset = page_offset(page);
1199 em_tree = &BTRFS_I(inode)->extent_tree;
1200 spin_lock(&em_tree->lock);
1201 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1202 spin_unlock(&em_tree->lock);
1204 __unplug_io_fn(bdi, page);
1208 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1209 free_extent_map(em);
1210 __unplug_io_fn(bdi, page);
1213 offset = offset - em->start;
1214 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1215 em->block_start + offset, page);
1216 free_extent_map(em);
1219 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1222 bdi->ra_pages = default_backing_dev_info.ra_pages;
1224 bdi->capabilities = default_backing_dev_info.capabilities;
1225 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1226 bdi->unplug_io_data = info;
1227 bdi->congested_fn = btrfs_congested_fn;
1228 bdi->congested_data = info;
1232 static int bio_ready_for_csum(struct bio *bio)
1238 struct extent_io_tree *io_tree = NULL;
1239 struct btrfs_fs_info *info = NULL;
1240 struct bio_vec *bvec;
1244 bio_for_each_segment(bvec, bio, i) {
1245 page = bvec->bv_page;
1246 if (page->private == EXTENT_PAGE_PRIVATE) {
1247 length += bvec->bv_len;
1250 if (!page->private) {
1251 length += bvec->bv_len;
1254 length = bvec->bv_len;
1255 buf_len = page->private >> 2;
1256 start = page_offset(page) + bvec->bv_offset;
1257 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1258 info = BTRFS_I(page->mapping->host)->root->fs_info;
1260 /* are we fully contained in this bio? */
1261 if (buf_len <= length)
1264 ret = extent_range_uptodate(io_tree, start + length,
1265 start + buf_len - 1);
1272 * called by the kthread helper functions to finally call the bio end_io
1273 * functions. This is where read checksum verification actually happens
1275 static void end_workqueue_fn(struct btrfs_work *work)
1278 struct end_io_wq *end_io_wq;
1279 struct btrfs_fs_info *fs_info;
1282 end_io_wq = container_of(work, struct end_io_wq, work);
1283 bio = end_io_wq->bio;
1284 fs_info = end_io_wq->info;
1286 /* metadata bios are special because the whole tree block must
1287 * be checksummed at once. This makes sure the entire block is in
1288 * ram and up to date before trying to verify things. For
1289 * blocksize <= pagesize, it is basically a noop
1291 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1292 btrfs_queue_worker(&fs_info->endio_workers,
1296 error = end_io_wq->error;
1297 bio->bi_private = end_io_wq->private;
1298 bio->bi_end_io = end_io_wq->end_io;
1300 bio_endio(bio, error);
1303 static int cleaner_kthread(void *arg)
1305 struct btrfs_root *root = arg;
1309 if (root->fs_info->closing)
1312 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1313 mutex_lock(&root->fs_info->cleaner_mutex);
1314 btrfs_clean_old_snapshots(root);
1315 mutex_unlock(&root->fs_info->cleaner_mutex);
1317 if (freezing(current)) {
1321 if (root->fs_info->closing)
1323 set_current_state(TASK_INTERRUPTIBLE);
1325 __set_current_state(TASK_RUNNING);
1327 } while (!kthread_should_stop());
1331 static int transaction_kthread(void *arg)
1333 struct btrfs_root *root = arg;
1334 struct btrfs_trans_handle *trans;
1335 struct btrfs_transaction *cur;
1337 unsigned long delay;
1342 if (root->fs_info->closing)
1346 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1347 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1349 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1350 printk("btrfs: total reference cache size %Lu\n",
1351 root->fs_info->total_ref_cache_size);
1354 mutex_lock(&root->fs_info->trans_mutex);
1355 cur = root->fs_info->running_transaction;
1357 mutex_unlock(&root->fs_info->trans_mutex);
1361 now = get_seconds();
1362 if (now < cur->start_time || now - cur->start_time < 30) {
1363 mutex_unlock(&root->fs_info->trans_mutex);
1367 mutex_unlock(&root->fs_info->trans_mutex);
1368 trans = btrfs_start_transaction(root, 1);
1369 ret = btrfs_commit_transaction(trans, root);
1371 wake_up_process(root->fs_info->cleaner_kthread);
1372 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1374 if (freezing(current)) {
1377 if (root->fs_info->closing)
1379 set_current_state(TASK_INTERRUPTIBLE);
1380 schedule_timeout(delay);
1381 __set_current_state(TASK_RUNNING);
1383 } while (!kthread_should_stop());
1387 struct btrfs_root *open_ctree(struct super_block *sb,
1388 struct btrfs_fs_devices *fs_devices,
1397 struct buffer_head *bh;
1398 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1400 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1402 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1404 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1406 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1408 struct btrfs_root *log_tree_root;
1413 struct btrfs_super_block *disk_super;
1415 if (!extent_root || !tree_root || !fs_info ||
1416 !chunk_root || !dev_root) {
1420 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1421 INIT_LIST_HEAD(&fs_info->trans_list);
1422 INIT_LIST_HEAD(&fs_info->dead_roots);
1423 INIT_LIST_HEAD(&fs_info->hashers);
1424 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1425 spin_lock_init(&fs_info->hash_lock);
1426 spin_lock_init(&fs_info->delalloc_lock);
1427 spin_lock_init(&fs_info->new_trans_lock);
1428 spin_lock_init(&fs_info->ref_cache_lock);
1430 init_completion(&fs_info->kobj_unregister);
1431 fs_info->tree_root = tree_root;
1432 fs_info->extent_root = extent_root;
1433 fs_info->chunk_root = chunk_root;
1434 fs_info->dev_root = dev_root;
1435 fs_info->fs_devices = fs_devices;
1436 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1437 INIT_LIST_HEAD(&fs_info->space_info);
1438 btrfs_mapping_init(&fs_info->mapping_tree);
1439 atomic_set(&fs_info->nr_async_submits, 0);
1440 atomic_set(&fs_info->async_submit_draining, 0);
1441 atomic_set(&fs_info->nr_async_bios, 0);
1442 atomic_set(&fs_info->throttles, 0);
1443 atomic_set(&fs_info->throttle_gen, 0);
1445 fs_info->max_extent = (u64)-1;
1446 fs_info->max_inline = 8192 * 1024;
1447 setup_bdi(fs_info, &fs_info->bdi);
1448 fs_info->btree_inode = new_inode(sb);
1449 fs_info->btree_inode->i_ino = 1;
1450 fs_info->btree_inode->i_nlink = 1;
1452 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1454 INIT_LIST_HEAD(&fs_info->ordered_extents);
1455 spin_lock_init(&fs_info->ordered_extent_lock);
1457 sb->s_blocksize = 4096;
1458 sb->s_blocksize_bits = blksize_bits(4096);
1461 * we set the i_size on the btree inode to the max possible int.
1462 * the real end of the address space is determined by all of
1463 * the devices in the system
1465 fs_info->btree_inode->i_size = OFFSET_MAX;
1466 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1467 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1469 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1470 fs_info->btree_inode->i_mapping,
1472 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1475 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1477 spin_lock_init(&fs_info->block_group_cache_lock);
1478 fs_info->block_group_cache_tree.rb_node = NULL;
1480 extent_io_tree_init(&fs_info->pinned_extents,
1481 fs_info->btree_inode->i_mapping, GFP_NOFS);
1482 extent_io_tree_init(&fs_info->pending_del,
1483 fs_info->btree_inode->i_mapping, GFP_NOFS);
1484 extent_io_tree_init(&fs_info->extent_ins,
1485 fs_info->btree_inode->i_mapping, GFP_NOFS);
1486 fs_info->do_barriers = 1;
1488 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1489 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1490 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1492 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1493 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1494 sizeof(struct btrfs_key));
1495 insert_inode_hash(fs_info->btree_inode);
1497 mutex_init(&fs_info->trans_mutex);
1498 mutex_init(&fs_info->tree_log_mutex);
1499 mutex_init(&fs_info->drop_mutex);
1500 mutex_init(&fs_info->extent_ins_mutex);
1501 mutex_init(&fs_info->pinned_mutex);
1502 mutex_init(&fs_info->chunk_mutex);
1503 mutex_init(&fs_info->transaction_kthread_mutex);
1504 mutex_init(&fs_info->cleaner_mutex);
1505 mutex_init(&fs_info->volume_mutex);
1506 mutex_init(&fs_info->tree_reloc_mutex);
1507 init_waitqueue_head(&fs_info->transaction_throttle);
1508 init_waitqueue_head(&fs_info->transaction_wait);
1509 init_waitqueue_head(&fs_info->async_submit_wait);
1510 init_waitqueue_head(&fs_info->tree_log_wait);
1511 atomic_set(&fs_info->tree_log_commit, 0);
1512 atomic_set(&fs_info->tree_log_writers, 0);
1513 fs_info->tree_log_transid = 0;
1516 ret = add_hasher(fs_info, "crc32c");
1518 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1523 __setup_root(4096, 4096, 4096, 4096, tree_root,
1524 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1527 bh = __bread(fs_devices->latest_bdev,
1528 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1532 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1535 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1537 disk_super = &fs_info->super_copy;
1538 if (!btrfs_super_root(disk_super))
1539 goto fail_sb_buffer;
1541 err = btrfs_parse_options(tree_root, options);
1543 goto fail_sb_buffer;
1546 * we need to start all the end_io workers up front because the
1547 * queue work function gets called at interrupt time, and so it
1548 * cannot dynamically grow.
1550 btrfs_init_workers(&fs_info->workers, "worker",
1551 fs_info->thread_pool_size);
1553 btrfs_init_workers(&fs_info->submit_workers, "submit",
1554 min_t(u64, fs_devices->num_devices,
1555 fs_info->thread_pool_size));
1557 /* a higher idle thresh on the submit workers makes it much more
1558 * likely that bios will be send down in a sane order to the
1561 fs_info->submit_workers.idle_thresh = 64;
1563 /* fs_info->workers is responsible for checksumming file data
1564 * blocks and metadata. Using a larger idle thresh allows each
1565 * worker thread to operate on things in roughly the order they
1566 * were sent by the writeback daemons, improving overall locality
1567 * of the IO going down the pipe.
1569 fs_info->workers.idle_thresh = 8;
1570 fs_info->workers.ordered = 1;
1572 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1573 btrfs_init_workers(&fs_info->endio_workers, "endio",
1574 fs_info->thread_pool_size);
1575 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1576 fs_info->thread_pool_size);
1579 * endios are largely parallel and should have a very
1582 fs_info->endio_workers.idle_thresh = 4;
1583 fs_info->endio_write_workers.idle_thresh = 64;
1585 btrfs_start_workers(&fs_info->workers, 1);
1586 btrfs_start_workers(&fs_info->submit_workers, 1);
1587 btrfs_start_workers(&fs_info->fixup_workers, 1);
1588 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1589 btrfs_start_workers(&fs_info->endio_write_workers,
1590 fs_info->thread_pool_size);
1593 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1594 printk("Btrfs: wanted %llu devices, but found %llu\n",
1595 (unsigned long long)btrfs_super_num_devices(disk_super),
1596 (unsigned long long)fs_devices->open_devices);
1597 if (btrfs_test_opt(tree_root, DEGRADED))
1598 printk("continuing in degraded mode\n");
1600 goto fail_sb_buffer;
1604 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1605 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1606 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1608 nodesize = btrfs_super_nodesize(disk_super);
1609 leafsize = btrfs_super_leafsize(disk_super);
1610 sectorsize = btrfs_super_sectorsize(disk_super);
1611 stripesize = btrfs_super_stripesize(disk_super);
1612 tree_root->nodesize = nodesize;
1613 tree_root->leafsize = leafsize;
1614 tree_root->sectorsize = sectorsize;
1615 tree_root->stripesize = stripesize;
1617 sb->s_blocksize = sectorsize;
1618 sb->s_blocksize_bits = blksize_bits(sectorsize);
1620 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1621 sizeof(disk_super->magic))) {
1622 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1623 goto fail_sb_buffer;
1626 mutex_lock(&fs_info->chunk_mutex);
1627 ret = btrfs_read_sys_array(tree_root);
1628 mutex_unlock(&fs_info->chunk_mutex);
1630 printk("btrfs: failed to read the system array on %s\n",
1632 goto fail_sys_array;
1635 blocksize = btrfs_level_size(tree_root,
1636 btrfs_super_chunk_root_level(disk_super));
1637 generation = btrfs_super_chunk_root_generation(disk_super);
1639 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1640 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1642 chunk_root->node = read_tree_block(chunk_root,
1643 btrfs_super_chunk_root(disk_super),
1644 blocksize, generation);
1645 BUG_ON(!chunk_root->node);
1647 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1648 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1651 mutex_lock(&fs_info->chunk_mutex);
1652 ret = btrfs_read_chunk_tree(chunk_root);
1653 mutex_unlock(&fs_info->chunk_mutex);
1656 btrfs_close_extra_devices(fs_devices);
1658 blocksize = btrfs_level_size(tree_root,
1659 btrfs_super_root_level(disk_super));
1660 generation = btrfs_super_generation(disk_super);
1662 tree_root->node = read_tree_block(tree_root,
1663 btrfs_super_root(disk_super),
1664 blocksize, generation);
1665 if (!tree_root->node)
1666 goto fail_sb_buffer;
1669 ret = find_and_setup_root(tree_root, fs_info,
1670 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1672 goto fail_tree_root;
1673 extent_root->track_dirty = 1;
1675 ret = find_and_setup_root(tree_root, fs_info,
1676 BTRFS_DEV_TREE_OBJECTID, dev_root);
1677 dev_root->track_dirty = 1;
1680 goto fail_extent_root;
1682 btrfs_read_block_groups(extent_root);
1684 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1685 fs_info->data_alloc_profile = (u64)-1;
1686 fs_info->metadata_alloc_profile = (u64)-1;
1687 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1688 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1690 if (!fs_info->cleaner_kthread)
1691 goto fail_extent_root;
1693 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1695 "btrfs-transaction");
1696 if (!fs_info->transaction_kthread)
1699 if (btrfs_super_log_root(disk_super) != 0) {
1701 u64 bytenr = btrfs_super_log_root(disk_super);
1704 btrfs_level_size(tree_root,
1705 btrfs_super_log_root_level(disk_super));
1707 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1710 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1711 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1713 log_tree_root->node = read_tree_block(tree_root, bytenr,
1716 ret = btrfs_recover_log_trees(log_tree_root);
1719 fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1721 ret = btrfs_cleanup_reloc_trees(tree_root);
1727 kthread_stop(fs_info->cleaner_kthread);
1729 free_extent_buffer(extent_root->node);
1731 free_extent_buffer(tree_root->node);
1734 btrfs_stop_workers(&fs_info->fixup_workers);
1735 btrfs_stop_workers(&fs_info->workers);
1736 btrfs_stop_workers(&fs_info->endio_workers);
1737 btrfs_stop_workers(&fs_info->endio_write_workers);
1738 btrfs_stop_workers(&fs_info->submit_workers);
1740 iput(fs_info->btree_inode);
1742 btrfs_close_devices(fs_info->fs_devices);
1743 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1747 bdi_destroy(&fs_info->bdi);
1751 return ERR_PTR(err);
1754 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1756 char b[BDEVNAME_SIZE];
1759 set_buffer_uptodate(bh);
1761 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1762 printk(KERN_WARNING "lost page write due to "
1763 "I/O error on %s\n",
1764 bdevname(bh->b_bdev, b));
1766 /* note, we dont' set_buffer_write_io_error because we have
1767 * our own ways of dealing with the IO errors
1769 clear_buffer_uptodate(bh);
1775 int write_all_supers(struct btrfs_root *root)
1777 struct list_head *cur;
1778 struct list_head *head = &root->fs_info->fs_devices->devices;
1779 struct btrfs_device *dev;
1780 struct btrfs_super_block *sb;
1781 struct btrfs_dev_item *dev_item;
1782 struct buffer_head *bh;
1786 int total_errors = 0;
1790 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1791 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1793 sb = &root->fs_info->super_for_commit;
1794 dev_item = &sb->dev_item;
1795 list_for_each(cur, head) {
1796 dev = list_entry(cur, struct btrfs_device, dev_list);
1801 if (!dev->in_fs_metadata)
1804 btrfs_set_stack_device_type(dev_item, dev->type);
1805 btrfs_set_stack_device_id(dev_item, dev->devid);
1806 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1807 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1808 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1809 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1810 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1811 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1812 flags = btrfs_super_flags(sb);
1813 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1817 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1818 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1819 btrfs_csum_final(crc, sb->csum);
1821 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1822 BTRFS_SUPER_INFO_SIZE);
1824 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1825 dev->pending_io = bh;
1828 set_buffer_uptodate(bh);
1830 bh->b_end_io = btrfs_end_buffer_write_sync;
1832 if (do_barriers && dev->barriers) {
1833 ret = submit_bh(WRITE_BARRIER, bh);
1834 if (ret == -EOPNOTSUPP) {
1835 printk("btrfs: disabling barriers on dev %s\n",
1837 set_buffer_uptodate(bh);
1841 ret = submit_bh(WRITE, bh);
1844 ret = submit_bh(WRITE, bh);
1849 if (total_errors > max_errors) {
1850 printk("btrfs: %d errors while writing supers\n", total_errors);
1855 list_for_each(cur, head) {
1856 dev = list_entry(cur, struct btrfs_device, dev_list);
1859 if (!dev->in_fs_metadata)
1862 BUG_ON(!dev->pending_io);
1863 bh = dev->pending_io;
1865 if (!buffer_uptodate(dev->pending_io)) {
1866 if (do_barriers && dev->barriers) {
1867 printk("btrfs: disabling barriers on dev %s\n",
1869 set_buffer_uptodate(bh);
1873 ret = submit_bh(WRITE, bh);
1876 if (!buffer_uptodate(bh))
1883 dev->pending_io = NULL;
1886 if (total_errors > max_errors) {
1887 printk("btrfs: %d errors while writing supers\n", total_errors);
1893 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1898 ret = write_all_supers(root);
1902 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1904 radix_tree_delete(&fs_info->fs_roots_radix,
1905 (unsigned long)root->root_key.objectid);
1907 btrfs_sysfs_del_root(root);
1911 free_extent_buffer(root->node);
1912 if (root->commit_root)
1913 free_extent_buffer(root->commit_root);
1920 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1923 struct btrfs_root *gang[8];
1927 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1932 for (i = 0; i < ret; i++)
1933 btrfs_free_fs_root(fs_info, gang[i]);
1938 int close_ctree(struct btrfs_root *root)
1941 struct btrfs_trans_handle *trans;
1942 struct btrfs_fs_info *fs_info = root->fs_info;
1944 fs_info->closing = 1;
1947 kthread_stop(root->fs_info->transaction_kthread);
1948 kthread_stop(root->fs_info->cleaner_kthread);
1950 btrfs_clean_old_snapshots(root);
1951 trans = btrfs_start_transaction(root, 1);
1952 ret = btrfs_commit_transaction(trans, root);
1953 /* run commit again to drop the original snapshot */
1954 trans = btrfs_start_transaction(root, 1);
1955 btrfs_commit_transaction(trans, root);
1956 ret = btrfs_write_and_wait_transaction(NULL, root);
1959 write_ctree_super(NULL, root);
1961 if (fs_info->delalloc_bytes) {
1962 printk("btrfs: at unmount delalloc count %Lu\n",
1963 fs_info->delalloc_bytes);
1965 if (fs_info->total_ref_cache_size) {
1966 printk("btrfs: at umount reference cache size %Lu\n",
1967 fs_info->total_ref_cache_size);
1970 if (fs_info->extent_root->node)
1971 free_extent_buffer(fs_info->extent_root->node);
1973 if (fs_info->tree_root->node)
1974 free_extent_buffer(fs_info->tree_root->node);
1976 if (root->fs_info->chunk_root->node);
1977 free_extent_buffer(root->fs_info->chunk_root->node);
1979 if (root->fs_info->dev_root->node);
1980 free_extent_buffer(root->fs_info->dev_root->node);
1982 btrfs_free_block_groups(root->fs_info);
1983 fs_info->closing = 2;
1984 del_fs_roots(fs_info);
1986 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1988 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1990 btrfs_stop_workers(&fs_info->fixup_workers);
1991 btrfs_stop_workers(&fs_info->workers);
1992 btrfs_stop_workers(&fs_info->endio_workers);
1993 btrfs_stop_workers(&fs_info->endio_write_workers);
1994 btrfs_stop_workers(&fs_info->submit_workers);
1996 iput(fs_info->btree_inode);
1998 while(!list_empty(&fs_info->hashers)) {
1999 struct btrfs_hasher *hasher;
2000 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2002 list_del(&hasher->hashers);
2003 crypto_free_hash(&fs_info->hash_tfm);
2007 btrfs_close_devices(fs_info->fs_devices);
2008 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2010 bdi_destroy(&fs_info->bdi);
2012 kfree(fs_info->extent_root);
2013 kfree(fs_info->tree_root);
2014 kfree(fs_info->chunk_root);
2015 kfree(fs_info->dev_root);
2019 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2022 struct inode *btree_inode = buf->first_page->mapping->host;
2024 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2028 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2033 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2035 struct inode *btree_inode = buf->first_page->mapping->host;
2036 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2040 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2042 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2043 u64 transid = btrfs_header_generation(buf);
2044 struct inode *btree_inode = root->fs_info->btree_inode;
2046 WARN_ON(!btrfs_tree_locked(buf));
2047 if (transid != root->fs_info->generation) {
2048 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2049 (unsigned long long)buf->start,
2050 transid, root->fs_info->generation);
2053 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2056 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2059 * looks as though older kernels can get into trouble with
2060 * this code, they end up stuck in balance_dirty_pages forever
2062 struct extent_io_tree *tree;
2065 unsigned long thresh = 96 * 1024 * 1024;
2066 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2068 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2071 num_dirty = count_range_bits(tree, &start, (u64)-1,
2072 thresh, EXTENT_DIRTY);
2073 if (num_dirty > thresh) {
2074 balance_dirty_pages_ratelimited_nr(
2075 root->fs_info->btree_inode->i_mapping, 1);
2080 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2082 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2084 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2086 buf->flags |= EXTENT_UPTODATE;
2091 int btree_lock_page_hook(struct page *page)
2093 struct inode *inode = page->mapping->host;
2094 struct btrfs_root *root = BTRFS_I(inode)->root;
2095 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2096 struct extent_buffer *eb;
2098 u64 bytenr = page_offset(page);
2100 if (page->private == EXTENT_PAGE_PRIVATE)
2103 len = page->private >> 2;
2104 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2108 btrfs_tree_lock(eb);
2109 spin_lock(&root->fs_info->hash_lock);
2110 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2111 spin_unlock(&root->fs_info->hash_lock);
2112 btrfs_tree_unlock(eb);
2113 free_extent_buffer(eb);
2119 static struct extent_io_ops btree_extent_io_ops = {
2120 .write_cache_pages_lock_hook = btree_lock_page_hook,
2121 .readpage_end_io_hook = btree_readpage_end_io_hook,
2122 .submit_bio_hook = btree_submit_bio_hook,
2123 /* note we're sharing with inode.c for the merge bio hook */
2124 .merge_bio_hook = btrfs_merge_bio_hook,