2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
48 static struct extent_io_ops btree_extent_io_ops;
49 static void end_workqueue_fn(struct btrfs_work *work);
50 static void free_fs_root(struct btrfs_root *root);
51 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
53 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
54 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_root *root);
57 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
58 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
60 struct extent_io_tree *dirty_pages,
62 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
63 struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_root *root);
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
75 struct btrfs_fs_info *info;
78 struct list_head list;
79 struct btrfs_work work;
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
87 struct async_submit_bio {
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
95 unsigned long bio_flags;
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
101 struct btrfs_work work;
105 * Lockdep class keys for extent_buffer->lock's in this root. For a given
106 * eb, the lockdep key is determined by the btrfs_root it belongs to and
107 * the level the eb occupies in the tree.
109 * Different roots are used for different purposes and may nest inside each
110 * other and they require separate keysets. As lockdep keys should be
111 * static, assign keysets according to the purpose of the root as indicated
112 * by btrfs_root->objectid. This ensures that all special purpose roots
113 * have separate keysets.
115 * Lock-nesting across peer nodes is always done with the immediate parent
116 * node locked thus preventing deadlock. As lockdep doesn't know this, use
117 * subclass to avoid triggering lockdep warning in such cases.
119 * The key is set by the readpage_end_io_hook after the buffer has passed
120 * csum validation but before the pages are unlocked. It is also set by
121 * btrfs_init_new_buffer on freshly allocated blocks.
123 * We also add a check to make sure the highest level of the tree is the
124 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
125 * needs update as well.
127 #ifdef CONFIG_DEBUG_LOCK_ALLOC
128 # if BTRFS_MAX_LEVEL != 8
132 static struct btrfs_lockdep_keyset {
133 u64 id; /* root objectid */
134 const char *name_stem; /* lock name stem */
135 char names[BTRFS_MAX_LEVEL + 1][20];
136 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
137 } btrfs_lockdep_keysets[] = {
138 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
139 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
140 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
141 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
142 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
143 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
144 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
145 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
146 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
147 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
148 { .id = 0, .name_stem = "tree" },
151 void __init btrfs_init_lockdep(void)
155 /* initialize lockdep class names */
156 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
157 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
159 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
160 snprintf(ks->names[j], sizeof(ks->names[j]),
161 "btrfs-%s-%02d", ks->name_stem, j);
165 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
168 struct btrfs_lockdep_keyset *ks;
170 BUG_ON(level >= ARRAY_SIZE(ks->keys));
172 /* find the matching keyset, id 0 is the default entry */
173 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
174 if (ks->id == objectid)
177 lockdep_set_class_and_name(&eb->lock,
178 &ks->keys[level], ks->names[level]);
184 * extents on the btree inode are pretty simple, there's one extent
185 * that covers the entire device
187 static struct extent_map *btree_get_extent(struct inode *inode,
188 struct page *page, size_t pg_offset, u64 start, u64 len,
191 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
192 struct extent_map *em;
195 read_lock(&em_tree->lock);
196 em = lookup_extent_mapping(em_tree, start, len);
199 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
200 read_unlock(&em_tree->lock);
203 read_unlock(&em_tree->lock);
205 em = alloc_extent_map();
207 em = ERR_PTR(-ENOMEM);
212 em->block_len = (u64)-1;
214 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
216 write_lock(&em_tree->lock);
217 ret = add_extent_mapping(em_tree, em);
218 if (ret == -EEXIST) {
219 u64 failed_start = em->start;
220 u64 failed_len = em->len;
223 em = lookup_extent_mapping(em_tree, start, len);
227 em = lookup_extent_mapping(em_tree, failed_start,
235 write_unlock(&em_tree->lock);
243 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
245 return crc32c(seed, data, len);
248 void btrfs_csum_final(u32 crc, char *result)
250 put_unaligned_le32(~crc, result);
254 * compute the csum for a btree block, and either verify it or write it
255 * into the csum field of the block.
257 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263 unsigned long cur_len;
264 unsigned long offset = BTRFS_CSUM_SIZE;
266 unsigned long map_start;
267 unsigned long map_len;
270 unsigned long inline_result;
272 len = buf->len - offset;
274 err = map_private_extent_buffer(buf, offset, 32,
275 &kaddr, &map_start, &map_len);
278 cur_len = min(len, map_len - (offset - map_start));
279 crc = btrfs_csum_data(root, kaddr + offset - map_start,
284 if (csum_size > sizeof(inline_result)) {
285 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 result = (char *)&inline_result;
292 btrfs_csum_final(crc, result);
295 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 memcpy(&found, result, csum_size);
300 read_extent_buffer(buf, &val, 0, csum_size);
301 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
302 "failed on %llu wanted %X found %X "
304 root->fs_info->sb->s_id,
305 (unsigned long long)buf->start, val, found,
306 btrfs_header_level(buf));
307 if (result != (char *)&inline_result)
312 write_extent_buffer(buf, result, 0, csum_size);
314 if (result != (char *)&inline_result)
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
325 static int verify_parent_transid(struct extent_io_tree *io_tree,
326 struct extent_buffer *eb, u64 parent_transid)
328 struct extent_state *cached_state = NULL;
331 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
335 0, &cached_state, GFP_NOFS);
336 if (extent_buffer_uptodate(eb) &&
337 btrfs_header_generation(eb) == parent_transid) {
341 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
343 (unsigned long long)eb->start,
344 (unsigned long long)parent_transid,
345 (unsigned long long)btrfs_header_generation(eb));
347 clear_extent_buffer_uptodate(eb);
349 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350 &cached_state, GFP_NOFS);
355 * helper to read a given tree block, doing retries as required when
356 * the checksums don't match and we have alternate mirrors to try.
358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359 struct extent_buffer *eb,
360 u64 start, u64 parent_transid)
362 struct extent_io_tree *io_tree;
367 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
368 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
370 ret = read_extent_buffer_pages(io_tree, eb, start,
372 btree_get_extent, mirror_num);
373 if (!ret && !verify_parent_transid(io_tree, eb, parent_transid))
377 * This buffer's crc is fine, but its contents are corrupted, so
378 * there is no reason to read the other copies, they won't be
381 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
384 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
390 if (mirror_num > num_copies)
397 * checksum a dirty tree block before IO. This has extra checks to make sure
398 * we only fill in the checksum field in the first page of a multi-page block
401 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
403 struct extent_io_tree *tree;
404 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
406 struct extent_buffer *eb;
408 tree = &BTRFS_I(page->mapping->host)->io_tree;
410 eb = (struct extent_buffer *)page->private;
411 if (page != eb->pages[0])
414 found_start = btrfs_header_bytenr(eb);
415 if (found_start != start) {
419 if (eb->pages[0] != page) {
423 if (!PageUptodate(page)) {
427 csum_tree_block(root, eb, 0);
431 static int check_tree_block_fsid(struct btrfs_root *root,
432 struct extent_buffer *eb)
434 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
435 u8 fsid[BTRFS_UUID_SIZE];
438 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
441 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
445 fs_devices = fs_devices->seed;
450 #define CORRUPT(reason, eb, root, slot) \
451 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
452 "root=%llu, slot=%d\n", reason, \
453 (unsigned long long)btrfs_header_bytenr(eb), \
454 (unsigned long long)root->objectid, slot)
456 static noinline int check_leaf(struct btrfs_root *root,
457 struct extent_buffer *leaf)
459 struct btrfs_key key;
460 struct btrfs_key leaf_key;
461 u32 nritems = btrfs_header_nritems(leaf);
467 /* Check the 0 item */
468 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
469 BTRFS_LEAF_DATA_SIZE(root)) {
470 CORRUPT("invalid item offset size pair", leaf, root, 0);
475 * Check to make sure each items keys are in the correct order and their
476 * offsets make sense. We only have to loop through nritems-1 because
477 * we check the current slot against the next slot, which verifies the
478 * next slot's offset+size makes sense and that the current's slot
481 for (slot = 0; slot < nritems - 1; slot++) {
482 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
483 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
485 /* Make sure the keys are in the right order */
486 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
487 CORRUPT("bad key order", leaf, root, slot);
492 * Make sure the offset and ends are right, remember that the
493 * item data starts at the end of the leaf and grows towards the
496 if (btrfs_item_offset_nr(leaf, slot) !=
497 btrfs_item_end_nr(leaf, slot + 1)) {
498 CORRUPT("slot offset bad", leaf, root, slot);
503 * Check to make sure that we don't point outside of the leaf,
504 * just incase all the items are consistent to eachother, but
505 * all point outside of the leaf.
507 if (btrfs_item_end_nr(leaf, slot) >
508 BTRFS_LEAF_DATA_SIZE(root)) {
509 CORRUPT("slot end outside of leaf", leaf, root, slot);
517 struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
518 struct page *page, int max_walk)
520 struct extent_buffer *eb;
521 u64 start = page_offset(page);
525 if (start < max_walk)
528 min_start = start - max_walk;
530 while (start >= min_start) {
531 eb = find_extent_buffer(tree, start, 0);
534 * we found an extent buffer and it contains our page
537 if (eb->start <= target &&
538 eb->start + eb->len > target)
541 /* we found an extent buffer that wasn't for us */
542 free_extent_buffer(eb);
547 start -= PAGE_CACHE_SIZE;
552 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
553 struct extent_state *state)
555 struct extent_io_tree *tree;
558 struct extent_buffer *eb;
559 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
566 tree = &BTRFS_I(page->mapping->host)->io_tree;
567 eb = (struct extent_buffer *)page->private;
569 /* the pending IO might have been the only thing that kept this buffer
570 * in memory. Make sure we have a ref for all this other checks
572 extent_buffer_get(eb);
574 reads_done = atomic_dec_and_test(&eb->io_pages);
578 found_start = btrfs_header_bytenr(eb);
579 if (found_start != eb->start) {
580 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
582 (unsigned long long)found_start,
583 (unsigned long long)eb->start);
587 if (check_tree_block_fsid(root, eb)) {
588 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
589 (unsigned long long)eb->start);
593 found_level = btrfs_header_level(eb);
595 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
598 ret = csum_tree_block(root, eb, 1);
605 * If this is a leaf block and it is corrupt, set the corrupt bit so
606 * that we don't try and read the other copies of this block, just
609 if (found_level == 0 && check_leaf(root, eb)) {
610 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
615 set_extent_buffer_uptodate(eb);
617 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
618 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
619 btree_readahead_hook(root, eb, eb->start, ret);
623 clear_extent_buffer_uptodate(eb);
624 free_extent_buffer(eb);
629 static int btree_io_failed_hook(struct bio *failed_bio,
630 struct page *page, u64 start, u64 end,
631 int mirror_num, struct extent_state *state)
633 struct extent_buffer *eb;
634 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
636 eb = (struct extent_buffer *)page->private;
637 if (page != eb->pages[0])
640 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
641 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
642 btree_readahead_hook(root, eb, eb->start, -EIO);
644 return -EIO; /* we fixed nothing */
647 static void end_workqueue_bio(struct bio *bio, int err)
649 struct end_io_wq *end_io_wq = bio->bi_private;
650 struct btrfs_fs_info *fs_info;
652 fs_info = end_io_wq->info;
653 end_io_wq->error = err;
654 end_io_wq->work.func = end_workqueue_fn;
655 end_io_wq->work.flags = 0;
657 if (bio->bi_rw & REQ_WRITE) {
658 if (end_io_wq->metadata == 1)
659 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
661 else if (end_io_wq->metadata == 2)
662 btrfs_queue_worker(&fs_info->endio_freespace_worker,
665 btrfs_queue_worker(&fs_info->endio_write_workers,
668 if (end_io_wq->metadata)
669 btrfs_queue_worker(&fs_info->endio_meta_workers,
672 btrfs_queue_worker(&fs_info->endio_workers,
678 * For the metadata arg you want
681 * 1 - if normal metadta
682 * 2 - if writing to the free space cache area
684 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
687 struct end_io_wq *end_io_wq;
688 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
692 end_io_wq->private = bio->bi_private;
693 end_io_wq->end_io = bio->bi_end_io;
694 end_io_wq->info = info;
695 end_io_wq->error = 0;
696 end_io_wq->bio = bio;
697 end_io_wq->metadata = metadata;
699 bio->bi_private = end_io_wq;
700 bio->bi_end_io = end_workqueue_bio;
704 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
706 unsigned long limit = min_t(unsigned long,
707 info->workers.max_workers,
708 info->fs_devices->open_devices);
712 static void run_one_async_start(struct btrfs_work *work)
714 struct async_submit_bio *async;
716 async = container_of(work, struct async_submit_bio, work);
717 async->submit_bio_start(async->inode, async->rw, async->bio,
718 async->mirror_num, async->bio_flags,
722 static void run_one_async_done(struct btrfs_work *work)
724 struct btrfs_fs_info *fs_info;
725 struct async_submit_bio *async;
728 async = container_of(work, struct async_submit_bio, work);
729 fs_info = BTRFS_I(async->inode)->root->fs_info;
731 limit = btrfs_async_submit_limit(fs_info);
732 limit = limit * 2 / 3;
734 atomic_dec(&fs_info->nr_async_submits);
736 if (atomic_read(&fs_info->nr_async_submits) < limit &&
737 waitqueue_active(&fs_info->async_submit_wait))
738 wake_up(&fs_info->async_submit_wait);
740 async->submit_bio_done(async->inode, async->rw, async->bio,
741 async->mirror_num, async->bio_flags,
745 static void run_one_async_free(struct btrfs_work *work)
747 struct async_submit_bio *async;
749 async = container_of(work, struct async_submit_bio, work);
753 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
754 int rw, struct bio *bio, int mirror_num,
755 unsigned long bio_flags,
757 extent_submit_bio_hook_t *submit_bio_start,
758 extent_submit_bio_hook_t *submit_bio_done)
760 struct async_submit_bio *async;
762 async = kmalloc(sizeof(*async), GFP_NOFS);
766 async->inode = inode;
769 async->mirror_num = mirror_num;
770 async->submit_bio_start = submit_bio_start;
771 async->submit_bio_done = submit_bio_done;
773 async->work.func = run_one_async_start;
774 async->work.ordered_func = run_one_async_done;
775 async->work.ordered_free = run_one_async_free;
777 async->work.flags = 0;
778 async->bio_flags = bio_flags;
779 async->bio_offset = bio_offset;
781 atomic_inc(&fs_info->nr_async_submits);
784 btrfs_set_work_high_prio(&async->work);
786 btrfs_queue_worker(&fs_info->workers, &async->work);
788 while (atomic_read(&fs_info->async_submit_draining) &&
789 atomic_read(&fs_info->nr_async_submits)) {
790 wait_event(fs_info->async_submit_wait,
791 (atomic_read(&fs_info->nr_async_submits) == 0));
797 static int btree_csum_one_bio(struct bio *bio)
799 struct bio_vec *bvec = bio->bi_io_vec;
801 struct btrfs_root *root;
803 WARN_ON(bio->bi_vcnt <= 0);
804 while (bio_index < bio->bi_vcnt) {
805 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
806 csum_dirty_buffer(root, bvec->bv_page);
813 static int __btree_submit_bio_start(struct inode *inode, int rw,
814 struct bio *bio, int mirror_num,
815 unsigned long bio_flags,
819 * when we're called for a write, we're already in the async
820 * submission context. Just jump into btrfs_map_bio
822 btree_csum_one_bio(bio);
826 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
827 int mirror_num, unsigned long bio_flags,
831 * when we're called for a write, we're already in the async
832 * submission context. Just jump into btrfs_map_bio
834 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
837 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
838 int mirror_num, unsigned long bio_flags,
843 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
847 if (!(rw & REQ_WRITE)) {
849 * called for a read, do the setup so that checksum validation
850 * can happen in the async kernel threads
852 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
857 * kthread helpers are used to submit writes so that checksumming
858 * can happen in parallel across all CPUs
860 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
861 inode, rw, bio, mirror_num, 0,
863 __btree_submit_bio_start,
864 __btree_submit_bio_done);
867 #ifdef CONFIG_MIGRATION
868 static int btree_migratepage(struct address_space *mapping,
869 struct page *newpage, struct page *page,
870 enum migrate_mode mode)
873 * we can't safely write a btree page from here,
874 * we haven't done the locking hook
879 * Buffers may be managed in a filesystem specific way.
880 * We must have no buffers or drop them.
882 if (page_has_private(page) &&
883 !try_to_release_page(page, GFP_KERNEL))
885 return migrate_page(mapping, newpage, page, mode);
890 static int btree_writepages(struct address_space *mapping,
891 struct writeback_control *wbc)
893 struct extent_io_tree *tree;
894 tree = &BTRFS_I(mapping->host)->io_tree;
895 if (wbc->sync_mode == WB_SYNC_NONE) {
896 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
898 unsigned long thresh = 32 * 1024 * 1024;
900 if (wbc->for_kupdate)
903 /* this is a bit racy, but that's ok */
904 num_dirty = root->fs_info->dirty_metadata_bytes;
905 if (num_dirty < thresh)
908 return btree_write_cache_pages(mapping, wbc);
911 static int btree_readpage(struct file *file, struct page *page)
913 struct extent_io_tree *tree;
914 tree = &BTRFS_I(page->mapping->host)->io_tree;
915 return extent_read_full_page(tree, page, btree_get_extent, 0);
918 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
920 if (PageWriteback(page) || PageDirty(page))
923 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
924 * slab allocation from alloc_extent_state down the callchain where
925 * it'd hit a BUG_ON as those flags are not allowed.
927 gfp_flags &= ~GFP_SLAB_BUG_MASK;
929 return try_release_extent_buffer(page, gfp_flags);
932 static void btree_invalidatepage(struct page *page, unsigned long offset)
934 struct extent_io_tree *tree;
935 tree = &BTRFS_I(page->mapping->host)->io_tree;
936 extent_invalidatepage(tree, page, offset);
937 btree_releasepage(page, GFP_NOFS);
938 if (PagePrivate(page)) {
939 printk(KERN_WARNING "btrfs warning page private not zero "
940 "on page %llu\n", (unsigned long long)page_offset(page));
941 ClearPagePrivate(page);
942 set_page_private(page, 0);
943 page_cache_release(page);
947 static int btree_set_page_dirty(struct page *page)
949 struct extent_buffer *eb;
951 BUG_ON(!PagePrivate(page));
952 eb = (struct extent_buffer *)page->private;
954 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
955 BUG_ON(!atomic_read(&eb->refs));
956 btrfs_assert_tree_locked(eb);
957 return __set_page_dirty_nobuffers(page);
960 static const struct address_space_operations btree_aops = {
961 .readpage = btree_readpage,
962 .writepages = btree_writepages,
963 .releasepage = btree_releasepage,
964 .invalidatepage = btree_invalidatepage,
965 #ifdef CONFIG_MIGRATION
966 .migratepage = btree_migratepage,
968 .set_page_dirty = btree_set_page_dirty,
971 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
974 struct extent_buffer *buf = NULL;
975 struct inode *btree_inode = root->fs_info->btree_inode;
978 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
981 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
982 buf, 0, WAIT_NONE, btree_get_extent, 0);
983 free_extent_buffer(buf);
987 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
988 int mirror_num, struct extent_buffer **eb)
990 struct extent_buffer *buf = NULL;
991 struct inode *btree_inode = root->fs_info->btree_inode;
992 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
995 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
999 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1001 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1002 btree_get_extent, mirror_num);
1004 free_extent_buffer(buf);
1008 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1009 free_extent_buffer(buf);
1011 } else if (extent_buffer_uptodate(buf)) {
1014 free_extent_buffer(buf);
1019 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1020 u64 bytenr, u32 blocksize)
1022 struct inode *btree_inode = root->fs_info->btree_inode;
1023 struct extent_buffer *eb;
1024 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1029 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1030 u64 bytenr, u32 blocksize)
1032 struct inode *btree_inode = root->fs_info->btree_inode;
1033 struct extent_buffer *eb;
1035 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1041 int btrfs_write_tree_block(struct extent_buffer *buf)
1043 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1044 buf->start + buf->len - 1);
1047 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1049 return filemap_fdatawait_range(buf->pages[0]->mapping,
1050 buf->start, buf->start + buf->len - 1);
1053 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1054 u32 blocksize, u64 parent_transid)
1056 struct extent_buffer *buf = NULL;
1059 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1063 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1068 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1069 struct extent_buffer *buf)
1071 if (btrfs_header_generation(buf) ==
1072 root->fs_info->running_transaction->transid) {
1073 btrfs_assert_tree_locked(buf);
1075 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1076 spin_lock(&root->fs_info->delalloc_lock);
1077 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1078 root->fs_info->dirty_metadata_bytes -= buf->len;
1081 spin_unlock(&root->fs_info->delalloc_lock);
1084 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1085 btrfs_set_lock_blocking(buf);
1086 clear_extent_buffer_dirty(buf);
1091 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1092 u32 stripesize, struct btrfs_root *root,
1093 struct btrfs_fs_info *fs_info,
1097 root->commit_root = NULL;
1098 root->sectorsize = sectorsize;
1099 root->nodesize = nodesize;
1100 root->leafsize = leafsize;
1101 root->stripesize = stripesize;
1103 root->track_dirty = 0;
1105 root->orphan_item_inserted = 0;
1106 root->orphan_cleanup_state = 0;
1108 root->objectid = objectid;
1109 root->last_trans = 0;
1110 root->highest_objectid = 0;
1112 root->inode_tree = RB_ROOT;
1113 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1114 root->block_rsv = NULL;
1115 root->orphan_block_rsv = NULL;
1117 INIT_LIST_HEAD(&root->dirty_list);
1118 INIT_LIST_HEAD(&root->orphan_list);
1119 INIT_LIST_HEAD(&root->root_list);
1120 spin_lock_init(&root->orphan_lock);
1121 spin_lock_init(&root->inode_lock);
1122 spin_lock_init(&root->accounting_lock);
1123 mutex_init(&root->objectid_mutex);
1124 mutex_init(&root->log_mutex);
1125 init_waitqueue_head(&root->log_writer_wait);
1126 init_waitqueue_head(&root->log_commit_wait[0]);
1127 init_waitqueue_head(&root->log_commit_wait[1]);
1128 atomic_set(&root->log_commit[0], 0);
1129 atomic_set(&root->log_commit[1], 0);
1130 atomic_set(&root->log_writers, 0);
1131 root->log_batch = 0;
1132 root->log_transid = 0;
1133 root->last_log_commit = 0;
1134 extent_io_tree_init(&root->dirty_log_pages,
1135 fs_info->btree_inode->i_mapping);
1137 memset(&root->root_key, 0, sizeof(root->root_key));
1138 memset(&root->root_item, 0, sizeof(root->root_item));
1139 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1140 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1141 root->defrag_trans_start = fs_info->generation;
1142 init_completion(&root->kobj_unregister);
1143 root->defrag_running = 0;
1144 root->root_key.objectid = objectid;
1149 static int find_and_setup_root(struct btrfs_root *tree_root,
1150 struct btrfs_fs_info *fs_info,
1152 struct btrfs_root *root)
1158 __setup_root(tree_root->nodesize, tree_root->leafsize,
1159 tree_root->sectorsize, tree_root->stripesize,
1160 root, fs_info, objectid);
1161 ret = btrfs_find_last_root(tree_root, objectid,
1162 &root->root_item, &root->root_key);
1167 generation = btrfs_root_generation(&root->root_item);
1168 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1169 root->commit_root = NULL;
1170 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1171 blocksize, generation);
1172 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1173 free_extent_buffer(root->node);
1177 root->commit_root = btrfs_root_node(root);
1181 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1183 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1185 root->fs_info = fs_info;
1189 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1190 struct btrfs_fs_info *fs_info)
1192 struct btrfs_root *root;
1193 struct btrfs_root *tree_root = fs_info->tree_root;
1194 struct extent_buffer *leaf;
1196 root = btrfs_alloc_root(fs_info);
1198 return ERR_PTR(-ENOMEM);
1200 __setup_root(tree_root->nodesize, tree_root->leafsize,
1201 tree_root->sectorsize, tree_root->stripesize,
1202 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1204 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1205 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1206 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1208 * log trees do not get reference counted because they go away
1209 * before a real commit is actually done. They do store pointers
1210 * to file data extents, and those reference counts still get
1211 * updated (along with back refs to the log tree).
1215 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1216 BTRFS_TREE_LOG_OBJECTID, NULL,
1220 return ERR_CAST(leaf);
1223 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1224 btrfs_set_header_bytenr(leaf, leaf->start);
1225 btrfs_set_header_generation(leaf, trans->transid);
1226 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1227 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1230 write_extent_buffer(root->node, root->fs_info->fsid,
1231 (unsigned long)btrfs_header_fsid(root->node),
1233 btrfs_mark_buffer_dirty(root->node);
1234 btrfs_tree_unlock(root->node);
1238 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1239 struct btrfs_fs_info *fs_info)
1241 struct btrfs_root *log_root;
1243 log_root = alloc_log_tree(trans, fs_info);
1244 if (IS_ERR(log_root))
1245 return PTR_ERR(log_root);
1246 WARN_ON(fs_info->log_root_tree);
1247 fs_info->log_root_tree = log_root;
1251 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1252 struct btrfs_root *root)
1254 struct btrfs_root *log_root;
1255 struct btrfs_inode_item *inode_item;
1257 log_root = alloc_log_tree(trans, root->fs_info);
1258 if (IS_ERR(log_root))
1259 return PTR_ERR(log_root);
1261 log_root->last_trans = trans->transid;
1262 log_root->root_key.offset = root->root_key.objectid;
1264 inode_item = &log_root->root_item.inode;
1265 inode_item->generation = cpu_to_le64(1);
1266 inode_item->size = cpu_to_le64(3);
1267 inode_item->nlink = cpu_to_le32(1);
1268 inode_item->nbytes = cpu_to_le64(root->leafsize);
1269 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1271 btrfs_set_root_node(&log_root->root_item, log_root->node);
1273 WARN_ON(root->log_root);
1274 root->log_root = log_root;
1275 root->log_transid = 0;
1276 root->last_log_commit = 0;
1280 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1281 struct btrfs_key *location)
1283 struct btrfs_root *root;
1284 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1285 struct btrfs_path *path;
1286 struct extent_buffer *l;
1291 root = btrfs_alloc_root(fs_info);
1293 return ERR_PTR(-ENOMEM);
1294 if (location->offset == (u64)-1) {
1295 ret = find_and_setup_root(tree_root, fs_info,
1296 location->objectid, root);
1299 return ERR_PTR(ret);
1304 __setup_root(tree_root->nodesize, tree_root->leafsize,
1305 tree_root->sectorsize, tree_root->stripesize,
1306 root, fs_info, location->objectid);
1308 path = btrfs_alloc_path();
1311 return ERR_PTR(-ENOMEM);
1313 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1316 read_extent_buffer(l, &root->root_item,
1317 btrfs_item_ptr_offset(l, path->slots[0]),
1318 sizeof(root->root_item));
1319 memcpy(&root->root_key, location, sizeof(*location));
1321 btrfs_free_path(path);
1326 return ERR_PTR(ret);
1329 generation = btrfs_root_generation(&root->root_item);
1330 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1331 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1332 blocksize, generation);
1333 root->commit_root = btrfs_root_node(root);
1334 BUG_ON(!root->node);
1336 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1338 btrfs_check_and_init_root_item(&root->root_item);
1344 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1345 struct btrfs_key *location)
1347 struct btrfs_root *root;
1350 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1351 return fs_info->tree_root;
1352 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1353 return fs_info->extent_root;
1354 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1355 return fs_info->chunk_root;
1356 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1357 return fs_info->dev_root;
1358 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1359 return fs_info->csum_root;
1361 spin_lock(&fs_info->fs_roots_radix_lock);
1362 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1363 (unsigned long)location->objectid);
1364 spin_unlock(&fs_info->fs_roots_radix_lock);
1368 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1372 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1373 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1375 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1380 btrfs_init_free_ino_ctl(root);
1381 mutex_init(&root->fs_commit_mutex);
1382 spin_lock_init(&root->cache_lock);
1383 init_waitqueue_head(&root->cache_wait);
1385 ret = get_anon_bdev(&root->anon_dev);
1389 if (btrfs_root_refs(&root->root_item) == 0) {
1394 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1398 root->orphan_item_inserted = 1;
1400 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1404 spin_lock(&fs_info->fs_roots_radix_lock);
1405 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1406 (unsigned long)root->root_key.objectid,
1411 spin_unlock(&fs_info->fs_roots_radix_lock);
1412 radix_tree_preload_end();
1414 if (ret == -EEXIST) {
1421 ret = btrfs_find_dead_roots(fs_info->tree_root,
1422 root->root_key.objectid);
1427 return ERR_PTR(ret);
1430 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1432 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1434 struct btrfs_device *device;
1435 struct backing_dev_info *bdi;
1438 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1441 bdi = blk_get_backing_dev_info(device->bdev);
1442 if (bdi && bdi_congested(bdi, bdi_bits)) {
1452 * If this fails, caller must call bdi_destroy() to get rid of the
1455 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1459 bdi->capabilities = BDI_CAP_MAP_COPY;
1460 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1464 bdi->ra_pages = default_backing_dev_info.ra_pages;
1465 bdi->congested_fn = btrfs_congested_fn;
1466 bdi->congested_data = info;
1471 * called by the kthread helper functions to finally call the bio end_io
1472 * functions. This is where read checksum verification actually happens
1474 static void end_workqueue_fn(struct btrfs_work *work)
1477 struct end_io_wq *end_io_wq;
1478 struct btrfs_fs_info *fs_info;
1481 end_io_wq = container_of(work, struct end_io_wq, work);
1482 bio = end_io_wq->bio;
1483 fs_info = end_io_wq->info;
1485 error = end_io_wq->error;
1486 bio->bi_private = end_io_wq->private;
1487 bio->bi_end_io = end_io_wq->end_io;
1489 bio_endio(bio, error);
1492 static int cleaner_kthread(void *arg)
1494 struct btrfs_root *root = arg;
1497 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1499 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1500 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1501 btrfs_run_delayed_iputs(root);
1502 btrfs_clean_old_snapshots(root);
1503 mutex_unlock(&root->fs_info->cleaner_mutex);
1504 btrfs_run_defrag_inodes(root->fs_info);
1507 if (!try_to_freeze()) {
1508 set_current_state(TASK_INTERRUPTIBLE);
1509 if (!kthread_should_stop())
1511 __set_current_state(TASK_RUNNING);
1513 } while (!kthread_should_stop());
1517 static int transaction_kthread(void *arg)
1519 struct btrfs_root *root = arg;
1520 struct btrfs_trans_handle *trans;
1521 struct btrfs_transaction *cur;
1524 unsigned long delay;
1529 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1530 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1532 spin_lock(&root->fs_info->trans_lock);
1533 cur = root->fs_info->running_transaction;
1535 spin_unlock(&root->fs_info->trans_lock);
1539 now = get_seconds();
1540 if (!cur->blocked &&
1541 (now < cur->start_time || now - cur->start_time < 30)) {
1542 spin_unlock(&root->fs_info->trans_lock);
1546 transid = cur->transid;
1547 spin_unlock(&root->fs_info->trans_lock);
1549 trans = btrfs_join_transaction(root);
1550 BUG_ON(IS_ERR(trans));
1551 if (transid == trans->transid) {
1552 ret = btrfs_commit_transaction(trans, root);
1555 btrfs_end_transaction(trans, root);
1558 wake_up_process(root->fs_info->cleaner_kthread);
1559 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1561 if (!try_to_freeze()) {
1562 set_current_state(TASK_INTERRUPTIBLE);
1563 if (!kthread_should_stop() &&
1564 !btrfs_transaction_blocked(root->fs_info))
1565 schedule_timeout(delay);
1566 __set_current_state(TASK_RUNNING);
1568 } while (!kthread_should_stop());
1573 * this will find the highest generation in the array of
1574 * root backups. The index of the highest array is returned,
1575 * or -1 if we can't find anything.
1577 * We check to make sure the array is valid by comparing the
1578 * generation of the latest root in the array with the generation
1579 * in the super block. If they don't match we pitch it.
1581 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1584 int newest_index = -1;
1585 struct btrfs_root_backup *root_backup;
1588 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1589 root_backup = info->super_copy->super_roots + i;
1590 cur = btrfs_backup_tree_root_gen(root_backup);
1591 if (cur == newest_gen)
1595 /* check to see if we actually wrapped around */
1596 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1597 root_backup = info->super_copy->super_roots;
1598 cur = btrfs_backup_tree_root_gen(root_backup);
1599 if (cur == newest_gen)
1602 return newest_index;
1607 * find the oldest backup so we know where to store new entries
1608 * in the backup array. This will set the backup_root_index
1609 * field in the fs_info struct
1611 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1614 int newest_index = -1;
1616 newest_index = find_newest_super_backup(info, newest_gen);
1617 /* if there was garbage in there, just move along */
1618 if (newest_index == -1) {
1619 info->backup_root_index = 0;
1621 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1626 * copy all the root pointers into the super backup array.
1627 * this will bump the backup pointer by one when it is
1630 static void backup_super_roots(struct btrfs_fs_info *info)
1633 struct btrfs_root_backup *root_backup;
1636 next_backup = info->backup_root_index;
1637 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1638 BTRFS_NUM_BACKUP_ROOTS;
1641 * just overwrite the last backup if we're at the same generation
1642 * this happens only at umount
1644 root_backup = info->super_for_commit->super_roots + last_backup;
1645 if (btrfs_backup_tree_root_gen(root_backup) ==
1646 btrfs_header_generation(info->tree_root->node))
1647 next_backup = last_backup;
1649 root_backup = info->super_for_commit->super_roots + next_backup;
1652 * make sure all of our padding and empty slots get zero filled
1653 * regardless of which ones we use today
1655 memset(root_backup, 0, sizeof(*root_backup));
1657 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1659 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1660 btrfs_set_backup_tree_root_gen(root_backup,
1661 btrfs_header_generation(info->tree_root->node));
1663 btrfs_set_backup_tree_root_level(root_backup,
1664 btrfs_header_level(info->tree_root->node));
1666 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1667 btrfs_set_backup_chunk_root_gen(root_backup,
1668 btrfs_header_generation(info->chunk_root->node));
1669 btrfs_set_backup_chunk_root_level(root_backup,
1670 btrfs_header_level(info->chunk_root->node));
1672 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1673 btrfs_set_backup_extent_root_gen(root_backup,
1674 btrfs_header_generation(info->extent_root->node));
1675 btrfs_set_backup_extent_root_level(root_backup,
1676 btrfs_header_level(info->extent_root->node));
1679 * we might commit during log recovery, which happens before we set
1680 * the fs_root. Make sure it is valid before we fill it in.
1682 if (info->fs_root && info->fs_root->node) {
1683 btrfs_set_backup_fs_root(root_backup,
1684 info->fs_root->node->start);
1685 btrfs_set_backup_fs_root_gen(root_backup,
1686 btrfs_header_generation(info->fs_root->node));
1687 btrfs_set_backup_fs_root_level(root_backup,
1688 btrfs_header_level(info->fs_root->node));
1691 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1692 btrfs_set_backup_dev_root_gen(root_backup,
1693 btrfs_header_generation(info->dev_root->node));
1694 btrfs_set_backup_dev_root_level(root_backup,
1695 btrfs_header_level(info->dev_root->node));
1697 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1698 btrfs_set_backup_csum_root_gen(root_backup,
1699 btrfs_header_generation(info->csum_root->node));
1700 btrfs_set_backup_csum_root_level(root_backup,
1701 btrfs_header_level(info->csum_root->node));
1703 btrfs_set_backup_total_bytes(root_backup,
1704 btrfs_super_total_bytes(info->super_copy));
1705 btrfs_set_backup_bytes_used(root_backup,
1706 btrfs_super_bytes_used(info->super_copy));
1707 btrfs_set_backup_num_devices(root_backup,
1708 btrfs_super_num_devices(info->super_copy));
1711 * if we don't copy this out to the super_copy, it won't get remembered
1712 * for the next commit
1714 memcpy(&info->super_copy->super_roots,
1715 &info->super_for_commit->super_roots,
1716 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1720 * this copies info out of the root backup array and back into
1721 * the in-memory super block. It is meant to help iterate through
1722 * the array, so you send it the number of backups you've already
1723 * tried and the last backup index you used.
1725 * this returns -1 when it has tried all the backups
1727 static noinline int next_root_backup(struct btrfs_fs_info *info,
1728 struct btrfs_super_block *super,
1729 int *num_backups_tried, int *backup_index)
1731 struct btrfs_root_backup *root_backup;
1732 int newest = *backup_index;
1734 if (*num_backups_tried == 0) {
1735 u64 gen = btrfs_super_generation(super);
1737 newest = find_newest_super_backup(info, gen);
1741 *backup_index = newest;
1742 *num_backups_tried = 1;
1743 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1744 /* we've tried all the backups, all done */
1747 /* jump to the next oldest backup */
1748 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1749 BTRFS_NUM_BACKUP_ROOTS;
1750 *backup_index = newest;
1751 *num_backups_tried += 1;
1753 root_backup = super->super_roots + newest;
1755 btrfs_set_super_generation(super,
1756 btrfs_backup_tree_root_gen(root_backup));
1757 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1758 btrfs_set_super_root_level(super,
1759 btrfs_backup_tree_root_level(root_backup));
1760 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1763 * fixme: the total bytes and num_devices need to match or we should
1766 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1767 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1771 /* helper to cleanup tree roots */
1772 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1774 free_extent_buffer(info->tree_root->node);
1775 free_extent_buffer(info->tree_root->commit_root);
1776 free_extent_buffer(info->dev_root->node);
1777 free_extent_buffer(info->dev_root->commit_root);
1778 free_extent_buffer(info->extent_root->node);
1779 free_extent_buffer(info->extent_root->commit_root);
1780 free_extent_buffer(info->csum_root->node);
1781 free_extent_buffer(info->csum_root->commit_root);
1783 info->tree_root->node = NULL;
1784 info->tree_root->commit_root = NULL;
1785 info->dev_root->node = NULL;
1786 info->dev_root->commit_root = NULL;
1787 info->extent_root->node = NULL;
1788 info->extent_root->commit_root = NULL;
1789 info->csum_root->node = NULL;
1790 info->csum_root->commit_root = NULL;
1793 free_extent_buffer(info->chunk_root->node);
1794 free_extent_buffer(info->chunk_root->commit_root);
1795 info->chunk_root->node = NULL;
1796 info->chunk_root->commit_root = NULL;
1801 int open_ctree(struct super_block *sb,
1802 struct btrfs_fs_devices *fs_devices,
1812 struct btrfs_key location;
1813 struct buffer_head *bh;
1814 struct btrfs_super_block *disk_super;
1815 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1816 struct btrfs_root *tree_root;
1817 struct btrfs_root *extent_root;
1818 struct btrfs_root *csum_root;
1819 struct btrfs_root *chunk_root;
1820 struct btrfs_root *dev_root;
1821 struct btrfs_root *log_tree_root;
1824 int num_backups_tried = 0;
1825 int backup_index = 0;
1827 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1828 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1829 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1830 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1831 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1833 if (!tree_root || !extent_root || !csum_root ||
1834 !chunk_root || !dev_root) {
1839 ret = init_srcu_struct(&fs_info->subvol_srcu);
1845 ret = setup_bdi(fs_info, &fs_info->bdi);
1851 fs_info->btree_inode = new_inode(sb);
1852 if (!fs_info->btree_inode) {
1857 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1859 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1860 INIT_LIST_HEAD(&fs_info->trans_list);
1861 INIT_LIST_HEAD(&fs_info->dead_roots);
1862 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1863 INIT_LIST_HEAD(&fs_info->hashers);
1864 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1865 INIT_LIST_HEAD(&fs_info->ordered_operations);
1866 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1867 spin_lock_init(&fs_info->delalloc_lock);
1868 spin_lock_init(&fs_info->trans_lock);
1869 spin_lock_init(&fs_info->ref_cache_lock);
1870 spin_lock_init(&fs_info->fs_roots_radix_lock);
1871 spin_lock_init(&fs_info->delayed_iput_lock);
1872 spin_lock_init(&fs_info->defrag_inodes_lock);
1873 spin_lock_init(&fs_info->free_chunk_lock);
1874 mutex_init(&fs_info->reloc_mutex);
1876 init_completion(&fs_info->kobj_unregister);
1877 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1878 INIT_LIST_HEAD(&fs_info->space_info);
1879 btrfs_mapping_init(&fs_info->mapping_tree);
1880 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1881 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1882 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1883 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1884 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1885 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1886 atomic_set(&fs_info->nr_async_submits, 0);
1887 atomic_set(&fs_info->async_delalloc_pages, 0);
1888 atomic_set(&fs_info->async_submit_draining, 0);
1889 atomic_set(&fs_info->nr_async_bios, 0);
1890 atomic_set(&fs_info->defrag_running, 0);
1892 fs_info->max_inline = 8192 * 1024;
1893 fs_info->metadata_ratio = 0;
1894 fs_info->defrag_inodes = RB_ROOT;
1895 fs_info->trans_no_join = 0;
1896 fs_info->free_chunk_space = 0;
1898 /* readahead state */
1899 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1900 spin_lock_init(&fs_info->reada_lock);
1902 fs_info->thread_pool_size = min_t(unsigned long,
1903 num_online_cpus() + 2, 8);
1905 INIT_LIST_HEAD(&fs_info->ordered_extents);
1906 spin_lock_init(&fs_info->ordered_extent_lock);
1907 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1909 if (!fs_info->delayed_root) {
1913 btrfs_init_delayed_root(fs_info->delayed_root);
1915 mutex_init(&fs_info->scrub_lock);
1916 atomic_set(&fs_info->scrubs_running, 0);
1917 atomic_set(&fs_info->scrub_pause_req, 0);
1918 atomic_set(&fs_info->scrubs_paused, 0);
1919 atomic_set(&fs_info->scrub_cancel_req, 0);
1920 init_waitqueue_head(&fs_info->scrub_pause_wait);
1921 init_rwsem(&fs_info->scrub_super_lock);
1922 fs_info->scrub_workers_refcnt = 0;
1923 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1924 fs_info->check_integrity_print_mask = 0;
1927 spin_lock_init(&fs_info->balance_lock);
1928 mutex_init(&fs_info->balance_mutex);
1929 atomic_set(&fs_info->balance_running, 0);
1930 atomic_set(&fs_info->balance_pause_req, 0);
1931 atomic_set(&fs_info->balance_cancel_req, 0);
1932 fs_info->balance_ctl = NULL;
1933 init_waitqueue_head(&fs_info->balance_wait_q);
1935 sb->s_blocksize = 4096;
1936 sb->s_blocksize_bits = blksize_bits(4096);
1937 sb->s_bdi = &fs_info->bdi;
1939 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1940 set_nlink(fs_info->btree_inode, 1);
1942 * we set the i_size on the btree inode to the max possible int.
1943 * the real end of the address space is determined by all of
1944 * the devices in the system
1946 fs_info->btree_inode->i_size = OFFSET_MAX;
1947 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1948 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1950 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1951 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1952 fs_info->btree_inode->i_mapping);
1953 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
1954 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1956 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1958 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1959 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1960 sizeof(struct btrfs_key));
1961 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1962 insert_inode_hash(fs_info->btree_inode);
1964 spin_lock_init(&fs_info->block_group_cache_lock);
1965 fs_info->block_group_cache_tree = RB_ROOT;
1967 extent_io_tree_init(&fs_info->freed_extents[0],
1968 fs_info->btree_inode->i_mapping);
1969 extent_io_tree_init(&fs_info->freed_extents[1],
1970 fs_info->btree_inode->i_mapping);
1971 fs_info->pinned_extents = &fs_info->freed_extents[0];
1972 fs_info->do_barriers = 1;
1975 mutex_init(&fs_info->ordered_operations_mutex);
1976 mutex_init(&fs_info->tree_log_mutex);
1977 mutex_init(&fs_info->chunk_mutex);
1978 mutex_init(&fs_info->transaction_kthread_mutex);
1979 mutex_init(&fs_info->cleaner_mutex);
1980 mutex_init(&fs_info->volume_mutex);
1981 init_rwsem(&fs_info->extent_commit_sem);
1982 init_rwsem(&fs_info->cleanup_work_sem);
1983 init_rwsem(&fs_info->subvol_sem);
1985 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1986 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1988 init_waitqueue_head(&fs_info->transaction_throttle);
1989 init_waitqueue_head(&fs_info->transaction_wait);
1990 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1991 init_waitqueue_head(&fs_info->async_submit_wait);
1993 __setup_root(4096, 4096, 4096, 4096, tree_root,
1994 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1996 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2002 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2003 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2004 sizeof(*fs_info->super_for_commit));
2007 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2009 disk_super = fs_info->super_copy;
2010 if (!btrfs_super_root(disk_super))
2013 /* check FS state, whether FS is broken. */
2014 fs_info->fs_state |= btrfs_super_flags(disk_super);
2016 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2019 * run through our array of backup supers and setup
2020 * our ring pointer to the oldest one
2022 generation = btrfs_super_generation(disk_super);
2023 find_oldest_super_backup(fs_info, generation);
2026 * In the long term, we'll store the compression type in the super
2027 * block, and it'll be used for per file compression control.
2029 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2031 ret = btrfs_parse_options(tree_root, options);
2037 features = btrfs_super_incompat_flags(disk_super) &
2038 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2040 printk(KERN_ERR "BTRFS: couldn't mount because of "
2041 "unsupported optional features (%Lx).\n",
2042 (unsigned long long)features);
2047 if (btrfs_super_leafsize(disk_super) !=
2048 btrfs_super_nodesize(disk_super)) {
2049 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2050 "blocksizes don't match. node %d leaf %d\n",
2051 btrfs_super_nodesize(disk_super),
2052 btrfs_super_leafsize(disk_super));
2056 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2057 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2058 "blocksize (%d) was too large\n",
2059 btrfs_super_leafsize(disk_super));
2064 features = btrfs_super_incompat_flags(disk_super);
2065 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2066 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2067 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2070 * flag our filesystem as having big metadata blocks if
2071 * they are bigger than the page size
2073 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2074 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2075 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2076 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2079 btrfs_set_super_incompat_flags(disk_super, features);
2081 features = btrfs_super_compat_ro_flags(disk_super) &
2082 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2083 if (!(sb->s_flags & MS_RDONLY) && features) {
2084 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2085 "unsupported option features (%Lx).\n",
2086 (unsigned long long)features);
2091 btrfs_init_workers(&fs_info->generic_worker,
2092 "genwork", 1, NULL);
2094 btrfs_init_workers(&fs_info->workers, "worker",
2095 fs_info->thread_pool_size,
2096 &fs_info->generic_worker);
2098 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2099 fs_info->thread_pool_size,
2100 &fs_info->generic_worker);
2102 btrfs_init_workers(&fs_info->submit_workers, "submit",
2103 min_t(u64, fs_devices->num_devices,
2104 fs_info->thread_pool_size),
2105 &fs_info->generic_worker);
2107 btrfs_init_workers(&fs_info->caching_workers, "cache",
2108 2, &fs_info->generic_worker);
2110 /* a higher idle thresh on the submit workers makes it much more
2111 * likely that bios will be send down in a sane order to the
2114 fs_info->submit_workers.idle_thresh = 64;
2116 fs_info->workers.idle_thresh = 16;
2117 fs_info->workers.ordered = 1;
2119 fs_info->delalloc_workers.idle_thresh = 2;
2120 fs_info->delalloc_workers.ordered = 1;
2122 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2123 &fs_info->generic_worker);
2124 btrfs_init_workers(&fs_info->endio_workers, "endio",
2125 fs_info->thread_pool_size,
2126 &fs_info->generic_worker);
2127 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2128 fs_info->thread_pool_size,
2129 &fs_info->generic_worker);
2130 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2131 "endio-meta-write", fs_info->thread_pool_size,
2132 &fs_info->generic_worker);
2133 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2134 fs_info->thread_pool_size,
2135 &fs_info->generic_worker);
2136 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2137 1, &fs_info->generic_worker);
2138 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2139 fs_info->thread_pool_size,
2140 &fs_info->generic_worker);
2141 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2142 fs_info->thread_pool_size,
2143 &fs_info->generic_worker);
2146 * endios are largely parallel and should have a very
2149 fs_info->endio_workers.idle_thresh = 4;
2150 fs_info->endio_meta_workers.idle_thresh = 4;
2152 fs_info->endio_write_workers.idle_thresh = 2;
2153 fs_info->endio_meta_write_workers.idle_thresh = 2;
2154 fs_info->readahead_workers.idle_thresh = 2;
2157 * btrfs_start_workers can really only fail because of ENOMEM so just
2158 * return -ENOMEM if any of these fail.
2160 ret = btrfs_start_workers(&fs_info->workers);
2161 ret |= btrfs_start_workers(&fs_info->generic_worker);
2162 ret |= btrfs_start_workers(&fs_info->submit_workers);
2163 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2164 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2165 ret |= btrfs_start_workers(&fs_info->endio_workers);
2166 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2167 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2168 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2169 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2170 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2171 ret |= btrfs_start_workers(&fs_info->caching_workers);
2172 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2175 goto fail_sb_buffer;
2178 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2179 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2180 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2182 nodesize = btrfs_super_nodesize(disk_super);
2183 leafsize = btrfs_super_leafsize(disk_super);
2184 sectorsize = btrfs_super_sectorsize(disk_super);
2185 stripesize = btrfs_super_stripesize(disk_super);
2186 tree_root->nodesize = nodesize;
2187 tree_root->leafsize = leafsize;
2188 tree_root->sectorsize = sectorsize;
2189 tree_root->stripesize = stripesize;
2191 sb->s_blocksize = sectorsize;
2192 sb->s_blocksize_bits = blksize_bits(sectorsize);
2194 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2195 sizeof(disk_super->magic))) {
2196 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2197 goto fail_sb_buffer;
2200 if (sectorsize < PAGE_SIZE) {
2201 printk(KERN_WARNING "btrfs: Incompatible sector size "
2202 "found on %s\n", sb->s_id);
2203 goto fail_sb_buffer;
2206 mutex_lock(&fs_info->chunk_mutex);
2207 ret = btrfs_read_sys_array(tree_root);
2208 mutex_unlock(&fs_info->chunk_mutex);
2210 printk(KERN_WARNING "btrfs: failed to read the system "
2211 "array on %s\n", sb->s_id);
2212 goto fail_sb_buffer;
2215 blocksize = btrfs_level_size(tree_root,
2216 btrfs_super_chunk_root_level(disk_super));
2217 generation = btrfs_super_chunk_root_generation(disk_super);
2219 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2220 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2222 chunk_root->node = read_tree_block(chunk_root,
2223 btrfs_super_chunk_root(disk_super),
2224 blocksize, generation);
2225 BUG_ON(!chunk_root->node);
2226 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2227 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2229 goto fail_tree_roots;
2231 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2232 chunk_root->commit_root = btrfs_root_node(chunk_root);
2234 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2235 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2238 ret = btrfs_read_chunk_tree(chunk_root);
2240 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2242 goto fail_tree_roots;
2245 btrfs_close_extra_devices(fs_devices);
2247 if (!fs_devices->latest_bdev) {
2248 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2250 goto fail_tree_roots;
2254 blocksize = btrfs_level_size(tree_root,
2255 btrfs_super_root_level(disk_super));
2256 generation = btrfs_super_generation(disk_super);
2258 tree_root->node = read_tree_block(tree_root,
2259 btrfs_super_root(disk_super),
2260 blocksize, generation);
2261 if (!tree_root->node ||
2262 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2263 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2266 goto recovery_tree_root;
2269 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2270 tree_root->commit_root = btrfs_root_node(tree_root);
2272 ret = find_and_setup_root(tree_root, fs_info,
2273 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2275 goto recovery_tree_root;
2276 extent_root->track_dirty = 1;
2278 ret = find_and_setup_root(tree_root, fs_info,
2279 BTRFS_DEV_TREE_OBJECTID, dev_root);
2281 goto recovery_tree_root;
2282 dev_root->track_dirty = 1;
2284 ret = find_and_setup_root(tree_root, fs_info,
2285 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2287 goto recovery_tree_root;
2289 csum_root->track_dirty = 1;
2291 fs_info->generation = generation;
2292 fs_info->last_trans_committed = generation;
2294 ret = btrfs_init_space_info(fs_info);
2296 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2297 goto fail_block_groups;
2300 ret = btrfs_read_block_groups(extent_root);
2302 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2303 goto fail_block_groups;
2306 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2308 if (IS_ERR(fs_info->cleaner_kthread))
2309 goto fail_block_groups;
2311 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2313 "btrfs-transaction");
2314 if (IS_ERR(fs_info->transaction_kthread))
2317 if (!btrfs_test_opt(tree_root, SSD) &&
2318 !btrfs_test_opt(tree_root, NOSSD) &&
2319 !fs_info->fs_devices->rotating) {
2320 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2322 btrfs_set_opt(fs_info->mount_opt, SSD);
2325 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2326 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2327 ret = btrfsic_mount(tree_root, fs_devices,
2328 btrfs_test_opt(tree_root,
2329 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2331 fs_info->check_integrity_print_mask);
2333 printk(KERN_WARNING "btrfs: failed to initialize"
2334 " integrity check module %s\n", sb->s_id);
2338 /* do not make disk changes in broken FS */
2339 if (btrfs_super_log_root(disk_super) != 0 &&
2340 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2341 u64 bytenr = btrfs_super_log_root(disk_super);
2343 if (fs_devices->rw_devices == 0) {
2344 printk(KERN_WARNING "Btrfs log replay required "
2347 goto fail_trans_kthread;
2350 btrfs_level_size(tree_root,
2351 btrfs_super_log_root_level(disk_super));
2353 log_tree_root = btrfs_alloc_root(fs_info);
2354 if (!log_tree_root) {
2356 goto fail_trans_kthread;
2359 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2360 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2362 log_tree_root->node = read_tree_block(tree_root, bytenr,
2365 ret = btrfs_recover_log_trees(log_tree_root);
2368 if (sb->s_flags & MS_RDONLY) {
2369 ret = btrfs_commit_super(tree_root);
2374 ret = btrfs_find_orphan_roots(tree_root);
2377 if (!(sb->s_flags & MS_RDONLY)) {
2378 ret = btrfs_cleanup_fs_roots(fs_info);
2381 ret = btrfs_recover_relocation(tree_root);
2384 "btrfs: failed to recover relocation\n");
2386 goto fail_trans_kthread;
2390 location.objectid = BTRFS_FS_TREE_OBJECTID;
2391 location.type = BTRFS_ROOT_ITEM_KEY;
2392 location.offset = (u64)-1;
2394 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2395 if (!fs_info->fs_root)
2396 goto fail_trans_kthread;
2397 if (IS_ERR(fs_info->fs_root)) {
2398 err = PTR_ERR(fs_info->fs_root);
2399 goto fail_trans_kthread;
2402 if (!(sb->s_flags & MS_RDONLY)) {
2403 down_read(&fs_info->cleanup_work_sem);
2404 err = btrfs_orphan_cleanup(fs_info->fs_root);
2406 err = btrfs_orphan_cleanup(fs_info->tree_root);
2407 up_read(&fs_info->cleanup_work_sem);
2410 err = btrfs_recover_balance(fs_info->tree_root);
2413 close_ctree(tree_root);
2421 kthread_stop(fs_info->transaction_kthread);
2423 kthread_stop(fs_info->cleaner_kthread);
2426 * make sure we're done with the btree inode before we stop our
2429 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2430 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2433 btrfs_free_block_groups(fs_info);
2436 free_root_pointers(fs_info, 1);
2439 btrfs_stop_workers(&fs_info->generic_worker);
2440 btrfs_stop_workers(&fs_info->readahead_workers);
2441 btrfs_stop_workers(&fs_info->fixup_workers);
2442 btrfs_stop_workers(&fs_info->delalloc_workers);
2443 btrfs_stop_workers(&fs_info->workers);
2444 btrfs_stop_workers(&fs_info->endio_workers);
2445 btrfs_stop_workers(&fs_info->endio_meta_workers);
2446 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2447 btrfs_stop_workers(&fs_info->endio_write_workers);
2448 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2449 btrfs_stop_workers(&fs_info->submit_workers);
2450 btrfs_stop_workers(&fs_info->delayed_workers);
2451 btrfs_stop_workers(&fs_info->caching_workers);
2454 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2456 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2457 iput(fs_info->btree_inode);
2459 bdi_destroy(&fs_info->bdi);
2461 cleanup_srcu_struct(&fs_info->subvol_srcu);
2463 btrfs_close_devices(fs_info->fs_devices);
2467 if (!btrfs_test_opt(tree_root, RECOVERY))
2468 goto fail_tree_roots;
2470 free_root_pointers(fs_info, 0);
2472 /* don't use the log in recovery mode, it won't be valid */
2473 btrfs_set_super_log_root(disk_super, 0);
2475 /* we can't trust the free space cache either */
2476 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2478 ret = next_root_backup(fs_info, fs_info->super_copy,
2479 &num_backups_tried, &backup_index);
2481 goto fail_block_groups;
2482 goto retry_root_backup;
2485 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2487 char b[BDEVNAME_SIZE];
2490 set_buffer_uptodate(bh);
2492 printk_ratelimited(KERN_WARNING "lost page write due to "
2493 "I/O error on %s\n",
2494 bdevname(bh->b_bdev, b));
2495 /* note, we dont' set_buffer_write_io_error because we have
2496 * our own ways of dealing with the IO errors
2498 clear_buffer_uptodate(bh);
2504 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2506 struct buffer_head *bh;
2507 struct buffer_head *latest = NULL;
2508 struct btrfs_super_block *super;
2513 /* we would like to check all the supers, but that would make
2514 * a btrfs mount succeed after a mkfs from a different FS.
2515 * So, we need to add a special mount option to scan for
2516 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2518 for (i = 0; i < 1; i++) {
2519 bytenr = btrfs_sb_offset(i);
2520 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2522 bh = __bread(bdev, bytenr / 4096, 4096);
2526 super = (struct btrfs_super_block *)bh->b_data;
2527 if (btrfs_super_bytenr(super) != bytenr ||
2528 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2529 sizeof(super->magic))) {
2534 if (!latest || btrfs_super_generation(super) > transid) {
2537 transid = btrfs_super_generation(super);
2546 * this should be called twice, once with wait == 0 and
2547 * once with wait == 1. When wait == 0 is done, all the buffer heads
2548 * we write are pinned.
2550 * They are released when wait == 1 is done.
2551 * max_mirrors must be the same for both runs, and it indicates how
2552 * many supers on this one device should be written.
2554 * max_mirrors == 0 means to write them all.
2556 static int write_dev_supers(struct btrfs_device *device,
2557 struct btrfs_super_block *sb,
2558 int do_barriers, int wait, int max_mirrors)
2560 struct buffer_head *bh;
2567 if (max_mirrors == 0)
2568 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2570 for (i = 0; i < max_mirrors; i++) {
2571 bytenr = btrfs_sb_offset(i);
2572 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2576 bh = __find_get_block(device->bdev, bytenr / 4096,
2577 BTRFS_SUPER_INFO_SIZE);
2580 if (!buffer_uptodate(bh))
2583 /* drop our reference */
2586 /* drop the reference from the wait == 0 run */
2590 btrfs_set_super_bytenr(sb, bytenr);
2593 crc = btrfs_csum_data(NULL, (char *)sb +
2594 BTRFS_CSUM_SIZE, crc,
2595 BTRFS_SUPER_INFO_SIZE -
2597 btrfs_csum_final(crc, sb->csum);
2600 * one reference for us, and we leave it for the
2603 bh = __getblk(device->bdev, bytenr / 4096,
2604 BTRFS_SUPER_INFO_SIZE);
2605 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2607 /* one reference for submit_bh */
2610 set_buffer_uptodate(bh);
2612 bh->b_end_io = btrfs_end_buffer_write_sync;
2616 * we fua the first super. The others we allow
2619 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2623 return errors < i ? 0 : -1;
2627 * endio for the write_dev_flush, this will wake anyone waiting
2628 * for the barrier when it is done
2630 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2633 if (err == -EOPNOTSUPP)
2634 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2635 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2637 if (bio->bi_private)
2638 complete(bio->bi_private);
2643 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2644 * sent down. With wait == 1, it waits for the previous flush.
2646 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2649 static int write_dev_flush(struct btrfs_device *device, int wait)
2654 if (device->nobarriers)
2658 bio = device->flush_bio;
2662 wait_for_completion(&device->flush_wait);
2664 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2665 printk("btrfs: disabling barriers on dev %s\n",
2667 device->nobarriers = 1;
2669 if (!bio_flagged(bio, BIO_UPTODATE)) {
2673 /* drop the reference from the wait == 0 run */
2675 device->flush_bio = NULL;
2681 * one reference for us, and we leave it for the
2684 device->flush_bio = NULL;;
2685 bio = bio_alloc(GFP_NOFS, 0);
2689 bio->bi_end_io = btrfs_end_empty_barrier;
2690 bio->bi_bdev = device->bdev;
2691 init_completion(&device->flush_wait);
2692 bio->bi_private = &device->flush_wait;
2693 device->flush_bio = bio;
2696 btrfsic_submit_bio(WRITE_FLUSH, bio);
2702 * send an empty flush down to each device in parallel,
2703 * then wait for them
2705 static int barrier_all_devices(struct btrfs_fs_info *info)
2707 struct list_head *head;
2708 struct btrfs_device *dev;
2712 /* send down all the barriers */
2713 head = &info->fs_devices->devices;
2714 list_for_each_entry_rcu(dev, head, dev_list) {
2719 if (!dev->in_fs_metadata || !dev->writeable)
2722 ret = write_dev_flush(dev, 0);
2727 /* wait for all the barriers */
2728 list_for_each_entry_rcu(dev, head, dev_list) {
2733 if (!dev->in_fs_metadata || !dev->writeable)
2736 ret = write_dev_flush(dev, 1);
2745 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2747 struct list_head *head;
2748 struct btrfs_device *dev;
2749 struct btrfs_super_block *sb;
2750 struct btrfs_dev_item *dev_item;
2754 int total_errors = 0;
2757 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2758 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2759 backup_super_roots(root->fs_info);
2761 sb = root->fs_info->super_for_commit;
2762 dev_item = &sb->dev_item;
2764 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2765 head = &root->fs_info->fs_devices->devices;
2768 barrier_all_devices(root->fs_info);
2770 list_for_each_entry_rcu(dev, head, dev_list) {
2775 if (!dev->in_fs_metadata || !dev->writeable)
2778 btrfs_set_stack_device_generation(dev_item, 0);
2779 btrfs_set_stack_device_type(dev_item, dev->type);
2780 btrfs_set_stack_device_id(dev_item, dev->devid);
2781 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2782 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2783 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2784 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2785 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2786 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2787 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2789 flags = btrfs_super_flags(sb);
2790 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2792 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2796 if (total_errors > max_errors) {
2797 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2803 list_for_each_entry_rcu(dev, head, dev_list) {
2806 if (!dev->in_fs_metadata || !dev->writeable)
2809 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2813 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2814 if (total_errors > max_errors) {
2815 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2822 int write_ctree_super(struct btrfs_trans_handle *trans,
2823 struct btrfs_root *root, int max_mirrors)
2827 ret = write_all_supers(root, max_mirrors);
2831 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2833 spin_lock(&fs_info->fs_roots_radix_lock);
2834 radix_tree_delete(&fs_info->fs_roots_radix,
2835 (unsigned long)root->root_key.objectid);
2836 spin_unlock(&fs_info->fs_roots_radix_lock);
2838 if (btrfs_root_refs(&root->root_item) == 0)
2839 synchronize_srcu(&fs_info->subvol_srcu);
2841 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2842 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2847 static void free_fs_root(struct btrfs_root *root)
2849 iput(root->cache_inode);
2850 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2852 free_anon_bdev(root->anon_dev);
2853 free_extent_buffer(root->node);
2854 free_extent_buffer(root->commit_root);
2855 kfree(root->free_ino_ctl);
2856 kfree(root->free_ino_pinned);
2861 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2864 struct btrfs_root *gang[8];
2867 while (!list_empty(&fs_info->dead_roots)) {
2868 gang[0] = list_entry(fs_info->dead_roots.next,
2869 struct btrfs_root, root_list);
2870 list_del(&gang[0]->root_list);
2872 if (gang[0]->in_radix) {
2873 btrfs_free_fs_root(fs_info, gang[0]);
2875 free_extent_buffer(gang[0]->node);
2876 free_extent_buffer(gang[0]->commit_root);
2882 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2887 for (i = 0; i < ret; i++)
2888 btrfs_free_fs_root(fs_info, gang[i]);
2893 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2895 u64 root_objectid = 0;
2896 struct btrfs_root *gang[8];
2901 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2902 (void **)gang, root_objectid,
2907 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2908 for (i = 0; i < ret; i++) {
2911 root_objectid = gang[i]->root_key.objectid;
2912 err = btrfs_orphan_cleanup(gang[i]);
2921 int btrfs_commit_super(struct btrfs_root *root)
2923 struct btrfs_trans_handle *trans;
2926 mutex_lock(&root->fs_info->cleaner_mutex);
2927 btrfs_run_delayed_iputs(root);
2928 btrfs_clean_old_snapshots(root);
2929 mutex_unlock(&root->fs_info->cleaner_mutex);
2931 /* wait until ongoing cleanup work done */
2932 down_write(&root->fs_info->cleanup_work_sem);
2933 up_write(&root->fs_info->cleanup_work_sem);
2935 trans = btrfs_join_transaction(root);
2937 return PTR_ERR(trans);
2938 ret = btrfs_commit_transaction(trans, root);
2940 /* run commit again to drop the original snapshot */
2941 trans = btrfs_join_transaction(root);
2943 return PTR_ERR(trans);
2944 btrfs_commit_transaction(trans, root);
2945 ret = btrfs_write_and_wait_transaction(NULL, root);
2948 ret = write_ctree_super(NULL, root, 0);
2952 int close_ctree(struct btrfs_root *root)
2954 struct btrfs_fs_info *fs_info = root->fs_info;
2957 fs_info->closing = 1;
2960 /* pause restriper - we want to resume on mount */
2961 btrfs_pause_balance(root->fs_info);
2963 btrfs_scrub_cancel(root);
2965 /* wait for any defraggers to finish */
2966 wait_event(fs_info->transaction_wait,
2967 (atomic_read(&fs_info->defrag_running) == 0));
2969 /* clear out the rbtree of defraggable inodes */
2970 btrfs_run_defrag_inodes(fs_info);
2973 * Here come 2 situations when btrfs is broken to flip readonly:
2975 * 1. when btrfs flips readonly somewhere else before
2976 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2977 * and btrfs will skip to write sb directly to keep
2978 * ERROR state on disk.
2980 * 2. when btrfs flips readonly just in btrfs_commit_super,
2981 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2982 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2983 * btrfs will cleanup all FS resources first and write sb then.
2985 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2986 ret = btrfs_commit_super(root);
2988 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2991 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2992 ret = btrfs_error_commit_super(root);
2994 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2997 btrfs_put_block_group_cache(fs_info);
2999 kthread_stop(fs_info->transaction_kthread);
3000 kthread_stop(fs_info->cleaner_kthread);
3002 fs_info->closing = 2;
3005 if (fs_info->delalloc_bytes) {
3006 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3007 (unsigned long long)fs_info->delalloc_bytes);
3009 if (fs_info->total_ref_cache_size) {
3010 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3011 (unsigned long long)fs_info->total_ref_cache_size);
3014 free_extent_buffer(fs_info->extent_root->node);
3015 free_extent_buffer(fs_info->extent_root->commit_root);
3016 free_extent_buffer(fs_info->tree_root->node);
3017 free_extent_buffer(fs_info->tree_root->commit_root);
3018 free_extent_buffer(fs_info->chunk_root->node);
3019 free_extent_buffer(fs_info->chunk_root->commit_root);
3020 free_extent_buffer(fs_info->dev_root->node);
3021 free_extent_buffer(fs_info->dev_root->commit_root);
3022 free_extent_buffer(fs_info->csum_root->node);
3023 free_extent_buffer(fs_info->csum_root->commit_root);
3025 btrfs_free_block_groups(fs_info);
3027 del_fs_roots(fs_info);
3029 iput(fs_info->btree_inode);
3031 btrfs_stop_workers(&fs_info->generic_worker);
3032 btrfs_stop_workers(&fs_info->fixup_workers);
3033 btrfs_stop_workers(&fs_info->delalloc_workers);
3034 btrfs_stop_workers(&fs_info->workers);
3035 btrfs_stop_workers(&fs_info->endio_workers);
3036 btrfs_stop_workers(&fs_info->endio_meta_workers);
3037 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3038 btrfs_stop_workers(&fs_info->endio_write_workers);
3039 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3040 btrfs_stop_workers(&fs_info->submit_workers);
3041 btrfs_stop_workers(&fs_info->delayed_workers);
3042 btrfs_stop_workers(&fs_info->caching_workers);
3043 btrfs_stop_workers(&fs_info->readahead_workers);
3045 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3046 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3047 btrfsic_unmount(root, fs_info->fs_devices);
3050 btrfs_close_devices(fs_info->fs_devices);
3051 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3053 bdi_destroy(&fs_info->bdi);
3054 cleanup_srcu_struct(&fs_info->subvol_srcu);
3059 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3062 struct inode *btree_inode = buf->pages[0]->mapping->host;
3064 ret = extent_buffer_uptodate(buf);
3068 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3073 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3075 return set_extent_buffer_uptodate(buf);
3078 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3080 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3081 u64 transid = btrfs_header_generation(buf);
3084 btrfs_assert_tree_locked(buf);
3085 if (transid != root->fs_info->generation) {
3086 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3087 "found %llu running %llu\n",
3088 (unsigned long long)buf->start,
3089 (unsigned long long)transid,
3090 (unsigned long long)root->fs_info->generation);
3093 was_dirty = set_extent_buffer_dirty(buf);
3095 spin_lock(&root->fs_info->delalloc_lock);
3096 root->fs_info->dirty_metadata_bytes += buf->len;
3097 spin_unlock(&root->fs_info->delalloc_lock);
3101 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3104 * looks as though older kernels can get into trouble with
3105 * this code, they end up stuck in balance_dirty_pages forever
3108 unsigned long thresh = 32 * 1024 * 1024;
3110 if (current->flags & PF_MEMALLOC)
3113 btrfs_balance_delayed_items(root);
3115 num_dirty = root->fs_info->dirty_metadata_bytes;
3117 if (num_dirty > thresh) {
3118 balance_dirty_pages_ratelimited_nr(
3119 root->fs_info->btree_inode->i_mapping, 1);
3124 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3127 * looks as though older kernels can get into trouble with
3128 * this code, they end up stuck in balance_dirty_pages forever
3131 unsigned long thresh = 32 * 1024 * 1024;
3133 if (current->flags & PF_MEMALLOC)
3136 num_dirty = root->fs_info->dirty_metadata_bytes;
3138 if (num_dirty > thresh) {
3139 balance_dirty_pages_ratelimited_nr(
3140 root->fs_info->btree_inode->i_mapping, 1);
3145 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3147 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3148 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3151 static int btree_lock_page_hook(struct page *page, void *data,
3152 void (*flush_fn)(void *))
3154 struct inode *inode = page->mapping->host;
3155 struct btrfs_root *root = BTRFS_I(inode)->root;
3156 struct extent_buffer *eb;
3159 * We culled this eb but the page is still hanging out on the mapping,
3162 if (!PagePrivate(page))
3165 eb = (struct extent_buffer *)page->private;
3170 if (page != eb->pages[0])
3173 if (!btrfs_try_tree_write_lock(eb)) {
3175 btrfs_tree_lock(eb);
3177 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3179 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3180 spin_lock(&root->fs_info->delalloc_lock);
3181 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3182 root->fs_info->dirty_metadata_bytes -= eb->len;
3185 spin_unlock(&root->fs_info->delalloc_lock);
3188 btrfs_tree_unlock(eb);
3190 if (!trylock_page(page)) {
3197 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3203 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3204 printk(KERN_WARNING "warning: mount fs with errors, "
3205 "running btrfsck is recommended\n");
3208 int btrfs_error_commit_super(struct btrfs_root *root)
3212 mutex_lock(&root->fs_info->cleaner_mutex);
3213 btrfs_run_delayed_iputs(root);
3214 mutex_unlock(&root->fs_info->cleaner_mutex);
3216 down_write(&root->fs_info->cleanup_work_sem);
3217 up_write(&root->fs_info->cleanup_work_sem);
3219 /* cleanup FS via transaction */
3220 btrfs_cleanup_transaction(root);
3222 ret = write_ctree_super(NULL, root, 0);
3227 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3229 struct btrfs_inode *btrfs_inode;
3230 struct list_head splice;
3232 INIT_LIST_HEAD(&splice);
3234 mutex_lock(&root->fs_info->ordered_operations_mutex);
3235 spin_lock(&root->fs_info->ordered_extent_lock);
3237 list_splice_init(&root->fs_info->ordered_operations, &splice);
3238 while (!list_empty(&splice)) {
3239 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3240 ordered_operations);
3242 list_del_init(&btrfs_inode->ordered_operations);
3244 btrfs_invalidate_inodes(btrfs_inode->root);
3247 spin_unlock(&root->fs_info->ordered_extent_lock);
3248 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3253 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3255 struct list_head splice;
3256 struct btrfs_ordered_extent *ordered;
3257 struct inode *inode;
3259 INIT_LIST_HEAD(&splice);
3261 spin_lock(&root->fs_info->ordered_extent_lock);
3263 list_splice_init(&root->fs_info->ordered_extents, &splice);
3264 while (!list_empty(&splice)) {
3265 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3268 list_del_init(&ordered->root_extent_list);
3269 atomic_inc(&ordered->refs);
3271 /* the inode may be getting freed (in sys_unlink path). */
3272 inode = igrab(ordered->inode);
3274 spin_unlock(&root->fs_info->ordered_extent_lock);
3278 atomic_set(&ordered->refs, 1);
3279 btrfs_put_ordered_extent(ordered);
3281 spin_lock(&root->fs_info->ordered_extent_lock);
3284 spin_unlock(&root->fs_info->ordered_extent_lock);
3289 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3290 struct btrfs_root *root)
3292 struct rb_node *node;
3293 struct btrfs_delayed_ref_root *delayed_refs;
3294 struct btrfs_delayed_ref_node *ref;
3297 delayed_refs = &trans->delayed_refs;
3299 spin_lock(&delayed_refs->lock);
3300 if (delayed_refs->num_entries == 0) {
3301 spin_unlock(&delayed_refs->lock);
3302 printk(KERN_INFO "delayed_refs has NO entry\n");
3306 node = rb_first(&delayed_refs->root);
3308 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3309 node = rb_next(node);
3312 rb_erase(&ref->rb_node, &delayed_refs->root);
3313 delayed_refs->num_entries--;
3315 atomic_set(&ref->refs, 1);
3316 if (btrfs_delayed_ref_is_head(ref)) {
3317 struct btrfs_delayed_ref_head *head;
3319 head = btrfs_delayed_node_to_head(ref);
3320 mutex_lock(&head->mutex);
3321 kfree(head->extent_op);
3322 delayed_refs->num_heads--;
3323 if (list_empty(&head->cluster))
3324 delayed_refs->num_heads_ready--;
3325 list_del_init(&head->cluster);
3326 mutex_unlock(&head->mutex);
3329 spin_unlock(&delayed_refs->lock);
3330 btrfs_put_delayed_ref(ref);
3333 spin_lock(&delayed_refs->lock);
3336 spin_unlock(&delayed_refs->lock);
3341 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3343 struct btrfs_pending_snapshot *snapshot;
3344 struct list_head splice;
3346 INIT_LIST_HEAD(&splice);
3348 list_splice_init(&t->pending_snapshots, &splice);
3350 while (!list_empty(&splice)) {
3351 snapshot = list_entry(splice.next,
3352 struct btrfs_pending_snapshot,
3355 list_del_init(&snapshot->list);
3363 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3365 struct btrfs_inode *btrfs_inode;
3366 struct list_head splice;
3368 INIT_LIST_HEAD(&splice);
3370 spin_lock(&root->fs_info->delalloc_lock);
3371 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3373 while (!list_empty(&splice)) {
3374 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3377 list_del_init(&btrfs_inode->delalloc_inodes);
3379 btrfs_invalidate_inodes(btrfs_inode->root);
3382 spin_unlock(&root->fs_info->delalloc_lock);
3387 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3388 struct extent_io_tree *dirty_pages,
3393 struct inode *btree_inode = root->fs_info->btree_inode;
3394 struct extent_buffer *eb;
3398 unsigned long index;
3401 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3406 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3407 while (start <= end) {
3408 index = start >> PAGE_CACHE_SHIFT;
3409 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3410 page = find_get_page(btree_inode->i_mapping, index);
3413 offset = page_offset(page);
3415 spin_lock(&dirty_pages->buffer_lock);
3416 eb = radix_tree_lookup(
3417 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3418 offset >> PAGE_CACHE_SHIFT);
3419 spin_unlock(&dirty_pages->buffer_lock);
3421 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3423 atomic_set(&eb->refs, 1);
3425 if (PageWriteback(page))
3426 end_page_writeback(page);
3429 if (PageDirty(page)) {
3430 clear_page_dirty_for_io(page);
3431 spin_lock_irq(&page->mapping->tree_lock);
3432 radix_tree_tag_clear(&page->mapping->page_tree,
3434 PAGECACHE_TAG_DIRTY);
3435 spin_unlock_irq(&page->mapping->tree_lock);
3438 page->mapping->a_ops->invalidatepage(page, 0);
3446 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3447 struct extent_io_tree *pinned_extents)
3449 struct extent_io_tree *unpin;
3454 unpin = pinned_extents;
3456 ret = find_first_extent_bit(unpin, 0, &start, &end,
3462 if (btrfs_test_opt(root, DISCARD))
3463 ret = btrfs_error_discard_extent(root, start,
3467 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3468 btrfs_error_unpin_extent_range(root, start, end);
3475 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3477 struct btrfs_transaction *t;
3482 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3484 spin_lock(&root->fs_info->trans_lock);
3485 list_splice_init(&root->fs_info->trans_list, &list);
3486 root->fs_info->trans_no_join = 1;
3487 spin_unlock(&root->fs_info->trans_lock);
3489 while (!list_empty(&list)) {
3490 t = list_entry(list.next, struct btrfs_transaction, list);
3494 btrfs_destroy_ordered_operations(root);
3496 btrfs_destroy_ordered_extents(root);
3498 btrfs_destroy_delayed_refs(t, root);
3500 btrfs_block_rsv_release(root,
3501 &root->fs_info->trans_block_rsv,
3502 t->dirty_pages.dirty_bytes);
3504 /* FIXME: cleanup wait for commit */
3507 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3508 wake_up(&root->fs_info->transaction_blocked_wait);
3511 if (waitqueue_active(&root->fs_info->transaction_wait))
3512 wake_up(&root->fs_info->transaction_wait);
3515 if (waitqueue_active(&t->commit_wait))
3516 wake_up(&t->commit_wait);
3518 btrfs_destroy_pending_snapshots(t);
3520 btrfs_destroy_delalloc_inodes(root);
3522 spin_lock(&root->fs_info->trans_lock);
3523 root->fs_info->running_transaction = NULL;
3524 spin_unlock(&root->fs_info->trans_lock);
3526 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3529 btrfs_destroy_pinned_extent(root,
3530 root->fs_info->pinned_extents);
3532 atomic_set(&t->use_count, 0);
3533 list_del_init(&t->list);
3534 memset(t, 0, sizeof(*t));
3535 kmem_cache_free(btrfs_transaction_cachep, t);
3538 spin_lock(&root->fs_info->trans_lock);
3539 root->fs_info->trans_no_join = 0;
3540 spin_unlock(&root->fs_info->trans_lock);
3541 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3546 static struct extent_io_ops btree_extent_io_ops = {
3547 .write_cache_pages_lock_hook = btree_lock_page_hook,
3548 .readpage_end_io_hook = btree_readpage_end_io_hook,
3549 .readpage_io_failed_hook = btree_io_failed_hook,
3550 .submit_bio_hook = btree_submit_bio_hook,
3551 /* note we're sharing with inode.c for the merge bio hook */
3552 .merge_bio_hook = btrfs_merge_bio_hook,