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 <linux/semaphore.h>
35 #include <asm/unaligned.h>
39 #include "transaction.h"
40 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "async-thread.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
48 #include "check-integrity.h"
49 #include "rcu-string.h"
50 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static struct extent_io_ops btree_extent_io_ops;
58 static void end_workqueue_fn(struct btrfs_work *work);
59 static void free_fs_root(struct btrfs_root *root);
60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
62 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
64 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
67 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
68 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
69 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
72 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
74 static int btrfs_cleanup_transaction(struct btrfs_root *root);
75 static void btrfs_error_commit_super(struct btrfs_root *root);
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_fs_info *info;
89 struct list_head list;
90 struct btrfs_work work;
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
98 struct async_submit_bio {
101 struct list_head list;
102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
106 unsigned long bio_flags;
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
112 struct btrfs_work work;
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
140 # if BTRFS_MAX_LEVEL != 8
144 static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149 } btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
160 { .id = 0, .name_stem = "tree" },
163 void __init btrfs_init_lockdep(void)
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
177 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
180 struct btrfs_lockdep_keyset *ks;
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
199 static struct extent_map *btree_get_extent(struct inode *inode,
200 struct page *page, size_t pg_offset, u64 start, u64 len,
203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
207 read_lock(&em_tree->lock);
208 em = lookup_extent_mapping(em_tree, start, len);
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212 read_unlock(&em_tree->lock);
215 read_unlock(&em_tree->lock);
217 em = alloc_extent_map();
219 em = ERR_PTR(-ENOMEM);
224 em->block_len = (u64)-1;
226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
228 write_lock(&em_tree->lock);
229 ret = add_extent_mapping(em_tree, em, 0);
230 if (ret == -EEXIST) {
232 em = lookup_extent_mapping(em_tree, start, len);
239 write_unlock(&em_tree->lock);
245 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
247 return crc32c(seed, data, len);
250 void btrfs_csum_final(u32 crc, char *result)
252 put_unaligned_le32(~crc, result);
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
259 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
268 unsigned long map_start;
269 unsigned long map_len;
272 unsigned long inline_result;
274 len = buf->len - offset;
276 err = map_private_extent_buffer(buf, offset, 32,
277 &kaddr, &map_start, &map_len);
280 cur_len = min(len, map_len - (offset - map_start));
281 crc = btrfs_csum_data(kaddr + offset - map_start,
286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
291 result = (char *)&inline_result;
294 btrfs_csum_final(crc, result);
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
300 memcpy(&found, result, csum_size);
302 read_extent_buffer(buf, &val, 0, csum_size);
303 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
304 "failed on %llu wanted %X found %X "
306 root->fs_info->sb->s_id,
307 (unsigned long long)buf->start, val, found,
308 btrfs_header_level(buf));
309 if (result != (char *)&inline_result)
314 write_extent_buffer(buf, result, 0, csum_size);
316 if (result != (char *)&inline_result)
322 * we can't consider a given block up to date unless the transid of the
323 * block matches the transid in the parent node's pointer. This is how we
324 * detect blocks that either didn't get written at all or got written
325 * in the wrong place.
327 static int verify_parent_transid(struct extent_io_tree *io_tree,
328 struct extent_buffer *eb, u64 parent_transid,
331 struct extent_state *cached_state = NULL;
334 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
340 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
342 if (extent_buffer_uptodate(eb) &&
343 btrfs_header_generation(eb) == parent_transid) {
347 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
349 (unsigned long long)eb->start,
350 (unsigned long long)parent_transid,
351 (unsigned long long)btrfs_header_generation(eb));
353 clear_extent_buffer_uptodate(eb);
355 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
356 &cached_state, GFP_NOFS);
361 * Return 0 if the superblock checksum type matches the checksum value of that
362 * algorithm. Pass the raw disk superblock data.
364 static int btrfs_check_super_csum(char *raw_disk_sb)
366 struct btrfs_super_block *disk_sb =
367 (struct btrfs_super_block *)raw_disk_sb;
368 u16 csum_type = btrfs_super_csum_type(disk_sb);
371 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
373 const int csum_size = sizeof(crc);
374 char result[csum_size];
377 * The super_block structure does not span the whole
378 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
379 * is filled with zeros and is included in the checkum.
381 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
382 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
383 btrfs_csum_final(crc, result);
385 if (memcmp(raw_disk_sb, result, csum_size))
388 if (ret && btrfs_super_generation(disk_sb) < 10) {
389 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
394 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
395 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
404 * helper to read a given tree block, doing retries as required when
405 * the checksums don't match and we have alternate mirrors to try.
407 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
408 struct extent_buffer *eb,
409 u64 start, u64 parent_transid)
411 struct extent_io_tree *io_tree;
416 int failed_mirror = 0;
418 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
419 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
421 ret = read_extent_buffer_pages(io_tree, eb, start,
423 btree_get_extent, mirror_num);
425 if (!verify_parent_transid(io_tree, eb,
433 * This buffer's crc is fine, but its contents are corrupted, so
434 * there is no reason to read the other copies, they won't be
437 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
440 num_copies = btrfs_num_copies(root->fs_info,
445 if (!failed_mirror) {
447 failed_mirror = eb->read_mirror;
451 if (mirror_num == failed_mirror)
454 if (mirror_num > num_copies)
458 if (failed && !ret && failed_mirror)
459 repair_eb_io_failure(root, eb, failed_mirror);
465 * checksum a dirty tree block before IO. This has extra checks to make sure
466 * we only fill in the checksum field in the first page of a multi-page block
469 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
471 struct extent_io_tree *tree;
472 u64 start = page_offset(page);
474 struct extent_buffer *eb;
476 tree = &BTRFS_I(page->mapping->host)->io_tree;
478 eb = (struct extent_buffer *)page->private;
479 if (page != eb->pages[0])
481 found_start = btrfs_header_bytenr(eb);
482 if (found_start != start) {
486 if (!PageUptodate(page)) {
490 csum_tree_block(root, eb, 0);
494 static int check_tree_block_fsid(struct btrfs_root *root,
495 struct extent_buffer *eb)
497 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
498 u8 fsid[BTRFS_UUID_SIZE];
501 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
504 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
508 fs_devices = fs_devices->seed;
513 #define CORRUPT(reason, eb, root, slot) \
514 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
515 "root=%llu, slot=%d\n", reason, \
516 (unsigned long long)btrfs_header_bytenr(eb), \
517 (unsigned long long)root->objectid, slot)
519 static noinline int check_leaf(struct btrfs_root *root,
520 struct extent_buffer *leaf)
522 struct btrfs_key key;
523 struct btrfs_key leaf_key;
524 u32 nritems = btrfs_header_nritems(leaf);
530 /* Check the 0 item */
531 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
532 BTRFS_LEAF_DATA_SIZE(root)) {
533 CORRUPT("invalid item offset size pair", leaf, root, 0);
538 * Check to make sure each items keys are in the correct order and their
539 * offsets make sense. We only have to loop through nritems-1 because
540 * we check the current slot against the next slot, which verifies the
541 * next slot's offset+size makes sense and that the current's slot
544 for (slot = 0; slot < nritems - 1; slot++) {
545 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
546 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
548 /* Make sure the keys are in the right order */
549 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
550 CORRUPT("bad key order", leaf, root, slot);
555 * Make sure the offset and ends are right, remember that the
556 * item data starts at the end of the leaf and grows towards the
559 if (btrfs_item_offset_nr(leaf, slot) !=
560 btrfs_item_end_nr(leaf, slot + 1)) {
561 CORRUPT("slot offset bad", leaf, root, slot);
566 * Check to make sure that we don't point outside of the leaf,
567 * just incase all the items are consistent to eachother, but
568 * all point outside of the leaf.
570 if (btrfs_item_end_nr(leaf, slot) >
571 BTRFS_LEAF_DATA_SIZE(root)) {
572 CORRUPT("slot end outside of leaf", leaf, root, slot);
580 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
581 u64 phy_offset, struct page *page,
582 u64 start, u64 end, int mirror)
584 struct extent_io_tree *tree;
587 struct extent_buffer *eb;
588 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
595 tree = &BTRFS_I(page->mapping->host)->io_tree;
596 eb = (struct extent_buffer *)page->private;
598 /* the pending IO might have been the only thing that kept this buffer
599 * in memory. Make sure we have a ref for all this other checks
601 extent_buffer_get(eb);
603 reads_done = atomic_dec_and_test(&eb->io_pages);
607 eb->read_mirror = mirror;
608 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
613 found_start = btrfs_header_bytenr(eb);
614 if (found_start != eb->start) {
615 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
617 (unsigned long long)found_start,
618 (unsigned long long)eb->start);
622 if (check_tree_block_fsid(root, eb)) {
623 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
624 (unsigned long long)eb->start);
628 found_level = btrfs_header_level(eb);
629 if (found_level >= BTRFS_MAX_LEVEL) {
630 btrfs_info(root->fs_info, "bad tree block level %d\n",
631 (int)btrfs_header_level(eb));
636 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
639 ret = csum_tree_block(root, eb, 1);
646 * If this is a leaf block and it is corrupt, set the corrupt bit so
647 * that we don't try and read the other copies of this block, just
650 if (found_level == 0 && check_leaf(root, eb)) {
651 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
656 set_extent_buffer_uptodate(eb);
659 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
660 btree_readahead_hook(root, eb, eb->start, ret);
664 * our io error hook is going to dec the io pages
665 * again, we have to make sure it has something
668 atomic_inc(&eb->io_pages);
669 clear_extent_buffer_uptodate(eb);
671 free_extent_buffer(eb);
676 static int btree_io_failed_hook(struct page *page, int failed_mirror)
678 struct extent_buffer *eb;
679 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
681 eb = (struct extent_buffer *)page->private;
682 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
683 eb->read_mirror = failed_mirror;
684 atomic_dec(&eb->io_pages);
685 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
686 btree_readahead_hook(root, eb, eb->start, -EIO);
687 return -EIO; /* we fixed nothing */
690 static void end_workqueue_bio(struct bio *bio, int err)
692 struct end_io_wq *end_io_wq = bio->bi_private;
693 struct btrfs_fs_info *fs_info;
695 fs_info = end_io_wq->info;
696 end_io_wq->error = err;
697 end_io_wq->work.func = end_workqueue_fn;
698 end_io_wq->work.flags = 0;
700 if (bio->bi_rw & REQ_WRITE) {
701 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
702 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
704 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
705 btrfs_queue_worker(&fs_info->endio_freespace_worker,
707 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
708 btrfs_queue_worker(&fs_info->endio_raid56_workers,
711 btrfs_queue_worker(&fs_info->endio_write_workers,
714 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
715 btrfs_queue_worker(&fs_info->endio_raid56_workers,
717 else if (end_io_wq->metadata)
718 btrfs_queue_worker(&fs_info->endio_meta_workers,
721 btrfs_queue_worker(&fs_info->endio_workers,
727 * For the metadata arg you want
730 * 1 - if normal metadta
731 * 2 - if writing to the free space cache area
732 * 3 - raid parity work
734 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
737 struct end_io_wq *end_io_wq;
738 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
742 end_io_wq->private = bio->bi_private;
743 end_io_wq->end_io = bio->bi_end_io;
744 end_io_wq->info = info;
745 end_io_wq->error = 0;
746 end_io_wq->bio = bio;
747 end_io_wq->metadata = metadata;
749 bio->bi_private = end_io_wq;
750 bio->bi_end_io = end_workqueue_bio;
754 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
756 unsigned long limit = min_t(unsigned long,
757 info->workers.max_workers,
758 info->fs_devices->open_devices);
762 static void run_one_async_start(struct btrfs_work *work)
764 struct async_submit_bio *async;
767 async = container_of(work, struct async_submit_bio, work);
768 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
769 async->mirror_num, async->bio_flags,
775 static void run_one_async_done(struct btrfs_work *work)
777 struct btrfs_fs_info *fs_info;
778 struct async_submit_bio *async;
781 async = container_of(work, struct async_submit_bio, work);
782 fs_info = BTRFS_I(async->inode)->root->fs_info;
784 limit = btrfs_async_submit_limit(fs_info);
785 limit = limit * 2 / 3;
787 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
788 waitqueue_active(&fs_info->async_submit_wait))
789 wake_up(&fs_info->async_submit_wait);
791 /* If an error occured we just want to clean up the bio and move on */
793 bio_endio(async->bio, async->error);
797 async->submit_bio_done(async->inode, async->rw, async->bio,
798 async->mirror_num, async->bio_flags,
802 static void run_one_async_free(struct btrfs_work *work)
804 struct async_submit_bio *async;
806 async = container_of(work, struct async_submit_bio, work);
810 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
811 int rw, struct bio *bio, int mirror_num,
812 unsigned long bio_flags,
814 extent_submit_bio_hook_t *submit_bio_start,
815 extent_submit_bio_hook_t *submit_bio_done)
817 struct async_submit_bio *async;
819 async = kmalloc(sizeof(*async), GFP_NOFS);
823 async->inode = inode;
826 async->mirror_num = mirror_num;
827 async->submit_bio_start = submit_bio_start;
828 async->submit_bio_done = submit_bio_done;
830 async->work.func = run_one_async_start;
831 async->work.ordered_func = run_one_async_done;
832 async->work.ordered_free = run_one_async_free;
834 async->work.flags = 0;
835 async->bio_flags = bio_flags;
836 async->bio_offset = bio_offset;
840 atomic_inc(&fs_info->nr_async_submits);
843 btrfs_set_work_high_prio(&async->work);
845 btrfs_queue_worker(&fs_info->workers, &async->work);
847 while (atomic_read(&fs_info->async_submit_draining) &&
848 atomic_read(&fs_info->nr_async_submits)) {
849 wait_event(fs_info->async_submit_wait,
850 (atomic_read(&fs_info->nr_async_submits) == 0));
856 static int btree_csum_one_bio(struct bio *bio)
858 struct bio_vec *bvec = bio->bi_io_vec;
860 struct btrfs_root *root;
863 WARN_ON(bio->bi_vcnt <= 0);
864 while (bio_index < bio->bi_vcnt) {
865 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
866 ret = csum_dirty_buffer(root, bvec->bv_page);
875 static int __btree_submit_bio_start(struct inode *inode, int rw,
876 struct bio *bio, int mirror_num,
877 unsigned long bio_flags,
881 * when we're called for a write, we're already in the async
882 * submission context. Just jump into btrfs_map_bio
884 return btree_csum_one_bio(bio);
887 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
888 int mirror_num, unsigned long bio_flags,
894 * when we're called for a write, we're already in the async
895 * submission context. Just jump into btrfs_map_bio
897 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
903 static int check_async_write(struct inode *inode, unsigned long bio_flags)
905 if (bio_flags & EXTENT_BIO_TREE_LOG)
914 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
915 int mirror_num, unsigned long bio_flags,
918 int async = check_async_write(inode, bio_flags);
921 if (!(rw & REQ_WRITE)) {
923 * called for a read, do the setup so that checksum validation
924 * can happen in the async kernel threads
926 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
930 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
933 ret = btree_csum_one_bio(bio);
936 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
940 * kthread helpers are used to submit writes so that
941 * checksumming can happen in parallel across all CPUs
943 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
944 inode, rw, bio, mirror_num, 0,
946 __btree_submit_bio_start,
947 __btree_submit_bio_done);
957 #ifdef CONFIG_MIGRATION
958 static int btree_migratepage(struct address_space *mapping,
959 struct page *newpage, struct page *page,
960 enum migrate_mode mode)
963 * we can't safely write a btree page from here,
964 * we haven't done the locking hook
969 * Buffers may be managed in a filesystem specific way.
970 * We must have no buffers or drop them.
972 if (page_has_private(page) &&
973 !try_to_release_page(page, GFP_KERNEL))
975 return migrate_page(mapping, newpage, page, mode);
980 static int btree_writepages(struct address_space *mapping,
981 struct writeback_control *wbc)
983 struct extent_io_tree *tree;
984 struct btrfs_fs_info *fs_info;
987 tree = &BTRFS_I(mapping->host)->io_tree;
988 if (wbc->sync_mode == WB_SYNC_NONE) {
990 if (wbc->for_kupdate)
993 fs_info = BTRFS_I(mapping->host)->root->fs_info;
994 /* this is a bit racy, but that's ok */
995 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
996 BTRFS_DIRTY_METADATA_THRESH);
1000 return btree_write_cache_pages(mapping, wbc);
1003 static int btree_readpage(struct file *file, struct page *page)
1005 struct extent_io_tree *tree;
1006 tree = &BTRFS_I(page->mapping->host)->io_tree;
1007 return extent_read_full_page(tree, page, btree_get_extent, 0);
1010 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1012 if (PageWriteback(page) || PageDirty(page))
1015 return try_release_extent_buffer(page);
1018 static void btree_invalidatepage(struct page *page, unsigned int offset,
1019 unsigned int length)
1021 struct extent_io_tree *tree;
1022 tree = &BTRFS_I(page->mapping->host)->io_tree;
1023 extent_invalidatepage(tree, page, offset);
1024 btree_releasepage(page, GFP_NOFS);
1025 if (PagePrivate(page)) {
1026 printk(KERN_WARNING "btrfs warning page private not zero "
1027 "on page %llu\n", (unsigned long long)page_offset(page));
1028 ClearPagePrivate(page);
1029 set_page_private(page, 0);
1030 page_cache_release(page);
1034 static int btree_set_page_dirty(struct page *page)
1037 struct extent_buffer *eb;
1039 BUG_ON(!PagePrivate(page));
1040 eb = (struct extent_buffer *)page->private;
1042 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1043 BUG_ON(!atomic_read(&eb->refs));
1044 btrfs_assert_tree_locked(eb);
1046 return __set_page_dirty_nobuffers(page);
1049 static const struct address_space_operations btree_aops = {
1050 .readpage = btree_readpage,
1051 .writepages = btree_writepages,
1052 .releasepage = btree_releasepage,
1053 .invalidatepage = btree_invalidatepage,
1054 #ifdef CONFIG_MIGRATION
1055 .migratepage = btree_migratepage,
1057 .set_page_dirty = btree_set_page_dirty,
1060 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1063 struct extent_buffer *buf = NULL;
1064 struct inode *btree_inode = root->fs_info->btree_inode;
1067 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1070 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1071 buf, 0, WAIT_NONE, btree_get_extent, 0);
1072 free_extent_buffer(buf);
1076 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1077 int mirror_num, struct extent_buffer **eb)
1079 struct extent_buffer *buf = NULL;
1080 struct inode *btree_inode = root->fs_info->btree_inode;
1081 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1084 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1088 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1090 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1091 btree_get_extent, mirror_num);
1093 free_extent_buffer(buf);
1097 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1098 free_extent_buffer(buf);
1100 } else if (extent_buffer_uptodate(buf)) {
1103 free_extent_buffer(buf);
1108 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1109 u64 bytenr, u32 blocksize)
1111 struct inode *btree_inode = root->fs_info->btree_inode;
1112 struct extent_buffer *eb;
1113 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1118 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1119 u64 bytenr, u32 blocksize)
1121 struct inode *btree_inode = root->fs_info->btree_inode;
1122 struct extent_buffer *eb;
1124 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1130 int btrfs_write_tree_block(struct extent_buffer *buf)
1132 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1133 buf->start + buf->len - 1);
1136 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1138 return filemap_fdatawait_range(buf->pages[0]->mapping,
1139 buf->start, buf->start + buf->len - 1);
1142 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1143 u32 blocksize, u64 parent_transid)
1145 struct extent_buffer *buf = NULL;
1148 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1152 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1154 free_extent_buffer(buf);
1161 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1162 struct extent_buffer *buf)
1164 struct btrfs_fs_info *fs_info = root->fs_info;
1166 if (btrfs_header_generation(buf) ==
1167 fs_info->running_transaction->transid) {
1168 btrfs_assert_tree_locked(buf);
1170 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1171 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1173 fs_info->dirty_metadata_batch);
1174 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1175 btrfs_set_lock_blocking(buf);
1176 clear_extent_buffer_dirty(buf);
1181 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1182 u32 stripesize, struct btrfs_root *root,
1183 struct btrfs_fs_info *fs_info,
1187 root->commit_root = NULL;
1188 root->sectorsize = sectorsize;
1189 root->nodesize = nodesize;
1190 root->leafsize = leafsize;
1191 root->stripesize = stripesize;
1193 root->track_dirty = 0;
1195 root->orphan_item_inserted = 0;
1196 root->orphan_cleanup_state = 0;
1198 root->objectid = objectid;
1199 root->last_trans = 0;
1200 root->highest_objectid = 0;
1201 root->nr_delalloc_inodes = 0;
1202 root->nr_ordered_extents = 0;
1204 root->inode_tree = RB_ROOT;
1205 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1206 root->block_rsv = NULL;
1207 root->orphan_block_rsv = NULL;
1209 INIT_LIST_HEAD(&root->dirty_list);
1210 INIT_LIST_HEAD(&root->root_list);
1211 INIT_LIST_HEAD(&root->delalloc_inodes);
1212 INIT_LIST_HEAD(&root->delalloc_root);
1213 INIT_LIST_HEAD(&root->ordered_extents);
1214 INIT_LIST_HEAD(&root->ordered_root);
1215 INIT_LIST_HEAD(&root->logged_list[0]);
1216 INIT_LIST_HEAD(&root->logged_list[1]);
1217 spin_lock_init(&root->orphan_lock);
1218 spin_lock_init(&root->inode_lock);
1219 spin_lock_init(&root->delalloc_lock);
1220 spin_lock_init(&root->ordered_extent_lock);
1221 spin_lock_init(&root->accounting_lock);
1222 spin_lock_init(&root->log_extents_lock[0]);
1223 spin_lock_init(&root->log_extents_lock[1]);
1224 mutex_init(&root->objectid_mutex);
1225 mutex_init(&root->log_mutex);
1226 init_waitqueue_head(&root->log_writer_wait);
1227 init_waitqueue_head(&root->log_commit_wait[0]);
1228 init_waitqueue_head(&root->log_commit_wait[1]);
1229 atomic_set(&root->log_commit[0], 0);
1230 atomic_set(&root->log_commit[1], 0);
1231 atomic_set(&root->log_writers, 0);
1232 atomic_set(&root->log_batch, 0);
1233 atomic_set(&root->orphan_inodes, 0);
1234 atomic_set(&root->refs, 1);
1235 root->log_transid = 0;
1236 root->last_log_commit = 0;
1237 extent_io_tree_init(&root->dirty_log_pages,
1238 fs_info->btree_inode->i_mapping);
1240 memset(&root->root_key, 0, sizeof(root->root_key));
1241 memset(&root->root_item, 0, sizeof(root->root_item));
1242 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1243 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1244 root->defrag_trans_start = fs_info->generation;
1245 init_completion(&root->kobj_unregister);
1246 root->defrag_running = 0;
1247 root->root_key.objectid = objectid;
1250 spin_lock_init(&root->root_item_lock);
1253 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1255 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1257 root->fs_info = fs_info;
1261 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1262 struct btrfs_fs_info *fs_info,
1265 struct extent_buffer *leaf;
1266 struct btrfs_root *tree_root = fs_info->tree_root;
1267 struct btrfs_root *root;
1268 struct btrfs_key key;
1273 root = btrfs_alloc_root(fs_info);
1275 return ERR_PTR(-ENOMEM);
1277 __setup_root(tree_root->nodesize, tree_root->leafsize,
1278 tree_root->sectorsize, tree_root->stripesize,
1279 root, fs_info, objectid);
1280 root->root_key.objectid = objectid;
1281 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1282 root->root_key.offset = 0;
1284 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1285 0, objectid, NULL, 0, 0, 0);
1287 ret = PTR_ERR(leaf);
1292 bytenr = leaf->start;
1293 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1294 btrfs_set_header_bytenr(leaf, leaf->start);
1295 btrfs_set_header_generation(leaf, trans->transid);
1296 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1297 btrfs_set_header_owner(leaf, objectid);
1300 write_extent_buffer(leaf, fs_info->fsid,
1301 (unsigned long)btrfs_header_fsid(leaf),
1303 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1304 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1306 btrfs_mark_buffer_dirty(leaf);
1308 root->commit_root = btrfs_root_node(root);
1309 root->track_dirty = 1;
1312 root->root_item.flags = 0;
1313 root->root_item.byte_limit = 0;
1314 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1315 btrfs_set_root_generation(&root->root_item, trans->transid);
1316 btrfs_set_root_level(&root->root_item, 0);
1317 btrfs_set_root_refs(&root->root_item, 1);
1318 btrfs_set_root_used(&root->root_item, leaf->len);
1319 btrfs_set_root_last_snapshot(&root->root_item, 0);
1320 btrfs_set_root_dirid(&root->root_item, 0);
1322 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1323 root->root_item.drop_level = 0;
1325 key.objectid = objectid;
1326 key.type = BTRFS_ROOT_ITEM_KEY;
1328 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1332 btrfs_tree_unlock(leaf);
1338 btrfs_tree_unlock(leaf);
1339 free_extent_buffer(leaf);
1343 return ERR_PTR(ret);
1346 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1347 struct btrfs_fs_info *fs_info)
1349 struct btrfs_root *root;
1350 struct btrfs_root *tree_root = fs_info->tree_root;
1351 struct extent_buffer *leaf;
1353 root = btrfs_alloc_root(fs_info);
1355 return ERR_PTR(-ENOMEM);
1357 __setup_root(tree_root->nodesize, tree_root->leafsize,
1358 tree_root->sectorsize, tree_root->stripesize,
1359 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1361 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1362 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1363 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1365 * log trees do not get reference counted because they go away
1366 * before a real commit is actually done. They do store pointers
1367 * to file data extents, and those reference counts still get
1368 * updated (along with back refs to the log tree).
1372 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1373 BTRFS_TREE_LOG_OBJECTID, NULL,
1377 return ERR_CAST(leaf);
1380 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1381 btrfs_set_header_bytenr(leaf, leaf->start);
1382 btrfs_set_header_generation(leaf, trans->transid);
1383 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1384 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1387 write_extent_buffer(root->node, root->fs_info->fsid,
1388 (unsigned long)btrfs_header_fsid(root->node),
1390 btrfs_mark_buffer_dirty(root->node);
1391 btrfs_tree_unlock(root->node);
1395 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1396 struct btrfs_fs_info *fs_info)
1398 struct btrfs_root *log_root;
1400 log_root = alloc_log_tree(trans, fs_info);
1401 if (IS_ERR(log_root))
1402 return PTR_ERR(log_root);
1403 WARN_ON(fs_info->log_root_tree);
1404 fs_info->log_root_tree = log_root;
1408 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1409 struct btrfs_root *root)
1411 struct btrfs_root *log_root;
1412 struct btrfs_inode_item *inode_item;
1414 log_root = alloc_log_tree(trans, root->fs_info);
1415 if (IS_ERR(log_root))
1416 return PTR_ERR(log_root);
1418 log_root->last_trans = trans->transid;
1419 log_root->root_key.offset = root->root_key.objectid;
1421 inode_item = &log_root->root_item.inode;
1422 btrfs_set_stack_inode_generation(inode_item, 1);
1423 btrfs_set_stack_inode_size(inode_item, 3);
1424 btrfs_set_stack_inode_nlink(inode_item, 1);
1425 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1426 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1428 btrfs_set_root_node(&log_root->root_item, log_root->node);
1430 WARN_ON(root->log_root);
1431 root->log_root = log_root;
1432 root->log_transid = 0;
1433 root->last_log_commit = 0;
1437 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1438 struct btrfs_key *key)
1440 struct btrfs_root *root;
1441 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1442 struct btrfs_path *path;
1447 path = btrfs_alloc_path();
1449 return ERR_PTR(-ENOMEM);
1451 root = btrfs_alloc_root(fs_info);
1457 __setup_root(tree_root->nodesize, tree_root->leafsize,
1458 tree_root->sectorsize, tree_root->stripesize,
1459 root, fs_info, key->objectid);
1461 ret = btrfs_find_root(tree_root, key, path,
1462 &root->root_item, &root->root_key);
1469 generation = btrfs_root_generation(&root->root_item);
1470 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1471 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1472 blocksize, generation);
1476 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1480 root->commit_root = btrfs_root_node(root);
1482 btrfs_free_path(path);
1486 free_extent_buffer(root->node);
1490 root = ERR_PTR(ret);
1494 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1495 struct btrfs_key *location)
1497 struct btrfs_root *root;
1499 root = btrfs_read_tree_root(tree_root, location);
1503 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1505 btrfs_check_and_init_root_item(&root->root_item);
1511 int btrfs_init_fs_root(struct btrfs_root *root)
1515 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1516 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1518 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1523 btrfs_init_free_ino_ctl(root);
1524 mutex_init(&root->fs_commit_mutex);
1525 spin_lock_init(&root->cache_lock);
1526 init_waitqueue_head(&root->cache_wait);
1528 ret = get_anon_bdev(&root->anon_dev);
1533 kfree(root->free_ino_ctl);
1534 kfree(root->free_ino_pinned);
1538 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1541 struct btrfs_root *root;
1543 spin_lock(&fs_info->fs_roots_radix_lock);
1544 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1545 (unsigned long)root_id);
1546 spin_unlock(&fs_info->fs_roots_radix_lock);
1550 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1551 struct btrfs_root *root)
1555 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1559 spin_lock(&fs_info->fs_roots_radix_lock);
1560 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1561 (unsigned long)root->root_key.objectid,
1565 spin_unlock(&fs_info->fs_roots_radix_lock);
1566 radix_tree_preload_end();
1571 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1572 struct btrfs_key *location)
1574 struct btrfs_root *root;
1577 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1578 return fs_info->tree_root;
1579 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1580 return fs_info->extent_root;
1581 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1582 return fs_info->chunk_root;
1583 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1584 return fs_info->dev_root;
1585 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1586 return fs_info->csum_root;
1587 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1588 return fs_info->quota_root ? fs_info->quota_root :
1590 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1591 return fs_info->uuid_root ? fs_info->uuid_root :
1594 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1598 root = btrfs_read_fs_root(fs_info->tree_root, location);
1602 if (btrfs_root_refs(&root->root_item) == 0) {
1607 ret = btrfs_init_fs_root(root);
1611 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1615 root->orphan_item_inserted = 1;
1617 ret = btrfs_insert_fs_root(fs_info, root);
1619 if (ret == -EEXIST) {
1628 return ERR_PTR(ret);
1631 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1633 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1635 struct btrfs_device *device;
1636 struct backing_dev_info *bdi;
1639 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1642 bdi = blk_get_backing_dev_info(device->bdev);
1643 if (bdi && bdi_congested(bdi, bdi_bits)) {
1653 * If this fails, caller must call bdi_destroy() to get rid of the
1656 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1660 bdi->capabilities = BDI_CAP_MAP_COPY;
1661 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1665 bdi->ra_pages = default_backing_dev_info.ra_pages;
1666 bdi->congested_fn = btrfs_congested_fn;
1667 bdi->congested_data = info;
1672 * called by the kthread helper functions to finally call the bio end_io
1673 * functions. This is where read checksum verification actually happens
1675 static void end_workqueue_fn(struct btrfs_work *work)
1678 struct end_io_wq *end_io_wq;
1679 struct btrfs_fs_info *fs_info;
1682 end_io_wq = container_of(work, struct end_io_wq, work);
1683 bio = end_io_wq->bio;
1684 fs_info = end_io_wq->info;
1686 error = end_io_wq->error;
1687 bio->bi_private = end_io_wq->private;
1688 bio->bi_end_io = end_io_wq->end_io;
1690 bio_endio(bio, error);
1693 static int cleaner_kthread(void *arg)
1695 struct btrfs_root *root = arg;
1701 /* Make the cleaner go to sleep early. */
1702 if (btrfs_need_cleaner_sleep(root))
1705 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1709 * Avoid the problem that we change the status of the fs
1710 * during the above check and trylock.
1712 if (btrfs_need_cleaner_sleep(root)) {
1713 mutex_unlock(&root->fs_info->cleaner_mutex);
1717 btrfs_run_delayed_iputs(root);
1718 again = btrfs_clean_one_deleted_snapshot(root);
1719 mutex_unlock(&root->fs_info->cleaner_mutex);
1722 * The defragger has dealt with the R/O remount and umount,
1723 * needn't do anything special here.
1725 btrfs_run_defrag_inodes(root->fs_info);
1727 if (!try_to_freeze() && !again) {
1728 set_current_state(TASK_INTERRUPTIBLE);
1729 if (!kthread_should_stop())
1731 __set_current_state(TASK_RUNNING);
1733 } while (!kthread_should_stop());
1737 static int transaction_kthread(void *arg)
1739 struct btrfs_root *root = arg;
1740 struct btrfs_trans_handle *trans;
1741 struct btrfs_transaction *cur;
1744 unsigned long delay;
1748 cannot_commit = false;
1749 delay = HZ * root->fs_info->commit_interval;
1750 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1752 spin_lock(&root->fs_info->trans_lock);
1753 cur = root->fs_info->running_transaction;
1755 spin_unlock(&root->fs_info->trans_lock);
1759 now = get_seconds();
1760 if (cur->state < TRANS_STATE_BLOCKED &&
1761 (now < cur->start_time ||
1762 now - cur->start_time < root->fs_info->commit_interval)) {
1763 spin_unlock(&root->fs_info->trans_lock);
1767 transid = cur->transid;
1768 spin_unlock(&root->fs_info->trans_lock);
1770 /* If the file system is aborted, this will always fail. */
1771 trans = btrfs_attach_transaction(root);
1772 if (IS_ERR(trans)) {
1773 if (PTR_ERR(trans) != -ENOENT)
1774 cannot_commit = true;
1777 if (transid == trans->transid) {
1778 btrfs_commit_transaction(trans, root);
1780 btrfs_end_transaction(trans, root);
1783 wake_up_process(root->fs_info->cleaner_kthread);
1784 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1786 if (!try_to_freeze()) {
1787 set_current_state(TASK_INTERRUPTIBLE);
1788 if (!kthread_should_stop() &&
1789 (!btrfs_transaction_blocked(root->fs_info) ||
1791 schedule_timeout(delay);
1792 __set_current_state(TASK_RUNNING);
1794 } while (!kthread_should_stop());
1799 * this will find the highest generation in the array of
1800 * root backups. The index of the highest array is returned,
1801 * or -1 if we can't find anything.
1803 * We check to make sure the array is valid by comparing the
1804 * generation of the latest root in the array with the generation
1805 * in the super block. If they don't match we pitch it.
1807 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1810 int newest_index = -1;
1811 struct btrfs_root_backup *root_backup;
1814 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1815 root_backup = info->super_copy->super_roots + i;
1816 cur = btrfs_backup_tree_root_gen(root_backup);
1817 if (cur == newest_gen)
1821 /* check to see if we actually wrapped around */
1822 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1823 root_backup = info->super_copy->super_roots;
1824 cur = btrfs_backup_tree_root_gen(root_backup);
1825 if (cur == newest_gen)
1828 return newest_index;
1833 * find the oldest backup so we know where to store new entries
1834 * in the backup array. This will set the backup_root_index
1835 * field in the fs_info struct
1837 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1840 int newest_index = -1;
1842 newest_index = find_newest_super_backup(info, newest_gen);
1843 /* if there was garbage in there, just move along */
1844 if (newest_index == -1) {
1845 info->backup_root_index = 0;
1847 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1852 * copy all the root pointers into the super backup array.
1853 * this will bump the backup pointer by one when it is
1856 static void backup_super_roots(struct btrfs_fs_info *info)
1859 struct btrfs_root_backup *root_backup;
1862 next_backup = info->backup_root_index;
1863 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1864 BTRFS_NUM_BACKUP_ROOTS;
1867 * just overwrite the last backup if we're at the same generation
1868 * this happens only at umount
1870 root_backup = info->super_for_commit->super_roots + last_backup;
1871 if (btrfs_backup_tree_root_gen(root_backup) ==
1872 btrfs_header_generation(info->tree_root->node))
1873 next_backup = last_backup;
1875 root_backup = info->super_for_commit->super_roots + next_backup;
1878 * make sure all of our padding and empty slots get zero filled
1879 * regardless of which ones we use today
1881 memset(root_backup, 0, sizeof(*root_backup));
1883 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1885 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1886 btrfs_set_backup_tree_root_gen(root_backup,
1887 btrfs_header_generation(info->tree_root->node));
1889 btrfs_set_backup_tree_root_level(root_backup,
1890 btrfs_header_level(info->tree_root->node));
1892 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1893 btrfs_set_backup_chunk_root_gen(root_backup,
1894 btrfs_header_generation(info->chunk_root->node));
1895 btrfs_set_backup_chunk_root_level(root_backup,
1896 btrfs_header_level(info->chunk_root->node));
1898 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1899 btrfs_set_backup_extent_root_gen(root_backup,
1900 btrfs_header_generation(info->extent_root->node));
1901 btrfs_set_backup_extent_root_level(root_backup,
1902 btrfs_header_level(info->extent_root->node));
1905 * we might commit during log recovery, which happens before we set
1906 * the fs_root. Make sure it is valid before we fill it in.
1908 if (info->fs_root && info->fs_root->node) {
1909 btrfs_set_backup_fs_root(root_backup,
1910 info->fs_root->node->start);
1911 btrfs_set_backup_fs_root_gen(root_backup,
1912 btrfs_header_generation(info->fs_root->node));
1913 btrfs_set_backup_fs_root_level(root_backup,
1914 btrfs_header_level(info->fs_root->node));
1917 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1918 btrfs_set_backup_dev_root_gen(root_backup,
1919 btrfs_header_generation(info->dev_root->node));
1920 btrfs_set_backup_dev_root_level(root_backup,
1921 btrfs_header_level(info->dev_root->node));
1923 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1924 btrfs_set_backup_csum_root_gen(root_backup,
1925 btrfs_header_generation(info->csum_root->node));
1926 btrfs_set_backup_csum_root_level(root_backup,
1927 btrfs_header_level(info->csum_root->node));
1929 btrfs_set_backup_total_bytes(root_backup,
1930 btrfs_super_total_bytes(info->super_copy));
1931 btrfs_set_backup_bytes_used(root_backup,
1932 btrfs_super_bytes_used(info->super_copy));
1933 btrfs_set_backup_num_devices(root_backup,
1934 btrfs_super_num_devices(info->super_copy));
1937 * if we don't copy this out to the super_copy, it won't get remembered
1938 * for the next commit
1940 memcpy(&info->super_copy->super_roots,
1941 &info->super_for_commit->super_roots,
1942 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1946 * this copies info out of the root backup array and back into
1947 * the in-memory super block. It is meant to help iterate through
1948 * the array, so you send it the number of backups you've already
1949 * tried and the last backup index you used.
1951 * this returns -1 when it has tried all the backups
1953 static noinline int next_root_backup(struct btrfs_fs_info *info,
1954 struct btrfs_super_block *super,
1955 int *num_backups_tried, int *backup_index)
1957 struct btrfs_root_backup *root_backup;
1958 int newest = *backup_index;
1960 if (*num_backups_tried == 0) {
1961 u64 gen = btrfs_super_generation(super);
1963 newest = find_newest_super_backup(info, gen);
1967 *backup_index = newest;
1968 *num_backups_tried = 1;
1969 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1970 /* we've tried all the backups, all done */
1973 /* jump to the next oldest backup */
1974 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1975 BTRFS_NUM_BACKUP_ROOTS;
1976 *backup_index = newest;
1977 *num_backups_tried += 1;
1979 root_backup = super->super_roots + newest;
1981 btrfs_set_super_generation(super,
1982 btrfs_backup_tree_root_gen(root_backup));
1983 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1984 btrfs_set_super_root_level(super,
1985 btrfs_backup_tree_root_level(root_backup));
1986 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1989 * fixme: the total bytes and num_devices need to match or we should
1992 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1993 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1997 /* helper to cleanup workers */
1998 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2000 btrfs_stop_workers(&fs_info->generic_worker);
2001 btrfs_stop_workers(&fs_info->fixup_workers);
2002 btrfs_stop_workers(&fs_info->delalloc_workers);
2003 btrfs_stop_workers(&fs_info->workers);
2004 btrfs_stop_workers(&fs_info->endio_workers);
2005 btrfs_stop_workers(&fs_info->endio_meta_workers);
2006 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2007 btrfs_stop_workers(&fs_info->rmw_workers);
2008 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2009 btrfs_stop_workers(&fs_info->endio_write_workers);
2010 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2011 btrfs_stop_workers(&fs_info->submit_workers);
2012 btrfs_stop_workers(&fs_info->delayed_workers);
2013 btrfs_stop_workers(&fs_info->caching_workers);
2014 btrfs_stop_workers(&fs_info->readahead_workers);
2015 btrfs_stop_workers(&fs_info->flush_workers);
2016 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2019 /* helper to cleanup tree roots */
2020 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2022 free_extent_buffer(info->tree_root->node);
2023 free_extent_buffer(info->tree_root->commit_root);
2024 info->tree_root->node = NULL;
2025 info->tree_root->commit_root = NULL;
2027 if (info->dev_root) {
2028 free_extent_buffer(info->dev_root->node);
2029 free_extent_buffer(info->dev_root->commit_root);
2030 info->dev_root->node = NULL;
2031 info->dev_root->commit_root = NULL;
2033 if (info->extent_root) {
2034 free_extent_buffer(info->extent_root->node);
2035 free_extent_buffer(info->extent_root->commit_root);
2036 info->extent_root->node = NULL;
2037 info->extent_root->commit_root = NULL;
2039 if (info->csum_root) {
2040 free_extent_buffer(info->csum_root->node);
2041 free_extent_buffer(info->csum_root->commit_root);
2042 info->csum_root->node = NULL;
2043 info->csum_root->commit_root = NULL;
2045 if (info->quota_root) {
2046 free_extent_buffer(info->quota_root->node);
2047 free_extent_buffer(info->quota_root->commit_root);
2048 info->quota_root->node = NULL;
2049 info->quota_root->commit_root = NULL;
2051 if (info->uuid_root) {
2052 free_extent_buffer(info->uuid_root->node);
2053 free_extent_buffer(info->uuid_root->commit_root);
2054 info->uuid_root->node = NULL;
2055 info->uuid_root->commit_root = NULL;
2058 free_extent_buffer(info->chunk_root->node);
2059 free_extent_buffer(info->chunk_root->commit_root);
2060 info->chunk_root->node = NULL;
2061 info->chunk_root->commit_root = NULL;
2065 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2068 struct btrfs_root *gang[8];
2071 while (!list_empty(&fs_info->dead_roots)) {
2072 gang[0] = list_entry(fs_info->dead_roots.next,
2073 struct btrfs_root, root_list);
2074 list_del(&gang[0]->root_list);
2076 if (gang[0]->in_radix) {
2077 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2079 free_extent_buffer(gang[0]->node);
2080 free_extent_buffer(gang[0]->commit_root);
2081 btrfs_put_fs_root(gang[0]);
2086 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2091 for (i = 0; i < ret; i++)
2092 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2096 int open_ctree(struct super_block *sb,
2097 struct btrfs_fs_devices *fs_devices,
2107 struct btrfs_key location;
2108 struct buffer_head *bh;
2109 struct btrfs_super_block *disk_super;
2110 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2111 struct btrfs_root *tree_root;
2112 struct btrfs_root *extent_root;
2113 struct btrfs_root *csum_root;
2114 struct btrfs_root *chunk_root;
2115 struct btrfs_root *dev_root;
2116 struct btrfs_root *quota_root;
2117 struct btrfs_root *uuid_root;
2118 struct btrfs_root *log_tree_root;
2121 int num_backups_tried = 0;
2122 int backup_index = 0;
2123 bool create_uuid_tree;
2124 bool check_uuid_tree;
2126 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2127 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2128 if (!tree_root || !chunk_root) {
2133 ret = init_srcu_struct(&fs_info->subvol_srcu);
2139 ret = setup_bdi(fs_info, &fs_info->bdi);
2145 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2150 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2151 (1 + ilog2(nr_cpu_ids));
2153 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2156 goto fail_dirty_metadata_bytes;
2159 fs_info->btree_inode = new_inode(sb);
2160 if (!fs_info->btree_inode) {
2162 goto fail_delalloc_bytes;
2165 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2167 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2168 INIT_LIST_HEAD(&fs_info->trans_list);
2169 INIT_LIST_HEAD(&fs_info->dead_roots);
2170 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2171 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2172 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2173 spin_lock_init(&fs_info->delalloc_root_lock);
2174 spin_lock_init(&fs_info->trans_lock);
2175 spin_lock_init(&fs_info->fs_roots_radix_lock);
2176 spin_lock_init(&fs_info->delayed_iput_lock);
2177 spin_lock_init(&fs_info->defrag_inodes_lock);
2178 spin_lock_init(&fs_info->free_chunk_lock);
2179 spin_lock_init(&fs_info->tree_mod_seq_lock);
2180 spin_lock_init(&fs_info->super_lock);
2181 rwlock_init(&fs_info->tree_mod_log_lock);
2182 mutex_init(&fs_info->reloc_mutex);
2183 seqlock_init(&fs_info->profiles_lock);
2185 init_completion(&fs_info->kobj_unregister);
2186 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2187 INIT_LIST_HEAD(&fs_info->space_info);
2188 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2189 btrfs_mapping_init(&fs_info->mapping_tree);
2190 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2191 BTRFS_BLOCK_RSV_GLOBAL);
2192 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2193 BTRFS_BLOCK_RSV_DELALLOC);
2194 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2195 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2196 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2197 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2198 BTRFS_BLOCK_RSV_DELOPS);
2199 atomic_set(&fs_info->nr_async_submits, 0);
2200 atomic_set(&fs_info->async_delalloc_pages, 0);
2201 atomic_set(&fs_info->async_submit_draining, 0);
2202 atomic_set(&fs_info->nr_async_bios, 0);
2203 atomic_set(&fs_info->defrag_running, 0);
2204 atomic64_set(&fs_info->tree_mod_seq, 0);
2206 fs_info->max_inline = 8192 * 1024;
2207 fs_info->metadata_ratio = 0;
2208 fs_info->defrag_inodes = RB_ROOT;
2209 fs_info->free_chunk_space = 0;
2210 fs_info->tree_mod_log = RB_ROOT;
2211 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2213 /* readahead state */
2214 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2215 spin_lock_init(&fs_info->reada_lock);
2217 fs_info->thread_pool_size = min_t(unsigned long,
2218 num_online_cpus() + 2, 8);
2220 INIT_LIST_HEAD(&fs_info->ordered_roots);
2221 spin_lock_init(&fs_info->ordered_root_lock);
2222 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2224 if (!fs_info->delayed_root) {
2228 btrfs_init_delayed_root(fs_info->delayed_root);
2230 mutex_init(&fs_info->scrub_lock);
2231 atomic_set(&fs_info->scrubs_running, 0);
2232 atomic_set(&fs_info->scrub_pause_req, 0);
2233 atomic_set(&fs_info->scrubs_paused, 0);
2234 atomic_set(&fs_info->scrub_cancel_req, 0);
2235 init_waitqueue_head(&fs_info->scrub_pause_wait);
2236 init_rwsem(&fs_info->scrub_super_lock);
2237 fs_info->scrub_workers_refcnt = 0;
2238 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2239 fs_info->check_integrity_print_mask = 0;
2242 spin_lock_init(&fs_info->balance_lock);
2243 mutex_init(&fs_info->balance_mutex);
2244 atomic_set(&fs_info->balance_running, 0);
2245 atomic_set(&fs_info->balance_pause_req, 0);
2246 atomic_set(&fs_info->balance_cancel_req, 0);
2247 fs_info->balance_ctl = NULL;
2248 init_waitqueue_head(&fs_info->balance_wait_q);
2250 sb->s_blocksize = 4096;
2251 sb->s_blocksize_bits = blksize_bits(4096);
2252 sb->s_bdi = &fs_info->bdi;
2254 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2255 set_nlink(fs_info->btree_inode, 1);
2257 * we set the i_size on the btree inode to the max possible int.
2258 * the real end of the address space is determined by all of
2259 * the devices in the system
2261 fs_info->btree_inode->i_size = OFFSET_MAX;
2262 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2263 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2265 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2266 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2267 fs_info->btree_inode->i_mapping);
2268 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2269 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2271 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2273 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2274 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2275 sizeof(struct btrfs_key));
2276 set_bit(BTRFS_INODE_DUMMY,
2277 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2278 insert_inode_hash(fs_info->btree_inode);
2280 spin_lock_init(&fs_info->block_group_cache_lock);
2281 fs_info->block_group_cache_tree = RB_ROOT;
2282 fs_info->first_logical_byte = (u64)-1;
2284 extent_io_tree_init(&fs_info->freed_extents[0],
2285 fs_info->btree_inode->i_mapping);
2286 extent_io_tree_init(&fs_info->freed_extents[1],
2287 fs_info->btree_inode->i_mapping);
2288 fs_info->pinned_extents = &fs_info->freed_extents[0];
2289 fs_info->do_barriers = 1;
2292 mutex_init(&fs_info->ordered_operations_mutex);
2293 mutex_init(&fs_info->ordered_extent_flush_mutex);
2294 mutex_init(&fs_info->tree_log_mutex);
2295 mutex_init(&fs_info->chunk_mutex);
2296 mutex_init(&fs_info->transaction_kthread_mutex);
2297 mutex_init(&fs_info->cleaner_mutex);
2298 mutex_init(&fs_info->volume_mutex);
2299 init_rwsem(&fs_info->extent_commit_sem);
2300 init_rwsem(&fs_info->cleanup_work_sem);
2301 init_rwsem(&fs_info->subvol_sem);
2302 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2303 fs_info->dev_replace.lock_owner = 0;
2304 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2305 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2306 mutex_init(&fs_info->dev_replace.lock_management_lock);
2307 mutex_init(&fs_info->dev_replace.lock);
2309 spin_lock_init(&fs_info->qgroup_lock);
2310 mutex_init(&fs_info->qgroup_ioctl_lock);
2311 fs_info->qgroup_tree = RB_ROOT;
2312 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2313 fs_info->qgroup_seq = 1;
2314 fs_info->quota_enabled = 0;
2315 fs_info->pending_quota_state = 0;
2316 fs_info->qgroup_ulist = NULL;
2317 mutex_init(&fs_info->qgroup_rescan_lock);
2319 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2320 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2322 init_waitqueue_head(&fs_info->transaction_throttle);
2323 init_waitqueue_head(&fs_info->transaction_wait);
2324 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2325 init_waitqueue_head(&fs_info->async_submit_wait);
2327 ret = btrfs_alloc_stripe_hash_table(fs_info);
2333 __setup_root(4096, 4096, 4096, 4096, tree_root,
2334 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2336 invalidate_bdev(fs_devices->latest_bdev);
2339 * Read super block and check the signature bytes only
2341 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2348 * We want to check superblock checksum, the type is stored inside.
2349 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2351 if (btrfs_check_super_csum(bh->b_data)) {
2352 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2358 * super_copy is zeroed at allocation time and we never touch the
2359 * following bytes up to INFO_SIZE, the checksum is calculated from
2360 * the whole block of INFO_SIZE
2362 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2363 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2364 sizeof(*fs_info->super_for_commit));
2367 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2369 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2371 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2376 disk_super = fs_info->super_copy;
2377 if (!btrfs_super_root(disk_super))
2380 /* check FS state, whether FS is broken. */
2381 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2382 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2385 * run through our array of backup supers and setup
2386 * our ring pointer to the oldest one
2388 generation = btrfs_super_generation(disk_super);
2389 find_oldest_super_backup(fs_info, generation);
2392 * In the long term, we'll store the compression type in the super
2393 * block, and it'll be used for per file compression control.
2395 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2397 ret = btrfs_parse_options(tree_root, options);
2403 features = btrfs_super_incompat_flags(disk_super) &
2404 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2406 printk(KERN_ERR "BTRFS: couldn't mount because of "
2407 "unsupported optional features (%Lx).\n",
2408 (unsigned long long)features);
2413 if (btrfs_super_leafsize(disk_super) !=
2414 btrfs_super_nodesize(disk_super)) {
2415 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2416 "blocksizes don't match. node %d leaf %d\n",
2417 btrfs_super_nodesize(disk_super),
2418 btrfs_super_leafsize(disk_super));
2422 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2423 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2424 "blocksize (%d) was too large\n",
2425 btrfs_super_leafsize(disk_super));
2430 features = btrfs_super_incompat_flags(disk_super);
2431 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2432 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2433 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2435 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2436 printk(KERN_ERR "btrfs: has skinny extents\n");
2439 * flag our filesystem as having big metadata blocks if
2440 * they are bigger than the page size
2442 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2443 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2444 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2445 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2448 nodesize = btrfs_super_nodesize(disk_super);
2449 leafsize = btrfs_super_leafsize(disk_super);
2450 sectorsize = btrfs_super_sectorsize(disk_super);
2451 stripesize = btrfs_super_stripesize(disk_super);
2452 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2453 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2456 * mixed block groups end up with duplicate but slightly offset
2457 * extent buffers for the same range. It leads to corruptions
2459 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2460 (sectorsize != leafsize)) {
2461 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2462 "are not allowed for mixed block groups on %s\n",
2468 * Needn't use the lock because there is no other task which will
2471 btrfs_set_super_incompat_flags(disk_super, features);
2473 features = btrfs_super_compat_ro_flags(disk_super) &
2474 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2475 if (!(sb->s_flags & MS_RDONLY) && features) {
2476 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2477 "unsupported option features (%Lx).\n",
2478 (unsigned long long)features);
2483 btrfs_init_workers(&fs_info->generic_worker,
2484 "genwork", 1, NULL);
2486 btrfs_init_workers(&fs_info->workers, "worker",
2487 fs_info->thread_pool_size,
2488 &fs_info->generic_worker);
2490 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2491 fs_info->thread_pool_size,
2492 &fs_info->generic_worker);
2494 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2495 fs_info->thread_pool_size,
2496 &fs_info->generic_worker);
2498 btrfs_init_workers(&fs_info->submit_workers, "submit",
2499 min_t(u64, fs_devices->num_devices,
2500 fs_info->thread_pool_size),
2501 &fs_info->generic_worker);
2503 btrfs_init_workers(&fs_info->caching_workers, "cache",
2504 2, &fs_info->generic_worker);
2506 /* a higher idle thresh on the submit workers makes it much more
2507 * likely that bios will be send down in a sane order to the
2510 fs_info->submit_workers.idle_thresh = 64;
2512 fs_info->workers.idle_thresh = 16;
2513 fs_info->workers.ordered = 1;
2515 fs_info->delalloc_workers.idle_thresh = 2;
2516 fs_info->delalloc_workers.ordered = 1;
2518 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2519 &fs_info->generic_worker);
2520 btrfs_init_workers(&fs_info->endio_workers, "endio",
2521 fs_info->thread_pool_size,
2522 &fs_info->generic_worker);
2523 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2524 fs_info->thread_pool_size,
2525 &fs_info->generic_worker);
2526 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2527 "endio-meta-write", fs_info->thread_pool_size,
2528 &fs_info->generic_worker);
2529 btrfs_init_workers(&fs_info->endio_raid56_workers,
2530 "endio-raid56", fs_info->thread_pool_size,
2531 &fs_info->generic_worker);
2532 btrfs_init_workers(&fs_info->rmw_workers,
2533 "rmw", fs_info->thread_pool_size,
2534 &fs_info->generic_worker);
2535 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2536 fs_info->thread_pool_size,
2537 &fs_info->generic_worker);
2538 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2539 1, &fs_info->generic_worker);
2540 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2541 fs_info->thread_pool_size,
2542 &fs_info->generic_worker);
2543 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2544 fs_info->thread_pool_size,
2545 &fs_info->generic_worker);
2546 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2547 &fs_info->generic_worker);
2550 * endios are largely parallel and should have a very
2553 fs_info->endio_workers.idle_thresh = 4;
2554 fs_info->endio_meta_workers.idle_thresh = 4;
2555 fs_info->endio_raid56_workers.idle_thresh = 4;
2556 fs_info->rmw_workers.idle_thresh = 2;
2558 fs_info->endio_write_workers.idle_thresh = 2;
2559 fs_info->endio_meta_write_workers.idle_thresh = 2;
2560 fs_info->readahead_workers.idle_thresh = 2;
2563 * btrfs_start_workers can really only fail because of ENOMEM so just
2564 * return -ENOMEM if any of these fail.
2566 ret = btrfs_start_workers(&fs_info->workers);
2567 ret |= btrfs_start_workers(&fs_info->generic_worker);
2568 ret |= btrfs_start_workers(&fs_info->submit_workers);
2569 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2570 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2571 ret |= btrfs_start_workers(&fs_info->endio_workers);
2572 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2573 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2574 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2575 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2576 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2577 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2578 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2579 ret |= btrfs_start_workers(&fs_info->caching_workers);
2580 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2581 ret |= btrfs_start_workers(&fs_info->flush_workers);
2582 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2585 goto fail_sb_buffer;
2588 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2589 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2590 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2592 tree_root->nodesize = nodesize;
2593 tree_root->leafsize = leafsize;
2594 tree_root->sectorsize = sectorsize;
2595 tree_root->stripesize = stripesize;
2597 sb->s_blocksize = sectorsize;
2598 sb->s_blocksize_bits = blksize_bits(sectorsize);
2600 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2601 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2602 goto fail_sb_buffer;
2605 if (sectorsize != PAGE_SIZE) {
2606 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2607 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2608 goto fail_sb_buffer;
2611 mutex_lock(&fs_info->chunk_mutex);
2612 ret = btrfs_read_sys_array(tree_root);
2613 mutex_unlock(&fs_info->chunk_mutex);
2615 printk(KERN_WARNING "btrfs: failed to read the system "
2616 "array on %s\n", sb->s_id);
2617 goto fail_sb_buffer;
2620 blocksize = btrfs_level_size(tree_root,
2621 btrfs_super_chunk_root_level(disk_super));
2622 generation = btrfs_super_chunk_root_generation(disk_super);
2624 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2625 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2627 chunk_root->node = read_tree_block(chunk_root,
2628 btrfs_super_chunk_root(disk_super),
2629 blocksize, generation);
2630 if (!chunk_root->node ||
2631 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2632 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2634 goto fail_tree_roots;
2636 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2637 chunk_root->commit_root = btrfs_root_node(chunk_root);
2639 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2640 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2643 ret = btrfs_read_chunk_tree(chunk_root);
2645 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2647 goto fail_tree_roots;
2651 * keep the device that is marked to be the target device for the
2652 * dev_replace procedure
2654 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2656 if (!fs_devices->latest_bdev) {
2657 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2659 goto fail_tree_roots;
2663 blocksize = btrfs_level_size(tree_root,
2664 btrfs_super_root_level(disk_super));
2665 generation = btrfs_super_generation(disk_super);
2667 tree_root->node = read_tree_block(tree_root,
2668 btrfs_super_root(disk_super),
2669 blocksize, generation);
2670 if (!tree_root->node ||
2671 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2672 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2675 goto recovery_tree_root;
2678 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2679 tree_root->commit_root = btrfs_root_node(tree_root);
2681 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2682 location.type = BTRFS_ROOT_ITEM_KEY;
2683 location.offset = 0;
2685 extent_root = btrfs_read_tree_root(tree_root, &location);
2686 if (IS_ERR(extent_root)) {
2687 ret = PTR_ERR(extent_root);
2688 goto recovery_tree_root;
2690 extent_root->track_dirty = 1;
2691 fs_info->extent_root = extent_root;
2693 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2694 dev_root = btrfs_read_tree_root(tree_root, &location);
2695 if (IS_ERR(dev_root)) {
2696 ret = PTR_ERR(dev_root);
2697 goto recovery_tree_root;
2699 dev_root->track_dirty = 1;
2700 fs_info->dev_root = dev_root;
2701 btrfs_init_devices_late(fs_info);
2703 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2704 csum_root = btrfs_read_tree_root(tree_root, &location);
2705 if (IS_ERR(csum_root)) {
2706 ret = PTR_ERR(csum_root);
2707 goto recovery_tree_root;
2709 csum_root->track_dirty = 1;
2710 fs_info->csum_root = csum_root;
2712 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2713 quota_root = btrfs_read_tree_root(tree_root, &location);
2714 if (!IS_ERR(quota_root)) {
2715 quota_root->track_dirty = 1;
2716 fs_info->quota_enabled = 1;
2717 fs_info->pending_quota_state = 1;
2718 fs_info->quota_root = quota_root;
2721 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2722 uuid_root = btrfs_read_tree_root(tree_root, &location);
2723 if (IS_ERR(uuid_root)) {
2724 ret = PTR_ERR(uuid_root);
2726 goto recovery_tree_root;
2727 create_uuid_tree = true;
2728 check_uuid_tree = false;
2730 uuid_root->track_dirty = 1;
2731 fs_info->uuid_root = uuid_root;
2732 create_uuid_tree = false;
2734 generation != btrfs_super_uuid_tree_generation(disk_super);
2737 fs_info->generation = generation;
2738 fs_info->last_trans_committed = generation;
2740 ret = btrfs_recover_balance(fs_info);
2742 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2743 goto fail_block_groups;
2746 ret = btrfs_init_dev_stats(fs_info);
2748 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2750 goto fail_block_groups;
2753 ret = btrfs_init_dev_replace(fs_info);
2755 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2756 goto fail_block_groups;
2759 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2761 ret = btrfs_init_space_info(fs_info);
2763 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2764 goto fail_block_groups;
2767 ret = btrfs_read_block_groups(extent_root);
2769 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2770 goto fail_block_groups;
2772 fs_info->num_tolerated_disk_barrier_failures =
2773 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2774 if (fs_info->fs_devices->missing_devices >
2775 fs_info->num_tolerated_disk_barrier_failures &&
2776 !(sb->s_flags & MS_RDONLY)) {
2778 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2779 goto fail_block_groups;
2782 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2784 if (IS_ERR(fs_info->cleaner_kthread))
2785 goto fail_block_groups;
2787 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2789 "btrfs-transaction");
2790 if (IS_ERR(fs_info->transaction_kthread))
2793 if (!btrfs_test_opt(tree_root, SSD) &&
2794 !btrfs_test_opt(tree_root, NOSSD) &&
2795 !fs_info->fs_devices->rotating) {
2796 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2798 btrfs_set_opt(fs_info->mount_opt, SSD);
2801 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2802 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2803 ret = btrfsic_mount(tree_root, fs_devices,
2804 btrfs_test_opt(tree_root,
2805 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2807 fs_info->check_integrity_print_mask);
2809 printk(KERN_WARNING "btrfs: failed to initialize"
2810 " integrity check module %s\n", sb->s_id);
2813 ret = btrfs_read_qgroup_config(fs_info);
2815 goto fail_trans_kthread;
2817 /* do not make disk changes in broken FS */
2818 if (btrfs_super_log_root(disk_super) != 0) {
2819 u64 bytenr = btrfs_super_log_root(disk_super);
2821 if (fs_devices->rw_devices == 0) {
2822 printk(KERN_WARNING "Btrfs log replay required "
2828 btrfs_level_size(tree_root,
2829 btrfs_super_log_root_level(disk_super));
2831 log_tree_root = btrfs_alloc_root(fs_info);
2832 if (!log_tree_root) {
2837 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2838 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2840 log_tree_root->node = read_tree_block(tree_root, bytenr,
2843 if (!log_tree_root->node ||
2844 !extent_buffer_uptodate(log_tree_root->node)) {
2845 printk(KERN_ERR "btrfs: failed to read log tree\n");
2846 free_extent_buffer(log_tree_root->node);
2847 kfree(log_tree_root);
2848 goto fail_trans_kthread;
2850 /* returns with log_tree_root freed on success */
2851 ret = btrfs_recover_log_trees(log_tree_root);
2853 btrfs_error(tree_root->fs_info, ret,
2854 "Failed to recover log tree");
2855 free_extent_buffer(log_tree_root->node);
2856 kfree(log_tree_root);
2857 goto fail_trans_kthread;
2860 if (sb->s_flags & MS_RDONLY) {
2861 ret = btrfs_commit_super(tree_root);
2863 goto fail_trans_kthread;
2867 ret = btrfs_find_orphan_roots(tree_root);
2869 goto fail_trans_kthread;
2871 if (!(sb->s_flags & MS_RDONLY)) {
2872 ret = btrfs_cleanup_fs_roots(fs_info);
2874 goto fail_trans_kthread;
2876 ret = btrfs_recover_relocation(tree_root);
2879 "btrfs: failed to recover relocation\n");
2885 location.objectid = BTRFS_FS_TREE_OBJECTID;
2886 location.type = BTRFS_ROOT_ITEM_KEY;
2887 location.offset = 0;
2889 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2890 if (IS_ERR(fs_info->fs_root)) {
2891 err = PTR_ERR(fs_info->fs_root);
2895 if (sb->s_flags & MS_RDONLY)
2898 down_read(&fs_info->cleanup_work_sem);
2899 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2900 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2901 up_read(&fs_info->cleanup_work_sem);
2902 close_ctree(tree_root);
2905 up_read(&fs_info->cleanup_work_sem);
2907 ret = btrfs_resume_balance_async(fs_info);
2909 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2910 close_ctree(tree_root);
2914 ret = btrfs_resume_dev_replace_async(fs_info);
2916 pr_warn("btrfs: failed to resume dev_replace\n");
2917 close_ctree(tree_root);
2921 btrfs_qgroup_rescan_resume(fs_info);
2923 if (create_uuid_tree) {
2924 pr_info("btrfs: creating UUID tree\n");
2925 ret = btrfs_create_uuid_tree(fs_info);
2927 pr_warn("btrfs: failed to create the UUID tree %d\n",
2929 close_ctree(tree_root);
2932 } else if (check_uuid_tree) {
2933 pr_info("btrfs: checking UUID tree\n");
2934 ret = btrfs_check_uuid_tree(fs_info);
2936 pr_warn("btrfs: failed to check the UUID tree %d\n",
2938 close_ctree(tree_root);
2942 fs_info->update_uuid_tree_gen = 1;
2948 btrfs_free_qgroup_config(fs_info);
2950 kthread_stop(fs_info->transaction_kthread);
2951 btrfs_cleanup_transaction(fs_info->tree_root);
2952 del_fs_roots(fs_info);
2954 kthread_stop(fs_info->cleaner_kthread);
2957 * make sure we're done with the btree inode before we stop our
2960 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2963 btrfs_put_block_group_cache(fs_info);
2964 btrfs_free_block_groups(fs_info);
2967 free_root_pointers(fs_info, 1);
2968 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2971 btrfs_stop_all_workers(fs_info);
2974 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2976 iput(fs_info->btree_inode);
2977 fail_delalloc_bytes:
2978 percpu_counter_destroy(&fs_info->delalloc_bytes);
2979 fail_dirty_metadata_bytes:
2980 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2982 bdi_destroy(&fs_info->bdi);
2984 cleanup_srcu_struct(&fs_info->subvol_srcu);
2986 btrfs_free_stripe_hash_table(fs_info);
2987 btrfs_close_devices(fs_info->fs_devices);
2991 if (!btrfs_test_opt(tree_root, RECOVERY))
2992 goto fail_tree_roots;
2994 free_root_pointers(fs_info, 0);
2996 /* don't use the log in recovery mode, it won't be valid */
2997 btrfs_set_super_log_root(disk_super, 0);
2999 /* we can't trust the free space cache either */
3000 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3002 ret = next_root_backup(fs_info, fs_info->super_copy,
3003 &num_backups_tried, &backup_index);
3005 goto fail_block_groups;
3006 goto retry_root_backup;
3009 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3012 set_buffer_uptodate(bh);
3014 struct btrfs_device *device = (struct btrfs_device *)
3017 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3018 "I/O error on %s\n",
3019 rcu_str_deref(device->name));
3020 /* note, we dont' set_buffer_write_io_error because we have
3021 * our own ways of dealing with the IO errors
3023 clear_buffer_uptodate(bh);
3024 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3030 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3032 struct buffer_head *bh;
3033 struct buffer_head *latest = NULL;
3034 struct btrfs_super_block *super;
3039 /* we would like to check all the supers, but that would make
3040 * a btrfs mount succeed after a mkfs from a different FS.
3041 * So, we need to add a special mount option to scan for
3042 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3044 for (i = 0; i < 1; i++) {
3045 bytenr = btrfs_sb_offset(i);
3046 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3047 i_size_read(bdev->bd_inode))
3049 bh = __bread(bdev, bytenr / 4096,
3050 BTRFS_SUPER_INFO_SIZE);
3054 super = (struct btrfs_super_block *)bh->b_data;
3055 if (btrfs_super_bytenr(super) != bytenr ||
3056 btrfs_super_magic(super) != BTRFS_MAGIC) {
3061 if (!latest || btrfs_super_generation(super) > transid) {
3064 transid = btrfs_super_generation(super);
3073 * this should be called twice, once with wait == 0 and
3074 * once with wait == 1. When wait == 0 is done, all the buffer heads
3075 * we write are pinned.
3077 * They are released when wait == 1 is done.
3078 * max_mirrors must be the same for both runs, and it indicates how
3079 * many supers on this one device should be written.
3081 * max_mirrors == 0 means to write them all.
3083 static int write_dev_supers(struct btrfs_device *device,
3084 struct btrfs_super_block *sb,
3085 int do_barriers, int wait, int max_mirrors)
3087 struct buffer_head *bh;
3094 if (max_mirrors == 0)
3095 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3097 for (i = 0; i < max_mirrors; i++) {
3098 bytenr = btrfs_sb_offset(i);
3099 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3103 bh = __find_get_block(device->bdev, bytenr / 4096,
3104 BTRFS_SUPER_INFO_SIZE);
3110 if (!buffer_uptodate(bh))
3113 /* drop our reference */
3116 /* drop the reference from the wait == 0 run */
3120 btrfs_set_super_bytenr(sb, bytenr);
3123 crc = btrfs_csum_data((char *)sb +
3124 BTRFS_CSUM_SIZE, crc,
3125 BTRFS_SUPER_INFO_SIZE -
3127 btrfs_csum_final(crc, sb->csum);
3130 * one reference for us, and we leave it for the
3133 bh = __getblk(device->bdev, bytenr / 4096,
3134 BTRFS_SUPER_INFO_SIZE);
3136 printk(KERN_ERR "btrfs: couldn't get super "
3137 "buffer head for bytenr %Lu\n", bytenr);
3142 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3144 /* one reference for submit_bh */
3147 set_buffer_uptodate(bh);
3149 bh->b_end_io = btrfs_end_buffer_write_sync;
3150 bh->b_private = device;
3154 * we fua the first super. The others we allow
3157 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3161 return errors < i ? 0 : -1;
3165 * endio for the write_dev_flush, this will wake anyone waiting
3166 * for the barrier when it is done
3168 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3171 if (err == -EOPNOTSUPP)
3172 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3173 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3175 if (bio->bi_private)
3176 complete(bio->bi_private);
3181 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3182 * sent down. With wait == 1, it waits for the previous flush.
3184 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3187 static int write_dev_flush(struct btrfs_device *device, int wait)
3192 if (device->nobarriers)
3196 bio = device->flush_bio;
3200 wait_for_completion(&device->flush_wait);
3202 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3203 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3204 rcu_str_deref(device->name));
3205 device->nobarriers = 1;
3206 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3208 btrfs_dev_stat_inc_and_print(device,
3209 BTRFS_DEV_STAT_FLUSH_ERRS);
3212 /* drop the reference from the wait == 0 run */
3214 device->flush_bio = NULL;
3220 * one reference for us, and we leave it for the
3223 device->flush_bio = NULL;
3224 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3228 bio->bi_end_io = btrfs_end_empty_barrier;
3229 bio->bi_bdev = device->bdev;
3230 init_completion(&device->flush_wait);
3231 bio->bi_private = &device->flush_wait;
3232 device->flush_bio = bio;
3235 btrfsic_submit_bio(WRITE_FLUSH, bio);
3241 * send an empty flush down to each device in parallel,
3242 * then wait for them
3244 static int barrier_all_devices(struct btrfs_fs_info *info)
3246 struct list_head *head;
3247 struct btrfs_device *dev;
3248 int errors_send = 0;
3249 int errors_wait = 0;
3252 /* send down all the barriers */
3253 head = &info->fs_devices->devices;
3254 list_for_each_entry_rcu(dev, head, dev_list) {
3259 if (!dev->in_fs_metadata || !dev->writeable)
3262 ret = write_dev_flush(dev, 0);
3267 /* wait for all the barriers */
3268 list_for_each_entry_rcu(dev, head, dev_list) {
3273 if (!dev->in_fs_metadata || !dev->writeable)
3276 ret = write_dev_flush(dev, 1);
3280 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3281 errors_wait > info->num_tolerated_disk_barrier_failures)
3286 int btrfs_calc_num_tolerated_disk_barrier_failures(
3287 struct btrfs_fs_info *fs_info)
3289 struct btrfs_ioctl_space_info space;
3290 struct btrfs_space_info *sinfo;
3291 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3292 BTRFS_BLOCK_GROUP_SYSTEM,
3293 BTRFS_BLOCK_GROUP_METADATA,
3294 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3298 int num_tolerated_disk_barrier_failures =
3299 (int)fs_info->fs_devices->num_devices;
3301 for (i = 0; i < num_types; i++) {
3302 struct btrfs_space_info *tmp;
3306 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3307 if (tmp->flags == types[i]) {
3317 down_read(&sinfo->groups_sem);
3318 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3319 if (!list_empty(&sinfo->block_groups[c])) {
3322 btrfs_get_block_group_info(
3323 &sinfo->block_groups[c], &space);
3324 if (space.total_bytes == 0 ||
3325 space.used_bytes == 0)
3327 flags = space.flags;
3330 * 0: if dup, single or RAID0 is configured for
3331 * any of metadata, system or data, else
3332 * 1: if RAID5 is configured, or if RAID1 or
3333 * RAID10 is configured and only two mirrors
3335 * 2: if RAID6 is configured, else
3336 * num_mirrors - 1: if RAID1 or RAID10 is
3337 * configured and more than
3338 * 2 mirrors are used.
3340 if (num_tolerated_disk_barrier_failures > 0 &&
3341 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3342 BTRFS_BLOCK_GROUP_RAID0)) ||
3343 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3345 num_tolerated_disk_barrier_failures = 0;
3346 else if (num_tolerated_disk_barrier_failures > 1) {
3347 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3348 BTRFS_BLOCK_GROUP_RAID5 |
3349 BTRFS_BLOCK_GROUP_RAID10)) {
3350 num_tolerated_disk_barrier_failures = 1;
3352 BTRFS_BLOCK_GROUP_RAID6) {
3353 num_tolerated_disk_barrier_failures = 2;
3358 up_read(&sinfo->groups_sem);
3361 return num_tolerated_disk_barrier_failures;
3364 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3366 struct list_head *head;
3367 struct btrfs_device *dev;
3368 struct btrfs_super_block *sb;
3369 struct btrfs_dev_item *dev_item;
3373 int total_errors = 0;
3376 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3377 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3378 backup_super_roots(root->fs_info);
3380 sb = root->fs_info->super_for_commit;
3381 dev_item = &sb->dev_item;
3383 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3384 head = &root->fs_info->fs_devices->devices;
3387 ret = barrier_all_devices(root->fs_info);
3390 &root->fs_info->fs_devices->device_list_mutex);
3391 btrfs_error(root->fs_info, ret,
3392 "errors while submitting device barriers.");
3397 list_for_each_entry_rcu(dev, head, dev_list) {
3402 if (!dev->in_fs_metadata || !dev->writeable)
3405 btrfs_set_stack_device_generation(dev_item, 0);
3406 btrfs_set_stack_device_type(dev_item, dev->type);
3407 btrfs_set_stack_device_id(dev_item, dev->devid);
3408 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3409 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3410 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3411 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3412 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3413 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3414 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3416 flags = btrfs_super_flags(sb);
3417 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3419 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3423 if (total_errors > max_errors) {
3424 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3427 /* This shouldn't happen. FUA is masked off if unsupported */
3432 list_for_each_entry_rcu(dev, head, dev_list) {
3435 if (!dev->in_fs_metadata || !dev->writeable)
3438 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3442 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3443 if (total_errors > max_errors) {
3444 btrfs_error(root->fs_info, -EIO,
3445 "%d errors while writing supers", total_errors);
3451 int write_ctree_super(struct btrfs_trans_handle *trans,
3452 struct btrfs_root *root, int max_mirrors)
3456 ret = write_all_supers(root, max_mirrors);
3460 /* Drop a fs root from the radix tree and free it. */
3461 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3462 struct btrfs_root *root)
3464 spin_lock(&fs_info->fs_roots_radix_lock);
3465 radix_tree_delete(&fs_info->fs_roots_radix,
3466 (unsigned long)root->root_key.objectid);
3467 spin_unlock(&fs_info->fs_roots_radix_lock);
3469 if (btrfs_root_refs(&root->root_item) == 0)
3470 synchronize_srcu(&fs_info->subvol_srcu);
3472 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3473 btrfs_free_log(NULL, root);
3474 btrfs_free_log_root_tree(NULL, fs_info);
3477 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3478 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3482 static void free_fs_root(struct btrfs_root *root)
3484 iput(root->cache_inode);
3485 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3487 free_anon_bdev(root->anon_dev);
3488 free_extent_buffer(root->node);
3489 free_extent_buffer(root->commit_root);
3490 kfree(root->free_ino_ctl);
3491 kfree(root->free_ino_pinned);
3493 btrfs_put_fs_root(root);
3496 void btrfs_free_fs_root(struct btrfs_root *root)
3501 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3503 u64 root_objectid = 0;
3504 struct btrfs_root *gang[8];
3509 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3510 (void **)gang, root_objectid,
3515 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3516 for (i = 0; i < ret; i++) {
3519 root_objectid = gang[i]->root_key.objectid;
3520 err = btrfs_orphan_cleanup(gang[i]);
3529 int btrfs_commit_super(struct btrfs_root *root)
3531 struct btrfs_trans_handle *trans;
3534 mutex_lock(&root->fs_info->cleaner_mutex);
3535 btrfs_run_delayed_iputs(root);
3536 mutex_unlock(&root->fs_info->cleaner_mutex);
3537 wake_up_process(root->fs_info->cleaner_kthread);
3539 /* wait until ongoing cleanup work done */
3540 down_write(&root->fs_info->cleanup_work_sem);
3541 up_write(&root->fs_info->cleanup_work_sem);
3543 trans = btrfs_join_transaction(root);
3545 return PTR_ERR(trans);
3546 ret = btrfs_commit_transaction(trans, root);
3549 /* run commit again to drop the original snapshot */
3550 trans = btrfs_join_transaction(root);
3552 return PTR_ERR(trans);
3553 ret = btrfs_commit_transaction(trans, root);
3556 ret = btrfs_write_and_wait_transaction(NULL, root);
3558 btrfs_error(root->fs_info, ret,
3559 "Failed to sync btree inode to disk.");
3563 ret = write_ctree_super(NULL, root, 0);
3567 int close_ctree(struct btrfs_root *root)
3569 struct btrfs_fs_info *fs_info = root->fs_info;
3572 fs_info->closing = 1;
3575 /* wait for the uuid_scan task to finish */
3576 down(&fs_info->uuid_tree_rescan_sem);
3577 /* avoid complains from lockdep et al., set sem back to initial state */
3578 up(&fs_info->uuid_tree_rescan_sem);
3580 /* pause restriper - we want to resume on mount */
3581 btrfs_pause_balance(fs_info);
3583 btrfs_dev_replace_suspend_for_unmount(fs_info);
3585 btrfs_scrub_cancel(fs_info);
3587 /* wait for any defraggers to finish */
3588 wait_event(fs_info->transaction_wait,
3589 (atomic_read(&fs_info->defrag_running) == 0));
3591 /* clear out the rbtree of defraggable inodes */
3592 btrfs_cleanup_defrag_inodes(fs_info);
3594 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3595 ret = btrfs_commit_super(root);
3597 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3600 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3601 btrfs_error_commit_super(root);
3603 btrfs_put_block_group_cache(fs_info);
3605 kthread_stop(fs_info->transaction_kthread);
3606 kthread_stop(fs_info->cleaner_kthread);
3608 fs_info->closing = 2;
3611 btrfs_free_qgroup_config(root->fs_info);
3613 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3614 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3615 percpu_counter_sum(&fs_info->delalloc_bytes));
3618 btrfs_free_block_groups(fs_info);
3620 btrfs_stop_all_workers(fs_info);
3622 del_fs_roots(fs_info);
3624 free_root_pointers(fs_info, 1);
3626 iput(fs_info->btree_inode);
3628 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3629 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3630 btrfsic_unmount(root, fs_info->fs_devices);
3633 btrfs_close_devices(fs_info->fs_devices);
3634 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3636 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3637 percpu_counter_destroy(&fs_info->delalloc_bytes);
3638 bdi_destroy(&fs_info->bdi);
3639 cleanup_srcu_struct(&fs_info->subvol_srcu);
3641 btrfs_free_stripe_hash_table(fs_info);
3646 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3650 struct inode *btree_inode = buf->pages[0]->mapping->host;
3652 ret = extent_buffer_uptodate(buf);
3656 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3657 parent_transid, atomic);
3663 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3665 return set_extent_buffer_uptodate(buf);
3668 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3670 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3671 u64 transid = btrfs_header_generation(buf);
3674 btrfs_assert_tree_locked(buf);
3675 if (transid != root->fs_info->generation)
3676 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3677 "found %llu running %llu\n",
3678 (unsigned long long)buf->start,
3679 (unsigned long long)transid,
3680 (unsigned long long)root->fs_info->generation);
3681 was_dirty = set_extent_buffer_dirty(buf);
3683 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3685 root->fs_info->dirty_metadata_batch);
3688 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3692 * looks as though older kernels can get into trouble with
3693 * this code, they end up stuck in balance_dirty_pages forever
3697 if (current->flags & PF_MEMALLOC)
3701 btrfs_balance_delayed_items(root);
3703 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3704 BTRFS_DIRTY_METADATA_THRESH);
3706 balance_dirty_pages_ratelimited(
3707 root->fs_info->btree_inode->i_mapping);
3712 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3714 __btrfs_btree_balance_dirty(root, 1);
3717 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3719 __btrfs_btree_balance_dirty(root, 0);
3722 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3724 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3725 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3728 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3732 * Placeholder for checks
3737 static void btrfs_error_commit_super(struct btrfs_root *root)
3739 mutex_lock(&root->fs_info->cleaner_mutex);
3740 btrfs_run_delayed_iputs(root);
3741 mutex_unlock(&root->fs_info->cleaner_mutex);
3743 down_write(&root->fs_info->cleanup_work_sem);
3744 up_write(&root->fs_info->cleanup_work_sem);
3746 /* cleanup FS via transaction */
3747 btrfs_cleanup_transaction(root);
3750 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3751 struct btrfs_root *root)
3753 struct btrfs_inode *btrfs_inode;
3754 struct list_head splice;
3756 INIT_LIST_HEAD(&splice);
3758 mutex_lock(&root->fs_info->ordered_operations_mutex);
3759 spin_lock(&root->fs_info->ordered_root_lock);
3761 list_splice_init(&t->ordered_operations, &splice);
3762 while (!list_empty(&splice)) {
3763 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3764 ordered_operations);
3766 list_del_init(&btrfs_inode->ordered_operations);
3767 spin_unlock(&root->fs_info->ordered_root_lock);
3769 btrfs_invalidate_inodes(btrfs_inode->root);
3771 spin_lock(&root->fs_info->ordered_root_lock);
3774 spin_unlock(&root->fs_info->ordered_root_lock);
3775 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3778 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3780 struct btrfs_ordered_extent *ordered;
3782 spin_lock(&root->ordered_extent_lock);
3784 * This will just short circuit the ordered completion stuff which will
3785 * make sure the ordered extent gets properly cleaned up.
3787 list_for_each_entry(ordered, &root->ordered_extents,
3789 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3790 spin_unlock(&root->ordered_extent_lock);
3793 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3795 struct btrfs_root *root;
3796 struct list_head splice;
3798 INIT_LIST_HEAD(&splice);
3800 spin_lock(&fs_info->ordered_root_lock);
3801 list_splice_init(&fs_info->ordered_roots, &splice);
3802 while (!list_empty(&splice)) {
3803 root = list_first_entry(&splice, struct btrfs_root,
3805 list_del_init(&root->ordered_root);
3807 btrfs_destroy_ordered_extents(root);
3809 cond_resched_lock(&fs_info->ordered_root_lock);
3811 spin_unlock(&fs_info->ordered_root_lock);
3814 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3815 struct btrfs_root *root)
3817 struct rb_node *node;
3818 struct btrfs_delayed_ref_root *delayed_refs;
3819 struct btrfs_delayed_ref_node *ref;
3822 delayed_refs = &trans->delayed_refs;
3824 spin_lock(&delayed_refs->lock);
3825 if (delayed_refs->num_entries == 0) {
3826 spin_unlock(&delayed_refs->lock);
3827 printk(KERN_INFO "delayed_refs has NO entry\n");
3831 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3832 struct btrfs_delayed_ref_head *head = NULL;
3833 bool pin_bytes = false;
3835 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3836 atomic_set(&ref->refs, 1);
3837 if (btrfs_delayed_ref_is_head(ref)) {
3839 head = btrfs_delayed_node_to_head(ref);
3840 if (!mutex_trylock(&head->mutex)) {
3841 atomic_inc(&ref->refs);
3842 spin_unlock(&delayed_refs->lock);
3844 /* Need to wait for the delayed ref to run */
3845 mutex_lock(&head->mutex);
3846 mutex_unlock(&head->mutex);
3847 btrfs_put_delayed_ref(ref);
3849 spin_lock(&delayed_refs->lock);
3853 if (head->must_insert_reserved)
3855 btrfs_free_delayed_extent_op(head->extent_op);
3856 delayed_refs->num_heads--;
3857 if (list_empty(&head->cluster))
3858 delayed_refs->num_heads_ready--;
3859 list_del_init(&head->cluster);
3863 rb_erase(&ref->rb_node, &delayed_refs->root);
3864 delayed_refs->num_entries--;
3865 spin_unlock(&delayed_refs->lock);
3868 btrfs_pin_extent(root, ref->bytenr,
3870 mutex_unlock(&head->mutex);
3872 btrfs_put_delayed_ref(ref);
3875 spin_lock(&delayed_refs->lock);
3878 spin_unlock(&delayed_refs->lock);
3883 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3885 struct btrfs_pending_snapshot *snapshot;
3886 struct list_head splice;
3888 INIT_LIST_HEAD(&splice);
3890 list_splice_init(&t->pending_snapshots, &splice);
3892 while (!list_empty(&splice)) {
3893 snapshot = list_entry(splice.next,
3894 struct btrfs_pending_snapshot,
3896 snapshot->error = -ECANCELED;
3897 list_del_init(&snapshot->list);
3901 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3903 struct btrfs_inode *btrfs_inode;
3904 struct list_head splice;
3906 INIT_LIST_HEAD(&splice);
3908 spin_lock(&root->delalloc_lock);
3909 list_splice_init(&root->delalloc_inodes, &splice);
3911 while (!list_empty(&splice)) {
3912 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3915 list_del_init(&btrfs_inode->delalloc_inodes);
3916 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3917 &btrfs_inode->runtime_flags);
3918 spin_unlock(&root->delalloc_lock);
3920 btrfs_invalidate_inodes(btrfs_inode->root);
3922 spin_lock(&root->delalloc_lock);
3925 spin_unlock(&root->delalloc_lock);
3928 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3930 struct btrfs_root *root;
3931 struct list_head splice;
3933 INIT_LIST_HEAD(&splice);
3935 spin_lock(&fs_info->delalloc_root_lock);
3936 list_splice_init(&fs_info->delalloc_roots, &splice);
3937 while (!list_empty(&splice)) {
3938 root = list_first_entry(&splice, struct btrfs_root,
3940 list_del_init(&root->delalloc_root);
3941 root = btrfs_grab_fs_root(root);
3943 spin_unlock(&fs_info->delalloc_root_lock);
3945 btrfs_destroy_delalloc_inodes(root);
3946 btrfs_put_fs_root(root);
3948 spin_lock(&fs_info->delalloc_root_lock);
3950 spin_unlock(&fs_info->delalloc_root_lock);
3953 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3954 struct extent_io_tree *dirty_pages,
3958 struct extent_buffer *eb;
3963 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3968 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3969 while (start <= end) {
3970 eb = btrfs_find_tree_block(root, start,
3972 start += root->leafsize;
3975 wait_on_extent_buffer_writeback(eb);
3977 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3979 clear_extent_buffer_dirty(eb);
3980 free_extent_buffer_stale(eb);
3987 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3988 struct extent_io_tree *pinned_extents)
3990 struct extent_io_tree *unpin;
3996 unpin = pinned_extents;
3999 ret = find_first_extent_bit(unpin, 0, &start, &end,
4000 EXTENT_DIRTY, NULL);
4005 if (btrfs_test_opt(root, DISCARD))
4006 ret = btrfs_error_discard_extent(root, start,
4010 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4011 btrfs_error_unpin_extent_range(root, start, end);
4016 if (unpin == &root->fs_info->freed_extents[0])
4017 unpin = &root->fs_info->freed_extents[1];
4019 unpin = &root->fs_info->freed_extents[0];
4027 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4028 struct btrfs_root *root)
4030 btrfs_destroy_delayed_refs(cur_trans, root);
4031 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4032 cur_trans->dirty_pages.dirty_bytes);
4034 cur_trans->state = TRANS_STATE_COMMIT_START;
4035 wake_up(&root->fs_info->transaction_blocked_wait);
4037 btrfs_evict_pending_snapshots(cur_trans);
4039 cur_trans->state = TRANS_STATE_UNBLOCKED;
4040 wake_up(&root->fs_info->transaction_wait);
4042 btrfs_destroy_delayed_inodes(root);
4043 btrfs_assert_delayed_root_empty(root);
4045 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4047 btrfs_destroy_pinned_extent(root,
4048 root->fs_info->pinned_extents);
4050 cur_trans->state =TRANS_STATE_COMPLETED;
4051 wake_up(&cur_trans->commit_wait);
4054 memset(cur_trans, 0, sizeof(*cur_trans));
4055 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4059 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4061 struct btrfs_transaction *t;
4064 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4066 spin_lock(&root->fs_info->trans_lock);
4067 list_splice_init(&root->fs_info->trans_list, &list);
4068 root->fs_info->running_transaction = NULL;
4069 spin_unlock(&root->fs_info->trans_lock);
4071 while (!list_empty(&list)) {
4072 t = list_entry(list.next, struct btrfs_transaction, list);
4074 btrfs_destroy_ordered_operations(t, root);
4076 btrfs_destroy_all_ordered_extents(root->fs_info);
4078 btrfs_destroy_delayed_refs(t, root);
4081 * FIXME: cleanup wait for commit
4082 * We needn't acquire the lock here, because we are during
4083 * the umount, there is no other task which will change it.
4085 t->state = TRANS_STATE_COMMIT_START;
4087 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4088 wake_up(&root->fs_info->transaction_blocked_wait);
4090 btrfs_evict_pending_snapshots(t);
4092 t->state = TRANS_STATE_UNBLOCKED;
4094 if (waitqueue_active(&root->fs_info->transaction_wait))
4095 wake_up(&root->fs_info->transaction_wait);
4097 btrfs_destroy_delayed_inodes(root);
4098 btrfs_assert_delayed_root_empty(root);
4100 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4102 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4105 btrfs_destroy_pinned_extent(root,
4106 root->fs_info->pinned_extents);
4108 t->state = TRANS_STATE_COMPLETED;
4110 if (waitqueue_active(&t->commit_wait))
4111 wake_up(&t->commit_wait);
4113 atomic_set(&t->use_count, 0);
4114 list_del_init(&t->list);
4115 memset(t, 0, sizeof(*t));
4116 kmem_cache_free(btrfs_transaction_cachep, t);
4119 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4124 static struct extent_io_ops btree_extent_io_ops = {
4125 .readpage_end_io_hook = btree_readpage_end_io_hook,
4126 .readpage_io_failed_hook = btree_io_failed_hook,
4127 .submit_bio_hook = btree_submit_bio_hook,
4128 /* note we're sharing with inode.c for the merge bio hook */
4129 .merge_bio_hook = btrfs_merge_bio_hook,