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.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
39 #include "async-thread.h"
42 #include "free-space-cache.h"
44 static struct extent_io_ops btree_extent_io_ops;
45 static void end_workqueue_fn(struct btrfs_work *work);
46 static void free_fs_root(struct btrfs_root *root);
49 * end_io_wq structs are used to do processing in task context when an IO is
50 * complete. This is used during reads to verify checksums, and it is used
51 * by writes to insert metadata for new file extents after IO is complete.
57 struct btrfs_fs_info *info;
60 struct list_head list;
61 struct btrfs_work work;
65 * async submit bios are used to offload expensive checksumming
66 * onto the worker threads. They checksum file and metadata bios
67 * just before they are sent down the IO stack.
69 struct async_submit_bio {
72 struct list_head list;
73 extent_submit_bio_hook_t *submit_bio_start;
74 extent_submit_bio_hook_t *submit_bio_done;
77 unsigned long bio_flags;
79 * bio_offset is optional, can be used if the pages in the bio
80 * can't tell us where in the file the bio should go
83 struct btrfs_work work;
86 /* These are used to set the lockdep class on the extent buffer locks.
87 * The class is set by the readpage_end_io_hook after the buffer has
88 * passed csum validation but before the pages are unlocked.
90 * The lockdep class is also set by btrfs_init_new_buffer on freshly
93 * The class is based on the level in the tree block, which allows lockdep
94 * to know that lower nodes nest inside the locks of higher nodes.
96 * We also add a check to make sure the highest level of the tree is
97 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
98 * code needs update as well.
100 #ifdef CONFIG_DEBUG_LOCK_ALLOC
101 # if BTRFS_MAX_LEVEL != 8
104 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
105 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
115 /* highest possible level */
121 * extents on the btree inode are pretty simple, there's one extent
122 * that covers the entire device
124 static struct extent_map *btree_get_extent(struct inode *inode,
125 struct page *page, size_t page_offset, u64 start, u64 len,
128 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
129 struct extent_map *em;
132 read_lock(&em_tree->lock);
133 em = lookup_extent_mapping(em_tree, start, len);
136 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
137 read_unlock(&em_tree->lock);
140 read_unlock(&em_tree->lock);
142 em = alloc_extent_map(GFP_NOFS);
144 em = ERR_PTR(-ENOMEM);
149 em->block_len = (u64)-1;
151 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
153 write_lock(&em_tree->lock);
154 ret = add_extent_mapping(em_tree, em);
155 if (ret == -EEXIST) {
156 u64 failed_start = em->start;
157 u64 failed_len = em->len;
160 em = lookup_extent_mapping(em_tree, start, len);
164 em = lookup_extent_mapping(em_tree, failed_start,
172 write_unlock(&em_tree->lock);
180 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
182 return crc32c(seed, data, len);
185 void btrfs_csum_final(u32 crc, char *result)
187 *(__le32 *)result = ~cpu_to_le32(crc);
191 * compute the csum for a btree block, and either verify it or write it
192 * into the csum field of the block.
194 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
198 btrfs_super_csum_size(&root->fs_info->super_copy);
201 unsigned long cur_len;
202 unsigned long offset = BTRFS_CSUM_SIZE;
203 char *map_token = NULL;
205 unsigned long map_start;
206 unsigned long map_len;
209 unsigned long inline_result;
211 len = buf->len - offset;
213 err = map_private_extent_buffer(buf, offset, 32,
215 &map_start, &map_len, KM_USER0);
218 cur_len = min(len, map_len - (offset - map_start));
219 crc = btrfs_csum_data(root, kaddr + offset - map_start,
223 unmap_extent_buffer(buf, map_token, KM_USER0);
225 if (csum_size > sizeof(inline_result)) {
226 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
230 result = (char *)&inline_result;
233 btrfs_csum_final(crc, result);
236 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
239 memcpy(&found, result, csum_size);
241 read_extent_buffer(buf, &val, 0, csum_size);
242 if (printk_ratelimit()) {
243 printk(KERN_INFO "btrfs: %s checksum verify "
244 "failed on %llu wanted %X found %X "
246 root->fs_info->sb->s_id,
247 (unsigned long long)buf->start, val, found,
248 btrfs_header_level(buf));
250 if (result != (char *)&inline_result)
255 write_extent_buffer(buf, result, 0, csum_size);
257 if (result != (char *)&inline_result)
263 * we can't consider a given block up to date unless the transid of the
264 * block matches the transid in the parent node's pointer. This is how we
265 * detect blocks that either didn't get written at all or got written
266 * in the wrong place.
268 static int verify_parent_transid(struct extent_io_tree *io_tree,
269 struct extent_buffer *eb, u64 parent_transid)
271 struct extent_state *cached_state = NULL;
274 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
277 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
278 0, &cached_state, GFP_NOFS);
279 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
280 btrfs_header_generation(eb) == parent_transid) {
284 if (printk_ratelimit()) {
285 printk("parent transid verify failed on %llu wanted %llu "
287 (unsigned long long)eb->start,
288 (unsigned long long)parent_transid,
289 (unsigned long long)btrfs_header_generation(eb));
292 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
294 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
295 &cached_state, GFP_NOFS);
300 * helper to read a given tree block, doing retries as required when
301 * the checksums don't match and we have alternate mirrors to try.
303 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
304 struct extent_buffer *eb,
305 u64 start, u64 parent_transid)
307 struct extent_io_tree *io_tree;
312 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
314 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
315 btree_get_extent, mirror_num);
317 !verify_parent_transid(io_tree, eb, parent_transid))
320 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
326 if (mirror_num > num_copies)
333 * checksum a dirty tree block before IO. This has extra checks to make sure
334 * we only fill in the checksum field in the first page of a multi-page block
337 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
339 struct extent_io_tree *tree;
340 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
343 struct extent_buffer *eb;
346 tree = &BTRFS_I(page->mapping->host)->io_tree;
348 if (page->private == EXTENT_PAGE_PRIVATE)
352 len = page->private >> 2;
355 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
356 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
357 btrfs_header_generation(eb));
359 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
361 found_start = btrfs_header_bytenr(eb);
362 if (found_start != start) {
366 if (eb->first_page != page) {
370 if (!PageUptodate(page)) {
374 csum_tree_block(root, eb, 0);
376 free_extent_buffer(eb);
381 static int check_tree_block_fsid(struct btrfs_root *root,
382 struct extent_buffer *eb)
384 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
385 u8 fsid[BTRFS_UUID_SIZE];
388 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
391 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
395 fs_devices = fs_devices->seed;
400 #ifdef CONFIG_DEBUG_LOCK_ALLOC
401 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
403 lockdep_set_class_and_name(&eb->lock,
404 &btrfs_eb_class[level],
405 btrfs_eb_name[level]);
409 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
410 struct extent_state *state)
412 struct extent_io_tree *tree;
416 struct extent_buffer *eb;
417 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
420 tree = &BTRFS_I(page->mapping->host)->io_tree;
421 if (page->private == EXTENT_PAGE_PRIVATE)
426 len = page->private >> 2;
429 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
431 found_start = btrfs_header_bytenr(eb);
432 if (found_start != start) {
433 if (printk_ratelimit()) {
434 printk(KERN_INFO "btrfs bad tree block start "
436 (unsigned long long)found_start,
437 (unsigned long long)eb->start);
442 if (eb->first_page != page) {
443 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
444 eb->first_page->index, page->index);
449 if (check_tree_block_fsid(root, eb)) {
450 if (printk_ratelimit()) {
451 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
452 (unsigned long long)eb->start);
457 found_level = btrfs_header_level(eb);
459 btrfs_set_buffer_lockdep_class(eb, found_level);
461 ret = csum_tree_block(root, eb, 1);
465 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
466 end = eb->start + end - 1;
468 free_extent_buffer(eb);
473 static void end_workqueue_bio(struct bio *bio, int err)
475 struct end_io_wq *end_io_wq = bio->bi_private;
476 struct btrfs_fs_info *fs_info;
478 fs_info = end_io_wq->info;
479 end_io_wq->error = err;
480 end_io_wq->work.func = end_workqueue_fn;
481 end_io_wq->work.flags = 0;
483 if (bio->bi_rw & REQ_WRITE) {
484 if (end_io_wq->metadata == 1)
485 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
487 else if (end_io_wq->metadata == 2)
488 btrfs_queue_worker(&fs_info->endio_freespace_worker,
491 btrfs_queue_worker(&fs_info->endio_write_workers,
494 if (end_io_wq->metadata)
495 btrfs_queue_worker(&fs_info->endio_meta_workers,
498 btrfs_queue_worker(&fs_info->endio_workers,
504 * For the metadata arg you want
507 * 1 - if normal metadta
508 * 2 - if writing to the free space cache area
510 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
513 struct end_io_wq *end_io_wq;
514 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
518 end_io_wq->private = bio->bi_private;
519 end_io_wq->end_io = bio->bi_end_io;
520 end_io_wq->info = info;
521 end_io_wq->error = 0;
522 end_io_wq->bio = bio;
523 end_io_wq->metadata = metadata;
525 bio->bi_private = end_io_wq;
526 bio->bi_end_io = end_workqueue_bio;
530 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
532 unsigned long limit = min_t(unsigned long,
533 info->workers.max_workers,
534 info->fs_devices->open_devices);
538 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
540 return atomic_read(&info->nr_async_bios) >
541 btrfs_async_submit_limit(info);
544 static void run_one_async_start(struct btrfs_work *work)
546 struct async_submit_bio *async;
548 async = container_of(work, struct async_submit_bio, work);
549 async->submit_bio_start(async->inode, async->rw, async->bio,
550 async->mirror_num, async->bio_flags,
554 static void run_one_async_done(struct btrfs_work *work)
556 struct btrfs_fs_info *fs_info;
557 struct async_submit_bio *async;
560 async = container_of(work, struct async_submit_bio, work);
561 fs_info = BTRFS_I(async->inode)->root->fs_info;
563 limit = btrfs_async_submit_limit(fs_info);
564 limit = limit * 2 / 3;
566 atomic_dec(&fs_info->nr_async_submits);
568 if (atomic_read(&fs_info->nr_async_submits) < limit &&
569 waitqueue_active(&fs_info->async_submit_wait))
570 wake_up(&fs_info->async_submit_wait);
572 async->submit_bio_done(async->inode, async->rw, async->bio,
573 async->mirror_num, async->bio_flags,
577 static void run_one_async_free(struct btrfs_work *work)
579 struct async_submit_bio *async;
581 async = container_of(work, struct async_submit_bio, work);
585 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
586 int rw, struct bio *bio, int mirror_num,
587 unsigned long bio_flags,
589 extent_submit_bio_hook_t *submit_bio_start,
590 extent_submit_bio_hook_t *submit_bio_done)
592 struct async_submit_bio *async;
594 async = kmalloc(sizeof(*async), GFP_NOFS);
598 async->inode = inode;
601 async->mirror_num = mirror_num;
602 async->submit_bio_start = submit_bio_start;
603 async->submit_bio_done = submit_bio_done;
605 async->work.func = run_one_async_start;
606 async->work.ordered_func = run_one_async_done;
607 async->work.ordered_free = run_one_async_free;
609 async->work.flags = 0;
610 async->bio_flags = bio_flags;
611 async->bio_offset = bio_offset;
613 atomic_inc(&fs_info->nr_async_submits);
616 btrfs_set_work_high_prio(&async->work);
618 btrfs_queue_worker(&fs_info->workers, &async->work);
620 while (atomic_read(&fs_info->async_submit_draining) &&
621 atomic_read(&fs_info->nr_async_submits)) {
622 wait_event(fs_info->async_submit_wait,
623 (atomic_read(&fs_info->nr_async_submits) == 0));
629 static int btree_csum_one_bio(struct bio *bio)
631 struct bio_vec *bvec = bio->bi_io_vec;
633 struct btrfs_root *root;
635 WARN_ON(bio->bi_vcnt <= 0);
636 while (bio_index < bio->bi_vcnt) {
637 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
638 csum_dirty_buffer(root, bvec->bv_page);
645 static int __btree_submit_bio_start(struct inode *inode, int rw,
646 struct bio *bio, int mirror_num,
647 unsigned long bio_flags,
651 * when we're called for a write, we're already in the async
652 * submission context. Just jump into btrfs_map_bio
654 btree_csum_one_bio(bio);
658 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
659 int mirror_num, unsigned long bio_flags,
663 * when we're called for a write, we're already in the async
664 * submission context. Just jump into btrfs_map_bio
666 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
669 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
670 int mirror_num, unsigned long bio_flags,
675 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
679 if (!(rw & REQ_WRITE)) {
681 * called for a read, do the setup so that checksum validation
682 * can happen in the async kernel threads
684 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
689 * kthread helpers are used to submit writes so that checksumming
690 * can happen in parallel across all CPUs
692 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
693 inode, rw, bio, mirror_num, 0,
695 __btree_submit_bio_start,
696 __btree_submit_bio_done);
699 static int btree_migratepage(struct address_space *mapping,
700 struct page *newpage, struct page *page)
703 * we can't safely write a btree page from here,
704 * we haven't done the locking hook
709 * Buffers may be managed in a filesystem specific way.
710 * We must have no buffers or drop them.
712 if (page_has_private(page) &&
713 !try_to_release_page(page, GFP_KERNEL))
716 return migrate_page(mapping, newpage, page);
719 static int btree_writepage(struct page *page, struct writeback_control *wbc)
721 struct extent_io_tree *tree;
722 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
723 struct extent_buffer *eb;
726 tree = &BTRFS_I(page->mapping->host)->io_tree;
727 if (!(current->flags & PF_MEMALLOC)) {
728 return extent_write_full_page(tree, page,
729 btree_get_extent, wbc);
732 redirty_page_for_writepage(wbc, page);
733 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
736 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
738 spin_lock(&root->fs_info->delalloc_lock);
739 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
740 spin_unlock(&root->fs_info->delalloc_lock);
742 free_extent_buffer(eb);
748 static int btree_writepages(struct address_space *mapping,
749 struct writeback_control *wbc)
751 struct extent_io_tree *tree;
752 tree = &BTRFS_I(mapping->host)->io_tree;
753 if (wbc->sync_mode == WB_SYNC_NONE) {
754 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
756 unsigned long thresh = 32 * 1024 * 1024;
758 if (wbc->for_kupdate)
761 /* this is a bit racy, but that's ok */
762 num_dirty = root->fs_info->dirty_metadata_bytes;
763 if (num_dirty < thresh)
766 return extent_writepages(tree, mapping, btree_get_extent, wbc);
769 static int btree_readpage(struct file *file, struct page *page)
771 struct extent_io_tree *tree;
772 tree = &BTRFS_I(page->mapping->host)->io_tree;
773 return extent_read_full_page(tree, page, btree_get_extent);
776 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
778 struct extent_io_tree *tree;
779 struct extent_map_tree *map;
782 if (PageWriteback(page) || PageDirty(page))
785 tree = &BTRFS_I(page->mapping->host)->io_tree;
786 map = &BTRFS_I(page->mapping->host)->extent_tree;
788 ret = try_release_extent_state(map, tree, page, gfp_flags);
792 ret = try_release_extent_buffer(tree, page);
794 ClearPagePrivate(page);
795 set_page_private(page, 0);
796 page_cache_release(page);
802 static void btree_invalidatepage(struct page *page, unsigned long offset)
804 struct extent_io_tree *tree;
805 tree = &BTRFS_I(page->mapping->host)->io_tree;
806 extent_invalidatepage(tree, page, offset);
807 btree_releasepage(page, GFP_NOFS);
808 if (PagePrivate(page)) {
809 printk(KERN_WARNING "btrfs warning page private not zero "
810 "on page %llu\n", (unsigned long long)page_offset(page));
811 ClearPagePrivate(page);
812 set_page_private(page, 0);
813 page_cache_release(page);
817 static const struct address_space_operations btree_aops = {
818 .readpage = btree_readpage,
819 .writepage = btree_writepage,
820 .writepages = btree_writepages,
821 .releasepage = btree_releasepage,
822 .invalidatepage = btree_invalidatepage,
823 .sync_page = block_sync_page,
824 .migratepage = btree_migratepage,
827 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
830 struct extent_buffer *buf = NULL;
831 struct inode *btree_inode = root->fs_info->btree_inode;
834 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
837 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
838 buf, 0, 0, btree_get_extent, 0);
839 free_extent_buffer(buf);
843 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
844 u64 bytenr, u32 blocksize)
846 struct inode *btree_inode = root->fs_info->btree_inode;
847 struct extent_buffer *eb;
848 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
849 bytenr, blocksize, GFP_NOFS);
853 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
854 u64 bytenr, u32 blocksize)
856 struct inode *btree_inode = root->fs_info->btree_inode;
857 struct extent_buffer *eb;
859 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
860 bytenr, blocksize, NULL, GFP_NOFS);
865 int btrfs_write_tree_block(struct extent_buffer *buf)
867 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
868 buf->start + buf->len - 1);
871 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
873 return filemap_fdatawait_range(buf->first_page->mapping,
874 buf->start, buf->start + buf->len - 1);
877 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
878 u32 blocksize, u64 parent_transid)
880 struct extent_buffer *buf = NULL;
883 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
887 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
890 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
895 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
896 struct extent_buffer *buf)
898 struct inode *btree_inode = root->fs_info->btree_inode;
899 if (btrfs_header_generation(buf) ==
900 root->fs_info->running_transaction->transid) {
901 btrfs_assert_tree_locked(buf);
903 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
904 spin_lock(&root->fs_info->delalloc_lock);
905 if (root->fs_info->dirty_metadata_bytes >= buf->len)
906 root->fs_info->dirty_metadata_bytes -= buf->len;
909 spin_unlock(&root->fs_info->delalloc_lock);
912 /* ugh, clear_extent_buffer_dirty needs to lock the page */
913 btrfs_set_lock_blocking(buf);
914 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
920 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
921 u32 stripesize, struct btrfs_root *root,
922 struct btrfs_fs_info *fs_info,
926 root->commit_root = NULL;
927 root->sectorsize = sectorsize;
928 root->nodesize = nodesize;
929 root->leafsize = leafsize;
930 root->stripesize = stripesize;
932 root->track_dirty = 0;
934 root->orphan_item_inserted = 0;
935 root->orphan_cleanup_state = 0;
937 root->fs_info = fs_info;
938 root->objectid = objectid;
939 root->last_trans = 0;
940 root->highest_objectid = 0;
943 root->inode_tree = RB_ROOT;
944 root->block_rsv = NULL;
945 root->orphan_block_rsv = NULL;
947 INIT_LIST_HEAD(&root->dirty_list);
948 INIT_LIST_HEAD(&root->orphan_list);
949 INIT_LIST_HEAD(&root->root_list);
950 spin_lock_init(&root->node_lock);
951 spin_lock_init(&root->orphan_lock);
952 spin_lock_init(&root->inode_lock);
953 spin_lock_init(&root->accounting_lock);
954 mutex_init(&root->objectid_mutex);
955 mutex_init(&root->log_mutex);
956 init_waitqueue_head(&root->log_writer_wait);
957 init_waitqueue_head(&root->log_commit_wait[0]);
958 init_waitqueue_head(&root->log_commit_wait[1]);
959 atomic_set(&root->log_commit[0], 0);
960 atomic_set(&root->log_commit[1], 0);
961 atomic_set(&root->log_writers, 0);
963 root->log_transid = 0;
964 root->last_log_commit = 0;
965 extent_io_tree_init(&root->dirty_log_pages,
966 fs_info->btree_inode->i_mapping, GFP_NOFS);
968 memset(&root->root_key, 0, sizeof(root->root_key));
969 memset(&root->root_item, 0, sizeof(root->root_item));
970 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
971 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
972 root->defrag_trans_start = fs_info->generation;
973 init_completion(&root->kobj_unregister);
974 root->defrag_running = 0;
975 root->root_key.objectid = objectid;
976 root->anon_super.s_root = NULL;
977 root->anon_super.s_dev = 0;
978 INIT_LIST_HEAD(&root->anon_super.s_list);
979 INIT_LIST_HEAD(&root->anon_super.s_instances);
980 init_rwsem(&root->anon_super.s_umount);
985 static int find_and_setup_root(struct btrfs_root *tree_root,
986 struct btrfs_fs_info *fs_info,
988 struct btrfs_root *root)
994 __setup_root(tree_root->nodesize, tree_root->leafsize,
995 tree_root->sectorsize, tree_root->stripesize,
996 root, fs_info, objectid);
997 ret = btrfs_find_last_root(tree_root, objectid,
998 &root->root_item, &root->root_key);
1003 generation = btrfs_root_generation(&root->root_item);
1004 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1005 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1006 blocksize, generation);
1007 BUG_ON(!root->node);
1008 root->commit_root = btrfs_root_node(root);
1012 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1013 struct btrfs_fs_info *fs_info)
1015 struct btrfs_root *root;
1016 struct btrfs_root *tree_root = fs_info->tree_root;
1017 struct extent_buffer *leaf;
1019 root = kzalloc(sizeof(*root), GFP_NOFS);
1021 return ERR_PTR(-ENOMEM);
1023 __setup_root(tree_root->nodesize, tree_root->leafsize,
1024 tree_root->sectorsize, tree_root->stripesize,
1025 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1027 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1028 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1029 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1031 * log trees do not get reference counted because they go away
1032 * before a real commit is actually done. They do store pointers
1033 * to file data extents, and those reference counts still get
1034 * updated (along with back refs to the log tree).
1038 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1039 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1042 return ERR_CAST(leaf);
1045 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1046 btrfs_set_header_bytenr(leaf, leaf->start);
1047 btrfs_set_header_generation(leaf, trans->transid);
1048 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1049 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1052 write_extent_buffer(root->node, root->fs_info->fsid,
1053 (unsigned long)btrfs_header_fsid(root->node),
1055 btrfs_mark_buffer_dirty(root->node);
1056 btrfs_tree_unlock(root->node);
1060 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1061 struct btrfs_fs_info *fs_info)
1063 struct btrfs_root *log_root;
1065 log_root = alloc_log_tree(trans, fs_info);
1066 if (IS_ERR(log_root))
1067 return PTR_ERR(log_root);
1068 WARN_ON(fs_info->log_root_tree);
1069 fs_info->log_root_tree = log_root;
1073 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1074 struct btrfs_root *root)
1076 struct btrfs_root *log_root;
1077 struct btrfs_inode_item *inode_item;
1079 log_root = alloc_log_tree(trans, root->fs_info);
1080 if (IS_ERR(log_root))
1081 return PTR_ERR(log_root);
1083 log_root->last_trans = trans->transid;
1084 log_root->root_key.offset = root->root_key.objectid;
1086 inode_item = &log_root->root_item.inode;
1087 inode_item->generation = cpu_to_le64(1);
1088 inode_item->size = cpu_to_le64(3);
1089 inode_item->nlink = cpu_to_le32(1);
1090 inode_item->nbytes = cpu_to_le64(root->leafsize);
1091 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1093 btrfs_set_root_node(&log_root->root_item, log_root->node);
1095 WARN_ON(root->log_root);
1096 root->log_root = log_root;
1097 root->log_transid = 0;
1098 root->last_log_commit = 0;
1102 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1103 struct btrfs_key *location)
1105 struct btrfs_root *root;
1106 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1107 struct btrfs_path *path;
1108 struct extent_buffer *l;
1113 root = kzalloc(sizeof(*root), GFP_NOFS);
1115 return ERR_PTR(-ENOMEM);
1116 if (location->offset == (u64)-1) {
1117 ret = find_and_setup_root(tree_root, fs_info,
1118 location->objectid, root);
1121 return ERR_PTR(ret);
1126 __setup_root(tree_root->nodesize, tree_root->leafsize,
1127 tree_root->sectorsize, tree_root->stripesize,
1128 root, fs_info, location->objectid);
1130 path = btrfs_alloc_path();
1132 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1135 read_extent_buffer(l, &root->root_item,
1136 btrfs_item_ptr_offset(l, path->slots[0]),
1137 sizeof(root->root_item));
1138 memcpy(&root->root_key, location, sizeof(*location));
1140 btrfs_free_path(path);
1144 return ERR_PTR(ret);
1147 generation = btrfs_root_generation(&root->root_item);
1148 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1149 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1150 blocksize, generation);
1151 root->commit_root = btrfs_root_node(root);
1152 BUG_ON(!root->node);
1154 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1160 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1163 struct btrfs_root *root;
1165 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1166 return fs_info->tree_root;
1167 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1168 return fs_info->extent_root;
1170 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1171 (unsigned long)root_objectid);
1175 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1176 struct btrfs_key *location)
1178 struct btrfs_root *root;
1181 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1182 return fs_info->tree_root;
1183 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1184 return fs_info->extent_root;
1185 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1186 return fs_info->chunk_root;
1187 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1188 return fs_info->dev_root;
1189 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1190 return fs_info->csum_root;
1192 spin_lock(&fs_info->fs_roots_radix_lock);
1193 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1194 (unsigned long)location->objectid);
1195 spin_unlock(&fs_info->fs_roots_radix_lock);
1199 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1203 set_anon_super(&root->anon_super, NULL);
1205 if (btrfs_root_refs(&root->root_item) == 0) {
1210 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1214 root->orphan_item_inserted = 1;
1216 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1220 spin_lock(&fs_info->fs_roots_radix_lock);
1221 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222 (unsigned long)root->root_key.objectid,
1227 spin_unlock(&fs_info->fs_roots_radix_lock);
1228 radix_tree_preload_end();
1230 if (ret == -EEXIST) {
1237 ret = btrfs_find_dead_roots(fs_info->tree_root,
1238 root->root_key.objectid);
1243 return ERR_PTR(ret);
1246 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1247 struct btrfs_key *location,
1248 const char *name, int namelen)
1250 return btrfs_read_fs_root_no_name(fs_info, location);
1252 struct btrfs_root *root;
1255 root = btrfs_read_fs_root_no_name(fs_info, location);
1262 ret = btrfs_set_root_name(root, name, namelen);
1264 free_extent_buffer(root->node);
1266 return ERR_PTR(ret);
1269 ret = btrfs_sysfs_add_root(root);
1271 free_extent_buffer(root->node);
1274 return ERR_PTR(ret);
1281 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1283 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1285 struct btrfs_device *device;
1286 struct backing_dev_info *bdi;
1288 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1291 bdi = blk_get_backing_dev_info(device->bdev);
1292 if (bdi && bdi_congested(bdi, bdi_bits)) {
1301 * this unplugs every device on the box, and it is only used when page
1304 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1306 struct btrfs_device *device;
1307 struct btrfs_fs_info *info;
1309 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1310 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1314 bdi = blk_get_backing_dev_info(device->bdev);
1315 if (bdi->unplug_io_fn)
1316 bdi->unplug_io_fn(bdi, page);
1320 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1322 struct inode *inode;
1323 struct extent_map_tree *em_tree;
1324 struct extent_map *em;
1325 struct address_space *mapping;
1328 /* the generic O_DIRECT read code does this */
1330 __unplug_io_fn(bdi, page);
1335 * page->mapping may change at any time. Get a consistent copy
1336 * and use that for everything below
1339 mapping = page->mapping;
1343 inode = mapping->host;
1346 * don't do the expensive searching for a small number of
1349 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1350 __unplug_io_fn(bdi, page);
1354 offset = page_offset(page);
1356 em_tree = &BTRFS_I(inode)->extent_tree;
1357 read_lock(&em_tree->lock);
1358 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1359 read_unlock(&em_tree->lock);
1361 __unplug_io_fn(bdi, page);
1365 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1366 free_extent_map(em);
1367 __unplug_io_fn(bdi, page);
1370 offset = offset - em->start;
1371 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1372 em->block_start + offset, page);
1373 free_extent_map(em);
1377 * If this fails, caller must call bdi_destroy() to get rid of the
1380 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1384 bdi->capabilities = BDI_CAP_MAP_COPY;
1385 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1389 bdi->ra_pages = default_backing_dev_info.ra_pages;
1390 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1391 bdi->unplug_io_data = info;
1392 bdi->congested_fn = btrfs_congested_fn;
1393 bdi->congested_data = info;
1397 static int bio_ready_for_csum(struct bio *bio)
1403 struct extent_io_tree *io_tree = NULL;
1404 struct bio_vec *bvec;
1408 bio_for_each_segment(bvec, bio, i) {
1409 page = bvec->bv_page;
1410 if (page->private == EXTENT_PAGE_PRIVATE) {
1411 length += bvec->bv_len;
1414 if (!page->private) {
1415 length += bvec->bv_len;
1418 length = bvec->bv_len;
1419 buf_len = page->private >> 2;
1420 start = page_offset(page) + bvec->bv_offset;
1421 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1423 /* are we fully contained in this bio? */
1424 if (buf_len <= length)
1427 ret = extent_range_uptodate(io_tree, start + length,
1428 start + buf_len - 1);
1433 * called by the kthread helper functions to finally call the bio end_io
1434 * functions. This is where read checksum verification actually happens
1436 static void end_workqueue_fn(struct btrfs_work *work)
1439 struct end_io_wq *end_io_wq;
1440 struct btrfs_fs_info *fs_info;
1443 end_io_wq = container_of(work, struct end_io_wq, work);
1444 bio = end_io_wq->bio;
1445 fs_info = end_io_wq->info;
1447 /* metadata bio reads are special because the whole tree block must
1448 * be checksummed at once. This makes sure the entire block is in
1449 * ram and up to date before trying to verify things. For
1450 * blocksize <= pagesize, it is basically a noop
1452 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1453 !bio_ready_for_csum(bio)) {
1454 btrfs_queue_worker(&fs_info->endio_meta_workers,
1458 error = end_io_wq->error;
1459 bio->bi_private = end_io_wq->private;
1460 bio->bi_end_io = end_io_wq->end_io;
1462 bio_endio(bio, error);
1465 static int cleaner_kthread(void *arg)
1467 struct btrfs_root *root = arg;
1470 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1472 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1473 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1474 btrfs_run_delayed_iputs(root);
1475 btrfs_clean_old_snapshots(root);
1476 mutex_unlock(&root->fs_info->cleaner_mutex);
1479 if (freezing(current)) {
1482 set_current_state(TASK_INTERRUPTIBLE);
1483 if (!kthread_should_stop())
1485 __set_current_state(TASK_RUNNING);
1487 } while (!kthread_should_stop());
1491 static int transaction_kthread(void *arg)
1493 struct btrfs_root *root = arg;
1494 struct btrfs_trans_handle *trans;
1495 struct btrfs_transaction *cur;
1498 unsigned long delay;
1503 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1504 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1506 spin_lock(&root->fs_info->new_trans_lock);
1507 cur = root->fs_info->running_transaction;
1509 spin_unlock(&root->fs_info->new_trans_lock);
1513 now = get_seconds();
1514 if (!cur->blocked &&
1515 (now < cur->start_time || now - cur->start_time < 30)) {
1516 spin_unlock(&root->fs_info->new_trans_lock);
1520 transid = cur->transid;
1521 spin_unlock(&root->fs_info->new_trans_lock);
1523 trans = btrfs_join_transaction(root, 1);
1524 if (transid == trans->transid) {
1525 ret = btrfs_commit_transaction(trans, root);
1528 btrfs_end_transaction(trans, root);
1531 wake_up_process(root->fs_info->cleaner_kthread);
1532 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1534 if (freezing(current)) {
1537 set_current_state(TASK_INTERRUPTIBLE);
1538 if (!kthread_should_stop() &&
1539 !btrfs_transaction_blocked(root->fs_info))
1540 schedule_timeout(delay);
1541 __set_current_state(TASK_RUNNING);
1543 } while (!kthread_should_stop());
1547 struct btrfs_root *open_ctree(struct super_block *sb,
1548 struct btrfs_fs_devices *fs_devices,
1558 struct btrfs_key location;
1559 struct buffer_head *bh;
1560 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1562 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1564 struct btrfs_root *tree_root = btrfs_sb(sb);
1565 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1566 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1568 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1570 struct btrfs_root *log_tree_root;
1575 struct btrfs_super_block *disk_super;
1577 if (!extent_root || !tree_root || !fs_info ||
1578 !chunk_root || !dev_root || !csum_root) {
1583 ret = init_srcu_struct(&fs_info->subvol_srcu);
1589 ret = setup_bdi(fs_info, &fs_info->bdi);
1595 fs_info->btree_inode = new_inode(sb);
1596 if (!fs_info->btree_inode) {
1601 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1602 INIT_LIST_HEAD(&fs_info->trans_list);
1603 INIT_LIST_HEAD(&fs_info->dead_roots);
1604 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1605 INIT_LIST_HEAD(&fs_info->hashers);
1606 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1607 INIT_LIST_HEAD(&fs_info->ordered_operations);
1608 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1609 spin_lock_init(&fs_info->delalloc_lock);
1610 spin_lock_init(&fs_info->new_trans_lock);
1611 spin_lock_init(&fs_info->ref_cache_lock);
1612 spin_lock_init(&fs_info->fs_roots_radix_lock);
1613 spin_lock_init(&fs_info->delayed_iput_lock);
1615 init_completion(&fs_info->kobj_unregister);
1616 fs_info->tree_root = tree_root;
1617 fs_info->extent_root = extent_root;
1618 fs_info->csum_root = csum_root;
1619 fs_info->chunk_root = chunk_root;
1620 fs_info->dev_root = dev_root;
1621 fs_info->fs_devices = fs_devices;
1622 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1623 INIT_LIST_HEAD(&fs_info->space_info);
1624 btrfs_mapping_init(&fs_info->mapping_tree);
1625 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1626 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1627 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1628 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1629 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1630 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1631 mutex_init(&fs_info->durable_block_rsv_mutex);
1632 atomic_set(&fs_info->nr_async_submits, 0);
1633 atomic_set(&fs_info->async_delalloc_pages, 0);
1634 atomic_set(&fs_info->async_submit_draining, 0);
1635 atomic_set(&fs_info->nr_async_bios, 0);
1637 fs_info->max_inline = 8192 * 1024;
1638 fs_info->metadata_ratio = 0;
1640 fs_info->thread_pool_size = min_t(unsigned long,
1641 num_online_cpus() + 2, 8);
1643 INIT_LIST_HEAD(&fs_info->ordered_extents);
1644 spin_lock_init(&fs_info->ordered_extent_lock);
1646 sb->s_blocksize = 4096;
1647 sb->s_blocksize_bits = blksize_bits(4096);
1648 sb->s_bdi = &fs_info->bdi;
1650 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1651 fs_info->btree_inode->i_nlink = 1;
1653 * we set the i_size on the btree inode to the max possible int.
1654 * the real end of the address space is determined by all of
1655 * the devices in the system
1657 fs_info->btree_inode->i_size = OFFSET_MAX;
1658 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1659 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1661 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1662 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1663 fs_info->btree_inode->i_mapping,
1665 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1668 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1670 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1671 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1672 sizeof(struct btrfs_key));
1673 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1674 insert_inode_hash(fs_info->btree_inode);
1676 spin_lock_init(&fs_info->block_group_cache_lock);
1677 fs_info->block_group_cache_tree = RB_ROOT;
1679 extent_io_tree_init(&fs_info->freed_extents[0],
1680 fs_info->btree_inode->i_mapping, GFP_NOFS);
1681 extent_io_tree_init(&fs_info->freed_extents[1],
1682 fs_info->btree_inode->i_mapping, GFP_NOFS);
1683 fs_info->pinned_extents = &fs_info->freed_extents[0];
1684 fs_info->do_barriers = 1;
1687 mutex_init(&fs_info->trans_mutex);
1688 mutex_init(&fs_info->ordered_operations_mutex);
1689 mutex_init(&fs_info->tree_log_mutex);
1690 mutex_init(&fs_info->chunk_mutex);
1691 mutex_init(&fs_info->transaction_kthread_mutex);
1692 mutex_init(&fs_info->cleaner_mutex);
1693 mutex_init(&fs_info->volume_mutex);
1694 init_rwsem(&fs_info->extent_commit_sem);
1695 init_rwsem(&fs_info->cleanup_work_sem);
1696 init_rwsem(&fs_info->subvol_sem);
1698 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1699 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1701 init_waitqueue_head(&fs_info->transaction_throttle);
1702 init_waitqueue_head(&fs_info->transaction_wait);
1703 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1704 init_waitqueue_head(&fs_info->async_submit_wait);
1706 __setup_root(4096, 4096, 4096, 4096, tree_root,
1707 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1709 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1713 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1714 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1715 sizeof(fs_info->super_for_commit));
1718 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1720 disk_super = &fs_info->super_copy;
1721 if (!btrfs_super_root(disk_super))
1724 ret = btrfs_parse_options(tree_root, options);
1730 features = btrfs_super_incompat_flags(disk_super) &
1731 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1733 printk(KERN_ERR "BTRFS: couldn't mount because of "
1734 "unsupported optional features (%Lx).\n",
1735 (unsigned long long)features);
1740 features = btrfs_super_incompat_flags(disk_super);
1741 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1742 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1743 btrfs_set_super_incompat_flags(disk_super, features);
1746 features = btrfs_super_compat_ro_flags(disk_super) &
1747 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1748 if (!(sb->s_flags & MS_RDONLY) && features) {
1749 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1750 "unsupported option features (%Lx).\n",
1751 (unsigned long long)features);
1756 btrfs_init_workers(&fs_info->generic_worker,
1757 "genwork", 1, NULL);
1759 btrfs_init_workers(&fs_info->workers, "worker",
1760 fs_info->thread_pool_size,
1761 &fs_info->generic_worker);
1763 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1764 fs_info->thread_pool_size,
1765 &fs_info->generic_worker);
1767 btrfs_init_workers(&fs_info->submit_workers, "submit",
1768 min_t(u64, fs_devices->num_devices,
1769 fs_info->thread_pool_size),
1770 &fs_info->generic_worker);
1772 /* a higher idle thresh on the submit workers makes it much more
1773 * likely that bios will be send down in a sane order to the
1776 fs_info->submit_workers.idle_thresh = 64;
1778 fs_info->workers.idle_thresh = 16;
1779 fs_info->workers.ordered = 1;
1781 fs_info->delalloc_workers.idle_thresh = 2;
1782 fs_info->delalloc_workers.ordered = 1;
1784 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1785 &fs_info->generic_worker);
1786 btrfs_init_workers(&fs_info->endio_workers, "endio",
1787 fs_info->thread_pool_size,
1788 &fs_info->generic_worker);
1789 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1790 fs_info->thread_pool_size,
1791 &fs_info->generic_worker);
1792 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1793 "endio-meta-write", fs_info->thread_pool_size,
1794 &fs_info->generic_worker);
1795 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1796 fs_info->thread_pool_size,
1797 &fs_info->generic_worker);
1798 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1799 1, &fs_info->generic_worker);
1802 * endios are largely parallel and should have a very
1805 fs_info->endio_workers.idle_thresh = 4;
1806 fs_info->endio_meta_workers.idle_thresh = 4;
1808 fs_info->endio_write_workers.idle_thresh = 2;
1809 fs_info->endio_meta_write_workers.idle_thresh = 2;
1811 btrfs_start_workers(&fs_info->workers, 1);
1812 btrfs_start_workers(&fs_info->generic_worker, 1);
1813 btrfs_start_workers(&fs_info->submit_workers, 1);
1814 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1815 btrfs_start_workers(&fs_info->fixup_workers, 1);
1816 btrfs_start_workers(&fs_info->endio_workers, 1);
1817 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1818 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1819 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1820 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1822 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1823 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1824 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1826 nodesize = btrfs_super_nodesize(disk_super);
1827 leafsize = btrfs_super_leafsize(disk_super);
1828 sectorsize = btrfs_super_sectorsize(disk_super);
1829 stripesize = btrfs_super_stripesize(disk_super);
1830 tree_root->nodesize = nodesize;
1831 tree_root->leafsize = leafsize;
1832 tree_root->sectorsize = sectorsize;
1833 tree_root->stripesize = stripesize;
1835 sb->s_blocksize = sectorsize;
1836 sb->s_blocksize_bits = blksize_bits(sectorsize);
1838 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1839 sizeof(disk_super->magic))) {
1840 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1841 goto fail_sb_buffer;
1844 mutex_lock(&fs_info->chunk_mutex);
1845 ret = btrfs_read_sys_array(tree_root);
1846 mutex_unlock(&fs_info->chunk_mutex);
1848 printk(KERN_WARNING "btrfs: failed to read the system "
1849 "array on %s\n", sb->s_id);
1850 goto fail_sb_buffer;
1853 blocksize = btrfs_level_size(tree_root,
1854 btrfs_super_chunk_root_level(disk_super));
1855 generation = btrfs_super_chunk_root_generation(disk_super);
1857 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1858 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1860 chunk_root->node = read_tree_block(chunk_root,
1861 btrfs_super_chunk_root(disk_super),
1862 blocksize, generation);
1863 BUG_ON(!chunk_root->node);
1864 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1865 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1867 goto fail_chunk_root;
1869 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1870 chunk_root->commit_root = btrfs_root_node(chunk_root);
1872 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1873 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1876 mutex_lock(&fs_info->chunk_mutex);
1877 ret = btrfs_read_chunk_tree(chunk_root);
1878 mutex_unlock(&fs_info->chunk_mutex);
1880 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1882 goto fail_chunk_root;
1885 btrfs_close_extra_devices(fs_devices);
1887 blocksize = btrfs_level_size(tree_root,
1888 btrfs_super_root_level(disk_super));
1889 generation = btrfs_super_generation(disk_super);
1891 tree_root->node = read_tree_block(tree_root,
1892 btrfs_super_root(disk_super),
1893 blocksize, generation);
1894 if (!tree_root->node)
1895 goto fail_chunk_root;
1896 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1897 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1899 goto fail_tree_root;
1901 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1902 tree_root->commit_root = btrfs_root_node(tree_root);
1904 ret = find_and_setup_root(tree_root, fs_info,
1905 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1907 goto fail_tree_root;
1908 extent_root->track_dirty = 1;
1910 ret = find_and_setup_root(tree_root, fs_info,
1911 BTRFS_DEV_TREE_OBJECTID, dev_root);
1913 goto fail_extent_root;
1914 dev_root->track_dirty = 1;
1916 ret = find_and_setup_root(tree_root, fs_info,
1917 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1921 csum_root->track_dirty = 1;
1923 fs_info->generation = generation;
1924 fs_info->last_trans_committed = generation;
1925 fs_info->data_alloc_profile = (u64)-1;
1926 fs_info->metadata_alloc_profile = (u64)-1;
1927 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1929 ret = btrfs_read_block_groups(extent_root);
1931 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1932 goto fail_block_groups;
1935 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1937 if (IS_ERR(fs_info->cleaner_kthread))
1938 goto fail_block_groups;
1940 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1942 "btrfs-transaction");
1943 if (IS_ERR(fs_info->transaction_kthread))
1946 if (!btrfs_test_opt(tree_root, SSD) &&
1947 !btrfs_test_opt(tree_root, NOSSD) &&
1948 !fs_info->fs_devices->rotating) {
1949 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1951 btrfs_set_opt(fs_info->mount_opt, SSD);
1954 if (btrfs_super_log_root(disk_super) != 0) {
1955 u64 bytenr = btrfs_super_log_root(disk_super);
1957 if (fs_devices->rw_devices == 0) {
1958 printk(KERN_WARNING "Btrfs log replay required "
1961 goto fail_trans_kthread;
1964 btrfs_level_size(tree_root,
1965 btrfs_super_log_root_level(disk_super));
1967 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1968 if (!log_tree_root) {
1970 goto fail_trans_kthread;
1973 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1974 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1976 log_tree_root->node = read_tree_block(tree_root, bytenr,
1979 ret = btrfs_recover_log_trees(log_tree_root);
1982 if (sb->s_flags & MS_RDONLY) {
1983 ret = btrfs_commit_super(tree_root);
1988 ret = btrfs_find_orphan_roots(tree_root);
1991 if (!(sb->s_flags & MS_RDONLY)) {
1992 ret = btrfs_cleanup_fs_roots(fs_info);
1995 ret = btrfs_recover_relocation(tree_root);
1998 "btrfs: failed to recover relocation\n");
2000 goto fail_trans_kthread;
2004 location.objectid = BTRFS_FS_TREE_OBJECTID;
2005 location.type = BTRFS_ROOT_ITEM_KEY;
2006 location.offset = (u64)-1;
2008 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2009 if (!fs_info->fs_root)
2010 goto fail_trans_kthread;
2011 if (IS_ERR(fs_info->fs_root)) {
2012 err = PTR_ERR(fs_info->fs_root);
2013 goto fail_trans_kthread;
2016 if (!(sb->s_flags & MS_RDONLY)) {
2017 down_read(&fs_info->cleanup_work_sem);
2018 btrfs_orphan_cleanup(fs_info->fs_root);
2019 btrfs_orphan_cleanup(fs_info->tree_root);
2020 up_read(&fs_info->cleanup_work_sem);
2026 kthread_stop(fs_info->transaction_kthread);
2028 kthread_stop(fs_info->cleaner_kthread);
2031 * make sure we're done with the btree inode before we stop our
2034 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2035 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2038 btrfs_free_block_groups(fs_info);
2039 free_extent_buffer(csum_root->node);
2040 free_extent_buffer(csum_root->commit_root);
2042 free_extent_buffer(dev_root->node);
2043 free_extent_buffer(dev_root->commit_root);
2045 free_extent_buffer(extent_root->node);
2046 free_extent_buffer(extent_root->commit_root);
2048 free_extent_buffer(tree_root->node);
2049 free_extent_buffer(tree_root->commit_root);
2051 free_extent_buffer(chunk_root->node);
2052 free_extent_buffer(chunk_root->commit_root);
2054 btrfs_stop_workers(&fs_info->generic_worker);
2055 btrfs_stop_workers(&fs_info->fixup_workers);
2056 btrfs_stop_workers(&fs_info->delalloc_workers);
2057 btrfs_stop_workers(&fs_info->workers);
2058 btrfs_stop_workers(&fs_info->endio_workers);
2059 btrfs_stop_workers(&fs_info->endio_meta_workers);
2060 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2061 btrfs_stop_workers(&fs_info->endio_write_workers);
2062 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2063 btrfs_stop_workers(&fs_info->submit_workers);
2065 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2066 iput(fs_info->btree_inode);
2068 btrfs_close_devices(fs_info->fs_devices);
2069 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2071 bdi_destroy(&fs_info->bdi);
2073 cleanup_srcu_struct(&fs_info->subvol_srcu);
2081 return ERR_PTR(err);
2084 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2086 char b[BDEVNAME_SIZE];
2089 set_buffer_uptodate(bh);
2091 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2092 printk(KERN_WARNING "lost page write due to "
2093 "I/O error on %s\n",
2094 bdevname(bh->b_bdev, b));
2096 /* note, we dont' set_buffer_write_io_error because we have
2097 * our own ways of dealing with the IO errors
2099 clear_buffer_uptodate(bh);
2105 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2107 struct buffer_head *bh;
2108 struct buffer_head *latest = NULL;
2109 struct btrfs_super_block *super;
2114 /* we would like to check all the supers, but that would make
2115 * a btrfs mount succeed after a mkfs from a different FS.
2116 * So, we need to add a special mount option to scan for
2117 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2119 for (i = 0; i < 1; i++) {
2120 bytenr = btrfs_sb_offset(i);
2121 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2123 bh = __bread(bdev, bytenr / 4096, 4096);
2127 super = (struct btrfs_super_block *)bh->b_data;
2128 if (btrfs_super_bytenr(super) != bytenr ||
2129 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2130 sizeof(super->magic))) {
2135 if (!latest || btrfs_super_generation(super) > transid) {
2138 transid = btrfs_super_generation(super);
2147 * this should be called twice, once with wait == 0 and
2148 * once with wait == 1. When wait == 0 is done, all the buffer heads
2149 * we write are pinned.
2151 * They are released when wait == 1 is done.
2152 * max_mirrors must be the same for both runs, and it indicates how
2153 * many supers on this one device should be written.
2155 * max_mirrors == 0 means to write them all.
2157 static int write_dev_supers(struct btrfs_device *device,
2158 struct btrfs_super_block *sb,
2159 int do_barriers, int wait, int max_mirrors)
2161 struct buffer_head *bh;
2167 int last_barrier = 0;
2169 if (max_mirrors == 0)
2170 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2172 /* make sure only the last submit_bh does a barrier */
2174 for (i = 0; i < max_mirrors; i++) {
2175 bytenr = btrfs_sb_offset(i);
2176 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2177 device->total_bytes)
2183 for (i = 0; i < max_mirrors; i++) {
2184 bytenr = btrfs_sb_offset(i);
2185 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2189 bh = __find_get_block(device->bdev, bytenr / 4096,
2190 BTRFS_SUPER_INFO_SIZE);
2193 if (!buffer_uptodate(bh))
2196 /* drop our reference */
2199 /* drop the reference from the wait == 0 run */
2203 btrfs_set_super_bytenr(sb, bytenr);
2206 crc = btrfs_csum_data(NULL, (char *)sb +
2207 BTRFS_CSUM_SIZE, crc,
2208 BTRFS_SUPER_INFO_SIZE -
2210 btrfs_csum_final(crc, sb->csum);
2213 * one reference for us, and we leave it for the
2216 bh = __getblk(device->bdev, bytenr / 4096,
2217 BTRFS_SUPER_INFO_SIZE);
2218 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2220 /* one reference for submit_bh */
2223 set_buffer_uptodate(bh);
2225 bh->b_end_io = btrfs_end_buffer_write_sync;
2228 if (i == last_barrier && do_barriers && device->barriers) {
2229 ret = submit_bh(WRITE_BARRIER, bh);
2230 if (ret == -EOPNOTSUPP) {
2231 printk("btrfs: disabling barriers on dev %s\n",
2233 set_buffer_uptodate(bh);
2234 device->barriers = 0;
2235 /* one reference for submit_bh */
2238 ret = submit_bh(WRITE_SYNC, bh);
2241 ret = submit_bh(WRITE_SYNC, bh);
2247 return errors < i ? 0 : -1;
2250 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2252 struct list_head *head;
2253 struct btrfs_device *dev;
2254 struct btrfs_super_block *sb;
2255 struct btrfs_dev_item *dev_item;
2259 int total_errors = 0;
2262 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2263 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2265 sb = &root->fs_info->super_for_commit;
2266 dev_item = &sb->dev_item;
2268 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2269 head = &root->fs_info->fs_devices->devices;
2270 list_for_each_entry(dev, head, dev_list) {
2275 if (!dev->in_fs_metadata || !dev->writeable)
2278 btrfs_set_stack_device_generation(dev_item, 0);
2279 btrfs_set_stack_device_type(dev_item, dev->type);
2280 btrfs_set_stack_device_id(dev_item, dev->devid);
2281 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2282 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2283 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2284 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2285 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2286 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2287 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2289 flags = btrfs_super_flags(sb);
2290 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2292 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2296 if (total_errors > max_errors) {
2297 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2303 list_for_each_entry(dev, head, dev_list) {
2306 if (!dev->in_fs_metadata || !dev->writeable)
2309 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2313 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2314 if (total_errors > max_errors) {
2315 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2322 int write_ctree_super(struct btrfs_trans_handle *trans,
2323 struct btrfs_root *root, int max_mirrors)
2327 ret = write_all_supers(root, max_mirrors);
2331 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2333 spin_lock(&fs_info->fs_roots_radix_lock);
2334 radix_tree_delete(&fs_info->fs_roots_radix,
2335 (unsigned long)root->root_key.objectid);
2336 spin_unlock(&fs_info->fs_roots_radix_lock);
2338 if (btrfs_root_refs(&root->root_item) == 0)
2339 synchronize_srcu(&fs_info->subvol_srcu);
2345 static void free_fs_root(struct btrfs_root *root)
2347 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2348 if (root->anon_super.s_dev) {
2349 down_write(&root->anon_super.s_umount);
2350 kill_anon_super(&root->anon_super);
2352 free_extent_buffer(root->node);
2353 free_extent_buffer(root->commit_root);
2358 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2361 struct btrfs_root *gang[8];
2364 while (!list_empty(&fs_info->dead_roots)) {
2365 gang[0] = list_entry(fs_info->dead_roots.next,
2366 struct btrfs_root, root_list);
2367 list_del(&gang[0]->root_list);
2369 if (gang[0]->in_radix) {
2370 btrfs_free_fs_root(fs_info, gang[0]);
2372 free_extent_buffer(gang[0]->node);
2373 free_extent_buffer(gang[0]->commit_root);
2379 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2384 for (i = 0; i < ret; i++)
2385 btrfs_free_fs_root(fs_info, gang[i]);
2390 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2392 u64 root_objectid = 0;
2393 struct btrfs_root *gang[8];
2398 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2399 (void **)gang, root_objectid,
2404 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2405 for (i = 0; i < ret; i++) {
2406 root_objectid = gang[i]->root_key.objectid;
2407 btrfs_orphan_cleanup(gang[i]);
2414 int btrfs_commit_super(struct btrfs_root *root)
2416 struct btrfs_trans_handle *trans;
2419 mutex_lock(&root->fs_info->cleaner_mutex);
2420 btrfs_run_delayed_iputs(root);
2421 btrfs_clean_old_snapshots(root);
2422 mutex_unlock(&root->fs_info->cleaner_mutex);
2424 /* wait until ongoing cleanup work done */
2425 down_write(&root->fs_info->cleanup_work_sem);
2426 up_write(&root->fs_info->cleanup_work_sem);
2428 trans = btrfs_join_transaction(root, 1);
2429 ret = btrfs_commit_transaction(trans, root);
2431 /* run commit again to drop the original snapshot */
2432 trans = btrfs_join_transaction(root, 1);
2433 btrfs_commit_transaction(trans, root);
2434 ret = btrfs_write_and_wait_transaction(NULL, root);
2437 ret = write_ctree_super(NULL, root, 0);
2441 int close_ctree(struct btrfs_root *root)
2443 struct btrfs_fs_info *fs_info = root->fs_info;
2446 fs_info->closing = 1;
2449 btrfs_put_block_group_cache(fs_info);
2450 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2451 ret = btrfs_commit_super(root);
2453 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2456 kthread_stop(root->fs_info->transaction_kthread);
2457 kthread_stop(root->fs_info->cleaner_kthread);
2459 fs_info->closing = 2;
2462 if (fs_info->delalloc_bytes) {
2463 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2464 (unsigned long long)fs_info->delalloc_bytes);
2466 if (fs_info->total_ref_cache_size) {
2467 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2468 (unsigned long long)fs_info->total_ref_cache_size);
2471 free_extent_buffer(fs_info->extent_root->node);
2472 free_extent_buffer(fs_info->extent_root->commit_root);
2473 free_extent_buffer(fs_info->tree_root->node);
2474 free_extent_buffer(fs_info->tree_root->commit_root);
2475 free_extent_buffer(root->fs_info->chunk_root->node);
2476 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2477 free_extent_buffer(root->fs_info->dev_root->node);
2478 free_extent_buffer(root->fs_info->dev_root->commit_root);
2479 free_extent_buffer(root->fs_info->csum_root->node);
2480 free_extent_buffer(root->fs_info->csum_root->commit_root);
2482 btrfs_free_block_groups(root->fs_info);
2484 del_fs_roots(fs_info);
2486 iput(fs_info->btree_inode);
2488 btrfs_stop_workers(&fs_info->generic_worker);
2489 btrfs_stop_workers(&fs_info->fixup_workers);
2490 btrfs_stop_workers(&fs_info->delalloc_workers);
2491 btrfs_stop_workers(&fs_info->workers);
2492 btrfs_stop_workers(&fs_info->endio_workers);
2493 btrfs_stop_workers(&fs_info->endio_meta_workers);
2494 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2495 btrfs_stop_workers(&fs_info->endio_write_workers);
2496 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2497 btrfs_stop_workers(&fs_info->submit_workers);
2499 btrfs_close_devices(fs_info->fs_devices);
2500 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2502 bdi_destroy(&fs_info->bdi);
2503 cleanup_srcu_struct(&fs_info->subvol_srcu);
2505 kfree(fs_info->extent_root);
2506 kfree(fs_info->tree_root);
2507 kfree(fs_info->chunk_root);
2508 kfree(fs_info->dev_root);
2509 kfree(fs_info->csum_root);
2513 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2516 struct inode *btree_inode = buf->first_page->mapping->host;
2518 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2523 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2528 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2530 struct inode *btree_inode = buf->first_page->mapping->host;
2531 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2535 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2537 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2538 u64 transid = btrfs_header_generation(buf);
2539 struct inode *btree_inode = root->fs_info->btree_inode;
2542 btrfs_assert_tree_locked(buf);
2543 if (transid != root->fs_info->generation) {
2544 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2545 "found %llu running %llu\n",
2546 (unsigned long long)buf->start,
2547 (unsigned long long)transid,
2548 (unsigned long long)root->fs_info->generation);
2551 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2554 spin_lock(&root->fs_info->delalloc_lock);
2555 root->fs_info->dirty_metadata_bytes += buf->len;
2556 spin_unlock(&root->fs_info->delalloc_lock);
2560 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2563 * looks as though older kernels can get into trouble with
2564 * this code, they end up stuck in balance_dirty_pages forever
2567 unsigned long thresh = 32 * 1024 * 1024;
2569 if (current->flags & PF_MEMALLOC)
2572 num_dirty = root->fs_info->dirty_metadata_bytes;
2574 if (num_dirty > thresh) {
2575 balance_dirty_pages_ratelimited_nr(
2576 root->fs_info->btree_inode->i_mapping, 1);
2581 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2583 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2585 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2587 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2591 int btree_lock_page_hook(struct page *page)
2593 struct inode *inode = page->mapping->host;
2594 struct btrfs_root *root = BTRFS_I(inode)->root;
2595 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2596 struct extent_buffer *eb;
2598 u64 bytenr = page_offset(page);
2600 if (page->private == EXTENT_PAGE_PRIVATE)
2603 len = page->private >> 2;
2604 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2608 btrfs_tree_lock(eb);
2609 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2611 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2612 spin_lock(&root->fs_info->delalloc_lock);
2613 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2614 root->fs_info->dirty_metadata_bytes -= eb->len;
2617 spin_unlock(&root->fs_info->delalloc_lock);
2620 btrfs_tree_unlock(eb);
2621 free_extent_buffer(eb);
2627 static struct extent_io_ops btree_extent_io_ops = {
2628 .write_cache_pages_lock_hook = btree_lock_page_hook,
2629 .readpage_end_io_hook = btree_readpage_end_io_hook,
2630 .submit_bio_hook = btree_submit_bio_hook,
2631 /* note we're sharing with inode.c for the merge bio hook */
2632 .merge_bio_hook = btrfs_merge_bio_hook,