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);
360 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
361 btrfs_header_generation(eb));
363 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
365 found_start = btrfs_header_bytenr(eb);
366 if (found_start != start) {
370 if (eb->first_page != page) {
374 if (!PageUptodate(page)) {
378 csum_tree_block(root, eb, 0);
380 free_extent_buffer(eb);
385 static int check_tree_block_fsid(struct btrfs_root *root,
386 struct extent_buffer *eb)
388 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
389 u8 fsid[BTRFS_UUID_SIZE];
392 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
395 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
399 fs_devices = fs_devices->seed;
404 #ifdef CONFIG_DEBUG_LOCK_ALLOC
405 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
407 lockdep_set_class_and_name(&eb->lock,
408 &btrfs_eb_class[level],
409 btrfs_eb_name[level]);
413 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
414 struct extent_state *state)
416 struct extent_io_tree *tree;
420 struct extent_buffer *eb;
421 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
424 tree = &BTRFS_I(page->mapping->host)->io_tree;
425 if (page->private == EXTENT_PAGE_PRIVATE)
430 len = page->private >> 2;
433 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
439 found_start = btrfs_header_bytenr(eb);
440 if (found_start != start) {
441 if (printk_ratelimit()) {
442 printk(KERN_INFO "btrfs bad tree block start "
444 (unsigned long long)found_start,
445 (unsigned long long)eb->start);
450 if (eb->first_page != page) {
451 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
452 eb->first_page->index, page->index);
457 if (check_tree_block_fsid(root, eb)) {
458 if (printk_ratelimit()) {
459 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
460 (unsigned long long)eb->start);
465 found_level = btrfs_header_level(eb);
467 btrfs_set_buffer_lockdep_class(eb, found_level);
469 ret = csum_tree_block(root, eb, 1);
473 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
474 end = eb->start + end - 1;
476 free_extent_buffer(eb);
481 static void end_workqueue_bio(struct bio *bio, int err)
483 struct end_io_wq *end_io_wq = bio->bi_private;
484 struct btrfs_fs_info *fs_info;
486 fs_info = end_io_wq->info;
487 end_io_wq->error = err;
488 end_io_wq->work.func = end_workqueue_fn;
489 end_io_wq->work.flags = 0;
491 if (bio->bi_rw & REQ_WRITE) {
492 if (end_io_wq->metadata == 1)
493 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
495 else if (end_io_wq->metadata == 2)
496 btrfs_queue_worker(&fs_info->endio_freespace_worker,
499 btrfs_queue_worker(&fs_info->endio_write_workers,
502 if (end_io_wq->metadata)
503 btrfs_queue_worker(&fs_info->endio_meta_workers,
506 btrfs_queue_worker(&fs_info->endio_workers,
512 * For the metadata arg you want
515 * 1 - if normal metadta
516 * 2 - if writing to the free space cache area
518 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
521 struct end_io_wq *end_io_wq;
522 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
526 end_io_wq->private = bio->bi_private;
527 end_io_wq->end_io = bio->bi_end_io;
528 end_io_wq->info = info;
529 end_io_wq->error = 0;
530 end_io_wq->bio = bio;
531 end_io_wq->metadata = metadata;
533 bio->bi_private = end_io_wq;
534 bio->bi_end_io = end_workqueue_bio;
538 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
540 unsigned long limit = min_t(unsigned long,
541 info->workers.max_workers,
542 info->fs_devices->open_devices);
546 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
548 return atomic_read(&info->nr_async_bios) >
549 btrfs_async_submit_limit(info);
552 static void run_one_async_start(struct btrfs_work *work)
554 struct async_submit_bio *async;
556 async = container_of(work, struct async_submit_bio, work);
557 async->submit_bio_start(async->inode, async->rw, async->bio,
558 async->mirror_num, async->bio_flags,
562 static void run_one_async_done(struct btrfs_work *work)
564 struct btrfs_fs_info *fs_info;
565 struct async_submit_bio *async;
568 async = container_of(work, struct async_submit_bio, work);
569 fs_info = BTRFS_I(async->inode)->root->fs_info;
571 limit = btrfs_async_submit_limit(fs_info);
572 limit = limit * 2 / 3;
574 atomic_dec(&fs_info->nr_async_submits);
576 if (atomic_read(&fs_info->nr_async_submits) < limit &&
577 waitqueue_active(&fs_info->async_submit_wait))
578 wake_up(&fs_info->async_submit_wait);
580 async->submit_bio_done(async->inode, async->rw, async->bio,
581 async->mirror_num, async->bio_flags,
585 static void run_one_async_free(struct btrfs_work *work)
587 struct async_submit_bio *async;
589 async = container_of(work, struct async_submit_bio, work);
593 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
594 int rw, struct bio *bio, int mirror_num,
595 unsigned long bio_flags,
597 extent_submit_bio_hook_t *submit_bio_start,
598 extent_submit_bio_hook_t *submit_bio_done)
600 struct async_submit_bio *async;
602 async = kmalloc(sizeof(*async), GFP_NOFS);
606 async->inode = inode;
609 async->mirror_num = mirror_num;
610 async->submit_bio_start = submit_bio_start;
611 async->submit_bio_done = submit_bio_done;
613 async->work.func = run_one_async_start;
614 async->work.ordered_func = run_one_async_done;
615 async->work.ordered_free = run_one_async_free;
617 async->work.flags = 0;
618 async->bio_flags = bio_flags;
619 async->bio_offset = bio_offset;
621 atomic_inc(&fs_info->nr_async_submits);
624 btrfs_set_work_high_prio(&async->work);
626 btrfs_queue_worker(&fs_info->workers, &async->work);
628 while (atomic_read(&fs_info->async_submit_draining) &&
629 atomic_read(&fs_info->nr_async_submits)) {
630 wait_event(fs_info->async_submit_wait,
631 (atomic_read(&fs_info->nr_async_submits) == 0));
637 static int btree_csum_one_bio(struct bio *bio)
639 struct bio_vec *bvec = bio->bi_io_vec;
641 struct btrfs_root *root;
643 WARN_ON(bio->bi_vcnt <= 0);
644 while (bio_index < bio->bi_vcnt) {
645 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
646 csum_dirty_buffer(root, bvec->bv_page);
653 static int __btree_submit_bio_start(struct inode *inode, int rw,
654 struct bio *bio, int mirror_num,
655 unsigned long bio_flags,
659 * when we're called for a write, we're already in the async
660 * submission context. Just jump into btrfs_map_bio
662 btree_csum_one_bio(bio);
666 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
667 int mirror_num, unsigned long bio_flags,
671 * when we're called for a write, we're already in the async
672 * submission context. Just jump into btrfs_map_bio
674 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
677 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
678 int mirror_num, unsigned long bio_flags,
683 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
687 if (!(rw & REQ_WRITE)) {
689 * called for a read, do the setup so that checksum validation
690 * can happen in the async kernel threads
692 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
697 * kthread helpers are used to submit writes so that checksumming
698 * can happen in parallel across all CPUs
700 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
701 inode, rw, bio, mirror_num, 0,
703 __btree_submit_bio_start,
704 __btree_submit_bio_done);
707 #ifdef CONFIG_MIGRATION
708 static int btree_migratepage(struct address_space *mapping,
709 struct page *newpage, struct page *page)
712 * we can't safely write a btree page from here,
713 * we haven't done the locking hook
718 * Buffers may be managed in a filesystem specific way.
719 * We must have no buffers or drop them.
721 if (page_has_private(page) &&
722 !try_to_release_page(page, GFP_KERNEL))
724 return migrate_page(mapping, newpage, page);
728 static int btree_writepage(struct page *page, struct writeback_control *wbc)
730 struct extent_io_tree *tree;
731 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
732 struct extent_buffer *eb;
735 tree = &BTRFS_I(page->mapping->host)->io_tree;
736 if (!(current->flags & PF_MEMALLOC)) {
737 return extent_write_full_page(tree, page,
738 btree_get_extent, wbc);
741 redirty_page_for_writepage(wbc, page);
742 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
745 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
747 spin_lock(&root->fs_info->delalloc_lock);
748 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
749 spin_unlock(&root->fs_info->delalloc_lock);
751 free_extent_buffer(eb);
757 static int btree_writepages(struct address_space *mapping,
758 struct writeback_control *wbc)
760 struct extent_io_tree *tree;
761 tree = &BTRFS_I(mapping->host)->io_tree;
762 if (wbc->sync_mode == WB_SYNC_NONE) {
763 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
765 unsigned long thresh = 32 * 1024 * 1024;
767 if (wbc->for_kupdate)
770 /* this is a bit racy, but that's ok */
771 num_dirty = root->fs_info->dirty_metadata_bytes;
772 if (num_dirty < thresh)
775 return extent_writepages(tree, mapping, btree_get_extent, wbc);
778 static int btree_readpage(struct file *file, struct page *page)
780 struct extent_io_tree *tree;
781 tree = &BTRFS_I(page->mapping->host)->io_tree;
782 return extent_read_full_page(tree, page, btree_get_extent);
785 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
787 struct extent_io_tree *tree;
788 struct extent_map_tree *map;
791 if (PageWriteback(page) || PageDirty(page))
794 tree = &BTRFS_I(page->mapping->host)->io_tree;
795 map = &BTRFS_I(page->mapping->host)->extent_tree;
797 ret = try_release_extent_state(map, tree, page, gfp_flags);
801 ret = try_release_extent_buffer(tree, page);
803 ClearPagePrivate(page);
804 set_page_private(page, 0);
805 page_cache_release(page);
811 static void btree_invalidatepage(struct page *page, unsigned long offset)
813 struct extent_io_tree *tree;
814 tree = &BTRFS_I(page->mapping->host)->io_tree;
815 extent_invalidatepage(tree, page, offset);
816 btree_releasepage(page, GFP_NOFS);
817 if (PagePrivate(page)) {
818 printk(KERN_WARNING "btrfs warning page private not zero "
819 "on page %llu\n", (unsigned long long)page_offset(page));
820 ClearPagePrivate(page);
821 set_page_private(page, 0);
822 page_cache_release(page);
826 static const struct address_space_operations btree_aops = {
827 .readpage = btree_readpage,
828 .writepage = btree_writepage,
829 .writepages = btree_writepages,
830 .releasepage = btree_releasepage,
831 .invalidatepage = btree_invalidatepage,
832 .sync_page = block_sync_page,
833 #ifdef CONFIG_MIGRATION
834 .migratepage = btree_migratepage,
838 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
841 struct extent_buffer *buf = NULL;
842 struct inode *btree_inode = root->fs_info->btree_inode;
845 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
848 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
849 buf, 0, 0, btree_get_extent, 0);
850 free_extent_buffer(buf);
854 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
855 u64 bytenr, u32 blocksize)
857 struct inode *btree_inode = root->fs_info->btree_inode;
858 struct extent_buffer *eb;
859 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
860 bytenr, blocksize, GFP_NOFS);
864 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
865 u64 bytenr, u32 blocksize)
867 struct inode *btree_inode = root->fs_info->btree_inode;
868 struct extent_buffer *eb;
870 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
871 bytenr, blocksize, NULL, GFP_NOFS);
876 int btrfs_write_tree_block(struct extent_buffer *buf)
878 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
879 buf->start + buf->len - 1);
882 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
884 return filemap_fdatawait_range(buf->first_page->mapping,
885 buf->start, buf->start + buf->len - 1);
888 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
889 u32 blocksize, u64 parent_transid)
891 struct extent_buffer *buf = NULL;
894 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
898 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
901 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
906 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
907 struct extent_buffer *buf)
909 struct inode *btree_inode = root->fs_info->btree_inode;
910 if (btrfs_header_generation(buf) ==
911 root->fs_info->running_transaction->transid) {
912 btrfs_assert_tree_locked(buf);
914 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
915 spin_lock(&root->fs_info->delalloc_lock);
916 if (root->fs_info->dirty_metadata_bytes >= buf->len)
917 root->fs_info->dirty_metadata_bytes -= buf->len;
920 spin_unlock(&root->fs_info->delalloc_lock);
923 /* ugh, clear_extent_buffer_dirty needs to lock the page */
924 btrfs_set_lock_blocking(buf);
925 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
931 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
932 u32 stripesize, struct btrfs_root *root,
933 struct btrfs_fs_info *fs_info,
937 root->commit_root = NULL;
938 root->sectorsize = sectorsize;
939 root->nodesize = nodesize;
940 root->leafsize = leafsize;
941 root->stripesize = stripesize;
943 root->track_dirty = 0;
945 root->orphan_item_inserted = 0;
946 root->orphan_cleanup_state = 0;
948 root->fs_info = fs_info;
949 root->objectid = objectid;
950 root->last_trans = 0;
951 root->highest_objectid = 0;
954 root->inode_tree = RB_ROOT;
955 root->block_rsv = NULL;
956 root->orphan_block_rsv = NULL;
958 INIT_LIST_HEAD(&root->dirty_list);
959 INIT_LIST_HEAD(&root->orphan_list);
960 INIT_LIST_HEAD(&root->root_list);
961 spin_lock_init(&root->node_lock);
962 spin_lock_init(&root->orphan_lock);
963 spin_lock_init(&root->inode_lock);
964 spin_lock_init(&root->accounting_lock);
965 mutex_init(&root->objectid_mutex);
966 mutex_init(&root->log_mutex);
967 init_waitqueue_head(&root->log_writer_wait);
968 init_waitqueue_head(&root->log_commit_wait[0]);
969 init_waitqueue_head(&root->log_commit_wait[1]);
970 atomic_set(&root->log_commit[0], 0);
971 atomic_set(&root->log_commit[1], 0);
972 atomic_set(&root->log_writers, 0);
974 root->log_transid = 0;
975 root->last_log_commit = 0;
976 extent_io_tree_init(&root->dirty_log_pages,
977 fs_info->btree_inode->i_mapping, GFP_NOFS);
979 memset(&root->root_key, 0, sizeof(root->root_key));
980 memset(&root->root_item, 0, sizeof(root->root_item));
981 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
982 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
983 root->defrag_trans_start = fs_info->generation;
984 init_completion(&root->kobj_unregister);
985 root->defrag_running = 0;
986 root->root_key.objectid = objectid;
987 root->anon_super.s_root = NULL;
988 root->anon_super.s_dev = 0;
989 INIT_LIST_HEAD(&root->anon_super.s_list);
990 INIT_LIST_HEAD(&root->anon_super.s_instances);
991 init_rwsem(&root->anon_super.s_umount);
996 static int find_and_setup_root(struct btrfs_root *tree_root,
997 struct btrfs_fs_info *fs_info,
999 struct btrfs_root *root)
1005 __setup_root(tree_root->nodesize, tree_root->leafsize,
1006 tree_root->sectorsize, tree_root->stripesize,
1007 root, fs_info, objectid);
1008 ret = btrfs_find_last_root(tree_root, objectid,
1009 &root->root_item, &root->root_key);
1014 generation = btrfs_root_generation(&root->root_item);
1015 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1016 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1017 blocksize, generation);
1018 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1019 free_extent_buffer(root->node);
1022 root->commit_root = btrfs_root_node(root);
1026 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1027 struct btrfs_fs_info *fs_info)
1029 struct btrfs_root *root;
1030 struct btrfs_root *tree_root = fs_info->tree_root;
1031 struct extent_buffer *leaf;
1033 root = kzalloc(sizeof(*root), GFP_NOFS);
1035 return ERR_PTR(-ENOMEM);
1037 __setup_root(tree_root->nodesize, tree_root->leafsize,
1038 tree_root->sectorsize, tree_root->stripesize,
1039 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1041 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1042 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1043 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1045 * log trees do not get reference counted because they go away
1046 * before a real commit is actually done. They do store pointers
1047 * to file data extents, and those reference counts still get
1048 * updated (along with back refs to the log tree).
1052 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1053 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1056 return ERR_CAST(leaf);
1059 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1060 btrfs_set_header_bytenr(leaf, leaf->start);
1061 btrfs_set_header_generation(leaf, trans->transid);
1062 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1063 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1066 write_extent_buffer(root->node, root->fs_info->fsid,
1067 (unsigned long)btrfs_header_fsid(root->node),
1069 btrfs_mark_buffer_dirty(root->node);
1070 btrfs_tree_unlock(root->node);
1074 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1075 struct btrfs_fs_info *fs_info)
1077 struct btrfs_root *log_root;
1079 log_root = alloc_log_tree(trans, fs_info);
1080 if (IS_ERR(log_root))
1081 return PTR_ERR(log_root);
1082 WARN_ON(fs_info->log_root_tree);
1083 fs_info->log_root_tree = log_root;
1087 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1088 struct btrfs_root *root)
1090 struct btrfs_root *log_root;
1091 struct btrfs_inode_item *inode_item;
1093 log_root = alloc_log_tree(trans, root->fs_info);
1094 if (IS_ERR(log_root))
1095 return PTR_ERR(log_root);
1097 log_root->last_trans = trans->transid;
1098 log_root->root_key.offset = root->root_key.objectid;
1100 inode_item = &log_root->root_item.inode;
1101 inode_item->generation = cpu_to_le64(1);
1102 inode_item->size = cpu_to_le64(3);
1103 inode_item->nlink = cpu_to_le32(1);
1104 inode_item->nbytes = cpu_to_le64(root->leafsize);
1105 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1107 btrfs_set_root_node(&log_root->root_item, log_root->node);
1109 WARN_ON(root->log_root);
1110 root->log_root = log_root;
1111 root->log_transid = 0;
1112 root->last_log_commit = 0;
1116 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1117 struct btrfs_key *location)
1119 struct btrfs_root *root;
1120 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1121 struct btrfs_path *path;
1122 struct extent_buffer *l;
1127 root = kzalloc(sizeof(*root), GFP_NOFS);
1129 return ERR_PTR(-ENOMEM);
1130 if (location->offset == (u64)-1) {
1131 ret = find_and_setup_root(tree_root, fs_info,
1132 location->objectid, root);
1135 return ERR_PTR(ret);
1140 __setup_root(tree_root->nodesize, tree_root->leafsize,
1141 tree_root->sectorsize, tree_root->stripesize,
1142 root, fs_info, location->objectid);
1144 path = btrfs_alloc_path();
1146 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1149 read_extent_buffer(l, &root->root_item,
1150 btrfs_item_ptr_offset(l, path->slots[0]),
1151 sizeof(root->root_item));
1152 memcpy(&root->root_key, location, sizeof(*location));
1154 btrfs_free_path(path);
1159 return ERR_PTR(ret);
1162 generation = btrfs_root_generation(&root->root_item);
1163 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1164 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1165 blocksize, generation);
1166 root->commit_root = btrfs_root_node(root);
1167 BUG_ON(!root->node);
1169 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1175 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1178 struct btrfs_root *root;
1180 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1181 return fs_info->tree_root;
1182 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1183 return fs_info->extent_root;
1185 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1186 (unsigned long)root_objectid);
1190 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1191 struct btrfs_key *location)
1193 struct btrfs_root *root;
1196 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1197 return fs_info->tree_root;
1198 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1199 return fs_info->extent_root;
1200 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1201 return fs_info->chunk_root;
1202 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1203 return fs_info->dev_root;
1204 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1205 return fs_info->csum_root;
1207 spin_lock(&fs_info->fs_roots_radix_lock);
1208 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1209 (unsigned long)location->objectid);
1210 spin_unlock(&fs_info->fs_roots_radix_lock);
1214 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1218 set_anon_super(&root->anon_super, NULL);
1220 if (btrfs_root_refs(&root->root_item) == 0) {
1225 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1229 root->orphan_item_inserted = 1;
1231 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1235 spin_lock(&fs_info->fs_roots_radix_lock);
1236 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1237 (unsigned long)root->root_key.objectid,
1242 spin_unlock(&fs_info->fs_roots_radix_lock);
1243 radix_tree_preload_end();
1245 if (ret == -EEXIST) {
1252 ret = btrfs_find_dead_roots(fs_info->tree_root,
1253 root->root_key.objectid);
1258 return ERR_PTR(ret);
1261 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1262 struct btrfs_key *location,
1263 const char *name, int namelen)
1265 return btrfs_read_fs_root_no_name(fs_info, location);
1267 struct btrfs_root *root;
1270 root = btrfs_read_fs_root_no_name(fs_info, location);
1277 ret = btrfs_set_root_name(root, name, namelen);
1279 free_extent_buffer(root->node);
1281 return ERR_PTR(ret);
1284 ret = btrfs_sysfs_add_root(root);
1286 free_extent_buffer(root->node);
1289 return ERR_PTR(ret);
1296 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1298 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1300 struct btrfs_device *device;
1301 struct backing_dev_info *bdi;
1303 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1306 bdi = blk_get_backing_dev_info(device->bdev);
1307 if (bdi && bdi_congested(bdi, bdi_bits)) {
1316 * this unplugs every device on the box, and it is only used when page
1319 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1321 struct btrfs_device *device;
1322 struct btrfs_fs_info *info;
1324 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1325 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1329 bdi = blk_get_backing_dev_info(device->bdev);
1330 if (bdi->unplug_io_fn)
1331 bdi->unplug_io_fn(bdi, page);
1335 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1337 struct inode *inode;
1338 struct extent_map_tree *em_tree;
1339 struct extent_map *em;
1340 struct address_space *mapping;
1343 /* the generic O_DIRECT read code does this */
1345 __unplug_io_fn(bdi, page);
1350 * page->mapping may change at any time. Get a consistent copy
1351 * and use that for everything below
1354 mapping = page->mapping;
1358 inode = mapping->host;
1361 * don't do the expensive searching for a small number of
1364 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1365 __unplug_io_fn(bdi, page);
1369 offset = page_offset(page);
1371 em_tree = &BTRFS_I(inode)->extent_tree;
1372 read_lock(&em_tree->lock);
1373 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1374 read_unlock(&em_tree->lock);
1376 __unplug_io_fn(bdi, page);
1380 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1381 free_extent_map(em);
1382 __unplug_io_fn(bdi, page);
1385 offset = offset - em->start;
1386 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1387 em->block_start + offset, page);
1388 free_extent_map(em);
1392 * If this fails, caller must call bdi_destroy() to get rid of the
1395 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1399 bdi->capabilities = BDI_CAP_MAP_COPY;
1400 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1404 bdi->ra_pages = default_backing_dev_info.ra_pages;
1405 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1406 bdi->unplug_io_data = info;
1407 bdi->congested_fn = btrfs_congested_fn;
1408 bdi->congested_data = info;
1412 static int bio_ready_for_csum(struct bio *bio)
1418 struct extent_io_tree *io_tree = NULL;
1419 struct bio_vec *bvec;
1423 bio_for_each_segment(bvec, bio, i) {
1424 page = bvec->bv_page;
1425 if (page->private == EXTENT_PAGE_PRIVATE) {
1426 length += bvec->bv_len;
1429 if (!page->private) {
1430 length += bvec->bv_len;
1433 length = bvec->bv_len;
1434 buf_len = page->private >> 2;
1435 start = page_offset(page) + bvec->bv_offset;
1436 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1438 /* are we fully contained in this bio? */
1439 if (buf_len <= length)
1442 ret = extent_range_uptodate(io_tree, start + length,
1443 start + buf_len - 1);
1448 * called by the kthread helper functions to finally call the bio end_io
1449 * functions. This is where read checksum verification actually happens
1451 static void end_workqueue_fn(struct btrfs_work *work)
1454 struct end_io_wq *end_io_wq;
1455 struct btrfs_fs_info *fs_info;
1458 end_io_wq = container_of(work, struct end_io_wq, work);
1459 bio = end_io_wq->bio;
1460 fs_info = end_io_wq->info;
1462 /* metadata bio reads are special because the whole tree block must
1463 * be checksummed at once. This makes sure the entire block is in
1464 * ram and up to date before trying to verify things. For
1465 * blocksize <= pagesize, it is basically a noop
1467 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1468 !bio_ready_for_csum(bio)) {
1469 btrfs_queue_worker(&fs_info->endio_meta_workers,
1473 error = end_io_wq->error;
1474 bio->bi_private = end_io_wq->private;
1475 bio->bi_end_io = end_io_wq->end_io;
1477 bio_endio(bio, error);
1480 static int cleaner_kthread(void *arg)
1482 struct btrfs_root *root = arg;
1485 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1487 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1488 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1489 btrfs_run_delayed_iputs(root);
1490 btrfs_clean_old_snapshots(root);
1491 mutex_unlock(&root->fs_info->cleaner_mutex);
1494 if (freezing(current)) {
1497 set_current_state(TASK_INTERRUPTIBLE);
1498 if (!kthread_should_stop())
1500 __set_current_state(TASK_RUNNING);
1502 } while (!kthread_should_stop());
1506 static int transaction_kthread(void *arg)
1508 struct btrfs_root *root = arg;
1509 struct btrfs_trans_handle *trans;
1510 struct btrfs_transaction *cur;
1513 unsigned long delay;
1518 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1519 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1521 spin_lock(&root->fs_info->new_trans_lock);
1522 cur = root->fs_info->running_transaction;
1524 spin_unlock(&root->fs_info->new_trans_lock);
1528 now = get_seconds();
1529 if (!cur->blocked &&
1530 (now < cur->start_time || now - cur->start_time < 30)) {
1531 spin_unlock(&root->fs_info->new_trans_lock);
1535 transid = cur->transid;
1536 spin_unlock(&root->fs_info->new_trans_lock);
1538 trans = btrfs_join_transaction(root, 1);
1539 if (transid == trans->transid) {
1540 ret = btrfs_commit_transaction(trans, root);
1543 btrfs_end_transaction(trans, root);
1546 wake_up_process(root->fs_info->cleaner_kthread);
1547 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1549 if (freezing(current)) {
1552 set_current_state(TASK_INTERRUPTIBLE);
1553 if (!kthread_should_stop() &&
1554 !btrfs_transaction_blocked(root->fs_info))
1555 schedule_timeout(delay);
1556 __set_current_state(TASK_RUNNING);
1558 } while (!kthread_should_stop());
1562 struct btrfs_root *open_ctree(struct super_block *sb,
1563 struct btrfs_fs_devices *fs_devices,
1573 struct btrfs_key location;
1574 struct buffer_head *bh;
1575 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1577 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1579 struct btrfs_root *tree_root = btrfs_sb(sb);
1580 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1581 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1583 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1585 struct btrfs_root *log_tree_root;
1590 struct btrfs_super_block *disk_super;
1592 if (!extent_root || !tree_root || !fs_info ||
1593 !chunk_root || !dev_root || !csum_root) {
1598 ret = init_srcu_struct(&fs_info->subvol_srcu);
1604 ret = setup_bdi(fs_info, &fs_info->bdi);
1610 fs_info->btree_inode = new_inode(sb);
1611 if (!fs_info->btree_inode) {
1616 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1617 INIT_LIST_HEAD(&fs_info->trans_list);
1618 INIT_LIST_HEAD(&fs_info->dead_roots);
1619 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1620 INIT_LIST_HEAD(&fs_info->hashers);
1621 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1622 INIT_LIST_HEAD(&fs_info->ordered_operations);
1623 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1624 spin_lock_init(&fs_info->delalloc_lock);
1625 spin_lock_init(&fs_info->new_trans_lock);
1626 spin_lock_init(&fs_info->ref_cache_lock);
1627 spin_lock_init(&fs_info->fs_roots_radix_lock);
1628 spin_lock_init(&fs_info->delayed_iput_lock);
1630 init_completion(&fs_info->kobj_unregister);
1631 fs_info->tree_root = tree_root;
1632 fs_info->extent_root = extent_root;
1633 fs_info->csum_root = csum_root;
1634 fs_info->chunk_root = chunk_root;
1635 fs_info->dev_root = dev_root;
1636 fs_info->fs_devices = fs_devices;
1637 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1638 INIT_LIST_HEAD(&fs_info->space_info);
1639 btrfs_mapping_init(&fs_info->mapping_tree);
1640 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1641 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1642 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1643 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1644 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1645 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1646 mutex_init(&fs_info->durable_block_rsv_mutex);
1647 atomic_set(&fs_info->nr_async_submits, 0);
1648 atomic_set(&fs_info->async_delalloc_pages, 0);
1649 atomic_set(&fs_info->async_submit_draining, 0);
1650 atomic_set(&fs_info->nr_async_bios, 0);
1652 fs_info->max_inline = 8192 * 1024;
1653 fs_info->metadata_ratio = 0;
1655 fs_info->thread_pool_size = min_t(unsigned long,
1656 num_online_cpus() + 2, 8);
1658 INIT_LIST_HEAD(&fs_info->ordered_extents);
1659 spin_lock_init(&fs_info->ordered_extent_lock);
1661 sb->s_blocksize = 4096;
1662 sb->s_blocksize_bits = blksize_bits(4096);
1663 sb->s_bdi = &fs_info->bdi;
1665 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1666 fs_info->btree_inode->i_nlink = 1;
1668 * we set the i_size on the btree inode to the max possible int.
1669 * the real end of the address space is determined by all of
1670 * the devices in the system
1672 fs_info->btree_inode->i_size = OFFSET_MAX;
1673 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1674 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1676 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1677 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1678 fs_info->btree_inode->i_mapping,
1680 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1683 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1685 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1686 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1687 sizeof(struct btrfs_key));
1688 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1689 insert_inode_hash(fs_info->btree_inode);
1691 spin_lock_init(&fs_info->block_group_cache_lock);
1692 fs_info->block_group_cache_tree = RB_ROOT;
1694 extent_io_tree_init(&fs_info->freed_extents[0],
1695 fs_info->btree_inode->i_mapping, GFP_NOFS);
1696 extent_io_tree_init(&fs_info->freed_extents[1],
1697 fs_info->btree_inode->i_mapping, GFP_NOFS);
1698 fs_info->pinned_extents = &fs_info->freed_extents[0];
1699 fs_info->do_barriers = 1;
1702 mutex_init(&fs_info->trans_mutex);
1703 mutex_init(&fs_info->ordered_operations_mutex);
1704 mutex_init(&fs_info->tree_log_mutex);
1705 mutex_init(&fs_info->chunk_mutex);
1706 mutex_init(&fs_info->transaction_kthread_mutex);
1707 mutex_init(&fs_info->cleaner_mutex);
1708 mutex_init(&fs_info->volume_mutex);
1709 init_rwsem(&fs_info->extent_commit_sem);
1710 init_rwsem(&fs_info->cleanup_work_sem);
1711 init_rwsem(&fs_info->subvol_sem);
1713 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1714 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1716 init_waitqueue_head(&fs_info->transaction_throttle);
1717 init_waitqueue_head(&fs_info->transaction_wait);
1718 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1719 init_waitqueue_head(&fs_info->async_submit_wait);
1721 __setup_root(4096, 4096, 4096, 4096, tree_root,
1722 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1724 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1730 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1731 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1732 sizeof(fs_info->super_for_commit));
1735 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1737 disk_super = &fs_info->super_copy;
1738 if (!btrfs_super_root(disk_super))
1741 ret = btrfs_parse_options(tree_root, options);
1747 features = btrfs_super_incompat_flags(disk_super) &
1748 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1750 printk(KERN_ERR "BTRFS: couldn't mount because of "
1751 "unsupported optional features (%Lx).\n",
1752 (unsigned long long)features);
1757 features = btrfs_super_incompat_flags(disk_super);
1758 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1759 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1760 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1761 btrfs_set_super_incompat_flags(disk_super, features);
1763 features = btrfs_super_compat_ro_flags(disk_super) &
1764 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1765 if (!(sb->s_flags & MS_RDONLY) && features) {
1766 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1767 "unsupported option features (%Lx).\n",
1768 (unsigned long long)features);
1773 btrfs_init_workers(&fs_info->generic_worker,
1774 "genwork", 1, NULL);
1776 btrfs_init_workers(&fs_info->workers, "worker",
1777 fs_info->thread_pool_size,
1778 &fs_info->generic_worker);
1780 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1781 fs_info->thread_pool_size,
1782 &fs_info->generic_worker);
1784 btrfs_init_workers(&fs_info->submit_workers, "submit",
1785 min_t(u64, fs_devices->num_devices,
1786 fs_info->thread_pool_size),
1787 &fs_info->generic_worker);
1789 /* a higher idle thresh on the submit workers makes it much more
1790 * likely that bios will be send down in a sane order to the
1793 fs_info->submit_workers.idle_thresh = 64;
1795 fs_info->workers.idle_thresh = 16;
1796 fs_info->workers.ordered = 1;
1798 fs_info->delalloc_workers.idle_thresh = 2;
1799 fs_info->delalloc_workers.ordered = 1;
1801 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1802 &fs_info->generic_worker);
1803 btrfs_init_workers(&fs_info->endio_workers, "endio",
1804 fs_info->thread_pool_size,
1805 &fs_info->generic_worker);
1806 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1807 fs_info->thread_pool_size,
1808 &fs_info->generic_worker);
1809 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1810 "endio-meta-write", fs_info->thread_pool_size,
1811 &fs_info->generic_worker);
1812 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1813 fs_info->thread_pool_size,
1814 &fs_info->generic_worker);
1815 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1816 1, &fs_info->generic_worker);
1819 * endios are largely parallel and should have a very
1822 fs_info->endio_workers.idle_thresh = 4;
1823 fs_info->endio_meta_workers.idle_thresh = 4;
1825 fs_info->endio_write_workers.idle_thresh = 2;
1826 fs_info->endio_meta_write_workers.idle_thresh = 2;
1828 btrfs_start_workers(&fs_info->workers, 1);
1829 btrfs_start_workers(&fs_info->generic_worker, 1);
1830 btrfs_start_workers(&fs_info->submit_workers, 1);
1831 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1832 btrfs_start_workers(&fs_info->fixup_workers, 1);
1833 btrfs_start_workers(&fs_info->endio_workers, 1);
1834 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1835 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1836 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1837 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1839 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1840 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1841 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1843 nodesize = btrfs_super_nodesize(disk_super);
1844 leafsize = btrfs_super_leafsize(disk_super);
1845 sectorsize = btrfs_super_sectorsize(disk_super);
1846 stripesize = btrfs_super_stripesize(disk_super);
1847 tree_root->nodesize = nodesize;
1848 tree_root->leafsize = leafsize;
1849 tree_root->sectorsize = sectorsize;
1850 tree_root->stripesize = stripesize;
1852 sb->s_blocksize = sectorsize;
1853 sb->s_blocksize_bits = blksize_bits(sectorsize);
1855 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1856 sizeof(disk_super->magic))) {
1857 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1858 goto fail_sb_buffer;
1861 mutex_lock(&fs_info->chunk_mutex);
1862 ret = btrfs_read_sys_array(tree_root);
1863 mutex_unlock(&fs_info->chunk_mutex);
1865 printk(KERN_WARNING "btrfs: failed to read the system "
1866 "array on %s\n", sb->s_id);
1867 goto fail_sb_buffer;
1870 blocksize = btrfs_level_size(tree_root,
1871 btrfs_super_chunk_root_level(disk_super));
1872 generation = btrfs_super_chunk_root_generation(disk_super);
1874 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1875 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1877 chunk_root->node = read_tree_block(chunk_root,
1878 btrfs_super_chunk_root(disk_super),
1879 blocksize, generation);
1880 BUG_ON(!chunk_root->node);
1881 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1882 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1884 goto fail_chunk_root;
1886 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1887 chunk_root->commit_root = btrfs_root_node(chunk_root);
1889 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1890 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1893 mutex_lock(&fs_info->chunk_mutex);
1894 ret = btrfs_read_chunk_tree(chunk_root);
1895 mutex_unlock(&fs_info->chunk_mutex);
1897 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1899 goto fail_chunk_root;
1902 btrfs_close_extra_devices(fs_devices);
1904 blocksize = btrfs_level_size(tree_root,
1905 btrfs_super_root_level(disk_super));
1906 generation = btrfs_super_generation(disk_super);
1908 tree_root->node = read_tree_block(tree_root,
1909 btrfs_super_root(disk_super),
1910 blocksize, generation);
1911 if (!tree_root->node)
1912 goto fail_chunk_root;
1913 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1914 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1916 goto fail_tree_root;
1918 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1919 tree_root->commit_root = btrfs_root_node(tree_root);
1921 ret = find_and_setup_root(tree_root, fs_info,
1922 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1924 goto fail_tree_root;
1925 extent_root->track_dirty = 1;
1927 ret = find_and_setup_root(tree_root, fs_info,
1928 BTRFS_DEV_TREE_OBJECTID, dev_root);
1930 goto fail_extent_root;
1931 dev_root->track_dirty = 1;
1933 ret = find_and_setup_root(tree_root, fs_info,
1934 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1938 csum_root->track_dirty = 1;
1940 fs_info->generation = generation;
1941 fs_info->last_trans_committed = generation;
1942 fs_info->data_alloc_profile = (u64)-1;
1943 fs_info->metadata_alloc_profile = (u64)-1;
1944 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1946 ret = btrfs_read_block_groups(extent_root);
1948 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1949 goto fail_block_groups;
1952 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1954 if (IS_ERR(fs_info->cleaner_kthread))
1955 goto fail_block_groups;
1957 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1959 "btrfs-transaction");
1960 if (IS_ERR(fs_info->transaction_kthread))
1963 if (!btrfs_test_opt(tree_root, SSD) &&
1964 !btrfs_test_opt(tree_root, NOSSD) &&
1965 !fs_info->fs_devices->rotating) {
1966 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1968 btrfs_set_opt(fs_info->mount_opt, SSD);
1971 if (btrfs_super_log_root(disk_super) != 0) {
1972 u64 bytenr = btrfs_super_log_root(disk_super);
1974 if (fs_devices->rw_devices == 0) {
1975 printk(KERN_WARNING "Btrfs log replay required "
1978 goto fail_trans_kthread;
1981 btrfs_level_size(tree_root,
1982 btrfs_super_log_root_level(disk_super));
1984 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1985 if (!log_tree_root) {
1987 goto fail_trans_kthread;
1990 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1991 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1993 log_tree_root->node = read_tree_block(tree_root, bytenr,
1996 ret = btrfs_recover_log_trees(log_tree_root);
1999 if (sb->s_flags & MS_RDONLY) {
2000 ret = btrfs_commit_super(tree_root);
2005 ret = btrfs_find_orphan_roots(tree_root);
2008 if (!(sb->s_flags & MS_RDONLY)) {
2009 ret = btrfs_cleanup_fs_roots(fs_info);
2012 ret = btrfs_recover_relocation(tree_root);
2015 "btrfs: failed to recover relocation\n");
2017 goto fail_trans_kthread;
2021 location.objectid = BTRFS_FS_TREE_OBJECTID;
2022 location.type = BTRFS_ROOT_ITEM_KEY;
2023 location.offset = (u64)-1;
2025 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2026 if (!fs_info->fs_root)
2027 goto fail_trans_kthread;
2028 if (IS_ERR(fs_info->fs_root)) {
2029 err = PTR_ERR(fs_info->fs_root);
2030 goto fail_trans_kthread;
2033 if (!(sb->s_flags & MS_RDONLY)) {
2034 down_read(&fs_info->cleanup_work_sem);
2035 btrfs_orphan_cleanup(fs_info->fs_root);
2036 btrfs_orphan_cleanup(fs_info->tree_root);
2037 up_read(&fs_info->cleanup_work_sem);
2043 kthread_stop(fs_info->transaction_kthread);
2045 kthread_stop(fs_info->cleaner_kthread);
2048 * make sure we're done with the btree inode before we stop our
2051 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2052 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2055 btrfs_free_block_groups(fs_info);
2056 free_extent_buffer(csum_root->node);
2057 free_extent_buffer(csum_root->commit_root);
2059 free_extent_buffer(dev_root->node);
2060 free_extent_buffer(dev_root->commit_root);
2062 free_extent_buffer(extent_root->node);
2063 free_extent_buffer(extent_root->commit_root);
2065 free_extent_buffer(tree_root->node);
2066 free_extent_buffer(tree_root->commit_root);
2068 free_extent_buffer(chunk_root->node);
2069 free_extent_buffer(chunk_root->commit_root);
2071 btrfs_stop_workers(&fs_info->generic_worker);
2072 btrfs_stop_workers(&fs_info->fixup_workers);
2073 btrfs_stop_workers(&fs_info->delalloc_workers);
2074 btrfs_stop_workers(&fs_info->workers);
2075 btrfs_stop_workers(&fs_info->endio_workers);
2076 btrfs_stop_workers(&fs_info->endio_meta_workers);
2077 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2078 btrfs_stop_workers(&fs_info->endio_write_workers);
2079 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2080 btrfs_stop_workers(&fs_info->submit_workers);
2082 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2083 iput(fs_info->btree_inode);
2085 btrfs_close_devices(fs_info->fs_devices);
2086 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2088 bdi_destroy(&fs_info->bdi);
2090 cleanup_srcu_struct(&fs_info->subvol_srcu);
2098 return ERR_PTR(err);
2101 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2103 char b[BDEVNAME_SIZE];
2106 set_buffer_uptodate(bh);
2108 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2109 printk(KERN_WARNING "lost page write due to "
2110 "I/O error on %s\n",
2111 bdevname(bh->b_bdev, b));
2113 /* note, we dont' set_buffer_write_io_error because we have
2114 * our own ways of dealing with the IO errors
2116 clear_buffer_uptodate(bh);
2122 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2124 struct buffer_head *bh;
2125 struct buffer_head *latest = NULL;
2126 struct btrfs_super_block *super;
2131 /* we would like to check all the supers, but that would make
2132 * a btrfs mount succeed after a mkfs from a different FS.
2133 * So, we need to add a special mount option to scan for
2134 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2136 for (i = 0; i < 1; i++) {
2137 bytenr = btrfs_sb_offset(i);
2138 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2140 bh = __bread(bdev, bytenr / 4096, 4096);
2144 super = (struct btrfs_super_block *)bh->b_data;
2145 if (btrfs_super_bytenr(super) != bytenr ||
2146 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2147 sizeof(super->magic))) {
2152 if (!latest || btrfs_super_generation(super) > transid) {
2155 transid = btrfs_super_generation(super);
2164 * this should be called twice, once with wait == 0 and
2165 * once with wait == 1. When wait == 0 is done, all the buffer heads
2166 * we write are pinned.
2168 * They are released when wait == 1 is done.
2169 * max_mirrors must be the same for both runs, and it indicates how
2170 * many supers on this one device should be written.
2172 * max_mirrors == 0 means to write them all.
2174 static int write_dev_supers(struct btrfs_device *device,
2175 struct btrfs_super_block *sb,
2176 int do_barriers, int wait, int max_mirrors)
2178 struct buffer_head *bh;
2184 int last_barrier = 0;
2186 if (max_mirrors == 0)
2187 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2189 /* make sure only the last submit_bh does a barrier */
2191 for (i = 0; i < max_mirrors; i++) {
2192 bytenr = btrfs_sb_offset(i);
2193 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2194 device->total_bytes)
2200 for (i = 0; i < max_mirrors; i++) {
2201 bytenr = btrfs_sb_offset(i);
2202 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2206 bh = __find_get_block(device->bdev, bytenr / 4096,
2207 BTRFS_SUPER_INFO_SIZE);
2210 if (!buffer_uptodate(bh))
2213 /* drop our reference */
2216 /* drop the reference from the wait == 0 run */
2220 btrfs_set_super_bytenr(sb, bytenr);
2223 crc = btrfs_csum_data(NULL, (char *)sb +
2224 BTRFS_CSUM_SIZE, crc,
2225 BTRFS_SUPER_INFO_SIZE -
2227 btrfs_csum_final(crc, sb->csum);
2230 * one reference for us, and we leave it for the
2233 bh = __getblk(device->bdev, bytenr / 4096,
2234 BTRFS_SUPER_INFO_SIZE);
2235 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2237 /* one reference for submit_bh */
2240 set_buffer_uptodate(bh);
2242 bh->b_end_io = btrfs_end_buffer_write_sync;
2245 if (i == last_barrier && do_barriers && device->barriers) {
2246 ret = submit_bh(WRITE_BARRIER, bh);
2247 if (ret == -EOPNOTSUPP) {
2248 printk("btrfs: disabling barriers on dev %s\n",
2250 set_buffer_uptodate(bh);
2251 device->barriers = 0;
2252 /* one reference for submit_bh */
2255 ret = submit_bh(WRITE_SYNC, bh);
2258 ret = submit_bh(WRITE_SYNC, bh);
2264 return errors < i ? 0 : -1;
2267 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2269 struct list_head *head;
2270 struct btrfs_device *dev;
2271 struct btrfs_super_block *sb;
2272 struct btrfs_dev_item *dev_item;
2276 int total_errors = 0;
2279 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2280 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2282 sb = &root->fs_info->super_for_commit;
2283 dev_item = &sb->dev_item;
2285 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2286 head = &root->fs_info->fs_devices->devices;
2287 list_for_each_entry(dev, head, dev_list) {
2292 if (!dev->in_fs_metadata || !dev->writeable)
2295 btrfs_set_stack_device_generation(dev_item, 0);
2296 btrfs_set_stack_device_type(dev_item, dev->type);
2297 btrfs_set_stack_device_id(dev_item, dev->devid);
2298 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2299 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2300 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2301 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2302 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2303 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2304 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2306 flags = btrfs_super_flags(sb);
2307 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2309 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2313 if (total_errors > max_errors) {
2314 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2320 list_for_each_entry(dev, head, dev_list) {
2323 if (!dev->in_fs_metadata || !dev->writeable)
2326 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2330 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2331 if (total_errors > max_errors) {
2332 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2339 int write_ctree_super(struct btrfs_trans_handle *trans,
2340 struct btrfs_root *root, int max_mirrors)
2344 ret = write_all_supers(root, max_mirrors);
2348 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2350 spin_lock(&fs_info->fs_roots_radix_lock);
2351 radix_tree_delete(&fs_info->fs_roots_radix,
2352 (unsigned long)root->root_key.objectid);
2353 spin_unlock(&fs_info->fs_roots_radix_lock);
2355 if (btrfs_root_refs(&root->root_item) == 0)
2356 synchronize_srcu(&fs_info->subvol_srcu);
2362 static void free_fs_root(struct btrfs_root *root)
2364 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2365 if (root->anon_super.s_dev) {
2366 down_write(&root->anon_super.s_umount);
2367 kill_anon_super(&root->anon_super);
2369 free_extent_buffer(root->node);
2370 free_extent_buffer(root->commit_root);
2375 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2378 struct btrfs_root *gang[8];
2381 while (!list_empty(&fs_info->dead_roots)) {
2382 gang[0] = list_entry(fs_info->dead_roots.next,
2383 struct btrfs_root, root_list);
2384 list_del(&gang[0]->root_list);
2386 if (gang[0]->in_radix) {
2387 btrfs_free_fs_root(fs_info, gang[0]);
2389 free_extent_buffer(gang[0]->node);
2390 free_extent_buffer(gang[0]->commit_root);
2396 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2401 for (i = 0; i < ret; i++)
2402 btrfs_free_fs_root(fs_info, gang[i]);
2407 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2409 u64 root_objectid = 0;
2410 struct btrfs_root *gang[8];
2415 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2416 (void **)gang, root_objectid,
2421 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2422 for (i = 0; i < ret; i++) {
2423 root_objectid = gang[i]->root_key.objectid;
2424 btrfs_orphan_cleanup(gang[i]);
2431 int btrfs_commit_super(struct btrfs_root *root)
2433 struct btrfs_trans_handle *trans;
2436 mutex_lock(&root->fs_info->cleaner_mutex);
2437 btrfs_run_delayed_iputs(root);
2438 btrfs_clean_old_snapshots(root);
2439 mutex_unlock(&root->fs_info->cleaner_mutex);
2441 /* wait until ongoing cleanup work done */
2442 down_write(&root->fs_info->cleanup_work_sem);
2443 up_write(&root->fs_info->cleanup_work_sem);
2445 trans = btrfs_join_transaction(root, 1);
2446 ret = btrfs_commit_transaction(trans, root);
2448 /* run commit again to drop the original snapshot */
2449 trans = btrfs_join_transaction(root, 1);
2450 btrfs_commit_transaction(trans, root);
2451 ret = btrfs_write_and_wait_transaction(NULL, root);
2454 ret = write_ctree_super(NULL, root, 0);
2458 int close_ctree(struct btrfs_root *root)
2460 struct btrfs_fs_info *fs_info = root->fs_info;
2463 fs_info->closing = 1;
2466 btrfs_put_block_group_cache(fs_info);
2467 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2468 ret = btrfs_commit_super(root);
2470 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2473 kthread_stop(root->fs_info->transaction_kthread);
2474 kthread_stop(root->fs_info->cleaner_kthread);
2476 fs_info->closing = 2;
2479 if (fs_info->delalloc_bytes) {
2480 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2481 (unsigned long long)fs_info->delalloc_bytes);
2483 if (fs_info->total_ref_cache_size) {
2484 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2485 (unsigned long long)fs_info->total_ref_cache_size);
2488 free_extent_buffer(fs_info->extent_root->node);
2489 free_extent_buffer(fs_info->extent_root->commit_root);
2490 free_extent_buffer(fs_info->tree_root->node);
2491 free_extent_buffer(fs_info->tree_root->commit_root);
2492 free_extent_buffer(root->fs_info->chunk_root->node);
2493 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2494 free_extent_buffer(root->fs_info->dev_root->node);
2495 free_extent_buffer(root->fs_info->dev_root->commit_root);
2496 free_extent_buffer(root->fs_info->csum_root->node);
2497 free_extent_buffer(root->fs_info->csum_root->commit_root);
2499 btrfs_free_block_groups(root->fs_info);
2501 del_fs_roots(fs_info);
2503 iput(fs_info->btree_inode);
2505 btrfs_stop_workers(&fs_info->generic_worker);
2506 btrfs_stop_workers(&fs_info->fixup_workers);
2507 btrfs_stop_workers(&fs_info->delalloc_workers);
2508 btrfs_stop_workers(&fs_info->workers);
2509 btrfs_stop_workers(&fs_info->endio_workers);
2510 btrfs_stop_workers(&fs_info->endio_meta_workers);
2511 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2512 btrfs_stop_workers(&fs_info->endio_write_workers);
2513 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2514 btrfs_stop_workers(&fs_info->submit_workers);
2516 btrfs_close_devices(fs_info->fs_devices);
2517 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2519 bdi_destroy(&fs_info->bdi);
2520 cleanup_srcu_struct(&fs_info->subvol_srcu);
2522 kfree(fs_info->extent_root);
2523 kfree(fs_info->tree_root);
2524 kfree(fs_info->chunk_root);
2525 kfree(fs_info->dev_root);
2526 kfree(fs_info->csum_root);
2530 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2533 struct inode *btree_inode = buf->first_page->mapping->host;
2535 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2540 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2545 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2547 struct inode *btree_inode = buf->first_page->mapping->host;
2548 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2552 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2554 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2555 u64 transid = btrfs_header_generation(buf);
2556 struct inode *btree_inode = root->fs_info->btree_inode;
2559 btrfs_assert_tree_locked(buf);
2560 if (transid != root->fs_info->generation) {
2561 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2562 "found %llu running %llu\n",
2563 (unsigned long long)buf->start,
2564 (unsigned long long)transid,
2565 (unsigned long long)root->fs_info->generation);
2568 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2571 spin_lock(&root->fs_info->delalloc_lock);
2572 root->fs_info->dirty_metadata_bytes += buf->len;
2573 spin_unlock(&root->fs_info->delalloc_lock);
2577 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2580 * looks as though older kernels can get into trouble with
2581 * this code, they end up stuck in balance_dirty_pages forever
2584 unsigned long thresh = 32 * 1024 * 1024;
2586 if (current->flags & PF_MEMALLOC)
2589 num_dirty = root->fs_info->dirty_metadata_bytes;
2591 if (num_dirty > thresh) {
2592 balance_dirty_pages_ratelimited_nr(
2593 root->fs_info->btree_inode->i_mapping, 1);
2598 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2600 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2602 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2604 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2608 int btree_lock_page_hook(struct page *page)
2610 struct inode *inode = page->mapping->host;
2611 struct btrfs_root *root = BTRFS_I(inode)->root;
2612 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2613 struct extent_buffer *eb;
2615 u64 bytenr = page_offset(page);
2617 if (page->private == EXTENT_PAGE_PRIVATE)
2620 len = page->private >> 2;
2621 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2625 btrfs_tree_lock(eb);
2626 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2628 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2629 spin_lock(&root->fs_info->delalloc_lock);
2630 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2631 root->fs_info->dirty_metadata_bytes -= eb->len;
2634 spin_unlock(&root->fs_info->delalloc_lock);
2637 btrfs_tree_unlock(eb);
2638 free_extent_buffer(eb);
2644 static struct extent_io_ops btree_extent_io_ops = {
2645 .write_cache_pages_lock_hook = btree_lock_page_hook,
2646 .readpage_end_io_hook = btree_readpage_end_io_hook,
2647 .submit_bio_hook = btree_submit_bio_hook,
2648 /* note we're sharing with inode.c for the merge bio hook */
2649 .merge_bio_hook = btrfs_merge_bio_hook,