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;
449 if (bio->bi_rw & (1 << BIO_RW)) {
450 btrfs_queue_worker(&fs_info->endio_write_workers,
453 if (end_io_wq->metadata)
454 btrfs_queue_worker(&fs_info->endio_meta_workers,
457 btrfs_queue_worker(&fs_info->endio_workers,
462 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
465 struct end_io_wq *end_io_wq;
466 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
470 end_io_wq->private = bio->bi_private;
471 end_io_wq->end_io = bio->bi_end_io;
472 end_io_wq->info = info;
473 end_io_wq->error = 0;
474 end_io_wq->bio = bio;
475 end_io_wq->metadata = metadata;
477 bio->bi_private = end_io_wq;
478 bio->bi_end_io = end_workqueue_bio;
482 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
484 unsigned long limit = min_t(unsigned long,
485 info->workers.max_workers,
486 info->fs_devices->open_devices);
490 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
492 return atomic_read(&info->nr_async_bios) >
493 btrfs_async_submit_limit(info);
496 static void run_one_async_start(struct btrfs_work *work)
498 struct btrfs_fs_info *fs_info;
499 struct async_submit_bio *async;
501 async = container_of(work, struct async_submit_bio, work);
502 fs_info = BTRFS_I(async->inode)->root->fs_info;
503 async->submit_bio_start(async->inode, async->rw, async->bio,
504 async->mirror_num, async->bio_flags);
507 static void run_one_async_done(struct btrfs_work *work)
509 struct btrfs_fs_info *fs_info;
510 struct async_submit_bio *async;
513 async = container_of(work, struct async_submit_bio, work);
514 fs_info = BTRFS_I(async->inode)->root->fs_info;
516 limit = btrfs_async_submit_limit(fs_info);
517 limit = limit * 2 / 3;
519 atomic_dec(&fs_info->nr_async_submits);
521 if (atomic_read(&fs_info->nr_async_submits) < limit &&
522 waitqueue_active(&fs_info->async_submit_wait))
523 wake_up(&fs_info->async_submit_wait);
525 async->submit_bio_done(async->inode, async->rw, async->bio,
526 async->mirror_num, async->bio_flags);
529 static void run_one_async_free(struct btrfs_work *work)
531 struct async_submit_bio *async;
533 async = container_of(work, struct async_submit_bio, work);
537 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
538 int rw, struct bio *bio, int mirror_num,
539 unsigned long bio_flags,
540 extent_submit_bio_hook_t *submit_bio_start,
541 extent_submit_bio_hook_t *submit_bio_done)
543 struct async_submit_bio *async;
545 async = kmalloc(sizeof(*async), GFP_NOFS);
549 async->inode = inode;
552 async->mirror_num = mirror_num;
553 async->submit_bio_start = submit_bio_start;
554 async->submit_bio_done = submit_bio_done;
556 async->work.func = run_one_async_start;
557 async->work.ordered_func = run_one_async_done;
558 async->work.ordered_free = run_one_async_free;
560 async->work.flags = 0;
561 async->bio_flags = bio_flags;
563 atomic_inc(&fs_info->nr_async_submits);
564 btrfs_queue_worker(&fs_info->workers, &async->work);
566 int limit = btrfs_async_submit_limit(fs_info);
567 if (atomic_read(&fs_info->nr_async_submits) > limit) {
568 wait_event_timeout(fs_info->async_submit_wait,
569 (atomic_read(&fs_info->nr_async_submits) < limit),
572 wait_event_timeout(fs_info->async_submit_wait,
573 (atomic_read(&fs_info->nr_async_bios) < limit),
577 while(atomic_read(&fs_info->async_submit_draining) &&
578 atomic_read(&fs_info->nr_async_submits)) {
579 wait_event(fs_info->async_submit_wait,
580 (atomic_read(&fs_info->nr_async_submits) == 0));
586 static int btree_csum_one_bio(struct bio *bio)
588 struct bio_vec *bvec = bio->bi_io_vec;
590 struct btrfs_root *root;
592 WARN_ON(bio->bi_vcnt <= 0);
593 while(bio_index < bio->bi_vcnt) {
594 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
595 csum_dirty_buffer(root, bvec->bv_page);
602 static int __btree_submit_bio_start(struct inode *inode, int rw,
603 struct bio *bio, int mirror_num,
604 unsigned long bio_flags)
607 * when we're called for a write, we're already in the async
608 * submission context. Just jump into btrfs_map_bio
610 btree_csum_one_bio(bio);
614 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
615 int mirror_num, unsigned long bio_flags)
618 * when we're called for a write, we're already in the async
619 * submission context. Just jump into btrfs_map_bio
621 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
624 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
625 int mirror_num, unsigned long bio_flags)
628 * kthread helpers are used to submit writes so that checksumming
629 * can happen in parallel across all CPUs
631 if (!(rw & (1 << BIO_RW))) {
634 * called for a read, do the setup so that checksum validation
635 * can happen in the async kernel threads
637 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
641 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
644 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
645 inode, rw, bio, mirror_num, 0,
646 __btree_submit_bio_start,
647 __btree_submit_bio_done);
650 static int btree_writepage(struct page *page, struct writeback_control *wbc)
652 struct extent_io_tree *tree;
653 tree = &BTRFS_I(page->mapping->host)->io_tree;
655 if (current->flags & PF_MEMALLOC) {
656 redirty_page_for_writepage(wbc, page);
660 return extent_write_full_page(tree, page, btree_get_extent, wbc);
663 static int btree_writepages(struct address_space *mapping,
664 struct writeback_control *wbc)
666 struct extent_io_tree *tree;
667 tree = &BTRFS_I(mapping->host)->io_tree;
668 if (wbc->sync_mode == WB_SYNC_NONE) {
671 unsigned long thresh = 32 * 1024 * 1024;
673 if (wbc->for_kupdate)
676 num_dirty = count_range_bits(tree, &start, (u64)-1,
677 thresh, EXTENT_DIRTY);
678 if (num_dirty < thresh) {
682 return extent_writepages(tree, mapping, btree_get_extent, wbc);
685 static int btree_readpage(struct file *file, struct page *page)
687 struct extent_io_tree *tree;
688 tree = &BTRFS_I(page->mapping->host)->io_tree;
689 return extent_read_full_page(tree, page, btree_get_extent);
692 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
694 struct extent_io_tree *tree;
695 struct extent_map_tree *map;
698 if (PageWriteback(page) || PageDirty(page))
701 tree = &BTRFS_I(page->mapping->host)->io_tree;
702 map = &BTRFS_I(page->mapping->host)->extent_tree;
704 ret = try_release_extent_state(map, tree, page, gfp_flags);
709 ret = try_release_extent_buffer(tree, page);
711 ClearPagePrivate(page);
712 set_page_private(page, 0);
713 page_cache_release(page);
719 static void btree_invalidatepage(struct page *page, unsigned long offset)
721 struct extent_io_tree *tree;
722 tree = &BTRFS_I(page->mapping->host)->io_tree;
723 extent_invalidatepage(tree, page, offset);
724 btree_releasepage(page, GFP_NOFS);
725 if (PagePrivate(page)) {
726 printk("warning page private not zero on page %Lu\n",
728 ClearPagePrivate(page);
729 set_page_private(page, 0);
730 page_cache_release(page);
735 static int btree_writepage(struct page *page, struct writeback_control *wbc)
737 struct buffer_head *bh;
738 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
739 struct buffer_head *head;
740 if (!page_has_buffers(page)) {
741 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
742 (1 << BH_Dirty)|(1 << BH_Uptodate));
744 head = page_buffers(page);
747 if (buffer_dirty(bh))
748 csum_tree_block(root, bh, 0);
749 bh = bh->b_this_page;
750 } while (bh != head);
751 return block_write_full_page(page, btree_get_block, wbc);
755 static struct address_space_operations btree_aops = {
756 .readpage = btree_readpage,
757 .writepage = btree_writepage,
758 .writepages = btree_writepages,
759 .releasepage = btree_releasepage,
760 .invalidatepage = btree_invalidatepage,
761 .sync_page = block_sync_page,
764 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
767 struct extent_buffer *buf = NULL;
768 struct inode *btree_inode = root->fs_info->btree_inode;
771 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
774 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
775 buf, 0, 0, btree_get_extent, 0);
776 free_extent_buffer(buf);
780 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
781 u64 bytenr, u32 blocksize)
783 struct inode *btree_inode = root->fs_info->btree_inode;
784 struct extent_buffer *eb;
785 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
786 bytenr, blocksize, GFP_NOFS);
790 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
791 u64 bytenr, u32 blocksize)
793 struct inode *btree_inode = root->fs_info->btree_inode;
794 struct extent_buffer *eb;
796 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
797 bytenr, blocksize, NULL, GFP_NOFS);
802 int btrfs_write_tree_block(struct extent_buffer *buf)
804 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
805 buf->start + buf->len - 1, WB_SYNC_ALL);
808 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
810 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
811 buf->start, buf->start + buf->len -1);
814 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
815 u32 blocksize, u64 parent_transid)
817 struct extent_buffer *buf = NULL;
818 struct inode *btree_inode = root->fs_info->btree_inode;
819 struct extent_io_tree *io_tree;
822 io_tree = &BTRFS_I(btree_inode)->io_tree;
824 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
828 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
831 buf->flags |= EXTENT_UPTODATE;
839 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
840 struct extent_buffer *buf)
842 struct inode *btree_inode = root->fs_info->btree_inode;
843 if (btrfs_header_generation(buf) ==
844 root->fs_info->running_transaction->transid) {
845 WARN_ON(!btrfs_tree_locked(buf));
846 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
852 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
853 u32 stripesize, struct btrfs_root *root,
854 struct btrfs_fs_info *fs_info,
858 root->commit_root = NULL;
859 root->ref_tree = NULL;
860 root->sectorsize = sectorsize;
861 root->nodesize = nodesize;
862 root->leafsize = leafsize;
863 root->stripesize = stripesize;
865 root->track_dirty = 0;
867 root->fs_info = fs_info;
868 root->objectid = objectid;
869 root->last_trans = 0;
870 root->highest_inode = 0;
871 root->last_inode_alloc = 0;
875 INIT_LIST_HEAD(&root->dirty_list);
876 INIT_LIST_HEAD(&root->orphan_list);
877 INIT_LIST_HEAD(&root->dead_list);
878 spin_lock_init(&root->node_lock);
879 spin_lock_init(&root->list_lock);
880 mutex_init(&root->objectid_mutex);
881 mutex_init(&root->log_mutex);
882 extent_io_tree_init(&root->dirty_log_pages,
883 fs_info->btree_inode->i_mapping, GFP_NOFS);
885 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
886 root->ref_tree = &root->ref_tree_struct;
888 memset(&root->root_key, 0, sizeof(root->root_key));
889 memset(&root->root_item, 0, sizeof(root->root_item));
890 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
891 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
892 root->defrag_trans_start = fs_info->generation;
893 init_completion(&root->kobj_unregister);
894 root->defrag_running = 0;
895 root->defrag_level = 0;
896 root->root_key.objectid = objectid;
897 root->anon_super.s_root = NULL;
898 root->anon_super.s_dev = 0;
899 INIT_LIST_HEAD(&root->anon_super.s_list);
900 INIT_LIST_HEAD(&root->anon_super.s_instances);
901 init_rwsem(&root->anon_super.s_umount);
906 static int find_and_setup_root(struct btrfs_root *tree_root,
907 struct btrfs_fs_info *fs_info,
909 struct btrfs_root *root)
915 __setup_root(tree_root->nodesize, tree_root->leafsize,
916 tree_root->sectorsize, tree_root->stripesize,
917 root, fs_info, objectid);
918 ret = btrfs_find_last_root(tree_root, objectid,
919 &root->root_item, &root->root_key);
922 generation = btrfs_root_generation(&root->root_item);
923 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
924 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
925 blocksize, generation);
930 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
931 struct btrfs_fs_info *fs_info)
933 struct extent_buffer *eb;
934 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
943 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
944 0, &start, &end, EXTENT_DIRTY);
948 clear_extent_dirty(&log_root_tree->dirty_log_pages,
949 start, end, GFP_NOFS);
951 eb = fs_info->log_root_tree->node;
953 WARN_ON(btrfs_header_level(eb) != 0);
954 WARN_ON(btrfs_header_nritems(eb) != 0);
956 ret = btrfs_free_reserved_extent(fs_info->tree_root,
960 free_extent_buffer(eb);
961 kfree(fs_info->log_root_tree);
962 fs_info->log_root_tree = NULL;
966 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
967 struct btrfs_fs_info *fs_info)
969 struct btrfs_root *root;
970 struct btrfs_root *tree_root = fs_info->tree_root;
972 root = kzalloc(sizeof(*root), GFP_NOFS);
976 __setup_root(tree_root->nodesize, tree_root->leafsize,
977 tree_root->sectorsize, tree_root->stripesize,
978 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
980 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
981 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
982 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
985 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
986 0, BTRFS_TREE_LOG_OBJECTID,
987 trans->transid, 0, 0, 0);
989 btrfs_set_header_nritems(root->node, 0);
990 btrfs_set_header_level(root->node, 0);
991 btrfs_set_header_bytenr(root->node, root->node->start);
992 btrfs_set_header_generation(root->node, trans->transid);
993 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
995 write_extent_buffer(root->node, root->fs_info->fsid,
996 (unsigned long)btrfs_header_fsid(root->node),
998 btrfs_mark_buffer_dirty(root->node);
999 btrfs_tree_unlock(root->node);
1000 fs_info->log_root_tree = root;
1004 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1005 struct btrfs_key *location)
1007 struct btrfs_root *root;
1008 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1009 struct btrfs_path *path;
1010 struct extent_buffer *l;
1016 root = kzalloc(sizeof(*root), GFP_NOFS);
1018 return ERR_PTR(-ENOMEM);
1019 if (location->offset == (u64)-1) {
1020 ret = find_and_setup_root(tree_root, fs_info,
1021 location->objectid, root);
1024 return ERR_PTR(ret);
1029 __setup_root(tree_root->nodesize, tree_root->leafsize,
1030 tree_root->sectorsize, tree_root->stripesize,
1031 root, fs_info, location->objectid);
1033 path = btrfs_alloc_path();
1035 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1042 read_extent_buffer(l, &root->root_item,
1043 btrfs_item_ptr_offset(l, path->slots[0]),
1044 sizeof(root->root_item));
1045 memcpy(&root->root_key, location, sizeof(*location));
1048 btrfs_release_path(root, path);
1049 btrfs_free_path(path);
1052 return ERR_PTR(ret);
1054 generation = btrfs_root_generation(&root->root_item);
1055 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1056 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1057 blocksize, generation);
1058 BUG_ON(!root->node);
1060 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1062 ret = btrfs_find_highest_inode(root, &highest_inode);
1064 root->highest_inode = highest_inode;
1065 root->last_inode_alloc = highest_inode;
1071 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1074 struct btrfs_root *root;
1076 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1077 return fs_info->tree_root;
1078 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1079 return fs_info->extent_root;
1081 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1082 (unsigned long)root_objectid);
1086 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1087 struct btrfs_key *location)
1089 struct btrfs_root *root;
1092 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1093 return fs_info->tree_root;
1094 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1095 return fs_info->extent_root;
1096 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1097 return fs_info->chunk_root;
1098 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1099 return fs_info->dev_root;
1101 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1102 (unsigned long)location->objectid);
1106 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1110 set_anon_super(&root->anon_super, NULL);
1112 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1113 (unsigned long)root->root_key.objectid,
1116 free_extent_buffer(root->node);
1118 return ERR_PTR(ret);
1120 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1121 ret = btrfs_find_dead_roots(fs_info->tree_root,
1122 root->root_key.objectid, root);
1124 btrfs_orphan_cleanup(root);
1129 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1130 struct btrfs_key *location,
1131 const char *name, int namelen)
1133 struct btrfs_root *root;
1136 root = btrfs_read_fs_root_no_name(fs_info, location);
1143 ret = btrfs_set_root_name(root, name, namelen);
1145 free_extent_buffer(root->node);
1147 return ERR_PTR(ret);
1150 ret = btrfs_sysfs_add_root(root);
1152 free_extent_buffer(root->node);
1155 return ERR_PTR(ret);
1162 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1163 struct btrfs_hasher *hasher;
1165 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1168 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1169 if (!hasher->hash_tfm) {
1173 spin_lock(&info->hash_lock);
1174 list_add(&hasher->list, &info->hashers);
1175 spin_unlock(&info->hash_lock);
1180 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1182 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1184 struct list_head *cur;
1185 struct btrfs_device *device;
1186 struct backing_dev_info *bdi;
1188 if ((bdi_bits & (1 << BDI_write_congested)) &&
1189 btrfs_congested_async(info, 0))
1192 list_for_each(cur, &info->fs_devices->devices) {
1193 device = list_entry(cur, struct btrfs_device, dev_list);
1196 bdi = blk_get_backing_dev_info(device->bdev);
1197 if (bdi && bdi_congested(bdi, bdi_bits)) {
1206 * this unplugs every device on the box, and it is only used when page
1209 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1211 struct list_head *cur;
1212 struct btrfs_device *device;
1213 struct btrfs_fs_info *info;
1215 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1216 list_for_each(cur, &info->fs_devices->devices) {
1217 device = list_entry(cur, struct btrfs_device, dev_list);
1221 bdi = blk_get_backing_dev_info(device->bdev);
1222 if (bdi->unplug_io_fn) {
1223 bdi->unplug_io_fn(bdi, page);
1228 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1230 struct inode *inode;
1231 struct extent_map_tree *em_tree;
1232 struct extent_map *em;
1233 struct address_space *mapping;
1236 /* the generic O_DIRECT read code does this */
1238 __unplug_io_fn(bdi, page);
1243 * page->mapping may change at any time. Get a consistent copy
1244 * and use that for everything below
1247 mapping = page->mapping;
1251 inode = mapping->host;
1254 * don't do the expensive searching for a small number of
1257 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1258 __unplug_io_fn(bdi, page);
1262 offset = page_offset(page);
1264 em_tree = &BTRFS_I(inode)->extent_tree;
1265 spin_lock(&em_tree->lock);
1266 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1267 spin_unlock(&em_tree->lock);
1269 __unplug_io_fn(bdi, page);
1273 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1274 free_extent_map(em);
1275 __unplug_io_fn(bdi, page);
1278 offset = offset - em->start;
1279 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1280 em->block_start + offset, page);
1281 free_extent_map(em);
1284 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1287 bdi->ra_pages = default_backing_dev_info.ra_pages;
1289 bdi->capabilities = default_backing_dev_info.capabilities;
1290 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1291 bdi->unplug_io_data = info;
1292 bdi->congested_fn = btrfs_congested_fn;
1293 bdi->congested_data = info;
1297 static int bio_ready_for_csum(struct bio *bio)
1303 struct extent_io_tree *io_tree = NULL;
1304 struct btrfs_fs_info *info = NULL;
1305 struct bio_vec *bvec;
1309 bio_for_each_segment(bvec, bio, i) {
1310 page = bvec->bv_page;
1311 if (page->private == EXTENT_PAGE_PRIVATE) {
1312 length += bvec->bv_len;
1315 if (!page->private) {
1316 length += bvec->bv_len;
1319 length = bvec->bv_len;
1320 buf_len = page->private >> 2;
1321 start = page_offset(page) + bvec->bv_offset;
1322 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1323 info = BTRFS_I(page->mapping->host)->root->fs_info;
1325 /* are we fully contained in this bio? */
1326 if (buf_len <= length)
1329 ret = extent_range_uptodate(io_tree, start + length,
1330 start + buf_len - 1);
1337 * called by the kthread helper functions to finally call the bio end_io
1338 * functions. This is where read checksum verification actually happens
1340 static void end_workqueue_fn(struct btrfs_work *work)
1343 struct end_io_wq *end_io_wq;
1344 struct btrfs_fs_info *fs_info;
1347 end_io_wq = container_of(work, struct end_io_wq, work);
1348 bio = end_io_wq->bio;
1349 fs_info = end_io_wq->info;
1351 /* metadata bios are special because the whole tree block must
1352 * be checksummed at once. This makes sure the entire block is in
1353 * ram and up to date before trying to verify things. For
1354 * blocksize <= pagesize, it is basically a noop
1356 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1357 btrfs_queue_worker(&fs_info->endio_meta_workers,
1361 error = end_io_wq->error;
1362 bio->bi_private = end_io_wq->private;
1363 bio->bi_end_io = end_io_wq->end_io;
1365 bio_endio(bio, error);
1368 static int cleaner_kthread(void *arg)
1370 struct btrfs_root *root = arg;
1374 if (root->fs_info->closing)
1377 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1378 mutex_lock(&root->fs_info->cleaner_mutex);
1379 btrfs_clean_old_snapshots(root);
1380 mutex_unlock(&root->fs_info->cleaner_mutex);
1382 if (freezing(current)) {
1386 if (root->fs_info->closing)
1388 set_current_state(TASK_INTERRUPTIBLE);
1390 __set_current_state(TASK_RUNNING);
1392 } while (!kthread_should_stop());
1396 static int transaction_kthread(void *arg)
1398 struct btrfs_root *root = arg;
1399 struct btrfs_trans_handle *trans;
1400 struct btrfs_transaction *cur;
1402 unsigned long delay;
1407 if (root->fs_info->closing)
1411 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1412 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1414 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1415 printk("btrfs: total reference cache size %Lu\n",
1416 root->fs_info->total_ref_cache_size);
1419 mutex_lock(&root->fs_info->trans_mutex);
1420 cur = root->fs_info->running_transaction;
1422 mutex_unlock(&root->fs_info->trans_mutex);
1426 now = get_seconds();
1427 if (now < cur->start_time || now - cur->start_time < 30) {
1428 mutex_unlock(&root->fs_info->trans_mutex);
1432 mutex_unlock(&root->fs_info->trans_mutex);
1433 trans = btrfs_start_transaction(root, 1);
1434 ret = btrfs_commit_transaction(trans, root);
1436 wake_up_process(root->fs_info->cleaner_kthread);
1437 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1439 if (freezing(current)) {
1442 if (root->fs_info->closing)
1444 set_current_state(TASK_INTERRUPTIBLE);
1445 schedule_timeout(delay);
1446 __set_current_state(TASK_RUNNING);
1448 } while (!kthread_should_stop());
1452 struct btrfs_root *open_ctree(struct super_block *sb,
1453 struct btrfs_fs_devices *fs_devices,
1463 struct btrfs_key location;
1464 struct buffer_head *bh;
1465 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1467 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1469 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1471 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1473 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1475 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1477 struct btrfs_root *log_tree_root;
1482 struct btrfs_super_block *disk_super;
1484 if (!extent_root || !tree_root || !fs_info ||
1485 !chunk_root || !dev_root || !csum_root) {
1489 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1490 INIT_LIST_HEAD(&fs_info->trans_list);
1491 INIT_LIST_HEAD(&fs_info->dead_roots);
1492 INIT_LIST_HEAD(&fs_info->hashers);
1493 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1494 spin_lock_init(&fs_info->hash_lock);
1495 spin_lock_init(&fs_info->delalloc_lock);
1496 spin_lock_init(&fs_info->new_trans_lock);
1497 spin_lock_init(&fs_info->ref_cache_lock);
1499 init_completion(&fs_info->kobj_unregister);
1500 fs_info->tree_root = tree_root;
1501 fs_info->extent_root = extent_root;
1502 fs_info->csum_root = csum_root;
1503 fs_info->chunk_root = chunk_root;
1504 fs_info->dev_root = dev_root;
1505 fs_info->fs_devices = fs_devices;
1506 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1507 INIT_LIST_HEAD(&fs_info->space_info);
1508 btrfs_mapping_init(&fs_info->mapping_tree);
1509 atomic_set(&fs_info->nr_async_submits, 0);
1510 atomic_set(&fs_info->async_delalloc_pages, 0);
1511 atomic_set(&fs_info->async_submit_draining, 0);
1512 atomic_set(&fs_info->nr_async_bios, 0);
1513 atomic_set(&fs_info->throttles, 0);
1514 atomic_set(&fs_info->throttle_gen, 0);
1516 fs_info->max_extent = (u64)-1;
1517 fs_info->max_inline = 8192 * 1024;
1518 setup_bdi(fs_info, &fs_info->bdi);
1519 fs_info->btree_inode = new_inode(sb);
1520 fs_info->btree_inode->i_ino = 1;
1521 fs_info->btree_inode->i_nlink = 1;
1523 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1525 INIT_LIST_HEAD(&fs_info->ordered_extents);
1526 spin_lock_init(&fs_info->ordered_extent_lock);
1528 sb->s_blocksize = 4096;
1529 sb->s_blocksize_bits = blksize_bits(4096);
1532 * we set the i_size on the btree inode to the max possible int.
1533 * the real end of the address space is determined by all of
1534 * the devices in the system
1536 fs_info->btree_inode->i_size = OFFSET_MAX;
1537 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1538 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1540 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1541 fs_info->btree_inode->i_mapping,
1543 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1546 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1548 spin_lock_init(&fs_info->block_group_cache_lock);
1549 fs_info->block_group_cache_tree.rb_node = NULL;
1551 extent_io_tree_init(&fs_info->pinned_extents,
1552 fs_info->btree_inode->i_mapping, GFP_NOFS);
1553 extent_io_tree_init(&fs_info->pending_del,
1554 fs_info->btree_inode->i_mapping, GFP_NOFS);
1555 extent_io_tree_init(&fs_info->extent_ins,
1556 fs_info->btree_inode->i_mapping, GFP_NOFS);
1557 fs_info->do_barriers = 1;
1559 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1560 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1561 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1563 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1564 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1565 sizeof(struct btrfs_key));
1566 insert_inode_hash(fs_info->btree_inode);
1568 mutex_init(&fs_info->trans_mutex);
1569 mutex_init(&fs_info->tree_log_mutex);
1570 mutex_init(&fs_info->drop_mutex);
1571 mutex_init(&fs_info->extent_ins_mutex);
1572 mutex_init(&fs_info->pinned_mutex);
1573 mutex_init(&fs_info->chunk_mutex);
1574 mutex_init(&fs_info->transaction_kthread_mutex);
1575 mutex_init(&fs_info->cleaner_mutex);
1576 mutex_init(&fs_info->volume_mutex);
1577 mutex_init(&fs_info->tree_reloc_mutex);
1578 init_waitqueue_head(&fs_info->transaction_throttle);
1579 init_waitqueue_head(&fs_info->transaction_wait);
1580 init_waitqueue_head(&fs_info->async_submit_wait);
1581 init_waitqueue_head(&fs_info->tree_log_wait);
1582 atomic_set(&fs_info->tree_log_commit, 0);
1583 atomic_set(&fs_info->tree_log_writers, 0);
1584 fs_info->tree_log_transid = 0;
1587 ret = add_hasher(fs_info, "crc32c");
1589 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1594 __setup_root(4096, 4096, 4096, 4096, tree_root,
1595 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1598 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1602 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1605 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1607 disk_super = &fs_info->super_copy;
1608 if (!btrfs_super_root(disk_super))
1611 ret = btrfs_parse_options(tree_root, options);
1617 features = btrfs_super_incompat_flags(disk_super) &
1618 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1620 printk(KERN_ERR "BTRFS: couldn't mount because of "
1621 "unsupported optional features (%Lx).\n",
1627 features = btrfs_super_compat_ro_flags(disk_super) &
1628 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1629 if (!(sb->s_flags & MS_RDONLY) && features) {
1630 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1631 "unsupported option features (%Lx).\n",
1638 * we need to start all the end_io workers up front because the
1639 * queue work function gets called at interrupt time, and so it
1640 * cannot dynamically grow.
1642 btrfs_init_workers(&fs_info->workers, "worker",
1643 fs_info->thread_pool_size);
1645 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1646 fs_info->thread_pool_size);
1648 btrfs_init_workers(&fs_info->submit_workers, "submit",
1649 min_t(u64, fs_devices->num_devices,
1650 fs_info->thread_pool_size));
1652 /* a higher idle thresh on the submit workers makes it much more
1653 * likely that bios will be send down in a sane order to the
1656 fs_info->submit_workers.idle_thresh = 64;
1658 fs_info->workers.idle_thresh = 16;
1659 fs_info->workers.ordered = 1;
1661 fs_info->delalloc_workers.idle_thresh = 2;
1662 fs_info->delalloc_workers.ordered = 1;
1664 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1665 btrfs_init_workers(&fs_info->endio_workers, "endio",
1666 fs_info->thread_pool_size);
1667 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1668 fs_info->thread_pool_size);
1669 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1670 fs_info->thread_pool_size);
1673 * endios are largely parallel and should have a very
1676 fs_info->endio_workers.idle_thresh = 4;
1677 fs_info->endio_write_workers.idle_thresh = 64;
1679 btrfs_start_workers(&fs_info->workers, 1);
1680 btrfs_start_workers(&fs_info->submit_workers, 1);
1681 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1682 btrfs_start_workers(&fs_info->fixup_workers, 1);
1683 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1684 btrfs_start_workers(&fs_info->endio_meta_workers,
1685 fs_info->thread_pool_size);
1686 btrfs_start_workers(&fs_info->endio_write_workers,
1687 fs_info->thread_pool_size);
1689 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1690 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1691 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1693 nodesize = btrfs_super_nodesize(disk_super);
1694 leafsize = btrfs_super_leafsize(disk_super);
1695 sectorsize = btrfs_super_sectorsize(disk_super);
1696 stripesize = btrfs_super_stripesize(disk_super);
1697 tree_root->nodesize = nodesize;
1698 tree_root->leafsize = leafsize;
1699 tree_root->sectorsize = sectorsize;
1700 tree_root->stripesize = stripesize;
1702 sb->s_blocksize = sectorsize;
1703 sb->s_blocksize_bits = blksize_bits(sectorsize);
1705 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1706 sizeof(disk_super->magic))) {
1707 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1708 goto fail_sb_buffer;
1711 mutex_lock(&fs_info->chunk_mutex);
1712 ret = btrfs_read_sys_array(tree_root, btrfs_super_bytenr(disk_super));
1713 mutex_unlock(&fs_info->chunk_mutex);
1715 printk("btrfs: failed to read the system array on %s\n",
1717 goto fail_sys_array;
1720 blocksize = btrfs_level_size(tree_root,
1721 btrfs_super_chunk_root_level(disk_super));
1722 generation = btrfs_super_chunk_root_generation(disk_super);
1724 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1725 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1727 chunk_root->node = read_tree_block(chunk_root,
1728 btrfs_super_chunk_root(disk_super),
1729 blocksize, generation);
1730 BUG_ON(!chunk_root->node);
1732 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1733 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1736 mutex_lock(&fs_info->chunk_mutex);
1737 ret = btrfs_read_chunk_tree(chunk_root);
1738 mutex_unlock(&fs_info->chunk_mutex);
1740 printk("btrfs: failed to read chunk tree on %s\n", sb->s_id);
1741 goto fail_chunk_root;
1744 btrfs_close_extra_devices(fs_devices);
1746 blocksize = btrfs_level_size(tree_root,
1747 btrfs_super_root_level(disk_super));
1748 generation = btrfs_super_generation(disk_super);
1750 tree_root->node = read_tree_block(tree_root,
1751 btrfs_super_root(disk_super),
1752 blocksize, generation);
1753 if (!tree_root->node)
1754 goto fail_chunk_root;
1757 ret = find_and_setup_root(tree_root, fs_info,
1758 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1760 goto fail_tree_root;
1761 extent_root->track_dirty = 1;
1763 ret = find_and_setup_root(tree_root, fs_info,
1764 BTRFS_DEV_TREE_OBJECTID, dev_root);
1765 dev_root->track_dirty = 1;
1768 goto fail_extent_root;
1770 ret = find_and_setup_root(tree_root, fs_info,
1771 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1773 goto fail_extent_root;
1775 csum_root->track_dirty = 1;
1777 btrfs_read_block_groups(extent_root);
1779 fs_info->generation = generation + 1;
1780 fs_info->last_trans_committed = generation;
1781 fs_info->data_alloc_profile = (u64)-1;
1782 fs_info->metadata_alloc_profile = (u64)-1;
1783 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1784 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1786 if (!fs_info->cleaner_kthread)
1787 goto fail_csum_root;
1789 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1791 "btrfs-transaction");
1792 if (!fs_info->transaction_kthread)
1795 if (btrfs_super_log_root(disk_super) != 0) {
1796 u64 bytenr = btrfs_super_log_root(disk_super);
1798 if (fs_devices->rw_devices == 0) {
1799 printk("Btrfs log replay required on RO media\n");
1801 goto fail_trans_kthread;
1804 btrfs_level_size(tree_root,
1805 btrfs_super_log_root_level(disk_super));
1807 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1810 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1811 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1813 log_tree_root->node = read_tree_block(tree_root, bytenr,
1816 ret = btrfs_recover_log_trees(log_tree_root);
1819 if (sb->s_flags & MS_RDONLY) {
1820 ret = btrfs_commit_super(tree_root);
1825 if (!(sb->s_flags & MS_RDONLY)) {
1826 ret = btrfs_cleanup_reloc_trees(tree_root);
1830 location.objectid = BTRFS_FS_TREE_OBJECTID;
1831 location.type = BTRFS_ROOT_ITEM_KEY;
1832 location.offset = (u64)-1;
1834 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1835 if (!fs_info->fs_root)
1836 goto fail_trans_kthread;
1840 kthread_stop(fs_info->transaction_kthread);
1842 kthread_stop(fs_info->cleaner_kthread);
1845 * make sure we're done with the btree inode before we stop our
1848 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1849 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1852 free_extent_buffer(csum_root->node);
1854 free_extent_buffer(extent_root->node);
1856 free_extent_buffer(tree_root->node);
1858 free_extent_buffer(chunk_root->node);
1860 free_extent_buffer(dev_root->node);
1862 btrfs_stop_workers(&fs_info->fixup_workers);
1863 btrfs_stop_workers(&fs_info->delalloc_workers);
1864 btrfs_stop_workers(&fs_info->workers);
1865 btrfs_stop_workers(&fs_info->endio_workers);
1866 btrfs_stop_workers(&fs_info->endio_meta_workers);
1867 btrfs_stop_workers(&fs_info->endio_write_workers);
1868 btrfs_stop_workers(&fs_info->submit_workers);
1870 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1871 iput(fs_info->btree_inode);
1873 btrfs_close_devices(fs_info->fs_devices);
1874 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1878 bdi_destroy(&fs_info->bdi);
1883 return ERR_PTR(err);
1886 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1888 char b[BDEVNAME_SIZE];
1891 set_buffer_uptodate(bh);
1893 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1894 printk(KERN_WARNING "lost page write due to "
1895 "I/O error on %s\n",
1896 bdevname(bh->b_bdev, b));
1898 /* note, we dont' set_buffer_write_io_error because we have
1899 * our own ways of dealing with the IO errors
1901 clear_buffer_uptodate(bh);
1907 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1909 struct buffer_head *bh;
1910 struct buffer_head *latest = NULL;
1911 struct btrfs_super_block *super;
1916 /* we would like to check all the supers, but that would make
1917 * a btrfs mount succeed after a mkfs from a different FS.
1918 * So, we need to add a special mount option to scan for
1919 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1921 for (i = 0; i < 1; i++) {
1922 bytenr = btrfs_sb_offset(i);
1923 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1925 bh = __bread(bdev, bytenr / 4096, 4096);
1929 super = (struct btrfs_super_block *)bh->b_data;
1930 if (btrfs_super_bytenr(super) != bytenr ||
1931 strncmp((char *)(&super->magic), BTRFS_MAGIC,
1932 sizeof(super->magic))) {
1937 if (!latest || btrfs_super_generation(super) > transid) {
1940 transid = btrfs_super_generation(super);
1948 static int write_dev_supers(struct btrfs_device *device,
1949 struct btrfs_super_block *sb,
1950 int do_barriers, int wait, int max_mirrors)
1952 struct buffer_head *bh;
1958 int last_barrier = 0;
1960 if (max_mirrors == 0)
1961 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
1963 /* make sure only the last submit_bh does a barrier */
1965 for (i = 0; i < max_mirrors; i++) {
1966 bytenr = btrfs_sb_offset(i);
1967 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1968 device->total_bytes)
1974 for (i = 0; i < max_mirrors; i++) {
1975 bytenr = btrfs_sb_offset(i);
1976 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1980 bh = __find_get_block(device->bdev, bytenr / 4096,
1981 BTRFS_SUPER_INFO_SIZE);
1985 if (buffer_uptodate(bh)) {
1990 btrfs_set_super_bytenr(sb, bytenr);
1993 crc = btrfs_csum_data(NULL, (char *)sb +
1994 BTRFS_CSUM_SIZE, crc,
1995 BTRFS_SUPER_INFO_SIZE -
1997 btrfs_csum_final(crc, sb->csum);
1999 bh = __getblk(device->bdev, bytenr / 4096,
2000 BTRFS_SUPER_INFO_SIZE);
2001 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2003 set_buffer_uptodate(bh);
2006 bh->b_end_io = btrfs_end_buffer_write_sync;
2009 if (i == last_barrier && do_barriers && device->barriers) {
2010 ret = submit_bh(WRITE_BARRIER, bh);
2011 if (ret == -EOPNOTSUPP) {
2012 printk("btrfs: disabling barriers on dev %s\n",
2014 set_buffer_uptodate(bh);
2015 device->barriers = 0;
2018 ret = submit_bh(WRITE, bh);
2021 ret = submit_bh(WRITE, bh);
2026 if (!buffer_uptodate(bh))
2034 return errors < i ? 0 : -1;
2037 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2039 struct list_head *cur;
2040 struct list_head *head = &root->fs_info->fs_devices->devices;
2041 struct btrfs_device *dev;
2042 struct btrfs_super_block *sb;
2043 struct btrfs_dev_item *dev_item;
2047 int total_errors = 0;
2050 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2051 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2053 sb = &root->fs_info->super_for_commit;
2054 dev_item = &sb->dev_item;
2055 list_for_each(cur, head) {
2056 dev = list_entry(cur, struct btrfs_device, dev_list);
2061 if (!dev->in_fs_metadata || !dev->writeable)
2064 btrfs_set_stack_device_generation(dev_item, 0);
2065 btrfs_set_stack_device_type(dev_item, dev->type);
2066 btrfs_set_stack_device_id(dev_item, dev->devid);
2067 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2068 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2069 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2070 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2071 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2072 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2073 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2075 flags = btrfs_super_flags(sb);
2076 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2078 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2082 if (total_errors > max_errors) {
2083 printk("btrfs: %d errors while writing supers\n", total_errors);
2088 list_for_each(cur, head) {
2089 dev = list_entry(cur, struct btrfs_device, dev_list);
2092 if (!dev->in_fs_metadata || !dev->writeable)
2095 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2099 if (total_errors > max_errors) {
2100 printk("btrfs: %d errors while writing supers\n", total_errors);
2106 int write_ctree_super(struct btrfs_trans_handle *trans,
2107 struct btrfs_root *root, int max_mirrors)
2111 ret = write_all_supers(root, max_mirrors);
2115 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2117 radix_tree_delete(&fs_info->fs_roots_radix,
2118 (unsigned long)root->root_key.objectid);
2119 if (root->anon_super.s_dev) {
2120 down_write(&root->anon_super.s_umount);
2121 kill_anon_super(&root->anon_super);
2125 btrfs_sysfs_del_root(root);
2128 free_extent_buffer(root->node);
2129 if (root->commit_root)
2130 free_extent_buffer(root->commit_root);
2137 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2140 struct btrfs_root *gang[8];
2144 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2149 for (i = 0; i < ret; i++)
2150 btrfs_free_fs_root(fs_info, gang[i]);
2155 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2157 u64 root_objectid = 0;
2158 struct btrfs_root *gang[8];
2163 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2164 (void **)gang, root_objectid,
2168 for (i = 0; i < ret; i++) {
2169 root_objectid = gang[i]->root_key.objectid;
2170 ret = btrfs_find_dead_roots(fs_info->tree_root,
2171 root_objectid, gang[i]);
2173 btrfs_orphan_cleanup(gang[i]);
2180 int btrfs_commit_super(struct btrfs_root *root)
2182 struct btrfs_trans_handle *trans;
2185 mutex_lock(&root->fs_info->cleaner_mutex);
2186 btrfs_clean_old_snapshots(root);
2187 mutex_unlock(&root->fs_info->cleaner_mutex);
2188 trans = btrfs_start_transaction(root, 1);
2189 ret = btrfs_commit_transaction(trans, root);
2191 /* run commit again to drop the original snapshot */
2192 trans = btrfs_start_transaction(root, 1);
2193 btrfs_commit_transaction(trans, root);
2194 ret = btrfs_write_and_wait_transaction(NULL, root);
2197 ret = write_ctree_super(NULL, root, 0);
2201 int close_ctree(struct btrfs_root *root)
2203 struct btrfs_fs_info *fs_info = root->fs_info;
2206 fs_info->closing = 1;
2209 kthread_stop(root->fs_info->transaction_kthread);
2210 kthread_stop(root->fs_info->cleaner_kthread);
2212 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2213 ret = btrfs_commit_super(root);
2215 printk("btrfs: commit super returns %d\n", ret);
2219 if (fs_info->delalloc_bytes) {
2220 printk("btrfs: at unmount delalloc count %Lu\n",
2221 fs_info->delalloc_bytes);
2223 if (fs_info->total_ref_cache_size) {
2224 printk("btrfs: at umount reference cache size %Lu\n",
2225 fs_info->total_ref_cache_size);
2228 if (fs_info->extent_root->node)
2229 free_extent_buffer(fs_info->extent_root->node);
2231 if (fs_info->tree_root->node)
2232 free_extent_buffer(fs_info->tree_root->node);
2234 if (root->fs_info->chunk_root->node);
2235 free_extent_buffer(root->fs_info->chunk_root->node);
2237 if (root->fs_info->dev_root->node);
2238 free_extent_buffer(root->fs_info->dev_root->node);
2240 if (root->fs_info->csum_root->node);
2241 free_extent_buffer(root->fs_info->csum_root->node);
2243 btrfs_free_block_groups(root->fs_info);
2245 del_fs_roots(fs_info);
2247 iput(fs_info->btree_inode);
2249 btrfs_stop_workers(&fs_info->fixup_workers);
2250 btrfs_stop_workers(&fs_info->delalloc_workers);
2251 btrfs_stop_workers(&fs_info->workers);
2252 btrfs_stop_workers(&fs_info->endio_workers);
2253 btrfs_stop_workers(&fs_info->endio_meta_workers);
2254 btrfs_stop_workers(&fs_info->endio_write_workers);
2255 btrfs_stop_workers(&fs_info->submit_workers);
2258 while(!list_empty(&fs_info->hashers)) {
2259 struct btrfs_hasher *hasher;
2260 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2262 list_del(&hasher->hashers);
2263 crypto_free_hash(&fs_info->hash_tfm);
2267 btrfs_close_devices(fs_info->fs_devices);
2268 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2270 bdi_destroy(&fs_info->bdi);
2272 kfree(fs_info->extent_root);
2273 kfree(fs_info->tree_root);
2274 kfree(fs_info->chunk_root);
2275 kfree(fs_info->dev_root);
2276 kfree(fs_info->csum_root);
2280 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2283 struct inode *btree_inode = buf->first_page->mapping->host;
2285 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2289 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2294 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2296 struct inode *btree_inode = buf->first_page->mapping->host;
2297 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2301 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2303 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2304 u64 transid = btrfs_header_generation(buf);
2305 struct inode *btree_inode = root->fs_info->btree_inode;
2307 WARN_ON(!btrfs_tree_locked(buf));
2308 if (transid != root->fs_info->generation) {
2309 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
2310 (unsigned long long)buf->start,
2311 transid, root->fs_info->generation);
2314 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2317 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2320 * looks as though older kernels can get into trouble with
2321 * this code, they end up stuck in balance_dirty_pages forever
2323 struct extent_io_tree *tree;
2326 unsigned long thresh = 32 * 1024 * 1024;
2327 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2329 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2332 num_dirty = count_range_bits(tree, &start, (u64)-1,
2333 thresh, EXTENT_DIRTY);
2334 if (num_dirty > thresh) {
2335 balance_dirty_pages_ratelimited_nr(
2336 root->fs_info->btree_inode->i_mapping, 1);
2341 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2343 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2345 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2347 buf->flags |= EXTENT_UPTODATE;
2352 int btree_lock_page_hook(struct page *page)
2354 struct inode *inode = page->mapping->host;
2355 struct btrfs_root *root = BTRFS_I(inode)->root;
2356 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2357 struct extent_buffer *eb;
2359 u64 bytenr = page_offset(page);
2361 if (page->private == EXTENT_PAGE_PRIVATE)
2364 len = page->private >> 2;
2365 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2369 btrfs_tree_lock(eb);
2370 spin_lock(&root->fs_info->hash_lock);
2371 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2372 spin_unlock(&root->fs_info->hash_lock);
2373 btrfs_tree_unlock(eb);
2374 free_extent_buffer(eb);
2380 static struct extent_io_ops btree_extent_io_ops = {
2381 .write_cache_pages_lock_hook = btree_lock_page_hook,
2382 .readpage_end_io_hook = btree_readpage_end_io_hook,
2383 .submit_bio_hook = btree_submit_bio_hook,
2384 /* note we're sharing with inode.c for the merge bio hook */
2385 .merge_bio_hook = btrfs_merge_bio_hook,