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;
220 char *map_token = NULL;
222 unsigned long map_start;
223 unsigned long map_len;
226 unsigned long inline_result;
228 len = buf->len - offset;
230 err = map_private_extent_buffer(buf, offset, 32,
232 &map_start, &map_len, KM_USER0);
235 cur_len = min(len, map_len - (offset - map_start));
236 crc = btrfs_csum_data(root, kaddr + offset - map_start,
240 unmap_extent_buffer(buf, map_token, KM_USER0);
242 if (csum_size > sizeof(inline_result)) {
243 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
247 result = (char *)&inline_result;
250 btrfs_csum_final(crc, result);
253 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
256 memcpy(&found, result, csum_size);
258 read_extent_buffer(buf, &val, 0, csum_size);
259 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
260 "failed on %llu wanted %X found %X "
262 root->fs_info->sb->s_id,
263 (unsigned long long)buf->start, val, found,
264 btrfs_header_level(buf));
265 if (result != (char *)&inline_result)
270 write_extent_buffer(buf, result, 0, csum_size);
272 if (result != (char *)&inline_result)
278 * we can't consider a given block up to date unless the transid of the
279 * block matches the transid in the parent node's pointer. This is how we
280 * detect blocks that either didn't get written at all or got written
281 * in the wrong place.
283 static int verify_parent_transid(struct extent_io_tree *io_tree,
284 struct extent_buffer *eb, u64 parent_transid)
286 struct extent_state *cached_state = NULL;
289 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
292 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
293 0, &cached_state, GFP_NOFS);
294 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
295 btrfs_header_generation(eb) == parent_transid) {
299 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
301 (unsigned long long)eb->start,
302 (unsigned long long)parent_transid,
303 (unsigned long long)btrfs_header_generation(eb));
305 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
307 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
308 &cached_state, GFP_NOFS);
313 * helper to read a given tree block, doing retries as required when
314 * the checksums don't match and we have alternate mirrors to try.
316 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
317 struct extent_buffer *eb,
318 u64 start, u64 parent_transid)
320 struct extent_io_tree *io_tree;
325 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
326 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
328 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
329 btree_get_extent, mirror_num);
331 !verify_parent_transid(io_tree, eb, parent_transid))
335 * This buffer's crc is fine, but its contents are corrupted, so
336 * there is no reason to read the other copies, they won't be
339 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
342 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
348 if (mirror_num > num_copies)
355 * checksum a dirty tree block before IO. This has extra checks to make sure
356 * we only fill in the checksum field in the first page of a multi-page block
359 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
361 struct extent_io_tree *tree;
362 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
365 struct extent_buffer *eb;
368 tree = &BTRFS_I(page->mapping->host)->io_tree;
370 if (page->private == EXTENT_PAGE_PRIVATE) {
374 if (!page->private) {
378 len = page->private >> 2;
381 eb = alloc_extent_buffer(tree, start, len, page);
386 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
387 btrfs_header_generation(eb));
389 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
391 found_start = btrfs_header_bytenr(eb);
392 if (found_start != start) {
396 if (eb->first_page != page) {
400 if (!PageUptodate(page)) {
404 csum_tree_block(root, eb, 0);
406 free_extent_buffer(eb);
411 static int check_tree_block_fsid(struct btrfs_root *root,
412 struct extent_buffer *eb)
414 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
415 u8 fsid[BTRFS_UUID_SIZE];
418 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
421 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
425 fs_devices = fs_devices->seed;
430 #define CORRUPT(reason, eb, root, slot) \
431 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
432 "root=%llu, slot=%d\n", reason, \
433 (unsigned long long)btrfs_header_bytenr(eb), \
434 (unsigned long long)root->objectid, slot)
436 static noinline int check_leaf(struct btrfs_root *root,
437 struct extent_buffer *leaf)
439 struct btrfs_key key;
440 struct btrfs_key leaf_key;
441 u32 nritems = btrfs_header_nritems(leaf);
447 /* Check the 0 item */
448 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
449 BTRFS_LEAF_DATA_SIZE(root)) {
450 CORRUPT("invalid item offset size pair", leaf, root, 0);
455 * Check to make sure each items keys are in the correct order and their
456 * offsets make sense. We only have to loop through nritems-1 because
457 * we check the current slot against the next slot, which verifies the
458 * next slot's offset+size makes sense and that the current's slot
461 for (slot = 0; slot < nritems - 1; slot++) {
462 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
463 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
465 /* Make sure the keys are in the right order */
466 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
467 CORRUPT("bad key order", leaf, root, slot);
472 * Make sure the offset and ends are right, remember that the
473 * item data starts at the end of the leaf and grows towards the
476 if (btrfs_item_offset_nr(leaf, slot) !=
477 btrfs_item_end_nr(leaf, slot + 1)) {
478 CORRUPT("slot offset bad", leaf, root, slot);
483 * Check to make sure that we don't point outside of the leaf,
484 * just incase all the items are consistent to eachother, but
485 * all point outside of the leaf.
487 if (btrfs_item_end_nr(leaf, slot) >
488 BTRFS_LEAF_DATA_SIZE(root)) {
489 CORRUPT("slot end outside of leaf", leaf, root, slot);
497 #ifdef CONFIG_DEBUG_LOCK_ALLOC
498 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
500 lockdep_set_class_and_name(&eb->lock,
501 &btrfs_eb_class[level],
502 btrfs_eb_name[level]);
506 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
507 struct extent_state *state)
509 struct extent_io_tree *tree;
513 struct extent_buffer *eb;
514 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
517 tree = &BTRFS_I(page->mapping->host)->io_tree;
518 if (page->private == EXTENT_PAGE_PRIVATE)
523 len = page->private >> 2;
526 eb = alloc_extent_buffer(tree, start, len, page);
532 found_start = btrfs_header_bytenr(eb);
533 if (found_start != start) {
534 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
536 (unsigned long long)found_start,
537 (unsigned long long)eb->start);
541 if (eb->first_page != page) {
542 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
543 eb->first_page->index, page->index);
548 if (check_tree_block_fsid(root, eb)) {
549 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
550 (unsigned long long)eb->start);
554 found_level = btrfs_header_level(eb);
556 btrfs_set_buffer_lockdep_class(eb, found_level);
558 ret = csum_tree_block(root, eb, 1);
565 * If this is a leaf block and it is corrupt, set the corrupt bit so
566 * that we don't try and read the other copies of this block, just
569 if (found_level == 0 && check_leaf(root, eb)) {
570 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
574 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
575 end = eb->start + end - 1;
577 free_extent_buffer(eb);
582 static void end_workqueue_bio(struct bio *bio, int err)
584 struct end_io_wq *end_io_wq = bio->bi_private;
585 struct btrfs_fs_info *fs_info;
587 fs_info = end_io_wq->info;
588 end_io_wq->error = err;
589 end_io_wq->work.func = end_workqueue_fn;
590 end_io_wq->work.flags = 0;
592 if (bio->bi_rw & REQ_WRITE) {
593 if (end_io_wq->metadata == 1)
594 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
596 else if (end_io_wq->metadata == 2)
597 btrfs_queue_worker(&fs_info->endio_freespace_worker,
600 btrfs_queue_worker(&fs_info->endio_write_workers,
603 if (end_io_wq->metadata)
604 btrfs_queue_worker(&fs_info->endio_meta_workers,
607 btrfs_queue_worker(&fs_info->endio_workers,
613 * For the metadata arg you want
616 * 1 - if normal metadta
617 * 2 - if writing to the free space cache area
619 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
622 struct end_io_wq *end_io_wq;
623 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
627 end_io_wq->private = bio->bi_private;
628 end_io_wq->end_io = bio->bi_end_io;
629 end_io_wq->info = info;
630 end_io_wq->error = 0;
631 end_io_wq->bio = bio;
632 end_io_wq->metadata = metadata;
634 bio->bi_private = end_io_wq;
635 bio->bi_end_io = end_workqueue_bio;
639 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
641 unsigned long limit = min_t(unsigned long,
642 info->workers.max_workers,
643 info->fs_devices->open_devices);
647 static void run_one_async_start(struct btrfs_work *work)
649 struct async_submit_bio *async;
651 async = container_of(work, struct async_submit_bio, work);
652 async->submit_bio_start(async->inode, async->rw, async->bio,
653 async->mirror_num, async->bio_flags,
657 static void run_one_async_done(struct btrfs_work *work)
659 struct btrfs_fs_info *fs_info;
660 struct async_submit_bio *async;
663 async = container_of(work, struct async_submit_bio, work);
664 fs_info = BTRFS_I(async->inode)->root->fs_info;
666 limit = btrfs_async_submit_limit(fs_info);
667 limit = limit * 2 / 3;
669 atomic_dec(&fs_info->nr_async_submits);
671 if (atomic_read(&fs_info->nr_async_submits) < limit &&
672 waitqueue_active(&fs_info->async_submit_wait))
673 wake_up(&fs_info->async_submit_wait);
675 async->submit_bio_done(async->inode, async->rw, async->bio,
676 async->mirror_num, async->bio_flags,
680 static void run_one_async_free(struct btrfs_work *work)
682 struct async_submit_bio *async;
684 async = container_of(work, struct async_submit_bio, work);
688 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
689 int rw, struct bio *bio, int mirror_num,
690 unsigned long bio_flags,
692 extent_submit_bio_hook_t *submit_bio_start,
693 extent_submit_bio_hook_t *submit_bio_done)
695 struct async_submit_bio *async;
697 async = kmalloc(sizeof(*async), GFP_NOFS);
701 async->inode = inode;
704 async->mirror_num = mirror_num;
705 async->submit_bio_start = submit_bio_start;
706 async->submit_bio_done = submit_bio_done;
708 async->work.func = run_one_async_start;
709 async->work.ordered_func = run_one_async_done;
710 async->work.ordered_free = run_one_async_free;
712 async->work.flags = 0;
713 async->bio_flags = bio_flags;
714 async->bio_offset = bio_offset;
716 atomic_inc(&fs_info->nr_async_submits);
719 btrfs_set_work_high_prio(&async->work);
721 btrfs_queue_worker(&fs_info->workers, &async->work);
723 while (atomic_read(&fs_info->async_submit_draining) &&
724 atomic_read(&fs_info->nr_async_submits)) {
725 wait_event(fs_info->async_submit_wait,
726 (atomic_read(&fs_info->nr_async_submits) == 0));
732 static int btree_csum_one_bio(struct bio *bio)
734 struct bio_vec *bvec = bio->bi_io_vec;
736 struct btrfs_root *root;
738 WARN_ON(bio->bi_vcnt <= 0);
739 while (bio_index < bio->bi_vcnt) {
740 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
741 csum_dirty_buffer(root, bvec->bv_page);
748 static int __btree_submit_bio_start(struct inode *inode, int rw,
749 struct bio *bio, int mirror_num,
750 unsigned long bio_flags,
754 * when we're called for a write, we're already in the async
755 * submission context. Just jump into btrfs_map_bio
757 btree_csum_one_bio(bio);
761 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
762 int mirror_num, unsigned long bio_flags,
766 * when we're called for a write, we're already in the async
767 * submission context. Just jump into btrfs_map_bio
769 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
772 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
773 int mirror_num, unsigned long bio_flags,
778 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
782 if (!(rw & REQ_WRITE)) {
784 * called for a read, do the setup so that checksum validation
785 * can happen in the async kernel threads
787 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
792 * kthread helpers are used to submit writes so that checksumming
793 * can happen in parallel across all CPUs
795 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
796 inode, rw, bio, mirror_num, 0,
798 __btree_submit_bio_start,
799 __btree_submit_bio_done);
802 #ifdef CONFIG_MIGRATION
803 static int btree_migratepage(struct address_space *mapping,
804 struct page *newpage, struct page *page)
807 * we can't safely write a btree page from here,
808 * we haven't done the locking hook
813 * Buffers may be managed in a filesystem specific way.
814 * We must have no buffers or drop them.
816 if (page_has_private(page) &&
817 !try_to_release_page(page, GFP_KERNEL))
819 return migrate_page(mapping, newpage, page);
823 static int btree_writepage(struct page *page, struct writeback_control *wbc)
825 struct extent_io_tree *tree;
826 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
827 struct extent_buffer *eb;
830 tree = &BTRFS_I(page->mapping->host)->io_tree;
831 if (!(current->flags & PF_MEMALLOC)) {
832 return extent_write_full_page(tree, page,
833 btree_get_extent, wbc);
836 redirty_page_for_writepage(wbc, page);
837 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
840 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
842 spin_lock(&root->fs_info->delalloc_lock);
843 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
844 spin_unlock(&root->fs_info->delalloc_lock);
846 free_extent_buffer(eb);
852 static int btree_writepages(struct address_space *mapping,
853 struct writeback_control *wbc)
855 struct extent_io_tree *tree;
856 tree = &BTRFS_I(mapping->host)->io_tree;
857 if (wbc->sync_mode == WB_SYNC_NONE) {
858 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
860 unsigned long thresh = 32 * 1024 * 1024;
862 if (wbc->for_kupdate)
865 /* this is a bit racy, but that's ok */
866 num_dirty = root->fs_info->dirty_metadata_bytes;
867 if (num_dirty < thresh)
870 return extent_writepages(tree, mapping, btree_get_extent, wbc);
873 static int btree_readpage(struct file *file, struct page *page)
875 struct extent_io_tree *tree;
876 tree = &BTRFS_I(page->mapping->host)->io_tree;
877 return extent_read_full_page(tree, page, btree_get_extent);
880 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
882 struct extent_io_tree *tree;
883 struct extent_map_tree *map;
886 if (PageWriteback(page) || PageDirty(page))
889 tree = &BTRFS_I(page->mapping->host)->io_tree;
890 map = &BTRFS_I(page->mapping->host)->extent_tree;
892 ret = try_release_extent_state(map, tree, page, gfp_flags);
896 ret = try_release_extent_buffer(tree, page);
898 ClearPagePrivate(page);
899 set_page_private(page, 0);
900 page_cache_release(page);
906 static void btree_invalidatepage(struct page *page, unsigned long offset)
908 struct extent_io_tree *tree;
909 tree = &BTRFS_I(page->mapping->host)->io_tree;
910 extent_invalidatepage(tree, page, offset);
911 btree_releasepage(page, GFP_NOFS);
912 if (PagePrivate(page)) {
913 printk(KERN_WARNING "btrfs warning page private not zero "
914 "on page %llu\n", (unsigned long long)page_offset(page));
915 ClearPagePrivate(page);
916 set_page_private(page, 0);
917 page_cache_release(page);
921 static const struct address_space_operations btree_aops = {
922 .readpage = btree_readpage,
923 .writepage = btree_writepage,
924 .writepages = btree_writepages,
925 .releasepage = btree_releasepage,
926 .invalidatepage = btree_invalidatepage,
927 #ifdef CONFIG_MIGRATION
928 .migratepage = btree_migratepage,
932 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
935 struct extent_buffer *buf = NULL;
936 struct inode *btree_inode = root->fs_info->btree_inode;
939 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
942 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
943 buf, 0, 0, btree_get_extent, 0);
944 free_extent_buffer(buf);
948 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
949 u64 bytenr, u32 blocksize)
951 struct inode *btree_inode = root->fs_info->btree_inode;
952 struct extent_buffer *eb;
953 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
958 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
959 u64 bytenr, u32 blocksize)
961 struct inode *btree_inode = root->fs_info->btree_inode;
962 struct extent_buffer *eb;
964 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
965 bytenr, blocksize, NULL);
970 int btrfs_write_tree_block(struct extent_buffer *buf)
972 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
973 buf->start + buf->len - 1);
976 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
978 return filemap_fdatawait_range(buf->first_page->mapping,
979 buf->start, buf->start + buf->len - 1);
982 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
983 u32 blocksize, u64 parent_transid)
985 struct extent_buffer *buf = NULL;
988 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
992 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
995 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1000 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1001 struct extent_buffer *buf)
1003 struct inode *btree_inode = root->fs_info->btree_inode;
1004 if (btrfs_header_generation(buf) ==
1005 root->fs_info->running_transaction->transid) {
1006 btrfs_assert_tree_locked(buf);
1008 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1009 spin_lock(&root->fs_info->delalloc_lock);
1010 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1011 root->fs_info->dirty_metadata_bytes -= buf->len;
1014 spin_unlock(&root->fs_info->delalloc_lock);
1017 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1018 btrfs_set_lock_blocking(buf);
1019 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1025 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1026 u32 stripesize, struct btrfs_root *root,
1027 struct btrfs_fs_info *fs_info,
1031 root->commit_root = NULL;
1032 root->sectorsize = sectorsize;
1033 root->nodesize = nodesize;
1034 root->leafsize = leafsize;
1035 root->stripesize = stripesize;
1037 root->track_dirty = 0;
1039 root->orphan_item_inserted = 0;
1040 root->orphan_cleanup_state = 0;
1042 root->fs_info = fs_info;
1043 root->objectid = objectid;
1044 root->last_trans = 0;
1045 root->highest_objectid = 0;
1048 root->inode_tree = RB_ROOT;
1049 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1050 root->block_rsv = NULL;
1051 root->orphan_block_rsv = NULL;
1053 INIT_LIST_HEAD(&root->dirty_list);
1054 INIT_LIST_HEAD(&root->orphan_list);
1055 INIT_LIST_HEAD(&root->root_list);
1056 spin_lock_init(&root->orphan_lock);
1057 spin_lock_init(&root->inode_lock);
1058 spin_lock_init(&root->accounting_lock);
1059 mutex_init(&root->objectid_mutex);
1060 mutex_init(&root->log_mutex);
1061 init_waitqueue_head(&root->log_writer_wait);
1062 init_waitqueue_head(&root->log_commit_wait[0]);
1063 init_waitqueue_head(&root->log_commit_wait[1]);
1064 atomic_set(&root->log_commit[0], 0);
1065 atomic_set(&root->log_commit[1], 0);
1066 atomic_set(&root->log_writers, 0);
1067 root->log_batch = 0;
1068 root->log_transid = 0;
1069 root->last_log_commit = 0;
1070 extent_io_tree_init(&root->dirty_log_pages,
1071 fs_info->btree_inode->i_mapping);
1073 memset(&root->root_key, 0, sizeof(root->root_key));
1074 memset(&root->root_item, 0, sizeof(root->root_item));
1075 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1076 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1077 root->defrag_trans_start = fs_info->generation;
1078 init_completion(&root->kobj_unregister);
1079 root->defrag_running = 0;
1080 root->root_key.objectid = objectid;
1081 root->anon_super.s_root = NULL;
1082 root->anon_super.s_dev = 0;
1083 INIT_LIST_HEAD(&root->anon_super.s_list);
1084 INIT_LIST_HEAD(&root->anon_super.s_instances);
1085 init_rwsem(&root->anon_super.s_umount);
1090 static int find_and_setup_root(struct btrfs_root *tree_root,
1091 struct btrfs_fs_info *fs_info,
1093 struct btrfs_root *root)
1099 __setup_root(tree_root->nodesize, tree_root->leafsize,
1100 tree_root->sectorsize, tree_root->stripesize,
1101 root, fs_info, objectid);
1102 ret = btrfs_find_last_root(tree_root, objectid,
1103 &root->root_item, &root->root_key);
1108 generation = btrfs_root_generation(&root->root_item);
1109 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1110 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1111 blocksize, generation);
1112 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1113 free_extent_buffer(root->node);
1116 root->commit_root = btrfs_root_node(root);
1120 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1121 struct btrfs_fs_info *fs_info)
1123 struct btrfs_root *root;
1124 struct btrfs_root *tree_root = fs_info->tree_root;
1125 struct extent_buffer *leaf;
1127 root = kzalloc(sizeof(*root), GFP_NOFS);
1129 return ERR_PTR(-ENOMEM);
1131 __setup_root(tree_root->nodesize, tree_root->leafsize,
1132 tree_root->sectorsize, tree_root->stripesize,
1133 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1135 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1136 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1137 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1139 * log trees do not get reference counted because they go away
1140 * before a real commit is actually done. They do store pointers
1141 * to file data extents, and those reference counts still get
1142 * updated (along with back refs to the log tree).
1146 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1147 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1150 return ERR_CAST(leaf);
1153 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1154 btrfs_set_header_bytenr(leaf, leaf->start);
1155 btrfs_set_header_generation(leaf, trans->transid);
1156 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1157 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1160 write_extent_buffer(root->node, root->fs_info->fsid,
1161 (unsigned long)btrfs_header_fsid(root->node),
1163 btrfs_mark_buffer_dirty(root->node);
1164 btrfs_tree_unlock(root->node);
1168 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1169 struct btrfs_fs_info *fs_info)
1171 struct btrfs_root *log_root;
1173 log_root = alloc_log_tree(trans, fs_info);
1174 if (IS_ERR(log_root))
1175 return PTR_ERR(log_root);
1176 WARN_ON(fs_info->log_root_tree);
1177 fs_info->log_root_tree = log_root;
1181 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1182 struct btrfs_root *root)
1184 struct btrfs_root *log_root;
1185 struct btrfs_inode_item *inode_item;
1187 log_root = alloc_log_tree(trans, root->fs_info);
1188 if (IS_ERR(log_root))
1189 return PTR_ERR(log_root);
1191 log_root->last_trans = trans->transid;
1192 log_root->root_key.offset = root->root_key.objectid;
1194 inode_item = &log_root->root_item.inode;
1195 inode_item->generation = cpu_to_le64(1);
1196 inode_item->size = cpu_to_le64(3);
1197 inode_item->nlink = cpu_to_le32(1);
1198 inode_item->nbytes = cpu_to_le64(root->leafsize);
1199 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1201 btrfs_set_root_node(&log_root->root_item, log_root->node);
1203 WARN_ON(root->log_root);
1204 root->log_root = log_root;
1205 root->log_transid = 0;
1206 root->last_log_commit = 0;
1210 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1211 struct btrfs_key *location)
1213 struct btrfs_root *root;
1214 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1215 struct btrfs_path *path;
1216 struct extent_buffer *l;
1221 root = kzalloc(sizeof(*root), GFP_NOFS);
1223 return ERR_PTR(-ENOMEM);
1224 if (location->offset == (u64)-1) {
1225 ret = find_and_setup_root(tree_root, fs_info,
1226 location->objectid, root);
1229 return ERR_PTR(ret);
1234 __setup_root(tree_root->nodesize, tree_root->leafsize,
1235 tree_root->sectorsize, tree_root->stripesize,
1236 root, fs_info, location->objectid);
1238 path = btrfs_alloc_path();
1241 return ERR_PTR(-ENOMEM);
1243 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1246 read_extent_buffer(l, &root->root_item,
1247 btrfs_item_ptr_offset(l, path->slots[0]),
1248 sizeof(root->root_item));
1249 memcpy(&root->root_key, location, sizeof(*location));
1251 btrfs_free_path(path);
1256 return ERR_PTR(ret);
1259 generation = btrfs_root_generation(&root->root_item);
1260 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1261 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1262 blocksize, generation);
1263 root->commit_root = btrfs_root_node(root);
1264 BUG_ON(!root->node);
1266 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1268 btrfs_check_and_init_root_item(&root->root_item);
1274 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1275 struct btrfs_key *location)
1277 struct btrfs_root *root;
1280 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1281 return fs_info->tree_root;
1282 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1283 return fs_info->extent_root;
1284 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1285 return fs_info->chunk_root;
1286 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1287 return fs_info->dev_root;
1288 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1289 return fs_info->csum_root;
1291 spin_lock(&fs_info->fs_roots_radix_lock);
1292 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1293 (unsigned long)location->objectid);
1294 spin_unlock(&fs_info->fs_roots_radix_lock);
1298 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1302 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1303 if (!root->free_ino_ctl)
1305 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1307 if (!root->free_ino_pinned)
1310 btrfs_init_free_ino_ctl(root);
1311 mutex_init(&root->fs_commit_mutex);
1312 spin_lock_init(&root->cache_lock);
1313 init_waitqueue_head(&root->cache_wait);
1315 ret = set_anon_super(&root->anon_super, NULL);
1319 if (btrfs_root_refs(&root->root_item) == 0) {
1324 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1328 root->orphan_item_inserted = 1;
1330 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1334 spin_lock(&fs_info->fs_roots_radix_lock);
1335 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1336 (unsigned long)root->root_key.objectid,
1341 spin_unlock(&fs_info->fs_roots_radix_lock);
1342 radix_tree_preload_end();
1344 if (ret == -EEXIST) {
1351 ret = btrfs_find_dead_roots(fs_info->tree_root,
1352 root->root_key.objectid);
1357 return ERR_PTR(ret);
1360 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1362 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1364 struct btrfs_device *device;
1365 struct backing_dev_info *bdi;
1368 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1371 bdi = blk_get_backing_dev_info(device->bdev);
1372 if (bdi && bdi_congested(bdi, bdi_bits)) {
1382 * If this fails, caller must call bdi_destroy() to get rid of the
1385 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1389 bdi->capabilities = BDI_CAP_MAP_COPY;
1390 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1394 bdi->ra_pages = default_backing_dev_info.ra_pages;
1395 bdi->congested_fn = btrfs_congested_fn;
1396 bdi->congested_data = info;
1400 static int bio_ready_for_csum(struct bio *bio)
1406 struct extent_io_tree *io_tree = NULL;
1407 struct bio_vec *bvec;
1411 bio_for_each_segment(bvec, bio, i) {
1412 page = bvec->bv_page;
1413 if (page->private == EXTENT_PAGE_PRIVATE) {
1414 length += bvec->bv_len;
1417 if (!page->private) {
1418 length += bvec->bv_len;
1421 length = bvec->bv_len;
1422 buf_len = page->private >> 2;
1423 start = page_offset(page) + bvec->bv_offset;
1424 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1426 /* are we fully contained in this bio? */
1427 if (buf_len <= length)
1430 ret = extent_range_uptodate(io_tree, start + length,
1431 start + buf_len - 1);
1436 * called by the kthread helper functions to finally call the bio end_io
1437 * functions. This is where read checksum verification actually happens
1439 static void end_workqueue_fn(struct btrfs_work *work)
1442 struct end_io_wq *end_io_wq;
1443 struct btrfs_fs_info *fs_info;
1446 end_io_wq = container_of(work, struct end_io_wq, work);
1447 bio = end_io_wq->bio;
1448 fs_info = end_io_wq->info;
1450 /* metadata bio reads are special because the whole tree block must
1451 * be checksummed at once. This makes sure the entire block is in
1452 * ram and up to date before trying to verify things. For
1453 * blocksize <= pagesize, it is basically a noop
1455 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1456 !bio_ready_for_csum(bio)) {
1457 btrfs_queue_worker(&fs_info->endio_meta_workers,
1461 error = end_io_wq->error;
1462 bio->bi_private = end_io_wq->private;
1463 bio->bi_end_io = end_io_wq->end_io;
1465 bio_endio(bio, error);
1468 static int cleaner_kthread(void *arg)
1470 struct btrfs_root *root = arg;
1473 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1475 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1476 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1477 btrfs_run_delayed_iputs(root);
1478 btrfs_clean_old_snapshots(root);
1479 mutex_unlock(&root->fs_info->cleaner_mutex);
1480 btrfs_run_defrag_inodes(root->fs_info);
1483 if (freezing(current)) {
1486 set_current_state(TASK_INTERRUPTIBLE);
1487 if (!kthread_should_stop())
1489 __set_current_state(TASK_RUNNING);
1491 } while (!kthread_should_stop());
1495 static int transaction_kthread(void *arg)
1497 struct btrfs_root *root = arg;
1498 struct btrfs_trans_handle *trans;
1499 struct btrfs_transaction *cur;
1502 unsigned long delay;
1507 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1508 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1510 spin_lock(&root->fs_info->trans_lock);
1511 cur = root->fs_info->running_transaction;
1513 spin_unlock(&root->fs_info->trans_lock);
1517 now = get_seconds();
1518 if (!cur->blocked &&
1519 (now < cur->start_time || now - cur->start_time < 30)) {
1520 spin_unlock(&root->fs_info->trans_lock);
1524 transid = cur->transid;
1525 spin_unlock(&root->fs_info->trans_lock);
1527 trans = btrfs_join_transaction(root);
1528 BUG_ON(IS_ERR(trans));
1529 if (transid == trans->transid) {
1530 ret = btrfs_commit_transaction(trans, root);
1533 btrfs_end_transaction(trans, root);
1536 wake_up_process(root->fs_info->cleaner_kthread);
1537 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1539 if (freezing(current)) {
1542 set_current_state(TASK_INTERRUPTIBLE);
1543 if (!kthread_should_stop() &&
1544 !btrfs_transaction_blocked(root->fs_info))
1545 schedule_timeout(delay);
1546 __set_current_state(TASK_RUNNING);
1548 } while (!kthread_should_stop());
1552 struct btrfs_root *open_ctree(struct super_block *sb,
1553 struct btrfs_fs_devices *fs_devices,
1563 struct btrfs_key location;
1564 struct buffer_head *bh;
1565 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1567 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1569 struct btrfs_root *tree_root = btrfs_sb(sb);
1570 struct btrfs_fs_info *fs_info = NULL;
1571 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1573 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1575 struct btrfs_root *log_tree_root;
1580 struct btrfs_super_block *disk_super;
1582 if (!extent_root || !tree_root || !tree_root->fs_info ||
1583 !chunk_root || !dev_root || !csum_root) {
1587 fs_info = tree_root->fs_info;
1589 ret = init_srcu_struct(&fs_info->subvol_srcu);
1595 ret = setup_bdi(fs_info, &fs_info->bdi);
1601 fs_info->btree_inode = new_inode(sb);
1602 if (!fs_info->btree_inode) {
1607 fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1609 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1610 INIT_LIST_HEAD(&fs_info->trans_list);
1611 INIT_LIST_HEAD(&fs_info->dead_roots);
1612 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1613 INIT_LIST_HEAD(&fs_info->hashers);
1614 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1615 INIT_LIST_HEAD(&fs_info->ordered_operations);
1616 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1617 spin_lock_init(&fs_info->delalloc_lock);
1618 spin_lock_init(&fs_info->trans_lock);
1619 spin_lock_init(&fs_info->ref_cache_lock);
1620 spin_lock_init(&fs_info->fs_roots_radix_lock);
1621 spin_lock_init(&fs_info->delayed_iput_lock);
1622 spin_lock_init(&fs_info->defrag_inodes_lock);
1623 mutex_init(&fs_info->reloc_mutex);
1625 init_completion(&fs_info->kobj_unregister);
1626 fs_info->tree_root = tree_root;
1627 fs_info->extent_root = extent_root;
1628 fs_info->csum_root = csum_root;
1629 fs_info->chunk_root = chunk_root;
1630 fs_info->dev_root = dev_root;
1631 fs_info->fs_devices = fs_devices;
1632 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1633 INIT_LIST_HEAD(&fs_info->space_info);
1634 btrfs_mapping_init(&fs_info->mapping_tree);
1635 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1636 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1637 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1638 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1639 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1640 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1641 mutex_init(&fs_info->durable_block_rsv_mutex);
1642 atomic_set(&fs_info->nr_async_submits, 0);
1643 atomic_set(&fs_info->async_delalloc_pages, 0);
1644 atomic_set(&fs_info->async_submit_draining, 0);
1645 atomic_set(&fs_info->nr_async_bios, 0);
1646 atomic_set(&fs_info->defrag_running, 0);
1648 fs_info->max_inline = 8192 * 1024;
1649 fs_info->metadata_ratio = 0;
1650 fs_info->defrag_inodes = RB_ROOT;
1651 fs_info->trans_no_join = 0;
1653 fs_info->thread_pool_size = min_t(unsigned long,
1654 num_online_cpus() + 2, 8);
1656 INIT_LIST_HEAD(&fs_info->ordered_extents);
1657 spin_lock_init(&fs_info->ordered_extent_lock);
1658 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1660 if (!fs_info->delayed_root) {
1664 btrfs_init_delayed_root(fs_info->delayed_root);
1666 mutex_init(&fs_info->scrub_lock);
1667 atomic_set(&fs_info->scrubs_running, 0);
1668 atomic_set(&fs_info->scrub_pause_req, 0);
1669 atomic_set(&fs_info->scrubs_paused, 0);
1670 atomic_set(&fs_info->scrub_cancel_req, 0);
1671 init_waitqueue_head(&fs_info->scrub_pause_wait);
1672 init_rwsem(&fs_info->scrub_super_lock);
1673 fs_info->scrub_workers_refcnt = 0;
1675 sb->s_blocksize = 4096;
1676 sb->s_blocksize_bits = blksize_bits(4096);
1677 sb->s_bdi = &fs_info->bdi;
1679 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1680 fs_info->btree_inode->i_nlink = 1;
1682 * we set the i_size on the btree inode to the max possible int.
1683 * the real end of the address space is determined by all of
1684 * the devices in the system
1686 fs_info->btree_inode->i_size = OFFSET_MAX;
1687 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1688 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1690 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1691 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1692 fs_info->btree_inode->i_mapping);
1693 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1695 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1697 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1698 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1699 sizeof(struct btrfs_key));
1700 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1701 insert_inode_hash(fs_info->btree_inode);
1703 spin_lock_init(&fs_info->block_group_cache_lock);
1704 fs_info->block_group_cache_tree = RB_ROOT;
1706 extent_io_tree_init(&fs_info->freed_extents[0],
1707 fs_info->btree_inode->i_mapping);
1708 extent_io_tree_init(&fs_info->freed_extents[1],
1709 fs_info->btree_inode->i_mapping);
1710 fs_info->pinned_extents = &fs_info->freed_extents[0];
1711 fs_info->do_barriers = 1;
1714 mutex_init(&fs_info->ordered_operations_mutex);
1715 mutex_init(&fs_info->tree_log_mutex);
1716 mutex_init(&fs_info->chunk_mutex);
1717 mutex_init(&fs_info->transaction_kthread_mutex);
1718 mutex_init(&fs_info->cleaner_mutex);
1719 mutex_init(&fs_info->volume_mutex);
1720 init_rwsem(&fs_info->extent_commit_sem);
1721 init_rwsem(&fs_info->cleanup_work_sem);
1722 init_rwsem(&fs_info->subvol_sem);
1724 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1725 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1727 init_waitqueue_head(&fs_info->transaction_throttle);
1728 init_waitqueue_head(&fs_info->transaction_wait);
1729 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1730 init_waitqueue_head(&fs_info->async_submit_wait);
1732 __setup_root(4096, 4096, 4096, 4096, tree_root,
1733 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1735 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1741 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1742 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1743 sizeof(fs_info->super_for_commit));
1746 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1748 disk_super = &fs_info->super_copy;
1749 if (!btrfs_super_root(disk_super))
1752 /* check FS state, whether FS is broken. */
1753 fs_info->fs_state |= btrfs_super_flags(disk_super);
1755 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1758 * In the long term, we'll store the compression type in the super
1759 * block, and it'll be used for per file compression control.
1761 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1763 ret = btrfs_parse_options(tree_root, options);
1769 features = btrfs_super_incompat_flags(disk_super) &
1770 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1772 printk(KERN_ERR "BTRFS: couldn't mount because of "
1773 "unsupported optional features (%Lx).\n",
1774 (unsigned long long)features);
1779 features = btrfs_super_incompat_flags(disk_super);
1780 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1781 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1782 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1783 btrfs_set_super_incompat_flags(disk_super, features);
1785 features = btrfs_super_compat_ro_flags(disk_super) &
1786 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1787 if (!(sb->s_flags & MS_RDONLY) && features) {
1788 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1789 "unsupported option features (%Lx).\n",
1790 (unsigned long long)features);
1795 btrfs_init_workers(&fs_info->generic_worker,
1796 "genwork", 1, NULL);
1798 btrfs_init_workers(&fs_info->workers, "worker",
1799 fs_info->thread_pool_size,
1800 &fs_info->generic_worker);
1802 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1803 fs_info->thread_pool_size,
1804 &fs_info->generic_worker);
1806 btrfs_init_workers(&fs_info->submit_workers, "submit",
1807 min_t(u64, fs_devices->num_devices,
1808 fs_info->thread_pool_size),
1809 &fs_info->generic_worker);
1811 /* a higher idle thresh on the submit workers makes it much more
1812 * likely that bios will be send down in a sane order to the
1815 fs_info->submit_workers.idle_thresh = 64;
1817 fs_info->workers.idle_thresh = 16;
1818 fs_info->workers.ordered = 1;
1820 fs_info->delalloc_workers.idle_thresh = 2;
1821 fs_info->delalloc_workers.ordered = 1;
1823 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1824 &fs_info->generic_worker);
1825 btrfs_init_workers(&fs_info->endio_workers, "endio",
1826 fs_info->thread_pool_size,
1827 &fs_info->generic_worker);
1828 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1829 fs_info->thread_pool_size,
1830 &fs_info->generic_worker);
1831 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1832 "endio-meta-write", fs_info->thread_pool_size,
1833 &fs_info->generic_worker);
1834 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1835 fs_info->thread_pool_size,
1836 &fs_info->generic_worker);
1837 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1838 1, &fs_info->generic_worker);
1839 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1840 fs_info->thread_pool_size,
1841 &fs_info->generic_worker);
1844 * endios are largely parallel and should have a very
1847 fs_info->endio_workers.idle_thresh = 4;
1848 fs_info->endio_meta_workers.idle_thresh = 4;
1850 fs_info->endio_write_workers.idle_thresh = 2;
1851 fs_info->endio_meta_write_workers.idle_thresh = 2;
1853 btrfs_start_workers(&fs_info->workers, 1);
1854 btrfs_start_workers(&fs_info->generic_worker, 1);
1855 btrfs_start_workers(&fs_info->submit_workers, 1);
1856 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1857 btrfs_start_workers(&fs_info->fixup_workers, 1);
1858 btrfs_start_workers(&fs_info->endio_workers, 1);
1859 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1860 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1861 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1862 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1863 btrfs_start_workers(&fs_info->delayed_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);
2122 kfree(fs_info->delayed_root);
2124 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2125 iput(fs_info->btree_inode);
2127 btrfs_close_devices(fs_info->fs_devices);
2128 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2130 bdi_destroy(&fs_info->bdi);
2132 cleanup_srcu_struct(&fs_info->subvol_srcu);
2140 return ERR_PTR(err);
2143 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2145 char b[BDEVNAME_SIZE];
2148 set_buffer_uptodate(bh);
2150 printk_ratelimited(KERN_WARNING "lost page write due to "
2151 "I/O error on %s\n",
2152 bdevname(bh->b_bdev, b));
2153 /* note, we dont' set_buffer_write_io_error because we have
2154 * our own ways of dealing with the IO errors
2156 clear_buffer_uptodate(bh);
2162 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2164 struct buffer_head *bh;
2165 struct buffer_head *latest = NULL;
2166 struct btrfs_super_block *super;
2171 /* we would like to check all the supers, but that would make
2172 * a btrfs mount succeed after a mkfs from a different FS.
2173 * So, we need to add a special mount option to scan for
2174 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2176 for (i = 0; i < 1; i++) {
2177 bytenr = btrfs_sb_offset(i);
2178 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2180 bh = __bread(bdev, bytenr / 4096, 4096);
2184 super = (struct btrfs_super_block *)bh->b_data;
2185 if (btrfs_super_bytenr(super) != bytenr ||
2186 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2187 sizeof(super->magic))) {
2192 if (!latest || btrfs_super_generation(super) > transid) {
2195 transid = btrfs_super_generation(super);
2204 * this should be called twice, once with wait == 0 and
2205 * once with wait == 1. When wait == 0 is done, all the buffer heads
2206 * we write are pinned.
2208 * They are released when wait == 1 is done.
2209 * max_mirrors must be the same for both runs, and it indicates how
2210 * many supers on this one device should be written.
2212 * max_mirrors == 0 means to write them all.
2214 static int write_dev_supers(struct btrfs_device *device,
2215 struct btrfs_super_block *sb,
2216 int do_barriers, int wait, int max_mirrors)
2218 struct buffer_head *bh;
2224 int last_barrier = 0;
2226 if (max_mirrors == 0)
2227 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2229 /* make sure only the last submit_bh does a barrier */
2231 for (i = 0; i < max_mirrors; i++) {
2232 bytenr = btrfs_sb_offset(i);
2233 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2234 device->total_bytes)
2240 for (i = 0; i < max_mirrors; i++) {
2241 bytenr = btrfs_sb_offset(i);
2242 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2246 bh = __find_get_block(device->bdev, bytenr / 4096,
2247 BTRFS_SUPER_INFO_SIZE);
2250 if (!buffer_uptodate(bh))
2253 /* drop our reference */
2256 /* drop the reference from the wait == 0 run */
2260 btrfs_set_super_bytenr(sb, bytenr);
2263 crc = btrfs_csum_data(NULL, (char *)sb +
2264 BTRFS_CSUM_SIZE, crc,
2265 BTRFS_SUPER_INFO_SIZE -
2267 btrfs_csum_final(crc, sb->csum);
2270 * one reference for us, and we leave it for the
2273 bh = __getblk(device->bdev, bytenr / 4096,
2274 BTRFS_SUPER_INFO_SIZE);
2275 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2277 /* one reference for submit_bh */
2280 set_buffer_uptodate(bh);
2282 bh->b_end_io = btrfs_end_buffer_write_sync;
2285 if (i == last_barrier && do_barriers)
2286 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2288 ret = submit_bh(WRITE_SYNC, bh);
2293 return errors < i ? 0 : -1;
2296 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2298 struct list_head *head;
2299 struct btrfs_device *dev;
2300 struct btrfs_super_block *sb;
2301 struct btrfs_dev_item *dev_item;
2305 int total_errors = 0;
2308 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2309 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2311 sb = &root->fs_info->super_for_commit;
2312 dev_item = &sb->dev_item;
2314 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2315 head = &root->fs_info->fs_devices->devices;
2316 list_for_each_entry_rcu(dev, head, dev_list) {
2321 if (!dev->in_fs_metadata || !dev->writeable)
2324 btrfs_set_stack_device_generation(dev_item, 0);
2325 btrfs_set_stack_device_type(dev_item, dev->type);
2326 btrfs_set_stack_device_id(dev_item, dev->devid);
2327 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2328 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2329 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2330 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2331 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2332 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2333 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2335 flags = btrfs_super_flags(sb);
2336 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2338 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2342 if (total_errors > max_errors) {
2343 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2349 list_for_each_entry_rcu(dev, head, dev_list) {
2352 if (!dev->in_fs_metadata || !dev->writeable)
2355 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2359 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2360 if (total_errors > max_errors) {
2361 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2368 int write_ctree_super(struct btrfs_trans_handle *trans,
2369 struct btrfs_root *root, int max_mirrors)
2373 ret = write_all_supers(root, max_mirrors);
2377 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2379 spin_lock(&fs_info->fs_roots_radix_lock);
2380 radix_tree_delete(&fs_info->fs_roots_radix,
2381 (unsigned long)root->root_key.objectid);
2382 spin_unlock(&fs_info->fs_roots_radix_lock);
2384 if (btrfs_root_refs(&root->root_item) == 0)
2385 synchronize_srcu(&fs_info->subvol_srcu);
2387 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2388 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2393 static void free_fs_root(struct btrfs_root *root)
2395 iput(root->cache_inode);
2396 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2397 if (root->anon_super.s_dev) {
2398 down_write(&root->anon_super.s_umount);
2399 kill_anon_super(&root->anon_super);
2401 free_extent_buffer(root->node);
2402 free_extent_buffer(root->commit_root);
2403 kfree(root->free_ino_ctl);
2404 kfree(root->free_ino_pinned);
2409 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2412 struct btrfs_root *gang[8];
2415 while (!list_empty(&fs_info->dead_roots)) {
2416 gang[0] = list_entry(fs_info->dead_roots.next,
2417 struct btrfs_root, root_list);
2418 list_del(&gang[0]->root_list);
2420 if (gang[0]->in_radix) {
2421 btrfs_free_fs_root(fs_info, gang[0]);
2423 free_extent_buffer(gang[0]->node);
2424 free_extent_buffer(gang[0]->commit_root);
2430 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2435 for (i = 0; i < ret; i++)
2436 btrfs_free_fs_root(fs_info, gang[i]);
2441 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2443 u64 root_objectid = 0;
2444 struct btrfs_root *gang[8];
2449 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2450 (void **)gang, root_objectid,
2455 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2456 for (i = 0; i < ret; i++) {
2459 root_objectid = gang[i]->root_key.objectid;
2460 err = btrfs_orphan_cleanup(gang[i]);
2469 int btrfs_commit_super(struct btrfs_root *root)
2471 struct btrfs_trans_handle *trans;
2474 mutex_lock(&root->fs_info->cleaner_mutex);
2475 btrfs_run_delayed_iputs(root);
2476 btrfs_clean_old_snapshots(root);
2477 mutex_unlock(&root->fs_info->cleaner_mutex);
2479 /* wait until ongoing cleanup work done */
2480 down_write(&root->fs_info->cleanup_work_sem);
2481 up_write(&root->fs_info->cleanup_work_sem);
2483 trans = btrfs_join_transaction(root);
2485 return PTR_ERR(trans);
2486 ret = btrfs_commit_transaction(trans, root);
2488 /* run commit again to drop the original snapshot */
2489 trans = btrfs_join_transaction(root);
2491 return PTR_ERR(trans);
2492 btrfs_commit_transaction(trans, root);
2493 ret = btrfs_write_and_wait_transaction(NULL, root);
2496 ret = write_ctree_super(NULL, root, 0);
2500 int close_ctree(struct btrfs_root *root)
2502 struct btrfs_fs_info *fs_info = root->fs_info;
2505 fs_info->closing = 1;
2508 btrfs_scrub_cancel(root);
2510 /* wait for any defraggers to finish */
2511 wait_event(fs_info->transaction_wait,
2512 (atomic_read(&fs_info->defrag_running) == 0));
2514 /* clear out the rbtree of defraggable inodes */
2515 btrfs_run_defrag_inodes(root->fs_info);
2517 btrfs_put_block_group_cache(fs_info);
2520 * Here come 2 situations when btrfs is broken to flip readonly:
2522 * 1. when btrfs flips readonly somewhere else before
2523 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2524 * and btrfs will skip to write sb directly to keep
2525 * ERROR state on disk.
2527 * 2. when btrfs flips readonly just in btrfs_commit_super,
2528 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2529 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2530 * btrfs will cleanup all FS resources first and write sb then.
2532 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2533 ret = btrfs_commit_super(root);
2535 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2538 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2539 ret = btrfs_error_commit_super(root);
2541 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2544 kthread_stop(root->fs_info->transaction_kthread);
2545 kthread_stop(root->fs_info->cleaner_kthread);
2547 fs_info->closing = 2;
2550 if (fs_info->delalloc_bytes) {
2551 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2552 (unsigned long long)fs_info->delalloc_bytes);
2554 if (fs_info->total_ref_cache_size) {
2555 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2556 (unsigned long long)fs_info->total_ref_cache_size);
2559 free_extent_buffer(fs_info->extent_root->node);
2560 free_extent_buffer(fs_info->extent_root->commit_root);
2561 free_extent_buffer(fs_info->tree_root->node);
2562 free_extent_buffer(fs_info->tree_root->commit_root);
2563 free_extent_buffer(root->fs_info->chunk_root->node);
2564 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2565 free_extent_buffer(root->fs_info->dev_root->node);
2566 free_extent_buffer(root->fs_info->dev_root->commit_root);
2567 free_extent_buffer(root->fs_info->csum_root->node);
2568 free_extent_buffer(root->fs_info->csum_root->commit_root);
2570 btrfs_free_block_groups(root->fs_info);
2572 del_fs_roots(fs_info);
2574 iput(fs_info->btree_inode);
2575 kfree(fs_info->delayed_root);
2577 btrfs_stop_workers(&fs_info->generic_worker);
2578 btrfs_stop_workers(&fs_info->fixup_workers);
2579 btrfs_stop_workers(&fs_info->delalloc_workers);
2580 btrfs_stop_workers(&fs_info->workers);
2581 btrfs_stop_workers(&fs_info->endio_workers);
2582 btrfs_stop_workers(&fs_info->endio_meta_workers);
2583 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2584 btrfs_stop_workers(&fs_info->endio_write_workers);
2585 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2586 btrfs_stop_workers(&fs_info->submit_workers);
2587 btrfs_stop_workers(&fs_info->delayed_workers);
2589 btrfs_close_devices(fs_info->fs_devices);
2590 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2592 bdi_destroy(&fs_info->bdi);
2593 cleanup_srcu_struct(&fs_info->subvol_srcu);
2595 kfree(fs_info->extent_root);
2596 kfree(fs_info->tree_root);
2597 kfree(fs_info->chunk_root);
2598 kfree(fs_info->dev_root);
2599 kfree(fs_info->csum_root);
2605 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2608 struct inode *btree_inode = buf->first_page->mapping->host;
2610 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2615 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2620 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2622 struct inode *btree_inode = buf->first_page->mapping->host;
2623 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2627 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2629 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2630 u64 transid = btrfs_header_generation(buf);
2631 struct inode *btree_inode = root->fs_info->btree_inode;
2634 btrfs_assert_tree_locked(buf);
2635 if (transid != root->fs_info->generation) {
2636 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2637 "found %llu running %llu\n",
2638 (unsigned long long)buf->start,
2639 (unsigned long long)transid,
2640 (unsigned long long)root->fs_info->generation);
2643 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2646 spin_lock(&root->fs_info->delalloc_lock);
2647 root->fs_info->dirty_metadata_bytes += buf->len;
2648 spin_unlock(&root->fs_info->delalloc_lock);
2652 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2655 * looks as though older kernels can get into trouble with
2656 * this code, they end up stuck in balance_dirty_pages forever
2659 unsigned long thresh = 32 * 1024 * 1024;
2661 if (current->flags & PF_MEMALLOC)
2664 btrfs_balance_delayed_items(root);
2666 num_dirty = root->fs_info->dirty_metadata_bytes;
2668 if (num_dirty > thresh) {
2669 balance_dirty_pages_ratelimited_nr(
2670 root->fs_info->btree_inode->i_mapping, 1);
2675 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2678 * looks as though older kernels can get into trouble with
2679 * this code, they end up stuck in balance_dirty_pages forever
2682 unsigned long thresh = 32 * 1024 * 1024;
2684 if (current->flags & PF_MEMALLOC)
2687 num_dirty = root->fs_info->dirty_metadata_bytes;
2689 if (num_dirty > thresh) {
2690 balance_dirty_pages_ratelimited_nr(
2691 root->fs_info->btree_inode->i_mapping, 1);
2696 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2698 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2700 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2702 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2706 int btree_lock_page_hook(struct page *page)
2708 struct inode *inode = page->mapping->host;
2709 struct btrfs_root *root = BTRFS_I(inode)->root;
2710 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2711 struct extent_buffer *eb;
2713 u64 bytenr = page_offset(page);
2715 if (page->private == EXTENT_PAGE_PRIVATE)
2718 len = page->private >> 2;
2719 eb = find_extent_buffer(io_tree, bytenr, len);
2723 btrfs_tree_lock(eb);
2724 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2726 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2727 spin_lock(&root->fs_info->delalloc_lock);
2728 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2729 root->fs_info->dirty_metadata_bytes -= eb->len;
2732 spin_unlock(&root->fs_info->delalloc_lock);
2735 btrfs_tree_unlock(eb);
2736 free_extent_buffer(eb);
2742 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2748 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2749 printk(KERN_WARNING "warning: mount fs with errors, "
2750 "running btrfsck is recommended\n");
2753 int btrfs_error_commit_super(struct btrfs_root *root)
2757 mutex_lock(&root->fs_info->cleaner_mutex);
2758 btrfs_run_delayed_iputs(root);
2759 mutex_unlock(&root->fs_info->cleaner_mutex);
2761 down_write(&root->fs_info->cleanup_work_sem);
2762 up_write(&root->fs_info->cleanup_work_sem);
2764 /* cleanup FS via transaction */
2765 btrfs_cleanup_transaction(root);
2767 ret = write_ctree_super(NULL, root, 0);
2772 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2774 struct btrfs_inode *btrfs_inode;
2775 struct list_head splice;
2777 INIT_LIST_HEAD(&splice);
2779 mutex_lock(&root->fs_info->ordered_operations_mutex);
2780 spin_lock(&root->fs_info->ordered_extent_lock);
2782 list_splice_init(&root->fs_info->ordered_operations, &splice);
2783 while (!list_empty(&splice)) {
2784 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2785 ordered_operations);
2787 list_del_init(&btrfs_inode->ordered_operations);
2789 btrfs_invalidate_inodes(btrfs_inode->root);
2792 spin_unlock(&root->fs_info->ordered_extent_lock);
2793 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2798 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2800 struct list_head splice;
2801 struct btrfs_ordered_extent *ordered;
2802 struct inode *inode;
2804 INIT_LIST_HEAD(&splice);
2806 spin_lock(&root->fs_info->ordered_extent_lock);
2808 list_splice_init(&root->fs_info->ordered_extents, &splice);
2809 while (!list_empty(&splice)) {
2810 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2813 list_del_init(&ordered->root_extent_list);
2814 atomic_inc(&ordered->refs);
2816 /* the inode may be getting freed (in sys_unlink path). */
2817 inode = igrab(ordered->inode);
2819 spin_unlock(&root->fs_info->ordered_extent_lock);
2823 atomic_set(&ordered->refs, 1);
2824 btrfs_put_ordered_extent(ordered);
2826 spin_lock(&root->fs_info->ordered_extent_lock);
2829 spin_unlock(&root->fs_info->ordered_extent_lock);
2834 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2835 struct btrfs_root *root)
2837 struct rb_node *node;
2838 struct btrfs_delayed_ref_root *delayed_refs;
2839 struct btrfs_delayed_ref_node *ref;
2842 delayed_refs = &trans->delayed_refs;
2844 spin_lock(&delayed_refs->lock);
2845 if (delayed_refs->num_entries == 0) {
2846 spin_unlock(&delayed_refs->lock);
2847 printk(KERN_INFO "delayed_refs has NO entry\n");
2851 node = rb_first(&delayed_refs->root);
2853 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2854 node = rb_next(node);
2857 rb_erase(&ref->rb_node, &delayed_refs->root);
2858 delayed_refs->num_entries--;
2860 atomic_set(&ref->refs, 1);
2861 if (btrfs_delayed_ref_is_head(ref)) {
2862 struct btrfs_delayed_ref_head *head;
2864 head = btrfs_delayed_node_to_head(ref);
2865 mutex_lock(&head->mutex);
2866 kfree(head->extent_op);
2867 delayed_refs->num_heads--;
2868 if (list_empty(&head->cluster))
2869 delayed_refs->num_heads_ready--;
2870 list_del_init(&head->cluster);
2871 mutex_unlock(&head->mutex);
2874 spin_unlock(&delayed_refs->lock);
2875 btrfs_put_delayed_ref(ref);
2878 spin_lock(&delayed_refs->lock);
2881 spin_unlock(&delayed_refs->lock);
2886 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2888 struct btrfs_pending_snapshot *snapshot;
2889 struct list_head splice;
2891 INIT_LIST_HEAD(&splice);
2893 list_splice_init(&t->pending_snapshots, &splice);
2895 while (!list_empty(&splice)) {
2896 snapshot = list_entry(splice.next,
2897 struct btrfs_pending_snapshot,
2900 list_del_init(&snapshot->list);
2908 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2910 struct btrfs_inode *btrfs_inode;
2911 struct list_head splice;
2913 INIT_LIST_HEAD(&splice);
2915 spin_lock(&root->fs_info->delalloc_lock);
2916 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2918 while (!list_empty(&splice)) {
2919 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2922 list_del_init(&btrfs_inode->delalloc_inodes);
2924 btrfs_invalidate_inodes(btrfs_inode->root);
2927 spin_unlock(&root->fs_info->delalloc_lock);
2932 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2933 struct extent_io_tree *dirty_pages,
2938 struct inode *btree_inode = root->fs_info->btree_inode;
2939 struct extent_buffer *eb;
2943 unsigned long index;
2946 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2951 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2952 while (start <= end) {
2953 index = start >> PAGE_CACHE_SHIFT;
2954 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2955 page = find_get_page(btree_inode->i_mapping, index);
2958 offset = page_offset(page);
2960 spin_lock(&dirty_pages->buffer_lock);
2961 eb = radix_tree_lookup(
2962 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2963 offset >> PAGE_CACHE_SHIFT);
2964 spin_unlock(&dirty_pages->buffer_lock);
2966 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2968 atomic_set(&eb->refs, 1);
2970 if (PageWriteback(page))
2971 end_page_writeback(page);
2974 if (PageDirty(page)) {
2975 clear_page_dirty_for_io(page);
2976 spin_lock_irq(&page->mapping->tree_lock);
2977 radix_tree_tag_clear(&page->mapping->page_tree,
2979 PAGECACHE_TAG_DIRTY);
2980 spin_unlock_irq(&page->mapping->tree_lock);
2983 page->mapping->a_ops->invalidatepage(page, 0);
2991 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2992 struct extent_io_tree *pinned_extents)
2994 struct extent_io_tree *unpin;
2999 unpin = pinned_extents;
3001 ret = find_first_extent_bit(unpin, 0, &start, &end,
3007 if (btrfs_test_opt(root, DISCARD))
3008 ret = btrfs_error_discard_extent(root, start,
3012 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3013 btrfs_error_unpin_extent_range(root, start, end);
3020 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3022 struct btrfs_transaction *t;
3027 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3029 spin_lock(&root->fs_info->trans_lock);
3030 list_splice_init(&root->fs_info->trans_list, &list);
3031 root->fs_info->trans_no_join = 1;
3032 spin_unlock(&root->fs_info->trans_lock);
3034 while (!list_empty(&list)) {
3035 t = list_entry(list.next, struct btrfs_transaction, list);
3039 btrfs_destroy_ordered_operations(root);
3041 btrfs_destroy_ordered_extents(root);
3043 btrfs_destroy_delayed_refs(t, root);
3045 btrfs_block_rsv_release(root,
3046 &root->fs_info->trans_block_rsv,
3047 t->dirty_pages.dirty_bytes);
3049 /* FIXME: cleanup wait for commit */
3052 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3053 wake_up(&root->fs_info->transaction_blocked_wait);
3056 if (waitqueue_active(&root->fs_info->transaction_wait))
3057 wake_up(&root->fs_info->transaction_wait);
3060 if (waitqueue_active(&t->commit_wait))
3061 wake_up(&t->commit_wait);
3063 btrfs_destroy_pending_snapshots(t);
3065 btrfs_destroy_delalloc_inodes(root);
3067 spin_lock(&root->fs_info->trans_lock);
3068 root->fs_info->running_transaction = NULL;
3069 spin_unlock(&root->fs_info->trans_lock);
3071 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3074 btrfs_destroy_pinned_extent(root,
3075 root->fs_info->pinned_extents);
3077 atomic_set(&t->use_count, 0);
3078 list_del_init(&t->list);
3079 memset(t, 0, sizeof(*t));
3080 kmem_cache_free(btrfs_transaction_cachep, t);
3083 spin_lock(&root->fs_info->trans_lock);
3084 root->fs_info->trans_no_join = 0;
3085 spin_unlock(&root->fs_info->trans_lock);
3086 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3091 static struct extent_io_ops btree_extent_io_ops = {
3092 .write_cache_pages_lock_hook = btree_lock_page_hook,
3093 .readpage_end_io_hook = btree_readpage_end_io_hook,
3094 .submit_bio_hook = btree_submit_bio_hook,
3095 /* note we're sharing with inode.c for the merge bio hook */
3096 .merge_bio_hook = btrfs_merge_bio_hook,