2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
39 #include "async-thread.h"
42 #include "free-space-cache.h"
44 static struct extent_io_ops btree_extent_io_ops;
45 static void end_workqueue_fn(struct btrfs_work *work);
46 static void free_fs_root(struct btrfs_root *root);
47 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
49 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
50 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
51 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
52 struct btrfs_root *root);
53 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
54 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
55 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
56 struct extent_io_tree *dirty_pages,
58 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
59 struct extent_io_tree *pinned_extents);
60 static int btrfs_cleanup_transaction(struct btrfs_root *root);
63 * end_io_wq structs are used to do processing in task context when an IO is
64 * complete. This is used during reads to verify checksums, and it is used
65 * by writes to insert metadata for new file extents after IO is complete.
71 struct btrfs_fs_info *info;
74 struct list_head list;
75 struct btrfs_work work;
79 * async submit bios are used to offload expensive checksumming
80 * onto the worker threads. They checksum file and metadata bios
81 * just before they are sent down the IO stack.
83 struct async_submit_bio {
86 struct list_head list;
87 extent_submit_bio_hook_t *submit_bio_start;
88 extent_submit_bio_hook_t *submit_bio_done;
91 unsigned long bio_flags;
93 * bio_offset is optional, can be used if the pages in the bio
94 * can't tell us where in the file the bio should go
97 struct btrfs_work work;
100 /* These are used to set the lockdep class on the extent buffer locks.
101 * The class is set by the readpage_end_io_hook after the buffer has
102 * passed csum validation but before the pages are unlocked.
104 * The lockdep class is also set by btrfs_init_new_buffer on freshly
107 * The class is based on the level in the tree block, which allows lockdep
108 * to know that lower nodes nest inside the locks of higher nodes.
110 * We also add a check to make sure the highest level of the tree is
111 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
112 * code needs update as well.
114 #ifdef CONFIG_DEBUG_LOCK_ALLOC
115 # if BTRFS_MAX_LEVEL != 8
118 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
119 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
129 /* highest possible level */
135 * extents on the btree inode are pretty simple, there's one extent
136 * that covers the entire device
138 static struct extent_map *btree_get_extent(struct inode *inode,
139 struct page *page, size_t page_offset, u64 start, u64 len,
142 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
143 struct extent_map *em;
146 read_lock(&em_tree->lock);
147 em = lookup_extent_mapping(em_tree, start, len);
150 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
151 read_unlock(&em_tree->lock);
154 read_unlock(&em_tree->lock);
156 em = alloc_extent_map(GFP_NOFS);
158 em = ERR_PTR(-ENOMEM);
163 em->block_len = (u64)-1;
165 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
167 write_lock(&em_tree->lock);
168 ret = add_extent_mapping(em_tree, em);
169 if (ret == -EEXIST) {
170 u64 failed_start = em->start;
171 u64 failed_len = em->len;
174 em = lookup_extent_mapping(em_tree, start, len);
178 em = lookup_extent_mapping(em_tree, failed_start,
186 write_unlock(&em_tree->lock);
194 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
196 return crc32c(seed, data, len);
199 void btrfs_csum_final(u32 crc, char *result)
201 *(__le32 *)result = ~cpu_to_le32(crc);
205 * compute the csum for a btree block, and either verify it or write it
206 * into the csum field of the block.
208 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
212 btrfs_super_csum_size(&root->fs_info->super_copy);
215 unsigned long cur_len;
216 unsigned long offset = BTRFS_CSUM_SIZE;
217 char *map_token = NULL;
219 unsigned long map_start;
220 unsigned long map_len;
223 unsigned long inline_result;
225 len = buf->len - offset;
227 err = map_private_extent_buffer(buf, offset, 32,
229 &map_start, &map_len, KM_USER0);
232 cur_len = min(len, map_len - (offset - map_start));
233 crc = btrfs_csum_data(root, kaddr + offset - map_start,
237 unmap_extent_buffer(buf, map_token, KM_USER0);
239 if (csum_size > sizeof(inline_result)) {
240 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
244 result = (char *)&inline_result;
247 btrfs_csum_final(crc, result);
250 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
253 memcpy(&found, result, csum_size);
255 read_extent_buffer(buf, &val, 0, csum_size);
256 if (printk_ratelimit()) {
257 printk(KERN_INFO "btrfs: %s checksum verify "
258 "failed on %llu wanted %X found %X "
260 root->fs_info->sb->s_id,
261 (unsigned long long)buf->start, val, found,
262 btrfs_header_level(buf));
264 if (result != (char *)&inline_result)
269 write_extent_buffer(buf, result, 0, csum_size);
271 if (result != (char *)&inline_result)
277 * we can't consider a given block up to date unless the transid of the
278 * block matches the transid in the parent node's pointer. This is how we
279 * detect blocks that either didn't get written at all or got written
280 * in the wrong place.
282 static int verify_parent_transid(struct extent_io_tree *io_tree,
283 struct extent_buffer *eb, u64 parent_transid)
285 struct extent_state *cached_state = NULL;
288 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
291 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
292 0, &cached_state, GFP_NOFS);
293 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
294 btrfs_header_generation(eb) == parent_transid) {
298 if (printk_ratelimit()) {
299 printk("parent transid verify failed on %llu wanted %llu "
301 (unsigned long long)eb->start,
302 (unsigned long long)parent_transid,
303 (unsigned long long)btrfs_header_generation(eb));
306 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
308 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
309 &cached_state, GFP_NOFS);
314 * helper to read a given tree block, doing retries as required when
315 * the checksums don't match and we have alternate mirrors to try.
317 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
318 struct extent_buffer *eb,
319 u64 start, u64 parent_transid)
321 struct extent_io_tree *io_tree;
326 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
327 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
329 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
330 btree_get_extent, mirror_num);
332 !verify_parent_transid(io_tree, eb, parent_transid))
336 * This buffer's crc is fine, but its contents are corrupted, so
337 * there is no reason to read the other copies, they won't be
340 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
343 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
349 if (mirror_num > num_copies)
356 * checksum a dirty tree block before IO. This has extra checks to make sure
357 * we only fill in the checksum field in the first page of a multi-page block
360 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
362 struct extent_io_tree *tree;
363 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
366 struct extent_buffer *eb;
369 tree = &BTRFS_I(page->mapping->host)->io_tree;
371 if (page->private == EXTENT_PAGE_PRIVATE) {
375 if (!page->private) {
379 len = page->private >> 2;
382 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
387 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
388 btrfs_header_generation(eb));
390 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
392 found_start = btrfs_header_bytenr(eb);
393 if (found_start != start) {
397 if (eb->first_page != page) {
401 if (!PageUptodate(page)) {
405 csum_tree_block(root, eb, 0);
407 free_extent_buffer(eb);
412 static int check_tree_block_fsid(struct btrfs_root *root,
413 struct extent_buffer *eb)
415 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
416 u8 fsid[BTRFS_UUID_SIZE];
419 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
422 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
426 fs_devices = fs_devices->seed;
431 #define CORRUPT(reason, eb, root, slot) \
432 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
433 "root=%llu, slot=%d\n", reason, \
434 (unsigned long long)btrfs_header_bytenr(eb), \
435 (unsigned long long)root->objectid, slot)
437 static noinline int check_leaf(struct btrfs_root *root,
438 struct extent_buffer *leaf)
440 struct btrfs_key key;
441 struct btrfs_key leaf_key;
442 u32 nritems = btrfs_header_nritems(leaf);
448 /* Check the 0 item */
449 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
450 BTRFS_LEAF_DATA_SIZE(root)) {
451 CORRUPT("invalid item offset size pair", leaf, root, 0);
456 * Check to make sure each items keys are in the correct order and their
457 * offsets make sense. We only have to loop through nritems-1 because
458 * we check the current slot against the next slot, which verifies the
459 * next slot's offset+size makes sense and that the current's slot
462 for (slot = 0; slot < nritems - 1; slot++) {
463 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
464 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
466 /* Make sure the keys are in the right order */
467 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
468 CORRUPT("bad key order", leaf, root, slot);
473 * Make sure the offset and ends are right, remember that the
474 * item data starts at the end of the leaf and grows towards the
477 if (btrfs_item_offset_nr(leaf, slot) !=
478 btrfs_item_end_nr(leaf, slot + 1)) {
479 CORRUPT("slot offset bad", leaf, root, slot);
484 * Check to make sure that we don't point outside of the leaf,
485 * just incase all the items are consistent to eachother, but
486 * all point outside of the leaf.
488 if (btrfs_item_end_nr(leaf, slot) >
489 BTRFS_LEAF_DATA_SIZE(root)) {
490 CORRUPT("slot end outside of leaf", leaf, root, slot);
498 #ifdef CONFIG_DEBUG_LOCK_ALLOC
499 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
501 lockdep_set_class_and_name(&eb->lock,
502 &btrfs_eb_class[level],
503 btrfs_eb_name[level]);
507 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
508 struct extent_state *state)
510 struct extent_io_tree *tree;
514 struct extent_buffer *eb;
515 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
518 tree = &BTRFS_I(page->mapping->host)->io_tree;
519 if (page->private == EXTENT_PAGE_PRIVATE)
524 len = page->private >> 2;
527 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
533 found_start = btrfs_header_bytenr(eb);
534 if (found_start != start) {
535 if (printk_ratelimit()) {
536 printk(KERN_INFO "btrfs bad tree block start "
538 (unsigned long long)found_start,
539 (unsigned long long)eb->start);
544 if (eb->first_page != page) {
545 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
546 eb->first_page->index, page->index);
551 if (check_tree_block_fsid(root, eb)) {
552 if (printk_ratelimit()) {
553 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
554 (unsigned long long)eb->start);
559 found_level = btrfs_header_level(eb);
561 btrfs_set_buffer_lockdep_class(eb, found_level);
563 ret = csum_tree_block(root, eb, 1);
570 * If this is a leaf block and it is corrupt, set the corrupt bit so
571 * that we don't try and read the other copies of this block, just
574 if (found_level == 0 && check_leaf(root, eb)) {
575 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
579 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
580 end = eb->start + end - 1;
582 free_extent_buffer(eb);
587 static void end_workqueue_bio(struct bio *bio, int err)
589 struct end_io_wq *end_io_wq = bio->bi_private;
590 struct btrfs_fs_info *fs_info;
592 fs_info = end_io_wq->info;
593 end_io_wq->error = err;
594 end_io_wq->work.func = end_workqueue_fn;
595 end_io_wq->work.flags = 0;
597 if (bio->bi_rw & REQ_WRITE) {
598 if (end_io_wq->metadata == 1)
599 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
601 else if (end_io_wq->metadata == 2)
602 btrfs_queue_worker(&fs_info->endio_freespace_worker,
605 btrfs_queue_worker(&fs_info->endio_write_workers,
608 if (end_io_wq->metadata)
609 btrfs_queue_worker(&fs_info->endio_meta_workers,
612 btrfs_queue_worker(&fs_info->endio_workers,
618 * For the metadata arg you want
621 * 1 - if normal metadta
622 * 2 - if writing to the free space cache area
624 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
627 struct end_io_wq *end_io_wq;
628 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
632 end_io_wq->private = bio->bi_private;
633 end_io_wq->end_io = bio->bi_end_io;
634 end_io_wq->info = info;
635 end_io_wq->error = 0;
636 end_io_wq->bio = bio;
637 end_io_wq->metadata = metadata;
639 bio->bi_private = end_io_wq;
640 bio->bi_end_io = end_workqueue_bio;
644 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
646 unsigned long limit = min_t(unsigned long,
647 info->workers.max_workers,
648 info->fs_devices->open_devices);
652 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
654 return atomic_read(&info->nr_async_bios) >
655 btrfs_async_submit_limit(info);
658 static void run_one_async_start(struct btrfs_work *work)
660 struct async_submit_bio *async;
662 async = container_of(work, struct async_submit_bio, work);
663 async->submit_bio_start(async->inode, async->rw, async->bio,
664 async->mirror_num, async->bio_flags,
668 static void run_one_async_done(struct btrfs_work *work)
670 struct btrfs_fs_info *fs_info;
671 struct async_submit_bio *async;
674 async = container_of(work, struct async_submit_bio, work);
675 fs_info = BTRFS_I(async->inode)->root->fs_info;
677 limit = btrfs_async_submit_limit(fs_info);
678 limit = limit * 2 / 3;
680 atomic_dec(&fs_info->nr_async_submits);
682 if (atomic_read(&fs_info->nr_async_submits) < limit &&
683 waitqueue_active(&fs_info->async_submit_wait))
684 wake_up(&fs_info->async_submit_wait);
686 async->submit_bio_done(async->inode, async->rw, async->bio,
687 async->mirror_num, async->bio_flags,
691 static void run_one_async_free(struct btrfs_work *work)
693 struct async_submit_bio *async;
695 async = container_of(work, struct async_submit_bio, work);
699 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
700 int rw, struct bio *bio, int mirror_num,
701 unsigned long bio_flags,
703 extent_submit_bio_hook_t *submit_bio_start,
704 extent_submit_bio_hook_t *submit_bio_done)
706 struct async_submit_bio *async;
708 async = kmalloc(sizeof(*async), GFP_NOFS);
712 async->inode = inode;
715 async->mirror_num = mirror_num;
716 async->submit_bio_start = submit_bio_start;
717 async->submit_bio_done = submit_bio_done;
719 async->work.func = run_one_async_start;
720 async->work.ordered_func = run_one_async_done;
721 async->work.ordered_free = run_one_async_free;
723 async->work.flags = 0;
724 async->bio_flags = bio_flags;
725 async->bio_offset = bio_offset;
727 atomic_inc(&fs_info->nr_async_submits);
730 btrfs_set_work_high_prio(&async->work);
732 btrfs_queue_worker(&fs_info->workers, &async->work);
734 while (atomic_read(&fs_info->async_submit_draining) &&
735 atomic_read(&fs_info->nr_async_submits)) {
736 wait_event(fs_info->async_submit_wait,
737 (atomic_read(&fs_info->nr_async_submits) == 0));
743 static int btree_csum_one_bio(struct bio *bio)
745 struct bio_vec *bvec = bio->bi_io_vec;
747 struct btrfs_root *root;
749 WARN_ON(bio->bi_vcnt <= 0);
750 while (bio_index < bio->bi_vcnt) {
751 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
752 csum_dirty_buffer(root, bvec->bv_page);
759 static int __btree_submit_bio_start(struct inode *inode, int rw,
760 struct bio *bio, int mirror_num,
761 unsigned long bio_flags,
765 * when we're called for a write, we're already in the async
766 * submission context. Just jump into btrfs_map_bio
768 btree_csum_one_bio(bio);
772 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
773 int mirror_num, unsigned long bio_flags,
777 * when we're called for a write, we're already in the async
778 * submission context. Just jump into btrfs_map_bio
780 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
783 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
784 int mirror_num, unsigned long bio_flags,
789 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
793 if (!(rw & REQ_WRITE)) {
795 * called for a read, do the setup so that checksum validation
796 * can happen in the async kernel threads
798 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
803 * kthread helpers are used to submit writes so that checksumming
804 * can happen in parallel across all CPUs
806 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
807 inode, rw, bio, mirror_num, 0,
809 __btree_submit_bio_start,
810 __btree_submit_bio_done);
813 #ifdef CONFIG_MIGRATION
814 static int btree_migratepage(struct address_space *mapping,
815 struct page *newpage, struct page *page)
818 * we can't safely write a btree page from here,
819 * we haven't done the locking hook
824 * Buffers may be managed in a filesystem specific way.
825 * We must have no buffers or drop them.
827 if (page_has_private(page) &&
828 !try_to_release_page(page, GFP_KERNEL))
830 return migrate_page(mapping, newpage, page);
834 static int btree_writepage(struct page *page, struct writeback_control *wbc)
836 struct extent_io_tree *tree;
837 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
838 struct extent_buffer *eb;
841 tree = &BTRFS_I(page->mapping->host)->io_tree;
842 if (!(current->flags & PF_MEMALLOC)) {
843 return extent_write_full_page(tree, page,
844 btree_get_extent, wbc);
847 redirty_page_for_writepage(wbc, page);
848 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
851 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
853 spin_lock(&root->fs_info->delalloc_lock);
854 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
855 spin_unlock(&root->fs_info->delalloc_lock);
857 free_extent_buffer(eb);
863 static int btree_writepages(struct address_space *mapping,
864 struct writeback_control *wbc)
866 struct extent_io_tree *tree;
867 tree = &BTRFS_I(mapping->host)->io_tree;
868 if (wbc->sync_mode == WB_SYNC_NONE) {
869 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
871 unsigned long thresh = 32 * 1024 * 1024;
873 if (wbc->for_kupdate)
876 /* this is a bit racy, but that's ok */
877 num_dirty = root->fs_info->dirty_metadata_bytes;
878 if (num_dirty < thresh)
881 return extent_writepages(tree, mapping, btree_get_extent, wbc);
884 static int btree_readpage(struct file *file, struct page *page)
886 struct extent_io_tree *tree;
887 tree = &BTRFS_I(page->mapping->host)->io_tree;
888 return extent_read_full_page(tree, page, btree_get_extent);
891 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
893 struct extent_io_tree *tree;
894 struct extent_map_tree *map;
897 if (PageWriteback(page) || PageDirty(page))
900 tree = &BTRFS_I(page->mapping->host)->io_tree;
901 map = &BTRFS_I(page->mapping->host)->extent_tree;
903 ret = try_release_extent_state(map, tree, page, gfp_flags);
907 ret = try_release_extent_buffer(tree, page);
909 ClearPagePrivate(page);
910 set_page_private(page, 0);
911 page_cache_release(page);
917 static void btree_invalidatepage(struct page *page, unsigned long offset)
919 struct extent_io_tree *tree;
920 tree = &BTRFS_I(page->mapping->host)->io_tree;
921 extent_invalidatepage(tree, page, offset);
922 btree_releasepage(page, GFP_NOFS);
923 if (PagePrivate(page)) {
924 printk(KERN_WARNING "btrfs warning page private not zero "
925 "on page %llu\n", (unsigned long long)page_offset(page));
926 ClearPagePrivate(page);
927 set_page_private(page, 0);
928 page_cache_release(page);
932 static const struct address_space_operations btree_aops = {
933 .readpage = btree_readpage,
934 .writepage = btree_writepage,
935 .writepages = btree_writepages,
936 .releasepage = btree_releasepage,
937 .invalidatepage = btree_invalidatepage,
938 .sync_page = block_sync_page,
939 #ifdef CONFIG_MIGRATION
940 .migratepage = btree_migratepage,
944 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
947 struct extent_buffer *buf = NULL;
948 struct inode *btree_inode = root->fs_info->btree_inode;
951 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
954 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
955 buf, 0, 0, btree_get_extent, 0);
956 free_extent_buffer(buf);
960 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
961 u64 bytenr, u32 blocksize)
963 struct inode *btree_inode = root->fs_info->btree_inode;
964 struct extent_buffer *eb;
965 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
966 bytenr, blocksize, GFP_NOFS);
970 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
971 u64 bytenr, u32 blocksize)
973 struct inode *btree_inode = root->fs_info->btree_inode;
974 struct extent_buffer *eb;
976 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
977 bytenr, blocksize, NULL, GFP_NOFS);
982 int btrfs_write_tree_block(struct extent_buffer *buf)
984 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
985 buf->start + buf->len - 1);
988 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
990 return filemap_fdatawait_range(buf->first_page->mapping,
991 buf->start, buf->start + buf->len - 1);
994 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
995 u32 blocksize, u64 parent_transid)
997 struct extent_buffer *buf = NULL;
1000 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1004 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1007 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1012 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1013 struct extent_buffer *buf)
1015 struct inode *btree_inode = root->fs_info->btree_inode;
1016 if (btrfs_header_generation(buf) ==
1017 root->fs_info->running_transaction->transid) {
1018 btrfs_assert_tree_locked(buf);
1020 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1021 spin_lock(&root->fs_info->delalloc_lock);
1022 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1023 root->fs_info->dirty_metadata_bytes -= buf->len;
1026 spin_unlock(&root->fs_info->delalloc_lock);
1029 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1030 btrfs_set_lock_blocking(buf);
1031 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1037 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1038 u32 stripesize, struct btrfs_root *root,
1039 struct btrfs_fs_info *fs_info,
1043 root->commit_root = NULL;
1044 root->sectorsize = sectorsize;
1045 root->nodesize = nodesize;
1046 root->leafsize = leafsize;
1047 root->stripesize = stripesize;
1049 root->track_dirty = 0;
1051 root->orphan_item_inserted = 0;
1052 root->orphan_cleanup_state = 0;
1054 root->fs_info = fs_info;
1055 root->objectid = objectid;
1056 root->last_trans = 0;
1057 root->highest_objectid = 0;
1060 root->inode_tree = RB_ROOT;
1061 root->block_rsv = NULL;
1062 root->orphan_block_rsv = NULL;
1064 INIT_LIST_HEAD(&root->dirty_list);
1065 INIT_LIST_HEAD(&root->orphan_list);
1066 INIT_LIST_HEAD(&root->root_list);
1067 spin_lock_init(&root->node_lock);
1068 spin_lock_init(&root->orphan_lock);
1069 spin_lock_init(&root->inode_lock);
1070 spin_lock_init(&root->accounting_lock);
1071 mutex_init(&root->objectid_mutex);
1072 mutex_init(&root->log_mutex);
1073 init_waitqueue_head(&root->log_writer_wait);
1074 init_waitqueue_head(&root->log_commit_wait[0]);
1075 init_waitqueue_head(&root->log_commit_wait[1]);
1076 atomic_set(&root->log_commit[0], 0);
1077 atomic_set(&root->log_commit[1], 0);
1078 atomic_set(&root->log_writers, 0);
1079 root->log_batch = 0;
1080 root->log_transid = 0;
1081 root->last_log_commit = 0;
1082 extent_io_tree_init(&root->dirty_log_pages,
1083 fs_info->btree_inode->i_mapping, GFP_NOFS);
1085 memset(&root->root_key, 0, sizeof(root->root_key));
1086 memset(&root->root_item, 0, sizeof(root->root_item));
1087 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1088 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1089 root->defrag_trans_start = fs_info->generation;
1090 init_completion(&root->kobj_unregister);
1091 root->defrag_running = 0;
1092 root->root_key.objectid = objectid;
1093 root->anon_super.s_root = NULL;
1094 root->anon_super.s_dev = 0;
1095 INIT_LIST_HEAD(&root->anon_super.s_list);
1096 INIT_LIST_HEAD(&root->anon_super.s_instances);
1097 init_rwsem(&root->anon_super.s_umount);
1102 static int find_and_setup_root(struct btrfs_root *tree_root,
1103 struct btrfs_fs_info *fs_info,
1105 struct btrfs_root *root)
1111 __setup_root(tree_root->nodesize, tree_root->leafsize,
1112 tree_root->sectorsize, tree_root->stripesize,
1113 root, fs_info, objectid);
1114 ret = btrfs_find_last_root(tree_root, objectid,
1115 &root->root_item, &root->root_key);
1120 generation = btrfs_root_generation(&root->root_item);
1121 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1122 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1123 blocksize, generation);
1124 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1125 free_extent_buffer(root->node);
1128 root->commit_root = btrfs_root_node(root);
1132 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1133 struct btrfs_fs_info *fs_info)
1135 struct btrfs_root *root;
1136 struct btrfs_root *tree_root = fs_info->tree_root;
1137 struct extent_buffer *leaf;
1139 root = kzalloc(sizeof(*root), GFP_NOFS);
1141 return ERR_PTR(-ENOMEM);
1143 __setup_root(tree_root->nodesize, tree_root->leafsize,
1144 tree_root->sectorsize, tree_root->stripesize,
1145 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1147 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1148 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1149 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1151 * log trees do not get reference counted because they go away
1152 * before a real commit is actually done. They do store pointers
1153 * to file data extents, and those reference counts still get
1154 * updated (along with back refs to the log tree).
1158 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1159 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1162 return ERR_CAST(leaf);
1165 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1166 btrfs_set_header_bytenr(leaf, leaf->start);
1167 btrfs_set_header_generation(leaf, trans->transid);
1168 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1169 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1172 write_extent_buffer(root->node, root->fs_info->fsid,
1173 (unsigned long)btrfs_header_fsid(root->node),
1175 btrfs_mark_buffer_dirty(root->node);
1176 btrfs_tree_unlock(root->node);
1180 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1181 struct btrfs_fs_info *fs_info)
1183 struct btrfs_root *log_root;
1185 log_root = alloc_log_tree(trans, fs_info);
1186 if (IS_ERR(log_root))
1187 return PTR_ERR(log_root);
1188 WARN_ON(fs_info->log_root_tree);
1189 fs_info->log_root_tree = log_root;
1193 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1194 struct btrfs_root *root)
1196 struct btrfs_root *log_root;
1197 struct btrfs_inode_item *inode_item;
1199 log_root = alloc_log_tree(trans, root->fs_info);
1200 if (IS_ERR(log_root))
1201 return PTR_ERR(log_root);
1203 log_root->last_trans = trans->transid;
1204 log_root->root_key.offset = root->root_key.objectid;
1206 inode_item = &log_root->root_item.inode;
1207 inode_item->generation = cpu_to_le64(1);
1208 inode_item->size = cpu_to_le64(3);
1209 inode_item->nlink = cpu_to_le32(1);
1210 inode_item->nbytes = cpu_to_le64(root->leafsize);
1211 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1213 btrfs_set_root_node(&log_root->root_item, log_root->node);
1215 WARN_ON(root->log_root);
1216 root->log_root = log_root;
1217 root->log_transid = 0;
1218 root->last_log_commit = 0;
1222 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1223 struct btrfs_key *location)
1225 struct btrfs_root *root;
1226 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1227 struct btrfs_path *path;
1228 struct extent_buffer *l;
1233 root = kzalloc(sizeof(*root), GFP_NOFS);
1235 return ERR_PTR(-ENOMEM);
1236 if (location->offset == (u64)-1) {
1237 ret = find_and_setup_root(tree_root, fs_info,
1238 location->objectid, root);
1241 return ERR_PTR(ret);
1246 __setup_root(tree_root->nodesize, tree_root->leafsize,
1247 tree_root->sectorsize, tree_root->stripesize,
1248 root, fs_info, location->objectid);
1250 path = btrfs_alloc_path();
1253 return ERR_PTR(-ENOMEM);
1255 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1258 read_extent_buffer(l, &root->root_item,
1259 btrfs_item_ptr_offset(l, path->slots[0]),
1260 sizeof(root->root_item));
1261 memcpy(&root->root_key, location, sizeof(*location));
1263 btrfs_free_path(path);
1268 return ERR_PTR(ret);
1271 generation = btrfs_root_generation(&root->root_item);
1272 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1273 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1274 blocksize, generation);
1275 root->commit_root = btrfs_root_node(root);
1276 BUG_ON(!root->node);
1278 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1284 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1287 struct btrfs_root *root;
1289 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1290 return fs_info->tree_root;
1291 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1292 return fs_info->extent_root;
1294 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1295 (unsigned long)root_objectid);
1299 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1300 struct btrfs_key *location)
1302 struct btrfs_root *root;
1305 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1306 return fs_info->tree_root;
1307 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1308 return fs_info->extent_root;
1309 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1310 return fs_info->chunk_root;
1311 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1312 return fs_info->dev_root;
1313 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1314 return fs_info->csum_root;
1316 spin_lock(&fs_info->fs_roots_radix_lock);
1317 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1318 (unsigned long)location->objectid);
1319 spin_unlock(&fs_info->fs_roots_radix_lock);
1323 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1327 set_anon_super(&root->anon_super, NULL);
1329 if (btrfs_root_refs(&root->root_item) == 0) {
1334 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1338 root->orphan_item_inserted = 1;
1340 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1344 spin_lock(&fs_info->fs_roots_radix_lock);
1345 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1346 (unsigned long)root->root_key.objectid,
1351 spin_unlock(&fs_info->fs_roots_radix_lock);
1352 radix_tree_preload_end();
1354 if (ret == -EEXIST) {
1361 ret = btrfs_find_dead_roots(fs_info->tree_root,
1362 root->root_key.objectid);
1367 return ERR_PTR(ret);
1370 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1371 struct btrfs_key *location,
1372 const char *name, int namelen)
1374 return btrfs_read_fs_root_no_name(fs_info, location);
1376 struct btrfs_root *root;
1379 root = btrfs_read_fs_root_no_name(fs_info, location);
1386 ret = btrfs_set_root_name(root, name, namelen);
1388 free_extent_buffer(root->node);
1390 return ERR_PTR(ret);
1393 ret = btrfs_sysfs_add_root(root);
1395 free_extent_buffer(root->node);
1398 return ERR_PTR(ret);
1405 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1407 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1409 struct btrfs_device *device;
1410 struct backing_dev_info *bdi;
1412 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1415 bdi = blk_get_backing_dev_info(device->bdev);
1416 if (bdi && bdi_congested(bdi, bdi_bits)) {
1425 * this unplugs every device on the box, and it is only used when page
1428 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1430 struct btrfs_device *device;
1431 struct btrfs_fs_info *info;
1433 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1434 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1438 bdi = blk_get_backing_dev_info(device->bdev);
1439 if (bdi->unplug_io_fn)
1440 bdi->unplug_io_fn(bdi, page);
1444 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1446 struct inode *inode;
1447 struct extent_map_tree *em_tree;
1448 struct extent_map *em;
1449 struct address_space *mapping;
1452 /* the generic O_DIRECT read code does this */
1454 __unplug_io_fn(bdi, page);
1459 * page->mapping may change at any time. Get a consistent copy
1460 * and use that for everything below
1463 mapping = page->mapping;
1467 inode = mapping->host;
1470 * don't do the expensive searching for a small number of
1473 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1474 __unplug_io_fn(bdi, page);
1478 offset = page_offset(page);
1480 em_tree = &BTRFS_I(inode)->extent_tree;
1481 read_lock(&em_tree->lock);
1482 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1483 read_unlock(&em_tree->lock);
1485 __unplug_io_fn(bdi, page);
1489 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1490 free_extent_map(em);
1491 __unplug_io_fn(bdi, page);
1494 offset = offset - em->start;
1495 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1496 em->block_start + offset, page);
1497 free_extent_map(em);
1501 * If this fails, caller must call bdi_destroy() to get rid of the
1504 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1508 bdi->capabilities = BDI_CAP_MAP_COPY;
1509 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1513 bdi->ra_pages = default_backing_dev_info.ra_pages;
1514 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1515 bdi->unplug_io_data = info;
1516 bdi->congested_fn = btrfs_congested_fn;
1517 bdi->congested_data = info;
1521 static int bio_ready_for_csum(struct bio *bio)
1527 struct extent_io_tree *io_tree = NULL;
1528 struct bio_vec *bvec;
1532 bio_for_each_segment(bvec, bio, i) {
1533 page = bvec->bv_page;
1534 if (page->private == EXTENT_PAGE_PRIVATE) {
1535 length += bvec->bv_len;
1538 if (!page->private) {
1539 length += bvec->bv_len;
1542 length = bvec->bv_len;
1543 buf_len = page->private >> 2;
1544 start = page_offset(page) + bvec->bv_offset;
1545 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1547 /* are we fully contained in this bio? */
1548 if (buf_len <= length)
1551 ret = extent_range_uptodate(io_tree, start + length,
1552 start + buf_len - 1);
1557 * called by the kthread helper functions to finally call the bio end_io
1558 * functions. This is where read checksum verification actually happens
1560 static void end_workqueue_fn(struct btrfs_work *work)
1563 struct end_io_wq *end_io_wq;
1564 struct btrfs_fs_info *fs_info;
1567 end_io_wq = container_of(work, struct end_io_wq, work);
1568 bio = end_io_wq->bio;
1569 fs_info = end_io_wq->info;
1571 /* metadata bio reads are special because the whole tree block must
1572 * be checksummed at once. This makes sure the entire block is in
1573 * ram and up to date before trying to verify things. For
1574 * blocksize <= pagesize, it is basically a noop
1576 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1577 !bio_ready_for_csum(bio)) {
1578 btrfs_queue_worker(&fs_info->endio_meta_workers,
1582 error = end_io_wq->error;
1583 bio->bi_private = end_io_wq->private;
1584 bio->bi_end_io = end_io_wq->end_io;
1586 bio_endio(bio, error);
1589 static int cleaner_kthread(void *arg)
1591 struct btrfs_root *root = arg;
1594 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1596 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1597 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1598 btrfs_run_delayed_iputs(root);
1599 btrfs_clean_old_snapshots(root);
1600 mutex_unlock(&root->fs_info->cleaner_mutex);
1603 if (freezing(current)) {
1606 set_current_state(TASK_INTERRUPTIBLE);
1607 if (!kthread_should_stop())
1609 __set_current_state(TASK_RUNNING);
1611 } while (!kthread_should_stop());
1615 static int transaction_kthread(void *arg)
1617 struct btrfs_root *root = arg;
1618 struct btrfs_trans_handle *trans;
1619 struct btrfs_transaction *cur;
1622 unsigned long delay;
1627 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1628 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1630 spin_lock(&root->fs_info->new_trans_lock);
1631 cur = root->fs_info->running_transaction;
1633 spin_unlock(&root->fs_info->new_trans_lock);
1637 now = get_seconds();
1638 if (!cur->blocked &&
1639 (now < cur->start_time || now - cur->start_time < 30)) {
1640 spin_unlock(&root->fs_info->new_trans_lock);
1644 transid = cur->transid;
1645 spin_unlock(&root->fs_info->new_trans_lock);
1647 trans = btrfs_join_transaction(root, 1);
1648 BUG_ON(IS_ERR(trans));
1649 if (transid == trans->transid) {
1650 ret = btrfs_commit_transaction(trans, root);
1653 btrfs_end_transaction(trans, root);
1656 wake_up_process(root->fs_info->cleaner_kthread);
1657 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1659 if (freezing(current)) {
1662 set_current_state(TASK_INTERRUPTIBLE);
1663 if (!kthread_should_stop() &&
1664 !btrfs_transaction_blocked(root->fs_info))
1665 schedule_timeout(delay);
1666 __set_current_state(TASK_RUNNING);
1668 } while (!kthread_should_stop());
1672 struct btrfs_root *open_ctree(struct super_block *sb,
1673 struct btrfs_fs_devices *fs_devices,
1683 struct btrfs_key location;
1684 struct buffer_head *bh;
1685 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1687 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1689 struct btrfs_root *tree_root = btrfs_sb(sb);
1690 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1691 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1693 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1695 struct btrfs_root *log_tree_root;
1700 struct btrfs_super_block *disk_super;
1702 if (!extent_root || !tree_root || !fs_info ||
1703 !chunk_root || !dev_root || !csum_root) {
1708 ret = init_srcu_struct(&fs_info->subvol_srcu);
1714 ret = setup_bdi(fs_info, &fs_info->bdi);
1720 fs_info->btree_inode = new_inode(sb);
1721 if (!fs_info->btree_inode) {
1726 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1727 INIT_LIST_HEAD(&fs_info->trans_list);
1728 INIT_LIST_HEAD(&fs_info->dead_roots);
1729 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1730 INIT_LIST_HEAD(&fs_info->hashers);
1731 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1732 INIT_LIST_HEAD(&fs_info->ordered_operations);
1733 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1734 spin_lock_init(&fs_info->delalloc_lock);
1735 spin_lock_init(&fs_info->new_trans_lock);
1736 spin_lock_init(&fs_info->ref_cache_lock);
1737 spin_lock_init(&fs_info->fs_roots_radix_lock);
1738 spin_lock_init(&fs_info->delayed_iput_lock);
1740 init_completion(&fs_info->kobj_unregister);
1741 fs_info->tree_root = tree_root;
1742 fs_info->extent_root = extent_root;
1743 fs_info->csum_root = csum_root;
1744 fs_info->chunk_root = chunk_root;
1745 fs_info->dev_root = dev_root;
1746 fs_info->fs_devices = fs_devices;
1747 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1748 INIT_LIST_HEAD(&fs_info->space_info);
1749 btrfs_mapping_init(&fs_info->mapping_tree);
1750 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1751 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1752 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1753 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1754 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1755 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1756 mutex_init(&fs_info->durable_block_rsv_mutex);
1757 atomic_set(&fs_info->nr_async_submits, 0);
1758 atomic_set(&fs_info->async_delalloc_pages, 0);
1759 atomic_set(&fs_info->async_submit_draining, 0);
1760 atomic_set(&fs_info->nr_async_bios, 0);
1762 fs_info->max_inline = 8192 * 1024;
1763 fs_info->metadata_ratio = 0;
1765 fs_info->thread_pool_size = min_t(unsigned long,
1766 num_online_cpus() + 2, 8);
1768 INIT_LIST_HEAD(&fs_info->ordered_extents);
1769 spin_lock_init(&fs_info->ordered_extent_lock);
1771 sb->s_blocksize = 4096;
1772 sb->s_blocksize_bits = blksize_bits(4096);
1773 sb->s_bdi = &fs_info->bdi;
1775 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1776 fs_info->btree_inode->i_nlink = 1;
1778 * we set the i_size on the btree inode to the max possible int.
1779 * the real end of the address space is determined by all of
1780 * the devices in the system
1782 fs_info->btree_inode->i_size = OFFSET_MAX;
1783 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1784 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1786 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1787 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1788 fs_info->btree_inode->i_mapping,
1790 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1793 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1795 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1796 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1797 sizeof(struct btrfs_key));
1798 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1799 insert_inode_hash(fs_info->btree_inode);
1801 spin_lock_init(&fs_info->block_group_cache_lock);
1802 fs_info->block_group_cache_tree = RB_ROOT;
1804 extent_io_tree_init(&fs_info->freed_extents[0],
1805 fs_info->btree_inode->i_mapping, GFP_NOFS);
1806 extent_io_tree_init(&fs_info->freed_extents[1],
1807 fs_info->btree_inode->i_mapping, GFP_NOFS);
1808 fs_info->pinned_extents = &fs_info->freed_extents[0];
1809 fs_info->do_barriers = 1;
1812 mutex_init(&fs_info->trans_mutex);
1813 mutex_init(&fs_info->ordered_operations_mutex);
1814 mutex_init(&fs_info->tree_log_mutex);
1815 mutex_init(&fs_info->chunk_mutex);
1816 mutex_init(&fs_info->transaction_kthread_mutex);
1817 mutex_init(&fs_info->cleaner_mutex);
1818 mutex_init(&fs_info->volume_mutex);
1819 init_rwsem(&fs_info->extent_commit_sem);
1820 init_rwsem(&fs_info->cleanup_work_sem);
1821 init_rwsem(&fs_info->subvol_sem);
1823 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1824 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1826 init_waitqueue_head(&fs_info->transaction_throttle);
1827 init_waitqueue_head(&fs_info->transaction_wait);
1828 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1829 init_waitqueue_head(&fs_info->async_submit_wait);
1831 __setup_root(4096, 4096, 4096, 4096, tree_root,
1832 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1834 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1840 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1841 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1842 sizeof(fs_info->super_for_commit));
1845 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1847 disk_super = &fs_info->super_copy;
1848 if (!btrfs_super_root(disk_super))
1851 /* check FS state, whether FS is broken. */
1852 fs_info->fs_state |= btrfs_super_flags(disk_super);
1854 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1856 ret = btrfs_parse_options(tree_root, options);
1862 features = btrfs_super_incompat_flags(disk_super) &
1863 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1865 printk(KERN_ERR "BTRFS: couldn't mount because of "
1866 "unsupported optional features (%Lx).\n",
1867 (unsigned long long)features);
1872 features = btrfs_super_incompat_flags(disk_super);
1873 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1874 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1875 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1876 btrfs_set_super_incompat_flags(disk_super, features);
1878 features = btrfs_super_compat_ro_flags(disk_super) &
1879 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1880 if (!(sb->s_flags & MS_RDONLY) && features) {
1881 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1882 "unsupported option features (%Lx).\n",
1883 (unsigned long long)features);
1888 btrfs_init_workers(&fs_info->generic_worker,
1889 "genwork", 1, NULL);
1891 btrfs_init_workers(&fs_info->workers, "worker",
1892 fs_info->thread_pool_size,
1893 &fs_info->generic_worker);
1895 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1896 fs_info->thread_pool_size,
1897 &fs_info->generic_worker);
1899 btrfs_init_workers(&fs_info->submit_workers, "submit",
1900 min_t(u64, fs_devices->num_devices,
1901 fs_info->thread_pool_size),
1902 &fs_info->generic_worker);
1904 /* a higher idle thresh on the submit workers makes it much more
1905 * likely that bios will be send down in a sane order to the
1908 fs_info->submit_workers.idle_thresh = 64;
1910 fs_info->workers.idle_thresh = 16;
1911 fs_info->workers.ordered = 1;
1913 fs_info->delalloc_workers.idle_thresh = 2;
1914 fs_info->delalloc_workers.ordered = 1;
1916 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1917 &fs_info->generic_worker);
1918 btrfs_init_workers(&fs_info->endio_workers, "endio",
1919 fs_info->thread_pool_size,
1920 &fs_info->generic_worker);
1921 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1922 fs_info->thread_pool_size,
1923 &fs_info->generic_worker);
1924 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1925 "endio-meta-write", fs_info->thread_pool_size,
1926 &fs_info->generic_worker);
1927 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1928 fs_info->thread_pool_size,
1929 &fs_info->generic_worker);
1930 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1931 1, &fs_info->generic_worker);
1934 * endios are largely parallel and should have a very
1937 fs_info->endio_workers.idle_thresh = 4;
1938 fs_info->endio_meta_workers.idle_thresh = 4;
1940 fs_info->endio_write_workers.idle_thresh = 2;
1941 fs_info->endio_meta_write_workers.idle_thresh = 2;
1943 btrfs_start_workers(&fs_info->workers, 1);
1944 btrfs_start_workers(&fs_info->generic_worker, 1);
1945 btrfs_start_workers(&fs_info->submit_workers, 1);
1946 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1947 btrfs_start_workers(&fs_info->fixup_workers, 1);
1948 btrfs_start_workers(&fs_info->endio_workers, 1);
1949 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1950 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1951 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1952 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1954 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1955 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1956 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1958 nodesize = btrfs_super_nodesize(disk_super);
1959 leafsize = btrfs_super_leafsize(disk_super);
1960 sectorsize = btrfs_super_sectorsize(disk_super);
1961 stripesize = btrfs_super_stripesize(disk_super);
1962 tree_root->nodesize = nodesize;
1963 tree_root->leafsize = leafsize;
1964 tree_root->sectorsize = sectorsize;
1965 tree_root->stripesize = stripesize;
1967 sb->s_blocksize = sectorsize;
1968 sb->s_blocksize_bits = blksize_bits(sectorsize);
1970 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1971 sizeof(disk_super->magic))) {
1972 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1973 goto fail_sb_buffer;
1976 mutex_lock(&fs_info->chunk_mutex);
1977 ret = btrfs_read_sys_array(tree_root);
1978 mutex_unlock(&fs_info->chunk_mutex);
1980 printk(KERN_WARNING "btrfs: failed to read the system "
1981 "array on %s\n", sb->s_id);
1982 goto fail_sb_buffer;
1985 blocksize = btrfs_level_size(tree_root,
1986 btrfs_super_chunk_root_level(disk_super));
1987 generation = btrfs_super_chunk_root_generation(disk_super);
1989 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1990 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1992 chunk_root->node = read_tree_block(chunk_root,
1993 btrfs_super_chunk_root(disk_super),
1994 blocksize, generation);
1995 BUG_ON(!chunk_root->node);
1996 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1997 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1999 goto fail_chunk_root;
2001 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2002 chunk_root->commit_root = btrfs_root_node(chunk_root);
2004 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2005 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2008 mutex_lock(&fs_info->chunk_mutex);
2009 ret = btrfs_read_chunk_tree(chunk_root);
2010 mutex_unlock(&fs_info->chunk_mutex);
2012 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2014 goto fail_chunk_root;
2017 btrfs_close_extra_devices(fs_devices);
2019 blocksize = btrfs_level_size(tree_root,
2020 btrfs_super_root_level(disk_super));
2021 generation = btrfs_super_generation(disk_super);
2023 tree_root->node = read_tree_block(tree_root,
2024 btrfs_super_root(disk_super),
2025 blocksize, generation);
2026 if (!tree_root->node)
2027 goto fail_chunk_root;
2028 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2029 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2031 goto fail_tree_root;
2033 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2034 tree_root->commit_root = btrfs_root_node(tree_root);
2036 ret = find_and_setup_root(tree_root, fs_info,
2037 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2039 goto fail_tree_root;
2040 extent_root->track_dirty = 1;
2042 ret = find_and_setup_root(tree_root, fs_info,
2043 BTRFS_DEV_TREE_OBJECTID, dev_root);
2045 goto fail_extent_root;
2046 dev_root->track_dirty = 1;
2048 ret = find_and_setup_root(tree_root, fs_info,
2049 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2053 csum_root->track_dirty = 1;
2055 fs_info->generation = generation;
2056 fs_info->last_trans_committed = generation;
2057 fs_info->data_alloc_profile = (u64)-1;
2058 fs_info->metadata_alloc_profile = (u64)-1;
2059 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2061 ret = btrfs_read_block_groups(extent_root);
2063 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2064 goto fail_block_groups;
2067 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2069 if (IS_ERR(fs_info->cleaner_kthread))
2070 goto fail_block_groups;
2072 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2074 "btrfs-transaction");
2075 if (IS_ERR(fs_info->transaction_kthread))
2078 if (!btrfs_test_opt(tree_root, SSD) &&
2079 !btrfs_test_opt(tree_root, NOSSD) &&
2080 !fs_info->fs_devices->rotating) {
2081 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2083 btrfs_set_opt(fs_info->mount_opt, SSD);
2086 /* do not make disk changes in broken FS */
2087 if (btrfs_super_log_root(disk_super) != 0 &&
2088 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2089 u64 bytenr = btrfs_super_log_root(disk_super);
2091 if (fs_devices->rw_devices == 0) {
2092 printk(KERN_WARNING "Btrfs log replay required "
2095 goto fail_trans_kthread;
2098 btrfs_level_size(tree_root,
2099 btrfs_super_log_root_level(disk_super));
2101 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2102 if (!log_tree_root) {
2104 goto fail_trans_kthread;
2107 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2108 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2110 log_tree_root->node = read_tree_block(tree_root, bytenr,
2113 ret = btrfs_recover_log_trees(log_tree_root);
2116 if (sb->s_flags & MS_RDONLY) {
2117 ret = btrfs_commit_super(tree_root);
2122 ret = btrfs_find_orphan_roots(tree_root);
2125 if (!(sb->s_flags & MS_RDONLY)) {
2126 ret = btrfs_cleanup_fs_roots(fs_info);
2129 ret = btrfs_recover_relocation(tree_root);
2132 "btrfs: failed to recover relocation\n");
2134 goto fail_trans_kthread;
2138 location.objectid = BTRFS_FS_TREE_OBJECTID;
2139 location.type = BTRFS_ROOT_ITEM_KEY;
2140 location.offset = (u64)-1;
2142 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2143 if (!fs_info->fs_root)
2144 goto fail_trans_kthread;
2145 if (IS_ERR(fs_info->fs_root)) {
2146 err = PTR_ERR(fs_info->fs_root);
2147 goto fail_trans_kthread;
2150 if (!(sb->s_flags & MS_RDONLY)) {
2151 down_read(&fs_info->cleanup_work_sem);
2152 err = btrfs_orphan_cleanup(fs_info->fs_root);
2154 err = btrfs_orphan_cleanup(fs_info->tree_root);
2155 up_read(&fs_info->cleanup_work_sem);
2157 close_ctree(tree_root);
2158 return ERR_PTR(err);
2165 kthread_stop(fs_info->transaction_kthread);
2167 kthread_stop(fs_info->cleaner_kthread);
2170 * make sure we're done with the btree inode before we stop our
2173 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2174 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2177 btrfs_free_block_groups(fs_info);
2178 free_extent_buffer(csum_root->node);
2179 free_extent_buffer(csum_root->commit_root);
2181 free_extent_buffer(dev_root->node);
2182 free_extent_buffer(dev_root->commit_root);
2184 free_extent_buffer(extent_root->node);
2185 free_extent_buffer(extent_root->commit_root);
2187 free_extent_buffer(tree_root->node);
2188 free_extent_buffer(tree_root->commit_root);
2190 free_extent_buffer(chunk_root->node);
2191 free_extent_buffer(chunk_root->commit_root);
2193 btrfs_stop_workers(&fs_info->generic_worker);
2194 btrfs_stop_workers(&fs_info->fixup_workers);
2195 btrfs_stop_workers(&fs_info->delalloc_workers);
2196 btrfs_stop_workers(&fs_info->workers);
2197 btrfs_stop_workers(&fs_info->endio_workers);
2198 btrfs_stop_workers(&fs_info->endio_meta_workers);
2199 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2200 btrfs_stop_workers(&fs_info->endio_write_workers);
2201 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2202 btrfs_stop_workers(&fs_info->submit_workers);
2204 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2205 iput(fs_info->btree_inode);
2207 btrfs_close_devices(fs_info->fs_devices);
2208 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2210 bdi_destroy(&fs_info->bdi);
2212 cleanup_srcu_struct(&fs_info->subvol_srcu);
2220 return ERR_PTR(err);
2223 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2225 char b[BDEVNAME_SIZE];
2228 set_buffer_uptodate(bh);
2230 if (printk_ratelimit()) {
2231 printk(KERN_WARNING "lost page write due to "
2232 "I/O error on %s\n",
2233 bdevname(bh->b_bdev, b));
2235 /* note, we dont' set_buffer_write_io_error because we have
2236 * our own ways of dealing with the IO errors
2238 clear_buffer_uptodate(bh);
2244 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2246 struct buffer_head *bh;
2247 struct buffer_head *latest = NULL;
2248 struct btrfs_super_block *super;
2253 /* we would like to check all the supers, but that would make
2254 * a btrfs mount succeed after a mkfs from a different FS.
2255 * So, we need to add a special mount option to scan for
2256 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2258 for (i = 0; i < 1; i++) {
2259 bytenr = btrfs_sb_offset(i);
2260 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2262 bh = __bread(bdev, bytenr / 4096, 4096);
2266 super = (struct btrfs_super_block *)bh->b_data;
2267 if (btrfs_super_bytenr(super) != bytenr ||
2268 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2269 sizeof(super->magic))) {
2274 if (!latest || btrfs_super_generation(super) > transid) {
2277 transid = btrfs_super_generation(super);
2286 * this should be called twice, once with wait == 0 and
2287 * once with wait == 1. When wait == 0 is done, all the buffer heads
2288 * we write are pinned.
2290 * They are released when wait == 1 is done.
2291 * max_mirrors must be the same for both runs, and it indicates how
2292 * many supers on this one device should be written.
2294 * max_mirrors == 0 means to write them all.
2296 static int write_dev_supers(struct btrfs_device *device,
2297 struct btrfs_super_block *sb,
2298 int do_barriers, int wait, int max_mirrors)
2300 struct buffer_head *bh;
2306 int last_barrier = 0;
2308 if (max_mirrors == 0)
2309 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2311 /* make sure only the last submit_bh does a barrier */
2313 for (i = 0; i < max_mirrors; i++) {
2314 bytenr = btrfs_sb_offset(i);
2315 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2316 device->total_bytes)
2322 for (i = 0; i < max_mirrors; i++) {
2323 bytenr = btrfs_sb_offset(i);
2324 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2328 bh = __find_get_block(device->bdev, bytenr / 4096,
2329 BTRFS_SUPER_INFO_SIZE);
2332 if (!buffer_uptodate(bh))
2335 /* drop our reference */
2338 /* drop the reference from the wait == 0 run */
2342 btrfs_set_super_bytenr(sb, bytenr);
2345 crc = btrfs_csum_data(NULL, (char *)sb +
2346 BTRFS_CSUM_SIZE, crc,
2347 BTRFS_SUPER_INFO_SIZE -
2349 btrfs_csum_final(crc, sb->csum);
2352 * one reference for us, and we leave it for the
2355 bh = __getblk(device->bdev, bytenr / 4096,
2356 BTRFS_SUPER_INFO_SIZE);
2357 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2359 /* one reference for submit_bh */
2362 set_buffer_uptodate(bh);
2364 bh->b_end_io = btrfs_end_buffer_write_sync;
2367 if (i == last_barrier && do_barriers)
2368 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2370 ret = submit_bh(WRITE_SYNC, bh);
2375 return errors < i ? 0 : -1;
2378 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2380 struct list_head *head;
2381 struct btrfs_device *dev;
2382 struct btrfs_super_block *sb;
2383 struct btrfs_dev_item *dev_item;
2387 int total_errors = 0;
2390 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2391 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2393 sb = &root->fs_info->super_for_commit;
2394 dev_item = &sb->dev_item;
2396 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2397 head = &root->fs_info->fs_devices->devices;
2398 list_for_each_entry(dev, head, dev_list) {
2403 if (!dev->in_fs_metadata || !dev->writeable)
2406 btrfs_set_stack_device_generation(dev_item, 0);
2407 btrfs_set_stack_device_type(dev_item, dev->type);
2408 btrfs_set_stack_device_id(dev_item, dev->devid);
2409 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2410 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2411 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2412 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2413 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2414 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2415 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2417 flags = btrfs_super_flags(sb);
2418 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2420 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2424 if (total_errors > max_errors) {
2425 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2431 list_for_each_entry(dev, head, dev_list) {
2434 if (!dev->in_fs_metadata || !dev->writeable)
2437 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2441 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2442 if (total_errors > max_errors) {
2443 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2450 int write_ctree_super(struct btrfs_trans_handle *trans,
2451 struct btrfs_root *root, int max_mirrors)
2455 ret = write_all_supers(root, max_mirrors);
2459 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2461 spin_lock(&fs_info->fs_roots_radix_lock);
2462 radix_tree_delete(&fs_info->fs_roots_radix,
2463 (unsigned long)root->root_key.objectid);
2464 spin_unlock(&fs_info->fs_roots_radix_lock);
2466 if (btrfs_root_refs(&root->root_item) == 0)
2467 synchronize_srcu(&fs_info->subvol_srcu);
2473 static void free_fs_root(struct btrfs_root *root)
2475 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2476 if (root->anon_super.s_dev) {
2477 down_write(&root->anon_super.s_umount);
2478 kill_anon_super(&root->anon_super);
2480 free_extent_buffer(root->node);
2481 free_extent_buffer(root->commit_root);
2486 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2489 struct btrfs_root *gang[8];
2492 while (!list_empty(&fs_info->dead_roots)) {
2493 gang[0] = list_entry(fs_info->dead_roots.next,
2494 struct btrfs_root, root_list);
2495 list_del(&gang[0]->root_list);
2497 if (gang[0]->in_radix) {
2498 btrfs_free_fs_root(fs_info, gang[0]);
2500 free_extent_buffer(gang[0]->node);
2501 free_extent_buffer(gang[0]->commit_root);
2507 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2512 for (i = 0; i < ret; i++)
2513 btrfs_free_fs_root(fs_info, gang[i]);
2518 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2520 u64 root_objectid = 0;
2521 struct btrfs_root *gang[8];
2526 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2527 (void **)gang, root_objectid,
2532 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2533 for (i = 0; i < ret; i++) {
2536 root_objectid = gang[i]->root_key.objectid;
2537 err = btrfs_orphan_cleanup(gang[i]);
2546 int btrfs_commit_super(struct btrfs_root *root)
2548 struct btrfs_trans_handle *trans;
2551 mutex_lock(&root->fs_info->cleaner_mutex);
2552 btrfs_run_delayed_iputs(root);
2553 btrfs_clean_old_snapshots(root);
2554 mutex_unlock(&root->fs_info->cleaner_mutex);
2556 /* wait until ongoing cleanup work done */
2557 down_write(&root->fs_info->cleanup_work_sem);
2558 up_write(&root->fs_info->cleanup_work_sem);
2560 trans = btrfs_join_transaction(root, 1);
2562 return PTR_ERR(trans);
2563 ret = btrfs_commit_transaction(trans, root);
2565 /* run commit again to drop the original snapshot */
2566 trans = btrfs_join_transaction(root, 1);
2568 return PTR_ERR(trans);
2569 btrfs_commit_transaction(trans, root);
2570 ret = btrfs_write_and_wait_transaction(NULL, root);
2573 ret = write_ctree_super(NULL, root, 0);
2577 int close_ctree(struct btrfs_root *root)
2579 struct btrfs_fs_info *fs_info = root->fs_info;
2582 fs_info->closing = 1;
2585 btrfs_put_block_group_cache(fs_info);
2588 * Here come 2 situations when btrfs is broken to flip readonly:
2590 * 1. when btrfs flips readonly somewhere else before
2591 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2592 * and btrfs will skip to write sb directly to keep
2593 * ERROR state on disk.
2595 * 2. when btrfs flips readonly just in btrfs_commit_super,
2596 * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2597 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2598 * btrfs will cleanup all FS resources first and write sb then.
2600 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2601 ret = btrfs_commit_super(root);
2603 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2606 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2607 ret = btrfs_error_commit_super(root);
2609 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2612 kthread_stop(root->fs_info->transaction_kthread);
2613 kthread_stop(root->fs_info->cleaner_kthread);
2615 fs_info->closing = 2;
2618 if (fs_info->delalloc_bytes) {
2619 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2620 (unsigned long long)fs_info->delalloc_bytes);
2622 if (fs_info->total_ref_cache_size) {
2623 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2624 (unsigned long long)fs_info->total_ref_cache_size);
2627 free_extent_buffer(fs_info->extent_root->node);
2628 free_extent_buffer(fs_info->extent_root->commit_root);
2629 free_extent_buffer(fs_info->tree_root->node);
2630 free_extent_buffer(fs_info->tree_root->commit_root);
2631 free_extent_buffer(root->fs_info->chunk_root->node);
2632 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2633 free_extent_buffer(root->fs_info->dev_root->node);
2634 free_extent_buffer(root->fs_info->dev_root->commit_root);
2635 free_extent_buffer(root->fs_info->csum_root->node);
2636 free_extent_buffer(root->fs_info->csum_root->commit_root);
2638 btrfs_free_block_groups(root->fs_info);
2640 del_fs_roots(fs_info);
2642 iput(fs_info->btree_inode);
2644 btrfs_stop_workers(&fs_info->generic_worker);
2645 btrfs_stop_workers(&fs_info->fixup_workers);
2646 btrfs_stop_workers(&fs_info->delalloc_workers);
2647 btrfs_stop_workers(&fs_info->workers);
2648 btrfs_stop_workers(&fs_info->endio_workers);
2649 btrfs_stop_workers(&fs_info->endio_meta_workers);
2650 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2651 btrfs_stop_workers(&fs_info->endio_write_workers);
2652 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2653 btrfs_stop_workers(&fs_info->submit_workers);
2655 btrfs_close_devices(fs_info->fs_devices);
2656 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2658 bdi_destroy(&fs_info->bdi);
2659 cleanup_srcu_struct(&fs_info->subvol_srcu);
2661 kfree(fs_info->extent_root);
2662 kfree(fs_info->tree_root);
2663 kfree(fs_info->chunk_root);
2664 kfree(fs_info->dev_root);
2665 kfree(fs_info->csum_root);
2671 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2674 struct inode *btree_inode = buf->first_page->mapping->host;
2676 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2681 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2686 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2688 struct inode *btree_inode = buf->first_page->mapping->host;
2689 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2693 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2695 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2696 u64 transid = btrfs_header_generation(buf);
2697 struct inode *btree_inode = root->fs_info->btree_inode;
2700 btrfs_assert_tree_locked(buf);
2701 if (transid != root->fs_info->generation) {
2702 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2703 "found %llu running %llu\n",
2704 (unsigned long long)buf->start,
2705 (unsigned long long)transid,
2706 (unsigned long long)root->fs_info->generation);
2709 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2712 spin_lock(&root->fs_info->delalloc_lock);
2713 root->fs_info->dirty_metadata_bytes += buf->len;
2714 spin_unlock(&root->fs_info->delalloc_lock);
2718 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2721 * looks as though older kernels can get into trouble with
2722 * this code, they end up stuck in balance_dirty_pages forever
2725 unsigned long thresh = 32 * 1024 * 1024;
2727 if (current->flags & PF_MEMALLOC)
2730 num_dirty = root->fs_info->dirty_metadata_bytes;
2732 if (num_dirty > thresh) {
2733 balance_dirty_pages_ratelimited_nr(
2734 root->fs_info->btree_inode->i_mapping, 1);
2739 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2741 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2743 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2745 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2749 int btree_lock_page_hook(struct page *page)
2751 struct inode *inode = page->mapping->host;
2752 struct btrfs_root *root = BTRFS_I(inode)->root;
2753 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2754 struct extent_buffer *eb;
2756 u64 bytenr = page_offset(page);
2758 if (page->private == EXTENT_PAGE_PRIVATE)
2761 len = page->private >> 2;
2762 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2766 btrfs_tree_lock(eb);
2767 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2769 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2770 spin_lock(&root->fs_info->delalloc_lock);
2771 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2772 root->fs_info->dirty_metadata_bytes -= eb->len;
2775 spin_unlock(&root->fs_info->delalloc_lock);
2778 btrfs_tree_unlock(eb);
2779 free_extent_buffer(eb);
2785 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2791 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2792 printk(KERN_WARNING "warning: mount fs with errors, "
2793 "running btrfsck is recommended\n");
2796 int btrfs_error_commit_super(struct btrfs_root *root)
2800 mutex_lock(&root->fs_info->cleaner_mutex);
2801 btrfs_run_delayed_iputs(root);
2802 mutex_unlock(&root->fs_info->cleaner_mutex);
2804 down_write(&root->fs_info->cleanup_work_sem);
2805 up_write(&root->fs_info->cleanup_work_sem);
2807 /* cleanup FS via transaction */
2808 btrfs_cleanup_transaction(root);
2810 ret = write_ctree_super(NULL, root, 0);
2815 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2817 struct btrfs_inode *btrfs_inode;
2818 struct list_head splice;
2820 INIT_LIST_HEAD(&splice);
2822 mutex_lock(&root->fs_info->ordered_operations_mutex);
2823 spin_lock(&root->fs_info->ordered_extent_lock);
2825 list_splice_init(&root->fs_info->ordered_operations, &splice);
2826 while (!list_empty(&splice)) {
2827 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2828 ordered_operations);
2830 list_del_init(&btrfs_inode->ordered_operations);
2832 btrfs_invalidate_inodes(btrfs_inode->root);
2835 spin_unlock(&root->fs_info->ordered_extent_lock);
2836 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2841 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2843 struct list_head splice;
2844 struct btrfs_ordered_extent *ordered;
2845 struct inode *inode;
2847 INIT_LIST_HEAD(&splice);
2849 spin_lock(&root->fs_info->ordered_extent_lock);
2851 list_splice_init(&root->fs_info->ordered_extents, &splice);
2852 while (!list_empty(&splice)) {
2853 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2856 list_del_init(&ordered->root_extent_list);
2857 atomic_inc(&ordered->refs);
2859 /* the inode may be getting freed (in sys_unlink path). */
2860 inode = igrab(ordered->inode);
2862 spin_unlock(&root->fs_info->ordered_extent_lock);
2866 atomic_set(&ordered->refs, 1);
2867 btrfs_put_ordered_extent(ordered);
2869 spin_lock(&root->fs_info->ordered_extent_lock);
2872 spin_unlock(&root->fs_info->ordered_extent_lock);
2877 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2878 struct btrfs_root *root)
2880 struct rb_node *node;
2881 struct btrfs_delayed_ref_root *delayed_refs;
2882 struct btrfs_delayed_ref_node *ref;
2885 delayed_refs = &trans->delayed_refs;
2887 spin_lock(&delayed_refs->lock);
2888 if (delayed_refs->num_entries == 0) {
2889 printk(KERN_INFO "delayed_refs has NO entry\n");
2893 node = rb_first(&delayed_refs->root);
2895 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2896 node = rb_next(node);
2899 rb_erase(&ref->rb_node, &delayed_refs->root);
2900 delayed_refs->num_entries--;
2902 atomic_set(&ref->refs, 1);
2903 if (btrfs_delayed_ref_is_head(ref)) {
2904 struct btrfs_delayed_ref_head *head;
2906 head = btrfs_delayed_node_to_head(ref);
2907 mutex_lock(&head->mutex);
2908 kfree(head->extent_op);
2909 delayed_refs->num_heads--;
2910 if (list_empty(&head->cluster))
2911 delayed_refs->num_heads_ready--;
2912 list_del_init(&head->cluster);
2913 mutex_unlock(&head->mutex);
2916 spin_unlock(&delayed_refs->lock);
2917 btrfs_put_delayed_ref(ref);
2920 spin_lock(&delayed_refs->lock);
2923 spin_unlock(&delayed_refs->lock);
2928 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2930 struct btrfs_pending_snapshot *snapshot;
2931 struct list_head splice;
2933 INIT_LIST_HEAD(&splice);
2935 list_splice_init(&t->pending_snapshots, &splice);
2937 while (!list_empty(&splice)) {
2938 snapshot = list_entry(splice.next,
2939 struct btrfs_pending_snapshot,
2942 list_del_init(&snapshot->list);
2950 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2952 struct btrfs_inode *btrfs_inode;
2953 struct list_head splice;
2955 INIT_LIST_HEAD(&splice);
2957 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2959 spin_lock(&root->fs_info->delalloc_lock);
2961 while (!list_empty(&splice)) {
2962 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2965 list_del_init(&btrfs_inode->delalloc_inodes);
2967 btrfs_invalidate_inodes(btrfs_inode->root);
2970 spin_unlock(&root->fs_info->delalloc_lock);
2975 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2976 struct extent_io_tree *dirty_pages,
2981 struct inode *btree_inode = root->fs_info->btree_inode;
2982 struct extent_buffer *eb;
2986 unsigned long index;
2989 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2994 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2995 while (start <= end) {
2996 index = start >> PAGE_CACHE_SHIFT;
2997 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2998 page = find_get_page(btree_inode->i_mapping, index);
3001 offset = page_offset(page);
3003 spin_lock(&dirty_pages->buffer_lock);
3004 eb = radix_tree_lookup(
3005 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3006 offset >> PAGE_CACHE_SHIFT);
3007 spin_unlock(&dirty_pages->buffer_lock);
3009 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3011 atomic_set(&eb->refs, 1);
3013 if (PageWriteback(page))
3014 end_page_writeback(page);
3017 if (PageDirty(page)) {
3018 clear_page_dirty_for_io(page);
3019 spin_lock_irq(&page->mapping->tree_lock);
3020 radix_tree_tag_clear(&page->mapping->page_tree,
3022 PAGECACHE_TAG_DIRTY);
3023 spin_unlock_irq(&page->mapping->tree_lock);
3026 page->mapping->a_ops->invalidatepage(page, 0);
3034 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3035 struct extent_io_tree *pinned_extents)
3037 struct extent_io_tree *unpin;
3042 unpin = pinned_extents;
3044 ret = find_first_extent_bit(unpin, 0, &start, &end,
3050 ret = btrfs_error_discard_extent(root, start, end + 1 - start);
3052 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3053 btrfs_error_unpin_extent_range(root, start, end);
3060 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3062 struct btrfs_transaction *t;
3067 mutex_lock(&root->fs_info->trans_mutex);
3068 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3070 list_splice_init(&root->fs_info->trans_list, &list);
3071 while (!list_empty(&list)) {
3072 t = list_entry(list.next, struct btrfs_transaction, list);
3076 btrfs_destroy_ordered_operations(root);
3078 btrfs_destroy_ordered_extents(root);
3080 btrfs_destroy_delayed_refs(t, root);
3082 btrfs_block_rsv_release(root,
3083 &root->fs_info->trans_block_rsv,
3084 t->dirty_pages.dirty_bytes);
3086 /* FIXME: cleanup wait for commit */
3089 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3090 wake_up(&root->fs_info->transaction_blocked_wait);
3093 if (waitqueue_active(&root->fs_info->transaction_wait))
3094 wake_up(&root->fs_info->transaction_wait);
3095 mutex_unlock(&root->fs_info->trans_mutex);
3097 mutex_lock(&root->fs_info->trans_mutex);
3099 if (waitqueue_active(&t->commit_wait))
3100 wake_up(&t->commit_wait);
3101 mutex_unlock(&root->fs_info->trans_mutex);
3103 mutex_lock(&root->fs_info->trans_mutex);
3105 btrfs_destroy_pending_snapshots(t);
3107 btrfs_destroy_delalloc_inodes(root);
3109 spin_lock(&root->fs_info->new_trans_lock);
3110 root->fs_info->running_transaction = NULL;
3111 spin_unlock(&root->fs_info->new_trans_lock);
3113 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3116 btrfs_destroy_pinned_extent(root,
3117 root->fs_info->pinned_extents);
3120 list_del_init(&t->list);
3121 memset(t, 0, sizeof(*t));
3122 kmem_cache_free(btrfs_transaction_cachep, t);
3125 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3126 mutex_unlock(&root->fs_info->trans_mutex);
3131 static struct extent_io_ops btree_extent_io_ops = {
3132 .write_cache_pages_lock_hook = btree_lock_page_hook,
3133 .readpage_end_io_hook = btree_readpage_end_io_hook,
3134 .submit_bio_hook = btree_submit_bio_hook,
3135 /* note we're sharing with inode.c for the merge bio hook */
3136 .merge_bio_hook = btrfs_merge_bio_hook,