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 static int check_tree_block_fsid(struct btrfs_root *root,
349 struct extent_buffer *eb)
351 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
352 u8 fsid[BTRFS_UUID_SIZE];
355 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
358 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
362 fs_devices = fs_devices->seed;
367 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
368 struct extent_state *state)
370 struct extent_io_tree *tree;
374 struct extent_buffer *eb;
375 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
378 tree = &BTRFS_I(page->mapping->host)->io_tree;
379 if (page->private == EXTENT_PAGE_PRIVATE)
383 len = page->private >> 2;
387 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
389 found_start = btrfs_header_bytenr(eb);
390 if (found_start != start) {
391 printk("bad tree block start %llu %llu\n",
392 (unsigned long long)found_start,
393 (unsigned long long)eb->start);
397 if (eb->first_page != page) {
398 printk("bad first page %lu %lu\n", eb->first_page->index,
404 if (check_tree_block_fsid(root, eb)) {
405 printk("bad fsid on block %Lu\n", eb->start);
409 found_level = btrfs_header_level(eb);
411 ret = csum_tree_block(root, eb, 1);
415 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
416 end = eb->start + end - 1;
418 free_extent_buffer(eb);
423 static void end_workqueue_bio(struct bio *bio, int err)
425 struct end_io_wq *end_io_wq = bio->bi_private;
426 struct btrfs_fs_info *fs_info;
428 fs_info = end_io_wq->info;
429 end_io_wq->error = err;
430 end_io_wq->work.func = end_workqueue_fn;
431 end_io_wq->work.flags = 0;
432 if (bio->bi_rw & (1 << BIO_RW))
433 btrfs_queue_worker(&fs_info->endio_write_workers,
436 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
439 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
442 struct end_io_wq *end_io_wq;
443 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
447 end_io_wq->private = bio->bi_private;
448 end_io_wq->end_io = bio->bi_end_io;
449 end_io_wq->info = info;
450 end_io_wq->error = 0;
451 end_io_wq->bio = bio;
452 end_io_wq->metadata = metadata;
454 bio->bi_private = end_io_wq;
455 bio->bi_end_io = end_workqueue_bio;
459 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
461 unsigned long limit = min_t(unsigned long,
462 info->workers.max_workers,
463 info->fs_devices->open_devices);
467 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
469 return atomic_read(&info->nr_async_bios) >
470 btrfs_async_submit_limit(info);
473 static void run_one_async_start(struct btrfs_work *work)
475 struct btrfs_fs_info *fs_info;
476 struct async_submit_bio *async;
478 async = container_of(work, struct async_submit_bio, work);
479 fs_info = BTRFS_I(async->inode)->root->fs_info;
480 async->submit_bio_start(async->inode, async->rw, async->bio,
481 async->mirror_num, async->bio_flags);
484 static void run_one_async_done(struct btrfs_work *work)
486 struct btrfs_fs_info *fs_info;
487 struct async_submit_bio *async;
490 async = container_of(work, struct async_submit_bio, work);
491 fs_info = BTRFS_I(async->inode)->root->fs_info;
493 limit = btrfs_async_submit_limit(fs_info);
494 limit = limit * 2 / 3;
496 atomic_dec(&fs_info->nr_async_submits);
498 if (atomic_read(&fs_info->nr_async_submits) < limit &&
499 waitqueue_active(&fs_info->async_submit_wait))
500 wake_up(&fs_info->async_submit_wait);
502 async->submit_bio_done(async->inode, async->rw, async->bio,
503 async->mirror_num, async->bio_flags);
506 static void run_one_async_free(struct btrfs_work *work)
508 struct async_submit_bio *async;
510 async = container_of(work, struct async_submit_bio, work);
514 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
515 int rw, struct bio *bio, int mirror_num,
516 unsigned long bio_flags,
517 extent_submit_bio_hook_t *submit_bio_start,
518 extent_submit_bio_hook_t *submit_bio_done)
520 struct async_submit_bio *async;
521 int limit = btrfs_async_submit_limit(fs_info);
523 async = kmalloc(sizeof(*async), GFP_NOFS);
527 async->inode = inode;
530 async->mirror_num = mirror_num;
531 async->submit_bio_start = submit_bio_start;
532 async->submit_bio_done = submit_bio_done;
534 async->work.func = run_one_async_start;
535 async->work.ordered_func = run_one_async_done;
536 async->work.ordered_free = run_one_async_free;
538 async->work.flags = 0;
539 async->bio_flags = bio_flags;
541 while(atomic_read(&fs_info->async_submit_draining) &&
542 atomic_read(&fs_info->nr_async_submits)) {
543 wait_event(fs_info->async_submit_wait,
544 (atomic_read(&fs_info->nr_async_submits) == 0));
547 atomic_inc(&fs_info->nr_async_submits);
548 btrfs_queue_worker(&fs_info->workers, &async->work);
550 if (atomic_read(&fs_info->nr_async_submits) > limit) {
551 wait_event_timeout(fs_info->async_submit_wait,
552 (atomic_read(&fs_info->nr_async_submits) < limit),
555 wait_event_timeout(fs_info->async_submit_wait,
556 (atomic_read(&fs_info->nr_async_bios) < limit),
560 while(atomic_read(&fs_info->async_submit_draining) &&
561 atomic_read(&fs_info->nr_async_submits)) {
562 wait_event(fs_info->async_submit_wait,
563 (atomic_read(&fs_info->nr_async_submits) == 0));
569 static int btree_csum_one_bio(struct bio *bio)
571 struct bio_vec *bvec = bio->bi_io_vec;
573 struct btrfs_root *root;
575 WARN_ON(bio->bi_vcnt <= 0);
576 while(bio_index < bio->bi_vcnt) {
577 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
578 csum_dirty_buffer(root, bvec->bv_page);
585 static int __btree_submit_bio_start(struct inode *inode, int rw,
586 struct bio *bio, int mirror_num,
587 unsigned long bio_flags)
590 * when we're called for a write, we're already in the async
591 * submission context. Just jump into btrfs_map_bio
593 btree_csum_one_bio(bio);
597 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
598 int mirror_num, unsigned long bio_flags)
601 * when we're called for a write, we're already in the async
602 * submission context. Just jump into btrfs_map_bio
604 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
607 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
608 int mirror_num, unsigned long bio_flags)
611 * kthread helpers are used to submit writes so that checksumming
612 * can happen in parallel across all CPUs
614 if (!(rw & (1 << BIO_RW))) {
617 * called for a read, do the setup so that checksum validation
618 * can happen in the async kernel threads
620 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
624 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
627 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
628 inode, rw, bio, mirror_num, 0,
629 __btree_submit_bio_start,
630 __btree_submit_bio_done);
633 static int btree_writepage(struct page *page, struct writeback_control *wbc)
635 struct extent_io_tree *tree;
636 tree = &BTRFS_I(page->mapping->host)->io_tree;
638 if (current->flags & PF_MEMALLOC) {
639 redirty_page_for_writepage(wbc, page);
643 return extent_write_full_page(tree, page, btree_get_extent, wbc);
646 static int btree_writepages(struct address_space *mapping,
647 struct writeback_control *wbc)
649 struct extent_io_tree *tree;
650 tree = &BTRFS_I(mapping->host)->io_tree;
651 if (wbc->sync_mode == WB_SYNC_NONE) {
654 unsigned long thresh = 32 * 1024 * 1024;
656 if (wbc->for_kupdate)
659 num_dirty = count_range_bits(tree, &start, (u64)-1,
660 thresh, EXTENT_DIRTY);
661 if (num_dirty < thresh) {
665 return extent_writepages(tree, mapping, btree_get_extent, wbc);
668 int btree_readpage(struct file *file, struct page *page)
670 struct extent_io_tree *tree;
671 tree = &BTRFS_I(page->mapping->host)->io_tree;
672 return extent_read_full_page(tree, page, btree_get_extent);
675 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
677 struct extent_io_tree *tree;
678 struct extent_map_tree *map;
681 if (PageWriteback(page) || PageDirty(page))
684 tree = &BTRFS_I(page->mapping->host)->io_tree;
685 map = &BTRFS_I(page->mapping->host)->extent_tree;
687 ret = try_release_extent_state(map, tree, page, gfp_flags);
692 ret = try_release_extent_buffer(tree, page);
694 ClearPagePrivate(page);
695 set_page_private(page, 0);
696 page_cache_release(page);
702 static void btree_invalidatepage(struct page *page, unsigned long offset)
704 struct extent_io_tree *tree;
705 tree = &BTRFS_I(page->mapping->host)->io_tree;
706 extent_invalidatepage(tree, page, offset);
707 btree_releasepage(page, GFP_NOFS);
708 if (PagePrivate(page)) {
709 printk("warning page private not zero on page %Lu\n",
711 ClearPagePrivate(page);
712 set_page_private(page, 0);
713 page_cache_release(page);
718 static int btree_writepage(struct page *page, struct writeback_control *wbc)
720 struct buffer_head *bh;
721 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
722 struct buffer_head *head;
723 if (!page_has_buffers(page)) {
724 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
725 (1 << BH_Dirty)|(1 << BH_Uptodate));
727 head = page_buffers(page);
730 if (buffer_dirty(bh))
731 csum_tree_block(root, bh, 0);
732 bh = bh->b_this_page;
733 } while (bh != head);
734 return block_write_full_page(page, btree_get_block, wbc);
738 static struct address_space_operations btree_aops = {
739 .readpage = btree_readpage,
740 .writepage = btree_writepage,
741 .writepages = btree_writepages,
742 .releasepage = btree_releasepage,
743 .invalidatepage = btree_invalidatepage,
744 .sync_page = block_sync_page,
747 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
750 struct extent_buffer *buf = NULL;
751 struct inode *btree_inode = root->fs_info->btree_inode;
754 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
757 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
758 buf, 0, 0, btree_get_extent, 0);
759 free_extent_buffer(buf);
763 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
764 u64 bytenr, u32 blocksize)
766 struct inode *btree_inode = root->fs_info->btree_inode;
767 struct extent_buffer *eb;
768 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
769 bytenr, blocksize, GFP_NOFS);
773 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
774 u64 bytenr, u32 blocksize)
776 struct inode *btree_inode = root->fs_info->btree_inode;
777 struct extent_buffer *eb;
779 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
780 bytenr, blocksize, NULL, GFP_NOFS);
785 int btrfs_write_tree_block(struct extent_buffer *buf)
787 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
788 buf->start + buf->len - 1, WB_SYNC_ALL);
791 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
793 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
794 buf->start, buf->start + buf->len -1);
797 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
798 u32 blocksize, u64 parent_transid)
800 struct extent_buffer *buf = NULL;
801 struct inode *btree_inode = root->fs_info->btree_inode;
802 struct extent_io_tree *io_tree;
805 io_tree = &BTRFS_I(btree_inode)->io_tree;
807 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
811 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
814 buf->flags |= EXTENT_UPTODATE;
822 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
823 struct extent_buffer *buf)
825 struct inode *btree_inode = root->fs_info->btree_inode;
826 if (btrfs_header_generation(buf) ==
827 root->fs_info->running_transaction->transid) {
828 WARN_ON(!btrfs_tree_locked(buf));
829 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
835 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
836 u32 stripesize, struct btrfs_root *root,
837 struct btrfs_fs_info *fs_info,
841 root->commit_root = NULL;
842 root->ref_tree = NULL;
843 root->sectorsize = sectorsize;
844 root->nodesize = nodesize;
845 root->leafsize = leafsize;
846 root->stripesize = stripesize;
848 root->track_dirty = 0;
850 root->fs_info = fs_info;
851 root->objectid = objectid;
852 root->last_trans = 0;
853 root->highest_inode = 0;
854 root->last_inode_alloc = 0;
858 INIT_LIST_HEAD(&root->dirty_list);
859 INIT_LIST_HEAD(&root->orphan_list);
860 INIT_LIST_HEAD(&root->dead_list);
861 spin_lock_init(&root->node_lock);
862 spin_lock_init(&root->list_lock);
863 mutex_init(&root->objectid_mutex);
864 mutex_init(&root->log_mutex);
865 extent_io_tree_init(&root->dirty_log_pages,
866 fs_info->btree_inode->i_mapping, GFP_NOFS);
868 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
869 root->ref_tree = &root->ref_tree_struct;
871 memset(&root->root_key, 0, sizeof(root->root_key));
872 memset(&root->root_item, 0, sizeof(root->root_item));
873 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
874 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
875 root->defrag_trans_start = fs_info->generation;
876 init_completion(&root->kobj_unregister);
877 root->defrag_running = 0;
878 root->defrag_level = 0;
879 root->root_key.objectid = objectid;
880 root->anon_super.s_root = NULL;
881 root->anon_super.s_dev = 0;
882 INIT_LIST_HEAD(&root->anon_super.s_list);
883 INIT_LIST_HEAD(&root->anon_super.s_instances);
884 init_rwsem(&root->anon_super.s_umount);
889 static int find_and_setup_root(struct btrfs_root *tree_root,
890 struct btrfs_fs_info *fs_info,
892 struct btrfs_root *root)
898 __setup_root(tree_root->nodesize, tree_root->leafsize,
899 tree_root->sectorsize, tree_root->stripesize,
900 root, fs_info, objectid);
901 ret = btrfs_find_last_root(tree_root, objectid,
902 &root->root_item, &root->root_key);
905 generation = btrfs_root_generation(&root->root_item);
906 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
907 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
908 blocksize, generation);
913 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
914 struct btrfs_fs_info *fs_info)
916 struct extent_buffer *eb;
917 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
926 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
927 0, &start, &end, EXTENT_DIRTY);
931 clear_extent_dirty(&log_root_tree->dirty_log_pages,
932 start, end, GFP_NOFS);
934 eb = fs_info->log_root_tree->node;
936 WARN_ON(btrfs_header_level(eb) != 0);
937 WARN_ON(btrfs_header_nritems(eb) != 0);
939 ret = btrfs_free_reserved_extent(fs_info->tree_root,
943 free_extent_buffer(eb);
944 kfree(fs_info->log_root_tree);
945 fs_info->log_root_tree = NULL;
949 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
950 struct btrfs_fs_info *fs_info)
952 struct btrfs_root *root;
953 struct btrfs_root *tree_root = fs_info->tree_root;
955 root = kzalloc(sizeof(*root), GFP_NOFS);
959 __setup_root(tree_root->nodesize, tree_root->leafsize,
960 tree_root->sectorsize, tree_root->stripesize,
961 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
963 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
964 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
965 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
968 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
969 0, BTRFS_TREE_LOG_OBJECTID,
970 trans->transid, 0, 0, 0);
972 btrfs_set_header_nritems(root->node, 0);
973 btrfs_set_header_level(root->node, 0);
974 btrfs_set_header_bytenr(root->node, root->node->start);
975 btrfs_set_header_generation(root->node, trans->transid);
976 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
978 write_extent_buffer(root->node, root->fs_info->fsid,
979 (unsigned long)btrfs_header_fsid(root->node),
981 btrfs_mark_buffer_dirty(root->node);
982 btrfs_tree_unlock(root->node);
983 fs_info->log_root_tree = root;
987 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
988 struct btrfs_key *location)
990 struct btrfs_root *root;
991 struct btrfs_fs_info *fs_info = tree_root->fs_info;
992 struct btrfs_path *path;
993 struct extent_buffer *l;
999 root = kzalloc(sizeof(*root), GFP_NOFS);
1001 return ERR_PTR(-ENOMEM);
1002 if (location->offset == (u64)-1) {
1003 ret = find_and_setup_root(tree_root, fs_info,
1004 location->objectid, root);
1007 return ERR_PTR(ret);
1012 __setup_root(tree_root->nodesize, tree_root->leafsize,
1013 tree_root->sectorsize, tree_root->stripesize,
1014 root, fs_info, location->objectid);
1016 path = btrfs_alloc_path();
1018 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1025 read_extent_buffer(l, &root->root_item,
1026 btrfs_item_ptr_offset(l, path->slots[0]),
1027 sizeof(root->root_item));
1028 memcpy(&root->root_key, location, sizeof(*location));
1031 btrfs_release_path(root, path);
1032 btrfs_free_path(path);
1035 return ERR_PTR(ret);
1037 generation = btrfs_root_generation(&root->root_item);
1038 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1039 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1040 blocksize, generation);
1041 BUG_ON(!root->node);
1043 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1045 ret = btrfs_find_highest_inode(root, &highest_inode);
1047 root->highest_inode = highest_inode;
1048 root->last_inode_alloc = highest_inode;
1054 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1057 struct btrfs_root *root;
1059 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1060 return fs_info->tree_root;
1061 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1062 return fs_info->extent_root;
1064 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1065 (unsigned long)root_objectid);
1069 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1070 struct btrfs_key *location)
1072 struct btrfs_root *root;
1075 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1076 return fs_info->tree_root;
1077 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1078 return fs_info->extent_root;
1079 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1080 return fs_info->chunk_root;
1081 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1082 return fs_info->dev_root;
1084 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1085 (unsigned long)location->objectid);
1089 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1093 set_anon_super(&root->anon_super, NULL);
1095 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1096 (unsigned long)root->root_key.objectid,
1099 free_extent_buffer(root->node);
1101 return ERR_PTR(ret);
1103 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1104 ret = btrfs_find_dead_roots(fs_info->tree_root,
1105 root->root_key.objectid, root);
1107 btrfs_orphan_cleanup(root);
1112 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1113 struct btrfs_key *location,
1114 const char *name, int namelen)
1116 struct btrfs_root *root;
1119 root = btrfs_read_fs_root_no_name(fs_info, location);
1126 ret = btrfs_set_root_name(root, name, namelen);
1128 free_extent_buffer(root->node);
1130 return ERR_PTR(ret);
1133 ret = btrfs_sysfs_add_root(root);
1135 free_extent_buffer(root->node);
1138 return ERR_PTR(ret);
1145 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1146 struct btrfs_hasher *hasher;
1148 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1151 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1152 if (!hasher->hash_tfm) {
1156 spin_lock(&info->hash_lock);
1157 list_add(&hasher->list, &info->hashers);
1158 spin_unlock(&info->hash_lock);
1163 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1165 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1167 struct list_head *cur;
1168 struct btrfs_device *device;
1169 struct backing_dev_info *bdi;
1171 if ((bdi_bits & (1 << BDI_write_congested)) &&
1172 btrfs_congested_async(info, 0))
1175 list_for_each(cur, &info->fs_devices->devices) {
1176 device = list_entry(cur, struct btrfs_device, dev_list);
1179 bdi = blk_get_backing_dev_info(device->bdev);
1180 if (bdi && bdi_congested(bdi, bdi_bits)) {
1189 * this unplugs every device on the box, and it is only used when page
1192 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1194 struct list_head *cur;
1195 struct btrfs_device *device;
1196 struct btrfs_fs_info *info;
1198 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1199 list_for_each(cur, &info->fs_devices->devices) {
1200 device = list_entry(cur, struct btrfs_device, dev_list);
1201 bdi = blk_get_backing_dev_info(device->bdev);
1202 if (bdi->unplug_io_fn) {
1203 bdi->unplug_io_fn(bdi, page);
1208 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1210 struct inode *inode;
1211 struct extent_map_tree *em_tree;
1212 struct extent_map *em;
1213 struct address_space *mapping;
1216 /* the generic O_DIRECT read code does this */
1218 __unplug_io_fn(bdi, page);
1223 * page->mapping may change at any time. Get a consistent copy
1224 * and use that for everything below
1227 mapping = page->mapping;
1231 inode = mapping->host;
1234 * don't do the expensive searching for a small number of
1237 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1238 __unplug_io_fn(bdi, page);
1242 offset = page_offset(page);
1244 em_tree = &BTRFS_I(inode)->extent_tree;
1245 spin_lock(&em_tree->lock);
1246 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1247 spin_unlock(&em_tree->lock);
1249 __unplug_io_fn(bdi, page);
1253 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1254 free_extent_map(em);
1255 __unplug_io_fn(bdi, page);
1258 offset = offset - em->start;
1259 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1260 em->block_start + offset, page);
1261 free_extent_map(em);
1264 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1267 bdi->ra_pages = default_backing_dev_info.ra_pages;
1269 bdi->capabilities = default_backing_dev_info.capabilities;
1270 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1271 bdi->unplug_io_data = info;
1272 bdi->congested_fn = btrfs_congested_fn;
1273 bdi->congested_data = info;
1277 static int bio_ready_for_csum(struct bio *bio)
1283 struct extent_io_tree *io_tree = NULL;
1284 struct btrfs_fs_info *info = NULL;
1285 struct bio_vec *bvec;
1289 bio_for_each_segment(bvec, bio, i) {
1290 page = bvec->bv_page;
1291 if (page->private == EXTENT_PAGE_PRIVATE) {
1292 length += bvec->bv_len;
1295 if (!page->private) {
1296 length += bvec->bv_len;
1299 length = bvec->bv_len;
1300 buf_len = page->private >> 2;
1301 start = page_offset(page) + bvec->bv_offset;
1302 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1303 info = BTRFS_I(page->mapping->host)->root->fs_info;
1305 /* are we fully contained in this bio? */
1306 if (buf_len <= length)
1309 ret = extent_range_uptodate(io_tree, start + length,
1310 start + buf_len - 1);
1317 * called by the kthread helper functions to finally call the bio end_io
1318 * functions. This is where read checksum verification actually happens
1320 static void end_workqueue_fn(struct btrfs_work *work)
1323 struct end_io_wq *end_io_wq;
1324 struct btrfs_fs_info *fs_info;
1327 end_io_wq = container_of(work, struct end_io_wq, work);
1328 bio = end_io_wq->bio;
1329 fs_info = end_io_wq->info;
1331 /* metadata bios are special because the whole tree block must
1332 * be checksummed at once. This makes sure the entire block is in
1333 * ram and up to date before trying to verify things. For
1334 * blocksize <= pagesize, it is basically a noop
1336 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1337 btrfs_queue_worker(&fs_info->endio_workers,
1341 error = end_io_wq->error;
1342 bio->bi_private = end_io_wq->private;
1343 bio->bi_end_io = end_io_wq->end_io;
1345 bio_endio(bio, error);
1348 static int cleaner_kthread(void *arg)
1350 struct btrfs_root *root = arg;
1354 if (root->fs_info->closing)
1357 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1358 mutex_lock(&root->fs_info->cleaner_mutex);
1359 btrfs_clean_old_snapshots(root);
1360 mutex_unlock(&root->fs_info->cleaner_mutex);
1362 if (freezing(current)) {
1366 if (root->fs_info->closing)
1368 set_current_state(TASK_INTERRUPTIBLE);
1370 __set_current_state(TASK_RUNNING);
1372 } while (!kthread_should_stop());
1376 static int transaction_kthread(void *arg)
1378 struct btrfs_root *root = arg;
1379 struct btrfs_trans_handle *trans;
1380 struct btrfs_transaction *cur;
1382 unsigned long delay;
1387 if (root->fs_info->closing)
1391 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1392 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1394 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1395 printk("btrfs: total reference cache size %Lu\n",
1396 root->fs_info->total_ref_cache_size);
1399 mutex_lock(&root->fs_info->trans_mutex);
1400 cur = root->fs_info->running_transaction;
1402 mutex_unlock(&root->fs_info->trans_mutex);
1406 now = get_seconds();
1407 if (now < cur->start_time || now - cur->start_time < 30) {
1408 mutex_unlock(&root->fs_info->trans_mutex);
1412 mutex_unlock(&root->fs_info->trans_mutex);
1413 trans = btrfs_start_transaction(root, 1);
1414 ret = btrfs_commit_transaction(trans, root);
1416 wake_up_process(root->fs_info->cleaner_kthread);
1417 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1419 if (freezing(current)) {
1422 if (root->fs_info->closing)
1424 set_current_state(TASK_INTERRUPTIBLE);
1425 schedule_timeout(delay);
1426 __set_current_state(TASK_RUNNING);
1428 } while (!kthread_should_stop());
1432 struct btrfs_root *open_ctree(struct super_block *sb,
1433 struct btrfs_fs_devices *fs_devices,
1442 struct btrfs_key location;
1443 struct buffer_head *bh;
1444 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1446 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1448 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1450 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1452 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1454 struct btrfs_root *log_tree_root;
1459 struct btrfs_super_block *disk_super;
1461 if (!extent_root || !tree_root || !fs_info ||
1462 !chunk_root || !dev_root) {
1466 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1467 INIT_LIST_HEAD(&fs_info->trans_list);
1468 INIT_LIST_HEAD(&fs_info->dead_roots);
1469 INIT_LIST_HEAD(&fs_info->hashers);
1470 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1471 spin_lock_init(&fs_info->hash_lock);
1472 spin_lock_init(&fs_info->delalloc_lock);
1473 spin_lock_init(&fs_info->new_trans_lock);
1474 spin_lock_init(&fs_info->ref_cache_lock);
1476 init_completion(&fs_info->kobj_unregister);
1477 fs_info->tree_root = tree_root;
1478 fs_info->extent_root = extent_root;
1479 fs_info->chunk_root = chunk_root;
1480 fs_info->dev_root = dev_root;
1481 fs_info->fs_devices = fs_devices;
1482 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1483 INIT_LIST_HEAD(&fs_info->space_info);
1484 btrfs_mapping_init(&fs_info->mapping_tree);
1485 atomic_set(&fs_info->nr_async_submits, 0);
1486 atomic_set(&fs_info->async_delalloc_pages, 0);
1487 atomic_set(&fs_info->async_submit_draining, 0);
1488 atomic_set(&fs_info->nr_async_bios, 0);
1489 atomic_set(&fs_info->throttles, 0);
1490 atomic_set(&fs_info->throttle_gen, 0);
1492 fs_info->max_extent = (u64)-1;
1493 fs_info->max_inline = 8192 * 1024;
1494 setup_bdi(fs_info, &fs_info->bdi);
1495 fs_info->btree_inode = new_inode(sb);
1496 fs_info->btree_inode->i_ino = 1;
1497 fs_info->btree_inode->i_nlink = 1;
1499 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1501 INIT_LIST_HEAD(&fs_info->ordered_extents);
1502 spin_lock_init(&fs_info->ordered_extent_lock);
1504 sb->s_blocksize = 4096;
1505 sb->s_blocksize_bits = blksize_bits(4096);
1508 * we set the i_size on the btree inode to the max possible int.
1509 * the real end of the address space is determined by all of
1510 * the devices in the system
1512 fs_info->btree_inode->i_size = OFFSET_MAX;
1513 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1514 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1516 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1517 fs_info->btree_inode->i_mapping,
1519 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1522 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1524 spin_lock_init(&fs_info->block_group_cache_lock);
1525 fs_info->block_group_cache_tree.rb_node = NULL;
1527 extent_io_tree_init(&fs_info->pinned_extents,
1528 fs_info->btree_inode->i_mapping, GFP_NOFS);
1529 extent_io_tree_init(&fs_info->pending_del,
1530 fs_info->btree_inode->i_mapping, GFP_NOFS);
1531 extent_io_tree_init(&fs_info->extent_ins,
1532 fs_info->btree_inode->i_mapping, GFP_NOFS);
1533 fs_info->do_barriers = 1;
1535 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1536 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1537 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1539 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1540 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1541 sizeof(struct btrfs_key));
1542 insert_inode_hash(fs_info->btree_inode);
1544 mutex_init(&fs_info->trans_mutex);
1545 mutex_init(&fs_info->tree_log_mutex);
1546 mutex_init(&fs_info->drop_mutex);
1547 mutex_init(&fs_info->extent_ins_mutex);
1548 mutex_init(&fs_info->pinned_mutex);
1549 mutex_init(&fs_info->chunk_mutex);
1550 mutex_init(&fs_info->transaction_kthread_mutex);
1551 mutex_init(&fs_info->cleaner_mutex);
1552 mutex_init(&fs_info->volume_mutex);
1553 mutex_init(&fs_info->tree_reloc_mutex);
1554 init_waitqueue_head(&fs_info->transaction_throttle);
1555 init_waitqueue_head(&fs_info->transaction_wait);
1556 init_waitqueue_head(&fs_info->async_submit_wait);
1557 init_waitqueue_head(&fs_info->tree_log_wait);
1558 atomic_set(&fs_info->tree_log_commit, 0);
1559 atomic_set(&fs_info->tree_log_writers, 0);
1560 fs_info->tree_log_transid = 0;
1563 ret = add_hasher(fs_info, "crc32c");
1565 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1570 __setup_root(4096, 4096, 4096, 4096, tree_root,
1571 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1574 bh = __bread(fs_devices->latest_bdev,
1575 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1579 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1582 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1584 disk_super = &fs_info->super_copy;
1585 if (!btrfs_super_root(disk_super))
1586 goto fail_sb_buffer;
1588 ret = btrfs_parse_options(tree_root, options);
1591 goto fail_sb_buffer;
1595 * we need to start all the end_io workers up front because the
1596 * queue work function gets called at interrupt time, and so it
1597 * cannot dynamically grow.
1599 btrfs_init_workers(&fs_info->workers, "worker",
1600 fs_info->thread_pool_size);
1602 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1603 fs_info->thread_pool_size);
1605 btrfs_init_workers(&fs_info->submit_workers, "submit",
1606 min_t(u64, fs_devices->num_devices,
1607 fs_info->thread_pool_size));
1609 /* a higher idle thresh on the submit workers makes it much more
1610 * likely that bios will be send down in a sane order to the
1613 fs_info->submit_workers.idle_thresh = 64;
1615 fs_info->workers.idle_thresh = 16;
1616 fs_info->workers.ordered = 1;
1618 fs_info->delalloc_workers.idle_thresh = 2;
1619 fs_info->delalloc_workers.ordered = 1;
1621 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1622 btrfs_init_workers(&fs_info->endio_workers, "endio",
1623 fs_info->thread_pool_size);
1624 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1625 fs_info->thread_pool_size);
1628 * endios are largely parallel and should have a very
1631 fs_info->endio_workers.idle_thresh = 4;
1632 fs_info->endio_write_workers.idle_thresh = 64;
1634 btrfs_start_workers(&fs_info->workers, 1);
1635 btrfs_start_workers(&fs_info->submit_workers, 1);
1636 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1637 btrfs_start_workers(&fs_info->fixup_workers, 1);
1638 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1639 btrfs_start_workers(&fs_info->endio_write_workers,
1640 fs_info->thread_pool_size);
1642 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1643 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1644 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1646 nodesize = btrfs_super_nodesize(disk_super);
1647 leafsize = btrfs_super_leafsize(disk_super);
1648 sectorsize = btrfs_super_sectorsize(disk_super);
1649 stripesize = btrfs_super_stripesize(disk_super);
1650 tree_root->nodesize = nodesize;
1651 tree_root->leafsize = leafsize;
1652 tree_root->sectorsize = sectorsize;
1653 tree_root->stripesize = stripesize;
1655 sb->s_blocksize = sectorsize;
1656 sb->s_blocksize_bits = blksize_bits(sectorsize);
1658 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1659 sizeof(disk_super->magic))) {
1660 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1661 goto fail_sb_buffer;
1664 mutex_lock(&fs_info->chunk_mutex);
1665 ret = btrfs_read_sys_array(tree_root);
1666 mutex_unlock(&fs_info->chunk_mutex);
1668 printk("btrfs: failed to read the system array on %s\n",
1670 goto fail_sys_array;
1673 blocksize = btrfs_level_size(tree_root,
1674 btrfs_super_chunk_root_level(disk_super));
1675 generation = btrfs_super_chunk_root_generation(disk_super);
1677 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1678 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1680 chunk_root->node = read_tree_block(chunk_root,
1681 btrfs_super_chunk_root(disk_super),
1682 blocksize, generation);
1683 BUG_ON(!chunk_root->node);
1685 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1686 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1689 mutex_lock(&fs_info->chunk_mutex);
1690 ret = btrfs_read_chunk_tree(chunk_root);
1691 mutex_unlock(&fs_info->chunk_mutex);
1693 printk("btrfs: failed to read chunk tree on %s\n", sb->s_id);
1694 goto fail_chunk_root;
1697 btrfs_close_extra_devices(fs_devices);
1699 blocksize = btrfs_level_size(tree_root,
1700 btrfs_super_root_level(disk_super));
1701 generation = btrfs_super_generation(disk_super);
1703 tree_root->node = read_tree_block(tree_root,
1704 btrfs_super_root(disk_super),
1705 blocksize, generation);
1706 if (!tree_root->node)
1707 goto fail_chunk_root;
1710 ret = find_and_setup_root(tree_root, fs_info,
1711 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1713 goto fail_tree_root;
1714 extent_root->track_dirty = 1;
1716 ret = find_and_setup_root(tree_root, fs_info,
1717 BTRFS_DEV_TREE_OBJECTID, dev_root);
1718 dev_root->track_dirty = 1;
1721 goto fail_extent_root;
1723 btrfs_read_block_groups(extent_root);
1725 fs_info->generation = generation + 1;
1726 fs_info->last_trans_committed = generation;
1727 fs_info->data_alloc_profile = (u64)-1;
1728 fs_info->metadata_alloc_profile = (u64)-1;
1729 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1730 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1732 if (!fs_info->cleaner_kthread)
1733 goto fail_extent_root;
1735 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1737 "btrfs-transaction");
1738 if (!fs_info->transaction_kthread)
1741 if (sb->s_flags & MS_RDONLY)
1744 if (btrfs_super_log_root(disk_super) != 0) {
1746 u64 bytenr = btrfs_super_log_root(disk_super);
1749 btrfs_level_size(tree_root,
1750 btrfs_super_log_root_level(disk_super));
1752 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1755 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1756 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1758 log_tree_root->node = read_tree_block(tree_root, bytenr,
1761 ret = btrfs_recover_log_trees(log_tree_root);
1765 ret = btrfs_cleanup_reloc_trees(tree_root);
1768 location.objectid = BTRFS_FS_TREE_OBJECTID;
1769 location.type = BTRFS_ROOT_ITEM_KEY;
1770 location.offset = (u64)-1;
1773 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1774 if (!fs_info->fs_root)
1779 kthread_stop(fs_info->cleaner_kthread);
1781 free_extent_buffer(extent_root->node);
1783 free_extent_buffer(tree_root->node);
1785 free_extent_buffer(chunk_root->node);
1788 btrfs_stop_workers(&fs_info->fixup_workers);
1789 btrfs_stop_workers(&fs_info->delalloc_workers);
1790 btrfs_stop_workers(&fs_info->workers);
1791 btrfs_stop_workers(&fs_info->endio_workers);
1792 btrfs_stop_workers(&fs_info->endio_write_workers);
1793 btrfs_stop_workers(&fs_info->submit_workers);
1795 iput(fs_info->btree_inode);
1797 btrfs_close_devices(fs_info->fs_devices);
1798 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1802 bdi_destroy(&fs_info->bdi);
1806 return ERR_PTR(err);
1809 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1811 char b[BDEVNAME_SIZE];
1814 set_buffer_uptodate(bh);
1816 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1817 printk(KERN_WARNING "lost page write due to "
1818 "I/O error on %s\n",
1819 bdevname(bh->b_bdev, b));
1821 /* note, we dont' set_buffer_write_io_error because we have
1822 * our own ways of dealing with the IO errors
1824 clear_buffer_uptodate(bh);
1830 int write_all_supers(struct btrfs_root *root)
1832 struct list_head *cur;
1833 struct list_head *head = &root->fs_info->fs_devices->devices;
1834 struct btrfs_device *dev;
1835 struct btrfs_super_block *sb;
1836 struct btrfs_dev_item *dev_item;
1837 struct buffer_head *bh;
1841 int total_errors = 0;
1845 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1846 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1848 sb = &root->fs_info->super_for_commit;
1849 dev_item = &sb->dev_item;
1850 list_for_each(cur, head) {
1851 dev = list_entry(cur, struct btrfs_device, dev_list);
1856 if (!dev->in_fs_metadata || !dev->writeable)
1859 btrfs_set_stack_device_generation(dev_item, 0);
1860 btrfs_set_stack_device_type(dev_item, dev->type);
1861 btrfs_set_stack_device_id(dev_item, dev->devid);
1862 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1863 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1864 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1865 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1866 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1867 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1868 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1869 flags = btrfs_super_flags(sb);
1870 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1874 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1875 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1876 btrfs_csum_final(crc, sb->csum);
1878 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1879 BTRFS_SUPER_INFO_SIZE);
1881 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1882 dev->pending_io = bh;
1885 set_buffer_uptodate(bh);
1887 bh->b_end_io = btrfs_end_buffer_write_sync;
1889 if (do_barriers && dev->barriers) {
1890 ret = submit_bh(WRITE_BARRIER, bh);
1891 if (ret == -EOPNOTSUPP) {
1892 printk("btrfs: disabling barriers on dev %s\n",
1894 set_buffer_uptodate(bh);
1898 ret = submit_bh(WRITE, bh);
1901 ret = submit_bh(WRITE, bh);
1906 if (total_errors > max_errors) {
1907 printk("btrfs: %d errors while writing supers\n", total_errors);
1912 list_for_each(cur, head) {
1913 dev = list_entry(cur, struct btrfs_device, dev_list);
1916 if (!dev->in_fs_metadata || !dev->writeable)
1919 BUG_ON(!dev->pending_io);
1920 bh = dev->pending_io;
1922 if (!buffer_uptodate(dev->pending_io)) {
1923 if (do_barriers && dev->barriers) {
1924 printk("btrfs: disabling barriers on dev %s\n",
1926 set_buffer_uptodate(bh);
1930 ret = submit_bh(WRITE, bh);
1933 if (!buffer_uptodate(bh))
1940 dev->pending_io = NULL;
1943 if (total_errors > max_errors) {
1944 printk("btrfs: %d errors while writing supers\n", total_errors);
1950 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1955 ret = write_all_supers(root);
1959 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1961 radix_tree_delete(&fs_info->fs_roots_radix,
1962 (unsigned long)root->root_key.objectid);
1963 if (root->anon_super.s_dev) {
1964 down_write(&root->anon_super.s_umount);
1965 kill_anon_super(&root->anon_super);
1969 btrfs_sysfs_del_root(root);
1972 free_extent_buffer(root->node);
1973 if (root->commit_root)
1974 free_extent_buffer(root->commit_root);
1981 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1984 struct btrfs_root *gang[8];
1988 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1993 for (i = 0; i < ret; i++)
1994 btrfs_free_fs_root(fs_info, gang[i]);
1999 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2001 u64 root_objectid = 0;
2002 struct btrfs_root *gang[8];
2007 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2008 (void **)gang, root_objectid,
2012 for (i = 0; i < ret; i++) {
2013 root_objectid = gang[i]->root_key.objectid;
2014 ret = btrfs_find_dead_roots(fs_info->tree_root,
2015 root_objectid, gang[i]);
2017 btrfs_orphan_cleanup(gang[i]);
2024 int btrfs_commit_super(struct btrfs_root *root)
2026 struct btrfs_trans_handle *trans;
2029 mutex_lock(&root->fs_info->cleaner_mutex);
2030 btrfs_clean_old_snapshots(root);
2031 mutex_unlock(&root->fs_info->cleaner_mutex);
2032 trans = btrfs_start_transaction(root, 1);
2033 ret = btrfs_commit_transaction(trans, root);
2035 /* run commit again to drop the original snapshot */
2036 trans = btrfs_start_transaction(root, 1);
2037 btrfs_commit_transaction(trans, root);
2038 ret = btrfs_write_and_wait_transaction(NULL, root);
2041 ret = write_ctree_super(NULL, root);
2045 int close_ctree(struct btrfs_root *root)
2047 struct btrfs_fs_info *fs_info = root->fs_info;
2050 fs_info->closing = 1;
2053 kthread_stop(root->fs_info->transaction_kthread);
2054 kthread_stop(root->fs_info->cleaner_kthread);
2056 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2057 ret = btrfs_commit_super(root);
2059 printk("btrfs: commit super returns %d\n", ret);
2063 if (fs_info->delalloc_bytes) {
2064 printk("btrfs: at unmount delalloc count %Lu\n",
2065 fs_info->delalloc_bytes);
2067 if (fs_info->total_ref_cache_size) {
2068 printk("btrfs: at umount reference cache size %Lu\n",
2069 fs_info->total_ref_cache_size);
2072 if (fs_info->extent_root->node)
2073 free_extent_buffer(fs_info->extent_root->node);
2075 if (fs_info->tree_root->node)
2076 free_extent_buffer(fs_info->tree_root->node);
2078 if (root->fs_info->chunk_root->node);
2079 free_extent_buffer(root->fs_info->chunk_root->node);
2081 if (root->fs_info->dev_root->node);
2082 free_extent_buffer(root->fs_info->dev_root->node);
2084 btrfs_free_block_groups(root->fs_info);
2086 del_fs_roots(fs_info);
2088 iput(fs_info->btree_inode);
2090 btrfs_stop_workers(&fs_info->fixup_workers);
2091 btrfs_stop_workers(&fs_info->delalloc_workers);
2092 btrfs_stop_workers(&fs_info->workers);
2093 btrfs_stop_workers(&fs_info->endio_workers);
2094 btrfs_stop_workers(&fs_info->endio_write_workers);
2095 btrfs_stop_workers(&fs_info->submit_workers);
2098 while(!list_empty(&fs_info->hashers)) {
2099 struct btrfs_hasher *hasher;
2100 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2102 list_del(&hasher->hashers);
2103 crypto_free_hash(&fs_info->hash_tfm);
2107 btrfs_close_devices(fs_info->fs_devices);
2108 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2110 bdi_destroy(&fs_info->bdi);
2112 kfree(fs_info->extent_root);
2113 kfree(fs_info->tree_root);
2114 kfree(fs_info->chunk_root);
2115 kfree(fs_info->dev_root);
2119 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2122 struct inode *btree_inode = buf->first_page->mapping->host;
2124 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2128 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2133 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2135 struct inode *btree_inode = buf->first_page->mapping->host;
2136 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2140 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2142 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2143 u64 transid = btrfs_header_generation(buf);
2144 struct inode *btree_inode = root->fs_info->btree_inode;
2146 WARN_ON(!btrfs_tree_locked(buf));
2147 if (transid != root->fs_info->generation) {
2148 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2149 (unsigned long long)buf->start,
2150 transid, root->fs_info->generation);
2153 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2156 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2159 * looks as though older kernels can get into trouble with
2160 * this code, they end up stuck in balance_dirty_pages forever
2162 struct extent_io_tree *tree;
2165 unsigned long thresh = 32 * 1024 * 1024;
2166 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2168 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2171 num_dirty = count_range_bits(tree, &start, (u64)-1,
2172 thresh, EXTENT_DIRTY);
2173 if (num_dirty > thresh) {
2174 balance_dirty_pages_ratelimited_nr(
2175 root->fs_info->btree_inode->i_mapping, 1);
2180 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2182 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2184 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2186 buf->flags |= EXTENT_UPTODATE;
2191 int btree_lock_page_hook(struct page *page)
2193 struct inode *inode = page->mapping->host;
2194 struct btrfs_root *root = BTRFS_I(inode)->root;
2195 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2196 struct extent_buffer *eb;
2198 u64 bytenr = page_offset(page);
2200 if (page->private == EXTENT_PAGE_PRIVATE)
2203 len = page->private >> 2;
2204 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2208 btrfs_tree_lock(eb);
2209 spin_lock(&root->fs_info->hash_lock);
2210 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2211 spin_unlock(&root->fs_info->hash_lock);
2212 btrfs_tree_unlock(eb);
2213 free_extent_buffer(eb);
2219 static struct extent_io_ops btree_extent_io_ops = {
2220 .write_cache_pages_lock_hook = btree_lock_page_hook,
2221 .readpage_end_io_hook = btree_readpage_end_io_hook,
2222 .submit_bio_hook = btree_submit_bio_hook,
2223 /* note we're sharing with inode.c for the merge bio hook */
2224 .merge_bio_hook = btrfs_merge_bio_hook,