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>
34 #include "transaction.h"
35 #include "btrfs_inode.h"
37 #include "print-tree.h"
38 #include "async-thread.h"
40 #include "ref-cache.h"
44 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
46 if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
47 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
48 (unsigned long long)extent_buffer_blocknr(buf),
49 (unsigned long long)btrfs_header_blocknr(buf));
56 static struct extent_io_ops btree_extent_io_ops;
57 static void end_workqueue_fn(struct btrfs_work *work);
60 * end_io_wq structs are used to do processing in task context when an IO is
61 * complete. This is used during reads to verify checksums, and it is used
62 * by writes to insert metadata for new file extents after IO is complete.
68 struct btrfs_fs_info *info;
71 struct list_head list;
72 struct btrfs_work work;
76 * async submit bios are used to offload expensive checksumming
77 * onto the worker threads. They checksum file and metadata bios
78 * just before they are sent down the IO stack.
80 struct async_submit_bio {
83 struct list_head list;
84 extent_submit_bio_hook_t *submit_bio_start;
85 extent_submit_bio_hook_t *submit_bio_done;
88 unsigned long bio_flags;
89 struct btrfs_work work;
93 * extents on the btree inode are pretty simple, there's one extent
94 * that covers the entire device
96 static struct extent_map *btree_get_extent(struct inode *inode,
97 struct page *page, size_t page_offset, u64 start, u64 len,
100 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
101 struct extent_map *em;
104 spin_lock(&em_tree->lock);
105 em = lookup_extent_mapping(em_tree, start, len);
108 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
109 spin_unlock(&em_tree->lock);
112 spin_unlock(&em_tree->lock);
114 em = alloc_extent_map(GFP_NOFS);
116 em = ERR_PTR(-ENOMEM);
121 em->block_len = (u64)-1;
123 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
125 spin_lock(&em_tree->lock);
126 ret = add_extent_mapping(em_tree, em);
127 if (ret == -EEXIST) {
128 u64 failed_start = em->start;
129 u64 failed_len = em->len;
131 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
132 em->start, em->len, em->block_start);
134 em = lookup_extent_mapping(em_tree, start, len);
136 printk("after failing, found %Lu %Lu %Lu\n",
137 em->start, em->len, em->block_start);
140 em = lookup_extent_mapping(em_tree, failed_start,
143 printk("double failure lookup gives us "
144 "%Lu %Lu -> %Lu\n", em->start,
145 em->len, em->block_start);
154 spin_unlock(&em_tree->lock);
162 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
164 return btrfs_crc32c(seed, data, len);
167 void btrfs_csum_final(u32 crc, char *result)
169 *(__le32 *)result = ~cpu_to_le32(crc);
173 * compute the csum for a btree block, and either verify it or write it
174 * into the csum field of the block.
176 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
180 btrfs_super_csum_size(&root->fs_info->super_copy);
183 unsigned long cur_len;
184 unsigned long offset = BTRFS_CSUM_SIZE;
185 char *map_token = NULL;
187 unsigned long map_start;
188 unsigned long map_len;
191 unsigned long inline_result;
193 len = buf->len - offset;
195 err = map_private_extent_buffer(buf, offset, 32,
197 &map_start, &map_len, KM_USER0);
199 printk("failed to map extent buffer! %lu\n",
203 cur_len = min(len, map_len - (offset - map_start));
204 crc = btrfs_csum_data(root, kaddr + offset - map_start,
208 unmap_extent_buffer(buf, map_token, KM_USER0);
210 if (csum_size > sizeof(inline_result)) {
211 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
215 result = (char *)&inline_result;
218 btrfs_csum_final(crc, result);
221 /* FIXME, this is not good */
222 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
225 memcpy(&found, result, csum_size);
227 read_extent_buffer(buf, &val, 0, csum_size);
228 printk("btrfs: %s checksum verify failed on %llu "
229 "wanted %X found %X level %d\n",
230 root->fs_info->sb->s_id,
231 buf->start, val, found, btrfs_header_level(buf));
232 if (result != (char *)&inline_result)
237 write_extent_buffer(buf, result, 0, csum_size);
239 if (result != (char *)&inline_result)
245 * we can't consider a given block up to date unless the transid of the
246 * block matches the transid in the parent node's pointer. This is how we
247 * detect blocks that either didn't get written at all or got written
248 * in the wrong place.
250 static int verify_parent_transid(struct extent_io_tree *io_tree,
251 struct extent_buffer *eb, u64 parent_transid)
255 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
258 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
259 if (extent_buffer_uptodate(io_tree, eb) &&
260 btrfs_header_generation(eb) == parent_transid) {
264 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
265 (unsigned long long)eb->start,
266 (unsigned long long)parent_transid,
267 (unsigned long long)btrfs_header_generation(eb));
269 clear_extent_buffer_uptodate(io_tree, eb);
271 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
277 * helper to read a given tree block, doing retries as required when
278 * the checksums don't match and we have alternate mirrors to try.
280 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
281 struct extent_buffer *eb,
282 u64 start, u64 parent_transid)
284 struct extent_io_tree *io_tree;
289 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
291 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
292 btree_get_extent, mirror_num);
294 !verify_parent_transid(io_tree, eb, parent_transid))
296 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
297 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
303 if (mirror_num > num_copies)
310 * checksum a dirty tree block before IO. This has extra checks to make
311 * sure we only fill in the checksum field in the first page of a multi-page block
313 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
315 struct extent_io_tree *tree;
316 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
320 struct extent_buffer *eb;
323 tree = &BTRFS_I(page->mapping->host)->io_tree;
325 if (page->private == EXTENT_PAGE_PRIVATE)
329 len = page->private >> 2;
333 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
334 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
335 btrfs_header_generation(eb));
337 found_start = btrfs_header_bytenr(eb);
338 if (found_start != start) {
339 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
340 start, found_start, len);
344 if (eb->first_page != page) {
345 printk("bad first page %lu %lu\n", eb->first_page->index,
350 if (!PageUptodate(page)) {
351 printk("csum not up to date page %lu\n", page->index);
355 found_level = btrfs_header_level(eb);
357 csum_tree_block(root, eb, 0);
359 free_extent_buffer(eb);
364 static int check_tree_block_fsid(struct btrfs_root *root,
365 struct extent_buffer *eb)
367 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
368 u8 fsid[BTRFS_UUID_SIZE];
371 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
374 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
378 fs_devices = fs_devices->seed;
383 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
384 struct extent_state *state)
386 struct extent_io_tree *tree;
390 struct extent_buffer *eb;
391 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
394 tree = &BTRFS_I(page->mapping->host)->io_tree;
395 if (page->private == EXTENT_PAGE_PRIVATE)
399 len = page->private >> 2;
403 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
405 found_start = btrfs_header_bytenr(eb);
406 if (found_start != start) {
407 printk("bad tree block start %llu %llu\n",
408 (unsigned long long)found_start,
409 (unsigned long long)eb->start);
413 if (eb->first_page != page) {
414 printk("bad first page %lu %lu\n", eb->first_page->index,
420 if (check_tree_block_fsid(root, eb)) {
421 printk("bad fsid on block %Lu\n", eb->start);
425 found_level = btrfs_header_level(eb);
427 ret = csum_tree_block(root, eb, 1);
431 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
432 end = eb->start + end - 1;
434 free_extent_buffer(eb);
439 static void end_workqueue_bio(struct bio *bio, int err)
441 struct end_io_wq *end_io_wq = bio->bi_private;
442 struct btrfs_fs_info *fs_info;
444 fs_info = end_io_wq->info;
445 end_io_wq->error = err;
446 end_io_wq->work.func = end_workqueue_fn;
447 end_io_wq->work.flags = 0;
448 if (bio->bi_rw & (1 << BIO_RW))
449 btrfs_queue_worker(&fs_info->endio_write_workers,
452 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
455 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
458 struct end_io_wq *end_io_wq;
459 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
463 end_io_wq->private = bio->bi_private;
464 end_io_wq->end_io = bio->bi_end_io;
465 end_io_wq->info = info;
466 end_io_wq->error = 0;
467 end_io_wq->bio = bio;
468 end_io_wq->metadata = metadata;
470 bio->bi_private = end_io_wq;
471 bio->bi_end_io = end_workqueue_bio;
475 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
477 unsigned long limit = min_t(unsigned long,
478 info->workers.max_workers,
479 info->fs_devices->open_devices);
483 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
485 return atomic_read(&info->nr_async_bios) >
486 btrfs_async_submit_limit(info);
489 static void run_one_async_start(struct btrfs_work *work)
491 struct btrfs_fs_info *fs_info;
492 struct async_submit_bio *async;
494 async = container_of(work, struct async_submit_bio, work);
495 fs_info = BTRFS_I(async->inode)->root->fs_info;
496 async->submit_bio_start(async->inode, async->rw, async->bio,
497 async->mirror_num, async->bio_flags);
500 static void run_one_async_done(struct btrfs_work *work)
502 struct btrfs_fs_info *fs_info;
503 struct async_submit_bio *async;
506 async = container_of(work, struct async_submit_bio, work);
507 fs_info = BTRFS_I(async->inode)->root->fs_info;
509 limit = btrfs_async_submit_limit(fs_info);
510 limit = limit * 2 / 3;
512 atomic_dec(&fs_info->nr_async_submits);
514 if (atomic_read(&fs_info->nr_async_submits) < limit &&
515 waitqueue_active(&fs_info->async_submit_wait))
516 wake_up(&fs_info->async_submit_wait);
518 async->submit_bio_done(async->inode, async->rw, async->bio,
519 async->mirror_num, async->bio_flags);
522 static void run_one_async_free(struct btrfs_work *work)
524 struct async_submit_bio *async;
526 async = container_of(work, struct async_submit_bio, work);
530 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
531 int rw, struct bio *bio, int mirror_num,
532 unsigned long bio_flags,
533 extent_submit_bio_hook_t *submit_bio_start,
534 extent_submit_bio_hook_t *submit_bio_done)
536 struct async_submit_bio *async;
538 async = kmalloc(sizeof(*async), GFP_NOFS);
542 async->inode = inode;
545 async->mirror_num = mirror_num;
546 async->submit_bio_start = submit_bio_start;
547 async->submit_bio_done = submit_bio_done;
549 async->work.func = run_one_async_start;
550 async->work.ordered_func = run_one_async_done;
551 async->work.ordered_free = run_one_async_free;
553 async->work.flags = 0;
554 async->bio_flags = bio_flags;
556 atomic_inc(&fs_info->nr_async_submits);
557 btrfs_queue_worker(&fs_info->workers, &async->work);
559 int limit = btrfs_async_submit_limit(fs_info);
560 if (atomic_read(&fs_info->nr_async_submits) > limit) {
561 wait_event_timeout(fs_info->async_submit_wait,
562 (atomic_read(&fs_info->nr_async_submits) < limit),
565 wait_event_timeout(fs_info->async_submit_wait,
566 (atomic_read(&fs_info->nr_async_bios) < limit),
570 while(atomic_read(&fs_info->async_submit_draining) &&
571 atomic_read(&fs_info->nr_async_submits)) {
572 wait_event(fs_info->async_submit_wait,
573 (atomic_read(&fs_info->nr_async_submits) == 0));
579 static int btree_csum_one_bio(struct bio *bio)
581 struct bio_vec *bvec = bio->bi_io_vec;
583 struct btrfs_root *root;
585 WARN_ON(bio->bi_vcnt <= 0);
586 while(bio_index < bio->bi_vcnt) {
587 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
588 csum_dirty_buffer(root, bvec->bv_page);
595 static int __btree_submit_bio_start(struct inode *inode, int rw,
596 struct bio *bio, int mirror_num,
597 unsigned long bio_flags)
600 * when we're called for a write, we're already in the async
601 * submission context. Just jump into btrfs_map_bio
603 btree_csum_one_bio(bio);
607 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
608 int mirror_num, unsigned long bio_flags)
611 * when we're called for a write, we're already in the async
612 * submission context. Just jump into btrfs_map_bio
614 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
617 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
618 int mirror_num, unsigned long bio_flags)
621 * kthread helpers are used to submit writes so that checksumming
622 * can happen in parallel across all CPUs
624 if (!(rw & (1 << BIO_RW))) {
627 * called for a read, do the setup so that checksum validation
628 * can happen in the async kernel threads
630 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
634 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
637 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
638 inode, rw, bio, mirror_num, 0,
639 __btree_submit_bio_start,
640 __btree_submit_bio_done);
643 static int btree_writepage(struct page *page, struct writeback_control *wbc)
645 struct extent_io_tree *tree;
646 tree = &BTRFS_I(page->mapping->host)->io_tree;
648 if (current->flags & PF_MEMALLOC) {
649 redirty_page_for_writepage(wbc, page);
653 return extent_write_full_page(tree, page, btree_get_extent, wbc);
656 static int btree_writepages(struct address_space *mapping,
657 struct writeback_control *wbc)
659 struct extent_io_tree *tree;
660 tree = &BTRFS_I(mapping->host)->io_tree;
661 if (wbc->sync_mode == WB_SYNC_NONE) {
664 unsigned long thresh = 32 * 1024 * 1024;
666 if (wbc->for_kupdate)
669 num_dirty = count_range_bits(tree, &start, (u64)-1,
670 thresh, EXTENT_DIRTY);
671 if (num_dirty < thresh) {
675 return extent_writepages(tree, mapping, btree_get_extent, wbc);
678 static int btree_readpage(struct file *file, struct page *page)
680 struct extent_io_tree *tree;
681 tree = &BTRFS_I(page->mapping->host)->io_tree;
682 return extent_read_full_page(tree, page, btree_get_extent);
685 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
687 struct extent_io_tree *tree;
688 struct extent_map_tree *map;
691 if (PageWriteback(page) || PageDirty(page))
694 tree = &BTRFS_I(page->mapping->host)->io_tree;
695 map = &BTRFS_I(page->mapping->host)->extent_tree;
697 ret = try_release_extent_state(map, tree, page, gfp_flags);
702 ret = try_release_extent_buffer(tree, page);
704 ClearPagePrivate(page);
705 set_page_private(page, 0);
706 page_cache_release(page);
712 static void btree_invalidatepage(struct page *page, unsigned long offset)
714 struct extent_io_tree *tree;
715 tree = &BTRFS_I(page->mapping->host)->io_tree;
716 extent_invalidatepage(tree, page, offset);
717 btree_releasepage(page, GFP_NOFS);
718 if (PagePrivate(page)) {
719 printk("warning page private not zero on page %Lu\n",
721 ClearPagePrivate(page);
722 set_page_private(page, 0);
723 page_cache_release(page);
728 static int btree_writepage(struct page *page, struct writeback_control *wbc)
730 struct buffer_head *bh;
731 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
732 struct buffer_head *head;
733 if (!page_has_buffers(page)) {
734 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
735 (1 << BH_Dirty)|(1 << BH_Uptodate));
737 head = page_buffers(page);
740 if (buffer_dirty(bh))
741 csum_tree_block(root, bh, 0);
742 bh = bh->b_this_page;
743 } while (bh != head);
744 return block_write_full_page(page, btree_get_block, wbc);
748 static struct address_space_operations btree_aops = {
749 .readpage = btree_readpage,
750 .writepage = btree_writepage,
751 .writepages = btree_writepages,
752 .releasepage = btree_releasepage,
753 .invalidatepage = btree_invalidatepage,
754 .sync_page = block_sync_page,
757 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
760 struct extent_buffer *buf = NULL;
761 struct inode *btree_inode = root->fs_info->btree_inode;
764 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
767 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
768 buf, 0, 0, btree_get_extent, 0);
769 free_extent_buffer(buf);
773 struct extent_buffer *btrfs_find_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;
778 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
779 bytenr, blocksize, GFP_NOFS);
783 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
784 u64 bytenr, u32 blocksize)
786 struct inode *btree_inode = root->fs_info->btree_inode;
787 struct extent_buffer *eb;
789 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
790 bytenr, blocksize, NULL, GFP_NOFS);
795 int btrfs_write_tree_block(struct extent_buffer *buf)
797 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
798 buf->start + buf->len - 1, WB_SYNC_ALL);
801 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
803 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
804 buf->start, buf->start + buf->len -1);
807 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
808 u32 blocksize, u64 parent_transid)
810 struct extent_buffer *buf = NULL;
811 struct inode *btree_inode = root->fs_info->btree_inode;
812 struct extent_io_tree *io_tree;
815 io_tree = &BTRFS_I(btree_inode)->io_tree;
817 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
821 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
824 buf->flags |= EXTENT_UPTODATE;
832 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
833 struct extent_buffer *buf)
835 struct inode *btree_inode = root->fs_info->btree_inode;
836 if (btrfs_header_generation(buf) ==
837 root->fs_info->running_transaction->transid) {
838 WARN_ON(!btrfs_tree_locked(buf));
839 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
845 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
846 u32 stripesize, struct btrfs_root *root,
847 struct btrfs_fs_info *fs_info,
851 root->commit_root = NULL;
852 root->ref_tree = NULL;
853 root->sectorsize = sectorsize;
854 root->nodesize = nodesize;
855 root->leafsize = leafsize;
856 root->stripesize = stripesize;
858 root->track_dirty = 0;
860 root->fs_info = fs_info;
861 root->objectid = objectid;
862 root->last_trans = 0;
863 root->highest_inode = 0;
864 root->last_inode_alloc = 0;
868 INIT_LIST_HEAD(&root->dirty_list);
869 INIT_LIST_HEAD(&root->orphan_list);
870 INIT_LIST_HEAD(&root->dead_list);
871 spin_lock_init(&root->node_lock);
872 spin_lock_init(&root->list_lock);
873 mutex_init(&root->objectid_mutex);
874 mutex_init(&root->log_mutex);
875 extent_io_tree_init(&root->dirty_log_pages,
876 fs_info->btree_inode->i_mapping, GFP_NOFS);
878 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
879 root->ref_tree = &root->ref_tree_struct;
881 memset(&root->root_key, 0, sizeof(root->root_key));
882 memset(&root->root_item, 0, sizeof(root->root_item));
883 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
884 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
885 root->defrag_trans_start = fs_info->generation;
886 init_completion(&root->kobj_unregister);
887 root->defrag_running = 0;
888 root->defrag_level = 0;
889 root->root_key.objectid = objectid;
890 root->anon_super.s_root = NULL;
891 root->anon_super.s_dev = 0;
892 INIT_LIST_HEAD(&root->anon_super.s_list);
893 INIT_LIST_HEAD(&root->anon_super.s_instances);
894 init_rwsem(&root->anon_super.s_umount);
899 static int find_and_setup_root(struct btrfs_root *tree_root,
900 struct btrfs_fs_info *fs_info,
902 struct btrfs_root *root)
908 __setup_root(tree_root->nodesize, tree_root->leafsize,
909 tree_root->sectorsize, tree_root->stripesize,
910 root, fs_info, objectid);
911 ret = btrfs_find_last_root(tree_root, objectid,
912 &root->root_item, &root->root_key);
915 generation = btrfs_root_generation(&root->root_item);
916 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
917 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
918 blocksize, generation);
923 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
924 struct btrfs_fs_info *fs_info)
926 struct extent_buffer *eb;
927 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
936 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
937 0, &start, &end, EXTENT_DIRTY);
941 clear_extent_dirty(&log_root_tree->dirty_log_pages,
942 start, end, GFP_NOFS);
944 eb = fs_info->log_root_tree->node;
946 WARN_ON(btrfs_header_level(eb) != 0);
947 WARN_ON(btrfs_header_nritems(eb) != 0);
949 ret = btrfs_free_reserved_extent(fs_info->tree_root,
953 free_extent_buffer(eb);
954 kfree(fs_info->log_root_tree);
955 fs_info->log_root_tree = NULL;
959 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
960 struct btrfs_fs_info *fs_info)
962 struct btrfs_root *root;
963 struct btrfs_root *tree_root = fs_info->tree_root;
965 root = kzalloc(sizeof(*root), GFP_NOFS);
969 __setup_root(tree_root->nodesize, tree_root->leafsize,
970 tree_root->sectorsize, tree_root->stripesize,
971 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
973 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
974 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
975 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
978 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
979 0, BTRFS_TREE_LOG_OBJECTID,
980 trans->transid, 0, 0, 0);
982 btrfs_set_header_nritems(root->node, 0);
983 btrfs_set_header_level(root->node, 0);
984 btrfs_set_header_bytenr(root->node, root->node->start);
985 btrfs_set_header_generation(root->node, trans->transid);
986 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
988 write_extent_buffer(root->node, root->fs_info->fsid,
989 (unsigned long)btrfs_header_fsid(root->node),
991 btrfs_mark_buffer_dirty(root->node);
992 btrfs_tree_unlock(root->node);
993 fs_info->log_root_tree = root;
997 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
998 struct btrfs_key *location)
1000 struct btrfs_root *root;
1001 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1002 struct btrfs_path *path;
1003 struct extent_buffer *l;
1009 root = kzalloc(sizeof(*root), GFP_NOFS);
1011 return ERR_PTR(-ENOMEM);
1012 if (location->offset == (u64)-1) {
1013 ret = find_and_setup_root(tree_root, fs_info,
1014 location->objectid, root);
1017 return ERR_PTR(ret);
1022 __setup_root(tree_root->nodesize, tree_root->leafsize,
1023 tree_root->sectorsize, tree_root->stripesize,
1024 root, fs_info, location->objectid);
1026 path = btrfs_alloc_path();
1028 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1035 read_extent_buffer(l, &root->root_item,
1036 btrfs_item_ptr_offset(l, path->slots[0]),
1037 sizeof(root->root_item));
1038 memcpy(&root->root_key, location, sizeof(*location));
1041 btrfs_release_path(root, path);
1042 btrfs_free_path(path);
1045 return ERR_PTR(ret);
1047 generation = btrfs_root_generation(&root->root_item);
1048 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1049 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1050 blocksize, generation);
1051 BUG_ON(!root->node);
1053 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1055 ret = btrfs_find_highest_inode(root, &highest_inode);
1057 root->highest_inode = highest_inode;
1058 root->last_inode_alloc = highest_inode;
1064 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1067 struct btrfs_root *root;
1069 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1070 return fs_info->tree_root;
1071 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1072 return fs_info->extent_root;
1074 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1075 (unsigned long)root_objectid);
1079 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1080 struct btrfs_key *location)
1082 struct btrfs_root *root;
1085 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1086 return fs_info->tree_root;
1087 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1088 return fs_info->extent_root;
1089 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1090 return fs_info->chunk_root;
1091 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1092 return fs_info->dev_root;
1094 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1095 (unsigned long)location->objectid);
1099 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1103 set_anon_super(&root->anon_super, NULL);
1105 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1106 (unsigned long)root->root_key.objectid,
1109 free_extent_buffer(root->node);
1111 return ERR_PTR(ret);
1113 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1114 ret = btrfs_find_dead_roots(fs_info->tree_root,
1115 root->root_key.objectid, root);
1117 btrfs_orphan_cleanup(root);
1122 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1123 struct btrfs_key *location,
1124 const char *name, int namelen)
1126 struct btrfs_root *root;
1129 root = btrfs_read_fs_root_no_name(fs_info, location);
1136 ret = btrfs_set_root_name(root, name, namelen);
1138 free_extent_buffer(root->node);
1140 return ERR_PTR(ret);
1143 ret = btrfs_sysfs_add_root(root);
1145 free_extent_buffer(root->node);
1148 return ERR_PTR(ret);
1155 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1156 struct btrfs_hasher *hasher;
1158 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1161 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1162 if (!hasher->hash_tfm) {
1166 spin_lock(&info->hash_lock);
1167 list_add(&hasher->list, &info->hashers);
1168 spin_unlock(&info->hash_lock);
1173 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1175 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1177 struct list_head *cur;
1178 struct btrfs_device *device;
1179 struct backing_dev_info *bdi;
1181 if ((bdi_bits & (1 << BDI_write_congested)) &&
1182 btrfs_congested_async(info, 0))
1185 list_for_each(cur, &info->fs_devices->devices) {
1186 device = list_entry(cur, struct btrfs_device, dev_list);
1189 bdi = blk_get_backing_dev_info(device->bdev);
1190 if (bdi && bdi_congested(bdi, bdi_bits)) {
1199 * this unplugs every device on the box, and it is only used when page
1202 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1204 struct list_head *cur;
1205 struct btrfs_device *device;
1206 struct btrfs_fs_info *info;
1208 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1209 list_for_each(cur, &info->fs_devices->devices) {
1210 device = list_entry(cur, struct btrfs_device, dev_list);
1211 bdi = blk_get_backing_dev_info(device->bdev);
1212 if (bdi->unplug_io_fn) {
1213 bdi->unplug_io_fn(bdi, page);
1218 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1220 struct inode *inode;
1221 struct extent_map_tree *em_tree;
1222 struct extent_map *em;
1223 struct address_space *mapping;
1226 /* the generic O_DIRECT read code does this */
1228 __unplug_io_fn(bdi, page);
1233 * page->mapping may change at any time. Get a consistent copy
1234 * and use that for everything below
1237 mapping = page->mapping;
1241 inode = mapping->host;
1244 * don't do the expensive searching for a small number of
1247 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1248 __unplug_io_fn(bdi, page);
1252 offset = page_offset(page);
1254 em_tree = &BTRFS_I(inode)->extent_tree;
1255 spin_lock(&em_tree->lock);
1256 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1257 spin_unlock(&em_tree->lock);
1259 __unplug_io_fn(bdi, page);
1263 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1264 free_extent_map(em);
1265 __unplug_io_fn(bdi, page);
1268 offset = offset - em->start;
1269 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1270 em->block_start + offset, page);
1271 free_extent_map(em);
1274 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1277 bdi->ra_pages = default_backing_dev_info.ra_pages;
1279 bdi->capabilities = default_backing_dev_info.capabilities;
1280 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1281 bdi->unplug_io_data = info;
1282 bdi->congested_fn = btrfs_congested_fn;
1283 bdi->congested_data = info;
1287 static int bio_ready_for_csum(struct bio *bio)
1293 struct extent_io_tree *io_tree = NULL;
1294 struct btrfs_fs_info *info = NULL;
1295 struct bio_vec *bvec;
1299 bio_for_each_segment(bvec, bio, i) {
1300 page = bvec->bv_page;
1301 if (page->private == EXTENT_PAGE_PRIVATE) {
1302 length += bvec->bv_len;
1305 if (!page->private) {
1306 length += bvec->bv_len;
1309 length = bvec->bv_len;
1310 buf_len = page->private >> 2;
1311 start = page_offset(page) + bvec->bv_offset;
1312 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1313 info = BTRFS_I(page->mapping->host)->root->fs_info;
1315 /* are we fully contained in this bio? */
1316 if (buf_len <= length)
1319 ret = extent_range_uptodate(io_tree, start + length,
1320 start + buf_len - 1);
1327 * called by the kthread helper functions to finally call the bio end_io
1328 * functions. This is where read checksum verification actually happens
1330 static void end_workqueue_fn(struct btrfs_work *work)
1333 struct end_io_wq *end_io_wq;
1334 struct btrfs_fs_info *fs_info;
1337 end_io_wq = container_of(work, struct end_io_wq, work);
1338 bio = end_io_wq->bio;
1339 fs_info = end_io_wq->info;
1341 /* metadata bios are special because the whole tree block must
1342 * be checksummed at once. This makes sure the entire block is in
1343 * ram and up to date before trying to verify things. For
1344 * blocksize <= pagesize, it is basically a noop
1346 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1347 btrfs_queue_worker(&fs_info->endio_workers,
1351 error = end_io_wq->error;
1352 bio->bi_private = end_io_wq->private;
1353 bio->bi_end_io = end_io_wq->end_io;
1355 bio_endio(bio, error);
1358 static int cleaner_kthread(void *arg)
1360 struct btrfs_root *root = arg;
1364 if (root->fs_info->closing)
1367 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1368 mutex_lock(&root->fs_info->cleaner_mutex);
1369 btrfs_clean_old_snapshots(root);
1370 mutex_unlock(&root->fs_info->cleaner_mutex);
1372 if (freezing(current)) {
1376 if (root->fs_info->closing)
1378 set_current_state(TASK_INTERRUPTIBLE);
1380 __set_current_state(TASK_RUNNING);
1382 } while (!kthread_should_stop());
1386 static int transaction_kthread(void *arg)
1388 struct btrfs_root *root = arg;
1389 struct btrfs_trans_handle *trans;
1390 struct btrfs_transaction *cur;
1392 unsigned long delay;
1397 if (root->fs_info->closing)
1401 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1402 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1404 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1405 printk("btrfs: total reference cache size %Lu\n",
1406 root->fs_info->total_ref_cache_size);
1409 mutex_lock(&root->fs_info->trans_mutex);
1410 cur = root->fs_info->running_transaction;
1412 mutex_unlock(&root->fs_info->trans_mutex);
1416 now = get_seconds();
1417 if (now < cur->start_time || now - cur->start_time < 30) {
1418 mutex_unlock(&root->fs_info->trans_mutex);
1422 mutex_unlock(&root->fs_info->trans_mutex);
1423 trans = btrfs_start_transaction(root, 1);
1424 ret = btrfs_commit_transaction(trans, root);
1426 wake_up_process(root->fs_info->cleaner_kthread);
1427 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1429 if (freezing(current)) {
1432 if (root->fs_info->closing)
1434 set_current_state(TASK_INTERRUPTIBLE);
1435 schedule_timeout(delay);
1436 __set_current_state(TASK_RUNNING);
1438 } while (!kthread_should_stop());
1442 struct btrfs_root *open_ctree(struct super_block *sb,
1443 struct btrfs_fs_devices *fs_devices,
1453 struct btrfs_key location;
1454 struct buffer_head *bh;
1455 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1457 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1459 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1461 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1463 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1465 struct btrfs_root *log_tree_root;
1470 struct btrfs_super_block *disk_super;
1472 if (!extent_root || !tree_root || !fs_info ||
1473 !chunk_root || !dev_root) {
1477 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1478 INIT_LIST_HEAD(&fs_info->trans_list);
1479 INIT_LIST_HEAD(&fs_info->dead_roots);
1480 INIT_LIST_HEAD(&fs_info->hashers);
1481 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1482 spin_lock_init(&fs_info->hash_lock);
1483 spin_lock_init(&fs_info->delalloc_lock);
1484 spin_lock_init(&fs_info->new_trans_lock);
1485 spin_lock_init(&fs_info->ref_cache_lock);
1487 init_completion(&fs_info->kobj_unregister);
1488 fs_info->tree_root = tree_root;
1489 fs_info->extent_root = extent_root;
1490 fs_info->chunk_root = chunk_root;
1491 fs_info->dev_root = dev_root;
1492 fs_info->fs_devices = fs_devices;
1493 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1494 INIT_LIST_HEAD(&fs_info->space_info);
1495 btrfs_mapping_init(&fs_info->mapping_tree);
1496 atomic_set(&fs_info->nr_async_submits, 0);
1497 atomic_set(&fs_info->async_delalloc_pages, 0);
1498 atomic_set(&fs_info->async_submit_draining, 0);
1499 atomic_set(&fs_info->nr_async_bios, 0);
1500 atomic_set(&fs_info->throttles, 0);
1501 atomic_set(&fs_info->throttle_gen, 0);
1503 fs_info->max_extent = (u64)-1;
1504 fs_info->max_inline = 8192 * 1024;
1505 setup_bdi(fs_info, &fs_info->bdi);
1506 fs_info->btree_inode = new_inode(sb);
1507 fs_info->btree_inode->i_ino = 1;
1508 fs_info->btree_inode->i_nlink = 1;
1510 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1512 INIT_LIST_HEAD(&fs_info->ordered_extents);
1513 spin_lock_init(&fs_info->ordered_extent_lock);
1515 sb->s_blocksize = 4096;
1516 sb->s_blocksize_bits = blksize_bits(4096);
1519 * we set the i_size on the btree inode to the max possible int.
1520 * the real end of the address space is determined by all of
1521 * the devices in the system
1523 fs_info->btree_inode->i_size = OFFSET_MAX;
1524 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1525 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1527 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1528 fs_info->btree_inode->i_mapping,
1530 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1533 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1535 spin_lock_init(&fs_info->block_group_cache_lock);
1536 fs_info->block_group_cache_tree.rb_node = NULL;
1538 extent_io_tree_init(&fs_info->pinned_extents,
1539 fs_info->btree_inode->i_mapping, GFP_NOFS);
1540 extent_io_tree_init(&fs_info->pending_del,
1541 fs_info->btree_inode->i_mapping, GFP_NOFS);
1542 extent_io_tree_init(&fs_info->extent_ins,
1543 fs_info->btree_inode->i_mapping, GFP_NOFS);
1544 fs_info->do_barriers = 1;
1546 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1547 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1548 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1550 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1551 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1552 sizeof(struct btrfs_key));
1553 insert_inode_hash(fs_info->btree_inode);
1555 mutex_init(&fs_info->trans_mutex);
1556 mutex_init(&fs_info->tree_log_mutex);
1557 mutex_init(&fs_info->drop_mutex);
1558 mutex_init(&fs_info->extent_ins_mutex);
1559 mutex_init(&fs_info->pinned_mutex);
1560 mutex_init(&fs_info->chunk_mutex);
1561 mutex_init(&fs_info->transaction_kthread_mutex);
1562 mutex_init(&fs_info->cleaner_mutex);
1563 mutex_init(&fs_info->volume_mutex);
1564 mutex_init(&fs_info->tree_reloc_mutex);
1565 init_waitqueue_head(&fs_info->transaction_throttle);
1566 init_waitqueue_head(&fs_info->transaction_wait);
1567 init_waitqueue_head(&fs_info->async_submit_wait);
1568 init_waitqueue_head(&fs_info->tree_log_wait);
1569 atomic_set(&fs_info->tree_log_commit, 0);
1570 atomic_set(&fs_info->tree_log_writers, 0);
1571 fs_info->tree_log_transid = 0;
1574 ret = add_hasher(fs_info, "crc32c");
1576 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1581 __setup_root(4096, 4096, 4096, 4096, tree_root,
1582 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1585 bh = __bread(fs_devices->latest_bdev,
1586 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1590 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1593 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1595 disk_super = &fs_info->super_copy;
1596 if (!btrfs_super_root(disk_super))
1599 ret = btrfs_parse_options(tree_root, options);
1605 features = btrfs_super_incompat_flags(disk_super) &
1606 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1608 printk(KERN_ERR "BTRFS: couldn't mount because of "
1609 "unsupported optional features (%Lx).\n",
1615 features = btrfs_super_compat_ro_flags(disk_super) &
1616 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1617 if (!(sb->s_flags & MS_RDONLY) && features) {
1618 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1619 "unsupported option features (%Lx).\n",
1626 * we need to start all the end_io workers up front because the
1627 * queue work function gets called at interrupt time, and so it
1628 * cannot dynamically grow.
1630 btrfs_init_workers(&fs_info->workers, "worker",
1631 fs_info->thread_pool_size);
1633 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1634 fs_info->thread_pool_size);
1636 btrfs_init_workers(&fs_info->submit_workers, "submit",
1637 min_t(u64, fs_devices->num_devices,
1638 fs_info->thread_pool_size));
1640 /* a higher idle thresh on the submit workers makes it much more
1641 * likely that bios will be send down in a sane order to the
1644 fs_info->submit_workers.idle_thresh = 64;
1646 fs_info->workers.idle_thresh = 16;
1647 fs_info->workers.ordered = 1;
1649 fs_info->delalloc_workers.idle_thresh = 2;
1650 fs_info->delalloc_workers.ordered = 1;
1652 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1653 btrfs_init_workers(&fs_info->endio_workers, "endio",
1654 fs_info->thread_pool_size);
1655 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1656 fs_info->thread_pool_size);
1659 * endios are largely parallel and should have a very
1662 fs_info->endio_workers.idle_thresh = 4;
1663 fs_info->endio_write_workers.idle_thresh = 64;
1665 btrfs_start_workers(&fs_info->workers, 1);
1666 btrfs_start_workers(&fs_info->submit_workers, 1);
1667 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1668 btrfs_start_workers(&fs_info->fixup_workers, 1);
1669 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1670 btrfs_start_workers(&fs_info->endio_write_workers,
1671 fs_info->thread_pool_size);
1673 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1674 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1675 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1677 nodesize = btrfs_super_nodesize(disk_super);
1678 leafsize = btrfs_super_leafsize(disk_super);
1679 sectorsize = btrfs_super_sectorsize(disk_super);
1680 stripesize = btrfs_super_stripesize(disk_super);
1681 tree_root->nodesize = nodesize;
1682 tree_root->leafsize = leafsize;
1683 tree_root->sectorsize = sectorsize;
1684 tree_root->stripesize = stripesize;
1686 sb->s_blocksize = sectorsize;
1687 sb->s_blocksize_bits = blksize_bits(sectorsize);
1689 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1690 sizeof(disk_super->magic))) {
1691 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1692 goto fail_sb_buffer;
1695 mutex_lock(&fs_info->chunk_mutex);
1696 ret = btrfs_read_sys_array(tree_root);
1697 mutex_unlock(&fs_info->chunk_mutex);
1699 printk("btrfs: failed to read the system array on %s\n",
1701 goto fail_sys_array;
1704 blocksize = btrfs_level_size(tree_root,
1705 btrfs_super_chunk_root_level(disk_super));
1706 generation = btrfs_super_chunk_root_generation(disk_super);
1708 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1709 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1711 chunk_root->node = read_tree_block(chunk_root,
1712 btrfs_super_chunk_root(disk_super),
1713 blocksize, generation);
1714 BUG_ON(!chunk_root->node);
1716 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1717 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1720 mutex_lock(&fs_info->chunk_mutex);
1721 ret = btrfs_read_chunk_tree(chunk_root);
1722 mutex_unlock(&fs_info->chunk_mutex);
1724 printk("btrfs: failed to read chunk tree on %s\n", sb->s_id);
1725 goto fail_chunk_root;
1728 btrfs_close_extra_devices(fs_devices);
1730 blocksize = btrfs_level_size(tree_root,
1731 btrfs_super_root_level(disk_super));
1732 generation = btrfs_super_generation(disk_super);
1734 tree_root->node = read_tree_block(tree_root,
1735 btrfs_super_root(disk_super),
1736 blocksize, generation);
1737 if (!tree_root->node)
1738 goto fail_chunk_root;
1741 ret = find_and_setup_root(tree_root, fs_info,
1742 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1744 goto fail_tree_root;
1745 extent_root->track_dirty = 1;
1747 ret = find_and_setup_root(tree_root, fs_info,
1748 BTRFS_DEV_TREE_OBJECTID, dev_root);
1749 dev_root->track_dirty = 1;
1752 goto fail_extent_root;
1754 btrfs_read_block_groups(extent_root);
1756 fs_info->generation = generation + 1;
1757 fs_info->last_trans_committed = generation;
1758 fs_info->data_alloc_profile = (u64)-1;
1759 fs_info->metadata_alloc_profile = (u64)-1;
1760 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1761 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1763 if (!fs_info->cleaner_kthread)
1764 goto fail_extent_root;
1766 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1768 "btrfs-transaction");
1769 if (!fs_info->transaction_kthread)
1772 if (btrfs_super_log_root(disk_super) != 0) {
1773 u64 bytenr = btrfs_super_log_root(disk_super);
1775 if (fs_devices->rw_devices == 0) {
1776 printk("Btrfs log replay required on RO media\n");
1778 goto fail_trans_kthread;
1781 btrfs_level_size(tree_root,
1782 btrfs_super_log_root_level(disk_super));
1784 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1787 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1788 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1790 log_tree_root->node = read_tree_block(tree_root, bytenr,
1793 ret = btrfs_recover_log_trees(log_tree_root);
1796 if (sb->s_flags & MS_RDONLY) {
1797 ret = btrfs_commit_super(tree_root);
1802 if (!(sb->s_flags & MS_RDONLY)) {
1803 ret = btrfs_cleanup_reloc_trees(tree_root);
1807 location.objectid = BTRFS_FS_TREE_OBJECTID;
1808 location.type = BTRFS_ROOT_ITEM_KEY;
1809 location.offset = (u64)-1;
1811 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1812 if (!fs_info->fs_root)
1813 goto fail_trans_kthread;
1817 kthread_stop(fs_info->transaction_kthread);
1819 kthread_stop(fs_info->cleaner_kthread);
1822 * make sure we're done with the btree inode before we stop our
1825 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1826 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1829 free_extent_buffer(extent_root->node);
1831 free_extent_buffer(tree_root->node);
1833 free_extent_buffer(chunk_root->node);
1835 free_extent_buffer(dev_root->node);
1837 btrfs_stop_workers(&fs_info->fixup_workers);
1838 btrfs_stop_workers(&fs_info->delalloc_workers);
1839 btrfs_stop_workers(&fs_info->workers);
1840 btrfs_stop_workers(&fs_info->endio_workers);
1841 btrfs_stop_workers(&fs_info->endio_write_workers);
1842 btrfs_stop_workers(&fs_info->submit_workers);
1844 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1845 iput(fs_info->btree_inode);
1847 btrfs_close_devices(fs_info->fs_devices);
1848 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1852 bdi_destroy(&fs_info->bdi);
1856 return ERR_PTR(err);
1859 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1861 char b[BDEVNAME_SIZE];
1864 set_buffer_uptodate(bh);
1866 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1867 printk(KERN_WARNING "lost page write due to "
1868 "I/O error on %s\n",
1869 bdevname(bh->b_bdev, b));
1871 /* note, we dont' set_buffer_write_io_error because we have
1872 * our own ways of dealing with the IO errors
1874 clear_buffer_uptodate(bh);
1880 static int write_all_supers(struct btrfs_root *root)
1882 struct list_head *cur;
1883 struct list_head *head = &root->fs_info->fs_devices->devices;
1884 struct btrfs_device *dev;
1885 struct btrfs_super_block *sb;
1886 struct btrfs_dev_item *dev_item;
1887 struct buffer_head *bh;
1891 int total_errors = 0;
1895 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1896 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1898 sb = &root->fs_info->super_for_commit;
1899 dev_item = &sb->dev_item;
1900 list_for_each(cur, head) {
1901 dev = list_entry(cur, struct btrfs_device, dev_list);
1906 if (!dev->in_fs_metadata || !dev->writeable)
1909 btrfs_set_stack_device_generation(dev_item, 0);
1910 btrfs_set_stack_device_type(dev_item, dev->type);
1911 btrfs_set_stack_device_id(dev_item, dev->devid);
1912 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1913 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1914 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1915 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1916 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1917 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1918 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1919 flags = btrfs_super_flags(sb);
1920 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1924 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1925 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1926 btrfs_csum_final(crc, sb->csum);
1928 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1929 BTRFS_SUPER_INFO_SIZE);
1931 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1932 dev->pending_io = bh;
1935 set_buffer_uptodate(bh);
1937 bh->b_end_io = btrfs_end_buffer_write_sync;
1939 if (do_barriers && dev->barriers) {
1940 ret = submit_bh(WRITE_BARRIER, bh);
1941 if (ret == -EOPNOTSUPP) {
1942 printk("btrfs: disabling barriers on dev %s\n",
1944 set_buffer_uptodate(bh);
1948 ret = submit_bh(WRITE, bh);
1951 ret = submit_bh(WRITE, bh);
1956 if (total_errors > max_errors) {
1957 printk("btrfs: %d errors while writing supers\n", total_errors);
1962 list_for_each(cur, head) {
1963 dev = list_entry(cur, struct btrfs_device, dev_list);
1966 if (!dev->in_fs_metadata || !dev->writeable)
1969 BUG_ON(!dev->pending_io);
1970 bh = dev->pending_io;
1972 if (!buffer_uptodate(dev->pending_io)) {
1973 if (do_barriers && dev->barriers) {
1974 printk("btrfs: disabling barriers on dev %s\n",
1976 set_buffer_uptodate(bh);
1980 ret = submit_bh(WRITE, bh);
1983 if (!buffer_uptodate(bh))
1990 dev->pending_io = NULL;
1993 if (total_errors > max_errors) {
1994 printk("btrfs: %d errors while writing supers\n", total_errors);
2000 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
2005 ret = write_all_supers(root);
2009 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2011 radix_tree_delete(&fs_info->fs_roots_radix,
2012 (unsigned long)root->root_key.objectid);
2013 if (root->anon_super.s_dev) {
2014 down_write(&root->anon_super.s_umount);
2015 kill_anon_super(&root->anon_super);
2019 btrfs_sysfs_del_root(root);
2022 free_extent_buffer(root->node);
2023 if (root->commit_root)
2024 free_extent_buffer(root->commit_root);
2031 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2034 struct btrfs_root *gang[8];
2038 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2043 for (i = 0; i < ret; i++)
2044 btrfs_free_fs_root(fs_info, gang[i]);
2049 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2051 u64 root_objectid = 0;
2052 struct btrfs_root *gang[8];
2057 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2058 (void **)gang, root_objectid,
2062 for (i = 0; i < ret; i++) {
2063 root_objectid = gang[i]->root_key.objectid;
2064 ret = btrfs_find_dead_roots(fs_info->tree_root,
2065 root_objectid, gang[i]);
2067 btrfs_orphan_cleanup(gang[i]);
2074 int btrfs_commit_super(struct btrfs_root *root)
2076 struct btrfs_trans_handle *trans;
2079 mutex_lock(&root->fs_info->cleaner_mutex);
2080 btrfs_clean_old_snapshots(root);
2081 mutex_unlock(&root->fs_info->cleaner_mutex);
2082 trans = btrfs_start_transaction(root, 1);
2083 ret = btrfs_commit_transaction(trans, root);
2085 /* run commit again to drop the original snapshot */
2086 trans = btrfs_start_transaction(root, 1);
2087 btrfs_commit_transaction(trans, root);
2088 ret = btrfs_write_and_wait_transaction(NULL, root);
2091 ret = write_ctree_super(NULL, root);
2095 int close_ctree(struct btrfs_root *root)
2097 struct btrfs_fs_info *fs_info = root->fs_info;
2100 fs_info->closing = 1;
2103 kthread_stop(root->fs_info->transaction_kthread);
2104 kthread_stop(root->fs_info->cleaner_kthread);
2106 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2107 ret = btrfs_commit_super(root);
2109 printk("btrfs: commit super returns %d\n", ret);
2113 if (fs_info->delalloc_bytes) {
2114 printk("btrfs: at unmount delalloc count %Lu\n",
2115 fs_info->delalloc_bytes);
2117 if (fs_info->total_ref_cache_size) {
2118 printk("btrfs: at umount reference cache size %Lu\n",
2119 fs_info->total_ref_cache_size);
2122 if (fs_info->extent_root->node)
2123 free_extent_buffer(fs_info->extent_root->node);
2125 if (fs_info->tree_root->node)
2126 free_extent_buffer(fs_info->tree_root->node);
2128 if (root->fs_info->chunk_root->node);
2129 free_extent_buffer(root->fs_info->chunk_root->node);
2131 if (root->fs_info->dev_root->node);
2132 free_extent_buffer(root->fs_info->dev_root->node);
2134 btrfs_free_block_groups(root->fs_info);
2136 del_fs_roots(fs_info);
2138 iput(fs_info->btree_inode);
2140 btrfs_stop_workers(&fs_info->fixup_workers);
2141 btrfs_stop_workers(&fs_info->delalloc_workers);
2142 btrfs_stop_workers(&fs_info->workers);
2143 btrfs_stop_workers(&fs_info->endio_workers);
2144 btrfs_stop_workers(&fs_info->endio_write_workers);
2145 btrfs_stop_workers(&fs_info->submit_workers);
2148 while(!list_empty(&fs_info->hashers)) {
2149 struct btrfs_hasher *hasher;
2150 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2152 list_del(&hasher->hashers);
2153 crypto_free_hash(&fs_info->hash_tfm);
2157 btrfs_close_devices(fs_info->fs_devices);
2158 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2160 bdi_destroy(&fs_info->bdi);
2162 kfree(fs_info->extent_root);
2163 kfree(fs_info->tree_root);
2164 kfree(fs_info->chunk_root);
2165 kfree(fs_info->dev_root);
2169 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2172 struct inode *btree_inode = buf->first_page->mapping->host;
2174 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2178 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2183 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2185 struct inode *btree_inode = buf->first_page->mapping->host;
2186 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2190 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2192 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2193 u64 transid = btrfs_header_generation(buf);
2194 struct inode *btree_inode = root->fs_info->btree_inode;
2196 WARN_ON(!btrfs_tree_locked(buf));
2197 if (transid != root->fs_info->generation) {
2198 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2199 (unsigned long long)buf->start,
2200 transid, root->fs_info->generation);
2203 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2206 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2209 * looks as though older kernels can get into trouble with
2210 * this code, they end up stuck in balance_dirty_pages forever
2212 struct extent_io_tree *tree;
2215 unsigned long thresh = 32 * 1024 * 1024;
2216 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2218 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2221 num_dirty = count_range_bits(tree, &start, (u64)-1,
2222 thresh, EXTENT_DIRTY);
2223 if (num_dirty > thresh) {
2224 balance_dirty_pages_ratelimited_nr(
2225 root->fs_info->btree_inode->i_mapping, 1);
2230 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2232 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2234 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2236 buf->flags |= EXTENT_UPTODATE;
2241 int btree_lock_page_hook(struct page *page)
2243 struct inode *inode = page->mapping->host;
2244 struct btrfs_root *root = BTRFS_I(inode)->root;
2245 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2246 struct extent_buffer *eb;
2248 u64 bytenr = page_offset(page);
2250 if (page->private == EXTENT_PAGE_PRIVATE)
2253 len = page->private >> 2;
2254 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2258 btrfs_tree_lock(eb);
2259 spin_lock(&root->fs_info->hash_lock);
2260 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2261 spin_unlock(&root->fs_info->hash_lock);
2262 btrfs_tree_unlock(eb);
2263 free_extent_buffer(eb);
2269 static struct extent_io_ops btree_extent_io_ops = {
2270 .write_cache_pages_lock_hook = btree_lock_page_hook,
2271 .readpage_end_io_hook = btree_readpage_end_io_hook,
2272 .submit_bio_hook = btree_submit_bio_hook,
2273 /* note we're sharing with inode.c for the merge bio hook */
2274 .merge_bio_hook = btrfs_merge_bio_hook,