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)
1285 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1288 struct btrfs_root *root;
1290 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1291 return fs_info->tree_root;
1292 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1293 return fs_info->extent_root;
1295 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1296 (unsigned long)root_objectid);
1300 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1301 struct btrfs_key *location)
1303 struct btrfs_root *root;
1306 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1307 return fs_info->tree_root;
1308 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1309 return fs_info->extent_root;
1310 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1311 return fs_info->chunk_root;
1312 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1313 return fs_info->dev_root;
1314 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1315 return fs_info->csum_root;
1317 spin_lock(&fs_info->fs_roots_radix_lock);
1318 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1319 (unsigned long)location->objectid);
1320 spin_unlock(&fs_info->fs_roots_radix_lock);
1324 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1328 set_anon_super(&root->anon_super, NULL);
1330 if (btrfs_root_refs(&root->root_item) == 0) {
1335 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1339 root->orphan_item_inserted = 1;
1341 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1345 spin_lock(&fs_info->fs_roots_radix_lock);
1346 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1347 (unsigned long)root->root_key.objectid,
1352 spin_unlock(&fs_info->fs_roots_radix_lock);
1353 radix_tree_preload_end();
1355 if (ret == -EEXIST) {
1362 ret = btrfs_find_dead_roots(fs_info->tree_root,
1363 root->root_key.objectid);
1368 return ERR_PTR(ret);
1371 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1372 struct btrfs_key *location,
1373 const char *name, int namelen)
1375 return btrfs_read_fs_root_no_name(fs_info, location);
1377 struct btrfs_root *root;
1380 root = btrfs_read_fs_root_no_name(fs_info, location);
1387 ret = btrfs_set_root_name(root, name, namelen);
1389 free_extent_buffer(root->node);
1391 return ERR_PTR(ret);
1394 ret = btrfs_sysfs_add_root(root);
1396 free_extent_buffer(root->node);
1399 return ERR_PTR(ret);
1406 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1408 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1410 struct btrfs_device *device;
1411 struct backing_dev_info *bdi;
1413 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1416 bdi = blk_get_backing_dev_info(device->bdev);
1417 if (bdi && bdi_congested(bdi, bdi_bits)) {
1426 * this unplugs every device on the box, and it is only used when page
1429 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1431 struct btrfs_device *device;
1432 struct btrfs_fs_info *info;
1434 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1435 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1439 bdi = blk_get_backing_dev_info(device->bdev);
1440 if (bdi->unplug_io_fn)
1441 bdi->unplug_io_fn(bdi, page);
1445 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1447 struct inode *inode;
1448 struct extent_map_tree *em_tree;
1449 struct extent_map *em;
1450 struct address_space *mapping;
1453 /* the generic O_DIRECT read code does this */
1455 __unplug_io_fn(bdi, page);
1460 * page->mapping may change at any time. Get a consistent copy
1461 * and use that for everything below
1464 mapping = page->mapping;
1468 inode = mapping->host;
1471 * don't do the expensive searching for a small number of
1474 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1475 __unplug_io_fn(bdi, page);
1479 offset = page_offset(page);
1481 em_tree = &BTRFS_I(inode)->extent_tree;
1482 read_lock(&em_tree->lock);
1483 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1484 read_unlock(&em_tree->lock);
1486 __unplug_io_fn(bdi, page);
1490 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1491 free_extent_map(em);
1492 __unplug_io_fn(bdi, page);
1495 offset = offset - em->start;
1496 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1497 em->block_start + offset, page);
1498 free_extent_map(em);
1502 * If this fails, caller must call bdi_destroy() to get rid of the
1505 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1509 bdi->capabilities = BDI_CAP_MAP_COPY;
1510 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1514 bdi->ra_pages = default_backing_dev_info.ra_pages;
1515 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1516 bdi->unplug_io_data = info;
1517 bdi->congested_fn = btrfs_congested_fn;
1518 bdi->congested_data = info;
1522 static int bio_ready_for_csum(struct bio *bio)
1528 struct extent_io_tree *io_tree = NULL;
1529 struct bio_vec *bvec;
1533 bio_for_each_segment(bvec, bio, i) {
1534 page = bvec->bv_page;
1535 if (page->private == EXTENT_PAGE_PRIVATE) {
1536 length += bvec->bv_len;
1539 if (!page->private) {
1540 length += bvec->bv_len;
1543 length = bvec->bv_len;
1544 buf_len = page->private >> 2;
1545 start = page_offset(page) + bvec->bv_offset;
1546 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1548 /* are we fully contained in this bio? */
1549 if (buf_len <= length)
1552 ret = extent_range_uptodate(io_tree, start + length,
1553 start + buf_len - 1);
1558 * called by the kthread helper functions to finally call the bio end_io
1559 * functions. This is where read checksum verification actually happens
1561 static void end_workqueue_fn(struct btrfs_work *work)
1564 struct end_io_wq *end_io_wq;
1565 struct btrfs_fs_info *fs_info;
1568 end_io_wq = container_of(work, struct end_io_wq, work);
1569 bio = end_io_wq->bio;
1570 fs_info = end_io_wq->info;
1572 /* metadata bio reads are special because the whole tree block must
1573 * be checksummed at once. This makes sure the entire block is in
1574 * ram and up to date before trying to verify things. For
1575 * blocksize <= pagesize, it is basically a noop
1577 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1578 !bio_ready_for_csum(bio)) {
1579 btrfs_queue_worker(&fs_info->endio_meta_workers,
1583 error = end_io_wq->error;
1584 bio->bi_private = end_io_wq->private;
1585 bio->bi_end_io = end_io_wq->end_io;
1587 bio_endio(bio, error);
1590 static int cleaner_kthread(void *arg)
1592 struct btrfs_root *root = arg;
1595 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1597 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1598 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1599 btrfs_run_delayed_iputs(root);
1600 btrfs_clean_old_snapshots(root);
1601 mutex_unlock(&root->fs_info->cleaner_mutex);
1604 if (freezing(current)) {
1607 set_current_state(TASK_INTERRUPTIBLE);
1608 if (!kthread_should_stop())
1610 __set_current_state(TASK_RUNNING);
1612 } while (!kthread_should_stop());
1616 static int transaction_kthread(void *arg)
1618 struct btrfs_root *root = arg;
1619 struct btrfs_trans_handle *trans;
1620 struct btrfs_transaction *cur;
1623 unsigned long delay;
1628 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1629 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1631 spin_lock(&root->fs_info->new_trans_lock);
1632 cur = root->fs_info->running_transaction;
1634 spin_unlock(&root->fs_info->new_trans_lock);
1638 now = get_seconds();
1639 if (!cur->blocked &&
1640 (now < cur->start_time || now - cur->start_time < 30)) {
1641 spin_unlock(&root->fs_info->new_trans_lock);
1645 transid = cur->transid;
1646 spin_unlock(&root->fs_info->new_trans_lock);
1648 trans = btrfs_join_transaction(root, 1);
1649 BUG_ON(IS_ERR(trans));
1650 if (transid == trans->transid) {
1651 ret = btrfs_commit_transaction(trans, root);
1654 btrfs_end_transaction(trans, root);
1657 wake_up_process(root->fs_info->cleaner_kthread);
1658 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1660 if (freezing(current)) {
1663 set_current_state(TASK_INTERRUPTIBLE);
1664 if (!kthread_should_stop() &&
1665 !btrfs_transaction_blocked(root->fs_info))
1666 schedule_timeout(delay);
1667 __set_current_state(TASK_RUNNING);
1669 } while (!kthread_should_stop());
1673 struct btrfs_root *open_ctree(struct super_block *sb,
1674 struct btrfs_fs_devices *fs_devices,
1684 struct btrfs_key location;
1685 struct buffer_head *bh;
1686 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1688 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1690 struct btrfs_root *tree_root = btrfs_sb(sb);
1691 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1692 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1694 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1696 struct btrfs_root *log_tree_root;
1701 struct btrfs_super_block *disk_super;
1703 if (!extent_root || !tree_root || !fs_info ||
1704 !chunk_root || !dev_root || !csum_root) {
1709 ret = init_srcu_struct(&fs_info->subvol_srcu);
1715 ret = setup_bdi(fs_info, &fs_info->bdi);
1721 fs_info->btree_inode = new_inode(sb);
1722 if (!fs_info->btree_inode) {
1727 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1728 INIT_LIST_HEAD(&fs_info->trans_list);
1729 INIT_LIST_HEAD(&fs_info->dead_roots);
1730 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1731 INIT_LIST_HEAD(&fs_info->hashers);
1732 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1733 INIT_LIST_HEAD(&fs_info->ordered_operations);
1734 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1735 spin_lock_init(&fs_info->delalloc_lock);
1736 spin_lock_init(&fs_info->new_trans_lock);
1737 spin_lock_init(&fs_info->ref_cache_lock);
1738 spin_lock_init(&fs_info->fs_roots_radix_lock);
1739 spin_lock_init(&fs_info->delayed_iput_lock);
1741 init_completion(&fs_info->kobj_unregister);
1742 fs_info->tree_root = tree_root;
1743 fs_info->extent_root = extent_root;
1744 fs_info->csum_root = csum_root;
1745 fs_info->chunk_root = chunk_root;
1746 fs_info->dev_root = dev_root;
1747 fs_info->fs_devices = fs_devices;
1748 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1749 INIT_LIST_HEAD(&fs_info->space_info);
1750 btrfs_mapping_init(&fs_info->mapping_tree);
1751 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1752 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1753 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1754 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1755 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1756 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1757 mutex_init(&fs_info->durable_block_rsv_mutex);
1758 atomic_set(&fs_info->nr_async_submits, 0);
1759 atomic_set(&fs_info->async_delalloc_pages, 0);
1760 atomic_set(&fs_info->async_submit_draining, 0);
1761 atomic_set(&fs_info->nr_async_bios, 0);
1763 fs_info->max_inline = 8192 * 1024;
1764 fs_info->metadata_ratio = 0;
1766 fs_info->thread_pool_size = min_t(unsigned long,
1767 num_online_cpus() + 2, 8);
1769 INIT_LIST_HEAD(&fs_info->ordered_extents);
1770 spin_lock_init(&fs_info->ordered_extent_lock);
1772 sb->s_blocksize = 4096;
1773 sb->s_blocksize_bits = blksize_bits(4096);
1774 sb->s_bdi = &fs_info->bdi;
1776 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1777 fs_info->btree_inode->i_nlink = 1;
1779 * we set the i_size on the btree inode to the max possible int.
1780 * the real end of the address space is determined by all of
1781 * the devices in the system
1783 fs_info->btree_inode->i_size = OFFSET_MAX;
1784 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1785 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1787 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1788 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1789 fs_info->btree_inode->i_mapping,
1791 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1794 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1796 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1797 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1798 sizeof(struct btrfs_key));
1799 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1800 insert_inode_hash(fs_info->btree_inode);
1802 spin_lock_init(&fs_info->block_group_cache_lock);
1803 fs_info->block_group_cache_tree = RB_ROOT;
1805 extent_io_tree_init(&fs_info->freed_extents[0],
1806 fs_info->btree_inode->i_mapping, GFP_NOFS);
1807 extent_io_tree_init(&fs_info->freed_extents[1],
1808 fs_info->btree_inode->i_mapping, GFP_NOFS);
1809 fs_info->pinned_extents = &fs_info->freed_extents[0];
1810 fs_info->do_barriers = 1;
1813 mutex_init(&fs_info->trans_mutex);
1814 mutex_init(&fs_info->ordered_operations_mutex);
1815 mutex_init(&fs_info->tree_log_mutex);
1816 mutex_init(&fs_info->chunk_mutex);
1817 mutex_init(&fs_info->transaction_kthread_mutex);
1818 mutex_init(&fs_info->cleaner_mutex);
1819 mutex_init(&fs_info->volume_mutex);
1820 init_rwsem(&fs_info->extent_commit_sem);
1821 init_rwsem(&fs_info->cleanup_work_sem);
1822 init_rwsem(&fs_info->subvol_sem);
1824 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1825 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1827 init_waitqueue_head(&fs_info->transaction_throttle);
1828 init_waitqueue_head(&fs_info->transaction_wait);
1829 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1830 init_waitqueue_head(&fs_info->async_submit_wait);
1832 __setup_root(4096, 4096, 4096, 4096, tree_root,
1833 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1835 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1841 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1842 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1843 sizeof(fs_info->super_for_commit));
1846 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1848 disk_super = &fs_info->super_copy;
1849 if (!btrfs_super_root(disk_super))
1852 /* check FS state, whether FS is broken. */
1853 fs_info->fs_state |= btrfs_super_flags(disk_super);
1855 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1858 * In the long term, we'll store the compression type in the super
1859 * block, and it'll be used for per file compression control.
1861 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1863 ret = btrfs_parse_options(tree_root, options);
1869 features = btrfs_super_incompat_flags(disk_super) &
1870 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1872 printk(KERN_ERR "BTRFS: couldn't mount because of "
1873 "unsupported optional features (%Lx).\n",
1874 (unsigned long long)features);
1879 features = btrfs_super_incompat_flags(disk_super);
1880 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1881 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1882 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1883 btrfs_set_super_incompat_flags(disk_super, features);
1885 features = btrfs_super_compat_ro_flags(disk_super) &
1886 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1887 if (!(sb->s_flags & MS_RDONLY) && features) {
1888 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1889 "unsupported option features (%Lx).\n",
1890 (unsigned long long)features);
1895 btrfs_init_workers(&fs_info->generic_worker,
1896 "genwork", 1, NULL);
1898 btrfs_init_workers(&fs_info->workers, "worker",
1899 fs_info->thread_pool_size,
1900 &fs_info->generic_worker);
1902 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1903 fs_info->thread_pool_size,
1904 &fs_info->generic_worker);
1906 btrfs_init_workers(&fs_info->submit_workers, "submit",
1907 min_t(u64, fs_devices->num_devices,
1908 fs_info->thread_pool_size),
1909 &fs_info->generic_worker);
1911 /* a higher idle thresh on the submit workers makes it much more
1912 * likely that bios will be send down in a sane order to the
1915 fs_info->submit_workers.idle_thresh = 64;
1917 fs_info->workers.idle_thresh = 16;
1918 fs_info->workers.ordered = 1;
1920 fs_info->delalloc_workers.idle_thresh = 2;
1921 fs_info->delalloc_workers.ordered = 1;
1923 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1924 &fs_info->generic_worker);
1925 btrfs_init_workers(&fs_info->endio_workers, "endio",
1926 fs_info->thread_pool_size,
1927 &fs_info->generic_worker);
1928 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1929 fs_info->thread_pool_size,
1930 &fs_info->generic_worker);
1931 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1932 "endio-meta-write", fs_info->thread_pool_size,
1933 &fs_info->generic_worker);
1934 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1935 fs_info->thread_pool_size,
1936 &fs_info->generic_worker);
1937 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1938 1, &fs_info->generic_worker);
1941 * endios are largely parallel and should have a very
1944 fs_info->endio_workers.idle_thresh = 4;
1945 fs_info->endio_meta_workers.idle_thresh = 4;
1947 fs_info->endio_write_workers.idle_thresh = 2;
1948 fs_info->endio_meta_write_workers.idle_thresh = 2;
1950 btrfs_start_workers(&fs_info->workers, 1);
1951 btrfs_start_workers(&fs_info->generic_worker, 1);
1952 btrfs_start_workers(&fs_info->submit_workers, 1);
1953 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1954 btrfs_start_workers(&fs_info->fixup_workers, 1);
1955 btrfs_start_workers(&fs_info->endio_workers, 1);
1956 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1957 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1958 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1959 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1961 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1962 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1963 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1965 nodesize = btrfs_super_nodesize(disk_super);
1966 leafsize = btrfs_super_leafsize(disk_super);
1967 sectorsize = btrfs_super_sectorsize(disk_super);
1968 stripesize = btrfs_super_stripesize(disk_super);
1969 tree_root->nodesize = nodesize;
1970 tree_root->leafsize = leafsize;
1971 tree_root->sectorsize = sectorsize;
1972 tree_root->stripesize = stripesize;
1974 sb->s_blocksize = sectorsize;
1975 sb->s_blocksize_bits = blksize_bits(sectorsize);
1977 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1978 sizeof(disk_super->magic))) {
1979 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1980 goto fail_sb_buffer;
1983 mutex_lock(&fs_info->chunk_mutex);
1984 ret = btrfs_read_sys_array(tree_root);
1985 mutex_unlock(&fs_info->chunk_mutex);
1987 printk(KERN_WARNING "btrfs: failed to read the system "
1988 "array on %s\n", sb->s_id);
1989 goto fail_sb_buffer;
1992 blocksize = btrfs_level_size(tree_root,
1993 btrfs_super_chunk_root_level(disk_super));
1994 generation = btrfs_super_chunk_root_generation(disk_super);
1996 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1997 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1999 chunk_root->node = read_tree_block(chunk_root,
2000 btrfs_super_chunk_root(disk_super),
2001 blocksize, generation);
2002 BUG_ON(!chunk_root->node);
2003 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2004 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2006 goto fail_chunk_root;
2008 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2009 chunk_root->commit_root = btrfs_root_node(chunk_root);
2011 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2012 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2015 mutex_lock(&fs_info->chunk_mutex);
2016 ret = btrfs_read_chunk_tree(chunk_root);
2017 mutex_unlock(&fs_info->chunk_mutex);
2019 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2021 goto fail_chunk_root;
2024 btrfs_close_extra_devices(fs_devices);
2026 blocksize = btrfs_level_size(tree_root,
2027 btrfs_super_root_level(disk_super));
2028 generation = btrfs_super_generation(disk_super);
2030 tree_root->node = read_tree_block(tree_root,
2031 btrfs_super_root(disk_super),
2032 blocksize, generation);
2033 if (!tree_root->node)
2034 goto fail_chunk_root;
2035 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2036 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2038 goto fail_tree_root;
2040 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2041 tree_root->commit_root = btrfs_root_node(tree_root);
2043 ret = find_and_setup_root(tree_root, fs_info,
2044 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2046 goto fail_tree_root;
2047 extent_root->track_dirty = 1;
2049 ret = find_and_setup_root(tree_root, fs_info,
2050 BTRFS_DEV_TREE_OBJECTID, dev_root);
2052 goto fail_extent_root;
2053 dev_root->track_dirty = 1;
2055 ret = find_and_setup_root(tree_root, fs_info,
2056 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2060 csum_root->track_dirty = 1;
2062 fs_info->generation = generation;
2063 fs_info->last_trans_committed = generation;
2064 fs_info->data_alloc_profile = (u64)-1;
2065 fs_info->metadata_alloc_profile = (u64)-1;
2066 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2068 ret = btrfs_read_block_groups(extent_root);
2070 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2071 goto fail_block_groups;
2074 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2076 if (IS_ERR(fs_info->cleaner_kthread))
2077 goto fail_block_groups;
2079 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2081 "btrfs-transaction");
2082 if (IS_ERR(fs_info->transaction_kthread))
2085 if (!btrfs_test_opt(tree_root, SSD) &&
2086 !btrfs_test_opt(tree_root, NOSSD) &&
2087 !fs_info->fs_devices->rotating) {
2088 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2090 btrfs_set_opt(fs_info->mount_opt, SSD);
2093 /* do not make disk changes in broken FS */
2094 if (btrfs_super_log_root(disk_super) != 0 &&
2095 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2096 u64 bytenr = btrfs_super_log_root(disk_super);
2098 if (fs_devices->rw_devices == 0) {
2099 printk(KERN_WARNING "Btrfs log replay required "
2102 goto fail_trans_kthread;
2105 btrfs_level_size(tree_root,
2106 btrfs_super_log_root_level(disk_super));
2108 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2109 if (!log_tree_root) {
2111 goto fail_trans_kthread;
2114 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2115 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2117 log_tree_root->node = read_tree_block(tree_root, bytenr,
2120 ret = btrfs_recover_log_trees(log_tree_root);
2123 if (sb->s_flags & MS_RDONLY) {
2124 ret = btrfs_commit_super(tree_root);
2129 ret = btrfs_find_orphan_roots(tree_root);
2132 if (!(sb->s_flags & MS_RDONLY)) {
2133 ret = btrfs_cleanup_fs_roots(fs_info);
2136 ret = btrfs_recover_relocation(tree_root);
2139 "btrfs: failed to recover relocation\n");
2141 goto fail_trans_kthread;
2145 location.objectid = BTRFS_FS_TREE_OBJECTID;
2146 location.type = BTRFS_ROOT_ITEM_KEY;
2147 location.offset = (u64)-1;
2149 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2150 if (!fs_info->fs_root)
2151 goto fail_trans_kthread;
2152 if (IS_ERR(fs_info->fs_root)) {
2153 err = PTR_ERR(fs_info->fs_root);
2154 goto fail_trans_kthread;
2157 if (!(sb->s_flags & MS_RDONLY)) {
2158 down_read(&fs_info->cleanup_work_sem);
2159 err = btrfs_orphan_cleanup(fs_info->fs_root);
2161 err = btrfs_orphan_cleanup(fs_info->tree_root);
2162 up_read(&fs_info->cleanup_work_sem);
2164 close_ctree(tree_root);
2165 return ERR_PTR(err);
2172 kthread_stop(fs_info->transaction_kthread);
2174 kthread_stop(fs_info->cleaner_kthread);
2177 * make sure we're done with the btree inode before we stop our
2180 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2181 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2184 btrfs_free_block_groups(fs_info);
2185 free_extent_buffer(csum_root->node);
2186 free_extent_buffer(csum_root->commit_root);
2188 free_extent_buffer(dev_root->node);
2189 free_extent_buffer(dev_root->commit_root);
2191 free_extent_buffer(extent_root->node);
2192 free_extent_buffer(extent_root->commit_root);
2194 free_extent_buffer(tree_root->node);
2195 free_extent_buffer(tree_root->commit_root);
2197 free_extent_buffer(chunk_root->node);
2198 free_extent_buffer(chunk_root->commit_root);
2200 btrfs_stop_workers(&fs_info->generic_worker);
2201 btrfs_stop_workers(&fs_info->fixup_workers);
2202 btrfs_stop_workers(&fs_info->delalloc_workers);
2203 btrfs_stop_workers(&fs_info->workers);
2204 btrfs_stop_workers(&fs_info->endio_workers);
2205 btrfs_stop_workers(&fs_info->endio_meta_workers);
2206 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2207 btrfs_stop_workers(&fs_info->endio_write_workers);
2208 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2209 btrfs_stop_workers(&fs_info->submit_workers);
2211 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2212 iput(fs_info->btree_inode);
2214 btrfs_close_devices(fs_info->fs_devices);
2215 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2217 bdi_destroy(&fs_info->bdi);
2219 cleanup_srcu_struct(&fs_info->subvol_srcu);
2227 return ERR_PTR(err);
2230 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2232 char b[BDEVNAME_SIZE];
2235 set_buffer_uptodate(bh);
2237 if (printk_ratelimit()) {
2238 printk(KERN_WARNING "lost page write due to "
2239 "I/O error on %s\n",
2240 bdevname(bh->b_bdev, b));
2242 /* note, we dont' set_buffer_write_io_error because we have
2243 * our own ways of dealing with the IO errors
2245 clear_buffer_uptodate(bh);
2251 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2253 struct buffer_head *bh;
2254 struct buffer_head *latest = NULL;
2255 struct btrfs_super_block *super;
2260 /* we would like to check all the supers, but that would make
2261 * a btrfs mount succeed after a mkfs from a different FS.
2262 * So, we need to add a special mount option to scan for
2263 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2265 for (i = 0; i < 1; i++) {
2266 bytenr = btrfs_sb_offset(i);
2267 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2269 bh = __bread(bdev, bytenr / 4096, 4096);
2273 super = (struct btrfs_super_block *)bh->b_data;
2274 if (btrfs_super_bytenr(super) != bytenr ||
2275 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2276 sizeof(super->magic))) {
2281 if (!latest || btrfs_super_generation(super) > transid) {
2284 transid = btrfs_super_generation(super);
2293 * this should be called twice, once with wait == 0 and
2294 * once with wait == 1. When wait == 0 is done, all the buffer heads
2295 * we write are pinned.
2297 * They are released when wait == 1 is done.
2298 * max_mirrors must be the same for both runs, and it indicates how
2299 * many supers on this one device should be written.
2301 * max_mirrors == 0 means to write them all.
2303 static int write_dev_supers(struct btrfs_device *device,
2304 struct btrfs_super_block *sb,
2305 int do_barriers, int wait, int max_mirrors)
2307 struct buffer_head *bh;
2313 int last_barrier = 0;
2315 if (max_mirrors == 0)
2316 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2318 /* make sure only the last submit_bh does a barrier */
2320 for (i = 0; i < max_mirrors; i++) {
2321 bytenr = btrfs_sb_offset(i);
2322 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2323 device->total_bytes)
2329 for (i = 0; i < max_mirrors; i++) {
2330 bytenr = btrfs_sb_offset(i);
2331 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2335 bh = __find_get_block(device->bdev, bytenr / 4096,
2336 BTRFS_SUPER_INFO_SIZE);
2339 if (!buffer_uptodate(bh))
2342 /* drop our reference */
2345 /* drop the reference from the wait == 0 run */
2349 btrfs_set_super_bytenr(sb, bytenr);
2352 crc = btrfs_csum_data(NULL, (char *)sb +
2353 BTRFS_CSUM_SIZE, crc,
2354 BTRFS_SUPER_INFO_SIZE -
2356 btrfs_csum_final(crc, sb->csum);
2359 * one reference for us, and we leave it for the
2362 bh = __getblk(device->bdev, bytenr / 4096,
2363 BTRFS_SUPER_INFO_SIZE);
2364 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2366 /* one reference for submit_bh */
2369 set_buffer_uptodate(bh);
2371 bh->b_end_io = btrfs_end_buffer_write_sync;
2374 if (i == last_barrier && do_barriers)
2375 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2377 ret = submit_bh(WRITE_SYNC, bh);
2382 return errors < i ? 0 : -1;
2385 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2387 struct list_head *head;
2388 struct btrfs_device *dev;
2389 struct btrfs_super_block *sb;
2390 struct btrfs_dev_item *dev_item;
2394 int total_errors = 0;
2397 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2398 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2400 sb = &root->fs_info->super_for_commit;
2401 dev_item = &sb->dev_item;
2403 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2404 head = &root->fs_info->fs_devices->devices;
2405 list_for_each_entry(dev, head, dev_list) {
2410 if (!dev->in_fs_metadata || !dev->writeable)
2413 btrfs_set_stack_device_generation(dev_item, 0);
2414 btrfs_set_stack_device_type(dev_item, dev->type);
2415 btrfs_set_stack_device_id(dev_item, dev->devid);
2416 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2417 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2418 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2419 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2420 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2421 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2422 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2424 flags = btrfs_super_flags(sb);
2425 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2427 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2431 if (total_errors > max_errors) {
2432 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2438 list_for_each_entry(dev, head, dev_list) {
2441 if (!dev->in_fs_metadata || !dev->writeable)
2444 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2448 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2449 if (total_errors > max_errors) {
2450 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2457 int write_ctree_super(struct btrfs_trans_handle *trans,
2458 struct btrfs_root *root, int max_mirrors)
2462 ret = write_all_supers(root, max_mirrors);
2466 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2468 spin_lock(&fs_info->fs_roots_radix_lock);
2469 radix_tree_delete(&fs_info->fs_roots_radix,
2470 (unsigned long)root->root_key.objectid);
2471 spin_unlock(&fs_info->fs_roots_radix_lock);
2473 if (btrfs_root_refs(&root->root_item) == 0)
2474 synchronize_srcu(&fs_info->subvol_srcu);
2480 static void free_fs_root(struct btrfs_root *root)
2482 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2483 if (root->anon_super.s_dev) {
2484 down_write(&root->anon_super.s_umount);
2485 kill_anon_super(&root->anon_super);
2487 free_extent_buffer(root->node);
2488 free_extent_buffer(root->commit_root);
2493 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2496 struct btrfs_root *gang[8];
2499 while (!list_empty(&fs_info->dead_roots)) {
2500 gang[0] = list_entry(fs_info->dead_roots.next,
2501 struct btrfs_root, root_list);
2502 list_del(&gang[0]->root_list);
2504 if (gang[0]->in_radix) {
2505 btrfs_free_fs_root(fs_info, gang[0]);
2507 free_extent_buffer(gang[0]->node);
2508 free_extent_buffer(gang[0]->commit_root);
2514 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2519 for (i = 0; i < ret; i++)
2520 btrfs_free_fs_root(fs_info, gang[i]);
2525 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2527 u64 root_objectid = 0;
2528 struct btrfs_root *gang[8];
2533 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2534 (void **)gang, root_objectid,
2539 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2540 for (i = 0; i < ret; i++) {
2543 root_objectid = gang[i]->root_key.objectid;
2544 err = btrfs_orphan_cleanup(gang[i]);
2553 int btrfs_commit_super(struct btrfs_root *root)
2555 struct btrfs_trans_handle *trans;
2558 mutex_lock(&root->fs_info->cleaner_mutex);
2559 btrfs_run_delayed_iputs(root);
2560 btrfs_clean_old_snapshots(root);
2561 mutex_unlock(&root->fs_info->cleaner_mutex);
2563 /* wait until ongoing cleanup work done */
2564 down_write(&root->fs_info->cleanup_work_sem);
2565 up_write(&root->fs_info->cleanup_work_sem);
2567 trans = btrfs_join_transaction(root, 1);
2569 return PTR_ERR(trans);
2570 ret = btrfs_commit_transaction(trans, root);
2572 /* run commit again to drop the original snapshot */
2573 trans = btrfs_join_transaction(root, 1);
2575 return PTR_ERR(trans);
2576 btrfs_commit_transaction(trans, root);
2577 ret = btrfs_write_and_wait_transaction(NULL, root);
2580 ret = write_ctree_super(NULL, root, 0);
2584 int close_ctree(struct btrfs_root *root)
2586 struct btrfs_fs_info *fs_info = root->fs_info;
2589 fs_info->closing = 1;
2592 btrfs_put_block_group_cache(fs_info);
2595 * Here come 2 situations when btrfs is broken to flip readonly:
2597 * 1. when btrfs flips readonly somewhere else before
2598 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2599 * and btrfs will skip to write sb directly to keep
2600 * ERROR state on disk.
2602 * 2. when btrfs flips readonly just in btrfs_commit_super,
2603 * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2604 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2605 * btrfs will cleanup all FS resources first and write sb then.
2607 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2608 ret = btrfs_commit_super(root);
2610 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2613 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2614 ret = btrfs_error_commit_super(root);
2616 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2619 kthread_stop(root->fs_info->transaction_kthread);
2620 kthread_stop(root->fs_info->cleaner_kthread);
2622 fs_info->closing = 2;
2625 if (fs_info->delalloc_bytes) {
2626 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2627 (unsigned long long)fs_info->delalloc_bytes);
2629 if (fs_info->total_ref_cache_size) {
2630 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2631 (unsigned long long)fs_info->total_ref_cache_size);
2634 free_extent_buffer(fs_info->extent_root->node);
2635 free_extent_buffer(fs_info->extent_root->commit_root);
2636 free_extent_buffer(fs_info->tree_root->node);
2637 free_extent_buffer(fs_info->tree_root->commit_root);
2638 free_extent_buffer(root->fs_info->chunk_root->node);
2639 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2640 free_extent_buffer(root->fs_info->dev_root->node);
2641 free_extent_buffer(root->fs_info->dev_root->commit_root);
2642 free_extent_buffer(root->fs_info->csum_root->node);
2643 free_extent_buffer(root->fs_info->csum_root->commit_root);
2645 btrfs_free_block_groups(root->fs_info);
2647 del_fs_roots(fs_info);
2649 iput(fs_info->btree_inode);
2651 btrfs_stop_workers(&fs_info->generic_worker);
2652 btrfs_stop_workers(&fs_info->fixup_workers);
2653 btrfs_stop_workers(&fs_info->delalloc_workers);
2654 btrfs_stop_workers(&fs_info->workers);
2655 btrfs_stop_workers(&fs_info->endio_workers);
2656 btrfs_stop_workers(&fs_info->endio_meta_workers);
2657 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2658 btrfs_stop_workers(&fs_info->endio_write_workers);
2659 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2660 btrfs_stop_workers(&fs_info->submit_workers);
2662 btrfs_close_devices(fs_info->fs_devices);
2663 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2665 bdi_destroy(&fs_info->bdi);
2666 cleanup_srcu_struct(&fs_info->subvol_srcu);
2668 kfree(fs_info->extent_root);
2669 kfree(fs_info->tree_root);
2670 kfree(fs_info->chunk_root);
2671 kfree(fs_info->dev_root);
2672 kfree(fs_info->csum_root);
2678 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2681 struct inode *btree_inode = buf->first_page->mapping->host;
2683 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2688 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2693 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2695 struct inode *btree_inode = buf->first_page->mapping->host;
2696 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2700 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2702 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2703 u64 transid = btrfs_header_generation(buf);
2704 struct inode *btree_inode = root->fs_info->btree_inode;
2707 btrfs_assert_tree_locked(buf);
2708 if (transid != root->fs_info->generation) {
2709 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2710 "found %llu running %llu\n",
2711 (unsigned long long)buf->start,
2712 (unsigned long long)transid,
2713 (unsigned long long)root->fs_info->generation);
2716 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2719 spin_lock(&root->fs_info->delalloc_lock);
2720 root->fs_info->dirty_metadata_bytes += buf->len;
2721 spin_unlock(&root->fs_info->delalloc_lock);
2725 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2728 * looks as though older kernels can get into trouble with
2729 * this code, they end up stuck in balance_dirty_pages forever
2732 unsigned long thresh = 32 * 1024 * 1024;
2734 if (current->flags & PF_MEMALLOC)
2737 num_dirty = root->fs_info->dirty_metadata_bytes;
2739 if (num_dirty > thresh) {
2740 balance_dirty_pages_ratelimited_nr(
2741 root->fs_info->btree_inode->i_mapping, 1);
2746 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2748 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2750 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2752 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2756 int btree_lock_page_hook(struct page *page)
2758 struct inode *inode = page->mapping->host;
2759 struct btrfs_root *root = BTRFS_I(inode)->root;
2760 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2761 struct extent_buffer *eb;
2763 u64 bytenr = page_offset(page);
2765 if (page->private == EXTENT_PAGE_PRIVATE)
2768 len = page->private >> 2;
2769 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2773 btrfs_tree_lock(eb);
2774 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2776 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2777 spin_lock(&root->fs_info->delalloc_lock);
2778 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2779 root->fs_info->dirty_metadata_bytes -= eb->len;
2782 spin_unlock(&root->fs_info->delalloc_lock);
2785 btrfs_tree_unlock(eb);
2786 free_extent_buffer(eb);
2792 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2798 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2799 printk(KERN_WARNING "warning: mount fs with errors, "
2800 "running btrfsck is recommended\n");
2803 int btrfs_error_commit_super(struct btrfs_root *root)
2807 mutex_lock(&root->fs_info->cleaner_mutex);
2808 btrfs_run_delayed_iputs(root);
2809 mutex_unlock(&root->fs_info->cleaner_mutex);
2811 down_write(&root->fs_info->cleanup_work_sem);
2812 up_write(&root->fs_info->cleanup_work_sem);
2814 /* cleanup FS via transaction */
2815 btrfs_cleanup_transaction(root);
2817 ret = write_ctree_super(NULL, root, 0);
2822 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2824 struct btrfs_inode *btrfs_inode;
2825 struct list_head splice;
2827 INIT_LIST_HEAD(&splice);
2829 mutex_lock(&root->fs_info->ordered_operations_mutex);
2830 spin_lock(&root->fs_info->ordered_extent_lock);
2832 list_splice_init(&root->fs_info->ordered_operations, &splice);
2833 while (!list_empty(&splice)) {
2834 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2835 ordered_operations);
2837 list_del_init(&btrfs_inode->ordered_operations);
2839 btrfs_invalidate_inodes(btrfs_inode->root);
2842 spin_unlock(&root->fs_info->ordered_extent_lock);
2843 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2848 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2850 struct list_head splice;
2851 struct btrfs_ordered_extent *ordered;
2852 struct inode *inode;
2854 INIT_LIST_HEAD(&splice);
2856 spin_lock(&root->fs_info->ordered_extent_lock);
2858 list_splice_init(&root->fs_info->ordered_extents, &splice);
2859 while (!list_empty(&splice)) {
2860 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2863 list_del_init(&ordered->root_extent_list);
2864 atomic_inc(&ordered->refs);
2866 /* the inode may be getting freed (in sys_unlink path). */
2867 inode = igrab(ordered->inode);
2869 spin_unlock(&root->fs_info->ordered_extent_lock);
2873 atomic_set(&ordered->refs, 1);
2874 btrfs_put_ordered_extent(ordered);
2876 spin_lock(&root->fs_info->ordered_extent_lock);
2879 spin_unlock(&root->fs_info->ordered_extent_lock);
2884 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2885 struct btrfs_root *root)
2887 struct rb_node *node;
2888 struct btrfs_delayed_ref_root *delayed_refs;
2889 struct btrfs_delayed_ref_node *ref;
2892 delayed_refs = &trans->delayed_refs;
2894 spin_lock(&delayed_refs->lock);
2895 if (delayed_refs->num_entries == 0) {
2896 printk(KERN_INFO "delayed_refs has NO entry\n");
2900 node = rb_first(&delayed_refs->root);
2902 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2903 node = rb_next(node);
2906 rb_erase(&ref->rb_node, &delayed_refs->root);
2907 delayed_refs->num_entries--;
2909 atomic_set(&ref->refs, 1);
2910 if (btrfs_delayed_ref_is_head(ref)) {
2911 struct btrfs_delayed_ref_head *head;
2913 head = btrfs_delayed_node_to_head(ref);
2914 mutex_lock(&head->mutex);
2915 kfree(head->extent_op);
2916 delayed_refs->num_heads--;
2917 if (list_empty(&head->cluster))
2918 delayed_refs->num_heads_ready--;
2919 list_del_init(&head->cluster);
2920 mutex_unlock(&head->mutex);
2923 spin_unlock(&delayed_refs->lock);
2924 btrfs_put_delayed_ref(ref);
2927 spin_lock(&delayed_refs->lock);
2930 spin_unlock(&delayed_refs->lock);
2935 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2937 struct btrfs_pending_snapshot *snapshot;
2938 struct list_head splice;
2940 INIT_LIST_HEAD(&splice);
2942 list_splice_init(&t->pending_snapshots, &splice);
2944 while (!list_empty(&splice)) {
2945 snapshot = list_entry(splice.next,
2946 struct btrfs_pending_snapshot,
2949 list_del_init(&snapshot->list);
2957 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2959 struct btrfs_inode *btrfs_inode;
2960 struct list_head splice;
2962 INIT_LIST_HEAD(&splice);
2964 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2966 spin_lock(&root->fs_info->delalloc_lock);
2968 while (!list_empty(&splice)) {
2969 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2972 list_del_init(&btrfs_inode->delalloc_inodes);
2974 btrfs_invalidate_inodes(btrfs_inode->root);
2977 spin_unlock(&root->fs_info->delalloc_lock);
2982 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2983 struct extent_io_tree *dirty_pages,
2988 struct inode *btree_inode = root->fs_info->btree_inode;
2989 struct extent_buffer *eb;
2993 unsigned long index;
2996 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3001 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3002 while (start <= end) {
3003 index = start >> PAGE_CACHE_SHIFT;
3004 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3005 page = find_get_page(btree_inode->i_mapping, index);
3008 offset = page_offset(page);
3010 spin_lock(&dirty_pages->buffer_lock);
3011 eb = radix_tree_lookup(
3012 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3013 offset >> PAGE_CACHE_SHIFT);
3014 spin_unlock(&dirty_pages->buffer_lock);
3016 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3018 atomic_set(&eb->refs, 1);
3020 if (PageWriteback(page))
3021 end_page_writeback(page);
3024 if (PageDirty(page)) {
3025 clear_page_dirty_for_io(page);
3026 spin_lock_irq(&page->mapping->tree_lock);
3027 radix_tree_tag_clear(&page->mapping->page_tree,
3029 PAGECACHE_TAG_DIRTY);
3030 spin_unlock_irq(&page->mapping->tree_lock);
3033 page->mapping->a_ops->invalidatepage(page, 0);
3041 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3042 struct extent_io_tree *pinned_extents)
3044 struct extent_io_tree *unpin;
3049 unpin = pinned_extents;
3051 ret = find_first_extent_bit(unpin, 0, &start, &end,
3057 ret = btrfs_error_discard_extent(root, start, end + 1 - start);
3059 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3060 btrfs_error_unpin_extent_range(root, start, end);
3067 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3069 struct btrfs_transaction *t;
3074 mutex_lock(&root->fs_info->trans_mutex);
3075 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3077 list_splice_init(&root->fs_info->trans_list, &list);
3078 while (!list_empty(&list)) {
3079 t = list_entry(list.next, struct btrfs_transaction, list);
3083 btrfs_destroy_ordered_operations(root);
3085 btrfs_destroy_ordered_extents(root);
3087 btrfs_destroy_delayed_refs(t, root);
3089 btrfs_block_rsv_release(root,
3090 &root->fs_info->trans_block_rsv,
3091 t->dirty_pages.dirty_bytes);
3093 /* FIXME: cleanup wait for commit */
3096 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3097 wake_up(&root->fs_info->transaction_blocked_wait);
3100 if (waitqueue_active(&root->fs_info->transaction_wait))
3101 wake_up(&root->fs_info->transaction_wait);
3102 mutex_unlock(&root->fs_info->trans_mutex);
3104 mutex_lock(&root->fs_info->trans_mutex);
3106 if (waitqueue_active(&t->commit_wait))
3107 wake_up(&t->commit_wait);
3108 mutex_unlock(&root->fs_info->trans_mutex);
3110 mutex_lock(&root->fs_info->trans_mutex);
3112 btrfs_destroy_pending_snapshots(t);
3114 btrfs_destroy_delalloc_inodes(root);
3116 spin_lock(&root->fs_info->new_trans_lock);
3117 root->fs_info->running_transaction = NULL;
3118 spin_unlock(&root->fs_info->new_trans_lock);
3120 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3123 btrfs_destroy_pinned_extent(root,
3124 root->fs_info->pinned_extents);
3127 list_del_init(&t->list);
3128 memset(t, 0, sizeof(*t));
3129 kmem_cache_free(btrfs_transaction_cachep, t);
3132 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3133 mutex_unlock(&root->fs_info->trans_mutex);
3138 static struct extent_io_ops btree_extent_io_ops = {
3139 .write_cache_pages_lock_hook = btree_lock_page_hook,
3140 .readpage_end_io_hook = btree_readpage_end_io_hook,
3141 .submit_bio_hook = btree_submit_bio_hook,
3142 /* note we're sharing with inode.c for the merge bio hook */
3143 .merge_bio_hook = btrfs_merge_bio_hook,