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 <linux/uuid.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
42 #include "async-thread.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
53 #include <asm/cpufeature.h>
56 static struct extent_io_ops btree_extent_io_ops;
57 static void end_workqueue_fn(struct btrfs_work *work);
58 static void free_fs_root(struct btrfs_root *root);
59 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
61 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
62 struct btrfs_root *root);
63 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_root *root);
66 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
74 static void btrfs_error_commit_super(struct btrfs_root *root);
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
85 struct btrfs_fs_info *info;
88 struct list_head list;
89 struct btrfs_work work;
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
97 struct async_submit_bio {
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
105 unsigned long bio_flags;
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
111 struct btrfs_work work;
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
139 # if BTRFS_MAX_LEVEL != 8
143 static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148 } btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = 0, .name_stem = "tree" },
162 void __init btrfs_init_lockdep(void)
166 /* initialize lockdep class names */
167 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
168 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
171 snprintf(ks->names[j], sizeof(ks->names[j]),
172 "btrfs-%s-%02d", ks->name_stem, j);
176 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
179 struct btrfs_lockdep_keyset *ks;
181 BUG_ON(level >= ARRAY_SIZE(ks->keys));
183 /* find the matching keyset, id 0 is the default entry */
184 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
185 if (ks->id == objectid)
188 lockdep_set_class_and_name(&eb->lock,
189 &ks->keys[level], ks->names[level]);
195 * extents on the btree inode are pretty simple, there's one extent
196 * that covers the entire device
198 static struct extent_map *btree_get_extent(struct inode *inode,
199 struct page *page, size_t pg_offset, u64 start, u64 len,
202 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
203 struct extent_map *em;
206 read_lock(&em_tree->lock);
207 em = lookup_extent_mapping(em_tree, start, len);
210 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
211 read_unlock(&em_tree->lock);
214 read_unlock(&em_tree->lock);
216 em = alloc_extent_map();
218 em = ERR_PTR(-ENOMEM);
223 em->block_len = (u64)-1;
225 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
227 write_lock(&em_tree->lock);
228 ret = add_extent_mapping(em_tree, em, 0);
229 if (ret == -EEXIST) {
231 em = lookup_extent_mapping(em_tree, start, len);
238 write_unlock(&em_tree->lock);
244 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
246 return crc32c(seed, data, len);
249 void btrfs_csum_final(u32 crc, char *result)
251 put_unaligned_le32(~crc, result);
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
258 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
267 unsigned long map_start;
268 unsigned long map_len;
271 unsigned long inline_result;
273 len = buf->len - offset;
275 err = map_private_extent_buffer(buf, offset, 32,
276 &kaddr, &map_start, &map_len);
279 cur_len = min(len, map_len - (offset - map_start));
280 crc = btrfs_csum_data(kaddr + offset - map_start,
285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
290 result = (char *)&inline_result;
293 btrfs_csum_final(crc, result);
296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
299 memcpy(&found, result, csum_size);
301 read_extent_buffer(buf, &val, 0, csum_size);
302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303 "failed on %llu wanted %X found %X "
305 root->fs_info->sb->s_id,
306 (unsigned long long)buf->start, val, found,
307 btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
313 write_extent_buffer(buf, result, 0, csum_size);
315 if (result != (char *)&inline_result)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
330 struct extent_state *cached_state = NULL;
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 (unsigned long long)eb->start,
349 (unsigned long long)parent_transid,
350 (unsigned long long)btrfs_header_generation(eb));
352 clear_extent_buffer_uptodate(eb);
354 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 &cached_state, GFP_NOFS);
360 * Return 0 if the superblock checksum type matches the checksum value of that
361 * algorithm. Pass the raw disk superblock data.
363 static int btrfs_check_super_csum(char *raw_disk_sb)
365 struct btrfs_super_block *disk_sb =
366 (struct btrfs_super_block *)raw_disk_sb;
367 u16 csum_type = btrfs_super_csum_type(disk_sb);
370 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
372 const int csum_size = sizeof(crc);
373 char result[csum_size];
376 * The super_block structure does not span the whole
377 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378 * is filled with zeros and is included in the checkum.
380 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
381 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382 btrfs_csum_final(crc, result);
384 if (memcmp(raw_disk_sb, result, csum_size))
387 if (ret && btrfs_super_generation(disk_sb) < 10) {
388 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
393 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
394 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
403 * helper to read a given tree block, doing retries as required when
404 * the checksums don't match and we have alternate mirrors to try.
406 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
407 struct extent_buffer *eb,
408 u64 start, u64 parent_transid)
410 struct extent_io_tree *io_tree;
415 int failed_mirror = 0;
417 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
418 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
420 ret = read_extent_buffer_pages(io_tree, eb, start,
422 btree_get_extent, mirror_num);
424 if (!verify_parent_transid(io_tree, eb,
432 * This buffer's crc is fine, but its contents are corrupted, so
433 * there is no reason to read the other copies, they won't be
436 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
439 num_copies = btrfs_num_copies(root->fs_info,
444 if (!failed_mirror) {
446 failed_mirror = eb->read_mirror;
450 if (mirror_num == failed_mirror)
453 if (mirror_num > num_copies)
457 if (failed && !ret && failed_mirror)
458 repair_eb_io_failure(root, eb, failed_mirror);
464 * checksum a dirty tree block before IO. This has extra checks to make sure
465 * we only fill in the checksum field in the first page of a multi-page block
468 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
470 struct extent_io_tree *tree;
471 u64 start = page_offset(page);
473 struct extent_buffer *eb;
475 tree = &BTRFS_I(page->mapping->host)->io_tree;
477 eb = (struct extent_buffer *)page->private;
478 if (page != eb->pages[0])
480 found_start = btrfs_header_bytenr(eb);
481 if (found_start != start) {
485 if (!PageUptodate(page)) {
489 csum_tree_block(root, eb, 0);
493 static int check_tree_block_fsid(struct btrfs_root *root,
494 struct extent_buffer *eb)
496 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
497 u8 fsid[BTRFS_UUID_SIZE];
500 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
503 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
507 fs_devices = fs_devices->seed;
512 #define CORRUPT(reason, eb, root, slot) \
513 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
514 "root=%llu, slot=%d\n", reason, \
515 (unsigned long long)btrfs_header_bytenr(eb), \
516 (unsigned long long)root->objectid, slot)
518 static noinline int check_leaf(struct btrfs_root *root,
519 struct extent_buffer *leaf)
521 struct btrfs_key key;
522 struct btrfs_key leaf_key;
523 u32 nritems = btrfs_header_nritems(leaf);
529 /* Check the 0 item */
530 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("invalid item offset size pair", leaf, root, 0);
537 * Check to make sure each items keys are in the correct order and their
538 * offsets make sense. We only have to loop through nritems-1 because
539 * we check the current slot against the next slot, which verifies the
540 * next slot's offset+size makes sense and that the current's slot
543 for (slot = 0; slot < nritems - 1; slot++) {
544 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
545 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
547 /* Make sure the keys are in the right order */
548 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
549 CORRUPT("bad key order", leaf, root, slot);
554 * Make sure the offset and ends are right, remember that the
555 * item data starts at the end of the leaf and grows towards the
558 if (btrfs_item_offset_nr(leaf, slot) !=
559 btrfs_item_end_nr(leaf, slot + 1)) {
560 CORRUPT("slot offset bad", leaf, root, slot);
565 * Check to make sure that we don't point outside of the leaf,
566 * just incase all the items are consistent to eachother, but
567 * all point outside of the leaf.
569 if (btrfs_item_end_nr(leaf, slot) >
570 BTRFS_LEAF_DATA_SIZE(root)) {
571 CORRUPT("slot end outside of leaf", leaf, root, slot);
579 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
580 u64 phy_offset, struct page *page,
581 u64 start, u64 end, int mirror)
583 struct extent_io_tree *tree;
586 struct extent_buffer *eb;
587 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
594 tree = &BTRFS_I(page->mapping->host)->io_tree;
595 eb = (struct extent_buffer *)page->private;
597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
600 extent_buffer_get(eb);
602 reads_done = atomic_dec_and_test(&eb->io_pages);
606 eb->read_mirror = mirror;
607 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
612 found_start = btrfs_header_bytenr(eb);
613 if (found_start != eb->start) {
614 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
616 (unsigned long long)found_start,
617 (unsigned long long)eb->start);
621 if (check_tree_block_fsid(root, eb)) {
622 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
623 (unsigned long long)eb->start);
627 found_level = btrfs_header_level(eb);
628 if (found_level >= BTRFS_MAX_LEVEL) {
629 btrfs_info(root->fs_info, "bad tree block level %d\n",
630 (int)btrfs_header_level(eb));
635 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
638 ret = csum_tree_block(root, eb, 1);
645 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 * that we don't try and read the other copies of this block, just
649 if (found_level == 0 && check_leaf(root, eb)) {
650 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
655 set_extent_buffer_uptodate(eb);
658 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
659 btree_readahead_hook(root, eb, eb->start, ret);
663 * our io error hook is going to dec the io pages
664 * again, we have to make sure it has something
667 atomic_inc(&eb->io_pages);
668 clear_extent_buffer_uptodate(eb);
670 free_extent_buffer(eb);
675 static int btree_io_failed_hook(struct page *page, int failed_mirror)
677 struct extent_buffer *eb;
678 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
680 eb = (struct extent_buffer *)page->private;
681 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
682 eb->read_mirror = failed_mirror;
683 atomic_dec(&eb->io_pages);
684 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
685 btree_readahead_hook(root, eb, eb->start, -EIO);
686 return -EIO; /* we fixed nothing */
689 static void end_workqueue_bio(struct bio *bio, int err)
691 struct end_io_wq *end_io_wq = bio->bi_private;
692 struct btrfs_fs_info *fs_info;
694 fs_info = end_io_wq->info;
695 end_io_wq->error = err;
696 end_io_wq->work.func = end_workqueue_fn;
697 end_io_wq->work.flags = 0;
699 if (bio->bi_rw & REQ_WRITE) {
700 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
701 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
703 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
704 btrfs_queue_worker(&fs_info->endio_freespace_worker,
706 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
707 btrfs_queue_worker(&fs_info->endio_raid56_workers,
710 btrfs_queue_worker(&fs_info->endio_write_workers,
713 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
714 btrfs_queue_worker(&fs_info->endio_raid56_workers,
716 else if (end_io_wq->metadata)
717 btrfs_queue_worker(&fs_info->endio_meta_workers,
720 btrfs_queue_worker(&fs_info->endio_workers,
726 * For the metadata arg you want
729 * 1 - if normal metadta
730 * 2 - if writing to the free space cache area
731 * 3 - raid parity work
733 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
736 struct end_io_wq *end_io_wq;
737 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
741 end_io_wq->private = bio->bi_private;
742 end_io_wq->end_io = bio->bi_end_io;
743 end_io_wq->info = info;
744 end_io_wq->error = 0;
745 end_io_wq->bio = bio;
746 end_io_wq->metadata = metadata;
748 bio->bi_private = end_io_wq;
749 bio->bi_end_io = end_workqueue_bio;
753 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
755 unsigned long limit = min_t(unsigned long,
756 info->workers.max_workers,
757 info->fs_devices->open_devices);
761 static void run_one_async_start(struct btrfs_work *work)
763 struct async_submit_bio *async;
766 async = container_of(work, struct async_submit_bio, work);
767 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
768 async->mirror_num, async->bio_flags,
774 static void run_one_async_done(struct btrfs_work *work)
776 struct btrfs_fs_info *fs_info;
777 struct async_submit_bio *async;
780 async = container_of(work, struct async_submit_bio, work);
781 fs_info = BTRFS_I(async->inode)->root->fs_info;
783 limit = btrfs_async_submit_limit(fs_info);
784 limit = limit * 2 / 3;
786 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
787 waitqueue_active(&fs_info->async_submit_wait))
788 wake_up(&fs_info->async_submit_wait);
790 /* If an error occured we just want to clean up the bio and move on */
792 bio_endio(async->bio, async->error);
796 async->submit_bio_done(async->inode, async->rw, async->bio,
797 async->mirror_num, async->bio_flags,
801 static void run_one_async_free(struct btrfs_work *work)
803 struct async_submit_bio *async;
805 async = container_of(work, struct async_submit_bio, work);
809 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
810 int rw, struct bio *bio, int mirror_num,
811 unsigned long bio_flags,
813 extent_submit_bio_hook_t *submit_bio_start,
814 extent_submit_bio_hook_t *submit_bio_done)
816 struct async_submit_bio *async;
818 async = kmalloc(sizeof(*async), GFP_NOFS);
822 async->inode = inode;
825 async->mirror_num = mirror_num;
826 async->submit_bio_start = submit_bio_start;
827 async->submit_bio_done = submit_bio_done;
829 async->work.func = run_one_async_start;
830 async->work.ordered_func = run_one_async_done;
831 async->work.ordered_free = run_one_async_free;
833 async->work.flags = 0;
834 async->bio_flags = bio_flags;
835 async->bio_offset = bio_offset;
839 atomic_inc(&fs_info->nr_async_submits);
842 btrfs_set_work_high_prio(&async->work);
844 btrfs_queue_worker(&fs_info->workers, &async->work);
846 while (atomic_read(&fs_info->async_submit_draining) &&
847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
855 static int btree_csum_one_bio(struct bio *bio)
857 struct bio_vec *bvec = bio->bi_io_vec;
859 struct btrfs_root *root;
862 WARN_ON(bio->bi_vcnt <= 0);
863 while (bio_index < bio->bi_vcnt) {
864 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
865 ret = csum_dirty_buffer(root, bvec->bv_page);
874 static int __btree_submit_bio_start(struct inode *inode, int rw,
875 struct bio *bio, int mirror_num,
876 unsigned long bio_flags,
880 * when we're called for a write, we're already in the async
881 * submission context. Just jump into btrfs_map_bio
883 return btree_csum_one_bio(bio);
886 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
887 int mirror_num, unsigned long bio_flags,
893 * when we're called for a write, we're already in the async
894 * submission context. Just jump into btrfs_map_bio
896 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
902 static int check_async_write(struct inode *inode, unsigned long bio_flags)
904 if (bio_flags & EXTENT_BIO_TREE_LOG)
913 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
914 int mirror_num, unsigned long bio_flags,
917 int async = check_async_write(inode, bio_flags);
920 if (!(rw & REQ_WRITE)) {
922 * called for a read, do the setup so that checksum validation
923 * can happen in the async kernel threads
925 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
929 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
932 ret = btree_csum_one_bio(bio);
935 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
942 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
943 inode, rw, bio, mirror_num, 0,
945 __btree_submit_bio_start,
946 __btree_submit_bio_done);
956 #ifdef CONFIG_MIGRATION
957 static int btree_migratepage(struct address_space *mapping,
958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
974 return migrate_page(mapping, newpage, page, mode);
979 static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
982 struct extent_io_tree *tree;
983 struct btrfs_fs_info *fs_info;
986 tree = &BTRFS_I(mapping->host)->io_tree;
987 if (wbc->sync_mode == WB_SYNC_NONE) {
989 if (wbc->for_kupdate)
992 fs_info = BTRFS_I(mapping->host)->root->fs_info;
993 /* this is a bit racy, but that's ok */
994 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
995 BTRFS_DIRTY_METADATA_THRESH);
999 return btree_write_cache_pages(mapping, wbc);
1002 static int btree_readpage(struct file *file, struct page *page)
1004 struct extent_io_tree *tree;
1005 tree = &BTRFS_I(page->mapping->host)->io_tree;
1006 return extent_read_full_page(tree, page, btree_get_extent, 0);
1009 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1011 if (PageWriteback(page) || PageDirty(page))
1014 return try_release_extent_buffer(page);
1017 static void btree_invalidatepage(struct page *page, unsigned int offset,
1018 unsigned int length)
1020 struct extent_io_tree *tree;
1021 tree = &BTRFS_I(page->mapping->host)->io_tree;
1022 extent_invalidatepage(tree, page, offset);
1023 btree_releasepage(page, GFP_NOFS);
1024 if (PagePrivate(page)) {
1025 printk(KERN_WARNING "btrfs warning page private not zero "
1026 "on page %llu\n", (unsigned long long)page_offset(page));
1027 ClearPagePrivate(page);
1028 set_page_private(page, 0);
1029 page_cache_release(page);
1033 static int btree_set_page_dirty(struct page *page)
1036 struct extent_buffer *eb;
1038 BUG_ON(!PagePrivate(page));
1039 eb = (struct extent_buffer *)page->private;
1041 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1042 BUG_ON(!atomic_read(&eb->refs));
1043 btrfs_assert_tree_locked(eb);
1045 return __set_page_dirty_nobuffers(page);
1048 static const struct address_space_operations btree_aops = {
1049 .readpage = btree_readpage,
1050 .writepages = btree_writepages,
1051 .releasepage = btree_releasepage,
1052 .invalidatepage = btree_invalidatepage,
1053 #ifdef CONFIG_MIGRATION
1054 .migratepage = btree_migratepage,
1056 .set_page_dirty = btree_set_page_dirty,
1059 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1062 struct extent_buffer *buf = NULL;
1063 struct inode *btree_inode = root->fs_info->btree_inode;
1066 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1069 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1070 buf, 0, WAIT_NONE, btree_get_extent, 0);
1071 free_extent_buffer(buf);
1075 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1076 int mirror_num, struct extent_buffer **eb)
1078 struct extent_buffer *buf = NULL;
1079 struct inode *btree_inode = root->fs_info->btree_inode;
1080 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1083 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1087 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1089 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1090 btree_get_extent, mirror_num);
1092 free_extent_buffer(buf);
1096 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1097 free_extent_buffer(buf);
1099 } else if (extent_buffer_uptodate(buf)) {
1102 free_extent_buffer(buf);
1107 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1108 u64 bytenr, u32 blocksize)
1110 struct inode *btree_inode = root->fs_info->btree_inode;
1111 struct extent_buffer *eb;
1112 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1117 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1118 u64 bytenr, u32 blocksize)
1120 struct inode *btree_inode = root->fs_info->btree_inode;
1121 struct extent_buffer *eb;
1123 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1129 int btrfs_write_tree_block(struct extent_buffer *buf)
1131 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1132 buf->start + buf->len - 1);
1135 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1137 return filemap_fdatawait_range(buf->pages[0]->mapping,
1138 buf->start, buf->start + buf->len - 1);
1141 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1142 u32 blocksize, u64 parent_transid)
1144 struct extent_buffer *buf = NULL;
1147 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1151 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1153 free_extent_buffer(buf);
1160 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1161 struct extent_buffer *buf)
1163 struct btrfs_fs_info *fs_info = root->fs_info;
1165 if (btrfs_header_generation(buf) ==
1166 fs_info->running_transaction->transid) {
1167 btrfs_assert_tree_locked(buf);
1169 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1170 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1172 fs_info->dirty_metadata_batch);
1173 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1174 btrfs_set_lock_blocking(buf);
1175 clear_extent_buffer_dirty(buf);
1180 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1181 u32 stripesize, struct btrfs_root *root,
1182 struct btrfs_fs_info *fs_info,
1186 root->commit_root = NULL;
1187 root->sectorsize = sectorsize;
1188 root->nodesize = nodesize;
1189 root->leafsize = leafsize;
1190 root->stripesize = stripesize;
1192 root->track_dirty = 0;
1194 root->orphan_item_inserted = 0;
1195 root->orphan_cleanup_state = 0;
1197 root->objectid = objectid;
1198 root->last_trans = 0;
1199 root->highest_objectid = 0;
1200 root->nr_delalloc_inodes = 0;
1201 root->nr_ordered_extents = 0;
1203 root->inode_tree = RB_ROOT;
1204 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1205 root->block_rsv = NULL;
1206 root->orphan_block_rsv = NULL;
1208 INIT_LIST_HEAD(&root->dirty_list);
1209 INIT_LIST_HEAD(&root->root_list);
1210 INIT_LIST_HEAD(&root->delalloc_inodes);
1211 INIT_LIST_HEAD(&root->delalloc_root);
1212 INIT_LIST_HEAD(&root->ordered_extents);
1213 INIT_LIST_HEAD(&root->ordered_root);
1214 INIT_LIST_HEAD(&root->logged_list[0]);
1215 INIT_LIST_HEAD(&root->logged_list[1]);
1216 spin_lock_init(&root->orphan_lock);
1217 spin_lock_init(&root->inode_lock);
1218 spin_lock_init(&root->delalloc_lock);
1219 spin_lock_init(&root->ordered_extent_lock);
1220 spin_lock_init(&root->accounting_lock);
1221 spin_lock_init(&root->log_extents_lock[0]);
1222 spin_lock_init(&root->log_extents_lock[1]);
1223 mutex_init(&root->objectid_mutex);
1224 mutex_init(&root->log_mutex);
1225 init_waitqueue_head(&root->log_writer_wait);
1226 init_waitqueue_head(&root->log_commit_wait[0]);
1227 init_waitqueue_head(&root->log_commit_wait[1]);
1228 atomic_set(&root->log_commit[0], 0);
1229 atomic_set(&root->log_commit[1], 0);
1230 atomic_set(&root->log_writers, 0);
1231 atomic_set(&root->log_batch, 0);
1232 atomic_set(&root->orphan_inodes, 0);
1233 atomic_set(&root->refs, 1);
1234 root->log_transid = 0;
1235 root->last_log_commit = 0;
1236 extent_io_tree_init(&root->dirty_log_pages,
1237 fs_info->btree_inode->i_mapping);
1239 memset(&root->root_key, 0, sizeof(root->root_key));
1240 memset(&root->root_item, 0, sizeof(root->root_item));
1241 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1242 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1243 root->defrag_trans_start = fs_info->generation;
1244 init_completion(&root->kobj_unregister);
1245 root->defrag_running = 0;
1246 root->root_key.objectid = objectid;
1249 spin_lock_init(&root->root_item_lock);
1252 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1254 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1256 root->fs_info = fs_info;
1260 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1261 struct btrfs_fs_info *fs_info,
1264 struct extent_buffer *leaf;
1265 struct btrfs_root *tree_root = fs_info->tree_root;
1266 struct btrfs_root *root;
1267 struct btrfs_key key;
1272 root = btrfs_alloc_root(fs_info);
1274 return ERR_PTR(-ENOMEM);
1276 __setup_root(tree_root->nodesize, tree_root->leafsize,
1277 tree_root->sectorsize, tree_root->stripesize,
1278 root, fs_info, objectid);
1279 root->root_key.objectid = objectid;
1280 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1281 root->root_key.offset = 0;
1283 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1284 0, objectid, NULL, 0, 0, 0);
1286 ret = PTR_ERR(leaf);
1291 bytenr = leaf->start;
1292 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1293 btrfs_set_header_bytenr(leaf, leaf->start);
1294 btrfs_set_header_generation(leaf, trans->transid);
1295 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1296 btrfs_set_header_owner(leaf, objectid);
1299 write_extent_buffer(leaf, fs_info->fsid,
1300 (unsigned long)btrfs_header_fsid(leaf),
1302 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1303 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1305 btrfs_mark_buffer_dirty(leaf);
1307 root->commit_root = btrfs_root_node(root);
1308 root->track_dirty = 1;
1311 root->root_item.flags = 0;
1312 root->root_item.byte_limit = 0;
1313 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1314 btrfs_set_root_generation(&root->root_item, trans->transid);
1315 btrfs_set_root_level(&root->root_item, 0);
1316 btrfs_set_root_refs(&root->root_item, 1);
1317 btrfs_set_root_used(&root->root_item, leaf->len);
1318 btrfs_set_root_last_snapshot(&root->root_item, 0);
1319 btrfs_set_root_dirid(&root->root_item, 0);
1321 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1322 root->root_item.drop_level = 0;
1324 key.objectid = objectid;
1325 key.type = BTRFS_ROOT_ITEM_KEY;
1327 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1331 btrfs_tree_unlock(leaf);
1337 btrfs_tree_unlock(leaf);
1338 free_extent_buffer(leaf);
1342 return ERR_PTR(ret);
1345 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1346 struct btrfs_fs_info *fs_info)
1348 struct btrfs_root *root;
1349 struct btrfs_root *tree_root = fs_info->tree_root;
1350 struct extent_buffer *leaf;
1352 root = btrfs_alloc_root(fs_info);
1354 return ERR_PTR(-ENOMEM);
1356 __setup_root(tree_root->nodesize, tree_root->leafsize,
1357 tree_root->sectorsize, tree_root->stripesize,
1358 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1360 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1361 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1362 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1364 * log trees do not get reference counted because they go away
1365 * before a real commit is actually done. They do store pointers
1366 * to file data extents, and those reference counts still get
1367 * updated (along with back refs to the log tree).
1371 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1372 BTRFS_TREE_LOG_OBJECTID, NULL,
1376 return ERR_CAST(leaf);
1379 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1380 btrfs_set_header_bytenr(leaf, leaf->start);
1381 btrfs_set_header_generation(leaf, trans->transid);
1382 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1383 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1386 write_extent_buffer(root->node, root->fs_info->fsid,
1387 (unsigned long)btrfs_header_fsid(root->node),
1389 btrfs_mark_buffer_dirty(root->node);
1390 btrfs_tree_unlock(root->node);
1394 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1395 struct btrfs_fs_info *fs_info)
1397 struct btrfs_root *log_root;
1399 log_root = alloc_log_tree(trans, fs_info);
1400 if (IS_ERR(log_root))
1401 return PTR_ERR(log_root);
1402 WARN_ON(fs_info->log_root_tree);
1403 fs_info->log_root_tree = log_root;
1407 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1408 struct btrfs_root *root)
1410 struct btrfs_root *log_root;
1411 struct btrfs_inode_item *inode_item;
1413 log_root = alloc_log_tree(trans, root->fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1417 log_root->last_trans = trans->transid;
1418 log_root->root_key.offset = root->root_key.objectid;
1420 inode_item = &log_root->root_item.inode;
1421 btrfs_set_stack_inode_generation(inode_item, 1);
1422 btrfs_set_stack_inode_size(inode_item, 3);
1423 btrfs_set_stack_inode_nlink(inode_item, 1);
1424 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1425 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1427 btrfs_set_root_node(&log_root->root_item, log_root->node);
1429 WARN_ON(root->log_root);
1430 root->log_root = log_root;
1431 root->log_transid = 0;
1432 root->last_log_commit = 0;
1436 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1437 struct btrfs_key *key)
1439 struct btrfs_root *root;
1440 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1441 struct btrfs_path *path;
1446 path = btrfs_alloc_path();
1448 return ERR_PTR(-ENOMEM);
1450 root = btrfs_alloc_root(fs_info);
1456 __setup_root(tree_root->nodesize, tree_root->leafsize,
1457 tree_root->sectorsize, tree_root->stripesize,
1458 root, fs_info, key->objectid);
1460 ret = btrfs_find_root(tree_root, key, path,
1461 &root->root_item, &root->root_key);
1468 generation = btrfs_root_generation(&root->root_item);
1469 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1470 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1471 blocksize, generation);
1475 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1479 root->commit_root = btrfs_root_node(root);
1481 btrfs_free_path(path);
1485 free_extent_buffer(root->node);
1489 root = ERR_PTR(ret);
1493 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1494 struct btrfs_key *location)
1496 struct btrfs_root *root;
1498 root = btrfs_read_tree_root(tree_root, location);
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1504 btrfs_check_and_init_root_item(&root->root_item);
1510 int btrfs_init_fs_root(struct btrfs_root *root)
1514 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1515 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1517 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1522 btrfs_init_free_ino_ctl(root);
1523 mutex_init(&root->fs_commit_mutex);
1524 spin_lock_init(&root->cache_lock);
1525 init_waitqueue_head(&root->cache_wait);
1527 ret = get_anon_bdev(&root->anon_dev);
1532 kfree(root->free_ino_ctl);
1533 kfree(root->free_ino_pinned);
1537 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1540 struct btrfs_root *root;
1542 spin_lock(&fs_info->fs_roots_radix_lock);
1543 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1544 (unsigned long)root_id);
1545 spin_unlock(&fs_info->fs_roots_radix_lock);
1549 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1550 struct btrfs_root *root)
1554 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1558 spin_lock(&fs_info->fs_roots_radix_lock);
1559 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1560 (unsigned long)root->root_key.objectid,
1564 spin_unlock(&fs_info->fs_roots_radix_lock);
1565 radix_tree_preload_end();
1570 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1571 struct btrfs_key *location)
1573 struct btrfs_root *root;
1576 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1577 return fs_info->tree_root;
1578 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1579 return fs_info->extent_root;
1580 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1581 return fs_info->chunk_root;
1582 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1583 return fs_info->dev_root;
1584 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1585 return fs_info->csum_root;
1586 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1587 return fs_info->quota_root ? fs_info->quota_root :
1590 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1594 root = btrfs_read_fs_root(fs_info->tree_root, location);
1598 if (btrfs_root_refs(&root->root_item) == 0) {
1603 ret = btrfs_init_fs_root(root);
1607 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1611 root->orphan_item_inserted = 1;
1613 ret = btrfs_insert_fs_root(fs_info, root);
1615 if (ret == -EEXIST) {
1624 return ERR_PTR(ret);
1627 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1629 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1631 struct btrfs_device *device;
1632 struct backing_dev_info *bdi;
1635 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1638 bdi = blk_get_backing_dev_info(device->bdev);
1639 if (bdi && bdi_congested(bdi, bdi_bits)) {
1649 * If this fails, caller must call bdi_destroy() to get rid of the
1652 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1656 bdi->capabilities = BDI_CAP_MAP_COPY;
1657 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1661 bdi->ra_pages = default_backing_dev_info.ra_pages;
1662 bdi->congested_fn = btrfs_congested_fn;
1663 bdi->congested_data = info;
1668 * called by the kthread helper functions to finally call the bio end_io
1669 * functions. This is where read checksum verification actually happens
1671 static void end_workqueue_fn(struct btrfs_work *work)
1674 struct end_io_wq *end_io_wq;
1675 struct btrfs_fs_info *fs_info;
1678 end_io_wq = container_of(work, struct end_io_wq, work);
1679 bio = end_io_wq->bio;
1680 fs_info = end_io_wq->info;
1682 error = end_io_wq->error;
1683 bio->bi_private = end_io_wq->private;
1684 bio->bi_end_io = end_io_wq->end_io;
1686 bio_endio(bio, error);
1689 static int cleaner_kthread(void *arg)
1691 struct btrfs_root *root = arg;
1697 /* Make the cleaner go to sleep early. */
1698 if (btrfs_need_cleaner_sleep(root))
1701 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1705 * Avoid the problem that we change the status of the fs
1706 * during the above check and trylock.
1708 if (btrfs_need_cleaner_sleep(root)) {
1709 mutex_unlock(&root->fs_info->cleaner_mutex);
1713 btrfs_run_delayed_iputs(root);
1714 again = btrfs_clean_one_deleted_snapshot(root);
1715 mutex_unlock(&root->fs_info->cleaner_mutex);
1718 * The defragger has dealt with the R/O remount and umount,
1719 * needn't do anything special here.
1721 btrfs_run_defrag_inodes(root->fs_info);
1723 if (!try_to_freeze() && !again) {
1724 set_current_state(TASK_INTERRUPTIBLE);
1725 if (!kthread_should_stop())
1727 __set_current_state(TASK_RUNNING);
1729 } while (!kthread_should_stop());
1733 static int transaction_kthread(void *arg)
1735 struct btrfs_root *root = arg;
1736 struct btrfs_trans_handle *trans;
1737 struct btrfs_transaction *cur;
1740 unsigned long delay;
1744 cannot_commit = false;
1745 delay = HZ * root->fs_info->commit_interval;
1746 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1748 spin_lock(&root->fs_info->trans_lock);
1749 cur = root->fs_info->running_transaction;
1751 spin_unlock(&root->fs_info->trans_lock);
1755 now = get_seconds();
1756 if (cur->state < TRANS_STATE_BLOCKED &&
1757 (now < cur->start_time ||
1758 now - cur->start_time < root->fs_info->commit_interval)) {
1759 spin_unlock(&root->fs_info->trans_lock);
1763 transid = cur->transid;
1764 spin_unlock(&root->fs_info->trans_lock);
1766 /* If the file system is aborted, this will always fail. */
1767 trans = btrfs_attach_transaction(root);
1768 if (IS_ERR(trans)) {
1769 if (PTR_ERR(trans) != -ENOENT)
1770 cannot_commit = true;
1773 if (transid == trans->transid) {
1774 btrfs_commit_transaction(trans, root);
1776 btrfs_end_transaction(trans, root);
1779 wake_up_process(root->fs_info->cleaner_kthread);
1780 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1782 if (!try_to_freeze()) {
1783 set_current_state(TASK_INTERRUPTIBLE);
1784 if (!kthread_should_stop() &&
1785 (!btrfs_transaction_blocked(root->fs_info) ||
1787 schedule_timeout(delay);
1788 __set_current_state(TASK_RUNNING);
1790 } while (!kthread_should_stop());
1795 * this will find the highest generation in the array of
1796 * root backups. The index of the highest array is returned,
1797 * or -1 if we can't find anything.
1799 * We check to make sure the array is valid by comparing the
1800 * generation of the latest root in the array with the generation
1801 * in the super block. If they don't match we pitch it.
1803 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1806 int newest_index = -1;
1807 struct btrfs_root_backup *root_backup;
1810 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811 root_backup = info->super_copy->super_roots + i;
1812 cur = btrfs_backup_tree_root_gen(root_backup);
1813 if (cur == newest_gen)
1817 /* check to see if we actually wrapped around */
1818 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819 root_backup = info->super_copy->super_roots;
1820 cur = btrfs_backup_tree_root_gen(root_backup);
1821 if (cur == newest_gen)
1824 return newest_index;
1829 * find the oldest backup so we know where to store new entries
1830 * in the backup array. This will set the backup_root_index
1831 * field in the fs_info struct
1833 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1836 int newest_index = -1;
1838 newest_index = find_newest_super_backup(info, newest_gen);
1839 /* if there was garbage in there, just move along */
1840 if (newest_index == -1) {
1841 info->backup_root_index = 0;
1843 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1852 static void backup_super_roots(struct btrfs_fs_info *info)
1855 struct btrfs_root_backup *root_backup;
1858 next_backup = info->backup_root_index;
1859 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860 BTRFS_NUM_BACKUP_ROOTS;
1863 * just overwrite the last backup if we're at the same generation
1864 * this happens only at umount
1866 root_backup = info->super_for_commit->super_roots + last_backup;
1867 if (btrfs_backup_tree_root_gen(root_backup) ==
1868 btrfs_header_generation(info->tree_root->node))
1869 next_backup = last_backup;
1871 root_backup = info->super_for_commit->super_roots + next_backup;
1874 * make sure all of our padding and empty slots get zero filled
1875 * regardless of which ones we use today
1877 memset(root_backup, 0, sizeof(*root_backup));
1879 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1881 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882 btrfs_set_backup_tree_root_gen(root_backup,
1883 btrfs_header_generation(info->tree_root->node));
1885 btrfs_set_backup_tree_root_level(root_backup,
1886 btrfs_header_level(info->tree_root->node));
1888 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889 btrfs_set_backup_chunk_root_gen(root_backup,
1890 btrfs_header_generation(info->chunk_root->node));
1891 btrfs_set_backup_chunk_root_level(root_backup,
1892 btrfs_header_level(info->chunk_root->node));
1894 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895 btrfs_set_backup_extent_root_gen(root_backup,
1896 btrfs_header_generation(info->extent_root->node));
1897 btrfs_set_backup_extent_root_level(root_backup,
1898 btrfs_header_level(info->extent_root->node));
1901 * we might commit during log recovery, which happens before we set
1902 * the fs_root. Make sure it is valid before we fill it in.
1904 if (info->fs_root && info->fs_root->node) {
1905 btrfs_set_backup_fs_root(root_backup,
1906 info->fs_root->node->start);
1907 btrfs_set_backup_fs_root_gen(root_backup,
1908 btrfs_header_generation(info->fs_root->node));
1909 btrfs_set_backup_fs_root_level(root_backup,
1910 btrfs_header_level(info->fs_root->node));
1913 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914 btrfs_set_backup_dev_root_gen(root_backup,
1915 btrfs_header_generation(info->dev_root->node));
1916 btrfs_set_backup_dev_root_level(root_backup,
1917 btrfs_header_level(info->dev_root->node));
1919 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920 btrfs_set_backup_csum_root_gen(root_backup,
1921 btrfs_header_generation(info->csum_root->node));
1922 btrfs_set_backup_csum_root_level(root_backup,
1923 btrfs_header_level(info->csum_root->node));
1925 btrfs_set_backup_total_bytes(root_backup,
1926 btrfs_super_total_bytes(info->super_copy));
1927 btrfs_set_backup_bytes_used(root_backup,
1928 btrfs_super_bytes_used(info->super_copy));
1929 btrfs_set_backup_num_devices(root_backup,
1930 btrfs_super_num_devices(info->super_copy));
1933 * if we don't copy this out to the super_copy, it won't get remembered
1934 * for the next commit
1936 memcpy(&info->super_copy->super_roots,
1937 &info->super_for_commit->super_roots,
1938 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1942 * this copies info out of the root backup array and back into
1943 * the in-memory super block. It is meant to help iterate through
1944 * the array, so you send it the number of backups you've already
1945 * tried and the last backup index you used.
1947 * this returns -1 when it has tried all the backups
1949 static noinline int next_root_backup(struct btrfs_fs_info *info,
1950 struct btrfs_super_block *super,
1951 int *num_backups_tried, int *backup_index)
1953 struct btrfs_root_backup *root_backup;
1954 int newest = *backup_index;
1956 if (*num_backups_tried == 0) {
1957 u64 gen = btrfs_super_generation(super);
1959 newest = find_newest_super_backup(info, gen);
1963 *backup_index = newest;
1964 *num_backups_tried = 1;
1965 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966 /* we've tried all the backups, all done */
1969 /* jump to the next oldest backup */
1970 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 BTRFS_NUM_BACKUP_ROOTS;
1972 *backup_index = newest;
1973 *num_backups_tried += 1;
1975 root_backup = super->super_roots + newest;
1977 btrfs_set_super_generation(super,
1978 btrfs_backup_tree_root_gen(root_backup));
1979 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980 btrfs_set_super_root_level(super,
1981 btrfs_backup_tree_root_level(root_backup));
1982 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1985 * fixme: the total bytes and num_devices need to match or we should
1988 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1993 /* helper to cleanup workers */
1994 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1996 btrfs_stop_workers(&fs_info->generic_worker);
1997 btrfs_stop_workers(&fs_info->fixup_workers);
1998 btrfs_stop_workers(&fs_info->delalloc_workers);
1999 btrfs_stop_workers(&fs_info->workers);
2000 btrfs_stop_workers(&fs_info->endio_workers);
2001 btrfs_stop_workers(&fs_info->endio_meta_workers);
2002 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2003 btrfs_stop_workers(&fs_info->rmw_workers);
2004 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2005 btrfs_stop_workers(&fs_info->endio_write_workers);
2006 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2007 btrfs_stop_workers(&fs_info->submit_workers);
2008 btrfs_stop_workers(&fs_info->delayed_workers);
2009 btrfs_stop_workers(&fs_info->caching_workers);
2010 btrfs_stop_workers(&fs_info->readahead_workers);
2011 btrfs_stop_workers(&fs_info->flush_workers);
2012 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2015 /* helper to cleanup tree roots */
2016 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2018 free_extent_buffer(info->tree_root->node);
2019 free_extent_buffer(info->tree_root->commit_root);
2020 info->tree_root->node = NULL;
2021 info->tree_root->commit_root = NULL;
2023 if (info->dev_root) {
2024 free_extent_buffer(info->dev_root->node);
2025 free_extent_buffer(info->dev_root->commit_root);
2026 info->dev_root->node = NULL;
2027 info->dev_root->commit_root = NULL;
2029 if (info->extent_root) {
2030 free_extent_buffer(info->extent_root->node);
2031 free_extent_buffer(info->extent_root->commit_root);
2032 info->extent_root->node = NULL;
2033 info->extent_root->commit_root = NULL;
2035 if (info->csum_root) {
2036 free_extent_buffer(info->csum_root->node);
2037 free_extent_buffer(info->csum_root->commit_root);
2038 info->csum_root->node = NULL;
2039 info->csum_root->commit_root = NULL;
2041 if (info->quota_root) {
2042 free_extent_buffer(info->quota_root->node);
2043 free_extent_buffer(info->quota_root->commit_root);
2044 info->quota_root->node = NULL;
2045 info->quota_root->commit_root = NULL;
2048 free_extent_buffer(info->chunk_root->node);
2049 free_extent_buffer(info->chunk_root->commit_root);
2050 info->chunk_root->node = NULL;
2051 info->chunk_root->commit_root = NULL;
2055 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2058 struct btrfs_root *gang[8];
2061 while (!list_empty(&fs_info->dead_roots)) {
2062 gang[0] = list_entry(fs_info->dead_roots.next,
2063 struct btrfs_root, root_list);
2064 list_del(&gang[0]->root_list);
2066 if (gang[0]->in_radix) {
2067 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2069 free_extent_buffer(gang[0]->node);
2070 free_extent_buffer(gang[0]->commit_root);
2071 btrfs_put_fs_root(gang[0]);
2076 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2081 for (i = 0; i < ret; i++)
2082 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2086 int open_ctree(struct super_block *sb,
2087 struct btrfs_fs_devices *fs_devices,
2097 struct btrfs_key location;
2098 struct buffer_head *bh;
2099 struct btrfs_super_block *disk_super;
2100 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2101 struct btrfs_root *tree_root;
2102 struct btrfs_root *extent_root;
2103 struct btrfs_root *csum_root;
2104 struct btrfs_root *chunk_root;
2105 struct btrfs_root *dev_root;
2106 struct btrfs_root *quota_root;
2107 struct btrfs_root *log_tree_root;
2110 int num_backups_tried = 0;
2111 int backup_index = 0;
2113 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2114 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2115 if (!tree_root || !chunk_root) {
2120 ret = init_srcu_struct(&fs_info->subvol_srcu);
2126 ret = setup_bdi(fs_info, &fs_info->bdi);
2132 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2137 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2138 (1 + ilog2(nr_cpu_ids));
2140 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2143 goto fail_dirty_metadata_bytes;
2146 fs_info->btree_inode = new_inode(sb);
2147 if (!fs_info->btree_inode) {
2149 goto fail_delalloc_bytes;
2152 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2154 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2155 INIT_LIST_HEAD(&fs_info->trans_list);
2156 INIT_LIST_HEAD(&fs_info->dead_roots);
2157 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2158 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2159 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2160 spin_lock_init(&fs_info->delalloc_root_lock);
2161 spin_lock_init(&fs_info->trans_lock);
2162 spin_lock_init(&fs_info->fs_roots_radix_lock);
2163 spin_lock_init(&fs_info->delayed_iput_lock);
2164 spin_lock_init(&fs_info->defrag_inodes_lock);
2165 spin_lock_init(&fs_info->free_chunk_lock);
2166 spin_lock_init(&fs_info->tree_mod_seq_lock);
2167 spin_lock_init(&fs_info->super_lock);
2168 rwlock_init(&fs_info->tree_mod_log_lock);
2169 mutex_init(&fs_info->reloc_mutex);
2170 seqlock_init(&fs_info->profiles_lock);
2172 init_completion(&fs_info->kobj_unregister);
2173 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2174 INIT_LIST_HEAD(&fs_info->space_info);
2175 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2176 btrfs_mapping_init(&fs_info->mapping_tree);
2177 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2178 BTRFS_BLOCK_RSV_GLOBAL);
2179 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2180 BTRFS_BLOCK_RSV_DELALLOC);
2181 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2182 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2183 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2184 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2185 BTRFS_BLOCK_RSV_DELOPS);
2186 atomic_set(&fs_info->nr_async_submits, 0);
2187 atomic_set(&fs_info->async_delalloc_pages, 0);
2188 atomic_set(&fs_info->async_submit_draining, 0);
2189 atomic_set(&fs_info->nr_async_bios, 0);
2190 atomic_set(&fs_info->defrag_running, 0);
2191 atomic64_set(&fs_info->tree_mod_seq, 0);
2193 fs_info->max_inline = 8192 * 1024;
2194 fs_info->metadata_ratio = 0;
2195 fs_info->defrag_inodes = RB_ROOT;
2196 fs_info->free_chunk_space = 0;
2197 fs_info->tree_mod_log = RB_ROOT;
2198 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2200 /* readahead state */
2201 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2202 spin_lock_init(&fs_info->reada_lock);
2204 fs_info->thread_pool_size = min_t(unsigned long,
2205 num_online_cpus() + 2, 8);
2207 INIT_LIST_HEAD(&fs_info->ordered_roots);
2208 spin_lock_init(&fs_info->ordered_root_lock);
2209 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2211 if (!fs_info->delayed_root) {
2215 btrfs_init_delayed_root(fs_info->delayed_root);
2217 mutex_init(&fs_info->scrub_lock);
2218 atomic_set(&fs_info->scrubs_running, 0);
2219 atomic_set(&fs_info->scrub_pause_req, 0);
2220 atomic_set(&fs_info->scrubs_paused, 0);
2221 atomic_set(&fs_info->scrub_cancel_req, 0);
2222 init_waitqueue_head(&fs_info->scrub_pause_wait);
2223 init_rwsem(&fs_info->scrub_super_lock);
2224 fs_info->scrub_workers_refcnt = 0;
2225 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2226 fs_info->check_integrity_print_mask = 0;
2229 spin_lock_init(&fs_info->balance_lock);
2230 mutex_init(&fs_info->balance_mutex);
2231 atomic_set(&fs_info->balance_running, 0);
2232 atomic_set(&fs_info->balance_pause_req, 0);
2233 atomic_set(&fs_info->balance_cancel_req, 0);
2234 fs_info->balance_ctl = NULL;
2235 init_waitqueue_head(&fs_info->balance_wait_q);
2237 sb->s_blocksize = 4096;
2238 sb->s_blocksize_bits = blksize_bits(4096);
2239 sb->s_bdi = &fs_info->bdi;
2241 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2242 set_nlink(fs_info->btree_inode, 1);
2244 * we set the i_size on the btree inode to the max possible int.
2245 * the real end of the address space is determined by all of
2246 * the devices in the system
2248 fs_info->btree_inode->i_size = OFFSET_MAX;
2249 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2250 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2252 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2253 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2254 fs_info->btree_inode->i_mapping);
2255 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2256 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2258 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2260 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2261 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2262 sizeof(struct btrfs_key));
2263 set_bit(BTRFS_INODE_DUMMY,
2264 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2265 insert_inode_hash(fs_info->btree_inode);
2267 spin_lock_init(&fs_info->block_group_cache_lock);
2268 fs_info->block_group_cache_tree = RB_ROOT;
2269 fs_info->first_logical_byte = (u64)-1;
2271 extent_io_tree_init(&fs_info->freed_extents[0],
2272 fs_info->btree_inode->i_mapping);
2273 extent_io_tree_init(&fs_info->freed_extents[1],
2274 fs_info->btree_inode->i_mapping);
2275 fs_info->pinned_extents = &fs_info->freed_extents[0];
2276 fs_info->do_barriers = 1;
2279 mutex_init(&fs_info->ordered_operations_mutex);
2280 mutex_init(&fs_info->ordered_extent_flush_mutex);
2281 mutex_init(&fs_info->tree_log_mutex);
2282 mutex_init(&fs_info->chunk_mutex);
2283 mutex_init(&fs_info->transaction_kthread_mutex);
2284 mutex_init(&fs_info->cleaner_mutex);
2285 mutex_init(&fs_info->volume_mutex);
2286 init_rwsem(&fs_info->extent_commit_sem);
2287 init_rwsem(&fs_info->cleanup_work_sem);
2288 init_rwsem(&fs_info->subvol_sem);
2289 fs_info->dev_replace.lock_owner = 0;
2290 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2291 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2292 mutex_init(&fs_info->dev_replace.lock_management_lock);
2293 mutex_init(&fs_info->dev_replace.lock);
2295 spin_lock_init(&fs_info->qgroup_lock);
2296 mutex_init(&fs_info->qgroup_ioctl_lock);
2297 fs_info->qgroup_tree = RB_ROOT;
2298 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2299 fs_info->qgroup_seq = 1;
2300 fs_info->quota_enabled = 0;
2301 fs_info->pending_quota_state = 0;
2302 fs_info->qgroup_ulist = NULL;
2303 mutex_init(&fs_info->qgroup_rescan_lock);
2305 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2306 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2308 init_waitqueue_head(&fs_info->transaction_throttle);
2309 init_waitqueue_head(&fs_info->transaction_wait);
2310 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2311 init_waitqueue_head(&fs_info->async_submit_wait);
2313 ret = btrfs_alloc_stripe_hash_table(fs_info);
2319 __setup_root(4096, 4096, 4096, 4096, tree_root,
2320 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2322 invalidate_bdev(fs_devices->latest_bdev);
2325 * Read super block and check the signature bytes only
2327 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2334 * We want to check superblock checksum, the type is stored inside.
2335 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2337 if (btrfs_check_super_csum(bh->b_data)) {
2338 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2344 * super_copy is zeroed at allocation time and we never touch the
2345 * following bytes up to INFO_SIZE, the checksum is calculated from
2346 * the whole block of INFO_SIZE
2348 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2349 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2350 sizeof(*fs_info->super_for_commit));
2353 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2355 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2357 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2362 disk_super = fs_info->super_copy;
2363 if (!btrfs_super_root(disk_super))
2366 /* check FS state, whether FS is broken. */
2367 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2368 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2371 * run through our array of backup supers and setup
2372 * our ring pointer to the oldest one
2374 generation = btrfs_super_generation(disk_super);
2375 find_oldest_super_backup(fs_info, generation);
2378 * In the long term, we'll store the compression type in the super
2379 * block, and it'll be used for per file compression control.
2381 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2383 ret = btrfs_parse_options(tree_root, options);
2389 features = btrfs_super_incompat_flags(disk_super) &
2390 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2392 printk(KERN_ERR "BTRFS: couldn't mount because of "
2393 "unsupported optional features (%Lx).\n",
2394 (unsigned long long)features);
2399 if (btrfs_super_leafsize(disk_super) !=
2400 btrfs_super_nodesize(disk_super)) {
2401 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2402 "blocksizes don't match. node %d leaf %d\n",
2403 btrfs_super_nodesize(disk_super),
2404 btrfs_super_leafsize(disk_super));
2408 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2409 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2410 "blocksize (%d) was too large\n",
2411 btrfs_super_leafsize(disk_super));
2416 features = btrfs_super_incompat_flags(disk_super);
2417 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2418 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2419 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2421 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2422 printk(KERN_ERR "btrfs: has skinny extents\n");
2425 * flag our filesystem as having big metadata blocks if
2426 * they are bigger than the page size
2428 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2429 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2430 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2431 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2434 nodesize = btrfs_super_nodesize(disk_super);
2435 leafsize = btrfs_super_leafsize(disk_super);
2436 sectorsize = btrfs_super_sectorsize(disk_super);
2437 stripesize = btrfs_super_stripesize(disk_super);
2438 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2439 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2442 * mixed block groups end up with duplicate but slightly offset
2443 * extent buffers for the same range. It leads to corruptions
2445 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2446 (sectorsize != leafsize)) {
2447 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2448 "are not allowed for mixed block groups on %s\n",
2454 * Needn't use the lock because there is no other task which will
2457 btrfs_set_super_incompat_flags(disk_super, features);
2459 features = btrfs_super_compat_ro_flags(disk_super) &
2460 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2461 if (!(sb->s_flags & MS_RDONLY) && features) {
2462 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2463 "unsupported option features (%Lx).\n",
2464 (unsigned long long)features);
2469 btrfs_init_workers(&fs_info->generic_worker,
2470 "genwork", 1, NULL);
2472 btrfs_init_workers(&fs_info->workers, "worker",
2473 fs_info->thread_pool_size,
2474 &fs_info->generic_worker);
2476 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2477 fs_info->thread_pool_size,
2478 &fs_info->generic_worker);
2480 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2481 fs_info->thread_pool_size,
2482 &fs_info->generic_worker);
2484 btrfs_init_workers(&fs_info->submit_workers, "submit",
2485 min_t(u64, fs_devices->num_devices,
2486 fs_info->thread_pool_size),
2487 &fs_info->generic_worker);
2489 btrfs_init_workers(&fs_info->caching_workers, "cache",
2490 2, &fs_info->generic_worker);
2492 /* a higher idle thresh on the submit workers makes it much more
2493 * likely that bios will be send down in a sane order to the
2496 fs_info->submit_workers.idle_thresh = 64;
2498 fs_info->workers.idle_thresh = 16;
2499 fs_info->workers.ordered = 1;
2501 fs_info->delalloc_workers.idle_thresh = 2;
2502 fs_info->delalloc_workers.ordered = 1;
2504 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2505 &fs_info->generic_worker);
2506 btrfs_init_workers(&fs_info->endio_workers, "endio",
2507 fs_info->thread_pool_size,
2508 &fs_info->generic_worker);
2509 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2510 fs_info->thread_pool_size,
2511 &fs_info->generic_worker);
2512 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2513 "endio-meta-write", fs_info->thread_pool_size,
2514 &fs_info->generic_worker);
2515 btrfs_init_workers(&fs_info->endio_raid56_workers,
2516 "endio-raid56", fs_info->thread_pool_size,
2517 &fs_info->generic_worker);
2518 btrfs_init_workers(&fs_info->rmw_workers,
2519 "rmw", fs_info->thread_pool_size,
2520 &fs_info->generic_worker);
2521 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2522 fs_info->thread_pool_size,
2523 &fs_info->generic_worker);
2524 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2525 1, &fs_info->generic_worker);
2526 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2527 fs_info->thread_pool_size,
2528 &fs_info->generic_worker);
2529 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2530 fs_info->thread_pool_size,
2531 &fs_info->generic_worker);
2532 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2533 &fs_info->generic_worker);
2536 * endios are largely parallel and should have a very
2539 fs_info->endio_workers.idle_thresh = 4;
2540 fs_info->endio_meta_workers.idle_thresh = 4;
2541 fs_info->endio_raid56_workers.idle_thresh = 4;
2542 fs_info->rmw_workers.idle_thresh = 2;
2544 fs_info->endio_write_workers.idle_thresh = 2;
2545 fs_info->endio_meta_write_workers.idle_thresh = 2;
2546 fs_info->readahead_workers.idle_thresh = 2;
2549 * btrfs_start_workers can really only fail because of ENOMEM so just
2550 * return -ENOMEM if any of these fail.
2552 ret = btrfs_start_workers(&fs_info->workers);
2553 ret |= btrfs_start_workers(&fs_info->generic_worker);
2554 ret |= btrfs_start_workers(&fs_info->submit_workers);
2555 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2556 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2557 ret |= btrfs_start_workers(&fs_info->endio_workers);
2558 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2559 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2560 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2561 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2562 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2563 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2564 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2565 ret |= btrfs_start_workers(&fs_info->caching_workers);
2566 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2567 ret |= btrfs_start_workers(&fs_info->flush_workers);
2568 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2571 goto fail_sb_buffer;
2574 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2575 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2576 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2578 tree_root->nodesize = nodesize;
2579 tree_root->leafsize = leafsize;
2580 tree_root->sectorsize = sectorsize;
2581 tree_root->stripesize = stripesize;
2583 sb->s_blocksize = sectorsize;
2584 sb->s_blocksize_bits = blksize_bits(sectorsize);
2586 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2587 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2588 goto fail_sb_buffer;
2591 if (sectorsize != PAGE_SIZE) {
2592 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2593 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2594 goto fail_sb_buffer;
2597 mutex_lock(&fs_info->chunk_mutex);
2598 ret = btrfs_read_sys_array(tree_root);
2599 mutex_unlock(&fs_info->chunk_mutex);
2601 printk(KERN_WARNING "btrfs: failed to read the system "
2602 "array on %s\n", sb->s_id);
2603 goto fail_sb_buffer;
2606 blocksize = btrfs_level_size(tree_root,
2607 btrfs_super_chunk_root_level(disk_super));
2608 generation = btrfs_super_chunk_root_generation(disk_super);
2610 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2611 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2613 chunk_root->node = read_tree_block(chunk_root,
2614 btrfs_super_chunk_root(disk_super),
2615 blocksize, generation);
2616 if (!chunk_root->node ||
2617 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2618 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2620 goto fail_tree_roots;
2622 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2623 chunk_root->commit_root = btrfs_root_node(chunk_root);
2625 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2626 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2629 ret = btrfs_read_chunk_tree(chunk_root);
2631 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2633 goto fail_tree_roots;
2637 * keep the device that is marked to be the target device for the
2638 * dev_replace procedure
2640 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2642 if (!fs_devices->latest_bdev) {
2643 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2645 goto fail_tree_roots;
2649 blocksize = btrfs_level_size(tree_root,
2650 btrfs_super_root_level(disk_super));
2651 generation = btrfs_super_generation(disk_super);
2653 tree_root->node = read_tree_block(tree_root,
2654 btrfs_super_root(disk_super),
2655 blocksize, generation);
2656 if (!tree_root->node ||
2657 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2658 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2661 goto recovery_tree_root;
2664 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2665 tree_root->commit_root = btrfs_root_node(tree_root);
2667 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2668 location.type = BTRFS_ROOT_ITEM_KEY;
2669 location.offset = 0;
2671 extent_root = btrfs_read_tree_root(tree_root, &location);
2672 if (IS_ERR(extent_root)) {
2673 ret = PTR_ERR(extent_root);
2674 goto recovery_tree_root;
2676 extent_root->track_dirty = 1;
2677 fs_info->extent_root = extent_root;
2679 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2680 dev_root = btrfs_read_tree_root(tree_root, &location);
2681 if (IS_ERR(dev_root)) {
2682 ret = PTR_ERR(dev_root);
2683 goto recovery_tree_root;
2685 dev_root->track_dirty = 1;
2686 fs_info->dev_root = dev_root;
2687 btrfs_init_devices_late(fs_info);
2689 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2690 csum_root = btrfs_read_tree_root(tree_root, &location);
2691 if (IS_ERR(csum_root)) {
2692 ret = PTR_ERR(csum_root);
2693 goto recovery_tree_root;
2695 csum_root->track_dirty = 1;
2696 fs_info->csum_root = csum_root;
2698 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2699 quota_root = btrfs_read_tree_root(tree_root, &location);
2700 if (!IS_ERR(quota_root)) {
2701 quota_root->track_dirty = 1;
2702 fs_info->quota_enabled = 1;
2703 fs_info->pending_quota_state = 1;
2704 fs_info->quota_root = quota_root;
2707 fs_info->generation = generation;
2708 fs_info->last_trans_committed = generation;
2710 ret = btrfs_recover_balance(fs_info);
2712 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2713 goto fail_block_groups;
2716 ret = btrfs_init_dev_stats(fs_info);
2718 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2720 goto fail_block_groups;
2723 ret = btrfs_init_dev_replace(fs_info);
2725 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2726 goto fail_block_groups;
2729 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2731 ret = btrfs_init_space_info(fs_info);
2733 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2734 goto fail_block_groups;
2737 ret = btrfs_read_block_groups(extent_root);
2739 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2740 goto fail_block_groups;
2742 fs_info->num_tolerated_disk_barrier_failures =
2743 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2744 if (fs_info->fs_devices->missing_devices >
2745 fs_info->num_tolerated_disk_barrier_failures &&
2746 !(sb->s_flags & MS_RDONLY)) {
2748 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2749 goto fail_block_groups;
2752 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2754 if (IS_ERR(fs_info->cleaner_kthread))
2755 goto fail_block_groups;
2757 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2759 "btrfs-transaction");
2760 if (IS_ERR(fs_info->transaction_kthread))
2763 if (!btrfs_test_opt(tree_root, SSD) &&
2764 !btrfs_test_opt(tree_root, NOSSD) &&
2765 !fs_info->fs_devices->rotating) {
2766 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2768 btrfs_set_opt(fs_info->mount_opt, SSD);
2771 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2772 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2773 ret = btrfsic_mount(tree_root, fs_devices,
2774 btrfs_test_opt(tree_root,
2775 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2777 fs_info->check_integrity_print_mask);
2779 printk(KERN_WARNING "btrfs: failed to initialize"
2780 " integrity check module %s\n", sb->s_id);
2783 ret = btrfs_read_qgroup_config(fs_info);
2785 goto fail_trans_kthread;
2787 /* do not make disk changes in broken FS */
2788 if (btrfs_super_log_root(disk_super) != 0) {
2789 u64 bytenr = btrfs_super_log_root(disk_super);
2791 if (fs_devices->rw_devices == 0) {
2792 printk(KERN_WARNING "Btrfs log replay required "
2798 btrfs_level_size(tree_root,
2799 btrfs_super_log_root_level(disk_super));
2801 log_tree_root = btrfs_alloc_root(fs_info);
2802 if (!log_tree_root) {
2807 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2808 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2810 log_tree_root->node = read_tree_block(tree_root, bytenr,
2813 if (!log_tree_root->node ||
2814 !extent_buffer_uptodate(log_tree_root->node)) {
2815 printk(KERN_ERR "btrfs: failed to read log tree\n");
2816 free_extent_buffer(log_tree_root->node);
2817 kfree(log_tree_root);
2818 goto fail_trans_kthread;
2820 /* returns with log_tree_root freed on success */
2821 ret = btrfs_recover_log_trees(log_tree_root);
2823 btrfs_error(tree_root->fs_info, ret,
2824 "Failed to recover log tree");
2825 free_extent_buffer(log_tree_root->node);
2826 kfree(log_tree_root);
2827 goto fail_trans_kthread;
2830 if (sb->s_flags & MS_RDONLY) {
2831 ret = btrfs_commit_super(tree_root);
2833 goto fail_trans_kthread;
2837 ret = btrfs_find_orphan_roots(tree_root);
2839 goto fail_trans_kthread;
2841 if (!(sb->s_flags & MS_RDONLY)) {
2842 ret = btrfs_cleanup_fs_roots(fs_info);
2844 goto fail_trans_kthread;
2846 ret = btrfs_recover_relocation(tree_root);
2849 "btrfs: failed to recover relocation\n");
2855 location.objectid = BTRFS_FS_TREE_OBJECTID;
2856 location.type = BTRFS_ROOT_ITEM_KEY;
2857 location.offset = 0;
2859 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2860 if (IS_ERR(fs_info->fs_root)) {
2861 err = PTR_ERR(fs_info->fs_root);
2865 if (sb->s_flags & MS_RDONLY)
2868 down_read(&fs_info->cleanup_work_sem);
2869 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2870 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2871 up_read(&fs_info->cleanup_work_sem);
2872 close_ctree(tree_root);
2875 up_read(&fs_info->cleanup_work_sem);
2877 ret = btrfs_resume_balance_async(fs_info);
2879 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2880 close_ctree(tree_root);
2884 ret = btrfs_resume_dev_replace_async(fs_info);
2886 pr_warn("btrfs: failed to resume dev_replace\n");
2887 close_ctree(tree_root);
2891 btrfs_qgroup_rescan_resume(fs_info);
2896 btrfs_free_qgroup_config(fs_info);
2898 kthread_stop(fs_info->transaction_kthread);
2899 btrfs_cleanup_transaction(fs_info->tree_root);
2900 del_fs_roots(fs_info);
2902 kthread_stop(fs_info->cleaner_kthread);
2905 * make sure we're done with the btree inode before we stop our
2908 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2911 btrfs_put_block_group_cache(fs_info);
2912 btrfs_free_block_groups(fs_info);
2915 free_root_pointers(fs_info, 1);
2916 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2919 btrfs_stop_all_workers(fs_info);
2922 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2924 iput(fs_info->btree_inode);
2925 fail_delalloc_bytes:
2926 percpu_counter_destroy(&fs_info->delalloc_bytes);
2927 fail_dirty_metadata_bytes:
2928 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2930 bdi_destroy(&fs_info->bdi);
2932 cleanup_srcu_struct(&fs_info->subvol_srcu);
2934 btrfs_free_stripe_hash_table(fs_info);
2935 btrfs_close_devices(fs_info->fs_devices);
2939 if (!btrfs_test_opt(tree_root, RECOVERY))
2940 goto fail_tree_roots;
2942 free_root_pointers(fs_info, 0);
2944 /* don't use the log in recovery mode, it won't be valid */
2945 btrfs_set_super_log_root(disk_super, 0);
2947 /* we can't trust the free space cache either */
2948 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2950 ret = next_root_backup(fs_info, fs_info->super_copy,
2951 &num_backups_tried, &backup_index);
2953 goto fail_block_groups;
2954 goto retry_root_backup;
2957 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2960 set_buffer_uptodate(bh);
2962 struct btrfs_device *device = (struct btrfs_device *)
2965 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2966 "I/O error on %s\n",
2967 rcu_str_deref(device->name));
2968 /* note, we dont' set_buffer_write_io_error because we have
2969 * our own ways of dealing with the IO errors
2971 clear_buffer_uptodate(bh);
2972 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2978 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2980 struct buffer_head *bh;
2981 struct buffer_head *latest = NULL;
2982 struct btrfs_super_block *super;
2987 /* we would like to check all the supers, but that would make
2988 * a btrfs mount succeed after a mkfs from a different FS.
2989 * So, we need to add a special mount option to scan for
2990 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2992 for (i = 0; i < 1; i++) {
2993 bytenr = btrfs_sb_offset(i);
2994 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2995 i_size_read(bdev->bd_inode))
2997 bh = __bread(bdev, bytenr / 4096,
2998 BTRFS_SUPER_INFO_SIZE);
3002 super = (struct btrfs_super_block *)bh->b_data;
3003 if (btrfs_super_bytenr(super) != bytenr ||
3004 btrfs_super_magic(super) != BTRFS_MAGIC) {
3009 if (!latest || btrfs_super_generation(super) > transid) {
3012 transid = btrfs_super_generation(super);
3021 * this should be called twice, once with wait == 0 and
3022 * once with wait == 1. When wait == 0 is done, all the buffer heads
3023 * we write are pinned.
3025 * They are released when wait == 1 is done.
3026 * max_mirrors must be the same for both runs, and it indicates how
3027 * many supers on this one device should be written.
3029 * max_mirrors == 0 means to write them all.
3031 static int write_dev_supers(struct btrfs_device *device,
3032 struct btrfs_super_block *sb,
3033 int do_barriers, int wait, int max_mirrors)
3035 struct buffer_head *bh;
3042 if (max_mirrors == 0)
3043 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3045 for (i = 0; i < max_mirrors; i++) {
3046 bytenr = btrfs_sb_offset(i);
3047 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3051 bh = __find_get_block(device->bdev, bytenr / 4096,
3052 BTRFS_SUPER_INFO_SIZE);
3058 if (!buffer_uptodate(bh))
3061 /* drop our reference */
3064 /* drop the reference from the wait == 0 run */
3068 btrfs_set_super_bytenr(sb, bytenr);
3071 crc = btrfs_csum_data((char *)sb +
3072 BTRFS_CSUM_SIZE, crc,
3073 BTRFS_SUPER_INFO_SIZE -
3075 btrfs_csum_final(crc, sb->csum);
3078 * one reference for us, and we leave it for the
3081 bh = __getblk(device->bdev, bytenr / 4096,
3082 BTRFS_SUPER_INFO_SIZE);
3084 printk(KERN_ERR "btrfs: couldn't get super "
3085 "buffer head for bytenr %Lu\n", bytenr);
3090 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3092 /* one reference for submit_bh */
3095 set_buffer_uptodate(bh);
3097 bh->b_end_io = btrfs_end_buffer_write_sync;
3098 bh->b_private = device;
3102 * we fua the first super. The others we allow
3105 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3109 return errors < i ? 0 : -1;
3113 * endio for the write_dev_flush, this will wake anyone waiting
3114 * for the barrier when it is done
3116 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3119 if (err == -EOPNOTSUPP)
3120 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3121 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3123 if (bio->bi_private)
3124 complete(bio->bi_private);
3129 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3130 * sent down. With wait == 1, it waits for the previous flush.
3132 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3135 static int write_dev_flush(struct btrfs_device *device, int wait)
3140 if (device->nobarriers)
3144 bio = device->flush_bio;
3148 wait_for_completion(&device->flush_wait);
3150 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3151 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3152 rcu_str_deref(device->name));
3153 device->nobarriers = 1;
3154 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3156 btrfs_dev_stat_inc_and_print(device,
3157 BTRFS_DEV_STAT_FLUSH_ERRS);
3160 /* drop the reference from the wait == 0 run */
3162 device->flush_bio = NULL;
3168 * one reference for us, and we leave it for the
3171 device->flush_bio = NULL;
3172 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3176 bio->bi_end_io = btrfs_end_empty_barrier;
3177 bio->bi_bdev = device->bdev;
3178 init_completion(&device->flush_wait);
3179 bio->bi_private = &device->flush_wait;
3180 device->flush_bio = bio;
3183 btrfsic_submit_bio(WRITE_FLUSH, bio);
3189 * send an empty flush down to each device in parallel,
3190 * then wait for them
3192 static int barrier_all_devices(struct btrfs_fs_info *info)
3194 struct list_head *head;
3195 struct btrfs_device *dev;
3196 int errors_send = 0;
3197 int errors_wait = 0;
3200 /* send down all the barriers */
3201 head = &info->fs_devices->devices;
3202 list_for_each_entry_rcu(dev, head, dev_list) {
3207 if (!dev->in_fs_metadata || !dev->writeable)
3210 ret = write_dev_flush(dev, 0);
3215 /* wait for all the barriers */
3216 list_for_each_entry_rcu(dev, head, dev_list) {
3221 if (!dev->in_fs_metadata || !dev->writeable)
3224 ret = write_dev_flush(dev, 1);
3228 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3229 errors_wait > info->num_tolerated_disk_barrier_failures)
3234 int btrfs_calc_num_tolerated_disk_barrier_failures(
3235 struct btrfs_fs_info *fs_info)
3237 struct btrfs_ioctl_space_info space;
3238 struct btrfs_space_info *sinfo;
3239 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3240 BTRFS_BLOCK_GROUP_SYSTEM,
3241 BTRFS_BLOCK_GROUP_METADATA,
3242 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3246 int num_tolerated_disk_barrier_failures =
3247 (int)fs_info->fs_devices->num_devices;
3249 for (i = 0; i < num_types; i++) {
3250 struct btrfs_space_info *tmp;
3254 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3255 if (tmp->flags == types[i]) {
3265 down_read(&sinfo->groups_sem);
3266 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3267 if (!list_empty(&sinfo->block_groups[c])) {
3270 btrfs_get_block_group_info(
3271 &sinfo->block_groups[c], &space);
3272 if (space.total_bytes == 0 ||
3273 space.used_bytes == 0)
3275 flags = space.flags;
3278 * 0: if dup, single or RAID0 is configured for
3279 * any of metadata, system or data, else
3280 * 1: if RAID5 is configured, or if RAID1 or
3281 * RAID10 is configured and only two mirrors
3283 * 2: if RAID6 is configured, else
3284 * num_mirrors - 1: if RAID1 or RAID10 is
3285 * configured and more than
3286 * 2 mirrors are used.
3288 if (num_tolerated_disk_barrier_failures > 0 &&
3289 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3290 BTRFS_BLOCK_GROUP_RAID0)) ||
3291 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3293 num_tolerated_disk_barrier_failures = 0;
3294 else if (num_tolerated_disk_barrier_failures > 1) {
3295 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3296 BTRFS_BLOCK_GROUP_RAID5 |
3297 BTRFS_BLOCK_GROUP_RAID10)) {
3298 num_tolerated_disk_barrier_failures = 1;
3300 BTRFS_BLOCK_GROUP_RAID6) {
3301 num_tolerated_disk_barrier_failures = 2;
3306 up_read(&sinfo->groups_sem);
3309 return num_tolerated_disk_barrier_failures;
3312 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3314 struct list_head *head;
3315 struct btrfs_device *dev;
3316 struct btrfs_super_block *sb;
3317 struct btrfs_dev_item *dev_item;
3321 int total_errors = 0;
3324 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3325 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3326 backup_super_roots(root->fs_info);
3328 sb = root->fs_info->super_for_commit;
3329 dev_item = &sb->dev_item;
3331 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3332 head = &root->fs_info->fs_devices->devices;
3335 ret = barrier_all_devices(root->fs_info);
3338 &root->fs_info->fs_devices->device_list_mutex);
3339 btrfs_error(root->fs_info, ret,
3340 "errors while submitting device barriers.");
3345 list_for_each_entry_rcu(dev, head, dev_list) {
3350 if (!dev->in_fs_metadata || !dev->writeable)
3353 btrfs_set_stack_device_generation(dev_item, 0);
3354 btrfs_set_stack_device_type(dev_item, dev->type);
3355 btrfs_set_stack_device_id(dev_item, dev->devid);
3356 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3357 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3358 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3359 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3360 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3361 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3362 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3364 flags = btrfs_super_flags(sb);
3365 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3367 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3371 if (total_errors > max_errors) {
3372 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3375 /* This shouldn't happen. FUA is masked off if unsupported */
3380 list_for_each_entry_rcu(dev, head, dev_list) {
3383 if (!dev->in_fs_metadata || !dev->writeable)
3386 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3390 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3391 if (total_errors > max_errors) {
3392 btrfs_error(root->fs_info, -EIO,
3393 "%d errors while writing supers", total_errors);
3399 int write_ctree_super(struct btrfs_trans_handle *trans,
3400 struct btrfs_root *root, int max_mirrors)
3404 ret = write_all_supers(root, max_mirrors);
3408 /* Drop a fs root from the radix tree and free it. */
3409 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3410 struct btrfs_root *root)
3412 spin_lock(&fs_info->fs_roots_radix_lock);
3413 radix_tree_delete(&fs_info->fs_roots_radix,
3414 (unsigned long)root->root_key.objectid);
3415 spin_unlock(&fs_info->fs_roots_radix_lock);
3417 if (btrfs_root_refs(&root->root_item) == 0)
3418 synchronize_srcu(&fs_info->subvol_srcu);
3420 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3421 btrfs_free_log(NULL, root);
3422 btrfs_free_log_root_tree(NULL, fs_info);
3425 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3426 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3430 static void free_fs_root(struct btrfs_root *root)
3432 iput(root->cache_inode);
3433 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3435 free_anon_bdev(root->anon_dev);
3436 free_extent_buffer(root->node);
3437 free_extent_buffer(root->commit_root);
3438 kfree(root->free_ino_ctl);
3439 kfree(root->free_ino_pinned);
3441 btrfs_put_fs_root(root);
3444 void btrfs_free_fs_root(struct btrfs_root *root)
3449 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3451 u64 root_objectid = 0;
3452 struct btrfs_root *gang[8];
3457 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3458 (void **)gang, root_objectid,
3463 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3464 for (i = 0; i < ret; i++) {
3467 root_objectid = gang[i]->root_key.objectid;
3468 err = btrfs_orphan_cleanup(gang[i]);
3477 int btrfs_commit_super(struct btrfs_root *root)
3479 struct btrfs_trans_handle *trans;
3482 mutex_lock(&root->fs_info->cleaner_mutex);
3483 btrfs_run_delayed_iputs(root);
3484 mutex_unlock(&root->fs_info->cleaner_mutex);
3485 wake_up_process(root->fs_info->cleaner_kthread);
3487 /* wait until ongoing cleanup work done */
3488 down_write(&root->fs_info->cleanup_work_sem);
3489 up_write(&root->fs_info->cleanup_work_sem);
3491 trans = btrfs_join_transaction(root);
3493 return PTR_ERR(trans);
3494 ret = btrfs_commit_transaction(trans, root);
3497 /* run commit again to drop the original snapshot */
3498 trans = btrfs_join_transaction(root);
3500 return PTR_ERR(trans);
3501 ret = btrfs_commit_transaction(trans, root);
3504 ret = btrfs_write_and_wait_transaction(NULL, root);
3506 btrfs_error(root->fs_info, ret,
3507 "Failed to sync btree inode to disk.");
3511 ret = write_ctree_super(NULL, root, 0);
3515 int close_ctree(struct btrfs_root *root)
3517 struct btrfs_fs_info *fs_info = root->fs_info;
3520 fs_info->closing = 1;
3523 /* pause restriper - we want to resume on mount */
3524 btrfs_pause_balance(fs_info);
3526 btrfs_dev_replace_suspend_for_unmount(fs_info);
3528 btrfs_scrub_cancel(fs_info);
3530 /* wait for any defraggers to finish */
3531 wait_event(fs_info->transaction_wait,
3532 (atomic_read(&fs_info->defrag_running) == 0));
3534 /* clear out the rbtree of defraggable inodes */
3535 btrfs_cleanup_defrag_inodes(fs_info);
3537 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3538 ret = btrfs_commit_super(root);
3540 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3543 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3544 btrfs_error_commit_super(root);
3546 btrfs_put_block_group_cache(fs_info);
3548 kthread_stop(fs_info->transaction_kthread);
3549 kthread_stop(fs_info->cleaner_kthread);
3551 fs_info->closing = 2;
3554 btrfs_free_qgroup_config(root->fs_info);
3556 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3557 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3558 percpu_counter_sum(&fs_info->delalloc_bytes));
3561 btrfs_free_block_groups(fs_info);
3563 btrfs_stop_all_workers(fs_info);
3565 del_fs_roots(fs_info);
3567 free_root_pointers(fs_info, 1);
3569 iput(fs_info->btree_inode);
3571 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3572 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3573 btrfsic_unmount(root, fs_info->fs_devices);
3576 btrfs_close_devices(fs_info->fs_devices);
3577 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3579 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3580 percpu_counter_destroy(&fs_info->delalloc_bytes);
3581 bdi_destroy(&fs_info->bdi);
3582 cleanup_srcu_struct(&fs_info->subvol_srcu);
3584 btrfs_free_stripe_hash_table(fs_info);
3589 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3593 struct inode *btree_inode = buf->pages[0]->mapping->host;
3595 ret = extent_buffer_uptodate(buf);
3599 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3600 parent_transid, atomic);
3606 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3608 return set_extent_buffer_uptodate(buf);
3611 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3613 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3614 u64 transid = btrfs_header_generation(buf);
3617 btrfs_assert_tree_locked(buf);
3618 if (transid != root->fs_info->generation)
3619 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3620 "found %llu running %llu\n",
3621 (unsigned long long)buf->start,
3622 (unsigned long long)transid,
3623 (unsigned long long)root->fs_info->generation);
3624 was_dirty = set_extent_buffer_dirty(buf);
3626 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3628 root->fs_info->dirty_metadata_batch);
3631 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3635 * looks as though older kernels can get into trouble with
3636 * this code, they end up stuck in balance_dirty_pages forever
3640 if (current->flags & PF_MEMALLOC)
3644 btrfs_balance_delayed_items(root);
3646 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3647 BTRFS_DIRTY_METADATA_THRESH);
3649 balance_dirty_pages_ratelimited(
3650 root->fs_info->btree_inode->i_mapping);
3655 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3657 __btrfs_btree_balance_dirty(root, 1);
3660 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3662 __btrfs_btree_balance_dirty(root, 0);
3665 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3667 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3668 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3671 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3675 * Placeholder for checks
3680 static void btrfs_error_commit_super(struct btrfs_root *root)
3682 mutex_lock(&root->fs_info->cleaner_mutex);
3683 btrfs_run_delayed_iputs(root);
3684 mutex_unlock(&root->fs_info->cleaner_mutex);
3686 down_write(&root->fs_info->cleanup_work_sem);
3687 up_write(&root->fs_info->cleanup_work_sem);
3689 /* cleanup FS via transaction */
3690 btrfs_cleanup_transaction(root);
3693 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3694 struct btrfs_root *root)
3696 struct btrfs_inode *btrfs_inode;
3697 struct list_head splice;
3699 INIT_LIST_HEAD(&splice);
3701 mutex_lock(&root->fs_info->ordered_operations_mutex);
3702 spin_lock(&root->fs_info->ordered_root_lock);
3704 list_splice_init(&t->ordered_operations, &splice);
3705 while (!list_empty(&splice)) {
3706 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3707 ordered_operations);
3709 list_del_init(&btrfs_inode->ordered_operations);
3710 spin_unlock(&root->fs_info->ordered_root_lock);
3712 btrfs_invalidate_inodes(btrfs_inode->root);
3714 spin_lock(&root->fs_info->ordered_root_lock);
3717 spin_unlock(&root->fs_info->ordered_root_lock);
3718 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3721 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3723 struct btrfs_ordered_extent *ordered;
3725 spin_lock(&root->ordered_extent_lock);
3727 * This will just short circuit the ordered completion stuff which will
3728 * make sure the ordered extent gets properly cleaned up.
3730 list_for_each_entry(ordered, &root->ordered_extents,
3732 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3733 spin_unlock(&root->ordered_extent_lock);
3736 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3738 struct btrfs_root *root;
3739 struct list_head splice;
3741 INIT_LIST_HEAD(&splice);
3743 spin_lock(&fs_info->ordered_root_lock);
3744 list_splice_init(&fs_info->ordered_roots, &splice);
3745 while (!list_empty(&splice)) {
3746 root = list_first_entry(&splice, struct btrfs_root,
3748 list_del_init(&root->ordered_root);
3750 btrfs_destroy_ordered_extents(root);
3752 cond_resched_lock(&fs_info->ordered_root_lock);
3754 spin_unlock(&fs_info->ordered_root_lock);
3757 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3758 struct btrfs_root *root)
3760 struct rb_node *node;
3761 struct btrfs_delayed_ref_root *delayed_refs;
3762 struct btrfs_delayed_ref_node *ref;
3765 delayed_refs = &trans->delayed_refs;
3767 spin_lock(&delayed_refs->lock);
3768 if (delayed_refs->num_entries == 0) {
3769 spin_unlock(&delayed_refs->lock);
3770 printk(KERN_INFO "delayed_refs has NO entry\n");
3774 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3775 struct btrfs_delayed_ref_head *head = NULL;
3776 bool pin_bytes = false;
3778 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3779 atomic_set(&ref->refs, 1);
3780 if (btrfs_delayed_ref_is_head(ref)) {
3782 head = btrfs_delayed_node_to_head(ref);
3783 if (!mutex_trylock(&head->mutex)) {
3784 atomic_inc(&ref->refs);
3785 spin_unlock(&delayed_refs->lock);
3787 /* Need to wait for the delayed ref to run */
3788 mutex_lock(&head->mutex);
3789 mutex_unlock(&head->mutex);
3790 btrfs_put_delayed_ref(ref);
3792 spin_lock(&delayed_refs->lock);
3796 if (head->must_insert_reserved)
3798 btrfs_free_delayed_extent_op(head->extent_op);
3799 delayed_refs->num_heads--;
3800 if (list_empty(&head->cluster))
3801 delayed_refs->num_heads_ready--;
3802 list_del_init(&head->cluster);
3806 rb_erase(&ref->rb_node, &delayed_refs->root);
3807 delayed_refs->num_entries--;
3808 spin_unlock(&delayed_refs->lock);
3811 btrfs_pin_extent(root, ref->bytenr,
3813 mutex_unlock(&head->mutex);
3815 btrfs_put_delayed_ref(ref);
3818 spin_lock(&delayed_refs->lock);
3821 spin_unlock(&delayed_refs->lock);
3826 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3828 struct btrfs_pending_snapshot *snapshot;
3829 struct list_head splice;
3831 INIT_LIST_HEAD(&splice);
3833 list_splice_init(&t->pending_snapshots, &splice);
3835 while (!list_empty(&splice)) {
3836 snapshot = list_entry(splice.next,
3837 struct btrfs_pending_snapshot,
3839 snapshot->error = -ECANCELED;
3840 list_del_init(&snapshot->list);
3844 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3846 struct btrfs_inode *btrfs_inode;
3847 struct list_head splice;
3849 INIT_LIST_HEAD(&splice);
3851 spin_lock(&root->delalloc_lock);
3852 list_splice_init(&root->delalloc_inodes, &splice);
3854 while (!list_empty(&splice)) {
3855 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3858 list_del_init(&btrfs_inode->delalloc_inodes);
3859 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3860 &btrfs_inode->runtime_flags);
3861 spin_unlock(&root->delalloc_lock);
3863 btrfs_invalidate_inodes(btrfs_inode->root);
3865 spin_lock(&root->delalloc_lock);
3868 spin_unlock(&root->delalloc_lock);
3871 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3873 struct btrfs_root *root;
3874 struct list_head splice;
3876 INIT_LIST_HEAD(&splice);
3878 spin_lock(&fs_info->delalloc_root_lock);
3879 list_splice_init(&fs_info->delalloc_roots, &splice);
3880 while (!list_empty(&splice)) {
3881 root = list_first_entry(&splice, struct btrfs_root,
3883 list_del_init(&root->delalloc_root);
3884 root = btrfs_grab_fs_root(root);
3886 spin_unlock(&fs_info->delalloc_root_lock);
3888 btrfs_destroy_delalloc_inodes(root);
3889 btrfs_put_fs_root(root);
3891 spin_lock(&fs_info->delalloc_root_lock);
3893 spin_unlock(&fs_info->delalloc_root_lock);
3896 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3897 struct extent_io_tree *dirty_pages,
3901 struct extent_buffer *eb;
3906 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3911 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3912 while (start <= end) {
3913 eb = btrfs_find_tree_block(root, start,
3915 start += root->leafsize;
3918 wait_on_extent_buffer_writeback(eb);
3920 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3922 clear_extent_buffer_dirty(eb);
3923 free_extent_buffer_stale(eb);
3930 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3931 struct extent_io_tree *pinned_extents)
3933 struct extent_io_tree *unpin;
3939 unpin = pinned_extents;
3942 ret = find_first_extent_bit(unpin, 0, &start, &end,
3943 EXTENT_DIRTY, NULL);
3948 if (btrfs_test_opt(root, DISCARD))
3949 ret = btrfs_error_discard_extent(root, start,
3953 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3954 btrfs_error_unpin_extent_range(root, start, end);
3959 if (unpin == &root->fs_info->freed_extents[0])
3960 unpin = &root->fs_info->freed_extents[1];
3962 unpin = &root->fs_info->freed_extents[0];
3970 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3971 struct btrfs_root *root)
3973 btrfs_destroy_delayed_refs(cur_trans, root);
3974 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3975 cur_trans->dirty_pages.dirty_bytes);
3977 cur_trans->state = TRANS_STATE_COMMIT_START;
3978 wake_up(&root->fs_info->transaction_blocked_wait);
3980 btrfs_evict_pending_snapshots(cur_trans);
3982 cur_trans->state = TRANS_STATE_UNBLOCKED;
3983 wake_up(&root->fs_info->transaction_wait);
3985 btrfs_destroy_delayed_inodes(root);
3986 btrfs_assert_delayed_root_empty(root);
3988 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3990 btrfs_destroy_pinned_extent(root,
3991 root->fs_info->pinned_extents);
3993 cur_trans->state =TRANS_STATE_COMPLETED;
3994 wake_up(&cur_trans->commit_wait);
3997 memset(cur_trans, 0, sizeof(*cur_trans));
3998 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4002 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4004 struct btrfs_transaction *t;
4007 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4009 spin_lock(&root->fs_info->trans_lock);
4010 list_splice_init(&root->fs_info->trans_list, &list);
4011 root->fs_info->running_transaction = NULL;
4012 spin_unlock(&root->fs_info->trans_lock);
4014 while (!list_empty(&list)) {
4015 t = list_entry(list.next, struct btrfs_transaction, list);
4017 btrfs_destroy_ordered_operations(t, root);
4019 btrfs_destroy_all_ordered_extents(root->fs_info);
4021 btrfs_destroy_delayed_refs(t, root);
4024 * FIXME: cleanup wait for commit
4025 * We needn't acquire the lock here, because we are during
4026 * the umount, there is no other task which will change it.
4028 t->state = TRANS_STATE_COMMIT_START;
4030 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4031 wake_up(&root->fs_info->transaction_blocked_wait);
4033 btrfs_evict_pending_snapshots(t);
4035 t->state = TRANS_STATE_UNBLOCKED;
4037 if (waitqueue_active(&root->fs_info->transaction_wait))
4038 wake_up(&root->fs_info->transaction_wait);
4040 btrfs_destroy_delayed_inodes(root);
4041 btrfs_assert_delayed_root_empty(root);
4043 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4045 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4048 btrfs_destroy_pinned_extent(root,
4049 root->fs_info->pinned_extents);
4051 t->state = TRANS_STATE_COMPLETED;
4053 if (waitqueue_active(&t->commit_wait))
4054 wake_up(&t->commit_wait);
4056 atomic_set(&t->use_count, 0);
4057 list_del_init(&t->list);
4058 memset(t, 0, sizeof(*t));
4059 kmem_cache_free(btrfs_transaction_cachep, t);
4062 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4067 static struct extent_io_ops btree_extent_io_ops = {
4068 .readpage_end_io_hook = btree_readpage_end_io_hook,
4069 .readpage_io_failed_hook = btree_io_failed_hook,
4070 .submit_bio_hook = btree_submit_bio_hook,
4071 /* note we're sharing with inode.c for the merge bio hook */
4072 .merge_bio_hook = btrfs_merge_bio_hook,