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>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
47 static struct extent_io_ops btree_extent_io_ops;
48 static void end_workqueue_fn(struct btrfs_work *work);
49 static void free_fs_root(struct btrfs_root *root);
50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 struct btrfs_root *root);
56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
59 struct extent_io_tree *dirty_pages,
61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
74 struct btrfs_fs_info *info;
77 struct list_head list;
78 struct btrfs_work work;
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
86 struct async_submit_bio {
89 struct list_head list;
90 extent_submit_bio_hook_t *submit_bio_start;
91 extent_submit_bio_hook_t *submit_bio_done;
94 unsigned long bio_flags;
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
100 struct btrfs_work work;
103 /* These are used to set the lockdep class on the extent buffer locks.
104 * The class is set by the readpage_end_io_hook after the buffer has
105 * passed csum validation but before the pages are unlocked.
107 * The lockdep class is also set by btrfs_init_new_buffer on freshly
110 * The class is based on the level in the tree block, which allows lockdep
111 * to know that lower nodes nest inside the locks of higher nodes.
113 * We also add a check to make sure the highest level of the tree is
114 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
115 * code needs update as well.
117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 # if BTRFS_MAX_LEVEL != 8
121 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
122 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
132 /* highest possible level */
138 * extents on the btree inode are pretty simple, there's one extent
139 * that covers the entire device
141 static struct extent_map *btree_get_extent(struct inode *inode,
142 struct page *page, size_t pg_offset, u64 start, u64 len,
145 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
146 struct extent_map *em;
149 read_lock(&em_tree->lock);
150 em = lookup_extent_mapping(em_tree, start, len);
153 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
154 read_unlock(&em_tree->lock);
157 read_unlock(&em_tree->lock);
159 em = alloc_extent_map();
161 em = ERR_PTR(-ENOMEM);
166 em->block_len = (u64)-1;
168 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
170 write_lock(&em_tree->lock);
171 ret = add_extent_mapping(em_tree, em);
172 if (ret == -EEXIST) {
173 u64 failed_start = em->start;
174 u64 failed_len = em->len;
177 em = lookup_extent_mapping(em_tree, start, len);
181 em = lookup_extent_mapping(em_tree, failed_start,
189 write_unlock(&em_tree->lock);
197 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
199 return crc32c(seed, data, len);
202 void btrfs_csum_final(u32 crc, char *result)
204 put_unaligned_le32(~crc, result);
208 * compute the csum for a btree block, and either verify it or write it
209 * into the csum field of the block.
211 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
215 btrfs_super_csum_size(&root->fs_info->super_copy);
218 unsigned long cur_len;
219 unsigned long offset = BTRFS_CSUM_SIZE;
221 unsigned long map_start;
222 unsigned long map_len;
225 unsigned long inline_result;
227 len = buf->len - offset;
229 err = map_private_extent_buffer(buf, offset, 32,
230 &kaddr, &map_start, &map_len);
233 cur_len = min(len, map_len - (offset - map_start));
234 crc = btrfs_csum_data(root, kaddr + offset - map_start,
239 if (csum_size > sizeof(inline_result)) {
240 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
244 result = (char *)&inline_result;
247 btrfs_csum_final(crc, result);
250 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
253 memcpy(&found, result, csum_size);
255 read_extent_buffer(buf, &val, 0, csum_size);
256 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
257 "failed on %llu wanted %X found %X "
259 root->fs_info->sb->s_id,
260 (unsigned long long)buf->start, val, found,
261 btrfs_header_level(buf));
262 if (result != (char *)&inline_result)
267 write_extent_buffer(buf, result, 0, csum_size);
269 if (result != (char *)&inline_result)
275 * we can't consider a given block up to date unless the transid of the
276 * block matches the transid in the parent node's pointer. This is how we
277 * detect blocks that either didn't get written at all or got written
278 * in the wrong place.
280 static int verify_parent_transid(struct extent_io_tree *io_tree,
281 struct extent_buffer *eb, u64 parent_transid)
283 struct extent_state *cached_state = NULL;
286 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
289 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
290 0, &cached_state, GFP_NOFS);
291 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
292 btrfs_header_generation(eb) == parent_transid) {
296 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
298 (unsigned long long)eb->start,
299 (unsigned long long)parent_transid,
300 (unsigned long long)btrfs_header_generation(eb));
302 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
304 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
305 &cached_state, GFP_NOFS);
310 * helper to read a given tree block, doing retries as required when
311 * the checksums don't match and we have alternate mirrors to try.
313 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
314 struct extent_buffer *eb,
315 u64 start, u64 parent_transid)
317 struct extent_io_tree *io_tree;
322 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
323 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
325 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
326 btree_get_extent, mirror_num);
328 !verify_parent_transid(io_tree, eb, parent_transid))
332 * This buffer's crc is fine, but its contents are corrupted, so
333 * there is no reason to read the other copies, they won't be
336 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
339 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
345 if (mirror_num > num_copies)
352 * checksum a dirty tree block before IO. This has extra checks to make sure
353 * we only fill in the checksum field in the first page of a multi-page block
356 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
358 struct extent_io_tree *tree;
359 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
362 struct extent_buffer *eb;
365 tree = &BTRFS_I(page->mapping->host)->io_tree;
367 if (page->private == EXTENT_PAGE_PRIVATE) {
371 if (!page->private) {
375 len = page->private >> 2;
378 eb = alloc_extent_buffer(tree, start, len, page);
383 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
384 btrfs_header_generation(eb));
386 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
388 found_start = btrfs_header_bytenr(eb);
389 if (found_start != start) {
393 if (eb->first_page != page) {
397 if (!PageUptodate(page)) {
401 csum_tree_block(root, eb, 0);
403 free_extent_buffer(eb);
408 static int check_tree_block_fsid(struct btrfs_root *root,
409 struct extent_buffer *eb)
411 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
412 u8 fsid[BTRFS_UUID_SIZE];
415 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
418 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
422 fs_devices = fs_devices->seed;
427 #define CORRUPT(reason, eb, root, slot) \
428 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
429 "root=%llu, slot=%d\n", reason, \
430 (unsigned long long)btrfs_header_bytenr(eb), \
431 (unsigned long long)root->objectid, slot)
433 static noinline int check_leaf(struct btrfs_root *root,
434 struct extent_buffer *leaf)
436 struct btrfs_key key;
437 struct btrfs_key leaf_key;
438 u32 nritems = btrfs_header_nritems(leaf);
444 /* Check the 0 item */
445 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
446 BTRFS_LEAF_DATA_SIZE(root)) {
447 CORRUPT("invalid item offset size pair", leaf, root, 0);
452 * Check to make sure each items keys are in the correct order and their
453 * offsets make sense. We only have to loop through nritems-1 because
454 * we check the current slot against the next slot, which verifies the
455 * next slot's offset+size makes sense and that the current's slot
458 for (slot = 0; slot < nritems - 1; slot++) {
459 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
460 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
462 /* Make sure the keys are in the right order */
463 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
464 CORRUPT("bad key order", leaf, root, slot);
469 * Make sure the offset and ends are right, remember that the
470 * item data starts at the end of the leaf and grows towards the
473 if (btrfs_item_offset_nr(leaf, slot) !=
474 btrfs_item_end_nr(leaf, slot + 1)) {
475 CORRUPT("slot offset bad", leaf, root, slot);
480 * Check to make sure that we don't point outside of the leaf,
481 * just incase all the items are consistent to eachother, but
482 * all point outside of the leaf.
484 if (btrfs_item_end_nr(leaf, slot) >
485 BTRFS_LEAF_DATA_SIZE(root)) {
486 CORRUPT("slot end outside of leaf", leaf, root, slot);
494 #ifdef CONFIG_DEBUG_LOCK_ALLOC
495 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
497 lockdep_set_class_and_name(&eb->lock,
498 &btrfs_eb_class[level],
499 btrfs_eb_name[level]);
503 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
504 struct extent_state *state)
506 struct extent_io_tree *tree;
510 struct extent_buffer *eb;
511 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
514 tree = &BTRFS_I(page->mapping->host)->io_tree;
515 if (page->private == EXTENT_PAGE_PRIVATE)
520 len = page->private >> 2;
523 eb = alloc_extent_buffer(tree, start, len, page);
529 found_start = btrfs_header_bytenr(eb);
530 if (found_start != start) {
531 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
533 (unsigned long long)found_start,
534 (unsigned long long)eb->start);
538 if (eb->first_page != page) {
539 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
540 eb->first_page->index, page->index);
545 if (check_tree_block_fsid(root, eb)) {
546 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
547 (unsigned long long)eb->start);
551 found_level = btrfs_header_level(eb);
553 btrfs_set_buffer_lockdep_class(eb, found_level);
555 ret = csum_tree_block(root, eb, 1);
562 * If this is a leaf block and it is corrupt, set the corrupt bit so
563 * that we don't try and read the other copies of this block, just
566 if (found_level == 0 && check_leaf(root, eb)) {
567 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
571 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
572 end = eb->start + end - 1;
574 free_extent_buffer(eb);
579 static void end_workqueue_bio(struct bio *bio, int err)
581 struct end_io_wq *end_io_wq = bio->bi_private;
582 struct btrfs_fs_info *fs_info;
584 fs_info = end_io_wq->info;
585 end_io_wq->error = err;
586 end_io_wq->work.func = end_workqueue_fn;
587 end_io_wq->work.flags = 0;
589 if (bio->bi_rw & REQ_WRITE) {
590 if (end_io_wq->metadata == 1)
591 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
593 else if (end_io_wq->metadata == 2)
594 btrfs_queue_worker(&fs_info->endio_freespace_worker,
597 btrfs_queue_worker(&fs_info->endio_write_workers,
600 if (end_io_wq->metadata)
601 btrfs_queue_worker(&fs_info->endio_meta_workers,
604 btrfs_queue_worker(&fs_info->endio_workers,
610 * For the metadata arg you want
613 * 1 - if normal metadta
614 * 2 - if writing to the free space cache area
616 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
619 struct end_io_wq *end_io_wq;
620 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
624 end_io_wq->private = bio->bi_private;
625 end_io_wq->end_io = bio->bi_end_io;
626 end_io_wq->info = info;
627 end_io_wq->error = 0;
628 end_io_wq->bio = bio;
629 end_io_wq->metadata = metadata;
631 bio->bi_private = end_io_wq;
632 bio->bi_end_io = end_workqueue_bio;
636 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
638 unsigned long limit = min_t(unsigned long,
639 info->workers.max_workers,
640 info->fs_devices->open_devices);
644 static void run_one_async_start(struct btrfs_work *work)
646 struct async_submit_bio *async;
648 async = container_of(work, struct async_submit_bio, work);
649 async->submit_bio_start(async->inode, async->rw, async->bio,
650 async->mirror_num, async->bio_flags,
654 static void run_one_async_done(struct btrfs_work *work)
656 struct btrfs_fs_info *fs_info;
657 struct async_submit_bio *async;
660 async = container_of(work, struct async_submit_bio, work);
661 fs_info = BTRFS_I(async->inode)->root->fs_info;
663 limit = btrfs_async_submit_limit(fs_info);
664 limit = limit * 2 / 3;
666 atomic_dec(&fs_info->nr_async_submits);
668 if (atomic_read(&fs_info->nr_async_submits) < limit &&
669 waitqueue_active(&fs_info->async_submit_wait))
670 wake_up(&fs_info->async_submit_wait);
672 async->submit_bio_done(async->inode, async->rw, async->bio,
673 async->mirror_num, async->bio_flags,
677 static void run_one_async_free(struct btrfs_work *work)
679 struct async_submit_bio *async;
681 async = container_of(work, struct async_submit_bio, work);
685 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
686 int rw, struct bio *bio, int mirror_num,
687 unsigned long bio_flags,
689 extent_submit_bio_hook_t *submit_bio_start,
690 extent_submit_bio_hook_t *submit_bio_done)
692 struct async_submit_bio *async;
694 async = kmalloc(sizeof(*async), GFP_NOFS);
698 async->inode = inode;
701 async->mirror_num = mirror_num;
702 async->submit_bio_start = submit_bio_start;
703 async->submit_bio_done = submit_bio_done;
705 async->work.func = run_one_async_start;
706 async->work.ordered_func = run_one_async_done;
707 async->work.ordered_free = run_one_async_free;
709 async->work.flags = 0;
710 async->bio_flags = bio_flags;
711 async->bio_offset = bio_offset;
713 atomic_inc(&fs_info->nr_async_submits);
716 btrfs_set_work_high_prio(&async->work);
718 btrfs_queue_worker(&fs_info->workers, &async->work);
720 while (atomic_read(&fs_info->async_submit_draining) &&
721 atomic_read(&fs_info->nr_async_submits)) {
722 wait_event(fs_info->async_submit_wait,
723 (atomic_read(&fs_info->nr_async_submits) == 0));
729 static int btree_csum_one_bio(struct bio *bio)
731 struct bio_vec *bvec = bio->bi_io_vec;
733 struct btrfs_root *root;
735 WARN_ON(bio->bi_vcnt <= 0);
736 while (bio_index < bio->bi_vcnt) {
737 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
738 csum_dirty_buffer(root, bvec->bv_page);
745 static int __btree_submit_bio_start(struct inode *inode, int rw,
746 struct bio *bio, int mirror_num,
747 unsigned long bio_flags,
751 * when we're called for a write, we're already in the async
752 * submission context. Just jump into btrfs_map_bio
754 btree_csum_one_bio(bio);
758 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
759 int mirror_num, unsigned long bio_flags,
763 * when we're called for a write, we're already in the async
764 * submission context. Just jump into btrfs_map_bio
766 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
769 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
770 int mirror_num, unsigned long bio_flags,
775 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
779 if (!(rw & REQ_WRITE)) {
781 * called for a read, do the setup so that checksum validation
782 * can happen in the async kernel threads
784 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
789 * kthread helpers are used to submit writes so that checksumming
790 * can happen in parallel across all CPUs
792 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
793 inode, rw, bio, mirror_num, 0,
795 __btree_submit_bio_start,
796 __btree_submit_bio_done);
799 #ifdef CONFIG_MIGRATION
800 static int btree_migratepage(struct address_space *mapping,
801 struct page *newpage, struct page *page)
804 * we can't safely write a btree page from here,
805 * we haven't done the locking hook
810 * Buffers may be managed in a filesystem specific way.
811 * We must have no buffers or drop them.
813 if (page_has_private(page) &&
814 !try_to_release_page(page, GFP_KERNEL))
816 return migrate_page(mapping, newpage, page);
820 static int btree_writepage(struct page *page, struct writeback_control *wbc)
822 struct extent_io_tree *tree;
823 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
824 struct extent_buffer *eb;
827 tree = &BTRFS_I(page->mapping->host)->io_tree;
828 if (!(current->flags & PF_MEMALLOC)) {
829 return extent_write_full_page(tree, page,
830 btree_get_extent, wbc);
833 redirty_page_for_writepage(wbc, page);
834 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
837 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
839 spin_lock(&root->fs_info->delalloc_lock);
840 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
841 spin_unlock(&root->fs_info->delalloc_lock);
843 free_extent_buffer(eb);
849 static int btree_writepages(struct address_space *mapping,
850 struct writeback_control *wbc)
852 struct extent_io_tree *tree;
853 tree = &BTRFS_I(mapping->host)->io_tree;
854 if (wbc->sync_mode == WB_SYNC_NONE) {
855 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
857 unsigned long thresh = 32 * 1024 * 1024;
859 if (wbc->for_kupdate)
862 /* this is a bit racy, but that's ok */
863 num_dirty = root->fs_info->dirty_metadata_bytes;
864 if (num_dirty < thresh)
867 return extent_writepages(tree, mapping, btree_get_extent, wbc);
870 static int btree_readpage(struct file *file, struct page *page)
872 struct extent_io_tree *tree;
873 tree = &BTRFS_I(page->mapping->host)->io_tree;
874 return extent_read_full_page(tree, page, btree_get_extent);
877 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
879 struct extent_io_tree *tree;
880 struct extent_map_tree *map;
883 if (PageWriteback(page) || PageDirty(page))
886 tree = &BTRFS_I(page->mapping->host)->io_tree;
887 map = &BTRFS_I(page->mapping->host)->extent_tree;
889 ret = try_release_extent_state(map, tree, page, gfp_flags);
893 ret = try_release_extent_buffer(tree, page);
895 ClearPagePrivate(page);
896 set_page_private(page, 0);
897 page_cache_release(page);
903 static void btree_invalidatepage(struct page *page, unsigned long offset)
905 struct extent_io_tree *tree;
906 tree = &BTRFS_I(page->mapping->host)->io_tree;
907 extent_invalidatepage(tree, page, offset);
908 btree_releasepage(page, GFP_NOFS);
909 if (PagePrivate(page)) {
910 printk(KERN_WARNING "btrfs warning page private not zero "
911 "on page %llu\n", (unsigned long long)page_offset(page));
912 ClearPagePrivate(page);
913 set_page_private(page, 0);
914 page_cache_release(page);
918 static const struct address_space_operations btree_aops = {
919 .readpage = btree_readpage,
920 .writepage = btree_writepage,
921 .writepages = btree_writepages,
922 .releasepage = btree_releasepage,
923 .invalidatepage = btree_invalidatepage,
924 #ifdef CONFIG_MIGRATION
925 .migratepage = btree_migratepage,
929 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
932 struct extent_buffer *buf = NULL;
933 struct inode *btree_inode = root->fs_info->btree_inode;
936 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
939 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
940 buf, 0, 0, btree_get_extent, 0);
941 free_extent_buffer(buf);
945 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
946 u64 bytenr, u32 blocksize)
948 struct inode *btree_inode = root->fs_info->btree_inode;
949 struct extent_buffer *eb;
950 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
955 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
956 u64 bytenr, u32 blocksize)
958 struct inode *btree_inode = root->fs_info->btree_inode;
959 struct extent_buffer *eb;
961 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
962 bytenr, blocksize, NULL);
967 int btrfs_write_tree_block(struct extent_buffer *buf)
969 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
970 buf->start + buf->len - 1);
973 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
975 return filemap_fdatawait_range(buf->first_page->mapping,
976 buf->start, buf->start + buf->len - 1);
979 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
980 u32 blocksize, u64 parent_transid)
982 struct extent_buffer *buf = NULL;
985 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
989 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
992 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
997 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
998 struct extent_buffer *buf)
1000 struct inode *btree_inode = root->fs_info->btree_inode;
1001 if (btrfs_header_generation(buf) ==
1002 root->fs_info->running_transaction->transid) {
1003 btrfs_assert_tree_locked(buf);
1005 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1006 spin_lock(&root->fs_info->delalloc_lock);
1007 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1008 root->fs_info->dirty_metadata_bytes -= buf->len;
1011 spin_unlock(&root->fs_info->delalloc_lock);
1014 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1015 btrfs_set_lock_blocking(buf);
1016 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1022 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1023 u32 stripesize, struct btrfs_root *root,
1024 struct btrfs_fs_info *fs_info,
1028 root->commit_root = NULL;
1029 root->sectorsize = sectorsize;
1030 root->nodesize = nodesize;
1031 root->leafsize = leafsize;
1032 root->stripesize = stripesize;
1034 root->track_dirty = 0;
1036 root->orphan_item_inserted = 0;
1037 root->orphan_cleanup_state = 0;
1039 root->fs_info = fs_info;
1040 root->objectid = objectid;
1041 root->last_trans = 0;
1042 root->highest_objectid = 0;
1044 root->inode_tree = RB_ROOT;
1045 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1046 root->block_rsv = NULL;
1047 root->orphan_block_rsv = NULL;
1049 INIT_LIST_HEAD(&root->dirty_list);
1050 INIT_LIST_HEAD(&root->orphan_list);
1051 INIT_LIST_HEAD(&root->root_list);
1052 spin_lock_init(&root->orphan_lock);
1053 spin_lock_init(&root->inode_lock);
1054 spin_lock_init(&root->accounting_lock);
1055 mutex_init(&root->objectid_mutex);
1056 mutex_init(&root->log_mutex);
1057 init_waitqueue_head(&root->log_writer_wait);
1058 init_waitqueue_head(&root->log_commit_wait[0]);
1059 init_waitqueue_head(&root->log_commit_wait[1]);
1060 atomic_set(&root->log_commit[0], 0);
1061 atomic_set(&root->log_commit[1], 0);
1062 atomic_set(&root->log_writers, 0);
1063 root->log_batch = 0;
1064 root->log_transid = 0;
1065 root->last_log_commit = 0;
1066 extent_io_tree_init(&root->dirty_log_pages,
1067 fs_info->btree_inode->i_mapping);
1069 memset(&root->root_key, 0, sizeof(root->root_key));
1070 memset(&root->root_item, 0, sizeof(root->root_item));
1071 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1072 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1073 root->defrag_trans_start = fs_info->generation;
1074 init_completion(&root->kobj_unregister);
1075 root->defrag_running = 0;
1076 root->root_key.objectid = objectid;
1077 root->anon_super.s_root = NULL;
1078 root->anon_super.s_dev = 0;
1079 INIT_LIST_HEAD(&root->anon_super.s_list);
1080 INIT_LIST_HEAD(&root->anon_super.s_instances);
1081 init_rwsem(&root->anon_super.s_umount);
1086 static int find_and_setup_root(struct btrfs_root *tree_root,
1087 struct btrfs_fs_info *fs_info,
1089 struct btrfs_root *root)
1095 __setup_root(tree_root->nodesize, tree_root->leafsize,
1096 tree_root->sectorsize, tree_root->stripesize,
1097 root, fs_info, objectid);
1098 ret = btrfs_find_last_root(tree_root, objectid,
1099 &root->root_item, &root->root_key);
1104 generation = btrfs_root_generation(&root->root_item);
1105 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1106 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1107 blocksize, generation);
1108 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1109 free_extent_buffer(root->node);
1112 root->commit_root = btrfs_root_node(root);
1116 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1117 struct btrfs_fs_info *fs_info)
1119 struct btrfs_root *root;
1120 struct btrfs_root *tree_root = fs_info->tree_root;
1121 struct extent_buffer *leaf;
1123 root = kzalloc(sizeof(*root), GFP_NOFS);
1125 return ERR_PTR(-ENOMEM);
1127 __setup_root(tree_root->nodesize, tree_root->leafsize,
1128 tree_root->sectorsize, tree_root->stripesize,
1129 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1131 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1132 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1133 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1135 * log trees do not get reference counted because they go away
1136 * before a real commit is actually done. They do store pointers
1137 * to file data extents, and those reference counts still get
1138 * updated (along with back refs to the log tree).
1142 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1143 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1146 return ERR_CAST(leaf);
1149 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1150 btrfs_set_header_bytenr(leaf, leaf->start);
1151 btrfs_set_header_generation(leaf, trans->transid);
1152 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1153 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1156 write_extent_buffer(root->node, root->fs_info->fsid,
1157 (unsigned long)btrfs_header_fsid(root->node),
1159 btrfs_mark_buffer_dirty(root->node);
1160 btrfs_tree_unlock(root->node);
1164 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1165 struct btrfs_fs_info *fs_info)
1167 struct btrfs_root *log_root;
1169 log_root = alloc_log_tree(trans, fs_info);
1170 if (IS_ERR(log_root))
1171 return PTR_ERR(log_root);
1172 WARN_ON(fs_info->log_root_tree);
1173 fs_info->log_root_tree = log_root;
1177 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1178 struct btrfs_root *root)
1180 struct btrfs_root *log_root;
1181 struct btrfs_inode_item *inode_item;
1183 log_root = alloc_log_tree(trans, root->fs_info);
1184 if (IS_ERR(log_root))
1185 return PTR_ERR(log_root);
1187 log_root->last_trans = trans->transid;
1188 log_root->root_key.offset = root->root_key.objectid;
1190 inode_item = &log_root->root_item.inode;
1191 inode_item->generation = cpu_to_le64(1);
1192 inode_item->size = cpu_to_le64(3);
1193 inode_item->nlink = cpu_to_le32(1);
1194 inode_item->nbytes = cpu_to_le64(root->leafsize);
1195 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1197 btrfs_set_root_node(&log_root->root_item, log_root->node);
1199 WARN_ON(root->log_root);
1200 root->log_root = log_root;
1201 root->log_transid = 0;
1202 root->last_log_commit = 0;
1206 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1207 struct btrfs_key *location)
1209 struct btrfs_root *root;
1210 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1211 struct btrfs_path *path;
1212 struct extent_buffer *l;
1217 root = kzalloc(sizeof(*root), GFP_NOFS);
1219 return ERR_PTR(-ENOMEM);
1220 if (location->offset == (u64)-1) {
1221 ret = find_and_setup_root(tree_root, fs_info,
1222 location->objectid, root);
1225 return ERR_PTR(ret);
1230 __setup_root(tree_root->nodesize, tree_root->leafsize,
1231 tree_root->sectorsize, tree_root->stripesize,
1232 root, fs_info, location->objectid);
1234 path = btrfs_alloc_path();
1237 return ERR_PTR(-ENOMEM);
1239 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1242 read_extent_buffer(l, &root->root_item,
1243 btrfs_item_ptr_offset(l, path->slots[0]),
1244 sizeof(root->root_item));
1245 memcpy(&root->root_key, location, sizeof(*location));
1247 btrfs_free_path(path);
1252 return ERR_PTR(ret);
1255 generation = btrfs_root_generation(&root->root_item);
1256 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1257 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1258 blocksize, generation);
1259 root->commit_root = btrfs_root_node(root);
1260 BUG_ON(!root->node);
1262 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1264 btrfs_check_and_init_root_item(&root->root_item);
1270 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1271 struct btrfs_key *location)
1273 struct btrfs_root *root;
1276 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1277 return fs_info->tree_root;
1278 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1279 return fs_info->extent_root;
1280 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1281 return fs_info->chunk_root;
1282 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1283 return fs_info->dev_root;
1284 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1285 return fs_info->csum_root;
1287 spin_lock(&fs_info->fs_roots_radix_lock);
1288 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1289 (unsigned long)location->objectid);
1290 spin_unlock(&fs_info->fs_roots_radix_lock);
1294 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1298 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1299 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1301 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1306 btrfs_init_free_ino_ctl(root);
1307 mutex_init(&root->fs_commit_mutex);
1308 spin_lock_init(&root->cache_lock);
1309 init_waitqueue_head(&root->cache_wait);
1311 ret = set_anon_super(&root->anon_super, NULL);
1315 if (btrfs_root_refs(&root->root_item) == 0) {
1320 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1324 root->orphan_item_inserted = 1;
1326 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1330 spin_lock(&fs_info->fs_roots_radix_lock);
1331 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1332 (unsigned long)root->root_key.objectid,
1337 spin_unlock(&fs_info->fs_roots_radix_lock);
1338 radix_tree_preload_end();
1340 if (ret == -EEXIST) {
1347 ret = btrfs_find_dead_roots(fs_info->tree_root,
1348 root->root_key.objectid);
1353 return ERR_PTR(ret);
1356 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1358 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1360 struct btrfs_device *device;
1361 struct backing_dev_info *bdi;
1364 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1367 bdi = blk_get_backing_dev_info(device->bdev);
1368 if (bdi && bdi_congested(bdi, bdi_bits)) {
1378 * If this fails, caller must call bdi_destroy() to get rid of the
1381 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1385 bdi->capabilities = BDI_CAP_MAP_COPY;
1386 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1390 bdi->ra_pages = default_backing_dev_info.ra_pages;
1391 bdi->congested_fn = btrfs_congested_fn;
1392 bdi->congested_data = info;
1396 static int bio_ready_for_csum(struct bio *bio)
1402 struct extent_io_tree *io_tree = NULL;
1403 struct bio_vec *bvec;
1407 bio_for_each_segment(bvec, bio, i) {
1408 page = bvec->bv_page;
1409 if (page->private == EXTENT_PAGE_PRIVATE) {
1410 length += bvec->bv_len;
1413 if (!page->private) {
1414 length += bvec->bv_len;
1417 length = bvec->bv_len;
1418 buf_len = page->private >> 2;
1419 start = page_offset(page) + bvec->bv_offset;
1420 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1422 /* are we fully contained in this bio? */
1423 if (buf_len <= length)
1426 ret = extent_range_uptodate(io_tree, start + length,
1427 start + buf_len - 1);
1432 * called by the kthread helper functions to finally call the bio end_io
1433 * functions. This is where read checksum verification actually happens
1435 static void end_workqueue_fn(struct btrfs_work *work)
1438 struct end_io_wq *end_io_wq;
1439 struct btrfs_fs_info *fs_info;
1442 end_io_wq = container_of(work, struct end_io_wq, work);
1443 bio = end_io_wq->bio;
1444 fs_info = end_io_wq->info;
1446 /* metadata bio reads are special because the whole tree block must
1447 * be checksummed at once. This makes sure the entire block is in
1448 * ram and up to date before trying to verify things. For
1449 * blocksize <= pagesize, it is basically a noop
1451 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1452 !bio_ready_for_csum(bio)) {
1453 btrfs_queue_worker(&fs_info->endio_meta_workers,
1457 error = end_io_wq->error;
1458 bio->bi_private = end_io_wq->private;
1459 bio->bi_end_io = end_io_wq->end_io;
1461 bio_endio(bio, error);
1464 static int cleaner_kthread(void *arg)
1466 struct btrfs_root *root = arg;
1469 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1471 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1472 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1473 btrfs_run_delayed_iputs(root);
1474 btrfs_clean_old_snapshots(root);
1475 mutex_unlock(&root->fs_info->cleaner_mutex);
1476 btrfs_run_defrag_inodes(root->fs_info);
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->trans_lock);
1507 cur = root->fs_info->running_transaction;
1509 spin_unlock(&root->fs_info->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->trans_lock);
1520 transid = cur->transid;
1521 spin_unlock(&root->fs_info->trans_lock);
1523 trans = btrfs_join_transaction(root);
1524 BUG_ON(IS_ERR(trans));
1525 if (transid == trans->transid) {
1526 ret = btrfs_commit_transaction(trans, root);
1529 btrfs_end_transaction(trans, root);
1532 wake_up_process(root->fs_info->cleaner_kthread);
1533 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1535 if (freezing(current)) {
1538 set_current_state(TASK_INTERRUPTIBLE);
1539 if (!kthread_should_stop() &&
1540 !btrfs_transaction_blocked(root->fs_info))
1541 schedule_timeout(delay);
1542 __set_current_state(TASK_RUNNING);
1544 } while (!kthread_should_stop());
1548 struct btrfs_root *open_ctree(struct super_block *sb,
1549 struct btrfs_fs_devices *fs_devices,
1559 struct btrfs_key location;
1560 struct buffer_head *bh;
1561 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1563 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1565 struct btrfs_root *tree_root = btrfs_sb(sb);
1566 struct btrfs_fs_info *fs_info = NULL;
1567 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1569 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1571 struct btrfs_root *log_tree_root;
1576 struct btrfs_super_block *disk_super;
1578 if (!extent_root || !tree_root || !tree_root->fs_info ||
1579 !chunk_root || !dev_root || !csum_root) {
1583 fs_info = tree_root->fs_info;
1585 ret = init_srcu_struct(&fs_info->subvol_srcu);
1591 ret = setup_bdi(fs_info, &fs_info->bdi);
1597 fs_info->btree_inode = new_inode(sb);
1598 if (!fs_info->btree_inode) {
1603 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1605 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1606 INIT_LIST_HEAD(&fs_info->trans_list);
1607 INIT_LIST_HEAD(&fs_info->dead_roots);
1608 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1609 INIT_LIST_HEAD(&fs_info->hashers);
1610 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1611 INIT_LIST_HEAD(&fs_info->ordered_operations);
1612 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1613 spin_lock_init(&fs_info->delalloc_lock);
1614 spin_lock_init(&fs_info->trans_lock);
1615 spin_lock_init(&fs_info->ref_cache_lock);
1616 spin_lock_init(&fs_info->fs_roots_radix_lock);
1617 spin_lock_init(&fs_info->delayed_iput_lock);
1618 spin_lock_init(&fs_info->defrag_inodes_lock);
1619 mutex_init(&fs_info->reloc_mutex);
1621 init_completion(&fs_info->kobj_unregister);
1622 fs_info->tree_root = tree_root;
1623 fs_info->extent_root = extent_root;
1624 fs_info->csum_root = csum_root;
1625 fs_info->chunk_root = chunk_root;
1626 fs_info->dev_root = dev_root;
1627 fs_info->fs_devices = fs_devices;
1628 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1629 INIT_LIST_HEAD(&fs_info->space_info);
1630 btrfs_mapping_init(&fs_info->mapping_tree);
1631 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1632 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1633 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1634 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1635 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1636 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1637 mutex_init(&fs_info->durable_block_rsv_mutex);
1638 atomic_set(&fs_info->nr_async_submits, 0);
1639 atomic_set(&fs_info->async_delalloc_pages, 0);
1640 atomic_set(&fs_info->async_submit_draining, 0);
1641 atomic_set(&fs_info->nr_async_bios, 0);
1642 atomic_set(&fs_info->defrag_running, 0);
1644 fs_info->max_inline = 8192 * 1024;
1645 fs_info->metadata_ratio = 0;
1646 fs_info->defrag_inodes = RB_ROOT;
1647 fs_info->trans_no_join = 0;
1649 fs_info->thread_pool_size = min_t(unsigned long,
1650 num_online_cpus() + 2, 8);
1652 INIT_LIST_HEAD(&fs_info->ordered_extents);
1653 spin_lock_init(&fs_info->ordered_extent_lock);
1654 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1656 if (!fs_info->delayed_root) {
1660 btrfs_init_delayed_root(fs_info->delayed_root);
1662 mutex_init(&fs_info->scrub_lock);
1663 atomic_set(&fs_info->scrubs_running, 0);
1664 atomic_set(&fs_info->scrub_pause_req, 0);
1665 atomic_set(&fs_info->scrubs_paused, 0);
1666 atomic_set(&fs_info->scrub_cancel_req, 0);
1667 init_waitqueue_head(&fs_info->scrub_pause_wait);
1668 init_rwsem(&fs_info->scrub_super_lock);
1669 fs_info->scrub_workers_refcnt = 0;
1671 sb->s_blocksize = 4096;
1672 sb->s_blocksize_bits = blksize_bits(4096);
1673 sb->s_bdi = &fs_info->bdi;
1675 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1676 fs_info->btree_inode->i_nlink = 1;
1678 * we set the i_size on the btree inode to the max possible int.
1679 * the real end of the address space is determined by all of
1680 * the devices in the system
1682 fs_info->btree_inode->i_size = OFFSET_MAX;
1683 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1684 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1686 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1687 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1688 fs_info->btree_inode->i_mapping);
1689 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1691 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1693 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1694 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1695 sizeof(struct btrfs_key));
1696 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1697 insert_inode_hash(fs_info->btree_inode);
1699 spin_lock_init(&fs_info->block_group_cache_lock);
1700 fs_info->block_group_cache_tree = RB_ROOT;
1702 extent_io_tree_init(&fs_info->freed_extents[0],
1703 fs_info->btree_inode->i_mapping);
1704 extent_io_tree_init(&fs_info->freed_extents[1],
1705 fs_info->btree_inode->i_mapping);
1706 fs_info->pinned_extents = &fs_info->freed_extents[0];
1707 fs_info->do_barriers = 1;
1710 mutex_init(&fs_info->ordered_operations_mutex);
1711 mutex_init(&fs_info->tree_log_mutex);
1712 mutex_init(&fs_info->chunk_mutex);
1713 mutex_init(&fs_info->transaction_kthread_mutex);
1714 mutex_init(&fs_info->cleaner_mutex);
1715 mutex_init(&fs_info->volume_mutex);
1716 init_rwsem(&fs_info->extent_commit_sem);
1717 init_rwsem(&fs_info->cleanup_work_sem);
1718 init_rwsem(&fs_info->subvol_sem);
1720 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1721 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1723 init_waitqueue_head(&fs_info->transaction_throttle);
1724 init_waitqueue_head(&fs_info->transaction_wait);
1725 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1726 init_waitqueue_head(&fs_info->async_submit_wait);
1728 __setup_root(4096, 4096, 4096, 4096, tree_root,
1729 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1731 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1737 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1738 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1739 sizeof(fs_info->super_for_commit));
1742 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1744 disk_super = &fs_info->super_copy;
1745 if (!btrfs_super_root(disk_super))
1748 /* check FS state, whether FS is broken. */
1749 fs_info->fs_state |= btrfs_super_flags(disk_super);
1751 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1754 * In the long term, we'll store the compression type in the super
1755 * block, and it'll be used for per file compression control.
1757 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1759 ret = btrfs_parse_options(tree_root, options);
1765 features = btrfs_super_incompat_flags(disk_super) &
1766 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1768 printk(KERN_ERR "BTRFS: couldn't mount because of "
1769 "unsupported optional features (%Lx).\n",
1770 (unsigned long long)features);
1775 features = btrfs_super_incompat_flags(disk_super);
1776 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1777 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1778 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1779 btrfs_set_super_incompat_flags(disk_super, features);
1781 features = btrfs_super_compat_ro_flags(disk_super) &
1782 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1783 if (!(sb->s_flags & MS_RDONLY) && features) {
1784 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1785 "unsupported option features (%Lx).\n",
1786 (unsigned long long)features);
1791 btrfs_init_workers(&fs_info->generic_worker,
1792 "genwork", 1, NULL);
1794 btrfs_init_workers(&fs_info->workers, "worker",
1795 fs_info->thread_pool_size,
1796 &fs_info->generic_worker);
1798 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1799 fs_info->thread_pool_size,
1800 &fs_info->generic_worker);
1802 btrfs_init_workers(&fs_info->submit_workers, "submit",
1803 min_t(u64, fs_devices->num_devices,
1804 fs_info->thread_pool_size),
1805 &fs_info->generic_worker);
1807 btrfs_init_workers(&fs_info->caching_workers, "cache",
1808 2, &fs_info->generic_worker);
1810 /* a higher idle thresh on the submit workers makes it much more
1811 * likely that bios will be send down in a sane order to the
1814 fs_info->submit_workers.idle_thresh = 64;
1816 fs_info->workers.idle_thresh = 16;
1817 fs_info->workers.ordered = 1;
1819 fs_info->delalloc_workers.idle_thresh = 2;
1820 fs_info->delalloc_workers.ordered = 1;
1822 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1823 &fs_info->generic_worker);
1824 btrfs_init_workers(&fs_info->endio_workers, "endio",
1825 fs_info->thread_pool_size,
1826 &fs_info->generic_worker);
1827 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1828 fs_info->thread_pool_size,
1829 &fs_info->generic_worker);
1830 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1831 "endio-meta-write", fs_info->thread_pool_size,
1832 &fs_info->generic_worker);
1833 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1834 fs_info->thread_pool_size,
1835 &fs_info->generic_worker);
1836 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1837 1, &fs_info->generic_worker);
1838 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1839 fs_info->thread_pool_size,
1840 &fs_info->generic_worker);
1843 * endios are largely parallel and should have a very
1846 fs_info->endio_workers.idle_thresh = 4;
1847 fs_info->endio_meta_workers.idle_thresh = 4;
1849 fs_info->endio_write_workers.idle_thresh = 2;
1850 fs_info->endio_meta_write_workers.idle_thresh = 2;
1852 btrfs_start_workers(&fs_info->workers, 1);
1853 btrfs_start_workers(&fs_info->generic_worker, 1);
1854 btrfs_start_workers(&fs_info->submit_workers, 1);
1855 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1856 btrfs_start_workers(&fs_info->fixup_workers, 1);
1857 btrfs_start_workers(&fs_info->endio_workers, 1);
1858 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1859 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1860 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1861 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1862 btrfs_start_workers(&fs_info->delayed_workers, 1);
1863 btrfs_start_workers(&fs_info->caching_workers, 1);
1865 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1866 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1867 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1869 nodesize = btrfs_super_nodesize(disk_super);
1870 leafsize = btrfs_super_leafsize(disk_super);
1871 sectorsize = btrfs_super_sectorsize(disk_super);
1872 stripesize = btrfs_super_stripesize(disk_super);
1873 tree_root->nodesize = nodesize;
1874 tree_root->leafsize = leafsize;
1875 tree_root->sectorsize = sectorsize;
1876 tree_root->stripesize = stripesize;
1878 sb->s_blocksize = sectorsize;
1879 sb->s_blocksize_bits = blksize_bits(sectorsize);
1881 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1882 sizeof(disk_super->magic))) {
1883 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1884 goto fail_sb_buffer;
1887 mutex_lock(&fs_info->chunk_mutex);
1888 ret = btrfs_read_sys_array(tree_root);
1889 mutex_unlock(&fs_info->chunk_mutex);
1891 printk(KERN_WARNING "btrfs: failed to read the system "
1892 "array on %s\n", sb->s_id);
1893 goto fail_sb_buffer;
1896 blocksize = btrfs_level_size(tree_root,
1897 btrfs_super_chunk_root_level(disk_super));
1898 generation = btrfs_super_chunk_root_generation(disk_super);
1900 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1901 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1903 chunk_root->node = read_tree_block(chunk_root,
1904 btrfs_super_chunk_root(disk_super),
1905 blocksize, generation);
1906 BUG_ON(!chunk_root->node);
1907 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1908 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1910 goto fail_chunk_root;
1912 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1913 chunk_root->commit_root = btrfs_root_node(chunk_root);
1915 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1916 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1919 mutex_lock(&fs_info->chunk_mutex);
1920 ret = btrfs_read_chunk_tree(chunk_root);
1921 mutex_unlock(&fs_info->chunk_mutex);
1923 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1925 goto fail_chunk_root;
1928 btrfs_close_extra_devices(fs_devices);
1930 blocksize = btrfs_level_size(tree_root,
1931 btrfs_super_root_level(disk_super));
1932 generation = btrfs_super_generation(disk_super);
1934 tree_root->node = read_tree_block(tree_root,
1935 btrfs_super_root(disk_super),
1936 blocksize, generation);
1937 if (!tree_root->node)
1938 goto fail_chunk_root;
1939 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1940 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1942 goto fail_tree_root;
1944 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1945 tree_root->commit_root = btrfs_root_node(tree_root);
1947 ret = find_and_setup_root(tree_root, fs_info,
1948 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1950 goto fail_tree_root;
1951 extent_root->track_dirty = 1;
1953 ret = find_and_setup_root(tree_root, fs_info,
1954 BTRFS_DEV_TREE_OBJECTID, dev_root);
1956 goto fail_extent_root;
1957 dev_root->track_dirty = 1;
1959 ret = find_and_setup_root(tree_root, fs_info,
1960 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1964 csum_root->track_dirty = 1;
1966 fs_info->generation = generation;
1967 fs_info->last_trans_committed = generation;
1968 fs_info->data_alloc_profile = (u64)-1;
1969 fs_info->metadata_alloc_profile = (u64)-1;
1970 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1972 ret = btrfs_init_space_info(fs_info);
1974 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
1975 goto fail_block_groups;
1978 ret = btrfs_read_block_groups(extent_root);
1980 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1981 goto fail_block_groups;
1984 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1986 if (IS_ERR(fs_info->cleaner_kthread))
1987 goto fail_block_groups;
1989 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1991 "btrfs-transaction");
1992 if (IS_ERR(fs_info->transaction_kthread))
1995 if (!btrfs_test_opt(tree_root, SSD) &&
1996 !btrfs_test_opt(tree_root, NOSSD) &&
1997 !fs_info->fs_devices->rotating) {
1998 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2000 btrfs_set_opt(fs_info->mount_opt, SSD);
2003 /* do not make disk changes in broken FS */
2004 if (btrfs_super_log_root(disk_super) != 0 &&
2005 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2006 u64 bytenr = btrfs_super_log_root(disk_super);
2008 if (fs_devices->rw_devices == 0) {
2009 printk(KERN_WARNING "Btrfs log replay required "
2012 goto fail_trans_kthread;
2015 btrfs_level_size(tree_root,
2016 btrfs_super_log_root_level(disk_super));
2018 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2019 if (!log_tree_root) {
2021 goto fail_trans_kthread;
2024 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2025 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2027 log_tree_root->node = read_tree_block(tree_root, bytenr,
2030 ret = btrfs_recover_log_trees(log_tree_root);
2033 if (sb->s_flags & MS_RDONLY) {
2034 ret = btrfs_commit_super(tree_root);
2039 ret = btrfs_find_orphan_roots(tree_root);
2042 if (!(sb->s_flags & MS_RDONLY)) {
2043 ret = btrfs_cleanup_fs_roots(fs_info);
2046 ret = btrfs_recover_relocation(tree_root);
2049 "btrfs: failed to recover relocation\n");
2051 goto fail_trans_kthread;
2055 location.objectid = BTRFS_FS_TREE_OBJECTID;
2056 location.type = BTRFS_ROOT_ITEM_KEY;
2057 location.offset = (u64)-1;
2059 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2060 if (!fs_info->fs_root)
2061 goto fail_trans_kthread;
2062 if (IS_ERR(fs_info->fs_root)) {
2063 err = PTR_ERR(fs_info->fs_root);
2064 goto fail_trans_kthread;
2067 if (!(sb->s_flags & MS_RDONLY)) {
2068 down_read(&fs_info->cleanup_work_sem);
2069 err = btrfs_orphan_cleanup(fs_info->fs_root);
2071 err = btrfs_orphan_cleanup(fs_info->tree_root);
2072 up_read(&fs_info->cleanup_work_sem);
2074 close_ctree(tree_root);
2075 return ERR_PTR(err);
2082 kthread_stop(fs_info->transaction_kthread);
2084 kthread_stop(fs_info->cleaner_kthread);
2087 * make sure we're done with the btree inode before we stop our
2090 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2091 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2094 btrfs_free_block_groups(fs_info);
2095 free_extent_buffer(csum_root->node);
2096 free_extent_buffer(csum_root->commit_root);
2098 free_extent_buffer(dev_root->node);
2099 free_extent_buffer(dev_root->commit_root);
2101 free_extent_buffer(extent_root->node);
2102 free_extent_buffer(extent_root->commit_root);
2104 free_extent_buffer(tree_root->node);
2105 free_extent_buffer(tree_root->commit_root);
2107 free_extent_buffer(chunk_root->node);
2108 free_extent_buffer(chunk_root->commit_root);
2110 btrfs_stop_workers(&fs_info->generic_worker);
2111 btrfs_stop_workers(&fs_info->fixup_workers);
2112 btrfs_stop_workers(&fs_info->delalloc_workers);
2113 btrfs_stop_workers(&fs_info->workers);
2114 btrfs_stop_workers(&fs_info->endio_workers);
2115 btrfs_stop_workers(&fs_info->endio_meta_workers);
2116 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2117 btrfs_stop_workers(&fs_info->endio_write_workers);
2118 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2119 btrfs_stop_workers(&fs_info->submit_workers);
2120 btrfs_stop_workers(&fs_info->delayed_workers);
2121 btrfs_stop_workers(&fs_info->caching_workers);
2123 kfree(fs_info->delayed_root);
2125 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2126 iput(fs_info->btree_inode);
2128 btrfs_close_devices(fs_info->fs_devices);
2129 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2131 bdi_destroy(&fs_info->bdi);
2133 cleanup_srcu_struct(&fs_info->subvol_srcu);
2141 return ERR_PTR(err);
2144 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2146 char b[BDEVNAME_SIZE];
2149 set_buffer_uptodate(bh);
2151 printk_ratelimited(KERN_WARNING "lost page write due to "
2152 "I/O error on %s\n",
2153 bdevname(bh->b_bdev, b));
2154 /* note, we dont' set_buffer_write_io_error because we have
2155 * our own ways of dealing with the IO errors
2157 clear_buffer_uptodate(bh);
2163 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2165 struct buffer_head *bh;
2166 struct buffer_head *latest = NULL;
2167 struct btrfs_super_block *super;
2172 /* we would like to check all the supers, but that would make
2173 * a btrfs mount succeed after a mkfs from a different FS.
2174 * So, we need to add a special mount option to scan for
2175 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2177 for (i = 0; i < 1; i++) {
2178 bytenr = btrfs_sb_offset(i);
2179 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2181 bh = __bread(bdev, bytenr / 4096, 4096);
2185 super = (struct btrfs_super_block *)bh->b_data;
2186 if (btrfs_super_bytenr(super) != bytenr ||
2187 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2188 sizeof(super->magic))) {
2193 if (!latest || btrfs_super_generation(super) > transid) {
2196 transid = btrfs_super_generation(super);
2205 * this should be called twice, once with wait == 0 and
2206 * once with wait == 1. When wait == 0 is done, all the buffer heads
2207 * we write are pinned.
2209 * They are released when wait == 1 is done.
2210 * max_mirrors must be the same for both runs, and it indicates how
2211 * many supers on this one device should be written.
2213 * max_mirrors == 0 means to write them all.
2215 static int write_dev_supers(struct btrfs_device *device,
2216 struct btrfs_super_block *sb,
2217 int do_barriers, int wait, int max_mirrors)
2219 struct buffer_head *bh;
2225 int last_barrier = 0;
2227 if (max_mirrors == 0)
2228 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2230 /* make sure only the last submit_bh does a barrier */
2232 for (i = 0; i < max_mirrors; i++) {
2233 bytenr = btrfs_sb_offset(i);
2234 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2235 device->total_bytes)
2241 for (i = 0; i < max_mirrors; i++) {
2242 bytenr = btrfs_sb_offset(i);
2243 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2247 bh = __find_get_block(device->bdev, bytenr / 4096,
2248 BTRFS_SUPER_INFO_SIZE);
2251 if (!buffer_uptodate(bh))
2254 /* drop our reference */
2257 /* drop the reference from the wait == 0 run */
2261 btrfs_set_super_bytenr(sb, bytenr);
2264 crc = btrfs_csum_data(NULL, (char *)sb +
2265 BTRFS_CSUM_SIZE, crc,
2266 BTRFS_SUPER_INFO_SIZE -
2268 btrfs_csum_final(crc, sb->csum);
2271 * one reference for us, and we leave it for the
2274 bh = __getblk(device->bdev, bytenr / 4096,
2275 BTRFS_SUPER_INFO_SIZE);
2276 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2278 /* one reference for submit_bh */
2281 set_buffer_uptodate(bh);
2283 bh->b_end_io = btrfs_end_buffer_write_sync;
2286 if (i == last_barrier && do_barriers)
2287 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2289 ret = submit_bh(WRITE_SYNC, bh);
2294 return errors < i ? 0 : -1;
2297 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2299 struct list_head *head;
2300 struct btrfs_device *dev;
2301 struct btrfs_super_block *sb;
2302 struct btrfs_dev_item *dev_item;
2306 int total_errors = 0;
2309 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2310 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2312 sb = &root->fs_info->super_for_commit;
2313 dev_item = &sb->dev_item;
2315 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2316 head = &root->fs_info->fs_devices->devices;
2317 list_for_each_entry_rcu(dev, head, dev_list) {
2322 if (!dev->in_fs_metadata || !dev->writeable)
2325 btrfs_set_stack_device_generation(dev_item, 0);
2326 btrfs_set_stack_device_type(dev_item, dev->type);
2327 btrfs_set_stack_device_id(dev_item, dev->devid);
2328 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2329 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2330 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2331 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2332 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2333 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2334 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2336 flags = btrfs_super_flags(sb);
2337 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2339 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2343 if (total_errors > max_errors) {
2344 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2350 list_for_each_entry_rcu(dev, head, dev_list) {
2353 if (!dev->in_fs_metadata || !dev->writeable)
2356 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2360 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2361 if (total_errors > max_errors) {
2362 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2369 int write_ctree_super(struct btrfs_trans_handle *trans,
2370 struct btrfs_root *root, int max_mirrors)
2374 ret = write_all_supers(root, max_mirrors);
2378 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2380 spin_lock(&fs_info->fs_roots_radix_lock);
2381 radix_tree_delete(&fs_info->fs_roots_radix,
2382 (unsigned long)root->root_key.objectid);
2383 spin_unlock(&fs_info->fs_roots_radix_lock);
2385 if (btrfs_root_refs(&root->root_item) == 0)
2386 synchronize_srcu(&fs_info->subvol_srcu);
2388 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2389 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2394 static void free_fs_root(struct btrfs_root *root)
2396 iput(root->cache_inode);
2397 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2398 if (root->anon_super.s_dev) {
2399 down_write(&root->anon_super.s_umount);
2400 kill_anon_super(&root->anon_super);
2402 free_extent_buffer(root->node);
2403 free_extent_buffer(root->commit_root);
2404 kfree(root->free_ino_ctl);
2405 kfree(root->free_ino_pinned);
2410 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2413 struct btrfs_root *gang[8];
2416 while (!list_empty(&fs_info->dead_roots)) {
2417 gang[0] = list_entry(fs_info->dead_roots.next,
2418 struct btrfs_root, root_list);
2419 list_del(&gang[0]->root_list);
2421 if (gang[0]->in_radix) {
2422 btrfs_free_fs_root(fs_info, gang[0]);
2424 free_extent_buffer(gang[0]->node);
2425 free_extent_buffer(gang[0]->commit_root);
2431 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2436 for (i = 0; i < ret; i++)
2437 btrfs_free_fs_root(fs_info, gang[i]);
2442 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2444 u64 root_objectid = 0;
2445 struct btrfs_root *gang[8];
2450 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2451 (void **)gang, root_objectid,
2456 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2457 for (i = 0; i < ret; i++) {
2460 root_objectid = gang[i]->root_key.objectid;
2461 err = btrfs_orphan_cleanup(gang[i]);
2470 int btrfs_commit_super(struct btrfs_root *root)
2472 struct btrfs_trans_handle *trans;
2475 mutex_lock(&root->fs_info->cleaner_mutex);
2476 btrfs_run_delayed_iputs(root);
2477 btrfs_clean_old_snapshots(root);
2478 mutex_unlock(&root->fs_info->cleaner_mutex);
2480 /* wait until ongoing cleanup work done */
2481 down_write(&root->fs_info->cleanup_work_sem);
2482 up_write(&root->fs_info->cleanup_work_sem);
2484 trans = btrfs_join_transaction(root);
2486 return PTR_ERR(trans);
2487 ret = btrfs_commit_transaction(trans, root);
2489 /* run commit again to drop the original snapshot */
2490 trans = btrfs_join_transaction(root);
2492 return PTR_ERR(trans);
2493 btrfs_commit_transaction(trans, root);
2494 ret = btrfs_write_and_wait_transaction(NULL, root);
2497 ret = write_ctree_super(NULL, root, 0);
2501 int close_ctree(struct btrfs_root *root)
2503 struct btrfs_fs_info *fs_info = root->fs_info;
2506 fs_info->closing = 1;
2509 btrfs_scrub_cancel(root);
2511 /* wait for any defraggers to finish */
2512 wait_event(fs_info->transaction_wait,
2513 (atomic_read(&fs_info->defrag_running) == 0));
2515 /* clear out the rbtree of defraggable inodes */
2516 btrfs_run_defrag_inodes(root->fs_info);
2518 btrfs_put_block_group_cache(fs_info);
2521 * Here come 2 situations when btrfs is broken to flip readonly:
2523 * 1. when btrfs flips readonly somewhere else before
2524 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2525 * and btrfs will skip to write sb directly to keep
2526 * ERROR state on disk.
2528 * 2. when btrfs flips readonly just in btrfs_commit_super,
2529 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2530 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2531 * btrfs will cleanup all FS resources first and write sb then.
2533 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2534 ret = btrfs_commit_super(root);
2536 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2539 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2540 ret = btrfs_error_commit_super(root);
2542 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2545 kthread_stop(root->fs_info->transaction_kthread);
2546 kthread_stop(root->fs_info->cleaner_kthread);
2548 fs_info->closing = 2;
2551 if (fs_info->delalloc_bytes) {
2552 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2553 (unsigned long long)fs_info->delalloc_bytes);
2555 if (fs_info->total_ref_cache_size) {
2556 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2557 (unsigned long long)fs_info->total_ref_cache_size);
2560 free_extent_buffer(fs_info->extent_root->node);
2561 free_extent_buffer(fs_info->extent_root->commit_root);
2562 free_extent_buffer(fs_info->tree_root->node);
2563 free_extent_buffer(fs_info->tree_root->commit_root);
2564 free_extent_buffer(root->fs_info->chunk_root->node);
2565 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2566 free_extent_buffer(root->fs_info->dev_root->node);
2567 free_extent_buffer(root->fs_info->dev_root->commit_root);
2568 free_extent_buffer(root->fs_info->csum_root->node);
2569 free_extent_buffer(root->fs_info->csum_root->commit_root);
2571 btrfs_free_block_groups(root->fs_info);
2573 del_fs_roots(fs_info);
2575 iput(fs_info->btree_inode);
2576 kfree(fs_info->delayed_root);
2578 btrfs_stop_workers(&fs_info->generic_worker);
2579 btrfs_stop_workers(&fs_info->fixup_workers);
2580 btrfs_stop_workers(&fs_info->delalloc_workers);
2581 btrfs_stop_workers(&fs_info->workers);
2582 btrfs_stop_workers(&fs_info->endio_workers);
2583 btrfs_stop_workers(&fs_info->endio_meta_workers);
2584 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2585 btrfs_stop_workers(&fs_info->endio_write_workers);
2586 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2587 btrfs_stop_workers(&fs_info->submit_workers);
2588 btrfs_stop_workers(&fs_info->delayed_workers);
2589 btrfs_stop_workers(&fs_info->caching_workers);
2591 btrfs_close_devices(fs_info->fs_devices);
2592 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2594 bdi_destroy(&fs_info->bdi);
2595 cleanup_srcu_struct(&fs_info->subvol_srcu);
2597 kfree(fs_info->extent_root);
2598 kfree(fs_info->tree_root);
2599 kfree(fs_info->chunk_root);
2600 kfree(fs_info->dev_root);
2601 kfree(fs_info->csum_root);
2607 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2610 struct inode *btree_inode = buf->first_page->mapping->host;
2612 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2617 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2622 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2624 struct inode *btree_inode = buf->first_page->mapping->host;
2625 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2629 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2631 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2632 u64 transid = btrfs_header_generation(buf);
2633 struct inode *btree_inode = root->fs_info->btree_inode;
2636 btrfs_assert_tree_locked(buf);
2637 if (transid != root->fs_info->generation) {
2638 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2639 "found %llu running %llu\n",
2640 (unsigned long long)buf->start,
2641 (unsigned long long)transid,
2642 (unsigned long long)root->fs_info->generation);
2645 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2648 spin_lock(&root->fs_info->delalloc_lock);
2649 root->fs_info->dirty_metadata_bytes += buf->len;
2650 spin_unlock(&root->fs_info->delalloc_lock);
2654 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2657 * looks as though older kernels can get into trouble with
2658 * this code, they end up stuck in balance_dirty_pages forever
2661 unsigned long thresh = 32 * 1024 * 1024;
2663 if (current->flags & PF_MEMALLOC)
2666 btrfs_balance_delayed_items(root);
2668 num_dirty = root->fs_info->dirty_metadata_bytes;
2670 if (num_dirty > thresh) {
2671 balance_dirty_pages_ratelimited_nr(
2672 root->fs_info->btree_inode->i_mapping, 1);
2677 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2680 * looks as though older kernels can get into trouble with
2681 * this code, they end up stuck in balance_dirty_pages forever
2684 unsigned long thresh = 32 * 1024 * 1024;
2686 if (current->flags & PF_MEMALLOC)
2689 num_dirty = root->fs_info->dirty_metadata_bytes;
2691 if (num_dirty > thresh) {
2692 balance_dirty_pages_ratelimited_nr(
2693 root->fs_info->btree_inode->i_mapping, 1);
2698 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2700 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2702 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2704 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2708 int btree_lock_page_hook(struct page *page)
2710 struct inode *inode = page->mapping->host;
2711 struct btrfs_root *root = BTRFS_I(inode)->root;
2712 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2713 struct extent_buffer *eb;
2715 u64 bytenr = page_offset(page);
2717 if (page->private == EXTENT_PAGE_PRIVATE)
2720 len = page->private >> 2;
2721 eb = find_extent_buffer(io_tree, bytenr, len);
2725 btrfs_tree_lock(eb);
2726 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2728 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2729 spin_lock(&root->fs_info->delalloc_lock);
2730 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2731 root->fs_info->dirty_metadata_bytes -= eb->len;
2734 spin_unlock(&root->fs_info->delalloc_lock);
2737 btrfs_tree_unlock(eb);
2738 free_extent_buffer(eb);
2744 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2750 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2751 printk(KERN_WARNING "warning: mount fs with errors, "
2752 "running btrfsck is recommended\n");
2755 int btrfs_error_commit_super(struct btrfs_root *root)
2759 mutex_lock(&root->fs_info->cleaner_mutex);
2760 btrfs_run_delayed_iputs(root);
2761 mutex_unlock(&root->fs_info->cleaner_mutex);
2763 down_write(&root->fs_info->cleanup_work_sem);
2764 up_write(&root->fs_info->cleanup_work_sem);
2766 /* cleanup FS via transaction */
2767 btrfs_cleanup_transaction(root);
2769 ret = write_ctree_super(NULL, root, 0);
2774 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2776 struct btrfs_inode *btrfs_inode;
2777 struct list_head splice;
2779 INIT_LIST_HEAD(&splice);
2781 mutex_lock(&root->fs_info->ordered_operations_mutex);
2782 spin_lock(&root->fs_info->ordered_extent_lock);
2784 list_splice_init(&root->fs_info->ordered_operations, &splice);
2785 while (!list_empty(&splice)) {
2786 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2787 ordered_operations);
2789 list_del_init(&btrfs_inode->ordered_operations);
2791 btrfs_invalidate_inodes(btrfs_inode->root);
2794 spin_unlock(&root->fs_info->ordered_extent_lock);
2795 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2800 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2802 struct list_head splice;
2803 struct btrfs_ordered_extent *ordered;
2804 struct inode *inode;
2806 INIT_LIST_HEAD(&splice);
2808 spin_lock(&root->fs_info->ordered_extent_lock);
2810 list_splice_init(&root->fs_info->ordered_extents, &splice);
2811 while (!list_empty(&splice)) {
2812 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2815 list_del_init(&ordered->root_extent_list);
2816 atomic_inc(&ordered->refs);
2818 /* the inode may be getting freed (in sys_unlink path). */
2819 inode = igrab(ordered->inode);
2821 spin_unlock(&root->fs_info->ordered_extent_lock);
2825 atomic_set(&ordered->refs, 1);
2826 btrfs_put_ordered_extent(ordered);
2828 spin_lock(&root->fs_info->ordered_extent_lock);
2831 spin_unlock(&root->fs_info->ordered_extent_lock);
2836 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2837 struct btrfs_root *root)
2839 struct rb_node *node;
2840 struct btrfs_delayed_ref_root *delayed_refs;
2841 struct btrfs_delayed_ref_node *ref;
2844 delayed_refs = &trans->delayed_refs;
2846 spin_lock(&delayed_refs->lock);
2847 if (delayed_refs->num_entries == 0) {
2848 spin_unlock(&delayed_refs->lock);
2849 printk(KERN_INFO "delayed_refs has NO entry\n");
2853 node = rb_first(&delayed_refs->root);
2855 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2856 node = rb_next(node);
2859 rb_erase(&ref->rb_node, &delayed_refs->root);
2860 delayed_refs->num_entries--;
2862 atomic_set(&ref->refs, 1);
2863 if (btrfs_delayed_ref_is_head(ref)) {
2864 struct btrfs_delayed_ref_head *head;
2866 head = btrfs_delayed_node_to_head(ref);
2867 mutex_lock(&head->mutex);
2868 kfree(head->extent_op);
2869 delayed_refs->num_heads--;
2870 if (list_empty(&head->cluster))
2871 delayed_refs->num_heads_ready--;
2872 list_del_init(&head->cluster);
2873 mutex_unlock(&head->mutex);
2876 spin_unlock(&delayed_refs->lock);
2877 btrfs_put_delayed_ref(ref);
2880 spin_lock(&delayed_refs->lock);
2883 spin_unlock(&delayed_refs->lock);
2888 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2890 struct btrfs_pending_snapshot *snapshot;
2891 struct list_head splice;
2893 INIT_LIST_HEAD(&splice);
2895 list_splice_init(&t->pending_snapshots, &splice);
2897 while (!list_empty(&splice)) {
2898 snapshot = list_entry(splice.next,
2899 struct btrfs_pending_snapshot,
2902 list_del_init(&snapshot->list);
2910 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2912 struct btrfs_inode *btrfs_inode;
2913 struct list_head splice;
2915 INIT_LIST_HEAD(&splice);
2917 spin_lock(&root->fs_info->delalloc_lock);
2918 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2920 while (!list_empty(&splice)) {
2921 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2924 list_del_init(&btrfs_inode->delalloc_inodes);
2926 btrfs_invalidate_inodes(btrfs_inode->root);
2929 spin_unlock(&root->fs_info->delalloc_lock);
2934 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2935 struct extent_io_tree *dirty_pages,
2940 struct inode *btree_inode = root->fs_info->btree_inode;
2941 struct extent_buffer *eb;
2945 unsigned long index;
2948 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2953 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2954 while (start <= end) {
2955 index = start >> PAGE_CACHE_SHIFT;
2956 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2957 page = find_get_page(btree_inode->i_mapping, index);
2960 offset = page_offset(page);
2962 spin_lock(&dirty_pages->buffer_lock);
2963 eb = radix_tree_lookup(
2964 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2965 offset >> PAGE_CACHE_SHIFT);
2966 spin_unlock(&dirty_pages->buffer_lock);
2968 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2970 atomic_set(&eb->refs, 1);
2972 if (PageWriteback(page))
2973 end_page_writeback(page);
2976 if (PageDirty(page)) {
2977 clear_page_dirty_for_io(page);
2978 spin_lock_irq(&page->mapping->tree_lock);
2979 radix_tree_tag_clear(&page->mapping->page_tree,
2981 PAGECACHE_TAG_DIRTY);
2982 spin_unlock_irq(&page->mapping->tree_lock);
2985 page->mapping->a_ops->invalidatepage(page, 0);
2993 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2994 struct extent_io_tree *pinned_extents)
2996 struct extent_io_tree *unpin;
3001 unpin = pinned_extents;
3003 ret = find_first_extent_bit(unpin, 0, &start, &end,
3009 if (btrfs_test_opt(root, DISCARD))
3010 ret = btrfs_error_discard_extent(root, start,
3014 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3015 btrfs_error_unpin_extent_range(root, start, end);
3022 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3024 struct btrfs_transaction *t;
3029 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3031 spin_lock(&root->fs_info->trans_lock);
3032 list_splice_init(&root->fs_info->trans_list, &list);
3033 root->fs_info->trans_no_join = 1;
3034 spin_unlock(&root->fs_info->trans_lock);
3036 while (!list_empty(&list)) {
3037 t = list_entry(list.next, struct btrfs_transaction, list);
3041 btrfs_destroy_ordered_operations(root);
3043 btrfs_destroy_ordered_extents(root);
3045 btrfs_destroy_delayed_refs(t, root);
3047 btrfs_block_rsv_release(root,
3048 &root->fs_info->trans_block_rsv,
3049 t->dirty_pages.dirty_bytes);
3051 /* FIXME: cleanup wait for commit */
3054 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3055 wake_up(&root->fs_info->transaction_blocked_wait);
3058 if (waitqueue_active(&root->fs_info->transaction_wait))
3059 wake_up(&root->fs_info->transaction_wait);
3062 if (waitqueue_active(&t->commit_wait))
3063 wake_up(&t->commit_wait);
3065 btrfs_destroy_pending_snapshots(t);
3067 btrfs_destroy_delalloc_inodes(root);
3069 spin_lock(&root->fs_info->trans_lock);
3070 root->fs_info->running_transaction = NULL;
3071 spin_unlock(&root->fs_info->trans_lock);
3073 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3076 btrfs_destroy_pinned_extent(root,
3077 root->fs_info->pinned_extents);
3079 atomic_set(&t->use_count, 0);
3080 list_del_init(&t->list);
3081 memset(t, 0, sizeof(*t));
3082 kmem_cache_free(btrfs_transaction_cachep, t);
3085 spin_lock(&root->fs_info->trans_lock);
3086 root->fs_info->trans_no_join = 0;
3087 spin_unlock(&root->fs_info->trans_lock);
3088 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3093 static struct extent_io_ops btree_extent_io_ops = {
3094 .write_cache_pages_lock_hook = btree_lock_page_hook,
3095 .readpage_end_io_hook = btree_readpage_end_io_hook,
3096 .submit_bio_hook = btree_submit_bio_hook,
3097 /* note we're sharing with inode.c for the merge bio hook */
3098 .merge_bio_hook = btrfs_merge_bio_hook,