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),
543 while(atomic_read(&fs_info->async_submit_draining) &&
544 atomic_read(&fs_info->nr_async_submits)) {
545 wait_event(fs_info->async_submit_wait,
546 (atomic_read(&fs_info->nr_async_submits) == 0));
552 static int btree_csum_one_bio(struct bio *bio)
554 struct bio_vec *bvec = bio->bi_io_vec;
556 struct btrfs_root *root;
558 WARN_ON(bio->bi_vcnt <= 0);
559 while(bio_index < bio->bi_vcnt) {
560 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
561 csum_dirty_buffer(root, bvec->bv_page);
568 static int __btree_submit_bio_start(struct inode *inode, int rw,
569 struct bio *bio, int mirror_num,
570 unsigned long bio_flags)
573 * when we're called for a write, we're already in the async
574 * submission context. Just jump into btrfs_map_bio
576 btree_csum_one_bio(bio);
580 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
581 int mirror_num, unsigned long bio_flags)
584 * when we're called for a write, we're already in the async
585 * submission context. Just jump into btrfs_map_bio
587 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
590 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
591 int mirror_num, unsigned long bio_flags)
594 * kthread helpers are used to submit writes so that checksumming
595 * can happen in parallel across all CPUs
597 if (!(rw & (1 << BIO_RW))) {
600 * called for a read, do the setup so that checksum validation
601 * can happen in the async kernel threads
603 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
607 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
610 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
611 inode, rw, bio, mirror_num, 0,
612 __btree_submit_bio_start,
613 __btree_submit_bio_done);
616 static int btree_writepage(struct page *page, struct writeback_control *wbc)
618 struct extent_io_tree *tree;
619 tree = &BTRFS_I(page->mapping->host)->io_tree;
621 if (current->flags & PF_MEMALLOC) {
622 redirty_page_for_writepage(wbc, page);
626 return extent_write_full_page(tree, page, btree_get_extent, wbc);
629 static int btree_writepages(struct address_space *mapping,
630 struct writeback_control *wbc)
632 struct extent_io_tree *tree;
633 tree = &BTRFS_I(mapping->host)->io_tree;
634 if (wbc->sync_mode == WB_SYNC_NONE) {
637 unsigned long thresh = 32 * 1024 * 1024;
639 if (wbc->for_kupdate)
642 num_dirty = count_range_bits(tree, &start, (u64)-1,
643 thresh, EXTENT_DIRTY);
644 if (num_dirty < thresh) {
648 return extent_writepages(tree, mapping, btree_get_extent, wbc);
651 int btree_readpage(struct file *file, struct page *page)
653 struct extent_io_tree *tree;
654 tree = &BTRFS_I(page->mapping->host)->io_tree;
655 return extent_read_full_page(tree, page, btree_get_extent);
658 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
660 struct extent_io_tree *tree;
661 struct extent_map_tree *map;
664 if (PageWriteback(page) || PageDirty(page))
667 tree = &BTRFS_I(page->mapping->host)->io_tree;
668 map = &BTRFS_I(page->mapping->host)->extent_tree;
670 ret = try_release_extent_state(map, tree, page, gfp_flags);
675 ret = try_release_extent_buffer(tree, page);
677 ClearPagePrivate(page);
678 set_page_private(page, 0);
679 page_cache_release(page);
685 static void btree_invalidatepage(struct page *page, unsigned long offset)
687 struct extent_io_tree *tree;
688 tree = &BTRFS_I(page->mapping->host)->io_tree;
689 extent_invalidatepage(tree, page, offset);
690 btree_releasepage(page, GFP_NOFS);
691 if (PagePrivate(page)) {
692 printk("warning page private not zero on page %Lu\n",
694 ClearPagePrivate(page);
695 set_page_private(page, 0);
696 page_cache_release(page);
701 static int btree_writepage(struct page *page, struct writeback_control *wbc)
703 struct buffer_head *bh;
704 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
705 struct buffer_head *head;
706 if (!page_has_buffers(page)) {
707 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
708 (1 << BH_Dirty)|(1 << BH_Uptodate));
710 head = page_buffers(page);
713 if (buffer_dirty(bh))
714 csum_tree_block(root, bh, 0);
715 bh = bh->b_this_page;
716 } while (bh != head);
717 return block_write_full_page(page, btree_get_block, wbc);
721 static struct address_space_operations btree_aops = {
722 .readpage = btree_readpage,
723 .writepage = btree_writepage,
724 .writepages = btree_writepages,
725 .releasepage = btree_releasepage,
726 .invalidatepage = btree_invalidatepage,
727 .sync_page = block_sync_page,
730 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
733 struct extent_buffer *buf = NULL;
734 struct inode *btree_inode = root->fs_info->btree_inode;
737 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
740 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
741 buf, 0, 0, btree_get_extent, 0);
742 free_extent_buffer(buf);
746 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
747 u64 bytenr, u32 blocksize)
749 struct inode *btree_inode = root->fs_info->btree_inode;
750 struct extent_buffer *eb;
751 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
752 bytenr, blocksize, GFP_NOFS);
756 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
757 u64 bytenr, u32 blocksize)
759 struct inode *btree_inode = root->fs_info->btree_inode;
760 struct extent_buffer *eb;
762 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
763 bytenr, blocksize, NULL, GFP_NOFS);
768 int btrfs_write_tree_block(struct extent_buffer *buf)
770 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
771 buf->start + buf->len - 1, WB_SYNC_ALL);
774 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
776 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
777 buf->start, buf->start + buf->len -1);
780 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
781 u32 blocksize, u64 parent_transid)
783 struct extent_buffer *buf = NULL;
784 struct inode *btree_inode = root->fs_info->btree_inode;
785 struct extent_io_tree *io_tree;
788 io_tree = &BTRFS_I(btree_inode)->io_tree;
790 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
794 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
797 buf->flags |= EXTENT_UPTODATE;
805 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
806 struct extent_buffer *buf)
808 struct inode *btree_inode = root->fs_info->btree_inode;
809 if (btrfs_header_generation(buf) ==
810 root->fs_info->running_transaction->transid) {
811 WARN_ON(!btrfs_tree_locked(buf));
812 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
818 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
819 u32 stripesize, struct btrfs_root *root,
820 struct btrfs_fs_info *fs_info,
825 root->commit_root = NULL;
826 root->ref_tree = NULL;
827 root->sectorsize = sectorsize;
828 root->nodesize = nodesize;
829 root->leafsize = leafsize;
830 root->stripesize = stripesize;
832 root->track_dirty = 0;
834 root->fs_info = fs_info;
835 root->objectid = objectid;
836 root->last_trans = 0;
837 root->highest_inode = 0;
838 root->last_inode_alloc = 0;
842 INIT_LIST_HEAD(&root->dirty_list);
843 INIT_LIST_HEAD(&root->orphan_list);
844 INIT_LIST_HEAD(&root->dead_list);
845 spin_lock_init(&root->node_lock);
846 spin_lock_init(&root->list_lock);
847 mutex_init(&root->objectid_mutex);
848 mutex_init(&root->log_mutex);
849 extent_io_tree_init(&root->dirty_log_pages,
850 fs_info->btree_inode->i_mapping, GFP_NOFS);
852 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
853 root->ref_tree = &root->ref_tree_struct;
855 memset(&root->root_key, 0, sizeof(root->root_key));
856 memset(&root->root_item, 0, sizeof(root->root_item));
857 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
858 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
859 root->defrag_trans_start = fs_info->generation;
860 init_completion(&root->kobj_unregister);
861 root->defrag_running = 0;
862 root->defrag_level = 0;
863 root->root_key.objectid = objectid;
867 static int find_and_setup_root(struct btrfs_root *tree_root,
868 struct btrfs_fs_info *fs_info,
870 struct btrfs_root *root)
876 __setup_root(tree_root->nodesize, tree_root->leafsize,
877 tree_root->sectorsize, tree_root->stripesize,
878 root, fs_info, objectid);
879 ret = btrfs_find_last_root(tree_root, objectid,
880 &root->root_item, &root->root_key);
883 generation = btrfs_root_generation(&root->root_item);
884 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
885 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
886 blocksize, generation);
891 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
892 struct btrfs_fs_info *fs_info)
894 struct extent_buffer *eb;
895 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
904 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
905 0, &start, &end, EXTENT_DIRTY);
909 clear_extent_dirty(&log_root_tree->dirty_log_pages,
910 start, end, GFP_NOFS);
912 eb = fs_info->log_root_tree->node;
914 WARN_ON(btrfs_header_level(eb) != 0);
915 WARN_ON(btrfs_header_nritems(eb) != 0);
917 ret = btrfs_free_reserved_extent(fs_info->tree_root,
921 free_extent_buffer(eb);
922 kfree(fs_info->log_root_tree);
923 fs_info->log_root_tree = NULL;
927 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
928 struct btrfs_fs_info *fs_info)
930 struct btrfs_root *root;
931 struct btrfs_root *tree_root = fs_info->tree_root;
933 root = kzalloc(sizeof(*root), GFP_NOFS);
937 __setup_root(tree_root->nodesize, tree_root->leafsize,
938 tree_root->sectorsize, tree_root->stripesize,
939 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
941 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
942 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
943 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
946 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
947 0, BTRFS_TREE_LOG_OBJECTID,
948 trans->transid, 0, 0, 0);
950 btrfs_set_header_nritems(root->node, 0);
951 btrfs_set_header_level(root->node, 0);
952 btrfs_set_header_bytenr(root->node, root->node->start);
953 btrfs_set_header_generation(root->node, trans->transid);
954 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
956 write_extent_buffer(root->node, root->fs_info->fsid,
957 (unsigned long)btrfs_header_fsid(root->node),
959 btrfs_mark_buffer_dirty(root->node);
960 btrfs_tree_unlock(root->node);
961 fs_info->log_root_tree = root;
965 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
966 struct btrfs_key *location)
968 struct btrfs_root *root;
969 struct btrfs_fs_info *fs_info = tree_root->fs_info;
970 struct btrfs_path *path;
971 struct extent_buffer *l;
977 root = kzalloc(sizeof(*root), GFP_NOFS);
979 return ERR_PTR(-ENOMEM);
980 if (location->offset == (u64)-1) {
981 ret = find_and_setup_root(tree_root, fs_info,
982 location->objectid, root);
990 __setup_root(tree_root->nodesize, tree_root->leafsize,
991 tree_root->sectorsize, tree_root->stripesize,
992 root, fs_info, location->objectid);
994 path = btrfs_alloc_path();
996 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1003 read_extent_buffer(l, &root->root_item,
1004 btrfs_item_ptr_offset(l, path->slots[0]),
1005 sizeof(root->root_item));
1006 memcpy(&root->root_key, location, sizeof(*location));
1009 btrfs_release_path(root, path);
1010 btrfs_free_path(path);
1013 return ERR_PTR(ret);
1015 generation = btrfs_root_generation(&root->root_item);
1016 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1017 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1018 blocksize, generation);
1019 BUG_ON(!root->node);
1021 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1023 ret = btrfs_find_highest_inode(root, &highest_inode);
1025 root->highest_inode = highest_inode;
1026 root->last_inode_alloc = highest_inode;
1032 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1035 struct btrfs_root *root;
1037 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1038 return fs_info->tree_root;
1039 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1040 return fs_info->extent_root;
1042 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1043 (unsigned long)root_objectid);
1047 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1048 struct btrfs_key *location)
1050 struct btrfs_root *root;
1053 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1054 return fs_info->tree_root;
1055 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1056 return fs_info->extent_root;
1057 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1058 return fs_info->chunk_root;
1059 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1060 return fs_info->dev_root;
1062 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1063 (unsigned long)location->objectid);
1067 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1070 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1071 (unsigned long)root->root_key.objectid,
1074 free_extent_buffer(root->node);
1076 return ERR_PTR(ret);
1078 ret = btrfs_find_dead_roots(fs_info->tree_root,
1079 root->root_key.objectid, root);
1085 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1086 struct btrfs_key *location,
1087 const char *name, int namelen)
1089 struct btrfs_root *root;
1092 root = btrfs_read_fs_root_no_name(fs_info, location);
1099 ret = btrfs_set_root_name(root, name, namelen);
1101 free_extent_buffer(root->node);
1103 return ERR_PTR(ret);
1106 ret = btrfs_sysfs_add_root(root);
1108 free_extent_buffer(root->node);
1111 return ERR_PTR(ret);
1117 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1118 struct btrfs_hasher *hasher;
1120 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1123 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1124 if (!hasher->hash_tfm) {
1128 spin_lock(&info->hash_lock);
1129 list_add(&hasher->list, &info->hashers);
1130 spin_unlock(&info->hash_lock);
1135 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1137 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1139 struct list_head *cur;
1140 struct btrfs_device *device;
1141 struct backing_dev_info *bdi;
1143 if ((bdi_bits & (1 << BDI_write_congested)) &&
1144 btrfs_congested_async(info, 0))
1147 list_for_each(cur, &info->fs_devices->devices) {
1148 device = list_entry(cur, struct btrfs_device, dev_list);
1151 bdi = blk_get_backing_dev_info(device->bdev);
1152 if (bdi && bdi_congested(bdi, bdi_bits)) {
1161 * this unplugs every device on the box, and it is only used when page
1164 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1166 struct list_head *cur;
1167 struct btrfs_device *device;
1168 struct btrfs_fs_info *info;
1170 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1171 list_for_each(cur, &info->fs_devices->devices) {
1172 device = list_entry(cur, struct btrfs_device, dev_list);
1173 bdi = blk_get_backing_dev_info(device->bdev);
1174 if (bdi->unplug_io_fn) {
1175 bdi->unplug_io_fn(bdi, page);
1180 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1182 struct inode *inode;
1183 struct extent_map_tree *em_tree;
1184 struct extent_map *em;
1185 struct address_space *mapping;
1188 /* the generic O_DIRECT read code does this */
1190 __unplug_io_fn(bdi, page);
1195 * page->mapping may change at any time. Get a consistent copy
1196 * and use that for everything below
1199 mapping = page->mapping;
1203 inode = mapping->host;
1204 offset = page_offset(page);
1206 em_tree = &BTRFS_I(inode)->extent_tree;
1207 spin_lock(&em_tree->lock);
1208 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1209 spin_unlock(&em_tree->lock);
1211 __unplug_io_fn(bdi, page);
1215 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1216 free_extent_map(em);
1217 __unplug_io_fn(bdi, page);
1220 offset = offset - em->start;
1221 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1222 em->block_start + offset, page);
1223 free_extent_map(em);
1226 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1229 bdi->ra_pages = default_backing_dev_info.ra_pages;
1231 bdi->capabilities = default_backing_dev_info.capabilities;
1232 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1233 bdi->unplug_io_data = info;
1234 bdi->congested_fn = btrfs_congested_fn;
1235 bdi->congested_data = info;
1239 static int bio_ready_for_csum(struct bio *bio)
1245 struct extent_io_tree *io_tree = NULL;
1246 struct btrfs_fs_info *info = NULL;
1247 struct bio_vec *bvec;
1251 bio_for_each_segment(bvec, bio, i) {
1252 page = bvec->bv_page;
1253 if (page->private == EXTENT_PAGE_PRIVATE) {
1254 length += bvec->bv_len;
1257 if (!page->private) {
1258 length += bvec->bv_len;
1261 length = bvec->bv_len;
1262 buf_len = page->private >> 2;
1263 start = page_offset(page) + bvec->bv_offset;
1264 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1265 info = BTRFS_I(page->mapping->host)->root->fs_info;
1267 /* are we fully contained in this bio? */
1268 if (buf_len <= length)
1271 ret = extent_range_uptodate(io_tree, start + length,
1272 start + buf_len - 1);
1279 * called by the kthread helper functions to finally call the bio end_io
1280 * functions. This is where read checksum verification actually happens
1282 static void end_workqueue_fn(struct btrfs_work *work)
1285 struct end_io_wq *end_io_wq;
1286 struct btrfs_fs_info *fs_info;
1289 end_io_wq = container_of(work, struct end_io_wq, work);
1290 bio = end_io_wq->bio;
1291 fs_info = end_io_wq->info;
1293 /* metadata bios are special because the whole tree block must
1294 * be checksummed at once. This makes sure the entire block is in
1295 * ram and up to date before trying to verify things. For
1296 * blocksize <= pagesize, it is basically a noop
1298 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1299 btrfs_queue_worker(&fs_info->endio_workers,
1303 error = end_io_wq->error;
1304 bio->bi_private = end_io_wq->private;
1305 bio->bi_end_io = end_io_wq->end_io;
1307 bio_endio(bio, error);
1310 static int cleaner_kthread(void *arg)
1312 struct btrfs_root *root = arg;
1316 if (root->fs_info->closing)
1319 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1320 mutex_lock(&root->fs_info->cleaner_mutex);
1321 btrfs_clean_old_snapshots(root);
1322 mutex_unlock(&root->fs_info->cleaner_mutex);
1324 if (freezing(current)) {
1328 if (root->fs_info->closing)
1330 set_current_state(TASK_INTERRUPTIBLE);
1332 __set_current_state(TASK_RUNNING);
1334 } while (!kthread_should_stop());
1338 static int transaction_kthread(void *arg)
1340 struct btrfs_root *root = arg;
1341 struct btrfs_trans_handle *trans;
1342 struct btrfs_transaction *cur;
1344 unsigned long delay;
1349 if (root->fs_info->closing)
1353 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1354 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1356 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1357 printk("btrfs: total reference cache size %Lu\n",
1358 root->fs_info->total_ref_cache_size);
1361 mutex_lock(&root->fs_info->trans_mutex);
1362 cur = root->fs_info->running_transaction;
1364 mutex_unlock(&root->fs_info->trans_mutex);
1368 now = get_seconds();
1369 if (now < cur->start_time || now - cur->start_time < 30) {
1370 mutex_unlock(&root->fs_info->trans_mutex);
1374 mutex_unlock(&root->fs_info->trans_mutex);
1375 trans = btrfs_start_transaction(root, 1);
1376 ret = btrfs_commit_transaction(trans, root);
1378 wake_up_process(root->fs_info->cleaner_kthread);
1379 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1381 if (freezing(current)) {
1384 if (root->fs_info->closing)
1386 set_current_state(TASK_INTERRUPTIBLE);
1387 schedule_timeout(delay);
1388 __set_current_state(TASK_RUNNING);
1390 } while (!kthread_should_stop());
1394 struct btrfs_root *open_ctree(struct super_block *sb,
1395 struct btrfs_fs_devices *fs_devices,
1404 struct buffer_head *bh;
1405 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1407 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1409 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1411 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1413 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1415 struct btrfs_root *log_tree_root;
1420 struct btrfs_super_block *disk_super;
1422 if (!extent_root || !tree_root || !fs_info ||
1423 !chunk_root || !dev_root) {
1427 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1428 INIT_LIST_HEAD(&fs_info->trans_list);
1429 INIT_LIST_HEAD(&fs_info->dead_roots);
1430 INIT_LIST_HEAD(&fs_info->hashers);
1431 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1432 spin_lock_init(&fs_info->hash_lock);
1433 spin_lock_init(&fs_info->delalloc_lock);
1434 spin_lock_init(&fs_info->new_trans_lock);
1435 spin_lock_init(&fs_info->ref_cache_lock);
1437 init_completion(&fs_info->kobj_unregister);
1438 fs_info->tree_root = tree_root;
1439 fs_info->extent_root = extent_root;
1440 fs_info->chunk_root = chunk_root;
1441 fs_info->dev_root = dev_root;
1442 fs_info->fs_devices = fs_devices;
1443 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1444 INIT_LIST_HEAD(&fs_info->space_info);
1445 btrfs_mapping_init(&fs_info->mapping_tree);
1446 atomic_set(&fs_info->nr_async_submits, 0);
1447 atomic_set(&fs_info->async_delalloc_pages, 0);
1448 atomic_set(&fs_info->async_submit_draining, 0);
1449 atomic_set(&fs_info->nr_async_bios, 0);
1450 atomic_set(&fs_info->throttles, 0);
1451 atomic_set(&fs_info->throttle_gen, 0);
1453 fs_info->max_extent = (u64)-1;
1454 fs_info->max_inline = 8192 * 1024;
1455 setup_bdi(fs_info, &fs_info->bdi);
1456 fs_info->btree_inode = new_inode(sb);
1457 fs_info->btree_inode->i_ino = 1;
1458 fs_info->btree_inode->i_nlink = 1;
1460 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1462 INIT_LIST_HEAD(&fs_info->ordered_extents);
1463 spin_lock_init(&fs_info->ordered_extent_lock);
1465 sb->s_blocksize = 4096;
1466 sb->s_blocksize_bits = blksize_bits(4096);
1469 * we set the i_size on the btree inode to the max possible int.
1470 * the real end of the address space is determined by all of
1471 * the devices in the system
1473 fs_info->btree_inode->i_size = OFFSET_MAX;
1474 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1475 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1477 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1478 fs_info->btree_inode->i_mapping,
1480 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1483 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1485 spin_lock_init(&fs_info->block_group_cache_lock);
1486 fs_info->block_group_cache_tree.rb_node = NULL;
1488 extent_io_tree_init(&fs_info->pinned_extents,
1489 fs_info->btree_inode->i_mapping, GFP_NOFS);
1490 extent_io_tree_init(&fs_info->pending_del,
1491 fs_info->btree_inode->i_mapping, GFP_NOFS);
1492 extent_io_tree_init(&fs_info->extent_ins,
1493 fs_info->btree_inode->i_mapping, GFP_NOFS);
1494 fs_info->do_barriers = 1;
1496 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1497 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1498 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1500 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1501 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1502 sizeof(struct btrfs_key));
1503 insert_inode_hash(fs_info->btree_inode);
1505 mutex_init(&fs_info->trans_mutex);
1506 mutex_init(&fs_info->tree_log_mutex);
1507 mutex_init(&fs_info->drop_mutex);
1508 mutex_init(&fs_info->extent_ins_mutex);
1509 mutex_init(&fs_info->pinned_mutex);
1510 mutex_init(&fs_info->chunk_mutex);
1511 mutex_init(&fs_info->transaction_kthread_mutex);
1512 mutex_init(&fs_info->cleaner_mutex);
1513 mutex_init(&fs_info->volume_mutex);
1514 mutex_init(&fs_info->tree_reloc_mutex);
1515 init_waitqueue_head(&fs_info->transaction_throttle);
1516 init_waitqueue_head(&fs_info->transaction_wait);
1517 init_waitqueue_head(&fs_info->async_submit_wait);
1518 init_waitqueue_head(&fs_info->tree_log_wait);
1519 atomic_set(&fs_info->tree_log_commit, 0);
1520 atomic_set(&fs_info->tree_log_writers, 0);
1521 fs_info->tree_log_transid = 0;
1524 ret = add_hasher(fs_info, "crc32c");
1526 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1531 __setup_root(4096, 4096, 4096, 4096, tree_root,
1532 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1535 bh = __bread(fs_devices->latest_bdev,
1536 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1540 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1543 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1545 disk_super = &fs_info->super_copy;
1546 if (!btrfs_super_root(disk_super))
1547 goto fail_sb_buffer;
1549 err = btrfs_parse_options(tree_root, options);
1551 goto fail_sb_buffer;
1554 * we need to start all the end_io workers up front because the
1555 * queue work function gets called at interrupt time, and so it
1556 * cannot dynamically grow.
1558 btrfs_init_workers(&fs_info->workers, "worker",
1559 fs_info->thread_pool_size);
1561 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1562 fs_info->thread_pool_size);
1564 btrfs_init_workers(&fs_info->submit_workers, "submit",
1565 min_t(u64, fs_devices->num_devices,
1566 fs_info->thread_pool_size));
1568 /* a higher idle thresh on the submit workers makes it much more
1569 * likely that bios will be send down in a sane order to the
1572 fs_info->submit_workers.idle_thresh = 64;
1574 fs_info->workers.idle_thresh = 16;
1575 fs_info->workers.ordered = 1;
1577 fs_info->delalloc_workers.idle_thresh = 2;
1578 fs_info->delalloc_workers.ordered = 1;
1580 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1581 btrfs_init_workers(&fs_info->endio_workers, "endio",
1582 fs_info->thread_pool_size);
1583 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1584 fs_info->thread_pool_size);
1587 * endios are largely parallel and should have a very
1590 fs_info->endio_workers.idle_thresh = 4;
1591 fs_info->endio_write_workers.idle_thresh = 64;
1593 btrfs_start_workers(&fs_info->workers, 1);
1594 btrfs_start_workers(&fs_info->submit_workers, 1);
1595 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1596 btrfs_start_workers(&fs_info->fixup_workers, 1);
1597 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1598 btrfs_start_workers(&fs_info->endio_write_workers,
1599 fs_info->thread_pool_size);
1602 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1603 printk("Btrfs: wanted %llu devices, but found %llu\n",
1604 (unsigned long long)btrfs_super_num_devices(disk_super),
1605 (unsigned long long)fs_devices->open_devices);
1606 if (btrfs_test_opt(tree_root, DEGRADED))
1607 printk("continuing in degraded mode\n");
1609 goto fail_sb_buffer;
1613 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1614 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1615 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1617 nodesize = btrfs_super_nodesize(disk_super);
1618 leafsize = btrfs_super_leafsize(disk_super);
1619 sectorsize = btrfs_super_sectorsize(disk_super);
1620 stripesize = btrfs_super_stripesize(disk_super);
1621 tree_root->nodesize = nodesize;
1622 tree_root->leafsize = leafsize;
1623 tree_root->sectorsize = sectorsize;
1624 tree_root->stripesize = stripesize;
1626 sb->s_blocksize = sectorsize;
1627 sb->s_blocksize_bits = blksize_bits(sectorsize);
1629 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1630 sizeof(disk_super->magic))) {
1631 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1632 goto fail_sb_buffer;
1635 mutex_lock(&fs_info->chunk_mutex);
1636 ret = btrfs_read_sys_array(tree_root);
1637 mutex_unlock(&fs_info->chunk_mutex);
1639 printk("btrfs: failed to read the system array on %s\n",
1641 goto fail_sys_array;
1644 blocksize = btrfs_level_size(tree_root,
1645 btrfs_super_chunk_root_level(disk_super));
1646 generation = btrfs_super_chunk_root_generation(disk_super);
1648 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1649 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1651 chunk_root->node = read_tree_block(chunk_root,
1652 btrfs_super_chunk_root(disk_super),
1653 blocksize, generation);
1654 BUG_ON(!chunk_root->node);
1656 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1657 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1660 mutex_lock(&fs_info->chunk_mutex);
1661 ret = btrfs_read_chunk_tree(chunk_root);
1662 mutex_unlock(&fs_info->chunk_mutex);
1665 btrfs_close_extra_devices(fs_devices);
1667 blocksize = btrfs_level_size(tree_root,
1668 btrfs_super_root_level(disk_super));
1669 generation = btrfs_super_generation(disk_super);
1671 tree_root->node = read_tree_block(tree_root,
1672 btrfs_super_root(disk_super),
1673 blocksize, generation);
1674 if (!tree_root->node)
1675 goto fail_sb_buffer;
1678 ret = find_and_setup_root(tree_root, fs_info,
1679 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1681 goto fail_tree_root;
1682 extent_root->track_dirty = 1;
1684 ret = find_and_setup_root(tree_root, fs_info,
1685 BTRFS_DEV_TREE_OBJECTID, dev_root);
1686 dev_root->track_dirty = 1;
1689 goto fail_extent_root;
1691 btrfs_read_block_groups(extent_root);
1693 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1694 fs_info->data_alloc_profile = (u64)-1;
1695 fs_info->metadata_alloc_profile = (u64)-1;
1696 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1697 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1699 if (!fs_info->cleaner_kthread)
1700 goto fail_extent_root;
1702 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1704 "btrfs-transaction");
1705 if (!fs_info->transaction_kthread)
1708 if (btrfs_super_log_root(disk_super) != 0) {
1710 u64 bytenr = btrfs_super_log_root(disk_super);
1713 btrfs_level_size(tree_root,
1714 btrfs_super_log_root_level(disk_super));
1716 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1719 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1720 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1722 log_tree_root->node = read_tree_block(tree_root, bytenr,
1725 ret = btrfs_recover_log_trees(log_tree_root);
1728 fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1730 ret = btrfs_cleanup_reloc_trees(tree_root);
1736 kthread_stop(fs_info->cleaner_kthread);
1738 free_extent_buffer(extent_root->node);
1740 free_extent_buffer(tree_root->node);
1743 btrfs_stop_workers(&fs_info->fixup_workers);
1744 btrfs_stop_workers(&fs_info->delalloc_workers);
1745 btrfs_stop_workers(&fs_info->workers);
1746 btrfs_stop_workers(&fs_info->endio_workers);
1747 btrfs_stop_workers(&fs_info->endio_write_workers);
1748 btrfs_stop_workers(&fs_info->submit_workers);
1750 iput(fs_info->btree_inode);
1752 btrfs_close_devices(fs_info->fs_devices);
1753 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1757 bdi_destroy(&fs_info->bdi);
1761 return ERR_PTR(err);
1764 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1766 char b[BDEVNAME_SIZE];
1769 set_buffer_uptodate(bh);
1771 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1772 printk(KERN_WARNING "lost page write due to "
1773 "I/O error on %s\n",
1774 bdevname(bh->b_bdev, b));
1776 /* note, we dont' set_buffer_write_io_error because we have
1777 * our own ways of dealing with the IO errors
1779 clear_buffer_uptodate(bh);
1785 int write_all_supers(struct btrfs_root *root)
1787 struct list_head *cur;
1788 struct list_head *head = &root->fs_info->fs_devices->devices;
1789 struct btrfs_device *dev;
1790 struct btrfs_super_block *sb;
1791 struct btrfs_dev_item *dev_item;
1792 struct buffer_head *bh;
1796 int total_errors = 0;
1800 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1801 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1803 sb = &root->fs_info->super_for_commit;
1804 dev_item = &sb->dev_item;
1805 list_for_each(cur, head) {
1806 dev = list_entry(cur, struct btrfs_device, dev_list);
1811 if (!dev->in_fs_metadata)
1814 btrfs_set_stack_device_type(dev_item, dev->type);
1815 btrfs_set_stack_device_id(dev_item, dev->devid);
1816 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1817 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1818 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1819 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1820 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1821 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1822 flags = btrfs_super_flags(sb);
1823 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1827 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1828 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1829 btrfs_csum_final(crc, sb->csum);
1831 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1832 BTRFS_SUPER_INFO_SIZE);
1834 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1835 dev->pending_io = bh;
1838 set_buffer_uptodate(bh);
1840 bh->b_end_io = btrfs_end_buffer_write_sync;
1842 if (do_barriers && dev->barriers) {
1843 ret = submit_bh(WRITE_BARRIER, bh);
1844 if (ret == -EOPNOTSUPP) {
1845 printk("btrfs: disabling barriers on dev %s\n",
1847 set_buffer_uptodate(bh);
1851 ret = submit_bh(WRITE, bh);
1854 ret = submit_bh(WRITE, bh);
1859 if (total_errors > max_errors) {
1860 printk("btrfs: %d errors while writing supers\n", total_errors);
1865 list_for_each(cur, head) {
1866 dev = list_entry(cur, struct btrfs_device, dev_list);
1869 if (!dev->in_fs_metadata)
1872 BUG_ON(!dev->pending_io);
1873 bh = dev->pending_io;
1875 if (!buffer_uptodate(dev->pending_io)) {
1876 if (do_barriers && dev->barriers) {
1877 printk("btrfs: disabling barriers on dev %s\n",
1879 set_buffer_uptodate(bh);
1883 ret = submit_bh(WRITE, bh);
1886 if (!buffer_uptodate(bh))
1893 dev->pending_io = NULL;
1896 if (total_errors > max_errors) {
1897 printk("btrfs: %d errors while writing supers\n", total_errors);
1903 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1908 ret = write_all_supers(root);
1912 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1914 radix_tree_delete(&fs_info->fs_roots_radix,
1915 (unsigned long)root->root_key.objectid);
1917 btrfs_sysfs_del_root(root);
1921 free_extent_buffer(root->node);
1922 if (root->commit_root)
1923 free_extent_buffer(root->commit_root);
1930 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1933 struct btrfs_root *gang[8];
1937 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1942 for (i = 0; i < ret; i++)
1943 btrfs_free_fs_root(fs_info, gang[i]);
1948 int close_ctree(struct btrfs_root *root)
1951 struct btrfs_trans_handle *trans;
1952 struct btrfs_fs_info *fs_info = root->fs_info;
1954 fs_info->closing = 1;
1957 kthread_stop(root->fs_info->transaction_kthread);
1958 kthread_stop(root->fs_info->cleaner_kthread);
1960 btrfs_clean_old_snapshots(root);
1961 trans = btrfs_start_transaction(root, 1);
1962 ret = btrfs_commit_transaction(trans, root);
1963 /* run commit again to drop the original snapshot */
1964 trans = btrfs_start_transaction(root, 1);
1965 btrfs_commit_transaction(trans, root);
1966 ret = btrfs_write_and_wait_transaction(NULL, root);
1969 write_ctree_super(NULL, root);
1971 if (fs_info->delalloc_bytes) {
1972 printk("btrfs: at unmount delalloc count %Lu\n",
1973 fs_info->delalloc_bytes);
1975 if (fs_info->total_ref_cache_size) {
1976 printk("btrfs: at umount reference cache size %Lu\n",
1977 fs_info->total_ref_cache_size);
1980 if (fs_info->extent_root->node)
1981 free_extent_buffer(fs_info->extent_root->node);
1983 if (fs_info->tree_root->node)
1984 free_extent_buffer(fs_info->tree_root->node);
1986 if (root->fs_info->chunk_root->node);
1987 free_extent_buffer(root->fs_info->chunk_root->node);
1989 if (root->fs_info->dev_root->node);
1990 free_extent_buffer(root->fs_info->dev_root->node);
1992 btrfs_free_block_groups(root->fs_info);
1993 fs_info->closing = 2;
1994 del_fs_roots(fs_info);
1996 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1998 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
2000 btrfs_stop_workers(&fs_info->fixup_workers);
2001 btrfs_stop_workers(&fs_info->delalloc_workers);
2002 btrfs_stop_workers(&fs_info->workers);
2003 btrfs_stop_workers(&fs_info->endio_workers);
2004 btrfs_stop_workers(&fs_info->endio_write_workers);
2005 btrfs_stop_workers(&fs_info->submit_workers);
2007 iput(fs_info->btree_inode);
2009 while(!list_empty(&fs_info->hashers)) {
2010 struct btrfs_hasher *hasher;
2011 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2013 list_del(&hasher->hashers);
2014 crypto_free_hash(&fs_info->hash_tfm);
2018 btrfs_close_devices(fs_info->fs_devices);
2019 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2021 bdi_destroy(&fs_info->bdi);
2023 kfree(fs_info->extent_root);
2024 kfree(fs_info->tree_root);
2025 kfree(fs_info->chunk_root);
2026 kfree(fs_info->dev_root);
2030 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2033 struct inode *btree_inode = buf->first_page->mapping->host;
2035 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2039 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2044 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2046 struct inode *btree_inode = buf->first_page->mapping->host;
2047 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2051 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2053 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2054 u64 transid = btrfs_header_generation(buf);
2055 struct inode *btree_inode = root->fs_info->btree_inode;
2057 WARN_ON(!btrfs_tree_locked(buf));
2058 if (transid != root->fs_info->generation) {
2059 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2060 (unsigned long long)buf->start,
2061 transid, root->fs_info->generation);
2064 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2067 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2070 * looks as though older kernels can get into trouble with
2071 * this code, they end up stuck in balance_dirty_pages forever
2073 struct extent_io_tree *tree;
2076 unsigned long thresh = 32 * 1024 * 1024;
2077 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2079 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2082 num_dirty = count_range_bits(tree, &start, (u64)-1,
2083 thresh, EXTENT_DIRTY);
2084 if (num_dirty > thresh) {
2085 balance_dirty_pages_ratelimited_nr(
2086 root->fs_info->btree_inode->i_mapping, 1);
2091 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2093 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2095 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2097 buf->flags |= EXTENT_UPTODATE;
2102 int btree_lock_page_hook(struct page *page)
2104 struct inode *inode = page->mapping->host;
2105 struct btrfs_root *root = BTRFS_I(inode)->root;
2106 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2107 struct extent_buffer *eb;
2109 u64 bytenr = page_offset(page);
2111 if (page->private == EXTENT_PAGE_PRIVATE)
2114 len = page->private >> 2;
2115 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2119 btrfs_tree_lock(eb);
2120 spin_lock(&root->fs_info->hash_lock);
2121 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2122 spin_unlock(&root->fs_info->hash_lock);
2123 btrfs_tree_unlock(eb);
2124 free_extent_buffer(eb);
2130 static struct extent_io_ops btree_extent_io_ops = {
2131 .write_cache_pages_lock_hook = btree_lock_page_hook,
2132 .readpage_end_io_hook = btree_readpage_end_io_hook,
2133 .submit_bio_hook = btree_submit_bio_hook,
2134 /* note we're sharing with inode.c for the merge bio hook */
2135 .merge_bio_hook = btrfs_merge_bio_hook,