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 <asm/unaligned.h>
36 #include "transaction.h"
37 #include "btrfs_inode.h"
39 #include "print-tree.h"
40 #include "async-thread.h"
43 #include "free-space-cache.h"
45 static struct extent_io_ops btree_extent_io_ops;
46 static void end_workqueue_fn(struct btrfs_work *work);
47 static void free_fs_root(struct btrfs_root *root);
48 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
50 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
51 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
52 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
53 struct btrfs_root *root);
54 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
55 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
56 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
57 struct extent_io_tree *dirty_pages,
59 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
60 struct extent_io_tree *pinned_extents);
61 static int btrfs_cleanup_transaction(struct btrfs_root *root);
64 * end_io_wq structs are used to do processing in task context when an IO is
65 * complete. This is used during reads to verify checksums, and it is used
66 * by writes to insert metadata for new file extents after IO is complete.
72 struct btrfs_fs_info *info;
75 struct list_head list;
76 struct btrfs_work work;
80 * async submit bios are used to offload expensive checksumming
81 * onto the worker threads. They checksum file and metadata bios
82 * just before they are sent down the IO stack.
84 struct async_submit_bio {
87 struct list_head list;
88 extent_submit_bio_hook_t *submit_bio_start;
89 extent_submit_bio_hook_t *submit_bio_done;
92 unsigned long bio_flags;
94 * bio_offset is optional, can be used if the pages in the bio
95 * can't tell us where in the file the bio should go
98 struct btrfs_work work;
101 /* These are used to set the lockdep class on the extent buffer locks.
102 * The class is set by the readpage_end_io_hook after the buffer has
103 * passed csum validation but before the pages are unlocked.
105 * The lockdep class is also set by btrfs_init_new_buffer on freshly
108 * The class is based on the level in the tree block, which allows lockdep
109 * to know that lower nodes nest inside the locks of higher nodes.
111 * We also add a check to make sure the highest level of the tree is
112 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
113 * code needs update as well.
115 #ifdef CONFIG_DEBUG_LOCK_ALLOC
116 # if BTRFS_MAX_LEVEL != 8
119 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
120 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
130 /* highest possible level */
136 * extents on the btree inode are pretty simple, there's one extent
137 * that covers the entire device
139 static struct extent_map *btree_get_extent(struct inode *inode,
140 struct page *page, size_t page_offset, u64 start, u64 len,
143 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
144 struct extent_map *em;
147 read_lock(&em_tree->lock);
148 em = lookup_extent_mapping(em_tree, start, len);
151 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
152 read_unlock(&em_tree->lock);
155 read_unlock(&em_tree->lock);
157 em = alloc_extent_map(GFP_NOFS);
159 em = ERR_PTR(-ENOMEM);
164 em->block_len = (u64)-1;
166 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
168 write_lock(&em_tree->lock);
169 ret = add_extent_mapping(em_tree, em);
170 if (ret == -EEXIST) {
171 u64 failed_start = em->start;
172 u64 failed_len = em->len;
175 em = lookup_extent_mapping(em_tree, start, len);
179 em = lookup_extent_mapping(em_tree, failed_start,
187 write_unlock(&em_tree->lock);
195 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
197 return crc32c(seed, data, len);
200 void btrfs_csum_final(u32 crc, char *result)
202 put_unaligned_le32(~crc, result);
206 * compute the csum for a btree block, and either verify it or write it
207 * into the csum field of the block.
209 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
213 btrfs_super_csum_size(&root->fs_info->super_copy);
216 unsigned long cur_len;
217 unsigned long offset = BTRFS_CSUM_SIZE;
218 char *map_token = NULL;
220 unsigned long map_start;
221 unsigned long map_len;
224 unsigned long inline_result;
226 len = buf->len - offset;
228 err = map_private_extent_buffer(buf, offset, 32,
230 &map_start, &map_len, KM_USER0);
233 cur_len = min(len, map_len - (offset - map_start));
234 crc = btrfs_csum_data(root, kaddr + offset - map_start,
238 unmap_extent_buffer(buf, map_token, KM_USER0);
240 if (csum_size > sizeof(inline_result)) {
241 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
245 result = (char *)&inline_result;
248 btrfs_csum_final(crc, result);
251 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
254 memcpy(&found, result, csum_size);
256 read_extent_buffer(buf, &val, 0, csum_size);
257 if (printk_ratelimit()) {
258 printk(KERN_INFO "btrfs: %s checksum verify "
259 "failed on %llu wanted %X found %X "
261 root->fs_info->sb->s_id,
262 (unsigned long long)buf->start, val, found,
263 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 if (printk_ratelimit()) {
300 printk("parent transid verify failed on %llu wanted %llu "
302 (unsigned long long)eb->start,
303 (unsigned long long)parent_transid,
304 (unsigned long long)btrfs_header_generation(eb));
307 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
309 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
310 &cached_state, GFP_NOFS);
315 * helper to read a given tree block, doing retries as required when
316 * the checksums don't match and we have alternate mirrors to try.
318 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
319 struct extent_buffer *eb,
320 u64 start, u64 parent_transid)
322 struct extent_io_tree *io_tree;
327 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
328 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
330 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
331 btree_get_extent, mirror_num);
333 !verify_parent_transid(io_tree, eb, parent_transid))
337 * This buffer's crc is fine, but its contents are corrupted, so
338 * there is no reason to read the other copies, they won't be
341 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
344 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
350 if (mirror_num > num_copies)
357 * checksum a dirty tree block before IO. This has extra checks to make sure
358 * we only fill in the checksum field in the first page of a multi-page block
361 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
363 struct extent_io_tree *tree;
364 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
367 struct extent_buffer *eb;
370 tree = &BTRFS_I(page->mapping->host)->io_tree;
372 if (page->private == EXTENT_PAGE_PRIVATE) {
376 if (!page->private) {
380 len = page->private >> 2;
383 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
388 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
389 btrfs_header_generation(eb));
391 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
393 found_start = btrfs_header_bytenr(eb);
394 if (found_start != start) {
398 if (eb->first_page != page) {
402 if (!PageUptodate(page)) {
406 csum_tree_block(root, eb, 0);
408 free_extent_buffer(eb);
413 static int check_tree_block_fsid(struct btrfs_root *root,
414 struct extent_buffer *eb)
416 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
417 u8 fsid[BTRFS_UUID_SIZE];
420 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
423 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
427 fs_devices = fs_devices->seed;
432 #define CORRUPT(reason, eb, root, slot) \
433 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
434 "root=%llu, slot=%d\n", reason, \
435 (unsigned long long)btrfs_header_bytenr(eb), \
436 (unsigned long long)root->objectid, slot)
438 static noinline int check_leaf(struct btrfs_root *root,
439 struct extent_buffer *leaf)
441 struct btrfs_key key;
442 struct btrfs_key leaf_key;
443 u32 nritems = btrfs_header_nritems(leaf);
449 /* Check the 0 item */
450 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
451 BTRFS_LEAF_DATA_SIZE(root)) {
452 CORRUPT("invalid item offset size pair", leaf, root, 0);
457 * Check to make sure each items keys are in the correct order and their
458 * offsets make sense. We only have to loop through nritems-1 because
459 * we check the current slot against the next slot, which verifies the
460 * next slot's offset+size makes sense and that the current's slot
463 for (slot = 0; slot < nritems - 1; slot++) {
464 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
465 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
467 /* Make sure the keys are in the right order */
468 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
469 CORRUPT("bad key order", leaf, root, slot);
474 * Make sure the offset and ends are right, remember that the
475 * item data starts at the end of the leaf and grows towards the
478 if (btrfs_item_offset_nr(leaf, slot) !=
479 btrfs_item_end_nr(leaf, slot + 1)) {
480 CORRUPT("slot offset bad", leaf, root, slot);
485 * Check to make sure that we don't point outside of the leaf,
486 * just incase all the items are consistent to eachother, but
487 * all point outside of the leaf.
489 if (btrfs_item_end_nr(leaf, slot) >
490 BTRFS_LEAF_DATA_SIZE(root)) {
491 CORRUPT("slot end outside of leaf", leaf, root, slot);
499 #ifdef CONFIG_DEBUG_LOCK_ALLOC
500 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
502 lockdep_set_class_and_name(&eb->lock,
503 &btrfs_eb_class[level],
504 btrfs_eb_name[level]);
508 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
509 struct extent_state *state)
511 struct extent_io_tree *tree;
515 struct extent_buffer *eb;
516 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
519 tree = &BTRFS_I(page->mapping->host)->io_tree;
520 if (page->private == EXTENT_PAGE_PRIVATE)
525 len = page->private >> 2;
528 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
534 found_start = btrfs_header_bytenr(eb);
535 if (found_start != start) {
536 if (printk_ratelimit()) {
537 printk(KERN_INFO "btrfs bad tree block start "
539 (unsigned long long)found_start,
540 (unsigned long long)eb->start);
545 if (eb->first_page != page) {
546 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
547 eb->first_page->index, page->index);
552 if (check_tree_block_fsid(root, eb)) {
553 if (printk_ratelimit()) {
554 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
555 (unsigned long long)eb->start);
560 found_level = btrfs_header_level(eb);
562 btrfs_set_buffer_lockdep_class(eb, found_level);
564 ret = csum_tree_block(root, eb, 1);
571 * If this is a leaf block and it is corrupt, set the corrupt bit so
572 * that we don't try and read the other copies of this block, just
575 if (found_level == 0 && check_leaf(root, eb)) {
576 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
580 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
581 end = eb->start + end - 1;
583 free_extent_buffer(eb);
588 static void end_workqueue_bio(struct bio *bio, int err)
590 struct end_io_wq *end_io_wq = bio->bi_private;
591 struct btrfs_fs_info *fs_info;
593 fs_info = end_io_wq->info;
594 end_io_wq->error = err;
595 end_io_wq->work.func = end_workqueue_fn;
596 end_io_wq->work.flags = 0;
598 if (bio->bi_rw & REQ_WRITE) {
599 if (end_io_wq->metadata == 1)
600 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
602 else if (end_io_wq->metadata == 2)
603 btrfs_queue_worker(&fs_info->endio_freespace_worker,
606 btrfs_queue_worker(&fs_info->endio_write_workers,
609 if (end_io_wq->metadata)
610 btrfs_queue_worker(&fs_info->endio_meta_workers,
613 btrfs_queue_worker(&fs_info->endio_workers,
619 * For the metadata arg you want
622 * 1 - if normal metadta
623 * 2 - if writing to the free space cache area
625 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
628 struct end_io_wq *end_io_wq;
629 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
633 end_io_wq->private = bio->bi_private;
634 end_io_wq->end_io = bio->bi_end_io;
635 end_io_wq->info = info;
636 end_io_wq->error = 0;
637 end_io_wq->bio = bio;
638 end_io_wq->metadata = metadata;
640 bio->bi_private = end_io_wq;
641 bio->bi_end_io = end_workqueue_bio;
645 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
647 unsigned long limit = min_t(unsigned long,
648 info->workers.max_workers,
649 info->fs_devices->open_devices);
653 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
655 return atomic_read(&info->nr_async_bios) >
656 btrfs_async_submit_limit(info);
659 static void run_one_async_start(struct btrfs_work *work)
661 struct async_submit_bio *async;
663 async = container_of(work, struct async_submit_bio, work);
664 async->submit_bio_start(async->inode, async->rw, async->bio,
665 async->mirror_num, async->bio_flags,
669 static void run_one_async_done(struct btrfs_work *work)
671 struct btrfs_fs_info *fs_info;
672 struct async_submit_bio *async;
675 async = container_of(work, struct async_submit_bio, work);
676 fs_info = BTRFS_I(async->inode)->root->fs_info;
678 limit = btrfs_async_submit_limit(fs_info);
679 limit = limit * 2 / 3;
681 atomic_dec(&fs_info->nr_async_submits);
683 if (atomic_read(&fs_info->nr_async_submits) < limit &&
684 waitqueue_active(&fs_info->async_submit_wait))
685 wake_up(&fs_info->async_submit_wait);
687 async->submit_bio_done(async->inode, async->rw, async->bio,
688 async->mirror_num, async->bio_flags,
692 static void run_one_async_free(struct btrfs_work *work)
694 struct async_submit_bio *async;
696 async = container_of(work, struct async_submit_bio, work);
700 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
701 int rw, struct bio *bio, int mirror_num,
702 unsigned long bio_flags,
704 extent_submit_bio_hook_t *submit_bio_start,
705 extent_submit_bio_hook_t *submit_bio_done)
707 struct async_submit_bio *async;
709 async = kmalloc(sizeof(*async), GFP_NOFS);
713 async->inode = inode;
716 async->mirror_num = mirror_num;
717 async->submit_bio_start = submit_bio_start;
718 async->submit_bio_done = submit_bio_done;
720 async->work.func = run_one_async_start;
721 async->work.ordered_func = run_one_async_done;
722 async->work.ordered_free = run_one_async_free;
724 async->work.flags = 0;
725 async->bio_flags = bio_flags;
726 async->bio_offset = bio_offset;
728 atomic_inc(&fs_info->nr_async_submits);
731 btrfs_set_work_high_prio(&async->work);
733 btrfs_queue_worker(&fs_info->workers, &async->work);
735 while (atomic_read(&fs_info->async_submit_draining) &&
736 atomic_read(&fs_info->nr_async_submits)) {
737 wait_event(fs_info->async_submit_wait,
738 (atomic_read(&fs_info->nr_async_submits) == 0));
744 static int btree_csum_one_bio(struct bio *bio)
746 struct bio_vec *bvec = bio->bi_io_vec;
748 struct btrfs_root *root;
750 WARN_ON(bio->bi_vcnt <= 0);
751 while (bio_index < bio->bi_vcnt) {
752 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
753 csum_dirty_buffer(root, bvec->bv_page);
760 static int __btree_submit_bio_start(struct inode *inode, int rw,
761 struct bio *bio, int mirror_num,
762 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 btree_csum_one_bio(bio);
773 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
774 int mirror_num, unsigned long bio_flags,
778 * when we're called for a write, we're already in the async
779 * submission context. Just jump into btrfs_map_bio
781 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
784 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
785 int mirror_num, unsigned long bio_flags,
790 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
794 if (!(rw & REQ_WRITE)) {
796 * called for a read, do the setup so that checksum validation
797 * can happen in the async kernel threads
799 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
804 * kthread helpers are used to submit writes so that checksumming
805 * can happen in parallel across all CPUs
807 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
808 inode, rw, bio, mirror_num, 0,
810 __btree_submit_bio_start,
811 __btree_submit_bio_done);
814 #ifdef CONFIG_MIGRATION
815 static int btree_migratepage(struct address_space *mapping,
816 struct page *newpage, struct page *page)
819 * we can't safely write a btree page from here,
820 * we haven't done the locking hook
825 * Buffers may be managed in a filesystem specific way.
826 * We must have no buffers or drop them.
828 if (page_has_private(page) &&
829 !try_to_release_page(page, GFP_KERNEL))
831 return migrate_page(mapping, newpage, page);
835 static int btree_writepage(struct page *page, struct writeback_control *wbc)
837 struct extent_io_tree *tree;
838 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
839 struct extent_buffer *eb;
842 tree = &BTRFS_I(page->mapping->host)->io_tree;
843 if (!(current->flags & PF_MEMALLOC)) {
844 return extent_write_full_page(tree, page,
845 btree_get_extent, wbc);
848 redirty_page_for_writepage(wbc, page);
849 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
852 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
854 spin_lock(&root->fs_info->delalloc_lock);
855 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
856 spin_unlock(&root->fs_info->delalloc_lock);
858 free_extent_buffer(eb);
864 static int btree_writepages(struct address_space *mapping,
865 struct writeback_control *wbc)
867 struct extent_io_tree *tree;
868 tree = &BTRFS_I(mapping->host)->io_tree;
869 if (wbc->sync_mode == WB_SYNC_NONE) {
870 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
872 unsigned long thresh = 32 * 1024 * 1024;
874 if (wbc->for_kupdate)
877 /* this is a bit racy, but that's ok */
878 num_dirty = root->fs_info->dirty_metadata_bytes;
879 if (num_dirty < thresh)
882 return extent_writepages(tree, mapping, btree_get_extent, wbc);
885 static int btree_readpage(struct file *file, struct page *page)
887 struct extent_io_tree *tree;
888 tree = &BTRFS_I(page->mapping->host)->io_tree;
889 return extent_read_full_page(tree, page, btree_get_extent);
892 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
894 struct extent_io_tree *tree;
895 struct extent_map_tree *map;
898 if (PageWriteback(page) || PageDirty(page))
901 tree = &BTRFS_I(page->mapping->host)->io_tree;
902 map = &BTRFS_I(page->mapping->host)->extent_tree;
904 ret = try_release_extent_state(map, tree, page, gfp_flags);
908 ret = try_release_extent_buffer(tree, page);
910 ClearPagePrivate(page);
911 set_page_private(page, 0);
912 page_cache_release(page);
918 static void btree_invalidatepage(struct page *page, unsigned long offset)
920 struct extent_io_tree *tree;
921 tree = &BTRFS_I(page->mapping->host)->io_tree;
922 extent_invalidatepage(tree, page, offset);
923 btree_releasepage(page, GFP_NOFS);
924 if (PagePrivate(page)) {
925 printk(KERN_WARNING "btrfs warning page private not zero "
926 "on page %llu\n", (unsigned long long)page_offset(page));
927 ClearPagePrivate(page);
928 set_page_private(page, 0);
929 page_cache_release(page);
933 static const struct address_space_operations btree_aops = {
934 .readpage = btree_readpage,
935 .writepage = btree_writepage,
936 .writepages = btree_writepages,
937 .releasepage = btree_releasepage,
938 .invalidatepage = btree_invalidatepage,
939 .sync_page = block_sync_page,
940 #ifdef CONFIG_MIGRATION
941 .migratepage = btree_migratepage,
945 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
948 struct extent_buffer *buf = NULL;
949 struct inode *btree_inode = root->fs_info->btree_inode;
952 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
955 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
956 buf, 0, 0, btree_get_extent, 0);
957 free_extent_buffer(buf);
961 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
962 u64 bytenr, u32 blocksize)
964 struct inode *btree_inode = root->fs_info->btree_inode;
965 struct extent_buffer *eb;
966 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
967 bytenr, blocksize, GFP_NOFS);
971 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
972 u64 bytenr, u32 blocksize)
974 struct inode *btree_inode = root->fs_info->btree_inode;
975 struct extent_buffer *eb;
977 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
978 bytenr, blocksize, NULL, GFP_NOFS);
983 int btrfs_write_tree_block(struct extent_buffer *buf)
985 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
986 buf->start + buf->len - 1);
989 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
991 return filemap_fdatawait_range(buf->first_page->mapping,
992 buf->start, buf->start + buf->len - 1);
995 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
996 u32 blocksize, u64 parent_transid)
998 struct extent_buffer *buf = NULL;
1001 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1005 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1008 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1013 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1014 struct extent_buffer *buf)
1016 struct inode *btree_inode = root->fs_info->btree_inode;
1017 if (btrfs_header_generation(buf) ==
1018 root->fs_info->running_transaction->transid) {
1019 btrfs_assert_tree_locked(buf);
1021 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1022 spin_lock(&root->fs_info->delalloc_lock);
1023 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1024 root->fs_info->dirty_metadata_bytes -= buf->len;
1027 spin_unlock(&root->fs_info->delalloc_lock);
1030 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1031 btrfs_set_lock_blocking(buf);
1032 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1038 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1039 u32 stripesize, struct btrfs_root *root,
1040 struct btrfs_fs_info *fs_info,
1044 root->commit_root = NULL;
1045 root->sectorsize = sectorsize;
1046 root->nodesize = nodesize;
1047 root->leafsize = leafsize;
1048 root->stripesize = stripesize;
1050 root->track_dirty = 0;
1052 root->orphan_item_inserted = 0;
1053 root->orphan_cleanup_state = 0;
1055 root->fs_info = fs_info;
1056 root->objectid = objectid;
1057 root->last_trans = 0;
1058 root->highest_objectid = 0;
1061 root->inode_tree = RB_ROOT;
1062 root->block_rsv = NULL;
1063 root->orphan_block_rsv = NULL;
1065 INIT_LIST_HEAD(&root->dirty_list);
1066 INIT_LIST_HEAD(&root->orphan_list);
1067 INIT_LIST_HEAD(&root->root_list);
1068 spin_lock_init(&root->node_lock);
1069 spin_lock_init(&root->orphan_lock);
1070 spin_lock_init(&root->inode_lock);
1071 spin_lock_init(&root->accounting_lock);
1072 mutex_init(&root->objectid_mutex);
1073 mutex_init(&root->log_mutex);
1074 init_waitqueue_head(&root->log_writer_wait);
1075 init_waitqueue_head(&root->log_commit_wait[0]);
1076 init_waitqueue_head(&root->log_commit_wait[1]);
1077 atomic_set(&root->log_commit[0], 0);
1078 atomic_set(&root->log_commit[1], 0);
1079 atomic_set(&root->log_writers, 0);
1080 root->log_batch = 0;
1081 root->log_transid = 0;
1082 root->last_log_commit = 0;
1083 extent_io_tree_init(&root->dirty_log_pages,
1084 fs_info->btree_inode->i_mapping, GFP_NOFS);
1086 memset(&root->root_key, 0, sizeof(root->root_key));
1087 memset(&root->root_item, 0, sizeof(root->root_item));
1088 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1089 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1090 root->defrag_trans_start = fs_info->generation;
1091 init_completion(&root->kobj_unregister);
1092 root->defrag_running = 0;
1093 root->root_key.objectid = objectid;
1094 root->anon_super.s_root = NULL;
1095 root->anon_super.s_dev = 0;
1096 INIT_LIST_HEAD(&root->anon_super.s_list);
1097 INIT_LIST_HEAD(&root->anon_super.s_instances);
1098 init_rwsem(&root->anon_super.s_umount);
1103 static int find_and_setup_root(struct btrfs_root *tree_root,
1104 struct btrfs_fs_info *fs_info,
1106 struct btrfs_root *root)
1112 __setup_root(tree_root->nodesize, tree_root->leafsize,
1113 tree_root->sectorsize, tree_root->stripesize,
1114 root, fs_info, objectid);
1115 ret = btrfs_find_last_root(tree_root, objectid,
1116 &root->root_item, &root->root_key);
1121 generation = btrfs_root_generation(&root->root_item);
1122 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1123 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1124 blocksize, generation);
1125 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1126 free_extent_buffer(root->node);
1129 root->commit_root = btrfs_root_node(root);
1133 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1134 struct btrfs_fs_info *fs_info)
1136 struct btrfs_root *root;
1137 struct btrfs_root *tree_root = fs_info->tree_root;
1138 struct extent_buffer *leaf;
1140 root = kzalloc(sizeof(*root), GFP_NOFS);
1142 return ERR_PTR(-ENOMEM);
1144 __setup_root(tree_root->nodesize, tree_root->leafsize,
1145 tree_root->sectorsize, tree_root->stripesize,
1146 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1148 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1149 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1150 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1152 * log trees do not get reference counted because they go away
1153 * before a real commit is actually done. They do store pointers
1154 * to file data extents, and those reference counts still get
1155 * updated (along with back refs to the log tree).
1159 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1160 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1163 return ERR_CAST(leaf);
1166 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1167 btrfs_set_header_bytenr(leaf, leaf->start);
1168 btrfs_set_header_generation(leaf, trans->transid);
1169 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1170 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1173 write_extent_buffer(root->node, root->fs_info->fsid,
1174 (unsigned long)btrfs_header_fsid(root->node),
1176 btrfs_mark_buffer_dirty(root->node);
1177 btrfs_tree_unlock(root->node);
1181 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1182 struct btrfs_fs_info *fs_info)
1184 struct btrfs_root *log_root;
1186 log_root = alloc_log_tree(trans, fs_info);
1187 if (IS_ERR(log_root))
1188 return PTR_ERR(log_root);
1189 WARN_ON(fs_info->log_root_tree);
1190 fs_info->log_root_tree = log_root;
1194 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1195 struct btrfs_root *root)
1197 struct btrfs_root *log_root;
1198 struct btrfs_inode_item *inode_item;
1200 log_root = alloc_log_tree(trans, root->fs_info);
1201 if (IS_ERR(log_root))
1202 return PTR_ERR(log_root);
1204 log_root->last_trans = trans->transid;
1205 log_root->root_key.offset = root->root_key.objectid;
1207 inode_item = &log_root->root_item.inode;
1208 inode_item->generation = cpu_to_le64(1);
1209 inode_item->size = cpu_to_le64(3);
1210 inode_item->nlink = cpu_to_le32(1);
1211 inode_item->nbytes = cpu_to_le64(root->leafsize);
1212 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1214 btrfs_set_root_node(&log_root->root_item, log_root->node);
1216 WARN_ON(root->log_root);
1217 root->log_root = log_root;
1218 root->log_transid = 0;
1219 root->last_log_commit = 0;
1223 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1224 struct btrfs_key *location)
1226 struct btrfs_root *root;
1227 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1228 struct btrfs_path *path;
1229 struct extent_buffer *l;
1234 root = kzalloc(sizeof(*root), GFP_NOFS);
1236 return ERR_PTR(-ENOMEM);
1237 if (location->offset == (u64)-1) {
1238 ret = find_and_setup_root(tree_root, fs_info,
1239 location->objectid, root);
1242 return ERR_PTR(ret);
1247 __setup_root(tree_root->nodesize, tree_root->leafsize,
1248 tree_root->sectorsize, tree_root->stripesize,
1249 root, fs_info, location->objectid);
1251 path = btrfs_alloc_path();
1254 return ERR_PTR(-ENOMEM);
1256 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1259 read_extent_buffer(l, &root->root_item,
1260 btrfs_item_ptr_offset(l, path->slots[0]),
1261 sizeof(root->root_item));
1262 memcpy(&root->root_key, location, sizeof(*location));
1264 btrfs_free_path(path);
1269 return ERR_PTR(ret);
1272 generation = btrfs_root_generation(&root->root_item);
1273 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1274 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1275 blocksize, generation);
1276 root->commit_root = btrfs_root_node(root);
1277 BUG_ON(!root->node);
1279 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1281 btrfs_check_and_init_root_item(&root->root_item);
1287 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1290 struct btrfs_root *root;
1292 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1293 return fs_info->tree_root;
1294 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1295 return fs_info->extent_root;
1297 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1298 (unsigned long)root_objectid);
1302 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1303 struct btrfs_key *location)
1305 struct btrfs_root *root;
1308 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1309 return fs_info->tree_root;
1310 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1311 return fs_info->extent_root;
1312 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1313 return fs_info->chunk_root;
1314 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1315 return fs_info->dev_root;
1316 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1317 return fs_info->csum_root;
1319 spin_lock(&fs_info->fs_roots_radix_lock);
1320 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1321 (unsigned long)location->objectid);
1322 spin_unlock(&fs_info->fs_roots_radix_lock);
1326 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1330 set_anon_super(&root->anon_super, NULL);
1332 if (btrfs_root_refs(&root->root_item) == 0) {
1337 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1341 root->orphan_item_inserted = 1;
1343 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1347 spin_lock(&fs_info->fs_roots_radix_lock);
1348 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1349 (unsigned long)root->root_key.objectid,
1354 spin_unlock(&fs_info->fs_roots_radix_lock);
1355 radix_tree_preload_end();
1357 if (ret == -EEXIST) {
1364 ret = btrfs_find_dead_roots(fs_info->tree_root,
1365 root->root_key.objectid);
1370 return ERR_PTR(ret);
1373 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1374 struct btrfs_key *location,
1375 const char *name, int namelen)
1377 return btrfs_read_fs_root_no_name(fs_info, location);
1379 struct btrfs_root *root;
1382 root = btrfs_read_fs_root_no_name(fs_info, location);
1389 ret = btrfs_set_root_name(root, name, namelen);
1391 free_extent_buffer(root->node);
1393 return ERR_PTR(ret);
1396 ret = btrfs_sysfs_add_root(root);
1398 free_extent_buffer(root->node);
1401 return ERR_PTR(ret);
1408 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1410 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1412 struct btrfs_device *device;
1413 struct backing_dev_info *bdi;
1415 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1418 bdi = blk_get_backing_dev_info(device->bdev);
1419 if (bdi && bdi_congested(bdi, bdi_bits)) {
1428 * this unplugs every device on the box, and it is only used when page
1431 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1433 struct btrfs_device *device;
1434 struct btrfs_fs_info *info;
1436 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1437 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1441 bdi = blk_get_backing_dev_info(device->bdev);
1442 if (bdi->unplug_io_fn)
1443 bdi->unplug_io_fn(bdi, page);
1447 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1449 struct inode *inode;
1450 struct extent_map_tree *em_tree;
1451 struct extent_map *em;
1452 struct address_space *mapping;
1455 /* the generic O_DIRECT read code does this */
1457 __unplug_io_fn(bdi, page);
1462 * page->mapping may change at any time. Get a consistent copy
1463 * and use that for everything below
1466 mapping = page->mapping;
1470 inode = mapping->host;
1473 * don't do the expensive searching for a small number of
1476 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1477 __unplug_io_fn(bdi, page);
1481 offset = page_offset(page);
1483 em_tree = &BTRFS_I(inode)->extent_tree;
1484 read_lock(&em_tree->lock);
1485 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1486 read_unlock(&em_tree->lock);
1488 __unplug_io_fn(bdi, page);
1492 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1493 free_extent_map(em);
1494 __unplug_io_fn(bdi, page);
1497 offset = offset - em->start;
1498 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1499 em->block_start + offset, page);
1500 free_extent_map(em);
1504 * If this fails, caller must call bdi_destroy() to get rid of the
1507 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1511 bdi->capabilities = BDI_CAP_MAP_COPY;
1512 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1516 bdi->ra_pages = default_backing_dev_info.ra_pages;
1517 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1518 bdi->unplug_io_data = info;
1519 bdi->congested_fn = btrfs_congested_fn;
1520 bdi->congested_data = info;
1524 static int bio_ready_for_csum(struct bio *bio)
1530 struct extent_io_tree *io_tree = NULL;
1531 struct bio_vec *bvec;
1535 bio_for_each_segment(bvec, bio, i) {
1536 page = bvec->bv_page;
1537 if (page->private == EXTENT_PAGE_PRIVATE) {
1538 length += bvec->bv_len;
1541 if (!page->private) {
1542 length += bvec->bv_len;
1545 length = bvec->bv_len;
1546 buf_len = page->private >> 2;
1547 start = page_offset(page) + bvec->bv_offset;
1548 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1550 /* are we fully contained in this bio? */
1551 if (buf_len <= length)
1554 ret = extent_range_uptodate(io_tree, start + length,
1555 start + buf_len - 1);
1560 * called by the kthread helper functions to finally call the bio end_io
1561 * functions. This is where read checksum verification actually happens
1563 static void end_workqueue_fn(struct btrfs_work *work)
1566 struct end_io_wq *end_io_wq;
1567 struct btrfs_fs_info *fs_info;
1570 end_io_wq = container_of(work, struct end_io_wq, work);
1571 bio = end_io_wq->bio;
1572 fs_info = end_io_wq->info;
1574 /* metadata bio reads are special because the whole tree block must
1575 * be checksummed at once. This makes sure the entire block is in
1576 * ram and up to date before trying to verify things. For
1577 * blocksize <= pagesize, it is basically a noop
1579 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1580 !bio_ready_for_csum(bio)) {
1581 btrfs_queue_worker(&fs_info->endio_meta_workers,
1585 error = end_io_wq->error;
1586 bio->bi_private = end_io_wq->private;
1587 bio->bi_end_io = end_io_wq->end_io;
1589 bio_endio(bio, error);
1592 static int cleaner_kthread(void *arg)
1594 struct btrfs_root *root = arg;
1597 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1599 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1600 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1601 btrfs_run_delayed_iputs(root);
1602 btrfs_clean_old_snapshots(root);
1603 mutex_unlock(&root->fs_info->cleaner_mutex);
1606 if (freezing(current)) {
1609 set_current_state(TASK_INTERRUPTIBLE);
1610 if (!kthread_should_stop())
1612 __set_current_state(TASK_RUNNING);
1614 } while (!kthread_should_stop());
1618 static int transaction_kthread(void *arg)
1620 struct btrfs_root *root = arg;
1621 struct btrfs_trans_handle *trans;
1622 struct btrfs_transaction *cur;
1625 unsigned long delay;
1630 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1631 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1633 spin_lock(&root->fs_info->new_trans_lock);
1634 cur = root->fs_info->running_transaction;
1636 spin_unlock(&root->fs_info->new_trans_lock);
1640 now = get_seconds();
1641 if (!cur->blocked &&
1642 (now < cur->start_time || now - cur->start_time < 30)) {
1643 spin_unlock(&root->fs_info->new_trans_lock);
1647 transid = cur->transid;
1648 spin_unlock(&root->fs_info->new_trans_lock);
1650 trans = btrfs_join_transaction(root, 1);
1651 BUG_ON(IS_ERR(trans));
1652 if (transid == trans->transid) {
1653 ret = btrfs_commit_transaction(trans, root);
1656 btrfs_end_transaction(trans, root);
1659 wake_up_process(root->fs_info->cleaner_kthread);
1660 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1662 if (freezing(current)) {
1665 set_current_state(TASK_INTERRUPTIBLE);
1666 if (!kthread_should_stop() &&
1667 !btrfs_transaction_blocked(root->fs_info))
1668 schedule_timeout(delay);
1669 __set_current_state(TASK_RUNNING);
1671 } while (!kthread_should_stop());
1675 struct btrfs_root *open_ctree(struct super_block *sb,
1676 struct btrfs_fs_devices *fs_devices,
1686 struct btrfs_key location;
1687 struct buffer_head *bh;
1688 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1690 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1692 struct btrfs_root *tree_root = btrfs_sb(sb);
1693 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1694 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1696 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1698 struct btrfs_root *log_tree_root;
1703 struct btrfs_super_block *disk_super;
1705 if (!extent_root || !tree_root || !fs_info ||
1706 !chunk_root || !dev_root || !csum_root) {
1711 ret = init_srcu_struct(&fs_info->subvol_srcu);
1717 ret = setup_bdi(fs_info, &fs_info->bdi);
1723 fs_info->btree_inode = new_inode(sb);
1724 if (!fs_info->btree_inode) {
1729 fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1731 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1732 INIT_LIST_HEAD(&fs_info->trans_list);
1733 INIT_LIST_HEAD(&fs_info->dead_roots);
1734 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1735 INIT_LIST_HEAD(&fs_info->hashers);
1736 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1737 INIT_LIST_HEAD(&fs_info->ordered_operations);
1738 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1739 spin_lock_init(&fs_info->delalloc_lock);
1740 spin_lock_init(&fs_info->new_trans_lock);
1741 spin_lock_init(&fs_info->ref_cache_lock);
1742 spin_lock_init(&fs_info->fs_roots_radix_lock);
1743 spin_lock_init(&fs_info->delayed_iput_lock);
1745 init_completion(&fs_info->kobj_unregister);
1746 fs_info->tree_root = tree_root;
1747 fs_info->extent_root = extent_root;
1748 fs_info->csum_root = csum_root;
1749 fs_info->chunk_root = chunk_root;
1750 fs_info->dev_root = dev_root;
1751 fs_info->fs_devices = fs_devices;
1752 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1753 INIT_LIST_HEAD(&fs_info->space_info);
1754 btrfs_mapping_init(&fs_info->mapping_tree);
1755 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1756 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1757 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1758 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1759 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1760 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1761 mutex_init(&fs_info->durable_block_rsv_mutex);
1762 atomic_set(&fs_info->nr_async_submits, 0);
1763 atomic_set(&fs_info->async_delalloc_pages, 0);
1764 atomic_set(&fs_info->async_submit_draining, 0);
1765 atomic_set(&fs_info->nr_async_bios, 0);
1767 fs_info->max_inline = 8192 * 1024;
1768 fs_info->metadata_ratio = 0;
1770 fs_info->thread_pool_size = min_t(unsigned long,
1771 num_online_cpus() + 2, 8);
1773 INIT_LIST_HEAD(&fs_info->ordered_extents);
1774 spin_lock_init(&fs_info->ordered_extent_lock);
1776 mutex_init(&fs_info->scrub_lock);
1777 atomic_set(&fs_info->scrubs_running, 0);
1778 atomic_set(&fs_info->scrub_pause_req, 0);
1779 atomic_set(&fs_info->scrubs_paused, 0);
1780 atomic_set(&fs_info->scrub_cancel_req, 0);
1781 init_waitqueue_head(&fs_info->scrub_pause_wait);
1782 init_rwsem(&fs_info->scrub_super_lock);
1783 fs_info->scrub_workers_refcnt = 0;
1784 btrfs_init_workers(&fs_info->scrub_workers, "scrub",
1785 fs_info->thread_pool_size, &fs_info->generic_worker);
1787 sb->s_blocksize = 4096;
1788 sb->s_blocksize_bits = blksize_bits(4096);
1789 sb->s_bdi = &fs_info->bdi;
1791 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1792 fs_info->btree_inode->i_nlink = 1;
1794 * we set the i_size on the btree inode to the max possible int.
1795 * the real end of the address space is determined by all of
1796 * the devices in the system
1798 fs_info->btree_inode->i_size = OFFSET_MAX;
1799 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1800 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1802 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1803 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1804 fs_info->btree_inode->i_mapping,
1806 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1809 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1811 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1812 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1813 sizeof(struct btrfs_key));
1814 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1815 insert_inode_hash(fs_info->btree_inode);
1817 spin_lock_init(&fs_info->block_group_cache_lock);
1818 fs_info->block_group_cache_tree = RB_ROOT;
1820 extent_io_tree_init(&fs_info->freed_extents[0],
1821 fs_info->btree_inode->i_mapping, GFP_NOFS);
1822 extent_io_tree_init(&fs_info->freed_extents[1],
1823 fs_info->btree_inode->i_mapping, GFP_NOFS);
1824 fs_info->pinned_extents = &fs_info->freed_extents[0];
1825 fs_info->do_barriers = 1;
1828 mutex_init(&fs_info->trans_mutex);
1829 mutex_init(&fs_info->ordered_operations_mutex);
1830 mutex_init(&fs_info->tree_log_mutex);
1831 mutex_init(&fs_info->chunk_mutex);
1832 mutex_init(&fs_info->transaction_kthread_mutex);
1833 mutex_init(&fs_info->cleaner_mutex);
1834 mutex_init(&fs_info->volume_mutex);
1835 init_rwsem(&fs_info->extent_commit_sem);
1836 init_rwsem(&fs_info->cleanup_work_sem);
1837 init_rwsem(&fs_info->subvol_sem);
1839 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1840 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1842 init_waitqueue_head(&fs_info->transaction_throttle);
1843 init_waitqueue_head(&fs_info->transaction_wait);
1844 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1845 init_waitqueue_head(&fs_info->async_submit_wait);
1847 __setup_root(4096, 4096, 4096, 4096, tree_root,
1848 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1850 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1856 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1857 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1858 sizeof(fs_info->super_for_commit));
1861 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1863 disk_super = &fs_info->super_copy;
1864 if (!btrfs_super_root(disk_super))
1867 /* check FS state, whether FS is broken. */
1868 fs_info->fs_state |= btrfs_super_flags(disk_super);
1870 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1873 * In the long term, we'll store the compression type in the super
1874 * block, and it'll be used for per file compression control.
1876 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1878 ret = btrfs_parse_options(tree_root, options);
1884 features = btrfs_super_incompat_flags(disk_super) &
1885 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1887 printk(KERN_ERR "BTRFS: couldn't mount because of "
1888 "unsupported optional features (%Lx).\n",
1889 (unsigned long long)features);
1894 features = btrfs_super_incompat_flags(disk_super);
1895 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1896 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1897 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1898 btrfs_set_super_incompat_flags(disk_super, features);
1900 features = btrfs_super_compat_ro_flags(disk_super) &
1901 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1902 if (!(sb->s_flags & MS_RDONLY) && features) {
1903 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1904 "unsupported option features (%Lx).\n",
1905 (unsigned long long)features);
1910 btrfs_init_workers(&fs_info->generic_worker,
1911 "genwork", 1, NULL);
1913 btrfs_init_workers(&fs_info->workers, "worker",
1914 fs_info->thread_pool_size,
1915 &fs_info->generic_worker);
1917 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1918 fs_info->thread_pool_size,
1919 &fs_info->generic_worker);
1921 btrfs_init_workers(&fs_info->submit_workers, "submit",
1922 min_t(u64, fs_devices->num_devices,
1923 fs_info->thread_pool_size),
1924 &fs_info->generic_worker);
1926 /* a higher idle thresh on the submit workers makes it much more
1927 * likely that bios will be send down in a sane order to the
1930 fs_info->submit_workers.idle_thresh = 64;
1932 fs_info->workers.idle_thresh = 16;
1933 fs_info->workers.ordered = 1;
1935 fs_info->delalloc_workers.idle_thresh = 2;
1936 fs_info->delalloc_workers.ordered = 1;
1938 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1939 &fs_info->generic_worker);
1940 btrfs_init_workers(&fs_info->endio_workers, "endio",
1941 fs_info->thread_pool_size,
1942 &fs_info->generic_worker);
1943 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1944 fs_info->thread_pool_size,
1945 &fs_info->generic_worker);
1946 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1947 "endio-meta-write", fs_info->thread_pool_size,
1948 &fs_info->generic_worker);
1949 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1950 fs_info->thread_pool_size,
1951 &fs_info->generic_worker);
1952 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1953 1, &fs_info->generic_worker);
1956 * endios are largely parallel and should have a very
1959 fs_info->endio_workers.idle_thresh = 4;
1960 fs_info->endio_meta_workers.idle_thresh = 4;
1962 fs_info->endio_write_workers.idle_thresh = 2;
1963 fs_info->endio_meta_write_workers.idle_thresh = 2;
1965 btrfs_start_workers(&fs_info->workers, 1);
1966 btrfs_start_workers(&fs_info->generic_worker, 1);
1967 btrfs_start_workers(&fs_info->submit_workers, 1);
1968 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1969 btrfs_start_workers(&fs_info->fixup_workers, 1);
1970 btrfs_start_workers(&fs_info->endio_workers, 1);
1971 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1972 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1973 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1974 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1976 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1977 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1978 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1980 nodesize = btrfs_super_nodesize(disk_super);
1981 leafsize = btrfs_super_leafsize(disk_super);
1982 sectorsize = btrfs_super_sectorsize(disk_super);
1983 stripesize = btrfs_super_stripesize(disk_super);
1984 tree_root->nodesize = nodesize;
1985 tree_root->leafsize = leafsize;
1986 tree_root->sectorsize = sectorsize;
1987 tree_root->stripesize = stripesize;
1989 sb->s_blocksize = sectorsize;
1990 sb->s_blocksize_bits = blksize_bits(sectorsize);
1992 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1993 sizeof(disk_super->magic))) {
1994 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1995 goto fail_sb_buffer;
1998 mutex_lock(&fs_info->chunk_mutex);
1999 ret = btrfs_read_sys_array(tree_root);
2000 mutex_unlock(&fs_info->chunk_mutex);
2002 printk(KERN_WARNING "btrfs: failed to read the system "
2003 "array on %s\n", sb->s_id);
2004 goto fail_sb_buffer;
2007 blocksize = btrfs_level_size(tree_root,
2008 btrfs_super_chunk_root_level(disk_super));
2009 generation = btrfs_super_chunk_root_generation(disk_super);
2011 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2012 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2014 chunk_root->node = read_tree_block(chunk_root,
2015 btrfs_super_chunk_root(disk_super),
2016 blocksize, generation);
2017 BUG_ON(!chunk_root->node);
2018 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2019 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2021 goto fail_chunk_root;
2023 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2024 chunk_root->commit_root = btrfs_root_node(chunk_root);
2026 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2027 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2030 mutex_lock(&fs_info->chunk_mutex);
2031 ret = btrfs_read_chunk_tree(chunk_root);
2032 mutex_unlock(&fs_info->chunk_mutex);
2034 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2036 goto fail_chunk_root;
2039 btrfs_close_extra_devices(fs_devices);
2041 blocksize = btrfs_level_size(tree_root,
2042 btrfs_super_root_level(disk_super));
2043 generation = btrfs_super_generation(disk_super);
2045 tree_root->node = read_tree_block(tree_root,
2046 btrfs_super_root(disk_super),
2047 blocksize, generation);
2048 if (!tree_root->node)
2049 goto fail_chunk_root;
2050 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2051 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2053 goto fail_tree_root;
2055 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2056 tree_root->commit_root = btrfs_root_node(tree_root);
2058 ret = find_and_setup_root(tree_root, fs_info,
2059 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2061 goto fail_tree_root;
2062 extent_root->track_dirty = 1;
2064 ret = find_and_setup_root(tree_root, fs_info,
2065 BTRFS_DEV_TREE_OBJECTID, dev_root);
2067 goto fail_extent_root;
2068 dev_root->track_dirty = 1;
2070 ret = find_and_setup_root(tree_root, fs_info,
2071 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2075 csum_root->track_dirty = 1;
2077 fs_info->generation = generation;
2078 fs_info->last_trans_committed = generation;
2079 fs_info->data_alloc_profile = (u64)-1;
2080 fs_info->metadata_alloc_profile = (u64)-1;
2081 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2083 ret = btrfs_init_space_info(fs_info);
2085 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2086 goto fail_block_groups;
2089 ret = btrfs_read_block_groups(extent_root);
2091 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2092 goto fail_block_groups;
2095 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2097 if (IS_ERR(fs_info->cleaner_kthread))
2098 goto fail_block_groups;
2100 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2102 "btrfs-transaction");
2103 if (IS_ERR(fs_info->transaction_kthread))
2106 if (!btrfs_test_opt(tree_root, SSD) &&
2107 !btrfs_test_opt(tree_root, NOSSD) &&
2108 !fs_info->fs_devices->rotating) {
2109 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2111 btrfs_set_opt(fs_info->mount_opt, SSD);
2114 /* do not make disk changes in broken FS */
2115 if (btrfs_super_log_root(disk_super) != 0 &&
2116 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2117 u64 bytenr = btrfs_super_log_root(disk_super);
2119 if (fs_devices->rw_devices == 0) {
2120 printk(KERN_WARNING "Btrfs log replay required "
2123 goto fail_trans_kthread;
2126 btrfs_level_size(tree_root,
2127 btrfs_super_log_root_level(disk_super));
2129 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2130 if (!log_tree_root) {
2132 goto fail_trans_kthread;
2135 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2136 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2138 log_tree_root->node = read_tree_block(tree_root, bytenr,
2141 ret = btrfs_recover_log_trees(log_tree_root);
2144 if (sb->s_flags & MS_RDONLY) {
2145 ret = btrfs_commit_super(tree_root);
2150 ret = btrfs_find_orphan_roots(tree_root);
2153 if (!(sb->s_flags & MS_RDONLY)) {
2154 ret = btrfs_cleanup_fs_roots(fs_info);
2157 ret = btrfs_recover_relocation(tree_root);
2160 "btrfs: failed to recover relocation\n");
2162 goto fail_trans_kthread;
2166 location.objectid = BTRFS_FS_TREE_OBJECTID;
2167 location.type = BTRFS_ROOT_ITEM_KEY;
2168 location.offset = (u64)-1;
2170 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2171 if (!fs_info->fs_root)
2172 goto fail_trans_kthread;
2173 if (IS_ERR(fs_info->fs_root)) {
2174 err = PTR_ERR(fs_info->fs_root);
2175 goto fail_trans_kthread;
2178 if (!(sb->s_flags & MS_RDONLY)) {
2179 down_read(&fs_info->cleanup_work_sem);
2180 err = btrfs_orphan_cleanup(fs_info->fs_root);
2182 err = btrfs_orphan_cleanup(fs_info->tree_root);
2183 up_read(&fs_info->cleanup_work_sem);
2185 close_ctree(tree_root);
2186 return ERR_PTR(err);
2193 kthread_stop(fs_info->transaction_kthread);
2195 kthread_stop(fs_info->cleaner_kthread);
2198 * make sure we're done with the btree inode before we stop our
2201 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2202 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2205 btrfs_free_block_groups(fs_info);
2206 free_extent_buffer(csum_root->node);
2207 free_extent_buffer(csum_root->commit_root);
2209 free_extent_buffer(dev_root->node);
2210 free_extent_buffer(dev_root->commit_root);
2212 free_extent_buffer(extent_root->node);
2213 free_extent_buffer(extent_root->commit_root);
2215 free_extent_buffer(tree_root->node);
2216 free_extent_buffer(tree_root->commit_root);
2218 free_extent_buffer(chunk_root->node);
2219 free_extent_buffer(chunk_root->commit_root);
2221 btrfs_stop_workers(&fs_info->generic_worker);
2222 btrfs_stop_workers(&fs_info->fixup_workers);
2223 btrfs_stop_workers(&fs_info->delalloc_workers);
2224 btrfs_stop_workers(&fs_info->workers);
2225 btrfs_stop_workers(&fs_info->endio_workers);
2226 btrfs_stop_workers(&fs_info->endio_meta_workers);
2227 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2228 btrfs_stop_workers(&fs_info->endio_write_workers);
2229 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2230 btrfs_stop_workers(&fs_info->submit_workers);
2232 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2233 iput(fs_info->btree_inode);
2235 btrfs_close_devices(fs_info->fs_devices);
2236 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2238 bdi_destroy(&fs_info->bdi);
2240 cleanup_srcu_struct(&fs_info->subvol_srcu);
2248 return ERR_PTR(err);
2251 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2253 char b[BDEVNAME_SIZE];
2256 set_buffer_uptodate(bh);
2258 if (printk_ratelimit()) {
2259 printk(KERN_WARNING "lost page write due to "
2260 "I/O error on %s\n",
2261 bdevname(bh->b_bdev, b));
2263 /* note, we dont' set_buffer_write_io_error because we have
2264 * our own ways of dealing with the IO errors
2266 clear_buffer_uptodate(bh);
2272 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2274 struct buffer_head *bh;
2275 struct buffer_head *latest = NULL;
2276 struct btrfs_super_block *super;
2281 /* we would like to check all the supers, but that would make
2282 * a btrfs mount succeed after a mkfs from a different FS.
2283 * So, we need to add a special mount option to scan for
2284 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2286 for (i = 0; i < 1; i++) {
2287 bytenr = btrfs_sb_offset(i);
2288 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2290 bh = __bread(bdev, bytenr / 4096, 4096);
2294 super = (struct btrfs_super_block *)bh->b_data;
2295 if (btrfs_super_bytenr(super) != bytenr ||
2296 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2297 sizeof(super->magic))) {
2302 if (!latest || btrfs_super_generation(super) > transid) {
2305 transid = btrfs_super_generation(super);
2314 * this should be called twice, once with wait == 0 and
2315 * once with wait == 1. When wait == 0 is done, all the buffer heads
2316 * we write are pinned.
2318 * They are released when wait == 1 is done.
2319 * max_mirrors must be the same for both runs, and it indicates how
2320 * many supers on this one device should be written.
2322 * max_mirrors == 0 means to write them all.
2324 static int write_dev_supers(struct btrfs_device *device,
2325 struct btrfs_super_block *sb,
2326 int do_barriers, int wait, int max_mirrors)
2328 struct buffer_head *bh;
2334 int last_barrier = 0;
2336 if (max_mirrors == 0)
2337 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2339 /* make sure only the last submit_bh does a barrier */
2341 for (i = 0; i < max_mirrors; i++) {
2342 bytenr = btrfs_sb_offset(i);
2343 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2344 device->total_bytes)
2350 for (i = 0; i < max_mirrors; i++) {
2351 bytenr = btrfs_sb_offset(i);
2352 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2356 bh = __find_get_block(device->bdev, bytenr / 4096,
2357 BTRFS_SUPER_INFO_SIZE);
2360 if (!buffer_uptodate(bh))
2363 /* drop our reference */
2366 /* drop the reference from the wait == 0 run */
2370 btrfs_set_super_bytenr(sb, bytenr);
2373 crc = btrfs_csum_data(NULL, (char *)sb +
2374 BTRFS_CSUM_SIZE, crc,
2375 BTRFS_SUPER_INFO_SIZE -
2377 btrfs_csum_final(crc, sb->csum);
2380 * one reference for us, and we leave it for the
2383 bh = __getblk(device->bdev, bytenr / 4096,
2384 BTRFS_SUPER_INFO_SIZE);
2385 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2387 /* one reference for submit_bh */
2390 set_buffer_uptodate(bh);
2392 bh->b_end_io = btrfs_end_buffer_write_sync;
2395 if (i == last_barrier && do_barriers)
2396 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2398 ret = submit_bh(WRITE_SYNC, bh);
2403 return errors < i ? 0 : -1;
2406 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2408 struct list_head *head;
2409 struct btrfs_device *dev;
2410 struct btrfs_super_block *sb;
2411 struct btrfs_dev_item *dev_item;
2415 int total_errors = 0;
2418 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2419 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2421 sb = &root->fs_info->super_for_commit;
2422 dev_item = &sb->dev_item;
2424 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2425 head = &root->fs_info->fs_devices->devices;
2426 list_for_each_entry(dev, head, dev_list) {
2431 if (!dev->in_fs_metadata || !dev->writeable)
2434 btrfs_set_stack_device_generation(dev_item, 0);
2435 btrfs_set_stack_device_type(dev_item, dev->type);
2436 btrfs_set_stack_device_id(dev_item, dev->devid);
2437 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2438 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2439 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2440 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2441 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2442 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2443 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2445 flags = btrfs_super_flags(sb);
2446 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2448 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2452 if (total_errors > max_errors) {
2453 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2459 list_for_each_entry(dev, head, dev_list) {
2462 if (!dev->in_fs_metadata || !dev->writeable)
2465 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2469 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2470 if (total_errors > max_errors) {
2471 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2478 int write_ctree_super(struct btrfs_trans_handle *trans,
2479 struct btrfs_root *root, int max_mirrors)
2483 ret = write_all_supers(root, max_mirrors);
2487 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2489 spin_lock(&fs_info->fs_roots_radix_lock);
2490 radix_tree_delete(&fs_info->fs_roots_radix,
2491 (unsigned long)root->root_key.objectid);
2492 spin_unlock(&fs_info->fs_roots_radix_lock);
2494 if (btrfs_root_refs(&root->root_item) == 0)
2495 synchronize_srcu(&fs_info->subvol_srcu);
2501 static void free_fs_root(struct btrfs_root *root)
2503 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2504 if (root->anon_super.s_dev) {
2505 down_write(&root->anon_super.s_umount);
2506 kill_anon_super(&root->anon_super);
2508 free_extent_buffer(root->node);
2509 free_extent_buffer(root->commit_root);
2514 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2517 struct btrfs_root *gang[8];
2520 while (!list_empty(&fs_info->dead_roots)) {
2521 gang[0] = list_entry(fs_info->dead_roots.next,
2522 struct btrfs_root, root_list);
2523 list_del(&gang[0]->root_list);
2525 if (gang[0]->in_radix) {
2526 btrfs_free_fs_root(fs_info, gang[0]);
2528 free_extent_buffer(gang[0]->node);
2529 free_extent_buffer(gang[0]->commit_root);
2535 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2540 for (i = 0; i < ret; i++)
2541 btrfs_free_fs_root(fs_info, gang[i]);
2546 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2548 u64 root_objectid = 0;
2549 struct btrfs_root *gang[8];
2554 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2555 (void **)gang, root_objectid,
2560 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2561 for (i = 0; i < ret; i++) {
2564 root_objectid = gang[i]->root_key.objectid;
2565 err = btrfs_orphan_cleanup(gang[i]);
2574 int btrfs_commit_super(struct btrfs_root *root)
2576 struct btrfs_trans_handle *trans;
2579 mutex_lock(&root->fs_info->cleaner_mutex);
2580 btrfs_run_delayed_iputs(root);
2581 btrfs_clean_old_snapshots(root);
2582 mutex_unlock(&root->fs_info->cleaner_mutex);
2584 /* wait until ongoing cleanup work done */
2585 down_write(&root->fs_info->cleanup_work_sem);
2586 up_write(&root->fs_info->cleanup_work_sem);
2588 trans = btrfs_join_transaction(root, 1);
2590 return PTR_ERR(trans);
2591 ret = btrfs_commit_transaction(trans, root);
2593 /* run commit again to drop the original snapshot */
2594 trans = btrfs_join_transaction(root, 1);
2596 return PTR_ERR(trans);
2597 btrfs_commit_transaction(trans, root);
2598 ret = btrfs_write_and_wait_transaction(NULL, root);
2601 ret = write_ctree_super(NULL, root, 0);
2605 int close_ctree(struct btrfs_root *root)
2607 struct btrfs_fs_info *fs_info = root->fs_info;
2610 fs_info->closing = 1;
2613 btrfs_scrub_cancel(root);
2614 btrfs_put_block_group_cache(fs_info);
2617 * Here come 2 situations when btrfs is broken to flip readonly:
2619 * 1. when btrfs flips readonly somewhere else before
2620 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2621 * and btrfs will skip to write sb directly to keep
2622 * ERROR state on disk.
2624 * 2. when btrfs flips readonly just in btrfs_commit_super,
2625 * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2626 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2627 * btrfs will cleanup all FS resources first and write sb then.
2629 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2630 ret = btrfs_commit_super(root);
2632 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2635 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2636 ret = btrfs_error_commit_super(root);
2638 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2641 kthread_stop(root->fs_info->transaction_kthread);
2642 kthread_stop(root->fs_info->cleaner_kthread);
2644 fs_info->closing = 2;
2647 if (fs_info->delalloc_bytes) {
2648 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2649 (unsigned long long)fs_info->delalloc_bytes);
2651 if (fs_info->total_ref_cache_size) {
2652 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2653 (unsigned long long)fs_info->total_ref_cache_size);
2656 free_extent_buffer(fs_info->extent_root->node);
2657 free_extent_buffer(fs_info->extent_root->commit_root);
2658 free_extent_buffer(fs_info->tree_root->node);
2659 free_extent_buffer(fs_info->tree_root->commit_root);
2660 free_extent_buffer(root->fs_info->chunk_root->node);
2661 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2662 free_extent_buffer(root->fs_info->dev_root->node);
2663 free_extent_buffer(root->fs_info->dev_root->commit_root);
2664 free_extent_buffer(root->fs_info->csum_root->node);
2665 free_extent_buffer(root->fs_info->csum_root->commit_root);
2667 btrfs_free_block_groups(root->fs_info);
2669 del_fs_roots(fs_info);
2671 iput(fs_info->btree_inode);
2673 btrfs_stop_workers(&fs_info->generic_worker);
2674 btrfs_stop_workers(&fs_info->fixup_workers);
2675 btrfs_stop_workers(&fs_info->delalloc_workers);
2676 btrfs_stop_workers(&fs_info->workers);
2677 btrfs_stop_workers(&fs_info->endio_workers);
2678 btrfs_stop_workers(&fs_info->endio_meta_workers);
2679 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2680 btrfs_stop_workers(&fs_info->endio_write_workers);
2681 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2682 btrfs_stop_workers(&fs_info->submit_workers);
2684 btrfs_close_devices(fs_info->fs_devices);
2685 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2687 bdi_destroy(&fs_info->bdi);
2688 cleanup_srcu_struct(&fs_info->subvol_srcu);
2690 kfree(fs_info->extent_root);
2691 kfree(fs_info->tree_root);
2692 kfree(fs_info->chunk_root);
2693 kfree(fs_info->dev_root);
2694 kfree(fs_info->csum_root);
2700 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2703 struct inode *btree_inode = buf->first_page->mapping->host;
2705 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2710 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2715 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2717 struct inode *btree_inode = buf->first_page->mapping->host;
2718 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2722 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2724 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2725 u64 transid = btrfs_header_generation(buf);
2726 struct inode *btree_inode = root->fs_info->btree_inode;
2729 btrfs_assert_tree_locked(buf);
2730 if (transid != root->fs_info->generation) {
2731 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2732 "found %llu running %llu\n",
2733 (unsigned long long)buf->start,
2734 (unsigned long long)transid,
2735 (unsigned long long)root->fs_info->generation);
2738 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2741 spin_lock(&root->fs_info->delalloc_lock);
2742 root->fs_info->dirty_metadata_bytes += buf->len;
2743 spin_unlock(&root->fs_info->delalloc_lock);
2747 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2750 * looks as though older kernels can get into trouble with
2751 * this code, they end up stuck in balance_dirty_pages forever
2754 unsigned long thresh = 32 * 1024 * 1024;
2756 if (current->flags & PF_MEMALLOC)
2759 num_dirty = root->fs_info->dirty_metadata_bytes;
2761 if (num_dirty > thresh) {
2762 balance_dirty_pages_ratelimited_nr(
2763 root->fs_info->btree_inode->i_mapping, 1);
2768 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2770 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2772 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2774 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2778 int btree_lock_page_hook(struct page *page)
2780 struct inode *inode = page->mapping->host;
2781 struct btrfs_root *root = BTRFS_I(inode)->root;
2782 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2783 struct extent_buffer *eb;
2785 u64 bytenr = page_offset(page);
2787 if (page->private == EXTENT_PAGE_PRIVATE)
2790 len = page->private >> 2;
2791 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2795 btrfs_tree_lock(eb);
2796 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2798 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2799 spin_lock(&root->fs_info->delalloc_lock);
2800 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2801 root->fs_info->dirty_metadata_bytes -= eb->len;
2804 spin_unlock(&root->fs_info->delalloc_lock);
2807 btrfs_tree_unlock(eb);
2808 free_extent_buffer(eb);
2814 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2820 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2821 printk(KERN_WARNING "warning: mount fs with errors, "
2822 "running btrfsck is recommended\n");
2825 int btrfs_error_commit_super(struct btrfs_root *root)
2829 mutex_lock(&root->fs_info->cleaner_mutex);
2830 btrfs_run_delayed_iputs(root);
2831 mutex_unlock(&root->fs_info->cleaner_mutex);
2833 down_write(&root->fs_info->cleanup_work_sem);
2834 up_write(&root->fs_info->cleanup_work_sem);
2836 /* cleanup FS via transaction */
2837 btrfs_cleanup_transaction(root);
2839 ret = write_ctree_super(NULL, root, 0);
2844 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2846 struct btrfs_inode *btrfs_inode;
2847 struct list_head splice;
2849 INIT_LIST_HEAD(&splice);
2851 mutex_lock(&root->fs_info->ordered_operations_mutex);
2852 spin_lock(&root->fs_info->ordered_extent_lock);
2854 list_splice_init(&root->fs_info->ordered_operations, &splice);
2855 while (!list_empty(&splice)) {
2856 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2857 ordered_operations);
2859 list_del_init(&btrfs_inode->ordered_operations);
2861 btrfs_invalidate_inodes(btrfs_inode->root);
2864 spin_unlock(&root->fs_info->ordered_extent_lock);
2865 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2870 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2872 struct list_head splice;
2873 struct btrfs_ordered_extent *ordered;
2874 struct inode *inode;
2876 INIT_LIST_HEAD(&splice);
2878 spin_lock(&root->fs_info->ordered_extent_lock);
2880 list_splice_init(&root->fs_info->ordered_extents, &splice);
2881 while (!list_empty(&splice)) {
2882 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2885 list_del_init(&ordered->root_extent_list);
2886 atomic_inc(&ordered->refs);
2888 /* the inode may be getting freed (in sys_unlink path). */
2889 inode = igrab(ordered->inode);
2891 spin_unlock(&root->fs_info->ordered_extent_lock);
2895 atomic_set(&ordered->refs, 1);
2896 btrfs_put_ordered_extent(ordered);
2898 spin_lock(&root->fs_info->ordered_extent_lock);
2901 spin_unlock(&root->fs_info->ordered_extent_lock);
2906 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2907 struct btrfs_root *root)
2909 struct rb_node *node;
2910 struct btrfs_delayed_ref_root *delayed_refs;
2911 struct btrfs_delayed_ref_node *ref;
2914 delayed_refs = &trans->delayed_refs;
2916 spin_lock(&delayed_refs->lock);
2917 if (delayed_refs->num_entries == 0) {
2918 spin_unlock(&delayed_refs->lock);
2919 printk(KERN_INFO "delayed_refs has NO entry\n");
2923 node = rb_first(&delayed_refs->root);
2925 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2926 node = rb_next(node);
2929 rb_erase(&ref->rb_node, &delayed_refs->root);
2930 delayed_refs->num_entries--;
2932 atomic_set(&ref->refs, 1);
2933 if (btrfs_delayed_ref_is_head(ref)) {
2934 struct btrfs_delayed_ref_head *head;
2936 head = btrfs_delayed_node_to_head(ref);
2937 mutex_lock(&head->mutex);
2938 kfree(head->extent_op);
2939 delayed_refs->num_heads--;
2940 if (list_empty(&head->cluster))
2941 delayed_refs->num_heads_ready--;
2942 list_del_init(&head->cluster);
2943 mutex_unlock(&head->mutex);
2946 spin_unlock(&delayed_refs->lock);
2947 btrfs_put_delayed_ref(ref);
2950 spin_lock(&delayed_refs->lock);
2953 spin_unlock(&delayed_refs->lock);
2958 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2960 struct btrfs_pending_snapshot *snapshot;
2961 struct list_head splice;
2963 INIT_LIST_HEAD(&splice);
2965 list_splice_init(&t->pending_snapshots, &splice);
2967 while (!list_empty(&splice)) {
2968 snapshot = list_entry(splice.next,
2969 struct btrfs_pending_snapshot,
2972 list_del_init(&snapshot->list);
2980 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2982 struct btrfs_inode *btrfs_inode;
2983 struct list_head splice;
2985 INIT_LIST_HEAD(&splice);
2987 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2989 spin_lock(&root->fs_info->delalloc_lock);
2991 while (!list_empty(&splice)) {
2992 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2995 list_del_init(&btrfs_inode->delalloc_inodes);
2997 btrfs_invalidate_inodes(btrfs_inode->root);
3000 spin_unlock(&root->fs_info->delalloc_lock);
3005 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3006 struct extent_io_tree *dirty_pages,
3011 struct inode *btree_inode = root->fs_info->btree_inode;
3012 struct extent_buffer *eb;
3016 unsigned long index;
3019 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3024 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3025 while (start <= end) {
3026 index = start >> PAGE_CACHE_SHIFT;
3027 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3028 page = find_get_page(btree_inode->i_mapping, index);
3031 offset = page_offset(page);
3033 spin_lock(&dirty_pages->buffer_lock);
3034 eb = radix_tree_lookup(
3035 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3036 offset >> PAGE_CACHE_SHIFT);
3037 spin_unlock(&dirty_pages->buffer_lock);
3039 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3041 atomic_set(&eb->refs, 1);
3043 if (PageWriteback(page))
3044 end_page_writeback(page);
3047 if (PageDirty(page)) {
3048 clear_page_dirty_for_io(page);
3049 spin_lock_irq(&page->mapping->tree_lock);
3050 radix_tree_tag_clear(&page->mapping->page_tree,
3052 PAGECACHE_TAG_DIRTY);
3053 spin_unlock_irq(&page->mapping->tree_lock);
3056 page->mapping->a_ops->invalidatepage(page, 0);
3064 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3065 struct extent_io_tree *pinned_extents)
3067 struct extent_io_tree *unpin;
3072 unpin = pinned_extents;
3074 ret = find_first_extent_bit(unpin, 0, &start, &end,
3080 if (btrfs_test_opt(root, DISCARD))
3081 ret = btrfs_error_discard_extent(root, start,
3085 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3086 btrfs_error_unpin_extent_range(root, start, end);
3093 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3095 struct btrfs_transaction *t;
3100 mutex_lock(&root->fs_info->trans_mutex);
3101 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3103 list_splice_init(&root->fs_info->trans_list, &list);
3104 while (!list_empty(&list)) {
3105 t = list_entry(list.next, struct btrfs_transaction, list);
3109 btrfs_destroy_ordered_operations(root);
3111 btrfs_destroy_ordered_extents(root);
3113 btrfs_destroy_delayed_refs(t, root);
3115 btrfs_block_rsv_release(root,
3116 &root->fs_info->trans_block_rsv,
3117 t->dirty_pages.dirty_bytes);
3119 /* FIXME: cleanup wait for commit */
3122 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3123 wake_up(&root->fs_info->transaction_blocked_wait);
3126 if (waitqueue_active(&root->fs_info->transaction_wait))
3127 wake_up(&root->fs_info->transaction_wait);
3128 mutex_unlock(&root->fs_info->trans_mutex);
3130 mutex_lock(&root->fs_info->trans_mutex);
3132 if (waitqueue_active(&t->commit_wait))
3133 wake_up(&t->commit_wait);
3134 mutex_unlock(&root->fs_info->trans_mutex);
3136 mutex_lock(&root->fs_info->trans_mutex);
3138 btrfs_destroy_pending_snapshots(t);
3140 btrfs_destroy_delalloc_inodes(root);
3142 spin_lock(&root->fs_info->new_trans_lock);
3143 root->fs_info->running_transaction = NULL;
3144 spin_unlock(&root->fs_info->new_trans_lock);
3146 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3149 btrfs_destroy_pinned_extent(root,
3150 root->fs_info->pinned_extents);
3152 atomic_set(&t->use_count, 0);
3153 list_del_init(&t->list);
3154 memset(t, 0, sizeof(*t));
3155 kmem_cache_free(btrfs_transaction_cachep, t);
3158 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3159 mutex_unlock(&root->fs_info->trans_mutex);
3164 static struct extent_io_ops btree_extent_io_ops = {
3165 .write_cache_pages_lock_hook = btree_lock_page_hook,
3166 .readpage_end_io_hook = btree_readpage_end_io_hook,
3167 .submit_bio_hook = btree_submit_bio_hook,
3168 /* note we're sharing with inode.c for the merge bio hook */
3169 .merge_bio_hook = btrfs_merge_bio_hook,